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
);
1209 * when nowcow writeback call back. This checks for snapshots or COW copies
1210 * of the extents that exist in the file, and COWs the file as required.
1212 * If no cow copies or snapshots exist, we write directly to the existing
1215 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1216 struct page
*locked_page
,
1217 u64 start
, u64 end
, int *page_started
, int force
,
1218 unsigned long *nr_written
)
1220 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1221 struct btrfs_trans_handle
*trans
;
1222 struct extent_buffer
*leaf
;
1223 struct btrfs_path
*path
;
1224 struct btrfs_file_extent_item
*fi
;
1225 struct btrfs_key found_key
;
1240 u64 ino
= btrfs_ino(inode
);
1242 path
= btrfs_alloc_path();
1244 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1245 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1246 EXTENT_DO_ACCOUNTING
|
1247 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1249 PAGE_SET_WRITEBACK
|
1250 PAGE_END_WRITEBACK
);
1254 nolock
= btrfs_is_free_space_inode(inode
);
1257 trans
= btrfs_join_transaction_nolock(root
);
1259 trans
= btrfs_join_transaction(root
);
1261 if (IS_ERR(trans
)) {
1262 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1263 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1264 EXTENT_DO_ACCOUNTING
|
1265 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1267 PAGE_SET_WRITEBACK
|
1268 PAGE_END_WRITEBACK
);
1269 btrfs_free_path(path
);
1270 return PTR_ERR(trans
);
1273 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1275 cow_start
= (u64
)-1;
1278 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1282 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1283 leaf
= path
->nodes
[0];
1284 btrfs_item_key_to_cpu(leaf
, &found_key
,
1285 path
->slots
[0] - 1);
1286 if (found_key
.objectid
== ino
&&
1287 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1292 leaf
= path
->nodes
[0];
1293 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1294 ret
= btrfs_next_leaf(root
, path
);
1299 leaf
= path
->nodes
[0];
1305 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1307 if (found_key
.objectid
> ino
)
1309 if (WARN_ON_ONCE(found_key
.objectid
< ino
) ||
1310 found_key
.type
< BTRFS_EXTENT_DATA_KEY
) {
1314 if (found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1315 found_key
.offset
> end
)
1318 if (found_key
.offset
> cur_offset
) {
1319 extent_end
= found_key
.offset
;
1324 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1325 struct btrfs_file_extent_item
);
1326 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1328 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
1329 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1330 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1331 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1332 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1333 extent_end
= found_key
.offset
+
1334 btrfs_file_extent_num_bytes(leaf
, fi
);
1336 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1337 if (extent_end
<= start
) {
1341 if (disk_bytenr
== 0)
1343 if (btrfs_file_extent_compression(leaf
, fi
) ||
1344 btrfs_file_extent_encryption(leaf
, fi
) ||
1345 btrfs_file_extent_other_encoding(leaf
, fi
))
1347 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1349 if (btrfs_extent_readonly(root
, disk_bytenr
))
1351 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1353 extent_offset
, disk_bytenr
))
1355 disk_bytenr
+= extent_offset
;
1356 disk_bytenr
+= cur_offset
- found_key
.offset
;
1357 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1359 * if there are pending snapshots for this root,
1360 * we fall into common COW way.
1363 err
= btrfs_start_write_no_snapshoting(root
);
1368 * force cow if csum exists in the range.
1369 * this ensure that csum for a given extent are
1370 * either valid or do not exist.
1372 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1375 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1376 extent_end
= found_key
.offset
+
1377 btrfs_file_extent_inline_len(leaf
,
1378 path
->slots
[0], fi
);
1379 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1384 if (extent_end
<= start
) {
1386 if (!nolock
&& nocow
)
1387 btrfs_end_write_no_snapshoting(root
);
1391 if (cow_start
== (u64
)-1)
1392 cow_start
= cur_offset
;
1393 cur_offset
= extent_end
;
1394 if (cur_offset
> end
)
1400 btrfs_release_path(path
);
1401 if (cow_start
!= (u64
)-1) {
1402 ret
= cow_file_range(inode
, locked_page
,
1403 cow_start
, found_key
.offset
- 1,
1404 page_started
, nr_written
, 1);
1406 if (!nolock
&& nocow
)
1407 btrfs_end_write_no_snapshoting(root
);
1410 cow_start
= (u64
)-1;
1413 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1414 struct extent_map
*em
;
1415 struct extent_map_tree
*em_tree
;
1416 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1417 em
= alloc_extent_map();
1418 BUG_ON(!em
); /* -ENOMEM */
1419 em
->start
= cur_offset
;
1420 em
->orig_start
= found_key
.offset
- extent_offset
;
1421 em
->len
= num_bytes
;
1422 em
->block_len
= num_bytes
;
1423 em
->block_start
= disk_bytenr
;
1424 em
->orig_block_len
= disk_num_bytes
;
1425 em
->ram_bytes
= ram_bytes
;
1426 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1427 em
->mod_start
= em
->start
;
1428 em
->mod_len
= em
->len
;
1429 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1430 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
1431 em
->generation
= -1;
1433 write_lock(&em_tree
->lock
);
1434 ret
= add_extent_mapping(em_tree
, em
, 1);
1435 write_unlock(&em_tree
->lock
);
1436 if (ret
!= -EEXIST
) {
1437 free_extent_map(em
);
1440 btrfs_drop_extent_cache(inode
, em
->start
,
1441 em
->start
+ em
->len
- 1, 0);
1443 type
= BTRFS_ORDERED_PREALLOC
;
1445 type
= BTRFS_ORDERED_NOCOW
;
1448 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1449 num_bytes
, num_bytes
, type
);
1450 BUG_ON(ret
); /* -ENOMEM */
1452 if (root
->root_key
.objectid
==
1453 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1454 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1457 if (!nolock
&& nocow
)
1458 btrfs_end_write_no_snapshoting(root
);
1463 extent_clear_unlock_delalloc(inode
, cur_offset
,
1464 cur_offset
+ num_bytes
- 1,
1465 locked_page
, EXTENT_LOCKED
|
1466 EXTENT_DELALLOC
, PAGE_UNLOCK
|
1468 if (!nolock
&& nocow
)
1469 btrfs_end_write_no_snapshoting(root
);
1470 cur_offset
= extent_end
;
1471 if (cur_offset
> end
)
1474 btrfs_release_path(path
);
1476 if (cur_offset
<= end
&& cow_start
== (u64
)-1) {
1477 cow_start
= cur_offset
;
1481 if (cow_start
!= (u64
)-1) {
1482 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1483 page_started
, nr_written
, 1);
1489 err
= btrfs_end_transaction(trans
, root
);
1493 if (ret
&& cur_offset
< end
)
1494 extent_clear_unlock_delalloc(inode
, cur_offset
, end
,
1495 locked_page
, EXTENT_LOCKED
|
1496 EXTENT_DELALLOC
| EXTENT_DEFRAG
|
1497 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
1499 PAGE_SET_WRITEBACK
|
1500 PAGE_END_WRITEBACK
);
1501 btrfs_free_path(path
);
1505 static inline int need_force_cow(struct inode
*inode
, u64 start
, u64 end
)
1508 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
1509 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
))
1513 * @defrag_bytes is a hint value, no spinlock held here,
1514 * if is not zero, it means the file is defragging.
1515 * Force cow if given extent needs to be defragged.
1517 if (BTRFS_I(inode
)->defrag_bytes
&&
1518 test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1519 EXTENT_DEFRAG
, 0, NULL
))
1526 * extent_io.c call back to do delayed allocation processing
1528 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1529 u64 start
, u64 end
, int *page_started
,
1530 unsigned long *nr_written
)
1533 int force_cow
= need_force_cow(inode
, start
, end
);
1535 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
&& !force_cow
) {
1536 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1537 page_started
, 1, nr_written
);
1538 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
&& !force_cow
) {
1539 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1540 page_started
, 0, nr_written
);
1541 } else if (!inode_need_compress(inode
)) {
1542 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1543 page_started
, nr_written
, 1);
1545 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1546 &BTRFS_I(inode
)->runtime_flags
);
1547 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1548 page_started
, nr_written
);
1553 static void btrfs_split_extent_hook(struct inode
*inode
,
1554 struct extent_state
*orig
, u64 split
)
1558 /* not delalloc, ignore it */
1559 if (!(orig
->state
& EXTENT_DELALLOC
))
1562 size
= orig
->end
- orig
->start
+ 1;
1563 if (size
> BTRFS_MAX_EXTENT_SIZE
) {
1568 * See the explanation in btrfs_merge_extent_hook, the same
1569 * applies here, just in reverse.
1571 new_size
= orig
->end
- split
+ 1;
1572 num_extents
= div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1573 BTRFS_MAX_EXTENT_SIZE
);
1574 new_size
= split
- orig
->start
;
1575 num_extents
+= div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1576 BTRFS_MAX_EXTENT_SIZE
);
1577 if (div64_u64(size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1578 BTRFS_MAX_EXTENT_SIZE
) >= num_extents
)
1582 spin_lock(&BTRFS_I(inode
)->lock
);
1583 BTRFS_I(inode
)->outstanding_extents
++;
1584 spin_unlock(&BTRFS_I(inode
)->lock
);
1588 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1589 * extents so we can keep track of new extents that are just merged onto old
1590 * extents, such as when we are doing sequential writes, so we can properly
1591 * account for the metadata space we'll need.
1593 static void btrfs_merge_extent_hook(struct inode
*inode
,
1594 struct extent_state
*new,
1595 struct extent_state
*other
)
1597 u64 new_size
, old_size
;
1600 /* not delalloc, ignore it */
1601 if (!(other
->state
& EXTENT_DELALLOC
))
1604 if (new->start
> other
->start
)
1605 new_size
= new->end
- other
->start
+ 1;
1607 new_size
= other
->end
- new->start
+ 1;
1609 /* we're not bigger than the max, unreserve the space and go */
1610 if (new_size
<= BTRFS_MAX_EXTENT_SIZE
) {
1611 spin_lock(&BTRFS_I(inode
)->lock
);
1612 BTRFS_I(inode
)->outstanding_extents
--;
1613 spin_unlock(&BTRFS_I(inode
)->lock
);
1618 * We have to add up either side to figure out how many extents were
1619 * accounted for before we merged into one big extent. If the number of
1620 * extents we accounted for is <= the amount we need for the new range
1621 * then we can return, otherwise drop. Think of it like this
1625 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1626 * need 2 outstanding extents, on one side we have 1 and the other side
1627 * we have 1 so they are == and we can return. But in this case
1629 * [MAX_SIZE+4k][MAX_SIZE+4k]
1631 * Each range on their own accounts for 2 extents, but merged together
1632 * they are only 3 extents worth of accounting, so we need to drop in
1635 old_size
= other
->end
- other
->start
+ 1;
1636 num_extents
= div64_u64(old_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1637 BTRFS_MAX_EXTENT_SIZE
);
1638 old_size
= new->end
- new->start
+ 1;
1639 num_extents
+= div64_u64(old_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1640 BTRFS_MAX_EXTENT_SIZE
);
1642 if (div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1643 BTRFS_MAX_EXTENT_SIZE
) >= num_extents
)
1646 spin_lock(&BTRFS_I(inode
)->lock
);
1647 BTRFS_I(inode
)->outstanding_extents
--;
1648 spin_unlock(&BTRFS_I(inode
)->lock
);
1651 static void btrfs_add_delalloc_inodes(struct btrfs_root
*root
,
1652 struct inode
*inode
)
1654 spin_lock(&root
->delalloc_lock
);
1655 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1656 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1657 &root
->delalloc_inodes
);
1658 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1659 &BTRFS_I(inode
)->runtime_flags
);
1660 root
->nr_delalloc_inodes
++;
1661 if (root
->nr_delalloc_inodes
== 1) {
1662 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1663 BUG_ON(!list_empty(&root
->delalloc_root
));
1664 list_add_tail(&root
->delalloc_root
,
1665 &root
->fs_info
->delalloc_roots
);
1666 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1669 spin_unlock(&root
->delalloc_lock
);
1672 static void btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1673 struct inode
*inode
)
1675 spin_lock(&root
->delalloc_lock
);
1676 if (!list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1677 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1678 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1679 &BTRFS_I(inode
)->runtime_flags
);
1680 root
->nr_delalloc_inodes
--;
1681 if (!root
->nr_delalloc_inodes
) {
1682 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1683 BUG_ON(list_empty(&root
->delalloc_root
));
1684 list_del_init(&root
->delalloc_root
);
1685 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1688 spin_unlock(&root
->delalloc_lock
);
1692 * extent_io.c set_bit_hook, used to track delayed allocation
1693 * bytes in this file, and to maintain the list of inodes that
1694 * have pending delalloc work to be done.
1696 static void btrfs_set_bit_hook(struct inode
*inode
,
1697 struct extent_state
*state
, unsigned *bits
)
1700 if ((*bits
& EXTENT_DEFRAG
) && !(*bits
& EXTENT_DELALLOC
))
1703 * set_bit and clear bit hooks normally require _irqsave/restore
1704 * but in this case, we are only testing for the DELALLOC
1705 * bit, which is only set or cleared with irqs on
1707 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1708 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1709 u64 len
= state
->end
+ 1 - state
->start
;
1710 bool do_list
= !btrfs_is_free_space_inode(inode
);
1712 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1713 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1715 spin_lock(&BTRFS_I(inode
)->lock
);
1716 BTRFS_I(inode
)->outstanding_extents
++;
1717 spin_unlock(&BTRFS_I(inode
)->lock
);
1720 /* For sanity tests */
1721 if (btrfs_test_is_dummy_root(root
))
1724 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, len
,
1725 root
->fs_info
->delalloc_batch
);
1726 spin_lock(&BTRFS_I(inode
)->lock
);
1727 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1728 if (*bits
& EXTENT_DEFRAG
)
1729 BTRFS_I(inode
)->defrag_bytes
+= len
;
1730 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1731 &BTRFS_I(inode
)->runtime_flags
))
1732 btrfs_add_delalloc_inodes(root
, inode
);
1733 spin_unlock(&BTRFS_I(inode
)->lock
);
1738 * extent_io.c clear_bit_hook, see set_bit_hook for why
1740 static void btrfs_clear_bit_hook(struct inode
*inode
,
1741 struct extent_state
*state
,
1744 u64 len
= state
->end
+ 1 - state
->start
;
1745 u64 num_extents
= div64_u64(len
+ BTRFS_MAX_EXTENT_SIZE
-1,
1746 BTRFS_MAX_EXTENT_SIZE
);
1748 spin_lock(&BTRFS_I(inode
)->lock
);
1749 if ((state
->state
& EXTENT_DEFRAG
) && (*bits
& EXTENT_DEFRAG
))
1750 BTRFS_I(inode
)->defrag_bytes
-= len
;
1751 spin_unlock(&BTRFS_I(inode
)->lock
);
1754 * set_bit and clear bit hooks normally require _irqsave/restore
1755 * but in this case, we are only testing for the DELALLOC
1756 * bit, which is only set or cleared with irqs on
1758 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1759 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1760 bool do_list
= !btrfs_is_free_space_inode(inode
);
1762 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1763 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1764 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1765 spin_lock(&BTRFS_I(inode
)->lock
);
1766 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
1767 spin_unlock(&BTRFS_I(inode
)->lock
);
1771 * We don't reserve metadata space for space cache inodes so we
1772 * don't need to call dellalloc_release_metadata if there is an
1775 if (*bits
& EXTENT_DO_ACCOUNTING
&&
1776 root
!= root
->fs_info
->tree_root
)
1777 btrfs_delalloc_release_metadata(inode
, len
);
1779 /* For sanity tests. */
1780 if (btrfs_test_is_dummy_root(root
))
1783 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1784 && do_list
&& !(state
->state
& EXTENT_NORESERVE
))
1785 btrfs_free_reserved_data_space_noquota(inode
,
1788 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, -len
,
1789 root
->fs_info
->delalloc_batch
);
1790 spin_lock(&BTRFS_I(inode
)->lock
);
1791 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1792 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1793 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1794 &BTRFS_I(inode
)->runtime_flags
))
1795 btrfs_del_delalloc_inode(root
, inode
);
1796 spin_unlock(&BTRFS_I(inode
)->lock
);
1801 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1802 * we don't create bios that span stripes or chunks
1804 int btrfs_merge_bio_hook(int rw
, struct page
*page
, unsigned long offset
,
1805 size_t size
, struct bio
*bio
,
1806 unsigned long bio_flags
)
1808 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1809 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
1814 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1817 length
= bio
->bi_iter
.bi_size
;
1818 map_length
= length
;
1819 ret
= btrfs_map_block(root
->fs_info
, rw
, logical
,
1820 &map_length
, NULL
, 0);
1821 /* Will always return 0 with map_multi == NULL */
1823 if (map_length
< length
+ size
)
1829 * in order to insert checksums into the metadata in large chunks,
1830 * we wait until bio submission time. All the pages in the bio are
1831 * checksummed and sums are attached onto the ordered extent record.
1833 * At IO completion time the cums attached on the ordered extent record
1834 * are inserted into the btree
1836 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1837 struct bio
*bio
, int mirror_num
,
1838 unsigned long bio_flags
,
1841 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1844 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1845 BUG_ON(ret
); /* -ENOMEM */
1850 * in order to insert checksums into the metadata in large chunks,
1851 * we wait until bio submission time. All the pages in the bio are
1852 * checksummed and sums are attached onto the ordered extent record.
1854 * At IO completion time the cums attached on the ordered extent record
1855 * are inserted into the btree
1857 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1858 int mirror_num
, unsigned long bio_flags
,
1861 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1864 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1866 bio
->bi_error
= ret
;
1873 * extent_io.c submission hook. This does the right thing for csum calculation
1874 * on write, or reading the csums from the tree before a read
1876 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1877 int mirror_num
, unsigned long bio_flags
,
1880 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1881 enum btrfs_wq_endio_type metadata
= BTRFS_WQ_ENDIO_DATA
;
1884 int async
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
1886 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1888 if (btrfs_is_free_space_inode(inode
))
1889 metadata
= BTRFS_WQ_ENDIO_FREE_SPACE
;
1891 if (!(rw
& REQ_WRITE
)) {
1892 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, metadata
);
1896 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1897 ret
= btrfs_submit_compressed_read(inode
, bio
,
1901 } else if (!skip_sum
) {
1902 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1907 } else if (async
&& !skip_sum
) {
1908 /* csum items have already been cloned */
1909 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1911 /* we're doing a write, do the async checksumming */
1912 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1913 inode
, rw
, bio
, mirror_num
,
1914 bio_flags
, bio_offset
,
1915 __btrfs_submit_bio_start
,
1916 __btrfs_submit_bio_done
);
1918 } else if (!skip_sum
) {
1919 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1925 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1929 bio
->bi_error
= ret
;
1936 * given a list of ordered sums record them in the inode. This happens
1937 * at IO completion time based on sums calculated at bio submission time.
1939 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1940 struct inode
*inode
, u64 file_offset
,
1941 struct list_head
*list
)
1943 struct btrfs_ordered_sum
*sum
;
1945 list_for_each_entry(sum
, list
, list
) {
1946 trans
->adding_csums
= 1;
1947 btrfs_csum_file_blocks(trans
,
1948 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1949 trans
->adding_csums
= 0;
1954 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1955 struct extent_state
**cached_state
)
1957 WARN_ON((end
& (PAGE_CACHE_SIZE
- 1)) == 0);
1958 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1959 cached_state
, GFP_NOFS
);
1962 /* see btrfs_writepage_start_hook for details on why this is required */
1963 struct btrfs_writepage_fixup
{
1965 struct btrfs_work work
;
1968 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1970 struct btrfs_writepage_fixup
*fixup
;
1971 struct btrfs_ordered_extent
*ordered
;
1972 struct extent_state
*cached_state
= NULL
;
1974 struct inode
*inode
;
1979 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1983 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1984 ClearPageChecked(page
);
1988 inode
= page
->mapping
->host
;
1989 page_start
= page_offset(page
);
1990 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1992 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1995 /* already ordered? We're done */
1996 if (PagePrivate2(page
))
1999 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
2001 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
2002 page_end
, &cached_state
, GFP_NOFS
);
2004 btrfs_start_ordered_extent(inode
, ordered
, 1);
2005 btrfs_put_ordered_extent(ordered
);
2009 ret
= btrfs_delalloc_reserve_space(inode
, page_start
,
2012 mapping_set_error(page
->mapping
, ret
);
2013 end_extent_writepage(page
, ret
, page_start
, page_end
);
2014 ClearPageChecked(page
);
2018 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
2019 ClearPageChecked(page
);
2020 set_page_dirty(page
);
2022 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2023 &cached_state
, GFP_NOFS
);
2026 page_cache_release(page
);
2031 * There are a few paths in the higher layers of the kernel that directly
2032 * set the page dirty bit without asking the filesystem if it is a
2033 * good idea. This causes problems because we want to make sure COW
2034 * properly happens and the data=ordered rules are followed.
2036 * In our case any range that doesn't have the ORDERED bit set
2037 * hasn't been properly setup for IO. We kick off an async process
2038 * to fix it up. The async helper will wait for ordered extents, set
2039 * the delalloc bit and make it safe to write the page.
2041 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
2043 struct inode
*inode
= page
->mapping
->host
;
2044 struct btrfs_writepage_fixup
*fixup
;
2045 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2047 /* this page is properly in the ordered list */
2048 if (TestClearPagePrivate2(page
))
2051 if (PageChecked(page
))
2054 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
2058 SetPageChecked(page
);
2059 page_cache_get(page
);
2060 btrfs_init_work(&fixup
->work
, btrfs_fixup_helper
,
2061 btrfs_writepage_fixup_worker
, NULL
, NULL
);
2063 btrfs_queue_work(root
->fs_info
->fixup_workers
, &fixup
->work
);
2067 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
2068 struct inode
*inode
, u64 file_pos
,
2069 u64 disk_bytenr
, u64 disk_num_bytes
,
2070 u64 num_bytes
, u64 ram_bytes
,
2071 u8 compression
, u8 encryption
,
2072 u16 other_encoding
, int extent_type
)
2074 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2075 struct btrfs_file_extent_item
*fi
;
2076 struct btrfs_path
*path
;
2077 struct extent_buffer
*leaf
;
2078 struct btrfs_key ins
;
2079 int extent_inserted
= 0;
2082 path
= btrfs_alloc_path();
2087 * we may be replacing one extent in the tree with another.
2088 * The new extent is pinned in the extent map, and we don't want
2089 * to drop it from the cache until it is completely in the btree.
2091 * So, tell btrfs_drop_extents to leave this extent in the cache.
2092 * the caller is expected to unpin it and allow it to be merged
2095 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, file_pos
,
2096 file_pos
+ num_bytes
, NULL
, 0,
2097 1, sizeof(*fi
), &extent_inserted
);
2101 if (!extent_inserted
) {
2102 ins
.objectid
= btrfs_ino(inode
);
2103 ins
.offset
= file_pos
;
2104 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
2106 path
->leave_spinning
= 1;
2107 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
2112 leaf
= path
->nodes
[0];
2113 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2114 struct btrfs_file_extent_item
);
2115 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
2116 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
2117 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
2118 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
2119 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2120 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2121 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
2122 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
2123 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
2124 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
2126 btrfs_mark_buffer_dirty(leaf
);
2127 btrfs_release_path(path
);
2129 inode_add_bytes(inode
, num_bytes
);
2131 ins
.objectid
= disk_bytenr
;
2132 ins
.offset
= disk_num_bytes
;
2133 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2134 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
2135 root
->root_key
.objectid
,
2136 btrfs_ino(inode
), file_pos
,
2139 * Release the reserved range from inode dirty range map, as it is
2140 * already moved into delayed_ref_head
2142 btrfs_qgroup_release_data(inode
, file_pos
, ram_bytes
);
2144 btrfs_free_path(path
);
2149 /* snapshot-aware defrag */
2150 struct sa_defrag_extent_backref
{
2151 struct rb_node node
;
2152 struct old_sa_defrag_extent
*old
;
2161 struct old_sa_defrag_extent
{
2162 struct list_head list
;
2163 struct new_sa_defrag_extent
*new;
2172 struct new_sa_defrag_extent
{
2173 struct rb_root root
;
2174 struct list_head head
;
2175 struct btrfs_path
*path
;
2176 struct inode
*inode
;
2184 static int backref_comp(struct sa_defrag_extent_backref
*b1
,
2185 struct sa_defrag_extent_backref
*b2
)
2187 if (b1
->root_id
< b2
->root_id
)
2189 else if (b1
->root_id
> b2
->root_id
)
2192 if (b1
->inum
< b2
->inum
)
2194 else if (b1
->inum
> b2
->inum
)
2197 if (b1
->file_pos
< b2
->file_pos
)
2199 else if (b1
->file_pos
> b2
->file_pos
)
2203 * [------------------------------] ===> (a range of space)
2204 * |<--->| |<---->| =============> (fs/file tree A)
2205 * |<---------------------------->| ===> (fs/file tree B)
2207 * A range of space can refer to two file extents in one tree while
2208 * refer to only one file extent in another tree.
2210 * So we may process a disk offset more than one time(two extents in A)
2211 * and locate at the same extent(one extent in B), then insert two same
2212 * backrefs(both refer to the extent in B).
2217 static void backref_insert(struct rb_root
*root
,
2218 struct sa_defrag_extent_backref
*backref
)
2220 struct rb_node
**p
= &root
->rb_node
;
2221 struct rb_node
*parent
= NULL
;
2222 struct sa_defrag_extent_backref
*entry
;
2227 entry
= rb_entry(parent
, struct sa_defrag_extent_backref
, node
);
2229 ret
= backref_comp(backref
, entry
);
2233 p
= &(*p
)->rb_right
;
2236 rb_link_node(&backref
->node
, parent
, p
);
2237 rb_insert_color(&backref
->node
, root
);
2241 * Note the backref might has changed, and in this case we just return 0.
2243 static noinline
int record_one_backref(u64 inum
, u64 offset
, u64 root_id
,
2246 struct btrfs_file_extent_item
*extent
;
2247 struct btrfs_fs_info
*fs_info
;
2248 struct old_sa_defrag_extent
*old
= ctx
;
2249 struct new_sa_defrag_extent
*new = old
->new;
2250 struct btrfs_path
*path
= new->path
;
2251 struct btrfs_key key
;
2252 struct btrfs_root
*root
;
2253 struct sa_defrag_extent_backref
*backref
;
2254 struct extent_buffer
*leaf
;
2255 struct inode
*inode
= new->inode
;
2261 if (BTRFS_I(inode
)->root
->root_key
.objectid
== root_id
&&
2262 inum
== btrfs_ino(inode
))
2265 key
.objectid
= root_id
;
2266 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2267 key
.offset
= (u64
)-1;
2269 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2270 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2272 if (PTR_ERR(root
) == -ENOENT
)
2275 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2276 inum
, offset
, root_id
);
2277 return PTR_ERR(root
);
2280 key
.objectid
= inum
;
2281 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2282 if (offset
> (u64
)-1 << 32)
2285 key
.offset
= offset
;
2287 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2288 if (WARN_ON(ret
< 0))
2295 leaf
= path
->nodes
[0];
2296 slot
= path
->slots
[0];
2298 if (slot
>= btrfs_header_nritems(leaf
)) {
2299 ret
= btrfs_next_leaf(root
, path
);
2302 } else if (ret
> 0) {
2311 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2313 if (key
.objectid
> inum
)
2316 if (key
.objectid
< inum
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2319 extent
= btrfs_item_ptr(leaf
, slot
,
2320 struct btrfs_file_extent_item
);
2322 if (btrfs_file_extent_disk_bytenr(leaf
, extent
) != old
->bytenr
)
2326 * 'offset' refers to the exact key.offset,
2327 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2328 * (key.offset - extent_offset).
2330 if (key
.offset
!= offset
)
2333 extent_offset
= btrfs_file_extent_offset(leaf
, extent
);
2334 num_bytes
= btrfs_file_extent_num_bytes(leaf
, extent
);
2336 if (extent_offset
>= old
->extent_offset
+ old
->offset
+
2337 old
->len
|| extent_offset
+ num_bytes
<=
2338 old
->extent_offset
+ old
->offset
)
2343 backref
= kmalloc(sizeof(*backref
), GFP_NOFS
);
2349 backref
->root_id
= root_id
;
2350 backref
->inum
= inum
;
2351 backref
->file_pos
= offset
;
2352 backref
->num_bytes
= num_bytes
;
2353 backref
->extent_offset
= extent_offset
;
2354 backref
->generation
= btrfs_file_extent_generation(leaf
, extent
);
2356 backref_insert(&new->root
, backref
);
2359 btrfs_release_path(path
);
2364 static noinline
bool record_extent_backrefs(struct btrfs_path
*path
,
2365 struct new_sa_defrag_extent
*new)
2367 struct btrfs_fs_info
*fs_info
= BTRFS_I(new->inode
)->root
->fs_info
;
2368 struct old_sa_defrag_extent
*old
, *tmp
;
2373 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2374 ret
= iterate_inodes_from_logical(old
->bytenr
+
2375 old
->extent_offset
, fs_info
,
2376 path
, record_one_backref
,
2378 if (ret
< 0 && ret
!= -ENOENT
)
2381 /* no backref to be processed for this extent */
2383 list_del(&old
->list
);
2388 if (list_empty(&new->head
))
2394 static int relink_is_mergable(struct extent_buffer
*leaf
,
2395 struct btrfs_file_extent_item
*fi
,
2396 struct new_sa_defrag_extent
*new)
2398 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) != new->bytenr
)
2401 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2404 if (btrfs_file_extent_compression(leaf
, fi
) != new->compress_type
)
2407 if (btrfs_file_extent_encryption(leaf
, fi
) ||
2408 btrfs_file_extent_other_encoding(leaf
, fi
))
2415 * Note the backref might has changed, and in this case we just return 0.
2417 static noinline
int relink_extent_backref(struct btrfs_path
*path
,
2418 struct sa_defrag_extent_backref
*prev
,
2419 struct sa_defrag_extent_backref
*backref
)
2421 struct btrfs_file_extent_item
*extent
;
2422 struct btrfs_file_extent_item
*item
;
2423 struct btrfs_ordered_extent
*ordered
;
2424 struct btrfs_trans_handle
*trans
;
2425 struct btrfs_fs_info
*fs_info
;
2426 struct btrfs_root
*root
;
2427 struct btrfs_key key
;
2428 struct extent_buffer
*leaf
;
2429 struct old_sa_defrag_extent
*old
= backref
->old
;
2430 struct new_sa_defrag_extent
*new = old
->new;
2431 struct inode
*src_inode
= new->inode
;
2432 struct inode
*inode
;
2433 struct extent_state
*cached
= NULL
;
2442 if (prev
&& prev
->root_id
== backref
->root_id
&&
2443 prev
->inum
== backref
->inum
&&
2444 prev
->file_pos
+ prev
->num_bytes
== backref
->file_pos
)
2447 /* step 1: get root */
2448 key
.objectid
= backref
->root_id
;
2449 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2450 key
.offset
= (u64
)-1;
2452 fs_info
= BTRFS_I(src_inode
)->root
->fs_info
;
2453 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
2455 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2457 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2458 if (PTR_ERR(root
) == -ENOENT
)
2460 return PTR_ERR(root
);
2463 if (btrfs_root_readonly(root
)) {
2464 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2468 /* step 2: get inode */
2469 key
.objectid
= backref
->inum
;
2470 key
.type
= BTRFS_INODE_ITEM_KEY
;
2473 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
2474 if (IS_ERR(inode
)) {
2475 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2479 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2481 /* step 3: relink backref */
2482 lock_start
= backref
->file_pos
;
2483 lock_end
= backref
->file_pos
+ backref
->num_bytes
- 1;
2484 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2487 ordered
= btrfs_lookup_first_ordered_extent(inode
, lock_end
);
2489 btrfs_put_ordered_extent(ordered
);
2493 trans
= btrfs_join_transaction(root
);
2494 if (IS_ERR(trans
)) {
2495 ret
= PTR_ERR(trans
);
2499 key
.objectid
= backref
->inum
;
2500 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2501 key
.offset
= backref
->file_pos
;
2503 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2506 } else if (ret
> 0) {
2511 extent
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2512 struct btrfs_file_extent_item
);
2514 if (btrfs_file_extent_generation(path
->nodes
[0], extent
) !=
2515 backref
->generation
)
2518 btrfs_release_path(path
);
2520 start
= backref
->file_pos
;
2521 if (backref
->extent_offset
< old
->extent_offset
+ old
->offset
)
2522 start
+= old
->extent_offset
+ old
->offset
-
2523 backref
->extent_offset
;
2525 len
= min(backref
->extent_offset
+ backref
->num_bytes
,
2526 old
->extent_offset
+ old
->offset
+ old
->len
);
2527 len
-= max(backref
->extent_offset
, old
->extent_offset
+ old
->offset
);
2529 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
2534 key
.objectid
= btrfs_ino(inode
);
2535 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2538 path
->leave_spinning
= 1;
2540 struct btrfs_file_extent_item
*fi
;
2542 struct btrfs_key found_key
;
2544 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2549 leaf
= path
->nodes
[0];
2550 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2552 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2553 struct btrfs_file_extent_item
);
2554 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
2556 if (extent_len
+ found_key
.offset
== start
&&
2557 relink_is_mergable(leaf
, fi
, new)) {
2558 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2560 btrfs_mark_buffer_dirty(leaf
);
2561 inode_add_bytes(inode
, len
);
2567 btrfs_release_path(path
);
2572 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2575 btrfs_abort_transaction(trans
, root
, ret
);
2579 leaf
= path
->nodes
[0];
2580 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2581 struct btrfs_file_extent_item
);
2582 btrfs_set_file_extent_disk_bytenr(leaf
, item
, new->bytenr
);
2583 btrfs_set_file_extent_disk_num_bytes(leaf
, item
, new->disk_len
);
2584 btrfs_set_file_extent_offset(leaf
, item
, start
- new->file_pos
);
2585 btrfs_set_file_extent_num_bytes(leaf
, item
, len
);
2586 btrfs_set_file_extent_ram_bytes(leaf
, item
, new->len
);
2587 btrfs_set_file_extent_generation(leaf
, item
, trans
->transid
);
2588 btrfs_set_file_extent_type(leaf
, item
, BTRFS_FILE_EXTENT_REG
);
2589 btrfs_set_file_extent_compression(leaf
, item
, new->compress_type
);
2590 btrfs_set_file_extent_encryption(leaf
, item
, 0);
2591 btrfs_set_file_extent_other_encoding(leaf
, item
, 0);
2593 btrfs_mark_buffer_dirty(leaf
);
2594 inode_add_bytes(inode
, len
);
2595 btrfs_release_path(path
);
2597 ret
= btrfs_inc_extent_ref(trans
, root
, new->bytenr
,
2599 backref
->root_id
, backref
->inum
,
2600 new->file_pos
); /* start - extent_offset */
2602 btrfs_abort_transaction(trans
, root
, ret
);
2608 btrfs_release_path(path
);
2609 path
->leave_spinning
= 0;
2610 btrfs_end_transaction(trans
, root
);
2612 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2618 static void free_sa_defrag_extent(struct new_sa_defrag_extent
*new)
2620 struct old_sa_defrag_extent
*old
, *tmp
;
2625 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2631 static void relink_file_extents(struct new_sa_defrag_extent
*new)
2633 struct btrfs_path
*path
;
2634 struct sa_defrag_extent_backref
*backref
;
2635 struct sa_defrag_extent_backref
*prev
= NULL
;
2636 struct inode
*inode
;
2637 struct btrfs_root
*root
;
2638 struct rb_node
*node
;
2642 root
= BTRFS_I(inode
)->root
;
2644 path
= btrfs_alloc_path();
2648 if (!record_extent_backrefs(path
, new)) {
2649 btrfs_free_path(path
);
2652 btrfs_release_path(path
);
2655 node
= rb_first(&new->root
);
2658 rb_erase(node
, &new->root
);
2660 backref
= rb_entry(node
, struct sa_defrag_extent_backref
, node
);
2662 ret
= relink_extent_backref(path
, prev
, backref
);
2675 btrfs_free_path(path
);
2677 free_sa_defrag_extent(new);
2679 atomic_dec(&root
->fs_info
->defrag_running
);
2680 wake_up(&root
->fs_info
->transaction_wait
);
2683 static struct new_sa_defrag_extent
*
2684 record_old_file_extents(struct inode
*inode
,
2685 struct btrfs_ordered_extent
*ordered
)
2687 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2688 struct btrfs_path
*path
;
2689 struct btrfs_key key
;
2690 struct old_sa_defrag_extent
*old
;
2691 struct new_sa_defrag_extent
*new;
2694 new = kmalloc(sizeof(*new), GFP_NOFS
);
2699 new->file_pos
= ordered
->file_offset
;
2700 new->len
= ordered
->len
;
2701 new->bytenr
= ordered
->start
;
2702 new->disk_len
= ordered
->disk_len
;
2703 new->compress_type
= ordered
->compress_type
;
2704 new->root
= RB_ROOT
;
2705 INIT_LIST_HEAD(&new->head
);
2707 path
= btrfs_alloc_path();
2711 key
.objectid
= btrfs_ino(inode
);
2712 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2713 key
.offset
= new->file_pos
;
2715 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2718 if (ret
> 0 && path
->slots
[0] > 0)
2721 /* find out all the old extents for the file range */
2723 struct btrfs_file_extent_item
*extent
;
2724 struct extent_buffer
*l
;
2733 slot
= path
->slots
[0];
2735 if (slot
>= btrfs_header_nritems(l
)) {
2736 ret
= btrfs_next_leaf(root
, path
);
2744 btrfs_item_key_to_cpu(l
, &key
, slot
);
2746 if (key
.objectid
!= btrfs_ino(inode
))
2748 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2750 if (key
.offset
>= new->file_pos
+ new->len
)
2753 extent
= btrfs_item_ptr(l
, slot
, struct btrfs_file_extent_item
);
2755 num_bytes
= btrfs_file_extent_num_bytes(l
, extent
);
2756 if (key
.offset
+ num_bytes
< new->file_pos
)
2759 disk_bytenr
= btrfs_file_extent_disk_bytenr(l
, extent
);
2763 extent_offset
= btrfs_file_extent_offset(l
, extent
);
2765 old
= kmalloc(sizeof(*old
), GFP_NOFS
);
2769 offset
= max(new->file_pos
, key
.offset
);
2770 end
= min(new->file_pos
+ new->len
, key
.offset
+ num_bytes
);
2772 old
->bytenr
= disk_bytenr
;
2773 old
->extent_offset
= extent_offset
;
2774 old
->offset
= offset
- key
.offset
;
2775 old
->len
= end
- offset
;
2778 list_add_tail(&old
->list
, &new->head
);
2784 btrfs_free_path(path
);
2785 atomic_inc(&root
->fs_info
->defrag_running
);
2790 btrfs_free_path(path
);
2792 free_sa_defrag_extent(new);
2796 static void btrfs_release_delalloc_bytes(struct btrfs_root
*root
,
2799 struct btrfs_block_group_cache
*cache
;
2801 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
2804 spin_lock(&cache
->lock
);
2805 cache
->delalloc_bytes
-= len
;
2806 spin_unlock(&cache
->lock
);
2808 btrfs_put_block_group(cache
);
2811 /* as ordered data IO finishes, this gets called so we can finish
2812 * an ordered extent if the range of bytes in the file it covers are
2815 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
2817 struct inode
*inode
= ordered_extent
->inode
;
2818 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2819 struct btrfs_trans_handle
*trans
= NULL
;
2820 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2821 struct extent_state
*cached_state
= NULL
;
2822 struct new_sa_defrag_extent
*new = NULL
;
2823 int compress_type
= 0;
2825 u64 logical_len
= ordered_extent
->len
;
2827 bool truncated
= false;
2829 nolock
= btrfs_is_free_space_inode(inode
);
2831 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
2836 btrfs_free_io_failure_record(inode
, ordered_extent
->file_offset
,
2837 ordered_extent
->file_offset
+
2838 ordered_extent
->len
- 1);
2840 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered_extent
->flags
)) {
2842 logical_len
= ordered_extent
->truncated_len
;
2843 /* Truncated the entire extent, don't bother adding */
2848 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
2849 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
2852 * For mwrite(mmap + memset to write) case, we still reserve
2853 * space for NOCOW range.
2854 * As NOCOW won't cause a new delayed ref, just free the space
2856 btrfs_qgroup_free_data(inode
, ordered_extent
->file_offset
,
2857 ordered_extent
->len
);
2858 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2860 trans
= btrfs_join_transaction_nolock(root
);
2862 trans
= btrfs_join_transaction(root
);
2863 if (IS_ERR(trans
)) {
2864 ret
= PTR_ERR(trans
);
2868 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2869 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2870 if (ret
) /* -ENOMEM or corruption */
2871 btrfs_abort_transaction(trans
, root
, ret
);
2875 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
2876 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2879 ret
= test_range_bit(io_tree
, ordered_extent
->file_offset
,
2880 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2881 EXTENT_DEFRAG
, 1, cached_state
);
2883 u64 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
2884 if (0 && last_snapshot
>= BTRFS_I(inode
)->generation
)
2885 /* the inode is shared */
2886 new = record_old_file_extents(inode
, ordered_extent
);
2888 clear_extent_bit(io_tree
, ordered_extent
->file_offset
,
2889 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2890 EXTENT_DEFRAG
, 0, 0, &cached_state
, GFP_NOFS
);
2894 trans
= btrfs_join_transaction_nolock(root
);
2896 trans
= btrfs_join_transaction(root
);
2897 if (IS_ERR(trans
)) {
2898 ret
= PTR_ERR(trans
);
2903 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2905 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
2906 compress_type
= ordered_extent
->compress_type
;
2907 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
2908 BUG_ON(compress_type
);
2909 ret
= btrfs_mark_extent_written(trans
, inode
,
2910 ordered_extent
->file_offset
,
2911 ordered_extent
->file_offset
+
2914 BUG_ON(root
== root
->fs_info
->tree_root
);
2915 ret
= insert_reserved_file_extent(trans
, inode
,
2916 ordered_extent
->file_offset
,
2917 ordered_extent
->start
,
2918 ordered_extent
->disk_len
,
2919 logical_len
, logical_len
,
2920 compress_type
, 0, 0,
2921 BTRFS_FILE_EXTENT_REG
);
2923 btrfs_release_delalloc_bytes(root
,
2924 ordered_extent
->start
,
2925 ordered_extent
->disk_len
);
2927 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
2928 ordered_extent
->file_offset
, ordered_extent
->len
,
2931 btrfs_abort_transaction(trans
, root
, ret
);
2935 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
2936 &ordered_extent
->list
);
2938 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2939 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2940 if (ret
) { /* -ENOMEM or corruption */
2941 btrfs_abort_transaction(trans
, root
, ret
);
2946 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
2947 ordered_extent
->file_offset
+
2948 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
2950 if (root
!= root
->fs_info
->tree_root
)
2951 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
2953 btrfs_end_transaction(trans
, root
);
2955 if (ret
|| truncated
) {
2959 start
= ordered_extent
->file_offset
+ logical_len
;
2961 start
= ordered_extent
->file_offset
;
2962 end
= ordered_extent
->file_offset
+ ordered_extent
->len
- 1;
2963 clear_extent_uptodate(io_tree
, start
, end
, NULL
, GFP_NOFS
);
2965 /* Drop the cache for the part of the extent we didn't write. */
2966 btrfs_drop_extent_cache(inode
, start
, end
, 0);
2969 * If the ordered extent had an IOERR or something else went
2970 * wrong we need to return the space for this ordered extent
2971 * back to the allocator. We only free the extent in the
2972 * truncated case if we didn't write out the extent at all.
2974 if ((ret
|| !logical_len
) &&
2975 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
2976 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
))
2977 btrfs_free_reserved_extent(root
, ordered_extent
->start
,
2978 ordered_extent
->disk_len
, 1);
2983 * This needs to be done to make sure anybody waiting knows we are done
2984 * updating everything for this ordered extent.
2986 btrfs_remove_ordered_extent(inode
, ordered_extent
);
2988 /* for snapshot-aware defrag */
2991 free_sa_defrag_extent(new);
2992 atomic_dec(&root
->fs_info
->defrag_running
);
2994 relink_file_extents(new);
2999 btrfs_put_ordered_extent(ordered_extent
);
3000 /* once for the tree */
3001 btrfs_put_ordered_extent(ordered_extent
);
3006 static void finish_ordered_fn(struct btrfs_work
*work
)
3008 struct btrfs_ordered_extent
*ordered_extent
;
3009 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
3010 btrfs_finish_ordered_io(ordered_extent
);
3013 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
3014 struct extent_state
*state
, int uptodate
)
3016 struct inode
*inode
= page
->mapping
->host
;
3017 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3018 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
3019 struct btrfs_workqueue
*wq
;
3020 btrfs_work_func_t func
;
3022 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
3024 ClearPagePrivate2(page
);
3025 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
3026 end
- start
+ 1, uptodate
))
3029 if (btrfs_is_free_space_inode(inode
)) {
3030 wq
= root
->fs_info
->endio_freespace_worker
;
3031 func
= btrfs_freespace_write_helper
;
3033 wq
= root
->fs_info
->endio_write_workers
;
3034 func
= btrfs_endio_write_helper
;
3037 btrfs_init_work(&ordered_extent
->work
, func
, finish_ordered_fn
, NULL
,
3039 btrfs_queue_work(wq
, &ordered_extent
->work
);
3044 static int __readpage_endio_check(struct inode
*inode
,
3045 struct btrfs_io_bio
*io_bio
,
3046 int icsum
, struct page
*page
,
3047 int pgoff
, u64 start
, size_t len
)
3053 csum_expected
= *(((u32
*)io_bio
->csum
) + icsum
);
3055 kaddr
= kmap_atomic(page
);
3056 csum
= btrfs_csum_data(kaddr
+ pgoff
, csum
, len
);
3057 btrfs_csum_final(csum
, (char *)&csum
);
3058 if (csum
!= csum_expected
)
3061 kunmap_atomic(kaddr
);
3064 btrfs_warn_rl(BTRFS_I(inode
)->root
->fs_info
,
3065 "csum failed ino %llu off %llu csum %u expected csum %u",
3066 btrfs_ino(inode
), start
, csum
, csum_expected
);
3067 memset(kaddr
+ pgoff
, 1, len
);
3068 flush_dcache_page(page
);
3069 kunmap_atomic(kaddr
);
3070 if (csum_expected
== 0)
3076 * when reads are done, we need to check csums to verify the data is correct
3077 * if there's a match, we allow the bio to finish. If not, the code in
3078 * extent_io.c will try to find good copies for us.
3080 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
3081 u64 phy_offset
, struct page
*page
,
3082 u64 start
, u64 end
, int mirror
)
3084 size_t offset
= start
- page_offset(page
);
3085 struct inode
*inode
= page
->mapping
->host
;
3086 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3087 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3089 if (PageChecked(page
)) {
3090 ClearPageChecked(page
);
3094 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
3097 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
3098 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
3099 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
3104 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
3105 return __readpage_endio_check(inode
, io_bio
, phy_offset
, page
, offset
,
3106 start
, (size_t)(end
- start
+ 1));
3109 struct delayed_iput
{
3110 struct list_head list
;
3111 struct inode
*inode
;
3114 /* JDM: If this is fs-wide, why can't we add a pointer to
3115 * btrfs_inode instead and avoid the allocation? */
3116 void btrfs_add_delayed_iput(struct inode
*inode
)
3118 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
3119 struct delayed_iput
*delayed
;
3121 if (atomic_add_unless(&inode
->i_count
, -1, 1))
3124 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
3125 delayed
->inode
= inode
;
3127 spin_lock(&fs_info
->delayed_iput_lock
);
3128 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
3129 spin_unlock(&fs_info
->delayed_iput_lock
);
3132 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
3135 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3136 struct delayed_iput
*delayed
;
3139 spin_lock(&fs_info
->delayed_iput_lock
);
3140 empty
= list_empty(&fs_info
->delayed_iputs
);
3141 spin_unlock(&fs_info
->delayed_iput_lock
);
3145 down_read(&fs_info
->delayed_iput_sem
);
3147 spin_lock(&fs_info
->delayed_iput_lock
);
3148 list_splice_init(&fs_info
->delayed_iputs
, &list
);
3149 spin_unlock(&fs_info
->delayed_iput_lock
);
3151 while (!list_empty(&list
)) {
3152 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
3153 list_del(&delayed
->list
);
3154 iput(delayed
->inode
);
3158 up_read(&root
->fs_info
->delayed_iput_sem
);
3162 * This is called in transaction commit time. If there are no orphan
3163 * files in the subvolume, it removes orphan item and frees block_rsv
3166 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
3167 struct btrfs_root
*root
)
3169 struct btrfs_block_rsv
*block_rsv
;
3172 if (atomic_read(&root
->orphan_inodes
) ||
3173 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
3176 spin_lock(&root
->orphan_lock
);
3177 if (atomic_read(&root
->orphan_inodes
)) {
3178 spin_unlock(&root
->orphan_lock
);
3182 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
3183 spin_unlock(&root
->orphan_lock
);
3187 block_rsv
= root
->orphan_block_rsv
;
3188 root
->orphan_block_rsv
= NULL
;
3189 spin_unlock(&root
->orphan_lock
);
3191 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
) &&
3192 btrfs_root_refs(&root
->root_item
) > 0) {
3193 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
3194 root
->root_key
.objectid
);
3196 btrfs_abort_transaction(trans
, root
, ret
);
3198 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
,
3203 WARN_ON(block_rsv
->size
> 0);
3204 btrfs_free_block_rsv(root
, block_rsv
);
3209 * This creates an orphan entry for the given inode in case something goes
3210 * wrong in the middle of an unlink/truncate.
3212 * NOTE: caller of this function should reserve 5 units of metadata for
3215 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
3217 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3218 struct btrfs_block_rsv
*block_rsv
= NULL
;
3223 if (!root
->orphan_block_rsv
) {
3224 block_rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
3229 spin_lock(&root
->orphan_lock
);
3230 if (!root
->orphan_block_rsv
) {
3231 root
->orphan_block_rsv
= block_rsv
;
3232 } else if (block_rsv
) {
3233 btrfs_free_block_rsv(root
, block_rsv
);
3237 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3238 &BTRFS_I(inode
)->runtime_flags
)) {
3241 * For proper ENOSPC handling, we should do orphan
3242 * cleanup when mounting. But this introduces backward
3243 * compatibility issue.
3245 if (!xchg(&root
->orphan_item_inserted
, 1))
3251 atomic_inc(&root
->orphan_inodes
);
3254 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3255 &BTRFS_I(inode
)->runtime_flags
))
3257 spin_unlock(&root
->orphan_lock
);
3259 /* grab metadata reservation from transaction handle */
3261 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
3262 BUG_ON(ret
); /* -ENOSPC in reservation; Logic error? JDM */
3265 /* insert an orphan item to track this unlinked/truncated file */
3267 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
3269 atomic_dec(&root
->orphan_inodes
);
3271 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3272 &BTRFS_I(inode
)->runtime_flags
);
3273 btrfs_orphan_release_metadata(inode
);
3275 if (ret
!= -EEXIST
) {
3276 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3277 &BTRFS_I(inode
)->runtime_flags
);
3278 btrfs_abort_transaction(trans
, root
, ret
);
3285 /* insert an orphan item to track subvolume contains orphan files */
3287 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
3288 root
->root_key
.objectid
);
3289 if (ret
&& ret
!= -EEXIST
) {
3290 btrfs_abort_transaction(trans
, root
, ret
);
3298 * We have done the truncate/delete so we can go ahead and remove the orphan
3299 * item for this particular inode.
3301 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3302 struct inode
*inode
)
3304 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3305 int delete_item
= 0;
3306 int release_rsv
= 0;
3309 spin_lock(&root
->orphan_lock
);
3310 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3311 &BTRFS_I(inode
)->runtime_flags
))
3314 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3315 &BTRFS_I(inode
)->runtime_flags
))
3317 spin_unlock(&root
->orphan_lock
);
3320 atomic_dec(&root
->orphan_inodes
);
3322 ret
= btrfs_del_orphan_item(trans
, root
,
3327 btrfs_orphan_release_metadata(inode
);
3333 * this cleans up any orphans that may be left on the list from the last use
3336 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3338 struct btrfs_path
*path
;
3339 struct extent_buffer
*leaf
;
3340 struct btrfs_key key
, found_key
;
3341 struct btrfs_trans_handle
*trans
;
3342 struct inode
*inode
;
3343 u64 last_objectid
= 0;
3344 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
3346 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
3349 path
= btrfs_alloc_path();
3356 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3357 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
3358 key
.offset
= (u64
)-1;
3361 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3366 * if ret == 0 means we found what we were searching for, which
3367 * is weird, but possible, so only screw with path if we didn't
3368 * find the key and see if we have stuff that matches
3372 if (path
->slots
[0] == 0)
3377 /* pull out the item */
3378 leaf
= path
->nodes
[0];
3379 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3381 /* make sure the item matches what we want */
3382 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3384 if (found_key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
3387 /* release the path since we're done with it */
3388 btrfs_release_path(path
);
3391 * this is where we are basically btrfs_lookup, without the
3392 * crossing root thing. we store the inode number in the
3393 * offset of the orphan item.
3396 if (found_key
.offset
== last_objectid
) {
3397 btrfs_err(root
->fs_info
,
3398 "Error removing orphan entry, stopping orphan cleanup");
3403 last_objectid
= found_key
.offset
;
3405 found_key
.objectid
= found_key
.offset
;
3406 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3407 found_key
.offset
= 0;
3408 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
3409 ret
= PTR_ERR_OR_ZERO(inode
);
3410 if (ret
&& ret
!= -ESTALE
)
3413 if (ret
== -ESTALE
&& root
== root
->fs_info
->tree_root
) {
3414 struct btrfs_root
*dead_root
;
3415 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3416 int is_dead_root
= 0;
3419 * this is an orphan in the tree root. Currently these
3420 * could come from 2 sources:
3421 * a) a snapshot deletion in progress
3422 * b) a free space cache inode
3423 * We need to distinguish those two, as the snapshot
3424 * orphan must not get deleted.
3425 * find_dead_roots already ran before us, so if this
3426 * is a snapshot deletion, we should find the root
3427 * in the dead_roots list
3429 spin_lock(&fs_info
->trans_lock
);
3430 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
3432 if (dead_root
->root_key
.objectid
==
3433 found_key
.objectid
) {
3438 spin_unlock(&fs_info
->trans_lock
);
3440 /* prevent this orphan from being found again */
3441 key
.offset
= found_key
.objectid
- 1;
3446 * Inode is already gone but the orphan item is still there,
3447 * kill the orphan item.
3449 if (ret
== -ESTALE
) {
3450 trans
= btrfs_start_transaction(root
, 1);
3451 if (IS_ERR(trans
)) {
3452 ret
= PTR_ERR(trans
);
3455 btrfs_debug(root
->fs_info
, "auto deleting %Lu",
3456 found_key
.objectid
);
3457 ret
= btrfs_del_orphan_item(trans
, root
,
3458 found_key
.objectid
);
3459 btrfs_end_transaction(trans
, root
);
3466 * add this inode to the orphan list so btrfs_orphan_del does
3467 * the proper thing when we hit it
3469 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3470 &BTRFS_I(inode
)->runtime_flags
);
3471 atomic_inc(&root
->orphan_inodes
);
3473 /* if we have links, this was a truncate, lets do that */
3474 if (inode
->i_nlink
) {
3475 if (WARN_ON(!S_ISREG(inode
->i_mode
))) {
3481 /* 1 for the orphan item deletion. */
3482 trans
= btrfs_start_transaction(root
, 1);
3483 if (IS_ERR(trans
)) {
3485 ret
= PTR_ERR(trans
);
3488 ret
= btrfs_orphan_add(trans
, inode
);
3489 btrfs_end_transaction(trans
, root
);
3495 ret
= btrfs_truncate(inode
);
3497 btrfs_orphan_del(NULL
, inode
);
3502 /* this will do delete_inode and everything for us */
3507 /* release the path since we're done with it */
3508 btrfs_release_path(path
);
3510 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
3512 if (root
->orphan_block_rsv
)
3513 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
3516 if (root
->orphan_block_rsv
||
3517 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
)) {
3518 trans
= btrfs_join_transaction(root
);
3520 btrfs_end_transaction(trans
, root
);
3524 btrfs_debug(root
->fs_info
, "unlinked %d orphans", nr_unlink
);
3526 btrfs_debug(root
->fs_info
, "truncated %d orphans", nr_truncate
);
3530 btrfs_err(root
->fs_info
,
3531 "could not do orphan cleanup %d", ret
);
3532 btrfs_free_path(path
);
3537 * very simple check to peek ahead in the leaf looking for xattrs. If we
3538 * don't find any xattrs, we know there can't be any acls.
3540 * slot is the slot the inode is in, objectid is the objectid of the inode
3542 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3543 int slot
, u64 objectid
,
3544 int *first_xattr_slot
)
3546 u32 nritems
= btrfs_header_nritems(leaf
);
3547 struct btrfs_key found_key
;
3548 static u64 xattr_access
= 0;
3549 static u64 xattr_default
= 0;
3552 if (!xattr_access
) {
3553 xattr_access
= btrfs_name_hash(POSIX_ACL_XATTR_ACCESS
,
3554 strlen(POSIX_ACL_XATTR_ACCESS
));
3555 xattr_default
= btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT
,
3556 strlen(POSIX_ACL_XATTR_DEFAULT
));
3560 *first_xattr_slot
= -1;
3561 while (slot
< nritems
) {
3562 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3564 /* we found a different objectid, there must not be acls */
3565 if (found_key
.objectid
!= objectid
)
3568 /* we found an xattr, assume we've got an acl */
3569 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
3570 if (*first_xattr_slot
== -1)
3571 *first_xattr_slot
= slot
;
3572 if (found_key
.offset
== xattr_access
||
3573 found_key
.offset
== xattr_default
)
3578 * we found a key greater than an xattr key, there can't
3579 * be any acls later on
3581 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3588 * it goes inode, inode backrefs, xattrs, extents,
3589 * so if there are a ton of hard links to an inode there can
3590 * be a lot of backrefs. Don't waste time searching too hard,
3591 * this is just an optimization
3596 /* we hit the end of the leaf before we found an xattr or
3597 * something larger than an xattr. We have to assume the inode
3600 if (*first_xattr_slot
== -1)
3601 *first_xattr_slot
= slot
;
3606 * read an inode from the btree into the in-memory inode
3608 static void btrfs_read_locked_inode(struct inode
*inode
)
3610 struct btrfs_path
*path
;
3611 struct extent_buffer
*leaf
;
3612 struct btrfs_inode_item
*inode_item
;
3613 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3614 struct btrfs_key location
;
3619 bool filled
= false;
3620 int first_xattr_slot
;
3622 ret
= btrfs_fill_inode(inode
, &rdev
);
3626 path
= btrfs_alloc_path();
3630 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3632 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3636 leaf
= path
->nodes
[0];
3641 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3642 struct btrfs_inode_item
);
3643 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3644 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3645 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3646 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3647 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
3649 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->atime
);
3650 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->atime
);
3652 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->mtime
);
3653 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->mtime
);
3655 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->ctime
);
3656 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->ctime
);
3658 BTRFS_I(inode
)->i_otime
.tv_sec
=
3659 btrfs_timespec_sec(leaf
, &inode_item
->otime
);
3660 BTRFS_I(inode
)->i_otime
.tv_nsec
=
3661 btrfs_timespec_nsec(leaf
, &inode_item
->otime
);
3663 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3664 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3665 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3667 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
3668 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3670 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3672 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3673 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
3677 * If we were modified in the current generation and evicted from memory
3678 * and then re-read we need to do a full sync since we don't have any
3679 * idea about which extents were modified before we were evicted from
3682 * This is required for both inode re-read from disk and delayed inode
3683 * in delayed_nodes_tree.
3685 if (BTRFS_I(inode
)->last_trans
== root
->fs_info
->generation
)
3686 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3687 &BTRFS_I(inode
)->runtime_flags
);
3690 * We don't persist the id of the transaction where an unlink operation
3691 * against the inode was last made. So here we assume the inode might
3692 * have been evicted, and therefore the exact value of last_unlink_trans
3693 * lost, and set it to last_trans to avoid metadata inconsistencies
3694 * between the inode and its parent if the inode is fsync'ed and the log
3695 * replayed. For example, in the scenario:
3698 * ln mydir/foo mydir/bar
3701 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3702 * xfs_io -c fsync mydir/foo
3704 * mount fs, triggers fsync log replay
3706 * We must make sure that when we fsync our inode foo we also log its
3707 * parent inode, otherwise after log replay the parent still has the
3708 * dentry with the "bar" name but our inode foo has a link count of 1
3709 * and doesn't have an inode ref with the name "bar" anymore.
3711 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3712 * but it guarantees correctness at the expense of ocassional full
3713 * transaction commits on fsync if our inode is a directory, or if our
3714 * inode is not a directory, logging its parent unnecessarily.
3716 BTRFS_I(inode
)->last_unlink_trans
= BTRFS_I(inode
)->last_trans
;
3719 if (inode
->i_nlink
!= 1 ||
3720 path
->slots
[0] >= btrfs_header_nritems(leaf
))
3723 btrfs_item_key_to_cpu(leaf
, &location
, path
->slots
[0]);
3724 if (location
.objectid
!= btrfs_ino(inode
))
3727 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3728 if (location
.type
== BTRFS_INODE_REF_KEY
) {
3729 struct btrfs_inode_ref
*ref
;
3731 ref
= (struct btrfs_inode_ref
*)ptr
;
3732 BTRFS_I(inode
)->dir_index
= btrfs_inode_ref_index(leaf
, ref
);
3733 } else if (location
.type
== BTRFS_INODE_EXTREF_KEY
) {
3734 struct btrfs_inode_extref
*extref
;
3736 extref
= (struct btrfs_inode_extref
*)ptr
;
3737 BTRFS_I(inode
)->dir_index
= btrfs_inode_extref_index(leaf
,
3742 * try to precache a NULL acl entry for files that don't have
3743 * any xattrs or acls
3745 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
3746 btrfs_ino(inode
), &first_xattr_slot
);
3747 if (first_xattr_slot
!= -1) {
3748 path
->slots
[0] = first_xattr_slot
;
3749 ret
= btrfs_load_inode_props(inode
, path
);
3751 btrfs_err(root
->fs_info
,
3752 "error loading props for ino %llu (root %llu): %d",
3754 root
->root_key
.objectid
, ret
);
3756 btrfs_free_path(path
);
3759 cache_no_acl(inode
);
3761 switch (inode
->i_mode
& S_IFMT
) {
3763 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3764 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3765 inode
->i_fop
= &btrfs_file_operations
;
3766 inode
->i_op
= &btrfs_file_inode_operations
;
3769 inode
->i_fop
= &btrfs_dir_file_operations
;
3770 if (root
== root
->fs_info
->tree_root
)
3771 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
3773 inode
->i_op
= &btrfs_dir_inode_operations
;
3776 inode
->i_op
= &btrfs_symlink_inode_operations
;
3777 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
3780 inode
->i_op
= &btrfs_special_inode_operations
;
3781 init_special_inode(inode
, inode
->i_mode
, rdev
);
3785 btrfs_update_iflags(inode
);
3789 btrfs_free_path(path
);
3790 make_bad_inode(inode
);
3794 * given a leaf and an inode, copy the inode fields into the leaf
3796 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3797 struct extent_buffer
*leaf
,
3798 struct btrfs_inode_item
*item
,
3799 struct inode
*inode
)
3801 struct btrfs_map_token token
;
3803 btrfs_init_map_token(&token
);
3805 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3806 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3807 btrfs_set_token_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
,
3809 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3810 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3812 btrfs_set_token_timespec_sec(leaf
, &item
->atime
,
3813 inode
->i_atime
.tv_sec
, &token
);
3814 btrfs_set_token_timespec_nsec(leaf
, &item
->atime
,
3815 inode
->i_atime
.tv_nsec
, &token
);
3817 btrfs_set_token_timespec_sec(leaf
, &item
->mtime
,
3818 inode
->i_mtime
.tv_sec
, &token
);
3819 btrfs_set_token_timespec_nsec(leaf
, &item
->mtime
,
3820 inode
->i_mtime
.tv_nsec
, &token
);
3822 btrfs_set_token_timespec_sec(leaf
, &item
->ctime
,
3823 inode
->i_ctime
.tv_sec
, &token
);
3824 btrfs_set_token_timespec_nsec(leaf
, &item
->ctime
,
3825 inode
->i_ctime
.tv_nsec
, &token
);
3827 btrfs_set_token_timespec_sec(leaf
, &item
->otime
,
3828 BTRFS_I(inode
)->i_otime
.tv_sec
, &token
);
3829 btrfs_set_token_timespec_nsec(leaf
, &item
->otime
,
3830 BTRFS_I(inode
)->i_otime
.tv_nsec
, &token
);
3832 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3834 btrfs_set_token_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
,
3836 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
3837 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3838 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3839 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3840 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3844 * copy everything in the in-memory inode into the btree.
3846 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
3847 struct btrfs_root
*root
, struct inode
*inode
)
3849 struct btrfs_inode_item
*inode_item
;
3850 struct btrfs_path
*path
;
3851 struct extent_buffer
*leaf
;
3854 path
= btrfs_alloc_path();
3858 path
->leave_spinning
= 1;
3859 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
3867 leaf
= path
->nodes
[0];
3868 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3869 struct btrfs_inode_item
);
3871 fill_inode_item(trans
, leaf
, inode_item
, inode
);
3872 btrfs_mark_buffer_dirty(leaf
);
3873 btrfs_set_inode_last_trans(trans
, inode
);
3876 btrfs_free_path(path
);
3881 * copy everything in the in-memory inode into the btree.
3883 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
3884 struct btrfs_root
*root
, struct inode
*inode
)
3889 * If the inode is a free space inode, we can deadlock during commit
3890 * if we put it into the delayed code.
3892 * The data relocation inode should also be directly updated
3895 if (!btrfs_is_free_space_inode(inode
)
3896 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
3897 && !root
->fs_info
->log_root_recovering
) {
3898 btrfs_update_root_times(trans
, root
);
3900 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
3902 btrfs_set_inode_last_trans(trans
, inode
);
3906 return btrfs_update_inode_item(trans
, root
, inode
);
3909 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
3910 struct btrfs_root
*root
,
3911 struct inode
*inode
)
3915 ret
= btrfs_update_inode(trans
, root
, inode
);
3917 return btrfs_update_inode_item(trans
, root
, inode
);
3922 * unlink helper that gets used here in inode.c and in the tree logging
3923 * recovery code. It remove a link in a directory with a given name, and
3924 * also drops the back refs in the inode to the directory
3926 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3927 struct btrfs_root
*root
,
3928 struct inode
*dir
, struct inode
*inode
,
3929 const char *name
, int name_len
)
3931 struct btrfs_path
*path
;
3933 struct extent_buffer
*leaf
;
3934 struct btrfs_dir_item
*di
;
3935 struct btrfs_key key
;
3937 u64 ino
= btrfs_ino(inode
);
3938 u64 dir_ino
= btrfs_ino(dir
);
3940 path
= btrfs_alloc_path();
3946 path
->leave_spinning
= 1;
3947 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3948 name
, name_len
, -1);
3957 leaf
= path
->nodes
[0];
3958 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3959 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3962 btrfs_release_path(path
);
3965 * If we don't have dir index, we have to get it by looking up
3966 * the inode ref, since we get the inode ref, remove it directly,
3967 * it is unnecessary to do delayed deletion.
3969 * But if we have dir index, needn't search inode ref to get it.
3970 * Since the inode ref is close to the inode item, it is better
3971 * that we delay to delete it, and just do this deletion when
3972 * we update the inode item.
3974 if (BTRFS_I(inode
)->dir_index
) {
3975 ret
= btrfs_delayed_delete_inode_ref(inode
);
3977 index
= BTRFS_I(inode
)->dir_index
;
3982 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
3985 btrfs_info(root
->fs_info
,
3986 "failed to delete reference to %.*s, inode %llu parent %llu",
3987 name_len
, name
, ino
, dir_ino
);
3988 btrfs_abort_transaction(trans
, root
, ret
);
3992 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3994 btrfs_abort_transaction(trans
, root
, ret
);
3998 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
4000 if (ret
!= 0 && ret
!= -ENOENT
) {
4001 btrfs_abort_transaction(trans
, root
, ret
);
4005 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
4010 btrfs_abort_transaction(trans
, root
, ret
);
4012 btrfs_free_path(path
);
4016 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
4017 inode_inc_iversion(inode
);
4018 inode_inc_iversion(dir
);
4019 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
4020 ret
= btrfs_update_inode(trans
, root
, dir
);
4025 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4026 struct btrfs_root
*root
,
4027 struct inode
*dir
, struct inode
*inode
,
4028 const char *name
, int name_len
)
4031 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
4034 ret
= btrfs_update_inode(trans
, root
, inode
);
4040 * helper to start transaction for unlink and rmdir.
4042 * unlink and rmdir are special in btrfs, they do not always free space, so
4043 * if we cannot make our reservations the normal way try and see if there is
4044 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4045 * allow the unlink to occur.
4047 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
)
4049 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4052 * 1 for the possible orphan item
4053 * 1 for the dir item
4054 * 1 for the dir index
4055 * 1 for the inode ref
4058 return btrfs_start_transaction_fallback_global_rsv(root
, 5, 5);
4061 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
4063 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4064 struct btrfs_trans_handle
*trans
;
4065 struct inode
*inode
= d_inode(dentry
);
4068 trans
= __unlink_start_trans(dir
);
4070 return PTR_ERR(trans
);
4072 btrfs_record_unlink_dir(trans
, dir
, d_inode(dentry
), 0);
4074 ret
= btrfs_unlink_inode(trans
, root
, dir
, d_inode(dentry
),
4075 dentry
->d_name
.name
, dentry
->d_name
.len
);
4079 if (inode
->i_nlink
== 0) {
4080 ret
= btrfs_orphan_add(trans
, inode
);
4086 btrfs_end_transaction(trans
, root
);
4087 btrfs_btree_balance_dirty(root
);
4091 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
4092 struct btrfs_root
*root
,
4093 struct inode
*dir
, u64 objectid
,
4094 const char *name
, int name_len
)
4096 struct btrfs_path
*path
;
4097 struct extent_buffer
*leaf
;
4098 struct btrfs_dir_item
*di
;
4099 struct btrfs_key key
;
4102 u64 dir_ino
= btrfs_ino(dir
);
4104 path
= btrfs_alloc_path();
4108 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4109 name
, name_len
, -1);
4110 if (IS_ERR_OR_NULL(di
)) {
4118 leaf
= path
->nodes
[0];
4119 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4120 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
4121 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4123 btrfs_abort_transaction(trans
, root
, ret
);
4126 btrfs_release_path(path
);
4128 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
4129 objectid
, root
->root_key
.objectid
,
4130 dir_ino
, &index
, name
, name_len
);
4132 if (ret
!= -ENOENT
) {
4133 btrfs_abort_transaction(trans
, root
, ret
);
4136 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
4138 if (IS_ERR_OR_NULL(di
)) {
4143 btrfs_abort_transaction(trans
, root
, ret
);
4147 leaf
= path
->nodes
[0];
4148 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4149 btrfs_release_path(path
);
4152 btrfs_release_path(path
);
4154 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
4156 btrfs_abort_transaction(trans
, root
, ret
);
4160 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
4161 inode_inc_iversion(dir
);
4162 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
4163 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
4165 btrfs_abort_transaction(trans
, root
, ret
);
4167 btrfs_free_path(path
);
4171 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
4173 struct inode
*inode
= d_inode(dentry
);
4175 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4176 struct btrfs_trans_handle
*trans
;
4178 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
4180 if (btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
4183 trans
= __unlink_start_trans(dir
);
4185 return PTR_ERR(trans
);
4187 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
4188 err
= btrfs_unlink_subvol(trans
, root
, dir
,
4189 BTRFS_I(inode
)->location
.objectid
,
4190 dentry
->d_name
.name
,
4191 dentry
->d_name
.len
);
4195 err
= btrfs_orphan_add(trans
, inode
);
4199 /* now the directory is empty */
4200 err
= btrfs_unlink_inode(trans
, root
, dir
, d_inode(dentry
),
4201 dentry
->d_name
.name
, dentry
->d_name
.len
);
4203 btrfs_i_size_write(inode
, 0);
4205 btrfs_end_transaction(trans
, root
);
4206 btrfs_btree_balance_dirty(root
);
4211 static int truncate_space_check(struct btrfs_trans_handle
*trans
,
4212 struct btrfs_root
*root
,
4217 bytes_deleted
= btrfs_csum_bytes_to_leaves(root
, bytes_deleted
);
4218 ret
= btrfs_block_rsv_add(root
, &root
->fs_info
->trans_block_rsv
,
4219 bytes_deleted
, BTRFS_RESERVE_NO_FLUSH
);
4221 trans
->bytes_reserved
+= bytes_deleted
;
4226 static int truncate_inline_extent(struct inode
*inode
,
4227 struct btrfs_path
*path
,
4228 struct btrfs_key
*found_key
,
4232 struct extent_buffer
*leaf
= path
->nodes
[0];
4233 int slot
= path
->slots
[0];
4234 struct btrfs_file_extent_item
*fi
;
4235 u32 size
= (u32
)(new_size
- found_key
->offset
);
4236 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4238 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
4240 if (btrfs_file_extent_compression(leaf
, fi
) != BTRFS_COMPRESS_NONE
) {
4241 loff_t offset
= new_size
;
4242 loff_t page_end
= ALIGN(offset
, PAGE_CACHE_SIZE
);
4245 * Zero out the remaining of the last page of our inline extent,
4246 * instead of directly truncating our inline extent here - that
4247 * would be much more complex (decompressing all the data, then
4248 * compressing the truncated data, which might be bigger than
4249 * the size of the inline extent, resize the extent, etc).
4250 * We release the path because to get the page we might need to
4251 * read the extent item from disk (data not in the page cache).
4253 btrfs_release_path(path
);
4254 return btrfs_truncate_page(inode
, offset
, page_end
- offset
, 0);
4257 btrfs_set_file_extent_ram_bytes(leaf
, fi
, size
);
4258 size
= btrfs_file_extent_calc_inline_size(size
);
4259 btrfs_truncate_item(root
, path
, size
, 1);
4261 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4262 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4268 * this can truncate away extent items, csum items and directory items.
4269 * It starts at a high offset and removes keys until it can't find
4270 * any higher than new_size
4272 * csum items that cross the new i_size are truncated to the new size
4275 * min_type is the minimum key type to truncate down to. If set to 0, this
4276 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4278 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
4279 struct btrfs_root
*root
,
4280 struct inode
*inode
,
4281 u64 new_size
, u32 min_type
)
4283 struct btrfs_path
*path
;
4284 struct extent_buffer
*leaf
;
4285 struct btrfs_file_extent_item
*fi
;
4286 struct btrfs_key key
;
4287 struct btrfs_key found_key
;
4288 u64 extent_start
= 0;
4289 u64 extent_num_bytes
= 0;
4290 u64 extent_offset
= 0;
4292 u64 last_size
= new_size
;
4293 u32 found_type
= (u8
)-1;
4296 int pending_del_nr
= 0;
4297 int pending_del_slot
= 0;
4298 int extent_type
= -1;
4301 u64 ino
= btrfs_ino(inode
);
4302 u64 bytes_deleted
= 0;
4304 bool should_throttle
= 0;
4305 bool should_end
= 0;
4307 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
4310 * for non-free space inodes and ref cows, we want to back off from
4313 if (!btrfs_is_free_space_inode(inode
) &&
4314 test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4317 path
= btrfs_alloc_path();
4323 * We want to drop from the next block forward in case this new size is
4324 * not block aligned since we will be keeping the last block of the
4325 * extent just the way it is.
4327 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4328 root
== root
->fs_info
->tree_root
)
4329 btrfs_drop_extent_cache(inode
, ALIGN(new_size
,
4330 root
->sectorsize
), (u64
)-1, 0);
4333 * This function is also used to drop the items in the log tree before
4334 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4335 * it is used to drop the loged items. So we shouldn't kill the delayed
4338 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
4339 btrfs_kill_delayed_inode_items(inode
);
4342 key
.offset
= (u64
)-1;
4347 * with a 16K leaf size and 128MB extents, you can actually queue
4348 * up a huge file in a single leaf. Most of the time that
4349 * bytes_deleted is > 0, it will be huge by the time we get here
4351 if (be_nice
&& bytes_deleted
> 32 * 1024 * 1024) {
4352 if (btrfs_should_end_transaction(trans
, root
)) {
4359 path
->leave_spinning
= 1;
4360 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
4367 /* there are no items in the tree for us to truncate, we're
4370 if (path
->slots
[0] == 0)
4377 leaf
= path
->nodes
[0];
4378 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4379 found_type
= found_key
.type
;
4381 if (found_key
.objectid
!= ino
)
4384 if (found_type
< min_type
)
4387 item_end
= found_key
.offset
;
4388 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
4389 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4390 struct btrfs_file_extent_item
);
4391 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4392 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4394 btrfs_file_extent_num_bytes(leaf
, fi
);
4395 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4396 item_end
+= btrfs_file_extent_inline_len(leaf
,
4397 path
->slots
[0], fi
);
4401 if (found_type
> min_type
) {
4404 if (item_end
< new_size
)
4406 if (found_key
.offset
>= new_size
)
4412 /* FIXME, shrink the extent if the ref count is only 1 */
4413 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
4417 last_size
= found_key
.offset
;
4419 last_size
= new_size
;
4421 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4423 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4425 u64 orig_num_bytes
=
4426 btrfs_file_extent_num_bytes(leaf
, fi
);
4427 extent_num_bytes
= ALIGN(new_size
-
4430 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4432 num_dec
= (orig_num_bytes
-
4434 if (test_bit(BTRFS_ROOT_REF_COWS
,
4437 inode_sub_bytes(inode
, num_dec
);
4438 btrfs_mark_buffer_dirty(leaf
);
4441 btrfs_file_extent_disk_num_bytes(leaf
,
4443 extent_offset
= found_key
.offset
-
4444 btrfs_file_extent_offset(leaf
, fi
);
4446 /* FIXME blocksize != 4096 */
4447 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
4448 if (extent_start
!= 0) {
4450 if (test_bit(BTRFS_ROOT_REF_COWS
,
4452 inode_sub_bytes(inode
, num_dec
);
4455 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4457 * we can't truncate inline items that have had
4461 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
4462 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
4465 * Need to release path in order to truncate a
4466 * compressed extent. So delete any accumulated
4467 * extent items so far.
4469 if (btrfs_file_extent_compression(leaf
, fi
) !=
4470 BTRFS_COMPRESS_NONE
&& pending_del_nr
) {
4471 err
= btrfs_del_items(trans
, root
, path
,
4475 btrfs_abort_transaction(trans
,
4483 err
= truncate_inline_extent(inode
, path
,
4488 btrfs_abort_transaction(trans
,
4492 } else if (test_bit(BTRFS_ROOT_REF_COWS
,
4494 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4499 if (!pending_del_nr
) {
4500 /* no pending yet, add ourselves */
4501 pending_del_slot
= path
->slots
[0];
4503 } else if (pending_del_nr
&&
4504 path
->slots
[0] + 1 == pending_del_slot
) {
4505 /* hop on the pending chunk */
4507 pending_del_slot
= path
->slots
[0];
4514 should_throttle
= 0;
4517 (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4518 root
== root
->fs_info
->tree_root
)) {
4519 btrfs_set_path_blocking(path
);
4520 bytes_deleted
+= extent_num_bytes
;
4521 ret
= btrfs_free_extent(trans
, root
, extent_start
,
4522 extent_num_bytes
, 0,
4523 btrfs_header_owner(leaf
),
4524 ino
, extent_offset
);
4526 if (btrfs_should_throttle_delayed_refs(trans
, root
))
4527 btrfs_async_run_delayed_refs(root
,
4528 trans
->delayed_ref_updates
* 2, 0);
4530 if (truncate_space_check(trans
, root
,
4531 extent_num_bytes
)) {
4534 if (btrfs_should_throttle_delayed_refs(trans
,
4536 should_throttle
= 1;
4541 if (found_type
== BTRFS_INODE_ITEM_KEY
)
4544 if (path
->slots
[0] == 0 ||
4545 path
->slots
[0] != pending_del_slot
||
4546 should_throttle
|| should_end
) {
4547 if (pending_del_nr
) {
4548 ret
= btrfs_del_items(trans
, root
, path
,
4552 btrfs_abort_transaction(trans
,
4558 btrfs_release_path(path
);
4559 if (should_throttle
) {
4560 unsigned long updates
= trans
->delayed_ref_updates
;
4562 trans
->delayed_ref_updates
= 0;
4563 ret
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
4569 * if we failed to refill our space rsv, bail out
4570 * and let the transaction restart
4582 if (pending_del_nr
) {
4583 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
4586 btrfs_abort_transaction(trans
, root
, ret
);
4589 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
4590 btrfs_ordered_update_i_size(inode
, last_size
, NULL
);
4592 btrfs_free_path(path
);
4594 if (be_nice
&& bytes_deleted
> 32 * 1024 * 1024) {
4595 unsigned long updates
= trans
->delayed_ref_updates
;
4597 trans
->delayed_ref_updates
= 0;
4598 ret
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
4607 * btrfs_truncate_page - read, zero a chunk and write a page
4608 * @inode - inode that we're zeroing
4609 * @from - the offset to start zeroing
4610 * @len - the length to zero, 0 to zero the entire range respective to the
4612 * @front - zero up to the offset instead of from the offset on
4614 * This will find the page for the "from" offset and cow the page and zero the
4615 * part we want to zero. This is used with truncate and hole punching.
4617 int btrfs_truncate_page(struct inode
*inode
, loff_t from
, loff_t len
,
4620 struct address_space
*mapping
= inode
->i_mapping
;
4621 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4622 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4623 struct btrfs_ordered_extent
*ordered
;
4624 struct extent_state
*cached_state
= NULL
;
4626 u32 blocksize
= root
->sectorsize
;
4627 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
4628 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
4630 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4635 if ((offset
& (blocksize
- 1)) == 0 &&
4636 (!len
|| ((len
& (blocksize
- 1)) == 0)))
4638 ret
= btrfs_delalloc_reserve_space(inode
,
4639 round_down(from
, PAGE_CACHE_SIZE
), PAGE_CACHE_SIZE
);
4644 page
= find_or_create_page(mapping
, index
, mask
);
4646 btrfs_delalloc_release_space(inode
,
4647 round_down(from
, PAGE_CACHE_SIZE
),
4653 page_start
= page_offset(page
);
4654 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4656 if (!PageUptodate(page
)) {
4657 ret
= btrfs_readpage(NULL
, page
);
4659 if (page
->mapping
!= mapping
) {
4661 page_cache_release(page
);
4664 if (!PageUptodate(page
)) {
4669 wait_on_page_writeback(page
);
4671 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
4672 set_page_extent_mapped(page
);
4674 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
4676 unlock_extent_cached(io_tree
, page_start
, page_end
,
4677 &cached_state
, GFP_NOFS
);
4679 page_cache_release(page
);
4680 btrfs_start_ordered_extent(inode
, ordered
, 1);
4681 btrfs_put_ordered_extent(ordered
);
4685 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
4686 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4687 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4688 0, 0, &cached_state
, GFP_NOFS
);
4690 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
4693 unlock_extent_cached(io_tree
, page_start
, page_end
,
4694 &cached_state
, GFP_NOFS
);
4698 if (offset
!= PAGE_CACHE_SIZE
) {
4700 len
= PAGE_CACHE_SIZE
- offset
;
4703 memset(kaddr
, 0, offset
);
4705 memset(kaddr
+ offset
, 0, len
);
4706 flush_dcache_page(page
);
4709 ClearPageChecked(page
);
4710 set_page_dirty(page
);
4711 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
4716 btrfs_delalloc_release_space(inode
, page_start
,
4719 page_cache_release(page
);
4724 static int maybe_insert_hole(struct btrfs_root
*root
, struct inode
*inode
,
4725 u64 offset
, u64 len
)
4727 struct btrfs_trans_handle
*trans
;
4731 * Still need to make sure the inode looks like it's been updated so
4732 * that any holes get logged if we fsync.
4734 if (btrfs_fs_incompat(root
->fs_info
, NO_HOLES
)) {
4735 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
4736 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
4737 BTRFS_I(inode
)->last_log_commit
= root
->last_log_commit
;
4742 * 1 - for the one we're dropping
4743 * 1 - for the one we're adding
4744 * 1 - for updating the inode.
4746 trans
= btrfs_start_transaction(root
, 3);
4748 return PTR_ERR(trans
);
4750 ret
= btrfs_drop_extents(trans
, root
, inode
, offset
, offset
+ len
, 1);
4752 btrfs_abort_transaction(trans
, root
, ret
);
4753 btrfs_end_transaction(trans
, root
);
4757 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(inode
), offset
,
4758 0, 0, len
, 0, len
, 0, 0, 0);
4760 btrfs_abort_transaction(trans
, root
, ret
);
4762 btrfs_update_inode(trans
, root
, inode
);
4763 btrfs_end_transaction(trans
, root
);
4768 * This function puts in dummy file extents for the area we're creating a hole
4769 * for. So if we are truncating this file to a larger size we need to insert
4770 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4771 * the range between oldsize and size
4773 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
4775 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4776 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4777 struct extent_map
*em
= NULL
;
4778 struct extent_state
*cached_state
= NULL
;
4779 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4780 u64 hole_start
= ALIGN(oldsize
, root
->sectorsize
);
4781 u64 block_end
= ALIGN(size
, root
->sectorsize
);
4788 * If our size started in the middle of a page we need to zero out the
4789 * rest of the page before we expand the i_size, otherwise we could
4790 * expose stale data.
4792 err
= btrfs_truncate_page(inode
, oldsize
, 0, 0);
4796 if (size
<= hole_start
)
4800 struct btrfs_ordered_extent
*ordered
;
4802 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
4804 ordered
= btrfs_lookup_ordered_range(inode
, hole_start
,
4805 block_end
- hole_start
);
4808 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
4809 &cached_state
, GFP_NOFS
);
4810 btrfs_start_ordered_extent(inode
, ordered
, 1);
4811 btrfs_put_ordered_extent(ordered
);
4814 cur_offset
= hole_start
;
4816 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4817 block_end
- cur_offset
, 0);
4823 last_byte
= min(extent_map_end(em
), block_end
);
4824 last_byte
= ALIGN(last_byte
, root
->sectorsize
);
4825 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
4826 struct extent_map
*hole_em
;
4827 hole_size
= last_byte
- cur_offset
;
4829 err
= maybe_insert_hole(root
, inode
, cur_offset
,
4833 btrfs_drop_extent_cache(inode
, cur_offset
,
4834 cur_offset
+ hole_size
- 1, 0);
4835 hole_em
= alloc_extent_map();
4837 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
4838 &BTRFS_I(inode
)->runtime_flags
);
4841 hole_em
->start
= cur_offset
;
4842 hole_em
->len
= hole_size
;
4843 hole_em
->orig_start
= cur_offset
;
4845 hole_em
->block_start
= EXTENT_MAP_HOLE
;
4846 hole_em
->block_len
= 0;
4847 hole_em
->orig_block_len
= 0;
4848 hole_em
->ram_bytes
= hole_size
;
4849 hole_em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4850 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
4851 hole_em
->generation
= root
->fs_info
->generation
;
4854 write_lock(&em_tree
->lock
);
4855 err
= add_extent_mapping(em_tree
, hole_em
, 1);
4856 write_unlock(&em_tree
->lock
);
4859 btrfs_drop_extent_cache(inode
, cur_offset
,
4863 free_extent_map(hole_em
);
4866 free_extent_map(em
);
4868 cur_offset
= last_byte
;
4869 if (cur_offset
>= block_end
)
4872 free_extent_map(em
);
4873 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
4878 static int wait_snapshoting_atomic_t(atomic_t
*a
)
4884 static void wait_for_snapshot_creation(struct btrfs_root
*root
)
4889 ret
= btrfs_start_write_no_snapshoting(root
);
4892 wait_on_atomic_t(&root
->will_be_snapshoted
,
4893 wait_snapshoting_atomic_t
,
4894 TASK_UNINTERRUPTIBLE
);
4898 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
4900 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4901 struct btrfs_trans_handle
*trans
;
4902 loff_t oldsize
= i_size_read(inode
);
4903 loff_t newsize
= attr
->ia_size
;
4904 int mask
= attr
->ia_valid
;
4908 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4909 * special case where we need to update the times despite not having
4910 * these flags set. For all other operations the VFS set these flags
4911 * explicitly if it wants a timestamp update.
4913 if (newsize
!= oldsize
) {
4914 inode_inc_iversion(inode
);
4915 if (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))
4916 inode
->i_ctime
= inode
->i_mtime
=
4917 current_fs_time(inode
->i_sb
);
4920 if (newsize
> oldsize
) {
4921 truncate_pagecache(inode
, newsize
);
4923 * Don't do an expanding truncate while snapshoting is ongoing.
4924 * This is to ensure the snapshot captures a fully consistent
4925 * state of this file - if the snapshot captures this expanding
4926 * truncation, it must capture all writes that happened before
4929 wait_for_snapshot_creation(root
);
4930 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
4932 btrfs_end_write_no_snapshoting(root
);
4936 trans
= btrfs_start_transaction(root
, 1);
4937 if (IS_ERR(trans
)) {
4938 btrfs_end_write_no_snapshoting(root
);
4939 return PTR_ERR(trans
);
4942 i_size_write(inode
, newsize
);
4943 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
4944 ret
= btrfs_update_inode(trans
, root
, inode
);
4945 btrfs_end_write_no_snapshoting(root
);
4946 btrfs_end_transaction(trans
, root
);
4950 * We're truncating a file that used to have good data down to
4951 * zero. Make sure it gets into the ordered flush list so that
4952 * any new writes get down to disk quickly.
4955 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
4956 &BTRFS_I(inode
)->runtime_flags
);
4959 * 1 for the orphan item we're going to add
4960 * 1 for the orphan item deletion.
4962 trans
= btrfs_start_transaction(root
, 2);
4964 return PTR_ERR(trans
);
4967 * We need to do this in case we fail at _any_ point during the
4968 * actual truncate. Once we do the truncate_setsize we could
4969 * invalidate pages which forces any outstanding ordered io to
4970 * be instantly completed which will give us extents that need
4971 * to be truncated. If we fail to get an orphan inode down we
4972 * could have left over extents that were never meant to live,
4973 * so we need to garuntee from this point on that everything
4974 * will be consistent.
4976 ret
= btrfs_orphan_add(trans
, inode
);
4977 btrfs_end_transaction(trans
, root
);
4981 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4982 truncate_setsize(inode
, newsize
);
4984 /* Disable nonlocked read DIO to avoid the end less truncate */
4985 btrfs_inode_block_unlocked_dio(inode
);
4986 inode_dio_wait(inode
);
4987 btrfs_inode_resume_unlocked_dio(inode
);
4989 ret
= btrfs_truncate(inode
);
4990 if (ret
&& inode
->i_nlink
) {
4994 * failed to truncate, disk_i_size is only adjusted down
4995 * as we remove extents, so it should represent the true
4996 * size of the inode, so reset the in memory size and
4997 * delete our orphan entry.
4999 trans
= btrfs_join_transaction(root
);
5000 if (IS_ERR(trans
)) {
5001 btrfs_orphan_del(NULL
, inode
);
5004 i_size_write(inode
, BTRFS_I(inode
)->disk_i_size
);
5005 err
= btrfs_orphan_del(trans
, inode
);
5007 btrfs_abort_transaction(trans
, root
, err
);
5008 btrfs_end_transaction(trans
, root
);
5015 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5017 struct inode
*inode
= d_inode(dentry
);
5018 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5021 if (btrfs_root_readonly(root
))
5024 err
= inode_change_ok(inode
, attr
);
5028 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
5029 err
= btrfs_setsize(inode
, attr
);
5034 if (attr
->ia_valid
) {
5035 setattr_copy(inode
, attr
);
5036 inode_inc_iversion(inode
);
5037 err
= btrfs_dirty_inode(inode
);
5039 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
5040 err
= posix_acl_chmod(inode
, inode
->i_mode
);
5047 * While truncating the inode pages during eviction, we get the VFS calling
5048 * btrfs_invalidatepage() against each page of the inode. This is slow because
5049 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5050 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5051 * extent_state structures over and over, wasting lots of time.
5053 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5054 * those expensive operations on a per page basis and do only the ordered io
5055 * finishing, while we release here the extent_map and extent_state structures,
5056 * without the excessive merging and splitting.
5058 static void evict_inode_truncate_pages(struct inode
*inode
)
5060 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5061 struct extent_map_tree
*map_tree
= &BTRFS_I(inode
)->extent_tree
;
5062 struct rb_node
*node
;
5064 ASSERT(inode
->i_state
& I_FREEING
);
5065 truncate_inode_pages_final(&inode
->i_data
);
5067 write_lock(&map_tree
->lock
);
5068 while (!RB_EMPTY_ROOT(&map_tree
->map
)) {
5069 struct extent_map
*em
;
5071 node
= rb_first(&map_tree
->map
);
5072 em
= rb_entry(node
, struct extent_map
, rb_node
);
5073 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5074 clear_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
5075 remove_extent_mapping(map_tree
, em
);
5076 free_extent_map(em
);
5077 if (need_resched()) {
5078 write_unlock(&map_tree
->lock
);
5080 write_lock(&map_tree
->lock
);
5083 write_unlock(&map_tree
->lock
);
5086 * Keep looping until we have no more ranges in the io tree.
5087 * We can have ongoing bios started by readpages (called from readahead)
5088 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5089 * still in progress (unlocked the pages in the bio but did not yet
5090 * unlocked the ranges in the io tree). Therefore this means some
5091 * ranges can still be locked and eviction started because before
5092 * submitting those bios, which are executed by a separate task (work
5093 * queue kthread), inode references (inode->i_count) were not taken
5094 * (which would be dropped in the end io callback of each bio).
5095 * Therefore here we effectively end up waiting for those bios and
5096 * anyone else holding locked ranges without having bumped the inode's
5097 * reference count - if we don't do it, when they access the inode's
5098 * io_tree to unlock a range it may be too late, leading to an
5099 * use-after-free issue.
5101 spin_lock(&io_tree
->lock
);
5102 while (!RB_EMPTY_ROOT(&io_tree
->state
)) {
5103 struct extent_state
*state
;
5104 struct extent_state
*cached_state
= NULL
;
5108 node
= rb_first(&io_tree
->state
);
5109 state
= rb_entry(node
, struct extent_state
, rb_node
);
5110 start
= state
->start
;
5112 spin_unlock(&io_tree
->lock
);
5114 lock_extent_bits(io_tree
, start
, end
, 0, &cached_state
);
5117 * If still has DELALLOC flag, the extent didn't reach disk,
5118 * and its reserved space won't be freed by delayed_ref.
5119 * So we need to free its reserved space here.
5120 * (Refer to comment in btrfs_invalidatepage, case 2)
5122 * Note, end is the bytenr of last byte, so we need + 1 here.
5124 if (state
->state
& EXTENT_DELALLOC
)
5125 btrfs_qgroup_free_data(inode
, start
, end
- start
+ 1);
5127 clear_extent_bit(io_tree
, start
, end
,
5128 EXTENT_LOCKED
| EXTENT_DIRTY
|
5129 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
5130 EXTENT_DEFRAG
, 1, 1,
5131 &cached_state
, GFP_NOFS
);
5134 spin_lock(&io_tree
->lock
);
5136 spin_unlock(&io_tree
->lock
);
5139 void btrfs_evict_inode(struct inode
*inode
)
5141 struct btrfs_trans_handle
*trans
;
5142 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5143 struct btrfs_block_rsv
*rsv
, *global_rsv
;
5144 int steal_from_global
= 0;
5145 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
5148 trace_btrfs_inode_evict(inode
);
5150 evict_inode_truncate_pages(inode
);
5152 if (inode
->i_nlink
&&
5153 ((btrfs_root_refs(&root
->root_item
) != 0 &&
5154 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
) ||
5155 btrfs_is_free_space_inode(inode
)))
5158 if (is_bad_inode(inode
)) {
5159 btrfs_orphan_del(NULL
, inode
);
5162 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5163 if (!special_file(inode
->i_mode
))
5164 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5166 btrfs_free_io_failure_record(inode
, 0, (u64
)-1);
5168 if (root
->fs_info
->log_root_recovering
) {
5169 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
5170 &BTRFS_I(inode
)->runtime_flags
));
5174 if (inode
->i_nlink
> 0) {
5175 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0 &&
5176 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
);
5180 ret
= btrfs_commit_inode_delayed_inode(inode
);
5182 btrfs_orphan_del(NULL
, inode
);
5186 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
5188 btrfs_orphan_del(NULL
, inode
);
5191 rsv
->size
= min_size
;
5193 global_rsv
= &root
->fs_info
->global_block_rsv
;
5195 btrfs_i_size_write(inode
, 0);
5198 * This is a bit simpler than btrfs_truncate since we've already
5199 * reserved our space for our orphan item in the unlink, so we just
5200 * need to reserve some slack space in case we add bytes and update
5201 * inode item when doing the truncate.
5204 ret
= btrfs_block_rsv_refill(root
, rsv
, min_size
,
5205 BTRFS_RESERVE_FLUSH_LIMIT
);
5208 * Try and steal from the global reserve since we will
5209 * likely not use this space anyway, we want to try as
5210 * hard as possible to get this to work.
5213 steal_from_global
++;
5215 steal_from_global
= 0;
5219 * steal_from_global == 0: we reserved stuff, hooray!
5220 * steal_from_global == 1: we didn't reserve stuff, boo!
5221 * steal_from_global == 2: we've committed, still not a lot of
5222 * room but maybe we'll have room in the global reserve this
5224 * steal_from_global == 3: abandon all hope!
5226 if (steal_from_global
> 2) {
5227 btrfs_warn(root
->fs_info
,
5228 "Could not get space for a delete, will truncate on mount %d",
5230 btrfs_orphan_del(NULL
, inode
);
5231 btrfs_free_block_rsv(root
, rsv
);
5235 trans
= btrfs_join_transaction(root
);
5236 if (IS_ERR(trans
)) {
5237 btrfs_orphan_del(NULL
, inode
);
5238 btrfs_free_block_rsv(root
, rsv
);
5243 * We can't just steal from the global reserve, we need tomake
5244 * sure there is room to do it, if not we need to commit and try
5247 if (steal_from_global
) {
5248 if (!btrfs_check_space_for_delayed_refs(trans
, root
))
5249 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
,
5256 * Couldn't steal from the global reserve, we have too much
5257 * pending stuff built up, commit the transaction and try it
5261 ret
= btrfs_commit_transaction(trans
, root
);
5263 btrfs_orphan_del(NULL
, inode
);
5264 btrfs_free_block_rsv(root
, rsv
);
5269 steal_from_global
= 0;
5272 trans
->block_rsv
= rsv
;
5274 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
5275 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
)
5278 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
5279 btrfs_end_transaction(trans
, root
);
5281 btrfs_btree_balance_dirty(root
);
5284 btrfs_free_block_rsv(root
, rsv
);
5287 * Errors here aren't a big deal, it just means we leave orphan items
5288 * in the tree. They will be cleaned up on the next mount.
5291 trans
->block_rsv
= root
->orphan_block_rsv
;
5292 btrfs_orphan_del(trans
, inode
);
5294 btrfs_orphan_del(NULL
, inode
);
5297 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
5298 if (!(root
== root
->fs_info
->tree_root
||
5299 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
5300 btrfs_return_ino(root
, btrfs_ino(inode
));
5302 btrfs_end_transaction(trans
, root
);
5303 btrfs_btree_balance_dirty(root
);
5305 btrfs_remove_delayed_node(inode
);
5311 * this returns the key found in the dir entry in the location pointer.
5312 * If no dir entries were found, location->objectid is 0.
5314 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
5315 struct btrfs_key
*location
)
5317 const char *name
= dentry
->d_name
.name
;
5318 int namelen
= dentry
->d_name
.len
;
5319 struct btrfs_dir_item
*di
;
5320 struct btrfs_path
*path
;
5321 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5324 path
= btrfs_alloc_path();
5328 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
5333 if (IS_ERR_OR_NULL(di
))
5336 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
5338 btrfs_free_path(path
);
5341 location
->objectid
= 0;
5346 * when we hit a tree root in a directory, the btrfs part of the inode
5347 * needs to be changed to reflect the root directory of the tree root. This
5348 * is kind of like crossing a mount point.
5350 static int fixup_tree_root_location(struct btrfs_root
*root
,
5352 struct dentry
*dentry
,
5353 struct btrfs_key
*location
,
5354 struct btrfs_root
**sub_root
)
5356 struct btrfs_path
*path
;
5357 struct btrfs_root
*new_root
;
5358 struct btrfs_root_ref
*ref
;
5359 struct extent_buffer
*leaf
;
5360 struct btrfs_key key
;
5364 path
= btrfs_alloc_path();
5371 key
.objectid
= BTRFS_I(dir
)->root
->root_key
.objectid
;
5372 key
.type
= BTRFS_ROOT_REF_KEY
;
5373 key
.offset
= location
->objectid
;
5375 ret
= btrfs_search_slot(NULL
, root
->fs_info
->tree_root
, &key
, path
,
5383 leaf
= path
->nodes
[0];
5384 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
5385 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
5386 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
5389 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
5390 (unsigned long)(ref
+ 1),
5391 dentry
->d_name
.len
);
5395 btrfs_release_path(path
);
5397 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
5398 if (IS_ERR(new_root
)) {
5399 err
= PTR_ERR(new_root
);
5403 *sub_root
= new_root
;
5404 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
5405 location
->type
= BTRFS_INODE_ITEM_KEY
;
5406 location
->offset
= 0;
5409 btrfs_free_path(path
);
5413 static void inode_tree_add(struct inode
*inode
)
5415 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5416 struct btrfs_inode
*entry
;
5418 struct rb_node
*parent
;
5419 struct rb_node
*new = &BTRFS_I(inode
)->rb_node
;
5420 u64 ino
= btrfs_ino(inode
);
5422 if (inode_unhashed(inode
))
5425 spin_lock(&root
->inode_lock
);
5426 p
= &root
->inode_tree
.rb_node
;
5429 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
5431 if (ino
< btrfs_ino(&entry
->vfs_inode
))
5432 p
= &parent
->rb_left
;
5433 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
5434 p
= &parent
->rb_right
;
5436 WARN_ON(!(entry
->vfs_inode
.i_state
&
5437 (I_WILL_FREE
| I_FREEING
)));
5438 rb_replace_node(parent
, new, &root
->inode_tree
);
5439 RB_CLEAR_NODE(parent
);
5440 spin_unlock(&root
->inode_lock
);
5444 rb_link_node(new, parent
, p
);
5445 rb_insert_color(new, &root
->inode_tree
);
5446 spin_unlock(&root
->inode_lock
);
5449 static void inode_tree_del(struct inode
*inode
)
5451 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5454 spin_lock(&root
->inode_lock
);
5455 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
5456 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
5457 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
5458 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5460 spin_unlock(&root
->inode_lock
);
5462 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0) {
5463 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
5464 spin_lock(&root
->inode_lock
);
5465 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5466 spin_unlock(&root
->inode_lock
);
5468 btrfs_add_dead_root(root
);
5472 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
5474 struct rb_node
*node
;
5475 struct rb_node
*prev
;
5476 struct btrfs_inode
*entry
;
5477 struct inode
*inode
;
5480 if (!test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
5481 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
5483 spin_lock(&root
->inode_lock
);
5485 node
= root
->inode_tree
.rb_node
;
5489 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5491 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
5492 node
= node
->rb_left
;
5493 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
5494 node
= node
->rb_right
;
5500 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
5501 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
5505 prev
= rb_next(prev
);
5509 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5510 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
5511 inode
= igrab(&entry
->vfs_inode
);
5513 spin_unlock(&root
->inode_lock
);
5514 if (atomic_read(&inode
->i_count
) > 1)
5515 d_prune_aliases(inode
);
5517 * btrfs_drop_inode will have it removed from
5518 * the inode cache when its usage count
5523 spin_lock(&root
->inode_lock
);
5527 if (cond_resched_lock(&root
->inode_lock
))
5530 node
= rb_next(node
);
5532 spin_unlock(&root
->inode_lock
);
5535 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
5537 struct btrfs_iget_args
*args
= p
;
5538 inode
->i_ino
= args
->location
->objectid
;
5539 memcpy(&BTRFS_I(inode
)->location
, args
->location
,
5540 sizeof(*args
->location
));
5541 BTRFS_I(inode
)->root
= args
->root
;
5545 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
5547 struct btrfs_iget_args
*args
= opaque
;
5548 return args
->location
->objectid
== BTRFS_I(inode
)->location
.objectid
&&
5549 args
->root
== BTRFS_I(inode
)->root
;
5552 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
5553 struct btrfs_key
*location
,
5554 struct btrfs_root
*root
)
5556 struct inode
*inode
;
5557 struct btrfs_iget_args args
;
5558 unsigned long hashval
= btrfs_inode_hash(location
->objectid
, root
);
5560 args
.location
= location
;
5563 inode
= iget5_locked(s
, hashval
, btrfs_find_actor
,
5564 btrfs_init_locked_inode
,
5569 /* Get an inode object given its location and corresponding root.
5570 * Returns in *is_new if the inode was read from disk
5572 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
5573 struct btrfs_root
*root
, int *new)
5575 struct inode
*inode
;
5577 inode
= btrfs_iget_locked(s
, location
, root
);
5579 return ERR_PTR(-ENOMEM
);
5581 if (inode
->i_state
& I_NEW
) {
5582 btrfs_read_locked_inode(inode
);
5583 if (!is_bad_inode(inode
)) {
5584 inode_tree_add(inode
);
5585 unlock_new_inode(inode
);
5589 unlock_new_inode(inode
);
5591 inode
= ERR_PTR(-ESTALE
);
5598 static struct inode
*new_simple_dir(struct super_block
*s
,
5599 struct btrfs_key
*key
,
5600 struct btrfs_root
*root
)
5602 struct inode
*inode
= new_inode(s
);
5605 return ERR_PTR(-ENOMEM
);
5607 BTRFS_I(inode
)->root
= root
;
5608 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
5609 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
5611 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
5612 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
5613 inode
->i_fop
= &simple_dir_operations
;
5614 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
5615 inode
->i_mtime
= CURRENT_TIME
;
5616 inode
->i_atime
= inode
->i_mtime
;
5617 inode
->i_ctime
= inode
->i_mtime
;
5618 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
5623 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
5625 struct inode
*inode
;
5626 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5627 struct btrfs_root
*sub_root
= root
;
5628 struct btrfs_key location
;
5632 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
5633 return ERR_PTR(-ENAMETOOLONG
);
5635 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
5637 return ERR_PTR(ret
);
5639 if (location
.objectid
== 0)
5640 return ERR_PTR(-ENOENT
);
5642 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
5643 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
5647 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
5649 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
5650 ret
= fixup_tree_root_location(root
, dir
, dentry
,
5651 &location
, &sub_root
);
5654 inode
= ERR_PTR(ret
);
5656 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
5658 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
5660 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
5662 if (!IS_ERR(inode
) && root
!= sub_root
) {
5663 down_read(&root
->fs_info
->cleanup_work_sem
);
5664 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
5665 ret
= btrfs_orphan_cleanup(sub_root
);
5666 up_read(&root
->fs_info
->cleanup_work_sem
);
5669 inode
= ERR_PTR(ret
);
5676 static int btrfs_dentry_delete(const struct dentry
*dentry
)
5678 struct btrfs_root
*root
;
5679 struct inode
*inode
= d_inode(dentry
);
5681 if (!inode
&& !IS_ROOT(dentry
))
5682 inode
= d_inode(dentry
->d_parent
);
5685 root
= BTRFS_I(inode
)->root
;
5686 if (btrfs_root_refs(&root
->root_item
) == 0)
5689 if (btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5695 static void btrfs_dentry_release(struct dentry
*dentry
)
5697 kfree(dentry
->d_fsdata
);
5700 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5703 struct inode
*inode
;
5705 inode
= btrfs_lookup_dentry(dir
, dentry
);
5706 if (IS_ERR(inode
)) {
5707 if (PTR_ERR(inode
) == -ENOENT
)
5710 return ERR_CAST(inode
);
5713 return d_splice_alias(inode
, dentry
);
5716 unsigned char btrfs_filetype_table
[] = {
5717 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
5720 static int btrfs_real_readdir(struct file
*file
, struct dir_context
*ctx
)
5722 struct inode
*inode
= file_inode(file
);
5723 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5724 struct btrfs_item
*item
;
5725 struct btrfs_dir_item
*di
;
5726 struct btrfs_key key
;
5727 struct btrfs_key found_key
;
5728 struct btrfs_path
*path
;
5729 struct list_head ins_list
;
5730 struct list_head del_list
;
5732 struct extent_buffer
*leaf
;
5734 unsigned char d_type
;
5739 int key_type
= BTRFS_DIR_INDEX_KEY
;
5743 int is_curr
= 0; /* ctx->pos points to the current index? */
5745 /* FIXME, use a real flag for deciding about the key type */
5746 if (root
->fs_info
->tree_root
== root
)
5747 key_type
= BTRFS_DIR_ITEM_KEY
;
5749 if (!dir_emit_dots(file
, ctx
))
5752 path
= btrfs_alloc_path();
5758 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5759 INIT_LIST_HEAD(&ins_list
);
5760 INIT_LIST_HEAD(&del_list
);
5761 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
5764 key
.type
= key_type
;
5765 key
.offset
= ctx
->pos
;
5766 key
.objectid
= btrfs_ino(inode
);
5768 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5773 leaf
= path
->nodes
[0];
5774 slot
= path
->slots
[0];
5775 if (slot
>= btrfs_header_nritems(leaf
)) {
5776 ret
= btrfs_next_leaf(root
, path
);
5784 item
= btrfs_item_nr(slot
);
5785 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5787 if (found_key
.objectid
!= key
.objectid
)
5789 if (found_key
.type
!= key_type
)
5791 if (found_key
.offset
< ctx
->pos
)
5793 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
5794 btrfs_should_delete_dir_index(&del_list
,
5798 ctx
->pos
= found_key
.offset
;
5801 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
5803 di_total
= btrfs_item_size(leaf
, item
);
5805 while (di_cur
< di_total
) {
5806 struct btrfs_key location
;
5808 if (verify_dir_item(root
, leaf
, di
))
5811 name_len
= btrfs_dir_name_len(leaf
, di
);
5812 if (name_len
<= sizeof(tmp_name
)) {
5813 name_ptr
= tmp_name
;
5815 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
5821 read_extent_buffer(leaf
, name_ptr
,
5822 (unsigned long)(di
+ 1), name_len
);
5824 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
5825 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
5828 /* is this a reference to our own snapshot? If so
5831 * In contrast to old kernels, we insert the snapshot's
5832 * dir item and dir index after it has been created, so
5833 * we won't find a reference to our own snapshot. We
5834 * still keep the following code for backward
5837 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
5838 location
.objectid
== root
->root_key
.objectid
) {
5842 over
= !dir_emit(ctx
, name_ptr
, name_len
,
5843 location
.objectid
, d_type
);
5846 if (name_ptr
!= tmp_name
)
5851 di_len
= btrfs_dir_name_len(leaf
, di
) +
5852 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
5854 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
5860 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5863 ret
= btrfs_readdir_delayed_dir_index(ctx
, &ins_list
);
5868 /* Reached end of directory/root. Bump pos past the last item. */
5872 * Stop new entries from being returned after we return the last
5875 * New directory entries are assigned a strictly increasing
5876 * offset. This means that new entries created during readdir
5877 * are *guaranteed* to be seen in the future by that readdir.
5878 * This has broken buggy programs which operate on names as
5879 * they're returned by readdir. Until we re-use freed offsets
5880 * we have this hack to stop new entries from being returned
5881 * under the assumption that they'll never reach this huge
5884 * This is being careful not to overflow 32bit loff_t unless the
5885 * last entry requires it because doing so has broken 32bit apps
5888 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5889 if (ctx
->pos
>= INT_MAX
)
5890 ctx
->pos
= LLONG_MAX
;
5897 if (key_type
== BTRFS_DIR_INDEX_KEY
)
5898 btrfs_put_delayed_items(&ins_list
, &del_list
);
5899 btrfs_free_path(path
);
5903 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5905 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5906 struct btrfs_trans_handle
*trans
;
5908 bool nolock
= false;
5910 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5913 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(inode
))
5916 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
5918 trans
= btrfs_join_transaction_nolock(root
);
5920 trans
= btrfs_join_transaction(root
);
5922 return PTR_ERR(trans
);
5923 ret
= btrfs_commit_transaction(trans
, root
);
5929 * This is somewhat expensive, updating the tree every time the
5930 * inode changes. But, it is most likely to find the inode in cache.
5931 * FIXME, needs more benchmarking...there are no reasons other than performance
5932 * to keep or drop this code.
5934 static int btrfs_dirty_inode(struct inode
*inode
)
5936 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5937 struct btrfs_trans_handle
*trans
;
5940 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5943 trans
= btrfs_join_transaction(root
);
5945 return PTR_ERR(trans
);
5947 ret
= btrfs_update_inode(trans
, root
, inode
);
5948 if (ret
&& ret
== -ENOSPC
) {
5949 /* whoops, lets try again with the full transaction */
5950 btrfs_end_transaction(trans
, root
);
5951 trans
= btrfs_start_transaction(root
, 1);
5953 return PTR_ERR(trans
);
5955 ret
= btrfs_update_inode(trans
, root
, inode
);
5957 btrfs_end_transaction(trans
, root
);
5958 if (BTRFS_I(inode
)->delayed_node
)
5959 btrfs_balance_delayed_items(root
);
5965 * This is a copy of file_update_time. We need this so we can return error on
5966 * ENOSPC for updating the inode in the case of file write and mmap writes.
5968 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
5971 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5973 if (btrfs_root_readonly(root
))
5976 if (flags
& S_VERSION
)
5977 inode_inc_iversion(inode
);
5978 if (flags
& S_CTIME
)
5979 inode
->i_ctime
= *now
;
5980 if (flags
& S_MTIME
)
5981 inode
->i_mtime
= *now
;
5982 if (flags
& S_ATIME
)
5983 inode
->i_atime
= *now
;
5984 return btrfs_dirty_inode(inode
);
5988 * find the highest existing sequence number in a directory
5989 * and then set the in-memory index_cnt variable to reflect
5990 * free sequence numbers
5992 static int btrfs_set_inode_index_count(struct inode
*inode
)
5994 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5995 struct btrfs_key key
, found_key
;
5996 struct btrfs_path
*path
;
5997 struct extent_buffer
*leaf
;
6000 key
.objectid
= btrfs_ino(inode
);
6001 key
.type
= BTRFS_DIR_INDEX_KEY
;
6002 key
.offset
= (u64
)-1;
6004 path
= btrfs_alloc_path();
6008 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6011 /* FIXME: we should be able to handle this */
6017 * MAGIC NUMBER EXPLANATION:
6018 * since we search a directory based on f_pos we have to start at 2
6019 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6020 * else has to start at 2
6022 if (path
->slots
[0] == 0) {
6023 BTRFS_I(inode
)->index_cnt
= 2;
6029 leaf
= path
->nodes
[0];
6030 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6032 if (found_key
.objectid
!= btrfs_ino(inode
) ||
6033 found_key
.type
!= BTRFS_DIR_INDEX_KEY
) {
6034 BTRFS_I(inode
)->index_cnt
= 2;
6038 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
6040 btrfs_free_path(path
);
6045 * helper to find a free sequence number in a given directory. This current
6046 * code is very simple, later versions will do smarter things in the btree
6048 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
6052 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
6053 ret
= btrfs_inode_delayed_dir_index_count(dir
);
6055 ret
= btrfs_set_inode_index_count(dir
);
6061 *index
= BTRFS_I(dir
)->index_cnt
;
6062 BTRFS_I(dir
)->index_cnt
++;
6067 static int btrfs_insert_inode_locked(struct inode
*inode
)
6069 struct btrfs_iget_args args
;
6070 args
.location
= &BTRFS_I(inode
)->location
;
6071 args
.root
= BTRFS_I(inode
)->root
;
6073 return insert_inode_locked4(inode
,
6074 btrfs_inode_hash(inode
->i_ino
, BTRFS_I(inode
)->root
),
6075 btrfs_find_actor
, &args
);
6078 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
6079 struct btrfs_root
*root
,
6081 const char *name
, int name_len
,
6082 u64 ref_objectid
, u64 objectid
,
6083 umode_t mode
, u64
*index
)
6085 struct inode
*inode
;
6086 struct btrfs_inode_item
*inode_item
;
6087 struct btrfs_key
*location
;
6088 struct btrfs_path
*path
;
6089 struct btrfs_inode_ref
*ref
;
6090 struct btrfs_key key
[2];
6092 int nitems
= name
? 2 : 1;
6096 path
= btrfs_alloc_path();
6098 return ERR_PTR(-ENOMEM
);
6100 inode
= new_inode(root
->fs_info
->sb
);
6102 btrfs_free_path(path
);
6103 return ERR_PTR(-ENOMEM
);
6107 * O_TMPFILE, set link count to 0, so that after this point,
6108 * we fill in an inode item with the correct link count.
6111 set_nlink(inode
, 0);
6114 * we have to initialize this early, so we can reclaim the inode
6115 * number if we fail afterwards in this function.
6117 inode
->i_ino
= objectid
;
6120 trace_btrfs_inode_request(dir
);
6122 ret
= btrfs_set_inode_index(dir
, index
);
6124 btrfs_free_path(path
);
6126 return ERR_PTR(ret
);
6132 * index_cnt is ignored for everything but a dir,
6133 * btrfs_get_inode_index_count has an explanation for the magic
6136 BTRFS_I(inode
)->index_cnt
= 2;
6137 BTRFS_I(inode
)->dir_index
= *index
;
6138 BTRFS_I(inode
)->root
= root
;
6139 BTRFS_I(inode
)->generation
= trans
->transid
;
6140 inode
->i_generation
= BTRFS_I(inode
)->generation
;
6143 * We could have gotten an inode number from somebody who was fsynced
6144 * and then removed in this same transaction, so let's just set full
6145 * sync since it will be a full sync anyway and this will blow away the
6146 * old info in the log.
6148 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
6150 key
[0].objectid
= objectid
;
6151 key
[0].type
= BTRFS_INODE_ITEM_KEY
;
6154 sizes
[0] = sizeof(struct btrfs_inode_item
);
6158 * Start new inodes with an inode_ref. This is slightly more
6159 * efficient for small numbers of hard links since they will
6160 * be packed into one item. Extended refs will kick in if we
6161 * add more hard links than can fit in the ref item.
6163 key
[1].objectid
= objectid
;
6164 key
[1].type
= BTRFS_INODE_REF_KEY
;
6165 key
[1].offset
= ref_objectid
;
6167 sizes
[1] = name_len
+ sizeof(*ref
);
6170 location
= &BTRFS_I(inode
)->location
;
6171 location
->objectid
= objectid
;
6172 location
->offset
= 0;
6173 location
->type
= BTRFS_INODE_ITEM_KEY
;
6175 ret
= btrfs_insert_inode_locked(inode
);
6179 path
->leave_spinning
= 1;
6180 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, nitems
);
6184 inode_init_owner(inode
, dir
, mode
);
6185 inode_set_bytes(inode
, 0);
6187 inode
->i_mtime
= CURRENT_TIME
;
6188 inode
->i_atime
= inode
->i_mtime
;
6189 inode
->i_ctime
= inode
->i_mtime
;
6190 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
6192 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
6193 struct btrfs_inode_item
);
6194 memset_extent_buffer(path
->nodes
[0], 0, (unsigned long)inode_item
,
6195 sizeof(*inode_item
));
6196 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
6199 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
6200 struct btrfs_inode_ref
);
6201 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
6202 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
6203 ptr
= (unsigned long)(ref
+ 1);
6204 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
6207 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6208 btrfs_free_path(path
);
6210 btrfs_inherit_iflags(inode
, dir
);
6212 if (S_ISREG(mode
)) {
6213 if (btrfs_test_opt(root
, NODATASUM
))
6214 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6215 if (btrfs_test_opt(root
, NODATACOW
))
6216 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
6217 BTRFS_INODE_NODATASUM
;
6220 inode_tree_add(inode
);
6222 trace_btrfs_inode_new(inode
);
6223 btrfs_set_inode_last_trans(trans
, inode
);
6225 btrfs_update_root_times(trans
, root
);
6227 ret
= btrfs_inode_inherit_props(trans
, inode
, dir
);
6229 btrfs_err(root
->fs_info
,
6230 "error inheriting props for ino %llu (root %llu): %d",
6231 btrfs_ino(inode
), root
->root_key
.objectid
, ret
);
6236 unlock_new_inode(inode
);
6239 BTRFS_I(dir
)->index_cnt
--;
6240 btrfs_free_path(path
);
6242 return ERR_PTR(ret
);
6245 static inline u8
btrfs_inode_type(struct inode
*inode
)
6247 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
6251 * utility function to add 'inode' into 'parent_inode' with
6252 * a give name and a given sequence number.
6253 * if 'add_backref' is true, also insert a backref from the
6254 * inode to the parent directory.
6256 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
6257 struct inode
*parent_inode
, struct inode
*inode
,
6258 const char *name
, int name_len
, int add_backref
, u64 index
)
6261 struct btrfs_key key
;
6262 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
6263 u64 ino
= btrfs_ino(inode
);
6264 u64 parent_ino
= btrfs_ino(parent_inode
);
6266 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6267 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
6270 key
.type
= BTRFS_INODE_ITEM_KEY
;
6274 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6275 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
6276 key
.objectid
, root
->root_key
.objectid
,
6277 parent_ino
, index
, name
, name_len
);
6278 } else if (add_backref
) {
6279 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
6283 /* Nothing to clean up yet */
6287 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
6289 btrfs_inode_type(inode
), index
);
6290 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
6293 btrfs_abort_transaction(trans
, root
, ret
);
6297 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
6299 inode_inc_iversion(parent_inode
);
6300 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
6301 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
6303 btrfs_abort_transaction(trans
, root
, ret
);
6307 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6310 err
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
6311 key
.objectid
, root
->root_key
.objectid
,
6312 parent_ino
, &local_index
, name
, name_len
);
6314 } else if (add_backref
) {
6318 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
6319 ino
, parent_ino
, &local_index
);
6324 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
6325 struct inode
*dir
, struct dentry
*dentry
,
6326 struct inode
*inode
, int backref
, u64 index
)
6328 int err
= btrfs_add_link(trans
, dir
, inode
,
6329 dentry
->d_name
.name
, dentry
->d_name
.len
,
6336 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
6337 umode_t mode
, dev_t rdev
)
6339 struct btrfs_trans_handle
*trans
;
6340 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6341 struct inode
*inode
= NULL
;
6347 if (!new_valid_dev(rdev
))
6351 * 2 for inode item and ref
6353 * 1 for xattr if selinux is on
6355 trans
= btrfs_start_transaction(root
, 5);
6357 return PTR_ERR(trans
);
6359 err
= btrfs_find_free_ino(root
, &objectid
);
6363 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6364 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6366 if (IS_ERR(inode
)) {
6367 err
= PTR_ERR(inode
);
6372 * If the active LSM wants to access the inode during
6373 * d_instantiate it needs these. Smack checks to see
6374 * if the filesystem supports xattrs by looking at the
6377 inode
->i_op
= &btrfs_special_inode_operations
;
6378 init_special_inode(inode
, inode
->i_mode
, rdev
);
6380 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6382 goto out_unlock_inode
;
6384 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
6386 goto out_unlock_inode
;
6388 btrfs_update_inode(trans
, root
, inode
);
6389 unlock_new_inode(inode
);
6390 d_instantiate(dentry
, inode
);
6394 btrfs_end_transaction(trans
, root
);
6395 btrfs_balance_delayed_items(root
);
6396 btrfs_btree_balance_dirty(root
);
6398 inode_dec_link_count(inode
);
6405 unlock_new_inode(inode
);
6410 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
6411 umode_t mode
, bool excl
)
6413 struct btrfs_trans_handle
*trans
;
6414 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6415 struct inode
*inode
= NULL
;
6416 int drop_inode_on_err
= 0;
6422 * 2 for inode item and ref
6424 * 1 for xattr if selinux is on
6426 trans
= btrfs_start_transaction(root
, 5);
6428 return PTR_ERR(trans
);
6430 err
= btrfs_find_free_ino(root
, &objectid
);
6434 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6435 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6437 if (IS_ERR(inode
)) {
6438 err
= PTR_ERR(inode
);
6441 drop_inode_on_err
= 1;
6443 * If the active LSM wants to access the inode during
6444 * d_instantiate it needs these. Smack checks to see
6445 * if the filesystem supports xattrs by looking at the
6448 inode
->i_fop
= &btrfs_file_operations
;
6449 inode
->i_op
= &btrfs_file_inode_operations
;
6450 inode
->i_mapping
->a_ops
= &btrfs_aops
;
6452 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6454 goto out_unlock_inode
;
6456 err
= btrfs_update_inode(trans
, root
, inode
);
6458 goto out_unlock_inode
;
6460 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
6462 goto out_unlock_inode
;
6464 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
6465 unlock_new_inode(inode
);
6466 d_instantiate(dentry
, inode
);
6469 btrfs_end_transaction(trans
, root
);
6470 if (err
&& drop_inode_on_err
) {
6471 inode_dec_link_count(inode
);
6474 btrfs_balance_delayed_items(root
);
6475 btrfs_btree_balance_dirty(root
);
6479 unlock_new_inode(inode
);
6484 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
6485 struct dentry
*dentry
)
6487 struct btrfs_trans_handle
*trans
;
6488 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6489 struct inode
*inode
= d_inode(old_dentry
);
6494 /* do not allow sys_link's with other subvols of the same device */
6495 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
6498 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
6501 err
= btrfs_set_inode_index(dir
, &index
);
6506 * 2 items for inode and inode ref
6507 * 2 items for dir items
6508 * 1 item for parent inode
6510 trans
= btrfs_start_transaction(root
, 5);
6511 if (IS_ERR(trans
)) {
6512 err
= PTR_ERR(trans
);
6516 /* There are several dir indexes for this inode, clear the cache. */
6517 BTRFS_I(inode
)->dir_index
= 0ULL;
6519 inode_inc_iversion(inode
);
6520 inode
->i_ctime
= CURRENT_TIME
;
6522 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
6524 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
6529 struct dentry
*parent
= dentry
->d_parent
;
6530 err
= btrfs_update_inode(trans
, root
, inode
);
6533 if (inode
->i_nlink
== 1) {
6535 * If new hard link count is 1, it's a file created
6536 * with open(2) O_TMPFILE flag.
6538 err
= btrfs_orphan_del(trans
, inode
);
6542 d_instantiate(dentry
, inode
);
6543 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
6546 btrfs_end_transaction(trans
, root
);
6547 btrfs_balance_delayed_items(root
);
6550 inode_dec_link_count(inode
);
6553 btrfs_btree_balance_dirty(root
);
6557 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
6559 struct inode
*inode
= NULL
;
6560 struct btrfs_trans_handle
*trans
;
6561 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6563 int drop_on_err
= 0;
6568 * 2 items for inode and ref
6569 * 2 items for dir items
6570 * 1 for xattr if selinux is on
6572 trans
= btrfs_start_transaction(root
, 5);
6574 return PTR_ERR(trans
);
6576 err
= btrfs_find_free_ino(root
, &objectid
);
6580 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6581 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6582 S_IFDIR
| mode
, &index
);
6583 if (IS_ERR(inode
)) {
6584 err
= PTR_ERR(inode
);
6589 /* these must be set before we unlock the inode */
6590 inode
->i_op
= &btrfs_dir_inode_operations
;
6591 inode
->i_fop
= &btrfs_dir_file_operations
;
6593 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6595 goto out_fail_inode
;
6597 btrfs_i_size_write(inode
, 0);
6598 err
= btrfs_update_inode(trans
, root
, inode
);
6600 goto out_fail_inode
;
6602 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
6603 dentry
->d_name
.len
, 0, index
);
6605 goto out_fail_inode
;
6607 d_instantiate(dentry
, inode
);
6609 * mkdir is special. We're unlocking after we call d_instantiate
6610 * to avoid a race with nfsd calling d_instantiate.
6612 unlock_new_inode(inode
);
6616 btrfs_end_transaction(trans
, root
);
6618 inode_dec_link_count(inode
);
6621 btrfs_balance_delayed_items(root
);
6622 btrfs_btree_balance_dirty(root
);
6626 unlock_new_inode(inode
);
6630 /* Find next extent map of a given extent map, caller needs to ensure locks */
6631 static struct extent_map
*next_extent_map(struct extent_map
*em
)
6633 struct rb_node
*next
;
6635 next
= rb_next(&em
->rb_node
);
6638 return container_of(next
, struct extent_map
, rb_node
);
6641 static struct extent_map
*prev_extent_map(struct extent_map
*em
)
6643 struct rb_node
*prev
;
6645 prev
= rb_prev(&em
->rb_node
);
6648 return container_of(prev
, struct extent_map
, rb_node
);
6651 /* helper for btfs_get_extent. Given an existing extent in the tree,
6652 * the existing extent is the nearest extent to map_start,
6653 * and an extent that you want to insert, deal with overlap and insert
6654 * the best fitted new extent into the tree.
6656 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
6657 struct extent_map
*existing
,
6658 struct extent_map
*em
,
6661 struct extent_map
*prev
;
6662 struct extent_map
*next
;
6667 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
6669 if (existing
->start
> map_start
) {
6671 prev
= prev_extent_map(next
);
6674 next
= next_extent_map(prev
);
6677 start
= prev
? extent_map_end(prev
) : em
->start
;
6678 start
= max_t(u64
, start
, em
->start
);
6679 end
= next
? next
->start
: extent_map_end(em
);
6680 end
= min_t(u64
, end
, extent_map_end(em
));
6681 start_diff
= start
- em
->start
;
6683 em
->len
= end
- start
;
6684 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
6685 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
6686 em
->block_start
+= start_diff
;
6687 em
->block_len
-= start_diff
;
6689 return add_extent_mapping(em_tree
, em
, 0);
6692 static noinline
int uncompress_inline(struct btrfs_path
*path
,
6693 struct inode
*inode
, struct page
*page
,
6694 size_t pg_offset
, u64 extent_offset
,
6695 struct btrfs_file_extent_item
*item
)
6698 struct extent_buffer
*leaf
= path
->nodes
[0];
6701 unsigned long inline_size
;
6705 WARN_ON(pg_offset
!= 0);
6706 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6707 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6708 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
6709 btrfs_item_nr(path
->slots
[0]));
6710 tmp
= kmalloc(inline_size
, GFP_NOFS
);
6713 ptr
= btrfs_file_extent_inline_start(item
);
6715 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
6717 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
6718 ret
= btrfs_decompress(compress_type
, tmp
, page
,
6719 extent_offset
, inline_size
, max_size
);
6725 * a bit scary, this does extent mapping from logical file offset to the disk.
6726 * the ugly parts come from merging extents from the disk with the in-ram
6727 * representation. This gets more complex because of the data=ordered code,
6728 * where the in-ram extents might be locked pending data=ordered completion.
6730 * This also copies inline extents directly into the page.
6733 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
6734 size_t pg_offset
, u64 start
, u64 len
,
6739 u64 extent_start
= 0;
6741 u64 objectid
= btrfs_ino(inode
);
6743 struct btrfs_path
*path
= NULL
;
6744 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6745 struct btrfs_file_extent_item
*item
;
6746 struct extent_buffer
*leaf
;
6747 struct btrfs_key found_key
;
6748 struct extent_map
*em
= NULL
;
6749 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
6750 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6751 struct btrfs_trans_handle
*trans
= NULL
;
6752 const bool new_inline
= !page
|| create
;
6755 read_lock(&em_tree
->lock
);
6756 em
= lookup_extent_mapping(em_tree
, start
, len
);
6758 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6759 read_unlock(&em_tree
->lock
);
6762 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
6763 free_extent_map(em
);
6764 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
6765 free_extent_map(em
);
6769 em
= alloc_extent_map();
6774 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6775 em
->start
= EXTENT_MAP_HOLE
;
6776 em
->orig_start
= EXTENT_MAP_HOLE
;
6778 em
->block_len
= (u64
)-1;
6781 path
= btrfs_alloc_path();
6787 * Chances are we'll be called again, so go ahead and do
6793 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
6794 objectid
, start
, trans
!= NULL
);
6801 if (path
->slots
[0] == 0)
6806 leaf
= path
->nodes
[0];
6807 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6808 struct btrfs_file_extent_item
);
6809 /* are we inside the extent that was found? */
6810 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6811 found_type
= found_key
.type
;
6812 if (found_key
.objectid
!= objectid
||
6813 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
6815 * If we backup past the first extent we want to move forward
6816 * and see if there is an extent in front of us, otherwise we'll
6817 * say there is a hole for our whole search range which can
6824 found_type
= btrfs_file_extent_type(leaf
, item
);
6825 extent_start
= found_key
.offset
;
6826 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6827 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6828 extent_end
= extent_start
+
6829 btrfs_file_extent_num_bytes(leaf
, item
);
6830 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6832 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6833 extent_end
= ALIGN(extent_start
+ size
, root
->sectorsize
);
6836 if (start
>= extent_end
) {
6838 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
6839 ret
= btrfs_next_leaf(root
, path
);
6846 leaf
= path
->nodes
[0];
6848 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6849 if (found_key
.objectid
!= objectid
||
6850 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6852 if (start
+ len
<= found_key
.offset
)
6854 if (start
> found_key
.offset
)
6857 em
->orig_start
= start
;
6858 em
->len
= found_key
.offset
- start
;
6862 btrfs_extent_item_to_extent_map(inode
, path
, item
, new_inline
, em
);
6864 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6865 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6867 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6871 size_t extent_offset
;
6877 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6878 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
6879 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
6880 size
- extent_offset
);
6881 em
->start
= extent_start
+ extent_offset
;
6882 em
->len
= ALIGN(copy_size
, root
->sectorsize
);
6883 em
->orig_block_len
= em
->len
;
6884 em
->orig_start
= em
->start
;
6885 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
6886 if (create
== 0 && !PageUptodate(page
)) {
6887 if (btrfs_file_extent_compression(leaf
, item
) !=
6888 BTRFS_COMPRESS_NONE
) {
6889 ret
= uncompress_inline(path
, inode
, page
,
6891 extent_offset
, item
);
6898 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6900 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
6901 memset(map
+ pg_offset
+ copy_size
, 0,
6902 PAGE_CACHE_SIZE
- pg_offset
-
6907 flush_dcache_page(page
);
6908 } else if (create
&& PageUptodate(page
)) {
6912 free_extent_map(em
);
6915 btrfs_release_path(path
);
6916 trans
= btrfs_join_transaction(root
);
6919 return ERR_CAST(trans
);
6923 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6926 btrfs_mark_buffer_dirty(leaf
);
6928 set_extent_uptodate(io_tree
, em
->start
,
6929 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
6934 em
->orig_start
= start
;
6937 em
->block_start
= EXTENT_MAP_HOLE
;
6938 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
6940 btrfs_release_path(path
);
6941 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
6942 btrfs_err(root
->fs_info
, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6943 em
->start
, em
->len
, start
, len
);
6949 write_lock(&em_tree
->lock
);
6950 ret
= add_extent_mapping(em_tree
, em
, 0);
6951 /* it is possible that someone inserted the extent into the tree
6952 * while we had the lock dropped. It is also possible that
6953 * an overlapping map exists in the tree
6955 if (ret
== -EEXIST
) {
6956 struct extent_map
*existing
;
6960 existing
= search_extent_mapping(em_tree
, start
, len
);
6962 * existing will always be non-NULL, since there must be
6963 * extent causing the -EEXIST.
6965 if (start
>= extent_map_end(existing
) ||
6966 start
<= existing
->start
) {
6968 * The existing extent map is the one nearest to
6969 * the [start, start + len) range which overlaps
6971 err
= merge_extent_mapping(em_tree
, existing
,
6973 free_extent_map(existing
);
6975 free_extent_map(em
);
6979 free_extent_map(em
);
6984 write_unlock(&em_tree
->lock
);
6987 trace_btrfs_get_extent(root
, em
);
6989 btrfs_free_path(path
);
6991 ret
= btrfs_end_transaction(trans
, root
);
6996 free_extent_map(em
);
6997 return ERR_PTR(err
);
6999 BUG_ON(!em
); /* Error is always set */
7003 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
7004 size_t pg_offset
, u64 start
, u64 len
,
7007 struct extent_map
*em
;
7008 struct extent_map
*hole_em
= NULL
;
7009 u64 range_start
= start
;
7015 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
7022 * - a pre-alloc extent,
7023 * there might actually be delalloc bytes behind it.
7025 if (em
->block_start
!= EXTENT_MAP_HOLE
&&
7026 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7032 /* check to see if we've wrapped (len == -1 or similar) */
7041 /* ok, we didn't find anything, lets look for delalloc */
7042 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
7043 end
, len
, EXTENT_DELALLOC
, 1);
7044 found_end
= range_start
+ found
;
7045 if (found_end
< range_start
)
7046 found_end
= (u64
)-1;
7049 * we didn't find anything useful, return
7050 * the original results from get_extent()
7052 if (range_start
> end
|| found_end
<= start
) {
7058 /* adjust the range_start to make sure it doesn't
7059 * go backwards from the start they passed in
7061 range_start
= max(start
, range_start
);
7062 found
= found_end
- range_start
;
7065 u64 hole_start
= start
;
7068 em
= alloc_extent_map();
7074 * when btrfs_get_extent can't find anything it
7075 * returns one huge hole
7077 * make sure what it found really fits our range, and
7078 * adjust to make sure it is based on the start from
7082 u64 calc_end
= extent_map_end(hole_em
);
7084 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
7085 free_extent_map(hole_em
);
7088 hole_start
= max(hole_em
->start
, start
);
7089 hole_len
= calc_end
- hole_start
;
7093 if (hole_em
&& range_start
> hole_start
) {
7094 /* our hole starts before our delalloc, so we
7095 * have to return just the parts of the hole
7096 * that go until the delalloc starts
7098 em
->len
= min(hole_len
,
7099 range_start
- hole_start
);
7100 em
->start
= hole_start
;
7101 em
->orig_start
= hole_start
;
7103 * don't adjust block start at all,
7104 * it is fixed at EXTENT_MAP_HOLE
7106 em
->block_start
= hole_em
->block_start
;
7107 em
->block_len
= hole_len
;
7108 if (test_bit(EXTENT_FLAG_PREALLOC
, &hole_em
->flags
))
7109 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
7111 em
->start
= range_start
;
7113 em
->orig_start
= range_start
;
7114 em
->block_start
= EXTENT_MAP_DELALLOC
;
7115 em
->block_len
= found
;
7117 } else if (hole_em
) {
7122 free_extent_map(hole_em
);
7124 free_extent_map(em
);
7125 return ERR_PTR(err
);
7130 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
7133 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7134 struct extent_map
*em
;
7135 struct btrfs_key ins
;
7139 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
7140 ret
= btrfs_reserve_extent(root
, len
, root
->sectorsize
, 0,
7141 alloc_hint
, &ins
, 1, 1);
7143 return ERR_PTR(ret
);
7145 em
= create_pinned_em(inode
, start
, ins
.offset
, start
, ins
.objectid
,
7146 ins
.offset
, ins
.offset
, ins
.offset
, 0);
7148 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
7152 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
7153 ins
.offset
, ins
.offset
, 0);
7155 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
7156 free_extent_map(em
);
7157 return ERR_PTR(ret
);
7164 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7165 * block must be cow'd
7167 noinline
int can_nocow_extent(struct inode
*inode
, u64 offset
, u64
*len
,
7168 u64
*orig_start
, u64
*orig_block_len
,
7171 struct btrfs_trans_handle
*trans
;
7172 struct btrfs_path
*path
;
7174 struct extent_buffer
*leaf
;
7175 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7176 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7177 struct btrfs_file_extent_item
*fi
;
7178 struct btrfs_key key
;
7185 bool nocow
= (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
);
7187 path
= btrfs_alloc_path();
7191 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, btrfs_ino(inode
),
7196 slot
= path
->slots
[0];
7199 /* can't find the item, must cow */
7206 leaf
= path
->nodes
[0];
7207 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
7208 if (key
.objectid
!= btrfs_ino(inode
) ||
7209 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7210 /* not our file or wrong item type, must cow */
7214 if (key
.offset
> offset
) {
7215 /* Wrong offset, must cow */
7219 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
7220 found_type
= btrfs_file_extent_type(leaf
, fi
);
7221 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
7222 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
7223 /* not a regular extent, must cow */
7227 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_REG
)
7230 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
7231 if (extent_end
<= offset
)
7234 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7235 if (disk_bytenr
== 0)
7238 if (btrfs_file_extent_compression(leaf
, fi
) ||
7239 btrfs_file_extent_encryption(leaf
, fi
) ||
7240 btrfs_file_extent_other_encoding(leaf
, fi
))
7243 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
7246 *orig_start
= key
.offset
- backref_offset
;
7247 *orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7248 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
7251 if (btrfs_extent_readonly(root
, disk_bytenr
))
7254 num_bytes
= min(offset
+ *len
, extent_end
) - offset
;
7255 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7258 range_end
= round_up(offset
+ num_bytes
, root
->sectorsize
) - 1;
7259 ret
= test_range_bit(io_tree
, offset
, range_end
,
7260 EXTENT_DELALLOC
, 0, NULL
);
7267 btrfs_release_path(path
);
7270 * look for other files referencing this extent, if we
7271 * find any we must cow
7273 trans
= btrfs_join_transaction(root
);
7274 if (IS_ERR(trans
)) {
7279 ret
= btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
7280 key
.offset
- backref_offset
, disk_bytenr
);
7281 btrfs_end_transaction(trans
, root
);
7288 * adjust disk_bytenr and num_bytes to cover just the bytes
7289 * in this extent we are about to write. If there
7290 * are any csums in that range we have to cow in order
7291 * to keep the csums correct
7293 disk_bytenr
+= backref_offset
;
7294 disk_bytenr
+= offset
- key
.offset
;
7295 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
7298 * all of the above have passed, it is safe to overwrite this extent
7304 btrfs_free_path(path
);
7308 bool btrfs_page_exists_in_range(struct inode
*inode
, loff_t start
, loff_t end
)
7310 struct radix_tree_root
*root
= &inode
->i_mapping
->page_tree
;
7312 void **pagep
= NULL
;
7313 struct page
*page
= NULL
;
7317 start_idx
= start
>> PAGE_CACHE_SHIFT
;
7320 * end is the last byte in the last page. end == start is legal
7322 end_idx
= end
>> PAGE_CACHE_SHIFT
;
7326 /* Most of the code in this while loop is lifted from
7327 * find_get_page. It's been modified to begin searching from a
7328 * page and return just the first page found in that range. If the
7329 * found idx is less than or equal to the end idx then we know that
7330 * a page exists. If no pages are found or if those pages are
7331 * outside of the range then we're fine (yay!) */
7332 while (page
== NULL
&&
7333 radix_tree_gang_lookup_slot(root
, &pagep
, NULL
, start_idx
, 1)) {
7334 page
= radix_tree_deref_slot(pagep
);
7335 if (unlikely(!page
))
7338 if (radix_tree_exception(page
)) {
7339 if (radix_tree_deref_retry(page
)) {
7344 * Otherwise, shmem/tmpfs must be storing a swap entry
7345 * here as an exceptional entry: so return it without
7346 * attempting to raise page count.
7349 break; /* TODO: Is this relevant for this use case? */
7352 if (!page_cache_get_speculative(page
)) {
7358 * Has the page moved?
7359 * This is part of the lockless pagecache protocol. See
7360 * include/linux/pagemap.h for details.
7362 if (unlikely(page
!= *pagep
)) {
7363 page_cache_release(page
);
7369 if (page
->index
<= end_idx
)
7371 page_cache_release(page
);
7378 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
7379 struct extent_state
**cached_state
, int writing
)
7381 struct btrfs_ordered_extent
*ordered
;
7385 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7388 * We're concerned with the entire range that we're going to be
7389 * doing DIO to, so we need to make sure theres no ordered
7390 * extents in this range.
7392 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
7393 lockend
- lockstart
+ 1);
7396 * We need to make sure there are no buffered pages in this
7397 * range either, we could have raced between the invalidate in
7398 * generic_file_direct_write and locking the extent. The
7399 * invalidate needs to happen so that reads after a write do not
7404 !btrfs_page_exists_in_range(inode
, lockstart
, lockend
)))
7407 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7408 cached_state
, GFP_NOFS
);
7411 btrfs_start_ordered_extent(inode
, ordered
, 1);
7412 btrfs_put_ordered_extent(ordered
);
7414 /* Screw you mmap */
7415 ret
= btrfs_fdatawrite_range(inode
, lockstart
, lockend
);
7418 ret
= filemap_fdatawait_range(inode
->i_mapping
,
7425 * If we found a page that couldn't be invalidated just
7426 * fall back to buffered.
7428 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
7429 lockstart
>> PAGE_CACHE_SHIFT
,
7430 lockend
>> PAGE_CACHE_SHIFT
);
7441 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
7442 u64 len
, u64 orig_start
,
7443 u64 block_start
, u64 block_len
,
7444 u64 orig_block_len
, u64 ram_bytes
,
7447 struct extent_map_tree
*em_tree
;
7448 struct extent_map
*em
;
7449 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7452 em_tree
= &BTRFS_I(inode
)->extent_tree
;
7453 em
= alloc_extent_map();
7455 return ERR_PTR(-ENOMEM
);
7458 em
->orig_start
= orig_start
;
7459 em
->mod_start
= start
;
7462 em
->block_len
= block_len
;
7463 em
->block_start
= block_start
;
7464 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
7465 em
->orig_block_len
= orig_block_len
;
7466 em
->ram_bytes
= ram_bytes
;
7467 em
->generation
= -1;
7468 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
7469 if (type
== BTRFS_ORDERED_PREALLOC
)
7470 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
7473 btrfs_drop_extent_cache(inode
, em
->start
,
7474 em
->start
+ em
->len
- 1, 0);
7475 write_lock(&em_tree
->lock
);
7476 ret
= add_extent_mapping(em_tree
, em
, 1);
7477 write_unlock(&em_tree
->lock
);
7478 } while (ret
== -EEXIST
);
7481 free_extent_map(em
);
7482 return ERR_PTR(ret
);
7488 struct btrfs_dio_data
{
7489 u64 outstanding_extents
;
7493 static void adjust_dio_outstanding_extents(struct inode
*inode
,
7494 struct btrfs_dio_data
*dio_data
,
7497 unsigned num_extents
;
7499 num_extents
= (unsigned) div64_u64(len
+ BTRFS_MAX_EXTENT_SIZE
- 1,
7500 BTRFS_MAX_EXTENT_SIZE
);
7502 * If we have an outstanding_extents count still set then we're
7503 * within our reservation, otherwise we need to adjust our inode
7504 * counter appropriately.
7506 if (dio_data
->outstanding_extents
) {
7507 dio_data
->outstanding_extents
-= num_extents
;
7509 spin_lock(&BTRFS_I(inode
)->lock
);
7510 BTRFS_I(inode
)->outstanding_extents
+= num_extents
;
7511 spin_unlock(&BTRFS_I(inode
)->lock
);
7515 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
7516 struct buffer_head
*bh_result
, int create
)
7518 struct extent_map
*em
;
7519 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7520 struct extent_state
*cached_state
= NULL
;
7521 struct btrfs_dio_data
*dio_data
= NULL
;
7522 u64 start
= iblock
<< inode
->i_blkbits
;
7523 u64 lockstart
, lockend
;
7524 u64 len
= bh_result
->b_size
;
7525 int unlock_bits
= EXTENT_LOCKED
;
7529 unlock_bits
|= EXTENT_DIRTY
;
7531 len
= min_t(u64
, len
, root
->sectorsize
);
7534 lockend
= start
+ len
- 1;
7536 if (current
->journal_info
) {
7538 * Need to pull our outstanding extents and set journal_info to NULL so
7539 * that anything that needs to check if there's a transction doesn't get
7542 dio_data
= current
->journal_info
;
7543 current
->journal_info
= NULL
;
7547 * If this errors out it's because we couldn't invalidate pagecache for
7548 * this range and we need to fallback to buffered.
7550 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
,
7556 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
7563 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7564 * io. INLINE is special, and we could probably kludge it in here, but
7565 * it's still buffered so for safety lets just fall back to the generic
7568 * For COMPRESSED we _have_ to read the entire extent in so we can
7569 * decompress it, so there will be buffering required no matter what we
7570 * do, so go ahead and fallback to buffered.
7572 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7573 * to buffered IO. Don't blame me, this is the price we pay for using
7576 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
7577 em
->block_start
== EXTENT_MAP_INLINE
) {
7578 free_extent_map(em
);
7583 /* Just a good old fashioned hole, return */
7584 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
7585 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
7586 free_extent_map(em
);
7591 * We don't allocate a new extent in the following cases
7593 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7595 * 2) The extent is marked as PREALLOC. We're good to go here and can
7596 * just use the extent.
7600 len
= min(len
, em
->len
- (start
- em
->start
));
7601 lockstart
= start
+ len
;
7605 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
7606 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
7607 em
->block_start
!= EXTENT_MAP_HOLE
)) {
7609 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
7611 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7612 type
= BTRFS_ORDERED_PREALLOC
;
7614 type
= BTRFS_ORDERED_NOCOW
;
7615 len
= min(len
, em
->len
- (start
- em
->start
));
7616 block_start
= em
->block_start
+ (start
- em
->start
);
7618 if (can_nocow_extent(inode
, start
, &len
, &orig_start
,
7619 &orig_block_len
, &ram_bytes
) == 1) {
7620 if (type
== BTRFS_ORDERED_PREALLOC
) {
7621 free_extent_map(em
);
7622 em
= create_pinned_em(inode
, start
, len
,
7633 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
7634 block_start
, len
, len
, type
);
7636 free_extent_map(em
);
7644 * this will cow the extent, reset the len in case we changed
7647 len
= bh_result
->b_size
;
7648 free_extent_map(em
);
7649 em
= btrfs_new_extent_direct(inode
, start
, len
);
7654 len
= min(len
, em
->len
- (start
- em
->start
));
7656 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
7658 bh_result
->b_size
= len
;
7659 bh_result
->b_bdev
= em
->bdev
;
7660 set_buffer_mapped(bh_result
);
7662 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7663 set_buffer_new(bh_result
);
7666 * Need to update the i_size under the extent lock so buffered
7667 * readers will get the updated i_size when we unlock.
7669 if (start
+ len
> i_size_read(inode
))
7670 i_size_write(inode
, start
+ len
);
7672 adjust_dio_outstanding_extents(inode
, dio_data
, len
);
7673 btrfs_free_reserved_data_space(inode
, start
, len
);
7674 WARN_ON(dio_data
->reserve
< len
);
7675 dio_data
->reserve
-= len
;
7676 current
->journal_info
= dio_data
;
7680 * In the case of write we need to clear and unlock the entire range,
7681 * in the case of read we need to unlock only the end area that we
7682 * aren't using if there is any left over space.
7684 if (lockstart
< lockend
) {
7685 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
7686 lockend
, unlock_bits
, 1, 0,
7687 &cached_state
, GFP_NOFS
);
7689 free_extent_state(cached_state
);
7692 free_extent_map(em
);
7697 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7698 unlock_bits
, 1, 0, &cached_state
, GFP_NOFS
);
7701 current
->journal_info
= dio_data
;
7703 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7704 * write less data then expected, so that we don't underflow our inode's
7705 * outstanding extents counter.
7707 if (create
&& dio_data
)
7708 adjust_dio_outstanding_extents(inode
, dio_data
, len
);
7713 static inline int submit_dio_repair_bio(struct inode
*inode
, struct bio
*bio
,
7714 int rw
, int mirror_num
)
7716 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7719 BUG_ON(rw
& REQ_WRITE
);
7723 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
,
7724 BTRFS_WQ_ENDIO_DIO_REPAIR
);
7728 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
7734 static int btrfs_check_dio_repairable(struct inode
*inode
,
7735 struct bio
*failed_bio
,
7736 struct io_failure_record
*failrec
,
7741 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
7742 failrec
->logical
, failrec
->len
);
7743 if (num_copies
== 1) {
7745 * we only have a single copy of the data, so don't bother with
7746 * all the retry and error correction code that follows. no
7747 * matter what the error is, it is very likely to persist.
7749 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7750 num_copies
, failrec
->this_mirror
, failed_mirror
);
7754 failrec
->failed_mirror
= failed_mirror
;
7755 failrec
->this_mirror
++;
7756 if (failrec
->this_mirror
== failed_mirror
)
7757 failrec
->this_mirror
++;
7759 if (failrec
->this_mirror
> num_copies
) {
7760 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7761 num_copies
, failrec
->this_mirror
, failed_mirror
);
7768 static int dio_read_error(struct inode
*inode
, struct bio
*failed_bio
,
7769 struct page
*page
, u64 start
, u64 end
,
7770 int failed_mirror
, bio_end_io_t
*repair_endio
,
7773 struct io_failure_record
*failrec
;
7779 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
7781 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
7785 ret
= btrfs_check_dio_repairable(inode
, failed_bio
, failrec
,
7788 free_io_failure(inode
, failrec
);
7792 if (failed_bio
->bi_vcnt
> 1)
7793 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
7795 read_mode
= READ_SYNC
;
7797 isector
= start
- btrfs_io_bio(failed_bio
)->logical
;
7798 isector
>>= inode
->i_sb
->s_blocksize_bits
;
7799 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
7800 0, isector
, repair_endio
, repair_arg
);
7802 free_io_failure(inode
, failrec
);
7806 btrfs_debug(BTRFS_I(inode
)->root
->fs_info
,
7807 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7808 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
7810 ret
= submit_dio_repair_bio(inode
, bio
, read_mode
,
7811 failrec
->this_mirror
);
7813 free_io_failure(inode
, failrec
);
7820 struct btrfs_retry_complete
{
7821 struct completion done
;
7822 struct inode
*inode
;
7827 static void btrfs_retry_endio_nocsum(struct bio
*bio
)
7829 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7830 struct bio_vec
*bvec
;
7837 bio_for_each_segment_all(bvec
, bio
, i
)
7838 clean_io_failure(done
->inode
, done
->start
, bvec
->bv_page
, 0);
7840 complete(&done
->done
);
7844 static int __btrfs_correct_data_nocsum(struct inode
*inode
,
7845 struct btrfs_io_bio
*io_bio
)
7847 struct bio_vec
*bvec
;
7848 struct btrfs_retry_complete done
;
7853 start
= io_bio
->logical
;
7856 bio_for_each_segment_all(bvec
, &io_bio
->bio
, i
) {
7860 init_completion(&done
.done
);
7862 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
->bv_page
, start
,
7863 start
+ bvec
->bv_len
- 1,
7865 btrfs_retry_endio_nocsum
, &done
);
7869 wait_for_completion(&done
.done
);
7871 if (!done
.uptodate
) {
7872 /* We might have another mirror, so try again */
7876 start
+= bvec
->bv_len
;
7882 static void btrfs_retry_endio(struct bio
*bio
)
7884 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7885 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
7886 struct bio_vec
*bvec
;
7895 bio_for_each_segment_all(bvec
, bio
, i
) {
7896 ret
= __readpage_endio_check(done
->inode
, io_bio
, i
,
7898 done
->start
, bvec
->bv_len
);
7900 clean_io_failure(done
->inode
, done
->start
,
7906 done
->uptodate
= uptodate
;
7908 complete(&done
->done
);
7912 static int __btrfs_subio_endio_read(struct inode
*inode
,
7913 struct btrfs_io_bio
*io_bio
, int err
)
7915 struct bio_vec
*bvec
;
7916 struct btrfs_retry_complete done
;
7923 start
= io_bio
->logical
;
7926 bio_for_each_segment_all(bvec
, &io_bio
->bio
, i
) {
7927 ret
= __readpage_endio_check(inode
, io_bio
, i
, bvec
->bv_page
,
7928 0, start
, bvec
->bv_len
);
7934 init_completion(&done
.done
);
7936 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
->bv_page
, start
,
7937 start
+ bvec
->bv_len
- 1,
7939 btrfs_retry_endio
, &done
);
7945 wait_for_completion(&done
.done
);
7947 if (!done
.uptodate
) {
7948 /* We might have another mirror, so try again */
7952 offset
+= bvec
->bv_len
;
7953 start
+= bvec
->bv_len
;
7959 static int btrfs_subio_endio_read(struct inode
*inode
,
7960 struct btrfs_io_bio
*io_bio
, int err
)
7962 bool skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
7966 return __btrfs_correct_data_nocsum(inode
, io_bio
);
7970 return __btrfs_subio_endio_read(inode
, io_bio
, err
);
7974 static void btrfs_endio_direct_read(struct bio
*bio
)
7976 struct btrfs_dio_private
*dip
= bio
->bi_private
;
7977 struct inode
*inode
= dip
->inode
;
7978 struct bio
*dio_bio
;
7979 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
7980 int err
= bio
->bi_error
;
7982 if (dip
->flags
& BTRFS_DIO_ORIG_BIO_SUBMITTED
)
7983 err
= btrfs_subio_endio_read(inode
, io_bio
, err
);
7985 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
7986 dip
->logical_offset
+ dip
->bytes
- 1);
7987 dio_bio
= dip
->dio_bio
;
7991 dio_end_io(dio_bio
, bio
->bi_error
);
7994 io_bio
->end_io(io_bio
, err
);
7998 static void btrfs_endio_direct_write(struct bio
*bio
)
8000 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8001 struct inode
*inode
= dip
->inode
;
8002 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8003 struct btrfs_ordered_extent
*ordered
= NULL
;
8004 u64 ordered_offset
= dip
->logical_offset
;
8005 u64 ordered_bytes
= dip
->bytes
;
8006 struct bio
*dio_bio
;
8010 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
8017 btrfs_init_work(&ordered
->work
, btrfs_endio_write_helper
,
8018 finish_ordered_fn
, NULL
, NULL
);
8019 btrfs_queue_work(root
->fs_info
->endio_write_workers
,
8023 * our bio might span multiple ordered extents. If we haven't
8024 * completed the accounting for the whole dio, go back and try again
8026 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
8027 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
8032 dio_bio
= dip
->dio_bio
;
8036 dio_end_io(dio_bio
, bio
->bi_error
);
8040 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
8041 struct bio
*bio
, int mirror_num
,
8042 unsigned long bio_flags
, u64 offset
)
8045 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8046 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
8047 BUG_ON(ret
); /* -ENOMEM */
8051 static void btrfs_end_dio_bio(struct bio
*bio
)
8053 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8054 int err
= bio
->bi_error
;
8057 btrfs_warn(BTRFS_I(dip
->inode
)->root
->fs_info
,
8058 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8059 btrfs_ino(dip
->inode
), bio
->bi_rw
,
8060 (unsigned long long)bio
->bi_iter
.bi_sector
,
8061 bio
->bi_iter
.bi_size
, err
);
8063 if (dip
->subio_endio
)
8064 err
= dip
->subio_endio(dip
->inode
, btrfs_io_bio(bio
), err
);
8070 * before atomic variable goto zero, we must make sure
8071 * dip->errors is perceived to be set.
8073 smp_mb__before_atomic();
8076 /* if there are more bios still pending for this dio, just exit */
8077 if (!atomic_dec_and_test(&dip
->pending_bios
))
8081 bio_io_error(dip
->orig_bio
);
8083 dip
->dio_bio
->bi_error
= 0;
8084 bio_endio(dip
->orig_bio
);
8090 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
8091 u64 first_sector
, gfp_t gfp_flags
)
8094 bio
= btrfs_bio_alloc(bdev
, first_sector
, BIO_MAX_PAGES
, gfp_flags
);
8096 bio_associate_current(bio
);
8100 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root
*root
,
8101 struct inode
*inode
,
8102 struct btrfs_dio_private
*dip
,
8106 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8107 struct btrfs_io_bio
*orig_io_bio
= btrfs_io_bio(dip
->orig_bio
);
8111 * We load all the csum data we need when we submit
8112 * the first bio to reduce the csum tree search and
8115 if (dip
->logical_offset
== file_offset
) {
8116 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, dip
->orig_bio
,
8122 if (bio
== dip
->orig_bio
)
8125 file_offset
-= dip
->logical_offset
;
8126 file_offset
>>= inode
->i_sb
->s_blocksize_bits
;
8127 io_bio
->csum
= (u8
*)(((u32
*)orig_io_bio
->csum
) + file_offset
);
8132 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
8133 int rw
, u64 file_offset
, int skip_sum
,
8136 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8137 int write
= rw
& REQ_WRITE
;
8138 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8142 async_submit
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
8147 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
,
8148 BTRFS_WQ_ENDIO_DATA
);
8156 if (write
&& async_submit
) {
8157 ret
= btrfs_wq_submit_bio(root
->fs_info
,
8158 inode
, rw
, bio
, 0, 0,
8160 __btrfs_submit_bio_start_direct_io
,
8161 __btrfs_submit_bio_done
);
8165 * If we aren't doing async submit, calculate the csum of the
8168 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
8172 ret
= btrfs_lookup_and_bind_dio_csum(root
, inode
, dip
, bio
,
8178 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
8184 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
8187 struct inode
*inode
= dip
->inode
;
8188 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8190 struct bio
*orig_bio
= dip
->orig_bio
;
8191 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
8192 u64 start_sector
= orig_bio
->bi_iter
.bi_sector
;
8193 u64 file_offset
= dip
->logical_offset
;
8198 int async_submit
= 0;
8200 map_length
= orig_bio
->bi_iter
.bi_size
;
8201 ret
= btrfs_map_block(root
->fs_info
, rw
, start_sector
<< 9,
8202 &map_length
, NULL
, 0);
8206 if (map_length
>= orig_bio
->bi_iter
.bi_size
) {
8208 dip
->flags
|= BTRFS_DIO_ORIG_BIO_SUBMITTED
;
8212 /* async crcs make it difficult to collect full stripe writes. */
8213 if (btrfs_get_alloc_profile(root
, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK
)
8218 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
8222 bio
->bi_private
= dip
;
8223 bio
->bi_end_io
= btrfs_end_dio_bio
;
8224 btrfs_io_bio(bio
)->logical
= file_offset
;
8225 atomic_inc(&dip
->pending_bios
);
8227 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
8228 if (map_length
< submit_len
+ bvec
->bv_len
||
8229 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
8230 bvec
->bv_offset
) < bvec
->bv_len
) {
8232 * inc the count before we submit the bio so
8233 * we know the end IO handler won't happen before
8234 * we inc the count. Otherwise, the dip might get freed
8235 * before we're done setting it up
8237 atomic_inc(&dip
->pending_bios
);
8238 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
8239 file_offset
, skip_sum
,
8243 atomic_dec(&dip
->pending_bios
);
8247 start_sector
+= submit_len
>> 9;
8248 file_offset
+= submit_len
;
8253 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
8254 start_sector
, GFP_NOFS
);
8257 bio
->bi_private
= dip
;
8258 bio
->bi_end_io
= btrfs_end_dio_bio
;
8259 btrfs_io_bio(bio
)->logical
= file_offset
;
8261 map_length
= orig_bio
->bi_iter
.bi_size
;
8262 ret
= btrfs_map_block(root
->fs_info
, rw
,
8264 &map_length
, NULL
, 0);
8270 submit_len
+= bvec
->bv_len
;
8277 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
8286 * before atomic variable goto zero, we must
8287 * make sure dip->errors is perceived to be set.
8289 smp_mb__before_atomic();
8290 if (atomic_dec_and_test(&dip
->pending_bios
))
8291 bio_io_error(dip
->orig_bio
);
8293 /* bio_end_io() will handle error, so we needn't return it */
8297 static void btrfs_submit_direct(int rw
, struct bio
*dio_bio
,
8298 struct inode
*inode
, loff_t file_offset
)
8300 struct btrfs_dio_private
*dip
= NULL
;
8301 struct bio
*io_bio
= NULL
;
8302 struct btrfs_io_bio
*btrfs_bio
;
8304 int write
= rw
& REQ_WRITE
;
8307 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
8309 io_bio
= btrfs_bio_clone(dio_bio
, GFP_NOFS
);
8315 dip
= kzalloc(sizeof(*dip
), GFP_NOFS
);
8321 dip
->private = dio_bio
->bi_private
;
8323 dip
->logical_offset
= file_offset
;
8324 dip
->bytes
= dio_bio
->bi_iter
.bi_size
;
8325 dip
->disk_bytenr
= (u64
)dio_bio
->bi_iter
.bi_sector
<< 9;
8326 io_bio
->bi_private
= dip
;
8327 dip
->orig_bio
= io_bio
;
8328 dip
->dio_bio
= dio_bio
;
8329 atomic_set(&dip
->pending_bios
, 0);
8330 btrfs_bio
= btrfs_io_bio(io_bio
);
8331 btrfs_bio
->logical
= file_offset
;
8334 io_bio
->bi_end_io
= btrfs_endio_direct_write
;
8336 io_bio
->bi_end_io
= btrfs_endio_direct_read
;
8337 dip
->subio_endio
= btrfs_subio_endio_read
;
8340 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
8344 if (btrfs_bio
->end_io
)
8345 btrfs_bio
->end_io(btrfs_bio
, ret
);
8349 * If we arrived here it means either we failed to submit the dip
8350 * or we either failed to clone the dio_bio or failed to allocate the
8351 * dip. If we cloned the dio_bio and allocated the dip, we can just
8352 * call bio_endio against our io_bio so that we get proper resource
8353 * cleanup if we fail to submit the dip, otherwise, we must do the
8354 * same as btrfs_endio_direct_[write|read] because we can't call these
8355 * callbacks - they require an allocated dip and a clone of dio_bio.
8357 if (io_bio
&& dip
) {
8358 io_bio
->bi_error
= -EIO
;
8361 * The end io callbacks free our dip, do the final put on io_bio
8362 * and all the cleanup and final put for dio_bio (through
8369 struct btrfs_ordered_extent
*ordered
;
8371 ordered
= btrfs_lookup_ordered_extent(inode
,
8373 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
8375 * Decrements our ref on the ordered extent and removes
8376 * the ordered extent from the inode's ordered tree,
8377 * doing all the proper resource cleanup such as for the
8378 * reserved space and waking up any waiters for this
8379 * ordered extent (through btrfs_remove_ordered_extent).
8381 btrfs_finish_ordered_io(ordered
);
8383 unlock_extent(&BTRFS_I(inode
)->io_tree
, file_offset
,
8384 file_offset
+ dio_bio
->bi_iter
.bi_size
- 1);
8386 dio_bio
->bi_error
= -EIO
;
8388 * Releases and cleans up our dio_bio, no need to bio_put()
8389 * nor bio_endio()/bio_io_error() against dio_bio.
8391 dio_end_io(dio_bio
, ret
);
8398 static ssize_t
check_direct_IO(struct btrfs_root
*root
, struct kiocb
*iocb
,
8399 const struct iov_iter
*iter
, loff_t offset
)
8403 unsigned blocksize_mask
= root
->sectorsize
- 1;
8404 ssize_t retval
= -EINVAL
;
8406 if (offset
& blocksize_mask
)
8409 if (iov_iter_alignment(iter
) & blocksize_mask
)
8412 /* If this is a write we don't need to check anymore */
8413 if (iov_iter_rw(iter
) == WRITE
)
8416 * Check to make sure we don't have duplicate iov_base's in this
8417 * iovec, if so return EINVAL, otherwise we'll get csum errors
8418 * when reading back.
8420 for (seg
= 0; seg
< iter
->nr_segs
; seg
++) {
8421 for (i
= seg
+ 1; i
< iter
->nr_segs
; i
++) {
8422 if (iter
->iov
[seg
].iov_base
== iter
->iov
[i
].iov_base
)
8431 static ssize_t
btrfs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
,
8434 struct file
*file
= iocb
->ki_filp
;
8435 struct inode
*inode
= file
->f_mapping
->host
;
8436 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8437 struct btrfs_dio_data dio_data
= { 0 };
8441 bool relock
= false;
8444 if (check_direct_IO(BTRFS_I(inode
)->root
, iocb
, iter
, offset
))
8447 inode_dio_begin(inode
);
8448 smp_mb__after_atomic();
8451 * The generic stuff only does filemap_write_and_wait_range, which
8452 * isn't enough if we've written compressed pages to this area, so
8453 * we need to flush the dirty pages again to make absolutely sure
8454 * that any outstanding dirty pages are on disk.
8456 count
= iov_iter_count(iter
);
8457 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
8458 &BTRFS_I(inode
)->runtime_flags
))
8459 filemap_fdatawrite_range(inode
->i_mapping
, offset
,
8460 offset
+ count
- 1);
8462 if (iov_iter_rw(iter
) == WRITE
) {
8464 * If the write DIO is beyond the EOF, we need update
8465 * the isize, but it is protected by i_mutex. So we can
8466 * not unlock the i_mutex at this case.
8468 if (offset
+ count
<= inode
->i_size
) {
8469 mutex_unlock(&inode
->i_mutex
);
8472 ret
= btrfs_delalloc_reserve_space(inode
, offset
, count
);
8475 dio_data
.outstanding_extents
= div64_u64(count
+
8476 BTRFS_MAX_EXTENT_SIZE
- 1,
8477 BTRFS_MAX_EXTENT_SIZE
);
8480 * We need to know how many extents we reserved so that we can
8481 * do the accounting properly if we go over the number we
8482 * originally calculated. Abuse current->journal_info for this.
8484 dio_data
.reserve
= round_up(count
, root
->sectorsize
);
8485 current
->journal_info
= &dio_data
;
8486 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK
,
8487 &BTRFS_I(inode
)->runtime_flags
)) {
8488 inode_dio_end(inode
);
8489 flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
8493 ret
= __blockdev_direct_IO(iocb
, inode
,
8494 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
8495 iter
, offset
, btrfs_get_blocks_direct
, NULL
,
8496 btrfs_submit_direct
, flags
);
8497 if (iov_iter_rw(iter
) == WRITE
) {
8498 current
->journal_info
= NULL
;
8499 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
8500 if (dio_data
.reserve
)
8501 btrfs_delalloc_release_space(inode
, offset
,
8503 } else if (ret
>= 0 && (size_t)ret
< count
)
8504 btrfs_delalloc_release_space(inode
, offset
,
8505 count
- (size_t)ret
);
8509 inode_dio_end(inode
);
8511 mutex_lock(&inode
->i_mutex
);
8516 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8518 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
8519 __u64 start
, __u64 len
)
8523 ret
= fiemap_check_flags(fieinfo
, BTRFS_FIEMAP_FLAGS
);
8527 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
8530 int btrfs_readpage(struct file
*file
, struct page
*page
)
8532 struct extent_io_tree
*tree
;
8533 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8534 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
8537 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
8539 struct extent_io_tree
*tree
;
8542 if (current
->flags
& PF_MEMALLOC
) {
8543 redirty_page_for_writepage(wbc
, page
);
8547 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8548 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
8551 static int btrfs_writepages(struct address_space
*mapping
,
8552 struct writeback_control
*wbc
)
8554 struct extent_io_tree
*tree
;
8556 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8557 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
8561 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
8562 struct list_head
*pages
, unsigned nr_pages
)
8564 struct extent_io_tree
*tree
;
8565 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8566 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
8569 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8571 struct extent_io_tree
*tree
;
8572 struct extent_map_tree
*map
;
8575 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8576 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
8577 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
8579 ClearPagePrivate(page
);
8580 set_page_private(page
, 0);
8581 page_cache_release(page
);
8586 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8588 if (PageWriteback(page
) || PageDirty(page
))
8590 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
8593 static void btrfs_invalidatepage(struct page
*page
, unsigned int offset
,
8594 unsigned int length
)
8596 struct inode
*inode
= page
->mapping
->host
;
8597 struct extent_io_tree
*tree
;
8598 struct btrfs_ordered_extent
*ordered
;
8599 struct extent_state
*cached_state
= NULL
;
8600 u64 page_start
= page_offset(page
);
8601 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
8602 int inode_evicting
= inode
->i_state
& I_FREEING
;
8605 * we have the page locked, so new writeback can't start,
8606 * and the dirty bit won't be cleared while we are here.
8608 * Wait for IO on this page so that we can safely clear
8609 * the PagePrivate2 bit and do ordered accounting
8611 wait_on_page_writeback(page
);
8613 tree
= &BTRFS_I(inode
)->io_tree
;
8615 btrfs_releasepage(page
, GFP_NOFS
);
8619 if (!inode_evicting
)
8620 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
);
8621 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
8624 * IO on this page will never be started, so we need
8625 * to account for any ordered extents now
8627 if (!inode_evicting
)
8628 clear_extent_bit(tree
, page_start
, page_end
,
8629 EXTENT_DIRTY
| EXTENT_DELALLOC
|
8630 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
8631 EXTENT_DEFRAG
, 1, 0, &cached_state
,
8634 * whoever cleared the private bit is responsible
8635 * for the finish_ordered_io
8637 if (TestClearPagePrivate2(page
)) {
8638 struct btrfs_ordered_inode_tree
*tree
;
8641 tree
= &BTRFS_I(inode
)->ordered_tree
;
8643 spin_lock_irq(&tree
->lock
);
8644 set_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
);
8645 new_len
= page_start
- ordered
->file_offset
;
8646 if (new_len
< ordered
->truncated_len
)
8647 ordered
->truncated_len
= new_len
;
8648 spin_unlock_irq(&tree
->lock
);
8650 if (btrfs_dec_test_ordered_pending(inode
, &ordered
,
8652 PAGE_CACHE_SIZE
, 1))
8653 btrfs_finish_ordered_io(ordered
);
8655 btrfs_put_ordered_extent(ordered
);
8656 if (!inode_evicting
) {
8657 cached_state
= NULL
;
8658 lock_extent_bits(tree
, page_start
, page_end
, 0,
8664 * Qgroup reserved space handler
8665 * Page here will be either
8666 * 1) Already written to disk
8667 * In this case, its reserved space is released from data rsv map
8668 * and will be freed by delayed_ref handler finally.
8669 * So even we call qgroup_free_data(), it won't decrease reserved
8671 * 2) Not written to disk
8672 * This means the reserved space should be freed here.
8674 btrfs_qgroup_free_data(inode
, page_start
, PAGE_CACHE_SIZE
);
8675 if (!inode_evicting
) {
8676 clear_extent_bit(tree
, page_start
, page_end
,
8677 EXTENT_LOCKED
| EXTENT_DIRTY
|
8678 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
8679 EXTENT_DEFRAG
, 1, 1,
8680 &cached_state
, GFP_NOFS
);
8682 __btrfs_releasepage(page
, GFP_NOFS
);
8685 ClearPageChecked(page
);
8686 if (PagePrivate(page
)) {
8687 ClearPagePrivate(page
);
8688 set_page_private(page
, 0);
8689 page_cache_release(page
);
8694 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8695 * called from a page fault handler when a page is first dirtied. Hence we must
8696 * be careful to check for EOF conditions here. We set the page up correctly
8697 * for a written page which means we get ENOSPC checking when writing into
8698 * holes and correct delalloc and unwritten extent mapping on filesystems that
8699 * support these features.
8701 * We are not allowed to take the i_mutex here so we have to play games to
8702 * protect against truncate races as the page could now be beyond EOF. Because
8703 * vmtruncate() writes the inode size before removing pages, once we have the
8704 * page lock we can determine safely if the page is beyond EOF. If it is not
8705 * beyond EOF, then the page is guaranteed safe against truncation until we
8708 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
8710 struct page
*page
= vmf
->page
;
8711 struct inode
*inode
= file_inode(vma
->vm_file
);
8712 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8713 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
8714 struct btrfs_ordered_extent
*ordered
;
8715 struct extent_state
*cached_state
= NULL
;
8717 unsigned long zero_start
;
8724 sb_start_pagefault(inode
->i_sb
);
8725 page_start
= page_offset(page
);
8726 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
8728 ret
= btrfs_delalloc_reserve_space(inode
, page_start
,
8731 ret
= file_update_time(vma
->vm_file
);
8737 else /* -ENOSPC, -EIO, etc */
8738 ret
= VM_FAULT_SIGBUS
;
8744 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
8747 size
= i_size_read(inode
);
8749 if ((page
->mapping
!= inode
->i_mapping
) ||
8750 (page_start
>= size
)) {
8751 /* page got truncated out from underneath us */
8754 wait_on_page_writeback(page
);
8756 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
8757 set_page_extent_mapped(page
);
8760 * we can't set the delalloc bits if there are pending ordered
8761 * extents. Drop our locks and wait for them to finish
8763 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
8765 unlock_extent_cached(io_tree
, page_start
, page_end
,
8766 &cached_state
, GFP_NOFS
);
8768 btrfs_start_ordered_extent(inode
, ordered
, 1);
8769 btrfs_put_ordered_extent(ordered
);
8774 * XXX - page_mkwrite gets called every time the page is dirtied, even
8775 * if it was already dirty, so for space accounting reasons we need to
8776 * clear any delalloc bits for the range we are fixing to save. There
8777 * is probably a better way to do this, but for now keep consistent with
8778 * prepare_pages in the normal write path.
8780 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
8781 EXTENT_DIRTY
| EXTENT_DELALLOC
|
8782 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
8783 0, 0, &cached_state
, GFP_NOFS
);
8785 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
8788 unlock_extent_cached(io_tree
, page_start
, page_end
,
8789 &cached_state
, GFP_NOFS
);
8790 ret
= VM_FAULT_SIGBUS
;
8795 /* page is wholly or partially inside EOF */
8796 if (page_start
+ PAGE_CACHE_SIZE
> size
)
8797 zero_start
= size
& ~PAGE_CACHE_MASK
;
8799 zero_start
= PAGE_CACHE_SIZE
;
8801 if (zero_start
!= PAGE_CACHE_SIZE
) {
8803 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
8804 flush_dcache_page(page
);
8807 ClearPageChecked(page
);
8808 set_page_dirty(page
);
8809 SetPageUptodate(page
);
8811 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
8812 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
8813 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
8815 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
8819 sb_end_pagefault(inode
->i_sb
);
8820 return VM_FAULT_LOCKED
;
8824 btrfs_delalloc_release_space(inode
, page_start
, PAGE_CACHE_SIZE
);
8826 sb_end_pagefault(inode
->i_sb
);
8830 static int btrfs_truncate(struct inode
*inode
)
8832 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8833 struct btrfs_block_rsv
*rsv
;
8836 struct btrfs_trans_handle
*trans
;
8837 u64 mask
= root
->sectorsize
- 1;
8838 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
8840 ret
= btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
),
8846 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8847 * 3 things going on here
8849 * 1) We need to reserve space for our orphan item and the space to
8850 * delete our orphan item. Lord knows we don't want to have a dangling
8851 * orphan item because we didn't reserve space to remove it.
8853 * 2) We need to reserve space to update our inode.
8855 * 3) We need to have something to cache all the space that is going to
8856 * be free'd up by the truncate operation, but also have some slack
8857 * space reserved in case it uses space during the truncate (thank you
8858 * very much snapshotting).
8860 * And we need these to all be seperate. The fact is we can use alot of
8861 * space doing the truncate, and we have no earthly idea how much space
8862 * we will use, so we need the truncate reservation to be seperate so it
8863 * doesn't end up using space reserved for updating the inode or
8864 * removing the orphan item. We also need to be able to stop the
8865 * transaction and start a new one, which means we need to be able to
8866 * update the inode several times, and we have no idea of knowing how
8867 * many times that will be, so we can't just reserve 1 item for the
8868 * entirety of the opration, so that has to be done seperately as well.
8869 * Then there is the orphan item, which does indeed need to be held on
8870 * to for the whole operation, and we need nobody to touch this reserved
8871 * space except the orphan code.
8873 * So that leaves us with
8875 * 1) root->orphan_block_rsv - for the orphan deletion.
8876 * 2) rsv - for the truncate reservation, which we will steal from the
8877 * transaction reservation.
8878 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8879 * updating the inode.
8881 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
8884 rsv
->size
= min_size
;
8888 * 1 for the truncate slack space
8889 * 1 for updating the inode.
8891 trans
= btrfs_start_transaction(root
, 2);
8892 if (IS_ERR(trans
)) {
8893 err
= PTR_ERR(trans
);
8897 /* Migrate the slack space for the truncate to our reserve */
8898 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
8903 * So if we truncate and then write and fsync we normally would just
8904 * write the extents that changed, which is a problem if we need to
8905 * first truncate that entire inode. So set this flag so we write out
8906 * all of the extents in the inode to the sync log so we're completely
8909 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
8910 trans
->block_rsv
= rsv
;
8913 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
8915 BTRFS_EXTENT_DATA_KEY
);
8916 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
) {
8921 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
8922 ret
= btrfs_update_inode(trans
, root
, inode
);
8928 btrfs_end_transaction(trans
, root
);
8929 btrfs_btree_balance_dirty(root
);
8931 trans
= btrfs_start_transaction(root
, 2);
8932 if (IS_ERR(trans
)) {
8933 ret
= err
= PTR_ERR(trans
);
8938 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
,
8940 BUG_ON(ret
); /* shouldn't happen */
8941 trans
->block_rsv
= rsv
;
8944 if (ret
== 0 && inode
->i_nlink
> 0) {
8945 trans
->block_rsv
= root
->orphan_block_rsv
;
8946 ret
= btrfs_orphan_del(trans
, inode
);
8952 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
8953 ret
= btrfs_update_inode(trans
, root
, inode
);
8957 ret
= btrfs_end_transaction(trans
, root
);
8958 btrfs_btree_balance_dirty(root
);
8962 btrfs_free_block_rsv(root
, rsv
);
8971 * create a new subvolume directory/inode (helper for the ioctl).
8973 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
8974 struct btrfs_root
*new_root
,
8975 struct btrfs_root
*parent_root
,
8978 struct inode
*inode
;
8982 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
8983 new_dirid
, new_dirid
,
8984 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
8987 return PTR_ERR(inode
);
8988 inode
->i_op
= &btrfs_dir_inode_operations
;
8989 inode
->i_fop
= &btrfs_dir_file_operations
;
8991 set_nlink(inode
, 1);
8992 btrfs_i_size_write(inode
, 0);
8993 unlock_new_inode(inode
);
8995 err
= btrfs_subvol_inherit_props(trans
, new_root
, parent_root
);
8997 btrfs_err(new_root
->fs_info
,
8998 "error inheriting subvolume %llu properties: %d",
8999 new_root
->root_key
.objectid
, err
);
9001 err
= btrfs_update_inode(trans
, new_root
, inode
);
9007 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
9009 struct btrfs_inode
*ei
;
9010 struct inode
*inode
;
9012 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
9019 ei
->last_sub_trans
= 0;
9020 ei
->logged_trans
= 0;
9021 ei
->delalloc_bytes
= 0;
9022 ei
->defrag_bytes
= 0;
9023 ei
->disk_i_size
= 0;
9026 ei
->index_cnt
= (u64
)-1;
9028 ei
->last_unlink_trans
= 0;
9029 ei
->last_log_commit
= 0;
9031 spin_lock_init(&ei
->lock
);
9032 ei
->outstanding_extents
= 0;
9033 ei
->reserved_extents
= 0;
9035 ei
->runtime_flags
= 0;
9036 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
9038 ei
->delayed_node
= NULL
;
9040 ei
->i_otime
.tv_sec
= 0;
9041 ei
->i_otime
.tv_nsec
= 0;
9043 inode
= &ei
->vfs_inode
;
9044 extent_map_tree_init(&ei
->extent_tree
);
9045 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
9046 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
9047 ei
->io_tree
.track_uptodate
= 1;
9048 ei
->io_failure_tree
.track_uptodate
= 1;
9049 atomic_set(&ei
->sync_writers
, 0);
9050 mutex_init(&ei
->log_mutex
);
9051 mutex_init(&ei
->delalloc_mutex
);
9052 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
9053 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
9054 RB_CLEAR_NODE(&ei
->rb_node
);
9059 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9060 void btrfs_test_destroy_inode(struct inode
*inode
)
9062 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
9063 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9067 static void btrfs_i_callback(struct rcu_head
*head
)
9069 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
9070 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9073 void btrfs_destroy_inode(struct inode
*inode
)
9075 struct btrfs_ordered_extent
*ordered
;
9076 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9078 WARN_ON(!hlist_empty(&inode
->i_dentry
));
9079 WARN_ON(inode
->i_data
.nrpages
);
9080 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
9081 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
9082 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
9083 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
9084 WARN_ON(BTRFS_I(inode
)->defrag_bytes
);
9087 * This can happen where we create an inode, but somebody else also
9088 * created the same inode and we need to destroy the one we already
9094 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
9095 &BTRFS_I(inode
)->runtime_flags
)) {
9096 btrfs_info(root
->fs_info
, "inode %llu still on the orphan list",
9098 atomic_dec(&root
->orphan_inodes
);
9102 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
9106 btrfs_err(root
->fs_info
, "found ordered extent %llu %llu on inode cleanup",
9107 ordered
->file_offset
, ordered
->len
);
9108 btrfs_remove_ordered_extent(inode
, ordered
);
9109 btrfs_put_ordered_extent(ordered
);
9110 btrfs_put_ordered_extent(ordered
);
9113 btrfs_qgroup_check_reserved_leak(inode
);
9114 inode_tree_del(inode
);
9115 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
9117 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
9120 int btrfs_drop_inode(struct inode
*inode
)
9122 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9127 /* the snap/subvol tree is on deleting */
9128 if (btrfs_root_refs(&root
->root_item
) == 0)
9131 return generic_drop_inode(inode
);
9134 static void init_once(void *foo
)
9136 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
9138 inode_init_once(&ei
->vfs_inode
);
9141 void btrfs_destroy_cachep(void)
9144 * Make sure all delayed rcu free inodes are flushed before we
9148 if (btrfs_inode_cachep
)
9149 kmem_cache_destroy(btrfs_inode_cachep
);
9150 if (btrfs_trans_handle_cachep
)
9151 kmem_cache_destroy(btrfs_trans_handle_cachep
);
9152 if (btrfs_transaction_cachep
)
9153 kmem_cache_destroy(btrfs_transaction_cachep
);
9154 if (btrfs_path_cachep
)
9155 kmem_cache_destroy(btrfs_path_cachep
);
9156 if (btrfs_free_space_cachep
)
9157 kmem_cache_destroy(btrfs_free_space_cachep
);
9158 if (btrfs_delalloc_work_cachep
)
9159 kmem_cache_destroy(btrfs_delalloc_work_cachep
);
9162 int btrfs_init_cachep(void)
9164 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
9165 sizeof(struct btrfs_inode
), 0,
9166 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
9167 if (!btrfs_inode_cachep
)
9170 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
9171 sizeof(struct btrfs_trans_handle
), 0,
9172 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9173 if (!btrfs_trans_handle_cachep
)
9176 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction",
9177 sizeof(struct btrfs_transaction
), 0,
9178 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9179 if (!btrfs_transaction_cachep
)
9182 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
9183 sizeof(struct btrfs_path
), 0,
9184 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9185 if (!btrfs_path_cachep
)
9188 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
9189 sizeof(struct btrfs_free_space
), 0,
9190 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9191 if (!btrfs_free_space_cachep
)
9194 btrfs_delalloc_work_cachep
= kmem_cache_create("btrfs_delalloc_work",
9195 sizeof(struct btrfs_delalloc_work
), 0,
9196 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
9198 if (!btrfs_delalloc_work_cachep
)
9203 btrfs_destroy_cachep();
9207 static int btrfs_getattr(struct vfsmount
*mnt
,
9208 struct dentry
*dentry
, struct kstat
*stat
)
9211 struct inode
*inode
= d_inode(dentry
);
9212 u32 blocksize
= inode
->i_sb
->s_blocksize
;
9214 generic_fillattr(inode
, stat
);
9215 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
9216 stat
->blksize
= PAGE_CACHE_SIZE
;
9218 spin_lock(&BTRFS_I(inode
)->lock
);
9219 delalloc_bytes
= BTRFS_I(inode
)->delalloc_bytes
;
9220 spin_unlock(&BTRFS_I(inode
)->lock
);
9221 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
9222 ALIGN(delalloc_bytes
, blocksize
)) >> 9;
9226 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
9227 struct inode
*new_dir
, struct dentry
*new_dentry
)
9229 struct btrfs_trans_handle
*trans
;
9230 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9231 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9232 struct inode
*new_inode
= d_inode(new_dentry
);
9233 struct inode
*old_inode
= d_inode(old_dentry
);
9234 struct timespec ctime
= CURRENT_TIME
;
9238 u64 old_ino
= btrfs_ino(old_inode
);
9240 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
9243 /* we only allow rename subvolume link between subvolumes */
9244 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9247 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
9248 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
9251 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
9252 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
9256 /* check for collisions, even if the name isn't there */
9257 ret
= btrfs_check_dir_item_collision(dest
, new_dir
->i_ino
,
9258 new_dentry
->d_name
.name
,
9259 new_dentry
->d_name
.len
);
9262 if (ret
== -EEXIST
) {
9264 * eexist without a new_inode */
9265 if (WARN_ON(!new_inode
)) {
9269 /* maybe -EOVERFLOW */
9276 * we're using rename to replace one file with another. Start IO on it
9277 * now so we don't add too much work to the end of the transaction
9279 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
)
9280 filemap_flush(old_inode
->i_mapping
);
9282 /* close the racy window with snapshot create/destroy ioctl */
9283 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9284 down_read(&root
->fs_info
->subvol_sem
);
9286 * We want to reserve the absolute worst case amount of items. So if
9287 * both inodes are subvols and we need to unlink them then that would
9288 * require 4 item modifications, but if they are both normal inodes it
9289 * would require 5 item modifications, so we'll assume their normal
9290 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9291 * should cover the worst case number of items we'll modify.
9293 trans
= btrfs_start_transaction(root
, 11);
9294 if (IS_ERR(trans
)) {
9295 ret
= PTR_ERR(trans
);
9300 btrfs_record_root_in_trans(trans
, dest
);
9302 ret
= btrfs_set_inode_index(new_dir
, &index
);
9306 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9307 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9308 /* force full log commit if subvolume involved. */
9309 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9311 ret
= btrfs_insert_inode_ref(trans
, dest
,
9312 new_dentry
->d_name
.name
,
9313 new_dentry
->d_name
.len
,
9315 btrfs_ino(new_dir
), index
);
9319 * this is an ugly little race, but the rename is required
9320 * to make sure that if we crash, the inode is either at the
9321 * old name or the new one. pinning the log transaction lets
9322 * us make sure we don't allow a log commit to come in after
9323 * we unlink the name but before we add the new name back in.
9325 btrfs_pin_log_trans(root
);
9328 inode_inc_iversion(old_dir
);
9329 inode_inc_iversion(new_dir
);
9330 inode_inc_iversion(old_inode
);
9331 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
9332 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
9333 old_inode
->i_ctime
= ctime
;
9335 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
9336 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
9338 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9339 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
9340 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
9341 old_dentry
->d_name
.name
,
9342 old_dentry
->d_name
.len
);
9344 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
9345 d_inode(old_dentry
),
9346 old_dentry
->d_name
.name
,
9347 old_dentry
->d_name
.len
);
9349 ret
= btrfs_update_inode(trans
, root
, old_inode
);
9352 btrfs_abort_transaction(trans
, root
, ret
);
9357 inode_inc_iversion(new_inode
);
9358 new_inode
->i_ctime
= CURRENT_TIME
;
9359 if (unlikely(btrfs_ino(new_inode
) ==
9360 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
9361 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
9362 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
9364 new_dentry
->d_name
.name
,
9365 new_dentry
->d_name
.len
);
9366 BUG_ON(new_inode
->i_nlink
== 0);
9368 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
9369 d_inode(new_dentry
),
9370 new_dentry
->d_name
.name
,
9371 new_dentry
->d_name
.len
);
9373 if (!ret
&& new_inode
->i_nlink
== 0)
9374 ret
= btrfs_orphan_add(trans
, d_inode(new_dentry
));
9376 btrfs_abort_transaction(trans
, root
, ret
);
9381 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
9382 new_dentry
->d_name
.name
,
9383 new_dentry
->d_name
.len
, 0, index
);
9385 btrfs_abort_transaction(trans
, root
, ret
);
9389 if (old_inode
->i_nlink
== 1)
9390 BTRFS_I(old_inode
)->dir_index
= index
;
9392 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
9393 struct dentry
*parent
= new_dentry
->d_parent
;
9394 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
9395 btrfs_end_log_trans(root
);
9398 btrfs_end_transaction(trans
, root
);
9400 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9401 up_read(&root
->fs_info
->subvol_sem
);
9406 static int btrfs_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
9407 struct inode
*new_dir
, struct dentry
*new_dentry
,
9410 if (flags
& ~RENAME_NOREPLACE
)
9413 return btrfs_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
9416 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
9418 struct btrfs_delalloc_work
*delalloc_work
;
9419 struct inode
*inode
;
9421 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
9423 inode
= delalloc_work
->inode
;
9424 if (delalloc_work
->wait
) {
9425 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
9427 filemap_flush(inode
->i_mapping
);
9428 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
9429 &BTRFS_I(inode
)->runtime_flags
))
9430 filemap_flush(inode
->i_mapping
);
9433 if (delalloc_work
->delay_iput
)
9434 btrfs_add_delayed_iput(inode
);
9437 complete(&delalloc_work
->completion
);
9440 struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
,
9441 int wait
, int delay_iput
)
9443 struct btrfs_delalloc_work
*work
;
9445 work
= kmem_cache_zalloc(btrfs_delalloc_work_cachep
, GFP_NOFS
);
9449 init_completion(&work
->completion
);
9450 INIT_LIST_HEAD(&work
->list
);
9451 work
->inode
= inode
;
9453 work
->delay_iput
= delay_iput
;
9454 WARN_ON_ONCE(!inode
);
9455 btrfs_init_work(&work
->work
, btrfs_flush_delalloc_helper
,
9456 btrfs_run_delalloc_work
, NULL
, NULL
);
9461 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work
*work
)
9463 wait_for_completion(&work
->completion
);
9464 kmem_cache_free(btrfs_delalloc_work_cachep
, work
);
9468 * some fairly slow code that needs optimization. This walks the list
9469 * of all the inodes with pending delalloc and forces them to disk.
9471 static int __start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
,
9474 struct btrfs_inode
*binode
;
9475 struct inode
*inode
;
9476 struct btrfs_delalloc_work
*work
, *next
;
9477 struct list_head works
;
9478 struct list_head splice
;
9481 INIT_LIST_HEAD(&works
);
9482 INIT_LIST_HEAD(&splice
);
9484 mutex_lock(&root
->delalloc_mutex
);
9485 spin_lock(&root
->delalloc_lock
);
9486 list_splice_init(&root
->delalloc_inodes
, &splice
);
9487 while (!list_empty(&splice
)) {
9488 binode
= list_entry(splice
.next
, struct btrfs_inode
,
9491 list_move_tail(&binode
->delalloc_inodes
,
9492 &root
->delalloc_inodes
);
9493 inode
= igrab(&binode
->vfs_inode
);
9495 cond_resched_lock(&root
->delalloc_lock
);
9498 spin_unlock(&root
->delalloc_lock
);
9500 work
= btrfs_alloc_delalloc_work(inode
, 0, delay_iput
);
9503 btrfs_add_delayed_iput(inode
);
9509 list_add_tail(&work
->list
, &works
);
9510 btrfs_queue_work(root
->fs_info
->flush_workers
,
9513 if (nr
!= -1 && ret
>= nr
)
9516 spin_lock(&root
->delalloc_lock
);
9518 spin_unlock(&root
->delalloc_lock
);
9521 list_for_each_entry_safe(work
, next
, &works
, list
) {
9522 list_del_init(&work
->list
);
9523 btrfs_wait_and_free_delalloc_work(work
);
9526 if (!list_empty_careful(&splice
)) {
9527 spin_lock(&root
->delalloc_lock
);
9528 list_splice_tail(&splice
, &root
->delalloc_inodes
);
9529 spin_unlock(&root
->delalloc_lock
);
9531 mutex_unlock(&root
->delalloc_mutex
);
9535 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
9539 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
9542 ret
= __start_delalloc_inodes(root
, delay_iput
, -1);
9546 * the filemap_flush will queue IO into the worker threads, but
9547 * we have to make sure the IO is actually started and that
9548 * ordered extents get created before we return
9550 atomic_inc(&root
->fs_info
->async_submit_draining
);
9551 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
9552 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
9553 wait_event(root
->fs_info
->async_submit_wait
,
9554 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
9555 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
9557 atomic_dec(&root
->fs_info
->async_submit_draining
);
9561 int btrfs_start_delalloc_roots(struct btrfs_fs_info
*fs_info
, int delay_iput
,
9564 struct btrfs_root
*root
;
9565 struct list_head splice
;
9568 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
9571 INIT_LIST_HEAD(&splice
);
9573 mutex_lock(&fs_info
->delalloc_root_mutex
);
9574 spin_lock(&fs_info
->delalloc_root_lock
);
9575 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
9576 while (!list_empty(&splice
) && nr
) {
9577 root
= list_first_entry(&splice
, struct btrfs_root
,
9579 root
= btrfs_grab_fs_root(root
);
9581 list_move_tail(&root
->delalloc_root
,
9582 &fs_info
->delalloc_roots
);
9583 spin_unlock(&fs_info
->delalloc_root_lock
);
9585 ret
= __start_delalloc_inodes(root
, delay_iput
, nr
);
9586 btrfs_put_fs_root(root
);
9594 spin_lock(&fs_info
->delalloc_root_lock
);
9596 spin_unlock(&fs_info
->delalloc_root_lock
);
9599 atomic_inc(&fs_info
->async_submit_draining
);
9600 while (atomic_read(&fs_info
->nr_async_submits
) ||
9601 atomic_read(&fs_info
->async_delalloc_pages
)) {
9602 wait_event(fs_info
->async_submit_wait
,
9603 (atomic_read(&fs_info
->nr_async_submits
) == 0 &&
9604 atomic_read(&fs_info
->async_delalloc_pages
) == 0));
9606 atomic_dec(&fs_info
->async_submit_draining
);
9608 if (!list_empty_careful(&splice
)) {
9609 spin_lock(&fs_info
->delalloc_root_lock
);
9610 list_splice_tail(&splice
, &fs_info
->delalloc_roots
);
9611 spin_unlock(&fs_info
->delalloc_root_lock
);
9613 mutex_unlock(&fs_info
->delalloc_root_mutex
);
9617 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
9618 const char *symname
)
9620 struct btrfs_trans_handle
*trans
;
9621 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
9622 struct btrfs_path
*path
;
9623 struct btrfs_key key
;
9624 struct inode
*inode
= NULL
;
9632 struct btrfs_file_extent_item
*ei
;
9633 struct extent_buffer
*leaf
;
9635 name_len
= strlen(symname
);
9636 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
9637 return -ENAMETOOLONG
;
9640 * 2 items for inode item and ref
9641 * 2 items for dir items
9642 * 1 item for xattr if selinux is on
9644 trans
= btrfs_start_transaction(root
, 5);
9646 return PTR_ERR(trans
);
9648 err
= btrfs_find_free_ino(root
, &objectid
);
9652 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
9653 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
9654 S_IFLNK
|S_IRWXUGO
, &index
);
9655 if (IS_ERR(inode
)) {
9656 err
= PTR_ERR(inode
);
9661 * If the active LSM wants to access the inode during
9662 * d_instantiate it needs these. Smack checks to see
9663 * if the filesystem supports xattrs by looking at the
9666 inode
->i_fop
= &btrfs_file_operations
;
9667 inode
->i_op
= &btrfs_file_inode_operations
;
9668 inode
->i_mapping
->a_ops
= &btrfs_aops
;
9669 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
9671 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
9673 goto out_unlock_inode
;
9675 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
9677 goto out_unlock_inode
;
9679 path
= btrfs_alloc_path();
9682 goto out_unlock_inode
;
9684 key
.objectid
= btrfs_ino(inode
);
9686 key
.type
= BTRFS_EXTENT_DATA_KEY
;
9687 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
9688 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
9691 btrfs_free_path(path
);
9692 goto out_unlock_inode
;
9694 leaf
= path
->nodes
[0];
9695 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
9696 struct btrfs_file_extent_item
);
9697 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
9698 btrfs_set_file_extent_type(leaf
, ei
,
9699 BTRFS_FILE_EXTENT_INLINE
);
9700 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
9701 btrfs_set_file_extent_compression(leaf
, ei
, 0);
9702 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
9703 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
9705 ptr
= btrfs_file_extent_inline_start(ei
);
9706 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
9707 btrfs_mark_buffer_dirty(leaf
);
9708 btrfs_free_path(path
);
9710 inode
->i_op
= &btrfs_symlink_inode_operations
;
9711 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
9712 inode_set_bytes(inode
, name_len
);
9713 btrfs_i_size_write(inode
, name_len
);
9714 err
= btrfs_update_inode(trans
, root
, inode
);
9717 goto out_unlock_inode
;
9720 unlock_new_inode(inode
);
9721 d_instantiate(dentry
, inode
);
9724 btrfs_end_transaction(trans
, root
);
9726 inode_dec_link_count(inode
);
9729 btrfs_btree_balance_dirty(root
);
9734 unlock_new_inode(inode
);
9738 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
9739 u64 start
, u64 num_bytes
, u64 min_size
,
9740 loff_t actual_len
, u64
*alloc_hint
,
9741 struct btrfs_trans_handle
*trans
)
9743 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
9744 struct extent_map
*em
;
9745 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9746 struct btrfs_key ins
;
9747 u64 cur_offset
= start
;
9750 u64 last_alloc
= (u64
)-1;
9752 bool own_trans
= true;
9756 while (num_bytes
> 0) {
9758 trans
= btrfs_start_transaction(root
, 3);
9759 if (IS_ERR(trans
)) {
9760 ret
= PTR_ERR(trans
);
9765 cur_bytes
= min(num_bytes
, 256ULL * 1024 * 1024);
9766 cur_bytes
= max(cur_bytes
, min_size
);
9768 * If we are severely fragmented we could end up with really
9769 * small allocations, so if the allocator is returning small
9770 * chunks lets make its job easier by only searching for those
9773 cur_bytes
= min(cur_bytes
, last_alloc
);
9774 ret
= btrfs_reserve_extent(root
, cur_bytes
, min_size
, 0,
9775 *alloc_hint
, &ins
, 1, 0);
9778 btrfs_end_transaction(trans
, root
);
9782 last_alloc
= ins
.offset
;
9783 ret
= insert_reserved_file_extent(trans
, inode
,
9784 cur_offset
, ins
.objectid
,
9785 ins
.offset
, ins
.offset
,
9786 ins
.offset
, 0, 0, 0,
9787 BTRFS_FILE_EXTENT_PREALLOC
);
9789 btrfs_free_reserved_extent(root
, ins
.objectid
,
9791 btrfs_abort_transaction(trans
, root
, ret
);
9793 btrfs_end_transaction(trans
, root
);
9797 btrfs_drop_extent_cache(inode
, cur_offset
,
9798 cur_offset
+ ins
.offset
-1, 0);
9800 em
= alloc_extent_map();
9802 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
9803 &BTRFS_I(inode
)->runtime_flags
);
9807 em
->start
= cur_offset
;
9808 em
->orig_start
= cur_offset
;
9809 em
->len
= ins
.offset
;
9810 em
->block_start
= ins
.objectid
;
9811 em
->block_len
= ins
.offset
;
9812 em
->orig_block_len
= ins
.offset
;
9813 em
->ram_bytes
= ins
.offset
;
9814 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
9815 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
9816 em
->generation
= trans
->transid
;
9819 write_lock(&em_tree
->lock
);
9820 ret
= add_extent_mapping(em_tree
, em
, 1);
9821 write_unlock(&em_tree
->lock
);
9824 btrfs_drop_extent_cache(inode
, cur_offset
,
9825 cur_offset
+ ins
.offset
- 1,
9828 free_extent_map(em
);
9830 num_bytes
-= ins
.offset
;
9831 cur_offset
+= ins
.offset
;
9832 *alloc_hint
= ins
.objectid
+ ins
.offset
;
9834 inode_inc_iversion(inode
);
9835 inode
->i_ctime
= CURRENT_TIME
;
9836 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
9837 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
9838 (actual_len
> inode
->i_size
) &&
9839 (cur_offset
> inode
->i_size
)) {
9840 if (cur_offset
> actual_len
)
9841 i_size
= actual_len
;
9843 i_size
= cur_offset
;
9844 i_size_write(inode
, i_size
);
9845 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
9848 ret
= btrfs_update_inode(trans
, root
, inode
);
9851 btrfs_abort_transaction(trans
, root
, ret
);
9853 btrfs_end_transaction(trans
, root
);
9858 btrfs_end_transaction(trans
, root
);
9863 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
9864 u64 start
, u64 num_bytes
, u64 min_size
,
9865 loff_t actual_len
, u64
*alloc_hint
)
9867 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
9868 min_size
, actual_len
, alloc_hint
,
9872 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
9873 struct btrfs_trans_handle
*trans
, int mode
,
9874 u64 start
, u64 num_bytes
, u64 min_size
,
9875 loff_t actual_len
, u64
*alloc_hint
)
9877 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
9878 min_size
, actual_len
, alloc_hint
, trans
);
9881 static int btrfs_set_page_dirty(struct page
*page
)
9883 return __set_page_dirty_nobuffers(page
);
9886 static int btrfs_permission(struct inode
*inode
, int mask
)
9888 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9889 umode_t mode
= inode
->i_mode
;
9891 if (mask
& MAY_WRITE
&&
9892 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
9893 if (btrfs_root_readonly(root
))
9895 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
9898 return generic_permission(inode
, mask
);
9901 static int btrfs_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
9903 struct btrfs_trans_handle
*trans
;
9904 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
9905 struct inode
*inode
= NULL
;
9911 * 5 units required for adding orphan entry
9913 trans
= btrfs_start_transaction(root
, 5);
9915 return PTR_ERR(trans
);
9917 ret
= btrfs_find_free_ino(root
, &objectid
);
9921 inode
= btrfs_new_inode(trans
, root
, dir
, NULL
, 0,
9922 btrfs_ino(dir
), objectid
, mode
, &index
);
9923 if (IS_ERR(inode
)) {
9924 ret
= PTR_ERR(inode
);
9929 inode
->i_fop
= &btrfs_file_operations
;
9930 inode
->i_op
= &btrfs_file_inode_operations
;
9932 inode
->i_mapping
->a_ops
= &btrfs_aops
;
9933 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
9935 ret
= btrfs_init_inode_security(trans
, inode
, dir
, NULL
);
9939 ret
= btrfs_update_inode(trans
, root
, inode
);
9942 ret
= btrfs_orphan_add(trans
, inode
);
9947 * We set number of links to 0 in btrfs_new_inode(), and here we set
9948 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9951 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9953 set_nlink(inode
, 1);
9954 unlock_new_inode(inode
);
9955 d_tmpfile(dentry
, inode
);
9956 mark_inode_dirty(inode
);
9959 btrfs_end_transaction(trans
, root
);
9962 btrfs_balance_delayed_items(root
);
9963 btrfs_btree_balance_dirty(root
);
9967 unlock_new_inode(inode
);
9972 /* Inspired by filemap_check_errors() */
9973 int btrfs_inode_check_errors(struct inode
*inode
)
9977 if (test_bit(AS_ENOSPC
, &inode
->i_mapping
->flags
) &&
9978 test_and_clear_bit(AS_ENOSPC
, &inode
->i_mapping
->flags
))
9980 if (test_bit(AS_EIO
, &inode
->i_mapping
->flags
) &&
9981 test_and_clear_bit(AS_EIO
, &inode
->i_mapping
->flags
))
9987 static const struct inode_operations btrfs_dir_inode_operations
= {
9988 .getattr
= btrfs_getattr
,
9989 .lookup
= btrfs_lookup
,
9990 .create
= btrfs_create
,
9991 .unlink
= btrfs_unlink
,
9993 .mkdir
= btrfs_mkdir
,
9994 .rmdir
= btrfs_rmdir
,
9995 .rename2
= btrfs_rename2
,
9996 .symlink
= btrfs_symlink
,
9997 .setattr
= btrfs_setattr
,
9998 .mknod
= btrfs_mknod
,
9999 .setxattr
= btrfs_setxattr
,
10000 .getxattr
= btrfs_getxattr
,
10001 .listxattr
= btrfs_listxattr
,
10002 .removexattr
= btrfs_removexattr
,
10003 .permission
= btrfs_permission
,
10004 .get_acl
= btrfs_get_acl
,
10005 .set_acl
= btrfs_set_acl
,
10006 .update_time
= btrfs_update_time
,
10007 .tmpfile
= btrfs_tmpfile
,
10009 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
10010 .lookup
= btrfs_lookup
,
10011 .permission
= btrfs_permission
,
10012 .get_acl
= btrfs_get_acl
,
10013 .set_acl
= btrfs_set_acl
,
10014 .update_time
= btrfs_update_time
,
10017 static const struct file_operations btrfs_dir_file_operations
= {
10018 .llseek
= generic_file_llseek
,
10019 .read
= generic_read_dir
,
10020 .iterate
= btrfs_real_readdir
,
10021 .unlocked_ioctl
= btrfs_ioctl
,
10022 #ifdef CONFIG_COMPAT
10023 .compat_ioctl
= btrfs_ioctl
,
10025 .release
= btrfs_release_file
,
10026 .fsync
= btrfs_sync_file
,
10029 static struct extent_io_ops btrfs_extent_io_ops
= {
10030 .fill_delalloc
= run_delalloc_range
,
10031 .submit_bio_hook
= btrfs_submit_bio_hook
,
10032 .merge_bio_hook
= btrfs_merge_bio_hook
,
10033 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
10034 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
10035 .writepage_start_hook
= btrfs_writepage_start_hook
,
10036 .set_bit_hook
= btrfs_set_bit_hook
,
10037 .clear_bit_hook
= btrfs_clear_bit_hook
,
10038 .merge_extent_hook
= btrfs_merge_extent_hook
,
10039 .split_extent_hook
= btrfs_split_extent_hook
,
10043 * btrfs doesn't support the bmap operation because swapfiles
10044 * use bmap to make a mapping of extents in the file. They assume
10045 * these extents won't change over the life of the file and they
10046 * use the bmap result to do IO directly to the drive.
10048 * the btrfs bmap call would return logical addresses that aren't
10049 * suitable for IO and they also will change frequently as COW
10050 * operations happen. So, swapfile + btrfs == corruption.
10052 * For now we're avoiding this by dropping bmap.
10054 static const struct address_space_operations btrfs_aops
= {
10055 .readpage
= btrfs_readpage
,
10056 .writepage
= btrfs_writepage
,
10057 .writepages
= btrfs_writepages
,
10058 .readpages
= btrfs_readpages
,
10059 .direct_IO
= btrfs_direct_IO
,
10060 .invalidatepage
= btrfs_invalidatepage
,
10061 .releasepage
= btrfs_releasepage
,
10062 .set_page_dirty
= btrfs_set_page_dirty
,
10063 .error_remove_page
= generic_error_remove_page
,
10066 static const struct address_space_operations btrfs_symlink_aops
= {
10067 .readpage
= btrfs_readpage
,
10068 .writepage
= btrfs_writepage
,
10069 .invalidatepage
= btrfs_invalidatepage
,
10070 .releasepage
= btrfs_releasepage
,
10073 static const struct inode_operations btrfs_file_inode_operations
= {
10074 .getattr
= btrfs_getattr
,
10075 .setattr
= btrfs_setattr
,
10076 .setxattr
= btrfs_setxattr
,
10077 .getxattr
= btrfs_getxattr
,
10078 .listxattr
= btrfs_listxattr
,
10079 .removexattr
= btrfs_removexattr
,
10080 .permission
= btrfs_permission
,
10081 .fiemap
= btrfs_fiemap
,
10082 .get_acl
= btrfs_get_acl
,
10083 .set_acl
= btrfs_set_acl
,
10084 .update_time
= btrfs_update_time
,
10086 static const struct inode_operations btrfs_special_inode_operations
= {
10087 .getattr
= btrfs_getattr
,
10088 .setattr
= btrfs_setattr
,
10089 .permission
= btrfs_permission
,
10090 .setxattr
= btrfs_setxattr
,
10091 .getxattr
= btrfs_getxattr
,
10092 .listxattr
= btrfs_listxattr
,
10093 .removexattr
= btrfs_removexattr
,
10094 .get_acl
= btrfs_get_acl
,
10095 .set_acl
= btrfs_set_acl
,
10096 .update_time
= btrfs_update_time
,
10098 static const struct inode_operations btrfs_symlink_inode_operations
= {
10099 .readlink
= generic_readlink
,
10100 .follow_link
= page_follow_link_light
,
10101 .put_link
= page_put_link
,
10102 .getattr
= btrfs_getattr
,
10103 .setattr
= btrfs_setattr
,
10104 .permission
= btrfs_permission
,
10105 .setxattr
= btrfs_setxattr
,
10106 .getxattr
= btrfs_getxattr
,
10107 .listxattr
= btrfs_listxattr
,
10108 .removexattr
= btrfs_removexattr
,
10109 .update_time
= btrfs_update_time
,
10112 const struct dentry_operations btrfs_dentry_operations
= {
10113 .d_delete
= btrfs_dentry_delete
,
10114 .d_release
= btrfs_dentry_release
,