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 static const struct inode_operations btrfs_dir_inode_operations
;
70 static const struct inode_operations btrfs_symlink_inode_operations
;
71 static const struct inode_operations btrfs_dir_ro_inode_operations
;
72 static const struct inode_operations btrfs_special_inode_operations
;
73 static const struct inode_operations btrfs_file_inode_operations
;
74 static const struct address_space_operations btrfs_aops
;
75 static const struct address_space_operations btrfs_symlink_aops
;
76 static const struct file_operations btrfs_dir_file_operations
;
77 static struct extent_io_ops btrfs_extent_io_ops
;
79 static struct kmem_cache
*btrfs_inode_cachep
;
80 static struct kmem_cache
*btrfs_delalloc_work_cachep
;
81 struct kmem_cache
*btrfs_trans_handle_cachep
;
82 struct kmem_cache
*btrfs_transaction_cachep
;
83 struct kmem_cache
*btrfs_path_cachep
;
84 struct kmem_cache
*btrfs_free_space_cachep
;
87 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
88 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
89 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
90 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
91 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
92 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
93 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
94 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
97 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
);
98 static int btrfs_truncate(struct inode
*inode
);
99 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
);
100 static noinline
int cow_file_range(struct inode
*inode
,
101 struct page
*locked_page
,
102 u64 start
, u64 end
, int *page_started
,
103 unsigned long *nr_written
, int unlock
);
104 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
105 u64 len
, u64 orig_start
,
106 u64 block_start
, u64 block_len
,
107 u64 orig_block_len
, u64 ram_bytes
,
110 static int btrfs_dirty_inode(struct inode
*inode
);
112 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
113 void btrfs_test_inode_set_ops(struct inode
*inode
)
115 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
119 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
120 struct inode
*inode
, struct inode
*dir
,
121 const struct qstr
*qstr
)
125 err
= btrfs_init_acl(trans
, inode
, dir
);
127 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
132 * this does all the hard work for inserting an inline extent into
133 * the btree. The caller should have done a btrfs_drop_extents so that
134 * no overlapping inline items exist in the btree
136 static int insert_inline_extent(struct btrfs_trans_handle
*trans
,
137 struct btrfs_path
*path
, int extent_inserted
,
138 struct btrfs_root
*root
, struct inode
*inode
,
139 u64 start
, size_t size
, size_t compressed_size
,
141 struct page
**compressed_pages
)
143 struct extent_buffer
*leaf
;
144 struct page
*page
= NULL
;
147 struct btrfs_file_extent_item
*ei
;
150 size_t cur_size
= size
;
151 unsigned long offset
;
153 if (compressed_size
&& compressed_pages
)
154 cur_size
= compressed_size
;
156 inode_add_bytes(inode
, size
);
158 if (!extent_inserted
) {
159 struct btrfs_key key
;
162 key
.objectid
= btrfs_ino(inode
);
164 key
.type
= BTRFS_EXTENT_DATA_KEY
;
166 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
167 path
->leave_spinning
= 1;
168 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
175 leaf
= path
->nodes
[0];
176 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
177 struct btrfs_file_extent_item
);
178 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
179 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
180 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
181 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
182 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
183 ptr
= btrfs_file_extent_inline_start(ei
);
185 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
188 while (compressed_size
> 0) {
189 cpage
= compressed_pages
[i
];
190 cur_size
= min_t(unsigned long, compressed_size
,
193 kaddr
= kmap_atomic(cpage
);
194 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
195 kunmap_atomic(kaddr
);
199 compressed_size
-= cur_size
;
201 btrfs_set_file_extent_compression(leaf
, ei
,
204 page
= find_get_page(inode
->i_mapping
,
205 start
>> PAGE_CACHE_SHIFT
);
206 btrfs_set_file_extent_compression(leaf
, ei
, 0);
207 kaddr
= kmap_atomic(page
);
208 offset
= start
& (PAGE_CACHE_SIZE
- 1);
209 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
210 kunmap_atomic(kaddr
);
211 page_cache_release(page
);
213 btrfs_mark_buffer_dirty(leaf
);
214 btrfs_release_path(path
);
217 * we're an inline extent, so nobody can
218 * extend the file past i_size without locking
219 * a page we already have locked.
221 * We must do any isize and inode updates
222 * before we unlock the pages. Otherwise we
223 * could end up racing with unlink.
225 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
226 ret
= btrfs_update_inode(trans
, root
, inode
);
235 * conditionally insert an inline extent into the file. This
236 * does the checks required to make sure the data is small enough
237 * to fit as an inline extent.
239 static noinline
int cow_file_range_inline(struct btrfs_root
*root
,
240 struct inode
*inode
, u64 start
,
241 u64 end
, size_t compressed_size
,
243 struct page
**compressed_pages
)
245 struct btrfs_trans_handle
*trans
;
246 u64 isize
= i_size_read(inode
);
247 u64 actual_end
= min(end
+ 1, isize
);
248 u64 inline_len
= actual_end
- start
;
249 u64 aligned_end
= ALIGN(end
, root
->sectorsize
);
250 u64 data_len
= inline_len
;
252 struct btrfs_path
*path
;
253 int extent_inserted
= 0;
254 u32 extent_item_size
;
257 data_len
= compressed_size
;
260 actual_end
> PAGE_CACHE_SIZE
||
261 data_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
263 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
265 data_len
> root
->fs_info
->max_inline
) {
269 path
= btrfs_alloc_path();
273 trans
= btrfs_join_transaction(root
);
275 btrfs_free_path(path
);
276 return PTR_ERR(trans
);
278 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
280 if (compressed_size
&& compressed_pages
)
281 extent_item_size
= btrfs_file_extent_calc_inline_size(
284 extent_item_size
= btrfs_file_extent_calc_inline_size(
287 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
,
288 start
, aligned_end
, NULL
,
289 1, 1, extent_item_size
, &extent_inserted
);
291 btrfs_abort_transaction(trans
, root
, ret
);
295 if (isize
> actual_end
)
296 inline_len
= min_t(u64
, isize
, actual_end
);
297 ret
= insert_inline_extent(trans
, path
, extent_inserted
,
299 inline_len
, compressed_size
,
300 compress_type
, compressed_pages
);
301 if (ret
&& ret
!= -ENOSPC
) {
302 btrfs_abort_transaction(trans
, root
, ret
);
304 } else if (ret
== -ENOSPC
) {
309 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
310 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
311 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
314 * Don't forget to free the reserved space, as for inlined extent
315 * it won't count as data extent, free them directly here.
316 * And at reserve time, it's always aligned to page size, so
317 * just free one page here.
319 btrfs_qgroup_free_data(inode
, 0, PAGE_CACHE_SIZE
);
320 btrfs_free_path(path
);
321 btrfs_end_transaction(trans
, root
);
325 struct async_extent
{
330 unsigned long nr_pages
;
332 struct list_head list
;
337 struct btrfs_root
*root
;
338 struct page
*locked_page
;
341 struct list_head extents
;
342 struct btrfs_work work
;
345 static noinline
int add_async_extent(struct async_cow
*cow
,
346 u64 start
, u64 ram_size
,
349 unsigned long nr_pages
,
352 struct async_extent
*async_extent
;
354 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
355 BUG_ON(!async_extent
); /* -ENOMEM */
356 async_extent
->start
= start
;
357 async_extent
->ram_size
= ram_size
;
358 async_extent
->compressed_size
= compressed_size
;
359 async_extent
->pages
= pages
;
360 async_extent
->nr_pages
= nr_pages
;
361 async_extent
->compress_type
= compress_type
;
362 list_add_tail(&async_extent
->list
, &cow
->extents
);
366 static inline int inode_need_compress(struct inode
*inode
)
368 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
371 if (btrfs_test_opt(root
, FORCE_COMPRESS
))
373 /* bad compression ratios */
374 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
376 if (btrfs_test_opt(root
, COMPRESS
) ||
377 BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
||
378 BTRFS_I(inode
)->force_compress
)
384 * we create compressed extents in two phases. The first
385 * phase compresses a range of pages that have already been
386 * locked (both pages and state bits are locked).
388 * This is done inside an ordered work queue, and the compression
389 * is spread across many cpus. The actual IO submission is step
390 * two, and the ordered work queue takes care of making sure that
391 * happens in the same order things were put onto the queue by
392 * writepages and friends.
394 * If this code finds it can't get good compression, it puts an
395 * entry onto the work queue to write the uncompressed bytes. This
396 * makes sure that both compressed inodes and uncompressed inodes
397 * are written in the same order that the flusher thread sent them
400 static noinline
void compress_file_range(struct inode
*inode
,
401 struct page
*locked_page
,
403 struct async_cow
*async_cow
,
406 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
408 u64 blocksize
= root
->sectorsize
;
410 u64 isize
= i_size_read(inode
);
412 struct page
**pages
= NULL
;
413 unsigned long nr_pages
;
414 unsigned long nr_pages_ret
= 0;
415 unsigned long total_compressed
= 0;
416 unsigned long total_in
= 0;
417 unsigned long max_compressed
= 128 * 1024;
418 unsigned long max_uncompressed
= 128 * 1024;
421 int compress_type
= root
->fs_info
->compress_type
;
424 /* if this is a small write inside eof, kick off a defrag */
425 if ((end
- start
+ 1) < 16 * 1024 &&
426 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
427 btrfs_add_inode_defrag(NULL
, inode
);
429 actual_end
= min_t(u64
, isize
, end
+ 1);
432 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
433 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
436 * we don't want to send crud past the end of i_size through
437 * compression, that's just a waste of CPU time. So, if the
438 * end of the file is before the start of our current
439 * requested range of bytes, we bail out to the uncompressed
440 * cleanup code that can deal with all of this.
442 * It isn't really the fastest way to fix things, but this is a
443 * very uncommon corner.
445 if (actual_end
<= start
)
446 goto cleanup_and_bail_uncompressed
;
448 total_compressed
= actual_end
- start
;
451 * skip compression for a small file range(<=blocksize) that
452 * isn't an inline extent, since it dosen't save disk space at all.
454 if (total_compressed
<= blocksize
&&
455 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
456 goto cleanup_and_bail_uncompressed
;
458 /* we want to make sure that amount of ram required to uncompress
459 * an extent is reasonable, so we limit the total size in ram
460 * of a compressed extent to 128k. This is a crucial number
461 * because it also controls how easily we can spread reads across
462 * cpus for decompression.
464 * We also want to make sure the amount of IO required to do
465 * a random read is reasonably small, so we limit the size of
466 * a compressed extent to 128k.
468 total_compressed
= min(total_compressed
, max_uncompressed
);
469 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
470 num_bytes
= max(blocksize
, num_bytes
);
475 * we do compression for mount -o compress and when the
476 * inode has not been flagged as nocompress. This flag can
477 * change at any time if we discover bad compression ratios.
479 if (inode_need_compress(inode
)) {
481 pages
= kcalloc(nr_pages
, sizeof(struct page
*), GFP_NOFS
);
483 /* just bail out to the uncompressed code */
487 if (BTRFS_I(inode
)->force_compress
)
488 compress_type
= BTRFS_I(inode
)->force_compress
;
491 * we need to call clear_page_dirty_for_io on each
492 * page in the range. Otherwise applications with the file
493 * mmap'd can wander in and change the page contents while
494 * we are compressing them.
496 * If the compression fails for any reason, we set the pages
497 * dirty again later on.
499 extent_range_clear_dirty_for_io(inode
, start
, end
);
501 ret
= btrfs_compress_pages(compress_type
,
502 inode
->i_mapping
, start
,
503 total_compressed
, pages
,
504 nr_pages
, &nr_pages_ret
,
510 unsigned long offset
= total_compressed
&
511 (PAGE_CACHE_SIZE
- 1);
512 struct page
*page
= pages
[nr_pages_ret
- 1];
515 /* zero the tail end of the last page, we might be
516 * sending it down to disk
519 kaddr
= kmap_atomic(page
);
520 memset(kaddr
+ offset
, 0,
521 PAGE_CACHE_SIZE
- offset
);
522 kunmap_atomic(kaddr
);
529 /* lets try to make an inline extent */
530 if (ret
|| total_in
< (actual_end
- start
)) {
531 /* we didn't compress the entire range, try
532 * to make an uncompressed inline extent.
534 ret
= cow_file_range_inline(root
, inode
, start
, end
,
537 /* try making a compressed inline extent */
538 ret
= cow_file_range_inline(root
, inode
, start
, end
,
540 compress_type
, pages
);
543 unsigned long clear_flags
= EXTENT_DELALLOC
|
545 unsigned long page_error_op
;
547 clear_flags
|= (ret
< 0) ? EXTENT_DO_ACCOUNTING
: 0;
548 page_error_op
= ret
< 0 ? PAGE_SET_ERROR
: 0;
551 * inline extent creation worked or returned error,
552 * we don't need to create any more async work items.
553 * Unlock and free up our temp pages.
555 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
,
556 clear_flags
, PAGE_UNLOCK
|
567 * we aren't doing an inline extent round the compressed size
568 * up to a block size boundary so the allocator does sane
571 total_compressed
= ALIGN(total_compressed
, blocksize
);
574 * one last check to make sure the compression is really a
575 * win, compare the page count read with the blocks on disk
577 total_in
= ALIGN(total_in
, PAGE_CACHE_SIZE
);
578 if (total_compressed
>= total_in
) {
581 num_bytes
= total_in
;
584 if (!will_compress
&& pages
) {
586 * the compression code ran but failed to make things smaller,
587 * free any pages it allocated and our page pointer array
589 for (i
= 0; i
< nr_pages_ret
; i
++) {
590 WARN_ON(pages
[i
]->mapping
);
591 page_cache_release(pages
[i
]);
595 total_compressed
= 0;
598 /* flag the file so we don't compress in the future */
599 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
600 !(BTRFS_I(inode
)->force_compress
)) {
601 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
607 /* the async work queues will take care of doing actual
608 * allocation on disk for these compressed pages,
609 * and will submit them to the elevator.
611 add_async_extent(async_cow
, start
, num_bytes
,
612 total_compressed
, pages
, nr_pages_ret
,
615 if (start
+ num_bytes
< end
) {
622 cleanup_and_bail_uncompressed
:
624 * No compression, but we still need to write the pages in
625 * the file we've been given so far. redirty the locked
626 * page if it corresponds to our extent and set things up
627 * for the async work queue to run cow_file_range to do
628 * the normal delalloc dance
630 if (page_offset(locked_page
) >= start
&&
631 page_offset(locked_page
) <= end
) {
632 __set_page_dirty_nobuffers(locked_page
);
633 /* unlocked later on in the async handlers */
636 extent_range_redirty_for_io(inode
, start
, end
);
637 add_async_extent(async_cow
, start
, end
- start
+ 1,
638 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
645 for (i
= 0; i
< nr_pages_ret
; i
++) {
646 WARN_ON(pages
[i
]->mapping
);
647 page_cache_release(pages
[i
]);
652 static void free_async_extent_pages(struct async_extent
*async_extent
)
656 if (!async_extent
->pages
)
659 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
660 WARN_ON(async_extent
->pages
[i
]->mapping
);
661 page_cache_release(async_extent
->pages
[i
]);
663 kfree(async_extent
->pages
);
664 async_extent
->nr_pages
= 0;
665 async_extent
->pages
= NULL
;
669 * phase two of compressed writeback. This is the ordered portion
670 * of the code, which only gets called in the order the work was
671 * queued. We walk all the async extents created by compress_file_range
672 * and send them down to the disk.
674 static noinline
void submit_compressed_extents(struct inode
*inode
,
675 struct async_cow
*async_cow
)
677 struct async_extent
*async_extent
;
679 struct btrfs_key ins
;
680 struct extent_map
*em
;
681 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
682 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
683 struct extent_io_tree
*io_tree
;
687 while (!list_empty(&async_cow
->extents
)) {
688 async_extent
= list_entry(async_cow
->extents
.next
,
689 struct async_extent
, list
);
690 list_del(&async_extent
->list
);
692 io_tree
= &BTRFS_I(inode
)->io_tree
;
695 /* did the compression code fall back to uncompressed IO? */
696 if (!async_extent
->pages
) {
697 int page_started
= 0;
698 unsigned long nr_written
= 0;
700 lock_extent(io_tree
, async_extent
->start
,
701 async_extent
->start
+
702 async_extent
->ram_size
- 1);
704 /* allocate blocks */
705 ret
= cow_file_range(inode
, async_cow
->locked_page
,
707 async_extent
->start
+
708 async_extent
->ram_size
- 1,
709 &page_started
, &nr_written
, 0);
714 * if page_started, cow_file_range inserted an
715 * inline extent and took care of all the unlocking
716 * and IO for us. Otherwise, we need to submit
717 * all those pages down to the drive.
719 if (!page_started
&& !ret
)
720 extent_write_locked_range(io_tree
,
721 inode
, async_extent
->start
,
722 async_extent
->start
+
723 async_extent
->ram_size
- 1,
727 unlock_page(async_cow
->locked_page
);
733 lock_extent(io_tree
, async_extent
->start
,
734 async_extent
->start
+ async_extent
->ram_size
- 1);
736 ret
= btrfs_reserve_extent(root
,
737 async_extent
->compressed_size
,
738 async_extent
->compressed_size
,
739 0, alloc_hint
, &ins
, 1, 1);
741 free_async_extent_pages(async_extent
);
743 if (ret
== -ENOSPC
) {
744 unlock_extent(io_tree
, async_extent
->start
,
745 async_extent
->start
+
746 async_extent
->ram_size
- 1);
749 * we need to redirty the pages if we decide to
750 * fallback to uncompressed IO, otherwise we
751 * will not submit these pages down to lower
754 extent_range_redirty_for_io(inode
,
756 async_extent
->start
+
757 async_extent
->ram_size
- 1);
764 * here we're doing allocation and writeback of the
767 btrfs_drop_extent_cache(inode
, async_extent
->start
,
768 async_extent
->start
+
769 async_extent
->ram_size
- 1, 0);
771 em
= alloc_extent_map();
774 goto out_free_reserve
;
776 em
->start
= async_extent
->start
;
777 em
->len
= async_extent
->ram_size
;
778 em
->orig_start
= em
->start
;
779 em
->mod_start
= em
->start
;
780 em
->mod_len
= em
->len
;
782 em
->block_start
= ins
.objectid
;
783 em
->block_len
= ins
.offset
;
784 em
->orig_block_len
= ins
.offset
;
785 em
->ram_bytes
= async_extent
->ram_size
;
786 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
787 em
->compress_type
= async_extent
->compress_type
;
788 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
789 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
793 write_lock(&em_tree
->lock
);
794 ret
= add_extent_mapping(em_tree
, em
, 1);
795 write_unlock(&em_tree
->lock
);
796 if (ret
!= -EEXIST
) {
800 btrfs_drop_extent_cache(inode
, async_extent
->start
,
801 async_extent
->start
+
802 async_extent
->ram_size
- 1, 0);
806 goto out_free_reserve
;
808 ret
= btrfs_add_ordered_extent_compress(inode
,
811 async_extent
->ram_size
,
813 BTRFS_ORDERED_COMPRESSED
,
814 async_extent
->compress_type
);
816 btrfs_drop_extent_cache(inode
, async_extent
->start
,
817 async_extent
->start
+
818 async_extent
->ram_size
- 1, 0);
819 goto out_free_reserve
;
823 * clear dirty, set writeback and unlock the pages.
825 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
826 async_extent
->start
+
827 async_extent
->ram_size
- 1,
828 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
,
829 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
831 ret
= btrfs_submit_compressed_write(inode
,
833 async_extent
->ram_size
,
835 ins
.offset
, async_extent
->pages
,
836 async_extent
->nr_pages
);
838 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
839 struct page
*p
= async_extent
->pages
[0];
840 const u64 start
= async_extent
->start
;
841 const u64 end
= start
+ async_extent
->ram_size
- 1;
843 p
->mapping
= inode
->i_mapping
;
844 tree
->ops
->writepage_end_io_hook(p
, start
, end
,
847 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
, 0,
850 free_async_extent_pages(async_extent
);
852 alloc_hint
= ins
.objectid
+ ins
.offset
;
858 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
860 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
861 async_extent
->start
+
862 async_extent
->ram_size
- 1,
863 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
|
864 EXTENT_DEFRAG
| EXTENT_DO_ACCOUNTING
,
865 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
866 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
|
868 free_async_extent_pages(async_extent
);
873 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
876 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
877 struct extent_map
*em
;
880 read_lock(&em_tree
->lock
);
881 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
884 * if block start isn't an actual block number then find the
885 * first block in this inode and use that as a hint. If that
886 * block is also bogus then just don't worry about it.
888 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
890 em
= search_extent_mapping(em_tree
, 0, 0);
891 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
892 alloc_hint
= em
->block_start
;
896 alloc_hint
= em
->block_start
;
900 read_unlock(&em_tree
->lock
);
906 * when extent_io.c finds a delayed allocation range in the file,
907 * the call backs end up in this code. The basic idea is to
908 * allocate extents on disk for the range, and create ordered data structs
909 * in ram to track those extents.
911 * locked_page is the page that writepage had locked already. We use
912 * it to make sure we don't do extra locks or unlocks.
914 * *page_started is set to one if we unlock locked_page and do everything
915 * required to start IO on it. It may be clean and already done with
918 static noinline
int cow_file_range(struct inode
*inode
,
919 struct page
*locked_page
,
920 u64 start
, u64 end
, int *page_started
,
921 unsigned long *nr_written
,
924 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
927 unsigned long ram_size
;
930 u64 blocksize
= root
->sectorsize
;
931 struct btrfs_key ins
;
932 struct extent_map
*em
;
933 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
936 if (btrfs_is_free_space_inode(inode
)) {
942 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
943 num_bytes
= max(blocksize
, num_bytes
);
944 disk_num_bytes
= num_bytes
;
946 /* if this is a small write inside eof, kick off defrag */
947 if (num_bytes
< 64 * 1024 &&
948 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
949 btrfs_add_inode_defrag(NULL
, inode
);
952 /* lets try to make an inline extent */
953 ret
= cow_file_range_inline(root
, inode
, start
, end
, 0, 0,
956 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
,
957 EXTENT_LOCKED
| EXTENT_DELALLOC
|
958 EXTENT_DEFRAG
, PAGE_UNLOCK
|
959 PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
962 *nr_written
= *nr_written
+
963 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
966 } else if (ret
< 0) {
971 BUG_ON(disk_num_bytes
>
972 btrfs_super_total_bytes(root
->fs_info
->super_copy
));
974 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
975 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
977 while (disk_num_bytes
> 0) {
980 cur_alloc_size
= disk_num_bytes
;
981 ret
= btrfs_reserve_extent(root
, cur_alloc_size
,
982 root
->sectorsize
, 0, alloc_hint
,
987 em
= alloc_extent_map();
993 em
->orig_start
= em
->start
;
994 ram_size
= ins
.offset
;
995 em
->len
= ins
.offset
;
996 em
->mod_start
= em
->start
;
997 em
->mod_len
= em
->len
;
999 em
->block_start
= ins
.objectid
;
1000 em
->block_len
= ins
.offset
;
1001 em
->orig_block_len
= ins
.offset
;
1002 em
->ram_bytes
= ram_size
;
1003 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1004 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1005 em
->generation
= -1;
1008 write_lock(&em_tree
->lock
);
1009 ret
= add_extent_mapping(em_tree
, em
, 1);
1010 write_unlock(&em_tree
->lock
);
1011 if (ret
!= -EEXIST
) {
1012 free_extent_map(em
);
1015 btrfs_drop_extent_cache(inode
, start
,
1016 start
+ ram_size
- 1, 0);
1021 cur_alloc_size
= ins
.offset
;
1022 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
1023 ram_size
, cur_alloc_size
, 0);
1025 goto out_drop_extent_cache
;
1027 if (root
->root_key
.objectid
==
1028 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1029 ret
= btrfs_reloc_clone_csums(inode
, start
,
1032 goto out_drop_extent_cache
;
1035 if (disk_num_bytes
< cur_alloc_size
)
1038 /* we're not doing compressed IO, don't unlock the first
1039 * page (which the caller expects to stay locked), don't
1040 * clear any dirty bits and don't set any writeback bits
1042 * Do set the Private2 bit so we know this page was properly
1043 * setup for writepage
1045 op
= unlock
? PAGE_UNLOCK
: 0;
1046 op
|= PAGE_SET_PRIVATE2
;
1048 extent_clear_unlock_delalloc(inode
, start
,
1049 start
+ ram_size
- 1, locked_page
,
1050 EXTENT_LOCKED
| EXTENT_DELALLOC
,
1052 disk_num_bytes
-= cur_alloc_size
;
1053 num_bytes
-= cur_alloc_size
;
1054 alloc_hint
= ins
.objectid
+ ins
.offset
;
1055 start
+= cur_alloc_size
;
1060 out_drop_extent_cache
:
1061 btrfs_drop_extent_cache(inode
, start
, start
+ ram_size
- 1, 0);
1063 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
1065 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1066 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
1067 EXTENT_DELALLOC
| EXTENT_DEFRAG
,
1068 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
1069 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
);
1074 * work queue call back to started compression on a file and pages
1076 static noinline
void async_cow_start(struct btrfs_work
*work
)
1078 struct async_cow
*async_cow
;
1080 async_cow
= container_of(work
, struct async_cow
, work
);
1082 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
1083 async_cow
->start
, async_cow
->end
, async_cow
,
1085 if (num_added
== 0) {
1086 btrfs_add_delayed_iput(async_cow
->inode
);
1087 async_cow
->inode
= NULL
;
1092 * work queue call back to submit previously compressed pages
1094 static noinline
void async_cow_submit(struct btrfs_work
*work
)
1096 struct async_cow
*async_cow
;
1097 struct btrfs_root
*root
;
1098 unsigned long nr_pages
;
1100 async_cow
= container_of(work
, struct async_cow
, work
);
1102 root
= async_cow
->root
;
1103 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
1107 * atomic_sub_return implies a barrier for waitqueue_active
1109 if (atomic_sub_return(nr_pages
, &root
->fs_info
->async_delalloc_pages
) <
1111 waitqueue_active(&root
->fs_info
->async_submit_wait
))
1112 wake_up(&root
->fs_info
->async_submit_wait
);
1114 if (async_cow
->inode
)
1115 submit_compressed_extents(async_cow
->inode
, async_cow
);
1118 static noinline
void async_cow_free(struct btrfs_work
*work
)
1120 struct async_cow
*async_cow
;
1121 async_cow
= container_of(work
, struct async_cow
, work
);
1122 if (async_cow
->inode
)
1123 btrfs_add_delayed_iput(async_cow
->inode
);
1127 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1128 u64 start
, u64 end
, int *page_started
,
1129 unsigned long *nr_written
)
1131 struct async_cow
*async_cow
;
1132 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1133 unsigned long nr_pages
;
1135 int limit
= 10 * 1024 * 1024;
1137 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1138 1, 0, NULL
, GFP_NOFS
);
1139 while (start
< end
) {
1140 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1141 BUG_ON(!async_cow
); /* -ENOMEM */
1142 async_cow
->inode
= igrab(inode
);
1143 async_cow
->root
= root
;
1144 async_cow
->locked_page
= locked_page
;
1145 async_cow
->start
= start
;
1147 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
&&
1148 !btrfs_test_opt(root
, FORCE_COMPRESS
))
1151 cur_end
= min(end
, start
+ 512 * 1024 - 1);
1153 async_cow
->end
= cur_end
;
1154 INIT_LIST_HEAD(&async_cow
->extents
);
1156 btrfs_init_work(&async_cow
->work
,
1157 btrfs_delalloc_helper
,
1158 async_cow_start
, async_cow_submit
,
1161 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
1163 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
1165 btrfs_queue_work(root
->fs_info
->delalloc_workers
,
1168 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
1169 wait_event(root
->fs_info
->async_submit_wait
,
1170 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
1174 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
1175 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
1176 wait_event(root
->fs_info
->async_submit_wait
,
1177 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
1181 *nr_written
+= nr_pages
;
1182 start
= cur_end
+ 1;
1188 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1189 u64 bytenr
, u64 num_bytes
)
1192 struct btrfs_ordered_sum
*sums
;
1195 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1196 bytenr
+ num_bytes
- 1, &list
, 0);
1197 if (ret
== 0 && list_empty(&list
))
1200 while (!list_empty(&list
)) {
1201 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1202 list_del(&sums
->list
);
1211 * when nowcow writeback call back. This checks for snapshots or COW copies
1212 * of the extents that exist in the file, and COWs the file as required.
1214 * If no cow copies or snapshots exist, we write directly to the existing
1217 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1218 struct page
*locked_page
,
1219 u64 start
, u64 end
, int *page_started
, int force
,
1220 unsigned long *nr_written
)
1222 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1223 struct btrfs_trans_handle
*trans
;
1224 struct extent_buffer
*leaf
;
1225 struct btrfs_path
*path
;
1226 struct btrfs_file_extent_item
*fi
;
1227 struct btrfs_key found_key
;
1242 u64 ino
= btrfs_ino(inode
);
1244 path
= btrfs_alloc_path();
1246 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1247 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1248 EXTENT_DO_ACCOUNTING
|
1249 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1251 PAGE_SET_WRITEBACK
|
1252 PAGE_END_WRITEBACK
);
1256 nolock
= btrfs_is_free_space_inode(inode
);
1259 trans
= btrfs_join_transaction_nolock(root
);
1261 trans
= btrfs_join_transaction(root
);
1263 if (IS_ERR(trans
)) {
1264 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1265 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1266 EXTENT_DO_ACCOUNTING
|
1267 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1269 PAGE_SET_WRITEBACK
|
1270 PAGE_END_WRITEBACK
);
1271 btrfs_free_path(path
);
1272 return PTR_ERR(trans
);
1275 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1277 cow_start
= (u64
)-1;
1280 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1284 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1285 leaf
= path
->nodes
[0];
1286 btrfs_item_key_to_cpu(leaf
, &found_key
,
1287 path
->slots
[0] - 1);
1288 if (found_key
.objectid
== ino
&&
1289 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1294 leaf
= path
->nodes
[0];
1295 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1296 ret
= btrfs_next_leaf(root
, path
);
1298 if (cow_start
!= (u64
)-1)
1299 cur_offset
= cow_start
;
1304 leaf
= path
->nodes
[0];
1310 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1312 if (found_key
.objectid
> ino
)
1314 if (WARN_ON_ONCE(found_key
.objectid
< ino
) ||
1315 found_key
.type
< BTRFS_EXTENT_DATA_KEY
) {
1319 if (found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1320 found_key
.offset
> end
)
1323 if (found_key
.offset
> cur_offset
) {
1324 extent_end
= found_key
.offset
;
1329 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1330 struct btrfs_file_extent_item
);
1331 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1333 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
1334 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1335 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1336 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1337 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1338 extent_end
= found_key
.offset
+
1339 btrfs_file_extent_num_bytes(leaf
, fi
);
1341 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1342 if (extent_end
<= start
) {
1346 if (disk_bytenr
== 0)
1348 if (btrfs_file_extent_compression(leaf
, fi
) ||
1349 btrfs_file_extent_encryption(leaf
, fi
) ||
1350 btrfs_file_extent_other_encoding(leaf
, fi
))
1352 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1354 if (btrfs_extent_readonly(root
, disk_bytenr
))
1356 ret
= btrfs_cross_ref_exist(trans
, root
, ino
,
1358 extent_offset
, disk_bytenr
);
1361 * ret could be -EIO if the above fails to read
1365 if (cow_start
!= (u64
)-1)
1366 cur_offset
= cow_start
;
1370 WARN_ON_ONCE(nolock
);
1373 disk_bytenr
+= extent_offset
;
1374 disk_bytenr
+= cur_offset
- found_key
.offset
;
1375 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1377 * if there are pending snapshots for this root,
1378 * we fall into common COW way.
1381 err
= btrfs_start_write_no_snapshoting(root
);
1386 * force cow if csum exists in the range.
1387 * this ensure that csum for a given extent are
1388 * either valid or do not exist.
1390 ret
= csum_exist_in_range(root
, disk_bytenr
, num_bytes
);
1393 * ret could be -EIO if the above fails to read
1397 if (cow_start
!= (u64
)-1)
1398 cur_offset
= cow_start
;
1401 WARN_ON_ONCE(nolock
);
1405 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1406 extent_end
= found_key
.offset
+
1407 btrfs_file_extent_inline_len(leaf
,
1408 path
->slots
[0], fi
);
1409 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1414 if (extent_end
<= start
) {
1416 if (!nolock
&& nocow
)
1417 btrfs_end_write_no_snapshoting(root
);
1421 if (cow_start
== (u64
)-1)
1422 cow_start
= cur_offset
;
1423 cur_offset
= extent_end
;
1424 if (cur_offset
> end
)
1430 btrfs_release_path(path
);
1431 if (cow_start
!= (u64
)-1) {
1432 ret
= cow_file_range(inode
, locked_page
,
1433 cow_start
, found_key
.offset
- 1,
1434 page_started
, nr_written
, 1);
1436 if (!nolock
&& nocow
)
1437 btrfs_end_write_no_snapshoting(root
);
1440 cow_start
= (u64
)-1;
1443 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1444 struct extent_map
*em
;
1445 struct extent_map_tree
*em_tree
;
1446 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1447 em
= alloc_extent_map();
1448 BUG_ON(!em
); /* -ENOMEM */
1449 em
->start
= cur_offset
;
1450 em
->orig_start
= found_key
.offset
- extent_offset
;
1451 em
->len
= num_bytes
;
1452 em
->block_len
= num_bytes
;
1453 em
->block_start
= disk_bytenr
;
1454 em
->orig_block_len
= disk_num_bytes
;
1455 em
->ram_bytes
= ram_bytes
;
1456 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1457 em
->mod_start
= em
->start
;
1458 em
->mod_len
= em
->len
;
1459 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1460 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
1461 em
->generation
= -1;
1463 write_lock(&em_tree
->lock
);
1464 ret
= add_extent_mapping(em_tree
, em
, 1);
1465 write_unlock(&em_tree
->lock
);
1466 if (ret
!= -EEXIST
) {
1467 free_extent_map(em
);
1470 btrfs_drop_extent_cache(inode
, em
->start
,
1471 em
->start
+ em
->len
- 1, 0);
1473 type
= BTRFS_ORDERED_PREALLOC
;
1475 type
= BTRFS_ORDERED_NOCOW
;
1478 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1479 num_bytes
, num_bytes
, type
);
1480 BUG_ON(ret
); /* -ENOMEM */
1482 if (root
->root_key
.objectid
==
1483 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1484 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1487 if (!nolock
&& nocow
)
1488 btrfs_end_write_no_snapshoting(root
);
1493 extent_clear_unlock_delalloc(inode
, cur_offset
,
1494 cur_offset
+ num_bytes
- 1,
1495 locked_page
, EXTENT_LOCKED
|
1496 EXTENT_DELALLOC
, PAGE_UNLOCK
|
1498 if (!nolock
&& nocow
)
1499 btrfs_end_write_no_snapshoting(root
);
1500 cur_offset
= extent_end
;
1501 if (cur_offset
> end
)
1504 btrfs_release_path(path
);
1506 if (cur_offset
<= end
&& cow_start
== (u64
)-1) {
1507 cow_start
= cur_offset
;
1511 if (cow_start
!= (u64
)-1) {
1512 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1513 page_started
, nr_written
, 1);
1519 err
= btrfs_end_transaction(trans
, root
);
1523 if (ret
&& cur_offset
< end
)
1524 extent_clear_unlock_delalloc(inode
, cur_offset
, end
,
1525 locked_page
, EXTENT_LOCKED
|
1526 EXTENT_DELALLOC
| EXTENT_DEFRAG
|
1527 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
1529 PAGE_SET_WRITEBACK
|
1530 PAGE_END_WRITEBACK
);
1531 btrfs_free_path(path
);
1535 static inline int need_force_cow(struct inode
*inode
, u64 start
, u64 end
)
1538 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
1539 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
))
1543 * @defrag_bytes is a hint value, no spinlock held here,
1544 * if is not zero, it means the file is defragging.
1545 * Force cow if given extent needs to be defragged.
1547 if (BTRFS_I(inode
)->defrag_bytes
&&
1548 test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1549 EXTENT_DEFRAG
, 0, NULL
))
1556 * extent_io.c call back to do delayed allocation processing
1558 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1559 u64 start
, u64 end
, int *page_started
,
1560 unsigned long *nr_written
)
1563 int force_cow
= need_force_cow(inode
, start
, end
);
1565 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
&& !force_cow
) {
1566 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1567 page_started
, 1, nr_written
);
1568 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
&& !force_cow
) {
1569 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1570 page_started
, 0, nr_written
);
1571 } else if (!inode_need_compress(inode
)) {
1572 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1573 page_started
, nr_written
, 1);
1575 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1576 &BTRFS_I(inode
)->runtime_flags
);
1577 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1578 page_started
, nr_written
);
1583 static void btrfs_split_extent_hook(struct inode
*inode
,
1584 struct extent_state
*orig
, u64 split
)
1588 /* not delalloc, ignore it */
1589 if (!(orig
->state
& EXTENT_DELALLOC
))
1592 size
= orig
->end
- orig
->start
+ 1;
1593 if (size
> BTRFS_MAX_EXTENT_SIZE
) {
1598 * See the explanation in btrfs_merge_extent_hook, the same
1599 * applies here, just in reverse.
1601 new_size
= orig
->end
- split
+ 1;
1602 num_extents
= div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1603 BTRFS_MAX_EXTENT_SIZE
);
1604 new_size
= split
- orig
->start
;
1605 num_extents
+= div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1606 BTRFS_MAX_EXTENT_SIZE
);
1607 if (div64_u64(size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1608 BTRFS_MAX_EXTENT_SIZE
) >= num_extents
)
1612 spin_lock(&BTRFS_I(inode
)->lock
);
1613 BTRFS_I(inode
)->outstanding_extents
++;
1614 spin_unlock(&BTRFS_I(inode
)->lock
);
1618 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1619 * extents so we can keep track of new extents that are just merged onto old
1620 * extents, such as when we are doing sequential writes, so we can properly
1621 * account for the metadata space we'll need.
1623 static void btrfs_merge_extent_hook(struct inode
*inode
,
1624 struct extent_state
*new,
1625 struct extent_state
*other
)
1627 u64 new_size
, old_size
;
1630 /* not delalloc, ignore it */
1631 if (!(other
->state
& EXTENT_DELALLOC
))
1634 if (new->start
> other
->start
)
1635 new_size
= new->end
- other
->start
+ 1;
1637 new_size
= other
->end
- new->start
+ 1;
1639 /* we're not bigger than the max, unreserve the space and go */
1640 if (new_size
<= BTRFS_MAX_EXTENT_SIZE
) {
1641 spin_lock(&BTRFS_I(inode
)->lock
);
1642 BTRFS_I(inode
)->outstanding_extents
--;
1643 spin_unlock(&BTRFS_I(inode
)->lock
);
1648 * We have to add up either side to figure out how many extents were
1649 * accounted for before we merged into one big extent. If the number of
1650 * extents we accounted for is <= the amount we need for the new range
1651 * then we can return, otherwise drop. Think of it like this
1655 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1656 * need 2 outstanding extents, on one side we have 1 and the other side
1657 * we have 1 so they are == and we can return. But in this case
1659 * [MAX_SIZE+4k][MAX_SIZE+4k]
1661 * Each range on their own accounts for 2 extents, but merged together
1662 * they are only 3 extents worth of accounting, so we need to drop in
1665 old_size
= other
->end
- other
->start
+ 1;
1666 num_extents
= div64_u64(old_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1667 BTRFS_MAX_EXTENT_SIZE
);
1668 old_size
= new->end
- new->start
+ 1;
1669 num_extents
+= div64_u64(old_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1670 BTRFS_MAX_EXTENT_SIZE
);
1672 if (div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1673 BTRFS_MAX_EXTENT_SIZE
) >= num_extents
)
1676 spin_lock(&BTRFS_I(inode
)->lock
);
1677 BTRFS_I(inode
)->outstanding_extents
--;
1678 spin_unlock(&BTRFS_I(inode
)->lock
);
1681 static void btrfs_add_delalloc_inodes(struct btrfs_root
*root
,
1682 struct inode
*inode
)
1684 spin_lock(&root
->delalloc_lock
);
1685 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1686 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1687 &root
->delalloc_inodes
);
1688 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1689 &BTRFS_I(inode
)->runtime_flags
);
1690 root
->nr_delalloc_inodes
++;
1691 if (root
->nr_delalloc_inodes
== 1) {
1692 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1693 BUG_ON(!list_empty(&root
->delalloc_root
));
1694 list_add_tail(&root
->delalloc_root
,
1695 &root
->fs_info
->delalloc_roots
);
1696 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1699 spin_unlock(&root
->delalloc_lock
);
1702 static void btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1703 struct inode
*inode
)
1705 spin_lock(&root
->delalloc_lock
);
1706 if (!list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1707 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1708 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1709 &BTRFS_I(inode
)->runtime_flags
);
1710 root
->nr_delalloc_inodes
--;
1711 if (!root
->nr_delalloc_inodes
) {
1712 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1713 BUG_ON(list_empty(&root
->delalloc_root
));
1714 list_del_init(&root
->delalloc_root
);
1715 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1718 spin_unlock(&root
->delalloc_lock
);
1722 * extent_io.c set_bit_hook, used to track delayed allocation
1723 * bytes in this file, and to maintain the list of inodes that
1724 * have pending delalloc work to be done.
1726 static void btrfs_set_bit_hook(struct inode
*inode
,
1727 struct extent_state
*state
, unsigned *bits
)
1730 if ((*bits
& EXTENT_DEFRAG
) && !(*bits
& EXTENT_DELALLOC
))
1733 * set_bit and clear bit hooks normally require _irqsave/restore
1734 * but in this case, we are only testing for the DELALLOC
1735 * bit, which is only set or cleared with irqs on
1737 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1738 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1739 u64 len
= state
->end
+ 1 - state
->start
;
1740 bool do_list
= !btrfs_is_free_space_inode(inode
);
1742 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1743 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1745 spin_lock(&BTRFS_I(inode
)->lock
);
1746 BTRFS_I(inode
)->outstanding_extents
++;
1747 spin_unlock(&BTRFS_I(inode
)->lock
);
1750 /* For sanity tests */
1751 if (btrfs_test_is_dummy_root(root
))
1754 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, len
,
1755 root
->fs_info
->delalloc_batch
);
1756 spin_lock(&BTRFS_I(inode
)->lock
);
1757 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1758 if (*bits
& EXTENT_DEFRAG
)
1759 BTRFS_I(inode
)->defrag_bytes
+= len
;
1760 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1761 &BTRFS_I(inode
)->runtime_flags
))
1762 btrfs_add_delalloc_inodes(root
, inode
);
1763 spin_unlock(&BTRFS_I(inode
)->lock
);
1768 * extent_io.c clear_bit_hook, see set_bit_hook for why
1770 static void btrfs_clear_bit_hook(struct inode
*inode
,
1771 struct extent_state
*state
,
1774 u64 len
= state
->end
+ 1 - state
->start
;
1775 u64 num_extents
= div64_u64(len
+ BTRFS_MAX_EXTENT_SIZE
-1,
1776 BTRFS_MAX_EXTENT_SIZE
);
1778 spin_lock(&BTRFS_I(inode
)->lock
);
1779 if ((state
->state
& EXTENT_DEFRAG
) && (*bits
& EXTENT_DEFRAG
))
1780 BTRFS_I(inode
)->defrag_bytes
-= len
;
1781 spin_unlock(&BTRFS_I(inode
)->lock
);
1784 * set_bit and clear bit hooks normally require _irqsave/restore
1785 * but in this case, we are only testing for the DELALLOC
1786 * bit, which is only set or cleared with irqs on
1788 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1789 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1790 bool do_list
= !btrfs_is_free_space_inode(inode
);
1792 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1793 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1794 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1795 spin_lock(&BTRFS_I(inode
)->lock
);
1796 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
1797 spin_unlock(&BTRFS_I(inode
)->lock
);
1801 * We don't reserve metadata space for space cache inodes so we
1802 * don't need to call dellalloc_release_metadata if there is an
1805 if (*bits
& EXTENT_DO_ACCOUNTING
&&
1806 root
!= root
->fs_info
->tree_root
)
1807 btrfs_delalloc_release_metadata(inode
, len
);
1809 /* For sanity tests. */
1810 if (btrfs_test_is_dummy_root(root
))
1813 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1814 && do_list
&& !(state
->state
& EXTENT_NORESERVE
))
1815 btrfs_free_reserved_data_space_noquota(inode
,
1818 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, -len
,
1819 root
->fs_info
->delalloc_batch
);
1820 spin_lock(&BTRFS_I(inode
)->lock
);
1821 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1822 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1823 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1824 &BTRFS_I(inode
)->runtime_flags
))
1825 btrfs_del_delalloc_inode(root
, inode
);
1826 spin_unlock(&BTRFS_I(inode
)->lock
);
1831 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1832 * we don't create bios that span stripes or chunks
1834 int btrfs_merge_bio_hook(int rw
, struct page
*page
, unsigned long offset
,
1835 size_t size
, struct bio
*bio
,
1836 unsigned long bio_flags
)
1838 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1839 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
1844 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1847 length
= bio
->bi_iter
.bi_size
;
1848 map_length
= length
;
1849 ret
= btrfs_map_block(root
->fs_info
, rw
, logical
,
1850 &map_length
, NULL
, 0);
1851 /* Will always return 0 with map_multi == NULL */
1853 if (map_length
< length
+ size
)
1859 * in order to insert checksums into the metadata in large chunks,
1860 * we wait until bio submission time. All the pages in the bio are
1861 * checksummed and sums are attached onto the ordered extent record.
1863 * At IO completion time the cums attached on the ordered extent record
1864 * are inserted into the btree
1866 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1867 struct bio
*bio
, int mirror_num
,
1868 unsigned long bio_flags
,
1871 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1874 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1875 BUG_ON(ret
); /* -ENOMEM */
1880 * in order to insert checksums into the metadata in large chunks,
1881 * we wait until bio submission time. All the pages in the bio are
1882 * checksummed and sums are attached onto the ordered extent record.
1884 * At IO completion time the cums attached on the ordered extent record
1885 * are inserted into the btree
1887 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1888 int mirror_num
, unsigned long bio_flags
,
1891 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1894 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1896 bio
->bi_error
= ret
;
1903 * extent_io.c submission hook. This does the right thing for csum calculation
1904 * on write, or reading the csums from the tree before a read
1906 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1907 int mirror_num
, unsigned long bio_flags
,
1910 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1911 enum btrfs_wq_endio_type metadata
= BTRFS_WQ_ENDIO_DATA
;
1914 int async
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
1916 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1918 if (btrfs_is_free_space_inode(inode
))
1919 metadata
= BTRFS_WQ_ENDIO_FREE_SPACE
;
1921 if (!(rw
& REQ_WRITE
)) {
1922 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, metadata
);
1926 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1927 ret
= btrfs_submit_compressed_read(inode
, bio
,
1931 } else if (!skip_sum
) {
1932 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1937 } else if (async
&& !skip_sum
) {
1938 /* csum items have already been cloned */
1939 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1941 /* we're doing a write, do the async checksumming */
1942 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1943 inode
, rw
, bio
, mirror_num
,
1944 bio_flags
, bio_offset
,
1945 __btrfs_submit_bio_start
,
1946 __btrfs_submit_bio_done
);
1948 } else if (!skip_sum
) {
1949 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1955 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1959 bio
->bi_error
= ret
;
1966 * given a list of ordered sums record them in the inode. This happens
1967 * at IO completion time based on sums calculated at bio submission time.
1969 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1970 struct inode
*inode
, u64 file_offset
,
1971 struct list_head
*list
)
1973 struct btrfs_ordered_sum
*sum
;
1975 list_for_each_entry(sum
, list
, list
) {
1976 trans
->adding_csums
= 1;
1977 btrfs_csum_file_blocks(trans
,
1978 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1979 trans
->adding_csums
= 0;
1984 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1985 struct extent_state
**cached_state
)
1987 WARN_ON((end
& (PAGE_CACHE_SIZE
- 1)) == 0);
1988 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1989 cached_state
, GFP_NOFS
);
1992 /* see btrfs_writepage_start_hook for details on why this is required */
1993 struct btrfs_writepage_fixup
{
1995 struct btrfs_work work
;
1998 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
2000 struct btrfs_writepage_fixup
*fixup
;
2001 struct btrfs_ordered_extent
*ordered
;
2002 struct extent_state
*cached_state
= NULL
;
2004 struct inode
*inode
;
2009 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
2013 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
2014 ClearPageChecked(page
);
2018 inode
= page
->mapping
->host
;
2019 page_start
= page_offset(page
);
2020 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
2022 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
2025 /* already ordered? We're done */
2026 if (PagePrivate2(page
))
2029 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
2031 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
2032 page_end
, &cached_state
, GFP_NOFS
);
2034 btrfs_start_ordered_extent(inode
, ordered
, 1);
2035 btrfs_put_ordered_extent(ordered
);
2039 ret
= btrfs_delalloc_reserve_space(inode
, page_start
,
2042 mapping_set_error(page
->mapping
, ret
);
2043 end_extent_writepage(page
, ret
, page_start
, page_end
);
2044 ClearPageChecked(page
);
2048 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
2051 mapping_set_error(page
->mapping
, ret
);
2052 end_extent_writepage(page
, ret
, page_start
, page_end
);
2053 ClearPageChecked(page
);
2057 ClearPageChecked(page
);
2058 set_page_dirty(page
);
2060 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2061 &cached_state
, GFP_NOFS
);
2064 page_cache_release(page
);
2069 * There are a few paths in the higher layers of the kernel that directly
2070 * set the page dirty bit without asking the filesystem if it is a
2071 * good idea. This causes problems because we want to make sure COW
2072 * properly happens and the data=ordered rules are followed.
2074 * In our case any range that doesn't have the ORDERED bit set
2075 * hasn't been properly setup for IO. We kick off an async process
2076 * to fix it up. The async helper will wait for ordered extents, set
2077 * the delalloc bit and make it safe to write the page.
2079 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
2081 struct inode
*inode
= page
->mapping
->host
;
2082 struct btrfs_writepage_fixup
*fixup
;
2083 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2085 /* this page is properly in the ordered list */
2086 if (TestClearPagePrivate2(page
))
2089 if (PageChecked(page
))
2092 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
2096 SetPageChecked(page
);
2097 page_cache_get(page
);
2098 btrfs_init_work(&fixup
->work
, btrfs_fixup_helper
,
2099 btrfs_writepage_fixup_worker
, NULL
, NULL
);
2101 btrfs_queue_work(root
->fs_info
->fixup_workers
, &fixup
->work
);
2105 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
2106 struct inode
*inode
, u64 file_pos
,
2107 u64 disk_bytenr
, u64 disk_num_bytes
,
2108 u64 num_bytes
, u64 ram_bytes
,
2109 u8 compression
, u8 encryption
,
2110 u16 other_encoding
, int extent_type
)
2112 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2113 struct btrfs_file_extent_item
*fi
;
2114 struct btrfs_path
*path
;
2115 struct extent_buffer
*leaf
;
2116 struct btrfs_key ins
;
2117 int extent_inserted
= 0;
2120 path
= btrfs_alloc_path();
2125 * we may be replacing one extent in the tree with another.
2126 * The new extent is pinned in the extent map, and we don't want
2127 * to drop it from the cache until it is completely in the btree.
2129 * So, tell btrfs_drop_extents to leave this extent in the cache.
2130 * the caller is expected to unpin it and allow it to be merged
2133 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, file_pos
,
2134 file_pos
+ num_bytes
, NULL
, 0,
2135 1, sizeof(*fi
), &extent_inserted
);
2139 if (!extent_inserted
) {
2140 ins
.objectid
= btrfs_ino(inode
);
2141 ins
.offset
= file_pos
;
2142 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
2144 path
->leave_spinning
= 1;
2145 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
2150 leaf
= path
->nodes
[0];
2151 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2152 struct btrfs_file_extent_item
);
2153 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
2154 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
2155 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
2156 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
2157 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2158 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2159 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
2160 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
2161 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
2162 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
2164 btrfs_mark_buffer_dirty(leaf
);
2165 btrfs_release_path(path
);
2167 inode_add_bytes(inode
, num_bytes
);
2169 ins
.objectid
= disk_bytenr
;
2170 ins
.offset
= disk_num_bytes
;
2171 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2172 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
2173 root
->root_key
.objectid
,
2174 btrfs_ino(inode
), file_pos
,
2177 * Release the reserved range from inode dirty range map, as it is
2178 * already moved into delayed_ref_head
2180 btrfs_qgroup_release_data(inode
, file_pos
, ram_bytes
);
2182 btrfs_free_path(path
);
2187 /* snapshot-aware defrag */
2188 struct sa_defrag_extent_backref
{
2189 struct rb_node node
;
2190 struct old_sa_defrag_extent
*old
;
2199 struct old_sa_defrag_extent
{
2200 struct list_head list
;
2201 struct new_sa_defrag_extent
*new;
2210 struct new_sa_defrag_extent
{
2211 struct rb_root root
;
2212 struct list_head head
;
2213 struct btrfs_path
*path
;
2214 struct inode
*inode
;
2222 static int backref_comp(struct sa_defrag_extent_backref
*b1
,
2223 struct sa_defrag_extent_backref
*b2
)
2225 if (b1
->root_id
< b2
->root_id
)
2227 else if (b1
->root_id
> b2
->root_id
)
2230 if (b1
->inum
< b2
->inum
)
2232 else if (b1
->inum
> b2
->inum
)
2235 if (b1
->file_pos
< b2
->file_pos
)
2237 else if (b1
->file_pos
> b2
->file_pos
)
2241 * [------------------------------] ===> (a range of space)
2242 * |<--->| |<---->| =============> (fs/file tree A)
2243 * |<---------------------------->| ===> (fs/file tree B)
2245 * A range of space can refer to two file extents in one tree while
2246 * refer to only one file extent in another tree.
2248 * So we may process a disk offset more than one time(two extents in A)
2249 * and locate at the same extent(one extent in B), then insert two same
2250 * backrefs(both refer to the extent in B).
2255 static void backref_insert(struct rb_root
*root
,
2256 struct sa_defrag_extent_backref
*backref
)
2258 struct rb_node
**p
= &root
->rb_node
;
2259 struct rb_node
*parent
= NULL
;
2260 struct sa_defrag_extent_backref
*entry
;
2265 entry
= rb_entry(parent
, struct sa_defrag_extent_backref
, node
);
2267 ret
= backref_comp(backref
, entry
);
2271 p
= &(*p
)->rb_right
;
2274 rb_link_node(&backref
->node
, parent
, p
);
2275 rb_insert_color(&backref
->node
, root
);
2279 * Note the backref might has changed, and in this case we just return 0.
2281 static noinline
int record_one_backref(u64 inum
, u64 offset
, u64 root_id
,
2284 struct btrfs_file_extent_item
*extent
;
2285 struct btrfs_fs_info
*fs_info
;
2286 struct old_sa_defrag_extent
*old
= ctx
;
2287 struct new_sa_defrag_extent
*new = old
->new;
2288 struct btrfs_path
*path
= new->path
;
2289 struct btrfs_key key
;
2290 struct btrfs_root
*root
;
2291 struct sa_defrag_extent_backref
*backref
;
2292 struct extent_buffer
*leaf
;
2293 struct inode
*inode
= new->inode
;
2299 if (BTRFS_I(inode
)->root
->root_key
.objectid
== root_id
&&
2300 inum
== btrfs_ino(inode
))
2303 key
.objectid
= root_id
;
2304 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2305 key
.offset
= (u64
)-1;
2307 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2308 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2310 if (PTR_ERR(root
) == -ENOENT
)
2313 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2314 inum
, offset
, root_id
);
2315 return PTR_ERR(root
);
2318 key
.objectid
= inum
;
2319 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2320 if (offset
> (u64
)-1 << 32)
2323 key
.offset
= offset
;
2325 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2326 if (WARN_ON(ret
< 0))
2333 leaf
= path
->nodes
[0];
2334 slot
= path
->slots
[0];
2336 if (slot
>= btrfs_header_nritems(leaf
)) {
2337 ret
= btrfs_next_leaf(root
, path
);
2340 } else if (ret
> 0) {
2349 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2351 if (key
.objectid
> inum
)
2354 if (key
.objectid
< inum
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2357 extent
= btrfs_item_ptr(leaf
, slot
,
2358 struct btrfs_file_extent_item
);
2360 if (btrfs_file_extent_disk_bytenr(leaf
, extent
) != old
->bytenr
)
2364 * 'offset' refers to the exact key.offset,
2365 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2366 * (key.offset - extent_offset).
2368 if (key
.offset
!= offset
)
2371 extent_offset
= btrfs_file_extent_offset(leaf
, extent
);
2372 num_bytes
= btrfs_file_extent_num_bytes(leaf
, extent
);
2374 if (extent_offset
>= old
->extent_offset
+ old
->offset
+
2375 old
->len
|| extent_offset
+ num_bytes
<=
2376 old
->extent_offset
+ old
->offset
)
2381 backref
= kmalloc(sizeof(*backref
), GFP_NOFS
);
2387 backref
->root_id
= root_id
;
2388 backref
->inum
= inum
;
2389 backref
->file_pos
= offset
;
2390 backref
->num_bytes
= num_bytes
;
2391 backref
->extent_offset
= extent_offset
;
2392 backref
->generation
= btrfs_file_extent_generation(leaf
, extent
);
2394 backref_insert(&new->root
, backref
);
2397 btrfs_release_path(path
);
2402 static noinline
bool record_extent_backrefs(struct btrfs_path
*path
,
2403 struct new_sa_defrag_extent
*new)
2405 struct btrfs_fs_info
*fs_info
= BTRFS_I(new->inode
)->root
->fs_info
;
2406 struct old_sa_defrag_extent
*old
, *tmp
;
2411 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2412 ret
= iterate_inodes_from_logical(old
->bytenr
+
2413 old
->extent_offset
, fs_info
,
2414 path
, record_one_backref
,
2416 if (ret
< 0 && ret
!= -ENOENT
)
2419 /* no backref to be processed for this extent */
2421 list_del(&old
->list
);
2426 if (list_empty(&new->head
))
2432 static int relink_is_mergable(struct extent_buffer
*leaf
,
2433 struct btrfs_file_extent_item
*fi
,
2434 struct new_sa_defrag_extent
*new)
2436 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) != new->bytenr
)
2439 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2442 if (btrfs_file_extent_compression(leaf
, fi
) != new->compress_type
)
2445 if (btrfs_file_extent_encryption(leaf
, fi
) ||
2446 btrfs_file_extent_other_encoding(leaf
, fi
))
2453 * Note the backref might has changed, and in this case we just return 0.
2455 static noinline
int relink_extent_backref(struct btrfs_path
*path
,
2456 struct sa_defrag_extent_backref
*prev
,
2457 struct sa_defrag_extent_backref
*backref
)
2459 struct btrfs_file_extent_item
*extent
;
2460 struct btrfs_file_extent_item
*item
;
2461 struct btrfs_ordered_extent
*ordered
;
2462 struct btrfs_trans_handle
*trans
;
2463 struct btrfs_fs_info
*fs_info
;
2464 struct btrfs_root
*root
;
2465 struct btrfs_key key
;
2466 struct extent_buffer
*leaf
;
2467 struct old_sa_defrag_extent
*old
= backref
->old
;
2468 struct new_sa_defrag_extent
*new = old
->new;
2469 struct inode
*src_inode
= new->inode
;
2470 struct inode
*inode
;
2471 struct extent_state
*cached
= NULL
;
2480 if (prev
&& prev
->root_id
== backref
->root_id
&&
2481 prev
->inum
== backref
->inum
&&
2482 prev
->file_pos
+ prev
->num_bytes
== backref
->file_pos
)
2485 /* step 1: get root */
2486 key
.objectid
= backref
->root_id
;
2487 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2488 key
.offset
= (u64
)-1;
2490 fs_info
= BTRFS_I(src_inode
)->root
->fs_info
;
2491 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
2493 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2495 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2496 if (PTR_ERR(root
) == -ENOENT
)
2498 return PTR_ERR(root
);
2501 if (btrfs_root_readonly(root
)) {
2502 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2506 /* step 2: get inode */
2507 key
.objectid
= backref
->inum
;
2508 key
.type
= BTRFS_INODE_ITEM_KEY
;
2511 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
2512 if (IS_ERR(inode
)) {
2513 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2517 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2519 /* step 3: relink backref */
2520 lock_start
= backref
->file_pos
;
2521 lock_end
= backref
->file_pos
+ backref
->num_bytes
- 1;
2522 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2525 ordered
= btrfs_lookup_first_ordered_extent(inode
, lock_end
);
2527 btrfs_put_ordered_extent(ordered
);
2531 trans
= btrfs_join_transaction(root
);
2532 if (IS_ERR(trans
)) {
2533 ret
= PTR_ERR(trans
);
2537 key
.objectid
= backref
->inum
;
2538 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2539 key
.offset
= backref
->file_pos
;
2541 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2544 } else if (ret
> 0) {
2549 extent
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2550 struct btrfs_file_extent_item
);
2552 if (btrfs_file_extent_generation(path
->nodes
[0], extent
) !=
2553 backref
->generation
)
2556 btrfs_release_path(path
);
2558 start
= backref
->file_pos
;
2559 if (backref
->extent_offset
< old
->extent_offset
+ old
->offset
)
2560 start
+= old
->extent_offset
+ old
->offset
-
2561 backref
->extent_offset
;
2563 len
= min(backref
->extent_offset
+ backref
->num_bytes
,
2564 old
->extent_offset
+ old
->offset
+ old
->len
);
2565 len
-= max(backref
->extent_offset
, old
->extent_offset
+ old
->offset
);
2567 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
2572 key
.objectid
= btrfs_ino(inode
);
2573 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2576 path
->leave_spinning
= 1;
2578 struct btrfs_file_extent_item
*fi
;
2580 struct btrfs_key found_key
;
2582 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2587 leaf
= path
->nodes
[0];
2588 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2590 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2591 struct btrfs_file_extent_item
);
2592 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
2594 if (extent_len
+ found_key
.offset
== start
&&
2595 relink_is_mergable(leaf
, fi
, new)) {
2596 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2598 btrfs_mark_buffer_dirty(leaf
);
2599 inode_add_bytes(inode
, len
);
2605 btrfs_release_path(path
);
2610 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2613 btrfs_abort_transaction(trans
, root
, ret
);
2617 leaf
= path
->nodes
[0];
2618 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2619 struct btrfs_file_extent_item
);
2620 btrfs_set_file_extent_disk_bytenr(leaf
, item
, new->bytenr
);
2621 btrfs_set_file_extent_disk_num_bytes(leaf
, item
, new->disk_len
);
2622 btrfs_set_file_extent_offset(leaf
, item
, start
- new->file_pos
);
2623 btrfs_set_file_extent_num_bytes(leaf
, item
, len
);
2624 btrfs_set_file_extent_ram_bytes(leaf
, item
, new->len
);
2625 btrfs_set_file_extent_generation(leaf
, item
, trans
->transid
);
2626 btrfs_set_file_extent_type(leaf
, item
, BTRFS_FILE_EXTENT_REG
);
2627 btrfs_set_file_extent_compression(leaf
, item
, new->compress_type
);
2628 btrfs_set_file_extent_encryption(leaf
, item
, 0);
2629 btrfs_set_file_extent_other_encoding(leaf
, item
, 0);
2631 btrfs_mark_buffer_dirty(leaf
);
2632 inode_add_bytes(inode
, len
);
2633 btrfs_release_path(path
);
2635 ret
= btrfs_inc_extent_ref(trans
, root
, new->bytenr
,
2637 backref
->root_id
, backref
->inum
,
2638 new->file_pos
); /* start - extent_offset */
2640 btrfs_abort_transaction(trans
, root
, ret
);
2646 btrfs_release_path(path
);
2647 path
->leave_spinning
= 0;
2648 btrfs_end_transaction(trans
, root
);
2650 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2656 static void free_sa_defrag_extent(struct new_sa_defrag_extent
*new)
2658 struct old_sa_defrag_extent
*old
, *tmp
;
2663 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2669 static void relink_file_extents(struct new_sa_defrag_extent
*new)
2671 struct btrfs_path
*path
;
2672 struct sa_defrag_extent_backref
*backref
;
2673 struct sa_defrag_extent_backref
*prev
= NULL
;
2674 struct inode
*inode
;
2675 struct btrfs_root
*root
;
2676 struct rb_node
*node
;
2680 root
= BTRFS_I(inode
)->root
;
2682 path
= btrfs_alloc_path();
2686 if (!record_extent_backrefs(path
, new)) {
2687 btrfs_free_path(path
);
2690 btrfs_release_path(path
);
2693 node
= rb_first(&new->root
);
2696 rb_erase(node
, &new->root
);
2698 backref
= rb_entry(node
, struct sa_defrag_extent_backref
, node
);
2700 ret
= relink_extent_backref(path
, prev
, backref
);
2713 btrfs_free_path(path
);
2715 free_sa_defrag_extent(new);
2717 atomic_dec(&root
->fs_info
->defrag_running
);
2718 wake_up(&root
->fs_info
->transaction_wait
);
2721 static struct new_sa_defrag_extent
*
2722 record_old_file_extents(struct inode
*inode
,
2723 struct btrfs_ordered_extent
*ordered
)
2725 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2726 struct btrfs_path
*path
;
2727 struct btrfs_key key
;
2728 struct old_sa_defrag_extent
*old
;
2729 struct new_sa_defrag_extent
*new;
2732 new = kmalloc(sizeof(*new), GFP_NOFS
);
2737 new->file_pos
= ordered
->file_offset
;
2738 new->len
= ordered
->len
;
2739 new->bytenr
= ordered
->start
;
2740 new->disk_len
= ordered
->disk_len
;
2741 new->compress_type
= ordered
->compress_type
;
2742 new->root
= RB_ROOT
;
2743 INIT_LIST_HEAD(&new->head
);
2745 path
= btrfs_alloc_path();
2749 key
.objectid
= btrfs_ino(inode
);
2750 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2751 key
.offset
= new->file_pos
;
2753 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2756 if (ret
> 0 && path
->slots
[0] > 0)
2759 /* find out all the old extents for the file range */
2761 struct btrfs_file_extent_item
*extent
;
2762 struct extent_buffer
*l
;
2771 slot
= path
->slots
[0];
2773 if (slot
>= btrfs_header_nritems(l
)) {
2774 ret
= btrfs_next_leaf(root
, path
);
2782 btrfs_item_key_to_cpu(l
, &key
, slot
);
2784 if (key
.objectid
!= btrfs_ino(inode
))
2786 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2788 if (key
.offset
>= new->file_pos
+ new->len
)
2791 extent
= btrfs_item_ptr(l
, slot
, struct btrfs_file_extent_item
);
2793 num_bytes
= btrfs_file_extent_num_bytes(l
, extent
);
2794 if (key
.offset
+ num_bytes
< new->file_pos
)
2797 disk_bytenr
= btrfs_file_extent_disk_bytenr(l
, extent
);
2801 extent_offset
= btrfs_file_extent_offset(l
, extent
);
2803 old
= kmalloc(sizeof(*old
), GFP_NOFS
);
2807 offset
= max(new->file_pos
, key
.offset
);
2808 end
= min(new->file_pos
+ new->len
, key
.offset
+ num_bytes
);
2810 old
->bytenr
= disk_bytenr
;
2811 old
->extent_offset
= extent_offset
;
2812 old
->offset
= offset
- key
.offset
;
2813 old
->len
= end
- offset
;
2816 list_add_tail(&old
->list
, &new->head
);
2822 btrfs_free_path(path
);
2823 atomic_inc(&root
->fs_info
->defrag_running
);
2828 btrfs_free_path(path
);
2830 free_sa_defrag_extent(new);
2834 static void btrfs_release_delalloc_bytes(struct btrfs_root
*root
,
2837 struct btrfs_block_group_cache
*cache
;
2839 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
2842 spin_lock(&cache
->lock
);
2843 cache
->delalloc_bytes
-= len
;
2844 spin_unlock(&cache
->lock
);
2846 btrfs_put_block_group(cache
);
2849 /* as ordered data IO finishes, this gets called so we can finish
2850 * an ordered extent if the range of bytes in the file it covers are
2853 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
2855 struct inode
*inode
= ordered_extent
->inode
;
2856 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2857 struct btrfs_trans_handle
*trans
= NULL
;
2858 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2859 struct extent_state
*cached_state
= NULL
;
2860 struct new_sa_defrag_extent
*new = NULL
;
2861 int compress_type
= 0;
2863 u64 logical_len
= ordered_extent
->len
;
2865 bool truncated
= false;
2867 nolock
= btrfs_is_free_space_inode(inode
);
2869 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
2874 btrfs_free_io_failure_record(inode
, ordered_extent
->file_offset
,
2875 ordered_extent
->file_offset
+
2876 ordered_extent
->len
- 1);
2878 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered_extent
->flags
)) {
2880 logical_len
= ordered_extent
->truncated_len
;
2881 /* Truncated the entire extent, don't bother adding */
2886 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
2887 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
2890 * For mwrite(mmap + memset to write) case, we still reserve
2891 * space for NOCOW range.
2892 * As NOCOW won't cause a new delayed ref, just free the space
2894 btrfs_qgroup_free_data(inode
, ordered_extent
->file_offset
,
2895 ordered_extent
->len
);
2896 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2898 trans
= btrfs_join_transaction_nolock(root
);
2900 trans
= btrfs_join_transaction(root
);
2901 if (IS_ERR(trans
)) {
2902 ret
= PTR_ERR(trans
);
2906 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2907 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2908 if (ret
) /* -ENOMEM or corruption */
2909 btrfs_abort_transaction(trans
, root
, ret
);
2913 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
2914 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2917 ret
= test_range_bit(io_tree
, ordered_extent
->file_offset
,
2918 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2919 EXTENT_DEFRAG
, 1, cached_state
);
2921 u64 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
2922 if (0 && last_snapshot
>= BTRFS_I(inode
)->generation
)
2923 /* the inode is shared */
2924 new = record_old_file_extents(inode
, ordered_extent
);
2926 clear_extent_bit(io_tree
, ordered_extent
->file_offset
,
2927 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2928 EXTENT_DEFRAG
, 0, 0, &cached_state
, GFP_NOFS
);
2932 trans
= btrfs_join_transaction_nolock(root
);
2934 trans
= btrfs_join_transaction(root
);
2935 if (IS_ERR(trans
)) {
2936 ret
= PTR_ERR(trans
);
2941 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2943 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
2944 compress_type
= ordered_extent
->compress_type
;
2945 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
2946 BUG_ON(compress_type
);
2947 ret
= btrfs_mark_extent_written(trans
, inode
,
2948 ordered_extent
->file_offset
,
2949 ordered_extent
->file_offset
+
2952 BUG_ON(root
== root
->fs_info
->tree_root
);
2953 ret
= insert_reserved_file_extent(trans
, inode
,
2954 ordered_extent
->file_offset
,
2955 ordered_extent
->start
,
2956 ordered_extent
->disk_len
,
2957 logical_len
, logical_len
,
2958 compress_type
, 0, 0,
2959 BTRFS_FILE_EXTENT_REG
);
2961 btrfs_release_delalloc_bytes(root
,
2962 ordered_extent
->start
,
2963 ordered_extent
->disk_len
);
2965 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
2966 ordered_extent
->file_offset
, ordered_extent
->len
,
2969 btrfs_abort_transaction(trans
, root
, ret
);
2973 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
2974 &ordered_extent
->list
);
2976 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2977 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2978 if (ret
) { /* -ENOMEM or corruption */
2979 btrfs_abort_transaction(trans
, root
, ret
);
2984 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
2985 ordered_extent
->file_offset
+
2986 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
2988 if (root
!= root
->fs_info
->tree_root
)
2989 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
2991 btrfs_end_transaction(trans
, root
);
2993 if (ret
|| truncated
) {
2997 start
= ordered_extent
->file_offset
+ logical_len
;
2999 start
= ordered_extent
->file_offset
;
3000 end
= ordered_extent
->file_offset
+ ordered_extent
->len
- 1;
3001 clear_extent_uptodate(io_tree
, start
, end
, NULL
, GFP_NOFS
);
3003 /* Drop the cache for the part of the extent we didn't write. */
3004 btrfs_drop_extent_cache(inode
, start
, end
, 0);
3007 * If the ordered extent had an IOERR or something else went
3008 * wrong we need to return the space for this ordered extent
3009 * back to the allocator. We only free the extent in the
3010 * truncated case if we didn't write out the extent at all.
3012 if ((ret
|| !logical_len
) &&
3013 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
3014 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
))
3015 btrfs_free_reserved_extent(root
, ordered_extent
->start
,
3016 ordered_extent
->disk_len
, 1);
3021 * This needs to be done to make sure anybody waiting knows we are done
3022 * updating everything for this ordered extent.
3024 btrfs_remove_ordered_extent(inode
, ordered_extent
);
3026 /* for snapshot-aware defrag */
3029 free_sa_defrag_extent(new);
3030 atomic_dec(&root
->fs_info
->defrag_running
);
3032 relink_file_extents(new);
3037 btrfs_put_ordered_extent(ordered_extent
);
3038 /* once for the tree */
3039 btrfs_put_ordered_extent(ordered_extent
);
3044 static void finish_ordered_fn(struct btrfs_work
*work
)
3046 struct btrfs_ordered_extent
*ordered_extent
;
3047 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
3048 btrfs_finish_ordered_io(ordered_extent
);
3051 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
3052 struct extent_state
*state
, int uptodate
)
3054 struct inode
*inode
= page
->mapping
->host
;
3055 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3056 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
3057 struct btrfs_workqueue
*wq
;
3058 btrfs_work_func_t func
;
3060 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
3062 ClearPagePrivate2(page
);
3063 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
3064 end
- start
+ 1, uptodate
))
3067 if (btrfs_is_free_space_inode(inode
)) {
3068 wq
= root
->fs_info
->endio_freespace_worker
;
3069 func
= btrfs_freespace_write_helper
;
3071 wq
= root
->fs_info
->endio_write_workers
;
3072 func
= btrfs_endio_write_helper
;
3075 btrfs_init_work(&ordered_extent
->work
, func
, finish_ordered_fn
, NULL
,
3077 btrfs_queue_work(wq
, &ordered_extent
->work
);
3082 static int __readpage_endio_check(struct inode
*inode
,
3083 struct btrfs_io_bio
*io_bio
,
3084 int icsum
, struct page
*page
,
3085 int pgoff
, u64 start
, size_t len
)
3091 csum_expected
= *(((u32
*)io_bio
->csum
) + icsum
);
3093 kaddr
= kmap_atomic(page
);
3094 csum
= btrfs_csum_data(kaddr
+ pgoff
, csum
, len
);
3095 btrfs_csum_final(csum
, (char *)&csum
);
3096 if (csum
!= csum_expected
)
3099 kunmap_atomic(kaddr
);
3102 btrfs_warn_rl(BTRFS_I(inode
)->root
->fs_info
,
3103 "csum failed ino %llu off %llu csum %u expected csum %u",
3104 btrfs_ino(inode
), start
, csum
, csum_expected
);
3105 memset(kaddr
+ pgoff
, 1, len
);
3106 flush_dcache_page(page
);
3107 kunmap_atomic(kaddr
);
3108 if (csum_expected
== 0)
3114 * when reads are done, we need to check csums to verify the data is correct
3115 * if there's a match, we allow the bio to finish. If not, the code in
3116 * extent_io.c will try to find good copies for us.
3118 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
3119 u64 phy_offset
, struct page
*page
,
3120 u64 start
, u64 end
, int mirror
)
3122 size_t offset
= start
- page_offset(page
);
3123 struct inode
*inode
= page
->mapping
->host
;
3124 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3125 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3127 if (PageChecked(page
)) {
3128 ClearPageChecked(page
);
3132 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
3135 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
3136 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
3137 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
3142 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
3143 return __readpage_endio_check(inode
, io_bio
, phy_offset
, page
, offset
,
3144 start
, (size_t)(end
- start
+ 1));
3147 struct delayed_iput
{
3148 struct list_head list
;
3149 struct inode
*inode
;
3152 /* JDM: If this is fs-wide, why can't we add a pointer to
3153 * btrfs_inode instead and avoid the allocation? */
3154 void btrfs_add_delayed_iput(struct inode
*inode
)
3156 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
3157 struct delayed_iput
*delayed
;
3159 if (atomic_add_unless(&inode
->i_count
, -1, 1))
3162 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
3163 delayed
->inode
= inode
;
3165 spin_lock(&fs_info
->delayed_iput_lock
);
3166 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
3167 spin_unlock(&fs_info
->delayed_iput_lock
);
3170 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
3173 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3174 struct delayed_iput
*delayed
;
3177 spin_lock(&fs_info
->delayed_iput_lock
);
3178 empty
= list_empty(&fs_info
->delayed_iputs
);
3179 spin_unlock(&fs_info
->delayed_iput_lock
);
3183 spin_lock(&fs_info
->delayed_iput_lock
);
3184 list_splice_init(&fs_info
->delayed_iputs
, &list
);
3185 spin_unlock(&fs_info
->delayed_iput_lock
);
3187 while (!list_empty(&list
)) {
3188 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
3189 list_del(&delayed
->list
);
3190 iput(delayed
->inode
);
3196 * This is called in transaction commit time. If there are no orphan
3197 * files in the subvolume, it removes orphan item and frees block_rsv
3200 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
3201 struct btrfs_root
*root
)
3203 struct btrfs_block_rsv
*block_rsv
;
3206 if (atomic_read(&root
->orphan_inodes
) ||
3207 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
3210 spin_lock(&root
->orphan_lock
);
3211 if (atomic_read(&root
->orphan_inodes
)) {
3212 spin_unlock(&root
->orphan_lock
);
3216 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
3217 spin_unlock(&root
->orphan_lock
);
3221 block_rsv
= root
->orphan_block_rsv
;
3222 root
->orphan_block_rsv
= NULL
;
3223 spin_unlock(&root
->orphan_lock
);
3225 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
) &&
3226 btrfs_root_refs(&root
->root_item
) > 0) {
3227 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
3228 root
->root_key
.objectid
);
3230 btrfs_abort_transaction(trans
, root
, ret
);
3232 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
,
3237 WARN_ON(block_rsv
->size
> 0);
3238 btrfs_free_block_rsv(root
, block_rsv
);
3243 * This creates an orphan entry for the given inode in case something goes
3244 * wrong in the middle of an unlink/truncate.
3246 * NOTE: caller of this function should reserve 5 units of metadata for
3249 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
3251 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3252 struct btrfs_block_rsv
*block_rsv
= NULL
;
3257 if (!root
->orphan_block_rsv
) {
3258 block_rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
3263 spin_lock(&root
->orphan_lock
);
3264 if (!root
->orphan_block_rsv
) {
3265 root
->orphan_block_rsv
= block_rsv
;
3266 } else if (block_rsv
) {
3267 btrfs_free_block_rsv(root
, block_rsv
);
3271 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3272 &BTRFS_I(inode
)->runtime_flags
)) {
3275 * For proper ENOSPC handling, we should do orphan
3276 * cleanup when mounting. But this introduces backward
3277 * compatibility issue.
3279 if (!xchg(&root
->orphan_item_inserted
, 1))
3285 atomic_inc(&root
->orphan_inodes
);
3288 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3289 &BTRFS_I(inode
)->runtime_flags
))
3291 spin_unlock(&root
->orphan_lock
);
3293 /* grab metadata reservation from transaction handle */
3295 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
3296 BUG_ON(ret
); /* -ENOSPC in reservation; Logic error? JDM */
3299 /* insert an orphan item to track this unlinked/truncated file */
3301 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
3303 atomic_dec(&root
->orphan_inodes
);
3305 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3306 &BTRFS_I(inode
)->runtime_flags
);
3307 btrfs_orphan_release_metadata(inode
);
3309 if (ret
!= -EEXIST
) {
3310 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3311 &BTRFS_I(inode
)->runtime_flags
);
3312 btrfs_abort_transaction(trans
, root
, ret
);
3319 /* insert an orphan item to track subvolume contains orphan files */
3321 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
3322 root
->root_key
.objectid
);
3323 if (ret
&& ret
!= -EEXIST
) {
3324 btrfs_abort_transaction(trans
, root
, ret
);
3332 * We have done the truncate/delete so we can go ahead and remove the orphan
3333 * item for this particular inode.
3335 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3336 struct inode
*inode
)
3338 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3339 int delete_item
= 0;
3340 int release_rsv
= 0;
3343 spin_lock(&root
->orphan_lock
);
3344 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3345 &BTRFS_I(inode
)->runtime_flags
))
3348 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3349 &BTRFS_I(inode
)->runtime_flags
))
3351 spin_unlock(&root
->orphan_lock
);
3354 atomic_dec(&root
->orphan_inodes
);
3356 ret
= btrfs_del_orphan_item(trans
, root
,
3361 btrfs_orphan_release_metadata(inode
);
3367 * this cleans up any orphans that may be left on the list from the last use
3370 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3372 struct btrfs_path
*path
;
3373 struct extent_buffer
*leaf
;
3374 struct btrfs_key key
, found_key
;
3375 struct btrfs_trans_handle
*trans
;
3376 struct inode
*inode
;
3377 u64 last_objectid
= 0;
3378 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
3380 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
3383 path
= btrfs_alloc_path();
3390 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3391 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
3392 key
.offset
= (u64
)-1;
3395 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3400 * if ret == 0 means we found what we were searching for, which
3401 * is weird, but possible, so only screw with path if we didn't
3402 * find the key and see if we have stuff that matches
3406 if (path
->slots
[0] == 0)
3411 /* pull out the item */
3412 leaf
= path
->nodes
[0];
3413 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3415 /* make sure the item matches what we want */
3416 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3418 if (found_key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
3421 /* release the path since we're done with it */
3422 btrfs_release_path(path
);
3425 * this is where we are basically btrfs_lookup, without the
3426 * crossing root thing. we store the inode number in the
3427 * offset of the orphan item.
3430 if (found_key
.offset
== last_objectid
) {
3431 btrfs_err(root
->fs_info
,
3432 "Error removing orphan entry, stopping orphan cleanup");
3437 last_objectid
= found_key
.offset
;
3439 found_key
.objectid
= found_key
.offset
;
3440 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3441 found_key
.offset
= 0;
3442 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
3443 ret
= PTR_ERR_OR_ZERO(inode
);
3444 if (ret
&& ret
!= -ESTALE
)
3447 if (ret
== -ESTALE
&& root
== root
->fs_info
->tree_root
) {
3448 struct btrfs_root
*dead_root
;
3449 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3450 int is_dead_root
= 0;
3453 * this is an orphan in the tree root. Currently these
3454 * could come from 2 sources:
3455 * a) a snapshot deletion in progress
3456 * b) a free space cache inode
3457 * We need to distinguish those two, as the snapshot
3458 * orphan must not get deleted.
3459 * find_dead_roots already ran before us, so if this
3460 * is a snapshot deletion, we should find the root
3461 * in the dead_roots list
3463 spin_lock(&fs_info
->trans_lock
);
3464 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
3466 if (dead_root
->root_key
.objectid
==
3467 found_key
.objectid
) {
3472 spin_unlock(&fs_info
->trans_lock
);
3474 /* prevent this orphan from being found again */
3475 key
.offset
= found_key
.objectid
- 1;
3480 * Inode is already gone but the orphan item is still there,
3481 * kill the orphan item.
3483 if (ret
== -ESTALE
) {
3484 trans
= btrfs_start_transaction(root
, 1);
3485 if (IS_ERR(trans
)) {
3486 ret
= PTR_ERR(trans
);
3489 btrfs_debug(root
->fs_info
, "auto deleting %Lu",
3490 found_key
.objectid
);
3491 ret
= btrfs_del_orphan_item(trans
, root
,
3492 found_key
.objectid
);
3493 btrfs_end_transaction(trans
, root
);
3500 * add this inode to the orphan list so btrfs_orphan_del does
3501 * the proper thing when we hit it
3503 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3504 &BTRFS_I(inode
)->runtime_flags
);
3505 atomic_inc(&root
->orphan_inodes
);
3507 /* if we have links, this was a truncate, lets do that */
3508 if (inode
->i_nlink
) {
3509 if (WARN_ON(!S_ISREG(inode
->i_mode
))) {
3515 /* 1 for the orphan item deletion. */
3516 trans
= btrfs_start_transaction(root
, 1);
3517 if (IS_ERR(trans
)) {
3519 ret
= PTR_ERR(trans
);
3522 ret
= btrfs_orphan_add(trans
, inode
);
3523 btrfs_end_transaction(trans
, root
);
3529 ret
= btrfs_truncate(inode
);
3531 btrfs_orphan_del(NULL
, inode
);
3536 /* this will do delete_inode and everything for us */
3541 /* release the path since we're done with it */
3542 btrfs_release_path(path
);
3544 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
3546 if (root
->orphan_block_rsv
)
3547 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
3550 if (root
->orphan_block_rsv
||
3551 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
)) {
3552 trans
= btrfs_join_transaction(root
);
3554 btrfs_end_transaction(trans
, root
);
3558 btrfs_debug(root
->fs_info
, "unlinked %d orphans", nr_unlink
);
3560 btrfs_debug(root
->fs_info
, "truncated %d orphans", nr_truncate
);
3564 btrfs_err(root
->fs_info
,
3565 "could not do orphan cleanup %d", ret
);
3566 btrfs_free_path(path
);
3571 * very simple check to peek ahead in the leaf looking for xattrs. If we
3572 * don't find any xattrs, we know there can't be any acls.
3574 * slot is the slot the inode is in, objectid is the objectid of the inode
3576 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3577 int slot
, u64 objectid
,
3578 int *first_xattr_slot
)
3580 u32 nritems
= btrfs_header_nritems(leaf
);
3581 struct btrfs_key found_key
;
3582 static u64 xattr_access
= 0;
3583 static u64 xattr_default
= 0;
3586 if (!xattr_access
) {
3587 xattr_access
= btrfs_name_hash(POSIX_ACL_XATTR_ACCESS
,
3588 strlen(POSIX_ACL_XATTR_ACCESS
));
3589 xattr_default
= btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT
,
3590 strlen(POSIX_ACL_XATTR_DEFAULT
));
3594 *first_xattr_slot
= -1;
3595 while (slot
< nritems
) {
3596 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3598 /* we found a different objectid, there must not be acls */
3599 if (found_key
.objectid
!= objectid
)
3602 /* we found an xattr, assume we've got an acl */
3603 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
3604 if (*first_xattr_slot
== -1)
3605 *first_xattr_slot
= slot
;
3606 if (found_key
.offset
== xattr_access
||
3607 found_key
.offset
== xattr_default
)
3612 * we found a key greater than an xattr key, there can't
3613 * be any acls later on
3615 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3622 * it goes inode, inode backrefs, xattrs, extents,
3623 * so if there are a ton of hard links to an inode there can
3624 * be a lot of backrefs. Don't waste time searching too hard,
3625 * this is just an optimization
3630 /* we hit the end of the leaf before we found an xattr or
3631 * something larger than an xattr. We have to assume the inode
3634 if (*first_xattr_slot
== -1)
3635 *first_xattr_slot
= slot
;
3640 * read an inode from the btree into the in-memory inode
3642 static void btrfs_read_locked_inode(struct inode
*inode
)
3644 struct btrfs_path
*path
;
3645 struct extent_buffer
*leaf
;
3646 struct btrfs_inode_item
*inode_item
;
3647 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3648 struct btrfs_key location
;
3653 bool filled
= false;
3654 int first_xattr_slot
;
3656 ret
= btrfs_fill_inode(inode
, &rdev
);
3660 path
= btrfs_alloc_path();
3664 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3666 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3670 leaf
= path
->nodes
[0];
3675 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3676 struct btrfs_inode_item
);
3677 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3678 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3679 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3680 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3681 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
3683 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->atime
);
3684 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->atime
);
3686 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->mtime
);
3687 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->mtime
);
3689 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->ctime
);
3690 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->ctime
);
3692 BTRFS_I(inode
)->i_otime
.tv_sec
=
3693 btrfs_timespec_sec(leaf
, &inode_item
->otime
);
3694 BTRFS_I(inode
)->i_otime
.tv_nsec
=
3695 btrfs_timespec_nsec(leaf
, &inode_item
->otime
);
3697 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3698 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3699 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3701 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
3702 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3704 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3706 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3707 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
3711 * If we were modified in the current generation and evicted from memory
3712 * and then re-read we need to do a full sync since we don't have any
3713 * idea about which extents were modified before we were evicted from
3716 * This is required for both inode re-read from disk and delayed inode
3717 * in delayed_nodes_tree.
3719 if (BTRFS_I(inode
)->last_trans
== root
->fs_info
->generation
)
3720 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3721 &BTRFS_I(inode
)->runtime_flags
);
3724 * We don't persist the id of the transaction where an unlink operation
3725 * against the inode was last made. So here we assume the inode might
3726 * have been evicted, and therefore the exact value of last_unlink_trans
3727 * lost, and set it to last_trans to avoid metadata inconsistencies
3728 * between the inode and its parent if the inode is fsync'ed and the log
3729 * replayed. For example, in the scenario:
3732 * ln mydir/foo mydir/bar
3735 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3736 * xfs_io -c fsync mydir/foo
3738 * mount fs, triggers fsync log replay
3740 * We must make sure that when we fsync our inode foo we also log its
3741 * parent inode, otherwise after log replay the parent still has the
3742 * dentry with the "bar" name but our inode foo has a link count of 1
3743 * and doesn't have an inode ref with the name "bar" anymore.
3745 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3746 * but it guarantees correctness at the expense of ocassional full
3747 * transaction commits on fsync if our inode is a directory, or if our
3748 * inode is not a directory, logging its parent unnecessarily.
3750 BTRFS_I(inode
)->last_unlink_trans
= BTRFS_I(inode
)->last_trans
;
3753 if (inode
->i_nlink
!= 1 ||
3754 path
->slots
[0] >= btrfs_header_nritems(leaf
))
3757 btrfs_item_key_to_cpu(leaf
, &location
, path
->slots
[0]);
3758 if (location
.objectid
!= btrfs_ino(inode
))
3761 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3762 if (location
.type
== BTRFS_INODE_REF_KEY
) {
3763 struct btrfs_inode_ref
*ref
;
3765 ref
= (struct btrfs_inode_ref
*)ptr
;
3766 BTRFS_I(inode
)->dir_index
= btrfs_inode_ref_index(leaf
, ref
);
3767 } else if (location
.type
== BTRFS_INODE_EXTREF_KEY
) {
3768 struct btrfs_inode_extref
*extref
;
3770 extref
= (struct btrfs_inode_extref
*)ptr
;
3771 BTRFS_I(inode
)->dir_index
= btrfs_inode_extref_index(leaf
,
3776 * try to precache a NULL acl entry for files that don't have
3777 * any xattrs or acls
3779 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
3780 btrfs_ino(inode
), &first_xattr_slot
);
3781 if (first_xattr_slot
!= -1) {
3782 path
->slots
[0] = first_xattr_slot
;
3783 ret
= btrfs_load_inode_props(inode
, path
);
3785 btrfs_err(root
->fs_info
,
3786 "error loading props for ino %llu (root %llu): %d",
3788 root
->root_key
.objectid
, ret
);
3790 btrfs_free_path(path
);
3793 cache_no_acl(inode
);
3795 switch (inode
->i_mode
& S_IFMT
) {
3797 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3798 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3799 inode
->i_fop
= &btrfs_file_operations
;
3800 inode
->i_op
= &btrfs_file_inode_operations
;
3803 inode
->i_fop
= &btrfs_dir_file_operations
;
3804 if (root
== root
->fs_info
->tree_root
)
3805 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
3807 inode
->i_op
= &btrfs_dir_inode_operations
;
3810 inode
->i_op
= &btrfs_symlink_inode_operations
;
3811 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
3814 inode
->i_op
= &btrfs_special_inode_operations
;
3815 init_special_inode(inode
, inode
->i_mode
, rdev
);
3819 btrfs_update_iflags(inode
);
3823 btrfs_free_path(path
);
3824 make_bad_inode(inode
);
3828 * given a leaf and an inode, copy the inode fields into the leaf
3830 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3831 struct extent_buffer
*leaf
,
3832 struct btrfs_inode_item
*item
,
3833 struct inode
*inode
)
3835 struct btrfs_map_token token
;
3837 btrfs_init_map_token(&token
);
3839 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3840 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3841 btrfs_set_token_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
,
3843 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3844 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3846 btrfs_set_token_timespec_sec(leaf
, &item
->atime
,
3847 inode
->i_atime
.tv_sec
, &token
);
3848 btrfs_set_token_timespec_nsec(leaf
, &item
->atime
,
3849 inode
->i_atime
.tv_nsec
, &token
);
3851 btrfs_set_token_timespec_sec(leaf
, &item
->mtime
,
3852 inode
->i_mtime
.tv_sec
, &token
);
3853 btrfs_set_token_timespec_nsec(leaf
, &item
->mtime
,
3854 inode
->i_mtime
.tv_nsec
, &token
);
3856 btrfs_set_token_timespec_sec(leaf
, &item
->ctime
,
3857 inode
->i_ctime
.tv_sec
, &token
);
3858 btrfs_set_token_timespec_nsec(leaf
, &item
->ctime
,
3859 inode
->i_ctime
.tv_nsec
, &token
);
3861 btrfs_set_token_timespec_sec(leaf
, &item
->otime
,
3862 BTRFS_I(inode
)->i_otime
.tv_sec
, &token
);
3863 btrfs_set_token_timespec_nsec(leaf
, &item
->otime
,
3864 BTRFS_I(inode
)->i_otime
.tv_nsec
, &token
);
3866 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3868 btrfs_set_token_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
,
3870 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
3871 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3872 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3873 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3874 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3878 * copy everything in the in-memory inode into the btree.
3880 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
3881 struct btrfs_root
*root
, struct inode
*inode
)
3883 struct btrfs_inode_item
*inode_item
;
3884 struct btrfs_path
*path
;
3885 struct extent_buffer
*leaf
;
3888 path
= btrfs_alloc_path();
3892 path
->leave_spinning
= 1;
3893 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
3901 leaf
= path
->nodes
[0];
3902 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3903 struct btrfs_inode_item
);
3905 fill_inode_item(trans
, leaf
, inode_item
, inode
);
3906 btrfs_mark_buffer_dirty(leaf
);
3907 btrfs_set_inode_last_trans(trans
, inode
);
3910 btrfs_free_path(path
);
3915 * copy everything in the in-memory inode into the btree.
3917 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
3918 struct btrfs_root
*root
, struct inode
*inode
)
3923 * If the inode is a free space inode, we can deadlock during commit
3924 * if we put it into the delayed code.
3926 * The data relocation inode should also be directly updated
3929 if (!btrfs_is_free_space_inode(inode
)
3930 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
3931 && !root
->fs_info
->log_root_recovering
) {
3932 btrfs_update_root_times(trans
, root
);
3934 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
3936 btrfs_set_inode_last_trans(trans
, inode
);
3940 return btrfs_update_inode_item(trans
, root
, inode
);
3943 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
3944 struct btrfs_root
*root
,
3945 struct inode
*inode
)
3949 ret
= btrfs_update_inode(trans
, root
, inode
);
3951 return btrfs_update_inode_item(trans
, root
, inode
);
3956 * unlink helper that gets used here in inode.c and in the tree logging
3957 * recovery code. It remove a link in a directory with a given name, and
3958 * also drops the back refs in the inode to the directory
3960 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3961 struct btrfs_root
*root
,
3962 struct inode
*dir
, struct inode
*inode
,
3963 const char *name
, int name_len
)
3965 struct btrfs_path
*path
;
3967 struct extent_buffer
*leaf
;
3968 struct btrfs_dir_item
*di
;
3969 struct btrfs_key key
;
3971 u64 ino
= btrfs_ino(inode
);
3972 u64 dir_ino
= btrfs_ino(dir
);
3974 path
= btrfs_alloc_path();
3980 path
->leave_spinning
= 1;
3981 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3982 name
, name_len
, -1);
3991 leaf
= path
->nodes
[0];
3992 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3993 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3996 btrfs_release_path(path
);
3999 * If we don't have dir index, we have to get it by looking up
4000 * the inode ref, since we get the inode ref, remove it directly,
4001 * it is unnecessary to do delayed deletion.
4003 * But if we have dir index, needn't search inode ref to get it.
4004 * Since the inode ref is close to the inode item, it is better
4005 * that we delay to delete it, and just do this deletion when
4006 * we update the inode item.
4008 if (BTRFS_I(inode
)->dir_index
) {
4009 ret
= btrfs_delayed_delete_inode_ref(inode
);
4011 index
= BTRFS_I(inode
)->dir_index
;
4016 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
4019 btrfs_info(root
->fs_info
,
4020 "failed to delete reference to %.*s, inode %llu parent %llu",
4021 name_len
, name
, ino
, dir_ino
);
4022 btrfs_abort_transaction(trans
, root
, ret
);
4026 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
4028 btrfs_abort_transaction(trans
, root
, ret
);
4032 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
4034 if (ret
!= 0 && ret
!= -ENOENT
) {
4035 btrfs_abort_transaction(trans
, root
, ret
);
4039 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
4044 btrfs_abort_transaction(trans
, root
, ret
);
4046 btrfs_free_path(path
);
4050 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
4051 inode_inc_iversion(inode
);
4052 inode_inc_iversion(dir
);
4053 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
4054 ret
= btrfs_update_inode(trans
, root
, dir
);
4059 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4060 struct btrfs_root
*root
,
4061 struct inode
*dir
, struct inode
*inode
,
4062 const char *name
, int name_len
)
4065 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
4068 ret
= btrfs_update_inode(trans
, root
, inode
);
4074 * helper to start transaction for unlink and rmdir.
4076 * unlink and rmdir are special in btrfs, they do not always free space, so
4077 * if we cannot make our reservations the normal way try and see if there is
4078 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4079 * allow the unlink to occur.
4081 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
)
4083 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4086 * 1 for the possible orphan item
4087 * 1 for the dir item
4088 * 1 for the dir index
4089 * 1 for the inode ref
4092 return btrfs_start_transaction_fallback_global_rsv(root
, 5, 5);
4095 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
4097 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4098 struct btrfs_trans_handle
*trans
;
4099 struct inode
*inode
= d_inode(dentry
);
4102 trans
= __unlink_start_trans(dir
);
4104 return PTR_ERR(trans
);
4106 btrfs_record_unlink_dir(trans
, dir
, d_inode(dentry
), 0);
4108 ret
= btrfs_unlink_inode(trans
, root
, dir
, d_inode(dentry
),
4109 dentry
->d_name
.name
, dentry
->d_name
.len
);
4113 if (inode
->i_nlink
== 0) {
4114 ret
= btrfs_orphan_add(trans
, inode
);
4120 btrfs_end_transaction(trans
, root
);
4121 btrfs_btree_balance_dirty(root
);
4125 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
4126 struct btrfs_root
*root
,
4127 struct inode
*dir
, u64 objectid
,
4128 const char *name
, int name_len
)
4130 struct btrfs_path
*path
;
4131 struct extent_buffer
*leaf
;
4132 struct btrfs_dir_item
*di
;
4133 struct btrfs_key key
;
4136 u64 dir_ino
= btrfs_ino(dir
);
4138 path
= btrfs_alloc_path();
4142 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4143 name
, name_len
, -1);
4144 if (IS_ERR_OR_NULL(di
)) {
4152 leaf
= path
->nodes
[0];
4153 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4154 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
4155 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4157 btrfs_abort_transaction(trans
, root
, ret
);
4160 btrfs_release_path(path
);
4162 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
4163 objectid
, root
->root_key
.objectid
,
4164 dir_ino
, &index
, name
, name_len
);
4166 if (ret
!= -ENOENT
) {
4167 btrfs_abort_transaction(trans
, root
, ret
);
4170 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
4172 if (IS_ERR_OR_NULL(di
)) {
4177 btrfs_abort_transaction(trans
, root
, ret
);
4181 leaf
= path
->nodes
[0];
4182 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4183 btrfs_release_path(path
);
4186 btrfs_release_path(path
);
4188 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
4190 btrfs_abort_transaction(trans
, root
, ret
);
4194 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
4195 inode_inc_iversion(dir
);
4196 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
4197 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
4199 btrfs_abort_transaction(trans
, root
, ret
);
4201 btrfs_free_path(path
);
4205 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
4207 struct inode
*inode
= d_inode(dentry
);
4209 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4210 struct btrfs_trans_handle
*trans
;
4212 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
4214 if (btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
4217 trans
= __unlink_start_trans(dir
);
4219 return PTR_ERR(trans
);
4221 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
4222 err
= btrfs_unlink_subvol(trans
, root
, dir
,
4223 BTRFS_I(inode
)->location
.objectid
,
4224 dentry
->d_name
.name
,
4225 dentry
->d_name
.len
);
4229 err
= btrfs_orphan_add(trans
, inode
);
4233 /* now the directory is empty */
4234 err
= btrfs_unlink_inode(trans
, root
, dir
, d_inode(dentry
),
4235 dentry
->d_name
.name
, dentry
->d_name
.len
);
4237 btrfs_i_size_write(inode
, 0);
4239 btrfs_end_transaction(trans
, root
);
4240 btrfs_btree_balance_dirty(root
);
4245 static int truncate_space_check(struct btrfs_trans_handle
*trans
,
4246 struct btrfs_root
*root
,
4251 bytes_deleted
= btrfs_csum_bytes_to_leaves(root
, bytes_deleted
);
4252 ret
= btrfs_block_rsv_add(root
, &root
->fs_info
->trans_block_rsv
,
4253 bytes_deleted
, BTRFS_RESERVE_NO_FLUSH
);
4255 trans
->bytes_reserved
+= bytes_deleted
;
4260 static int truncate_inline_extent(struct inode
*inode
,
4261 struct btrfs_path
*path
,
4262 struct btrfs_key
*found_key
,
4266 struct extent_buffer
*leaf
= path
->nodes
[0];
4267 int slot
= path
->slots
[0];
4268 struct btrfs_file_extent_item
*fi
;
4269 u32 size
= (u32
)(new_size
- found_key
->offset
);
4270 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4272 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
4274 if (btrfs_file_extent_compression(leaf
, fi
) != BTRFS_COMPRESS_NONE
) {
4275 loff_t offset
= new_size
;
4276 loff_t page_end
= ALIGN(offset
, PAGE_CACHE_SIZE
);
4279 * Zero out the remaining of the last page of our inline extent,
4280 * instead of directly truncating our inline extent here - that
4281 * would be much more complex (decompressing all the data, then
4282 * compressing the truncated data, which might be bigger than
4283 * the size of the inline extent, resize the extent, etc).
4284 * We release the path because to get the page we might need to
4285 * read the extent item from disk (data not in the page cache).
4287 btrfs_release_path(path
);
4288 return btrfs_truncate_page(inode
, offset
, page_end
- offset
, 0);
4291 btrfs_set_file_extent_ram_bytes(leaf
, fi
, size
);
4292 size
= btrfs_file_extent_calc_inline_size(size
);
4293 btrfs_truncate_item(root
, path
, size
, 1);
4295 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4296 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4302 * this can truncate away extent items, csum items and directory items.
4303 * It starts at a high offset and removes keys until it can't find
4304 * any higher than new_size
4306 * csum items that cross the new i_size are truncated to the new size
4309 * min_type is the minimum key type to truncate down to. If set to 0, this
4310 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4312 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
4313 struct btrfs_root
*root
,
4314 struct inode
*inode
,
4315 u64 new_size
, u32 min_type
)
4317 struct btrfs_path
*path
;
4318 struct extent_buffer
*leaf
;
4319 struct btrfs_file_extent_item
*fi
;
4320 struct btrfs_key key
;
4321 struct btrfs_key found_key
;
4322 u64 extent_start
= 0;
4323 u64 extent_num_bytes
= 0;
4324 u64 extent_offset
= 0;
4326 u64 last_size
= new_size
;
4327 u32 found_type
= (u8
)-1;
4330 int pending_del_nr
= 0;
4331 int pending_del_slot
= 0;
4332 int extent_type
= -1;
4335 u64 ino
= btrfs_ino(inode
);
4336 u64 bytes_deleted
= 0;
4338 bool should_throttle
= 0;
4339 bool should_end
= 0;
4341 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
4344 * for non-free space inodes and ref cows, we want to back off from
4347 if (!btrfs_is_free_space_inode(inode
) &&
4348 test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4351 path
= btrfs_alloc_path();
4357 * We want to drop from the next block forward in case this new size is
4358 * not block aligned since we will be keeping the last block of the
4359 * extent just the way it is.
4361 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4362 root
== root
->fs_info
->tree_root
)
4363 btrfs_drop_extent_cache(inode
, ALIGN(new_size
,
4364 root
->sectorsize
), (u64
)-1, 0);
4367 * This function is also used to drop the items in the log tree before
4368 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4369 * it is used to drop the loged items. So we shouldn't kill the delayed
4372 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
4373 btrfs_kill_delayed_inode_items(inode
);
4376 key
.offset
= (u64
)-1;
4381 * with a 16K leaf size and 128MB extents, you can actually queue
4382 * up a huge file in a single leaf. Most of the time that
4383 * bytes_deleted is > 0, it will be huge by the time we get here
4385 if (be_nice
&& bytes_deleted
> 32 * 1024 * 1024) {
4386 if (btrfs_should_end_transaction(trans
, root
)) {
4393 path
->leave_spinning
= 1;
4394 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
4401 /* there are no items in the tree for us to truncate, we're
4404 if (path
->slots
[0] == 0)
4411 leaf
= path
->nodes
[0];
4412 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4413 found_type
= found_key
.type
;
4415 if (found_key
.objectid
!= ino
)
4418 if (found_type
< min_type
)
4421 item_end
= found_key
.offset
;
4422 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
4423 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4424 struct btrfs_file_extent_item
);
4425 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4426 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4428 btrfs_file_extent_num_bytes(leaf
, fi
);
4429 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4430 item_end
+= btrfs_file_extent_inline_len(leaf
,
4431 path
->slots
[0], fi
);
4435 if (found_type
> min_type
) {
4438 if (item_end
< new_size
) {
4440 * With NO_HOLES mode, for the following mapping
4442 * [0-4k][hole][8k-12k]
4444 * if truncating isize down to 6k, it ends up
4447 if (btrfs_fs_incompat(root
->fs_info
, NO_HOLES
))
4448 last_size
= new_size
;
4451 if (found_key
.offset
>= new_size
)
4457 /* FIXME, shrink the extent if the ref count is only 1 */
4458 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
4462 last_size
= found_key
.offset
;
4464 last_size
= new_size
;
4466 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4468 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4470 u64 orig_num_bytes
=
4471 btrfs_file_extent_num_bytes(leaf
, fi
);
4472 extent_num_bytes
= ALIGN(new_size
-
4475 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4477 num_dec
= (orig_num_bytes
-
4479 if (test_bit(BTRFS_ROOT_REF_COWS
,
4482 inode_sub_bytes(inode
, num_dec
);
4483 btrfs_mark_buffer_dirty(leaf
);
4486 btrfs_file_extent_disk_num_bytes(leaf
,
4488 extent_offset
= found_key
.offset
-
4489 btrfs_file_extent_offset(leaf
, fi
);
4491 /* FIXME blocksize != 4096 */
4492 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
4493 if (extent_start
!= 0) {
4495 if (test_bit(BTRFS_ROOT_REF_COWS
,
4497 inode_sub_bytes(inode
, num_dec
);
4500 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4502 * we can't truncate inline items that have had
4506 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
4507 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
4510 * Need to release path in order to truncate a
4511 * compressed extent. So delete any accumulated
4512 * extent items so far.
4514 if (btrfs_file_extent_compression(leaf
, fi
) !=
4515 BTRFS_COMPRESS_NONE
&& pending_del_nr
) {
4516 err
= btrfs_del_items(trans
, root
, path
,
4520 btrfs_abort_transaction(trans
,
4528 err
= truncate_inline_extent(inode
, path
,
4533 btrfs_abort_transaction(trans
,
4537 } else if (test_bit(BTRFS_ROOT_REF_COWS
,
4539 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4544 if (!pending_del_nr
) {
4545 /* no pending yet, add ourselves */
4546 pending_del_slot
= path
->slots
[0];
4548 } else if (pending_del_nr
&&
4549 path
->slots
[0] + 1 == pending_del_slot
) {
4550 /* hop on the pending chunk */
4552 pending_del_slot
= path
->slots
[0];
4559 should_throttle
= 0;
4562 (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4563 root
== root
->fs_info
->tree_root
)) {
4564 btrfs_set_path_blocking(path
);
4565 bytes_deleted
+= extent_num_bytes
;
4566 ret
= btrfs_free_extent(trans
, root
, extent_start
,
4567 extent_num_bytes
, 0,
4568 btrfs_header_owner(leaf
),
4569 ino
, extent_offset
);
4571 if (btrfs_should_throttle_delayed_refs(trans
, root
))
4572 btrfs_async_run_delayed_refs(root
,
4573 trans
->delayed_ref_updates
* 2, 0);
4575 if (truncate_space_check(trans
, root
,
4576 extent_num_bytes
)) {
4579 if (btrfs_should_throttle_delayed_refs(trans
,
4581 should_throttle
= 1;
4586 if (found_type
== BTRFS_INODE_ITEM_KEY
)
4589 if (path
->slots
[0] == 0 ||
4590 path
->slots
[0] != pending_del_slot
||
4591 should_throttle
|| should_end
) {
4592 if (pending_del_nr
) {
4593 ret
= btrfs_del_items(trans
, root
, path
,
4597 btrfs_abort_transaction(trans
,
4603 btrfs_release_path(path
);
4604 if (should_throttle
) {
4605 unsigned long updates
= trans
->delayed_ref_updates
;
4607 trans
->delayed_ref_updates
= 0;
4608 ret
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
4614 * if we failed to refill our space rsv, bail out
4615 * and let the transaction restart
4627 if (pending_del_nr
) {
4628 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
4631 btrfs_abort_transaction(trans
, root
, ret
);
4634 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
4635 btrfs_ordered_update_i_size(inode
, last_size
, NULL
);
4637 btrfs_free_path(path
);
4639 if (be_nice
&& bytes_deleted
> 32 * 1024 * 1024) {
4640 unsigned long updates
= trans
->delayed_ref_updates
;
4642 trans
->delayed_ref_updates
= 0;
4643 ret
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
4652 * btrfs_truncate_page - read, zero a chunk and write a page
4653 * @inode - inode that we're zeroing
4654 * @from - the offset to start zeroing
4655 * @len - the length to zero, 0 to zero the entire range respective to the
4657 * @front - zero up to the offset instead of from the offset on
4659 * This will find the page for the "from" offset and cow the page and zero the
4660 * part we want to zero. This is used with truncate and hole punching.
4662 int btrfs_truncate_page(struct inode
*inode
, loff_t from
, loff_t len
,
4665 struct address_space
*mapping
= inode
->i_mapping
;
4666 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4667 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4668 struct btrfs_ordered_extent
*ordered
;
4669 struct extent_state
*cached_state
= NULL
;
4671 u32 blocksize
= root
->sectorsize
;
4672 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
4673 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
4675 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4680 if ((offset
& (blocksize
- 1)) == 0 &&
4681 (!len
|| ((len
& (blocksize
- 1)) == 0)))
4683 ret
= btrfs_delalloc_reserve_space(inode
,
4684 round_down(from
, PAGE_CACHE_SIZE
), PAGE_CACHE_SIZE
);
4689 page
= find_or_create_page(mapping
, index
, mask
);
4691 btrfs_delalloc_release_space(inode
,
4692 round_down(from
, PAGE_CACHE_SIZE
),
4698 page_start
= page_offset(page
);
4699 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4701 if (!PageUptodate(page
)) {
4702 ret
= btrfs_readpage(NULL
, page
);
4704 if (page
->mapping
!= mapping
) {
4706 page_cache_release(page
);
4709 if (!PageUptodate(page
)) {
4714 wait_on_page_writeback(page
);
4716 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
4717 set_page_extent_mapped(page
);
4719 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
4721 unlock_extent_cached(io_tree
, page_start
, page_end
,
4722 &cached_state
, GFP_NOFS
);
4724 page_cache_release(page
);
4725 btrfs_start_ordered_extent(inode
, ordered
, 1);
4726 btrfs_put_ordered_extent(ordered
);
4730 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
4731 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4732 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4733 0, 0, &cached_state
, GFP_NOFS
);
4735 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
4738 unlock_extent_cached(io_tree
, page_start
, page_end
,
4739 &cached_state
, GFP_NOFS
);
4743 if (offset
!= PAGE_CACHE_SIZE
) {
4745 len
= PAGE_CACHE_SIZE
- offset
;
4748 memset(kaddr
, 0, offset
);
4750 memset(kaddr
+ offset
, 0, len
);
4751 flush_dcache_page(page
);
4754 ClearPageChecked(page
);
4755 set_page_dirty(page
);
4756 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
4761 btrfs_delalloc_release_space(inode
, page_start
,
4764 page_cache_release(page
);
4769 static int maybe_insert_hole(struct btrfs_root
*root
, struct inode
*inode
,
4770 u64 offset
, u64 len
)
4772 struct btrfs_trans_handle
*trans
;
4776 * Still need to make sure the inode looks like it's been updated so
4777 * that any holes get logged if we fsync.
4779 if (btrfs_fs_incompat(root
->fs_info
, NO_HOLES
)) {
4780 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
4781 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
4782 BTRFS_I(inode
)->last_log_commit
= root
->last_log_commit
;
4787 * 1 - for the one we're dropping
4788 * 1 - for the one we're adding
4789 * 1 - for updating the inode.
4791 trans
= btrfs_start_transaction(root
, 3);
4793 return PTR_ERR(trans
);
4795 ret
= btrfs_drop_extents(trans
, root
, inode
, offset
, offset
+ len
, 1);
4797 btrfs_abort_transaction(trans
, root
, ret
);
4798 btrfs_end_transaction(trans
, root
);
4802 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(inode
), offset
,
4803 0, 0, len
, 0, len
, 0, 0, 0);
4805 btrfs_abort_transaction(trans
, root
, ret
);
4807 btrfs_update_inode(trans
, root
, inode
);
4808 btrfs_end_transaction(trans
, root
);
4813 * This function puts in dummy file extents for the area we're creating a hole
4814 * for. So if we are truncating this file to a larger size we need to insert
4815 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4816 * the range between oldsize and size
4818 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
4820 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4821 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4822 struct extent_map
*em
= NULL
;
4823 struct extent_state
*cached_state
= NULL
;
4824 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4825 u64 hole_start
= ALIGN(oldsize
, root
->sectorsize
);
4826 u64 block_end
= ALIGN(size
, root
->sectorsize
);
4833 * If our size started in the middle of a page we need to zero out the
4834 * rest of the page before we expand the i_size, otherwise we could
4835 * expose stale data.
4837 err
= btrfs_truncate_page(inode
, oldsize
, 0, 0);
4841 if (size
<= hole_start
)
4845 struct btrfs_ordered_extent
*ordered
;
4847 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
4849 ordered
= btrfs_lookup_ordered_range(inode
, hole_start
,
4850 block_end
- hole_start
);
4853 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
4854 &cached_state
, GFP_NOFS
);
4855 btrfs_start_ordered_extent(inode
, ordered
, 1);
4856 btrfs_put_ordered_extent(ordered
);
4859 cur_offset
= hole_start
;
4861 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4862 block_end
- cur_offset
, 0);
4868 last_byte
= min(extent_map_end(em
), block_end
);
4869 last_byte
= ALIGN(last_byte
, root
->sectorsize
);
4870 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
4871 struct extent_map
*hole_em
;
4872 hole_size
= last_byte
- cur_offset
;
4874 err
= maybe_insert_hole(root
, inode
, cur_offset
,
4878 btrfs_drop_extent_cache(inode
, cur_offset
,
4879 cur_offset
+ hole_size
- 1, 0);
4880 hole_em
= alloc_extent_map();
4882 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
4883 &BTRFS_I(inode
)->runtime_flags
);
4886 hole_em
->start
= cur_offset
;
4887 hole_em
->len
= hole_size
;
4888 hole_em
->orig_start
= cur_offset
;
4890 hole_em
->block_start
= EXTENT_MAP_HOLE
;
4891 hole_em
->block_len
= 0;
4892 hole_em
->orig_block_len
= 0;
4893 hole_em
->ram_bytes
= hole_size
;
4894 hole_em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4895 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
4896 hole_em
->generation
= root
->fs_info
->generation
;
4899 write_lock(&em_tree
->lock
);
4900 err
= add_extent_mapping(em_tree
, hole_em
, 1);
4901 write_unlock(&em_tree
->lock
);
4904 btrfs_drop_extent_cache(inode
, cur_offset
,
4908 free_extent_map(hole_em
);
4911 free_extent_map(em
);
4913 cur_offset
= last_byte
;
4914 if (cur_offset
>= block_end
)
4917 free_extent_map(em
);
4918 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
4923 static int wait_snapshoting_atomic_t(atomic_t
*a
)
4929 static void wait_for_snapshot_creation(struct btrfs_root
*root
)
4934 ret
= btrfs_start_write_no_snapshoting(root
);
4937 wait_on_atomic_t(&root
->will_be_snapshoted
,
4938 wait_snapshoting_atomic_t
,
4939 TASK_UNINTERRUPTIBLE
);
4943 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
4945 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4946 struct btrfs_trans_handle
*trans
;
4947 loff_t oldsize
= i_size_read(inode
);
4948 loff_t newsize
= attr
->ia_size
;
4949 int mask
= attr
->ia_valid
;
4953 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4954 * special case where we need to update the times despite not having
4955 * these flags set. For all other operations the VFS set these flags
4956 * explicitly if it wants a timestamp update.
4958 if (newsize
!= oldsize
) {
4959 inode_inc_iversion(inode
);
4960 if (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))
4961 inode
->i_ctime
= inode
->i_mtime
=
4962 current_fs_time(inode
->i_sb
);
4965 if (newsize
> oldsize
) {
4966 truncate_pagecache(inode
, newsize
);
4968 * Don't do an expanding truncate while snapshoting is ongoing.
4969 * This is to ensure the snapshot captures a fully consistent
4970 * state of this file - if the snapshot captures this expanding
4971 * truncation, it must capture all writes that happened before
4974 wait_for_snapshot_creation(root
);
4975 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
4977 btrfs_end_write_no_snapshoting(root
);
4981 trans
= btrfs_start_transaction(root
, 1);
4982 if (IS_ERR(trans
)) {
4983 btrfs_end_write_no_snapshoting(root
);
4984 return PTR_ERR(trans
);
4987 i_size_write(inode
, newsize
);
4988 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
4989 ret
= btrfs_update_inode(trans
, root
, inode
);
4990 btrfs_end_write_no_snapshoting(root
);
4991 btrfs_end_transaction(trans
, root
);
4995 * We're truncating a file that used to have good data down to
4996 * zero. Make sure it gets into the ordered flush list so that
4997 * any new writes get down to disk quickly.
5000 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
5001 &BTRFS_I(inode
)->runtime_flags
);
5004 * 1 for the orphan item we're going to add
5005 * 1 for the orphan item deletion.
5007 trans
= btrfs_start_transaction(root
, 2);
5009 return PTR_ERR(trans
);
5012 * We need to do this in case we fail at _any_ point during the
5013 * actual truncate. Once we do the truncate_setsize we could
5014 * invalidate pages which forces any outstanding ordered io to
5015 * be instantly completed which will give us extents that need
5016 * to be truncated. If we fail to get an orphan inode down we
5017 * could have left over extents that were never meant to live,
5018 * so we need to garuntee from this point on that everything
5019 * will be consistent.
5021 ret
= btrfs_orphan_add(trans
, inode
);
5022 btrfs_end_transaction(trans
, root
);
5026 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5027 truncate_setsize(inode
, newsize
);
5029 /* Disable nonlocked read DIO to avoid the end less truncate */
5030 btrfs_inode_block_unlocked_dio(inode
);
5031 inode_dio_wait(inode
);
5032 btrfs_inode_resume_unlocked_dio(inode
);
5034 ret
= btrfs_truncate(inode
);
5035 if (ret
&& inode
->i_nlink
) {
5039 * failed to truncate, disk_i_size is only adjusted down
5040 * as we remove extents, so it should represent the true
5041 * size of the inode, so reset the in memory size and
5042 * delete our orphan entry.
5044 trans
= btrfs_join_transaction(root
);
5045 if (IS_ERR(trans
)) {
5046 btrfs_orphan_del(NULL
, inode
);
5049 i_size_write(inode
, BTRFS_I(inode
)->disk_i_size
);
5050 err
= btrfs_orphan_del(trans
, inode
);
5052 btrfs_abort_transaction(trans
, root
, err
);
5053 btrfs_end_transaction(trans
, root
);
5060 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5062 struct inode
*inode
= d_inode(dentry
);
5063 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5066 if (btrfs_root_readonly(root
))
5069 err
= inode_change_ok(inode
, attr
);
5073 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
5074 err
= btrfs_setsize(inode
, attr
);
5079 if (attr
->ia_valid
) {
5080 setattr_copy(inode
, attr
);
5081 inode_inc_iversion(inode
);
5082 err
= btrfs_dirty_inode(inode
);
5084 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
5085 err
= posix_acl_chmod(inode
, inode
->i_mode
);
5092 * While truncating the inode pages during eviction, we get the VFS calling
5093 * btrfs_invalidatepage() against each page of the inode. This is slow because
5094 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5095 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5096 * extent_state structures over and over, wasting lots of time.
5098 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5099 * those expensive operations on a per page basis and do only the ordered io
5100 * finishing, while we release here the extent_map and extent_state structures,
5101 * without the excessive merging and splitting.
5103 static void evict_inode_truncate_pages(struct inode
*inode
)
5105 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5106 struct extent_map_tree
*map_tree
= &BTRFS_I(inode
)->extent_tree
;
5107 struct rb_node
*node
;
5109 ASSERT(inode
->i_state
& I_FREEING
);
5110 truncate_inode_pages_final(&inode
->i_data
);
5112 write_lock(&map_tree
->lock
);
5113 while (!RB_EMPTY_ROOT(&map_tree
->map
)) {
5114 struct extent_map
*em
;
5116 node
= rb_first(&map_tree
->map
);
5117 em
= rb_entry(node
, struct extent_map
, rb_node
);
5118 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5119 clear_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
5120 remove_extent_mapping(map_tree
, em
);
5121 free_extent_map(em
);
5122 if (need_resched()) {
5123 write_unlock(&map_tree
->lock
);
5125 write_lock(&map_tree
->lock
);
5128 write_unlock(&map_tree
->lock
);
5131 * Keep looping until we have no more ranges in the io tree.
5132 * We can have ongoing bios started by readpages (called from readahead)
5133 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5134 * still in progress (unlocked the pages in the bio but did not yet
5135 * unlocked the ranges in the io tree). Therefore this means some
5136 * ranges can still be locked and eviction started because before
5137 * submitting those bios, which are executed by a separate task (work
5138 * queue kthread), inode references (inode->i_count) were not taken
5139 * (which would be dropped in the end io callback of each bio).
5140 * Therefore here we effectively end up waiting for those bios and
5141 * anyone else holding locked ranges without having bumped the inode's
5142 * reference count - if we don't do it, when they access the inode's
5143 * io_tree to unlock a range it may be too late, leading to an
5144 * use-after-free issue.
5146 spin_lock(&io_tree
->lock
);
5147 while (!RB_EMPTY_ROOT(&io_tree
->state
)) {
5148 struct extent_state
*state
;
5149 struct extent_state
*cached_state
= NULL
;
5153 node
= rb_first(&io_tree
->state
);
5154 state
= rb_entry(node
, struct extent_state
, rb_node
);
5155 start
= state
->start
;
5157 spin_unlock(&io_tree
->lock
);
5159 lock_extent_bits(io_tree
, start
, end
, 0, &cached_state
);
5162 * If still has DELALLOC flag, the extent didn't reach disk,
5163 * and its reserved space won't be freed by delayed_ref.
5164 * So we need to free its reserved space here.
5165 * (Refer to comment in btrfs_invalidatepage, case 2)
5167 * Note, end is the bytenr of last byte, so we need + 1 here.
5169 if (state
->state
& EXTENT_DELALLOC
)
5170 btrfs_qgroup_free_data(inode
, start
, end
- start
+ 1);
5172 clear_extent_bit(io_tree
, start
, end
,
5173 EXTENT_LOCKED
| EXTENT_DIRTY
|
5174 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
5175 EXTENT_DEFRAG
, 1, 1,
5176 &cached_state
, GFP_NOFS
);
5179 spin_lock(&io_tree
->lock
);
5181 spin_unlock(&io_tree
->lock
);
5184 void btrfs_evict_inode(struct inode
*inode
)
5186 struct btrfs_trans_handle
*trans
;
5187 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5188 struct btrfs_block_rsv
*rsv
, *global_rsv
;
5189 int steal_from_global
= 0;
5190 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
5193 trace_btrfs_inode_evict(inode
);
5195 evict_inode_truncate_pages(inode
);
5197 if (inode
->i_nlink
&&
5198 ((btrfs_root_refs(&root
->root_item
) != 0 &&
5199 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
) ||
5200 btrfs_is_free_space_inode(inode
)))
5203 if (is_bad_inode(inode
)) {
5204 btrfs_orphan_del(NULL
, inode
);
5207 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5208 if (!special_file(inode
->i_mode
))
5209 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5211 btrfs_free_io_failure_record(inode
, 0, (u64
)-1);
5213 if (root
->fs_info
->log_root_recovering
) {
5214 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
5215 &BTRFS_I(inode
)->runtime_flags
));
5219 if (inode
->i_nlink
> 0) {
5220 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0 &&
5221 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
);
5225 ret
= btrfs_commit_inode_delayed_inode(inode
);
5227 btrfs_orphan_del(NULL
, inode
);
5231 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
5233 btrfs_orphan_del(NULL
, inode
);
5236 rsv
->size
= min_size
;
5238 global_rsv
= &root
->fs_info
->global_block_rsv
;
5240 btrfs_i_size_write(inode
, 0);
5243 * This is a bit simpler than btrfs_truncate since we've already
5244 * reserved our space for our orphan item in the unlink, so we just
5245 * need to reserve some slack space in case we add bytes and update
5246 * inode item when doing the truncate.
5249 ret
= btrfs_block_rsv_refill(root
, rsv
, min_size
,
5250 BTRFS_RESERVE_FLUSH_LIMIT
);
5253 * Try and steal from the global reserve since we will
5254 * likely not use this space anyway, we want to try as
5255 * hard as possible to get this to work.
5258 steal_from_global
++;
5260 steal_from_global
= 0;
5264 * steal_from_global == 0: we reserved stuff, hooray!
5265 * steal_from_global == 1: we didn't reserve stuff, boo!
5266 * steal_from_global == 2: we've committed, still not a lot of
5267 * room but maybe we'll have room in the global reserve this
5269 * steal_from_global == 3: abandon all hope!
5271 if (steal_from_global
> 2) {
5272 btrfs_warn(root
->fs_info
,
5273 "Could not get space for a delete, will truncate on mount %d",
5275 btrfs_orphan_del(NULL
, inode
);
5276 btrfs_free_block_rsv(root
, rsv
);
5280 trans
= btrfs_join_transaction(root
);
5281 if (IS_ERR(trans
)) {
5282 btrfs_orphan_del(NULL
, inode
);
5283 btrfs_free_block_rsv(root
, rsv
);
5288 * We can't just steal from the global reserve, we need tomake
5289 * sure there is room to do it, if not we need to commit and try
5292 if (steal_from_global
) {
5293 if (!btrfs_check_space_for_delayed_refs(trans
, root
))
5294 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
,
5301 * Couldn't steal from the global reserve, we have too much
5302 * pending stuff built up, commit the transaction and try it
5306 ret
= btrfs_commit_transaction(trans
, root
);
5308 btrfs_orphan_del(NULL
, inode
);
5309 btrfs_free_block_rsv(root
, rsv
);
5314 steal_from_global
= 0;
5317 trans
->block_rsv
= rsv
;
5319 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
5320 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
)
5323 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
5324 btrfs_end_transaction(trans
, root
);
5326 btrfs_btree_balance_dirty(root
);
5329 btrfs_free_block_rsv(root
, rsv
);
5332 * Errors here aren't a big deal, it just means we leave orphan items
5333 * in the tree. They will be cleaned up on the next mount.
5336 trans
->block_rsv
= root
->orphan_block_rsv
;
5337 btrfs_orphan_del(trans
, inode
);
5339 btrfs_orphan_del(NULL
, inode
);
5342 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
5343 if (!(root
== root
->fs_info
->tree_root
||
5344 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
5345 btrfs_return_ino(root
, btrfs_ino(inode
));
5347 btrfs_end_transaction(trans
, root
);
5348 btrfs_btree_balance_dirty(root
);
5350 btrfs_remove_delayed_node(inode
);
5356 * this returns the key found in the dir entry in the location pointer.
5357 * If no dir entries were found, location->objectid is 0.
5359 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
5360 struct btrfs_key
*location
)
5362 const char *name
= dentry
->d_name
.name
;
5363 int namelen
= dentry
->d_name
.len
;
5364 struct btrfs_dir_item
*di
;
5365 struct btrfs_path
*path
;
5366 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5369 path
= btrfs_alloc_path();
5373 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
5378 if (IS_ERR_OR_NULL(di
))
5381 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
5383 btrfs_free_path(path
);
5386 location
->objectid
= 0;
5391 * when we hit a tree root in a directory, the btrfs part of the inode
5392 * needs to be changed to reflect the root directory of the tree root. This
5393 * is kind of like crossing a mount point.
5395 static int fixup_tree_root_location(struct btrfs_root
*root
,
5397 struct dentry
*dentry
,
5398 struct btrfs_key
*location
,
5399 struct btrfs_root
**sub_root
)
5401 struct btrfs_path
*path
;
5402 struct btrfs_root
*new_root
;
5403 struct btrfs_root_ref
*ref
;
5404 struct extent_buffer
*leaf
;
5405 struct btrfs_key key
;
5409 path
= btrfs_alloc_path();
5416 key
.objectid
= BTRFS_I(dir
)->root
->root_key
.objectid
;
5417 key
.type
= BTRFS_ROOT_REF_KEY
;
5418 key
.offset
= location
->objectid
;
5420 ret
= btrfs_search_slot(NULL
, root
->fs_info
->tree_root
, &key
, path
,
5428 leaf
= path
->nodes
[0];
5429 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
5430 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
5431 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
5434 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
5435 (unsigned long)(ref
+ 1),
5436 dentry
->d_name
.len
);
5440 btrfs_release_path(path
);
5442 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
5443 if (IS_ERR(new_root
)) {
5444 err
= PTR_ERR(new_root
);
5448 *sub_root
= new_root
;
5449 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
5450 location
->type
= BTRFS_INODE_ITEM_KEY
;
5451 location
->offset
= 0;
5454 btrfs_free_path(path
);
5458 static void inode_tree_add(struct inode
*inode
)
5460 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5461 struct btrfs_inode
*entry
;
5463 struct rb_node
*parent
;
5464 struct rb_node
*new = &BTRFS_I(inode
)->rb_node
;
5465 u64 ino
= btrfs_ino(inode
);
5467 if (inode_unhashed(inode
))
5470 spin_lock(&root
->inode_lock
);
5471 p
= &root
->inode_tree
.rb_node
;
5474 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
5476 if (ino
< btrfs_ino(&entry
->vfs_inode
))
5477 p
= &parent
->rb_left
;
5478 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
5479 p
= &parent
->rb_right
;
5481 WARN_ON(!(entry
->vfs_inode
.i_state
&
5482 (I_WILL_FREE
| I_FREEING
)));
5483 rb_replace_node(parent
, new, &root
->inode_tree
);
5484 RB_CLEAR_NODE(parent
);
5485 spin_unlock(&root
->inode_lock
);
5489 rb_link_node(new, parent
, p
);
5490 rb_insert_color(new, &root
->inode_tree
);
5491 spin_unlock(&root
->inode_lock
);
5494 static void inode_tree_del(struct inode
*inode
)
5496 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5499 spin_lock(&root
->inode_lock
);
5500 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
5501 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
5502 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
5503 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5505 spin_unlock(&root
->inode_lock
);
5507 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0) {
5508 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
5509 spin_lock(&root
->inode_lock
);
5510 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5511 spin_unlock(&root
->inode_lock
);
5513 btrfs_add_dead_root(root
);
5517 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
5519 struct rb_node
*node
;
5520 struct rb_node
*prev
;
5521 struct btrfs_inode
*entry
;
5522 struct inode
*inode
;
5525 if (!test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
5526 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
5528 spin_lock(&root
->inode_lock
);
5530 node
= root
->inode_tree
.rb_node
;
5534 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5536 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
5537 node
= node
->rb_left
;
5538 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
5539 node
= node
->rb_right
;
5545 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
5546 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
5550 prev
= rb_next(prev
);
5554 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5555 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
5556 inode
= igrab(&entry
->vfs_inode
);
5558 spin_unlock(&root
->inode_lock
);
5559 if (atomic_read(&inode
->i_count
) > 1)
5560 d_prune_aliases(inode
);
5562 * btrfs_drop_inode will have it removed from
5563 * the inode cache when its usage count
5568 spin_lock(&root
->inode_lock
);
5572 if (cond_resched_lock(&root
->inode_lock
))
5575 node
= rb_next(node
);
5577 spin_unlock(&root
->inode_lock
);
5580 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
5582 struct btrfs_iget_args
*args
= p
;
5583 inode
->i_ino
= args
->location
->objectid
;
5584 memcpy(&BTRFS_I(inode
)->location
, args
->location
,
5585 sizeof(*args
->location
));
5586 BTRFS_I(inode
)->root
= args
->root
;
5590 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
5592 struct btrfs_iget_args
*args
= opaque
;
5593 return args
->location
->objectid
== BTRFS_I(inode
)->location
.objectid
&&
5594 args
->root
== BTRFS_I(inode
)->root
;
5597 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
5598 struct btrfs_key
*location
,
5599 struct btrfs_root
*root
)
5601 struct inode
*inode
;
5602 struct btrfs_iget_args args
;
5603 unsigned long hashval
= btrfs_inode_hash(location
->objectid
, root
);
5605 args
.location
= location
;
5608 inode
= iget5_locked(s
, hashval
, btrfs_find_actor
,
5609 btrfs_init_locked_inode
,
5614 /* Get an inode object given its location and corresponding root.
5615 * Returns in *is_new if the inode was read from disk
5617 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
5618 struct btrfs_root
*root
, int *new)
5620 struct inode
*inode
;
5622 inode
= btrfs_iget_locked(s
, location
, root
);
5624 return ERR_PTR(-ENOMEM
);
5626 if (inode
->i_state
& I_NEW
) {
5627 btrfs_read_locked_inode(inode
);
5628 if (!is_bad_inode(inode
)) {
5629 inode_tree_add(inode
);
5630 unlock_new_inode(inode
);
5634 unlock_new_inode(inode
);
5636 inode
= ERR_PTR(-ESTALE
);
5643 static struct inode
*new_simple_dir(struct super_block
*s
,
5644 struct btrfs_key
*key
,
5645 struct btrfs_root
*root
)
5647 struct inode
*inode
= new_inode(s
);
5650 return ERR_PTR(-ENOMEM
);
5652 BTRFS_I(inode
)->root
= root
;
5653 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
5654 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
5656 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
5657 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
5658 inode
->i_fop
= &simple_dir_operations
;
5659 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
5660 inode
->i_mtime
= CURRENT_TIME
;
5661 inode
->i_atime
= inode
->i_mtime
;
5662 inode
->i_ctime
= inode
->i_mtime
;
5663 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
5668 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
5670 struct inode
*inode
;
5671 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5672 struct btrfs_root
*sub_root
= root
;
5673 struct btrfs_key location
;
5677 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
5678 return ERR_PTR(-ENAMETOOLONG
);
5680 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
5682 return ERR_PTR(ret
);
5684 if (location
.objectid
== 0)
5685 return ERR_PTR(-ENOENT
);
5687 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
5688 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
5692 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
5694 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
5695 ret
= fixup_tree_root_location(root
, dir
, dentry
,
5696 &location
, &sub_root
);
5699 inode
= ERR_PTR(ret
);
5701 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
5703 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
5705 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
5707 if (!IS_ERR(inode
) && root
!= sub_root
) {
5708 down_read(&root
->fs_info
->cleanup_work_sem
);
5709 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
5710 ret
= btrfs_orphan_cleanup(sub_root
);
5711 up_read(&root
->fs_info
->cleanup_work_sem
);
5714 inode
= ERR_PTR(ret
);
5721 static int btrfs_dentry_delete(const struct dentry
*dentry
)
5723 struct btrfs_root
*root
;
5724 struct inode
*inode
= d_inode(dentry
);
5726 if (!inode
&& !IS_ROOT(dentry
))
5727 inode
= d_inode(dentry
->d_parent
);
5730 root
= BTRFS_I(inode
)->root
;
5731 if (btrfs_root_refs(&root
->root_item
) == 0)
5734 if (btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5740 static void btrfs_dentry_release(struct dentry
*dentry
)
5742 kfree(dentry
->d_fsdata
);
5745 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5748 struct inode
*inode
;
5750 inode
= btrfs_lookup_dentry(dir
, dentry
);
5751 if (IS_ERR(inode
)) {
5752 if (PTR_ERR(inode
) == -ENOENT
)
5755 return ERR_CAST(inode
);
5758 return d_splice_alias(inode
, dentry
);
5761 unsigned char btrfs_filetype_table
[] = {
5762 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
5765 static int btrfs_real_readdir(struct file
*file
, struct dir_context
*ctx
)
5767 struct inode
*inode
= file_inode(file
);
5768 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5769 struct btrfs_item
*item
;
5770 struct btrfs_dir_item
*di
;
5771 struct btrfs_key key
;
5772 struct btrfs_key found_key
;
5773 struct btrfs_path
*path
;
5774 struct list_head ins_list
;
5775 struct list_head del_list
;
5777 struct extent_buffer
*leaf
;
5779 unsigned char d_type
;
5784 int key_type
= BTRFS_DIR_INDEX_KEY
;
5788 int is_curr
= 0; /* ctx->pos points to the current index? */
5791 /* FIXME, use a real flag for deciding about the key type */
5792 if (root
->fs_info
->tree_root
== root
)
5793 key_type
= BTRFS_DIR_ITEM_KEY
;
5795 if (!dir_emit_dots(file
, ctx
))
5798 path
= btrfs_alloc_path();
5804 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5805 INIT_LIST_HEAD(&ins_list
);
5806 INIT_LIST_HEAD(&del_list
);
5807 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
5810 key
.type
= key_type
;
5811 key
.offset
= ctx
->pos
;
5812 key
.objectid
= btrfs_ino(inode
);
5814 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5820 leaf
= path
->nodes
[0];
5821 slot
= path
->slots
[0];
5822 if (slot
>= btrfs_header_nritems(leaf
)) {
5823 ret
= btrfs_next_leaf(root
, path
);
5831 item
= btrfs_item_nr(slot
);
5832 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5834 if (found_key
.objectid
!= key
.objectid
)
5836 if (found_key
.type
!= key_type
)
5838 if (found_key
.offset
< ctx
->pos
)
5840 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
5841 btrfs_should_delete_dir_index(&del_list
,
5845 ctx
->pos
= found_key
.offset
;
5848 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
5850 di_total
= btrfs_item_size(leaf
, item
);
5852 while (di_cur
< di_total
) {
5853 struct btrfs_key location
;
5855 if (verify_dir_item(root
, leaf
, di
))
5858 name_len
= btrfs_dir_name_len(leaf
, di
);
5859 if (name_len
<= sizeof(tmp_name
)) {
5860 name_ptr
= tmp_name
;
5862 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
5868 read_extent_buffer(leaf
, name_ptr
,
5869 (unsigned long)(di
+ 1), name_len
);
5871 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
5872 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
5875 /* is this a reference to our own snapshot? If so
5878 * In contrast to old kernels, we insert the snapshot's
5879 * dir item and dir index after it has been created, so
5880 * we won't find a reference to our own snapshot. We
5881 * still keep the following code for backward
5884 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
5885 location
.objectid
== root
->root_key
.objectid
) {
5889 over
= !dir_emit(ctx
, name_ptr
, name_len
,
5890 location
.objectid
, d_type
);
5893 if (name_ptr
!= tmp_name
)
5899 di_len
= btrfs_dir_name_len(leaf
, di
) +
5900 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
5902 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
5908 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5911 ret
= btrfs_readdir_delayed_dir_index(ctx
, &ins_list
, &emitted
);
5917 * If we haven't emitted any dir entry, we must not touch ctx->pos as
5918 * it was was set to the termination value in previous call. We assume
5919 * that "." and ".." were emitted if we reach this point and set the
5920 * termination value as well for an empty directory.
5922 if (ctx
->pos
> 2 && !emitted
)
5925 /* Reached end of directory/root. Bump pos past the last item. */
5929 * Stop new entries from being returned after we return the last
5932 * New directory entries are assigned a strictly increasing
5933 * offset. This means that new entries created during readdir
5934 * are *guaranteed* to be seen in the future by that readdir.
5935 * This has broken buggy programs which operate on names as
5936 * they're returned by readdir. Until we re-use freed offsets
5937 * we have this hack to stop new entries from being returned
5938 * under the assumption that they'll never reach this huge
5941 * This is being careful not to overflow 32bit loff_t unless the
5942 * last entry requires it because doing so has broken 32bit apps
5945 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5946 if (ctx
->pos
>= INT_MAX
)
5947 ctx
->pos
= LLONG_MAX
;
5954 if (key_type
== BTRFS_DIR_INDEX_KEY
)
5955 btrfs_put_delayed_items(&ins_list
, &del_list
);
5956 btrfs_free_path(path
);
5960 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5962 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5963 struct btrfs_trans_handle
*trans
;
5965 bool nolock
= false;
5967 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5970 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(inode
))
5973 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
5975 trans
= btrfs_join_transaction_nolock(root
);
5977 trans
= btrfs_join_transaction(root
);
5979 return PTR_ERR(trans
);
5980 ret
= btrfs_commit_transaction(trans
, root
);
5986 * This is somewhat expensive, updating the tree every time the
5987 * inode changes. But, it is most likely to find the inode in cache.
5988 * FIXME, needs more benchmarking...there are no reasons other than performance
5989 * to keep or drop this code.
5991 static int btrfs_dirty_inode(struct inode
*inode
)
5993 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5994 struct btrfs_trans_handle
*trans
;
5997 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
6000 trans
= btrfs_join_transaction(root
);
6002 return PTR_ERR(trans
);
6004 ret
= btrfs_update_inode(trans
, root
, inode
);
6005 if (ret
&& ret
== -ENOSPC
) {
6006 /* whoops, lets try again with the full transaction */
6007 btrfs_end_transaction(trans
, root
);
6008 trans
= btrfs_start_transaction(root
, 1);
6010 return PTR_ERR(trans
);
6012 ret
= btrfs_update_inode(trans
, root
, inode
);
6014 btrfs_end_transaction(trans
, root
);
6015 if (BTRFS_I(inode
)->delayed_node
)
6016 btrfs_balance_delayed_items(root
);
6022 * This is a copy of file_update_time. We need this so we can return error on
6023 * ENOSPC for updating the inode in the case of file write and mmap writes.
6025 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
6028 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6030 if (btrfs_root_readonly(root
))
6033 if (flags
& S_VERSION
)
6034 inode_inc_iversion(inode
);
6035 if (flags
& S_CTIME
)
6036 inode
->i_ctime
= *now
;
6037 if (flags
& S_MTIME
)
6038 inode
->i_mtime
= *now
;
6039 if (flags
& S_ATIME
)
6040 inode
->i_atime
= *now
;
6041 return btrfs_dirty_inode(inode
);
6045 * find the highest existing sequence number in a directory
6046 * and then set the in-memory index_cnt variable to reflect
6047 * free sequence numbers
6049 static int btrfs_set_inode_index_count(struct inode
*inode
)
6051 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6052 struct btrfs_key key
, found_key
;
6053 struct btrfs_path
*path
;
6054 struct extent_buffer
*leaf
;
6057 key
.objectid
= btrfs_ino(inode
);
6058 key
.type
= BTRFS_DIR_INDEX_KEY
;
6059 key
.offset
= (u64
)-1;
6061 path
= btrfs_alloc_path();
6065 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6068 /* FIXME: we should be able to handle this */
6074 * MAGIC NUMBER EXPLANATION:
6075 * since we search a directory based on f_pos we have to start at 2
6076 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6077 * else has to start at 2
6079 if (path
->slots
[0] == 0) {
6080 BTRFS_I(inode
)->index_cnt
= 2;
6086 leaf
= path
->nodes
[0];
6087 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6089 if (found_key
.objectid
!= btrfs_ino(inode
) ||
6090 found_key
.type
!= BTRFS_DIR_INDEX_KEY
) {
6091 BTRFS_I(inode
)->index_cnt
= 2;
6095 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
6097 btrfs_free_path(path
);
6102 * helper to find a free sequence number in a given directory. This current
6103 * code is very simple, later versions will do smarter things in the btree
6105 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
6109 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
6110 ret
= btrfs_inode_delayed_dir_index_count(dir
);
6112 ret
= btrfs_set_inode_index_count(dir
);
6118 *index
= BTRFS_I(dir
)->index_cnt
;
6119 BTRFS_I(dir
)->index_cnt
++;
6124 static int btrfs_insert_inode_locked(struct inode
*inode
)
6126 struct btrfs_iget_args args
;
6127 args
.location
= &BTRFS_I(inode
)->location
;
6128 args
.root
= BTRFS_I(inode
)->root
;
6130 return insert_inode_locked4(inode
,
6131 btrfs_inode_hash(inode
->i_ino
, BTRFS_I(inode
)->root
),
6132 btrfs_find_actor
, &args
);
6135 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
6136 struct btrfs_root
*root
,
6138 const char *name
, int name_len
,
6139 u64 ref_objectid
, u64 objectid
,
6140 umode_t mode
, u64
*index
)
6142 struct inode
*inode
;
6143 struct btrfs_inode_item
*inode_item
;
6144 struct btrfs_key
*location
;
6145 struct btrfs_path
*path
;
6146 struct btrfs_inode_ref
*ref
;
6147 struct btrfs_key key
[2];
6149 int nitems
= name
? 2 : 1;
6153 path
= btrfs_alloc_path();
6155 return ERR_PTR(-ENOMEM
);
6157 inode
= new_inode(root
->fs_info
->sb
);
6159 btrfs_free_path(path
);
6160 return ERR_PTR(-ENOMEM
);
6164 * O_TMPFILE, set link count to 0, so that after this point,
6165 * we fill in an inode item with the correct link count.
6168 set_nlink(inode
, 0);
6171 * we have to initialize this early, so we can reclaim the inode
6172 * number if we fail afterwards in this function.
6174 inode
->i_ino
= objectid
;
6177 trace_btrfs_inode_request(dir
);
6179 ret
= btrfs_set_inode_index(dir
, index
);
6181 btrfs_free_path(path
);
6183 return ERR_PTR(ret
);
6189 * index_cnt is ignored for everything but a dir,
6190 * btrfs_get_inode_index_count has an explanation for the magic
6193 BTRFS_I(inode
)->index_cnt
= 2;
6194 BTRFS_I(inode
)->dir_index
= *index
;
6195 BTRFS_I(inode
)->root
= root
;
6196 BTRFS_I(inode
)->generation
= trans
->transid
;
6197 inode
->i_generation
= BTRFS_I(inode
)->generation
;
6200 * We could have gotten an inode number from somebody who was fsynced
6201 * and then removed in this same transaction, so let's just set full
6202 * sync since it will be a full sync anyway and this will blow away the
6203 * old info in the log.
6205 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
6207 key
[0].objectid
= objectid
;
6208 key
[0].type
= BTRFS_INODE_ITEM_KEY
;
6211 sizes
[0] = sizeof(struct btrfs_inode_item
);
6215 * Start new inodes with an inode_ref. This is slightly more
6216 * efficient for small numbers of hard links since they will
6217 * be packed into one item. Extended refs will kick in if we
6218 * add more hard links than can fit in the ref item.
6220 key
[1].objectid
= objectid
;
6221 key
[1].type
= BTRFS_INODE_REF_KEY
;
6222 key
[1].offset
= ref_objectid
;
6224 sizes
[1] = name_len
+ sizeof(*ref
);
6227 location
= &BTRFS_I(inode
)->location
;
6228 location
->objectid
= objectid
;
6229 location
->offset
= 0;
6230 location
->type
= BTRFS_INODE_ITEM_KEY
;
6232 ret
= btrfs_insert_inode_locked(inode
);
6236 path
->leave_spinning
= 1;
6237 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, nitems
);
6241 inode_init_owner(inode
, dir
, mode
);
6242 inode_set_bytes(inode
, 0);
6244 inode
->i_mtime
= CURRENT_TIME
;
6245 inode
->i_atime
= inode
->i_mtime
;
6246 inode
->i_ctime
= inode
->i_mtime
;
6247 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
6249 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
6250 struct btrfs_inode_item
);
6251 memset_extent_buffer(path
->nodes
[0], 0, (unsigned long)inode_item
,
6252 sizeof(*inode_item
));
6253 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
6256 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
6257 struct btrfs_inode_ref
);
6258 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
6259 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
6260 ptr
= (unsigned long)(ref
+ 1);
6261 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
6264 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6265 btrfs_free_path(path
);
6267 btrfs_inherit_iflags(inode
, dir
);
6269 if (S_ISREG(mode
)) {
6270 if (btrfs_test_opt(root
, NODATASUM
))
6271 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6272 if (btrfs_test_opt(root
, NODATACOW
))
6273 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
6274 BTRFS_INODE_NODATASUM
;
6277 inode_tree_add(inode
);
6279 trace_btrfs_inode_new(inode
);
6280 btrfs_set_inode_last_trans(trans
, inode
);
6282 btrfs_update_root_times(trans
, root
);
6284 ret
= btrfs_inode_inherit_props(trans
, inode
, dir
);
6286 btrfs_err(root
->fs_info
,
6287 "error inheriting props for ino %llu (root %llu): %d",
6288 btrfs_ino(inode
), root
->root_key
.objectid
, ret
);
6293 unlock_new_inode(inode
);
6296 BTRFS_I(dir
)->index_cnt
--;
6297 btrfs_free_path(path
);
6299 return ERR_PTR(ret
);
6302 static inline u8
btrfs_inode_type(struct inode
*inode
)
6304 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
6308 * utility function to add 'inode' into 'parent_inode' with
6309 * a give name and a given sequence number.
6310 * if 'add_backref' is true, also insert a backref from the
6311 * inode to the parent directory.
6313 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
6314 struct inode
*parent_inode
, struct inode
*inode
,
6315 const char *name
, int name_len
, int add_backref
, u64 index
)
6318 struct btrfs_key key
;
6319 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
6320 u64 ino
= btrfs_ino(inode
);
6321 u64 parent_ino
= btrfs_ino(parent_inode
);
6323 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6324 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
6327 key
.type
= BTRFS_INODE_ITEM_KEY
;
6331 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6332 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
6333 key
.objectid
, root
->root_key
.objectid
,
6334 parent_ino
, index
, name
, name_len
);
6335 } else if (add_backref
) {
6336 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
6340 /* Nothing to clean up yet */
6344 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
6346 btrfs_inode_type(inode
), index
);
6347 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
6350 btrfs_abort_transaction(trans
, root
, ret
);
6354 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
6356 inode_inc_iversion(parent_inode
);
6357 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
6358 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
6360 btrfs_abort_transaction(trans
, root
, ret
);
6364 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6367 err
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
6368 key
.objectid
, root
->root_key
.objectid
,
6369 parent_ino
, &local_index
, name
, name_len
);
6371 } else if (add_backref
) {
6375 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
6376 ino
, parent_ino
, &local_index
);
6381 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
6382 struct inode
*dir
, struct dentry
*dentry
,
6383 struct inode
*inode
, int backref
, u64 index
)
6385 int err
= btrfs_add_link(trans
, dir
, inode
,
6386 dentry
->d_name
.name
, dentry
->d_name
.len
,
6393 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
6394 umode_t mode
, dev_t rdev
)
6396 struct btrfs_trans_handle
*trans
;
6397 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6398 struct inode
*inode
= NULL
;
6405 * 2 for inode item and ref
6407 * 1 for xattr if selinux is on
6409 trans
= btrfs_start_transaction(root
, 5);
6411 return PTR_ERR(trans
);
6413 err
= btrfs_find_free_ino(root
, &objectid
);
6417 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6418 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6420 if (IS_ERR(inode
)) {
6421 err
= PTR_ERR(inode
);
6426 * If the active LSM wants to access the inode during
6427 * d_instantiate it needs these. Smack checks to see
6428 * if the filesystem supports xattrs by looking at the
6431 inode
->i_op
= &btrfs_special_inode_operations
;
6432 init_special_inode(inode
, inode
->i_mode
, rdev
);
6434 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6436 goto out_unlock_inode
;
6438 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
6440 goto out_unlock_inode
;
6442 btrfs_update_inode(trans
, root
, inode
);
6443 d_instantiate_new(dentry
, inode
);
6447 btrfs_end_transaction(trans
, root
);
6448 btrfs_balance_delayed_items(root
);
6449 btrfs_btree_balance_dirty(root
);
6451 inode_dec_link_count(inode
);
6458 unlock_new_inode(inode
);
6463 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
6464 umode_t mode
, bool excl
)
6466 struct btrfs_trans_handle
*trans
;
6467 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6468 struct inode
*inode
= NULL
;
6469 int drop_inode_on_err
= 0;
6475 * 2 for inode item and ref
6477 * 1 for xattr if selinux is on
6479 trans
= btrfs_start_transaction(root
, 5);
6481 return PTR_ERR(trans
);
6483 err
= btrfs_find_free_ino(root
, &objectid
);
6487 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6488 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6490 if (IS_ERR(inode
)) {
6491 err
= PTR_ERR(inode
);
6494 drop_inode_on_err
= 1;
6496 * If the active LSM wants to access the inode during
6497 * d_instantiate it needs these. Smack checks to see
6498 * if the filesystem supports xattrs by looking at the
6501 inode
->i_fop
= &btrfs_file_operations
;
6502 inode
->i_op
= &btrfs_file_inode_operations
;
6503 inode
->i_mapping
->a_ops
= &btrfs_aops
;
6505 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6507 goto out_unlock_inode
;
6509 err
= btrfs_update_inode(trans
, root
, inode
);
6511 goto out_unlock_inode
;
6513 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
6515 goto out_unlock_inode
;
6517 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
6518 d_instantiate_new(dentry
, inode
);
6521 btrfs_end_transaction(trans
, root
);
6522 if (err
&& drop_inode_on_err
) {
6523 inode_dec_link_count(inode
);
6526 btrfs_balance_delayed_items(root
);
6527 btrfs_btree_balance_dirty(root
);
6531 unlock_new_inode(inode
);
6536 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
6537 struct dentry
*dentry
)
6539 struct btrfs_trans_handle
*trans
= NULL
;
6540 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6541 struct inode
*inode
= d_inode(old_dentry
);
6546 /* do not allow sys_link's with other subvols of the same device */
6547 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
6550 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
6553 err
= btrfs_set_inode_index(dir
, &index
);
6558 * 2 items for inode and inode ref
6559 * 2 items for dir items
6560 * 1 item for parent inode
6562 trans
= btrfs_start_transaction(root
, 5);
6563 if (IS_ERR(trans
)) {
6564 err
= PTR_ERR(trans
);
6569 /* There are several dir indexes for this inode, clear the cache. */
6570 BTRFS_I(inode
)->dir_index
= 0ULL;
6572 inode_inc_iversion(inode
);
6573 inode
->i_ctime
= CURRENT_TIME
;
6575 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
6577 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
6582 struct dentry
*parent
= dentry
->d_parent
;
6583 err
= btrfs_update_inode(trans
, root
, inode
);
6586 if (inode
->i_nlink
== 1) {
6588 * If new hard link count is 1, it's a file created
6589 * with open(2) O_TMPFILE flag.
6591 err
= btrfs_orphan_del(trans
, inode
);
6595 d_instantiate(dentry
, inode
);
6596 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
6599 btrfs_balance_delayed_items(root
);
6602 btrfs_end_transaction(trans
, root
);
6604 inode_dec_link_count(inode
);
6607 btrfs_btree_balance_dirty(root
);
6611 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
6613 struct inode
*inode
= NULL
;
6614 struct btrfs_trans_handle
*trans
;
6615 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6617 int drop_on_err
= 0;
6622 * 2 items for inode and ref
6623 * 2 items for dir items
6624 * 1 for xattr if selinux is on
6626 trans
= btrfs_start_transaction(root
, 5);
6628 return PTR_ERR(trans
);
6630 err
= btrfs_find_free_ino(root
, &objectid
);
6634 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6635 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6636 S_IFDIR
| mode
, &index
);
6637 if (IS_ERR(inode
)) {
6638 err
= PTR_ERR(inode
);
6643 /* these must be set before we unlock the inode */
6644 inode
->i_op
= &btrfs_dir_inode_operations
;
6645 inode
->i_fop
= &btrfs_dir_file_operations
;
6647 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6649 goto out_fail_inode
;
6651 btrfs_i_size_write(inode
, 0);
6652 err
= btrfs_update_inode(trans
, root
, inode
);
6654 goto out_fail_inode
;
6656 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
6657 dentry
->d_name
.len
, 0, index
);
6659 goto out_fail_inode
;
6661 d_instantiate_new(dentry
, inode
);
6665 btrfs_end_transaction(trans
, root
);
6667 inode_dec_link_count(inode
);
6670 btrfs_balance_delayed_items(root
);
6671 btrfs_btree_balance_dirty(root
);
6675 unlock_new_inode(inode
);
6679 /* Find next extent map of a given extent map, caller needs to ensure locks */
6680 static struct extent_map
*next_extent_map(struct extent_map
*em
)
6682 struct rb_node
*next
;
6684 next
= rb_next(&em
->rb_node
);
6687 return container_of(next
, struct extent_map
, rb_node
);
6690 static struct extent_map
*prev_extent_map(struct extent_map
*em
)
6692 struct rb_node
*prev
;
6694 prev
= rb_prev(&em
->rb_node
);
6697 return container_of(prev
, struct extent_map
, rb_node
);
6700 /* helper for btfs_get_extent. Given an existing extent in the tree,
6701 * the existing extent is the nearest extent to map_start,
6702 * and an extent that you want to insert, deal with overlap and insert
6703 * the best fitted new extent into the tree.
6705 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
6706 struct extent_map
*existing
,
6707 struct extent_map
*em
,
6710 struct extent_map
*prev
;
6711 struct extent_map
*next
;
6716 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
6718 if (existing
->start
> map_start
) {
6720 prev
= prev_extent_map(next
);
6723 next
= next_extent_map(prev
);
6726 start
= prev
? extent_map_end(prev
) : em
->start
;
6727 start
= max_t(u64
, start
, em
->start
);
6728 end
= next
? next
->start
: extent_map_end(em
);
6729 end
= min_t(u64
, end
, extent_map_end(em
));
6730 start_diff
= start
- em
->start
;
6732 em
->len
= end
- start
;
6733 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
6734 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
6735 em
->block_start
+= start_diff
;
6736 em
->block_len
-= start_diff
;
6738 return add_extent_mapping(em_tree
, em
, 0);
6741 static noinline
int uncompress_inline(struct btrfs_path
*path
,
6742 struct inode
*inode
, struct page
*page
,
6743 size_t pg_offset
, u64 extent_offset
,
6744 struct btrfs_file_extent_item
*item
)
6747 struct extent_buffer
*leaf
= path
->nodes
[0];
6750 unsigned long inline_size
;
6754 WARN_ON(pg_offset
!= 0);
6755 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6756 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6757 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
6758 btrfs_item_nr(path
->slots
[0]));
6759 tmp
= kmalloc(inline_size
, GFP_NOFS
);
6762 ptr
= btrfs_file_extent_inline_start(item
);
6764 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
6766 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
6767 ret
= btrfs_decompress(compress_type
, tmp
, page
,
6768 extent_offset
, inline_size
, max_size
);
6771 * decompression code contains a memset to fill in any space between the end
6772 * of the uncompressed data and the end of max_size in case the decompressed
6773 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6774 * the end of an inline extent and the beginning of the next block, so we
6775 * cover that region here.
6778 if (max_size
+ pg_offset
< PAGE_SIZE
) {
6779 char *map
= kmap(page
);
6780 memset(map
+ pg_offset
+ max_size
, 0, PAGE_SIZE
- max_size
- pg_offset
);
6788 * a bit scary, this does extent mapping from logical file offset to the disk.
6789 * the ugly parts come from merging extents from the disk with the in-ram
6790 * representation. This gets more complex because of the data=ordered code,
6791 * where the in-ram extents might be locked pending data=ordered completion.
6793 * This also copies inline extents directly into the page.
6796 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
6797 size_t pg_offset
, u64 start
, u64 len
,
6802 u64 extent_start
= 0;
6804 u64 objectid
= btrfs_ino(inode
);
6806 struct btrfs_path
*path
= NULL
;
6807 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6808 struct btrfs_file_extent_item
*item
;
6809 struct extent_buffer
*leaf
;
6810 struct btrfs_key found_key
;
6811 struct extent_map
*em
= NULL
;
6812 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
6813 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6814 struct btrfs_trans_handle
*trans
= NULL
;
6815 const bool new_inline
= !page
|| create
;
6818 read_lock(&em_tree
->lock
);
6819 em
= lookup_extent_mapping(em_tree
, start
, len
);
6821 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6822 read_unlock(&em_tree
->lock
);
6825 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
6826 free_extent_map(em
);
6827 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
6828 free_extent_map(em
);
6832 em
= alloc_extent_map();
6837 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6838 em
->start
= EXTENT_MAP_HOLE
;
6839 em
->orig_start
= EXTENT_MAP_HOLE
;
6841 em
->block_len
= (u64
)-1;
6844 path
= btrfs_alloc_path();
6850 * Chances are we'll be called again, so go ahead and do
6856 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
6857 objectid
, start
, trans
!= NULL
);
6864 if (path
->slots
[0] == 0)
6869 leaf
= path
->nodes
[0];
6870 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6871 struct btrfs_file_extent_item
);
6872 /* are we inside the extent that was found? */
6873 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6874 found_type
= found_key
.type
;
6875 if (found_key
.objectid
!= objectid
||
6876 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
6878 * If we backup past the first extent we want to move forward
6879 * and see if there is an extent in front of us, otherwise we'll
6880 * say there is a hole for our whole search range which can
6887 found_type
= btrfs_file_extent_type(leaf
, item
);
6888 extent_start
= found_key
.offset
;
6889 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6890 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6891 extent_end
= extent_start
+
6892 btrfs_file_extent_num_bytes(leaf
, item
);
6893 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6895 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6896 extent_end
= ALIGN(extent_start
+ size
, root
->sectorsize
);
6899 if (start
>= extent_end
) {
6901 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
6902 ret
= btrfs_next_leaf(root
, path
);
6909 leaf
= path
->nodes
[0];
6911 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6912 if (found_key
.objectid
!= objectid
||
6913 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6915 if (start
+ len
<= found_key
.offset
)
6917 if (start
> found_key
.offset
)
6920 em
->orig_start
= start
;
6921 em
->len
= found_key
.offset
- start
;
6925 btrfs_extent_item_to_extent_map(inode
, path
, item
, new_inline
, em
);
6927 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6928 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6930 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6934 size_t extent_offset
;
6940 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6941 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
6942 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
6943 size
- extent_offset
);
6944 em
->start
= extent_start
+ extent_offset
;
6945 em
->len
= ALIGN(copy_size
, root
->sectorsize
);
6946 em
->orig_block_len
= em
->len
;
6947 em
->orig_start
= em
->start
;
6948 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
6949 if (create
== 0 && !PageUptodate(page
)) {
6950 if (btrfs_file_extent_compression(leaf
, item
) !=
6951 BTRFS_COMPRESS_NONE
) {
6952 ret
= uncompress_inline(path
, inode
, page
,
6954 extent_offset
, item
);
6961 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6963 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
6964 memset(map
+ pg_offset
+ copy_size
, 0,
6965 PAGE_CACHE_SIZE
- pg_offset
-
6970 flush_dcache_page(page
);
6971 } else if (create
&& PageUptodate(page
)) {
6975 free_extent_map(em
);
6978 btrfs_release_path(path
);
6979 trans
= btrfs_join_transaction(root
);
6982 return ERR_CAST(trans
);
6986 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6989 btrfs_mark_buffer_dirty(leaf
);
6991 set_extent_uptodate(io_tree
, em
->start
,
6992 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
6997 em
->orig_start
= start
;
7000 em
->block_start
= EXTENT_MAP_HOLE
;
7001 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
7003 btrfs_release_path(path
);
7004 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
7005 btrfs_err(root
->fs_info
, "bad extent! em: [%llu %llu] passed [%llu %llu]",
7006 em
->start
, em
->len
, start
, len
);
7012 write_lock(&em_tree
->lock
);
7013 ret
= add_extent_mapping(em_tree
, em
, 0);
7014 /* it is possible that someone inserted the extent into the tree
7015 * while we had the lock dropped. It is also possible that
7016 * an overlapping map exists in the tree
7018 if (ret
== -EEXIST
) {
7019 struct extent_map
*existing
;
7023 existing
= search_extent_mapping(em_tree
, start
, len
);
7025 * existing will always be non-NULL, since there must be
7026 * extent causing the -EEXIST.
7028 if (start
>= extent_map_end(existing
) ||
7029 start
<= existing
->start
) {
7031 * The existing extent map is the one nearest to
7032 * the [start, start + len) range which overlaps
7034 err
= merge_extent_mapping(em_tree
, existing
,
7036 free_extent_map(existing
);
7038 free_extent_map(em
);
7042 free_extent_map(em
);
7047 write_unlock(&em_tree
->lock
);
7050 trace_btrfs_get_extent(root
, em
);
7052 btrfs_free_path(path
);
7054 ret
= btrfs_end_transaction(trans
, root
);
7059 free_extent_map(em
);
7060 return ERR_PTR(err
);
7062 BUG_ON(!em
); /* Error is always set */
7066 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
7067 size_t pg_offset
, u64 start
, u64 len
,
7070 struct extent_map
*em
;
7071 struct extent_map
*hole_em
= NULL
;
7072 u64 range_start
= start
;
7078 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
7085 * - a pre-alloc extent,
7086 * there might actually be delalloc bytes behind it.
7088 if (em
->block_start
!= EXTENT_MAP_HOLE
&&
7089 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7095 /* check to see if we've wrapped (len == -1 or similar) */
7104 /* ok, we didn't find anything, lets look for delalloc */
7105 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
7106 end
, len
, EXTENT_DELALLOC
, 1);
7107 found_end
= range_start
+ found
;
7108 if (found_end
< range_start
)
7109 found_end
= (u64
)-1;
7112 * we didn't find anything useful, return
7113 * the original results from get_extent()
7115 if (range_start
> end
|| found_end
<= start
) {
7121 /* adjust the range_start to make sure it doesn't
7122 * go backwards from the start they passed in
7124 range_start
= max(start
, range_start
);
7125 found
= found_end
- range_start
;
7128 u64 hole_start
= start
;
7131 em
= alloc_extent_map();
7137 * when btrfs_get_extent can't find anything it
7138 * returns one huge hole
7140 * make sure what it found really fits our range, and
7141 * adjust to make sure it is based on the start from
7145 u64 calc_end
= extent_map_end(hole_em
);
7147 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
7148 free_extent_map(hole_em
);
7151 hole_start
= max(hole_em
->start
, start
);
7152 hole_len
= calc_end
- hole_start
;
7156 if (hole_em
&& range_start
> hole_start
) {
7157 /* our hole starts before our delalloc, so we
7158 * have to return just the parts of the hole
7159 * that go until the delalloc starts
7161 em
->len
= min(hole_len
,
7162 range_start
- hole_start
);
7163 em
->start
= hole_start
;
7164 em
->orig_start
= hole_start
;
7166 * don't adjust block start at all,
7167 * it is fixed at EXTENT_MAP_HOLE
7169 em
->block_start
= hole_em
->block_start
;
7170 em
->block_len
= hole_len
;
7171 if (test_bit(EXTENT_FLAG_PREALLOC
, &hole_em
->flags
))
7172 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
7174 em
->start
= range_start
;
7176 em
->orig_start
= range_start
;
7177 em
->block_start
= EXTENT_MAP_DELALLOC
;
7178 em
->block_len
= found
;
7180 } else if (hole_em
) {
7185 free_extent_map(hole_em
);
7187 free_extent_map(em
);
7188 return ERR_PTR(err
);
7193 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
7196 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7197 struct extent_map
*em
;
7198 struct btrfs_key ins
;
7202 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
7203 ret
= btrfs_reserve_extent(root
, len
, root
->sectorsize
, 0,
7204 alloc_hint
, &ins
, 1, 1);
7206 return ERR_PTR(ret
);
7208 em
= create_pinned_em(inode
, start
, ins
.offset
, start
, ins
.objectid
,
7209 ins
.offset
, ins
.offset
, ins
.offset
, 0);
7211 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
7215 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
7216 ins
.offset
, ins
.offset
, 0);
7218 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
7219 free_extent_map(em
);
7220 return ERR_PTR(ret
);
7227 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7228 * block must be cow'd
7230 noinline
int can_nocow_extent(struct inode
*inode
, u64 offset
, u64
*len
,
7231 u64
*orig_start
, u64
*orig_block_len
,
7234 struct btrfs_trans_handle
*trans
;
7235 struct btrfs_path
*path
;
7237 struct extent_buffer
*leaf
;
7238 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7239 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7240 struct btrfs_file_extent_item
*fi
;
7241 struct btrfs_key key
;
7248 bool nocow
= (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
);
7250 path
= btrfs_alloc_path();
7254 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, btrfs_ino(inode
),
7259 slot
= path
->slots
[0];
7262 /* can't find the item, must cow */
7269 leaf
= path
->nodes
[0];
7270 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
7271 if (key
.objectid
!= btrfs_ino(inode
) ||
7272 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7273 /* not our file or wrong item type, must cow */
7277 if (key
.offset
> offset
) {
7278 /* Wrong offset, must cow */
7282 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
7283 found_type
= btrfs_file_extent_type(leaf
, fi
);
7284 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
7285 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
7286 /* not a regular extent, must cow */
7290 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_REG
)
7293 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
7294 if (extent_end
<= offset
)
7297 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7298 if (disk_bytenr
== 0)
7301 if (btrfs_file_extent_compression(leaf
, fi
) ||
7302 btrfs_file_extent_encryption(leaf
, fi
) ||
7303 btrfs_file_extent_other_encoding(leaf
, fi
))
7306 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
7309 *orig_start
= key
.offset
- backref_offset
;
7310 *orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7311 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
7314 if (btrfs_extent_readonly(root
, disk_bytenr
))
7317 num_bytes
= min(offset
+ *len
, extent_end
) - offset
;
7318 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7321 range_end
= round_up(offset
+ num_bytes
, root
->sectorsize
) - 1;
7322 ret
= test_range_bit(io_tree
, offset
, range_end
,
7323 EXTENT_DELALLOC
, 0, NULL
);
7330 btrfs_release_path(path
);
7333 * look for other files referencing this extent, if we
7334 * find any we must cow
7336 trans
= btrfs_join_transaction(root
);
7337 if (IS_ERR(trans
)) {
7342 ret
= btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
7343 key
.offset
- backref_offset
, disk_bytenr
);
7344 btrfs_end_transaction(trans
, root
);
7351 * adjust disk_bytenr and num_bytes to cover just the bytes
7352 * in this extent we are about to write. If there
7353 * are any csums in that range we have to cow in order
7354 * to keep the csums correct
7356 disk_bytenr
+= backref_offset
;
7357 disk_bytenr
+= offset
- key
.offset
;
7358 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
7361 * all of the above have passed, it is safe to overwrite this extent
7367 btrfs_free_path(path
);
7371 bool btrfs_page_exists_in_range(struct inode
*inode
, loff_t start
, loff_t end
)
7373 struct radix_tree_root
*root
= &inode
->i_mapping
->page_tree
;
7375 void **pagep
= NULL
;
7376 struct page
*page
= NULL
;
7377 unsigned long start_idx
;
7378 unsigned long end_idx
;
7380 start_idx
= start
>> PAGE_CACHE_SHIFT
;
7383 * end is the last byte in the last page. end == start is legal
7385 end_idx
= end
>> PAGE_CACHE_SHIFT
;
7389 /* Most of the code in this while loop is lifted from
7390 * find_get_page. It's been modified to begin searching from a
7391 * page and return just the first page found in that range. If the
7392 * found idx is less than or equal to the end idx then we know that
7393 * a page exists. If no pages are found or if those pages are
7394 * outside of the range then we're fine (yay!) */
7395 while (page
== NULL
&&
7396 radix_tree_gang_lookup_slot(root
, &pagep
, NULL
, start_idx
, 1)) {
7397 page
= radix_tree_deref_slot(pagep
);
7398 if (unlikely(!page
))
7401 if (radix_tree_exception(page
)) {
7402 if (radix_tree_deref_retry(page
)) {
7407 * Otherwise, shmem/tmpfs must be storing a swap entry
7408 * here as an exceptional entry: so return it without
7409 * attempting to raise page count.
7412 break; /* TODO: Is this relevant for this use case? */
7415 if (!page_cache_get_speculative(page
)) {
7421 * Has the page moved?
7422 * This is part of the lockless pagecache protocol. See
7423 * include/linux/pagemap.h for details.
7425 if (unlikely(page
!= *pagep
)) {
7426 page_cache_release(page
);
7432 if (page
->index
<= end_idx
)
7434 page_cache_release(page
);
7441 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
7442 struct extent_state
**cached_state
, int writing
)
7444 struct btrfs_ordered_extent
*ordered
;
7448 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7451 * We're concerned with the entire range that we're going to be
7452 * doing DIO to, so we need to make sure theres no ordered
7453 * extents in this range.
7455 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
7456 lockend
- lockstart
+ 1);
7459 * We need to make sure there are no buffered pages in this
7460 * range either, we could have raced between the invalidate in
7461 * generic_file_direct_write and locking the extent. The
7462 * invalidate needs to happen so that reads after a write do not
7467 !btrfs_page_exists_in_range(inode
, lockstart
, lockend
)))
7470 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7471 cached_state
, GFP_NOFS
);
7474 btrfs_start_ordered_extent(inode
, ordered
, 1);
7475 btrfs_put_ordered_extent(ordered
);
7477 /* Screw you mmap */
7478 ret
= btrfs_fdatawrite_range(inode
, lockstart
, lockend
);
7481 ret
= filemap_fdatawait_range(inode
->i_mapping
,
7488 * If we found a page that couldn't be invalidated just
7489 * fall back to buffered.
7491 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
7492 lockstart
>> PAGE_CACHE_SHIFT
,
7493 lockend
>> PAGE_CACHE_SHIFT
);
7504 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
7505 u64 len
, u64 orig_start
,
7506 u64 block_start
, u64 block_len
,
7507 u64 orig_block_len
, u64 ram_bytes
,
7510 struct extent_map_tree
*em_tree
;
7511 struct extent_map
*em
;
7512 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7515 em_tree
= &BTRFS_I(inode
)->extent_tree
;
7516 em
= alloc_extent_map();
7518 return ERR_PTR(-ENOMEM
);
7521 em
->orig_start
= orig_start
;
7522 em
->mod_start
= start
;
7525 em
->block_len
= block_len
;
7526 em
->block_start
= block_start
;
7527 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
7528 em
->orig_block_len
= orig_block_len
;
7529 em
->ram_bytes
= ram_bytes
;
7530 em
->generation
= -1;
7531 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
7532 if (type
== BTRFS_ORDERED_PREALLOC
)
7533 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
7536 btrfs_drop_extent_cache(inode
, em
->start
,
7537 em
->start
+ em
->len
- 1, 0);
7538 write_lock(&em_tree
->lock
);
7539 ret
= add_extent_mapping(em_tree
, em
, 1);
7540 write_unlock(&em_tree
->lock
);
7541 } while (ret
== -EEXIST
);
7544 free_extent_map(em
);
7545 return ERR_PTR(ret
);
7551 struct btrfs_dio_data
{
7552 u64 outstanding_extents
;
7556 static void adjust_dio_outstanding_extents(struct inode
*inode
,
7557 struct btrfs_dio_data
*dio_data
,
7560 unsigned num_extents
;
7562 num_extents
= (unsigned) div64_u64(len
+ BTRFS_MAX_EXTENT_SIZE
- 1,
7563 BTRFS_MAX_EXTENT_SIZE
);
7565 * If we have an outstanding_extents count still set then we're
7566 * within our reservation, otherwise we need to adjust our inode
7567 * counter appropriately.
7569 if (dio_data
->outstanding_extents
>= num_extents
) {
7570 dio_data
->outstanding_extents
-= num_extents
;
7573 * If dio write length has been split due to no large enough
7574 * contiguous space, we need to compensate our inode counter
7577 u64 num_needed
= num_extents
- dio_data
->outstanding_extents
;
7579 spin_lock(&BTRFS_I(inode
)->lock
);
7580 BTRFS_I(inode
)->outstanding_extents
+= num_needed
;
7581 spin_unlock(&BTRFS_I(inode
)->lock
);
7585 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
7586 struct buffer_head
*bh_result
, int create
)
7588 struct extent_map
*em
;
7589 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7590 struct extent_state
*cached_state
= NULL
;
7591 struct btrfs_dio_data
*dio_data
= NULL
;
7592 u64 start
= iblock
<< inode
->i_blkbits
;
7593 u64 lockstart
, lockend
;
7594 u64 len
= bh_result
->b_size
;
7595 int unlock_bits
= EXTENT_LOCKED
;
7599 unlock_bits
|= EXTENT_DIRTY
;
7601 len
= min_t(u64
, len
, root
->sectorsize
);
7604 lockend
= start
+ len
- 1;
7606 if (current
->journal_info
) {
7608 * Need to pull our outstanding extents and set journal_info to NULL so
7609 * that anything that needs to check if there's a transction doesn't get
7612 dio_data
= current
->journal_info
;
7613 current
->journal_info
= NULL
;
7617 * If this errors out it's because we couldn't invalidate pagecache for
7618 * this range and we need to fallback to buffered.
7620 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
,
7626 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
7633 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7634 * io. INLINE is special, and we could probably kludge it in here, but
7635 * it's still buffered so for safety lets just fall back to the generic
7638 * For COMPRESSED we _have_ to read the entire extent in so we can
7639 * decompress it, so there will be buffering required no matter what we
7640 * do, so go ahead and fallback to buffered.
7642 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7643 * to buffered IO. Don't blame me, this is the price we pay for using
7646 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
7647 em
->block_start
== EXTENT_MAP_INLINE
) {
7648 free_extent_map(em
);
7653 /* Just a good old fashioned hole, return */
7654 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
7655 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
7656 free_extent_map(em
);
7661 * We don't allocate a new extent in the following cases
7663 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7665 * 2) The extent is marked as PREALLOC. We're good to go here and can
7666 * just use the extent.
7670 len
= min(len
, em
->len
- (start
- em
->start
));
7671 lockstart
= start
+ len
;
7675 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
7676 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
7677 em
->block_start
!= EXTENT_MAP_HOLE
)) {
7679 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
7681 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7682 type
= BTRFS_ORDERED_PREALLOC
;
7684 type
= BTRFS_ORDERED_NOCOW
;
7685 len
= min(len
, em
->len
- (start
- em
->start
));
7686 block_start
= em
->block_start
+ (start
- em
->start
);
7688 if (can_nocow_extent(inode
, start
, &len
, &orig_start
,
7689 &orig_block_len
, &ram_bytes
) == 1) {
7690 if (type
== BTRFS_ORDERED_PREALLOC
) {
7691 free_extent_map(em
);
7692 em
= create_pinned_em(inode
, start
, len
,
7703 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
7704 block_start
, len
, len
, type
);
7706 free_extent_map(em
);
7714 * this will cow the extent, reset the len in case we changed
7717 len
= bh_result
->b_size
;
7718 free_extent_map(em
);
7719 em
= btrfs_new_extent_direct(inode
, start
, len
);
7724 len
= min(len
, em
->len
- (start
- em
->start
));
7726 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
7728 bh_result
->b_size
= len
;
7729 bh_result
->b_bdev
= em
->bdev
;
7730 set_buffer_mapped(bh_result
);
7732 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7733 set_buffer_new(bh_result
);
7736 * Need to update the i_size under the extent lock so buffered
7737 * readers will get the updated i_size when we unlock.
7739 if (start
+ len
> i_size_read(inode
))
7740 i_size_write(inode
, start
+ len
);
7742 adjust_dio_outstanding_extents(inode
, dio_data
, len
);
7743 btrfs_free_reserved_data_space(inode
, start
, len
);
7744 WARN_ON(dio_data
->reserve
< len
);
7745 dio_data
->reserve
-= len
;
7746 current
->journal_info
= dio_data
;
7750 * In the case of write we need to clear and unlock the entire range,
7751 * in the case of read we need to unlock only the end area that we
7752 * aren't using if there is any left over space.
7754 if (lockstart
< lockend
) {
7755 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
7756 lockend
, unlock_bits
, 1, 0,
7757 &cached_state
, GFP_NOFS
);
7759 free_extent_state(cached_state
);
7762 free_extent_map(em
);
7767 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7768 unlock_bits
, 1, 0, &cached_state
, GFP_NOFS
);
7771 current
->journal_info
= dio_data
;
7773 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7774 * write less data then expected, so that we don't underflow our inode's
7775 * outstanding extents counter.
7777 if (create
&& dio_data
)
7778 adjust_dio_outstanding_extents(inode
, dio_data
, len
);
7783 static inline int submit_dio_repair_bio(struct inode
*inode
, struct bio
*bio
,
7784 int rw
, int mirror_num
)
7786 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7789 BUG_ON(rw
& REQ_WRITE
);
7793 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
,
7794 BTRFS_WQ_ENDIO_DIO_REPAIR
);
7798 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
7804 static int btrfs_check_dio_repairable(struct inode
*inode
,
7805 struct bio
*failed_bio
,
7806 struct io_failure_record
*failrec
,
7811 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
7812 failrec
->logical
, failrec
->len
);
7813 if (num_copies
== 1) {
7815 * we only have a single copy of the data, so don't bother with
7816 * all the retry and error correction code that follows. no
7817 * matter what the error is, it is very likely to persist.
7819 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7820 num_copies
, failrec
->this_mirror
, failed_mirror
);
7824 failrec
->failed_mirror
= failed_mirror
;
7825 failrec
->this_mirror
++;
7826 if (failrec
->this_mirror
== failed_mirror
)
7827 failrec
->this_mirror
++;
7829 if (failrec
->this_mirror
> num_copies
) {
7830 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7831 num_copies
, failrec
->this_mirror
, failed_mirror
);
7838 static int dio_read_error(struct inode
*inode
, struct bio
*failed_bio
,
7839 struct page
*page
, u64 start
, u64 end
,
7840 int failed_mirror
, bio_end_io_t
*repair_endio
,
7843 struct io_failure_record
*failrec
;
7849 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
7851 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
7855 ret
= btrfs_check_dio_repairable(inode
, failed_bio
, failrec
,
7858 free_io_failure(inode
, failrec
);
7862 if (failed_bio
->bi_vcnt
> 1)
7863 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
7865 read_mode
= READ_SYNC
;
7867 isector
= start
- btrfs_io_bio(failed_bio
)->logical
;
7868 isector
>>= inode
->i_sb
->s_blocksize_bits
;
7869 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
7870 0, isector
, repair_endio
, repair_arg
);
7872 free_io_failure(inode
, failrec
);
7876 btrfs_debug(BTRFS_I(inode
)->root
->fs_info
,
7877 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7878 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
7880 ret
= submit_dio_repair_bio(inode
, bio
, read_mode
,
7881 failrec
->this_mirror
);
7883 free_io_failure(inode
, failrec
);
7890 struct btrfs_retry_complete
{
7891 struct completion done
;
7892 struct inode
*inode
;
7897 static void btrfs_retry_endio_nocsum(struct bio
*bio
)
7899 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7900 struct bio_vec
*bvec
;
7907 bio_for_each_segment_all(bvec
, bio
, i
)
7908 clean_io_failure(done
->inode
, done
->start
, bvec
->bv_page
, 0);
7910 complete(&done
->done
);
7914 static int __btrfs_correct_data_nocsum(struct inode
*inode
,
7915 struct btrfs_io_bio
*io_bio
)
7917 struct bio_vec
*bvec
;
7918 struct btrfs_retry_complete done
;
7923 start
= io_bio
->logical
;
7926 bio_for_each_segment_all(bvec
, &io_bio
->bio
, i
) {
7930 init_completion(&done
.done
);
7932 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
->bv_page
, start
,
7933 start
+ bvec
->bv_len
- 1,
7935 btrfs_retry_endio_nocsum
, &done
);
7939 wait_for_completion(&done
.done
);
7941 if (!done
.uptodate
) {
7942 /* We might have another mirror, so try again */
7946 start
+= bvec
->bv_len
;
7952 static void btrfs_retry_endio(struct bio
*bio
)
7954 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7955 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
7956 struct bio_vec
*bvec
;
7965 bio_for_each_segment_all(bvec
, bio
, i
) {
7966 ret
= __readpage_endio_check(done
->inode
, io_bio
, i
,
7968 done
->start
, bvec
->bv_len
);
7970 clean_io_failure(done
->inode
, done
->start
,
7976 done
->uptodate
= uptodate
;
7978 complete(&done
->done
);
7982 static int __btrfs_subio_endio_read(struct inode
*inode
,
7983 struct btrfs_io_bio
*io_bio
, int err
)
7985 struct bio_vec
*bvec
;
7986 struct btrfs_retry_complete done
;
7993 start
= io_bio
->logical
;
7996 bio_for_each_segment_all(bvec
, &io_bio
->bio
, i
) {
7997 ret
= __readpage_endio_check(inode
, io_bio
, i
, bvec
->bv_page
,
7998 0, start
, bvec
->bv_len
);
8004 init_completion(&done
.done
);
8006 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
->bv_page
, start
,
8007 start
+ bvec
->bv_len
- 1,
8009 btrfs_retry_endio
, &done
);
8015 wait_for_completion(&done
.done
);
8017 if (!done
.uptodate
) {
8018 /* We might have another mirror, so try again */
8022 offset
+= bvec
->bv_len
;
8023 start
+= bvec
->bv_len
;
8029 static int btrfs_subio_endio_read(struct inode
*inode
,
8030 struct btrfs_io_bio
*io_bio
, int err
)
8032 bool skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
8036 return __btrfs_correct_data_nocsum(inode
, io_bio
);
8040 return __btrfs_subio_endio_read(inode
, io_bio
, err
);
8044 static void btrfs_endio_direct_read(struct bio
*bio
)
8046 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8047 struct inode
*inode
= dip
->inode
;
8048 struct bio
*dio_bio
;
8049 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8050 int err
= bio
->bi_error
;
8052 if (dip
->flags
& BTRFS_DIO_ORIG_BIO_SUBMITTED
)
8053 err
= btrfs_subio_endio_read(inode
, io_bio
, err
);
8055 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
8056 dip
->logical_offset
+ dip
->bytes
- 1);
8057 dio_bio
= dip
->dio_bio
;
8061 dio_bio
->bi_error
= bio
->bi_error
;
8062 dio_end_io(dio_bio
, bio
->bi_error
);
8065 io_bio
->end_io(io_bio
, err
);
8069 static void btrfs_endio_direct_write(struct bio
*bio
)
8071 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8072 struct inode
*inode
= dip
->inode
;
8073 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8074 struct btrfs_ordered_extent
*ordered
= NULL
;
8075 u64 ordered_offset
= dip
->logical_offset
;
8076 u64 ordered_bytes
= dip
->bytes
;
8077 struct bio
*dio_bio
;
8081 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
8088 btrfs_init_work(&ordered
->work
, btrfs_endio_write_helper
,
8089 finish_ordered_fn
, NULL
, NULL
);
8090 btrfs_queue_work(root
->fs_info
->endio_write_workers
,
8094 * our bio might span multiple ordered extents. If we haven't
8095 * completed the accounting for the whole dio, go back and try again
8097 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
8098 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
8103 dio_bio
= dip
->dio_bio
;
8107 dio_bio
->bi_error
= bio
->bi_error
;
8108 dio_end_io(dio_bio
, bio
->bi_error
);
8112 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
8113 struct bio
*bio
, int mirror_num
,
8114 unsigned long bio_flags
, u64 offset
)
8117 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8118 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
8119 BUG_ON(ret
); /* -ENOMEM */
8123 static void btrfs_end_dio_bio(struct bio
*bio
)
8125 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8126 int err
= bio
->bi_error
;
8129 btrfs_warn(BTRFS_I(dip
->inode
)->root
->fs_info
,
8130 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8131 btrfs_ino(dip
->inode
), bio
->bi_rw
,
8132 (unsigned long long)bio
->bi_iter
.bi_sector
,
8133 bio
->bi_iter
.bi_size
, err
);
8135 if (dip
->subio_endio
)
8136 err
= dip
->subio_endio(dip
->inode
, btrfs_io_bio(bio
), err
);
8142 * before atomic variable goto zero, we must make sure
8143 * dip->errors is perceived to be set.
8145 smp_mb__before_atomic();
8148 /* if there are more bios still pending for this dio, just exit */
8149 if (!atomic_dec_and_test(&dip
->pending_bios
))
8153 bio_io_error(dip
->orig_bio
);
8155 dip
->dio_bio
->bi_error
= 0;
8156 bio_endio(dip
->orig_bio
);
8162 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
8163 u64 first_sector
, gfp_t gfp_flags
)
8166 bio
= btrfs_bio_alloc(bdev
, first_sector
, BIO_MAX_PAGES
, gfp_flags
);
8168 bio_associate_current(bio
);
8172 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root
*root
,
8173 struct inode
*inode
,
8174 struct btrfs_dio_private
*dip
,
8178 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8179 struct btrfs_io_bio
*orig_io_bio
= btrfs_io_bio(dip
->orig_bio
);
8183 * We load all the csum data we need when we submit
8184 * the first bio to reduce the csum tree search and
8187 if (dip
->logical_offset
== file_offset
) {
8188 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, dip
->orig_bio
,
8194 if (bio
== dip
->orig_bio
)
8197 file_offset
-= dip
->logical_offset
;
8198 file_offset
>>= inode
->i_sb
->s_blocksize_bits
;
8199 io_bio
->csum
= (u8
*)(((u32
*)orig_io_bio
->csum
) + file_offset
);
8204 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
8205 int rw
, u64 file_offset
, int skip_sum
,
8208 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8209 int write
= rw
& REQ_WRITE
;
8210 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8214 async_submit
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
8219 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
,
8220 BTRFS_WQ_ENDIO_DATA
);
8228 if (write
&& async_submit
) {
8229 ret
= btrfs_wq_submit_bio(root
->fs_info
,
8230 inode
, rw
, bio
, 0, 0,
8232 __btrfs_submit_bio_start_direct_io
,
8233 __btrfs_submit_bio_done
);
8237 * If we aren't doing async submit, calculate the csum of the
8240 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
8244 ret
= btrfs_lookup_and_bind_dio_csum(root
, inode
, dip
, bio
,
8250 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
8256 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
8259 struct inode
*inode
= dip
->inode
;
8260 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8262 struct bio
*orig_bio
= dip
->orig_bio
;
8263 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
8264 u64 start_sector
= orig_bio
->bi_iter
.bi_sector
;
8265 u64 file_offset
= dip
->logical_offset
;
8270 int async_submit
= 0;
8272 map_length
= orig_bio
->bi_iter
.bi_size
;
8273 ret
= btrfs_map_block(root
->fs_info
, rw
, start_sector
<< 9,
8274 &map_length
, NULL
, 0);
8278 if (map_length
>= orig_bio
->bi_iter
.bi_size
) {
8280 dip
->flags
|= BTRFS_DIO_ORIG_BIO_SUBMITTED
;
8284 /* async crcs make it difficult to collect full stripe writes. */
8285 if (btrfs_get_alloc_profile(root
, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK
)
8290 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
8294 bio
->bi_private
= dip
;
8295 bio
->bi_end_io
= btrfs_end_dio_bio
;
8296 btrfs_io_bio(bio
)->logical
= file_offset
;
8297 atomic_inc(&dip
->pending_bios
);
8299 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
8300 if (map_length
< submit_len
+ bvec
->bv_len
||
8301 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
8302 bvec
->bv_offset
) < bvec
->bv_len
) {
8304 * inc the count before we submit the bio so
8305 * we know the end IO handler won't happen before
8306 * we inc the count. Otherwise, the dip might get freed
8307 * before we're done setting it up
8309 atomic_inc(&dip
->pending_bios
);
8310 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
8311 file_offset
, skip_sum
,
8315 atomic_dec(&dip
->pending_bios
);
8319 start_sector
+= submit_len
>> 9;
8320 file_offset
+= submit_len
;
8325 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
8326 start_sector
, GFP_NOFS
);
8329 bio
->bi_private
= dip
;
8330 bio
->bi_end_io
= btrfs_end_dio_bio
;
8331 btrfs_io_bio(bio
)->logical
= file_offset
;
8333 map_length
= orig_bio
->bi_iter
.bi_size
;
8334 ret
= btrfs_map_block(root
->fs_info
, rw
,
8336 &map_length
, NULL
, 0);
8342 submit_len
+= bvec
->bv_len
;
8349 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
8358 * before atomic variable goto zero, we must
8359 * make sure dip->errors is perceived to be set.
8361 smp_mb__before_atomic();
8362 if (atomic_dec_and_test(&dip
->pending_bios
))
8363 bio_io_error(dip
->orig_bio
);
8365 /* bio_end_io() will handle error, so we needn't return it */
8369 static void btrfs_submit_direct(int rw
, struct bio
*dio_bio
,
8370 struct inode
*inode
, loff_t file_offset
)
8372 struct btrfs_dio_private
*dip
= NULL
;
8373 struct bio
*io_bio
= NULL
;
8374 struct btrfs_io_bio
*btrfs_bio
;
8376 int write
= rw
& REQ_WRITE
;
8379 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
8381 io_bio
= btrfs_bio_clone(dio_bio
, GFP_NOFS
);
8387 dip
= kzalloc(sizeof(*dip
), GFP_NOFS
);
8393 dip
->private = dio_bio
->bi_private
;
8395 dip
->logical_offset
= file_offset
;
8396 dip
->bytes
= dio_bio
->bi_iter
.bi_size
;
8397 dip
->disk_bytenr
= (u64
)dio_bio
->bi_iter
.bi_sector
<< 9;
8398 io_bio
->bi_private
= dip
;
8399 dip
->orig_bio
= io_bio
;
8400 dip
->dio_bio
= dio_bio
;
8401 atomic_set(&dip
->pending_bios
, 0);
8402 btrfs_bio
= btrfs_io_bio(io_bio
);
8403 btrfs_bio
->logical
= file_offset
;
8406 io_bio
->bi_end_io
= btrfs_endio_direct_write
;
8408 io_bio
->bi_end_io
= btrfs_endio_direct_read
;
8409 dip
->subio_endio
= btrfs_subio_endio_read
;
8412 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
8416 if (btrfs_bio
->end_io
)
8417 btrfs_bio
->end_io(btrfs_bio
, ret
);
8421 * If we arrived here it means either we failed to submit the dip
8422 * or we either failed to clone the dio_bio or failed to allocate the
8423 * dip. If we cloned the dio_bio and allocated the dip, we can just
8424 * call bio_endio against our io_bio so that we get proper resource
8425 * cleanup if we fail to submit the dip, otherwise, we must do the
8426 * same as btrfs_endio_direct_[write|read] because we can't call these
8427 * callbacks - they require an allocated dip and a clone of dio_bio.
8429 if (io_bio
&& dip
) {
8430 io_bio
->bi_error
= -EIO
;
8433 * The end io callbacks free our dip, do the final put on io_bio
8434 * and all the cleanup and final put for dio_bio (through
8441 struct btrfs_ordered_extent
*ordered
;
8443 ordered
= btrfs_lookup_ordered_extent(inode
,
8445 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
8447 * Decrements our ref on the ordered extent and removes
8448 * the ordered extent from the inode's ordered tree,
8449 * doing all the proper resource cleanup such as for the
8450 * reserved space and waking up any waiters for this
8451 * ordered extent (through btrfs_remove_ordered_extent).
8453 btrfs_finish_ordered_io(ordered
);
8455 unlock_extent(&BTRFS_I(inode
)->io_tree
, file_offset
,
8456 file_offset
+ dio_bio
->bi_iter
.bi_size
- 1);
8458 dio_bio
->bi_error
= -EIO
;
8460 * Releases and cleans up our dio_bio, no need to bio_put()
8461 * nor bio_endio()/bio_io_error() against dio_bio.
8463 dio_end_io(dio_bio
, ret
);
8470 static ssize_t
check_direct_IO(struct btrfs_root
*root
, struct kiocb
*iocb
,
8471 const struct iov_iter
*iter
, loff_t offset
)
8475 unsigned blocksize_mask
= root
->sectorsize
- 1;
8476 ssize_t retval
= -EINVAL
;
8478 if (offset
& blocksize_mask
)
8481 if (iov_iter_alignment(iter
) & blocksize_mask
)
8484 /* If this is a write we don't need to check anymore */
8485 if (iov_iter_rw(iter
) == WRITE
)
8488 * Check to make sure we don't have duplicate iov_base's in this
8489 * iovec, if so return EINVAL, otherwise we'll get csum errors
8490 * when reading back.
8492 for (seg
= 0; seg
< iter
->nr_segs
; seg
++) {
8493 for (i
= seg
+ 1; i
< iter
->nr_segs
; i
++) {
8494 if (iter
->iov
[seg
].iov_base
== iter
->iov
[i
].iov_base
)
8503 static ssize_t
btrfs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
,
8506 struct file
*file
= iocb
->ki_filp
;
8507 struct inode
*inode
= file
->f_mapping
->host
;
8508 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8509 struct btrfs_dio_data dio_data
= { 0 };
8513 bool relock
= false;
8516 if (check_direct_IO(BTRFS_I(inode
)->root
, iocb
, iter
, offset
))
8519 inode_dio_begin(inode
);
8520 smp_mb__after_atomic();
8523 * The generic stuff only does filemap_write_and_wait_range, which
8524 * isn't enough if we've written compressed pages to this area, so
8525 * we need to flush the dirty pages again to make absolutely sure
8526 * that any outstanding dirty pages are on disk.
8528 count
= iov_iter_count(iter
);
8529 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
8530 &BTRFS_I(inode
)->runtime_flags
))
8531 filemap_fdatawrite_range(inode
->i_mapping
, offset
,
8532 offset
+ count
- 1);
8534 if (iov_iter_rw(iter
) == WRITE
) {
8536 * If the write DIO is beyond the EOF, we need update
8537 * the isize, but it is protected by i_mutex. So we can
8538 * not unlock the i_mutex at this case.
8540 if (offset
+ count
<= inode
->i_size
) {
8541 mutex_unlock(&inode
->i_mutex
);
8544 ret
= btrfs_delalloc_reserve_space(inode
, offset
, count
);
8547 dio_data
.outstanding_extents
= div64_u64(count
+
8548 BTRFS_MAX_EXTENT_SIZE
- 1,
8549 BTRFS_MAX_EXTENT_SIZE
);
8552 * We need to know how many extents we reserved so that we can
8553 * do the accounting properly if we go over the number we
8554 * originally calculated. Abuse current->journal_info for this.
8556 dio_data
.reserve
= round_up(count
, root
->sectorsize
);
8557 current
->journal_info
= &dio_data
;
8558 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK
,
8559 &BTRFS_I(inode
)->runtime_flags
)) {
8560 inode_dio_end(inode
);
8561 flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
8565 ret
= __blockdev_direct_IO(iocb
, inode
,
8566 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
8567 iter
, offset
, btrfs_get_blocks_direct
, NULL
,
8568 btrfs_submit_direct
, flags
);
8569 if (iov_iter_rw(iter
) == WRITE
) {
8570 current
->journal_info
= NULL
;
8571 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
8572 if (dio_data
.reserve
)
8573 btrfs_delalloc_release_space(inode
, offset
,
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;
8687 int inode_evicting
= inode
->i_state
& I_FREEING
;
8690 * we have the page locked, so new writeback can't start,
8691 * and the dirty bit won't be cleared while we are here.
8693 * Wait for IO on this page so that we can safely clear
8694 * the PagePrivate2 bit and do ordered accounting
8696 wait_on_page_writeback(page
);
8698 tree
= &BTRFS_I(inode
)->io_tree
;
8700 btrfs_releasepage(page
, GFP_NOFS
);
8704 if (!inode_evicting
)
8705 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
);
8706 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
8709 * IO on this page will never be started, so we need
8710 * to account for any ordered extents now
8712 if (!inode_evicting
)
8713 clear_extent_bit(tree
, page_start
, page_end
,
8714 EXTENT_DIRTY
| EXTENT_DELALLOC
|
8715 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
8716 EXTENT_DEFRAG
, 1, 0, &cached_state
,
8719 * whoever cleared the private bit is responsible
8720 * for the finish_ordered_io
8722 if (TestClearPagePrivate2(page
)) {
8723 struct btrfs_ordered_inode_tree
*tree
;
8726 tree
= &BTRFS_I(inode
)->ordered_tree
;
8728 spin_lock_irq(&tree
->lock
);
8729 set_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
);
8730 new_len
= page_start
- ordered
->file_offset
;
8731 if (new_len
< ordered
->truncated_len
)
8732 ordered
->truncated_len
= new_len
;
8733 spin_unlock_irq(&tree
->lock
);
8735 if (btrfs_dec_test_ordered_pending(inode
, &ordered
,
8737 PAGE_CACHE_SIZE
, 1))
8738 btrfs_finish_ordered_io(ordered
);
8740 btrfs_put_ordered_extent(ordered
);
8741 if (!inode_evicting
) {
8742 cached_state
= NULL
;
8743 lock_extent_bits(tree
, page_start
, page_end
, 0,
8749 * Qgroup reserved space handler
8750 * Page here will be either
8751 * 1) Already written to disk
8752 * In this case, its reserved space is released from data rsv map
8753 * and will be freed by delayed_ref handler finally.
8754 * So even we call qgroup_free_data(), it won't decrease reserved
8756 * 2) Not written to disk
8757 * This means the reserved space should be freed here. However,
8758 * if a truncate invalidates the page (by clearing PageDirty)
8759 * and the page is accounted for while allocating extent
8760 * in btrfs_check_data_free_space() we let delayed_ref to
8761 * free the entire extent.
8763 if (PageDirty(page
))
8764 btrfs_qgroup_free_data(inode
, page_start
, PAGE_SIZE
);
8765 if (!inode_evicting
) {
8766 clear_extent_bit(tree
, page_start
, page_end
,
8767 EXTENT_LOCKED
| EXTENT_DIRTY
|
8768 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
8769 EXTENT_DEFRAG
, 1, 1,
8770 &cached_state
, GFP_NOFS
);
8772 __btrfs_releasepage(page
, GFP_NOFS
);
8775 ClearPageChecked(page
);
8776 if (PagePrivate(page
)) {
8777 ClearPagePrivate(page
);
8778 set_page_private(page
, 0);
8779 page_cache_release(page
);
8784 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8785 * called from a page fault handler when a page is first dirtied. Hence we must
8786 * be careful to check for EOF conditions here. We set the page up correctly
8787 * for a written page which means we get ENOSPC checking when writing into
8788 * holes and correct delalloc and unwritten extent mapping on filesystems that
8789 * support these features.
8791 * We are not allowed to take the i_mutex here so we have to play games to
8792 * protect against truncate races as the page could now be beyond EOF. Because
8793 * vmtruncate() writes the inode size before removing pages, once we have the
8794 * page lock we can determine safely if the page is beyond EOF. If it is not
8795 * beyond EOF, then the page is guaranteed safe against truncation until we
8798 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
8800 struct page
*page
= vmf
->page
;
8801 struct inode
*inode
= file_inode(vma
->vm_file
);
8802 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8803 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
8804 struct btrfs_ordered_extent
*ordered
;
8805 struct extent_state
*cached_state
= NULL
;
8807 unsigned long zero_start
;
8814 sb_start_pagefault(inode
->i_sb
);
8815 page_start
= page_offset(page
);
8816 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
8818 ret
= btrfs_delalloc_reserve_space(inode
, page_start
,
8821 ret
= file_update_time(vma
->vm_file
);
8827 else /* -ENOSPC, -EIO, etc */
8828 ret
= VM_FAULT_SIGBUS
;
8834 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
8837 size
= i_size_read(inode
);
8839 if ((page
->mapping
!= inode
->i_mapping
) ||
8840 (page_start
>= size
)) {
8841 /* page got truncated out from underneath us */
8844 wait_on_page_writeback(page
);
8846 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
8847 set_page_extent_mapped(page
);
8850 * we can't set the delalloc bits if there are pending ordered
8851 * extents. Drop our locks and wait for them to finish
8853 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
8855 unlock_extent_cached(io_tree
, page_start
, page_end
,
8856 &cached_state
, GFP_NOFS
);
8858 btrfs_start_ordered_extent(inode
, ordered
, 1);
8859 btrfs_put_ordered_extent(ordered
);
8864 * XXX - page_mkwrite gets called every time the page is dirtied, even
8865 * if it was already dirty, so for space accounting reasons we need to
8866 * clear any delalloc bits for the range we are fixing to save. There
8867 * is probably a better way to do this, but for now keep consistent with
8868 * prepare_pages in the normal write path.
8870 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
8871 EXTENT_DIRTY
| EXTENT_DELALLOC
|
8872 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
8873 0, 0, &cached_state
, GFP_NOFS
);
8875 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
8878 unlock_extent_cached(io_tree
, page_start
, page_end
,
8879 &cached_state
, GFP_NOFS
);
8880 ret
= VM_FAULT_SIGBUS
;
8885 /* page is wholly or partially inside EOF */
8886 if (page_start
+ PAGE_CACHE_SIZE
> size
)
8887 zero_start
= size
& ~PAGE_CACHE_MASK
;
8889 zero_start
= PAGE_CACHE_SIZE
;
8891 if (zero_start
!= PAGE_CACHE_SIZE
) {
8893 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
8894 flush_dcache_page(page
);
8897 ClearPageChecked(page
);
8898 set_page_dirty(page
);
8899 SetPageUptodate(page
);
8901 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
8902 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
8903 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
8905 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
8909 sb_end_pagefault(inode
->i_sb
);
8910 return VM_FAULT_LOCKED
;
8914 btrfs_delalloc_release_space(inode
, page_start
, PAGE_CACHE_SIZE
);
8916 sb_end_pagefault(inode
->i_sb
);
8920 static int btrfs_truncate(struct inode
*inode
)
8922 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8923 struct btrfs_block_rsv
*rsv
;
8926 struct btrfs_trans_handle
*trans
;
8927 u64 mask
= root
->sectorsize
- 1;
8928 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
8930 ret
= btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
),
8936 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8937 * 3 things going on here
8939 * 1) We need to reserve space for our orphan item and the space to
8940 * delete our orphan item. Lord knows we don't want to have a dangling
8941 * orphan item because we didn't reserve space to remove it.
8943 * 2) We need to reserve space to update our inode.
8945 * 3) We need to have something to cache all the space that is going to
8946 * be free'd up by the truncate operation, but also have some slack
8947 * space reserved in case it uses space during the truncate (thank you
8948 * very much snapshotting).
8950 * And we need these to all be seperate. The fact is we can use alot of
8951 * space doing the truncate, and we have no earthly idea how much space
8952 * we will use, so we need the truncate reservation to be seperate so it
8953 * doesn't end up using space reserved for updating the inode or
8954 * removing the orphan item. We also need to be able to stop the
8955 * transaction and start a new one, which means we need to be able to
8956 * update the inode several times, and we have no idea of knowing how
8957 * many times that will be, so we can't just reserve 1 item for the
8958 * entirety of the opration, so that has to be done seperately as well.
8959 * Then there is the orphan item, which does indeed need to be held on
8960 * to for the whole operation, and we need nobody to touch this reserved
8961 * space except the orphan code.
8963 * So that leaves us with
8965 * 1) root->orphan_block_rsv - for the orphan deletion.
8966 * 2) rsv - for the truncate reservation, which we will steal from the
8967 * transaction reservation.
8968 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8969 * updating the inode.
8971 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
8974 rsv
->size
= min_size
;
8978 * 1 for the truncate slack space
8979 * 1 for updating the inode.
8981 trans
= btrfs_start_transaction(root
, 2);
8982 if (IS_ERR(trans
)) {
8983 err
= PTR_ERR(trans
);
8987 /* Migrate the slack space for the truncate to our reserve */
8988 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
8993 * So if we truncate and then write and fsync we normally would just
8994 * write the extents that changed, which is a problem if we need to
8995 * first truncate that entire inode. So set this flag so we write out
8996 * all of the extents in the inode to the sync log so we're completely
8999 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
9000 trans
->block_rsv
= rsv
;
9003 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
9005 BTRFS_EXTENT_DATA_KEY
);
9006 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
) {
9011 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
9012 ret
= btrfs_update_inode(trans
, root
, inode
);
9018 btrfs_end_transaction(trans
, root
);
9019 btrfs_btree_balance_dirty(root
);
9021 trans
= btrfs_start_transaction(root
, 2);
9022 if (IS_ERR(trans
)) {
9023 ret
= err
= PTR_ERR(trans
);
9028 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
,
9030 BUG_ON(ret
); /* shouldn't happen */
9031 trans
->block_rsv
= rsv
;
9034 if (ret
== 0 && inode
->i_nlink
> 0) {
9035 trans
->block_rsv
= root
->orphan_block_rsv
;
9036 ret
= btrfs_orphan_del(trans
, inode
);
9042 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
9043 ret
= btrfs_update_inode(trans
, root
, inode
);
9047 ret
= btrfs_end_transaction(trans
, root
);
9048 btrfs_btree_balance_dirty(root
);
9052 btrfs_free_block_rsv(root
, rsv
);
9061 * create a new subvolume directory/inode (helper for the ioctl).
9063 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
9064 struct btrfs_root
*new_root
,
9065 struct btrfs_root
*parent_root
,
9068 struct inode
*inode
;
9072 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
9073 new_dirid
, new_dirid
,
9074 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
9077 return PTR_ERR(inode
);
9078 inode
->i_op
= &btrfs_dir_inode_operations
;
9079 inode
->i_fop
= &btrfs_dir_file_operations
;
9081 set_nlink(inode
, 1);
9082 btrfs_i_size_write(inode
, 0);
9083 unlock_new_inode(inode
);
9085 err
= btrfs_subvol_inherit_props(trans
, new_root
, parent_root
);
9087 btrfs_err(new_root
->fs_info
,
9088 "error inheriting subvolume %llu properties: %d",
9089 new_root
->root_key
.objectid
, err
);
9091 err
= btrfs_update_inode(trans
, new_root
, inode
);
9097 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
9099 struct btrfs_inode
*ei
;
9100 struct inode
*inode
;
9102 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
9109 ei
->last_sub_trans
= 0;
9110 ei
->logged_trans
= 0;
9111 ei
->delalloc_bytes
= 0;
9112 ei
->defrag_bytes
= 0;
9113 ei
->disk_i_size
= 0;
9116 ei
->index_cnt
= (u64
)-1;
9118 ei
->last_unlink_trans
= 0;
9119 ei
->last_log_commit
= 0;
9121 spin_lock_init(&ei
->lock
);
9122 ei
->outstanding_extents
= 0;
9123 ei
->reserved_extents
= 0;
9125 ei
->runtime_flags
= 0;
9126 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
9128 ei
->delayed_node
= NULL
;
9130 ei
->i_otime
.tv_sec
= 0;
9131 ei
->i_otime
.tv_nsec
= 0;
9133 inode
= &ei
->vfs_inode
;
9134 extent_map_tree_init(&ei
->extent_tree
);
9135 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
9136 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
9137 ei
->io_tree
.track_uptodate
= 1;
9138 ei
->io_failure_tree
.track_uptodate
= 1;
9139 atomic_set(&ei
->sync_writers
, 0);
9140 mutex_init(&ei
->log_mutex
);
9141 mutex_init(&ei
->delalloc_mutex
);
9142 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
9143 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
9144 RB_CLEAR_NODE(&ei
->rb_node
);
9149 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9150 void btrfs_test_destroy_inode(struct inode
*inode
)
9152 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
9153 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9157 static void btrfs_i_callback(struct rcu_head
*head
)
9159 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
9160 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9163 void btrfs_destroy_inode(struct inode
*inode
)
9165 struct btrfs_ordered_extent
*ordered
;
9166 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9168 WARN_ON(!hlist_empty(&inode
->i_dentry
));
9169 WARN_ON(inode
->i_data
.nrpages
);
9170 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
9171 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
9172 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
9173 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
9174 WARN_ON(BTRFS_I(inode
)->defrag_bytes
);
9177 * This can happen where we create an inode, but somebody else also
9178 * created the same inode and we need to destroy the one we already
9184 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
9185 &BTRFS_I(inode
)->runtime_flags
)) {
9186 btrfs_info(root
->fs_info
, "inode %llu still on the orphan list",
9188 atomic_dec(&root
->orphan_inodes
);
9192 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
9196 btrfs_err(root
->fs_info
, "found ordered extent %llu %llu on inode cleanup",
9197 ordered
->file_offset
, ordered
->len
);
9198 btrfs_remove_ordered_extent(inode
, ordered
);
9199 btrfs_put_ordered_extent(ordered
);
9200 btrfs_put_ordered_extent(ordered
);
9203 btrfs_qgroup_check_reserved_leak(inode
);
9204 inode_tree_del(inode
);
9205 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
9207 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
9210 int btrfs_drop_inode(struct inode
*inode
)
9212 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9217 /* the snap/subvol tree is on deleting */
9218 if (btrfs_root_refs(&root
->root_item
) == 0)
9221 return generic_drop_inode(inode
);
9224 static void init_once(void *foo
)
9226 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
9228 inode_init_once(&ei
->vfs_inode
);
9231 void btrfs_destroy_cachep(void)
9234 * Make sure all delayed rcu free inodes are flushed before we
9238 if (btrfs_inode_cachep
)
9239 kmem_cache_destroy(btrfs_inode_cachep
);
9240 if (btrfs_trans_handle_cachep
)
9241 kmem_cache_destroy(btrfs_trans_handle_cachep
);
9242 if (btrfs_transaction_cachep
)
9243 kmem_cache_destroy(btrfs_transaction_cachep
);
9244 if (btrfs_path_cachep
)
9245 kmem_cache_destroy(btrfs_path_cachep
);
9246 if (btrfs_free_space_cachep
)
9247 kmem_cache_destroy(btrfs_free_space_cachep
);
9248 if (btrfs_delalloc_work_cachep
)
9249 kmem_cache_destroy(btrfs_delalloc_work_cachep
);
9252 int btrfs_init_cachep(void)
9254 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
9255 sizeof(struct btrfs_inode
), 0,
9256 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
9257 if (!btrfs_inode_cachep
)
9260 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
9261 sizeof(struct btrfs_trans_handle
), 0,
9262 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9263 if (!btrfs_trans_handle_cachep
)
9266 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction",
9267 sizeof(struct btrfs_transaction
), 0,
9268 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9269 if (!btrfs_transaction_cachep
)
9272 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
9273 sizeof(struct btrfs_path
), 0,
9274 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9275 if (!btrfs_path_cachep
)
9278 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
9279 sizeof(struct btrfs_free_space
), 0,
9280 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9281 if (!btrfs_free_space_cachep
)
9284 btrfs_delalloc_work_cachep
= kmem_cache_create("btrfs_delalloc_work",
9285 sizeof(struct btrfs_delalloc_work
), 0,
9286 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
9288 if (!btrfs_delalloc_work_cachep
)
9293 btrfs_destroy_cachep();
9297 static int btrfs_getattr(struct vfsmount
*mnt
,
9298 struct dentry
*dentry
, struct kstat
*stat
)
9301 struct inode
*inode
= d_inode(dentry
);
9302 u32 blocksize
= inode
->i_sb
->s_blocksize
;
9304 generic_fillattr(inode
, stat
);
9305 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
9306 stat
->blksize
= PAGE_CACHE_SIZE
;
9308 spin_lock(&BTRFS_I(inode
)->lock
);
9309 delalloc_bytes
= BTRFS_I(inode
)->delalloc_bytes
;
9310 spin_unlock(&BTRFS_I(inode
)->lock
);
9311 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
9312 ALIGN(delalloc_bytes
, blocksize
)) >> 9;
9316 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
9317 struct inode
*new_dir
, struct dentry
*new_dentry
)
9319 struct btrfs_trans_handle
*trans
;
9320 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9321 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9322 struct inode
*new_inode
= d_inode(new_dentry
);
9323 struct inode
*old_inode
= d_inode(old_dentry
);
9324 struct timespec ctime
= CURRENT_TIME
;
9328 u64 old_ino
= btrfs_ino(old_inode
);
9330 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
9333 /* we only allow rename subvolume link between subvolumes */
9334 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9337 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
9338 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
9341 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
9342 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
9346 /* check for collisions, even if the name isn't there */
9347 ret
= btrfs_check_dir_item_collision(dest
, new_dir
->i_ino
,
9348 new_dentry
->d_name
.name
,
9349 new_dentry
->d_name
.len
);
9352 if (ret
== -EEXIST
) {
9354 * eexist without a new_inode */
9355 if (WARN_ON(!new_inode
)) {
9359 /* maybe -EOVERFLOW */
9366 * we're using rename to replace one file with another. Start IO on it
9367 * now so we don't add too much work to the end of the transaction
9369 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
)
9370 filemap_flush(old_inode
->i_mapping
);
9372 /* close the racy window with snapshot create/destroy ioctl */
9373 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9374 down_read(&root
->fs_info
->subvol_sem
);
9376 * We want to reserve the absolute worst case amount of items. So if
9377 * both inodes are subvols and we need to unlink them then that would
9378 * require 4 item modifications, but if they are both normal inodes it
9379 * would require 5 item modifications, so we'll assume their normal
9380 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9381 * should cover the worst case number of items we'll modify.
9383 trans
= btrfs_start_transaction(root
, 11);
9384 if (IS_ERR(trans
)) {
9385 ret
= PTR_ERR(trans
);
9390 btrfs_record_root_in_trans(trans
, dest
);
9392 ret
= btrfs_set_inode_index(new_dir
, &index
);
9396 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9397 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9398 /* force full log commit if subvolume involved. */
9399 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9401 ret
= btrfs_insert_inode_ref(trans
, dest
,
9402 new_dentry
->d_name
.name
,
9403 new_dentry
->d_name
.len
,
9405 btrfs_ino(new_dir
), index
);
9409 * this is an ugly little race, but the rename is required
9410 * to make sure that if we crash, the inode is either at the
9411 * old name or the new one. pinning the log transaction lets
9412 * us make sure we don't allow a log commit to come in after
9413 * we unlink the name but before we add the new name back in.
9415 btrfs_pin_log_trans(root
);
9418 inode_inc_iversion(old_dir
);
9419 inode_inc_iversion(new_dir
);
9420 inode_inc_iversion(old_inode
);
9421 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
9422 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
9423 old_inode
->i_ctime
= ctime
;
9425 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
9426 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
9428 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9429 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
9430 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
9431 old_dentry
->d_name
.name
,
9432 old_dentry
->d_name
.len
);
9434 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
9435 d_inode(old_dentry
),
9436 old_dentry
->d_name
.name
,
9437 old_dentry
->d_name
.len
);
9439 ret
= btrfs_update_inode(trans
, root
, old_inode
);
9442 btrfs_abort_transaction(trans
, root
, ret
);
9447 inode_inc_iversion(new_inode
);
9448 new_inode
->i_ctime
= CURRENT_TIME
;
9449 if (unlikely(btrfs_ino(new_inode
) ==
9450 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
9451 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
9452 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
9454 new_dentry
->d_name
.name
,
9455 new_dentry
->d_name
.len
);
9456 BUG_ON(new_inode
->i_nlink
== 0);
9458 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
9459 d_inode(new_dentry
),
9460 new_dentry
->d_name
.name
,
9461 new_dentry
->d_name
.len
);
9463 if (!ret
&& new_inode
->i_nlink
== 0)
9464 ret
= btrfs_orphan_add(trans
, d_inode(new_dentry
));
9466 btrfs_abort_transaction(trans
, root
, ret
);
9471 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
9472 new_dentry
->d_name
.name
,
9473 new_dentry
->d_name
.len
, 0, index
);
9475 btrfs_abort_transaction(trans
, root
, ret
);
9479 if (old_inode
->i_nlink
== 1)
9480 BTRFS_I(old_inode
)->dir_index
= index
;
9482 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
9483 struct dentry
*parent
= new_dentry
->d_parent
;
9484 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
9485 btrfs_end_log_trans(root
);
9488 btrfs_end_transaction(trans
, root
);
9490 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9491 up_read(&root
->fs_info
->subvol_sem
);
9496 static int btrfs_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
9497 struct inode
*new_dir
, struct dentry
*new_dentry
,
9500 if (flags
& ~RENAME_NOREPLACE
)
9503 return btrfs_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
9506 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
9508 struct btrfs_delalloc_work
*delalloc_work
;
9509 struct inode
*inode
;
9511 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
9513 inode
= delalloc_work
->inode
;
9514 if (delalloc_work
->wait
) {
9515 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
9517 filemap_flush(inode
->i_mapping
);
9518 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
9519 &BTRFS_I(inode
)->runtime_flags
))
9520 filemap_flush(inode
->i_mapping
);
9523 if (delalloc_work
->delay_iput
)
9524 btrfs_add_delayed_iput(inode
);
9527 complete(&delalloc_work
->completion
);
9530 struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
,
9531 int wait
, int delay_iput
)
9533 struct btrfs_delalloc_work
*work
;
9535 work
= kmem_cache_zalloc(btrfs_delalloc_work_cachep
, GFP_NOFS
);
9539 init_completion(&work
->completion
);
9540 INIT_LIST_HEAD(&work
->list
);
9541 work
->inode
= inode
;
9543 work
->delay_iput
= delay_iput
;
9544 WARN_ON_ONCE(!inode
);
9545 btrfs_init_work(&work
->work
, btrfs_flush_delalloc_helper
,
9546 btrfs_run_delalloc_work
, NULL
, NULL
);
9551 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work
*work
)
9553 wait_for_completion(&work
->completion
);
9554 kmem_cache_free(btrfs_delalloc_work_cachep
, work
);
9558 * some fairly slow code that needs optimization. This walks the list
9559 * of all the inodes with pending delalloc and forces them to disk.
9561 static int __start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
,
9564 struct btrfs_inode
*binode
;
9565 struct inode
*inode
;
9566 struct btrfs_delalloc_work
*work
, *next
;
9567 struct list_head works
;
9568 struct list_head splice
;
9571 INIT_LIST_HEAD(&works
);
9572 INIT_LIST_HEAD(&splice
);
9574 mutex_lock(&root
->delalloc_mutex
);
9575 spin_lock(&root
->delalloc_lock
);
9576 list_splice_init(&root
->delalloc_inodes
, &splice
);
9577 while (!list_empty(&splice
)) {
9578 binode
= list_entry(splice
.next
, struct btrfs_inode
,
9581 list_move_tail(&binode
->delalloc_inodes
,
9582 &root
->delalloc_inodes
);
9583 inode
= igrab(&binode
->vfs_inode
);
9585 cond_resched_lock(&root
->delalloc_lock
);
9588 spin_unlock(&root
->delalloc_lock
);
9590 work
= btrfs_alloc_delalloc_work(inode
, 0, delay_iput
);
9593 btrfs_add_delayed_iput(inode
);
9599 list_add_tail(&work
->list
, &works
);
9600 btrfs_queue_work(root
->fs_info
->flush_workers
,
9603 if (nr
!= -1 && ret
>= nr
)
9606 spin_lock(&root
->delalloc_lock
);
9608 spin_unlock(&root
->delalloc_lock
);
9611 list_for_each_entry_safe(work
, next
, &works
, list
) {
9612 list_del_init(&work
->list
);
9613 btrfs_wait_and_free_delalloc_work(work
);
9616 if (!list_empty_careful(&splice
)) {
9617 spin_lock(&root
->delalloc_lock
);
9618 list_splice_tail(&splice
, &root
->delalloc_inodes
);
9619 spin_unlock(&root
->delalloc_lock
);
9621 mutex_unlock(&root
->delalloc_mutex
);
9625 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
9629 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
9632 ret
= __start_delalloc_inodes(root
, delay_iput
, -1);
9636 * the filemap_flush will queue IO into the worker threads, but
9637 * we have to make sure the IO is actually started and that
9638 * ordered extents get created before we return
9640 atomic_inc(&root
->fs_info
->async_submit_draining
);
9641 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
9642 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
9643 wait_event(root
->fs_info
->async_submit_wait
,
9644 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
9645 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
9647 atomic_dec(&root
->fs_info
->async_submit_draining
);
9651 int btrfs_start_delalloc_roots(struct btrfs_fs_info
*fs_info
, int delay_iput
,
9654 struct btrfs_root
*root
;
9655 struct list_head splice
;
9658 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
9661 INIT_LIST_HEAD(&splice
);
9663 mutex_lock(&fs_info
->delalloc_root_mutex
);
9664 spin_lock(&fs_info
->delalloc_root_lock
);
9665 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
9666 while (!list_empty(&splice
) && nr
) {
9667 root
= list_first_entry(&splice
, struct btrfs_root
,
9669 root
= btrfs_grab_fs_root(root
);
9671 list_move_tail(&root
->delalloc_root
,
9672 &fs_info
->delalloc_roots
);
9673 spin_unlock(&fs_info
->delalloc_root_lock
);
9675 ret
= __start_delalloc_inodes(root
, delay_iput
, nr
);
9676 btrfs_put_fs_root(root
);
9684 spin_lock(&fs_info
->delalloc_root_lock
);
9686 spin_unlock(&fs_info
->delalloc_root_lock
);
9689 atomic_inc(&fs_info
->async_submit_draining
);
9690 while (atomic_read(&fs_info
->nr_async_submits
) ||
9691 atomic_read(&fs_info
->async_delalloc_pages
)) {
9692 wait_event(fs_info
->async_submit_wait
,
9693 (atomic_read(&fs_info
->nr_async_submits
) == 0 &&
9694 atomic_read(&fs_info
->async_delalloc_pages
) == 0));
9696 atomic_dec(&fs_info
->async_submit_draining
);
9698 if (!list_empty_careful(&splice
)) {
9699 spin_lock(&fs_info
->delalloc_root_lock
);
9700 list_splice_tail(&splice
, &fs_info
->delalloc_roots
);
9701 spin_unlock(&fs_info
->delalloc_root_lock
);
9703 mutex_unlock(&fs_info
->delalloc_root_mutex
);
9707 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
9708 const char *symname
)
9710 struct btrfs_trans_handle
*trans
;
9711 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
9712 struct btrfs_path
*path
;
9713 struct btrfs_key key
;
9714 struct inode
*inode
= NULL
;
9722 struct btrfs_file_extent_item
*ei
;
9723 struct extent_buffer
*leaf
;
9725 name_len
= strlen(symname
);
9726 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
9727 return -ENAMETOOLONG
;
9730 * 2 items for inode item and ref
9731 * 2 items for dir items
9732 * 1 item for updating parent inode item
9733 * 1 item for the inline extent item
9734 * 1 item for xattr if selinux is on
9736 trans
= btrfs_start_transaction(root
, 7);
9738 return PTR_ERR(trans
);
9740 err
= btrfs_find_free_ino(root
, &objectid
);
9744 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
9745 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
9746 S_IFLNK
|S_IRWXUGO
, &index
);
9747 if (IS_ERR(inode
)) {
9748 err
= PTR_ERR(inode
);
9753 * If the active LSM wants to access the inode during
9754 * d_instantiate it needs these. Smack checks to see
9755 * if the filesystem supports xattrs by looking at the
9758 inode
->i_fop
= &btrfs_file_operations
;
9759 inode
->i_op
= &btrfs_file_inode_operations
;
9760 inode
->i_mapping
->a_ops
= &btrfs_aops
;
9761 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
9763 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
9765 goto out_unlock_inode
;
9767 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
9769 goto out_unlock_inode
;
9771 path
= btrfs_alloc_path();
9774 goto out_unlock_inode
;
9776 key
.objectid
= btrfs_ino(inode
);
9778 key
.type
= BTRFS_EXTENT_DATA_KEY
;
9779 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
9780 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
9783 btrfs_free_path(path
);
9784 goto out_unlock_inode
;
9786 leaf
= path
->nodes
[0];
9787 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
9788 struct btrfs_file_extent_item
);
9789 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
9790 btrfs_set_file_extent_type(leaf
, ei
,
9791 BTRFS_FILE_EXTENT_INLINE
);
9792 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
9793 btrfs_set_file_extent_compression(leaf
, ei
, 0);
9794 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
9795 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
9797 ptr
= btrfs_file_extent_inline_start(ei
);
9798 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
9799 btrfs_mark_buffer_dirty(leaf
);
9800 btrfs_free_path(path
);
9802 inode
->i_op
= &btrfs_symlink_inode_operations
;
9803 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
9804 inode_set_bytes(inode
, name_len
);
9805 btrfs_i_size_write(inode
, name_len
);
9806 err
= btrfs_update_inode(trans
, root
, inode
);
9809 goto out_unlock_inode
;
9812 d_instantiate_new(dentry
, inode
);
9815 btrfs_end_transaction(trans
, root
);
9817 inode_dec_link_count(inode
);
9820 btrfs_btree_balance_dirty(root
);
9825 unlock_new_inode(inode
);
9829 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
9830 u64 start
, u64 num_bytes
, u64 min_size
,
9831 loff_t actual_len
, u64
*alloc_hint
,
9832 struct btrfs_trans_handle
*trans
)
9834 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
9835 struct extent_map
*em
;
9836 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9837 struct btrfs_key ins
;
9838 u64 cur_offset
= start
;
9841 u64 last_alloc
= (u64
)-1;
9843 bool own_trans
= true;
9847 while (num_bytes
> 0) {
9849 trans
= btrfs_start_transaction(root
, 3);
9850 if (IS_ERR(trans
)) {
9851 ret
= PTR_ERR(trans
);
9856 cur_bytes
= min(num_bytes
, 256ULL * 1024 * 1024);
9857 cur_bytes
= max(cur_bytes
, min_size
);
9859 * If we are severely fragmented we could end up with really
9860 * small allocations, so if the allocator is returning small
9861 * chunks lets make its job easier by only searching for those
9864 cur_bytes
= min(cur_bytes
, last_alloc
);
9865 ret
= btrfs_reserve_extent(root
, cur_bytes
, min_size
, 0,
9866 *alloc_hint
, &ins
, 1, 0);
9869 btrfs_end_transaction(trans
, root
);
9873 last_alloc
= ins
.offset
;
9874 ret
= insert_reserved_file_extent(trans
, inode
,
9875 cur_offset
, ins
.objectid
,
9876 ins
.offset
, ins
.offset
,
9877 ins
.offset
, 0, 0, 0,
9878 BTRFS_FILE_EXTENT_PREALLOC
);
9880 btrfs_free_reserved_extent(root
, ins
.objectid
,
9882 btrfs_abort_transaction(trans
, root
, ret
);
9884 btrfs_end_transaction(trans
, root
);
9888 btrfs_drop_extent_cache(inode
, cur_offset
,
9889 cur_offset
+ ins
.offset
-1, 0);
9891 em
= alloc_extent_map();
9893 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
9894 &BTRFS_I(inode
)->runtime_flags
);
9898 em
->start
= cur_offset
;
9899 em
->orig_start
= cur_offset
;
9900 em
->len
= ins
.offset
;
9901 em
->block_start
= ins
.objectid
;
9902 em
->block_len
= ins
.offset
;
9903 em
->orig_block_len
= ins
.offset
;
9904 em
->ram_bytes
= ins
.offset
;
9905 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
9906 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
9907 em
->generation
= trans
->transid
;
9910 write_lock(&em_tree
->lock
);
9911 ret
= add_extent_mapping(em_tree
, em
, 1);
9912 write_unlock(&em_tree
->lock
);
9915 btrfs_drop_extent_cache(inode
, cur_offset
,
9916 cur_offset
+ ins
.offset
- 1,
9919 free_extent_map(em
);
9921 num_bytes
-= ins
.offset
;
9922 cur_offset
+= ins
.offset
;
9923 *alloc_hint
= ins
.objectid
+ ins
.offset
;
9925 inode_inc_iversion(inode
);
9926 inode
->i_ctime
= CURRENT_TIME
;
9927 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
9928 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
9929 (actual_len
> inode
->i_size
) &&
9930 (cur_offset
> inode
->i_size
)) {
9931 if (cur_offset
> actual_len
)
9932 i_size
= actual_len
;
9934 i_size
= cur_offset
;
9935 i_size_write(inode
, i_size
);
9936 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
9939 ret
= btrfs_update_inode(trans
, root
, inode
);
9942 btrfs_abort_transaction(trans
, root
, ret
);
9944 btrfs_end_transaction(trans
, root
);
9949 btrfs_end_transaction(trans
, root
);
9954 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
9955 u64 start
, u64 num_bytes
, u64 min_size
,
9956 loff_t actual_len
, u64
*alloc_hint
)
9958 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
9959 min_size
, actual_len
, alloc_hint
,
9963 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
9964 struct btrfs_trans_handle
*trans
, int mode
,
9965 u64 start
, u64 num_bytes
, u64 min_size
,
9966 loff_t actual_len
, u64
*alloc_hint
)
9968 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
9969 min_size
, actual_len
, alloc_hint
, trans
);
9972 static int btrfs_set_page_dirty(struct page
*page
)
9974 return __set_page_dirty_nobuffers(page
);
9977 static int btrfs_permission(struct inode
*inode
, int mask
)
9979 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9980 umode_t mode
= inode
->i_mode
;
9982 if (mask
& MAY_WRITE
&&
9983 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
9984 if (btrfs_root_readonly(root
))
9986 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
9989 return generic_permission(inode
, mask
);
9992 static int btrfs_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
9994 struct btrfs_trans_handle
*trans
;
9995 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
9996 struct inode
*inode
= NULL
;
10002 * 5 units required for adding orphan entry
10004 trans
= btrfs_start_transaction(root
, 5);
10006 return PTR_ERR(trans
);
10008 ret
= btrfs_find_free_ino(root
, &objectid
);
10012 inode
= btrfs_new_inode(trans
, root
, dir
, NULL
, 0,
10013 btrfs_ino(dir
), objectid
, mode
, &index
);
10014 if (IS_ERR(inode
)) {
10015 ret
= PTR_ERR(inode
);
10020 inode
->i_fop
= &btrfs_file_operations
;
10021 inode
->i_op
= &btrfs_file_inode_operations
;
10023 inode
->i_mapping
->a_ops
= &btrfs_aops
;
10024 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
10026 ret
= btrfs_init_inode_security(trans
, inode
, dir
, NULL
);
10030 ret
= btrfs_update_inode(trans
, root
, inode
);
10033 ret
= btrfs_orphan_add(trans
, inode
);
10038 * We set number of links to 0 in btrfs_new_inode(), and here we set
10039 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10042 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10044 set_nlink(inode
, 1);
10045 unlock_new_inode(inode
);
10046 d_tmpfile(dentry
, inode
);
10047 mark_inode_dirty(inode
);
10050 btrfs_end_transaction(trans
, root
);
10053 btrfs_balance_delayed_items(root
);
10054 btrfs_btree_balance_dirty(root
);
10058 unlock_new_inode(inode
);
10063 /* Inspired by filemap_check_errors() */
10064 int btrfs_inode_check_errors(struct inode
*inode
)
10068 if (test_bit(AS_ENOSPC
, &inode
->i_mapping
->flags
) &&
10069 test_and_clear_bit(AS_ENOSPC
, &inode
->i_mapping
->flags
))
10071 if (test_bit(AS_EIO
, &inode
->i_mapping
->flags
) &&
10072 test_and_clear_bit(AS_EIO
, &inode
->i_mapping
->flags
))
10078 static const struct inode_operations btrfs_dir_inode_operations
= {
10079 .getattr
= btrfs_getattr
,
10080 .lookup
= btrfs_lookup
,
10081 .create
= btrfs_create
,
10082 .unlink
= btrfs_unlink
,
10083 .link
= btrfs_link
,
10084 .mkdir
= btrfs_mkdir
,
10085 .rmdir
= btrfs_rmdir
,
10086 .rename2
= btrfs_rename2
,
10087 .symlink
= btrfs_symlink
,
10088 .setattr
= btrfs_setattr
,
10089 .mknod
= btrfs_mknod
,
10090 .setxattr
= btrfs_setxattr
,
10091 .getxattr
= btrfs_getxattr
,
10092 .listxattr
= btrfs_listxattr
,
10093 .removexattr
= btrfs_removexattr
,
10094 .permission
= btrfs_permission
,
10095 .get_acl
= btrfs_get_acl
,
10096 .set_acl
= btrfs_set_acl
,
10097 .update_time
= btrfs_update_time
,
10098 .tmpfile
= btrfs_tmpfile
,
10100 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
10101 .lookup
= btrfs_lookup
,
10102 .permission
= btrfs_permission
,
10103 .get_acl
= btrfs_get_acl
,
10104 .set_acl
= btrfs_set_acl
,
10105 .update_time
= btrfs_update_time
,
10108 static const struct file_operations btrfs_dir_file_operations
= {
10109 .llseek
= generic_file_llseek
,
10110 .read
= generic_read_dir
,
10111 .iterate
= btrfs_real_readdir
,
10112 .unlocked_ioctl
= btrfs_ioctl
,
10113 #ifdef CONFIG_COMPAT
10114 .compat_ioctl
= btrfs_ioctl
,
10116 .release
= btrfs_release_file
,
10117 .fsync
= btrfs_sync_file
,
10120 static struct extent_io_ops btrfs_extent_io_ops
= {
10121 .fill_delalloc
= run_delalloc_range
,
10122 .submit_bio_hook
= btrfs_submit_bio_hook
,
10123 .merge_bio_hook
= btrfs_merge_bio_hook
,
10124 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
10125 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
10126 .writepage_start_hook
= btrfs_writepage_start_hook
,
10127 .set_bit_hook
= btrfs_set_bit_hook
,
10128 .clear_bit_hook
= btrfs_clear_bit_hook
,
10129 .merge_extent_hook
= btrfs_merge_extent_hook
,
10130 .split_extent_hook
= btrfs_split_extent_hook
,
10134 * btrfs doesn't support the bmap operation because swapfiles
10135 * use bmap to make a mapping of extents in the file. They assume
10136 * these extents won't change over the life of the file and they
10137 * use the bmap result to do IO directly to the drive.
10139 * the btrfs bmap call would return logical addresses that aren't
10140 * suitable for IO and they also will change frequently as COW
10141 * operations happen. So, swapfile + btrfs == corruption.
10143 * For now we're avoiding this by dropping bmap.
10145 static const struct address_space_operations btrfs_aops
= {
10146 .readpage
= btrfs_readpage
,
10147 .writepage
= btrfs_writepage
,
10148 .writepages
= btrfs_writepages
,
10149 .readpages
= btrfs_readpages
,
10150 .direct_IO
= btrfs_direct_IO
,
10151 .invalidatepage
= btrfs_invalidatepage
,
10152 .releasepage
= btrfs_releasepage
,
10153 .set_page_dirty
= btrfs_set_page_dirty
,
10154 .error_remove_page
= generic_error_remove_page
,
10157 static const struct address_space_operations btrfs_symlink_aops
= {
10158 .readpage
= btrfs_readpage
,
10159 .writepage
= btrfs_writepage
,
10160 .invalidatepage
= btrfs_invalidatepage
,
10161 .releasepage
= btrfs_releasepage
,
10164 static const struct inode_operations btrfs_file_inode_operations
= {
10165 .getattr
= btrfs_getattr
,
10166 .setattr
= btrfs_setattr
,
10167 .setxattr
= btrfs_setxattr
,
10168 .getxattr
= btrfs_getxattr
,
10169 .listxattr
= btrfs_listxattr
,
10170 .removexattr
= btrfs_removexattr
,
10171 .permission
= btrfs_permission
,
10172 .fiemap
= btrfs_fiemap
,
10173 .get_acl
= btrfs_get_acl
,
10174 .set_acl
= btrfs_set_acl
,
10175 .update_time
= btrfs_update_time
,
10177 static const struct inode_operations btrfs_special_inode_operations
= {
10178 .getattr
= btrfs_getattr
,
10179 .setattr
= btrfs_setattr
,
10180 .permission
= btrfs_permission
,
10181 .setxattr
= btrfs_setxattr
,
10182 .getxattr
= btrfs_getxattr
,
10183 .listxattr
= btrfs_listxattr
,
10184 .removexattr
= btrfs_removexattr
,
10185 .get_acl
= btrfs_get_acl
,
10186 .set_acl
= btrfs_set_acl
,
10187 .update_time
= btrfs_update_time
,
10189 static const struct inode_operations btrfs_symlink_inode_operations
= {
10190 .readlink
= generic_readlink
,
10191 .follow_link
= page_follow_link_light
,
10192 .put_link
= page_put_link
,
10193 .getattr
= btrfs_getattr
,
10194 .setattr
= btrfs_setattr
,
10195 .permission
= btrfs_permission
,
10196 .setxattr
= btrfs_setxattr
,
10197 .getxattr
= btrfs_getxattr
,
10198 .listxattr
= btrfs_listxattr
,
10199 .removexattr
= btrfs_removexattr
,
10200 .update_time
= btrfs_update_time
,
10203 const struct dentry_operations btrfs_dentry_operations
= {
10204 .d_delete
= btrfs_dentry_delete
,
10205 .d_release
= btrfs_dentry_release
,