2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
42 #include <linux/btrfs.h>
43 #include <linux/blkdev.h>
44 #include <linux/posix_acl_xattr.h>
45 #include <linux/uio.h>
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
55 #include "compression.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
64 struct btrfs_iget_args
{
65 struct btrfs_key
*location
;
66 struct btrfs_root
*root
;
69 struct btrfs_dio_data
{
70 u64 outstanding_extents
;
72 u64 unsubmitted_oe_range_start
;
73 u64 unsubmitted_oe_range_end
;
76 static const struct inode_operations btrfs_dir_inode_operations
;
77 static const struct inode_operations btrfs_symlink_inode_operations
;
78 static const struct inode_operations btrfs_dir_ro_inode_operations
;
79 static const struct inode_operations btrfs_special_inode_operations
;
80 static const struct inode_operations btrfs_file_inode_operations
;
81 static const struct address_space_operations btrfs_aops
;
82 static const struct address_space_operations btrfs_symlink_aops
;
83 static const struct file_operations btrfs_dir_file_operations
;
84 static const struct extent_io_ops btrfs_extent_io_ops
;
86 static struct kmem_cache
*btrfs_inode_cachep
;
87 struct kmem_cache
*btrfs_trans_handle_cachep
;
88 struct kmem_cache
*btrfs_transaction_cachep
;
89 struct kmem_cache
*btrfs_path_cachep
;
90 struct kmem_cache
*btrfs_free_space_cachep
;
93 static const unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
94 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
95 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
96 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
97 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
98 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
99 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
100 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
103 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
);
104 static int btrfs_truncate(struct inode
*inode
);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
);
106 static noinline
int cow_file_range(struct inode
*inode
,
107 struct page
*locked_page
,
108 u64 start
, u64 end
, int *page_started
,
109 unsigned long *nr_written
, int unlock
);
110 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
111 u64 len
, u64 orig_start
,
112 u64 block_start
, u64 block_len
,
113 u64 orig_block_len
, u64 ram_bytes
,
116 static int btrfs_dirty_inode(struct inode
*inode
);
118 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
119 void btrfs_test_inode_set_ops(struct inode
*inode
)
121 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
125 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
126 struct inode
*inode
, struct inode
*dir
,
127 const struct qstr
*qstr
)
131 err
= btrfs_init_acl(trans
, inode
, dir
);
133 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
138 * this does all the hard work for inserting an inline extent into
139 * the btree. The caller should have done a btrfs_drop_extents so that
140 * no overlapping inline items exist in the btree
142 static int insert_inline_extent(struct btrfs_trans_handle
*trans
,
143 struct btrfs_path
*path
, int extent_inserted
,
144 struct btrfs_root
*root
, struct inode
*inode
,
145 u64 start
, size_t size
, size_t compressed_size
,
147 struct page
**compressed_pages
)
149 struct extent_buffer
*leaf
;
150 struct page
*page
= NULL
;
153 struct btrfs_file_extent_item
*ei
;
156 size_t cur_size
= size
;
157 unsigned long offset
;
159 if (compressed_size
&& compressed_pages
)
160 cur_size
= compressed_size
;
162 inode_add_bytes(inode
, size
);
164 if (!extent_inserted
) {
165 struct btrfs_key key
;
168 key
.objectid
= btrfs_ino(inode
);
170 key
.type
= BTRFS_EXTENT_DATA_KEY
;
172 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
173 path
->leave_spinning
= 1;
174 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
181 leaf
= path
->nodes
[0];
182 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
183 struct btrfs_file_extent_item
);
184 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
185 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
186 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
187 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
188 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
189 ptr
= btrfs_file_extent_inline_start(ei
);
191 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
194 while (compressed_size
> 0) {
195 cpage
= compressed_pages
[i
];
196 cur_size
= min_t(unsigned long, compressed_size
,
199 kaddr
= kmap_atomic(cpage
);
200 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
201 kunmap_atomic(kaddr
);
205 compressed_size
-= cur_size
;
207 btrfs_set_file_extent_compression(leaf
, ei
,
210 page
= find_get_page(inode
->i_mapping
,
211 start
>> PAGE_CACHE_SHIFT
);
212 btrfs_set_file_extent_compression(leaf
, ei
, 0);
213 kaddr
= kmap_atomic(page
);
214 offset
= start
& (PAGE_CACHE_SIZE
- 1);
215 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
216 kunmap_atomic(kaddr
);
217 page_cache_release(page
);
219 btrfs_mark_buffer_dirty(leaf
);
220 btrfs_release_path(path
);
223 * we're an inline extent, so nobody can
224 * extend the file past i_size without locking
225 * a page we already have locked.
227 * We must do any isize and inode updates
228 * before we unlock the pages. Otherwise we
229 * could end up racing with unlink.
231 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
232 ret
= btrfs_update_inode(trans
, root
, inode
);
241 * conditionally insert an inline extent into the file. This
242 * does the checks required to make sure the data is small enough
243 * to fit as an inline extent.
245 static noinline
int cow_file_range_inline(struct btrfs_root
*root
,
246 struct inode
*inode
, u64 start
,
247 u64 end
, size_t compressed_size
,
249 struct page
**compressed_pages
)
251 struct btrfs_trans_handle
*trans
;
252 u64 isize
= i_size_read(inode
);
253 u64 actual_end
= min(end
+ 1, isize
);
254 u64 inline_len
= actual_end
- start
;
255 u64 aligned_end
= ALIGN(end
, root
->sectorsize
);
256 u64 data_len
= inline_len
;
258 struct btrfs_path
*path
;
259 int extent_inserted
= 0;
260 u32 extent_item_size
;
263 data_len
= compressed_size
;
266 actual_end
> PAGE_CACHE_SIZE
||
267 data_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
269 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
271 data_len
> root
->fs_info
->max_inline
) {
275 path
= btrfs_alloc_path();
279 trans
= btrfs_join_transaction(root
);
281 btrfs_free_path(path
);
282 return PTR_ERR(trans
);
284 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
286 if (compressed_size
&& compressed_pages
)
287 extent_item_size
= btrfs_file_extent_calc_inline_size(
290 extent_item_size
= btrfs_file_extent_calc_inline_size(
293 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
,
294 start
, aligned_end
, NULL
,
295 1, 1, extent_item_size
, &extent_inserted
);
297 btrfs_abort_transaction(trans
, root
, ret
);
301 if (isize
> actual_end
)
302 inline_len
= min_t(u64
, isize
, actual_end
);
303 ret
= insert_inline_extent(trans
, path
, extent_inserted
,
305 inline_len
, compressed_size
,
306 compress_type
, compressed_pages
);
307 if (ret
&& ret
!= -ENOSPC
) {
308 btrfs_abort_transaction(trans
, root
, ret
);
310 } else if (ret
== -ENOSPC
) {
315 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
316 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
317 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
320 * Don't forget to free the reserved space, as for inlined extent
321 * it won't count as data extent, free them directly here.
322 * And at reserve time, it's always aligned to page size, so
323 * just free one page here.
325 btrfs_qgroup_free_data(inode
, 0, PAGE_CACHE_SIZE
);
326 btrfs_free_path(path
);
327 btrfs_end_transaction(trans
, root
);
331 struct async_extent
{
336 unsigned long nr_pages
;
338 struct list_head list
;
343 struct btrfs_root
*root
;
344 struct page
*locked_page
;
347 struct list_head extents
;
348 struct btrfs_work work
;
351 static noinline
int add_async_extent(struct async_cow
*cow
,
352 u64 start
, u64 ram_size
,
355 unsigned long nr_pages
,
358 struct async_extent
*async_extent
;
360 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
361 BUG_ON(!async_extent
); /* -ENOMEM */
362 async_extent
->start
= start
;
363 async_extent
->ram_size
= ram_size
;
364 async_extent
->compressed_size
= compressed_size
;
365 async_extent
->pages
= pages
;
366 async_extent
->nr_pages
= nr_pages
;
367 async_extent
->compress_type
= compress_type
;
368 list_add_tail(&async_extent
->list
, &cow
->extents
);
372 static inline int inode_need_compress(struct inode
*inode
)
374 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
377 if (btrfs_test_opt(root
, FORCE_COMPRESS
))
379 /* bad compression ratios */
380 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
382 if (btrfs_test_opt(root
, COMPRESS
) ||
383 BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
||
384 BTRFS_I(inode
)->force_compress
)
390 * we create compressed extents in two phases. The first
391 * phase compresses a range of pages that have already been
392 * locked (both pages and state bits are locked).
394 * This is done inside an ordered work queue, and the compression
395 * is spread across many cpus. The actual IO submission is step
396 * two, and the ordered work queue takes care of making sure that
397 * happens in the same order things were put onto the queue by
398 * writepages and friends.
400 * If this code finds it can't get good compression, it puts an
401 * entry onto the work queue to write the uncompressed bytes. This
402 * makes sure that both compressed inodes and uncompressed inodes
403 * are written in the same order that the flusher thread sent them
406 static noinline
void compress_file_range(struct inode
*inode
,
407 struct page
*locked_page
,
409 struct async_cow
*async_cow
,
412 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
414 u64 blocksize
= root
->sectorsize
;
416 u64 isize
= i_size_read(inode
);
418 struct page
**pages
= NULL
;
419 unsigned long nr_pages
;
420 unsigned long nr_pages_ret
= 0;
421 unsigned long total_compressed
= 0;
422 unsigned long total_in
= 0;
423 unsigned long max_compressed
= SZ_128K
;
424 unsigned long max_uncompressed
= SZ_128K
;
427 int compress_type
= root
->fs_info
->compress_type
;
430 /* if this is a small write inside eof, kick off a defrag */
431 if ((end
- start
+ 1) < SZ_16K
&&
432 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
433 btrfs_add_inode_defrag(NULL
, inode
);
435 actual_end
= min_t(u64
, isize
, end
+ 1);
438 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
439 nr_pages
= min_t(unsigned long, nr_pages
, SZ_128K
/ PAGE_CACHE_SIZE
);
442 * we don't want to send crud past the end of i_size through
443 * compression, that's just a waste of CPU time. So, if the
444 * end of the file is before the start of our current
445 * requested range of bytes, we bail out to the uncompressed
446 * cleanup code that can deal with all of this.
448 * It isn't really the fastest way to fix things, but this is a
449 * very uncommon corner.
451 if (actual_end
<= start
)
452 goto cleanup_and_bail_uncompressed
;
454 total_compressed
= actual_end
- start
;
457 * skip compression for a small file range(<=blocksize) that
458 * isn't an inline extent, since it dosen't save disk space at all.
460 if (total_compressed
<= blocksize
&&
461 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
462 goto cleanup_and_bail_uncompressed
;
464 /* we want to make sure that amount of ram required to uncompress
465 * an extent is reasonable, so we limit the total size in ram
466 * of a compressed extent to 128k. This is a crucial number
467 * because it also controls how easily we can spread reads across
468 * cpus for decompression.
470 * We also want to make sure the amount of IO required to do
471 * a random read is reasonably small, so we limit the size of
472 * a compressed extent to 128k.
474 total_compressed
= min(total_compressed
, max_uncompressed
);
475 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
476 num_bytes
= max(blocksize
, num_bytes
);
481 * we do compression for mount -o compress and when the
482 * inode has not been flagged as nocompress. This flag can
483 * change at any time if we discover bad compression ratios.
485 if (inode_need_compress(inode
)) {
487 pages
= kcalloc(nr_pages
, sizeof(struct page
*), GFP_NOFS
);
489 /* just bail out to the uncompressed code */
493 if (BTRFS_I(inode
)->force_compress
)
494 compress_type
= BTRFS_I(inode
)->force_compress
;
497 * we need to call clear_page_dirty_for_io on each
498 * page in the range. Otherwise applications with the file
499 * mmap'd can wander in and change the page contents while
500 * we are compressing them.
502 * If the compression fails for any reason, we set the pages
503 * dirty again later on.
505 extent_range_clear_dirty_for_io(inode
, start
, end
);
507 ret
= btrfs_compress_pages(compress_type
,
508 inode
->i_mapping
, start
,
509 total_compressed
, pages
,
510 nr_pages
, &nr_pages_ret
,
516 unsigned long offset
= total_compressed
&
517 (PAGE_CACHE_SIZE
- 1);
518 struct page
*page
= pages
[nr_pages_ret
- 1];
521 /* zero the tail end of the last page, we might be
522 * sending it down to disk
525 kaddr
= kmap_atomic(page
);
526 memset(kaddr
+ offset
, 0,
527 PAGE_CACHE_SIZE
- offset
);
528 kunmap_atomic(kaddr
);
535 /* lets try to make an inline extent */
536 if (ret
|| total_in
< (actual_end
- start
)) {
537 /* we didn't compress the entire range, try
538 * to make an uncompressed inline extent.
540 ret
= cow_file_range_inline(root
, inode
, start
, end
,
543 /* try making a compressed inline extent */
544 ret
= cow_file_range_inline(root
, inode
, start
, end
,
546 compress_type
, pages
);
549 unsigned long clear_flags
= EXTENT_DELALLOC
|
551 unsigned long page_error_op
;
553 clear_flags
|= (ret
< 0) ? EXTENT_DO_ACCOUNTING
: 0;
554 page_error_op
= ret
< 0 ? PAGE_SET_ERROR
: 0;
557 * inline extent creation worked or returned error,
558 * we don't need to create any more async work items.
559 * Unlock and free up our temp pages.
561 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
,
562 clear_flags
, PAGE_UNLOCK
|
573 * we aren't doing an inline extent round the compressed size
574 * up to a block size boundary so the allocator does sane
577 total_compressed
= ALIGN(total_compressed
, blocksize
);
580 * one last check to make sure the compression is really a
581 * win, compare the page count read with the blocks on disk
583 total_in
= ALIGN(total_in
, PAGE_CACHE_SIZE
);
584 if (total_compressed
>= total_in
) {
587 num_bytes
= total_in
;
590 if (!will_compress
&& pages
) {
592 * the compression code ran but failed to make things smaller,
593 * free any pages it allocated and our page pointer array
595 for (i
= 0; i
< nr_pages_ret
; i
++) {
596 WARN_ON(pages
[i
]->mapping
);
597 page_cache_release(pages
[i
]);
601 total_compressed
= 0;
604 /* flag the file so we don't compress in the future */
605 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
606 !(BTRFS_I(inode
)->force_compress
)) {
607 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
613 /* the async work queues will take care of doing actual
614 * allocation on disk for these compressed pages,
615 * and will submit them to the elevator.
617 add_async_extent(async_cow
, start
, num_bytes
,
618 total_compressed
, pages
, nr_pages_ret
,
621 if (start
+ num_bytes
< end
) {
628 cleanup_and_bail_uncompressed
:
630 * No compression, but we still need to write the pages in
631 * the file we've been given so far. redirty the locked
632 * page if it corresponds to our extent and set things up
633 * for the async work queue to run cow_file_range to do
634 * the normal delalloc dance
636 if (page_offset(locked_page
) >= start
&&
637 page_offset(locked_page
) <= end
) {
638 __set_page_dirty_nobuffers(locked_page
);
639 /* unlocked later on in the async handlers */
642 extent_range_redirty_for_io(inode
, start
, end
);
643 add_async_extent(async_cow
, start
, end
- start
+ 1,
644 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
651 for (i
= 0; i
< nr_pages_ret
; i
++) {
652 WARN_ON(pages
[i
]->mapping
);
653 page_cache_release(pages
[i
]);
658 static void free_async_extent_pages(struct async_extent
*async_extent
)
662 if (!async_extent
->pages
)
665 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
666 WARN_ON(async_extent
->pages
[i
]->mapping
);
667 page_cache_release(async_extent
->pages
[i
]);
669 kfree(async_extent
->pages
);
670 async_extent
->nr_pages
= 0;
671 async_extent
->pages
= NULL
;
675 * phase two of compressed writeback. This is the ordered portion
676 * of the code, which only gets called in the order the work was
677 * queued. We walk all the async extents created by compress_file_range
678 * and send them down to the disk.
680 static noinline
void submit_compressed_extents(struct inode
*inode
,
681 struct async_cow
*async_cow
)
683 struct async_extent
*async_extent
;
685 struct btrfs_key ins
;
686 struct extent_map
*em
;
687 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
688 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
689 struct extent_io_tree
*io_tree
;
693 while (!list_empty(&async_cow
->extents
)) {
694 async_extent
= list_entry(async_cow
->extents
.next
,
695 struct async_extent
, list
);
696 list_del(&async_extent
->list
);
698 io_tree
= &BTRFS_I(inode
)->io_tree
;
701 /* did the compression code fall back to uncompressed IO? */
702 if (!async_extent
->pages
) {
703 int page_started
= 0;
704 unsigned long nr_written
= 0;
706 lock_extent(io_tree
, async_extent
->start
,
707 async_extent
->start
+
708 async_extent
->ram_size
- 1);
710 /* allocate blocks */
711 ret
= cow_file_range(inode
, async_cow
->locked_page
,
713 async_extent
->start
+
714 async_extent
->ram_size
- 1,
715 &page_started
, &nr_written
, 0);
720 * if page_started, cow_file_range inserted an
721 * inline extent and took care of all the unlocking
722 * and IO for us. Otherwise, we need to submit
723 * all those pages down to the drive.
725 if (!page_started
&& !ret
)
726 extent_write_locked_range(io_tree
,
727 inode
, async_extent
->start
,
728 async_extent
->start
+
729 async_extent
->ram_size
- 1,
733 unlock_page(async_cow
->locked_page
);
739 lock_extent(io_tree
, async_extent
->start
,
740 async_extent
->start
+ async_extent
->ram_size
- 1);
742 ret
= btrfs_reserve_extent(root
,
743 async_extent
->compressed_size
,
744 async_extent
->compressed_size
,
745 0, alloc_hint
, &ins
, 1, 1);
747 free_async_extent_pages(async_extent
);
749 if (ret
== -ENOSPC
) {
750 unlock_extent(io_tree
, async_extent
->start
,
751 async_extent
->start
+
752 async_extent
->ram_size
- 1);
755 * we need to redirty the pages if we decide to
756 * fallback to uncompressed IO, otherwise we
757 * will not submit these pages down to lower
760 extent_range_redirty_for_io(inode
,
762 async_extent
->start
+
763 async_extent
->ram_size
- 1);
770 * here we're doing allocation and writeback of the
773 btrfs_drop_extent_cache(inode
, async_extent
->start
,
774 async_extent
->start
+
775 async_extent
->ram_size
- 1, 0);
777 em
= alloc_extent_map();
780 goto out_free_reserve
;
782 em
->start
= async_extent
->start
;
783 em
->len
= async_extent
->ram_size
;
784 em
->orig_start
= em
->start
;
785 em
->mod_start
= em
->start
;
786 em
->mod_len
= em
->len
;
788 em
->block_start
= ins
.objectid
;
789 em
->block_len
= ins
.offset
;
790 em
->orig_block_len
= ins
.offset
;
791 em
->ram_bytes
= async_extent
->ram_size
;
792 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
793 em
->compress_type
= async_extent
->compress_type
;
794 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
795 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
799 write_lock(&em_tree
->lock
);
800 ret
= add_extent_mapping(em_tree
, em
, 1);
801 write_unlock(&em_tree
->lock
);
802 if (ret
!= -EEXIST
) {
806 btrfs_drop_extent_cache(inode
, async_extent
->start
,
807 async_extent
->start
+
808 async_extent
->ram_size
- 1, 0);
812 goto out_free_reserve
;
814 ret
= btrfs_add_ordered_extent_compress(inode
,
817 async_extent
->ram_size
,
819 BTRFS_ORDERED_COMPRESSED
,
820 async_extent
->compress_type
);
822 btrfs_drop_extent_cache(inode
, async_extent
->start
,
823 async_extent
->start
+
824 async_extent
->ram_size
- 1, 0);
825 goto out_free_reserve
;
829 * clear dirty, set writeback and unlock the pages.
831 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
832 async_extent
->start
+
833 async_extent
->ram_size
- 1,
834 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
,
835 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
837 ret
= btrfs_submit_compressed_write(inode
,
839 async_extent
->ram_size
,
841 ins
.offset
, async_extent
->pages
,
842 async_extent
->nr_pages
);
844 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
845 struct page
*p
= async_extent
->pages
[0];
846 const u64 start
= async_extent
->start
;
847 const u64 end
= start
+ async_extent
->ram_size
- 1;
849 p
->mapping
= inode
->i_mapping
;
850 tree
->ops
->writepage_end_io_hook(p
, start
, end
,
853 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
, 0,
856 free_async_extent_pages(async_extent
);
858 alloc_hint
= ins
.objectid
+ ins
.offset
;
864 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
866 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
867 async_extent
->start
+
868 async_extent
->ram_size
- 1,
869 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
|
870 EXTENT_DEFRAG
| EXTENT_DO_ACCOUNTING
,
871 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
872 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
|
874 free_async_extent_pages(async_extent
);
879 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
882 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
883 struct extent_map
*em
;
886 read_lock(&em_tree
->lock
);
887 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
890 * if block start isn't an actual block number then find the
891 * first block in this inode and use that as a hint. If that
892 * block is also bogus then just don't worry about it.
894 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
896 em
= search_extent_mapping(em_tree
, 0, 0);
897 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
898 alloc_hint
= em
->block_start
;
902 alloc_hint
= em
->block_start
;
906 read_unlock(&em_tree
->lock
);
912 * when extent_io.c finds a delayed allocation range in the file,
913 * the call backs end up in this code. The basic idea is to
914 * allocate extents on disk for the range, and create ordered data structs
915 * in ram to track those extents.
917 * locked_page is the page that writepage had locked already. We use
918 * it to make sure we don't do extra locks or unlocks.
920 * *page_started is set to one if we unlock locked_page and do everything
921 * required to start IO on it. It may be clean and already done with
924 static noinline
int cow_file_range(struct inode
*inode
,
925 struct page
*locked_page
,
926 u64 start
, u64 end
, int *page_started
,
927 unsigned long *nr_written
,
930 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
933 unsigned long ram_size
;
936 u64 blocksize
= root
->sectorsize
;
937 struct btrfs_key ins
;
938 struct extent_map
*em
;
939 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
942 if (btrfs_is_free_space_inode(inode
)) {
948 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
949 num_bytes
= max(blocksize
, num_bytes
);
950 disk_num_bytes
= num_bytes
;
952 /* if this is a small write inside eof, kick off defrag */
953 if (num_bytes
< SZ_64K
&&
954 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
955 btrfs_add_inode_defrag(NULL
, inode
);
958 /* lets try to make an inline extent */
959 ret
= cow_file_range_inline(root
, inode
, start
, end
, 0, 0,
962 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
,
963 EXTENT_LOCKED
| EXTENT_DELALLOC
|
964 EXTENT_DEFRAG
, PAGE_UNLOCK
|
965 PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
968 *nr_written
= *nr_written
+
969 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
972 } else if (ret
< 0) {
977 BUG_ON(disk_num_bytes
>
978 btrfs_super_total_bytes(root
->fs_info
->super_copy
));
980 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
981 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
983 while (disk_num_bytes
> 0) {
986 cur_alloc_size
= disk_num_bytes
;
987 ret
= btrfs_reserve_extent(root
, cur_alloc_size
,
988 root
->sectorsize
, 0, alloc_hint
,
993 em
= alloc_extent_map();
999 em
->orig_start
= em
->start
;
1000 ram_size
= ins
.offset
;
1001 em
->len
= ins
.offset
;
1002 em
->mod_start
= em
->start
;
1003 em
->mod_len
= em
->len
;
1005 em
->block_start
= ins
.objectid
;
1006 em
->block_len
= ins
.offset
;
1007 em
->orig_block_len
= ins
.offset
;
1008 em
->ram_bytes
= ram_size
;
1009 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1010 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1011 em
->generation
= -1;
1014 write_lock(&em_tree
->lock
);
1015 ret
= add_extent_mapping(em_tree
, em
, 1);
1016 write_unlock(&em_tree
->lock
);
1017 if (ret
!= -EEXIST
) {
1018 free_extent_map(em
);
1021 btrfs_drop_extent_cache(inode
, start
,
1022 start
+ ram_size
- 1, 0);
1027 cur_alloc_size
= ins
.offset
;
1028 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
1029 ram_size
, cur_alloc_size
, 0);
1031 goto out_drop_extent_cache
;
1033 if (root
->root_key
.objectid
==
1034 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1035 ret
= btrfs_reloc_clone_csums(inode
, start
,
1038 goto out_drop_extent_cache
;
1041 if (disk_num_bytes
< cur_alloc_size
)
1044 /* we're not doing compressed IO, don't unlock the first
1045 * page (which the caller expects to stay locked), don't
1046 * clear any dirty bits and don't set any writeback bits
1048 * Do set the Private2 bit so we know this page was properly
1049 * setup for writepage
1051 op
= unlock
? PAGE_UNLOCK
: 0;
1052 op
|= PAGE_SET_PRIVATE2
;
1054 extent_clear_unlock_delalloc(inode
, start
,
1055 start
+ ram_size
- 1, locked_page
,
1056 EXTENT_LOCKED
| EXTENT_DELALLOC
,
1058 disk_num_bytes
-= cur_alloc_size
;
1059 num_bytes
-= cur_alloc_size
;
1060 alloc_hint
= ins
.objectid
+ ins
.offset
;
1061 start
+= cur_alloc_size
;
1066 out_drop_extent_cache
:
1067 btrfs_drop_extent_cache(inode
, start
, start
+ ram_size
- 1, 0);
1069 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
1071 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1072 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
1073 EXTENT_DELALLOC
| EXTENT_DEFRAG
,
1074 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
1075 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
);
1080 * work queue call back to started compression on a file and pages
1082 static noinline
void async_cow_start(struct btrfs_work
*work
)
1084 struct async_cow
*async_cow
;
1086 async_cow
= container_of(work
, struct async_cow
, work
);
1088 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
1089 async_cow
->start
, async_cow
->end
, async_cow
,
1091 if (num_added
== 0) {
1092 btrfs_add_delayed_iput(async_cow
->inode
);
1093 async_cow
->inode
= NULL
;
1098 * work queue call back to submit previously compressed pages
1100 static noinline
void async_cow_submit(struct btrfs_work
*work
)
1102 struct async_cow
*async_cow
;
1103 struct btrfs_root
*root
;
1104 unsigned long nr_pages
;
1106 async_cow
= container_of(work
, struct async_cow
, work
);
1108 root
= async_cow
->root
;
1109 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
1113 * atomic_sub_return implies a barrier for waitqueue_active
1115 if (atomic_sub_return(nr_pages
, &root
->fs_info
->async_delalloc_pages
) <
1117 waitqueue_active(&root
->fs_info
->async_submit_wait
))
1118 wake_up(&root
->fs_info
->async_submit_wait
);
1120 if (async_cow
->inode
)
1121 submit_compressed_extents(async_cow
->inode
, async_cow
);
1124 static noinline
void async_cow_free(struct btrfs_work
*work
)
1126 struct async_cow
*async_cow
;
1127 async_cow
= container_of(work
, struct async_cow
, work
);
1128 if (async_cow
->inode
)
1129 btrfs_add_delayed_iput(async_cow
->inode
);
1133 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1134 u64 start
, u64 end
, int *page_started
,
1135 unsigned long *nr_written
)
1137 struct async_cow
*async_cow
;
1138 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1139 unsigned long nr_pages
;
1141 int limit
= 10 * SZ_1M
;
1143 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1144 1, 0, NULL
, GFP_NOFS
);
1145 while (start
< end
) {
1146 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1147 BUG_ON(!async_cow
); /* -ENOMEM */
1148 async_cow
->inode
= igrab(inode
);
1149 async_cow
->root
= root
;
1150 async_cow
->locked_page
= locked_page
;
1151 async_cow
->start
= start
;
1153 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
&&
1154 !btrfs_test_opt(root
, FORCE_COMPRESS
))
1157 cur_end
= min(end
, start
+ SZ_512K
- 1);
1159 async_cow
->end
= cur_end
;
1160 INIT_LIST_HEAD(&async_cow
->extents
);
1162 btrfs_init_work(&async_cow
->work
,
1163 btrfs_delalloc_helper
,
1164 async_cow_start
, async_cow_submit
,
1167 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
1169 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
1171 btrfs_queue_work(root
->fs_info
->delalloc_workers
,
1174 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
1175 wait_event(root
->fs_info
->async_submit_wait
,
1176 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
1180 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
1181 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
1182 wait_event(root
->fs_info
->async_submit_wait
,
1183 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
1187 *nr_written
+= nr_pages
;
1188 start
= cur_end
+ 1;
1194 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1195 u64 bytenr
, u64 num_bytes
)
1198 struct btrfs_ordered_sum
*sums
;
1201 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1202 bytenr
+ num_bytes
- 1, &list
, 0);
1203 if (ret
== 0 && list_empty(&list
))
1206 while (!list_empty(&list
)) {
1207 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1208 list_del(&sums
->list
);
1215 * when nowcow writeback call back. This checks for snapshots or COW copies
1216 * of the extents that exist in the file, and COWs the file as required.
1218 * If no cow copies or snapshots exist, we write directly to the existing
1221 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1222 struct page
*locked_page
,
1223 u64 start
, u64 end
, int *page_started
, int force
,
1224 unsigned long *nr_written
)
1226 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1227 struct btrfs_trans_handle
*trans
;
1228 struct extent_buffer
*leaf
;
1229 struct btrfs_path
*path
;
1230 struct btrfs_file_extent_item
*fi
;
1231 struct btrfs_key found_key
;
1246 u64 ino
= btrfs_ino(inode
);
1248 path
= btrfs_alloc_path();
1250 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1251 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1252 EXTENT_DO_ACCOUNTING
|
1253 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1255 PAGE_SET_WRITEBACK
|
1256 PAGE_END_WRITEBACK
);
1260 nolock
= btrfs_is_free_space_inode(inode
);
1263 trans
= btrfs_join_transaction_nolock(root
);
1265 trans
= btrfs_join_transaction(root
);
1267 if (IS_ERR(trans
)) {
1268 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1269 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1270 EXTENT_DO_ACCOUNTING
|
1271 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1273 PAGE_SET_WRITEBACK
|
1274 PAGE_END_WRITEBACK
);
1275 btrfs_free_path(path
);
1276 return PTR_ERR(trans
);
1279 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1281 cow_start
= (u64
)-1;
1284 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1288 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1289 leaf
= path
->nodes
[0];
1290 btrfs_item_key_to_cpu(leaf
, &found_key
,
1291 path
->slots
[0] - 1);
1292 if (found_key
.objectid
== ino
&&
1293 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1298 leaf
= path
->nodes
[0];
1299 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1300 ret
= btrfs_next_leaf(root
, path
);
1305 leaf
= path
->nodes
[0];
1311 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1313 if (found_key
.objectid
> ino
)
1315 if (WARN_ON_ONCE(found_key
.objectid
< ino
) ||
1316 found_key
.type
< BTRFS_EXTENT_DATA_KEY
) {
1320 if (found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1321 found_key
.offset
> end
)
1324 if (found_key
.offset
> cur_offset
) {
1325 extent_end
= found_key
.offset
;
1330 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1331 struct btrfs_file_extent_item
);
1332 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1334 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
1335 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1336 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1337 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1338 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1339 extent_end
= found_key
.offset
+
1340 btrfs_file_extent_num_bytes(leaf
, fi
);
1342 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1343 if (extent_end
<= start
) {
1347 if (disk_bytenr
== 0)
1349 if (btrfs_file_extent_compression(leaf
, fi
) ||
1350 btrfs_file_extent_encryption(leaf
, fi
) ||
1351 btrfs_file_extent_other_encoding(leaf
, fi
))
1353 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1355 if (btrfs_extent_readonly(root
, disk_bytenr
))
1357 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1359 extent_offset
, disk_bytenr
))
1361 disk_bytenr
+= extent_offset
;
1362 disk_bytenr
+= cur_offset
- found_key
.offset
;
1363 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1365 * if there are pending snapshots for this root,
1366 * we fall into common COW way.
1369 err
= btrfs_start_write_no_snapshoting(root
);
1374 * force cow if csum exists in the range.
1375 * this ensure that csum for a given extent are
1376 * either valid or do not exist.
1378 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1381 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1382 extent_end
= found_key
.offset
+
1383 btrfs_file_extent_inline_len(leaf
,
1384 path
->slots
[0], fi
);
1385 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1390 if (extent_end
<= start
) {
1392 if (!nolock
&& nocow
)
1393 btrfs_end_write_no_snapshoting(root
);
1397 if (cow_start
== (u64
)-1)
1398 cow_start
= cur_offset
;
1399 cur_offset
= extent_end
;
1400 if (cur_offset
> end
)
1406 btrfs_release_path(path
);
1407 if (cow_start
!= (u64
)-1) {
1408 ret
= cow_file_range(inode
, locked_page
,
1409 cow_start
, found_key
.offset
- 1,
1410 page_started
, nr_written
, 1);
1412 if (!nolock
&& nocow
)
1413 btrfs_end_write_no_snapshoting(root
);
1416 cow_start
= (u64
)-1;
1419 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1420 struct extent_map
*em
;
1421 struct extent_map_tree
*em_tree
;
1422 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1423 em
= alloc_extent_map();
1424 BUG_ON(!em
); /* -ENOMEM */
1425 em
->start
= cur_offset
;
1426 em
->orig_start
= found_key
.offset
- extent_offset
;
1427 em
->len
= num_bytes
;
1428 em
->block_len
= num_bytes
;
1429 em
->block_start
= disk_bytenr
;
1430 em
->orig_block_len
= disk_num_bytes
;
1431 em
->ram_bytes
= ram_bytes
;
1432 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1433 em
->mod_start
= em
->start
;
1434 em
->mod_len
= em
->len
;
1435 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1436 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
1437 em
->generation
= -1;
1439 write_lock(&em_tree
->lock
);
1440 ret
= add_extent_mapping(em_tree
, em
, 1);
1441 write_unlock(&em_tree
->lock
);
1442 if (ret
!= -EEXIST
) {
1443 free_extent_map(em
);
1446 btrfs_drop_extent_cache(inode
, em
->start
,
1447 em
->start
+ em
->len
- 1, 0);
1449 type
= BTRFS_ORDERED_PREALLOC
;
1451 type
= BTRFS_ORDERED_NOCOW
;
1454 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1455 num_bytes
, num_bytes
, type
);
1456 BUG_ON(ret
); /* -ENOMEM */
1458 if (root
->root_key
.objectid
==
1459 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1460 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1463 if (!nolock
&& nocow
)
1464 btrfs_end_write_no_snapshoting(root
);
1469 extent_clear_unlock_delalloc(inode
, cur_offset
,
1470 cur_offset
+ num_bytes
- 1,
1471 locked_page
, EXTENT_LOCKED
|
1472 EXTENT_DELALLOC
, PAGE_UNLOCK
|
1474 if (!nolock
&& nocow
)
1475 btrfs_end_write_no_snapshoting(root
);
1476 cur_offset
= extent_end
;
1477 if (cur_offset
> end
)
1480 btrfs_release_path(path
);
1482 if (cur_offset
<= end
&& cow_start
== (u64
)-1) {
1483 cow_start
= cur_offset
;
1487 if (cow_start
!= (u64
)-1) {
1488 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1489 page_started
, nr_written
, 1);
1495 err
= btrfs_end_transaction(trans
, root
);
1499 if (ret
&& cur_offset
< end
)
1500 extent_clear_unlock_delalloc(inode
, cur_offset
, end
,
1501 locked_page
, EXTENT_LOCKED
|
1502 EXTENT_DELALLOC
| EXTENT_DEFRAG
|
1503 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
1505 PAGE_SET_WRITEBACK
|
1506 PAGE_END_WRITEBACK
);
1507 btrfs_free_path(path
);
1511 static inline int need_force_cow(struct inode
*inode
, u64 start
, u64 end
)
1514 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
1515 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
))
1519 * @defrag_bytes is a hint value, no spinlock held here,
1520 * if is not zero, it means the file is defragging.
1521 * Force cow if given extent needs to be defragged.
1523 if (BTRFS_I(inode
)->defrag_bytes
&&
1524 test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1525 EXTENT_DEFRAG
, 0, NULL
))
1532 * extent_io.c call back to do delayed allocation processing
1534 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1535 u64 start
, u64 end
, int *page_started
,
1536 unsigned long *nr_written
)
1539 int force_cow
= need_force_cow(inode
, start
, end
);
1541 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
&& !force_cow
) {
1542 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1543 page_started
, 1, nr_written
);
1544 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
&& !force_cow
) {
1545 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1546 page_started
, 0, nr_written
);
1547 } else if (!inode_need_compress(inode
)) {
1548 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1549 page_started
, nr_written
, 1);
1551 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1552 &BTRFS_I(inode
)->runtime_flags
);
1553 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1554 page_started
, nr_written
);
1559 static void btrfs_split_extent_hook(struct inode
*inode
,
1560 struct extent_state
*orig
, u64 split
)
1564 /* not delalloc, ignore it */
1565 if (!(orig
->state
& EXTENT_DELALLOC
))
1568 size
= orig
->end
- orig
->start
+ 1;
1569 if (size
> BTRFS_MAX_EXTENT_SIZE
) {
1574 * See the explanation in btrfs_merge_extent_hook, the same
1575 * applies here, just in reverse.
1577 new_size
= orig
->end
- split
+ 1;
1578 num_extents
= div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1579 BTRFS_MAX_EXTENT_SIZE
);
1580 new_size
= split
- orig
->start
;
1581 num_extents
+= div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1582 BTRFS_MAX_EXTENT_SIZE
);
1583 if (div64_u64(size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1584 BTRFS_MAX_EXTENT_SIZE
) >= num_extents
)
1588 spin_lock(&BTRFS_I(inode
)->lock
);
1589 BTRFS_I(inode
)->outstanding_extents
++;
1590 spin_unlock(&BTRFS_I(inode
)->lock
);
1594 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1595 * extents so we can keep track of new extents that are just merged onto old
1596 * extents, such as when we are doing sequential writes, so we can properly
1597 * account for the metadata space we'll need.
1599 static void btrfs_merge_extent_hook(struct inode
*inode
,
1600 struct extent_state
*new,
1601 struct extent_state
*other
)
1603 u64 new_size
, old_size
;
1606 /* not delalloc, ignore it */
1607 if (!(other
->state
& EXTENT_DELALLOC
))
1610 if (new->start
> other
->start
)
1611 new_size
= new->end
- other
->start
+ 1;
1613 new_size
= other
->end
- new->start
+ 1;
1615 /* we're not bigger than the max, unreserve the space and go */
1616 if (new_size
<= BTRFS_MAX_EXTENT_SIZE
) {
1617 spin_lock(&BTRFS_I(inode
)->lock
);
1618 BTRFS_I(inode
)->outstanding_extents
--;
1619 spin_unlock(&BTRFS_I(inode
)->lock
);
1624 * We have to add up either side to figure out how many extents were
1625 * accounted for before we merged into one big extent. If the number of
1626 * extents we accounted for is <= the amount we need for the new range
1627 * then we can return, otherwise drop. Think of it like this
1631 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1632 * need 2 outstanding extents, on one side we have 1 and the other side
1633 * we have 1 so they are == and we can return. But in this case
1635 * [MAX_SIZE+4k][MAX_SIZE+4k]
1637 * Each range on their own accounts for 2 extents, but merged together
1638 * they are only 3 extents worth of accounting, so we need to drop in
1641 old_size
= other
->end
- other
->start
+ 1;
1642 num_extents
= div64_u64(old_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1643 BTRFS_MAX_EXTENT_SIZE
);
1644 old_size
= new->end
- new->start
+ 1;
1645 num_extents
+= div64_u64(old_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1646 BTRFS_MAX_EXTENT_SIZE
);
1648 if (div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1649 BTRFS_MAX_EXTENT_SIZE
) >= num_extents
)
1652 spin_lock(&BTRFS_I(inode
)->lock
);
1653 BTRFS_I(inode
)->outstanding_extents
--;
1654 spin_unlock(&BTRFS_I(inode
)->lock
);
1657 static void btrfs_add_delalloc_inodes(struct btrfs_root
*root
,
1658 struct inode
*inode
)
1660 spin_lock(&root
->delalloc_lock
);
1661 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1662 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1663 &root
->delalloc_inodes
);
1664 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1665 &BTRFS_I(inode
)->runtime_flags
);
1666 root
->nr_delalloc_inodes
++;
1667 if (root
->nr_delalloc_inodes
== 1) {
1668 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1669 BUG_ON(!list_empty(&root
->delalloc_root
));
1670 list_add_tail(&root
->delalloc_root
,
1671 &root
->fs_info
->delalloc_roots
);
1672 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1675 spin_unlock(&root
->delalloc_lock
);
1678 static void btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1679 struct inode
*inode
)
1681 spin_lock(&root
->delalloc_lock
);
1682 if (!list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1683 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1684 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1685 &BTRFS_I(inode
)->runtime_flags
);
1686 root
->nr_delalloc_inodes
--;
1687 if (!root
->nr_delalloc_inodes
) {
1688 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1689 BUG_ON(list_empty(&root
->delalloc_root
));
1690 list_del_init(&root
->delalloc_root
);
1691 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1694 spin_unlock(&root
->delalloc_lock
);
1698 * extent_io.c set_bit_hook, used to track delayed allocation
1699 * bytes in this file, and to maintain the list of inodes that
1700 * have pending delalloc work to be done.
1702 static void btrfs_set_bit_hook(struct inode
*inode
,
1703 struct extent_state
*state
, unsigned *bits
)
1706 if ((*bits
& EXTENT_DEFRAG
) && !(*bits
& EXTENT_DELALLOC
))
1709 * set_bit and clear bit hooks normally require _irqsave/restore
1710 * but in this case, we are only testing for the DELALLOC
1711 * bit, which is only set or cleared with irqs on
1713 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1714 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1715 u64 len
= state
->end
+ 1 - state
->start
;
1716 bool do_list
= !btrfs_is_free_space_inode(inode
);
1718 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1719 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1721 spin_lock(&BTRFS_I(inode
)->lock
);
1722 BTRFS_I(inode
)->outstanding_extents
++;
1723 spin_unlock(&BTRFS_I(inode
)->lock
);
1726 /* For sanity tests */
1727 if (btrfs_test_is_dummy_root(root
))
1730 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, len
,
1731 root
->fs_info
->delalloc_batch
);
1732 spin_lock(&BTRFS_I(inode
)->lock
);
1733 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1734 if (*bits
& EXTENT_DEFRAG
)
1735 BTRFS_I(inode
)->defrag_bytes
+= len
;
1736 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1737 &BTRFS_I(inode
)->runtime_flags
))
1738 btrfs_add_delalloc_inodes(root
, inode
);
1739 spin_unlock(&BTRFS_I(inode
)->lock
);
1744 * extent_io.c clear_bit_hook, see set_bit_hook for why
1746 static void btrfs_clear_bit_hook(struct inode
*inode
,
1747 struct extent_state
*state
,
1750 u64 len
= state
->end
+ 1 - state
->start
;
1751 u64 num_extents
= div64_u64(len
+ BTRFS_MAX_EXTENT_SIZE
-1,
1752 BTRFS_MAX_EXTENT_SIZE
);
1754 spin_lock(&BTRFS_I(inode
)->lock
);
1755 if ((state
->state
& EXTENT_DEFRAG
) && (*bits
& EXTENT_DEFRAG
))
1756 BTRFS_I(inode
)->defrag_bytes
-= len
;
1757 spin_unlock(&BTRFS_I(inode
)->lock
);
1760 * set_bit and clear bit hooks normally require _irqsave/restore
1761 * but in this case, we are only testing for the DELALLOC
1762 * bit, which is only set or cleared with irqs on
1764 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1765 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1766 bool do_list
= !btrfs_is_free_space_inode(inode
);
1768 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1769 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1770 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1771 spin_lock(&BTRFS_I(inode
)->lock
);
1772 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
1773 spin_unlock(&BTRFS_I(inode
)->lock
);
1777 * We don't reserve metadata space for space cache inodes so we
1778 * don't need to call dellalloc_release_metadata if there is an
1781 if (*bits
& EXTENT_DO_ACCOUNTING
&&
1782 root
!= root
->fs_info
->tree_root
)
1783 btrfs_delalloc_release_metadata(inode
, len
);
1785 /* For sanity tests. */
1786 if (btrfs_test_is_dummy_root(root
))
1789 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1790 && do_list
&& !(state
->state
& EXTENT_NORESERVE
))
1791 btrfs_free_reserved_data_space_noquota(inode
,
1794 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, -len
,
1795 root
->fs_info
->delalloc_batch
);
1796 spin_lock(&BTRFS_I(inode
)->lock
);
1797 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1798 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1799 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1800 &BTRFS_I(inode
)->runtime_flags
))
1801 btrfs_del_delalloc_inode(root
, inode
);
1802 spin_unlock(&BTRFS_I(inode
)->lock
);
1807 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1808 * we don't create bios that span stripes or chunks
1810 int btrfs_merge_bio_hook(int rw
, struct page
*page
, unsigned long offset
,
1811 size_t size
, struct bio
*bio
,
1812 unsigned long bio_flags
)
1814 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1815 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
1820 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1823 length
= bio
->bi_iter
.bi_size
;
1824 map_length
= length
;
1825 ret
= btrfs_map_block(root
->fs_info
, rw
, logical
,
1826 &map_length
, NULL
, 0);
1827 /* Will always return 0 with map_multi == NULL */
1829 if (map_length
< length
+ size
)
1835 * in order to insert checksums into the metadata in large chunks,
1836 * we wait until bio submission time. All the pages in the bio are
1837 * checksummed and sums are attached onto the ordered extent record.
1839 * At IO completion time the cums attached on the ordered extent record
1840 * are inserted into the btree
1842 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1843 struct bio
*bio
, int mirror_num
,
1844 unsigned long bio_flags
,
1847 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1850 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1851 BUG_ON(ret
); /* -ENOMEM */
1856 * in order to insert checksums into the metadata in large chunks,
1857 * we wait until bio submission time. All the pages in the bio are
1858 * checksummed and sums are attached onto the ordered extent record.
1860 * At IO completion time the cums attached on the ordered extent record
1861 * are inserted into the btree
1863 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1864 int mirror_num
, unsigned long bio_flags
,
1867 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1870 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1872 bio
->bi_error
= ret
;
1879 * extent_io.c submission hook. This does the right thing for csum calculation
1880 * on write, or reading the csums from the tree before a read
1882 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1883 int mirror_num
, unsigned long bio_flags
,
1886 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1887 enum btrfs_wq_endio_type metadata
= BTRFS_WQ_ENDIO_DATA
;
1890 int async
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
1892 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1894 if (btrfs_is_free_space_inode(inode
))
1895 metadata
= BTRFS_WQ_ENDIO_FREE_SPACE
;
1897 if (!(rw
& REQ_WRITE
)) {
1898 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, metadata
);
1902 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1903 ret
= btrfs_submit_compressed_read(inode
, bio
,
1907 } else if (!skip_sum
) {
1908 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1913 } else if (async
&& !skip_sum
) {
1914 /* csum items have already been cloned */
1915 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1917 /* we're doing a write, do the async checksumming */
1918 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1919 inode
, rw
, bio
, mirror_num
,
1920 bio_flags
, bio_offset
,
1921 __btrfs_submit_bio_start
,
1922 __btrfs_submit_bio_done
);
1924 } else if (!skip_sum
) {
1925 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1931 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1935 bio
->bi_error
= ret
;
1942 * given a list of ordered sums record them in the inode. This happens
1943 * at IO completion time based on sums calculated at bio submission time.
1945 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1946 struct inode
*inode
, u64 file_offset
,
1947 struct list_head
*list
)
1949 struct btrfs_ordered_sum
*sum
;
1951 list_for_each_entry(sum
, list
, list
) {
1952 trans
->adding_csums
= 1;
1953 btrfs_csum_file_blocks(trans
,
1954 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1955 trans
->adding_csums
= 0;
1960 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1961 struct extent_state
**cached_state
)
1963 WARN_ON((end
& (PAGE_CACHE_SIZE
- 1)) == 0);
1964 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1965 cached_state
, GFP_NOFS
);
1968 /* see btrfs_writepage_start_hook for details on why this is required */
1969 struct btrfs_writepage_fixup
{
1971 struct btrfs_work work
;
1974 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1976 struct btrfs_writepage_fixup
*fixup
;
1977 struct btrfs_ordered_extent
*ordered
;
1978 struct extent_state
*cached_state
= NULL
;
1980 struct inode
*inode
;
1985 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1989 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1990 ClearPageChecked(page
);
1994 inode
= page
->mapping
->host
;
1995 page_start
= page_offset(page
);
1996 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1998 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2001 /* already ordered? We're done */
2002 if (PagePrivate2(page
))
2005 ordered
= btrfs_lookup_ordered_range(inode
, page_start
,
2008 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
2009 page_end
, &cached_state
, GFP_NOFS
);
2011 btrfs_start_ordered_extent(inode
, ordered
, 1);
2012 btrfs_put_ordered_extent(ordered
);
2016 ret
= btrfs_delalloc_reserve_space(inode
, page_start
,
2019 mapping_set_error(page
->mapping
, ret
);
2020 end_extent_writepage(page
, ret
, page_start
, page_end
);
2021 ClearPageChecked(page
);
2025 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
2026 ClearPageChecked(page
);
2027 set_page_dirty(page
);
2029 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2030 &cached_state
, GFP_NOFS
);
2033 page_cache_release(page
);
2038 * There are a few paths in the higher layers of the kernel that directly
2039 * set the page dirty bit without asking the filesystem if it is a
2040 * good idea. This causes problems because we want to make sure COW
2041 * properly happens and the data=ordered rules are followed.
2043 * In our case any range that doesn't have the ORDERED bit set
2044 * hasn't been properly setup for IO. We kick off an async process
2045 * to fix it up. The async helper will wait for ordered extents, set
2046 * the delalloc bit and make it safe to write the page.
2048 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
2050 struct inode
*inode
= page
->mapping
->host
;
2051 struct btrfs_writepage_fixup
*fixup
;
2052 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2054 /* this page is properly in the ordered list */
2055 if (TestClearPagePrivate2(page
))
2058 if (PageChecked(page
))
2061 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
2065 SetPageChecked(page
);
2066 page_cache_get(page
);
2067 btrfs_init_work(&fixup
->work
, btrfs_fixup_helper
,
2068 btrfs_writepage_fixup_worker
, NULL
, NULL
);
2070 btrfs_queue_work(root
->fs_info
->fixup_workers
, &fixup
->work
);
2074 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
2075 struct inode
*inode
, u64 file_pos
,
2076 u64 disk_bytenr
, u64 disk_num_bytes
,
2077 u64 num_bytes
, u64 ram_bytes
,
2078 u8 compression
, u8 encryption
,
2079 u16 other_encoding
, int extent_type
)
2081 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2082 struct btrfs_file_extent_item
*fi
;
2083 struct btrfs_path
*path
;
2084 struct extent_buffer
*leaf
;
2085 struct btrfs_key ins
;
2086 int extent_inserted
= 0;
2089 path
= btrfs_alloc_path();
2094 * we may be replacing one extent in the tree with another.
2095 * The new extent is pinned in the extent map, and we don't want
2096 * to drop it from the cache until it is completely in the btree.
2098 * So, tell btrfs_drop_extents to leave this extent in the cache.
2099 * the caller is expected to unpin it and allow it to be merged
2102 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, file_pos
,
2103 file_pos
+ num_bytes
, NULL
, 0,
2104 1, sizeof(*fi
), &extent_inserted
);
2108 if (!extent_inserted
) {
2109 ins
.objectid
= btrfs_ino(inode
);
2110 ins
.offset
= file_pos
;
2111 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
2113 path
->leave_spinning
= 1;
2114 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
2119 leaf
= path
->nodes
[0];
2120 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2121 struct btrfs_file_extent_item
);
2122 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
2123 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
2124 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
2125 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
2126 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2127 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2128 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
2129 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
2130 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
2131 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
2133 btrfs_mark_buffer_dirty(leaf
);
2134 btrfs_release_path(path
);
2136 inode_add_bytes(inode
, num_bytes
);
2138 ins
.objectid
= disk_bytenr
;
2139 ins
.offset
= disk_num_bytes
;
2140 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2141 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
2142 root
->root_key
.objectid
,
2143 btrfs_ino(inode
), file_pos
,
2146 * Release the reserved range from inode dirty range map, as it is
2147 * already moved into delayed_ref_head
2149 btrfs_qgroup_release_data(inode
, file_pos
, ram_bytes
);
2151 btrfs_free_path(path
);
2156 /* snapshot-aware defrag */
2157 struct sa_defrag_extent_backref
{
2158 struct rb_node node
;
2159 struct old_sa_defrag_extent
*old
;
2168 struct old_sa_defrag_extent
{
2169 struct list_head list
;
2170 struct new_sa_defrag_extent
*new;
2179 struct new_sa_defrag_extent
{
2180 struct rb_root root
;
2181 struct list_head head
;
2182 struct btrfs_path
*path
;
2183 struct inode
*inode
;
2191 static int backref_comp(struct sa_defrag_extent_backref
*b1
,
2192 struct sa_defrag_extent_backref
*b2
)
2194 if (b1
->root_id
< b2
->root_id
)
2196 else if (b1
->root_id
> b2
->root_id
)
2199 if (b1
->inum
< b2
->inum
)
2201 else if (b1
->inum
> b2
->inum
)
2204 if (b1
->file_pos
< b2
->file_pos
)
2206 else if (b1
->file_pos
> b2
->file_pos
)
2210 * [------------------------------] ===> (a range of space)
2211 * |<--->| |<---->| =============> (fs/file tree A)
2212 * |<---------------------------->| ===> (fs/file tree B)
2214 * A range of space can refer to two file extents in one tree while
2215 * refer to only one file extent in another tree.
2217 * So we may process a disk offset more than one time(two extents in A)
2218 * and locate at the same extent(one extent in B), then insert two same
2219 * backrefs(both refer to the extent in B).
2224 static void backref_insert(struct rb_root
*root
,
2225 struct sa_defrag_extent_backref
*backref
)
2227 struct rb_node
**p
= &root
->rb_node
;
2228 struct rb_node
*parent
= NULL
;
2229 struct sa_defrag_extent_backref
*entry
;
2234 entry
= rb_entry(parent
, struct sa_defrag_extent_backref
, node
);
2236 ret
= backref_comp(backref
, entry
);
2240 p
= &(*p
)->rb_right
;
2243 rb_link_node(&backref
->node
, parent
, p
);
2244 rb_insert_color(&backref
->node
, root
);
2248 * Note the backref might has changed, and in this case we just return 0.
2250 static noinline
int record_one_backref(u64 inum
, u64 offset
, u64 root_id
,
2253 struct btrfs_file_extent_item
*extent
;
2254 struct btrfs_fs_info
*fs_info
;
2255 struct old_sa_defrag_extent
*old
= ctx
;
2256 struct new_sa_defrag_extent
*new = old
->new;
2257 struct btrfs_path
*path
= new->path
;
2258 struct btrfs_key key
;
2259 struct btrfs_root
*root
;
2260 struct sa_defrag_extent_backref
*backref
;
2261 struct extent_buffer
*leaf
;
2262 struct inode
*inode
= new->inode
;
2268 if (BTRFS_I(inode
)->root
->root_key
.objectid
== root_id
&&
2269 inum
== btrfs_ino(inode
))
2272 key
.objectid
= root_id
;
2273 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2274 key
.offset
= (u64
)-1;
2276 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2277 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2279 if (PTR_ERR(root
) == -ENOENT
)
2282 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2283 inum
, offset
, root_id
);
2284 return PTR_ERR(root
);
2287 key
.objectid
= inum
;
2288 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2289 if (offset
> (u64
)-1 << 32)
2292 key
.offset
= offset
;
2294 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2295 if (WARN_ON(ret
< 0))
2302 leaf
= path
->nodes
[0];
2303 slot
= path
->slots
[0];
2305 if (slot
>= btrfs_header_nritems(leaf
)) {
2306 ret
= btrfs_next_leaf(root
, path
);
2309 } else if (ret
> 0) {
2318 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2320 if (key
.objectid
> inum
)
2323 if (key
.objectid
< inum
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2326 extent
= btrfs_item_ptr(leaf
, slot
,
2327 struct btrfs_file_extent_item
);
2329 if (btrfs_file_extent_disk_bytenr(leaf
, extent
) != old
->bytenr
)
2333 * 'offset' refers to the exact key.offset,
2334 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2335 * (key.offset - extent_offset).
2337 if (key
.offset
!= offset
)
2340 extent_offset
= btrfs_file_extent_offset(leaf
, extent
);
2341 num_bytes
= btrfs_file_extent_num_bytes(leaf
, extent
);
2343 if (extent_offset
>= old
->extent_offset
+ old
->offset
+
2344 old
->len
|| extent_offset
+ num_bytes
<=
2345 old
->extent_offset
+ old
->offset
)
2350 backref
= kmalloc(sizeof(*backref
), GFP_NOFS
);
2356 backref
->root_id
= root_id
;
2357 backref
->inum
= inum
;
2358 backref
->file_pos
= offset
;
2359 backref
->num_bytes
= num_bytes
;
2360 backref
->extent_offset
= extent_offset
;
2361 backref
->generation
= btrfs_file_extent_generation(leaf
, extent
);
2363 backref_insert(&new->root
, backref
);
2366 btrfs_release_path(path
);
2371 static noinline
bool record_extent_backrefs(struct btrfs_path
*path
,
2372 struct new_sa_defrag_extent
*new)
2374 struct btrfs_fs_info
*fs_info
= BTRFS_I(new->inode
)->root
->fs_info
;
2375 struct old_sa_defrag_extent
*old
, *tmp
;
2380 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2381 ret
= iterate_inodes_from_logical(old
->bytenr
+
2382 old
->extent_offset
, fs_info
,
2383 path
, record_one_backref
,
2385 if (ret
< 0 && ret
!= -ENOENT
)
2388 /* no backref to be processed for this extent */
2390 list_del(&old
->list
);
2395 if (list_empty(&new->head
))
2401 static int relink_is_mergable(struct extent_buffer
*leaf
,
2402 struct btrfs_file_extent_item
*fi
,
2403 struct new_sa_defrag_extent
*new)
2405 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) != new->bytenr
)
2408 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2411 if (btrfs_file_extent_compression(leaf
, fi
) != new->compress_type
)
2414 if (btrfs_file_extent_encryption(leaf
, fi
) ||
2415 btrfs_file_extent_other_encoding(leaf
, fi
))
2422 * Note the backref might has changed, and in this case we just return 0.
2424 static noinline
int relink_extent_backref(struct btrfs_path
*path
,
2425 struct sa_defrag_extent_backref
*prev
,
2426 struct sa_defrag_extent_backref
*backref
)
2428 struct btrfs_file_extent_item
*extent
;
2429 struct btrfs_file_extent_item
*item
;
2430 struct btrfs_ordered_extent
*ordered
;
2431 struct btrfs_trans_handle
*trans
;
2432 struct btrfs_fs_info
*fs_info
;
2433 struct btrfs_root
*root
;
2434 struct btrfs_key key
;
2435 struct extent_buffer
*leaf
;
2436 struct old_sa_defrag_extent
*old
= backref
->old
;
2437 struct new_sa_defrag_extent
*new = old
->new;
2438 struct inode
*src_inode
= new->inode
;
2439 struct inode
*inode
;
2440 struct extent_state
*cached
= NULL
;
2449 if (prev
&& prev
->root_id
== backref
->root_id
&&
2450 prev
->inum
== backref
->inum
&&
2451 prev
->file_pos
+ prev
->num_bytes
== backref
->file_pos
)
2454 /* step 1: get root */
2455 key
.objectid
= backref
->root_id
;
2456 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2457 key
.offset
= (u64
)-1;
2459 fs_info
= BTRFS_I(src_inode
)->root
->fs_info
;
2460 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
2462 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2464 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2465 if (PTR_ERR(root
) == -ENOENT
)
2467 return PTR_ERR(root
);
2470 if (btrfs_root_readonly(root
)) {
2471 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2475 /* step 2: get inode */
2476 key
.objectid
= backref
->inum
;
2477 key
.type
= BTRFS_INODE_ITEM_KEY
;
2480 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
2481 if (IS_ERR(inode
)) {
2482 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2486 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2488 /* step 3: relink backref */
2489 lock_start
= backref
->file_pos
;
2490 lock_end
= backref
->file_pos
+ backref
->num_bytes
- 1;
2491 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2494 ordered
= btrfs_lookup_first_ordered_extent(inode
, lock_end
);
2496 btrfs_put_ordered_extent(ordered
);
2500 trans
= btrfs_join_transaction(root
);
2501 if (IS_ERR(trans
)) {
2502 ret
= PTR_ERR(trans
);
2506 key
.objectid
= backref
->inum
;
2507 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2508 key
.offset
= backref
->file_pos
;
2510 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2513 } else if (ret
> 0) {
2518 extent
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2519 struct btrfs_file_extent_item
);
2521 if (btrfs_file_extent_generation(path
->nodes
[0], extent
) !=
2522 backref
->generation
)
2525 btrfs_release_path(path
);
2527 start
= backref
->file_pos
;
2528 if (backref
->extent_offset
< old
->extent_offset
+ old
->offset
)
2529 start
+= old
->extent_offset
+ old
->offset
-
2530 backref
->extent_offset
;
2532 len
= min(backref
->extent_offset
+ backref
->num_bytes
,
2533 old
->extent_offset
+ old
->offset
+ old
->len
);
2534 len
-= max(backref
->extent_offset
, old
->extent_offset
+ old
->offset
);
2536 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
2541 key
.objectid
= btrfs_ino(inode
);
2542 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2545 path
->leave_spinning
= 1;
2547 struct btrfs_file_extent_item
*fi
;
2549 struct btrfs_key found_key
;
2551 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2556 leaf
= path
->nodes
[0];
2557 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2559 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2560 struct btrfs_file_extent_item
);
2561 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
2563 if (extent_len
+ found_key
.offset
== start
&&
2564 relink_is_mergable(leaf
, fi
, new)) {
2565 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2567 btrfs_mark_buffer_dirty(leaf
);
2568 inode_add_bytes(inode
, len
);
2574 btrfs_release_path(path
);
2579 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2582 btrfs_abort_transaction(trans
, root
, ret
);
2586 leaf
= path
->nodes
[0];
2587 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2588 struct btrfs_file_extent_item
);
2589 btrfs_set_file_extent_disk_bytenr(leaf
, item
, new->bytenr
);
2590 btrfs_set_file_extent_disk_num_bytes(leaf
, item
, new->disk_len
);
2591 btrfs_set_file_extent_offset(leaf
, item
, start
- new->file_pos
);
2592 btrfs_set_file_extent_num_bytes(leaf
, item
, len
);
2593 btrfs_set_file_extent_ram_bytes(leaf
, item
, new->len
);
2594 btrfs_set_file_extent_generation(leaf
, item
, trans
->transid
);
2595 btrfs_set_file_extent_type(leaf
, item
, BTRFS_FILE_EXTENT_REG
);
2596 btrfs_set_file_extent_compression(leaf
, item
, new->compress_type
);
2597 btrfs_set_file_extent_encryption(leaf
, item
, 0);
2598 btrfs_set_file_extent_other_encoding(leaf
, item
, 0);
2600 btrfs_mark_buffer_dirty(leaf
);
2601 inode_add_bytes(inode
, len
);
2602 btrfs_release_path(path
);
2604 ret
= btrfs_inc_extent_ref(trans
, root
, new->bytenr
,
2606 backref
->root_id
, backref
->inum
,
2607 new->file_pos
); /* start - extent_offset */
2609 btrfs_abort_transaction(trans
, root
, ret
);
2615 btrfs_release_path(path
);
2616 path
->leave_spinning
= 0;
2617 btrfs_end_transaction(trans
, root
);
2619 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2625 static void free_sa_defrag_extent(struct new_sa_defrag_extent
*new)
2627 struct old_sa_defrag_extent
*old
, *tmp
;
2632 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2638 static void relink_file_extents(struct new_sa_defrag_extent
*new)
2640 struct btrfs_path
*path
;
2641 struct sa_defrag_extent_backref
*backref
;
2642 struct sa_defrag_extent_backref
*prev
= NULL
;
2643 struct inode
*inode
;
2644 struct btrfs_root
*root
;
2645 struct rb_node
*node
;
2649 root
= BTRFS_I(inode
)->root
;
2651 path
= btrfs_alloc_path();
2655 if (!record_extent_backrefs(path
, new)) {
2656 btrfs_free_path(path
);
2659 btrfs_release_path(path
);
2662 node
= rb_first(&new->root
);
2665 rb_erase(node
, &new->root
);
2667 backref
= rb_entry(node
, struct sa_defrag_extent_backref
, node
);
2669 ret
= relink_extent_backref(path
, prev
, backref
);
2682 btrfs_free_path(path
);
2684 free_sa_defrag_extent(new);
2686 atomic_dec(&root
->fs_info
->defrag_running
);
2687 wake_up(&root
->fs_info
->transaction_wait
);
2690 static struct new_sa_defrag_extent
*
2691 record_old_file_extents(struct inode
*inode
,
2692 struct btrfs_ordered_extent
*ordered
)
2694 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2695 struct btrfs_path
*path
;
2696 struct btrfs_key key
;
2697 struct old_sa_defrag_extent
*old
;
2698 struct new_sa_defrag_extent
*new;
2701 new = kmalloc(sizeof(*new), GFP_NOFS
);
2706 new->file_pos
= ordered
->file_offset
;
2707 new->len
= ordered
->len
;
2708 new->bytenr
= ordered
->start
;
2709 new->disk_len
= ordered
->disk_len
;
2710 new->compress_type
= ordered
->compress_type
;
2711 new->root
= RB_ROOT
;
2712 INIT_LIST_HEAD(&new->head
);
2714 path
= btrfs_alloc_path();
2718 key
.objectid
= btrfs_ino(inode
);
2719 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2720 key
.offset
= new->file_pos
;
2722 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2725 if (ret
> 0 && path
->slots
[0] > 0)
2728 /* find out all the old extents for the file range */
2730 struct btrfs_file_extent_item
*extent
;
2731 struct extent_buffer
*l
;
2740 slot
= path
->slots
[0];
2742 if (slot
>= btrfs_header_nritems(l
)) {
2743 ret
= btrfs_next_leaf(root
, path
);
2751 btrfs_item_key_to_cpu(l
, &key
, slot
);
2753 if (key
.objectid
!= btrfs_ino(inode
))
2755 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2757 if (key
.offset
>= new->file_pos
+ new->len
)
2760 extent
= btrfs_item_ptr(l
, slot
, struct btrfs_file_extent_item
);
2762 num_bytes
= btrfs_file_extent_num_bytes(l
, extent
);
2763 if (key
.offset
+ num_bytes
< new->file_pos
)
2766 disk_bytenr
= btrfs_file_extent_disk_bytenr(l
, extent
);
2770 extent_offset
= btrfs_file_extent_offset(l
, extent
);
2772 old
= kmalloc(sizeof(*old
), GFP_NOFS
);
2776 offset
= max(new->file_pos
, key
.offset
);
2777 end
= min(new->file_pos
+ new->len
, key
.offset
+ num_bytes
);
2779 old
->bytenr
= disk_bytenr
;
2780 old
->extent_offset
= extent_offset
;
2781 old
->offset
= offset
- key
.offset
;
2782 old
->len
= end
- offset
;
2785 list_add_tail(&old
->list
, &new->head
);
2791 btrfs_free_path(path
);
2792 atomic_inc(&root
->fs_info
->defrag_running
);
2797 btrfs_free_path(path
);
2799 free_sa_defrag_extent(new);
2803 static void btrfs_release_delalloc_bytes(struct btrfs_root
*root
,
2806 struct btrfs_block_group_cache
*cache
;
2808 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
2811 spin_lock(&cache
->lock
);
2812 cache
->delalloc_bytes
-= len
;
2813 spin_unlock(&cache
->lock
);
2815 btrfs_put_block_group(cache
);
2818 /* as ordered data IO finishes, this gets called so we can finish
2819 * an ordered extent if the range of bytes in the file it covers are
2822 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
2824 struct inode
*inode
= ordered_extent
->inode
;
2825 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2826 struct btrfs_trans_handle
*trans
= NULL
;
2827 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2828 struct extent_state
*cached_state
= NULL
;
2829 struct new_sa_defrag_extent
*new = NULL
;
2830 int compress_type
= 0;
2832 u64 logical_len
= ordered_extent
->len
;
2834 bool truncated
= false;
2836 nolock
= btrfs_is_free_space_inode(inode
);
2838 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
2843 btrfs_free_io_failure_record(inode
, ordered_extent
->file_offset
,
2844 ordered_extent
->file_offset
+
2845 ordered_extent
->len
- 1);
2847 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered_extent
->flags
)) {
2849 logical_len
= ordered_extent
->truncated_len
;
2850 /* Truncated the entire extent, don't bother adding */
2855 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
2856 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
2859 * For mwrite(mmap + memset to write) case, we still reserve
2860 * space for NOCOW range.
2861 * As NOCOW won't cause a new delayed ref, just free the space
2863 btrfs_qgroup_free_data(inode
, ordered_extent
->file_offset
,
2864 ordered_extent
->len
);
2865 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2867 trans
= btrfs_join_transaction_nolock(root
);
2869 trans
= btrfs_join_transaction(root
);
2870 if (IS_ERR(trans
)) {
2871 ret
= PTR_ERR(trans
);
2875 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2876 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2877 if (ret
) /* -ENOMEM or corruption */
2878 btrfs_abort_transaction(trans
, root
, ret
);
2882 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
2883 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2886 ret
= test_range_bit(io_tree
, ordered_extent
->file_offset
,
2887 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2888 EXTENT_DEFRAG
, 1, cached_state
);
2890 u64 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
2891 if (0 && last_snapshot
>= BTRFS_I(inode
)->generation
)
2892 /* the inode is shared */
2893 new = record_old_file_extents(inode
, ordered_extent
);
2895 clear_extent_bit(io_tree
, ordered_extent
->file_offset
,
2896 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2897 EXTENT_DEFRAG
, 0, 0, &cached_state
, GFP_NOFS
);
2901 trans
= btrfs_join_transaction_nolock(root
);
2903 trans
= btrfs_join_transaction(root
);
2904 if (IS_ERR(trans
)) {
2905 ret
= PTR_ERR(trans
);
2910 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2912 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
2913 compress_type
= ordered_extent
->compress_type
;
2914 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
2915 BUG_ON(compress_type
);
2916 ret
= btrfs_mark_extent_written(trans
, inode
,
2917 ordered_extent
->file_offset
,
2918 ordered_extent
->file_offset
+
2921 BUG_ON(root
== root
->fs_info
->tree_root
);
2922 ret
= insert_reserved_file_extent(trans
, inode
,
2923 ordered_extent
->file_offset
,
2924 ordered_extent
->start
,
2925 ordered_extent
->disk_len
,
2926 logical_len
, logical_len
,
2927 compress_type
, 0, 0,
2928 BTRFS_FILE_EXTENT_REG
);
2930 btrfs_release_delalloc_bytes(root
,
2931 ordered_extent
->start
,
2932 ordered_extent
->disk_len
);
2934 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
2935 ordered_extent
->file_offset
, ordered_extent
->len
,
2938 btrfs_abort_transaction(trans
, root
, ret
);
2942 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
2943 &ordered_extent
->list
);
2945 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2946 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2947 if (ret
) { /* -ENOMEM or corruption */
2948 btrfs_abort_transaction(trans
, root
, ret
);
2953 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
2954 ordered_extent
->file_offset
+
2955 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
2957 if (root
!= root
->fs_info
->tree_root
)
2958 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
2960 btrfs_end_transaction(trans
, root
);
2962 if (ret
|| truncated
) {
2966 start
= ordered_extent
->file_offset
+ logical_len
;
2968 start
= ordered_extent
->file_offset
;
2969 end
= ordered_extent
->file_offset
+ ordered_extent
->len
- 1;
2970 clear_extent_uptodate(io_tree
, start
, end
, NULL
, GFP_NOFS
);
2972 /* Drop the cache for the part of the extent we didn't write. */
2973 btrfs_drop_extent_cache(inode
, start
, end
, 0);
2976 * If the ordered extent had an IOERR or something else went
2977 * wrong we need to return the space for this ordered extent
2978 * back to the allocator. We only free the extent in the
2979 * truncated case if we didn't write out the extent at all.
2981 if ((ret
|| !logical_len
) &&
2982 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
2983 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
))
2984 btrfs_free_reserved_extent(root
, ordered_extent
->start
,
2985 ordered_extent
->disk_len
, 1);
2990 * This needs to be done to make sure anybody waiting knows we are done
2991 * updating everything for this ordered extent.
2993 btrfs_remove_ordered_extent(inode
, ordered_extent
);
2995 /* for snapshot-aware defrag */
2998 free_sa_defrag_extent(new);
2999 atomic_dec(&root
->fs_info
->defrag_running
);
3001 relink_file_extents(new);
3006 btrfs_put_ordered_extent(ordered_extent
);
3007 /* once for the tree */
3008 btrfs_put_ordered_extent(ordered_extent
);
3013 static void finish_ordered_fn(struct btrfs_work
*work
)
3015 struct btrfs_ordered_extent
*ordered_extent
;
3016 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
3017 btrfs_finish_ordered_io(ordered_extent
);
3020 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
3021 struct extent_state
*state
, int uptodate
)
3023 struct inode
*inode
= page
->mapping
->host
;
3024 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3025 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
3026 struct btrfs_workqueue
*wq
;
3027 btrfs_work_func_t func
;
3029 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
3031 ClearPagePrivate2(page
);
3032 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
3033 end
- start
+ 1, uptodate
))
3036 if (btrfs_is_free_space_inode(inode
)) {
3037 wq
= root
->fs_info
->endio_freespace_worker
;
3038 func
= btrfs_freespace_write_helper
;
3040 wq
= root
->fs_info
->endio_write_workers
;
3041 func
= btrfs_endio_write_helper
;
3044 btrfs_init_work(&ordered_extent
->work
, func
, finish_ordered_fn
, NULL
,
3046 btrfs_queue_work(wq
, &ordered_extent
->work
);
3051 static int __readpage_endio_check(struct inode
*inode
,
3052 struct btrfs_io_bio
*io_bio
,
3053 int icsum
, struct page
*page
,
3054 int pgoff
, u64 start
, size_t len
)
3060 csum_expected
= *(((u32
*)io_bio
->csum
) + icsum
);
3062 kaddr
= kmap_atomic(page
);
3063 csum
= btrfs_csum_data(kaddr
+ pgoff
, csum
, len
);
3064 btrfs_csum_final(csum
, (char *)&csum
);
3065 if (csum
!= csum_expected
)
3068 kunmap_atomic(kaddr
);
3071 btrfs_warn_rl(BTRFS_I(inode
)->root
->fs_info
,
3072 "csum failed ino %llu off %llu csum %u expected csum %u",
3073 btrfs_ino(inode
), start
, csum
, csum_expected
);
3074 memset(kaddr
+ pgoff
, 1, len
);
3075 flush_dcache_page(page
);
3076 kunmap_atomic(kaddr
);
3077 if (csum_expected
== 0)
3083 * when reads are done, we need to check csums to verify the data is correct
3084 * if there's a match, we allow the bio to finish. If not, the code in
3085 * extent_io.c will try to find good copies for us.
3087 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
3088 u64 phy_offset
, struct page
*page
,
3089 u64 start
, u64 end
, int mirror
)
3091 size_t offset
= start
- page_offset(page
);
3092 struct inode
*inode
= page
->mapping
->host
;
3093 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3094 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3096 if (PageChecked(page
)) {
3097 ClearPageChecked(page
);
3101 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
3104 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
3105 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
3106 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
3111 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
3112 return __readpage_endio_check(inode
, io_bio
, phy_offset
, page
, offset
,
3113 start
, (size_t)(end
- start
+ 1));
3116 void btrfs_add_delayed_iput(struct inode
*inode
)
3118 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
3119 struct btrfs_inode
*binode
= BTRFS_I(inode
);
3121 if (atomic_add_unless(&inode
->i_count
, -1, 1))
3124 spin_lock(&fs_info
->delayed_iput_lock
);
3125 if (binode
->delayed_iput_count
== 0) {
3126 ASSERT(list_empty(&binode
->delayed_iput
));
3127 list_add_tail(&binode
->delayed_iput
, &fs_info
->delayed_iputs
);
3129 binode
->delayed_iput_count
++;
3131 spin_unlock(&fs_info
->delayed_iput_lock
);
3134 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
3136 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3138 spin_lock(&fs_info
->delayed_iput_lock
);
3139 while (!list_empty(&fs_info
->delayed_iputs
)) {
3140 struct btrfs_inode
*inode
;
3142 inode
= list_first_entry(&fs_info
->delayed_iputs
,
3143 struct btrfs_inode
, delayed_iput
);
3144 if (inode
->delayed_iput_count
) {
3145 inode
->delayed_iput_count
--;
3146 list_move_tail(&inode
->delayed_iput
,
3147 &fs_info
->delayed_iputs
);
3149 list_del_init(&inode
->delayed_iput
);
3151 spin_unlock(&fs_info
->delayed_iput_lock
);
3152 iput(&inode
->vfs_inode
);
3153 spin_lock(&fs_info
->delayed_iput_lock
);
3155 spin_unlock(&fs_info
->delayed_iput_lock
);
3159 * This is called in transaction commit time. If there are no orphan
3160 * files in the subvolume, it removes orphan item and frees block_rsv
3163 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
3164 struct btrfs_root
*root
)
3166 struct btrfs_block_rsv
*block_rsv
;
3169 if (atomic_read(&root
->orphan_inodes
) ||
3170 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
3173 spin_lock(&root
->orphan_lock
);
3174 if (atomic_read(&root
->orphan_inodes
)) {
3175 spin_unlock(&root
->orphan_lock
);
3179 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
3180 spin_unlock(&root
->orphan_lock
);
3184 block_rsv
= root
->orphan_block_rsv
;
3185 root
->orphan_block_rsv
= NULL
;
3186 spin_unlock(&root
->orphan_lock
);
3188 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
) &&
3189 btrfs_root_refs(&root
->root_item
) > 0) {
3190 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
3191 root
->root_key
.objectid
);
3193 btrfs_abort_transaction(trans
, root
, ret
);
3195 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
,
3200 WARN_ON(block_rsv
->size
> 0);
3201 btrfs_free_block_rsv(root
, block_rsv
);
3206 * This creates an orphan entry for the given inode in case something goes
3207 * wrong in the middle of an unlink/truncate.
3209 * NOTE: caller of this function should reserve 5 units of metadata for
3212 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
3214 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3215 struct btrfs_block_rsv
*block_rsv
= NULL
;
3220 if (!root
->orphan_block_rsv
) {
3221 block_rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
3226 spin_lock(&root
->orphan_lock
);
3227 if (!root
->orphan_block_rsv
) {
3228 root
->orphan_block_rsv
= block_rsv
;
3229 } else if (block_rsv
) {
3230 btrfs_free_block_rsv(root
, block_rsv
);
3234 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3235 &BTRFS_I(inode
)->runtime_flags
)) {
3238 * For proper ENOSPC handling, we should do orphan
3239 * cleanup when mounting. But this introduces backward
3240 * compatibility issue.
3242 if (!xchg(&root
->orphan_item_inserted
, 1))
3248 atomic_inc(&root
->orphan_inodes
);
3251 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3252 &BTRFS_I(inode
)->runtime_flags
))
3254 spin_unlock(&root
->orphan_lock
);
3256 /* grab metadata reservation from transaction handle */
3258 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
3259 BUG_ON(ret
); /* -ENOSPC in reservation; Logic error? JDM */
3262 /* insert an orphan item to track this unlinked/truncated file */
3264 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
3266 atomic_dec(&root
->orphan_inodes
);
3268 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3269 &BTRFS_I(inode
)->runtime_flags
);
3270 btrfs_orphan_release_metadata(inode
);
3272 if (ret
!= -EEXIST
) {
3273 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3274 &BTRFS_I(inode
)->runtime_flags
);
3275 btrfs_abort_transaction(trans
, root
, ret
);
3282 /* insert an orphan item to track subvolume contains orphan files */
3284 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
3285 root
->root_key
.objectid
);
3286 if (ret
&& ret
!= -EEXIST
) {
3287 btrfs_abort_transaction(trans
, root
, ret
);
3295 * We have done the truncate/delete so we can go ahead and remove the orphan
3296 * item for this particular inode.
3298 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3299 struct inode
*inode
)
3301 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3302 int delete_item
= 0;
3303 int release_rsv
= 0;
3306 spin_lock(&root
->orphan_lock
);
3307 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3308 &BTRFS_I(inode
)->runtime_flags
))
3311 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3312 &BTRFS_I(inode
)->runtime_flags
))
3314 spin_unlock(&root
->orphan_lock
);
3317 atomic_dec(&root
->orphan_inodes
);
3319 ret
= btrfs_del_orphan_item(trans
, root
,
3324 btrfs_orphan_release_metadata(inode
);
3330 * this cleans up any orphans that may be left on the list from the last use
3333 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3335 struct btrfs_path
*path
;
3336 struct extent_buffer
*leaf
;
3337 struct btrfs_key key
, found_key
;
3338 struct btrfs_trans_handle
*trans
;
3339 struct inode
*inode
;
3340 u64 last_objectid
= 0;
3341 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
3343 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
3346 path
= btrfs_alloc_path();
3351 path
->reada
= READA_BACK
;
3353 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3354 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
3355 key
.offset
= (u64
)-1;
3358 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3363 * if ret == 0 means we found what we were searching for, which
3364 * is weird, but possible, so only screw with path if we didn't
3365 * find the key and see if we have stuff that matches
3369 if (path
->slots
[0] == 0)
3374 /* pull out the item */
3375 leaf
= path
->nodes
[0];
3376 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3378 /* make sure the item matches what we want */
3379 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3381 if (found_key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
3384 /* release the path since we're done with it */
3385 btrfs_release_path(path
);
3388 * this is where we are basically btrfs_lookup, without the
3389 * crossing root thing. we store the inode number in the
3390 * offset of the orphan item.
3393 if (found_key
.offset
== last_objectid
) {
3394 btrfs_err(root
->fs_info
,
3395 "Error removing orphan entry, stopping orphan cleanup");
3400 last_objectid
= found_key
.offset
;
3402 found_key
.objectid
= found_key
.offset
;
3403 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3404 found_key
.offset
= 0;
3405 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
3406 ret
= PTR_ERR_OR_ZERO(inode
);
3407 if (ret
&& ret
!= -ESTALE
)
3410 if (ret
== -ESTALE
&& root
== root
->fs_info
->tree_root
) {
3411 struct btrfs_root
*dead_root
;
3412 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3413 int is_dead_root
= 0;
3416 * this is an orphan in the tree root. Currently these
3417 * could come from 2 sources:
3418 * a) a snapshot deletion in progress
3419 * b) a free space cache inode
3420 * We need to distinguish those two, as the snapshot
3421 * orphan must not get deleted.
3422 * find_dead_roots already ran before us, so if this
3423 * is a snapshot deletion, we should find the root
3424 * in the dead_roots list
3426 spin_lock(&fs_info
->trans_lock
);
3427 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
3429 if (dead_root
->root_key
.objectid
==
3430 found_key
.objectid
) {
3435 spin_unlock(&fs_info
->trans_lock
);
3437 /* prevent this orphan from being found again */
3438 key
.offset
= found_key
.objectid
- 1;
3443 * Inode is already gone but the orphan item is still there,
3444 * kill the orphan item.
3446 if (ret
== -ESTALE
) {
3447 trans
= btrfs_start_transaction(root
, 1);
3448 if (IS_ERR(trans
)) {
3449 ret
= PTR_ERR(trans
);
3452 btrfs_debug(root
->fs_info
, "auto deleting %Lu",
3453 found_key
.objectid
);
3454 ret
= btrfs_del_orphan_item(trans
, root
,
3455 found_key
.objectid
);
3456 btrfs_end_transaction(trans
, root
);
3463 * add this inode to the orphan list so btrfs_orphan_del does
3464 * the proper thing when we hit it
3466 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3467 &BTRFS_I(inode
)->runtime_flags
);
3468 atomic_inc(&root
->orphan_inodes
);
3470 /* if we have links, this was a truncate, lets do that */
3471 if (inode
->i_nlink
) {
3472 if (WARN_ON(!S_ISREG(inode
->i_mode
))) {
3478 /* 1 for the orphan item deletion. */
3479 trans
= btrfs_start_transaction(root
, 1);
3480 if (IS_ERR(trans
)) {
3482 ret
= PTR_ERR(trans
);
3485 ret
= btrfs_orphan_add(trans
, inode
);
3486 btrfs_end_transaction(trans
, root
);
3492 ret
= btrfs_truncate(inode
);
3494 btrfs_orphan_del(NULL
, inode
);
3499 /* this will do delete_inode and everything for us */
3504 /* release the path since we're done with it */
3505 btrfs_release_path(path
);
3507 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
3509 if (root
->orphan_block_rsv
)
3510 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
3513 if (root
->orphan_block_rsv
||
3514 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
)) {
3515 trans
= btrfs_join_transaction(root
);
3517 btrfs_end_transaction(trans
, root
);
3521 btrfs_debug(root
->fs_info
, "unlinked %d orphans", nr_unlink
);
3523 btrfs_debug(root
->fs_info
, "truncated %d orphans", nr_truncate
);
3527 btrfs_err(root
->fs_info
,
3528 "could not do orphan cleanup %d", ret
);
3529 btrfs_free_path(path
);
3534 * very simple check to peek ahead in the leaf looking for xattrs. If we
3535 * don't find any xattrs, we know there can't be any acls.
3537 * slot is the slot the inode is in, objectid is the objectid of the inode
3539 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3540 int slot
, u64 objectid
,
3541 int *first_xattr_slot
)
3543 u32 nritems
= btrfs_header_nritems(leaf
);
3544 struct btrfs_key found_key
;
3545 static u64 xattr_access
= 0;
3546 static u64 xattr_default
= 0;
3549 if (!xattr_access
) {
3550 xattr_access
= btrfs_name_hash(POSIX_ACL_XATTR_ACCESS
,
3551 strlen(POSIX_ACL_XATTR_ACCESS
));
3552 xattr_default
= btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT
,
3553 strlen(POSIX_ACL_XATTR_DEFAULT
));
3557 *first_xattr_slot
= -1;
3558 while (slot
< nritems
) {
3559 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3561 /* we found a different objectid, there must not be acls */
3562 if (found_key
.objectid
!= objectid
)
3565 /* we found an xattr, assume we've got an acl */
3566 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
3567 if (*first_xattr_slot
== -1)
3568 *first_xattr_slot
= slot
;
3569 if (found_key
.offset
== xattr_access
||
3570 found_key
.offset
== xattr_default
)
3575 * we found a key greater than an xattr key, there can't
3576 * be any acls later on
3578 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3585 * it goes inode, inode backrefs, xattrs, extents,
3586 * so if there are a ton of hard links to an inode there can
3587 * be a lot of backrefs. Don't waste time searching too hard,
3588 * this is just an optimization
3593 /* we hit the end of the leaf before we found an xattr or
3594 * something larger than an xattr. We have to assume the inode
3597 if (*first_xattr_slot
== -1)
3598 *first_xattr_slot
= slot
;
3603 * read an inode from the btree into the in-memory inode
3605 static void btrfs_read_locked_inode(struct inode
*inode
)
3607 struct btrfs_path
*path
;
3608 struct extent_buffer
*leaf
;
3609 struct btrfs_inode_item
*inode_item
;
3610 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3611 struct btrfs_key location
;
3616 bool filled
= false;
3617 int first_xattr_slot
;
3619 ret
= btrfs_fill_inode(inode
, &rdev
);
3623 path
= btrfs_alloc_path();
3627 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3629 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3633 leaf
= path
->nodes
[0];
3638 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3639 struct btrfs_inode_item
);
3640 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3641 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3642 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3643 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3644 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
3646 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->atime
);
3647 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->atime
);
3649 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->mtime
);
3650 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->mtime
);
3652 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->ctime
);
3653 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->ctime
);
3655 BTRFS_I(inode
)->i_otime
.tv_sec
=
3656 btrfs_timespec_sec(leaf
, &inode_item
->otime
);
3657 BTRFS_I(inode
)->i_otime
.tv_nsec
=
3658 btrfs_timespec_nsec(leaf
, &inode_item
->otime
);
3660 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3661 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3662 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3664 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
3665 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3667 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3669 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3670 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
3674 * If we were modified in the current generation and evicted from memory
3675 * and then re-read we need to do a full sync since we don't have any
3676 * idea about which extents were modified before we were evicted from
3679 * This is required for both inode re-read from disk and delayed inode
3680 * in delayed_nodes_tree.
3682 if (BTRFS_I(inode
)->last_trans
== root
->fs_info
->generation
)
3683 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3684 &BTRFS_I(inode
)->runtime_flags
);
3687 * We don't persist the id of the transaction where an unlink operation
3688 * against the inode was last made. So here we assume the inode might
3689 * have been evicted, and therefore the exact value of last_unlink_trans
3690 * lost, and set it to last_trans to avoid metadata inconsistencies
3691 * between the inode and its parent if the inode is fsync'ed and the log
3692 * replayed. For example, in the scenario:
3695 * ln mydir/foo mydir/bar
3698 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3699 * xfs_io -c fsync mydir/foo
3701 * mount fs, triggers fsync log replay
3703 * We must make sure that when we fsync our inode foo we also log its
3704 * parent inode, otherwise after log replay the parent still has the
3705 * dentry with the "bar" name but our inode foo has a link count of 1
3706 * and doesn't have an inode ref with the name "bar" anymore.
3708 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3709 * but it guarantees correctness at the expense of ocassional full
3710 * transaction commits on fsync if our inode is a directory, or if our
3711 * inode is not a directory, logging its parent unnecessarily.
3713 BTRFS_I(inode
)->last_unlink_trans
= BTRFS_I(inode
)->last_trans
;
3716 if (inode
->i_nlink
!= 1 ||
3717 path
->slots
[0] >= btrfs_header_nritems(leaf
))
3720 btrfs_item_key_to_cpu(leaf
, &location
, path
->slots
[0]);
3721 if (location
.objectid
!= btrfs_ino(inode
))
3724 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3725 if (location
.type
== BTRFS_INODE_REF_KEY
) {
3726 struct btrfs_inode_ref
*ref
;
3728 ref
= (struct btrfs_inode_ref
*)ptr
;
3729 BTRFS_I(inode
)->dir_index
= btrfs_inode_ref_index(leaf
, ref
);
3730 } else if (location
.type
== BTRFS_INODE_EXTREF_KEY
) {
3731 struct btrfs_inode_extref
*extref
;
3733 extref
= (struct btrfs_inode_extref
*)ptr
;
3734 BTRFS_I(inode
)->dir_index
= btrfs_inode_extref_index(leaf
,
3739 * try to precache a NULL acl entry for files that don't have
3740 * any xattrs or acls
3742 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
3743 btrfs_ino(inode
), &first_xattr_slot
);
3744 if (first_xattr_slot
!= -1) {
3745 path
->slots
[0] = first_xattr_slot
;
3746 ret
= btrfs_load_inode_props(inode
, path
);
3748 btrfs_err(root
->fs_info
,
3749 "error loading props for ino %llu (root %llu): %d",
3751 root
->root_key
.objectid
, ret
);
3753 btrfs_free_path(path
);
3756 cache_no_acl(inode
);
3758 switch (inode
->i_mode
& S_IFMT
) {
3760 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3761 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3762 inode
->i_fop
= &btrfs_file_operations
;
3763 inode
->i_op
= &btrfs_file_inode_operations
;
3766 inode
->i_fop
= &btrfs_dir_file_operations
;
3767 if (root
== root
->fs_info
->tree_root
)
3768 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
3770 inode
->i_op
= &btrfs_dir_inode_operations
;
3773 inode
->i_op
= &btrfs_symlink_inode_operations
;
3774 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
3777 inode
->i_op
= &btrfs_special_inode_operations
;
3778 init_special_inode(inode
, inode
->i_mode
, rdev
);
3782 btrfs_update_iflags(inode
);
3786 btrfs_free_path(path
);
3787 make_bad_inode(inode
);
3791 * given a leaf and an inode, copy the inode fields into the leaf
3793 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3794 struct extent_buffer
*leaf
,
3795 struct btrfs_inode_item
*item
,
3796 struct inode
*inode
)
3798 struct btrfs_map_token token
;
3800 btrfs_init_map_token(&token
);
3802 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3803 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3804 btrfs_set_token_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
,
3806 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3807 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3809 btrfs_set_token_timespec_sec(leaf
, &item
->atime
,
3810 inode
->i_atime
.tv_sec
, &token
);
3811 btrfs_set_token_timespec_nsec(leaf
, &item
->atime
,
3812 inode
->i_atime
.tv_nsec
, &token
);
3814 btrfs_set_token_timespec_sec(leaf
, &item
->mtime
,
3815 inode
->i_mtime
.tv_sec
, &token
);
3816 btrfs_set_token_timespec_nsec(leaf
, &item
->mtime
,
3817 inode
->i_mtime
.tv_nsec
, &token
);
3819 btrfs_set_token_timespec_sec(leaf
, &item
->ctime
,
3820 inode
->i_ctime
.tv_sec
, &token
);
3821 btrfs_set_token_timespec_nsec(leaf
, &item
->ctime
,
3822 inode
->i_ctime
.tv_nsec
, &token
);
3824 btrfs_set_token_timespec_sec(leaf
, &item
->otime
,
3825 BTRFS_I(inode
)->i_otime
.tv_sec
, &token
);
3826 btrfs_set_token_timespec_nsec(leaf
, &item
->otime
,
3827 BTRFS_I(inode
)->i_otime
.tv_nsec
, &token
);
3829 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3831 btrfs_set_token_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
,
3833 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
3834 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3835 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3836 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3837 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3841 * copy everything in the in-memory inode into the btree.
3843 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
3844 struct btrfs_root
*root
, struct inode
*inode
)
3846 struct btrfs_inode_item
*inode_item
;
3847 struct btrfs_path
*path
;
3848 struct extent_buffer
*leaf
;
3851 path
= btrfs_alloc_path();
3855 path
->leave_spinning
= 1;
3856 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
3864 leaf
= path
->nodes
[0];
3865 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3866 struct btrfs_inode_item
);
3868 fill_inode_item(trans
, leaf
, inode_item
, inode
);
3869 btrfs_mark_buffer_dirty(leaf
);
3870 btrfs_set_inode_last_trans(trans
, inode
);
3873 btrfs_free_path(path
);
3878 * copy everything in the in-memory inode into the btree.
3880 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
3881 struct btrfs_root
*root
, struct inode
*inode
)
3886 * If the inode is a free space inode, we can deadlock during commit
3887 * if we put it into the delayed code.
3889 * The data relocation inode should also be directly updated
3892 if (!btrfs_is_free_space_inode(inode
)
3893 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
3894 && !root
->fs_info
->log_root_recovering
) {
3895 btrfs_update_root_times(trans
, root
);
3897 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
3899 btrfs_set_inode_last_trans(trans
, inode
);
3903 return btrfs_update_inode_item(trans
, root
, inode
);
3906 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
3907 struct btrfs_root
*root
,
3908 struct inode
*inode
)
3912 ret
= btrfs_update_inode(trans
, root
, inode
);
3914 return btrfs_update_inode_item(trans
, root
, inode
);
3919 * unlink helper that gets used here in inode.c and in the tree logging
3920 * recovery code. It remove a link in a directory with a given name, and
3921 * also drops the back refs in the inode to the directory
3923 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3924 struct btrfs_root
*root
,
3925 struct inode
*dir
, struct inode
*inode
,
3926 const char *name
, int name_len
)
3928 struct btrfs_path
*path
;
3930 struct extent_buffer
*leaf
;
3931 struct btrfs_dir_item
*di
;
3932 struct btrfs_key key
;
3934 u64 ino
= btrfs_ino(inode
);
3935 u64 dir_ino
= btrfs_ino(dir
);
3937 path
= btrfs_alloc_path();
3943 path
->leave_spinning
= 1;
3944 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3945 name
, name_len
, -1);
3954 leaf
= path
->nodes
[0];
3955 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3956 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3959 btrfs_release_path(path
);
3962 * If we don't have dir index, we have to get it by looking up
3963 * the inode ref, since we get the inode ref, remove it directly,
3964 * it is unnecessary to do delayed deletion.
3966 * But if we have dir index, needn't search inode ref to get it.
3967 * Since the inode ref is close to the inode item, it is better
3968 * that we delay to delete it, and just do this deletion when
3969 * we update the inode item.
3971 if (BTRFS_I(inode
)->dir_index
) {
3972 ret
= btrfs_delayed_delete_inode_ref(inode
);
3974 index
= BTRFS_I(inode
)->dir_index
;
3979 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
3982 btrfs_info(root
->fs_info
,
3983 "failed to delete reference to %.*s, inode %llu parent %llu",
3984 name_len
, name
, ino
, dir_ino
);
3985 btrfs_abort_transaction(trans
, root
, ret
);
3989 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3991 btrfs_abort_transaction(trans
, root
, ret
);
3995 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
3997 if (ret
!= 0 && ret
!= -ENOENT
) {
3998 btrfs_abort_transaction(trans
, root
, ret
);
4002 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
4007 btrfs_abort_transaction(trans
, root
, ret
);
4009 btrfs_free_path(path
);
4013 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
4014 inode_inc_iversion(inode
);
4015 inode_inc_iversion(dir
);
4016 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
4017 ret
= btrfs_update_inode(trans
, root
, dir
);
4022 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4023 struct btrfs_root
*root
,
4024 struct inode
*dir
, struct inode
*inode
,
4025 const char *name
, int name_len
)
4028 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
4031 ret
= btrfs_update_inode(trans
, root
, inode
);
4037 * helper to start transaction for unlink and rmdir.
4039 * unlink and rmdir are special in btrfs, they do not always free space, so
4040 * if we cannot make our reservations the normal way try and see if there is
4041 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4042 * allow the unlink to occur.
4044 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
)
4046 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4049 * 1 for the possible orphan item
4050 * 1 for the dir item
4051 * 1 for the dir index
4052 * 1 for the inode ref
4055 return btrfs_start_transaction_fallback_global_rsv(root
, 5, 5);
4058 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
4060 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4061 struct btrfs_trans_handle
*trans
;
4062 struct inode
*inode
= d_inode(dentry
);
4065 trans
= __unlink_start_trans(dir
);
4067 return PTR_ERR(trans
);
4069 btrfs_record_unlink_dir(trans
, dir
, d_inode(dentry
), 0);
4071 ret
= btrfs_unlink_inode(trans
, root
, dir
, d_inode(dentry
),
4072 dentry
->d_name
.name
, dentry
->d_name
.len
);
4076 if (inode
->i_nlink
== 0) {
4077 ret
= btrfs_orphan_add(trans
, inode
);
4083 btrfs_end_transaction(trans
, root
);
4084 btrfs_btree_balance_dirty(root
);
4088 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
4089 struct btrfs_root
*root
,
4090 struct inode
*dir
, u64 objectid
,
4091 const char *name
, int name_len
)
4093 struct btrfs_path
*path
;
4094 struct extent_buffer
*leaf
;
4095 struct btrfs_dir_item
*di
;
4096 struct btrfs_key key
;
4099 u64 dir_ino
= btrfs_ino(dir
);
4101 path
= btrfs_alloc_path();
4105 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4106 name
, name_len
, -1);
4107 if (IS_ERR_OR_NULL(di
)) {
4115 leaf
= path
->nodes
[0];
4116 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4117 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
4118 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4120 btrfs_abort_transaction(trans
, root
, ret
);
4123 btrfs_release_path(path
);
4125 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
4126 objectid
, root
->root_key
.objectid
,
4127 dir_ino
, &index
, name
, name_len
);
4129 if (ret
!= -ENOENT
) {
4130 btrfs_abort_transaction(trans
, root
, ret
);
4133 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
4135 if (IS_ERR_OR_NULL(di
)) {
4140 btrfs_abort_transaction(trans
, root
, ret
);
4144 leaf
= path
->nodes
[0];
4145 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4146 btrfs_release_path(path
);
4149 btrfs_release_path(path
);
4151 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
4153 btrfs_abort_transaction(trans
, root
, ret
);
4157 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
4158 inode_inc_iversion(dir
);
4159 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
4160 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
4162 btrfs_abort_transaction(trans
, root
, ret
);
4164 btrfs_free_path(path
);
4168 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
4170 struct inode
*inode
= d_inode(dentry
);
4172 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4173 struct btrfs_trans_handle
*trans
;
4175 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
4177 if (btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
4180 trans
= __unlink_start_trans(dir
);
4182 return PTR_ERR(trans
);
4184 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
4185 err
= btrfs_unlink_subvol(trans
, root
, dir
,
4186 BTRFS_I(inode
)->location
.objectid
,
4187 dentry
->d_name
.name
,
4188 dentry
->d_name
.len
);
4192 err
= btrfs_orphan_add(trans
, inode
);
4196 /* now the directory is empty */
4197 err
= btrfs_unlink_inode(trans
, root
, dir
, d_inode(dentry
),
4198 dentry
->d_name
.name
, dentry
->d_name
.len
);
4200 btrfs_i_size_write(inode
, 0);
4202 btrfs_end_transaction(trans
, root
);
4203 btrfs_btree_balance_dirty(root
);
4208 static int truncate_space_check(struct btrfs_trans_handle
*trans
,
4209 struct btrfs_root
*root
,
4214 bytes_deleted
= btrfs_csum_bytes_to_leaves(root
, bytes_deleted
);
4215 ret
= btrfs_block_rsv_add(root
, &root
->fs_info
->trans_block_rsv
,
4216 bytes_deleted
, BTRFS_RESERVE_NO_FLUSH
);
4218 trans
->bytes_reserved
+= bytes_deleted
;
4223 static int truncate_inline_extent(struct inode
*inode
,
4224 struct btrfs_path
*path
,
4225 struct btrfs_key
*found_key
,
4229 struct extent_buffer
*leaf
= path
->nodes
[0];
4230 int slot
= path
->slots
[0];
4231 struct btrfs_file_extent_item
*fi
;
4232 u32 size
= (u32
)(new_size
- found_key
->offset
);
4233 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4235 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
4237 if (btrfs_file_extent_compression(leaf
, fi
) != BTRFS_COMPRESS_NONE
) {
4238 loff_t offset
= new_size
;
4239 loff_t page_end
= ALIGN(offset
, PAGE_CACHE_SIZE
);
4242 * Zero out the remaining of the last page of our inline extent,
4243 * instead of directly truncating our inline extent here - that
4244 * would be much more complex (decompressing all the data, then
4245 * compressing the truncated data, which might be bigger than
4246 * the size of the inline extent, resize the extent, etc).
4247 * We release the path because to get the page we might need to
4248 * read the extent item from disk (data not in the page cache).
4250 btrfs_release_path(path
);
4251 return btrfs_truncate_block(inode
, offset
, page_end
- offset
,
4255 btrfs_set_file_extent_ram_bytes(leaf
, fi
, size
);
4256 size
= btrfs_file_extent_calc_inline_size(size
);
4257 btrfs_truncate_item(root
, path
, size
, 1);
4259 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4260 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4266 * this can truncate away extent items, csum items and directory items.
4267 * It starts at a high offset and removes keys until it can't find
4268 * any higher than new_size
4270 * csum items that cross the new i_size are truncated to the new size
4273 * min_type is the minimum key type to truncate down to. If set to 0, this
4274 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4276 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
4277 struct btrfs_root
*root
,
4278 struct inode
*inode
,
4279 u64 new_size
, u32 min_type
)
4281 struct btrfs_path
*path
;
4282 struct extent_buffer
*leaf
;
4283 struct btrfs_file_extent_item
*fi
;
4284 struct btrfs_key key
;
4285 struct btrfs_key found_key
;
4286 u64 extent_start
= 0;
4287 u64 extent_num_bytes
= 0;
4288 u64 extent_offset
= 0;
4290 u64 last_size
= new_size
;
4291 u32 found_type
= (u8
)-1;
4294 int pending_del_nr
= 0;
4295 int pending_del_slot
= 0;
4296 int extent_type
= -1;
4299 u64 ino
= btrfs_ino(inode
);
4300 u64 bytes_deleted
= 0;
4302 bool should_throttle
= 0;
4303 bool should_end
= 0;
4305 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
4308 * for non-free space inodes and ref cows, we want to back off from
4311 if (!btrfs_is_free_space_inode(inode
) &&
4312 test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4315 path
= btrfs_alloc_path();
4318 path
->reada
= READA_BACK
;
4321 * We want to drop from the next block forward in case this new size is
4322 * not block aligned since we will be keeping the last block of the
4323 * extent just the way it is.
4325 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4326 root
== root
->fs_info
->tree_root
)
4327 btrfs_drop_extent_cache(inode
, ALIGN(new_size
,
4328 root
->sectorsize
), (u64
)-1, 0);
4331 * This function is also used to drop the items in the log tree before
4332 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4333 * it is used to drop the loged items. So we shouldn't kill the delayed
4336 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
4337 btrfs_kill_delayed_inode_items(inode
);
4340 key
.offset
= (u64
)-1;
4345 * with a 16K leaf size and 128MB extents, you can actually queue
4346 * up a huge file in a single leaf. Most of the time that
4347 * bytes_deleted is > 0, it will be huge by the time we get here
4349 if (be_nice
&& bytes_deleted
> SZ_32M
) {
4350 if (btrfs_should_end_transaction(trans
, root
)) {
4357 path
->leave_spinning
= 1;
4358 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
4365 /* there are no items in the tree for us to truncate, we're
4368 if (path
->slots
[0] == 0)
4375 leaf
= path
->nodes
[0];
4376 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4377 found_type
= found_key
.type
;
4379 if (found_key
.objectid
!= ino
)
4382 if (found_type
< min_type
)
4385 item_end
= found_key
.offset
;
4386 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
4387 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4388 struct btrfs_file_extent_item
);
4389 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4390 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4392 btrfs_file_extent_num_bytes(leaf
, fi
);
4393 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4394 item_end
+= btrfs_file_extent_inline_len(leaf
,
4395 path
->slots
[0], fi
);
4399 if (found_type
> min_type
) {
4402 if (item_end
< new_size
)
4404 if (found_key
.offset
>= new_size
)
4410 /* FIXME, shrink the extent if the ref count is only 1 */
4411 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
4415 last_size
= found_key
.offset
;
4417 last_size
= new_size
;
4419 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4421 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4423 u64 orig_num_bytes
=
4424 btrfs_file_extent_num_bytes(leaf
, fi
);
4425 extent_num_bytes
= ALIGN(new_size
-
4428 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4430 num_dec
= (orig_num_bytes
-
4432 if (test_bit(BTRFS_ROOT_REF_COWS
,
4435 inode_sub_bytes(inode
, num_dec
);
4436 btrfs_mark_buffer_dirty(leaf
);
4439 btrfs_file_extent_disk_num_bytes(leaf
,
4441 extent_offset
= found_key
.offset
-
4442 btrfs_file_extent_offset(leaf
, fi
);
4444 /* FIXME blocksize != 4096 */
4445 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
4446 if (extent_start
!= 0) {
4448 if (test_bit(BTRFS_ROOT_REF_COWS
,
4450 inode_sub_bytes(inode
, num_dec
);
4453 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4455 * we can't truncate inline items that have had
4459 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
4460 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
4463 * Need to release path in order to truncate a
4464 * compressed extent. So delete any accumulated
4465 * extent items so far.
4467 if (btrfs_file_extent_compression(leaf
, fi
) !=
4468 BTRFS_COMPRESS_NONE
&& pending_del_nr
) {
4469 err
= btrfs_del_items(trans
, root
, path
,
4473 btrfs_abort_transaction(trans
,
4481 err
= truncate_inline_extent(inode
, path
,
4486 btrfs_abort_transaction(trans
,
4490 } else if (test_bit(BTRFS_ROOT_REF_COWS
,
4492 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4497 if (!pending_del_nr
) {
4498 /* no pending yet, add ourselves */
4499 pending_del_slot
= path
->slots
[0];
4501 } else if (pending_del_nr
&&
4502 path
->slots
[0] + 1 == pending_del_slot
) {
4503 /* hop on the pending chunk */
4505 pending_del_slot
= path
->slots
[0];
4512 should_throttle
= 0;
4515 (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4516 root
== root
->fs_info
->tree_root
)) {
4517 btrfs_set_path_blocking(path
);
4518 bytes_deleted
+= extent_num_bytes
;
4519 ret
= btrfs_free_extent(trans
, root
, extent_start
,
4520 extent_num_bytes
, 0,
4521 btrfs_header_owner(leaf
),
4522 ino
, extent_offset
);
4524 if (btrfs_should_throttle_delayed_refs(trans
, root
))
4525 btrfs_async_run_delayed_refs(root
,
4526 trans
->delayed_ref_updates
* 2, 0);
4528 if (truncate_space_check(trans
, root
,
4529 extent_num_bytes
)) {
4532 if (btrfs_should_throttle_delayed_refs(trans
,
4534 should_throttle
= 1;
4539 if (found_type
== BTRFS_INODE_ITEM_KEY
)
4542 if (path
->slots
[0] == 0 ||
4543 path
->slots
[0] != pending_del_slot
||
4544 should_throttle
|| should_end
) {
4545 if (pending_del_nr
) {
4546 ret
= btrfs_del_items(trans
, root
, path
,
4550 btrfs_abort_transaction(trans
,
4556 btrfs_release_path(path
);
4557 if (should_throttle
) {
4558 unsigned long updates
= trans
->delayed_ref_updates
;
4560 trans
->delayed_ref_updates
= 0;
4561 ret
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
4567 * if we failed to refill our space rsv, bail out
4568 * and let the transaction restart
4580 if (pending_del_nr
) {
4581 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
4584 btrfs_abort_transaction(trans
, root
, ret
);
4587 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
4588 btrfs_ordered_update_i_size(inode
, last_size
, NULL
);
4590 btrfs_free_path(path
);
4592 if (be_nice
&& bytes_deleted
> SZ_32M
) {
4593 unsigned long updates
= trans
->delayed_ref_updates
;
4595 trans
->delayed_ref_updates
= 0;
4596 ret
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
4605 * btrfs_truncate_block - read, zero a chunk and write a block
4606 * @inode - inode that we're zeroing
4607 * @from - the offset to start zeroing
4608 * @len - the length to zero, 0 to zero the entire range respective to the
4610 * @front - zero up to the offset instead of from the offset on
4612 * This will find the block for the "from" offset and cow the block and zero the
4613 * part we want to zero. This is used with truncate and hole punching.
4615 int btrfs_truncate_block(struct inode
*inode
, loff_t from
, loff_t len
,
4618 struct address_space
*mapping
= inode
->i_mapping
;
4619 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4620 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4621 struct btrfs_ordered_extent
*ordered
;
4622 struct extent_state
*cached_state
= NULL
;
4624 u32 blocksize
= root
->sectorsize
;
4625 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
4626 unsigned offset
= from
& (blocksize
- 1);
4628 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4633 if ((offset
& (blocksize
- 1)) == 0 &&
4634 (!len
|| ((len
& (blocksize
- 1)) == 0)))
4637 ret
= btrfs_delalloc_reserve_space(inode
,
4638 round_down(from
, blocksize
), blocksize
);
4643 page
= find_or_create_page(mapping
, index
, mask
);
4645 btrfs_delalloc_release_space(inode
,
4646 round_down(from
, blocksize
),
4652 block_start
= round_down(from
, blocksize
);
4653 block_end
= block_start
+ blocksize
- 1;
4655 if (!PageUptodate(page
)) {
4656 ret
= btrfs_readpage(NULL
, page
);
4658 if (page
->mapping
!= mapping
) {
4660 page_cache_release(page
);
4663 if (!PageUptodate(page
)) {
4668 wait_on_page_writeback(page
);
4670 lock_extent_bits(io_tree
, block_start
, block_end
, &cached_state
);
4671 set_page_extent_mapped(page
);
4673 ordered
= btrfs_lookup_ordered_extent(inode
, block_start
);
4675 unlock_extent_cached(io_tree
, block_start
, block_end
,
4676 &cached_state
, GFP_NOFS
);
4678 page_cache_release(page
);
4679 btrfs_start_ordered_extent(inode
, ordered
, 1);
4680 btrfs_put_ordered_extent(ordered
);
4684 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, block_start
, block_end
,
4685 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4686 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4687 0, 0, &cached_state
, GFP_NOFS
);
4689 ret
= btrfs_set_extent_delalloc(inode
, block_start
, block_end
,
4692 unlock_extent_cached(io_tree
, block_start
, block_end
,
4693 &cached_state
, GFP_NOFS
);
4697 if (offset
!= blocksize
) {
4699 len
= blocksize
- offset
;
4702 memset(kaddr
+ (block_start
- page_offset(page
)),
4705 memset(kaddr
+ (block_start
- page_offset(page
)) + offset
,
4707 flush_dcache_page(page
);
4710 ClearPageChecked(page
);
4711 set_page_dirty(page
);
4712 unlock_extent_cached(io_tree
, block_start
, block_end
, &cached_state
,
4717 btrfs_delalloc_release_space(inode
, block_start
,
4720 page_cache_release(page
);
4725 static int maybe_insert_hole(struct btrfs_root
*root
, struct inode
*inode
,
4726 u64 offset
, u64 len
)
4728 struct btrfs_trans_handle
*trans
;
4732 * Still need to make sure the inode looks like it's been updated so
4733 * that any holes get logged if we fsync.
4735 if (btrfs_fs_incompat(root
->fs_info
, NO_HOLES
)) {
4736 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
4737 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
4738 BTRFS_I(inode
)->last_log_commit
= root
->last_log_commit
;
4743 * 1 - for the one we're dropping
4744 * 1 - for the one we're adding
4745 * 1 - for updating the inode.
4747 trans
= btrfs_start_transaction(root
, 3);
4749 return PTR_ERR(trans
);
4751 ret
= btrfs_drop_extents(trans
, root
, inode
, offset
, offset
+ len
, 1);
4753 btrfs_abort_transaction(trans
, root
, ret
);
4754 btrfs_end_transaction(trans
, root
);
4758 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(inode
), offset
,
4759 0, 0, len
, 0, len
, 0, 0, 0);
4761 btrfs_abort_transaction(trans
, root
, ret
);
4763 btrfs_update_inode(trans
, root
, inode
);
4764 btrfs_end_transaction(trans
, root
);
4769 * This function puts in dummy file extents for the area we're creating a hole
4770 * for. So if we are truncating this file to a larger size we need to insert
4771 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4772 * the range between oldsize and size
4774 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
4776 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4777 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4778 struct extent_map
*em
= NULL
;
4779 struct extent_state
*cached_state
= NULL
;
4780 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4781 u64 hole_start
= ALIGN(oldsize
, root
->sectorsize
);
4782 u64 block_end
= ALIGN(size
, root
->sectorsize
);
4789 * If our size started in the middle of a block we need to zero out the
4790 * rest of the block before we expand the i_size, otherwise we could
4791 * expose stale data.
4793 err
= btrfs_truncate_block(inode
, oldsize
, 0, 0);
4797 if (size
<= hole_start
)
4801 struct btrfs_ordered_extent
*ordered
;
4803 lock_extent_bits(io_tree
, hole_start
, block_end
- 1,
4805 ordered
= btrfs_lookup_ordered_range(inode
, hole_start
,
4806 block_end
- hole_start
);
4809 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
4810 &cached_state
, GFP_NOFS
);
4811 btrfs_start_ordered_extent(inode
, ordered
, 1);
4812 btrfs_put_ordered_extent(ordered
);
4815 cur_offset
= hole_start
;
4817 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4818 block_end
- cur_offset
, 0);
4824 last_byte
= min(extent_map_end(em
), block_end
);
4825 last_byte
= ALIGN(last_byte
, root
->sectorsize
);
4826 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
4827 struct extent_map
*hole_em
;
4828 hole_size
= last_byte
- cur_offset
;
4830 err
= maybe_insert_hole(root
, inode
, cur_offset
,
4834 btrfs_drop_extent_cache(inode
, cur_offset
,
4835 cur_offset
+ hole_size
- 1, 0);
4836 hole_em
= alloc_extent_map();
4838 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
4839 &BTRFS_I(inode
)->runtime_flags
);
4842 hole_em
->start
= cur_offset
;
4843 hole_em
->len
= hole_size
;
4844 hole_em
->orig_start
= cur_offset
;
4846 hole_em
->block_start
= EXTENT_MAP_HOLE
;
4847 hole_em
->block_len
= 0;
4848 hole_em
->orig_block_len
= 0;
4849 hole_em
->ram_bytes
= hole_size
;
4850 hole_em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4851 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
4852 hole_em
->generation
= root
->fs_info
->generation
;
4855 write_lock(&em_tree
->lock
);
4856 err
= add_extent_mapping(em_tree
, hole_em
, 1);
4857 write_unlock(&em_tree
->lock
);
4860 btrfs_drop_extent_cache(inode
, cur_offset
,
4864 free_extent_map(hole_em
);
4867 free_extent_map(em
);
4869 cur_offset
= last_byte
;
4870 if (cur_offset
>= block_end
)
4873 free_extent_map(em
);
4874 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
4879 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
4881 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4882 struct btrfs_trans_handle
*trans
;
4883 loff_t oldsize
= i_size_read(inode
);
4884 loff_t newsize
= attr
->ia_size
;
4885 int mask
= attr
->ia_valid
;
4889 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4890 * special case where we need to update the times despite not having
4891 * these flags set. For all other operations the VFS set these flags
4892 * explicitly if it wants a timestamp update.
4894 if (newsize
!= oldsize
) {
4895 inode_inc_iversion(inode
);
4896 if (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))
4897 inode
->i_ctime
= inode
->i_mtime
=
4898 current_fs_time(inode
->i_sb
);
4901 if (newsize
> oldsize
) {
4902 truncate_pagecache(inode
, newsize
);
4904 * Don't do an expanding truncate while snapshoting is ongoing.
4905 * This is to ensure the snapshot captures a fully consistent
4906 * state of this file - if the snapshot captures this expanding
4907 * truncation, it must capture all writes that happened before
4910 btrfs_wait_for_snapshot_creation(root
);
4911 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
4913 btrfs_end_write_no_snapshoting(root
);
4917 trans
= btrfs_start_transaction(root
, 1);
4918 if (IS_ERR(trans
)) {
4919 btrfs_end_write_no_snapshoting(root
);
4920 return PTR_ERR(trans
);
4923 i_size_write(inode
, newsize
);
4924 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
4925 ret
= btrfs_update_inode(trans
, root
, inode
);
4926 btrfs_end_write_no_snapshoting(root
);
4927 btrfs_end_transaction(trans
, root
);
4931 * We're truncating a file that used to have good data down to
4932 * zero. Make sure it gets into the ordered flush list so that
4933 * any new writes get down to disk quickly.
4936 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
4937 &BTRFS_I(inode
)->runtime_flags
);
4940 * 1 for the orphan item we're going to add
4941 * 1 for the orphan item deletion.
4943 trans
= btrfs_start_transaction(root
, 2);
4945 return PTR_ERR(trans
);
4948 * We need to do this in case we fail at _any_ point during the
4949 * actual truncate. Once we do the truncate_setsize we could
4950 * invalidate pages which forces any outstanding ordered io to
4951 * be instantly completed which will give us extents that need
4952 * to be truncated. If we fail to get an orphan inode down we
4953 * could have left over extents that were never meant to live,
4954 * so we need to garuntee from this point on that everything
4955 * will be consistent.
4957 ret
= btrfs_orphan_add(trans
, inode
);
4958 btrfs_end_transaction(trans
, root
);
4962 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4963 truncate_setsize(inode
, newsize
);
4965 /* Disable nonlocked read DIO to avoid the end less truncate */
4966 btrfs_inode_block_unlocked_dio(inode
);
4967 inode_dio_wait(inode
);
4968 btrfs_inode_resume_unlocked_dio(inode
);
4970 ret
= btrfs_truncate(inode
);
4971 if (ret
&& inode
->i_nlink
) {
4975 * failed to truncate, disk_i_size is only adjusted down
4976 * as we remove extents, so it should represent the true
4977 * size of the inode, so reset the in memory size and
4978 * delete our orphan entry.
4980 trans
= btrfs_join_transaction(root
);
4981 if (IS_ERR(trans
)) {
4982 btrfs_orphan_del(NULL
, inode
);
4985 i_size_write(inode
, BTRFS_I(inode
)->disk_i_size
);
4986 err
= btrfs_orphan_del(trans
, inode
);
4988 btrfs_abort_transaction(trans
, root
, err
);
4989 btrfs_end_transaction(trans
, root
);
4996 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4998 struct inode
*inode
= d_inode(dentry
);
4999 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5002 if (btrfs_root_readonly(root
))
5005 err
= inode_change_ok(inode
, attr
);
5009 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
5010 err
= btrfs_setsize(inode
, attr
);
5015 if (attr
->ia_valid
) {
5016 setattr_copy(inode
, attr
);
5017 inode_inc_iversion(inode
);
5018 err
= btrfs_dirty_inode(inode
);
5020 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
5021 err
= posix_acl_chmod(inode
, inode
->i_mode
);
5028 * While truncating the inode pages during eviction, we get the VFS calling
5029 * btrfs_invalidatepage() against each page of the inode. This is slow because
5030 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5031 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5032 * extent_state structures over and over, wasting lots of time.
5034 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5035 * those expensive operations on a per page basis and do only the ordered io
5036 * finishing, while we release here the extent_map and extent_state structures,
5037 * without the excessive merging and splitting.
5039 static void evict_inode_truncate_pages(struct inode
*inode
)
5041 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5042 struct extent_map_tree
*map_tree
= &BTRFS_I(inode
)->extent_tree
;
5043 struct rb_node
*node
;
5045 ASSERT(inode
->i_state
& I_FREEING
);
5046 truncate_inode_pages_final(&inode
->i_data
);
5048 write_lock(&map_tree
->lock
);
5049 while (!RB_EMPTY_ROOT(&map_tree
->map
)) {
5050 struct extent_map
*em
;
5052 node
= rb_first(&map_tree
->map
);
5053 em
= rb_entry(node
, struct extent_map
, rb_node
);
5054 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5055 clear_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
5056 remove_extent_mapping(map_tree
, em
);
5057 free_extent_map(em
);
5058 if (need_resched()) {
5059 write_unlock(&map_tree
->lock
);
5061 write_lock(&map_tree
->lock
);
5064 write_unlock(&map_tree
->lock
);
5067 * Keep looping until we have no more ranges in the io tree.
5068 * We can have ongoing bios started by readpages (called from readahead)
5069 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5070 * still in progress (unlocked the pages in the bio but did not yet
5071 * unlocked the ranges in the io tree). Therefore this means some
5072 * ranges can still be locked and eviction started because before
5073 * submitting those bios, which are executed by a separate task (work
5074 * queue kthread), inode references (inode->i_count) were not taken
5075 * (which would be dropped in the end io callback of each bio).
5076 * Therefore here we effectively end up waiting for those bios and
5077 * anyone else holding locked ranges without having bumped the inode's
5078 * reference count - if we don't do it, when they access the inode's
5079 * io_tree to unlock a range it may be too late, leading to an
5080 * use-after-free issue.
5082 spin_lock(&io_tree
->lock
);
5083 while (!RB_EMPTY_ROOT(&io_tree
->state
)) {
5084 struct extent_state
*state
;
5085 struct extent_state
*cached_state
= NULL
;
5089 node
= rb_first(&io_tree
->state
);
5090 state
= rb_entry(node
, struct extent_state
, rb_node
);
5091 start
= state
->start
;
5093 spin_unlock(&io_tree
->lock
);
5095 lock_extent_bits(io_tree
, start
, end
, &cached_state
);
5098 * If still has DELALLOC flag, the extent didn't reach disk,
5099 * and its reserved space won't be freed by delayed_ref.
5100 * So we need to free its reserved space here.
5101 * (Refer to comment in btrfs_invalidatepage, case 2)
5103 * Note, end is the bytenr of last byte, so we need + 1 here.
5105 if (state
->state
& EXTENT_DELALLOC
)
5106 btrfs_qgroup_free_data(inode
, start
, end
- start
+ 1);
5108 clear_extent_bit(io_tree
, start
, end
,
5109 EXTENT_LOCKED
| EXTENT_DIRTY
|
5110 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
5111 EXTENT_DEFRAG
, 1, 1,
5112 &cached_state
, GFP_NOFS
);
5115 spin_lock(&io_tree
->lock
);
5117 spin_unlock(&io_tree
->lock
);
5120 void btrfs_evict_inode(struct inode
*inode
)
5122 struct btrfs_trans_handle
*trans
;
5123 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5124 struct btrfs_block_rsv
*rsv
, *global_rsv
;
5125 int steal_from_global
= 0;
5126 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
5129 trace_btrfs_inode_evict(inode
);
5131 evict_inode_truncate_pages(inode
);
5133 if (inode
->i_nlink
&&
5134 ((btrfs_root_refs(&root
->root_item
) != 0 &&
5135 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
) ||
5136 btrfs_is_free_space_inode(inode
)))
5139 if (is_bad_inode(inode
)) {
5140 btrfs_orphan_del(NULL
, inode
);
5143 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5144 if (!special_file(inode
->i_mode
))
5145 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5147 btrfs_free_io_failure_record(inode
, 0, (u64
)-1);
5149 if (root
->fs_info
->log_root_recovering
) {
5150 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
5151 &BTRFS_I(inode
)->runtime_flags
));
5155 if (inode
->i_nlink
> 0) {
5156 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0 &&
5157 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
);
5161 ret
= btrfs_commit_inode_delayed_inode(inode
);
5163 btrfs_orphan_del(NULL
, inode
);
5167 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
5169 btrfs_orphan_del(NULL
, inode
);
5172 rsv
->size
= min_size
;
5174 global_rsv
= &root
->fs_info
->global_block_rsv
;
5176 btrfs_i_size_write(inode
, 0);
5179 * This is a bit simpler than btrfs_truncate since we've already
5180 * reserved our space for our orphan item in the unlink, so we just
5181 * need to reserve some slack space in case we add bytes and update
5182 * inode item when doing the truncate.
5185 ret
= btrfs_block_rsv_refill(root
, rsv
, min_size
,
5186 BTRFS_RESERVE_FLUSH_LIMIT
);
5189 * Try and steal from the global reserve since we will
5190 * likely not use this space anyway, we want to try as
5191 * hard as possible to get this to work.
5194 steal_from_global
++;
5196 steal_from_global
= 0;
5200 * steal_from_global == 0: we reserved stuff, hooray!
5201 * steal_from_global == 1: we didn't reserve stuff, boo!
5202 * steal_from_global == 2: we've committed, still not a lot of
5203 * room but maybe we'll have room in the global reserve this
5205 * steal_from_global == 3: abandon all hope!
5207 if (steal_from_global
> 2) {
5208 btrfs_warn(root
->fs_info
,
5209 "Could not get space for a delete, will truncate on mount %d",
5211 btrfs_orphan_del(NULL
, inode
);
5212 btrfs_free_block_rsv(root
, rsv
);
5216 trans
= btrfs_join_transaction(root
);
5217 if (IS_ERR(trans
)) {
5218 btrfs_orphan_del(NULL
, inode
);
5219 btrfs_free_block_rsv(root
, rsv
);
5224 * We can't just steal from the global reserve, we need tomake
5225 * sure there is room to do it, if not we need to commit and try
5228 if (steal_from_global
) {
5229 if (!btrfs_check_space_for_delayed_refs(trans
, root
))
5230 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
,
5237 * Couldn't steal from the global reserve, we have too much
5238 * pending stuff built up, commit the transaction and try it
5242 ret
= btrfs_commit_transaction(trans
, root
);
5244 btrfs_orphan_del(NULL
, inode
);
5245 btrfs_free_block_rsv(root
, rsv
);
5250 steal_from_global
= 0;
5253 trans
->block_rsv
= rsv
;
5255 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
5256 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
)
5259 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
5260 btrfs_end_transaction(trans
, root
);
5262 btrfs_btree_balance_dirty(root
);
5265 btrfs_free_block_rsv(root
, rsv
);
5268 * Errors here aren't a big deal, it just means we leave orphan items
5269 * in the tree. They will be cleaned up on the next mount.
5272 trans
->block_rsv
= root
->orphan_block_rsv
;
5273 btrfs_orphan_del(trans
, inode
);
5275 btrfs_orphan_del(NULL
, inode
);
5278 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
5279 if (!(root
== root
->fs_info
->tree_root
||
5280 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
5281 btrfs_return_ino(root
, btrfs_ino(inode
));
5283 btrfs_end_transaction(trans
, root
);
5284 btrfs_btree_balance_dirty(root
);
5286 btrfs_remove_delayed_node(inode
);
5291 * this returns the key found in the dir entry in the location pointer.
5292 * If no dir entries were found, location->objectid is 0.
5294 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
5295 struct btrfs_key
*location
)
5297 const char *name
= dentry
->d_name
.name
;
5298 int namelen
= dentry
->d_name
.len
;
5299 struct btrfs_dir_item
*di
;
5300 struct btrfs_path
*path
;
5301 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5304 path
= btrfs_alloc_path();
5308 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
5313 if (IS_ERR_OR_NULL(di
))
5316 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
5318 btrfs_free_path(path
);
5321 location
->objectid
= 0;
5326 * when we hit a tree root in a directory, the btrfs part of the inode
5327 * needs to be changed to reflect the root directory of the tree root. This
5328 * is kind of like crossing a mount point.
5330 static int fixup_tree_root_location(struct btrfs_root
*root
,
5332 struct dentry
*dentry
,
5333 struct btrfs_key
*location
,
5334 struct btrfs_root
**sub_root
)
5336 struct btrfs_path
*path
;
5337 struct btrfs_root
*new_root
;
5338 struct btrfs_root_ref
*ref
;
5339 struct extent_buffer
*leaf
;
5340 struct btrfs_key key
;
5344 path
= btrfs_alloc_path();
5351 key
.objectid
= BTRFS_I(dir
)->root
->root_key
.objectid
;
5352 key
.type
= BTRFS_ROOT_REF_KEY
;
5353 key
.offset
= location
->objectid
;
5355 ret
= btrfs_search_slot(NULL
, root
->fs_info
->tree_root
, &key
, path
,
5363 leaf
= path
->nodes
[0];
5364 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
5365 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
5366 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
5369 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
5370 (unsigned long)(ref
+ 1),
5371 dentry
->d_name
.len
);
5375 btrfs_release_path(path
);
5377 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
5378 if (IS_ERR(new_root
)) {
5379 err
= PTR_ERR(new_root
);
5383 *sub_root
= new_root
;
5384 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
5385 location
->type
= BTRFS_INODE_ITEM_KEY
;
5386 location
->offset
= 0;
5389 btrfs_free_path(path
);
5393 static void inode_tree_add(struct inode
*inode
)
5395 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5396 struct btrfs_inode
*entry
;
5398 struct rb_node
*parent
;
5399 struct rb_node
*new = &BTRFS_I(inode
)->rb_node
;
5400 u64 ino
= btrfs_ino(inode
);
5402 if (inode_unhashed(inode
))
5405 spin_lock(&root
->inode_lock
);
5406 p
= &root
->inode_tree
.rb_node
;
5409 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
5411 if (ino
< btrfs_ino(&entry
->vfs_inode
))
5412 p
= &parent
->rb_left
;
5413 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
5414 p
= &parent
->rb_right
;
5416 WARN_ON(!(entry
->vfs_inode
.i_state
&
5417 (I_WILL_FREE
| I_FREEING
)));
5418 rb_replace_node(parent
, new, &root
->inode_tree
);
5419 RB_CLEAR_NODE(parent
);
5420 spin_unlock(&root
->inode_lock
);
5424 rb_link_node(new, parent
, p
);
5425 rb_insert_color(new, &root
->inode_tree
);
5426 spin_unlock(&root
->inode_lock
);
5429 static void inode_tree_del(struct inode
*inode
)
5431 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5434 spin_lock(&root
->inode_lock
);
5435 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
5436 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
5437 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
5438 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5440 spin_unlock(&root
->inode_lock
);
5442 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0) {
5443 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
5444 spin_lock(&root
->inode_lock
);
5445 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5446 spin_unlock(&root
->inode_lock
);
5448 btrfs_add_dead_root(root
);
5452 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
5454 struct rb_node
*node
;
5455 struct rb_node
*prev
;
5456 struct btrfs_inode
*entry
;
5457 struct inode
*inode
;
5460 if (!test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
5461 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
5463 spin_lock(&root
->inode_lock
);
5465 node
= root
->inode_tree
.rb_node
;
5469 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5471 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
5472 node
= node
->rb_left
;
5473 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
5474 node
= node
->rb_right
;
5480 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
5481 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
5485 prev
= rb_next(prev
);
5489 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5490 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
5491 inode
= igrab(&entry
->vfs_inode
);
5493 spin_unlock(&root
->inode_lock
);
5494 if (atomic_read(&inode
->i_count
) > 1)
5495 d_prune_aliases(inode
);
5497 * btrfs_drop_inode will have it removed from
5498 * the inode cache when its usage count
5503 spin_lock(&root
->inode_lock
);
5507 if (cond_resched_lock(&root
->inode_lock
))
5510 node
= rb_next(node
);
5512 spin_unlock(&root
->inode_lock
);
5515 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
5517 struct btrfs_iget_args
*args
= p
;
5518 inode
->i_ino
= args
->location
->objectid
;
5519 memcpy(&BTRFS_I(inode
)->location
, args
->location
,
5520 sizeof(*args
->location
));
5521 BTRFS_I(inode
)->root
= args
->root
;
5525 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
5527 struct btrfs_iget_args
*args
= opaque
;
5528 return args
->location
->objectid
== BTRFS_I(inode
)->location
.objectid
&&
5529 args
->root
== BTRFS_I(inode
)->root
;
5532 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
5533 struct btrfs_key
*location
,
5534 struct btrfs_root
*root
)
5536 struct inode
*inode
;
5537 struct btrfs_iget_args args
;
5538 unsigned long hashval
= btrfs_inode_hash(location
->objectid
, root
);
5540 args
.location
= location
;
5543 inode
= iget5_locked(s
, hashval
, btrfs_find_actor
,
5544 btrfs_init_locked_inode
,
5549 /* Get an inode object given its location and corresponding root.
5550 * Returns in *is_new if the inode was read from disk
5552 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
5553 struct btrfs_root
*root
, int *new)
5555 struct inode
*inode
;
5557 inode
= btrfs_iget_locked(s
, location
, root
);
5559 return ERR_PTR(-ENOMEM
);
5561 if (inode
->i_state
& I_NEW
) {
5562 btrfs_read_locked_inode(inode
);
5563 if (!is_bad_inode(inode
)) {
5564 inode_tree_add(inode
);
5565 unlock_new_inode(inode
);
5569 unlock_new_inode(inode
);
5571 inode
= ERR_PTR(-ESTALE
);
5578 static struct inode
*new_simple_dir(struct super_block
*s
,
5579 struct btrfs_key
*key
,
5580 struct btrfs_root
*root
)
5582 struct inode
*inode
= new_inode(s
);
5585 return ERR_PTR(-ENOMEM
);
5587 BTRFS_I(inode
)->root
= root
;
5588 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
5589 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
5591 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
5592 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
5593 inode
->i_fop
= &simple_dir_operations
;
5594 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
5595 inode
->i_mtime
= CURRENT_TIME
;
5596 inode
->i_atime
= inode
->i_mtime
;
5597 inode
->i_ctime
= inode
->i_mtime
;
5598 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
5603 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
5605 struct inode
*inode
;
5606 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5607 struct btrfs_root
*sub_root
= root
;
5608 struct btrfs_key location
;
5612 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
5613 return ERR_PTR(-ENAMETOOLONG
);
5615 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
5617 return ERR_PTR(ret
);
5619 if (location
.objectid
== 0)
5620 return ERR_PTR(-ENOENT
);
5622 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
5623 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
5627 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
5629 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
5630 ret
= fixup_tree_root_location(root
, dir
, dentry
,
5631 &location
, &sub_root
);
5634 inode
= ERR_PTR(ret
);
5636 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
5638 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
5640 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
5642 if (!IS_ERR(inode
) && root
!= sub_root
) {
5643 down_read(&root
->fs_info
->cleanup_work_sem
);
5644 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
5645 ret
= btrfs_orphan_cleanup(sub_root
);
5646 up_read(&root
->fs_info
->cleanup_work_sem
);
5649 inode
= ERR_PTR(ret
);
5656 static int btrfs_dentry_delete(const struct dentry
*dentry
)
5658 struct btrfs_root
*root
;
5659 struct inode
*inode
= d_inode(dentry
);
5661 if (!inode
&& !IS_ROOT(dentry
))
5662 inode
= d_inode(dentry
->d_parent
);
5665 root
= BTRFS_I(inode
)->root
;
5666 if (btrfs_root_refs(&root
->root_item
) == 0)
5669 if (btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5675 static void btrfs_dentry_release(struct dentry
*dentry
)
5677 kfree(dentry
->d_fsdata
);
5680 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5683 struct inode
*inode
;
5685 inode
= btrfs_lookup_dentry(dir
, dentry
);
5686 if (IS_ERR(inode
)) {
5687 if (PTR_ERR(inode
) == -ENOENT
)
5690 return ERR_CAST(inode
);
5693 return d_splice_alias(inode
, dentry
);
5696 unsigned char btrfs_filetype_table
[] = {
5697 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
5700 static int btrfs_real_readdir(struct file
*file
, struct dir_context
*ctx
)
5702 struct inode
*inode
= file_inode(file
);
5703 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5704 struct btrfs_item
*item
;
5705 struct btrfs_dir_item
*di
;
5706 struct btrfs_key key
;
5707 struct btrfs_key found_key
;
5708 struct btrfs_path
*path
;
5709 struct list_head ins_list
;
5710 struct list_head del_list
;
5712 struct extent_buffer
*leaf
;
5714 unsigned char d_type
;
5719 int key_type
= BTRFS_DIR_INDEX_KEY
;
5723 int is_curr
= 0; /* ctx->pos points to the current index? */
5725 /* FIXME, use a real flag for deciding about the key type */
5726 if (root
->fs_info
->tree_root
== root
)
5727 key_type
= BTRFS_DIR_ITEM_KEY
;
5729 if (!dir_emit_dots(file
, ctx
))
5732 path
= btrfs_alloc_path();
5736 path
->reada
= READA_FORWARD
;
5738 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5739 INIT_LIST_HEAD(&ins_list
);
5740 INIT_LIST_HEAD(&del_list
);
5741 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
5744 key
.type
= key_type
;
5745 key
.offset
= ctx
->pos
;
5746 key
.objectid
= btrfs_ino(inode
);
5748 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5753 leaf
= path
->nodes
[0];
5754 slot
= path
->slots
[0];
5755 if (slot
>= btrfs_header_nritems(leaf
)) {
5756 ret
= btrfs_next_leaf(root
, path
);
5764 item
= btrfs_item_nr(slot
);
5765 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5767 if (found_key
.objectid
!= key
.objectid
)
5769 if (found_key
.type
!= key_type
)
5771 if (found_key
.offset
< ctx
->pos
)
5773 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
5774 btrfs_should_delete_dir_index(&del_list
,
5778 ctx
->pos
= found_key
.offset
;
5781 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
5783 di_total
= btrfs_item_size(leaf
, item
);
5785 while (di_cur
< di_total
) {
5786 struct btrfs_key location
;
5788 if (verify_dir_item(root
, leaf
, di
))
5791 name_len
= btrfs_dir_name_len(leaf
, di
);
5792 if (name_len
<= sizeof(tmp_name
)) {
5793 name_ptr
= tmp_name
;
5795 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
5801 read_extent_buffer(leaf
, name_ptr
,
5802 (unsigned long)(di
+ 1), name_len
);
5804 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
5805 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
5808 /* is this a reference to our own snapshot? If so
5811 * In contrast to old kernels, we insert the snapshot's
5812 * dir item and dir index after it has been created, so
5813 * we won't find a reference to our own snapshot. We
5814 * still keep the following code for backward
5817 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
5818 location
.objectid
== root
->root_key
.objectid
) {
5822 over
= !dir_emit(ctx
, name_ptr
, name_len
,
5823 location
.objectid
, d_type
);
5826 if (name_ptr
!= tmp_name
)
5831 di_len
= btrfs_dir_name_len(leaf
, di
) +
5832 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
5834 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
5840 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5843 ret
= btrfs_readdir_delayed_dir_index(ctx
, &ins_list
);
5848 /* Reached end of directory/root. Bump pos past the last item. */
5852 * Stop new entries from being returned after we return the last
5855 * New directory entries are assigned a strictly increasing
5856 * offset. This means that new entries created during readdir
5857 * are *guaranteed* to be seen in the future by that readdir.
5858 * This has broken buggy programs which operate on names as
5859 * they're returned by readdir. Until we re-use freed offsets
5860 * we have this hack to stop new entries from being returned
5861 * under the assumption that they'll never reach this huge
5864 * This is being careful not to overflow 32bit loff_t unless the
5865 * last entry requires it because doing so has broken 32bit apps
5868 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5869 if (ctx
->pos
>= INT_MAX
)
5870 ctx
->pos
= LLONG_MAX
;
5877 if (key_type
== BTRFS_DIR_INDEX_KEY
)
5878 btrfs_put_delayed_items(&ins_list
, &del_list
);
5879 btrfs_free_path(path
);
5883 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5885 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5886 struct btrfs_trans_handle
*trans
;
5888 bool nolock
= false;
5890 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5893 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(inode
))
5896 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
5898 trans
= btrfs_join_transaction_nolock(root
);
5900 trans
= btrfs_join_transaction(root
);
5902 return PTR_ERR(trans
);
5903 ret
= btrfs_commit_transaction(trans
, root
);
5909 * This is somewhat expensive, updating the tree every time the
5910 * inode changes. But, it is most likely to find the inode in cache.
5911 * FIXME, needs more benchmarking...there are no reasons other than performance
5912 * to keep or drop this code.
5914 static int btrfs_dirty_inode(struct inode
*inode
)
5916 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5917 struct btrfs_trans_handle
*trans
;
5920 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5923 trans
= btrfs_join_transaction(root
);
5925 return PTR_ERR(trans
);
5927 ret
= btrfs_update_inode(trans
, root
, inode
);
5928 if (ret
&& ret
== -ENOSPC
) {
5929 /* whoops, lets try again with the full transaction */
5930 btrfs_end_transaction(trans
, root
);
5931 trans
= btrfs_start_transaction(root
, 1);
5933 return PTR_ERR(trans
);
5935 ret
= btrfs_update_inode(trans
, root
, inode
);
5937 btrfs_end_transaction(trans
, root
);
5938 if (BTRFS_I(inode
)->delayed_node
)
5939 btrfs_balance_delayed_items(root
);
5945 * This is a copy of file_update_time. We need this so we can return error on
5946 * ENOSPC for updating the inode in the case of file write and mmap writes.
5948 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
5951 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5953 if (btrfs_root_readonly(root
))
5956 if (flags
& S_VERSION
)
5957 inode_inc_iversion(inode
);
5958 if (flags
& S_CTIME
)
5959 inode
->i_ctime
= *now
;
5960 if (flags
& S_MTIME
)
5961 inode
->i_mtime
= *now
;
5962 if (flags
& S_ATIME
)
5963 inode
->i_atime
= *now
;
5964 return btrfs_dirty_inode(inode
);
5968 * find the highest existing sequence number in a directory
5969 * and then set the in-memory index_cnt variable to reflect
5970 * free sequence numbers
5972 static int btrfs_set_inode_index_count(struct inode
*inode
)
5974 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5975 struct btrfs_key key
, found_key
;
5976 struct btrfs_path
*path
;
5977 struct extent_buffer
*leaf
;
5980 key
.objectid
= btrfs_ino(inode
);
5981 key
.type
= BTRFS_DIR_INDEX_KEY
;
5982 key
.offset
= (u64
)-1;
5984 path
= btrfs_alloc_path();
5988 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5991 /* FIXME: we should be able to handle this */
5997 * MAGIC NUMBER EXPLANATION:
5998 * since we search a directory based on f_pos we have to start at 2
5999 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6000 * else has to start at 2
6002 if (path
->slots
[0] == 0) {
6003 BTRFS_I(inode
)->index_cnt
= 2;
6009 leaf
= path
->nodes
[0];
6010 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6012 if (found_key
.objectid
!= btrfs_ino(inode
) ||
6013 found_key
.type
!= BTRFS_DIR_INDEX_KEY
) {
6014 BTRFS_I(inode
)->index_cnt
= 2;
6018 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
6020 btrfs_free_path(path
);
6025 * helper to find a free sequence number in a given directory. This current
6026 * code is very simple, later versions will do smarter things in the btree
6028 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
6032 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
6033 ret
= btrfs_inode_delayed_dir_index_count(dir
);
6035 ret
= btrfs_set_inode_index_count(dir
);
6041 *index
= BTRFS_I(dir
)->index_cnt
;
6042 BTRFS_I(dir
)->index_cnt
++;
6047 static int btrfs_insert_inode_locked(struct inode
*inode
)
6049 struct btrfs_iget_args args
;
6050 args
.location
= &BTRFS_I(inode
)->location
;
6051 args
.root
= BTRFS_I(inode
)->root
;
6053 return insert_inode_locked4(inode
,
6054 btrfs_inode_hash(inode
->i_ino
, BTRFS_I(inode
)->root
),
6055 btrfs_find_actor
, &args
);
6058 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
6059 struct btrfs_root
*root
,
6061 const char *name
, int name_len
,
6062 u64 ref_objectid
, u64 objectid
,
6063 umode_t mode
, u64
*index
)
6065 struct inode
*inode
;
6066 struct btrfs_inode_item
*inode_item
;
6067 struct btrfs_key
*location
;
6068 struct btrfs_path
*path
;
6069 struct btrfs_inode_ref
*ref
;
6070 struct btrfs_key key
[2];
6072 int nitems
= name
? 2 : 1;
6076 path
= btrfs_alloc_path();
6078 return ERR_PTR(-ENOMEM
);
6080 inode
= new_inode(root
->fs_info
->sb
);
6082 btrfs_free_path(path
);
6083 return ERR_PTR(-ENOMEM
);
6087 * O_TMPFILE, set link count to 0, so that after this point,
6088 * we fill in an inode item with the correct link count.
6091 set_nlink(inode
, 0);
6094 * we have to initialize this early, so we can reclaim the inode
6095 * number if we fail afterwards in this function.
6097 inode
->i_ino
= objectid
;
6100 trace_btrfs_inode_request(dir
);
6102 ret
= btrfs_set_inode_index(dir
, index
);
6104 btrfs_free_path(path
);
6106 return ERR_PTR(ret
);
6112 * index_cnt is ignored for everything but a dir,
6113 * btrfs_get_inode_index_count has an explanation for the magic
6116 BTRFS_I(inode
)->index_cnt
= 2;
6117 BTRFS_I(inode
)->dir_index
= *index
;
6118 BTRFS_I(inode
)->root
= root
;
6119 BTRFS_I(inode
)->generation
= trans
->transid
;
6120 inode
->i_generation
= BTRFS_I(inode
)->generation
;
6123 * We could have gotten an inode number from somebody who was fsynced
6124 * and then removed in this same transaction, so let's just set full
6125 * sync since it will be a full sync anyway and this will blow away the
6126 * old info in the log.
6128 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
6130 key
[0].objectid
= objectid
;
6131 key
[0].type
= BTRFS_INODE_ITEM_KEY
;
6134 sizes
[0] = sizeof(struct btrfs_inode_item
);
6138 * Start new inodes with an inode_ref. This is slightly more
6139 * efficient for small numbers of hard links since they will
6140 * be packed into one item. Extended refs will kick in if we
6141 * add more hard links than can fit in the ref item.
6143 key
[1].objectid
= objectid
;
6144 key
[1].type
= BTRFS_INODE_REF_KEY
;
6145 key
[1].offset
= ref_objectid
;
6147 sizes
[1] = name_len
+ sizeof(*ref
);
6150 location
= &BTRFS_I(inode
)->location
;
6151 location
->objectid
= objectid
;
6152 location
->offset
= 0;
6153 location
->type
= BTRFS_INODE_ITEM_KEY
;
6155 ret
= btrfs_insert_inode_locked(inode
);
6159 path
->leave_spinning
= 1;
6160 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, nitems
);
6164 inode_init_owner(inode
, dir
, mode
);
6165 inode_set_bytes(inode
, 0);
6167 inode
->i_mtime
= CURRENT_TIME
;
6168 inode
->i_atime
= inode
->i_mtime
;
6169 inode
->i_ctime
= inode
->i_mtime
;
6170 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
6172 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
6173 struct btrfs_inode_item
);
6174 memset_extent_buffer(path
->nodes
[0], 0, (unsigned long)inode_item
,
6175 sizeof(*inode_item
));
6176 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
6179 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
6180 struct btrfs_inode_ref
);
6181 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
6182 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
6183 ptr
= (unsigned long)(ref
+ 1);
6184 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
6187 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6188 btrfs_free_path(path
);
6190 btrfs_inherit_iflags(inode
, dir
);
6192 if (S_ISREG(mode
)) {
6193 if (btrfs_test_opt(root
, NODATASUM
))
6194 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6195 if (btrfs_test_opt(root
, NODATACOW
))
6196 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
6197 BTRFS_INODE_NODATASUM
;
6200 inode_tree_add(inode
);
6202 trace_btrfs_inode_new(inode
);
6203 btrfs_set_inode_last_trans(trans
, inode
);
6205 btrfs_update_root_times(trans
, root
);
6207 ret
= btrfs_inode_inherit_props(trans
, inode
, dir
);
6209 btrfs_err(root
->fs_info
,
6210 "error inheriting props for ino %llu (root %llu): %d",
6211 btrfs_ino(inode
), root
->root_key
.objectid
, ret
);
6216 unlock_new_inode(inode
);
6219 BTRFS_I(dir
)->index_cnt
--;
6220 btrfs_free_path(path
);
6222 return ERR_PTR(ret
);
6225 static inline u8
btrfs_inode_type(struct inode
*inode
)
6227 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
6231 * utility function to add 'inode' into 'parent_inode' with
6232 * a give name and a given sequence number.
6233 * if 'add_backref' is true, also insert a backref from the
6234 * inode to the parent directory.
6236 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
6237 struct inode
*parent_inode
, struct inode
*inode
,
6238 const char *name
, int name_len
, int add_backref
, u64 index
)
6241 struct btrfs_key key
;
6242 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
6243 u64 ino
= btrfs_ino(inode
);
6244 u64 parent_ino
= btrfs_ino(parent_inode
);
6246 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6247 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
6250 key
.type
= BTRFS_INODE_ITEM_KEY
;
6254 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6255 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
6256 key
.objectid
, root
->root_key
.objectid
,
6257 parent_ino
, index
, name
, name_len
);
6258 } else if (add_backref
) {
6259 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
6263 /* Nothing to clean up yet */
6267 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
6269 btrfs_inode_type(inode
), index
);
6270 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
6273 btrfs_abort_transaction(trans
, root
, ret
);
6277 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
6279 inode_inc_iversion(parent_inode
);
6280 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
6281 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
6283 btrfs_abort_transaction(trans
, root
, ret
);
6287 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6290 err
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
6291 key
.objectid
, root
->root_key
.objectid
,
6292 parent_ino
, &local_index
, name
, name_len
);
6294 } else if (add_backref
) {
6298 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
6299 ino
, parent_ino
, &local_index
);
6304 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
6305 struct inode
*dir
, struct dentry
*dentry
,
6306 struct inode
*inode
, int backref
, u64 index
)
6308 int err
= btrfs_add_link(trans
, dir
, inode
,
6309 dentry
->d_name
.name
, dentry
->d_name
.len
,
6316 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
6317 umode_t mode
, dev_t rdev
)
6319 struct btrfs_trans_handle
*trans
;
6320 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6321 struct inode
*inode
= NULL
;
6328 * 2 for inode item and ref
6330 * 1 for xattr if selinux is on
6332 trans
= btrfs_start_transaction(root
, 5);
6334 return PTR_ERR(trans
);
6336 err
= btrfs_find_free_ino(root
, &objectid
);
6340 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6341 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6343 if (IS_ERR(inode
)) {
6344 err
= PTR_ERR(inode
);
6349 * If the active LSM wants to access the inode during
6350 * d_instantiate it needs these. Smack checks to see
6351 * if the filesystem supports xattrs by looking at the
6354 inode
->i_op
= &btrfs_special_inode_operations
;
6355 init_special_inode(inode
, inode
->i_mode
, rdev
);
6357 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6359 goto out_unlock_inode
;
6361 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
6363 goto out_unlock_inode
;
6365 btrfs_update_inode(trans
, root
, inode
);
6366 unlock_new_inode(inode
);
6367 d_instantiate(dentry
, inode
);
6371 btrfs_end_transaction(trans
, root
);
6372 btrfs_balance_delayed_items(root
);
6373 btrfs_btree_balance_dirty(root
);
6375 inode_dec_link_count(inode
);
6382 unlock_new_inode(inode
);
6387 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
6388 umode_t mode
, bool excl
)
6390 struct btrfs_trans_handle
*trans
;
6391 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6392 struct inode
*inode
= NULL
;
6393 int drop_inode_on_err
= 0;
6399 * 2 for inode item and ref
6401 * 1 for xattr if selinux is on
6403 trans
= btrfs_start_transaction(root
, 5);
6405 return PTR_ERR(trans
);
6407 err
= btrfs_find_free_ino(root
, &objectid
);
6411 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6412 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6414 if (IS_ERR(inode
)) {
6415 err
= PTR_ERR(inode
);
6418 drop_inode_on_err
= 1;
6420 * If the active LSM wants to access the inode during
6421 * d_instantiate it needs these. Smack checks to see
6422 * if the filesystem supports xattrs by looking at the
6425 inode
->i_fop
= &btrfs_file_operations
;
6426 inode
->i_op
= &btrfs_file_inode_operations
;
6427 inode
->i_mapping
->a_ops
= &btrfs_aops
;
6429 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6431 goto out_unlock_inode
;
6433 err
= btrfs_update_inode(trans
, root
, inode
);
6435 goto out_unlock_inode
;
6437 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
6439 goto out_unlock_inode
;
6441 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
6442 unlock_new_inode(inode
);
6443 d_instantiate(dentry
, inode
);
6446 btrfs_end_transaction(trans
, root
);
6447 if (err
&& drop_inode_on_err
) {
6448 inode_dec_link_count(inode
);
6451 btrfs_balance_delayed_items(root
);
6452 btrfs_btree_balance_dirty(root
);
6456 unlock_new_inode(inode
);
6461 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
6462 struct dentry
*dentry
)
6464 struct btrfs_trans_handle
*trans
= NULL
;
6465 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6466 struct inode
*inode
= d_inode(old_dentry
);
6471 /* do not allow sys_link's with other subvols of the same device */
6472 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
6475 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
6478 err
= btrfs_set_inode_index(dir
, &index
);
6483 * 2 items for inode and inode ref
6484 * 2 items for dir items
6485 * 1 item for parent inode
6487 trans
= btrfs_start_transaction(root
, 5);
6488 if (IS_ERR(trans
)) {
6489 err
= PTR_ERR(trans
);
6494 /* There are several dir indexes for this inode, clear the cache. */
6495 BTRFS_I(inode
)->dir_index
= 0ULL;
6497 inode_inc_iversion(inode
);
6498 inode
->i_ctime
= CURRENT_TIME
;
6500 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
6502 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
6507 struct dentry
*parent
= dentry
->d_parent
;
6508 err
= btrfs_update_inode(trans
, root
, inode
);
6511 if (inode
->i_nlink
== 1) {
6513 * If new hard link count is 1, it's a file created
6514 * with open(2) O_TMPFILE flag.
6516 err
= btrfs_orphan_del(trans
, inode
);
6520 d_instantiate(dentry
, inode
);
6521 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
6524 btrfs_balance_delayed_items(root
);
6527 btrfs_end_transaction(trans
, root
);
6529 inode_dec_link_count(inode
);
6532 btrfs_btree_balance_dirty(root
);
6536 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
6538 struct inode
*inode
= NULL
;
6539 struct btrfs_trans_handle
*trans
;
6540 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6542 int drop_on_err
= 0;
6547 * 2 items for inode and ref
6548 * 2 items for dir items
6549 * 1 for xattr if selinux is on
6551 trans
= btrfs_start_transaction(root
, 5);
6553 return PTR_ERR(trans
);
6555 err
= btrfs_find_free_ino(root
, &objectid
);
6559 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6560 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6561 S_IFDIR
| mode
, &index
);
6562 if (IS_ERR(inode
)) {
6563 err
= PTR_ERR(inode
);
6568 /* these must be set before we unlock the inode */
6569 inode
->i_op
= &btrfs_dir_inode_operations
;
6570 inode
->i_fop
= &btrfs_dir_file_operations
;
6572 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6574 goto out_fail_inode
;
6576 btrfs_i_size_write(inode
, 0);
6577 err
= btrfs_update_inode(trans
, root
, inode
);
6579 goto out_fail_inode
;
6581 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
6582 dentry
->d_name
.len
, 0, index
);
6584 goto out_fail_inode
;
6586 d_instantiate(dentry
, inode
);
6588 * mkdir is special. We're unlocking after we call d_instantiate
6589 * to avoid a race with nfsd calling d_instantiate.
6591 unlock_new_inode(inode
);
6595 btrfs_end_transaction(trans
, root
);
6597 inode_dec_link_count(inode
);
6600 btrfs_balance_delayed_items(root
);
6601 btrfs_btree_balance_dirty(root
);
6605 unlock_new_inode(inode
);
6609 /* Find next extent map of a given extent map, caller needs to ensure locks */
6610 static struct extent_map
*next_extent_map(struct extent_map
*em
)
6612 struct rb_node
*next
;
6614 next
= rb_next(&em
->rb_node
);
6617 return container_of(next
, struct extent_map
, rb_node
);
6620 static struct extent_map
*prev_extent_map(struct extent_map
*em
)
6622 struct rb_node
*prev
;
6624 prev
= rb_prev(&em
->rb_node
);
6627 return container_of(prev
, struct extent_map
, rb_node
);
6630 /* helper for btfs_get_extent. Given an existing extent in the tree,
6631 * the existing extent is the nearest extent to map_start,
6632 * and an extent that you want to insert, deal with overlap and insert
6633 * the best fitted new extent into the tree.
6635 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
6636 struct extent_map
*existing
,
6637 struct extent_map
*em
,
6640 struct extent_map
*prev
;
6641 struct extent_map
*next
;
6646 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
6648 if (existing
->start
> map_start
) {
6650 prev
= prev_extent_map(next
);
6653 next
= next_extent_map(prev
);
6656 start
= prev
? extent_map_end(prev
) : em
->start
;
6657 start
= max_t(u64
, start
, em
->start
);
6658 end
= next
? next
->start
: extent_map_end(em
);
6659 end
= min_t(u64
, end
, extent_map_end(em
));
6660 start_diff
= start
- em
->start
;
6662 em
->len
= end
- start
;
6663 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
6664 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
6665 em
->block_start
+= start_diff
;
6666 em
->block_len
-= start_diff
;
6668 return add_extent_mapping(em_tree
, em
, 0);
6671 static noinline
int uncompress_inline(struct btrfs_path
*path
,
6673 size_t pg_offset
, u64 extent_offset
,
6674 struct btrfs_file_extent_item
*item
)
6677 struct extent_buffer
*leaf
= path
->nodes
[0];
6680 unsigned long inline_size
;
6684 WARN_ON(pg_offset
!= 0);
6685 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6686 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6687 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
6688 btrfs_item_nr(path
->slots
[0]));
6689 tmp
= kmalloc(inline_size
, GFP_NOFS
);
6692 ptr
= btrfs_file_extent_inline_start(item
);
6694 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
6696 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
6697 ret
= btrfs_decompress(compress_type
, tmp
, page
,
6698 extent_offset
, inline_size
, max_size
);
6704 * a bit scary, this does extent mapping from logical file offset to the disk.
6705 * the ugly parts come from merging extents from the disk with the in-ram
6706 * representation. This gets more complex because of the data=ordered code,
6707 * where the in-ram extents might be locked pending data=ordered completion.
6709 * This also copies inline extents directly into the page.
6712 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
6713 size_t pg_offset
, u64 start
, u64 len
,
6718 u64 extent_start
= 0;
6720 u64 objectid
= btrfs_ino(inode
);
6722 struct btrfs_path
*path
= NULL
;
6723 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6724 struct btrfs_file_extent_item
*item
;
6725 struct extent_buffer
*leaf
;
6726 struct btrfs_key found_key
;
6727 struct extent_map
*em
= NULL
;
6728 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
6729 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6730 struct btrfs_trans_handle
*trans
= NULL
;
6731 const bool new_inline
= !page
|| create
;
6734 read_lock(&em_tree
->lock
);
6735 em
= lookup_extent_mapping(em_tree
, start
, len
);
6737 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6738 read_unlock(&em_tree
->lock
);
6741 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
6742 free_extent_map(em
);
6743 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
6744 free_extent_map(em
);
6748 em
= alloc_extent_map();
6753 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6754 em
->start
= EXTENT_MAP_HOLE
;
6755 em
->orig_start
= EXTENT_MAP_HOLE
;
6757 em
->block_len
= (u64
)-1;
6760 path
= btrfs_alloc_path();
6766 * Chances are we'll be called again, so go ahead and do
6769 path
->reada
= READA_FORWARD
;
6772 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
6773 objectid
, start
, trans
!= NULL
);
6780 if (path
->slots
[0] == 0)
6785 leaf
= path
->nodes
[0];
6786 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6787 struct btrfs_file_extent_item
);
6788 /* are we inside the extent that was found? */
6789 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6790 found_type
= found_key
.type
;
6791 if (found_key
.objectid
!= objectid
||
6792 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
6794 * If we backup past the first extent we want to move forward
6795 * and see if there is an extent in front of us, otherwise we'll
6796 * say there is a hole for our whole search range which can
6803 found_type
= btrfs_file_extent_type(leaf
, item
);
6804 extent_start
= found_key
.offset
;
6805 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6806 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6807 extent_end
= extent_start
+
6808 btrfs_file_extent_num_bytes(leaf
, item
);
6809 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6811 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6812 extent_end
= ALIGN(extent_start
+ size
, root
->sectorsize
);
6815 if (start
>= extent_end
) {
6817 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
6818 ret
= btrfs_next_leaf(root
, path
);
6825 leaf
= path
->nodes
[0];
6827 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6828 if (found_key
.objectid
!= objectid
||
6829 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6831 if (start
+ len
<= found_key
.offset
)
6833 if (start
> found_key
.offset
)
6836 em
->orig_start
= start
;
6837 em
->len
= found_key
.offset
- start
;
6841 btrfs_extent_item_to_extent_map(inode
, path
, item
, new_inline
, em
);
6843 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6844 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6846 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6850 size_t extent_offset
;
6856 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6857 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
6858 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
6859 size
- extent_offset
);
6860 em
->start
= extent_start
+ extent_offset
;
6861 em
->len
= ALIGN(copy_size
, root
->sectorsize
);
6862 em
->orig_block_len
= em
->len
;
6863 em
->orig_start
= em
->start
;
6864 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
6865 if (create
== 0 && !PageUptodate(page
)) {
6866 if (btrfs_file_extent_compression(leaf
, item
) !=
6867 BTRFS_COMPRESS_NONE
) {
6868 ret
= uncompress_inline(path
, page
, pg_offset
,
6869 extent_offset
, item
);
6876 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6878 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
6879 memset(map
+ pg_offset
+ copy_size
, 0,
6880 PAGE_CACHE_SIZE
- pg_offset
-
6885 flush_dcache_page(page
);
6886 } else if (create
&& PageUptodate(page
)) {
6890 free_extent_map(em
);
6893 btrfs_release_path(path
);
6894 trans
= btrfs_join_transaction(root
);
6897 return ERR_CAST(trans
);
6901 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6904 btrfs_mark_buffer_dirty(leaf
);
6906 set_extent_uptodate(io_tree
, em
->start
,
6907 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
6912 em
->orig_start
= start
;
6915 em
->block_start
= EXTENT_MAP_HOLE
;
6916 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
6918 btrfs_release_path(path
);
6919 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
6920 btrfs_err(root
->fs_info
, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6921 em
->start
, em
->len
, start
, len
);
6927 write_lock(&em_tree
->lock
);
6928 ret
= add_extent_mapping(em_tree
, em
, 0);
6929 /* it is possible that someone inserted the extent into the tree
6930 * while we had the lock dropped. It is also possible that
6931 * an overlapping map exists in the tree
6933 if (ret
== -EEXIST
) {
6934 struct extent_map
*existing
;
6938 existing
= search_extent_mapping(em_tree
, start
, len
);
6940 * existing will always be non-NULL, since there must be
6941 * extent causing the -EEXIST.
6943 if (start
>= extent_map_end(existing
) ||
6944 start
<= existing
->start
) {
6946 * The existing extent map is the one nearest to
6947 * the [start, start + len) range which overlaps
6949 err
= merge_extent_mapping(em_tree
, existing
,
6951 free_extent_map(existing
);
6953 free_extent_map(em
);
6957 free_extent_map(em
);
6962 write_unlock(&em_tree
->lock
);
6965 trace_btrfs_get_extent(root
, em
);
6967 btrfs_free_path(path
);
6969 ret
= btrfs_end_transaction(trans
, root
);
6974 free_extent_map(em
);
6975 return ERR_PTR(err
);
6977 BUG_ON(!em
); /* Error is always set */
6981 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
6982 size_t pg_offset
, u64 start
, u64 len
,
6985 struct extent_map
*em
;
6986 struct extent_map
*hole_em
= NULL
;
6987 u64 range_start
= start
;
6993 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
7000 * - a pre-alloc extent,
7001 * there might actually be delalloc bytes behind it.
7003 if (em
->block_start
!= EXTENT_MAP_HOLE
&&
7004 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7010 /* check to see if we've wrapped (len == -1 or similar) */
7019 /* ok, we didn't find anything, lets look for delalloc */
7020 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
7021 end
, len
, EXTENT_DELALLOC
, 1);
7022 found_end
= range_start
+ found
;
7023 if (found_end
< range_start
)
7024 found_end
= (u64
)-1;
7027 * we didn't find anything useful, return
7028 * the original results from get_extent()
7030 if (range_start
> end
|| found_end
<= start
) {
7036 /* adjust the range_start to make sure it doesn't
7037 * go backwards from the start they passed in
7039 range_start
= max(start
, range_start
);
7040 found
= found_end
- range_start
;
7043 u64 hole_start
= start
;
7046 em
= alloc_extent_map();
7052 * when btrfs_get_extent can't find anything it
7053 * returns one huge hole
7055 * make sure what it found really fits our range, and
7056 * adjust to make sure it is based on the start from
7060 u64 calc_end
= extent_map_end(hole_em
);
7062 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
7063 free_extent_map(hole_em
);
7066 hole_start
= max(hole_em
->start
, start
);
7067 hole_len
= calc_end
- hole_start
;
7071 if (hole_em
&& range_start
> hole_start
) {
7072 /* our hole starts before our delalloc, so we
7073 * have to return just the parts of the hole
7074 * that go until the delalloc starts
7076 em
->len
= min(hole_len
,
7077 range_start
- hole_start
);
7078 em
->start
= hole_start
;
7079 em
->orig_start
= hole_start
;
7081 * don't adjust block start at all,
7082 * it is fixed at EXTENT_MAP_HOLE
7084 em
->block_start
= hole_em
->block_start
;
7085 em
->block_len
= hole_len
;
7086 if (test_bit(EXTENT_FLAG_PREALLOC
, &hole_em
->flags
))
7087 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
7089 em
->start
= range_start
;
7091 em
->orig_start
= range_start
;
7092 em
->block_start
= EXTENT_MAP_DELALLOC
;
7093 em
->block_len
= found
;
7095 } else if (hole_em
) {
7100 free_extent_map(hole_em
);
7102 free_extent_map(em
);
7103 return ERR_PTR(err
);
7108 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
7111 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7112 struct extent_map
*em
;
7113 struct btrfs_key ins
;
7117 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
7118 ret
= btrfs_reserve_extent(root
, len
, root
->sectorsize
, 0,
7119 alloc_hint
, &ins
, 1, 1);
7121 return ERR_PTR(ret
);
7124 * Create the ordered extent before the extent map. This is to avoid
7125 * races with the fast fsync path that would lead to it logging file
7126 * extent items that point to disk extents that were not yet written to.
7127 * The fast fsync path collects ordered extents into a local list and
7128 * then collects all the new extent maps, so we must create the ordered
7129 * extent first and make sure the fast fsync path collects any new
7130 * ordered extents after collecting new extent maps as well.
7131 * The fsync path simply can not rely on inode_dio_wait() because it
7132 * causes deadlock with AIO.
7134 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
7135 ins
.offset
, ins
.offset
, 0);
7137 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
7138 return ERR_PTR(ret
);
7141 em
= create_pinned_em(inode
, start
, ins
.offset
, start
, ins
.objectid
,
7142 ins
.offset
, ins
.offset
, ins
.offset
, 0);
7144 struct btrfs_ordered_extent
*oe
;
7146 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
7147 oe
= btrfs_lookup_ordered_extent(inode
, start
);
7151 set_bit(BTRFS_ORDERED_IOERR
, &oe
->flags
);
7152 set_bit(BTRFS_ORDERED_IO_DONE
, &oe
->flags
);
7153 btrfs_remove_ordered_extent(inode
, oe
);
7154 /* Once for our lookup and once for the ordered extents tree. */
7155 btrfs_put_ordered_extent(oe
);
7156 btrfs_put_ordered_extent(oe
);
7162 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7163 * block must be cow'd
7165 noinline
int can_nocow_extent(struct inode
*inode
, u64 offset
, u64
*len
,
7166 u64
*orig_start
, u64
*orig_block_len
,
7169 struct btrfs_trans_handle
*trans
;
7170 struct btrfs_path
*path
;
7172 struct extent_buffer
*leaf
;
7173 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7174 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7175 struct btrfs_file_extent_item
*fi
;
7176 struct btrfs_key key
;
7183 bool nocow
= (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
);
7185 path
= btrfs_alloc_path();
7189 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, btrfs_ino(inode
),
7194 slot
= path
->slots
[0];
7197 /* can't find the item, must cow */
7204 leaf
= path
->nodes
[0];
7205 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
7206 if (key
.objectid
!= btrfs_ino(inode
) ||
7207 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7208 /* not our file or wrong item type, must cow */
7212 if (key
.offset
> offset
) {
7213 /* Wrong offset, must cow */
7217 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
7218 found_type
= btrfs_file_extent_type(leaf
, fi
);
7219 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
7220 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
7221 /* not a regular extent, must cow */
7225 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_REG
)
7228 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
7229 if (extent_end
<= offset
)
7232 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7233 if (disk_bytenr
== 0)
7236 if (btrfs_file_extent_compression(leaf
, fi
) ||
7237 btrfs_file_extent_encryption(leaf
, fi
) ||
7238 btrfs_file_extent_other_encoding(leaf
, fi
))
7241 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
7244 *orig_start
= key
.offset
- backref_offset
;
7245 *orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7246 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
7249 if (btrfs_extent_readonly(root
, disk_bytenr
))
7252 num_bytes
= min(offset
+ *len
, extent_end
) - offset
;
7253 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7256 range_end
= round_up(offset
+ num_bytes
, root
->sectorsize
) - 1;
7257 ret
= test_range_bit(io_tree
, offset
, range_end
,
7258 EXTENT_DELALLOC
, 0, NULL
);
7265 btrfs_release_path(path
);
7268 * look for other files referencing this extent, if we
7269 * find any we must cow
7271 trans
= btrfs_join_transaction(root
);
7272 if (IS_ERR(trans
)) {
7277 ret
= btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
7278 key
.offset
- backref_offset
, disk_bytenr
);
7279 btrfs_end_transaction(trans
, root
);
7286 * adjust disk_bytenr and num_bytes to cover just the bytes
7287 * in this extent we are about to write. If there
7288 * are any csums in that range we have to cow in order
7289 * to keep the csums correct
7291 disk_bytenr
+= backref_offset
;
7292 disk_bytenr
+= offset
- key
.offset
;
7293 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
7296 * all of the above have passed, it is safe to overwrite this extent
7302 btrfs_free_path(path
);
7306 bool btrfs_page_exists_in_range(struct inode
*inode
, loff_t start
, loff_t end
)
7308 struct radix_tree_root
*root
= &inode
->i_mapping
->page_tree
;
7310 void **pagep
= NULL
;
7311 struct page
*page
= NULL
;
7315 start_idx
= start
>> PAGE_CACHE_SHIFT
;
7318 * end is the last byte in the last page. end == start is legal
7320 end_idx
= end
>> PAGE_CACHE_SHIFT
;
7324 /* Most of the code in this while loop is lifted from
7325 * find_get_page. It's been modified to begin searching from a
7326 * page and return just the first page found in that range. If the
7327 * found idx is less than or equal to the end idx then we know that
7328 * a page exists. If no pages are found or if those pages are
7329 * outside of the range then we're fine (yay!) */
7330 while (page
== NULL
&&
7331 radix_tree_gang_lookup_slot(root
, &pagep
, NULL
, start_idx
, 1)) {
7332 page
= radix_tree_deref_slot(pagep
);
7333 if (unlikely(!page
))
7336 if (radix_tree_exception(page
)) {
7337 if (radix_tree_deref_retry(page
)) {
7342 * Otherwise, shmem/tmpfs must be storing a swap entry
7343 * here as an exceptional entry: so return it without
7344 * attempting to raise page count.
7347 break; /* TODO: Is this relevant for this use case? */
7350 if (!page_cache_get_speculative(page
)) {
7356 * Has the page moved?
7357 * This is part of the lockless pagecache protocol. See
7358 * include/linux/pagemap.h for details.
7360 if (unlikely(page
!= *pagep
)) {
7361 page_cache_release(page
);
7367 if (page
->index
<= end_idx
)
7369 page_cache_release(page
);
7376 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
7377 struct extent_state
**cached_state
, int writing
)
7379 struct btrfs_ordered_extent
*ordered
;
7383 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7386 * We're concerned with the entire range that we're going to be
7387 * doing DIO to, so we need to make sure theres no ordered
7388 * extents in this range.
7390 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
7391 lockend
- lockstart
+ 1);
7394 * We need to make sure there are no buffered pages in this
7395 * range either, we could have raced between the invalidate in
7396 * generic_file_direct_write and locking the extent. The
7397 * invalidate needs to happen so that reads after a write do not
7402 !btrfs_page_exists_in_range(inode
, lockstart
, lockend
)))
7405 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7406 cached_state
, GFP_NOFS
);
7409 btrfs_start_ordered_extent(inode
, ordered
, 1);
7410 btrfs_put_ordered_extent(ordered
);
7413 * We could trigger writeback for this range (and wait
7414 * for it to complete) and then invalidate the pages for
7415 * this range (through invalidate_inode_pages2_range()),
7416 * but that can lead us to a deadlock with a concurrent
7417 * call to readpages() (a buffered read or a defrag call
7418 * triggered a readahead) on a page lock due to an
7419 * ordered dio extent we created before but did not have
7420 * yet a corresponding bio submitted (whence it can not
7421 * complete), which makes readpages() wait for that
7422 * ordered extent to complete while holding a lock on
7435 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
7436 u64 len
, u64 orig_start
,
7437 u64 block_start
, u64 block_len
,
7438 u64 orig_block_len
, u64 ram_bytes
,
7441 struct extent_map_tree
*em_tree
;
7442 struct extent_map
*em
;
7443 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7446 em_tree
= &BTRFS_I(inode
)->extent_tree
;
7447 em
= alloc_extent_map();
7449 return ERR_PTR(-ENOMEM
);
7452 em
->orig_start
= orig_start
;
7453 em
->mod_start
= start
;
7456 em
->block_len
= block_len
;
7457 em
->block_start
= block_start
;
7458 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
7459 em
->orig_block_len
= orig_block_len
;
7460 em
->ram_bytes
= ram_bytes
;
7461 em
->generation
= -1;
7462 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
7463 if (type
== BTRFS_ORDERED_PREALLOC
)
7464 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
7467 btrfs_drop_extent_cache(inode
, em
->start
,
7468 em
->start
+ em
->len
- 1, 0);
7469 write_lock(&em_tree
->lock
);
7470 ret
= add_extent_mapping(em_tree
, em
, 1);
7471 write_unlock(&em_tree
->lock
);
7472 } while (ret
== -EEXIST
);
7475 free_extent_map(em
);
7476 return ERR_PTR(ret
);
7482 static void adjust_dio_outstanding_extents(struct inode
*inode
,
7483 struct btrfs_dio_data
*dio_data
,
7486 unsigned num_extents
;
7488 num_extents
= (unsigned) div64_u64(len
+ BTRFS_MAX_EXTENT_SIZE
- 1,
7489 BTRFS_MAX_EXTENT_SIZE
);
7491 * If we have an outstanding_extents count still set then we're
7492 * within our reservation, otherwise we need to adjust our inode
7493 * counter appropriately.
7495 if (dio_data
->outstanding_extents
) {
7496 dio_data
->outstanding_extents
-= num_extents
;
7498 spin_lock(&BTRFS_I(inode
)->lock
);
7499 BTRFS_I(inode
)->outstanding_extents
+= num_extents
;
7500 spin_unlock(&BTRFS_I(inode
)->lock
);
7504 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
7505 struct buffer_head
*bh_result
, int create
)
7507 struct extent_map
*em
;
7508 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7509 struct extent_state
*cached_state
= NULL
;
7510 struct btrfs_dio_data
*dio_data
= NULL
;
7511 u64 start
= iblock
<< inode
->i_blkbits
;
7512 u64 lockstart
, lockend
;
7513 u64 len
= bh_result
->b_size
;
7514 int unlock_bits
= EXTENT_LOCKED
;
7518 unlock_bits
|= EXTENT_DIRTY
;
7520 len
= min_t(u64
, len
, root
->sectorsize
);
7523 lockend
= start
+ len
- 1;
7525 if (current
->journal_info
) {
7527 * Need to pull our outstanding extents and set journal_info to NULL so
7528 * that anything that needs to check if there's a transction doesn't get
7531 dio_data
= current
->journal_info
;
7532 current
->journal_info
= NULL
;
7536 * If this errors out it's because we couldn't invalidate pagecache for
7537 * this range and we need to fallback to buffered.
7539 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
,
7545 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
7552 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7553 * io. INLINE is special, and we could probably kludge it in here, but
7554 * it's still buffered so for safety lets just fall back to the generic
7557 * For COMPRESSED we _have_ to read the entire extent in so we can
7558 * decompress it, so there will be buffering required no matter what we
7559 * do, so go ahead and fallback to buffered.
7561 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7562 * to buffered IO. Don't blame me, this is the price we pay for using
7565 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
7566 em
->block_start
== EXTENT_MAP_INLINE
) {
7567 free_extent_map(em
);
7572 /* Just a good old fashioned hole, return */
7573 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
7574 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
7575 free_extent_map(em
);
7580 * We don't allocate a new extent in the following cases
7582 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7584 * 2) The extent is marked as PREALLOC. We're good to go here and can
7585 * just use the extent.
7589 len
= min(len
, em
->len
- (start
- em
->start
));
7590 lockstart
= start
+ len
;
7594 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
7595 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
7596 em
->block_start
!= EXTENT_MAP_HOLE
)) {
7598 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
7600 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7601 type
= BTRFS_ORDERED_PREALLOC
;
7603 type
= BTRFS_ORDERED_NOCOW
;
7604 len
= min(len
, em
->len
- (start
- em
->start
));
7605 block_start
= em
->block_start
+ (start
- em
->start
);
7607 if (can_nocow_extent(inode
, start
, &len
, &orig_start
,
7608 &orig_block_len
, &ram_bytes
) == 1) {
7609 if (type
== BTRFS_ORDERED_PREALLOC
) {
7610 free_extent_map(em
);
7611 em
= create_pinned_em(inode
, start
, len
,
7622 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
7623 block_start
, len
, len
, type
);
7625 free_extent_map(em
);
7633 * this will cow the extent, reset the len in case we changed
7636 len
= bh_result
->b_size
;
7637 free_extent_map(em
);
7638 em
= btrfs_new_extent_direct(inode
, start
, len
);
7643 len
= min(len
, em
->len
- (start
- em
->start
));
7645 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
7647 bh_result
->b_size
= len
;
7648 bh_result
->b_bdev
= em
->bdev
;
7649 set_buffer_mapped(bh_result
);
7651 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7652 set_buffer_new(bh_result
);
7655 * Need to update the i_size under the extent lock so buffered
7656 * readers will get the updated i_size when we unlock.
7658 if (start
+ len
> i_size_read(inode
))
7659 i_size_write(inode
, start
+ len
);
7661 adjust_dio_outstanding_extents(inode
, dio_data
, len
);
7662 btrfs_free_reserved_data_space(inode
, start
, len
);
7663 WARN_ON(dio_data
->reserve
< len
);
7664 dio_data
->reserve
-= len
;
7665 dio_data
->unsubmitted_oe_range_end
= start
+ len
;
7666 current
->journal_info
= dio_data
;
7670 * In the case of write we need to clear and unlock the entire range,
7671 * in the case of read we need to unlock only the end area that we
7672 * aren't using if there is any left over space.
7674 if (lockstart
< lockend
) {
7675 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
7676 lockend
, unlock_bits
, 1, 0,
7677 &cached_state
, GFP_NOFS
);
7679 free_extent_state(cached_state
);
7682 free_extent_map(em
);
7687 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7688 unlock_bits
, 1, 0, &cached_state
, GFP_NOFS
);
7691 current
->journal_info
= dio_data
;
7693 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7694 * write less data then expected, so that we don't underflow our inode's
7695 * outstanding extents counter.
7697 if (create
&& dio_data
)
7698 adjust_dio_outstanding_extents(inode
, dio_data
, len
);
7703 static inline int submit_dio_repair_bio(struct inode
*inode
, struct bio
*bio
,
7704 int rw
, int mirror_num
)
7706 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7709 BUG_ON(rw
& REQ_WRITE
);
7713 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
,
7714 BTRFS_WQ_ENDIO_DIO_REPAIR
);
7718 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
7724 static int btrfs_check_dio_repairable(struct inode
*inode
,
7725 struct bio
*failed_bio
,
7726 struct io_failure_record
*failrec
,
7731 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
7732 failrec
->logical
, failrec
->len
);
7733 if (num_copies
== 1) {
7735 * we only have a single copy of the data, so don't bother with
7736 * all the retry and error correction code that follows. no
7737 * matter what the error is, it is very likely to persist.
7739 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7740 num_copies
, failrec
->this_mirror
, failed_mirror
);
7744 failrec
->failed_mirror
= failed_mirror
;
7745 failrec
->this_mirror
++;
7746 if (failrec
->this_mirror
== failed_mirror
)
7747 failrec
->this_mirror
++;
7749 if (failrec
->this_mirror
> num_copies
) {
7750 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7751 num_copies
, failrec
->this_mirror
, failed_mirror
);
7758 static int dio_read_error(struct inode
*inode
, struct bio
*failed_bio
,
7759 struct page
*page
, unsigned int pgoff
,
7760 u64 start
, u64 end
, int failed_mirror
,
7761 bio_end_io_t
*repair_endio
, void *repair_arg
)
7763 struct io_failure_record
*failrec
;
7769 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
7771 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
7775 ret
= btrfs_check_dio_repairable(inode
, failed_bio
, failrec
,
7778 free_io_failure(inode
, failrec
);
7782 if ((failed_bio
->bi_vcnt
> 1)
7783 || (failed_bio
->bi_io_vec
->bv_len
7784 > BTRFS_I(inode
)->root
->sectorsize
))
7785 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
7787 read_mode
= READ_SYNC
;
7789 isector
= start
- btrfs_io_bio(failed_bio
)->logical
;
7790 isector
>>= inode
->i_sb
->s_blocksize_bits
;
7791 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
7792 pgoff
, isector
, repair_endio
, repair_arg
);
7794 free_io_failure(inode
, failrec
);
7798 btrfs_debug(BTRFS_I(inode
)->root
->fs_info
,
7799 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7800 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
7802 ret
= submit_dio_repair_bio(inode
, bio
, read_mode
,
7803 failrec
->this_mirror
);
7805 free_io_failure(inode
, failrec
);
7812 struct btrfs_retry_complete
{
7813 struct completion done
;
7814 struct inode
*inode
;
7819 static void btrfs_retry_endio_nocsum(struct bio
*bio
)
7821 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7822 struct inode
*inode
;
7823 struct bio_vec
*bvec
;
7829 ASSERT(bio
->bi_vcnt
== 1);
7830 inode
= bio
->bi_io_vec
->bv_page
->mapping
->host
;
7831 ASSERT(bio
->bi_io_vec
->bv_len
== BTRFS_I(inode
)->root
->sectorsize
);
7834 bio_for_each_segment_all(bvec
, bio
, i
)
7835 clean_io_failure(done
->inode
, done
->start
, bvec
->bv_page
, 0);
7837 complete(&done
->done
);
7841 static int __btrfs_correct_data_nocsum(struct inode
*inode
,
7842 struct btrfs_io_bio
*io_bio
)
7844 struct btrfs_fs_info
*fs_info
;
7845 struct bio_vec
*bvec
;
7846 struct btrfs_retry_complete done
;
7854 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
7855 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
7857 start
= io_bio
->logical
;
7860 bio_for_each_segment_all(bvec
, &io_bio
->bio
, i
) {
7861 nr_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, bvec
->bv_len
);
7862 pgoff
= bvec
->bv_offset
;
7864 next_block_or_try_again
:
7867 init_completion(&done
.done
);
7869 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
->bv_page
,
7870 pgoff
, start
, start
+ sectorsize
- 1,
7872 btrfs_retry_endio_nocsum
, &done
);
7876 wait_for_completion(&done
.done
);
7878 if (!done
.uptodate
) {
7879 /* We might have another mirror, so try again */
7880 goto next_block_or_try_again
;
7883 start
+= sectorsize
;
7886 pgoff
+= sectorsize
;
7887 goto next_block_or_try_again
;
7894 static void btrfs_retry_endio(struct bio
*bio
)
7896 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7897 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
7898 struct inode
*inode
;
7899 struct bio_vec
*bvec
;
7910 start
= done
->start
;
7912 ASSERT(bio
->bi_vcnt
== 1);
7913 inode
= bio
->bi_io_vec
->bv_page
->mapping
->host
;
7914 ASSERT(bio
->bi_io_vec
->bv_len
== BTRFS_I(inode
)->root
->sectorsize
);
7916 bio_for_each_segment_all(bvec
, bio
, i
) {
7917 ret
= __readpage_endio_check(done
->inode
, io_bio
, i
,
7918 bvec
->bv_page
, bvec
->bv_offset
,
7919 done
->start
, bvec
->bv_len
);
7921 clean_io_failure(done
->inode
, done
->start
,
7922 bvec
->bv_page
, bvec
->bv_offset
);
7927 done
->uptodate
= uptodate
;
7929 complete(&done
->done
);
7933 static int __btrfs_subio_endio_read(struct inode
*inode
,
7934 struct btrfs_io_bio
*io_bio
, int err
)
7936 struct btrfs_fs_info
*fs_info
;
7937 struct bio_vec
*bvec
;
7938 struct btrfs_retry_complete done
;
7948 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
7949 sectorsize
= BTRFS_I(inode
)->root
->sectorsize
;
7952 start
= io_bio
->logical
;
7955 bio_for_each_segment_all(bvec
, &io_bio
->bio
, i
) {
7956 nr_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, bvec
->bv_len
);
7958 pgoff
= bvec
->bv_offset
;
7960 csum_pos
= BTRFS_BYTES_TO_BLKS(fs_info
, offset
);
7961 ret
= __readpage_endio_check(inode
, io_bio
, csum_pos
,
7962 bvec
->bv_page
, pgoff
, start
,
7969 init_completion(&done
.done
);
7971 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
->bv_page
,
7972 pgoff
, start
, start
+ sectorsize
- 1,
7974 btrfs_retry_endio
, &done
);
7980 wait_for_completion(&done
.done
);
7982 if (!done
.uptodate
) {
7983 /* We might have another mirror, so try again */
7987 offset
+= sectorsize
;
7988 start
+= sectorsize
;
7993 pgoff
+= sectorsize
;
8001 static int btrfs_subio_endio_read(struct inode
*inode
,
8002 struct btrfs_io_bio
*io_bio
, int err
)
8004 bool skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
8008 return __btrfs_correct_data_nocsum(inode
, io_bio
);
8012 return __btrfs_subio_endio_read(inode
, io_bio
, err
);
8016 static void btrfs_endio_direct_read(struct bio
*bio
)
8018 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8019 struct inode
*inode
= dip
->inode
;
8020 struct bio
*dio_bio
;
8021 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8022 int err
= bio
->bi_error
;
8024 if (dip
->flags
& BTRFS_DIO_ORIG_BIO_SUBMITTED
)
8025 err
= btrfs_subio_endio_read(inode
, io_bio
, err
);
8027 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
8028 dip
->logical_offset
+ dip
->bytes
- 1);
8029 dio_bio
= dip
->dio_bio
;
8033 dio_end_io(dio_bio
, bio
->bi_error
);
8036 io_bio
->end_io(io_bio
, err
);
8040 static void btrfs_endio_direct_write_update_ordered(struct inode
*inode
,
8045 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8046 struct btrfs_ordered_extent
*ordered
= NULL
;
8047 u64 ordered_offset
= offset
;
8048 u64 ordered_bytes
= bytes
;
8052 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
8059 btrfs_init_work(&ordered
->work
, btrfs_endio_write_helper
,
8060 finish_ordered_fn
, NULL
, NULL
);
8061 btrfs_queue_work(root
->fs_info
->endio_write_workers
,
8065 * our bio might span multiple ordered extents. If we haven't
8066 * completed the accounting for the whole dio, go back and try again
8068 if (ordered_offset
< offset
+ bytes
) {
8069 ordered_bytes
= offset
+ bytes
- ordered_offset
;
8075 static void btrfs_endio_direct_write(struct bio
*bio
)
8077 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8078 struct bio
*dio_bio
= dip
->dio_bio
;
8080 btrfs_endio_direct_write_update_ordered(dip
->inode
,
8081 dip
->logical_offset
,
8087 dio_end_io(dio_bio
, bio
->bi_error
);
8091 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
8092 struct bio
*bio
, int mirror_num
,
8093 unsigned long bio_flags
, u64 offset
)
8096 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8097 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
8098 BUG_ON(ret
); /* -ENOMEM */
8102 static void btrfs_end_dio_bio(struct bio
*bio
)
8104 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8105 int err
= bio
->bi_error
;
8108 btrfs_warn(BTRFS_I(dip
->inode
)->root
->fs_info
,
8109 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8110 btrfs_ino(dip
->inode
), bio
->bi_rw
,
8111 (unsigned long long)bio
->bi_iter
.bi_sector
,
8112 bio
->bi_iter
.bi_size
, err
);
8114 if (dip
->subio_endio
)
8115 err
= dip
->subio_endio(dip
->inode
, btrfs_io_bio(bio
), err
);
8121 * before atomic variable goto zero, we must make sure
8122 * dip->errors is perceived to be set.
8124 smp_mb__before_atomic();
8127 /* if there are more bios still pending for this dio, just exit */
8128 if (!atomic_dec_and_test(&dip
->pending_bios
))
8132 bio_io_error(dip
->orig_bio
);
8134 dip
->dio_bio
->bi_error
= 0;
8135 bio_endio(dip
->orig_bio
);
8141 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
8142 u64 first_sector
, gfp_t gfp_flags
)
8145 bio
= btrfs_bio_alloc(bdev
, first_sector
, BIO_MAX_PAGES
, gfp_flags
);
8147 bio_associate_current(bio
);
8151 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root
*root
,
8152 struct inode
*inode
,
8153 struct btrfs_dio_private
*dip
,
8157 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8158 struct btrfs_io_bio
*orig_io_bio
= btrfs_io_bio(dip
->orig_bio
);
8162 * We load all the csum data we need when we submit
8163 * the first bio to reduce the csum tree search and
8166 if (dip
->logical_offset
== file_offset
) {
8167 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, dip
->orig_bio
,
8173 if (bio
== dip
->orig_bio
)
8176 file_offset
-= dip
->logical_offset
;
8177 file_offset
>>= inode
->i_sb
->s_blocksize_bits
;
8178 io_bio
->csum
= (u8
*)(((u32
*)orig_io_bio
->csum
) + file_offset
);
8183 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
8184 int rw
, u64 file_offset
, int skip_sum
,
8187 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8188 int write
= rw
& REQ_WRITE
;
8189 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8193 async_submit
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
8198 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
,
8199 BTRFS_WQ_ENDIO_DATA
);
8207 if (write
&& async_submit
) {
8208 ret
= btrfs_wq_submit_bio(root
->fs_info
,
8209 inode
, rw
, bio
, 0, 0,
8211 __btrfs_submit_bio_start_direct_io
,
8212 __btrfs_submit_bio_done
);
8216 * If we aren't doing async submit, calculate the csum of the
8219 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
8223 ret
= btrfs_lookup_and_bind_dio_csum(root
, inode
, dip
, bio
,
8229 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
8235 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
8238 struct inode
*inode
= dip
->inode
;
8239 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8241 struct bio
*orig_bio
= dip
->orig_bio
;
8242 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
8243 u64 start_sector
= orig_bio
->bi_iter
.bi_sector
;
8244 u64 file_offset
= dip
->logical_offset
;
8249 int async_submit
= 0;
8251 map_length
= orig_bio
->bi_iter
.bi_size
;
8252 ret
= btrfs_map_block(root
->fs_info
, rw
, start_sector
<< 9,
8253 &map_length
, NULL
, 0);
8257 if (map_length
>= orig_bio
->bi_iter
.bi_size
) {
8259 dip
->flags
|= BTRFS_DIO_ORIG_BIO_SUBMITTED
;
8263 /* async crcs make it difficult to collect full stripe writes. */
8264 if (btrfs_get_alloc_profile(root
, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK
)
8269 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
8273 bio
->bi_private
= dip
;
8274 bio
->bi_end_io
= btrfs_end_dio_bio
;
8275 btrfs_io_bio(bio
)->logical
= file_offset
;
8276 atomic_inc(&dip
->pending_bios
);
8278 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
8279 if (map_length
< submit_len
+ bvec
->bv_len
||
8280 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
8281 bvec
->bv_offset
) < bvec
->bv_len
) {
8283 * inc the count before we submit the bio so
8284 * we know the end IO handler won't happen before
8285 * we inc the count. Otherwise, the dip might get freed
8286 * before we're done setting it up
8288 atomic_inc(&dip
->pending_bios
);
8289 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
8290 file_offset
, skip_sum
,
8294 atomic_dec(&dip
->pending_bios
);
8298 start_sector
+= submit_len
>> 9;
8299 file_offset
+= submit_len
;
8304 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
8305 start_sector
, GFP_NOFS
);
8308 bio
->bi_private
= dip
;
8309 bio
->bi_end_io
= btrfs_end_dio_bio
;
8310 btrfs_io_bio(bio
)->logical
= file_offset
;
8312 map_length
= orig_bio
->bi_iter
.bi_size
;
8313 ret
= btrfs_map_block(root
->fs_info
, rw
,
8315 &map_length
, NULL
, 0);
8321 submit_len
+= bvec
->bv_len
;
8328 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
8337 * before atomic variable goto zero, we must
8338 * make sure dip->errors is perceived to be set.
8340 smp_mb__before_atomic();
8341 if (atomic_dec_and_test(&dip
->pending_bios
))
8342 bio_io_error(dip
->orig_bio
);
8344 /* bio_end_io() will handle error, so we needn't return it */
8348 static void btrfs_submit_direct(int rw
, struct bio
*dio_bio
,
8349 struct inode
*inode
, loff_t file_offset
)
8351 struct btrfs_dio_private
*dip
= NULL
;
8352 struct bio
*io_bio
= NULL
;
8353 struct btrfs_io_bio
*btrfs_bio
;
8355 int write
= rw
& REQ_WRITE
;
8358 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
8360 io_bio
= btrfs_bio_clone(dio_bio
, GFP_NOFS
);
8366 dip
= kzalloc(sizeof(*dip
), GFP_NOFS
);
8372 dip
->private = dio_bio
->bi_private
;
8374 dip
->logical_offset
= file_offset
;
8375 dip
->bytes
= dio_bio
->bi_iter
.bi_size
;
8376 dip
->disk_bytenr
= (u64
)dio_bio
->bi_iter
.bi_sector
<< 9;
8377 io_bio
->bi_private
= dip
;
8378 dip
->orig_bio
= io_bio
;
8379 dip
->dio_bio
= dio_bio
;
8380 atomic_set(&dip
->pending_bios
, 0);
8381 btrfs_bio
= btrfs_io_bio(io_bio
);
8382 btrfs_bio
->logical
= file_offset
;
8385 io_bio
->bi_end_io
= btrfs_endio_direct_write
;
8387 io_bio
->bi_end_io
= btrfs_endio_direct_read
;
8388 dip
->subio_endio
= btrfs_subio_endio_read
;
8392 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8393 * even if we fail to submit a bio, because in such case we do the
8394 * corresponding error handling below and it must not be done a second
8395 * time by btrfs_direct_IO().
8398 struct btrfs_dio_data
*dio_data
= current
->journal_info
;
8400 dio_data
->unsubmitted_oe_range_end
= dip
->logical_offset
+
8402 dio_data
->unsubmitted_oe_range_start
=
8403 dio_data
->unsubmitted_oe_range_end
;
8406 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
8410 if (btrfs_bio
->end_io
)
8411 btrfs_bio
->end_io(btrfs_bio
, ret
);
8415 * If we arrived here it means either we failed to submit the dip
8416 * or we either failed to clone the dio_bio or failed to allocate the
8417 * dip. If we cloned the dio_bio and allocated the dip, we can just
8418 * call bio_endio against our io_bio so that we get proper resource
8419 * cleanup if we fail to submit the dip, otherwise, we must do the
8420 * same as btrfs_endio_direct_[write|read] because we can't call these
8421 * callbacks - they require an allocated dip and a clone of dio_bio.
8423 if (io_bio
&& dip
) {
8424 io_bio
->bi_error
= -EIO
;
8427 * The end io callbacks free our dip, do the final put on io_bio
8428 * and all the cleanup and final put for dio_bio (through
8435 btrfs_endio_direct_write_update_ordered(inode
,
8437 dio_bio
->bi_iter
.bi_size
,
8440 unlock_extent(&BTRFS_I(inode
)->io_tree
, file_offset
,
8441 file_offset
+ dio_bio
->bi_iter
.bi_size
- 1);
8443 dio_bio
->bi_error
= -EIO
;
8445 * Releases and cleans up our dio_bio, no need to bio_put()
8446 * nor bio_endio()/bio_io_error() against dio_bio.
8448 dio_end_io(dio_bio
, ret
);
8455 static ssize_t
check_direct_IO(struct btrfs_root
*root
, struct kiocb
*iocb
,
8456 const struct iov_iter
*iter
, loff_t offset
)
8460 unsigned blocksize_mask
= root
->sectorsize
- 1;
8461 ssize_t retval
= -EINVAL
;
8463 if (offset
& blocksize_mask
)
8466 if (iov_iter_alignment(iter
) & blocksize_mask
)
8469 /* If this is a write we don't need to check anymore */
8470 if (iov_iter_rw(iter
) == WRITE
)
8473 * Check to make sure we don't have duplicate iov_base's in this
8474 * iovec, if so return EINVAL, otherwise we'll get csum errors
8475 * when reading back.
8477 for (seg
= 0; seg
< iter
->nr_segs
; seg
++) {
8478 for (i
= seg
+ 1; i
< iter
->nr_segs
; i
++) {
8479 if (iter
->iov
[seg
].iov_base
== iter
->iov
[i
].iov_base
)
8488 static ssize_t
btrfs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
,
8491 struct file
*file
= iocb
->ki_filp
;
8492 struct inode
*inode
= file
->f_mapping
->host
;
8493 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8494 struct btrfs_dio_data dio_data
= { 0 };
8498 bool relock
= false;
8501 if (check_direct_IO(BTRFS_I(inode
)->root
, iocb
, iter
, offset
))
8504 inode_dio_begin(inode
);
8505 smp_mb__after_atomic();
8508 * The generic stuff only does filemap_write_and_wait_range, which
8509 * isn't enough if we've written compressed pages to this area, so
8510 * we need to flush the dirty pages again to make absolutely sure
8511 * that any outstanding dirty pages are on disk.
8513 count
= iov_iter_count(iter
);
8514 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
8515 &BTRFS_I(inode
)->runtime_flags
))
8516 filemap_fdatawrite_range(inode
->i_mapping
, offset
,
8517 offset
+ count
- 1);
8519 if (iov_iter_rw(iter
) == WRITE
) {
8521 * If the write DIO is beyond the EOF, we need update
8522 * the isize, but it is protected by i_mutex. So we can
8523 * not unlock the i_mutex at this case.
8525 if (offset
+ count
<= inode
->i_size
) {
8526 mutex_unlock(&inode
->i_mutex
);
8529 ret
= btrfs_delalloc_reserve_space(inode
, offset
, count
);
8532 dio_data
.outstanding_extents
= div64_u64(count
+
8533 BTRFS_MAX_EXTENT_SIZE
- 1,
8534 BTRFS_MAX_EXTENT_SIZE
);
8537 * We need to know how many extents we reserved so that we can
8538 * do the accounting properly if we go over the number we
8539 * originally calculated. Abuse current->journal_info for this.
8541 dio_data
.reserve
= round_up(count
, root
->sectorsize
);
8542 dio_data
.unsubmitted_oe_range_start
= (u64
)offset
;
8543 dio_data
.unsubmitted_oe_range_end
= (u64
)offset
;
8544 current
->journal_info
= &dio_data
;
8545 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK
,
8546 &BTRFS_I(inode
)->runtime_flags
)) {
8547 inode_dio_end(inode
);
8548 flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
8552 ret
= __blockdev_direct_IO(iocb
, inode
,
8553 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
8554 iter
, offset
, btrfs_get_blocks_direct
, NULL
,
8555 btrfs_submit_direct
, flags
);
8556 if (iov_iter_rw(iter
) == WRITE
) {
8557 current
->journal_info
= NULL
;
8558 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
8559 if (dio_data
.reserve
)
8560 btrfs_delalloc_release_space(inode
, offset
,
8563 * On error we might have left some ordered extents
8564 * without submitting corresponding bios for them, so
8565 * cleanup them up to avoid other tasks getting them
8566 * and waiting for them to complete forever.
8568 if (dio_data
.unsubmitted_oe_range_start
<
8569 dio_data
.unsubmitted_oe_range_end
)
8570 btrfs_endio_direct_write_update_ordered(inode
,
8571 dio_data
.unsubmitted_oe_range_start
,
8572 dio_data
.unsubmitted_oe_range_end
-
8573 dio_data
.unsubmitted_oe_range_start
,
8575 } else if (ret
>= 0 && (size_t)ret
< count
)
8576 btrfs_delalloc_release_space(inode
, offset
,
8577 count
- (size_t)ret
);
8581 inode_dio_end(inode
);
8583 mutex_lock(&inode
->i_mutex
);
8588 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8590 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
8591 __u64 start
, __u64 len
)
8595 ret
= fiemap_check_flags(fieinfo
, BTRFS_FIEMAP_FLAGS
);
8599 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
8602 int btrfs_readpage(struct file
*file
, struct page
*page
)
8604 struct extent_io_tree
*tree
;
8605 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8606 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
8609 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
8611 struct extent_io_tree
*tree
;
8612 struct inode
*inode
= page
->mapping
->host
;
8615 if (current
->flags
& PF_MEMALLOC
) {
8616 redirty_page_for_writepage(wbc
, page
);
8622 * If we are under memory pressure we will call this directly from the
8623 * VM, we need to make sure we have the inode referenced for the ordered
8624 * extent. If not just return like we didn't do anything.
8626 if (!igrab(inode
)) {
8627 redirty_page_for_writepage(wbc
, page
);
8628 return AOP_WRITEPAGE_ACTIVATE
;
8630 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8631 ret
= extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
8632 btrfs_add_delayed_iput(inode
);
8636 static int btrfs_writepages(struct address_space
*mapping
,
8637 struct writeback_control
*wbc
)
8639 struct extent_io_tree
*tree
;
8641 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8642 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
8646 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
8647 struct list_head
*pages
, unsigned nr_pages
)
8649 struct extent_io_tree
*tree
;
8650 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8651 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
8654 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8656 struct extent_io_tree
*tree
;
8657 struct extent_map_tree
*map
;
8660 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8661 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
8662 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
8664 ClearPagePrivate(page
);
8665 set_page_private(page
, 0);
8666 page_cache_release(page
);
8671 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8673 if (PageWriteback(page
) || PageDirty(page
))
8675 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
8678 static void btrfs_invalidatepage(struct page
*page
, unsigned int offset
,
8679 unsigned int length
)
8681 struct inode
*inode
= page
->mapping
->host
;
8682 struct extent_io_tree
*tree
;
8683 struct btrfs_ordered_extent
*ordered
;
8684 struct extent_state
*cached_state
= NULL
;
8685 u64 page_start
= page_offset(page
);
8686 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
8689 int inode_evicting
= inode
->i_state
& I_FREEING
;
8692 * we have the page locked, so new writeback can't start,
8693 * and the dirty bit won't be cleared while we are here.
8695 * Wait for IO on this page so that we can safely clear
8696 * the PagePrivate2 bit and do ordered accounting
8698 wait_on_page_writeback(page
);
8700 tree
= &BTRFS_I(inode
)->io_tree
;
8702 btrfs_releasepage(page
, GFP_NOFS
);
8706 if (!inode_evicting
)
8707 lock_extent_bits(tree
, page_start
, page_end
, &cached_state
);
8710 ordered
= btrfs_lookup_ordered_range(inode
, start
,
8711 page_end
- start
+ 1);
8713 end
= min(page_end
, ordered
->file_offset
+ ordered
->len
- 1);
8715 * IO on this page will never be started, so we need
8716 * to account for any ordered extents now
8718 if (!inode_evicting
)
8719 clear_extent_bit(tree
, start
, end
,
8720 EXTENT_DIRTY
| EXTENT_DELALLOC
|
8721 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
8722 EXTENT_DEFRAG
, 1, 0, &cached_state
,
8725 * whoever cleared the private bit is responsible
8726 * for the finish_ordered_io
8728 if (TestClearPagePrivate2(page
)) {
8729 struct btrfs_ordered_inode_tree
*tree
;
8732 tree
= &BTRFS_I(inode
)->ordered_tree
;
8734 spin_lock_irq(&tree
->lock
);
8735 set_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
);
8736 new_len
= start
- ordered
->file_offset
;
8737 if (new_len
< ordered
->truncated_len
)
8738 ordered
->truncated_len
= new_len
;
8739 spin_unlock_irq(&tree
->lock
);
8741 if (btrfs_dec_test_ordered_pending(inode
, &ordered
,
8743 end
- start
+ 1, 1))
8744 btrfs_finish_ordered_io(ordered
);
8746 btrfs_put_ordered_extent(ordered
);
8747 if (!inode_evicting
) {
8748 cached_state
= NULL
;
8749 lock_extent_bits(tree
, start
, end
,
8754 if (start
< page_end
)
8759 * Qgroup reserved space handler
8760 * Page here will be either
8761 * 1) Already written to disk
8762 * In this case, its reserved space is released from data rsv map
8763 * and will be freed by delayed_ref handler finally.
8764 * So even we call qgroup_free_data(), it won't decrease reserved
8766 * 2) Not written to disk
8767 * This means the reserved space should be freed here.
8769 btrfs_qgroup_free_data(inode
, page_start
, PAGE_CACHE_SIZE
);
8770 if (!inode_evicting
) {
8771 clear_extent_bit(tree
, page_start
, page_end
,
8772 EXTENT_LOCKED
| EXTENT_DIRTY
|
8773 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
8774 EXTENT_DEFRAG
, 1, 1,
8775 &cached_state
, GFP_NOFS
);
8777 __btrfs_releasepage(page
, GFP_NOFS
);
8780 ClearPageChecked(page
);
8781 if (PagePrivate(page
)) {
8782 ClearPagePrivate(page
);
8783 set_page_private(page
, 0);
8784 page_cache_release(page
);
8789 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8790 * called from a page fault handler when a page is first dirtied. Hence we must
8791 * be careful to check for EOF conditions here. We set the page up correctly
8792 * for a written page which means we get ENOSPC checking when writing into
8793 * holes and correct delalloc and unwritten extent mapping on filesystems that
8794 * support these features.
8796 * We are not allowed to take the i_mutex here so we have to play games to
8797 * protect against truncate races as the page could now be beyond EOF. Because
8798 * vmtruncate() writes the inode size before removing pages, once we have the
8799 * page lock we can determine safely if the page is beyond EOF. If it is not
8800 * beyond EOF, then the page is guaranteed safe against truncation until we
8803 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
8805 struct page
*page
= vmf
->page
;
8806 struct inode
*inode
= file_inode(vma
->vm_file
);
8807 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8808 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
8809 struct btrfs_ordered_extent
*ordered
;
8810 struct extent_state
*cached_state
= NULL
;
8812 unsigned long zero_start
;
8821 reserved_space
= PAGE_CACHE_SIZE
;
8823 sb_start_pagefault(inode
->i_sb
);
8824 page_start
= page_offset(page
);
8825 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
8829 * Reserving delalloc space after obtaining the page lock can lead to
8830 * deadlock. For example, if a dirty page is locked by this function
8831 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8832 * dirty page write out, then the btrfs_writepage() function could
8833 * end up waiting indefinitely to get a lock on the page currently
8834 * being processed by btrfs_page_mkwrite() function.
8836 ret
= btrfs_delalloc_reserve_space(inode
, page_start
,
8839 ret
= file_update_time(vma
->vm_file
);
8845 else /* -ENOSPC, -EIO, etc */
8846 ret
= VM_FAULT_SIGBUS
;
8852 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
8855 size
= i_size_read(inode
);
8857 if ((page
->mapping
!= inode
->i_mapping
) ||
8858 (page_start
>= size
)) {
8859 /* page got truncated out from underneath us */
8862 wait_on_page_writeback(page
);
8864 lock_extent_bits(io_tree
, page_start
, page_end
, &cached_state
);
8865 set_page_extent_mapped(page
);
8868 * we can't set the delalloc bits if there are pending ordered
8869 * extents. Drop our locks and wait for them to finish
8871 ordered
= btrfs_lookup_ordered_range(inode
, page_start
, page_end
);
8873 unlock_extent_cached(io_tree
, page_start
, page_end
,
8874 &cached_state
, GFP_NOFS
);
8876 btrfs_start_ordered_extent(inode
, ordered
, 1);
8877 btrfs_put_ordered_extent(ordered
);
8881 if (page
->index
== ((size
- 1) >> PAGE_CACHE_SHIFT
)) {
8882 reserved_space
= round_up(size
- page_start
, root
->sectorsize
);
8883 if (reserved_space
< PAGE_CACHE_SIZE
) {
8884 end
= page_start
+ reserved_space
- 1;
8885 spin_lock(&BTRFS_I(inode
)->lock
);
8886 BTRFS_I(inode
)->outstanding_extents
++;
8887 spin_unlock(&BTRFS_I(inode
)->lock
);
8888 btrfs_delalloc_release_space(inode
, page_start
,
8889 PAGE_CACHE_SIZE
- reserved_space
);
8894 * XXX - page_mkwrite gets called every time the page is dirtied, even
8895 * if it was already dirty, so for space accounting reasons we need to
8896 * clear any delalloc bits for the range we are fixing to save. There
8897 * is probably a better way to do this, but for now keep consistent with
8898 * prepare_pages in the normal write path.
8900 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, end
,
8901 EXTENT_DIRTY
| EXTENT_DELALLOC
|
8902 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
8903 0, 0, &cached_state
, GFP_NOFS
);
8905 ret
= btrfs_set_extent_delalloc(inode
, page_start
, end
,
8908 unlock_extent_cached(io_tree
, page_start
, page_end
,
8909 &cached_state
, GFP_NOFS
);
8910 ret
= VM_FAULT_SIGBUS
;
8915 /* page is wholly or partially inside EOF */
8916 if (page_start
+ PAGE_CACHE_SIZE
> size
)
8917 zero_start
= size
& ~PAGE_CACHE_MASK
;
8919 zero_start
= PAGE_CACHE_SIZE
;
8921 if (zero_start
!= PAGE_CACHE_SIZE
) {
8923 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
8924 flush_dcache_page(page
);
8927 ClearPageChecked(page
);
8928 set_page_dirty(page
);
8929 SetPageUptodate(page
);
8931 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
8932 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
8933 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
8935 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
8939 sb_end_pagefault(inode
->i_sb
);
8940 return VM_FAULT_LOCKED
;
8944 btrfs_delalloc_release_space(inode
, page_start
, reserved_space
);
8946 sb_end_pagefault(inode
->i_sb
);
8950 static int btrfs_truncate(struct inode
*inode
)
8952 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8953 struct btrfs_block_rsv
*rsv
;
8956 struct btrfs_trans_handle
*trans
;
8957 u64 mask
= root
->sectorsize
- 1;
8958 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
8960 ret
= btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
),
8966 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8967 * 3 things going on here
8969 * 1) We need to reserve space for our orphan item and the space to
8970 * delete our orphan item. Lord knows we don't want to have a dangling
8971 * orphan item because we didn't reserve space to remove it.
8973 * 2) We need to reserve space to update our inode.
8975 * 3) We need to have something to cache all the space that is going to
8976 * be free'd up by the truncate operation, but also have some slack
8977 * space reserved in case it uses space during the truncate (thank you
8978 * very much snapshotting).
8980 * And we need these to all be seperate. The fact is we can use alot of
8981 * space doing the truncate, and we have no earthly idea how much space
8982 * we will use, so we need the truncate reservation to be seperate so it
8983 * doesn't end up using space reserved for updating the inode or
8984 * removing the orphan item. We also need to be able to stop the
8985 * transaction and start a new one, which means we need to be able to
8986 * update the inode several times, and we have no idea of knowing how
8987 * many times that will be, so we can't just reserve 1 item for the
8988 * entirety of the opration, so that has to be done seperately as well.
8989 * Then there is the orphan item, which does indeed need to be held on
8990 * to for the whole operation, and we need nobody to touch this reserved
8991 * space except the orphan code.
8993 * So that leaves us with
8995 * 1) root->orphan_block_rsv - for the orphan deletion.
8996 * 2) rsv - for the truncate reservation, which we will steal from the
8997 * transaction reservation.
8998 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8999 * updating the inode.
9001 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
9004 rsv
->size
= min_size
;
9008 * 1 for the truncate slack space
9009 * 1 for updating the inode.
9011 trans
= btrfs_start_transaction(root
, 2);
9012 if (IS_ERR(trans
)) {
9013 err
= PTR_ERR(trans
);
9017 /* Migrate the slack space for the truncate to our reserve */
9018 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
9023 * So if we truncate and then write and fsync we normally would just
9024 * write the extents that changed, which is a problem if we need to
9025 * first truncate that entire inode. So set this flag so we write out
9026 * all of the extents in the inode to the sync log so we're completely
9029 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
9030 trans
->block_rsv
= rsv
;
9033 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
9035 BTRFS_EXTENT_DATA_KEY
);
9036 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
) {
9041 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
9042 ret
= btrfs_update_inode(trans
, root
, inode
);
9048 btrfs_end_transaction(trans
, root
);
9049 btrfs_btree_balance_dirty(root
);
9051 trans
= btrfs_start_transaction(root
, 2);
9052 if (IS_ERR(trans
)) {
9053 ret
= err
= PTR_ERR(trans
);
9058 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
,
9060 BUG_ON(ret
); /* shouldn't happen */
9061 trans
->block_rsv
= rsv
;
9064 if (ret
== 0 && inode
->i_nlink
> 0) {
9065 trans
->block_rsv
= root
->orphan_block_rsv
;
9066 ret
= btrfs_orphan_del(trans
, inode
);
9072 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
9073 ret
= btrfs_update_inode(trans
, root
, inode
);
9077 ret
= btrfs_end_transaction(trans
, root
);
9078 btrfs_btree_balance_dirty(root
);
9082 btrfs_free_block_rsv(root
, rsv
);
9091 * create a new subvolume directory/inode (helper for the ioctl).
9093 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
9094 struct btrfs_root
*new_root
,
9095 struct btrfs_root
*parent_root
,
9098 struct inode
*inode
;
9102 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
9103 new_dirid
, new_dirid
,
9104 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
9107 return PTR_ERR(inode
);
9108 inode
->i_op
= &btrfs_dir_inode_operations
;
9109 inode
->i_fop
= &btrfs_dir_file_operations
;
9111 set_nlink(inode
, 1);
9112 btrfs_i_size_write(inode
, 0);
9113 unlock_new_inode(inode
);
9115 err
= btrfs_subvol_inherit_props(trans
, new_root
, parent_root
);
9117 btrfs_err(new_root
->fs_info
,
9118 "error inheriting subvolume %llu properties: %d",
9119 new_root
->root_key
.objectid
, err
);
9121 err
= btrfs_update_inode(trans
, new_root
, inode
);
9127 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
9129 struct btrfs_inode
*ei
;
9130 struct inode
*inode
;
9132 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
9139 ei
->last_sub_trans
= 0;
9140 ei
->logged_trans
= 0;
9141 ei
->delalloc_bytes
= 0;
9142 ei
->defrag_bytes
= 0;
9143 ei
->disk_i_size
= 0;
9146 ei
->index_cnt
= (u64
)-1;
9148 ei
->last_unlink_trans
= 0;
9149 ei
->last_log_commit
= 0;
9150 ei
->delayed_iput_count
= 0;
9152 spin_lock_init(&ei
->lock
);
9153 ei
->outstanding_extents
= 0;
9154 ei
->reserved_extents
= 0;
9156 ei
->runtime_flags
= 0;
9157 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
9159 ei
->delayed_node
= NULL
;
9161 ei
->i_otime
.tv_sec
= 0;
9162 ei
->i_otime
.tv_nsec
= 0;
9164 inode
= &ei
->vfs_inode
;
9165 extent_map_tree_init(&ei
->extent_tree
);
9166 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
9167 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
9168 ei
->io_tree
.track_uptodate
= 1;
9169 ei
->io_failure_tree
.track_uptodate
= 1;
9170 atomic_set(&ei
->sync_writers
, 0);
9171 mutex_init(&ei
->log_mutex
);
9172 mutex_init(&ei
->delalloc_mutex
);
9173 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
9174 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
9175 INIT_LIST_HEAD(&ei
->delayed_iput
);
9176 RB_CLEAR_NODE(&ei
->rb_node
);
9181 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9182 void btrfs_test_destroy_inode(struct inode
*inode
)
9184 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
9185 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9189 static void btrfs_i_callback(struct rcu_head
*head
)
9191 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
9192 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9195 void btrfs_destroy_inode(struct inode
*inode
)
9197 struct btrfs_ordered_extent
*ordered
;
9198 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9200 WARN_ON(!hlist_empty(&inode
->i_dentry
));
9201 WARN_ON(inode
->i_data
.nrpages
);
9202 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
9203 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
9204 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
9205 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
9206 WARN_ON(BTRFS_I(inode
)->defrag_bytes
);
9209 * This can happen where we create an inode, but somebody else also
9210 * created the same inode and we need to destroy the one we already
9216 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
9217 &BTRFS_I(inode
)->runtime_flags
)) {
9218 btrfs_info(root
->fs_info
, "inode %llu still on the orphan list",
9220 atomic_dec(&root
->orphan_inodes
);
9224 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
9228 btrfs_err(root
->fs_info
, "found ordered extent %llu %llu on inode cleanup",
9229 ordered
->file_offset
, ordered
->len
);
9230 btrfs_remove_ordered_extent(inode
, ordered
);
9231 btrfs_put_ordered_extent(ordered
);
9232 btrfs_put_ordered_extent(ordered
);
9235 btrfs_qgroup_check_reserved_leak(inode
);
9236 inode_tree_del(inode
);
9237 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
9239 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
9242 int btrfs_drop_inode(struct inode
*inode
)
9244 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9249 /* the snap/subvol tree is on deleting */
9250 if (btrfs_root_refs(&root
->root_item
) == 0)
9253 return generic_drop_inode(inode
);
9256 static void init_once(void *foo
)
9258 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
9260 inode_init_once(&ei
->vfs_inode
);
9263 void btrfs_destroy_cachep(void)
9266 * Make sure all delayed rcu free inodes are flushed before we
9270 if (btrfs_inode_cachep
)
9271 kmem_cache_destroy(btrfs_inode_cachep
);
9272 if (btrfs_trans_handle_cachep
)
9273 kmem_cache_destroy(btrfs_trans_handle_cachep
);
9274 if (btrfs_transaction_cachep
)
9275 kmem_cache_destroy(btrfs_transaction_cachep
);
9276 if (btrfs_path_cachep
)
9277 kmem_cache_destroy(btrfs_path_cachep
);
9278 if (btrfs_free_space_cachep
)
9279 kmem_cache_destroy(btrfs_free_space_cachep
);
9282 int btrfs_init_cachep(void)
9284 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
9285 sizeof(struct btrfs_inode
), 0,
9286 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
9287 if (!btrfs_inode_cachep
)
9290 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
9291 sizeof(struct btrfs_trans_handle
), 0,
9292 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9293 if (!btrfs_trans_handle_cachep
)
9296 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction",
9297 sizeof(struct btrfs_transaction
), 0,
9298 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9299 if (!btrfs_transaction_cachep
)
9302 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
9303 sizeof(struct btrfs_path
), 0,
9304 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9305 if (!btrfs_path_cachep
)
9308 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
9309 sizeof(struct btrfs_free_space
), 0,
9310 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9311 if (!btrfs_free_space_cachep
)
9316 btrfs_destroy_cachep();
9320 static int btrfs_getattr(struct vfsmount
*mnt
,
9321 struct dentry
*dentry
, struct kstat
*stat
)
9324 struct inode
*inode
= d_inode(dentry
);
9325 u32 blocksize
= inode
->i_sb
->s_blocksize
;
9327 generic_fillattr(inode
, stat
);
9328 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
9329 stat
->blksize
= PAGE_CACHE_SIZE
;
9331 spin_lock(&BTRFS_I(inode
)->lock
);
9332 delalloc_bytes
= BTRFS_I(inode
)->delalloc_bytes
;
9333 spin_unlock(&BTRFS_I(inode
)->lock
);
9334 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
9335 ALIGN(delalloc_bytes
, blocksize
)) >> 9;
9339 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
9340 struct inode
*new_dir
, struct dentry
*new_dentry
)
9342 struct btrfs_trans_handle
*trans
;
9343 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9344 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9345 struct inode
*new_inode
= d_inode(new_dentry
);
9346 struct inode
*old_inode
= d_inode(old_dentry
);
9347 struct timespec ctime
= CURRENT_TIME
;
9351 u64 old_ino
= btrfs_ino(old_inode
);
9353 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
9356 /* we only allow rename subvolume link between subvolumes */
9357 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9360 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
9361 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
9364 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
9365 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
9369 /* check for collisions, even if the name isn't there */
9370 ret
= btrfs_check_dir_item_collision(dest
, new_dir
->i_ino
,
9371 new_dentry
->d_name
.name
,
9372 new_dentry
->d_name
.len
);
9375 if (ret
== -EEXIST
) {
9377 * eexist without a new_inode */
9378 if (WARN_ON(!new_inode
)) {
9382 /* maybe -EOVERFLOW */
9389 * we're using rename to replace one file with another. Start IO on it
9390 * now so we don't add too much work to the end of the transaction
9392 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
)
9393 filemap_flush(old_inode
->i_mapping
);
9395 /* close the racy window with snapshot create/destroy ioctl */
9396 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9397 down_read(&root
->fs_info
->subvol_sem
);
9399 * We want to reserve the absolute worst case amount of items. So if
9400 * both inodes are subvols and we need to unlink them then that would
9401 * require 4 item modifications, but if they are both normal inodes it
9402 * would require 5 item modifications, so we'll assume their normal
9403 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9404 * should cover the worst case number of items we'll modify.
9406 trans
= btrfs_start_transaction(root
, 11);
9407 if (IS_ERR(trans
)) {
9408 ret
= PTR_ERR(trans
);
9413 btrfs_record_root_in_trans(trans
, dest
);
9415 ret
= btrfs_set_inode_index(new_dir
, &index
);
9419 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9420 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9421 /* force full log commit if subvolume involved. */
9422 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9424 ret
= btrfs_insert_inode_ref(trans
, dest
,
9425 new_dentry
->d_name
.name
,
9426 new_dentry
->d_name
.len
,
9428 btrfs_ino(new_dir
), index
);
9432 * this is an ugly little race, but the rename is required
9433 * to make sure that if we crash, the inode is either at the
9434 * old name or the new one. pinning the log transaction lets
9435 * us make sure we don't allow a log commit to come in after
9436 * we unlink the name but before we add the new name back in.
9438 btrfs_pin_log_trans(root
);
9441 inode_inc_iversion(old_dir
);
9442 inode_inc_iversion(new_dir
);
9443 inode_inc_iversion(old_inode
);
9444 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
9445 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
9446 old_inode
->i_ctime
= ctime
;
9448 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
9449 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
9451 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9452 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
9453 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
9454 old_dentry
->d_name
.name
,
9455 old_dentry
->d_name
.len
);
9457 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
9458 d_inode(old_dentry
),
9459 old_dentry
->d_name
.name
,
9460 old_dentry
->d_name
.len
);
9462 ret
= btrfs_update_inode(trans
, root
, old_inode
);
9465 btrfs_abort_transaction(trans
, root
, ret
);
9470 inode_inc_iversion(new_inode
);
9471 new_inode
->i_ctime
= CURRENT_TIME
;
9472 if (unlikely(btrfs_ino(new_inode
) ==
9473 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
9474 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
9475 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
9477 new_dentry
->d_name
.name
,
9478 new_dentry
->d_name
.len
);
9479 BUG_ON(new_inode
->i_nlink
== 0);
9481 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
9482 d_inode(new_dentry
),
9483 new_dentry
->d_name
.name
,
9484 new_dentry
->d_name
.len
);
9486 if (!ret
&& new_inode
->i_nlink
== 0)
9487 ret
= btrfs_orphan_add(trans
, d_inode(new_dentry
));
9489 btrfs_abort_transaction(trans
, root
, ret
);
9494 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
9495 new_dentry
->d_name
.name
,
9496 new_dentry
->d_name
.len
, 0, index
);
9498 btrfs_abort_transaction(trans
, root
, ret
);
9502 if (old_inode
->i_nlink
== 1)
9503 BTRFS_I(old_inode
)->dir_index
= index
;
9505 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
9506 struct dentry
*parent
= new_dentry
->d_parent
;
9507 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
9508 btrfs_end_log_trans(root
);
9511 btrfs_end_transaction(trans
, root
);
9513 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9514 up_read(&root
->fs_info
->subvol_sem
);
9519 static int btrfs_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
9520 struct inode
*new_dir
, struct dentry
*new_dentry
,
9523 if (flags
& ~RENAME_NOREPLACE
)
9526 return btrfs_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
9529 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
9531 struct btrfs_delalloc_work
*delalloc_work
;
9532 struct inode
*inode
;
9534 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
9536 inode
= delalloc_work
->inode
;
9537 filemap_flush(inode
->i_mapping
);
9538 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
9539 &BTRFS_I(inode
)->runtime_flags
))
9540 filemap_flush(inode
->i_mapping
);
9542 if (delalloc_work
->delay_iput
)
9543 btrfs_add_delayed_iput(inode
);
9546 complete(&delalloc_work
->completion
);
9549 struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
,
9552 struct btrfs_delalloc_work
*work
;
9554 work
= kmalloc(sizeof(*work
), GFP_NOFS
);
9558 init_completion(&work
->completion
);
9559 INIT_LIST_HEAD(&work
->list
);
9560 work
->inode
= inode
;
9561 work
->delay_iput
= delay_iput
;
9562 WARN_ON_ONCE(!inode
);
9563 btrfs_init_work(&work
->work
, btrfs_flush_delalloc_helper
,
9564 btrfs_run_delalloc_work
, NULL
, NULL
);
9569 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work
*work
)
9571 wait_for_completion(&work
->completion
);
9576 * some fairly slow code that needs optimization. This walks the list
9577 * of all the inodes with pending delalloc and forces them to disk.
9579 static int __start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
,
9582 struct btrfs_inode
*binode
;
9583 struct inode
*inode
;
9584 struct btrfs_delalloc_work
*work
, *next
;
9585 struct list_head works
;
9586 struct list_head splice
;
9589 INIT_LIST_HEAD(&works
);
9590 INIT_LIST_HEAD(&splice
);
9592 mutex_lock(&root
->delalloc_mutex
);
9593 spin_lock(&root
->delalloc_lock
);
9594 list_splice_init(&root
->delalloc_inodes
, &splice
);
9595 while (!list_empty(&splice
)) {
9596 binode
= list_entry(splice
.next
, struct btrfs_inode
,
9599 list_move_tail(&binode
->delalloc_inodes
,
9600 &root
->delalloc_inodes
);
9601 inode
= igrab(&binode
->vfs_inode
);
9603 cond_resched_lock(&root
->delalloc_lock
);
9606 spin_unlock(&root
->delalloc_lock
);
9608 work
= btrfs_alloc_delalloc_work(inode
, delay_iput
);
9611 btrfs_add_delayed_iput(inode
);
9617 list_add_tail(&work
->list
, &works
);
9618 btrfs_queue_work(root
->fs_info
->flush_workers
,
9621 if (nr
!= -1 && ret
>= nr
)
9624 spin_lock(&root
->delalloc_lock
);
9626 spin_unlock(&root
->delalloc_lock
);
9629 list_for_each_entry_safe(work
, next
, &works
, list
) {
9630 list_del_init(&work
->list
);
9631 btrfs_wait_and_free_delalloc_work(work
);
9634 if (!list_empty_careful(&splice
)) {
9635 spin_lock(&root
->delalloc_lock
);
9636 list_splice_tail(&splice
, &root
->delalloc_inodes
);
9637 spin_unlock(&root
->delalloc_lock
);
9639 mutex_unlock(&root
->delalloc_mutex
);
9643 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
9647 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
9650 ret
= __start_delalloc_inodes(root
, delay_iput
, -1);
9654 * the filemap_flush will queue IO into the worker threads, but
9655 * we have to make sure the IO is actually started and that
9656 * ordered extents get created before we return
9658 atomic_inc(&root
->fs_info
->async_submit_draining
);
9659 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
9660 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
9661 wait_event(root
->fs_info
->async_submit_wait
,
9662 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
9663 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
9665 atomic_dec(&root
->fs_info
->async_submit_draining
);
9669 int btrfs_start_delalloc_roots(struct btrfs_fs_info
*fs_info
, int delay_iput
,
9672 struct btrfs_root
*root
;
9673 struct list_head splice
;
9676 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
9679 INIT_LIST_HEAD(&splice
);
9681 mutex_lock(&fs_info
->delalloc_root_mutex
);
9682 spin_lock(&fs_info
->delalloc_root_lock
);
9683 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
9684 while (!list_empty(&splice
) && nr
) {
9685 root
= list_first_entry(&splice
, struct btrfs_root
,
9687 root
= btrfs_grab_fs_root(root
);
9689 list_move_tail(&root
->delalloc_root
,
9690 &fs_info
->delalloc_roots
);
9691 spin_unlock(&fs_info
->delalloc_root_lock
);
9693 ret
= __start_delalloc_inodes(root
, delay_iput
, nr
);
9694 btrfs_put_fs_root(root
);
9702 spin_lock(&fs_info
->delalloc_root_lock
);
9704 spin_unlock(&fs_info
->delalloc_root_lock
);
9707 atomic_inc(&fs_info
->async_submit_draining
);
9708 while (atomic_read(&fs_info
->nr_async_submits
) ||
9709 atomic_read(&fs_info
->async_delalloc_pages
)) {
9710 wait_event(fs_info
->async_submit_wait
,
9711 (atomic_read(&fs_info
->nr_async_submits
) == 0 &&
9712 atomic_read(&fs_info
->async_delalloc_pages
) == 0));
9714 atomic_dec(&fs_info
->async_submit_draining
);
9716 if (!list_empty_careful(&splice
)) {
9717 spin_lock(&fs_info
->delalloc_root_lock
);
9718 list_splice_tail(&splice
, &fs_info
->delalloc_roots
);
9719 spin_unlock(&fs_info
->delalloc_root_lock
);
9721 mutex_unlock(&fs_info
->delalloc_root_mutex
);
9725 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
9726 const char *symname
)
9728 struct btrfs_trans_handle
*trans
;
9729 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
9730 struct btrfs_path
*path
;
9731 struct btrfs_key key
;
9732 struct inode
*inode
= NULL
;
9740 struct btrfs_file_extent_item
*ei
;
9741 struct extent_buffer
*leaf
;
9743 name_len
= strlen(symname
);
9744 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
9745 return -ENAMETOOLONG
;
9748 * 2 items for inode item and ref
9749 * 2 items for dir items
9750 * 1 item for updating parent inode item
9751 * 1 item for the inline extent item
9752 * 1 item for xattr if selinux is on
9754 trans
= btrfs_start_transaction(root
, 7);
9756 return PTR_ERR(trans
);
9758 err
= btrfs_find_free_ino(root
, &objectid
);
9762 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
9763 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
9764 S_IFLNK
|S_IRWXUGO
, &index
);
9765 if (IS_ERR(inode
)) {
9766 err
= PTR_ERR(inode
);
9771 * If the active LSM wants to access the inode during
9772 * d_instantiate it needs these. Smack checks to see
9773 * if the filesystem supports xattrs by looking at the
9776 inode
->i_fop
= &btrfs_file_operations
;
9777 inode
->i_op
= &btrfs_file_inode_operations
;
9778 inode
->i_mapping
->a_ops
= &btrfs_aops
;
9779 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
9781 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
9783 goto out_unlock_inode
;
9785 path
= btrfs_alloc_path();
9788 goto out_unlock_inode
;
9790 key
.objectid
= btrfs_ino(inode
);
9792 key
.type
= BTRFS_EXTENT_DATA_KEY
;
9793 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
9794 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
9797 btrfs_free_path(path
);
9798 goto out_unlock_inode
;
9800 leaf
= path
->nodes
[0];
9801 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
9802 struct btrfs_file_extent_item
);
9803 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
9804 btrfs_set_file_extent_type(leaf
, ei
,
9805 BTRFS_FILE_EXTENT_INLINE
);
9806 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
9807 btrfs_set_file_extent_compression(leaf
, ei
, 0);
9808 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
9809 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
9811 ptr
= btrfs_file_extent_inline_start(ei
);
9812 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
9813 btrfs_mark_buffer_dirty(leaf
);
9814 btrfs_free_path(path
);
9816 inode
->i_op
= &btrfs_symlink_inode_operations
;
9817 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
9818 inode_set_bytes(inode
, name_len
);
9819 btrfs_i_size_write(inode
, name_len
);
9820 err
= btrfs_update_inode(trans
, root
, inode
);
9822 * Last step, add directory indexes for our symlink inode. This is the
9823 * last step to avoid extra cleanup of these indexes if an error happens
9827 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
9830 goto out_unlock_inode
;
9833 unlock_new_inode(inode
);
9834 d_instantiate(dentry
, inode
);
9837 btrfs_end_transaction(trans
, root
);
9839 inode_dec_link_count(inode
);
9842 btrfs_btree_balance_dirty(root
);
9847 unlock_new_inode(inode
);
9851 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
9852 u64 start
, u64 num_bytes
, u64 min_size
,
9853 loff_t actual_len
, u64
*alloc_hint
,
9854 struct btrfs_trans_handle
*trans
)
9856 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
9857 struct extent_map
*em
;
9858 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9859 struct btrfs_key ins
;
9860 u64 cur_offset
= start
;
9863 u64 last_alloc
= (u64
)-1;
9865 bool own_trans
= true;
9869 while (num_bytes
> 0) {
9871 trans
= btrfs_start_transaction(root
, 3);
9872 if (IS_ERR(trans
)) {
9873 ret
= PTR_ERR(trans
);
9878 cur_bytes
= min_t(u64
, num_bytes
, SZ_256M
);
9879 cur_bytes
= max(cur_bytes
, min_size
);
9881 * If we are severely fragmented we could end up with really
9882 * small allocations, so if the allocator is returning small
9883 * chunks lets make its job easier by only searching for those
9886 cur_bytes
= min(cur_bytes
, last_alloc
);
9887 ret
= btrfs_reserve_extent(root
, cur_bytes
, min_size
, 0,
9888 *alloc_hint
, &ins
, 1, 0);
9891 btrfs_end_transaction(trans
, root
);
9895 last_alloc
= ins
.offset
;
9896 ret
= insert_reserved_file_extent(trans
, inode
,
9897 cur_offset
, ins
.objectid
,
9898 ins
.offset
, ins
.offset
,
9899 ins
.offset
, 0, 0, 0,
9900 BTRFS_FILE_EXTENT_PREALLOC
);
9902 btrfs_free_reserved_extent(root
, ins
.objectid
,
9904 btrfs_abort_transaction(trans
, root
, ret
);
9906 btrfs_end_transaction(trans
, root
);
9910 btrfs_drop_extent_cache(inode
, cur_offset
,
9911 cur_offset
+ ins
.offset
-1, 0);
9913 em
= alloc_extent_map();
9915 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
9916 &BTRFS_I(inode
)->runtime_flags
);
9920 em
->start
= cur_offset
;
9921 em
->orig_start
= cur_offset
;
9922 em
->len
= ins
.offset
;
9923 em
->block_start
= ins
.objectid
;
9924 em
->block_len
= ins
.offset
;
9925 em
->orig_block_len
= ins
.offset
;
9926 em
->ram_bytes
= ins
.offset
;
9927 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
9928 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
9929 em
->generation
= trans
->transid
;
9932 write_lock(&em_tree
->lock
);
9933 ret
= add_extent_mapping(em_tree
, em
, 1);
9934 write_unlock(&em_tree
->lock
);
9937 btrfs_drop_extent_cache(inode
, cur_offset
,
9938 cur_offset
+ ins
.offset
- 1,
9941 free_extent_map(em
);
9943 num_bytes
-= ins
.offset
;
9944 cur_offset
+= ins
.offset
;
9945 *alloc_hint
= ins
.objectid
+ ins
.offset
;
9947 inode_inc_iversion(inode
);
9948 inode
->i_ctime
= CURRENT_TIME
;
9949 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
9950 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
9951 (actual_len
> inode
->i_size
) &&
9952 (cur_offset
> inode
->i_size
)) {
9953 if (cur_offset
> actual_len
)
9954 i_size
= actual_len
;
9956 i_size
= cur_offset
;
9957 i_size_write(inode
, i_size
);
9958 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
9961 ret
= btrfs_update_inode(trans
, root
, inode
);
9964 btrfs_abort_transaction(trans
, root
, ret
);
9966 btrfs_end_transaction(trans
, root
);
9971 btrfs_end_transaction(trans
, root
);
9976 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
9977 u64 start
, u64 num_bytes
, u64 min_size
,
9978 loff_t actual_len
, u64
*alloc_hint
)
9980 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
9981 min_size
, actual_len
, alloc_hint
,
9985 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
9986 struct btrfs_trans_handle
*trans
, int mode
,
9987 u64 start
, u64 num_bytes
, u64 min_size
,
9988 loff_t actual_len
, u64
*alloc_hint
)
9990 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
9991 min_size
, actual_len
, alloc_hint
, trans
);
9994 static int btrfs_set_page_dirty(struct page
*page
)
9996 return __set_page_dirty_nobuffers(page
);
9999 static int btrfs_permission(struct inode
*inode
, int mask
)
10001 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
10002 umode_t mode
= inode
->i_mode
;
10004 if (mask
& MAY_WRITE
&&
10005 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
10006 if (btrfs_root_readonly(root
))
10008 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
10011 return generic_permission(inode
, mask
);
10014 static int btrfs_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
10016 struct btrfs_trans_handle
*trans
;
10017 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
10018 struct inode
*inode
= NULL
;
10024 * 5 units required for adding orphan entry
10026 trans
= btrfs_start_transaction(root
, 5);
10028 return PTR_ERR(trans
);
10030 ret
= btrfs_find_free_ino(root
, &objectid
);
10034 inode
= btrfs_new_inode(trans
, root
, dir
, NULL
, 0,
10035 btrfs_ino(dir
), objectid
, mode
, &index
);
10036 if (IS_ERR(inode
)) {
10037 ret
= PTR_ERR(inode
);
10042 inode
->i_fop
= &btrfs_file_operations
;
10043 inode
->i_op
= &btrfs_file_inode_operations
;
10045 inode
->i_mapping
->a_ops
= &btrfs_aops
;
10046 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
10048 ret
= btrfs_init_inode_security(trans
, inode
, dir
, NULL
);
10052 ret
= btrfs_update_inode(trans
, root
, inode
);
10055 ret
= btrfs_orphan_add(trans
, inode
);
10060 * We set number of links to 0 in btrfs_new_inode(), and here we set
10061 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10064 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10066 set_nlink(inode
, 1);
10067 unlock_new_inode(inode
);
10068 d_tmpfile(dentry
, inode
);
10069 mark_inode_dirty(inode
);
10072 btrfs_end_transaction(trans
, root
);
10075 btrfs_balance_delayed_items(root
);
10076 btrfs_btree_balance_dirty(root
);
10080 unlock_new_inode(inode
);
10085 /* Inspired by filemap_check_errors() */
10086 int btrfs_inode_check_errors(struct inode
*inode
)
10090 if (test_bit(AS_ENOSPC
, &inode
->i_mapping
->flags
) &&
10091 test_and_clear_bit(AS_ENOSPC
, &inode
->i_mapping
->flags
))
10093 if (test_bit(AS_EIO
, &inode
->i_mapping
->flags
) &&
10094 test_and_clear_bit(AS_EIO
, &inode
->i_mapping
->flags
))
10100 static const struct inode_operations btrfs_dir_inode_operations
= {
10101 .getattr
= btrfs_getattr
,
10102 .lookup
= btrfs_lookup
,
10103 .create
= btrfs_create
,
10104 .unlink
= btrfs_unlink
,
10105 .link
= btrfs_link
,
10106 .mkdir
= btrfs_mkdir
,
10107 .rmdir
= btrfs_rmdir
,
10108 .rename2
= btrfs_rename2
,
10109 .symlink
= btrfs_symlink
,
10110 .setattr
= btrfs_setattr
,
10111 .mknod
= btrfs_mknod
,
10112 .setxattr
= btrfs_setxattr
,
10113 .getxattr
= btrfs_getxattr
,
10114 .listxattr
= btrfs_listxattr
,
10115 .removexattr
= btrfs_removexattr
,
10116 .permission
= btrfs_permission
,
10117 .get_acl
= btrfs_get_acl
,
10118 .set_acl
= btrfs_set_acl
,
10119 .update_time
= btrfs_update_time
,
10120 .tmpfile
= btrfs_tmpfile
,
10122 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
10123 .lookup
= btrfs_lookup
,
10124 .permission
= btrfs_permission
,
10125 .get_acl
= btrfs_get_acl
,
10126 .set_acl
= btrfs_set_acl
,
10127 .update_time
= btrfs_update_time
,
10130 static const struct file_operations btrfs_dir_file_operations
= {
10131 .llseek
= generic_file_llseek
,
10132 .read
= generic_read_dir
,
10133 .iterate
= btrfs_real_readdir
,
10134 .unlocked_ioctl
= btrfs_ioctl
,
10135 #ifdef CONFIG_COMPAT
10136 .compat_ioctl
= btrfs_ioctl
,
10138 .release
= btrfs_release_file
,
10139 .fsync
= btrfs_sync_file
,
10142 static const struct extent_io_ops btrfs_extent_io_ops
= {
10143 .fill_delalloc
= run_delalloc_range
,
10144 .submit_bio_hook
= btrfs_submit_bio_hook
,
10145 .merge_bio_hook
= btrfs_merge_bio_hook
,
10146 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
10147 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
10148 .writepage_start_hook
= btrfs_writepage_start_hook
,
10149 .set_bit_hook
= btrfs_set_bit_hook
,
10150 .clear_bit_hook
= btrfs_clear_bit_hook
,
10151 .merge_extent_hook
= btrfs_merge_extent_hook
,
10152 .split_extent_hook
= btrfs_split_extent_hook
,
10156 * btrfs doesn't support the bmap operation because swapfiles
10157 * use bmap to make a mapping of extents in the file. They assume
10158 * these extents won't change over the life of the file and they
10159 * use the bmap result to do IO directly to the drive.
10161 * the btrfs bmap call would return logical addresses that aren't
10162 * suitable for IO and they also will change frequently as COW
10163 * operations happen. So, swapfile + btrfs == corruption.
10165 * For now we're avoiding this by dropping bmap.
10167 static const struct address_space_operations btrfs_aops
= {
10168 .readpage
= btrfs_readpage
,
10169 .writepage
= btrfs_writepage
,
10170 .writepages
= btrfs_writepages
,
10171 .readpages
= btrfs_readpages
,
10172 .direct_IO
= btrfs_direct_IO
,
10173 .invalidatepage
= btrfs_invalidatepage
,
10174 .releasepage
= btrfs_releasepage
,
10175 .set_page_dirty
= btrfs_set_page_dirty
,
10176 .error_remove_page
= generic_error_remove_page
,
10179 static const struct address_space_operations btrfs_symlink_aops
= {
10180 .readpage
= btrfs_readpage
,
10181 .writepage
= btrfs_writepage
,
10182 .invalidatepage
= btrfs_invalidatepage
,
10183 .releasepage
= btrfs_releasepage
,
10186 static const struct inode_operations btrfs_file_inode_operations
= {
10187 .getattr
= btrfs_getattr
,
10188 .setattr
= btrfs_setattr
,
10189 .setxattr
= btrfs_setxattr
,
10190 .getxattr
= btrfs_getxattr
,
10191 .listxattr
= btrfs_listxattr
,
10192 .removexattr
= btrfs_removexattr
,
10193 .permission
= btrfs_permission
,
10194 .fiemap
= btrfs_fiemap
,
10195 .get_acl
= btrfs_get_acl
,
10196 .set_acl
= btrfs_set_acl
,
10197 .update_time
= btrfs_update_time
,
10199 static const struct inode_operations btrfs_special_inode_operations
= {
10200 .getattr
= btrfs_getattr
,
10201 .setattr
= btrfs_setattr
,
10202 .permission
= btrfs_permission
,
10203 .setxattr
= btrfs_setxattr
,
10204 .getxattr
= btrfs_getxattr
,
10205 .listxattr
= btrfs_listxattr
,
10206 .removexattr
= btrfs_removexattr
,
10207 .get_acl
= btrfs_get_acl
,
10208 .set_acl
= btrfs_set_acl
,
10209 .update_time
= btrfs_update_time
,
10211 static const struct inode_operations btrfs_symlink_inode_operations
= {
10212 .readlink
= generic_readlink
,
10213 .follow_link
= page_follow_link_light
,
10214 .put_link
= page_put_link
,
10215 .getattr
= btrfs_getattr
,
10216 .setattr
= btrfs_setattr
,
10217 .permission
= btrfs_permission
,
10218 .setxattr
= btrfs_setxattr
,
10219 .getxattr
= btrfs_getxattr
,
10220 .listxattr
= btrfs_listxattr
,
10221 .removexattr
= btrfs_removexattr
,
10222 .update_time
= btrfs_update_time
,
10225 const struct dentry_operations btrfs_dentry_operations
= {
10226 .d_delete
= btrfs_dentry_delete
,
10227 .d_release
= btrfs_dentry_release
,