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/compat.h>
34 #include <linux/bit_spinlock.h>
35 #include <linux/xattr.h>
36 #include <linux/posix_acl.h>
37 #include <linux/falloc.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/mount.h>
41 #include <linux/btrfs.h>
42 #include <linux/blkdev.h>
43 #include <linux/posix_acl_xattr.h>
44 #include <linux/uio.h>
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "print-tree.h"
50 #include "ordered-data.h"
54 #include "compression.h"
56 #include "free-space-cache.h"
57 #include "inode-map.h"
64 struct btrfs_iget_args
{
65 struct btrfs_key
*location
;
66 struct btrfs_root
*root
;
69 struct btrfs_dio_data
{
70 u64 outstanding_extents
;
72 u64 unsubmitted_oe_range_start
;
73 u64 unsubmitted_oe_range_end
;
77 static const struct inode_operations btrfs_dir_inode_operations
;
78 static const struct inode_operations btrfs_symlink_inode_operations
;
79 static const struct inode_operations btrfs_dir_ro_inode_operations
;
80 static const struct inode_operations btrfs_special_inode_operations
;
81 static const struct inode_operations btrfs_file_inode_operations
;
82 static const struct address_space_operations btrfs_aops
;
83 static const struct address_space_operations btrfs_symlink_aops
;
84 static const struct file_operations btrfs_dir_file_operations
;
85 static const struct extent_io_ops btrfs_extent_io_ops
;
87 static struct kmem_cache
*btrfs_inode_cachep
;
88 struct kmem_cache
*btrfs_trans_handle_cachep
;
89 struct kmem_cache
*btrfs_path_cachep
;
90 struct kmem_cache
*btrfs_free_space_cachep
;
93 static const unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
94 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
95 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
96 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
97 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
98 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
99 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
100 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
103 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
);
104 static int btrfs_truncate(struct inode
*inode
);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
);
106 static noinline
int cow_file_range(struct inode
*inode
,
107 struct page
*locked_page
,
108 u64 start
, u64 end
, u64 delalloc_end
,
109 int *page_started
, unsigned long *nr_written
,
110 int unlock
, struct btrfs_dedupe_hash
*hash
);
111 static struct extent_map
*create_io_em(struct inode
*inode
, u64 start
, u64 len
,
112 u64 orig_start
, u64 block_start
,
113 u64 block_len
, u64 orig_block_len
,
114 u64 ram_bytes
, int compress_type
,
117 static void __endio_write_update_ordered(struct inode
*inode
,
118 const u64 offset
, const u64 bytes
,
119 const bool uptodate
);
122 * Cleanup all submitted ordered extents in specified range to handle errors
123 * from the fill_dellaloc() callback.
125 * NOTE: caller must ensure that when an error happens, it can not call
126 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
127 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
128 * to be released, which we want to happen only when finishing the ordered
129 * extent (btrfs_finish_ordered_io()). Also note that the caller of the
130 * fill_delalloc() callback already does proper cleanup for the first page of
131 * the range, that is, it invokes the callback writepage_end_io_hook() for the
132 * range of the first page.
134 static inline void btrfs_cleanup_ordered_extents(struct inode
*inode
,
138 unsigned long index
= offset
>> PAGE_SHIFT
;
139 unsigned long end_index
= (offset
+ bytes
- 1) >> PAGE_SHIFT
;
142 while (index
<= end_index
) {
143 page
= find_get_page(inode
->i_mapping
, index
);
147 ClearPagePrivate2(page
);
150 return __endio_write_update_ordered(inode
, offset
+ PAGE_SIZE
,
151 bytes
- PAGE_SIZE
, false);
154 static int btrfs_dirty_inode(struct inode
*inode
);
156 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
157 void btrfs_test_inode_set_ops(struct inode
*inode
)
159 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
163 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
164 struct inode
*inode
, struct inode
*dir
,
165 const struct qstr
*qstr
)
169 err
= btrfs_init_acl(trans
, inode
, dir
);
171 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
176 * this does all the hard work for inserting an inline extent into
177 * the btree. The caller should have done a btrfs_drop_extents so that
178 * no overlapping inline items exist in the btree
180 static int insert_inline_extent(struct btrfs_trans_handle
*trans
,
181 struct btrfs_path
*path
, int extent_inserted
,
182 struct btrfs_root
*root
, struct inode
*inode
,
183 u64 start
, size_t size
, size_t compressed_size
,
185 struct page
**compressed_pages
)
187 struct extent_buffer
*leaf
;
188 struct page
*page
= NULL
;
191 struct btrfs_file_extent_item
*ei
;
193 size_t cur_size
= size
;
194 unsigned long offset
;
196 if (compressed_size
&& compressed_pages
)
197 cur_size
= compressed_size
;
199 inode_add_bytes(inode
, size
);
201 if (!extent_inserted
) {
202 struct btrfs_key key
;
205 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
207 key
.type
= BTRFS_EXTENT_DATA_KEY
;
209 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
210 path
->leave_spinning
= 1;
211 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
216 leaf
= path
->nodes
[0];
217 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
218 struct btrfs_file_extent_item
);
219 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
220 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
221 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
222 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
223 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
224 ptr
= btrfs_file_extent_inline_start(ei
);
226 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
229 while (compressed_size
> 0) {
230 cpage
= compressed_pages
[i
];
231 cur_size
= min_t(unsigned long, compressed_size
,
234 kaddr
= kmap_atomic(cpage
);
235 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
236 kunmap_atomic(kaddr
);
240 compressed_size
-= cur_size
;
242 btrfs_set_file_extent_compression(leaf
, ei
,
245 page
= find_get_page(inode
->i_mapping
,
246 start
>> PAGE_SHIFT
);
247 btrfs_set_file_extent_compression(leaf
, ei
, 0);
248 kaddr
= kmap_atomic(page
);
249 offset
= start
& (PAGE_SIZE
- 1);
250 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
251 kunmap_atomic(kaddr
);
254 btrfs_mark_buffer_dirty(leaf
);
255 btrfs_release_path(path
);
258 * we're an inline extent, so nobody can
259 * extend the file past i_size without locking
260 * a page we already have locked.
262 * We must do any isize and inode updates
263 * before we unlock the pages. Otherwise we
264 * could end up racing with unlink.
266 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
267 ret
= btrfs_update_inode(trans
, root
, inode
);
275 * conditionally insert an inline extent into the file. This
276 * does the checks required to make sure the data is small enough
277 * to fit as an inline extent.
279 static noinline
int cow_file_range_inline(struct btrfs_root
*root
,
280 struct inode
*inode
, u64 start
,
281 u64 end
, size_t compressed_size
,
283 struct page
**compressed_pages
)
285 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
286 struct btrfs_trans_handle
*trans
;
287 u64 isize
= i_size_read(inode
);
288 u64 actual_end
= min(end
+ 1, isize
);
289 u64 inline_len
= actual_end
- start
;
290 u64 aligned_end
= ALIGN(end
, fs_info
->sectorsize
);
291 u64 data_len
= inline_len
;
293 struct btrfs_path
*path
;
294 int extent_inserted
= 0;
295 u32 extent_item_size
;
298 data_len
= compressed_size
;
301 actual_end
> fs_info
->sectorsize
||
302 data_len
> BTRFS_MAX_INLINE_DATA_SIZE(fs_info
) ||
304 (actual_end
& (fs_info
->sectorsize
- 1)) == 0) ||
306 data_len
> fs_info
->max_inline
) {
310 path
= btrfs_alloc_path();
314 trans
= btrfs_join_transaction(root
);
316 btrfs_free_path(path
);
317 return PTR_ERR(trans
);
319 trans
->block_rsv
= &fs_info
->delalloc_block_rsv
;
321 if (compressed_size
&& compressed_pages
)
322 extent_item_size
= btrfs_file_extent_calc_inline_size(
325 extent_item_size
= btrfs_file_extent_calc_inline_size(
328 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
,
329 start
, aligned_end
, NULL
,
330 1, 1, extent_item_size
, &extent_inserted
);
332 btrfs_abort_transaction(trans
, ret
);
336 if (isize
> actual_end
)
337 inline_len
= min_t(u64
, isize
, actual_end
);
338 ret
= insert_inline_extent(trans
, path
, extent_inserted
,
340 inline_len
, compressed_size
,
341 compress_type
, compressed_pages
);
342 if (ret
&& ret
!= -ENOSPC
) {
343 btrfs_abort_transaction(trans
, ret
);
345 } else if (ret
== -ENOSPC
) {
350 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
351 btrfs_delalloc_release_metadata(BTRFS_I(inode
), end
+ 1 - start
);
352 btrfs_drop_extent_cache(BTRFS_I(inode
), start
, aligned_end
- 1, 0);
355 * Don't forget to free the reserved space, as for inlined extent
356 * it won't count as data extent, free them directly here.
357 * And at reserve time, it's always aligned to page size, so
358 * just free one page here.
360 btrfs_qgroup_free_data(inode
, NULL
, 0, PAGE_SIZE
);
361 btrfs_free_path(path
);
362 btrfs_end_transaction(trans
);
366 struct async_extent
{
371 unsigned long nr_pages
;
373 struct list_head list
;
378 struct btrfs_root
*root
;
379 struct page
*locked_page
;
382 struct list_head extents
;
383 struct btrfs_work work
;
386 static noinline
int add_async_extent(struct async_cow
*cow
,
387 u64 start
, u64 ram_size
,
390 unsigned long nr_pages
,
393 struct async_extent
*async_extent
;
395 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
396 BUG_ON(!async_extent
); /* -ENOMEM */
397 async_extent
->start
= start
;
398 async_extent
->ram_size
= ram_size
;
399 async_extent
->compressed_size
= compressed_size
;
400 async_extent
->pages
= pages
;
401 async_extent
->nr_pages
= nr_pages
;
402 async_extent
->compress_type
= compress_type
;
403 list_add_tail(&async_extent
->list
, &cow
->extents
);
407 static inline int inode_need_compress(struct inode
*inode
, u64 start
, u64 end
)
409 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
412 if (btrfs_test_opt(fs_info
, FORCE_COMPRESS
))
415 if (BTRFS_I(inode
)->defrag_compress
)
417 /* bad compression ratios */
418 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
420 if (btrfs_test_opt(fs_info
, COMPRESS
) ||
421 BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
||
422 BTRFS_I(inode
)->prop_compress
)
423 return btrfs_compress_heuristic(inode
, start
, end
);
427 static inline void inode_should_defrag(struct btrfs_inode
*inode
,
428 u64 start
, u64 end
, u64 num_bytes
, u64 small_write
)
430 /* If this is a small write inside eof, kick off a defrag */
431 if (num_bytes
< small_write
&&
432 (start
> 0 || end
+ 1 < inode
->disk_i_size
))
433 btrfs_add_inode_defrag(NULL
, inode
);
437 * we create compressed extents in two phases. The first
438 * phase compresses a range of pages that have already been
439 * locked (both pages and state bits are locked).
441 * This is done inside an ordered work queue, and the compression
442 * is spread across many cpus. The actual IO submission is step
443 * two, and the ordered work queue takes care of making sure that
444 * happens in the same order things were put onto the queue by
445 * writepages and friends.
447 * If this code finds it can't get good compression, it puts an
448 * entry onto the work queue to write the uncompressed bytes. This
449 * makes sure that both compressed inodes and uncompressed inodes
450 * are written in the same order that the flusher thread sent them
453 static noinline
void compress_file_range(struct inode
*inode
,
454 struct page
*locked_page
,
456 struct async_cow
*async_cow
,
459 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
460 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
462 u64 blocksize
= fs_info
->sectorsize
;
464 u64 isize
= i_size_read(inode
);
466 struct page
**pages
= NULL
;
467 unsigned long nr_pages
;
468 unsigned long total_compressed
= 0;
469 unsigned long total_in
= 0;
472 int compress_type
= fs_info
->compress_type
;
475 inode_should_defrag(BTRFS_I(inode
), start
, end
, end
- start
+ 1,
478 actual_end
= min_t(u64
, isize
, end
+ 1);
481 nr_pages
= (end
>> PAGE_SHIFT
) - (start
>> PAGE_SHIFT
) + 1;
482 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED
% PAGE_SIZE
) != 0);
483 nr_pages
= min_t(unsigned long, nr_pages
,
484 BTRFS_MAX_COMPRESSED
/ PAGE_SIZE
);
487 * we don't want to send crud past the end of i_size through
488 * compression, that's just a waste of CPU time. So, if the
489 * end of the file is before the start of our current
490 * requested range of bytes, we bail out to the uncompressed
491 * cleanup code that can deal with all of this.
493 * It isn't really the fastest way to fix things, but this is a
494 * very uncommon corner.
496 if (actual_end
<= start
)
497 goto cleanup_and_bail_uncompressed
;
499 total_compressed
= actual_end
- start
;
502 * skip compression for a small file range(<=blocksize) that
503 * isn't an inline extent, since it doesn't save disk space at all.
505 if (total_compressed
<= blocksize
&&
506 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
507 goto cleanup_and_bail_uncompressed
;
509 total_compressed
= min_t(unsigned long, total_compressed
,
510 BTRFS_MAX_UNCOMPRESSED
);
511 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
512 num_bytes
= max(blocksize
, num_bytes
);
517 * we do compression for mount -o compress and when the
518 * inode has not been flagged as nocompress. This flag can
519 * change at any time if we discover bad compression ratios.
521 if (inode_need_compress(inode
, start
, end
)) {
523 pages
= kcalloc(nr_pages
, sizeof(struct page
*), GFP_NOFS
);
525 /* just bail out to the uncompressed code */
529 if (BTRFS_I(inode
)->defrag_compress
)
530 compress_type
= BTRFS_I(inode
)->defrag_compress
;
531 else if (BTRFS_I(inode
)->prop_compress
)
532 compress_type
= BTRFS_I(inode
)->prop_compress
;
535 * we need to call clear_page_dirty_for_io on each
536 * page in the range. Otherwise applications with the file
537 * mmap'd can wander in and change the page contents while
538 * we are compressing them.
540 * If the compression fails for any reason, we set the pages
541 * dirty again later on.
543 extent_range_clear_dirty_for_io(inode
, start
, end
);
545 ret
= btrfs_compress_pages(compress_type
,
546 inode
->i_mapping
, start
,
553 unsigned long offset
= total_compressed
&
555 struct page
*page
= pages
[nr_pages
- 1];
558 /* zero the tail end of the last page, we might be
559 * sending it down to disk
562 kaddr
= kmap_atomic(page
);
563 memset(kaddr
+ offset
, 0,
565 kunmap_atomic(kaddr
);
572 /* lets try to make an inline extent */
573 if (ret
|| total_in
< (actual_end
- start
)) {
574 /* we didn't compress the entire range, try
575 * to make an uncompressed inline extent.
577 ret
= cow_file_range_inline(root
, inode
, start
, end
,
578 0, BTRFS_COMPRESS_NONE
, NULL
);
580 /* try making a compressed inline extent */
581 ret
= cow_file_range_inline(root
, inode
, start
, end
,
583 compress_type
, pages
);
586 unsigned long clear_flags
= EXTENT_DELALLOC
|
587 EXTENT_DELALLOC_NEW
| EXTENT_DEFRAG
;
588 unsigned long page_error_op
;
590 clear_flags
|= (ret
< 0) ? EXTENT_DO_ACCOUNTING
: 0;
591 page_error_op
= ret
< 0 ? PAGE_SET_ERROR
: 0;
594 * inline extent creation worked or returned error,
595 * we don't need to create any more async work items.
596 * Unlock and free up our temp pages.
598 extent_clear_unlock_delalloc(inode
, start
, end
, end
,
606 btrfs_free_reserved_data_space_noquota(inode
,
615 * we aren't doing an inline extent round the compressed size
616 * up to a block size boundary so the allocator does sane
619 total_compressed
= ALIGN(total_compressed
, blocksize
);
622 * one last check to make sure the compression is really a
623 * win, compare the page count read with the blocks on disk,
624 * compression must free at least one sector size
626 total_in
= ALIGN(total_in
, PAGE_SIZE
);
627 if (total_compressed
+ blocksize
<= total_in
) {
628 num_bytes
= total_in
;
632 * The async work queues will take care of doing actual
633 * allocation on disk for these compressed pages, and
634 * will submit them to the elevator.
636 add_async_extent(async_cow
, start
, num_bytes
,
637 total_compressed
, pages
, nr_pages
,
640 if (start
+ num_bytes
< end
) {
651 * the compression code ran but failed to make things smaller,
652 * free any pages it allocated and our page pointer array
654 for (i
= 0; i
< nr_pages
; i
++) {
655 WARN_ON(pages
[i
]->mapping
);
660 total_compressed
= 0;
663 /* flag the file so we don't compress in the future */
664 if (!btrfs_test_opt(fs_info
, FORCE_COMPRESS
) &&
665 !(BTRFS_I(inode
)->prop_compress
)) {
666 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
669 cleanup_and_bail_uncompressed
:
671 * No compression, but we still need to write the pages in the file
672 * we've been given so far. redirty the locked page if it corresponds
673 * to our extent and set things up for the async work queue to run
674 * cow_file_range to do the normal delalloc dance.
676 if (page_offset(locked_page
) >= start
&&
677 page_offset(locked_page
) <= end
)
678 __set_page_dirty_nobuffers(locked_page
);
679 /* unlocked later on in the async handlers */
682 extent_range_redirty_for_io(inode
, start
, end
);
683 add_async_extent(async_cow
, start
, end
- start
+ 1, 0, NULL
, 0,
684 BTRFS_COMPRESS_NONE
);
690 for (i
= 0; i
< nr_pages
; i
++) {
691 WARN_ON(pages
[i
]->mapping
);
697 static void free_async_extent_pages(struct async_extent
*async_extent
)
701 if (!async_extent
->pages
)
704 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
705 WARN_ON(async_extent
->pages
[i
]->mapping
);
706 put_page(async_extent
->pages
[i
]);
708 kfree(async_extent
->pages
);
709 async_extent
->nr_pages
= 0;
710 async_extent
->pages
= NULL
;
714 * phase two of compressed writeback. This is the ordered portion
715 * of the code, which only gets called in the order the work was
716 * queued. We walk all the async extents created by compress_file_range
717 * and send them down to the disk.
719 static noinline
void submit_compressed_extents(struct inode
*inode
,
720 struct async_cow
*async_cow
)
722 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
723 struct async_extent
*async_extent
;
725 struct btrfs_key ins
;
726 struct extent_map
*em
;
727 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
728 struct extent_io_tree
*io_tree
;
732 while (!list_empty(&async_cow
->extents
)) {
733 async_extent
= list_entry(async_cow
->extents
.next
,
734 struct async_extent
, list
);
735 list_del(&async_extent
->list
);
737 io_tree
= &BTRFS_I(inode
)->io_tree
;
740 /* did the compression code fall back to uncompressed IO? */
741 if (!async_extent
->pages
) {
742 int page_started
= 0;
743 unsigned long nr_written
= 0;
745 lock_extent(io_tree
, async_extent
->start
,
746 async_extent
->start
+
747 async_extent
->ram_size
- 1);
749 /* allocate blocks */
750 ret
= cow_file_range(inode
, async_cow
->locked_page
,
752 async_extent
->start
+
753 async_extent
->ram_size
- 1,
754 async_extent
->start
+
755 async_extent
->ram_size
- 1,
756 &page_started
, &nr_written
, 0,
762 * if page_started, cow_file_range inserted an
763 * inline extent and took care of all the unlocking
764 * and IO for us. Otherwise, we need to submit
765 * all those pages down to the drive.
767 if (!page_started
&& !ret
)
768 extent_write_locked_range(io_tree
,
769 inode
, async_extent
->start
,
770 async_extent
->start
+
771 async_extent
->ram_size
- 1,
775 unlock_page(async_cow
->locked_page
);
781 lock_extent(io_tree
, async_extent
->start
,
782 async_extent
->start
+ async_extent
->ram_size
- 1);
784 ret
= btrfs_reserve_extent(root
, async_extent
->ram_size
,
785 async_extent
->compressed_size
,
786 async_extent
->compressed_size
,
787 0, alloc_hint
, &ins
, 1, 1);
789 free_async_extent_pages(async_extent
);
791 if (ret
== -ENOSPC
) {
792 unlock_extent(io_tree
, async_extent
->start
,
793 async_extent
->start
+
794 async_extent
->ram_size
- 1);
797 * we need to redirty the pages if we decide to
798 * fallback to uncompressed IO, otherwise we
799 * will not submit these pages down to lower
802 extent_range_redirty_for_io(inode
,
804 async_extent
->start
+
805 async_extent
->ram_size
- 1);
812 * here we're doing allocation and writeback of the
815 em
= create_io_em(inode
, async_extent
->start
,
816 async_extent
->ram_size
, /* len */
817 async_extent
->start
, /* orig_start */
818 ins
.objectid
, /* block_start */
819 ins
.offset
, /* block_len */
820 ins
.offset
, /* orig_block_len */
821 async_extent
->ram_size
, /* ram_bytes */
822 async_extent
->compress_type
,
823 BTRFS_ORDERED_COMPRESSED
);
825 /* ret value is not necessary due to void function */
826 goto out_free_reserve
;
829 ret
= btrfs_add_ordered_extent_compress(inode
,
832 async_extent
->ram_size
,
834 BTRFS_ORDERED_COMPRESSED
,
835 async_extent
->compress_type
);
837 btrfs_drop_extent_cache(BTRFS_I(inode
),
839 async_extent
->start
+
840 async_extent
->ram_size
- 1, 0);
841 goto out_free_reserve
;
843 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
846 * clear dirty, set writeback and unlock the pages.
848 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
849 async_extent
->start
+
850 async_extent
->ram_size
- 1,
851 async_extent
->start
+
852 async_extent
->ram_size
- 1,
853 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
,
854 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
856 if (btrfs_submit_compressed_write(inode
,
858 async_extent
->ram_size
,
860 ins
.offset
, async_extent
->pages
,
861 async_extent
->nr_pages
)) {
862 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
863 struct page
*p
= async_extent
->pages
[0];
864 const u64 start
= async_extent
->start
;
865 const u64 end
= start
+ async_extent
->ram_size
- 1;
867 p
->mapping
= inode
->i_mapping
;
868 tree
->ops
->writepage_end_io_hook(p
, start
, end
,
871 extent_clear_unlock_delalloc(inode
, start
, end
, end
,
875 free_async_extent_pages(async_extent
);
877 alloc_hint
= ins
.objectid
+ ins
.offset
;
883 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
884 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 1);
886 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
887 async_extent
->start
+
888 async_extent
->ram_size
- 1,
889 async_extent
->start
+
890 async_extent
->ram_size
- 1,
891 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
|
892 EXTENT_DELALLOC_NEW
|
893 EXTENT_DEFRAG
| EXTENT_DO_ACCOUNTING
,
894 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
895 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
|
897 free_async_extent_pages(async_extent
);
902 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
905 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
906 struct extent_map
*em
;
909 read_lock(&em_tree
->lock
);
910 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
913 * if block start isn't an actual block number then find the
914 * first block in this inode and use that as a hint. If that
915 * block is also bogus then just don't worry about it.
917 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
919 em
= search_extent_mapping(em_tree
, 0, 0);
920 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
921 alloc_hint
= em
->block_start
;
925 alloc_hint
= em
->block_start
;
929 read_unlock(&em_tree
->lock
);
935 * when extent_io.c finds a delayed allocation range in the file,
936 * the call backs end up in this code. The basic idea is to
937 * allocate extents on disk for the range, and create ordered data structs
938 * in ram to track those extents.
940 * locked_page is the page that writepage had locked already. We use
941 * it to make sure we don't do extra locks or unlocks.
943 * *page_started is set to one if we unlock locked_page and do everything
944 * required to start IO on it. It may be clean and already done with
947 static noinline
int cow_file_range(struct inode
*inode
,
948 struct page
*locked_page
,
949 u64 start
, u64 end
, u64 delalloc_end
,
950 int *page_started
, unsigned long *nr_written
,
951 int unlock
, struct btrfs_dedupe_hash
*hash
)
953 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
954 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
957 unsigned long ram_size
;
959 u64 cur_alloc_size
= 0;
960 u64 blocksize
= fs_info
->sectorsize
;
961 struct btrfs_key ins
;
962 struct extent_map
*em
;
964 unsigned long page_ops
;
965 bool extent_reserved
= false;
968 if (btrfs_is_free_space_inode(BTRFS_I(inode
))) {
974 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
975 num_bytes
= max(blocksize
, num_bytes
);
976 disk_num_bytes
= num_bytes
;
978 inode_should_defrag(BTRFS_I(inode
), start
, end
, num_bytes
, SZ_64K
);
981 /* lets try to make an inline extent */
982 ret
= cow_file_range_inline(root
, inode
, start
, end
, 0,
983 BTRFS_COMPRESS_NONE
, NULL
);
985 extent_clear_unlock_delalloc(inode
, start
, end
,
987 EXTENT_LOCKED
| EXTENT_DELALLOC
|
988 EXTENT_DELALLOC_NEW
|
989 EXTENT_DEFRAG
, PAGE_UNLOCK
|
990 PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
992 btrfs_free_reserved_data_space_noquota(inode
, start
,
994 *nr_written
= *nr_written
+
995 (end
- start
+ PAGE_SIZE
) / PAGE_SIZE
;
998 } else if (ret
< 0) {
1003 BUG_ON(disk_num_bytes
>
1004 btrfs_super_total_bytes(fs_info
->super_copy
));
1006 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
1007 btrfs_drop_extent_cache(BTRFS_I(inode
), start
,
1008 start
+ num_bytes
- 1, 0);
1010 while (disk_num_bytes
> 0) {
1011 cur_alloc_size
= disk_num_bytes
;
1012 ret
= btrfs_reserve_extent(root
, cur_alloc_size
, cur_alloc_size
,
1013 fs_info
->sectorsize
, 0, alloc_hint
,
1017 cur_alloc_size
= ins
.offset
;
1018 extent_reserved
= true;
1020 ram_size
= ins
.offset
;
1021 em
= create_io_em(inode
, start
, ins
.offset
, /* len */
1022 start
, /* orig_start */
1023 ins
.objectid
, /* block_start */
1024 ins
.offset
, /* block_len */
1025 ins
.offset
, /* orig_block_len */
1026 ram_size
, /* ram_bytes */
1027 BTRFS_COMPRESS_NONE
, /* compress_type */
1028 BTRFS_ORDERED_REGULAR
/* type */);
1031 free_extent_map(em
);
1033 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
1034 ram_size
, cur_alloc_size
, 0);
1036 goto out_drop_extent_cache
;
1038 if (root
->root_key
.objectid
==
1039 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1040 ret
= btrfs_reloc_clone_csums(inode
, start
,
1043 * Only drop cache here, and process as normal.
1045 * We must not allow extent_clear_unlock_delalloc()
1046 * at out_unlock label to free meta of this ordered
1047 * extent, as its meta should be freed by
1048 * btrfs_finish_ordered_io().
1050 * So we must continue until @start is increased to
1051 * skip current ordered extent.
1054 btrfs_drop_extent_cache(BTRFS_I(inode
), start
,
1055 start
+ ram_size
- 1, 0);
1058 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
1060 /* we're not doing compressed IO, don't unlock the first
1061 * page (which the caller expects to stay locked), don't
1062 * clear any dirty bits and don't set any writeback bits
1064 * Do set the Private2 bit so we know this page was properly
1065 * setup for writepage
1067 page_ops
= unlock
? PAGE_UNLOCK
: 0;
1068 page_ops
|= PAGE_SET_PRIVATE2
;
1070 extent_clear_unlock_delalloc(inode
, start
,
1071 start
+ ram_size
- 1,
1072 delalloc_end
, locked_page
,
1073 EXTENT_LOCKED
| EXTENT_DELALLOC
,
1075 if (disk_num_bytes
< cur_alloc_size
)
1078 disk_num_bytes
-= cur_alloc_size
;
1079 num_bytes
-= cur_alloc_size
;
1080 alloc_hint
= ins
.objectid
+ ins
.offset
;
1081 start
+= cur_alloc_size
;
1082 extent_reserved
= false;
1085 * btrfs_reloc_clone_csums() error, since start is increased
1086 * extent_clear_unlock_delalloc() at out_unlock label won't
1087 * free metadata of current ordered extent, we're OK to exit.
1095 out_drop_extent_cache
:
1096 btrfs_drop_extent_cache(BTRFS_I(inode
), start
, start
+ ram_size
- 1, 0);
1098 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
1099 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 1);
1101 clear_bits
= EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DELALLOC_NEW
|
1102 EXTENT_DEFRAG
| EXTENT_CLEAR_META_RESV
;
1103 page_ops
= PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
1106 * If we reserved an extent for our delalloc range (or a subrange) and
1107 * failed to create the respective ordered extent, then it means that
1108 * when we reserved the extent we decremented the extent's size from
1109 * the data space_info's bytes_may_use counter and incremented the
1110 * space_info's bytes_reserved counter by the same amount. We must make
1111 * sure extent_clear_unlock_delalloc() does not try to decrement again
1112 * the data space_info's bytes_may_use counter, therefore we do not pass
1113 * it the flag EXTENT_CLEAR_DATA_RESV.
1115 if (extent_reserved
) {
1116 extent_clear_unlock_delalloc(inode
, start
,
1117 start
+ cur_alloc_size
,
1118 start
+ cur_alloc_size
,
1122 start
+= cur_alloc_size
;
1126 extent_clear_unlock_delalloc(inode
, start
, end
, delalloc_end
,
1128 clear_bits
| EXTENT_CLEAR_DATA_RESV
,
1134 * work queue call back to started compression on a file and pages
1136 static noinline
void async_cow_start(struct btrfs_work
*work
)
1138 struct async_cow
*async_cow
;
1140 async_cow
= container_of(work
, struct async_cow
, work
);
1142 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
1143 async_cow
->start
, async_cow
->end
, async_cow
,
1145 if (num_added
== 0) {
1146 btrfs_add_delayed_iput(async_cow
->inode
);
1147 async_cow
->inode
= NULL
;
1152 * work queue call back to submit previously compressed pages
1154 static noinline
void async_cow_submit(struct btrfs_work
*work
)
1156 struct btrfs_fs_info
*fs_info
;
1157 struct async_cow
*async_cow
;
1158 struct btrfs_root
*root
;
1159 unsigned long nr_pages
;
1161 async_cow
= container_of(work
, struct async_cow
, work
);
1163 root
= async_cow
->root
;
1164 fs_info
= root
->fs_info
;
1165 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_SIZE
) >>
1169 * atomic_sub_return implies a barrier for waitqueue_active
1171 if (atomic_sub_return(nr_pages
, &fs_info
->async_delalloc_pages
) <
1173 waitqueue_active(&fs_info
->async_submit_wait
))
1174 wake_up(&fs_info
->async_submit_wait
);
1176 if (async_cow
->inode
)
1177 submit_compressed_extents(async_cow
->inode
, async_cow
);
1180 static noinline
void async_cow_free(struct btrfs_work
*work
)
1182 struct async_cow
*async_cow
;
1183 async_cow
= container_of(work
, struct async_cow
, work
);
1184 if (async_cow
->inode
)
1185 btrfs_add_delayed_iput(async_cow
->inode
);
1189 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1190 u64 start
, u64 end
, int *page_started
,
1191 unsigned long *nr_written
)
1193 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1194 struct async_cow
*async_cow
;
1195 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1196 unsigned long nr_pages
;
1199 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1200 1, 0, NULL
, GFP_NOFS
);
1201 while (start
< end
) {
1202 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1203 BUG_ON(!async_cow
); /* -ENOMEM */
1204 async_cow
->inode
= igrab(inode
);
1205 async_cow
->root
= root
;
1206 async_cow
->locked_page
= locked_page
;
1207 async_cow
->start
= start
;
1209 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
&&
1210 !btrfs_test_opt(fs_info
, FORCE_COMPRESS
))
1213 cur_end
= min(end
, start
+ SZ_512K
- 1);
1215 async_cow
->end
= cur_end
;
1216 INIT_LIST_HEAD(&async_cow
->extents
);
1218 btrfs_init_work(&async_cow
->work
,
1219 btrfs_delalloc_helper
,
1220 async_cow_start
, async_cow_submit
,
1223 nr_pages
= (cur_end
- start
+ PAGE_SIZE
) >>
1225 atomic_add(nr_pages
, &fs_info
->async_delalloc_pages
);
1227 btrfs_queue_work(fs_info
->delalloc_workers
, &async_cow
->work
);
1229 while (atomic_read(&fs_info
->async_submit_draining
) &&
1230 atomic_read(&fs_info
->async_delalloc_pages
)) {
1231 wait_event(fs_info
->async_submit_wait
,
1232 (atomic_read(&fs_info
->async_delalloc_pages
) ==
1236 *nr_written
+= nr_pages
;
1237 start
= cur_end
+ 1;
1243 static noinline
int csum_exist_in_range(struct btrfs_fs_info
*fs_info
,
1244 u64 bytenr
, u64 num_bytes
)
1247 struct btrfs_ordered_sum
*sums
;
1250 ret
= btrfs_lookup_csums_range(fs_info
->csum_root
, bytenr
,
1251 bytenr
+ num_bytes
- 1, &list
, 0);
1252 if (ret
== 0 && list_empty(&list
))
1255 while (!list_empty(&list
)) {
1256 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1257 list_del(&sums
->list
);
1264 * when nowcow writeback call back. This checks for snapshots or COW copies
1265 * of the extents that exist in the file, and COWs the file as required.
1267 * If no cow copies or snapshots exist, we write directly to the existing
1270 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1271 struct page
*locked_page
,
1272 u64 start
, u64 end
, int *page_started
, int force
,
1273 unsigned long *nr_written
)
1275 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1276 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1277 struct extent_buffer
*leaf
;
1278 struct btrfs_path
*path
;
1279 struct btrfs_file_extent_item
*fi
;
1280 struct btrfs_key found_key
;
1281 struct extent_map
*em
;
1296 u64 ino
= btrfs_ino(BTRFS_I(inode
));
1298 path
= btrfs_alloc_path();
1300 extent_clear_unlock_delalloc(inode
, start
, end
, end
,
1302 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1303 EXTENT_DO_ACCOUNTING
|
1304 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1306 PAGE_SET_WRITEBACK
|
1307 PAGE_END_WRITEBACK
);
1311 nolock
= btrfs_is_free_space_inode(BTRFS_I(inode
));
1313 cow_start
= (u64
)-1;
1316 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, ino
,
1320 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1321 leaf
= path
->nodes
[0];
1322 btrfs_item_key_to_cpu(leaf
, &found_key
,
1323 path
->slots
[0] - 1);
1324 if (found_key
.objectid
== ino
&&
1325 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1330 leaf
= path
->nodes
[0];
1331 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1332 ret
= btrfs_next_leaf(root
, path
);
1337 leaf
= path
->nodes
[0];
1343 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1345 if (found_key
.objectid
> ino
)
1347 if (WARN_ON_ONCE(found_key
.objectid
< ino
) ||
1348 found_key
.type
< BTRFS_EXTENT_DATA_KEY
) {
1352 if (found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1353 found_key
.offset
> end
)
1356 if (found_key
.offset
> cur_offset
) {
1357 extent_end
= found_key
.offset
;
1362 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1363 struct btrfs_file_extent_item
);
1364 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1366 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
1367 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1368 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1369 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1370 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1371 extent_end
= found_key
.offset
+
1372 btrfs_file_extent_num_bytes(leaf
, fi
);
1374 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1375 if (extent_end
<= start
) {
1379 if (disk_bytenr
== 0)
1381 if (btrfs_file_extent_compression(leaf
, fi
) ||
1382 btrfs_file_extent_encryption(leaf
, fi
) ||
1383 btrfs_file_extent_other_encoding(leaf
, fi
))
1385 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1387 if (btrfs_extent_readonly(fs_info
, disk_bytenr
))
1389 if (btrfs_cross_ref_exist(root
, ino
,
1391 extent_offset
, disk_bytenr
))
1393 disk_bytenr
+= extent_offset
;
1394 disk_bytenr
+= cur_offset
- found_key
.offset
;
1395 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1397 * if there are pending snapshots for this root,
1398 * we fall into common COW way.
1401 err
= btrfs_start_write_no_snapshotting(root
);
1406 * force cow if csum exists in the range.
1407 * this ensure that csum for a given extent are
1408 * either valid or do not exist.
1410 if (csum_exist_in_range(fs_info
, disk_bytenr
,
1413 btrfs_end_write_no_snapshotting(root
);
1416 if (!btrfs_inc_nocow_writers(fs_info
, disk_bytenr
)) {
1418 btrfs_end_write_no_snapshotting(root
);
1422 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1423 extent_end
= found_key
.offset
+
1424 btrfs_file_extent_inline_len(leaf
,
1425 path
->slots
[0], fi
);
1426 extent_end
= ALIGN(extent_end
,
1427 fs_info
->sectorsize
);
1432 if (extent_end
<= start
) {
1434 if (!nolock
&& nocow
)
1435 btrfs_end_write_no_snapshotting(root
);
1437 btrfs_dec_nocow_writers(fs_info
, disk_bytenr
);
1441 if (cow_start
== (u64
)-1)
1442 cow_start
= cur_offset
;
1443 cur_offset
= extent_end
;
1444 if (cur_offset
> end
)
1450 btrfs_release_path(path
);
1451 if (cow_start
!= (u64
)-1) {
1452 ret
= cow_file_range(inode
, locked_page
,
1453 cow_start
, found_key
.offset
- 1,
1454 end
, page_started
, nr_written
, 1,
1457 if (!nolock
&& nocow
)
1458 btrfs_end_write_no_snapshotting(root
);
1460 btrfs_dec_nocow_writers(fs_info
,
1464 cow_start
= (u64
)-1;
1467 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1468 u64 orig_start
= found_key
.offset
- extent_offset
;
1470 em
= create_io_em(inode
, cur_offset
, num_bytes
,
1472 disk_bytenr
, /* block_start */
1473 num_bytes
, /* block_len */
1474 disk_num_bytes
, /* orig_block_len */
1475 ram_bytes
, BTRFS_COMPRESS_NONE
,
1476 BTRFS_ORDERED_PREALLOC
);
1478 if (!nolock
&& nocow
)
1479 btrfs_end_write_no_snapshotting(root
);
1481 btrfs_dec_nocow_writers(fs_info
,
1486 free_extent_map(em
);
1489 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1490 type
= BTRFS_ORDERED_PREALLOC
;
1492 type
= BTRFS_ORDERED_NOCOW
;
1495 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1496 num_bytes
, num_bytes
, type
);
1498 btrfs_dec_nocow_writers(fs_info
, disk_bytenr
);
1499 BUG_ON(ret
); /* -ENOMEM */
1501 if (root
->root_key
.objectid
==
1502 BTRFS_DATA_RELOC_TREE_OBJECTID
)
1504 * Error handled later, as we must prevent
1505 * extent_clear_unlock_delalloc() in error handler
1506 * from freeing metadata of created ordered extent.
1508 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1511 extent_clear_unlock_delalloc(inode
, cur_offset
,
1512 cur_offset
+ num_bytes
- 1, end
,
1513 locked_page
, EXTENT_LOCKED
|
1515 EXTENT_CLEAR_DATA_RESV
,
1516 PAGE_UNLOCK
| PAGE_SET_PRIVATE2
);
1518 if (!nolock
&& nocow
)
1519 btrfs_end_write_no_snapshotting(root
);
1520 cur_offset
= extent_end
;
1523 * btrfs_reloc_clone_csums() error, now we're OK to call error
1524 * handler, as metadata for created ordered extent will only
1525 * be freed by btrfs_finish_ordered_io().
1529 if (cur_offset
> end
)
1532 btrfs_release_path(path
);
1534 if (cur_offset
<= end
&& cow_start
== (u64
)-1) {
1535 cow_start
= cur_offset
;
1539 if (cow_start
!= (u64
)-1) {
1540 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
, end
,
1541 page_started
, nr_written
, 1, NULL
);
1547 if (ret
&& cur_offset
< end
)
1548 extent_clear_unlock_delalloc(inode
, cur_offset
, end
, end
,
1549 locked_page
, EXTENT_LOCKED
|
1550 EXTENT_DELALLOC
| EXTENT_DEFRAG
|
1551 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
1553 PAGE_SET_WRITEBACK
|
1554 PAGE_END_WRITEBACK
);
1555 btrfs_free_path(path
);
1559 static inline int need_force_cow(struct inode
*inode
, u64 start
, u64 end
)
1562 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
1563 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
))
1567 * @defrag_bytes is a hint value, no spinlock held here,
1568 * if is not zero, it means the file is defragging.
1569 * Force cow if given extent needs to be defragged.
1571 if (BTRFS_I(inode
)->defrag_bytes
&&
1572 test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1573 EXTENT_DEFRAG
, 0, NULL
))
1580 * extent_io.c call back to do delayed allocation processing
1582 static int run_delalloc_range(void *private_data
, struct page
*locked_page
,
1583 u64 start
, u64 end
, int *page_started
,
1584 unsigned long *nr_written
)
1586 struct inode
*inode
= private_data
;
1588 int force_cow
= need_force_cow(inode
, start
, end
);
1590 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
&& !force_cow
) {
1591 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1592 page_started
, 1, nr_written
);
1593 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
&& !force_cow
) {
1594 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1595 page_started
, 0, nr_written
);
1596 } else if (!inode_need_compress(inode
, start
, end
)) {
1597 ret
= cow_file_range(inode
, locked_page
, start
, end
, end
,
1598 page_started
, nr_written
, 1, NULL
);
1600 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1601 &BTRFS_I(inode
)->runtime_flags
);
1602 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1603 page_started
, nr_written
);
1606 btrfs_cleanup_ordered_extents(inode
, start
, end
- start
+ 1);
1610 static void btrfs_split_extent_hook(void *private_data
,
1611 struct extent_state
*orig
, u64 split
)
1613 struct inode
*inode
= private_data
;
1616 /* not delalloc, ignore it */
1617 if (!(orig
->state
& EXTENT_DELALLOC
))
1620 size
= orig
->end
- orig
->start
+ 1;
1621 if (size
> BTRFS_MAX_EXTENT_SIZE
) {
1626 * See the explanation in btrfs_merge_extent_hook, the same
1627 * applies here, just in reverse.
1629 new_size
= orig
->end
- split
+ 1;
1630 num_extents
= count_max_extents(new_size
);
1631 new_size
= split
- orig
->start
;
1632 num_extents
+= count_max_extents(new_size
);
1633 if (count_max_extents(size
) >= num_extents
)
1637 spin_lock(&BTRFS_I(inode
)->lock
);
1638 BTRFS_I(inode
)->outstanding_extents
++;
1639 spin_unlock(&BTRFS_I(inode
)->lock
);
1643 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1644 * extents so we can keep track of new extents that are just merged onto old
1645 * extents, such as when we are doing sequential writes, so we can properly
1646 * account for the metadata space we'll need.
1648 static void btrfs_merge_extent_hook(void *private_data
,
1649 struct extent_state
*new,
1650 struct extent_state
*other
)
1652 struct inode
*inode
= private_data
;
1653 u64 new_size
, old_size
;
1656 /* not delalloc, ignore it */
1657 if (!(other
->state
& EXTENT_DELALLOC
))
1660 if (new->start
> other
->start
)
1661 new_size
= new->end
- other
->start
+ 1;
1663 new_size
= other
->end
- new->start
+ 1;
1665 /* we're not bigger than the max, unreserve the space and go */
1666 if (new_size
<= BTRFS_MAX_EXTENT_SIZE
) {
1667 spin_lock(&BTRFS_I(inode
)->lock
);
1668 BTRFS_I(inode
)->outstanding_extents
--;
1669 spin_unlock(&BTRFS_I(inode
)->lock
);
1674 * We have to add up either side to figure out how many extents were
1675 * accounted for before we merged into one big extent. If the number of
1676 * extents we accounted for is <= the amount we need for the new range
1677 * then we can return, otherwise drop. Think of it like this
1681 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1682 * need 2 outstanding extents, on one side we have 1 and the other side
1683 * we have 1 so they are == and we can return. But in this case
1685 * [MAX_SIZE+4k][MAX_SIZE+4k]
1687 * Each range on their own accounts for 2 extents, but merged together
1688 * they are only 3 extents worth of accounting, so we need to drop in
1691 old_size
= other
->end
- other
->start
+ 1;
1692 num_extents
= count_max_extents(old_size
);
1693 old_size
= new->end
- new->start
+ 1;
1694 num_extents
+= count_max_extents(old_size
);
1695 if (count_max_extents(new_size
) >= num_extents
)
1698 spin_lock(&BTRFS_I(inode
)->lock
);
1699 BTRFS_I(inode
)->outstanding_extents
--;
1700 spin_unlock(&BTRFS_I(inode
)->lock
);
1703 static void btrfs_add_delalloc_inodes(struct btrfs_root
*root
,
1704 struct inode
*inode
)
1706 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1708 spin_lock(&root
->delalloc_lock
);
1709 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1710 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1711 &root
->delalloc_inodes
);
1712 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1713 &BTRFS_I(inode
)->runtime_flags
);
1714 root
->nr_delalloc_inodes
++;
1715 if (root
->nr_delalloc_inodes
== 1) {
1716 spin_lock(&fs_info
->delalloc_root_lock
);
1717 BUG_ON(!list_empty(&root
->delalloc_root
));
1718 list_add_tail(&root
->delalloc_root
,
1719 &fs_info
->delalloc_roots
);
1720 spin_unlock(&fs_info
->delalloc_root_lock
);
1723 spin_unlock(&root
->delalloc_lock
);
1726 static void btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1727 struct btrfs_inode
*inode
)
1729 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
1731 spin_lock(&root
->delalloc_lock
);
1732 if (!list_empty(&inode
->delalloc_inodes
)) {
1733 list_del_init(&inode
->delalloc_inodes
);
1734 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1735 &inode
->runtime_flags
);
1736 root
->nr_delalloc_inodes
--;
1737 if (!root
->nr_delalloc_inodes
) {
1738 spin_lock(&fs_info
->delalloc_root_lock
);
1739 BUG_ON(list_empty(&root
->delalloc_root
));
1740 list_del_init(&root
->delalloc_root
);
1741 spin_unlock(&fs_info
->delalloc_root_lock
);
1744 spin_unlock(&root
->delalloc_lock
);
1748 * extent_io.c set_bit_hook, used to track delayed allocation
1749 * bytes in this file, and to maintain the list of inodes that
1750 * have pending delalloc work to be done.
1752 static void btrfs_set_bit_hook(void *private_data
,
1753 struct extent_state
*state
, unsigned *bits
)
1755 struct inode
*inode
= private_data
;
1757 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1759 if ((*bits
& EXTENT_DEFRAG
) && !(*bits
& EXTENT_DELALLOC
))
1762 * set_bit and clear bit hooks normally require _irqsave/restore
1763 * but in this case, we are only testing for the DELALLOC
1764 * bit, which is only set or cleared with irqs on
1766 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1767 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1768 u64 len
= state
->end
+ 1 - state
->start
;
1769 bool do_list
= !btrfs_is_free_space_inode(BTRFS_I(inode
));
1771 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1772 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1774 spin_lock(&BTRFS_I(inode
)->lock
);
1775 BTRFS_I(inode
)->outstanding_extents
++;
1776 spin_unlock(&BTRFS_I(inode
)->lock
);
1779 /* For sanity tests */
1780 if (btrfs_is_testing(fs_info
))
1783 percpu_counter_add_batch(&fs_info
->delalloc_bytes
, len
,
1784 fs_info
->delalloc_batch
);
1785 spin_lock(&BTRFS_I(inode
)->lock
);
1786 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1787 if (*bits
& EXTENT_DEFRAG
)
1788 BTRFS_I(inode
)->defrag_bytes
+= len
;
1789 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1790 &BTRFS_I(inode
)->runtime_flags
))
1791 btrfs_add_delalloc_inodes(root
, inode
);
1792 spin_unlock(&BTRFS_I(inode
)->lock
);
1795 if (!(state
->state
& EXTENT_DELALLOC_NEW
) &&
1796 (*bits
& EXTENT_DELALLOC_NEW
)) {
1797 spin_lock(&BTRFS_I(inode
)->lock
);
1798 BTRFS_I(inode
)->new_delalloc_bytes
+= state
->end
+ 1 -
1800 spin_unlock(&BTRFS_I(inode
)->lock
);
1805 * extent_io.c clear_bit_hook, see set_bit_hook for why
1807 static void btrfs_clear_bit_hook(void *private_data
,
1808 struct extent_state
*state
,
1811 struct btrfs_inode
*inode
= BTRFS_I((struct inode
*)private_data
);
1812 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
1813 u64 len
= state
->end
+ 1 - state
->start
;
1814 u32 num_extents
= count_max_extents(len
);
1816 if ((state
->state
& EXTENT_DEFRAG
) && (*bits
& EXTENT_DEFRAG
)) {
1817 spin_lock(&inode
->lock
);
1818 inode
->defrag_bytes
-= len
;
1819 spin_unlock(&inode
->lock
);
1823 * set_bit and clear bit hooks normally require _irqsave/restore
1824 * but in this case, we are only testing for the DELALLOC
1825 * bit, which is only set or cleared with irqs on
1827 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1828 struct btrfs_root
*root
= inode
->root
;
1829 bool do_list
= !btrfs_is_free_space_inode(inode
);
1831 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1832 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1833 } else if (!(*bits
& EXTENT_CLEAR_META_RESV
)) {
1834 spin_lock(&inode
->lock
);
1835 inode
->outstanding_extents
-= num_extents
;
1836 spin_unlock(&inode
->lock
);
1840 * We don't reserve metadata space for space cache inodes so we
1841 * don't need to call dellalloc_release_metadata if there is an
1844 if (*bits
& EXTENT_CLEAR_META_RESV
&&
1845 root
!= fs_info
->tree_root
)
1846 btrfs_delalloc_release_metadata(inode
, len
);
1848 /* For sanity tests. */
1849 if (btrfs_is_testing(fs_info
))
1852 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1853 do_list
&& !(state
->state
& EXTENT_NORESERVE
) &&
1854 (*bits
& EXTENT_CLEAR_DATA_RESV
))
1855 btrfs_free_reserved_data_space_noquota(
1859 percpu_counter_add_batch(&fs_info
->delalloc_bytes
, -len
,
1860 fs_info
->delalloc_batch
);
1861 spin_lock(&inode
->lock
);
1862 inode
->delalloc_bytes
-= len
;
1863 if (do_list
&& inode
->delalloc_bytes
== 0 &&
1864 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1865 &inode
->runtime_flags
))
1866 btrfs_del_delalloc_inode(root
, inode
);
1867 spin_unlock(&inode
->lock
);
1870 if ((state
->state
& EXTENT_DELALLOC_NEW
) &&
1871 (*bits
& EXTENT_DELALLOC_NEW
)) {
1872 spin_lock(&inode
->lock
);
1873 ASSERT(inode
->new_delalloc_bytes
>= len
);
1874 inode
->new_delalloc_bytes
-= len
;
1875 spin_unlock(&inode
->lock
);
1880 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1881 * we don't create bios that span stripes or chunks
1883 * return 1 if page cannot be merged to bio
1884 * return 0 if page can be merged to bio
1885 * return error otherwise
1887 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1888 size_t size
, struct bio
*bio
,
1889 unsigned long bio_flags
)
1891 struct inode
*inode
= page
->mapping
->host
;
1892 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1893 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
1898 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1901 length
= bio
->bi_iter
.bi_size
;
1902 map_length
= length
;
1903 ret
= btrfs_map_block(fs_info
, btrfs_op(bio
), logical
, &map_length
,
1907 if (map_length
< length
+ size
)
1913 * in order to insert checksums into the metadata in large chunks,
1914 * we wait until bio submission time. All the pages in the bio are
1915 * checksummed and sums are attached onto the ordered extent record.
1917 * At IO completion time the cums attached on the ordered extent record
1918 * are inserted into the btree
1920 static blk_status_t
__btrfs_submit_bio_start(void *private_data
, struct bio
*bio
,
1921 int mirror_num
, unsigned long bio_flags
,
1924 struct inode
*inode
= private_data
;
1925 blk_status_t ret
= 0;
1927 ret
= btrfs_csum_one_bio(inode
, bio
, 0, 0);
1928 BUG_ON(ret
); /* -ENOMEM */
1933 * in order to insert checksums into the metadata in large chunks,
1934 * we wait until bio submission time. All the pages in the bio are
1935 * checksummed and sums are attached onto the ordered extent record.
1937 * At IO completion time the cums attached on the ordered extent record
1938 * are inserted into the btree
1940 static blk_status_t
__btrfs_submit_bio_done(void *private_data
, struct bio
*bio
,
1941 int mirror_num
, unsigned long bio_flags
,
1944 struct inode
*inode
= private_data
;
1945 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1948 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 1);
1950 bio
->bi_status
= ret
;
1957 * extent_io.c submission hook. This does the right thing for csum calculation
1958 * on write, or reading the csums from the tree before a read
1960 static blk_status_t
btrfs_submit_bio_hook(void *private_data
, struct bio
*bio
,
1961 int mirror_num
, unsigned long bio_flags
,
1964 struct inode
*inode
= private_data
;
1965 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1966 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1967 enum btrfs_wq_endio_type metadata
= BTRFS_WQ_ENDIO_DATA
;
1968 blk_status_t ret
= 0;
1970 int async
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
1972 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1974 if (btrfs_is_free_space_inode(BTRFS_I(inode
)))
1975 metadata
= BTRFS_WQ_ENDIO_FREE_SPACE
;
1977 if (bio_op(bio
) != REQ_OP_WRITE
) {
1978 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, metadata
);
1982 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1983 ret
= btrfs_submit_compressed_read(inode
, bio
,
1987 } else if (!skip_sum
) {
1988 ret
= btrfs_lookup_bio_sums(inode
, bio
, NULL
);
1993 } else if (async
&& !skip_sum
) {
1994 /* csum items have already been cloned */
1995 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1997 /* we're doing a write, do the async checksumming */
1998 ret
= btrfs_wq_submit_bio(fs_info
, bio
, mirror_num
, bio_flags
,
2000 __btrfs_submit_bio_start
,
2001 __btrfs_submit_bio_done
);
2003 } else if (!skip_sum
) {
2004 ret
= btrfs_csum_one_bio(inode
, bio
, 0, 0);
2010 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
2014 bio
->bi_status
= ret
;
2021 * given a list of ordered sums record them in the inode. This happens
2022 * at IO completion time based on sums calculated at bio submission time.
2024 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
2025 struct inode
*inode
, struct list_head
*list
)
2027 struct btrfs_ordered_sum
*sum
;
2029 list_for_each_entry(sum
, list
, list
) {
2030 trans
->adding_csums
= 1;
2031 btrfs_csum_file_blocks(trans
,
2032 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
2033 trans
->adding_csums
= 0;
2038 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
2039 struct extent_state
**cached_state
, int dedupe
)
2041 WARN_ON((end
& (PAGE_SIZE
- 1)) == 0);
2042 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
2046 /* see btrfs_writepage_start_hook for details on why this is required */
2047 struct btrfs_writepage_fixup
{
2049 struct btrfs_work work
;
2052 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
2054 struct btrfs_writepage_fixup
*fixup
;
2055 struct btrfs_ordered_extent
*ordered
;
2056 struct extent_state
*cached_state
= NULL
;
2057 struct extent_changeset
*data_reserved
= NULL
;
2059 struct inode
*inode
;
2064 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
2068 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
2069 ClearPageChecked(page
);
2073 inode
= page
->mapping
->host
;
2074 page_start
= page_offset(page
);
2075 page_end
= page_offset(page
) + PAGE_SIZE
- 1;
2077 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2080 /* already ordered? We're done */
2081 if (PagePrivate2(page
))
2084 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), page_start
,
2087 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
2088 page_end
, &cached_state
, GFP_NOFS
);
2090 btrfs_start_ordered_extent(inode
, ordered
, 1);
2091 btrfs_put_ordered_extent(ordered
);
2095 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
, page_start
,
2098 mapping_set_error(page
->mapping
, ret
);
2099 end_extent_writepage(page
, ret
, page_start
, page_end
);
2100 ClearPageChecked(page
);
2104 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
,
2106 ClearPageChecked(page
);
2107 set_page_dirty(page
);
2109 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2110 &cached_state
, GFP_NOFS
);
2115 extent_changeset_free(data_reserved
);
2119 * There are a few paths in the higher layers of the kernel that directly
2120 * set the page dirty bit without asking the filesystem if it is a
2121 * good idea. This causes problems because we want to make sure COW
2122 * properly happens and the data=ordered rules are followed.
2124 * In our case any range that doesn't have the ORDERED bit set
2125 * hasn't been properly setup for IO. We kick off an async process
2126 * to fix it up. The async helper will wait for ordered extents, set
2127 * the delalloc bit and make it safe to write the page.
2129 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
2131 struct inode
*inode
= page
->mapping
->host
;
2132 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2133 struct btrfs_writepage_fixup
*fixup
;
2135 /* this page is properly in the ordered list */
2136 if (TestClearPagePrivate2(page
))
2139 if (PageChecked(page
))
2142 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
2146 SetPageChecked(page
);
2148 btrfs_init_work(&fixup
->work
, btrfs_fixup_helper
,
2149 btrfs_writepage_fixup_worker
, NULL
, NULL
);
2151 btrfs_queue_work(fs_info
->fixup_workers
, &fixup
->work
);
2155 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
2156 struct inode
*inode
, u64 file_pos
,
2157 u64 disk_bytenr
, u64 disk_num_bytes
,
2158 u64 num_bytes
, u64 ram_bytes
,
2159 u8 compression
, u8 encryption
,
2160 u16 other_encoding
, int extent_type
)
2162 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2163 struct btrfs_file_extent_item
*fi
;
2164 struct btrfs_path
*path
;
2165 struct extent_buffer
*leaf
;
2166 struct btrfs_key ins
;
2168 int extent_inserted
= 0;
2171 path
= btrfs_alloc_path();
2176 * we may be replacing one extent in the tree with another.
2177 * The new extent is pinned in the extent map, and we don't want
2178 * to drop it from the cache until it is completely in the btree.
2180 * So, tell btrfs_drop_extents to leave this extent in the cache.
2181 * the caller is expected to unpin it and allow it to be merged
2184 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, file_pos
,
2185 file_pos
+ num_bytes
, NULL
, 0,
2186 1, sizeof(*fi
), &extent_inserted
);
2190 if (!extent_inserted
) {
2191 ins
.objectid
= btrfs_ino(BTRFS_I(inode
));
2192 ins
.offset
= file_pos
;
2193 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
2195 path
->leave_spinning
= 1;
2196 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
2201 leaf
= path
->nodes
[0];
2202 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2203 struct btrfs_file_extent_item
);
2204 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
2205 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
2206 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
2207 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
2208 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2209 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2210 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
2211 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
2212 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
2213 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
2215 btrfs_mark_buffer_dirty(leaf
);
2216 btrfs_release_path(path
);
2218 inode_add_bytes(inode
, num_bytes
);
2220 ins
.objectid
= disk_bytenr
;
2221 ins
.offset
= disk_num_bytes
;
2222 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2225 * Release the reserved range from inode dirty range map, as it is
2226 * already moved into delayed_ref_head
2228 ret
= btrfs_qgroup_release_data(inode
, file_pos
, ram_bytes
);
2232 ret
= btrfs_alloc_reserved_file_extent(trans
, root
->root_key
.objectid
,
2233 btrfs_ino(BTRFS_I(inode
)), file_pos
, qg_released
, &ins
);
2235 btrfs_free_path(path
);
2240 /* snapshot-aware defrag */
2241 struct sa_defrag_extent_backref
{
2242 struct rb_node node
;
2243 struct old_sa_defrag_extent
*old
;
2252 struct old_sa_defrag_extent
{
2253 struct list_head list
;
2254 struct new_sa_defrag_extent
*new;
2263 struct new_sa_defrag_extent
{
2264 struct rb_root root
;
2265 struct list_head head
;
2266 struct btrfs_path
*path
;
2267 struct inode
*inode
;
2275 static int backref_comp(struct sa_defrag_extent_backref
*b1
,
2276 struct sa_defrag_extent_backref
*b2
)
2278 if (b1
->root_id
< b2
->root_id
)
2280 else if (b1
->root_id
> b2
->root_id
)
2283 if (b1
->inum
< b2
->inum
)
2285 else if (b1
->inum
> b2
->inum
)
2288 if (b1
->file_pos
< b2
->file_pos
)
2290 else if (b1
->file_pos
> b2
->file_pos
)
2294 * [------------------------------] ===> (a range of space)
2295 * |<--->| |<---->| =============> (fs/file tree A)
2296 * |<---------------------------->| ===> (fs/file tree B)
2298 * A range of space can refer to two file extents in one tree while
2299 * refer to only one file extent in another tree.
2301 * So we may process a disk offset more than one time(two extents in A)
2302 * and locate at the same extent(one extent in B), then insert two same
2303 * backrefs(both refer to the extent in B).
2308 static void backref_insert(struct rb_root
*root
,
2309 struct sa_defrag_extent_backref
*backref
)
2311 struct rb_node
**p
= &root
->rb_node
;
2312 struct rb_node
*parent
= NULL
;
2313 struct sa_defrag_extent_backref
*entry
;
2318 entry
= rb_entry(parent
, struct sa_defrag_extent_backref
, node
);
2320 ret
= backref_comp(backref
, entry
);
2324 p
= &(*p
)->rb_right
;
2327 rb_link_node(&backref
->node
, parent
, p
);
2328 rb_insert_color(&backref
->node
, root
);
2332 * Note the backref might has changed, and in this case we just return 0.
2334 static noinline
int record_one_backref(u64 inum
, u64 offset
, u64 root_id
,
2337 struct btrfs_file_extent_item
*extent
;
2338 struct old_sa_defrag_extent
*old
= ctx
;
2339 struct new_sa_defrag_extent
*new = old
->new;
2340 struct btrfs_path
*path
= new->path
;
2341 struct btrfs_key key
;
2342 struct btrfs_root
*root
;
2343 struct sa_defrag_extent_backref
*backref
;
2344 struct extent_buffer
*leaf
;
2345 struct inode
*inode
= new->inode
;
2346 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2352 if (BTRFS_I(inode
)->root
->root_key
.objectid
== root_id
&&
2353 inum
== btrfs_ino(BTRFS_I(inode
)))
2356 key
.objectid
= root_id
;
2357 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2358 key
.offset
= (u64
)-1;
2360 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2362 if (PTR_ERR(root
) == -ENOENT
)
2365 btrfs_debug(fs_info
, "inum=%llu, offset=%llu, root_id=%llu",
2366 inum
, offset
, root_id
);
2367 return PTR_ERR(root
);
2370 key
.objectid
= inum
;
2371 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2372 if (offset
> (u64
)-1 << 32)
2375 key
.offset
= offset
;
2377 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2378 if (WARN_ON(ret
< 0))
2385 leaf
= path
->nodes
[0];
2386 slot
= path
->slots
[0];
2388 if (slot
>= btrfs_header_nritems(leaf
)) {
2389 ret
= btrfs_next_leaf(root
, path
);
2392 } else if (ret
> 0) {
2401 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2403 if (key
.objectid
> inum
)
2406 if (key
.objectid
< inum
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2409 extent
= btrfs_item_ptr(leaf
, slot
,
2410 struct btrfs_file_extent_item
);
2412 if (btrfs_file_extent_disk_bytenr(leaf
, extent
) != old
->bytenr
)
2416 * 'offset' refers to the exact key.offset,
2417 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2418 * (key.offset - extent_offset).
2420 if (key
.offset
!= offset
)
2423 extent_offset
= btrfs_file_extent_offset(leaf
, extent
);
2424 num_bytes
= btrfs_file_extent_num_bytes(leaf
, extent
);
2426 if (extent_offset
>= old
->extent_offset
+ old
->offset
+
2427 old
->len
|| extent_offset
+ num_bytes
<=
2428 old
->extent_offset
+ old
->offset
)
2433 backref
= kmalloc(sizeof(*backref
), GFP_NOFS
);
2439 backref
->root_id
= root_id
;
2440 backref
->inum
= inum
;
2441 backref
->file_pos
= offset
;
2442 backref
->num_bytes
= num_bytes
;
2443 backref
->extent_offset
= extent_offset
;
2444 backref
->generation
= btrfs_file_extent_generation(leaf
, extent
);
2446 backref_insert(&new->root
, backref
);
2449 btrfs_release_path(path
);
2454 static noinline
bool record_extent_backrefs(struct btrfs_path
*path
,
2455 struct new_sa_defrag_extent
*new)
2457 struct btrfs_fs_info
*fs_info
= btrfs_sb(new->inode
->i_sb
);
2458 struct old_sa_defrag_extent
*old
, *tmp
;
2463 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2464 ret
= iterate_inodes_from_logical(old
->bytenr
+
2465 old
->extent_offset
, fs_info
,
2466 path
, record_one_backref
,
2468 if (ret
< 0 && ret
!= -ENOENT
)
2471 /* no backref to be processed for this extent */
2473 list_del(&old
->list
);
2478 if (list_empty(&new->head
))
2484 static int relink_is_mergable(struct extent_buffer
*leaf
,
2485 struct btrfs_file_extent_item
*fi
,
2486 struct new_sa_defrag_extent
*new)
2488 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) != new->bytenr
)
2491 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2494 if (btrfs_file_extent_compression(leaf
, fi
) != new->compress_type
)
2497 if (btrfs_file_extent_encryption(leaf
, fi
) ||
2498 btrfs_file_extent_other_encoding(leaf
, fi
))
2505 * Note the backref might has changed, and in this case we just return 0.
2507 static noinline
int relink_extent_backref(struct btrfs_path
*path
,
2508 struct sa_defrag_extent_backref
*prev
,
2509 struct sa_defrag_extent_backref
*backref
)
2511 struct btrfs_file_extent_item
*extent
;
2512 struct btrfs_file_extent_item
*item
;
2513 struct btrfs_ordered_extent
*ordered
;
2514 struct btrfs_trans_handle
*trans
;
2515 struct btrfs_root
*root
;
2516 struct btrfs_key key
;
2517 struct extent_buffer
*leaf
;
2518 struct old_sa_defrag_extent
*old
= backref
->old
;
2519 struct new_sa_defrag_extent
*new = old
->new;
2520 struct btrfs_fs_info
*fs_info
= btrfs_sb(new->inode
->i_sb
);
2521 struct inode
*inode
;
2522 struct extent_state
*cached
= NULL
;
2531 if (prev
&& prev
->root_id
== backref
->root_id
&&
2532 prev
->inum
== backref
->inum
&&
2533 prev
->file_pos
+ prev
->num_bytes
== backref
->file_pos
)
2536 /* step 1: get root */
2537 key
.objectid
= backref
->root_id
;
2538 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2539 key
.offset
= (u64
)-1;
2541 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
2543 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2545 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2546 if (PTR_ERR(root
) == -ENOENT
)
2548 return PTR_ERR(root
);
2551 if (btrfs_root_readonly(root
)) {
2552 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2556 /* step 2: get inode */
2557 key
.objectid
= backref
->inum
;
2558 key
.type
= BTRFS_INODE_ITEM_KEY
;
2561 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
2562 if (IS_ERR(inode
)) {
2563 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2567 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2569 /* step 3: relink backref */
2570 lock_start
= backref
->file_pos
;
2571 lock_end
= backref
->file_pos
+ backref
->num_bytes
- 1;
2572 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2575 ordered
= btrfs_lookup_first_ordered_extent(inode
, lock_end
);
2577 btrfs_put_ordered_extent(ordered
);
2581 trans
= btrfs_join_transaction(root
);
2582 if (IS_ERR(trans
)) {
2583 ret
= PTR_ERR(trans
);
2587 key
.objectid
= backref
->inum
;
2588 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2589 key
.offset
= backref
->file_pos
;
2591 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2594 } else if (ret
> 0) {
2599 extent
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2600 struct btrfs_file_extent_item
);
2602 if (btrfs_file_extent_generation(path
->nodes
[0], extent
) !=
2603 backref
->generation
)
2606 btrfs_release_path(path
);
2608 start
= backref
->file_pos
;
2609 if (backref
->extent_offset
< old
->extent_offset
+ old
->offset
)
2610 start
+= old
->extent_offset
+ old
->offset
-
2611 backref
->extent_offset
;
2613 len
= min(backref
->extent_offset
+ backref
->num_bytes
,
2614 old
->extent_offset
+ old
->offset
+ old
->len
);
2615 len
-= max(backref
->extent_offset
, old
->extent_offset
+ old
->offset
);
2617 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
2622 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
2623 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2626 path
->leave_spinning
= 1;
2628 struct btrfs_file_extent_item
*fi
;
2630 struct btrfs_key found_key
;
2632 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2637 leaf
= path
->nodes
[0];
2638 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2640 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2641 struct btrfs_file_extent_item
);
2642 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
2644 if (extent_len
+ found_key
.offset
== start
&&
2645 relink_is_mergable(leaf
, fi
, new)) {
2646 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2648 btrfs_mark_buffer_dirty(leaf
);
2649 inode_add_bytes(inode
, len
);
2655 btrfs_release_path(path
);
2660 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2663 btrfs_abort_transaction(trans
, ret
);
2667 leaf
= path
->nodes
[0];
2668 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2669 struct btrfs_file_extent_item
);
2670 btrfs_set_file_extent_disk_bytenr(leaf
, item
, new->bytenr
);
2671 btrfs_set_file_extent_disk_num_bytes(leaf
, item
, new->disk_len
);
2672 btrfs_set_file_extent_offset(leaf
, item
, start
- new->file_pos
);
2673 btrfs_set_file_extent_num_bytes(leaf
, item
, len
);
2674 btrfs_set_file_extent_ram_bytes(leaf
, item
, new->len
);
2675 btrfs_set_file_extent_generation(leaf
, item
, trans
->transid
);
2676 btrfs_set_file_extent_type(leaf
, item
, BTRFS_FILE_EXTENT_REG
);
2677 btrfs_set_file_extent_compression(leaf
, item
, new->compress_type
);
2678 btrfs_set_file_extent_encryption(leaf
, item
, 0);
2679 btrfs_set_file_extent_other_encoding(leaf
, item
, 0);
2681 btrfs_mark_buffer_dirty(leaf
);
2682 inode_add_bytes(inode
, len
);
2683 btrfs_release_path(path
);
2685 ret
= btrfs_inc_extent_ref(trans
, fs_info
, new->bytenr
,
2687 backref
->root_id
, backref
->inum
,
2688 new->file_pos
); /* start - extent_offset */
2690 btrfs_abort_transaction(trans
, ret
);
2696 btrfs_release_path(path
);
2697 path
->leave_spinning
= 0;
2698 btrfs_end_transaction(trans
);
2700 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2706 static void free_sa_defrag_extent(struct new_sa_defrag_extent
*new)
2708 struct old_sa_defrag_extent
*old
, *tmp
;
2713 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2719 static void relink_file_extents(struct new_sa_defrag_extent
*new)
2721 struct btrfs_fs_info
*fs_info
= btrfs_sb(new->inode
->i_sb
);
2722 struct btrfs_path
*path
;
2723 struct sa_defrag_extent_backref
*backref
;
2724 struct sa_defrag_extent_backref
*prev
= NULL
;
2725 struct inode
*inode
;
2726 struct btrfs_root
*root
;
2727 struct rb_node
*node
;
2731 root
= BTRFS_I(inode
)->root
;
2733 path
= btrfs_alloc_path();
2737 if (!record_extent_backrefs(path
, new)) {
2738 btrfs_free_path(path
);
2741 btrfs_release_path(path
);
2744 node
= rb_first(&new->root
);
2747 rb_erase(node
, &new->root
);
2749 backref
= rb_entry(node
, struct sa_defrag_extent_backref
, node
);
2751 ret
= relink_extent_backref(path
, prev
, backref
);
2764 btrfs_free_path(path
);
2766 free_sa_defrag_extent(new);
2768 atomic_dec(&fs_info
->defrag_running
);
2769 wake_up(&fs_info
->transaction_wait
);
2772 static struct new_sa_defrag_extent
*
2773 record_old_file_extents(struct inode
*inode
,
2774 struct btrfs_ordered_extent
*ordered
)
2776 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2777 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2778 struct btrfs_path
*path
;
2779 struct btrfs_key key
;
2780 struct old_sa_defrag_extent
*old
;
2781 struct new_sa_defrag_extent
*new;
2784 new = kmalloc(sizeof(*new), GFP_NOFS
);
2789 new->file_pos
= ordered
->file_offset
;
2790 new->len
= ordered
->len
;
2791 new->bytenr
= ordered
->start
;
2792 new->disk_len
= ordered
->disk_len
;
2793 new->compress_type
= ordered
->compress_type
;
2794 new->root
= RB_ROOT
;
2795 INIT_LIST_HEAD(&new->head
);
2797 path
= btrfs_alloc_path();
2801 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
2802 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2803 key
.offset
= new->file_pos
;
2805 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2808 if (ret
> 0 && path
->slots
[0] > 0)
2811 /* find out all the old extents for the file range */
2813 struct btrfs_file_extent_item
*extent
;
2814 struct extent_buffer
*l
;
2823 slot
= path
->slots
[0];
2825 if (slot
>= btrfs_header_nritems(l
)) {
2826 ret
= btrfs_next_leaf(root
, path
);
2834 btrfs_item_key_to_cpu(l
, &key
, slot
);
2836 if (key
.objectid
!= btrfs_ino(BTRFS_I(inode
)))
2838 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2840 if (key
.offset
>= new->file_pos
+ new->len
)
2843 extent
= btrfs_item_ptr(l
, slot
, struct btrfs_file_extent_item
);
2845 num_bytes
= btrfs_file_extent_num_bytes(l
, extent
);
2846 if (key
.offset
+ num_bytes
< new->file_pos
)
2849 disk_bytenr
= btrfs_file_extent_disk_bytenr(l
, extent
);
2853 extent_offset
= btrfs_file_extent_offset(l
, extent
);
2855 old
= kmalloc(sizeof(*old
), GFP_NOFS
);
2859 offset
= max(new->file_pos
, key
.offset
);
2860 end
= min(new->file_pos
+ new->len
, key
.offset
+ num_bytes
);
2862 old
->bytenr
= disk_bytenr
;
2863 old
->extent_offset
= extent_offset
;
2864 old
->offset
= offset
- key
.offset
;
2865 old
->len
= end
- offset
;
2868 list_add_tail(&old
->list
, &new->head
);
2874 btrfs_free_path(path
);
2875 atomic_inc(&fs_info
->defrag_running
);
2880 btrfs_free_path(path
);
2882 free_sa_defrag_extent(new);
2886 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info
*fs_info
,
2889 struct btrfs_block_group_cache
*cache
;
2891 cache
= btrfs_lookup_block_group(fs_info
, start
);
2894 spin_lock(&cache
->lock
);
2895 cache
->delalloc_bytes
-= len
;
2896 spin_unlock(&cache
->lock
);
2898 btrfs_put_block_group(cache
);
2901 /* as ordered data IO finishes, this gets called so we can finish
2902 * an ordered extent if the range of bytes in the file it covers are
2905 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
2907 struct inode
*inode
= ordered_extent
->inode
;
2908 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2909 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2910 struct btrfs_trans_handle
*trans
= NULL
;
2911 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2912 struct extent_state
*cached_state
= NULL
;
2913 struct new_sa_defrag_extent
*new = NULL
;
2914 int compress_type
= 0;
2916 u64 logical_len
= ordered_extent
->len
;
2918 bool truncated
= false;
2919 bool range_locked
= false;
2920 bool clear_new_delalloc_bytes
= false;
2922 if (!test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
2923 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
) &&
2924 !test_bit(BTRFS_ORDERED_DIRECT
, &ordered_extent
->flags
))
2925 clear_new_delalloc_bytes
= true;
2927 nolock
= btrfs_is_free_space_inode(BTRFS_I(inode
));
2929 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
2934 btrfs_free_io_failure_record(BTRFS_I(inode
),
2935 ordered_extent
->file_offset
,
2936 ordered_extent
->file_offset
+
2937 ordered_extent
->len
- 1);
2939 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered_extent
->flags
)) {
2941 logical_len
= ordered_extent
->truncated_len
;
2942 /* Truncated the entire extent, don't bother adding */
2947 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
2948 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
2951 * For mwrite(mmap + memset to write) case, we still reserve
2952 * space for NOCOW range.
2953 * As NOCOW won't cause a new delayed ref, just free the space
2955 btrfs_qgroup_free_data(inode
, NULL
, ordered_extent
->file_offset
,
2956 ordered_extent
->len
);
2957 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2959 trans
= btrfs_join_transaction_nolock(root
);
2961 trans
= btrfs_join_transaction(root
);
2962 if (IS_ERR(trans
)) {
2963 ret
= PTR_ERR(trans
);
2967 trans
->block_rsv
= &fs_info
->delalloc_block_rsv
;
2968 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2969 if (ret
) /* -ENOMEM or corruption */
2970 btrfs_abort_transaction(trans
, ret
);
2974 range_locked
= true;
2975 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
2976 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2979 ret
= test_range_bit(io_tree
, ordered_extent
->file_offset
,
2980 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2981 EXTENT_DEFRAG
, 0, cached_state
);
2983 u64 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
2984 if (0 && last_snapshot
>= BTRFS_I(inode
)->generation
)
2985 /* the inode is shared */
2986 new = record_old_file_extents(inode
, ordered_extent
);
2988 clear_extent_bit(io_tree
, ordered_extent
->file_offset
,
2989 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2990 EXTENT_DEFRAG
, 0, 0, &cached_state
, GFP_NOFS
);
2994 trans
= btrfs_join_transaction_nolock(root
);
2996 trans
= btrfs_join_transaction(root
);
2997 if (IS_ERR(trans
)) {
2998 ret
= PTR_ERR(trans
);
3003 trans
->block_rsv
= &fs_info
->delalloc_block_rsv
;
3005 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
3006 compress_type
= ordered_extent
->compress_type
;
3007 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
3008 BUG_ON(compress_type
);
3009 ret
= btrfs_mark_extent_written(trans
, BTRFS_I(inode
),
3010 ordered_extent
->file_offset
,
3011 ordered_extent
->file_offset
+
3014 BUG_ON(root
== fs_info
->tree_root
);
3015 ret
= insert_reserved_file_extent(trans
, inode
,
3016 ordered_extent
->file_offset
,
3017 ordered_extent
->start
,
3018 ordered_extent
->disk_len
,
3019 logical_len
, logical_len
,
3020 compress_type
, 0, 0,
3021 BTRFS_FILE_EXTENT_REG
);
3023 btrfs_release_delalloc_bytes(fs_info
,
3024 ordered_extent
->start
,
3025 ordered_extent
->disk_len
);
3027 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
3028 ordered_extent
->file_offset
, ordered_extent
->len
,
3031 btrfs_abort_transaction(trans
, ret
);
3035 add_pending_csums(trans
, inode
, &ordered_extent
->list
);
3037 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
3038 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
3039 if (ret
) { /* -ENOMEM or corruption */
3040 btrfs_abort_transaction(trans
, ret
);
3045 if (range_locked
|| clear_new_delalloc_bytes
) {
3046 unsigned int clear_bits
= 0;
3049 clear_bits
|= EXTENT_LOCKED
;
3050 if (clear_new_delalloc_bytes
)
3051 clear_bits
|= EXTENT_DELALLOC_NEW
;
3052 clear_extent_bit(&BTRFS_I(inode
)->io_tree
,
3053 ordered_extent
->file_offset
,
3054 ordered_extent
->file_offset
+
3055 ordered_extent
->len
- 1,
3057 (clear_bits
& EXTENT_LOCKED
) ? 1 : 0,
3058 0, &cached_state
, GFP_NOFS
);
3061 if (root
!= fs_info
->tree_root
)
3062 btrfs_delalloc_release_metadata(BTRFS_I(inode
),
3063 ordered_extent
->len
);
3065 btrfs_end_transaction(trans
);
3067 if (ret
|| truncated
) {
3071 start
= ordered_extent
->file_offset
+ logical_len
;
3073 start
= ordered_extent
->file_offset
;
3074 end
= ordered_extent
->file_offset
+ ordered_extent
->len
- 1;
3075 clear_extent_uptodate(io_tree
, start
, end
, NULL
, GFP_NOFS
);
3077 /* Drop the cache for the part of the extent we didn't write. */
3078 btrfs_drop_extent_cache(BTRFS_I(inode
), start
, end
, 0);
3081 * If the ordered extent had an IOERR or something else went
3082 * wrong we need to return the space for this ordered extent
3083 * back to the allocator. We only free the extent in the
3084 * truncated case if we didn't write out the extent at all.
3086 if ((ret
|| !logical_len
) &&
3087 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
3088 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
))
3089 btrfs_free_reserved_extent(fs_info
,
3090 ordered_extent
->start
,
3091 ordered_extent
->disk_len
, 1);
3096 * This needs to be done to make sure anybody waiting knows we are done
3097 * updating everything for this ordered extent.
3099 btrfs_remove_ordered_extent(inode
, ordered_extent
);
3101 /* for snapshot-aware defrag */
3104 free_sa_defrag_extent(new);
3105 atomic_dec(&fs_info
->defrag_running
);
3107 relink_file_extents(new);
3112 btrfs_put_ordered_extent(ordered_extent
);
3113 /* once for the tree */
3114 btrfs_put_ordered_extent(ordered_extent
);
3119 static void finish_ordered_fn(struct btrfs_work
*work
)
3121 struct btrfs_ordered_extent
*ordered_extent
;
3122 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
3123 btrfs_finish_ordered_io(ordered_extent
);
3126 static void btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
3127 struct extent_state
*state
, int uptodate
)
3129 struct inode
*inode
= page
->mapping
->host
;
3130 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3131 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
3132 struct btrfs_workqueue
*wq
;
3133 btrfs_work_func_t func
;
3135 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
3137 ClearPagePrivate2(page
);
3138 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
3139 end
- start
+ 1, uptodate
))
3142 if (btrfs_is_free_space_inode(BTRFS_I(inode
))) {
3143 wq
= fs_info
->endio_freespace_worker
;
3144 func
= btrfs_freespace_write_helper
;
3146 wq
= fs_info
->endio_write_workers
;
3147 func
= btrfs_endio_write_helper
;
3150 btrfs_init_work(&ordered_extent
->work
, func
, finish_ordered_fn
, NULL
,
3152 btrfs_queue_work(wq
, &ordered_extent
->work
);
3155 static int __readpage_endio_check(struct inode
*inode
,
3156 struct btrfs_io_bio
*io_bio
,
3157 int icsum
, struct page
*page
,
3158 int pgoff
, u64 start
, size_t len
)
3164 csum_expected
= *(((u32
*)io_bio
->csum
) + icsum
);
3166 kaddr
= kmap_atomic(page
);
3167 csum
= btrfs_csum_data(kaddr
+ pgoff
, csum
, len
);
3168 btrfs_csum_final(csum
, (u8
*)&csum
);
3169 if (csum
!= csum_expected
)
3172 kunmap_atomic(kaddr
);
3175 btrfs_print_data_csum_error(BTRFS_I(inode
), start
, csum
, csum_expected
,
3176 io_bio
->mirror_num
);
3177 memset(kaddr
+ pgoff
, 1, len
);
3178 flush_dcache_page(page
);
3179 kunmap_atomic(kaddr
);
3184 * when reads are done, we need to check csums to verify the data is correct
3185 * if there's a match, we allow the bio to finish. If not, the code in
3186 * extent_io.c will try to find good copies for us.
3188 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
3189 u64 phy_offset
, struct page
*page
,
3190 u64 start
, u64 end
, int mirror
)
3192 size_t offset
= start
- page_offset(page
);
3193 struct inode
*inode
= page
->mapping
->host
;
3194 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3195 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3197 if (PageChecked(page
)) {
3198 ClearPageChecked(page
);
3202 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
3205 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
3206 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
3207 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
);
3211 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
3212 return __readpage_endio_check(inode
, io_bio
, phy_offset
, page
, offset
,
3213 start
, (size_t)(end
- start
+ 1));
3216 void btrfs_add_delayed_iput(struct inode
*inode
)
3218 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3219 struct btrfs_inode
*binode
= BTRFS_I(inode
);
3221 if (atomic_add_unless(&inode
->i_count
, -1, 1))
3224 spin_lock(&fs_info
->delayed_iput_lock
);
3225 if (binode
->delayed_iput_count
== 0) {
3226 ASSERT(list_empty(&binode
->delayed_iput
));
3227 list_add_tail(&binode
->delayed_iput
, &fs_info
->delayed_iputs
);
3229 binode
->delayed_iput_count
++;
3231 spin_unlock(&fs_info
->delayed_iput_lock
);
3234 void btrfs_run_delayed_iputs(struct btrfs_fs_info
*fs_info
)
3237 spin_lock(&fs_info
->delayed_iput_lock
);
3238 while (!list_empty(&fs_info
->delayed_iputs
)) {
3239 struct btrfs_inode
*inode
;
3241 inode
= list_first_entry(&fs_info
->delayed_iputs
,
3242 struct btrfs_inode
, delayed_iput
);
3243 if (inode
->delayed_iput_count
) {
3244 inode
->delayed_iput_count
--;
3245 list_move_tail(&inode
->delayed_iput
,
3246 &fs_info
->delayed_iputs
);
3248 list_del_init(&inode
->delayed_iput
);
3250 spin_unlock(&fs_info
->delayed_iput_lock
);
3251 iput(&inode
->vfs_inode
);
3252 spin_lock(&fs_info
->delayed_iput_lock
);
3254 spin_unlock(&fs_info
->delayed_iput_lock
);
3258 * This is called in transaction commit time. If there are no orphan
3259 * files in the subvolume, it removes orphan item and frees block_rsv
3262 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
3263 struct btrfs_root
*root
)
3265 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3266 struct btrfs_block_rsv
*block_rsv
;
3269 if (atomic_read(&root
->orphan_inodes
) ||
3270 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
3273 spin_lock(&root
->orphan_lock
);
3274 if (atomic_read(&root
->orphan_inodes
)) {
3275 spin_unlock(&root
->orphan_lock
);
3279 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
3280 spin_unlock(&root
->orphan_lock
);
3284 block_rsv
= root
->orphan_block_rsv
;
3285 root
->orphan_block_rsv
= NULL
;
3286 spin_unlock(&root
->orphan_lock
);
3288 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
) &&
3289 btrfs_root_refs(&root
->root_item
) > 0) {
3290 ret
= btrfs_del_orphan_item(trans
, fs_info
->tree_root
,
3291 root
->root_key
.objectid
);
3293 btrfs_abort_transaction(trans
, ret
);
3295 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
,
3300 WARN_ON(block_rsv
->size
> 0);
3301 btrfs_free_block_rsv(fs_info
, block_rsv
);
3306 * This creates an orphan entry for the given inode in case something goes
3307 * wrong in the middle of an unlink/truncate.
3309 * NOTE: caller of this function should reserve 5 units of metadata for
3312 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
,
3313 struct btrfs_inode
*inode
)
3315 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
3316 struct btrfs_root
*root
= inode
->root
;
3317 struct btrfs_block_rsv
*block_rsv
= NULL
;
3322 if (!root
->orphan_block_rsv
) {
3323 block_rsv
= btrfs_alloc_block_rsv(fs_info
,
3324 BTRFS_BLOCK_RSV_TEMP
);
3329 spin_lock(&root
->orphan_lock
);
3330 if (!root
->orphan_block_rsv
) {
3331 root
->orphan_block_rsv
= block_rsv
;
3332 } else if (block_rsv
) {
3333 btrfs_free_block_rsv(fs_info
, block_rsv
);
3337 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3338 &inode
->runtime_flags
)) {
3341 * For proper ENOSPC handling, we should do orphan
3342 * cleanup when mounting. But this introduces backward
3343 * compatibility issue.
3345 if (!xchg(&root
->orphan_item_inserted
, 1))
3351 atomic_inc(&root
->orphan_inodes
);
3354 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3355 &inode
->runtime_flags
))
3357 spin_unlock(&root
->orphan_lock
);
3359 /* grab metadata reservation from transaction handle */
3361 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
3364 atomic_dec(&root
->orphan_inodes
);
3365 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3366 &inode
->runtime_flags
);
3368 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3369 &inode
->runtime_flags
);
3374 /* insert an orphan item to track this unlinked/truncated file */
3376 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
3378 atomic_dec(&root
->orphan_inodes
);
3380 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3381 &inode
->runtime_flags
);
3382 btrfs_orphan_release_metadata(inode
);
3384 if (ret
!= -EEXIST
) {
3385 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3386 &inode
->runtime_flags
);
3387 btrfs_abort_transaction(trans
, ret
);
3394 /* insert an orphan item to track subvolume contains orphan files */
3396 ret
= btrfs_insert_orphan_item(trans
, fs_info
->tree_root
,
3397 root
->root_key
.objectid
);
3398 if (ret
&& ret
!= -EEXIST
) {
3399 btrfs_abort_transaction(trans
, ret
);
3407 * We have done the truncate/delete so we can go ahead and remove the orphan
3408 * item for this particular inode.
3410 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3411 struct btrfs_inode
*inode
)
3413 struct btrfs_root
*root
= inode
->root
;
3414 int delete_item
= 0;
3415 int release_rsv
= 0;
3418 spin_lock(&root
->orphan_lock
);
3419 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3420 &inode
->runtime_flags
))
3423 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3424 &inode
->runtime_flags
))
3426 spin_unlock(&root
->orphan_lock
);
3429 atomic_dec(&root
->orphan_inodes
);
3431 ret
= btrfs_del_orphan_item(trans
, root
,
3436 btrfs_orphan_release_metadata(inode
);
3442 * this cleans up any orphans that may be left on the list from the last use
3445 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3447 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3448 struct btrfs_path
*path
;
3449 struct extent_buffer
*leaf
;
3450 struct btrfs_key key
, found_key
;
3451 struct btrfs_trans_handle
*trans
;
3452 struct inode
*inode
;
3453 u64 last_objectid
= 0;
3454 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
3456 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
3459 path
= btrfs_alloc_path();
3464 path
->reada
= READA_BACK
;
3466 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3467 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
3468 key
.offset
= (u64
)-1;
3471 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3476 * if ret == 0 means we found what we were searching for, which
3477 * is weird, but possible, so only screw with path if we didn't
3478 * find the key and see if we have stuff that matches
3482 if (path
->slots
[0] == 0)
3487 /* pull out the item */
3488 leaf
= path
->nodes
[0];
3489 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3491 /* make sure the item matches what we want */
3492 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3494 if (found_key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
3497 /* release the path since we're done with it */
3498 btrfs_release_path(path
);
3501 * this is where we are basically btrfs_lookup, without the
3502 * crossing root thing. we store the inode number in the
3503 * offset of the orphan item.
3506 if (found_key
.offset
== last_objectid
) {
3508 "Error removing orphan entry, stopping orphan cleanup");
3513 last_objectid
= found_key
.offset
;
3515 found_key
.objectid
= found_key
.offset
;
3516 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3517 found_key
.offset
= 0;
3518 inode
= btrfs_iget(fs_info
->sb
, &found_key
, root
, NULL
);
3519 ret
= PTR_ERR_OR_ZERO(inode
);
3520 if (ret
&& ret
!= -ENOENT
)
3523 if (ret
== -ENOENT
&& root
== fs_info
->tree_root
) {
3524 struct btrfs_root
*dead_root
;
3525 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3526 int is_dead_root
= 0;
3529 * this is an orphan in the tree root. Currently these
3530 * could come from 2 sources:
3531 * a) a snapshot deletion in progress
3532 * b) a free space cache inode
3533 * We need to distinguish those two, as the snapshot
3534 * orphan must not get deleted.
3535 * find_dead_roots already ran before us, so if this
3536 * is a snapshot deletion, we should find the root
3537 * in the dead_roots list
3539 spin_lock(&fs_info
->trans_lock
);
3540 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
3542 if (dead_root
->root_key
.objectid
==
3543 found_key
.objectid
) {
3548 spin_unlock(&fs_info
->trans_lock
);
3550 /* prevent this orphan from being found again */
3551 key
.offset
= found_key
.objectid
- 1;
3556 * Inode is already gone but the orphan item is still there,
3557 * kill the orphan item.
3559 if (ret
== -ENOENT
) {
3560 trans
= btrfs_start_transaction(root
, 1);
3561 if (IS_ERR(trans
)) {
3562 ret
= PTR_ERR(trans
);
3565 btrfs_debug(fs_info
, "auto deleting %Lu",
3566 found_key
.objectid
);
3567 ret
= btrfs_del_orphan_item(trans
, root
,
3568 found_key
.objectid
);
3569 btrfs_end_transaction(trans
);
3576 * add this inode to the orphan list so btrfs_orphan_del does
3577 * the proper thing when we hit it
3579 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3580 &BTRFS_I(inode
)->runtime_flags
);
3581 atomic_inc(&root
->orphan_inodes
);
3583 /* if we have links, this was a truncate, lets do that */
3584 if (inode
->i_nlink
) {
3585 if (WARN_ON(!S_ISREG(inode
->i_mode
))) {
3591 /* 1 for the orphan item deletion. */
3592 trans
= btrfs_start_transaction(root
, 1);
3593 if (IS_ERR(trans
)) {
3595 ret
= PTR_ERR(trans
);
3598 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
3599 btrfs_end_transaction(trans
);
3605 ret
= btrfs_truncate(inode
);
3607 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
3612 /* this will do delete_inode and everything for us */
3617 /* release the path since we're done with it */
3618 btrfs_release_path(path
);
3620 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
3622 if (root
->orphan_block_rsv
)
3623 btrfs_block_rsv_release(fs_info
, root
->orphan_block_rsv
,
3626 if (root
->orphan_block_rsv
||
3627 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
)) {
3628 trans
= btrfs_join_transaction(root
);
3630 btrfs_end_transaction(trans
);
3634 btrfs_debug(fs_info
, "unlinked %d orphans", nr_unlink
);
3636 btrfs_debug(fs_info
, "truncated %d orphans", nr_truncate
);
3640 btrfs_err(fs_info
, "could not do orphan cleanup %d", ret
);
3641 btrfs_free_path(path
);
3646 * very simple check to peek ahead in the leaf looking for xattrs. If we
3647 * don't find any xattrs, we know there can't be any acls.
3649 * slot is the slot the inode is in, objectid is the objectid of the inode
3651 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3652 int slot
, u64 objectid
,
3653 int *first_xattr_slot
)
3655 u32 nritems
= btrfs_header_nritems(leaf
);
3656 struct btrfs_key found_key
;
3657 static u64 xattr_access
= 0;
3658 static u64 xattr_default
= 0;
3661 if (!xattr_access
) {
3662 xattr_access
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS
,
3663 strlen(XATTR_NAME_POSIX_ACL_ACCESS
));
3664 xattr_default
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT
,
3665 strlen(XATTR_NAME_POSIX_ACL_DEFAULT
));
3669 *first_xattr_slot
= -1;
3670 while (slot
< nritems
) {
3671 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3673 /* we found a different objectid, there must not be acls */
3674 if (found_key
.objectid
!= objectid
)
3677 /* we found an xattr, assume we've got an acl */
3678 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
3679 if (*first_xattr_slot
== -1)
3680 *first_xattr_slot
= slot
;
3681 if (found_key
.offset
== xattr_access
||
3682 found_key
.offset
== xattr_default
)
3687 * we found a key greater than an xattr key, there can't
3688 * be any acls later on
3690 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3697 * it goes inode, inode backrefs, xattrs, extents,
3698 * so if there are a ton of hard links to an inode there can
3699 * be a lot of backrefs. Don't waste time searching too hard,
3700 * this is just an optimization
3705 /* we hit the end of the leaf before we found an xattr or
3706 * something larger than an xattr. We have to assume the inode
3709 if (*first_xattr_slot
== -1)
3710 *first_xattr_slot
= slot
;
3715 * read an inode from the btree into the in-memory inode
3717 static int btrfs_read_locked_inode(struct inode
*inode
)
3719 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3720 struct btrfs_path
*path
;
3721 struct extent_buffer
*leaf
;
3722 struct btrfs_inode_item
*inode_item
;
3723 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3724 struct btrfs_key location
;
3729 bool filled
= false;
3730 int first_xattr_slot
;
3732 ret
= btrfs_fill_inode(inode
, &rdev
);
3736 path
= btrfs_alloc_path();
3742 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3744 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3751 leaf
= path
->nodes
[0];
3756 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3757 struct btrfs_inode_item
);
3758 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3759 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3760 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3761 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3762 btrfs_i_size_write(BTRFS_I(inode
), btrfs_inode_size(leaf
, inode_item
));
3764 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->atime
);
3765 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->atime
);
3767 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->mtime
);
3768 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->mtime
);
3770 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->ctime
);
3771 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->ctime
);
3773 BTRFS_I(inode
)->i_otime
.tv_sec
=
3774 btrfs_timespec_sec(leaf
, &inode_item
->otime
);
3775 BTRFS_I(inode
)->i_otime
.tv_nsec
=
3776 btrfs_timespec_nsec(leaf
, &inode_item
->otime
);
3778 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3779 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3780 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3782 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
3783 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3785 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3787 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3788 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
3792 * If we were modified in the current generation and evicted from memory
3793 * and then re-read we need to do a full sync since we don't have any
3794 * idea about which extents were modified before we were evicted from
3797 * This is required for both inode re-read from disk and delayed inode
3798 * in delayed_nodes_tree.
3800 if (BTRFS_I(inode
)->last_trans
== fs_info
->generation
)
3801 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3802 &BTRFS_I(inode
)->runtime_flags
);
3805 * We don't persist the id of the transaction where an unlink operation
3806 * against the inode was last made. So here we assume the inode might
3807 * have been evicted, and therefore the exact value of last_unlink_trans
3808 * lost, and set it to last_trans to avoid metadata inconsistencies
3809 * between the inode and its parent if the inode is fsync'ed and the log
3810 * replayed. For example, in the scenario:
3813 * ln mydir/foo mydir/bar
3816 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3817 * xfs_io -c fsync mydir/foo
3819 * mount fs, triggers fsync log replay
3821 * We must make sure that when we fsync our inode foo we also log its
3822 * parent inode, otherwise after log replay the parent still has the
3823 * dentry with the "bar" name but our inode foo has a link count of 1
3824 * and doesn't have an inode ref with the name "bar" anymore.
3826 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3827 * but it guarantees correctness at the expense of occasional full
3828 * transaction commits on fsync if our inode is a directory, or if our
3829 * inode is not a directory, logging its parent unnecessarily.
3831 BTRFS_I(inode
)->last_unlink_trans
= BTRFS_I(inode
)->last_trans
;
3834 if (inode
->i_nlink
!= 1 ||
3835 path
->slots
[0] >= btrfs_header_nritems(leaf
))
3838 btrfs_item_key_to_cpu(leaf
, &location
, path
->slots
[0]);
3839 if (location
.objectid
!= btrfs_ino(BTRFS_I(inode
)))
3842 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3843 if (location
.type
== BTRFS_INODE_REF_KEY
) {
3844 struct btrfs_inode_ref
*ref
;
3846 ref
= (struct btrfs_inode_ref
*)ptr
;
3847 BTRFS_I(inode
)->dir_index
= btrfs_inode_ref_index(leaf
, ref
);
3848 } else if (location
.type
== BTRFS_INODE_EXTREF_KEY
) {
3849 struct btrfs_inode_extref
*extref
;
3851 extref
= (struct btrfs_inode_extref
*)ptr
;
3852 BTRFS_I(inode
)->dir_index
= btrfs_inode_extref_index(leaf
,
3857 * try to precache a NULL acl entry for files that don't have
3858 * any xattrs or acls
3860 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
3861 btrfs_ino(BTRFS_I(inode
)), &first_xattr_slot
);
3862 if (first_xattr_slot
!= -1) {
3863 path
->slots
[0] = first_xattr_slot
;
3864 ret
= btrfs_load_inode_props(inode
, path
);
3867 "error loading props for ino %llu (root %llu): %d",
3868 btrfs_ino(BTRFS_I(inode
)),
3869 root
->root_key
.objectid
, ret
);
3871 btrfs_free_path(path
);
3874 cache_no_acl(inode
);
3876 switch (inode
->i_mode
& S_IFMT
) {
3878 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3879 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3880 inode
->i_fop
= &btrfs_file_operations
;
3881 inode
->i_op
= &btrfs_file_inode_operations
;
3884 inode
->i_fop
= &btrfs_dir_file_operations
;
3885 inode
->i_op
= &btrfs_dir_inode_operations
;
3888 inode
->i_op
= &btrfs_symlink_inode_operations
;
3889 inode_nohighmem(inode
);
3890 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
3893 inode
->i_op
= &btrfs_special_inode_operations
;
3894 init_special_inode(inode
, inode
->i_mode
, rdev
);
3898 btrfs_update_iflags(inode
);
3902 btrfs_free_path(path
);
3903 make_bad_inode(inode
);
3908 * given a leaf and an inode, copy the inode fields into the leaf
3910 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3911 struct extent_buffer
*leaf
,
3912 struct btrfs_inode_item
*item
,
3913 struct inode
*inode
)
3915 struct btrfs_map_token token
;
3917 btrfs_init_map_token(&token
);
3919 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3920 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3921 btrfs_set_token_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
,
3923 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3924 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3926 btrfs_set_token_timespec_sec(leaf
, &item
->atime
,
3927 inode
->i_atime
.tv_sec
, &token
);
3928 btrfs_set_token_timespec_nsec(leaf
, &item
->atime
,
3929 inode
->i_atime
.tv_nsec
, &token
);
3931 btrfs_set_token_timespec_sec(leaf
, &item
->mtime
,
3932 inode
->i_mtime
.tv_sec
, &token
);
3933 btrfs_set_token_timespec_nsec(leaf
, &item
->mtime
,
3934 inode
->i_mtime
.tv_nsec
, &token
);
3936 btrfs_set_token_timespec_sec(leaf
, &item
->ctime
,
3937 inode
->i_ctime
.tv_sec
, &token
);
3938 btrfs_set_token_timespec_nsec(leaf
, &item
->ctime
,
3939 inode
->i_ctime
.tv_nsec
, &token
);
3941 btrfs_set_token_timespec_sec(leaf
, &item
->otime
,
3942 BTRFS_I(inode
)->i_otime
.tv_sec
, &token
);
3943 btrfs_set_token_timespec_nsec(leaf
, &item
->otime
,
3944 BTRFS_I(inode
)->i_otime
.tv_nsec
, &token
);
3946 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3948 btrfs_set_token_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
,
3950 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
3951 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3952 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3953 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3954 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3958 * copy everything in the in-memory inode into the btree.
3960 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
3961 struct btrfs_root
*root
, struct inode
*inode
)
3963 struct btrfs_inode_item
*inode_item
;
3964 struct btrfs_path
*path
;
3965 struct extent_buffer
*leaf
;
3968 path
= btrfs_alloc_path();
3972 path
->leave_spinning
= 1;
3973 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
3981 leaf
= path
->nodes
[0];
3982 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3983 struct btrfs_inode_item
);
3985 fill_inode_item(trans
, leaf
, inode_item
, inode
);
3986 btrfs_mark_buffer_dirty(leaf
);
3987 btrfs_set_inode_last_trans(trans
, inode
);
3990 btrfs_free_path(path
);
3995 * copy everything in the in-memory inode into the btree.
3997 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
3998 struct btrfs_root
*root
, struct inode
*inode
)
4000 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4004 * If the inode is a free space inode, we can deadlock during commit
4005 * if we put it into the delayed code.
4007 * The data relocation inode should also be directly updated
4010 if (!btrfs_is_free_space_inode(BTRFS_I(inode
))
4011 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
4012 && !test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
)) {
4013 btrfs_update_root_times(trans
, root
);
4015 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
4017 btrfs_set_inode_last_trans(trans
, inode
);
4021 return btrfs_update_inode_item(trans
, root
, inode
);
4024 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
4025 struct btrfs_root
*root
,
4026 struct inode
*inode
)
4030 ret
= btrfs_update_inode(trans
, root
, inode
);
4032 return btrfs_update_inode_item(trans
, root
, inode
);
4037 * unlink helper that gets used here in inode.c and in the tree logging
4038 * recovery code. It remove a link in a directory with a given name, and
4039 * also drops the back refs in the inode to the directory
4041 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4042 struct btrfs_root
*root
,
4043 struct btrfs_inode
*dir
,
4044 struct btrfs_inode
*inode
,
4045 const char *name
, int name_len
)
4047 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4048 struct btrfs_path
*path
;
4050 struct extent_buffer
*leaf
;
4051 struct btrfs_dir_item
*di
;
4052 struct btrfs_key key
;
4054 u64 ino
= btrfs_ino(inode
);
4055 u64 dir_ino
= btrfs_ino(dir
);
4057 path
= btrfs_alloc_path();
4063 path
->leave_spinning
= 1;
4064 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4065 name
, name_len
, -1);
4074 leaf
= path
->nodes
[0];
4075 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4076 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4079 btrfs_release_path(path
);
4082 * If we don't have dir index, we have to get it by looking up
4083 * the inode ref, since we get the inode ref, remove it directly,
4084 * it is unnecessary to do delayed deletion.
4086 * But if we have dir index, needn't search inode ref to get it.
4087 * Since the inode ref is close to the inode item, it is better
4088 * that we delay to delete it, and just do this deletion when
4089 * we update the inode item.
4091 if (inode
->dir_index
) {
4092 ret
= btrfs_delayed_delete_inode_ref(inode
);
4094 index
= inode
->dir_index
;
4099 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
4103 "failed to delete reference to %.*s, inode %llu parent %llu",
4104 name_len
, name
, ino
, dir_ino
);
4105 btrfs_abort_transaction(trans
, ret
);
4109 ret
= btrfs_delete_delayed_dir_index(trans
, fs_info
, dir
, index
);
4111 btrfs_abort_transaction(trans
, ret
);
4115 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
, inode
,
4117 if (ret
!= 0 && ret
!= -ENOENT
) {
4118 btrfs_abort_transaction(trans
, ret
);
4122 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
, dir
,
4127 btrfs_abort_transaction(trans
, ret
);
4129 btrfs_free_path(path
);
4133 btrfs_i_size_write(dir
, dir
->vfs_inode
.i_size
- name_len
* 2);
4134 inode_inc_iversion(&inode
->vfs_inode
);
4135 inode_inc_iversion(&dir
->vfs_inode
);
4136 inode
->vfs_inode
.i_ctime
= dir
->vfs_inode
.i_mtime
=
4137 dir
->vfs_inode
.i_ctime
= current_time(&inode
->vfs_inode
);
4138 ret
= btrfs_update_inode(trans
, root
, &dir
->vfs_inode
);
4143 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4144 struct btrfs_root
*root
,
4145 struct btrfs_inode
*dir
, struct btrfs_inode
*inode
,
4146 const char *name
, int name_len
)
4149 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
4151 drop_nlink(&inode
->vfs_inode
);
4152 ret
= btrfs_update_inode(trans
, root
, &inode
->vfs_inode
);
4158 * helper to start transaction for unlink and rmdir.
4160 * unlink and rmdir are special in btrfs, they do not always free space, so
4161 * if we cannot make our reservations the normal way try and see if there is
4162 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4163 * allow the unlink to occur.
4165 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
)
4167 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4170 * 1 for the possible orphan item
4171 * 1 for the dir item
4172 * 1 for the dir index
4173 * 1 for the inode ref
4176 return btrfs_start_transaction_fallback_global_rsv(root
, 5, 5);
4179 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
4181 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4182 struct btrfs_trans_handle
*trans
;
4183 struct inode
*inode
= d_inode(dentry
);
4186 trans
= __unlink_start_trans(dir
);
4188 return PTR_ERR(trans
);
4190 btrfs_record_unlink_dir(trans
, BTRFS_I(dir
), BTRFS_I(d_inode(dentry
)),
4193 ret
= btrfs_unlink_inode(trans
, root
, BTRFS_I(dir
),
4194 BTRFS_I(d_inode(dentry
)), dentry
->d_name
.name
,
4195 dentry
->d_name
.len
);
4199 if (inode
->i_nlink
== 0) {
4200 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
4206 btrfs_end_transaction(trans
);
4207 btrfs_btree_balance_dirty(root
->fs_info
);
4211 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
4212 struct btrfs_root
*root
,
4213 struct inode
*dir
, u64 objectid
,
4214 const char *name
, int name_len
)
4216 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4217 struct btrfs_path
*path
;
4218 struct extent_buffer
*leaf
;
4219 struct btrfs_dir_item
*di
;
4220 struct btrfs_key key
;
4223 u64 dir_ino
= btrfs_ino(BTRFS_I(dir
));
4225 path
= btrfs_alloc_path();
4229 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4230 name
, name_len
, -1);
4231 if (IS_ERR_OR_NULL(di
)) {
4239 leaf
= path
->nodes
[0];
4240 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4241 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
4242 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4244 btrfs_abort_transaction(trans
, ret
);
4247 btrfs_release_path(path
);
4249 ret
= btrfs_del_root_ref(trans
, fs_info
, objectid
,
4250 root
->root_key
.objectid
, dir_ino
,
4251 &index
, name
, name_len
);
4253 if (ret
!= -ENOENT
) {
4254 btrfs_abort_transaction(trans
, ret
);
4257 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
4259 if (IS_ERR_OR_NULL(di
)) {
4264 btrfs_abort_transaction(trans
, ret
);
4268 leaf
= path
->nodes
[0];
4269 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4270 btrfs_release_path(path
);
4273 btrfs_release_path(path
);
4275 ret
= btrfs_delete_delayed_dir_index(trans
, fs_info
, BTRFS_I(dir
), index
);
4277 btrfs_abort_transaction(trans
, ret
);
4281 btrfs_i_size_write(BTRFS_I(dir
), dir
->i_size
- name_len
* 2);
4282 inode_inc_iversion(dir
);
4283 dir
->i_mtime
= dir
->i_ctime
= current_time(dir
);
4284 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
4286 btrfs_abort_transaction(trans
, ret
);
4288 btrfs_free_path(path
);
4292 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
4294 struct inode
*inode
= d_inode(dentry
);
4296 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4297 struct btrfs_trans_handle
*trans
;
4298 u64 last_unlink_trans
;
4300 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
4302 if (btrfs_ino(BTRFS_I(inode
)) == BTRFS_FIRST_FREE_OBJECTID
)
4305 trans
= __unlink_start_trans(dir
);
4307 return PTR_ERR(trans
);
4309 if (unlikely(btrfs_ino(BTRFS_I(inode
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
4310 err
= btrfs_unlink_subvol(trans
, root
, dir
,
4311 BTRFS_I(inode
)->location
.objectid
,
4312 dentry
->d_name
.name
,
4313 dentry
->d_name
.len
);
4317 err
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
4321 last_unlink_trans
= BTRFS_I(inode
)->last_unlink_trans
;
4323 /* now the directory is empty */
4324 err
= btrfs_unlink_inode(trans
, root
, BTRFS_I(dir
),
4325 BTRFS_I(d_inode(dentry
)), dentry
->d_name
.name
,
4326 dentry
->d_name
.len
);
4328 btrfs_i_size_write(BTRFS_I(inode
), 0);
4330 * Propagate the last_unlink_trans value of the deleted dir to
4331 * its parent directory. This is to prevent an unrecoverable
4332 * log tree in the case we do something like this:
4334 * 2) create snapshot under dir foo
4335 * 3) delete the snapshot
4338 * 6) fsync foo or some file inside foo
4340 if (last_unlink_trans
>= trans
->transid
)
4341 BTRFS_I(dir
)->last_unlink_trans
= last_unlink_trans
;
4344 btrfs_end_transaction(trans
);
4345 btrfs_btree_balance_dirty(root
->fs_info
);
4350 static int truncate_space_check(struct btrfs_trans_handle
*trans
,
4351 struct btrfs_root
*root
,
4354 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4358 * This is only used to apply pressure to the enospc system, we don't
4359 * intend to use this reservation at all.
4361 bytes_deleted
= btrfs_csum_bytes_to_leaves(fs_info
, bytes_deleted
);
4362 bytes_deleted
*= fs_info
->nodesize
;
4363 ret
= btrfs_block_rsv_add(root
, &fs_info
->trans_block_rsv
,
4364 bytes_deleted
, BTRFS_RESERVE_NO_FLUSH
);
4366 trace_btrfs_space_reservation(fs_info
, "transaction",
4369 trans
->bytes_reserved
+= bytes_deleted
;
4375 static int truncate_inline_extent(struct inode
*inode
,
4376 struct btrfs_path
*path
,
4377 struct btrfs_key
*found_key
,
4381 struct extent_buffer
*leaf
= path
->nodes
[0];
4382 int slot
= path
->slots
[0];
4383 struct btrfs_file_extent_item
*fi
;
4384 u32 size
= (u32
)(new_size
- found_key
->offset
);
4385 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4387 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
4389 if (btrfs_file_extent_compression(leaf
, fi
) != BTRFS_COMPRESS_NONE
) {
4390 loff_t offset
= new_size
;
4391 loff_t page_end
= ALIGN(offset
, PAGE_SIZE
);
4394 * Zero out the remaining of the last page of our inline extent,
4395 * instead of directly truncating our inline extent here - that
4396 * would be much more complex (decompressing all the data, then
4397 * compressing the truncated data, which might be bigger than
4398 * the size of the inline extent, resize the extent, etc).
4399 * We release the path because to get the page we might need to
4400 * read the extent item from disk (data not in the page cache).
4402 btrfs_release_path(path
);
4403 return btrfs_truncate_block(inode
, offset
, page_end
- offset
,
4407 btrfs_set_file_extent_ram_bytes(leaf
, fi
, size
);
4408 size
= btrfs_file_extent_calc_inline_size(size
);
4409 btrfs_truncate_item(root
->fs_info
, path
, size
, 1);
4411 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4412 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4418 * this can truncate away extent items, csum items and directory items.
4419 * It starts at a high offset and removes keys until it can't find
4420 * any higher than new_size
4422 * csum items that cross the new i_size are truncated to the new size
4425 * min_type is the minimum key type to truncate down to. If set to 0, this
4426 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4428 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
4429 struct btrfs_root
*root
,
4430 struct inode
*inode
,
4431 u64 new_size
, u32 min_type
)
4433 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4434 struct btrfs_path
*path
;
4435 struct extent_buffer
*leaf
;
4436 struct btrfs_file_extent_item
*fi
;
4437 struct btrfs_key key
;
4438 struct btrfs_key found_key
;
4439 u64 extent_start
= 0;
4440 u64 extent_num_bytes
= 0;
4441 u64 extent_offset
= 0;
4443 u64 last_size
= new_size
;
4444 u32 found_type
= (u8
)-1;
4447 int pending_del_nr
= 0;
4448 int pending_del_slot
= 0;
4449 int extent_type
= -1;
4452 u64 ino
= btrfs_ino(BTRFS_I(inode
));
4453 u64 bytes_deleted
= 0;
4455 bool should_throttle
= 0;
4456 bool should_end
= 0;
4458 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
4461 * for non-free space inodes and ref cows, we want to back off from
4464 if (!btrfs_is_free_space_inode(BTRFS_I(inode
)) &&
4465 test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4468 path
= btrfs_alloc_path();
4471 path
->reada
= READA_BACK
;
4474 * We want to drop from the next block forward in case this new size is
4475 * not block aligned since we will be keeping the last block of the
4476 * extent just the way it is.
4478 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4479 root
== fs_info
->tree_root
)
4480 btrfs_drop_extent_cache(BTRFS_I(inode
), ALIGN(new_size
,
4481 fs_info
->sectorsize
),
4485 * This function is also used to drop the items in the log tree before
4486 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4487 * it is used to drop the loged items. So we shouldn't kill the delayed
4490 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
4491 btrfs_kill_delayed_inode_items(BTRFS_I(inode
));
4494 key
.offset
= (u64
)-1;
4499 * with a 16K leaf size and 128MB extents, you can actually queue
4500 * up a huge file in a single leaf. Most of the time that
4501 * bytes_deleted is > 0, it will be huge by the time we get here
4503 if (be_nice
&& bytes_deleted
> SZ_32M
) {
4504 if (btrfs_should_end_transaction(trans
)) {
4511 path
->leave_spinning
= 1;
4512 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
4519 /* there are no items in the tree for us to truncate, we're
4522 if (path
->slots
[0] == 0)
4529 leaf
= path
->nodes
[0];
4530 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4531 found_type
= found_key
.type
;
4533 if (found_key
.objectid
!= ino
)
4536 if (found_type
< min_type
)
4539 item_end
= found_key
.offset
;
4540 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
4541 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4542 struct btrfs_file_extent_item
);
4543 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4544 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4546 btrfs_file_extent_num_bytes(leaf
, fi
);
4548 trace_btrfs_truncate_show_fi_regular(
4549 BTRFS_I(inode
), leaf
, fi
,
4551 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4552 item_end
+= btrfs_file_extent_inline_len(leaf
,
4553 path
->slots
[0], fi
);
4555 trace_btrfs_truncate_show_fi_inline(
4556 BTRFS_I(inode
), leaf
, fi
, path
->slots
[0],
4561 if (found_type
> min_type
) {
4564 if (item_end
< new_size
)
4566 if (found_key
.offset
>= new_size
)
4572 /* FIXME, shrink the extent if the ref count is only 1 */
4573 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
4577 last_size
= found_key
.offset
;
4579 last_size
= new_size
;
4581 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4583 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4585 u64 orig_num_bytes
=
4586 btrfs_file_extent_num_bytes(leaf
, fi
);
4587 extent_num_bytes
= ALIGN(new_size
-
4589 fs_info
->sectorsize
);
4590 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4592 num_dec
= (orig_num_bytes
-
4594 if (test_bit(BTRFS_ROOT_REF_COWS
,
4597 inode_sub_bytes(inode
, num_dec
);
4598 btrfs_mark_buffer_dirty(leaf
);
4601 btrfs_file_extent_disk_num_bytes(leaf
,
4603 extent_offset
= found_key
.offset
-
4604 btrfs_file_extent_offset(leaf
, fi
);
4606 /* FIXME blocksize != 4096 */
4607 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
4608 if (extent_start
!= 0) {
4610 if (test_bit(BTRFS_ROOT_REF_COWS
,
4612 inode_sub_bytes(inode
, num_dec
);
4615 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4617 * we can't truncate inline items that have had
4621 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
4622 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
4625 * Need to release path in order to truncate a
4626 * compressed extent. So delete any accumulated
4627 * extent items so far.
4629 if (btrfs_file_extent_compression(leaf
, fi
) !=
4630 BTRFS_COMPRESS_NONE
&& pending_del_nr
) {
4631 err
= btrfs_del_items(trans
, root
, path
,
4635 btrfs_abort_transaction(trans
,
4642 err
= truncate_inline_extent(inode
, path
,
4647 btrfs_abort_transaction(trans
, err
);
4650 } else if (test_bit(BTRFS_ROOT_REF_COWS
,
4652 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4657 if (!pending_del_nr
) {
4658 /* no pending yet, add ourselves */
4659 pending_del_slot
= path
->slots
[0];
4661 } else if (pending_del_nr
&&
4662 path
->slots
[0] + 1 == pending_del_slot
) {
4663 /* hop on the pending chunk */
4665 pending_del_slot
= path
->slots
[0];
4672 should_throttle
= 0;
4675 (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4676 root
== fs_info
->tree_root
)) {
4677 btrfs_set_path_blocking(path
);
4678 bytes_deleted
+= extent_num_bytes
;
4679 ret
= btrfs_free_extent(trans
, fs_info
, extent_start
,
4680 extent_num_bytes
, 0,
4681 btrfs_header_owner(leaf
),
4682 ino
, extent_offset
);
4684 if (btrfs_should_throttle_delayed_refs(trans
, fs_info
))
4685 btrfs_async_run_delayed_refs(fs_info
,
4686 trans
->delayed_ref_updates
* 2,
4689 if (truncate_space_check(trans
, root
,
4690 extent_num_bytes
)) {
4693 if (btrfs_should_throttle_delayed_refs(trans
,
4695 should_throttle
= 1;
4699 if (found_type
== BTRFS_INODE_ITEM_KEY
)
4702 if (path
->slots
[0] == 0 ||
4703 path
->slots
[0] != pending_del_slot
||
4704 should_throttle
|| should_end
) {
4705 if (pending_del_nr
) {
4706 ret
= btrfs_del_items(trans
, root
, path
,
4710 btrfs_abort_transaction(trans
, ret
);
4715 btrfs_release_path(path
);
4716 if (should_throttle
) {
4717 unsigned long updates
= trans
->delayed_ref_updates
;
4719 trans
->delayed_ref_updates
= 0;
4720 ret
= btrfs_run_delayed_refs(trans
,
4728 * if we failed to refill our space rsv, bail out
4729 * and let the transaction restart
4741 if (pending_del_nr
) {
4742 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
4745 btrfs_abort_transaction(trans
, ret
);
4748 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4749 ASSERT(last_size
>= new_size
);
4750 if (!err
&& last_size
> new_size
)
4751 last_size
= new_size
;
4752 btrfs_ordered_update_i_size(inode
, last_size
, NULL
);
4755 btrfs_free_path(path
);
4757 if (be_nice
&& bytes_deleted
> SZ_32M
) {
4758 unsigned long updates
= trans
->delayed_ref_updates
;
4760 trans
->delayed_ref_updates
= 0;
4761 ret
= btrfs_run_delayed_refs(trans
, fs_info
,
4771 * btrfs_truncate_block - read, zero a chunk and write a block
4772 * @inode - inode that we're zeroing
4773 * @from - the offset to start zeroing
4774 * @len - the length to zero, 0 to zero the entire range respective to the
4776 * @front - zero up to the offset instead of from the offset on
4778 * This will find the block for the "from" offset and cow the block and zero the
4779 * part we want to zero. This is used with truncate and hole punching.
4781 int btrfs_truncate_block(struct inode
*inode
, loff_t from
, loff_t len
,
4784 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4785 struct address_space
*mapping
= inode
->i_mapping
;
4786 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4787 struct btrfs_ordered_extent
*ordered
;
4788 struct extent_state
*cached_state
= NULL
;
4789 struct extent_changeset
*data_reserved
= NULL
;
4791 u32 blocksize
= fs_info
->sectorsize
;
4792 pgoff_t index
= from
>> PAGE_SHIFT
;
4793 unsigned offset
= from
& (blocksize
- 1);
4795 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4800 if ((offset
& (blocksize
- 1)) == 0 &&
4801 (!len
|| ((len
& (blocksize
- 1)) == 0)))
4804 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
,
4805 round_down(from
, blocksize
), blocksize
);
4810 page
= find_or_create_page(mapping
, index
, mask
);
4812 btrfs_delalloc_release_space(inode
, data_reserved
,
4813 round_down(from
, blocksize
),
4819 block_start
= round_down(from
, blocksize
);
4820 block_end
= block_start
+ blocksize
- 1;
4822 if (!PageUptodate(page
)) {
4823 ret
= btrfs_readpage(NULL
, page
);
4825 if (page
->mapping
!= mapping
) {
4830 if (!PageUptodate(page
)) {
4835 wait_on_page_writeback(page
);
4837 lock_extent_bits(io_tree
, block_start
, block_end
, &cached_state
);
4838 set_page_extent_mapped(page
);
4840 ordered
= btrfs_lookup_ordered_extent(inode
, block_start
);
4842 unlock_extent_cached(io_tree
, block_start
, block_end
,
4843 &cached_state
, GFP_NOFS
);
4846 btrfs_start_ordered_extent(inode
, ordered
, 1);
4847 btrfs_put_ordered_extent(ordered
);
4851 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, block_start
, block_end
,
4852 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4853 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4854 0, 0, &cached_state
, GFP_NOFS
);
4856 ret
= btrfs_set_extent_delalloc(inode
, block_start
, block_end
,
4859 unlock_extent_cached(io_tree
, block_start
, block_end
,
4860 &cached_state
, GFP_NOFS
);
4864 if (offset
!= blocksize
) {
4866 len
= blocksize
- offset
;
4869 memset(kaddr
+ (block_start
- page_offset(page
)),
4872 memset(kaddr
+ (block_start
- page_offset(page
)) + offset
,
4874 flush_dcache_page(page
);
4877 ClearPageChecked(page
);
4878 set_page_dirty(page
);
4879 unlock_extent_cached(io_tree
, block_start
, block_end
, &cached_state
,
4884 btrfs_delalloc_release_space(inode
, data_reserved
, block_start
,
4889 extent_changeset_free(data_reserved
);
4893 static int maybe_insert_hole(struct btrfs_root
*root
, struct inode
*inode
,
4894 u64 offset
, u64 len
)
4896 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4897 struct btrfs_trans_handle
*trans
;
4901 * Still need to make sure the inode looks like it's been updated so
4902 * that any holes get logged if we fsync.
4904 if (btrfs_fs_incompat(fs_info
, NO_HOLES
)) {
4905 BTRFS_I(inode
)->last_trans
= fs_info
->generation
;
4906 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
4907 BTRFS_I(inode
)->last_log_commit
= root
->last_log_commit
;
4912 * 1 - for the one we're dropping
4913 * 1 - for the one we're adding
4914 * 1 - for updating the inode.
4916 trans
= btrfs_start_transaction(root
, 3);
4918 return PTR_ERR(trans
);
4920 ret
= btrfs_drop_extents(trans
, root
, inode
, offset
, offset
+ len
, 1);
4922 btrfs_abort_transaction(trans
, ret
);
4923 btrfs_end_transaction(trans
);
4927 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(BTRFS_I(inode
)),
4928 offset
, 0, 0, len
, 0, len
, 0, 0, 0);
4930 btrfs_abort_transaction(trans
, ret
);
4932 btrfs_update_inode(trans
, root
, inode
);
4933 btrfs_end_transaction(trans
);
4938 * This function puts in dummy file extents for the area we're creating a hole
4939 * for. So if we are truncating this file to a larger size we need to insert
4940 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4941 * the range between oldsize and size
4943 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
4945 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4946 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4947 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4948 struct extent_map
*em
= NULL
;
4949 struct extent_state
*cached_state
= NULL
;
4950 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4951 u64 hole_start
= ALIGN(oldsize
, fs_info
->sectorsize
);
4952 u64 block_end
= ALIGN(size
, fs_info
->sectorsize
);
4959 * If our size started in the middle of a block we need to zero out the
4960 * rest of the block before we expand the i_size, otherwise we could
4961 * expose stale data.
4963 err
= btrfs_truncate_block(inode
, oldsize
, 0, 0);
4967 if (size
<= hole_start
)
4971 struct btrfs_ordered_extent
*ordered
;
4973 lock_extent_bits(io_tree
, hole_start
, block_end
- 1,
4975 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), hole_start
,
4976 block_end
- hole_start
);
4979 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
4980 &cached_state
, GFP_NOFS
);
4981 btrfs_start_ordered_extent(inode
, ordered
, 1);
4982 btrfs_put_ordered_extent(ordered
);
4985 cur_offset
= hole_start
;
4987 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0, cur_offset
,
4988 block_end
- cur_offset
, 0);
4994 last_byte
= min(extent_map_end(em
), block_end
);
4995 last_byte
= ALIGN(last_byte
, fs_info
->sectorsize
);
4996 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
4997 struct extent_map
*hole_em
;
4998 hole_size
= last_byte
- cur_offset
;
5000 err
= maybe_insert_hole(root
, inode
, cur_offset
,
5004 btrfs_drop_extent_cache(BTRFS_I(inode
), cur_offset
,
5005 cur_offset
+ hole_size
- 1, 0);
5006 hole_em
= alloc_extent_map();
5008 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
5009 &BTRFS_I(inode
)->runtime_flags
);
5012 hole_em
->start
= cur_offset
;
5013 hole_em
->len
= hole_size
;
5014 hole_em
->orig_start
= cur_offset
;
5016 hole_em
->block_start
= EXTENT_MAP_HOLE
;
5017 hole_em
->block_len
= 0;
5018 hole_em
->orig_block_len
= 0;
5019 hole_em
->ram_bytes
= hole_size
;
5020 hole_em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
5021 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
5022 hole_em
->generation
= fs_info
->generation
;
5025 write_lock(&em_tree
->lock
);
5026 err
= add_extent_mapping(em_tree
, hole_em
, 1);
5027 write_unlock(&em_tree
->lock
);
5030 btrfs_drop_extent_cache(BTRFS_I(inode
),
5035 free_extent_map(hole_em
);
5038 free_extent_map(em
);
5040 cur_offset
= last_byte
;
5041 if (cur_offset
>= block_end
)
5044 free_extent_map(em
);
5045 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
5050 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
5052 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5053 struct btrfs_trans_handle
*trans
;
5054 loff_t oldsize
= i_size_read(inode
);
5055 loff_t newsize
= attr
->ia_size
;
5056 int mask
= attr
->ia_valid
;
5060 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5061 * special case where we need to update the times despite not having
5062 * these flags set. For all other operations the VFS set these flags
5063 * explicitly if it wants a timestamp update.
5065 if (newsize
!= oldsize
) {
5066 inode_inc_iversion(inode
);
5067 if (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))
5068 inode
->i_ctime
= inode
->i_mtime
=
5069 current_time(inode
);
5072 if (newsize
> oldsize
) {
5074 * Don't do an expanding truncate while snapshotting is ongoing.
5075 * This is to ensure the snapshot captures a fully consistent
5076 * state of this file - if the snapshot captures this expanding
5077 * truncation, it must capture all writes that happened before
5080 btrfs_wait_for_snapshot_creation(root
);
5081 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
5083 btrfs_end_write_no_snapshotting(root
);
5087 trans
= btrfs_start_transaction(root
, 1);
5088 if (IS_ERR(trans
)) {
5089 btrfs_end_write_no_snapshotting(root
);
5090 return PTR_ERR(trans
);
5093 i_size_write(inode
, newsize
);
5094 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
5095 pagecache_isize_extended(inode
, oldsize
, newsize
);
5096 ret
= btrfs_update_inode(trans
, root
, inode
);
5097 btrfs_end_write_no_snapshotting(root
);
5098 btrfs_end_transaction(trans
);
5102 * We're truncating a file that used to have good data down to
5103 * zero. Make sure it gets into the ordered flush list so that
5104 * any new writes get down to disk quickly.
5107 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
5108 &BTRFS_I(inode
)->runtime_flags
);
5111 * 1 for the orphan item we're going to add
5112 * 1 for the orphan item deletion.
5114 trans
= btrfs_start_transaction(root
, 2);
5116 return PTR_ERR(trans
);
5119 * We need to do this in case we fail at _any_ point during the
5120 * actual truncate. Once we do the truncate_setsize we could
5121 * invalidate pages which forces any outstanding ordered io to
5122 * be instantly completed which will give us extents that need
5123 * to be truncated. If we fail to get an orphan inode down we
5124 * could have left over extents that were never meant to live,
5125 * so we need to guarantee from this point on that everything
5126 * will be consistent.
5128 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
5129 btrfs_end_transaction(trans
);
5133 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5134 truncate_setsize(inode
, newsize
);
5136 /* Disable nonlocked read DIO to avoid the end less truncate */
5137 btrfs_inode_block_unlocked_dio(BTRFS_I(inode
));
5138 inode_dio_wait(inode
);
5139 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode
));
5141 ret
= btrfs_truncate(inode
);
5142 if (ret
&& inode
->i_nlink
) {
5145 /* To get a stable disk_i_size */
5146 err
= btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5148 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5153 * failed to truncate, disk_i_size is only adjusted down
5154 * as we remove extents, so it should represent the true
5155 * size of the inode, so reset the in memory size and
5156 * delete our orphan entry.
5158 trans
= btrfs_join_transaction(root
);
5159 if (IS_ERR(trans
)) {
5160 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5163 i_size_write(inode
, BTRFS_I(inode
)->disk_i_size
);
5164 err
= btrfs_orphan_del(trans
, BTRFS_I(inode
));
5166 btrfs_abort_transaction(trans
, err
);
5167 btrfs_end_transaction(trans
);
5174 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5176 struct inode
*inode
= d_inode(dentry
);
5177 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5180 if (btrfs_root_readonly(root
))
5183 err
= setattr_prepare(dentry
, attr
);
5187 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
5188 err
= btrfs_setsize(inode
, attr
);
5193 if (attr
->ia_valid
) {
5194 setattr_copy(inode
, attr
);
5195 inode_inc_iversion(inode
);
5196 err
= btrfs_dirty_inode(inode
);
5198 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
5199 err
= posix_acl_chmod(inode
, inode
->i_mode
);
5206 * While truncating the inode pages during eviction, we get the VFS calling
5207 * btrfs_invalidatepage() against each page of the inode. This is slow because
5208 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5209 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5210 * extent_state structures over and over, wasting lots of time.
5212 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5213 * those expensive operations on a per page basis and do only the ordered io
5214 * finishing, while we release here the extent_map and extent_state structures,
5215 * without the excessive merging and splitting.
5217 static void evict_inode_truncate_pages(struct inode
*inode
)
5219 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5220 struct extent_map_tree
*map_tree
= &BTRFS_I(inode
)->extent_tree
;
5221 struct rb_node
*node
;
5223 ASSERT(inode
->i_state
& I_FREEING
);
5224 truncate_inode_pages_final(&inode
->i_data
);
5226 write_lock(&map_tree
->lock
);
5227 while (!RB_EMPTY_ROOT(&map_tree
->map
)) {
5228 struct extent_map
*em
;
5230 node
= rb_first(&map_tree
->map
);
5231 em
= rb_entry(node
, struct extent_map
, rb_node
);
5232 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5233 clear_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
5234 remove_extent_mapping(map_tree
, em
);
5235 free_extent_map(em
);
5236 if (need_resched()) {
5237 write_unlock(&map_tree
->lock
);
5239 write_lock(&map_tree
->lock
);
5242 write_unlock(&map_tree
->lock
);
5245 * Keep looping until we have no more ranges in the io tree.
5246 * We can have ongoing bios started by readpages (called from readahead)
5247 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5248 * still in progress (unlocked the pages in the bio but did not yet
5249 * unlocked the ranges in the io tree). Therefore this means some
5250 * ranges can still be locked and eviction started because before
5251 * submitting those bios, which are executed by a separate task (work
5252 * queue kthread), inode references (inode->i_count) were not taken
5253 * (which would be dropped in the end io callback of each bio).
5254 * Therefore here we effectively end up waiting for those bios and
5255 * anyone else holding locked ranges without having bumped the inode's
5256 * reference count - if we don't do it, when they access the inode's
5257 * io_tree to unlock a range it may be too late, leading to an
5258 * use-after-free issue.
5260 spin_lock(&io_tree
->lock
);
5261 while (!RB_EMPTY_ROOT(&io_tree
->state
)) {
5262 struct extent_state
*state
;
5263 struct extent_state
*cached_state
= NULL
;
5267 node
= rb_first(&io_tree
->state
);
5268 state
= rb_entry(node
, struct extent_state
, rb_node
);
5269 start
= state
->start
;
5271 spin_unlock(&io_tree
->lock
);
5273 lock_extent_bits(io_tree
, start
, end
, &cached_state
);
5276 * If still has DELALLOC flag, the extent didn't reach disk,
5277 * and its reserved space won't be freed by delayed_ref.
5278 * So we need to free its reserved space here.
5279 * (Refer to comment in btrfs_invalidatepage, case 2)
5281 * Note, end is the bytenr of last byte, so we need + 1 here.
5283 if (state
->state
& EXTENT_DELALLOC
)
5284 btrfs_qgroup_free_data(inode
, NULL
, start
, end
- start
+ 1);
5286 clear_extent_bit(io_tree
, start
, end
,
5287 EXTENT_LOCKED
| EXTENT_DIRTY
|
5288 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
5289 EXTENT_DEFRAG
, 1, 1,
5290 &cached_state
, GFP_NOFS
);
5293 spin_lock(&io_tree
->lock
);
5295 spin_unlock(&io_tree
->lock
);
5298 void btrfs_evict_inode(struct inode
*inode
)
5300 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5301 struct btrfs_trans_handle
*trans
;
5302 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5303 struct btrfs_block_rsv
*rsv
, *global_rsv
;
5304 int steal_from_global
= 0;
5308 trace_btrfs_inode_evict(inode
);
5311 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
5315 min_size
= btrfs_calc_trunc_metadata_size(fs_info
, 1);
5317 evict_inode_truncate_pages(inode
);
5319 if (inode
->i_nlink
&&
5320 ((btrfs_root_refs(&root
->root_item
) != 0 &&
5321 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
) ||
5322 btrfs_is_free_space_inode(BTRFS_I(inode
))))
5325 if (is_bad_inode(inode
)) {
5326 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5329 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5330 if (!special_file(inode
->i_mode
))
5331 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5333 btrfs_free_io_failure_record(BTRFS_I(inode
), 0, (u64
)-1);
5335 if (test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
)) {
5336 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
5337 &BTRFS_I(inode
)->runtime_flags
));
5341 if (inode
->i_nlink
> 0) {
5342 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0 &&
5343 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
);
5347 ret
= btrfs_commit_inode_delayed_inode(BTRFS_I(inode
));
5349 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5353 rsv
= btrfs_alloc_block_rsv(fs_info
, BTRFS_BLOCK_RSV_TEMP
);
5355 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5358 rsv
->size
= min_size
;
5360 global_rsv
= &fs_info
->global_block_rsv
;
5362 btrfs_i_size_write(BTRFS_I(inode
), 0);
5365 * This is a bit simpler than btrfs_truncate since we've already
5366 * reserved our space for our orphan item in the unlink, so we just
5367 * need to reserve some slack space in case we add bytes and update
5368 * inode item when doing the truncate.
5371 ret
= btrfs_block_rsv_refill(root
, rsv
, min_size
,
5372 BTRFS_RESERVE_FLUSH_LIMIT
);
5375 * Try and steal from the global reserve since we will
5376 * likely not use this space anyway, we want to try as
5377 * hard as possible to get this to work.
5380 steal_from_global
++;
5382 steal_from_global
= 0;
5386 * steal_from_global == 0: we reserved stuff, hooray!
5387 * steal_from_global == 1: we didn't reserve stuff, boo!
5388 * steal_from_global == 2: we've committed, still not a lot of
5389 * room but maybe we'll have room in the global reserve this
5391 * steal_from_global == 3: abandon all hope!
5393 if (steal_from_global
> 2) {
5395 "Could not get space for a delete, will truncate on mount %d",
5397 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5398 btrfs_free_block_rsv(fs_info
, rsv
);
5402 trans
= btrfs_join_transaction(root
);
5403 if (IS_ERR(trans
)) {
5404 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5405 btrfs_free_block_rsv(fs_info
, rsv
);
5410 * We can't just steal from the global reserve, we need to make
5411 * sure there is room to do it, if not we need to commit and try
5414 if (steal_from_global
) {
5415 if (!btrfs_check_space_for_delayed_refs(trans
, fs_info
))
5416 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
,
5423 * Couldn't steal from the global reserve, we have too much
5424 * pending stuff built up, commit the transaction and try it
5428 ret
= btrfs_commit_transaction(trans
);
5430 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5431 btrfs_free_block_rsv(fs_info
, rsv
);
5436 steal_from_global
= 0;
5439 trans
->block_rsv
= rsv
;
5441 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
5442 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
)
5445 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
5446 btrfs_end_transaction(trans
);
5448 btrfs_btree_balance_dirty(fs_info
);
5451 btrfs_free_block_rsv(fs_info
, rsv
);
5454 * Errors here aren't a big deal, it just means we leave orphan items
5455 * in the tree. They will be cleaned up on the next mount.
5458 trans
->block_rsv
= root
->orphan_block_rsv
;
5459 btrfs_orphan_del(trans
, BTRFS_I(inode
));
5461 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5464 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
5465 if (!(root
== fs_info
->tree_root
||
5466 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
5467 btrfs_return_ino(root
, btrfs_ino(BTRFS_I(inode
)));
5469 btrfs_end_transaction(trans
);
5470 btrfs_btree_balance_dirty(fs_info
);
5472 btrfs_remove_delayed_node(BTRFS_I(inode
));
5477 * this returns the key found in the dir entry in the location pointer.
5478 * If no dir entries were found, location->objectid is 0.
5480 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
5481 struct btrfs_key
*location
)
5483 const char *name
= dentry
->d_name
.name
;
5484 int namelen
= dentry
->d_name
.len
;
5485 struct btrfs_dir_item
*di
;
5486 struct btrfs_path
*path
;
5487 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5490 path
= btrfs_alloc_path();
5494 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(BTRFS_I(dir
)),
5499 if (IS_ERR_OR_NULL(di
))
5502 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
5504 btrfs_free_path(path
);
5507 location
->objectid
= 0;
5512 * when we hit a tree root in a directory, the btrfs part of the inode
5513 * needs to be changed to reflect the root directory of the tree root. This
5514 * is kind of like crossing a mount point.
5516 static int fixup_tree_root_location(struct btrfs_fs_info
*fs_info
,
5518 struct dentry
*dentry
,
5519 struct btrfs_key
*location
,
5520 struct btrfs_root
**sub_root
)
5522 struct btrfs_path
*path
;
5523 struct btrfs_root
*new_root
;
5524 struct btrfs_root_ref
*ref
;
5525 struct extent_buffer
*leaf
;
5526 struct btrfs_key key
;
5530 path
= btrfs_alloc_path();
5537 key
.objectid
= BTRFS_I(dir
)->root
->root_key
.objectid
;
5538 key
.type
= BTRFS_ROOT_REF_KEY
;
5539 key
.offset
= location
->objectid
;
5541 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
5548 leaf
= path
->nodes
[0];
5549 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
5550 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(BTRFS_I(dir
)) ||
5551 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
5554 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
5555 (unsigned long)(ref
+ 1),
5556 dentry
->d_name
.len
);
5560 btrfs_release_path(path
);
5562 new_root
= btrfs_read_fs_root_no_name(fs_info
, location
);
5563 if (IS_ERR(new_root
)) {
5564 err
= PTR_ERR(new_root
);
5568 *sub_root
= new_root
;
5569 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
5570 location
->type
= BTRFS_INODE_ITEM_KEY
;
5571 location
->offset
= 0;
5574 btrfs_free_path(path
);
5578 static void inode_tree_add(struct inode
*inode
)
5580 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5581 struct btrfs_inode
*entry
;
5583 struct rb_node
*parent
;
5584 struct rb_node
*new = &BTRFS_I(inode
)->rb_node
;
5585 u64 ino
= btrfs_ino(BTRFS_I(inode
));
5587 if (inode_unhashed(inode
))
5590 spin_lock(&root
->inode_lock
);
5591 p
= &root
->inode_tree
.rb_node
;
5594 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
5596 if (ino
< btrfs_ino(BTRFS_I(&entry
->vfs_inode
)))
5597 p
= &parent
->rb_left
;
5598 else if (ino
> btrfs_ino(BTRFS_I(&entry
->vfs_inode
)))
5599 p
= &parent
->rb_right
;
5601 WARN_ON(!(entry
->vfs_inode
.i_state
&
5602 (I_WILL_FREE
| I_FREEING
)));
5603 rb_replace_node(parent
, new, &root
->inode_tree
);
5604 RB_CLEAR_NODE(parent
);
5605 spin_unlock(&root
->inode_lock
);
5609 rb_link_node(new, parent
, p
);
5610 rb_insert_color(new, &root
->inode_tree
);
5611 spin_unlock(&root
->inode_lock
);
5614 static void inode_tree_del(struct inode
*inode
)
5616 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5617 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5620 spin_lock(&root
->inode_lock
);
5621 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
5622 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
5623 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
5624 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5626 spin_unlock(&root
->inode_lock
);
5628 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0) {
5629 synchronize_srcu(&fs_info
->subvol_srcu
);
5630 spin_lock(&root
->inode_lock
);
5631 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5632 spin_unlock(&root
->inode_lock
);
5634 btrfs_add_dead_root(root
);
5638 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
5640 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5641 struct rb_node
*node
;
5642 struct rb_node
*prev
;
5643 struct btrfs_inode
*entry
;
5644 struct inode
*inode
;
5647 if (!test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
5648 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
5650 spin_lock(&root
->inode_lock
);
5652 node
= root
->inode_tree
.rb_node
;
5656 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5658 if (objectid
< btrfs_ino(BTRFS_I(&entry
->vfs_inode
)))
5659 node
= node
->rb_left
;
5660 else if (objectid
> btrfs_ino(BTRFS_I(&entry
->vfs_inode
)))
5661 node
= node
->rb_right
;
5667 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
5668 if (objectid
<= btrfs_ino(BTRFS_I(&entry
->vfs_inode
))) {
5672 prev
= rb_next(prev
);
5676 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5677 objectid
= btrfs_ino(BTRFS_I(&entry
->vfs_inode
)) + 1;
5678 inode
= igrab(&entry
->vfs_inode
);
5680 spin_unlock(&root
->inode_lock
);
5681 if (atomic_read(&inode
->i_count
) > 1)
5682 d_prune_aliases(inode
);
5684 * btrfs_drop_inode will have it removed from
5685 * the inode cache when its usage count
5690 spin_lock(&root
->inode_lock
);
5694 if (cond_resched_lock(&root
->inode_lock
))
5697 node
= rb_next(node
);
5699 spin_unlock(&root
->inode_lock
);
5702 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
5704 struct btrfs_iget_args
*args
= p
;
5705 inode
->i_ino
= args
->location
->objectid
;
5706 memcpy(&BTRFS_I(inode
)->location
, args
->location
,
5707 sizeof(*args
->location
));
5708 BTRFS_I(inode
)->root
= args
->root
;
5712 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
5714 struct btrfs_iget_args
*args
= opaque
;
5715 return args
->location
->objectid
== BTRFS_I(inode
)->location
.objectid
&&
5716 args
->root
== BTRFS_I(inode
)->root
;
5719 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
5720 struct btrfs_key
*location
,
5721 struct btrfs_root
*root
)
5723 struct inode
*inode
;
5724 struct btrfs_iget_args args
;
5725 unsigned long hashval
= btrfs_inode_hash(location
->objectid
, root
);
5727 args
.location
= location
;
5730 inode
= iget5_locked(s
, hashval
, btrfs_find_actor
,
5731 btrfs_init_locked_inode
,
5736 /* Get an inode object given its location and corresponding root.
5737 * Returns in *is_new if the inode was read from disk
5739 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
5740 struct btrfs_root
*root
, int *new)
5742 struct inode
*inode
;
5744 inode
= btrfs_iget_locked(s
, location
, root
);
5746 return ERR_PTR(-ENOMEM
);
5748 if (inode
->i_state
& I_NEW
) {
5751 ret
= btrfs_read_locked_inode(inode
);
5752 if (!is_bad_inode(inode
)) {
5753 inode_tree_add(inode
);
5754 unlock_new_inode(inode
);
5758 unlock_new_inode(inode
);
5761 inode
= ERR_PTR(ret
< 0 ? ret
: -ESTALE
);
5768 static struct inode
*new_simple_dir(struct super_block
*s
,
5769 struct btrfs_key
*key
,
5770 struct btrfs_root
*root
)
5772 struct inode
*inode
= new_inode(s
);
5775 return ERR_PTR(-ENOMEM
);
5777 BTRFS_I(inode
)->root
= root
;
5778 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
5779 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
5781 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
5782 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
5783 inode
->i_opflags
&= ~IOP_XATTR
;
5784 inode
->i_fop
= &simple_dir_operations
;
5785 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
5786 inode
->i_mtime
= current_time(inode
);
5787 inode
->i_atime
= inode
->i_mtime
;
5788 inode
->i_ctime
= inode
->i_mtime
;
5789 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
5794 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
5796 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
5797 struct inode
*inode
;
5798 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5799 struct btrfs_root
*sub_root
= root
;
5800 struct btrfs_key location
;
5804 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
5805 return ERR_PTR(-ENAMETOOLONG
);
5807 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
5809 return ERR_PTR(ret
);
5811 if (location
.objectid
== 0)
5812 return ERR_PTR(-ENOENT
);
5814 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
5815 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
5819 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
5821 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
5822 ret
= fixup_tree_root_location(fs_info
, dir
, dentry
,
5823 &location
, &sub_root
);
5826 inode
= ERR_PTR(ret
);
5828 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
5830 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
5832 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5834 if (!IS_ERR(inode
) && root
!= sub_root
) {
5835 down_read(&fs_info
->cleanup_work_sem
);
5836 if (!sb_rdonly(inode
->i_sb
))
5837 ret
= btrfs_orphan_cleanup(sub_root
);
5838 up_read(&fs_info
->cleanup_work_sem
);
5841 inode
= ERR_PTR(ret
);
5848 static int btrfs_dentry_delete(const struct dentry
*dentry
)
5850 struct btrfs_root
*root
;
5851 struct inode
*inode
= d_inode(dentry
);
5853 if (!inode
&& !IS_ROOT(dentry
))
5854 inode
= d_inode(dentry
->d_parent
);
5857 root
= BTRFS_I(inode
)->root
;
5858 if (btrfs_root_refs(&root
->root_item
) == 0)
5861 if (btrfs_ino(BTRFS_I(inode
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5867 static void btrfs_dentry_release(struct dentry
*dentry
)
5869 kfree(dentry
->d_fsdata
);
5872 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5875 struct inode
*inode
;
5877 inode
= btrfs_lookup_dentry(dir
, dentry
);
5878 if (IS_ERR(inode
)) {
5879 if (PTR_ERR(inode
) == -ENOENT
)
5882 return ERR_CAST(inode
);
5885 return d_splice_alias(inode
, dentry
);
5888 unsigned char btrfs_filetype_table
[] = {
5889 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
5893 * All this infrastructure exists because dir_emit can fault, and we are holding
5894 * the tree lock when doing readdir. For now just allocate a buffer and copy
5895 * our information into that, and then dir_emit from the buffer. This is
5896 * similar to what NFS does, only we don't keep the buffer around in pagecache
5897 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5898 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5901 static int btrfs_opendir(struct inode
*inode
, struct file
*file
)
5903 struct btrfs_file_private
*private;
5905 private = kzalloc(sizeof(struct btrfs_file_private
), GFP_KERNEL
);
5908 private->filldir_buf
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
5909 if (!private->filldir_buf
) {
5913 file
->private_data
= private;
5924 static int btrfs_filldir(void *addr
, int entries
, struct dir_context
*ctx
)
5927 struct dir_entry
*entry
= addr
;
5928 char *name
= (char *)(entry
+ 1);
5930 ctx
->pos
= entry
->offset
;
5931 if (!dir_emit(ctx
, name
, entry
->name_len
, entry
->ino
,
5934 addr
+= sizeof(struct dir_entry
) + entry
->name_len
;
5940 static int btrfs_real_readdir(struct file
*file
, struct dir_context
*ctx
)
5942 struct inode
*inode
= file_inode(file
);
5943 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5944 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5945 struct btrfs_file_private
*private = file
->private_data
;
5946 struct btrfs_dir_item
*di
;
5947 struct btrfs_key key
;
5948 struct btrfs_key found_key
;
5949 struct btrfs_path
*path
;
5951 struct list_head ins_list
;
5952 struct list_head del_list
;
5954 struct extent_buffer
*leaf
;
5961 struct btrfs_key location
;
5963 if (!dir_emit_dots(file
, ctx
))
5966 path
= btrfs_alloc_path();
5970 addr
= private->filldir_buf
;
5971 path
->reada
= READA_FORWARD
;
5973 INIT_LIST_HEAD(&ins_list
);
5974 INIT_LIST_HEAD(&del_list
);
5975 put
= btrfs_readdir_get_delayed_items(inode
, &ins_list
, &del_list
);
5978 key
.type
= BTRFS_DIR_INDEX_KEY
;
5979 key
.offset
= ctx
->pos
;
5980 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
5982 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5987 struct dir_entry
*entry
;
5989 leaf
= path
->nodes
[0];
5990 slot
= path
->slots
[0];
5991 if (slot
>= btrfs_header_nritems(leaf
)) {
5992 ret
= btrfs_next_leaf(root
, path
);
6000 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6002 if (found_key
.objectid
!= key
.objectid
)
6004 if (found_key
.type
!= BTRFS_DIR_INDEX_KEY
)
6006 if (found_key
.offset
< ctx
->pos
)
6008 if (btrfs_should_delete_dir_index(&del_list
, found_key
.offset
))
6010 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
6011 if (verify_dir_item(fs_info
, leaf
, slot
, di
))
6014 name_len
= btrfs_dir_name_len(leaf
, di
);
6015 if ((total_len
+ sizeof(struct dir_entry
) + name_len
) >=
6017 btrfs_release_path(path
);
6018 ret
= btrfs_filldir(private->filldir_buf
, entries
, ctx
);
6021 addr
= private->filldir_buf
;
6028 entry
->name_len
= name_len
;
6029 name_ptr
= (char *)(entry
+ 1);
6030 read_extent_buffer(leaf
, name_ptr
, (unsigned long)(di
+ 1),
6032 entry
->type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
6033 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
6034 entry
->ino
= location
.objectid
;
6035 entry
->offset
= found_key
.offset
;
6037 addr
+= sizeof(struct dir_entry
) + name_len
;
6038 total_len
+= sizeof(struct dir_entry
) + name_len
;
6042 btrfs_release_path(path
);
6044 ret
= btrfs_filldir(private->filldir_buf
, entries
, ctx
);
6048 ret
= btrfs_readdir_delayed_dir_index(ctx
, &ins_list
);
6053 * Stop new entries from being returned after we return the last
6056 * New directory entries are assigned a strictly increasing
6057 * offset. This means that new entries created during readdir
6058 * are *guaranteed* to be seen in the future by that readdir.
6059 * This has broken buggy programs which operate on names as
6060 * they're returned by readdir. Until we re-use freed offsets
6061 * we have this hack to stop new entries from being returned
6062 * under the assumption that they'll never reach this huge
6065 * This is being careful not to overflow 32bit loff_t unless the
6066 * last entry requires it because doing so has broken 32bit apps
6069 if (ctx
->pos
>= INT_MAX
)
6070 ctx
->pos
= LLONG_MAX
;
6077 btrfs_readdir_put_delayed_items(inode
, &ins_list
, &del_list
);
6078 btrfs_free_path(path
);
6082 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
6084 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6085 struct btrfs_trans_handle
*trans
;
6087 bool nolock
= false;
6089 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
6092 if (btrfs_fs_closing(root
->fs_info
) &&
6093 btrfs_is_free_space_inode(BTRFS_I(inode
)))
6096 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
6098 trans
= btrfs_join_transaction_nolock(root
);
6100 trans
= btrfs_join_transaction(root
);
6102 return PTR_ERR(trans
);
6103 ret
= btrfs_commit_transaction(trans
);
6109 * This is somewhat expensive, updating the tree every time the
6110 * inode changes. But, it is most likely to find the inode in cache.
6111 * FIXME, needs more benchmarking...there are no reasons other than performance
6112 * to keep or drop this code.
6114 static int btrfs_dirty_inode(struct inode
*inode
)
6116 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
6117 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6118 struct btrfs_trans_handle
*trans
;
6121 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
6124 trans
= btrfs_join_transaction(root
);
6126 return PTR_ERR(trans
);
6128 ret
= btrfs_update_inode(trans
, root
, inode
);
6129 if (ret
&& ret
== -ENOSPC
) {
6130 /* whoops, lets try again with the full transaction */
6131 btrfs_end_transaction(trans
);
6132 trans
= btrfs_start_transaction(root
, 1);
6134 return PTR_ERR(trans
);
6136 ret
= btrfs_update_inode(trans
, root
, inode
);
6138 btrfs_end_transaction(trans
);
6139 if (BTRFS_I(inode
)->delayed_node
)
6140 btrfs_balance_delayed_items(fs_info
);
6146 * This is a copy of file_update_time. We need this so we can return error on
6147 * ENOSPC for updating the inode in the case of file write and mmap writes.
6149 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
6152 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6154 if (btrfs_root_readonly(root
))
6157 if (flags
& S_VERSION
)
6158 inode_inc_iversion(inode
);
6159 if (flags
& S_CTIME
)
6160 inode
->i_ctime
= *now
;
6161 if (flags
& S_MTIME
)
6162 inode
->i_mtime
= *now
;
6163 if (flags
& S_ATIME
)
6164 inode
->i_atime
= *now
;
6165 return btrfs_dirty_inode(inode
);
6169 * find the highest existing sequence number in a directory
6170 * and then set the in-memory index_cnt variable to reflect
6171 * free sequence numbers
6173 static int btrfs_set_inode_index_count(struct btrfs_inode
*inode
)
6175 struct btrfs_root
*root
= inode
->root
;
6176 struct btrfs_key key
, found_key
;
6177 struct btrfs_path
*path
;
6178 struct extent_buffer
*leaf
;
6181 key
.objectid
= btrfs_ino(inode
);
6182 key
.type
= BTRFS_DIR_INDEX_KEY
;
6183 key
.offset
= (u64
)-1;
6185 path
= btrfs_alloc_path();
6189 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6192 /* FIXME: we should be able to handle this */
6198 * MAGIC NUMBER EXPLANATION:
6199 * since we search a directory based on f_pos we have to start at 2
6200 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6201 * else has to start at 2
6203 if (path
->slots
[0] == 0) {
6204 inode
->index_cnt
= 2;
6210 leaf
= path
->nodes
[0];
6211 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6213 if (found_key
.objectid
!= btrfs_ino(inode
) ||
6214 found_key
.type
!= BTRFS_DIR_INDEX_KEY
) {
6215 inode
->index_cnt
= 2;
6219 inode
->index_cnt
= found_key
.offset
+ 1;
6221 btrfs_free_path(path
);
6226 * helper to find a free sequence number in a given directory. This current
6227 * code is very simple, later versions will do smarter things in the btree
6229 int btrfs_set_inode_index(struct btrfs_inode
*dir
, u64
*index
)
6233 if (dir
->index_cnt
== (u64
)-1) {
6234 ret
= btrfs_inode_delayed_dir_index_count(dir
);
6236 ret
= btrfs_set_inode_index_count(dir
);
6242 *index
= dir
->index_cnt
;
6248 static int btrfs_insert_inode_locked(struct inode
*inode
)
6250 struct btrfs_iget_args args
;
6251 args
.location
= &BTRFS_I(inode
)->location
;
6252 args
.root
= BTRFS_I(inode
)->root
;
6254 return insert_inode_locked4(inode
,
6255 btrfs_inode_hash(inode
->i_ino
, BTRFS_I(inode
)->root
),
6256 btrfs_find_actor
, &args
);
6260 * Inherit flags from the parent inode.
6262 * Currently only the compression flags and the cow flags are inherited.
6264 static void btrfs_inherit_iflags(struct inode
*inode
, struct inode
*dir
)
6271 flags
= BTRFS_I(dir
)->flags
;
6273 if (flags
& BTRFS_INODE_NOCOMPRESS
) {
6274 BTRFS_I(inode
)->flags
&= ~BTRFS_INODE_COMPRESS
;
6275 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
6276 } else if (flags
& BTRFS_INODE_COMPRESS
) {
6277 BTRFS_I(inode
)->flags
&= ~BTRFS_INODE_NOCOMPRESS
;
6278 BTRFS_I(inode
)->flags
|= BTRFS_INODE_COMPRESS
;
6281 if (flags
& BTRFS_INODE_NODATACOW
) {
6282 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
6283 if (S_ISREG(inode
->i_mode
))
6284 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6287 btrfs_update_iflags(inode
);
6290 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
6291 struct btrfs_root
*root
,
6293 const char *name
, int name_len
,
6294 u64 ref_objectid
, u64 objectid
,
6295 umode_t mode
, u64
*index
)
6297 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6298 struct inode
*inode
;
6299 struct btrfs_inode_item
*inode_item
;
6300 struct btrfs_key
*location
;
6301 struct btrfs_path
*path
;
6302 struct btrfs_inode_ref
*ref
;
6303 struct btrfs_key key
[2];
6305 int nitems
= name
? 2 : 1;
6309 path
= btrfs_alloc_path();
6311 return ERR_PTR(-ENOMEM
);
6313 inode
= new_inode(fs_info
->sb
);
6315 btrfs_free_path(path
);
6316 return ERR_PTR(-ENOMEM
);
6320 * O_TMPFILE, set link count to 0, so that after this point,
6321 * we fill in an inode item with the correct link count.
6324 set_nlink(inode
, 0);
6327 * we have to initialize this early, so we can reclaim the inode
6328 * number if we fail afterwards in this function.
6330 inode
->i_ino
= objectid
;
6333 trace_btrfs_inode_request(dir
);
6335 ret
= btrfs_set_inode_index(BTRFS_I(dir
), index
);
6337 btrfs_free_path(path
);
6339 return ERR_PTR(ret
);
6345 * index_cnt is ignored for everything but a dir,
6346 * btrfs_get_inode_index_count has an explanation for the magic
6349 BTRFS_I(inode
)->index_cnt
= 2;
6350 BTRFS_I(inode
)->dir_index
= *index
;
6351 BTRFS_I(inode
)->root
= root
;
6352 BTRFS_I(inode
)->generation
= trans
->transid
;
6353 inode
->i_generation
= BTRFS_I(inode
)->generation
;
6356 * We could have gotten an inode number from somebody who was fsynced
6357 * and then removed in this same transaction, so let's just set full
6358 * sync since it will be a full sync anyway and this will blow away the
6359 * old info in the log.
6361 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
6363 key
[0].objectid
= objectid
;
6364 key
[0].type
= BTRFS_INODE_ITEM_KEY
;
6367 sizes
[0] = sizeof(struct btrfs_inode_item
);
6371 * Start new inodes with an inode_ref. This is slightly more
6372 * efficient for small numbers of hard links since they will
6373 * be packed into one item. Extended refs will kick in if we
6374 * add more hard links than can fit in the ref item.
6376 key
[1].objectid
= objectid
;
6377 key
[1].type
= BTRFS_INODE_REF_KEY
;
6378 key
[1].offset
= ref_objectid
;
6380 sizes
[1] = name_len
+ sizeof(*ref
);
6383 location
= &BTRFS_I(inode
)->location
;
6384 location
->objectid
= objectid
;
6385 location
->offset
= 0;
6386 location
->type
= BTRFS_INODE_ITEM_KEY
;
6388 ret
= btrfs_insert_inode_locked(inode
);
6392 path
->leave_spinning
= 1;
6393 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, nitems
);
6397 inode_init_owner(inode
, dir
, mode
);
6398 inode_set_bytes(inode
, 0);
6400 inode
->i_mtime
= current_time(inode
);
6401 inode
->i_atime
= inode
->i_mtime
;
6402 inode
->i_ctime
= inode
->i_mtime
;
6403 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
6405 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
6406 struct btrfs_inode_item
);
6407 memzero_extent_buffer(path
->nodes
[0], (unsigned long)inode_item
,
6408 sizeof(*inode_item
));
6409 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
6412 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
6413 struct btrfs_inode_ref
);
6414 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
6415 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
6416 ptr
= (unsigned long)(ref
+ 1);
6417 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
6420 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6421 btrfs_free_path(path
);
6423 btrfs_inherit_iflags(inode
, dir
);
6425 if (S_ISREG(mode
)) {
6426 if (btrfs_test_opt(fs_info
, NODATASUM
))
6427 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6428 if (btrfs_test_opt(fs_info
, NODATACOW
))
6429 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
6430 BTRFS_INODE_NODATASUM
;
6433 inode_tree_add(inode
);
6435 trace_btrfs_inode_new(inode
);
6436 btrfs_set_inode_last_trans(trans
, inode
);
6438 btrfs_update_root_times(trans
, root
);
6440 ret
= btrfs_inode_inherit_props(trans
, inode
, dir
);
6443 "error inheriting props for ino %llu (root %llu): %d",
6444 btrfs_ino(BTRFS_I(inode
)), root
->root_key
.objectid
, ret
);
6449 unlock_new_inode(inode
);
6452 BTRFS_I(dir
)->index_cnt
--;
6453 btrfs_free_path(path
);
6455 return ERR_PTR(ret
);
6458 static inline u8
btrfs_inode_type(struct inode
*inode
)
6460 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
6464 * utility function to add 'inode' into 'parent_inode' with
6465 * a give name and a given sequence number.
6466 * if 'add_backref' is true, also insert a backref from the
6467 * inode to the parent directory.
6469 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
6470 struct btrfs_inode
*parent_inode
, struct btrfs_inode
*inode
,
6471 const char *name
, int name_len
, int add_backref
, u64 index
)
6473 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6475 struct btrfs_key key
;
6476 struct btrfs_root
*root
= parent_inode
->root
;
6477 u64 ino
= btrfs_ino(inode
);
6478 u64 parent_ino
= btrfs_ino(parent_inode
);
6480 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6481 memcpy(&key
, &inode
->root
->root_key
, sizeof(key
));
6484 key
.type
= BTRFS_INODE_ITEM_KEY
;
6488 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6489 ret
= btrfs_add_root_ref(trans
, fs_info
, key
.objectid
,
6490 root
->root_key
.objectid
, parent_ino
,
6491 index
, name
, name_len
);
6492 } else if (add_backref
) {
6493 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
6497 /* Nothing to clean up yet */
6501 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
6503 btrfs_inode_type(&inode
->vfs_inode
), index
);
6504 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
6507 btrfs_abort_transaction(trans
, ret
);
6511 btrfs_i_size_write(parent_inode
, parent_inode
->vfs_inode
.i_size
+
6513 inode_inc_iversion(&parent_inode
->vfs_inode
);
6514 parent_inode
->vfs_inode
.i_mtime
= parent_inode
->vfs_inode
.i_ctime
=
6515 current_time(&parent_inode
->vfs_inode
);
6516 ret
= btrfs_update_inode(trans
, root
, &parent_inode
->vfs_inode
);
6518 btrfs_abort_transaction(trans
, ret
);
6522 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6525 err
= btrfs_del_root_ref(trans
, fs_info
, key
.objectid
,
6526 root
->root_key
.objectid
, parent_ino
,
6527 &local_index
, name
, name_len
);
6529 } else if (add_backref
) {
6533 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
6534 ino
, parent_ino
, &local_index
);
6539 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
6540 struct btrfs_inode
*dir
, struct dentry
*dentry
,
6541 struct btrfs_inode
*inode
, int backref
, u64 index
)
6543 int err
= btrfs_add_link(trans
, dir
, inode
,
6544 dentry
->d_name
.name
, dentry
->d_name
.len
,
6551 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
6552 umode_t mode
, dev_t rdev
)
6554 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6555 struct btrfs_trans_handle
*trans
;
6556 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6557 struct inode
*inode
= NULL
;
6564 * 2 for inode item and ref
6566 * 1 for xattr if selinux is on
6568 trans
= btrfs_start_transaction(root
, 5);
6570 return PTR_ERR(trans
);
6572 err
= btrfs_find_free_ino(root
, &objectid
);
6576 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6577 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)), objectid
,
6579 if (IS_ERR(inode
)) {
6580 err
= PTR_ERR(inode
);
6585 * If the active LSM wants to access the inode during
6586 * d_instantiate it needs these. Smack checks to see
6587 * if the filesystem supports xattrs by looking at the
6590 inode
->i_op
= &btrfs_special_inode_operations
;
6591 init_special_inode(inode
, inode
->i_mode
, rdev
);
6593 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6595 goto out_unlock_inode
;
6597 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
, BTRFS_I(inode
),
6600 goto out_unlock_inode
;
6602 btrfs_update_inode(trans
, root
, inode
);
6603 unlock_new_inode(inode
);
6604 d_instantiate(dentry
, inode
);
6608 btrfs_end_transaction(trans
);
6609 btrfs_balance_delayed_items(fs_info
);
6610 btrfs_btree_balance_dirty(fs_info
);
6612 inode_dec_link_count(inode
);
6619 unlock_new_inode(inode
);
6624 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
6625 umode_t mode
, bool excl
)
6627 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6628 struct btrfs_trans_handle
*trans
;
6629 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6630 struct inode
*inode
= NULL
;
6631 int drop_inode_on_err
= 0;
6637 * 2 for inode item and ref
6639 * 1 for xattr if selinux is on
6641 trans
= btrfs_start_transaction(root
, 5);
6643 return PTR_ERR(trans
);
6645 err
= btrfs_find_free_ino(root
, &objectid
);
6649 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6650 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)), objectid
,
6652 if (IS_ERR(inode
)) {
6653 err
= PTR_ERR(inode
);
6656 drop_inode_on_err
= 1;
6658 * If the active LSM wants to access the inode during
6659 * d_instantiate it needs these. Smack checks to see
6660 * if the filesystem supports xattrs by looking at the
6663 inode
->i_fop
= &btrfs_file_operations
;
6664 inode
->i_op
= &btrfs_file_inode_operations
;
6665 inode
->i_mapping
->a_ops
= &btrfs_aops
;
6667 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6669 goto out_unlock_inode
;
6671 err
= btrfs_update_inode(trans
, root
, inode
);
6673 goto out_unlock_inode
;
6675 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
, BTRFS_I(inode
),
6678 goto out_unlock_inode
;
6680 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
6681 unlock_new_inode(inode
);
6682 d_instantiate(dentry
, inode
);
6685 btrfs_end_transaction(trans
);
6686 if (err
&& drop_inode_on_err
) {
6687 inode_dec_link_count(inode
);
6690 btrfs_balance_delayed_items(fs_info
);
6691 btrfs_btree_balance_dirty(fs_info
);
6695 unlock_new_inode(inode
);
6700 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
6701 struct dentry
*dentry
)
6703 struct btrfs_trans_handle
*trans
= NULL
;
6704 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6705 struct inode
*inode
= d_inode(old_dentry
);
6706 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
6711 /* do not allow sys_link's with other subvols of the same device */
6712 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
6715 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
6718 err
= btrfs_set_inode_index(BTRFS_I(dir
), &index
);
6723 * 2 items for inode and inode ref
6724 * 2 items for dir items
6725 * 1 item for parent inode
6727 trans
= btrfs_start_transaction(root
, 5);
6728 if (IS_ERR(trans
)) {
6729 err
= PTR_ERR(trans
);
6734 /* There are several dir indexes for this inode, clear the cache. */
6735 BTRFS_I(inode
)->dir_index
= 0ULL;
6737 inode_inc_iversion(inode
);
6738 inode
->i_ctime
= current_time(inode
);
6740 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
6742 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
, BTRFS_I(inode
),
6748 struct dentry
*parent
= dentry
->d_parent
;
6749 err
= btrfs_update_inode(trans
, root
, inode
);
6752 if (inode
->i_nlink
== 1) {
6754 * If new hard link count is 1, it's a file created
6755 * with open(2) O_TMPFILE flag.
6757 err
= btrfs_orphan_del(trans
, BTRFS_I(inode
));
6761 d_instantiate(dentry
, inode
);
6762 btrfs_log_new_name(trans
, BTRFS_I(inode
), NULL
, parent
);
6765 btrfs_balance_delayed_items(fs_info
);
6768 btrfs_end_transaction(trans
);
6770 inode_dec_link_count(inode
);
6773 btrfs_btree_balance_dirty(fs_info
);
6777 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
6779 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6780 struct inode
*inode
= NULL
;
6781 struct btrfs_trans_handle
*trans
;
6782 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6784 int drop_on_err
= 0;
6789 * 2 items for inode and ref
6790 * 2 items for dir items
6791 * 1 for xattr if selinux is on
6793 trans
= btrfs_start_transaction(root
, 5);
6795 return PTR_ERR(trans
);
6797 err
= btrfs_find_free_ino(root
, &objectid
);
6801 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6802 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)), objectid
,
6803 S_IFDIR
| mode
, &index
);
6804 if (IS_ERR(inode
)) {
6805 err
= PTR_ERR(inode
);
6810 /* these must be set before we unlock the inode */
6811 inode
->i_op
= &btrfs_dir_inode_operations
;
6812 inode
->i_fop
= &btrfs_dir_file_operations
;
6814 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6816 goto out_fail_inode
;
6818 btrfs_i_size_write(BTRFS_I(inode
), 0);
6819 err
= btrfs_update_inode(trans
, root
, inode
);
6821 goto out_fail_inode
;
6823 err
= btrfs_add_link(trans
, BTRFS_I(dir
), BTRFS_I(inode
),
6824 dentry
->d_name
.name
,
6825 dentry
->d_name
.len
, 0, index
);
6827 goto out_fail_inode
;
6829 d_instantiate(dentry
, inode
);
6831 * mkdir is special. We're unlocking after we call d_instantiate
6832 * to avoid a race with nfsd calling d_instantiate.
6834 unlock_new_inode(inode
);
6838 btrfs_end_transaction(trans
);
6840 inode_dec_link_count(inode
);
6843 btrfs_balance_delayed_items(fs_info
);
6844 btrfs_btree_balance_dirty(fs_info
);
6848 unlock_new_inode(inode
);
6852 /* Find next extent map of a given extent map, caller needs to ensure locks */
6853 static struct extent_map
*next_extent_map(struct extent_map
*em
)
6855 struct rb_node
*next
;
6857 next
= rb_next(&em
->rb_node
);
6860 return container_of(next
, struct extent_map
, rb_node
);
6863 static struct extent_map
*prev_extent_map(struct extent_map
*em
)
6865 struct rb_node
*prev
;
6867 prev
= rb_prev(&em
->rb_node
);
6870 return container_of(prev
, struct extent_map
, rb_node
);
6873 /* helper for btfs_get_extent. Given an existing extent in the tree,
6874 * the existing extent is the nearest extent to map_start,
6875 * and an extent that you want to insert, deal with overlap and insert
6876 * the best fitted new extent into the tree.
6878 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
6879 struct extent_map
*existing
,
6880 struct extent_map
*em
,
6883 struct extent_map
*prev
;
6884 struct extent_map
*next
;
6889 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
6891 if (existing
->start
> map_start
) {
6893 prev
= prev_extent_map(next
);
6896 next
= next_extent_map(prev
);
6899 start
= prev
? extent_map_end(prev
) : em
->start
;
6900 start
= max_t(u64
, start
, em
->start
);
6901 end
= next
? next
->start
: extent_map_end(em
);
6902 end
= min_t(u64
, end
, extent_map_end(em
));
6903 start_diff
= start
- em
->start
;
6905 em
->len
= end
- start
;
6906 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
6907 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
6908 em
->block_start
+= start_diff
;
6909 em
->block_len
-= start_diff
;
6911 return add_extent_mapping(em_tree
, em
, 0);
6914 static noinline
int uncompress_inline(struct btrfs_path
*path
,
6916 size_t pg_offset
, u64 extent_offset
,
6917 struct btrfs_file_extent_item
*item
)
6920 struct extent_buffer
*leaf
= path
->nodes
[0];
6923 unsigned long inline_size
;
6927 WARN_ON(pg_offset
!= 0);
6928 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6929 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6930 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
6931 btrfs_item_nr(path
->slots
[0]));
6932 tmp
= kmalloc(inline_size
, GFP_NOFS
);
6935 ptr
= btrfs_file_extent_inline_start(item
);
6937 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
6939 max_size
= min_t(unsigned long, PAGE_SIZE
, max_size
);
6940 ret
= btrfs_decompress(compress_type
, tmp
, page
,
6941 extent_offset
, inline_size
, max_size
);
6944 * decompression code contains a memset to fill in any space between the end
6945 * of the uncompressed data and the end of max_size in case the decompressed
6946 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6947 * the end of an inline extent and the beginning of the next block, so we
6948 * cover that region here.
6951 if (max_size
+ pg_offset
< PAGE_SIZE
) {
6952 char *map
= kmap(page
);
6953 memset(map
+ pg_offset
+ max_size
, 0, PAGE_SIZE
- max_size
- pg_offset
);
6961 * a bit scary, this does extent mapping from logical file offset to the disk.
6962 * the ugly parts come from merging extents from the disk with the in-ram
6963 * representation. This gets more complex because of the data=ordered code,
6964 * where the in-ram extents might be locked pending data=ordered completion.
6966 * This also copies inline extents directly into the page.
6968 struct extent_map
*btrfs_get_extent(struct btrfs_inode
*inode
,
6970 size_t pg_offset
, u64 start
, u64 len
,
6973 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6976 u64 extent_start
= 0;
6978 u64 objectid
= btrfs_ino(inode
);
6980 struct btrfs_path
*path
= NULL
;
6981 struct btrfs_root
*root
= inode
->root
;
6982 struct btrfs_file_extent_item
*item
;
6983 struct extent_buffer
*leaf
;
6984 struct btrfs_key found_key
;
6985 struct extent_map
*em
= NULL
;
6986 struct extent_map_tree
*em_tree
= &inode
->extent_tree
;
6987 struct extent_io_tree
*io_tree
= &inode
->io_tree
;
6988 struct btrfs_trans_handle
*trans
= NULL
;
6989 const bool new_inline
= !page
|| create
;
6992 read_lock(&em_tree
->lock
);
6993 em
= lookup_extent_mapping(em_tree
, start
, len
);
6995 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
6996 read_unlock(&em_tree
->lock
);
6999 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
7000 free_extent_map(em
);
7001 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
7002 free_extent_map(em
);
7006 em
= alloc_extent_map();
7011 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
7012 em
->start
= EXTENT_MAP_HOLE
;
7013 em
->orig_start
= EXTENT_MAP_HOLE
;
7015 em
->block_len
= (u64
)-1;
7018 path
= btrfs_alloc_path();
7024 * Chances are we'll be called again, so go ahead and do
7027 path
->reada
= READA_FORWARD
;
7030 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
7031 objectid
, start
, trans
!= NULL
);
7038 if (path
->slots
[0] == 0)
7043 leaf
= path
->nodes
[0];
7044 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7045 struct btrfs_file_extent_item
);
7046 /* are we inside the extent that was found? */
7047 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7048 found_type
= found_key
.type
;
7049 if (found_key
.objectid
!= objectid
||
7050 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
7052 * If we backup past the first extent we want to move forward
7053 * and see if there is an extent in front of us, otherwise we'll
7054 * say there is a hole for our whole search range which can
7061 found_type
= btrfs_file_extent_type(leaf
, item
);
7062 extent_start
= found_key
.offset
;
7063 if (found_type
== BTRFS_FILE_EXTENT_REG
||
7064 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7065 extent_end
= extent_start
+
7066 btrfs_file_extent_num_bytes(leaf
, item
);
7068 trace_btrfs_get_extent_show_fi_regular(inode
, leaf
, item
,
7070 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
7072 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
7073 extent_end
= ALIGN(extent_start
+ size
,
7074 fs_info
->sectorsize
);
7076 trace_btrfs_get_extent_show_fi_inline(inode
, leaf
, item
,
7081 if (start
>= extent_end
) {
7083 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
7084 ret
= btrfs_next_leaf(root
, path
);
7091 leaf
= path
->nodes
[0];
7093 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7094 if (found_key
.objectid
!= objectid
||
7095 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7097 if (start
+ len
<= found_key
.offset
)
7099 if (start
> found_key
.offset
)
7102 em
->orig_start
= start
;
7103 em
->len
= found_key
.offset
- start
;
7107 btrfs_extent_item_to_extent_map(inode
, path
, item
,
7110 if (found_type
== BTRFS_FILE_EXTENT_REG
||
7111 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7113 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
7117 size_t extent_offset
;
7123 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
7124 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
7125 copy_size
= min_t(u64
, PAGE_SIZE
- pg_offset
,
7126 size
- extent_offset
);
7127 em
->start
= extent_start
+ extent_offset
;
7128 em
->len
= ALIGN(copy_size
, fs_info
->sectorsize
);
7129 em
->orig_block_len
= em
->len
;
7130 em
->orig_start
= em
->start
;
7131 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
7132 if (create
== 0 && !PageUptodate(page
)) {
7133 if (btrfs_file_extent_compression(leaf
, item
) !=
7134 BTRFS_COMPRESS_NONE
) {
7135 ret
= uncompress_inline(path
, page
, pg_offset
,
7136 extent_offset
, item
);
7143 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
7145 if (pg_offset
+ copy_size
< PAGE_SIZE
) {
7146 memset(map
+ pg_offset
+ copy_size
, 0,
7147 PAGE_SIZE
- pg_offset
-
7152 flush_dcache_page(page
);
7153 } else if (create
&& PageUptodate(page
)) {
7157 free_extent_map(em
);
7160 btrfs_release_path(path
);
7161 trans
= btrfs_join_transaction(root
);
7164 return ERR_CAST(trans
);
7168 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
7171 btrfs_mark_buffer_dirty(leaf
);
7173 set_extent_uptodate(io_tree
, em
->start
,
7174 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
7179 em
->orig_start
= start
;
7182 em
->block_start
= EXTENT_MAP_HOLE
;
7183 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
7185 btrfs_release_path(path
);
7186 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
7188 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7189 em
->start
, em
->len
, start
, len
);
7195 write_lock(&em_tree
->lock
);
7196 ret
= add_extent_mapping(em_tree
, em
, 0);
7197 /* it is possible that someone inserted the extent into the tree
7198 * while we had the lock dropped. It is also possible that
7199 * an overlapping map exists in the tree
7201 if (ret
== -EEXIST
) {
7202 struct extent_map
*existing
;
7206 existing
= search_extent_mapping(em_tree
, start
, len
);
7208 * existing will always be non-NULL, since there must be
7209 * extent causing the -EEXIST.
7211 if (existing
->start
== em
->start
&&
7212 extent_map_end(existing
) >= extent_map_end(em
) &&
7213 em
->block_start
== existing
->block_start
) {
7215 * The existing extent map already encompasses the
7216 * entire extent map we tried to add.
7218 free_extent_map(em
);
7222 } else if (start
>= extent_map_end(existing
) ||
7223 start
<= existing
->start
) {
7225 * The existing extent map is the one nearest to
7226 * the [start, start + len) range which overlaps
7228 err
= merge_extent_mapping(em_tree
, existing
,
7230 free_extent_map(existing
);
7232 free_extent_map(em
);
7236 free_extent_map(em
);
7241 write_unlock(&em_tree
->lock
);
7244 trace_btrfs_get_extent(root
, inode
, em
);
7246 btrfs_free_path(path
);
7248 ret
= btrfs_end_transaction(trans
);
7253 free_extent_map(em
);
7254 return ERR_PTR(err
);
7256 BUG_ON(!em
); /* Error is always set */
7260 struct extent_map
*btrfs_get_extent_fiemap(struct btrfs_inode
*inode
,
7262 size_t pg_offset
, u64 start
, u64 len
,
7265 struct extent_map
*em
;
7266 struct extent_map
*hole_em
= NULL
;
7267 u64 range_start
= start
;
7273 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
7277 * If our em maps to:
7279 * - a pre-alloc extent,
7280 * there might actually be delalloc bytes behind it.
7282 if (em
->block_start
!= EXTENT_MAP_HOLE
&&
7283 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7288 /* check to see if we've wrapped (len == -1 or similar) */
7297 /* ok, we didn't find anything, lets look for delalloc */
7298 found
= count_range_bits(&inode
->io_tree
, &range_start
,
7299 end
, len
, EXTENT_DELALLOC
, 1);
7300 found_end
= range_start
+ found
;
7301 if (found_end
< range_start
)
7302 found_end
= (u64
)-1;
7305 * we didn't find anything useful, return
7306 * the original results from get_extent()
7308 if (range_start
> end
|| found_end
<= start
) {
7314 /* adjust the range_start to make sure it doesn't
7315 * go backwards from the start they passed in
7317 range_start
= max(start
, range_start
);
7318 found
= found_end
- range_start
;
7321 u64 hole_start
= start
;
7324 em
= alloc_extent_map();
7330 * when btrfs_get_extent can't find anything it
7331 * returns one huge hole
7333 * make sure what it found really fits our range, and
7334 * adjust to make sure it is based on the start from
7338 u64 calc_end
= extent_map_end(hole_em
);
7340 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
7341 free_extent_map(hole_em
);
7344 hole_start
= max(hole_em
->start
, start
);
7345 hole_len
= calc_end
- hole_start
;
7349 if (hole_em
&& range_start
> hole_start
) {
7350 /* our hole starts before our delalloc, so we
7351 * have to return just the parts of the hole
7352 * that go until the delalloc starts
7354 em
->len
= min(hole_len
,
7355 range_start
- hole_start
);
7356 em
->start
= hole_start
;
7357 em
->orig_start
= hole_start
;
7359 * don't adjust block start at all,
7360 * it is fixed at EXTENT_MAP_HOLE
7362 em
->block_start
= hole_em
->block_start
;
7363 em
->block_len
= hole_len
;
7364 if (test_bit(EXTENT_FLAG_PREALLOC
, &hole_em
->flags
))
7365 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
7367 em
->start
= range_start
;
7369 em
->orig_start
= range_start
;
7370 em
->block_start
= EXTENT_MAP_DELALLOC
;
7371 em
->block_len
= found
;
7373 } else if (hole_em
) {
7378 free_extent_map(hole_em
);
7380 free_extent_map(em
);
7381 return ERR_PTR(err
);
7386 static struct extent_map
*btrfs_create_dio_extent(struct inode
*inode
,
7389 const u64 orig_start
,
7390 const u64 block_start
,
7391 const u64 block_len
,
7392 const u64 orig_block_len
,
7393 const u64 ram_bytes
,
7396 struct extent_map
*em
= NULL
;
7399 if (type
!= BTRFS_ORDERED_NOCOW
) {
7400 em
= create_io_em(inode
, start
, len
, orig_start
,
7401 block_start
, block_len
, orig_block_len
,
7403 BTRFS_COMPRESS_NONE
, /* compress_type */
7408 ret
= btrfs_add_ordered_extent_dio(inode
, start
, block_start
,
7409 len
, block_len
, type
);
7412 free_extent_map(em
);
7413 btrfs_drop_extent_cache(BTRFS_I(inode
), start
,
7414 start
+ len
- 1, 0);
7423 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
7426 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7427 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7428 struct extent_map
*em
;
7429 struct btrfs_key ins
;
7433 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
7434 ret
= btrfs_reserve_extent(root
, len
, len
, fs_info
->sectorsize
,
7435 0, alloc_hint
, &ins
, 1, 1);
7437 return ERR_PTR(ret
);
7439 em
= btrfs_create_dio_extent(inode
, start
, ins
.offset
, start
,
7440 ins
.objectid
, ins
.offset
, ins
.offset
,
7441 ins
.offset
, BTRFS_ORDERED_REGULAR
);
7442 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
7444 btrfs_free_reserved_extent(fs_info
, ins
.objectid
,
7451 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7452 * block must be cow'd
7454 noinline
int can_nocow_extent(struct inode
*inode
, u64 offset
, u64
*len
,
7455 u64
*orig_start
, u64
*orig_block_len
,
7458 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7459 struct btrfs_path
*path
;
7461 struct extent_buffer
*leaf
;
7462 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7463 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7464 struct btrfs_file_extent_item
*fi
;
7465 struct btrfs_key key
;
7472 bool nocow
= (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
);
7474 path
= btrfs_alloc_path();
7478 ret
= btrfs_lookup_file_extent(NULL
, root
, path
,
7479 btrfs_ino(BTRFS_I(inode
)), offset
, 0);
7483 slot
= path
->slots
[0];
7486 /* can't find the item, must cow */
7493 leaf
= path
->nodes
[0];
7494 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
7495 if (key
.objectid
!= btrfs_ino(BTRFS_I(inode
)) ||
7496 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7497 /* not our file or wrong item type, must cow */
7501 if (key
.offset
> offset
) {
7502 /* Wrong offset, must cow */
7506 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
7507 found_type
= btrfs_file_extent_type(leaf
, fi
);
7508 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
7509 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
7510 /* not a regular extent, must cow */
7514 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_REG
)
7517 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
7518 if (extent_end
<= offset
)
7521 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7522 if (disk_bytenr
== 0)
7525 if (btrfs_file_extent_compression(leaf
, fi
) ||
7526 btrfs_file_extent_encryption(leaf
, fi
) ||
7527 btrfs_file_extent_other_encoding(leaf
, fi
))
7530 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
7533 *orig_start
= key
.offset
- backref_offset
;
7534 *orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7535 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
7538 if (btrfs_extent_readonly(fs_info
, disk_bytenr
))
7541 num_bytes
= min(offset
+ *len
, extent_end
) - offset
;
7542 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7545 range_end
= round_up(offset
+ num_bytes
,
7546 root
->fs_info
->sectorsize
) - 1;
7547 ret
= test_range_bit(io_tree
, offset
, range_end
,
7548 EXTENT_DELALLOC
, 0, NULL
);
7555 btrfs_release_path(path
);
7558 * look for other files referencing this extent, if we
7559 * find any we must cow
7562 ret
= btrfs_cross_ref_exist(root
, btrfs_ino(BTRFS_I(inode
)),
7563 key
.offset
- backref_offset
, disk_bytenr
);
7570 * adjust disk_bytenr and num_bytes to cover just the bytes
7571 * in this extent we are about to write. If there
7572 * are any csums in that range we have to cow in order
7573 * to keep the csums correct
7575 disk_bytenr
+= backref_offset
;
7576 disk_bytenr
+= offset
- key
.offset
;
7577 if (csum_exist_in_range(fs_info
, disk_bytenr
, num_bytes
))
7580 * all of the above have passed, it is safe to overwrite this extent
7586 btrfs_free_path(path
);
7590 bool btrfs_page_exists_in_range(struct inode
*inode
, loff_t start
, loff_t end
)
7592 struct radix_tree_root
*root
= &inode
->i_mapping
->page_tree
;
7594 void **pagep
= NULL
;
7595 struct page
*page
= NULL
;
7596 unsigned long start_idx
;
7597 unsigned long end_idx
;
7599 start_idx
= start
>> PAGE_SHIFT
;
7602 * end is the last byte in the last page. end == start is legal
7604 end_idx
= end
>> PAGE_SHIFT
;
7608 /* Most of the code in this while loop is lifted from
7609 * find_get_page. It's been modified to begin searching from a
7610 * page and return just the first page found in that range. If the
7611 * found idx is less than or equal to the end idx then we know that
7612 * a page exists. If no pages are found or if those pages are
7613 * outside of the range then we're fine (yay!) */
7614 while (page
== NULL
&&
7615 radix_tree_gang_lookup_slot(root
, &pagep
, NULL
, start_idx
, 1)) {
7616 page
= radix_tree_deref_slot(pagep
);
7617 if (unlikely(!page
))
7620 if (radix_tree_exception(page
)) {
7621 if (radix_tree_deref_retry(page
)) {
7626 * Otherwise, shmem/tmpfs must be storing a swap entry
7627 * here as an exceptional entry: so return it without
7628 * attempting to raise page count.
7631 break; /* TODO: Is this relevant for this use case? */
7634 if (!page_cache_get_speculative(page
)) {
7640 * Has the page moved?
7641 * This is part of the lockless pagecache protocol. See
7642 * include/linux/pagemap.h for details.
7644 if (unlikely(page
!= *pagep
)) {
7651 if (page
->index
<= end_idx
)
7660 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
7661 struct extent_state
**cached_state
, int writing
)
7663 struct btrfs_ordered_extent
*ordered
;
7667 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7670 * We're concerned with the entire range that we're going to be
7671 * doing DIO to, so we need to make sure there's no ordered
7672 * extents in this range.
7674 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), lockstart
,
7675 lockend
- lockstart
+ 1);
7678 * We need to make sure there are no buffered pages in this
7679 * range either, we could have raced between the invalidate in
7680 * generic_file_direct_write and locking the extent. The
7681 * invalidate needs to happen so that reads after a write do not
7686 !btrfs_page_exists_in_range(inode
, lockstart
, lockend
)))
7689 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7690 cached_state
, GFP_NOFS
);
7694 * If we are doing a DIO read and the ordered extent we
7695 * found is for a buffered write, we can not wait for it
7696 * to complete and retry, because if we do so we can
7697 * deadlock with concurrent buffered writes on page
7698 * locks. This happens only if our DIO read covers more
7699 * than one extent map, if at this point has already
7700 * created an ordered extent for a previous extent map
7701 * and locked its range in the inode's io tree, and a
7702 * concurrent write against that previous extent map's
7703 * range and this range started (we unlock the ranges
7704 * in the io tree only when the bios complete and
7705 * buffered writes always lock pages before attempting
7706 * to lock range in the io tree).
7709 test_bit(BTRFS_ORDERED_DIRECT
, &ordered
->flags
))
7710 btrfs_start_ordered_extent(inode
, ordered
, 1);
7713 btrfs_put_ordered_extent(ordered
);
7716 * We could trigger writeback for this range (and wait
7717 * for it to complete) and then invalidate the pages for
7718 * this range (through invalidate_inode_pages2_range()),
7719 * but that can lead us to a deadlock with a concurrent
7720 * call to readpages() (a buffered read or a defrag call
7721 * triggered a readahead) on a page lock due to an
7722 * ordered dio extent we created before but did not have
7723 * yet a corresponding bio submitted (whence it can not
7724 * complete), which makes readpages() wait for that
7725 * ordered extent to complete while holding a lock on
7740 /* The callers of this must take lock_extent() */
7741 static struct extent_map
*create_io_em(struct inode
*inode
, u64 start
, u64 len
,
7742 u64 orig_start
, u64 block_start
,
7743 u64 block_len
, u64 orig_block_len
,
7744 u64 ram_bytes
, int compress_type
,
7747 struct extent_map_tree
*em_tree
;
7748 struct extent_map
*em
;
7749 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7752 ASSERT(type
== BTRFS_ORDERED_PREALLOC
||
7753 type
== BTRFS_ORDERED_COMPRESSED
||
7754 type
== BTRFS_ORDERED_NOCOW
||
7755 type
== BTRFS_ORDERED_REGULAR
);
7757 em_tree
= &BTRFS_I(inode
)->extent_tree
;
7758 em
= alloc_extent_map();
7760 return ERR_PTR(-ENOMEM
);
7763 em
->orig_start
= orig_start
;
7765 em
->block_len
= block_len
;
7766 em
->block_start
= block_start
;
7767 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
7768 em
->orig_block_len
= orig_block_len
;
7769 em
->ram_bytes
= ram_bytes
;
7770 em
->generation
= -1;
7771 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
7772 if (type
== BTRFS_ORDERED_PREALLOC
) {
7773 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
7774 } else if (type
== BTRFS_ORDERED_COMPRESSED
) {
7775 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
7776 em
->compress_type
= compress_type
;
7780 btrfs_drop_extent_cache(BTRFS_I(inode
), em
->start
,
7781 em
->start
+ em
->len
- 1, 0);
7782 write_lock(&em_tree
->lock
);
7783 ret
= add_extent_mapping(em_tree
, em
, 1);
7784 write_unlock(&em_tree
->lock
);
7786 * The caller has taken lock_extent(), who could race with us
7789 } while (ret
== -EEXIST
);
7792 free_extent_map(em
);
7793 return ERR_PTR(ret
);
7796 /* em got 2 refs now, callers needs to do free_extent_map once. */
7800 static void adjust_dio_outstanding_extents(struct inode
*inode
,
7801 struct btrfs_dio_data
*dio_data
,
7804 unsigned num_extents
= count_max_extents(len
);
7807 * If we have an outstanding_extents count still set then we're
7808 * within our reservation, otherwise we need to adjust our inode
7809 * counter appropriately.
7811 if (dio_data
->outstanding_extents
>= num_extents
) {
7812 dio_data
->outstanding_extents
-= num_extents
;
7815 * If dio write length has been split due to no large enough
7816 * contiguous space, we need to compensate our inode counter
7819 u64 num_needed
= num_extents
- dio_data
->outstanding_extents
;
7821 spin_lock(&BTRFS_I(inode
)->lock
);
7822 BTRFS_I(inode
)->outstanding_extents
+= num_needed
;
7823 spin_unlock(&BTRFS_I(inode
)->lock
);
7827 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
7828 struct buffer_head
*bh_result
, int create
)
7830 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7831 struct extent_map
*em
;
7832 struct extent_state
*cached_state
= NULL
;
7833 struct btrfs_dio_data
*dio_data
= NULL
;
7834 u64 start
= iblock
<< inode
->i_blkbits
;
7835 u64 lockstart
, lockend
;
7836 u64 len
= bh_result
->b_size
;
7837 int unlock_bits
= EXTENT_LOCKED
;
7841 unlock_bits
|= EXTENT_DIRTY
;
7843 len
= min_t(u64
, len
, fs_info
->sectorsize
);
7846 lockend
= start
+ len
- 1;
7848 if (current
->journal_info
) {
7850 * Need to pull our outstanding extents and set journal_info to NULL so
7851 * that anything that needs to check if there's a transaction doesn't get
7854 dio_data
= current
->journal_info
;
7855 current
->journal_info
= NULL
;
7859 * If this errors out it's because we couldn't invalidate pagecache for
7860 * this range and we need to fallback to buffered.
7862 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
,
7868 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0, start
, len
, 0);
7875 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7876 * io. INLINE is special, and we could probably kludge it in here, but
7877 * it's still buffered so for safety lets just fall back to the generic
7880 * For COMPRESSED we _have_ to read the entire extent in so we can
7881 * decompress it, so there will be buffering required no matter what we
7882 * do, so go ahead and fallback to buffered.
7884 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7885 * to buffered IO. Don't blame me, this is the price we pay for using
7888 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
7889 em
->block_start
== EXTENT_MAP_INLINE
) {
7890 free_extent_map(em
);
7895 /* Just a good old fashioned hole, return */
7896 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
7897 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
7898 free_extent_map(em
);
7903 * We don't allocate a new extent in the following cases
7905 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7907 * 2) The extent is marked as PREALLOC. We're good to go here and can
7908 * just use the extent.
7912 len
= min(len
, em
->len
- (start
- em
->start
));
7913 lockstart
= start
+ len
;
7917 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
7918 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
7919 em
->block_start
!= EXTENT_MAP_HOLE
)) {
7921 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
7923 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7924 type
= BTRFS_ORDERED_PREALLOC
;
7926 type
= BTRFS_ORDERED_NOCOW
;
7927 len
= min(len
, em
->len
- (start
- em
->start
));
7928 block_start
= em
->block_start
+ (start
- em
->start
);
7930 if (can_nocow_extent(inode
, start
, &len
, &orig_start
,
7931 &orig_block_len
, &ram_bytes
) == 1 &&
7932 btrfs_inc_nocow_writers(fs_info
, block_start
)) {
7933 struct extent_map
*em2
;
7935 em2
= btrfs_create_dio_extent(inode
, start
, len
,
7936 orig_start
, block_start
,
7937 len
, orig_block_len
,
7939 btrfs_dec_nocow_writers(fs_info
, block_start
);
7940 if (type
== BTRFS_ORDERED_PREALLOC
) {
7941 free_extent_map(em
);
7944 if (em2
&& IS_ERR(em2
)) {
7949 * For inode marked NODATACOW or extent marked PREALLOC,
7950 * use the existing or preallocated extent, so does not
7951 * need to adjust btrfs_space_info's bytes_may_use.
7953 btrfs_free_reserved_data_space_noquota(inode
,
7960 * this will cow the extent, reset the len in case we changed
7963 len
= bh_result
->b_size
;
7964 free_extent_map(em
);
7965 em
= btrfs_new_extent_direct(inode
, start
, len
);
7970 len
= min(len
, em
->len
- (start
- em
->start
));
7972 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
7974 bh_result
->b_size
= len
;
7975 bh_result
->b_bdev
= em
->bdev
;
7976 set_buffer_mapped(bh_result
);
7978 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7979 set_buffer_new(bh_result
);
7982 * Need to update the i_size under the extent lock so buffered
7983 * readers will get the updated i_size when we unlock.
7985 if (!dio_data
->overwrite
&& start
+ len
> i_size_read(inode
))
7986 i_size_write(inode
, start
+ len
);
7988 adjust_dio_outstanding_extents(inode
, dio_data
, len
);
7989 WARN_ON(dio_data
->reserve
< len
);
7990 dio_data
->reserve
-= len
;
7991 dio_data
->unsubmitted_oe_range_end
= start
+ len
;
7992 current
->journal_info
= dio_data
;
7996 * In the case of write we need to clear and unlock the entire range,
7997 * in the case of read we need to unlock only the end area that we
7998 * aren't using if there is any left over space.
8000 if (lockstart
< lockend
) {
8001 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
8002 lockend
, unlock_bits
, 1, 0,
8003 &cached_state
, GFP_NOFS
);
8005 free_extent_state(cached_state
);
8008 free_extent_map(em
);
8013 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
8014 unlock_bits
, 1, 0, &cached_state
, GFP_NOFS
);
8017 current
->journal_info
= dio_data
;
8019 * Compensate the delalloc release we do in btrfs_direct_IO() when we
8020 * write less data then expected, so that we don't underflow our inode's
8021 * outstanding extents counter.
8023 if (create
&& dio_data
)
8024 adjust_dio_outstanding_extents(inode
, dio_data
, len
);
8029 static inline blk_status_t
submit_dio_repair_bio(struct inode
*inode
,
8033 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8036 BUG_ON(bio_op(bio
) == REQ_OP_WRITE
);
8040 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, BTRFS_WQ_ENDIO_DIO_REPAIR
);
8044 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
8050 static int btrfs_check_dio_repairable(struct inode
*inode
,
8051 struct bio
*failed_bio
,
8052 struct io_failure_record
*failrec
,
8055 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8058 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
, failrec
->len
);
8059 if (num_copies
== 1) {
8061 * we only have a single copy of the data, so don't bother with
8062 * all the retry and error correction code that follows. no
8063 * matter what the error is, it is very likely to persist.
8065 btrfs_debug(fs_info
,
8066 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
8067 num_copies
, failrec
->this_mirror
, failed_mirror
);
8071 failrec
->failed_mirror
= failed_mirror
;
8072 failrec
->this_mirror
++;
8073 if (failrec
->this_mirror
== failed_mirror
)
8074 failrec
->this_mirror
++;
8076 if (failrec
->this_mirror
> num_copies
) {
8077 btrfs_debug(fs_info
,
8078 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
8079 num_copies
, failrec
->this_mirror
, failed_mirror
);
8086 static blk_status_t
dio_read_error(struct inode
*inode
, struct bio
*failed_bio
,
8087 struct page
*page
, unsigned int pgoff
,
8088 u64 start
, u64 end
, int failed_mirror
,
8089 bio_end_io_t
*repair_endio
, void *repair_arg
)
8091 struct io_failure_record
*failrec
;
8092 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
8093 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
8096 unsigned int read_mode
= 0;
8099 blk_status_t status
;
8101 BUG_ON(bio_op(failed_bio
) == REQ_OP_WRITE
);
8103 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
8105 return errno_to_blk_status(ret
);
8107 ret
= btrfs_check_dio_repairable(inode
, failed_bio
, failrec
,
8110 free_io_failure(failure_tree
, io_tree
, failrec
);
8111 return BLK_STS_IOERR
;
8114 segs
= bio_segments(failed_bio
);
8116 (failed_bio
->bi_io_vec
->bv_len
> btrfs_inode_sectorsize(inode
)))
8117 read_mode
|= REQ_FAILFAST_DEV
;
8119 isector
= start
- btrfs_io_bio(failed_bio
)->logical
;
8120 isector
>>= inode
->i_sb
->s_blocksize_bits
;
8121 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
8122 pgoff
, isector
, repair_endio
, repair_arg
);
8123 bio_set_op_attrs(bio
, REQ_OP_READ
, read_mode
);
8125 btrfs_debug(BTRFS_I(inode
)->root
->fs_info
,
8126 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
8127 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
8129 status
= submit_dio_repair_bio(inode
, bio
, failrec
->this_mirror
);
8131 free_io_failure(failure_tree
, io_tree
, failrec
);
8138 struct btrfs_retry_complete
{
8139 struct completion done
;
8140 struct inode
*inode
;
8145 static void btrfs_retry_endio_nocsum(struct bio
*bio
)
8147 struct btrfs_retry_complete
*done
= bio
->bi_private
;
8148 struct inode
*inode
= done
->inode
;
8149 struct bio_vec
*bvec
;
8150 struct extent_io_tree
*io_tree
, *failure_tree
;
8156 ASSERT(bio
->bi_vcnt
== 1);
8157 io_tree
= &BTRFS_I(inode
)->io_tree
;
8158 failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
8159 ASSERT(bio
->bi_io_vec
->bv_len
== btrfs_inode_sectorsize(inode
));
8162 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
8163 bio_for_each_segment_all(bvec
, bio
, i
)
8164 clean_io_failure(BTRFS_I(inode
)->root
->fs_info
, failure_tree
,
8165 io_tree
, done
->start
, bvec
->bv_page
,
8166 btrfs_ino(BTRFS_I(inode
)), 0);
8168 complete(&done
->done
);
8172 static blk_status_t
__btrfs_correct_data_nocsum(struct inode
*inode
,
8173 struct btrfs_io_bio
*io_bio
)
8175 struct btrfs_fs_info
*fs_info
;
8176 struct bio_vec bvec
;
8177 struct bvec_iter iter
;
8178 struct btrfs_retry_complete done
;
8184 blk_status_t err
= BLK_STS_OK
;
8186 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
8187 sectorsize
= fs_info
->sectorsize
;
8189 start
= io_bio
->logical
;
8191 io_bio
->bio
.bi_iter
= io_bio
->iter
;
8193 bio_for_each_segment(bvec
, &io_bio
->bio
, iter
) {
8194 nr_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, bvec
.bv_len
);
8195 pgoff
= bvec
.bv_offset
;
8197 next_block_or_try_again
:
8200 init_completion(&done
.done
);
8202 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
.bv_page
,
8203 pgoff
, start
, start
+ sectorsize
- 1,
8205 btrfs_retry_endio_nocsum
, &done
);
8211 wait_for_completion_io(&done
.done
);
8213 if (!done
.uptodate
) {
8214 /* We might have another mirror, so try again */
8215 goto next_block_or_try_again
;
8219 start
+= sectorsize
;
8223 pgoff
+= sectorsize
;
8224 ASSERT(pgoff
< PAGE_SIZE
);
8225 goto next_block_or_try_again
;
8232 static void btrfs_retry_endio(struct bio
*bio
)
8234 struct btrfs_retry_complete
*done
= bio
->bi_private
;
8235 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8236 struct extent_io_tree
*io_tree
, *failure_tree
;
8237 struct inode
*inode
= done
->inode
;
8238 struct bio_vec
*bvec
;
8248 ASSERT(bio
->bi_vcnt
== 1);
8249 ASSERT(bio
->bi_io_vec
->bv_len
== btrfs_inode_sectorsize(done
->inode
));
8251 io_tree
= &BTRFS_I(inode
)->io_tree
;
8252 failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
8254 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
8255 bio_for_each_segment_all(bvec
, bio
, i
) {
8256 ret
= __readpage_endio_check(inode
, io_bio
, i
, bvec
->bv_page
,
8257 bvec
->bv_offset
, done
->start
,
8260 clean_io_failure(BTRFS_I(inode
)->root
->fs_info
,
8261 failure_tree
, io_tree
, done
->start
,
8263 btrfs_ino(BTRFS_I(inode
)),
8269 done
->uptodate
= uptodate
;
8271 complete(&done
->done
);
8275 static blk_status_t
__btrfs_subio_endio_read(struct inode
*inode
,
8276 struct btrfs_io_bio
*io_bio
, blk_status_t err
)
8278 struct btrfs_fs_info
*fs_info
;
8279 struct bio_vec bvec
;
8280 struct bvec_iter iter
;
8281 struct btrfs_retry_complete done
;
8288 bool uptodate
= (err
== 0);
8290 blk_status_t status
;
8292 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
8293 sectorsize
= fs_info
->sectorsize
;
8296 start
= io_bio
->logical
;
8298 io_bio
->bio
.bi_iter
= io_bio
->iter
;
8300 bio_for_each_segment(bvec
, &io_bio
->bio
, iter
) {
8301 nr_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, bvec
.bv_len
);
8303 pgoff
= bvec
.bv_offset
;
8306 csum_pos
= BTRFS_BYTES_TO_BLKS(fs_info
, offset
);
8307 ret
= __readpage_endio_check(inode
, io_bio
, csum_pos
,
8308 bvec
.bv_page
, pgoff
, start
, sectorsize
);
8315 init_completion(&done
.done
);
8317 status
= dio_read_error(inode
, &io_bio
->bio
, bvec
.bv_page
,
8318 pgoff
, start
, start
+ sectorsize
- 1,
8319 io_bio
->mirror_num
, btrfs_retry_endio
,
8326 wait_for_completion_io(&done
.done
);
8328 if (!done
.uptodate
) {
8329 /* We might have another mirror, so try again */
8333 offset
+= sectorsize
;
8334 start
+= sectorsize
;
8340 pgoff
+= sectorsize
;
8341 ASSERT(pgoff
< PAGE_SIZE
);
8349 static blk_status_t
btrfs_subio_endio_read(struct inode
*inode
,
8350 struct btrfs_io_bio
*io_bio
, blk_status_t err
)
8352 bool skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
8356 return __btrfs_correct_data_nocsum(inode
, io_bio
);
8360 return __btrfs_subio_endio_read(inode
, io_bio
, err
);
8364 static void btrfs_endio_direct_read(struct bio
*bio
)
8366 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8367 struct inode
*inode
= dip
->inode
;
8368 struct bio
*dio_bio
;
8369 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8370 blk_status_t err
= bio
->bi_status
;
8372 if (dip
->flags
& BTRFS_DIO_ORIG_BIO_SUBMITTED
)
8373 err
= btrfs_subio_endio_read(inode
, io_bio
, err
);
8375 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
8376 dip
->logical_offset
+ dip
->bytes
- 1);
8377 dio_bio
= dip
->dio_bio
;
8381 dio_bio
->bi_status
= err
;
8382 dio_end_io(dio_bio
);
8385 io_bio
->end_io(io_bio
, blk_status_to_errno(err
));
8389 static void __endio_write_update_ordered(struct inode
*inode
,
8390 const u64 offset
, const u64 bytes
,
8391 const bool uptodate
)
8393 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8394 struct btrfs_ordered_extent
*ordered
= NULL
;
8395 struct btrfs_workqueue
*wq
;
8396 btrfs_work_func_t func
;
8397 u64 ordered_offset
= offset
;
8398 u64 ordered_bytes
= bytes
;
8402 if (btrfs_is_free_space_inode(BTRFS_I(inode
))) {
8403 wq
= fs_info
->endio_freespace_worker
;
8404 func
= btrfs_freespace_write_helper
;
8406 wq
= fs_info
->endio_write_workers
;
8407 func
= btrfs_endio_write_helper
;
8411 last_offset
= ordered_offset
;
8412 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
8419 btrfs_init_work(&ordered
->work
, func
, finish_ordered_fn
, NULL
, NULL
);
8420 btrfs_queue_work(wq
, &ordered
->work
);
8423 * If btrfs_dec_test_ordered_pending does not find any ordered extent
8424 * in the range, we can exit.
8426 if (ordered_offset
== last_offset
)
8429 * our bio might span multiple ordered extents. If we haven't
8430 * completed the accounting for the whole dio, go back and try again
8432 if (ordered_offset
< offset
+ bytes
) {
8433 ordered_bytes
= offset
+ bytes
- ordered_offset
;
8439 static void btrfs_endio_direct_write(struct bio
*bio
)
8441 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8442 struct bio
*dio_bio
= dip
->dio_bio
;
8444 __endio_write_update_ordered(dip
->inode
, dip
->logical_offset
,
8445 dip
->bytes
, !bio
->bi_status
);
8449 dio_bio
->bi_status
= bio
->bi_status
;
8450 dio_end_io(dio_bio
);
8454 static blk_status_t
__btrfs_submit_bio_start_direct_io(void *private_data
,
8455 struct bio
*bio
, int mirror_num
,
8456 unsigned long bio_flags
, u64 offset
)
8458 struct inode
*inode
= private_data
;
8460 ret
= btrfs_csum_one_bio(inode
, bio
, offset
, 1);
8461 BUG_ON(ret
); /* -ENOMEM */
8465 static void btrfs_end_dio_bio(struct bio
*bio
)
8467 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8468 blk_status_t err
= bio
->bi_status
;
8471 btrfs_warn(BTRFS_I(dip
->inode
)->root
->fs_info
,
8472 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8473 btrfs_ino(BTRFS_I(dip
->inode
)), bio_op(bio
),
8475 (unsigned long long)bio
->bi_iter
.bi_sector
,
8476 bio
->bi_iter
.bi_size
, err
);
8478 if (dip
->subio_endio
)
8479 err
= dip
->subio_endio(dip
->inode
, btrfs_io_bio(bio
), err
);
8485 * before atomic variable goto zero, we must make sure
8486 * dip->errors is perceived to be set.
8488 smp_mb__before_atomic();
8491 /* if there are more bios still pending for this dio, just exit */
8492 if (!atomic_dec_and_test(&dip
->pending_bios
))
8496 bio_io_error(dip
->orig_bio
);
8498 dip
->dio_bio
->bi_status
= 0;
8499 bio_endio(dip
->orig_bio
);
8505 static inline blk_status_t
btrfs_lookup_and_bind_dio_csum(struct inode
*inode
,
8506 struct btrfs_dio_private
*dip
,
8510 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8511 struct btrfs_io_bio
*orig_io_bio
= btrfs_io_bio(dip
->orig_bio
);
8515 * We load all the csum data we need when we submit
8516 * the first bio to reduce the csum tree search and
8519 if (dip
->logical_offset
== file_offset
) {
8520 ret
= btrfs_lookup_bio_sums_dio(inode
, dip
->orig_bio
,
8526 if (bio
== dip
->orig_bio
)
8529 file_offset
-= dip
->logical_offset
;
8530 file_offset
>>= inode
->i_sb
->s_blocksize_bits
;
8531 io_bio
->csum
= (u8
*)(((u32
*)orig_io_bio
->csum
) + file_offset
);
8536 static inline blk_status_t
8537 __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
, u64 file_offset
,
8540 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8541 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8542 bool write
= bio_op(bio
) == REQ_OP_WRITE
;
8546 async_submit
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
8551 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, BTRFS_WQ_ENDIO_DATA
);
8556 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
8559 if (write
&& async_submit
) {
8560 ret
= btrfs_wq_submit_bio(fs_info
, bio
, 0, 0,
8562 __btrfs_submit_bio_start_direct_io
,
8563 __btrfs_submit_bio_done
);
8567 * If we aren't doing async submit, calculate the csum of the
8570 ret
= btrfs_csum_one_bio(inode
, bio
, file_offset
, 1);
8574 ret
= btrfs_lookup_and_bind_dio_csum(inode
, dip
, bio
,
8580 ret
= btrfs_map_bio(fs_info
, bio
, 0, async_submit
);
8586 static int btrfs_submit_direct_hook(struct btrfs_dio_private
*dip
)
8588 struct inode
*inode
= dip
->inode
;
8589 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8591 struct bio
*orig_bio
= dip
->orig_bio
;
8592 u64 start_sector
= orig_bio
->bi_iter
.bi_sector
;
8593 u64 file_offset
= dip
->logical_offset
;
8595 int async_submit
= 0;
8597 int clone_offset
= 0;
8600 blk_status_t status
;
8602 map_length
= orig_bio
->bi_iter
.bi_size
;
8603 submit_len
= map_length
;
8604 ret
= btrfs_map_block(fs_info
, btrfs_op(orig_bio
), start_sector
<< 9,
8605 &map_length
, NULL
, 0);
8609 if (map_length
>= submit_len
) {
8611 dip
->flags
|= BTRFS_DIO_ORIG_BIO_SUBMITTED
;
8615 /* async crcs make it difficult to collect full stripe writes. */
8616 if (btrfs_data_alloc_profile(fs_info
) & BTRFS_BLOCK_GROUP_RAID56_MASK
)
8622 ASSERT(map_length
<= INT_MAX
);
8623 atomic_inc(&dip
->pending_bios
);
8625 clone_len
= min_t(int, submit_len
, map_length
);
8628 * This will never fail as it's passing GPF_NOFS and
8629 * the allocation is backed by btrfs_bioset.
8631 bio
= btrfs_bio_clone_partial(orig_bio
, clone_offset
,
8633 bio
->bi_private
= dip
;
8634 bio
->bi_end_io
= btrfs_end_dio_bio
;
8635 btrfs_io_bio(bio
)->logical
= file_offset
;
8637 ASSERT(submit_len
>= clone_len
);
8638 submit_len
-= clone_len
;
8639 if (submit_len
== 0)
8643 * Increase the count before we submit the bio so we know
8644 * the end IO handler won't happen before we increase the
8645 * count. Otherwise, the dip might get freed before we're
8646 * done setting it up.
8648 atomic_inc(&dip
->pending_bios
);
8650 status
= __btrfs_submit_dio_bio(bio
, inode
, file_offset
,
8654 atomic_dec(&dip
->pending_bios
);
8658 clone_offset
+= clone_len
;
8659 start_sector
+= clone_len
>> 9;
8660 file_offset
+= clone_len
;
8662 map_length
= submit_len
;
8663 ret
= btrfs_map_block(fs_info
, btrfs_op(orig_bio
),
8664 start_sector
<< 9, &map_length
, NULL
, 0);
8667 } while (submit_len
> 0);
8670 status
= __btrfs_submit_dio_bio(bio
, inode
, file_offset
, async_submit
);
8678 * before atomic variable goto zero, we must
8679 * make sure dip->errors is perceived to be set.
8681 smp_mb__before_atomic();
8682 if (atomic_dec_and_test(&dip
->pending_bios
))
8683 bio_io_error(dip
->orig_bio
);
8685 /* bio_end_io() will handle error, so we needn't return it */
8689 static void btrfs_submit_direct(struct bio
*dio_bio
, struct inode
*inode
,
8692 struct btrfs_dio_private
*dip
= NULL
;
8693 struct bio
*bio
= NULL
;
8694 struct btrfs_io_bio
*io_bio
;
8695 bool write
= (bio_op(dio_bio
) == REQ_OP_WRITE
);
8698 bio
= btrfs_bio_clone(dio_bio
);
8700 dip
= kzalloc(sizeof(*dip
), GFP_NOFS
);
8706 dip
->private = dio_bio
->bi_private
;
8708 dip
->logical_offset
= file_offset
;
8709 dip
->bytes
= dio_bio
->bi_iter
.bi_size
;
8710 dip
->disk_bytenr
= (u64
)dio_bio
->bi_iter
.bi_sector
<< 9;
8711 bio
->bi_private
= dip
;
8712 dip
->orig_bio
= bio
;
8713 dip
->dio_bio
= dio_bio
;
8714 atomic_set(&dip
->pending_bios
, 0);
8715 io_bio
= btrfs_io_bio(bio
);
8716 io_bio
->logical
= file_offset
;
8719 bio
->bi_end_io
= btrfs_endio_direct_write
;
8721 bio
->bi_end_io
= btrfs_endio_direct_read
;
8722 dip
->subio_endio
= btrfs_subio_endio_read
;
8726 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8727 * even if we fail to submit a bio, because in such case we do the
8728 * corresponding error handling below and it must not be done a second
8729 * time by btrfs_direct_IO().
8732 struct btrfs_dio_data
*dio_data
= current
->journal_info
;
8734 dio_data
->unsubmitted_oe_range_end
= dip
->logical_offset
+
8736 dio_data
->unsubmitted_oe_range_start
=
8737 dio_data
->unsubmitted_oe_range_end
;
8740 ret
= btrfs_submit_direct_hook(dip
);
8745 io_bio
->end_io(io_bio
, ret
);
8749 * If we arrived here it means either we failed to submit the dip
8750 * or we either failed to clone the dio_bio or failed to allocate the
8751 * dip. If we cloned the dio_bio and allocated the dip, we can just
8752 * call bio_endio against our io_bio so that we get proper resource
8753 * cleanup if we fail to submit the dip, otherwise, we must do the
8754 * same as btrfs_endio_direct_[write|read] because we can't call these
8755 * callbacks - they require an allocated dip and a clone of dio_bio.
8760 * The end io callbacks free our dip, do the final put on bio
8761 * and all the cleanup and final put for dio_bio (through
8768 __endio_write_update_ordered(inode
,
8770 dio_bio
->bi_iter
.bi_size
,
8773 unlock_extent(&BTRFS_I(inode
)->io_tree
, file_offset
,
8774 file_offset
+ dio_bio
->bi_iter
.bi_size
- 1);
8776 dio_bio
->bi_status
= BLK_STS_IOERR
;
8778 * Releases and cleans up our dio_bio, no need to bio_put()
8779 * nor bio_endio()/bio_io_error() against dio_bio.
8781 dio_end_io(dio_bio
);
8788 static ssize_t
check_direct_IO(struct btrfs_fs_info
*fs_info
,
8790 const struct iov_iter
*iter
, loff_t offset
)
8794 unsigned int blocksize_mask
= fs_info
->sectorsize
- 1;
8795 ssize_t retval
= -EINVAL
;
8797 if (offset
& blocksize_mask
)
8800 if (iov_iter_alignment(iter
) & blocksize_mask
)
8803 /* If this is a write we don't need to check anymore */
8804 if (iov_iter_rw(iter
) != READ
|| !iter_is_iovec(iter
))
8807 * Check to make sure we don't have duplicate iov_base's in this
8808 * iovec, if so return EINVAL, otherwise we'll get csum errors
8809 * when reading back.
8811 for (seg
= 0; seg
< iter
->nr_segs
; seg
++) {
8812 for (i
= seg
+ 1; i
< iter
->nr_segs
; i
++) {
8813 if (iter
->iov
[seg
].iov_base
== iter
->iov
[i
].iov_base
)
8822 static ssize_t
btrfs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
8824 struct file
*file
= iocb
->ki_filp
;
8825 struct inode
*inode
= file
->f_mapping
->host
;
8826 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8827 struct btrfs_dio_data dio_data
= { 0 };
8828 struct extent_changeset
*data_reserved
= NULL
;
8829 loff_t offset
= iocb
->ki_pos
;
8833 bool relock
= false;
8836 if (check_direct_IO(fs_info
, iocb
, iter
, offset
))
8839 inode_dio_begin(inode
);
8842 * The generic stuff only does filemap_write_and_wait_range, which
8843 * isn't enough if we've written compressed pages to this area, so
8844 * we need to flush the dirty pages again to make absolutely sure
8845 * that any outstanding dirty pages are on disk.
8847 count
= iov_iter_count(iter
);
8848 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
8849 &BTRFS_I(inode
)->runtime_flags
))
8850 filemap_fdatawrite_range(inode
->i_mapping
, offset
,
8851 offset
+ count
- 1);
8853 if (iov_iter_rw(iter
) == WRITE
) {
8855 * If the write DIO is beyond the EOF, we need update
8856 * the isize, but it is protected by i_mutex. So we can
8857 * not unlock the i_mutex at this case.
8859 if (offset
+ count
<= inode
->i_size
) {
8860 dio_data
.overwrite
= 1;
8861 inode_unlock(inode
);
8863 } else if (iocb
->ki_flags
& IOCB_NOWAIT
) {
8867 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
,
8871 dio_data
.outstanding_extents
= count_max_extents(count
);
8874 * We need to know how many extents we reserved so that we can
8875 * do the accounting properly if we go over the number we
8876 * originally calculated. Abuse current->journal_info for this.
8878 dio_data
.reserve
= round_up(count
,
8879 fs_info
->sectorsize
);
8880 dio_data
.unsubmitted_oe_range_start
= (u64
)offset
;
8881 dio_data
.unsubmitted_oe_range_end
= (u64
)offset
;
8882 current
->journal_info
= &dio_data
;
8883 down_read(&BTRFS_I(inode
)->dio_sem
);
8884 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK
,
8885 &BTRFS_I(inode
)->runtime_flags
)) {
8886 inode_dio_end(inode
);
8887 flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
8891 ret
= __blockdev_direct_IO(iocb
, inode
,
8892 fs_info
->fs_devices
->latest_bdev
,
8893 iter
, btrfs_get_blocks_direct
, NULL
,
8894 btrfs_submit_direct
, flags
);
8895 if (iov_iter_rw(iter
) == WRITE
) {
8896 up_read(&BTRFS_I(inode
)->dio_sem
);
8897 current
->journal_info
= NULL
;
8898 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
8899 if (dio_data
.reserve
)
8900 btrfs_delalloc_release_space(inode
, data_reserved
,
8901 offset
, dio_data
.reserve
);
8903 * On error we might have left some ordered extents
8904 * without submitting corresponding bios for them, so
8905 * cleanup them up to avoid other tasks getting them
8906 * and waiting for them to complete forever.
8908 if (dio_data
.unsubmitted_oe_range_start
<
8909 dio_data
.unsubmitted_oe_range_end
)
8910 __endio_write_update_ordered(inode
,
8911 dio_data
.unsubmitted_oe_range_start
,
8912 dio_data
.unsubmitted_oe_range_end
-
8913 dio_data
.unsubmitted_oe_range_start
,
8915 } else if (ret
>= 0 && (size_t)ret
< count
)
8916 btrfs_delalloc_release_space(inode
, data_reserved
,
8917 offset
, count
- (size_t)ret
);
8921 inode_dio_end(inode
);
8925 extent_changeset_free(data_reserved
);
8929 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8931 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
8932 __u64 start
, __u64 len
)
8936 ret
= fiemap_check_flags(fieinfo
, BTRFS_FIEMAP_FLAGS
);
8940 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
8943 int btrfs_readpage(struct file
*file
, struct page
*page
)
8945 struct extent_io_tree
*tree
;
8946 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8947 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
8950 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
8952 struct extent_io_tree
*tree
;
8953 struct inode
*inode
= page
->mapping
->host
;
8956 if (current
->flags
& PF_MEMALLOC
) {
8957 redirty_page_for_writepage(wbc
, page
);
8963 * If we are under memory pressure we will call this directly from the
8964 * VM, we need to make sure we have the inode referenced for the ordered
8965 * extent. If not just return like we didn't do anything.
8967 if (!igrab(inode
)) {
8968 redirty_page_for_writepage(wbc
, page
);
8969 return AOP_WRITEPAGE_ACTIVATE
;
8971 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8972 ret
= extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
8973 btrfs_add_delayed_iput(inode
);
8977 static int btrfs_writepages(struct address_space
*mapping
,
8978 struct writeback_control
*wbc
)
8980 struct extent_io_tree
*tree
;
8982 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8983 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
8987 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
8988 struct list_head
*pages
, unsigned nr_pages
)
8990 struct extent_io_tree
*tree
;
8991 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8992 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
8995 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8997 struct extent_io_tree
*tree
;
8998 struct extent_map_tree
*map
;
9001 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
9002 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
9003 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
9005 ClearPagePrivate(page
);
9006 set_page_private(page
, 0);
9012 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
9014 if (PageWriteback(page
) || PageDirty(page
))
9016 return __btrfs_releasepage(page
, gfp_flags
);
9019 static void btrfs_invalidatepage(struct page
*page
, unsigned int offset
,
9020 unsigned int length
)
9022 struct inode
*inode
= page
->mapping
->host
;
9023 struct extent_io_tree
*tree
;
9024 struct btrfs_ordered_extent
*ordered
;
9025 struct extent_state
*cached_state
= NULL
;
9026 u64 page_start
= page_offset(page
);
9027 u64 page_end
= page_start
+ PAGE_SIZE
- 1;
9030 int inode_evicting
= inode
->i_state
& I_FREEING
;
9033 * we have the page locked, so new writeback can't start,
9034 * and the dirty bit won't be cleared while we are here.
9036 * Wait for IO on this page so that we can safely clear
9037 * the PagePrivate2 bit and do ordered accounting
9039 wait_on_page_writeback(page
);
9041 tree
= &BTRFS_I(inode
)->io_tree
;
9043 btrfs_releasepage(page
, GFP_NOFS
);
9047 if (!inode_evicting
)
9048 lock_extent_bits(tree
, page_start
, page_end
, &cached_state
);
9051 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), start
,
9052 page_end
- start
+ 1);
9054 end
= min(page_end
, ordered
->file_offset
+ ordered
->len
- 1);
9056 * IO on this page will never be started, so we need
9057 * to account for any ordered extents now
9059 if (!inode_evicting
)
9060 clear_extent_bit(tree
, start
, end
,
9061 EXTENT_DIRTY
| EXTENT_DELALLOC
|
9062 EXTENT_DELALLOC_NEW
|
9063 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
9064 EXTENT_DEFRAG
, 1, 0, &cached_state
,
9067 * whoever cleared the private bit is responsible
9068 * for the finish_ordered_io
9070 if (TestClearPagePrivate2(page
)) {
9071 struct btrfs_ordered_inode_tree
*tree
;
9074 tree
= &BTRFS_I(inode
)->ordered_tree
;
9076 spin_lock_irq(&tree
->lock
);
9077 set_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
);
9078 new_len
= start
- ordered
->file_offset
;
9079 if (new_len
< ordered
->truncated_len
)
9080 ordered
->truncated_len
= new_len
;
9081 spin_unlock_irq(&tree
->lock
);
9083 if (btrfs_dec_test_ordered_pending(inode
, &ordered
,
9085 end
- start
+ 1, 1))
9086 btrfs_finish_ordered_io(ordered
);
9088 btrfs_put_ordered_extent(ordered
);
9089 if (!inode_evicting
) {
9090 cached_state
= NULL
;
9091 lock_extent_bits(tree
, start
, end
,
9096 if (start
< page_end
)
9101 * Qgroup reserved space handler
9102 * Page here will be either
9103 * 1) Already written to disk
9104 * In this case, its reserved space is released from data rsv map
9105 * and will be freed by delayed_ref handler finally.
9106 * So even we call qgroup_free_data(), it won't decrease reserved
9108 * 2) Not written to disk
9109 * This means the reserved space should be freed here. However,
9110 * if a truncate invalidates the page (by clearing PageDirty)
9111 * and the page is accounted for while allocating extent
9112 * in btrfs_check_data_free_space() we let delayed_ref to
9113 * free the entire extent.
9115 if (PageDirty(page
))
9116 btrfs_qgroup_free_data(inode
, NULL
, page_start
, PAGE_SIZE
);
9117 if (!inode_evicting
) {
9118 clear_extent_bit(tree
, page_start
, page_end
,
9119 EXTENT_LOCKED
| EXTENT_DIRTY
|
9120 EXTENT_DELALLOC
| EXTENT_DELALLOC_NEW
|
9121 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
, 1, 1,
9122 &cached_state
, GFP_NOFS
);
9124 __btrfs_releasepage(page
, GFP_NOFS
);
9127 ClearPageChecked(page
);
9128 if (PagePrivate(page
)) {
9129 ClearPagePrivate(page
);
9130 set_page_private(page
, 0);
9136 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
9137 * called from a page fault handler when a page is first dirtied. Hence we must
9138 * be careful to check for EOF conditions here. We set the page up correctly
9139 * for a written page which means we get ENOSPC checking when writing into
9140 * holes and correct delalloc and unwritten extent mapping on filesystems that
9141 * support these features.
9143 * We are not allowed to take the i_mutex here so we have to play games to
9144 * protect against truncate races as the page could now be beyond EOF. Because
9145 * vmtruncate() writes the inode size before removing pages, once we have the
9146 * page lock we can determine safely if the page is beyond EOF. If it is not
9147 * beyond EOF, then the page is guaranteed safe against truncation until we
9150 int btrfs_page_mkwrite(struct vm_fault
*vmf
)
9152 struct page
*page
= vmf
->page
;
9153 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
9154 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
9155 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
9156 struct btrfs_ordered_extent
*ordered
;
9157 struct extent_state
*cached_state
= NULL
;
9158 struct extent_changeset
*data_reserved
= NULL
;
9160 unsigned long zero_start
;
9169 reserved_space
= PAGE_SIZE
;
9171 sb_start_pagefault(inode
->i_sb
);
9172 page_start
= page_offset(page
);
9173 page_end
= page_start
+ PAGE_SIZE
- 1;
9177 * Reserving delalloc space after obtaining the page lock can lead to
9178 * deadlock. For example, if a dirty page is locked by this function
9179 * and the call to btrfs_delalloc_reserve_space() ends up triggering
9180 * dirty page write out, then the btrfs_writepage() function could
9181 * end up waiting indefinitely to get a lock on the page currently
9182 * being processed by btrfs_page_mkwrite() function.
9184 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
, page_start
,
9187 ret
= file_update_time(vmf
->vma
->vm_file
);
9193 else /* -ENOSPC, -EIO, etc */
9194 ret
= VM_FAULT_SIGBUS
;
9200 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
9203 size
= i_size_read(inode
);
9205 if ((page
->mapping
!= inode
->i_mapping
) ||
9206 (page_start
>= size
)) {
9207 /* page got truncated out from underneath us */
9210 wait_on_page_writeback(page
);
9212 lock_extent_bits(io_tree
, page_start
, page_end
, &cached_state
);
9213 set_page_extent_mapped(page
);
9216 * we can't set the delalloc bits if there are pending ordered
9217 * extents. Drop our locks and wait for them to finish
9219 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), page_start
,
9222 unlock_extent_cached(io_tree
, page_start
, page_end
,
9223 &cached_state
, GFP_NOFS
);
9225 btrfs_start_ordered_extent(inode
, ordered
, 1);
9226 btrfs_put_ordered_extent(ordered
);
9230 if (page
->index
== ((size
- 1) >> PAGE_SHIFT
)) {
9231 reserved_space
= round_up(size
- page_start
,
9232 fs_info
->sectorsize
);
9233 if (reserved_space
< PAGE_SIZE
) {
9234 end
= page_start
+ reserved_space
- 1;
9235 spin_lock(&BTRFS_I(inode
)->lock
);
9236 BTRFS_I(inode
)->outstanding_extents
++;
9237 spin_unlock(&BTRFS_I(inode
)->lock
);
9238 btrfs_delalloc_release_space(inode
, data_reserved
,
9239 page_start
, PAGE_SIZE
- reserved_space
);
9244 * page_mkwrite gets called when the page is firstly dirtied after it's
9245 * faulted in, but write(2) could also dirty a page and set delalloc
9246 * bits, thus in this case for space account reason, we still need to
9247 * clear any delalloc bits within this page range since we have to
9248 * reserve data&meta space before lock_page() (see above comments).
9250 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, end
,
9251 EXTENT_DIRTY
| EXTENT_DELALLOC
|
9252 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
9253 0, 0, &cached_state
, GFP_NOFS
);
9255 ret
= btrfs_set_extent_delalloc(inode
, page_start
, end
,
9258 unlock_extent_cached(io_tree
, page_start
, page_end
,
9259 &cached_state
, GFP_NOFS
);
9260 ret
= VM_FAULT_SIGBUS
;
9265 /* page is wholly or partially inside EOF */
9266 if (page_start
+ PAGE_SIZE
> size
)
9267 zero_start
= size
& ~PAGE_MASK
;
9269 zero_start
= PAGE_SIZE
;
9271 if (zero_start
!= PAGE_SIZE
) {
9273 memset(kaddr
+ zero_start
, 0, PAGE_SIZE
- zero_start
);
9274 flush_dcache_page(page
);
9277 ClearPageChecked(page
);
9278 set_page_dirty(page
);
9279 SetPageUptodate(page
);
9281 BTRFS_I(inode
)->last_trans
= fs_info
->generation
;
9282 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
9283 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
9285 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
9289 sb_end_pagefault(inode
->i_sb
);
9290 extent_changeset_free(data_reserved
);
9291 return VM_FAULT_LOCKED
;
9295 btrfs_delalloc_release_space(inode
, data_reserved
, page_start
,
9298 sb_end_pagefault(inode
->i_sb
);
9299 extent_changeset_free(data_reserved
);
9303 static int btrfs_truncate(struct inode
*inode
)
9305 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
9306 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9307 struct btrfs_block_rsv
*rsv
;
9310 struct btrfs_trans_handle
*trans
;
9311 u64 mask
= fs_info
->sectorsize
- 1;
9312 u64 min_size
= btrfs_calc_trunc_metadata_size(fs_info
, 1);
9314 ret
= btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
),
9320 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9321 * 3 things going on here
9323 * 1) We need to reserve space for our orphan item and the space to
9324 * delete our orphan item. Lord knows we don't want to have a dangling
9325 * orphan item because we didn't reserve space to remove it.
9327 * 2) We need to reserve space to update our inode.
9329 * 3) We need to have something to cache all the space that is going to
9330 * be free'd up by the truncate operation, but also have some slack
9331 * space reserved in case it uses space during the truncate (thank you
9332 * very much snapshotting).
9334 * And we need these to all be separate. The fact is we can use a lot of
9335 * space doing the truncate, and we have no earthly idea how much space
9336 * we will use, so we need the truncate reservation to be separate so it
9337 * doesn't end up using space reserved for updating the inode or
9338 * removing the orphan item. We also need to be able to stop the
9339 * transaction and start a new one, which means we need to be able to
9340 * update the inode several times, and we have no idea of knowing how
9341 * many times that will be, so we can't just reserve 1 item for the
9342 * entirety of the operation, so that has to be done separately as well.
9343 * Then there is the orphan item, which does indeed need to be held on
9344 * to for the whole operation, and we need nobody to touch this reserved
9345 * space except the orphan code.
9347 * So that leaves us with
9349 * 1) root->orphan_block_rsv - for the orphan deletion.
9350 * 2) rsv - for the truncate reservation, which we will steal from the
9351 * transaction reservation.
9352 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9353 * updating the inode.
9355 rsv
= btrfs_alloc_block_rsv(fs_info
, BTRFS_BLOCK_RSV_TEMP
);
9358 rsv
->size
= min_size
;
9362 * 1 for the truncate slack space
9363 * 1 for updating the inode.
9365 trans
= btrfs_start_transaction(root
, 2);
9366 if (IS_ERR(trans
)) {
9367 err
= PTR_ERR(trans
);
9371 /* Migrate the slack space for the truncate to our reserve */
9372 ret
= btrfs_block_rsv_migrate(&fs_info
->trans_block_rsv
, rsv
,
9377 * So if we truncate and then write and fsync we normally would just
9378 * write the extents that changed, which is a problem if we need to
9379 * first truncate that entire inode. So set this flag so we write out
9380 * all of the extents in the inode to the sync log so we're completely
9383 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
9384 trans
->block_rsv
= rsv
;
9387 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
9389 BTRFS_EXTENT_DATA_KEY
);
9390 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
) {
9395 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
9396 ret
= btrfs_update_inode(trans
, root
, inode
);
9402 btrfs_end_transaction(trans
);
9403 btrfs_btree_balance_dirty(fs_info
);
9405 trans
= btrfs_start_transaction(root
, 2);
9406 if (IS_ERR(trans
)) {
9407 ret
= err
= PTR_ERR(trans
);
9412 btrfs_block_rsv_release(fs_info
, rsv
, -1);
9413 ret
= btrfs_block_rsv_migrate(&fs_info
->trans_block_rsv
,
9415 BUG_ON(ret
); /* shouldn't happen */
9416 trans
->block_rsv
= rsv
;
9419 if (ret
== 0 && inode
->i_nlink
> 0) {
9420 trans
->block_rsv
= root
->orphan_block_rsv
;
9421 ret
= btrfs_orphan_del(trans
, BTRFS_I(inode
));
9427 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
9428 ret
= btrfs_update_inode(trans
, root
, inode
);
9432 ret
= btrfs_end_transaction(trans
);
9433 btrfs_btree_balance_dirty(fs_info
);
9436 btrfs_free_block_rsv(fs_info
, rsv
);
9445 * create a new subvolume directory/inode (helper for the ioctl).
9447 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
9448 struct btrfs_root
*new_root
,
9449 struct btrfs_root
*parent_root
,
9452 struct inode
*inode
;
9456 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
9457 new_dirid
, new_dirid
,
9458 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
9461 return PTR_ERR(inode
);
9462 inode
->i_op
= &btrfs_dir_inode_operations
;
9463 inode
->i_fop
= &btrfs_dir_file_operations
;
9465 set_nlink(inode
, 1);
9466 btrfs_i_size_write(BTRFS_I(inode
), 0);
9467 unlock_new_inode(inode
);
9469 err
= btrfs_subvol_inherit_props(trans
, new_root
, parent_root
);
9471 btrfs_err(new_root
->fs_info
,
9472 "error inheriting subvolume %llu properties: %d",
9473 new_root
->root_key
.objectid
, err
);
9475 err
= btrfs_update_inode(trans
, new_root
, inode
);
9481 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
9483 struct btrfs_inode
*ei
;
9484 struct inode
*inode
;
9486 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
9493 ei
->last_sub_trans
= 0;
9494 ei
->logged_trans
= 0;
9495 ei
->delalloc_bytes
= 0;
9496 ei
->new_delalloc_bytes
= 0;
9497 ei
->defrag_bytes
= 0;
9498 ei
->disk_i_size
= 0;
9501 ei
->index_cnt
= (u64
)-1;
9503 ei
->last_unlink_trans
= 0;
9504 ei
->last_log_commit
= 0;
9505 ei
->delayed_iput_count
= 0;
9507 spin_lock_init(&ei
->lock
);
9508 ei
->outstanding_extents
= 0;
9509 ei
->reserved_extents
= 0;
9511 ei
->runtime_flags
= 0;
9512 ei
->prop_compress
= BTRFS_COMPRESS_NONE
;
9513 ei
->defrag_compress
= BTRFS_COMPRESS_NONE
;
9515 ei
->delayed_node
= NULL
;
9517 ei
->i_otime
.tv_sec
= 0;
9518 ei
->i_otime
.tv_nsec
= 0;
9520 inode
= &ei
->vfs_inode
;
9521 extent_map_tree_init(&ei
->extent_tree
);
9522 extent_io_tree_init(&ei
->io_tree
, inode
);
9523 extent_io_tree_init(&ei
->io_failure_tree
, inode
);
9524 ei
->io_tree
.track_uptodate
= 1;
9525 ei
->io_failure_tree
.track_uptodate
= 1;
9526 atomic_set(&ei
->sync_writers
, 0);
9527 mutex_init(&ei
->log_mutex
);
9528 mutex_init(&ei
->delalloc_mutex
);
9529 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
9530 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
9531 INIT_LIST_HEAD(&ei
->delayed_iput
);
9532 RB_CLEAR_NODE(&ei
->rb_node
);
9533 init_rwsem(&ei
->dio_sem
);
9538 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9539 void btrfs_test_destroy_inode(struct inode
*inode
)
9541 btrfs_drop_extent_cache(BTRFS_I(inode
), 0, (u64
)-1, 0);
9542 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9546 static void btrfs_i_callback(struct rcu_head
*head
)
9548 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
9549 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9552 void btrfs_destroy_inode(struct inode
*inode
)
9554 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
9555 struct btrfs_ordered_extent
*ordered
;
9556 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9558 WARN_ON(!hlist_empty(&inode
->i_dentry
));
9559 WARN_ON(inode
->i_data
.nrpages
);
9560 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
9561 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
9562 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
9563 WARN_ON(BTRFS_I(inode
)->new_delalloc_bytes
);
9564 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
9565 WARN_ON(BTRFS_I(inode
)->defrag_bytes
);
9568 * This can happen where we create an inode, but somebody else also
9569 * created the same inode and we need to destroy the one we already
9575 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
9576 &BTRFS_I(inode
)->runtime_flags
)) {
9577 btrfs_info(fs_info
, "inode %llu still on the orphan list",
9578 btrfs_ino(BTRFS_I(inode
)));
9579 atomic_dec(&root
->orphan_inodes
);
9583 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
9588 "found ordered extent %llu %llu on inode cleanup",
9589 ordered
->file_offset
, ordered
->len
);
9590 btrfs_remove_ordered_extent(inode
, ordered
);
9591 btrfs_put_ordered_extent(ordered
);
9592 btrfs_put_ordered_extent(ordered
);
9595 btrfs_qgroup_check_reserved_leak(inode
);
9596 inode_tree_del(inode
);
9597 btrfs_drop_extent_cache(BTRFS_I(inode
), 0, (u64
)-1, 0);
9599 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
9602 int btrfs_drop_inode(struct inode
*inode
)
9604 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9609 /* the snap/subvol tree is on deleting */
9610 if (btrfs_root_refs(&root
->root_item
) == 0)
9613 return generic_drop_inode(inode
);
9616 static void init_once(void *foo
)
9618 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
9620 inode_init_once(&ei
->vfs_inode
);
9623 void btrfs_destroy_cachep(void)
9626 * Make sure all delayed rcu free inodes are flushed before we
9630 kmem_cache_destroy(btrfs_inode_cachep
);
9631 kmem_cache_destroy(btrfs_trans_handle_cachep
);
9632 kmem_cache_destroy(btrfs_path_cachep
);
9633 kmem_cache_destroy(btrfs_free_space_cachep
);
9636 int btrfs_init_cachep(void)
9638 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
9639 sizeof(struct btrfs_inode
), 0,
9640 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
| SLAB_ACCOUNT
,
9642 if (!btrfs_inode_cachep
)
9645 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
9646 sizeof(struct btrfs_trans_handle
), 0,
9647 SLAB_TEMPORARY
| SLAB_MEM_SPREAD
, NULL
);
9648 if (!btrfs_trans_handle_cachep
)
9651 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
9652 sizeof(struct btrfs_path
), 0,
9653 SLAB_MEM_SPREAD
, NULL
);
9654 if (!btrfs_path_cachep
)
9657 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
9658 sizeof(struct btrfs_free_space
), 0,
9659 SLAB_MEM_SPREAD
, NULL
);
9660 if (!btrfs_free_space_cachep
)
9665 btrfs_destroy_cachep();
9669 static int btrfs_getattr(const struct path
*path
, struct kstat
*stat
,
9670 u32 request_mask
, unsigned int flags
)
9673 struct inode
*inode
= d_inode(path
->dentry
);
9674 u32 blocksize
= inode
->i_sb
->s_blocksize
;
9675 u32 bi_flags
= BTRFS_I(inode
)->flags
;
9677 stat
->result_mask
|= STATX_BTIME
;
9678 stat
->btime
.tv_sec
= BTRFS_I(inode
)->i_otime
.tv_sec
;
9679 stat
->btime
.tv_nsec
= BTRFS_I(inode
)->i_otime
.tv_nsec
;
9680 if (bi_flags
& BTRFS_INODE_APPEND
)
9681 stat
->attributes
|= STATX_ATTR_APPEND
;
9682 if (bi_flags
& BTRFS_INODE_COMPRESS
)
9683 stat
->attributes
|= STATX_ATTR_COMPRESSED
;
9684 if (bi_flags
& BTRFS_INODE_IMMUTABLE
)
9685 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
9686 if (bi_flags
& BTRFS_INODE_NODUMP
)
9687 stat
->attributes
|= STATX_ATTR_NODUMP
;
9689 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
9690 STATX_ATTR_COMPRESSED
|
9691 STATX_ATTR_IMMUTABLE
|
9694 generic_fillattr(inode
, stat
);
9695 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
9697 spin_lock(&BTRFS_I(inode
)->lock
);
9698 delalloc_bytes
= BTRFS_I(inode
)->new_delalloc_bytes
;
9699 spin_unlock(&BTRFS_I(inode
)->lock
);
9700 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
9701 ALIGN(delalloc_bytes
, blocksize
)) >> 9;
9705 static int btrfs_rename_exchange(struct inode
*old_dir
,
9706 struct dentry
*old_dentry
,
9707 struct inode
*new_dir
,
9708 struct dentry
*new_dentry
)
9710 struct btrfs_fs_info
*fs_info
= btrfs_sb(old_dir
->i_sb
);
9711 struct btrfs_trans_handle
*trans
;
9712 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9713 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9714 struct inode
*new_inode
= new_dentry
->d_inode
;
9715 struct inode
*old_inode
= old_dentry
->d_inode
;
9716 struct timespec ctime
= current_time(old_inode
);
9717 struct dentry
*parent
;
9718 u64 old_ino
= btrfs_ino(BTRFS_I(old_inode
));
9719 u64 new_ino
= btrfs_ino(BTRFS_I(new_inode
));
9724 bool root_log_pinned
= false;
9725 bool dest_log_pinned
= false;
9727 /* we only allow rename subvolume link between subvolumes */
9728 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9731 /* close the race window with snapshot create/destroy ioctl */
9732 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9733 down_read(&fs_info
->subvol_sem
);
9734 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9735 down_read(&fs_info
->subvol_sem
);
9738 * We want to reserve the absolute worst case amount of items. So if
9739 * both inodes are subvols and we need to unlink them then that would
9740 * require 4 item modifications, but if they are both normal inodes it
9741 * would require 5 item modifications, so we'll assume their normal
9742 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9743 * should cover the worst case number of items we'll modify.
9745 trans
= btrfs_start_transaction(root
, 12);
9746 if (IS_ERR(trans
)) {
9747 ret
= PTR_ERR(trans
);
9752 * We need to find a free sequence number both in the source and
9753 * in the destination directory for the exchange.
9755 ret
= btrfs_set_inode_index(BTRFS_I(new_dir
), &old_idx
);
9758 ret
= btrfs_set_inode_index(BTRFS_I(old_dir
), &new_idx
);
9762 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9763 BTRFS_I(new_inode
)->dir_index
= 0ULL;
9765 /* Reference for the source. */
9766 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9767 /* force full log commit if subvolume involved. */
9768 btrfs_set_log_full_commit(fs_info
, trans
);
9770 btrfs_pin_log_trans(root
);
9771 root_log_pinned
= true;
9772 ret
= btrfs_insert_inode_ref(trans
, dest
,
9773 new_dentry
->d_name
.name
,
9774 new_dentry
->d_name
.len
,
9776 btrfs_ino(BTRFS_I(new_dir
)),
9782 /* And now for the dest. */
9783 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9784 /* force full log commit if subvolume involved. */
9785 btrfs_set_log_full_commit(fs_info
, trans
);
9787 btrfs_pin_log_trans(dest
);
9788 dest_log_pinned
= true;
9789 ret
= btrfs_insert_inode_ref(trans
, root
,
9790 old_dentry
->d_name
.name
,
9791 old_dentry
->d_name
.len
,
9793 btrfs_ino(BTRFS_I(old_dir
)),
9799 /* Update inode version and ctime/mtime. */
9800 inode_inc_iversion(old_dir
);
9801 inode_inc_iversion(new_dir
);
9802 inode_inc_iversion(old_inode
);
9803 inode_inc_iversion(new_inode
);
9804 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
9805 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
9806 old_inode
->i_ctime
= ctime
;
9807 new_inode
->i_ctime
= ctime
;
9809 if (old_dentry
->d_parent
!= new_dentry
->d_parent
) {
9810 btrfs_record_unlink_dir(trans
, BTRFS_I(old_dir
),
9811 BTRFS_I(old_inode
), 1);
9812 btrfs_record_unlink_dir(trans
, BTRFS_I(new_dir
),
9813 BTRFS_I(new_inode
), 1);
9816 /* src is a subvolume */
9817 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9818 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
9819 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
,
9821 old_dentry
->d_name
.name
,
9822 old_dentry
->d_name
.len
);
9823 } else { /* src is an inode */
9824 ret
= __btrfs_unlink_inode(trans
, root
, BTRFS_I(old_dir
),
9825 BTRFS_I(old_dentry
->d_inode
),
9826 old_dentry
->d_name
.name
,
9827 old_dentry
->d_name
.len
);
9829 ret
= btrfs_update_inode(trans
, root
, old_inode
);
9832 btrfs_abort_transaction(trans
, ret
);
9836 /* dest is a subvolume */
9837 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9838 root_objectid
= BTRFS_I(new_inode
)->root
->root_key
.objectid
;
9839 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
9841 new_dentry
->d_name
.name
,
9842 new_dentry
->d_name
.len
);
9843 } else { /* dest is an inode */
9844 ret
= __btrfs_unlink_inode(trans
, dest
, BTRFS_I(new_dir
),
9845 BTRFS_I(new_dentry
->d_inode
),
9846 new_dentry
->d_name
.name
,
9847 new_dentry
->d_name
.len
);
9849 ret
= btrfs_update_inode(trans
, dest
, new_inode
);
9852 btrfs_abort_transaction(trans
, ret
);
9856 ret
= btrfs_add_link(trans
, BTRFS_I(new_dir
), BTRFS_I(old_inode
),
9857 new_dentry
->d_name
.name
,
9858 new_dentry
->d_name
.len
, 0, old_idx
);
9860 btrfs_abort_transaction(trans
, ret
);
9864 ret
= btrfs_add_link(trans
, BTRFS_I(old_dir
), BTRFS_I(new_inode
),
9865 old_dentry
->d_name
.name
,
9866 old_dentry
->d_name
.len
, 0, new_idx
);
9868 btrfs_abort_transaction(trans
, ret
);
9872 if (old_inode
->i_nlink
== 1)
9873 BTRFS_I(old_inode
)->dir_index
= old_idx
;
9874 if (new_inode
->i_nlink
== 1)
9875 BTRFS_I(new_inode
)->dir_index
= new_idx
;
9877 if (root_log_pinned
) {
9878 parent
= new_dentry
->d_parent
;
9879 btrfs_log_new_name(trans
, BTRFS_I(old_inode
), BTRFS_I(old_dir
),
9881 btrfs_end_log_trans(root
);
9882 root_log_pinned
= false;
9884 if (dest_log_pinned
) {
9885 parent
= old_dentry
->d_parent
;
9886 btrfs_log_new_name(trans
, BTRFS_I(new_inode
), BTRFS_I(new_dir
),
9888 btrfs_end_log_trans(dest
);
9889 dest_log_pinned
= false;
9893 * If we have pinned a log and an error happened, we unpin tasks
9894 * trying to sync the log and force them to fallback to a transaction
9895 * commit if the log currently contains any of the inodes involved in
9896 * this rename operation (to ensure we do not persist a log with an
9897 * inconsistent state for any of these inodes or leading to any
9898 * inconsistencies when replayed). If the transaction was aborted, the
9899 * abortion reason is propagated to userspace when attempting to commit
9900 * the transaction. If the log does not contain any of these inodes, we
9901 * allow the tasks to sync it.
9903 if (ret
&& (root_log_pinned
|| dest_log_pinned
)) {
9904 if (btrfs_inode_in_log(BTRFS_I(old_dir
), fs_info
->generation
) ||
9905 btrfs_inode_in_log(BTRFS_I(new_dir
), fs_info
->generation
) ||
9906 btrfs_inode_in_log(BTRFS_I(old_inode
), fs_info
->generation
) ||
9908 btrfs_inode_in_log(BTRFS_I(new_inode
), fs_info
->generation
)))
9909 btrfs_set_log_full_commit(fs_info
, trans
);
9911 if (root_log_pinned
) {
9912 btrfs_end_log_trans(root
);
9913 root_log_pinned
= false;
9915 if (dest_log_pinned
) {
9916 btrfs_end_log_trans(dest
);
9917 dest_log_pinned
= false;
9920 ret
= btrfs_end_transaction(trans
);
9922 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9923 up_read(&fs_info
->subvol_sem
);
9924 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9925 up_read(&fs_info
->subvol_sem
);
9930 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle
*trans
,
9931 struct btrfs_root
*root
,
9933 struct dentry
*dentry
)
9936 struct inode
*inode
;
9940 ret
= btrfs_find_free_ino(root
, &objectid
);
9944 inode
= btrfs_new_inode(trans
, root
, dir
,
9945 dentry
->d_name
.name
,
9947 btrfs_ino(BTRFS_I(dir
)),
9949 S_IFCHR
| WHITEOUT_MODE
,
9952 if (IS_ERR(inode
)) {
9953 ret
= PTR_ERR(inode
);
9957 inode
->i_op
= &btrfs_special_inode_operations
;
9958 init_special_inode(inode
, inode
->i_mode
,
9961 ret
= btrfs_init_inode_security(trans
, inode
, dir
,
9966 ret
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
,
9967 BTRFS_I(inode
), 0, index
);
9971 ret
= btrfs_update_inode(trans
, root
, inode
);
9973 unlock_new_inode(inode
);
9975 inode_dec_link_count(inode
);
9981 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
9982 struct inode
*new_dir
, struct dentry
*new_dentry
,
9985 struct btrfs_fs_info
*fs_info
= btrfs_sb(old_dir
->i_sb
);
9986 struct btrfs_trans_handle
*trans
;
9987 unsigned int trans_num_items
;
9988 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9989 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9990 struct inode
*new_inode
= d_inode(new_dentry
);
9991 struct inode
*old_inode
= d_inode(old_dentry
);
9995 u64 old_ino
= btrfs_ino(BTRFS_I(old_inode
));
9996 bool log_pinned
= false;
9998 if (btrfs_ino(BTRFS_I(new_dir
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
10001 /* we only allow rename subvolume link between subvolumes */
10002 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
10005 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
10006 (new_inode
&& btrfs_ino(BTRFS_I(new_inode
)) == BTRFS_FIRST_FREE_OBJECTID
))
10009 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
10010 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
10014 /* check for collisions, even if the name isn't there */
10015 ret
= btrfs_check_dir_item_collision(dest
, new_dir
->i_ino
,
10016 new_dentry
->d_name
.name
,
10017 new_dentry
->d_name
.len
);
10020 if (ret
== -EEXIST
) {
10021 /* we shouldn't get
10022 * eexist without a new_inode */
10023 if (WARN_ON(!new_inode
)) {
10027 /* maybe -EOVERFLOW */
10034 * we're using rename to replace one file with another. Start IO on it
10035 * now so we don't add too much work to the end of the transaction
10037 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
)
10038 filemap_flush(old_inode
->i_mapping
);
10040 /* close the racy window with snapshot create/destroy ioctl */
10041 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
10042 down_read(&fs_info
->subvol_sem
);
10044 * We want to reserve the absolute worst case amount of items. So if
10045 * both inodes are subvols and we need to unlink them then that would
10046 * require 4 item modifications, but if they are both normal inodes it
10047 * would require 5 item modifications, so we'll assume they are normal
10048 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
10049 * should cover the worst case number of items we'll modify.
10050 * If our rename has the whiteout flag, we need more 5 units for the
10051 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
10052 * when selinux is enabled).
10054 trans_num_items
= 11;
10055 if (flags
& RENAME_WHITEOUT
)
10056 trans_num_items
+= 5;
10057 trans
= btrfs_start_transaction(root
, trans_num_items
);
10058 if (IS_ERR(trans
)) {
10059 ret
= PTR_ERR(trans
);
10064 btrfs_record_root_in_trans(trans
, dest
);
10066 ret
= btrfs_set_inode_index(BTRFS_I(new_dir
), &index
);
10070 BTRFS_I(old_inode
)->dir_index
= 0ULL;
10071 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
10072 /* force full log commit if subvolume involved. */
10073 btrfs_set_log_full_commit(fs_info
, trans
);
10075 btrfs_pin_log_trans(root
);
10077 ret
= btrfs_insert_inode_ref(trans
, dest
,
10078 new_dentry
->d_name
.name
,
10079 new_dentry
->d_name
.len
,
10081 btrfs_ino(BTRFS_I(new_dir
)), index
);
10086 inode_inc_iversion(old_dir
);
10087 inode_inc_iversion(new_dir
);
10088 inode_inc_iversion(old_inode
);
10089 old_dir
->i_ctime
= old_dir
->i_mtime
=
10090 new_dir
->i_ctime
= new_dir
->i_mtime
=
10091 old_inode
->i_ctime
= current_time(old_dir
);
10093 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
10094 btrfs_record_unlink_dir(trans
, BTRFS_I(old_dir
),
10095 BTRFS_I(old_inode
), 1);
10097 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
10098 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
10099 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
10100 old_dentry
->d_name
.name
,
10101 old_dentry
->d_name
.len
);
10103 ret
= __btrfs_unlink_inode(trans
, root
, BTRFS_I(old_dir
),
10104 BTRFS_I(d_inode(old_dentry
)),
10105 old_dentry
->d_name
.name
,
10106 old_dentry
->d_name
.len
);
10108 ret
= btrfs_update_inode(trans
, root
, old_inode
);
10111 btrfs_abort_transaction(trans
, ret
);
10116 inode_inc_iversion(new_inode
);
10117 new_inode
->i_ctime
= current_time(new_inode
);
10118 if (unlikely(btrfs_ino(BTRFS_I(new_inode
)) ==
10119 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
10120 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
10121 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
10123 new_dentry
->d_name
.name
,
10124 new_dentry
->d_name
.len
);
10125 BUG_ON(new_inode
->i_nlink
== 0);
10127 ret
= btrfs_unlink_inode(trans
, dest
, BTRFS_I(new_dir
),
10128 BTRFS_I(d_inode(new_dentry
)),
10129 new_dentry
->d_name
.name
,
10130 new_dentry
->d_name
.len
);
10132 if (!ret
&& new_inode
->i_nlink
== 0)
10133 ret
= btrfs_orphan_add(trans
,
10134 BTRFS_I(d_inode(new_dentry
)));
10136 btrfs_abort_transaction(trans
, ret
);
10141 ret
= btrfs_add_link(trans
, BTRFS_I(new_dir
), BTRFS_I(old_inode
),
10142 new_dentry
->d_name
.name
,
10143 new_dentry
->d_name
.len
, 0, index
);
10145 btrfs_abort_transaction(trans
, ret
);
10149 if (old_inode
->i_nlink
== 1)
10150 BTRFS_I(old_inode
)->dir_index
= index
;
10153 struct dentry
*parent
= new_dentry
->d_parent
;
10155 btrfs_log_new_name(trans
, BTRFS_I(old_inode
), BTRFS_I(old_dir
),
10157 btrfs_end_log_trans(root
);
10158 log_pinned
= false;
10161 if (flags
& RENAME_WHITEOUT
) {
10162 ret
= btrfs_whiteout_for_rename(trans
, root
, old_dir
,
10166 btrfs_abort_transaction(trans
, ret
);
10172 * If we have pinned the log and an error happened, we unpin tasks
10173 * trying to sync the log and force them to fallback to a transaction
10174 * commit if the log currently contains any of the inodes involved in
10175 * this rename operation (to ensure we do not persist a log with an
10176 * inconsistent state for any of these inodes or leading to any
10177 * inconsistencies when replayed). If the transaction was aborted, the
10178 * abortion reason is propagated to userspace when attempting to commit
10179 * the transaction. If the log does not contain any of these inodes, we
10180 * allow the tasks to sync it.
10182 if (ret
&& log_pinned
) {
10183 if (btrfs_inode_in_log(BTRFS_I(old_dir
), fs_info
->generation
) ||
10184 btrfs_inode_in_log(BTRFS_I(new_dir
), fs_info
->generation
) ||
10185 btrfs_inode_in_log(BTRFS_I(old_inode
), fs_info
->generation
) ||
10187 btrfs_inode_in_log(BTRFS_I(new_inode
), fs_info
->generation
)))
10188 btrfs_set_log_full_commit(fs_info
, trans
);
10190 btrfs_end_log_trans(root
);
10191 log_pinned
= false;
10193 btrfs_end_transaction(trans
);
10195 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
10196 up_read(&fs_info
->subvol_sem
);
10201 static int btrfs_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
10202 struct inode
*new_dir
, struct dentry
*new_dentry
,
10203 unsigned int flags
)
10205 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
10208 if (flags
& RENAME_EXCHANGE
)
10209 return btrfs_rename_exchange(old_dir
, old_dentry
, new_dir
,
10212 return btrfs_rename(old_dir
, old_dentry
, new_dir
, new_dentry
, flags
);
10215 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
10217 struct btrfs_delalloc_work
*delalloc_work
;
10218 struct inode
*inode
;
10220 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
10222 inode
= delalloc_work
->inode
;
10223 filemap_flush(inode
->i_mapping
);
10224 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
10225 &BTRFS_I(inode
)->runtime_flags
))
10226 filemap_flush(inode
->i_mapping
);
10228 if (delalloc_work
->delay_iput
)
10229 btrfs_add_delayed_iput(inode
);
10232 complete(&delalloc_work
->completion
);
10235 struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
,
10238 struct btrfs_delalloc_work
*work
;
10240 work
= kmalloc(sizeof(*work
), GFP_NOFS
);
10244 init_completion(&work
->completion
);
10245 INIT_LIST_HEAD(&work
->list
);
10246 work
->inode
= inode
;
10247 work
->delay_iput
= delay_iput
;
10248 WARN_ON_ONCE(!inode
);
10249 btrfs_init_work(&work
->work
, btrfs_flush_delalloc_helper
,
10250 btrfs_run_delalloc_work
, NULL
, NULL
);
10255 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work
*work
)
10257 wait_for_completion(&work
->completion
);
10262 * some fairly slow code that needs optimization. This walks the list
10263 * of all the inodes with pending delalloc and forces them to disk.
10265 static int __start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
,
10268 struct btrfs_inode
*binode
;
10269 struct inode
*inode
;
10270 struct btrfs_delalloc_work
*work
, *next
;
10271 struct list_head works
;
10272 struct list_head splice
;
10275 INIT_LIST_HEAD(&works
);
10276 INIT_LIST_HEAD(&splice
);
10278 mutex_lock(&root
->delalloc_mutex
);
10279 spin_lock(&root
->delalloc_lock
);
10280 list_splice_init(&root
->delalloc_inodes
, &splice
);
10281 while (!list_empty(&splice
)) {
10282 binode
= list_entry(splice
.next
, struct btrfs_inode
,
10285 list_move_tail(&binode
->delalloc_inodes
,
10286 &root
->delalloc_inodes
);
10287 inode
= igrab(&binode
->vfs_inode
);
10289 cond_resched_lock(&root
->delalloc_lock
);
10292 spin_unlock(&root
->delalloc_lock
);
10294 work
= btrfs_alloc_delalloc_work(inode
, delay_iput
);
10297 btrfs_add_delayed_iput(inode
);
10303 list_add_tail(&work
->list
, &works
);
10304 btrfs_queue_work(root
->fs_info
->flush_workers
,
10307 if (nr
!= -1 && ret
>= nr
)
10310 spin_lock(&root
->delalloc_lock
);
10312 spin_unlock(&root
->delalloc_lock
);
10315 list_for_each_entry_safe(work
, next
, &works
, list
) {
10316 list_del_init(&work
->list
);
10317 btrfs_wait_and_free_delalloc_work(work
);
10320 if (!list_empty_careful(&splice
)) {
10321 spin_lock(&root
->delalloc_lock
);
10322 list_splice_tail(&splice
, &root
->delalloc_inodes
);
10323 spin_unlock(&root
->delalloc_lock
);
10325 mutex_unlock(&root
->delalloc_mutex
);
10329 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
10331 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
10334 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
10337 ret
= __start_delalloc_inodes(root
, delay_iput
, -1);
10341 * the filemap_flush will queue IO into the worker threads, but
10342 * we have to make sure the IO is actually started and that
10343 * ordered extents get created before we return
10345 atomic_inc(&fs_info
->async_submit_draining
);
10346 while (atomic_read(&fs_info
->nr_async_submits
) ||
10347 atomic_read(&fs_info
->async_delalloc_pages
)) {
10348 wait_event(fs_info
->async_submit_wait
,
10349 (atomic_read(&fs_info
->nr_async_submits
) == 0 &&
10350 atomic_read(&fs_info
->async_delalloc_pages
) == 0));
10352 atomic_dec(&fs_info
->async_submit_draining
);
10356 int btrfs_start_delalloc_roots(struct btrfs_fs_info
*fs_info
, int delay_iput
,
10359 struct btrfs_root
*root
;
10360 struct list_head splice
;
10363 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
10366 INIT_LIST_HEAD(&splice
);
10368 mutex_lock(&fs_info
->delalloc_root_mutex
);
10369 spin_lock(&fs_info
->delalloc_root_lock
);
10370 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
10371 while (!list_empty(&splice
) && nr
) {
10372 root
= list_first_entry(&splice
, struct btrfs_root
,
10374 root
= btrfs_grab_fs_root(root
);
10376 list_move_tail(&root
->delalloc_root
,
10377 &fs_info
->delalloc_roots
);
10378 spin_unlock(&fs_info
->delalloc_root_lock
);
10380 ret
= __start_delalloc_inodes(root
, delay_iput
, nr
);
10381 btrfs_put_fs_root(root
);
10389 spin_lock(&fs_info
->delalloc_root_lock
);
10391 spin_unlock(&fs_info
->delalloc_root_lock
);
10394 atomic_inc(&fs_info
->async_submit_draining
);
10395 while (atomic_read(&fs_info
->nr_async_submits
) ||
10396 atomic_read(&fs_info
->async_delalloc_pages
)) {
10397 wait_event(fs_info
->async_submit_wait
,
10398 (atomic_read(&fs_info
->nr_async_submits
) == 0 &&
10399 atomic_read(&fs_info
->async_delalloc_pages
) == 0));
10401 atomic_dec(&fs_info
->async_submit_draining
);
10403 if (!list_empty_careful(&splice
)) {
10404 spin_lock(&fs_info
->delalloc_root_lock
);
10405 list_splice_tail(&splice
, &fs_info
->delalloc_roots
);
10406 spin_unlock(&fs_info
->delalloc_root_lock
);
10408 mutex_unlock(&fs_info
->delalloc_root_mutex
);
10412 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
10413 const char *symname
)
10415 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
10416 struct btrfs_trans_handle
*trans
;
10417 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
10418 struct btrfs_path
*path
;
10419 struct btrfs_key key
;
10420 struct inode
*inode
= NULL
;
10422 int drop_inode
= 0;
10428 struct btrfs_file_extent_item
*ei
;
10429 struct extent_buffer
*leaf
;
10431 name_len
= strlen(symname
);
10432 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(fs_info
))
10433 return -ENAMETOOLONG
;
10436 * 2 items for inode item and ref
10437 * 2 items for dir items
10438 * 1 item for updating parent inode item
10439 * 1 item for the inline extent item
10440 * 1 item for xattr if selinux is on
10442 trans
= btrfs_start_transaction(root
, 7);
10444 return PTR_ERR(trans
);
10446 err
= btrfs_find_free_ino(root
, &objectid
);
10450 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
10451 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)),
10452 objectid
, S_IFLNK
|S_IRWXUGO
, &index
);
10453 if (IS_ERR(inode
)) {
10454 err
= PTR_ERR(inode
);
10459 * If the active LSM wants to access the inode during
10460 * d_instantiate it needs these. Smack checks to see
10461 * if the filesystem supports xattrs by looking at the
10464 inode
->i_fop
= &btrfs_file_operations
;
10465 inode
->i_op
= &btrfs_file_inode_operations
;
10466 inode
->i_mapping
->a_ops
= &btrfs_aops
;
10467 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
10469 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
10471 goto out_unlock_inode
;
10473 path
= btrfs_alloc_path();
10476 goto out_unlock_inode
;
10478 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
10480 key
.type
= BTRFS_EXTENT_DATA_KEY
;
10481 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
10482 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
10485 btrfs_free_path(path
);
10486 goto out_unlock_inode
;
10488 leaf
= path
->nodes
[0];
10489 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
10490 struct btrfs_file_extent_item
);
10491 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
10492 btrfs_set_file_extent_type(leaf
, ei
,
10493 BTRFS_FILE_EXTENT_INLINE
);
10494 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
10495 btrfs_set_file_extent_compression(leaf
, ei
, 0);
10496 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
10497 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
10499 ptr
= btrfs_file_extent_inline_start(ei
);
10500 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
10501 btrfs_mark_buffer_dirty(leaf
);
10502 btrfs_free_path(path
);
10504 inode
->i_op
= &btrfs_symlink_inode_operations
;
10505 inode_nohighmem(inode
);
10506 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
10507 inode_set_bytes(inode
, name_len
);
10508 btrfs_i_size_write(BTRFS_I(inode
), name_len
);
10509 err
= btrfs_update_inode(trans
, root
, inode
);
10511 * Last step, add directory indexes for our symlink inode. This is the
10512 * last step to avoid extra cleanup of these indexes if an error happens
10516 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
,
10517 BTRFS_I(inode
), 0, index
);
10520 goto out_unlock_inode
;
10523 unlock_new_inode(inode
);
10524 d_instantiate(dentry
, inode
);
10527 btrfs_end_transaction(trans
);
10529 inode_dec_link_count(inode
);
10532 btrfs_btree_balance_dirty(fs_info
);
10537 unlock_new_inode(inode
);
10541 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
10542 u64 start
, u64 num_bytes
, u64 min_size
,
10543 loff_t actual_len
, u64
*alloc_hint
,
10544 struct btrfs_trans_handle
*trans
)
10546 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
10547 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
10548 struct extent_map
*em
;
10549 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
10550 struct btrfs_key ins
;
10551 u64 cur_offset
= start
;
10554 u64 last_alloc
= (u64
)-1;
10556 bool own_trans
= true;
10557 u64 end
= start
+ num_bytes
- 1;
10561 while (num_bytes
> 0) {
10563 trans
= btrfs_start_transaction(root
, 3);
10564 if (IS_ERR(trans
)) {
10565 ret
= PTR_ERR(trans
);
10570 cur_bytes
= min_t(u64
, num_bytes
, SZ_256M
);
10571 cur_bytes
= max(cur_bytes
, min_size
);
10573 * If we are severely fragmented we could end up with really
10574 * small allocations, so if the allocator is returning small
10575 * chunks lets make its job easier by only searching for those
10578 cur_bytes
= min(cur_bytes
, last_alloc
);
10579 ret
= btrfs_reserve_extent(root
, cur_bytes
, cur_bytes
,
10580 min_size
, 0, *alloc_hint
, &ins
, 1, 0);
10583 btrfs_end_transaction(trans
);
10586 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
10588 last_alloc
= ins
.offset
;
10589 ret
= insert_reserved_file_extent(trans
, inode
,
10590 cur_offset
, ins
.objectid
,
10591 ins
.offset
, ins
.offset
,
10592 ins
.offset
, 0, 0, 0,
10593 BTRFS_FILE_EXTENT_PREALLOC
);
10595 btrfs_free_reserved_extent(fs_info
, ins
.objectid
,
10597 btrfs_abort_transaction(trans
, ret
);
10599 btrfs_end_transaction(trans
);
10603 btrfs_drop_extent_cache(BTRFS_I(inode
), cur_offset
,
10604 cur_offset
+ ins
.offset
-1, 0);
10606 em
= alloc_extent_map();
10608 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
10609 &BTRFS_I(inode
)->runtime_flags
);
10613 em
->start
= cur_offset
;
10614 em
->orig_start
= cur_offset
;
10615 em
->len
= ins
.offset
;
10616 em
->block_start
= ins
.objectid
;
10617 em
->block_len
= ins
.offset
;
10618 em
->orig_block_len
= ins
.offset
;
10619 em
->ram_bytes
= ins
.offset
;
10620 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
10621 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
10622 em
->generation
= trans
->transid
;
10625 write_lock(&em_tree
->lock
);
10626 ret
= add_extent_mapping(em_tree
, em
, 1);
10627 write_unlock(&em_tree
->lock
);
10628 if (ret
!= -EEXIST
)
10630 btrfs_drop_extent_cache(BTRFS_I(inode
), cur_offset
,
10631 cur_offset
+ ins
.offset
- 1,
10634 free_extent_map(em
);
10636 num_bytes
-= ins
.offset
;
10637 cur_offset
+= ins
.offset
;
10638 *alloc_hint
= ins
.objectid
+ ins
.offset
;
10640 inode_inc_iversion(inode
);
10641 inode
->i_ctime
= current_time(inode
);
10642 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
10643 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
10644 (actual_len
> inode
->i_size
) &&
10645 (cur_offset
> inode
->i_size
)) {
10646 if (cur_offset
> actual_len
)
10647 i_size
= actual_len
;
10649 i_size
= cur_offset
;
10650 i_size_write(inode
, i_size
);
10651 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
10654 ret
= btrfs_update_inode(trans
, root
, inode
);
10657 btrfs_abort_transaction(trans
, ret
);
10659 btrfs_end_transaction(trans
);
10664 btrfs_end_transaction(trans
);
10666 if (cur_offset
< end
)
10667 btrfs_free_reserved_data_space(inode
, NULL
, cur_offset
,
10668 end
- cur_offset
+ 1);
10672 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
10673 u64 start
, u64 num_bytes
, u64 min_size
,
10674 loff_t actual_len
, u64
*alloc_hint
)
10676 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
10677 min_size
, actual_len
, alloc_hint
,
10681 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
10682 struct btrfs_trans_handle
*trans
, int mode
,
10683 u64 start
, u64 num_bytes
, u64 min_size
,
10684 loff_t actual_len
, u64
*alloc_hint
)
10686 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
10687 min_size
, actual_len
, alloc_hint
, trans
);
10690 static int btrfs_set_page_dirty(struct page
*page
)
10692 return __set_page_dirty_nobuffers(page
);
10695 static int btrfs_permission(struct inode
*inode
, int mask
)
10697 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
10698 umode_t mode
= inode
->i_mode
;
10700 if (mask
& MAY_WRITE
&&
10701 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
10702 if (btrfs_root_readonly(root
))
10704 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
10707 return generic_permission(inode
, mask
);
10710 static int btrfs_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
10712 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
10713 struct btrfs_trans_handle
*trans
;
10714 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
10715 struct inode
*inode
= NULL
;
10721 * 5 units required for adding orphan entry
10723 trans
= btrfs_start_transaction(root
, 5);
10725 return PTR_ERR(trans
);
10727 ret
= btrfs_find_free_ino(root
, &objectid
);
10731 inode
= btrfs_new_inode(trans
, root
, dir
, NULL
, 0,
10732 btrfs_ino(BTRFS_I(dir
)), objectid
, mode
, &index
);
10733 if (IS_ERR(inode
)) {
10734 ret
= PTR_ERR(inode
);
10739 inode
->i_fop
= &btrfs_file_operations
;
10740 inode
->i_op
= &btrfs_file_inode_operations
;
10742 inode
->i_mapping
->a_ops
= &btrfs_aops
;
10743 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
10745 ret
= btrfs_init_inode_security(trans
, inode
, dir
, NULL
);
10749 ret
= btrfs_update_inode(trans
, root
, inode
);
10752 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
10757 * We set number of links to 0 in btrfs_new_inode(), and here we set
10758 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10761 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10763 set_nlink(inode
, 1);
10764 unlock_new_inode(inode
);
10765 d_tmpfile(dentry
, inode
);
10766 mark_inode_dirty(inode
);
10769 btrfs_end_transaction(trans
);
10772 btrfs_balance_delayed_items(fs_info
);
10773 btrfs_btree_balance_dirty(fs_info
);
10777 unlock_new_inode(inode
);
10782 __attribute__((const))
10783 static int btrfs_readpage_io_failed_hook(struct page
*page
, int failed_mirror
)
10788 static struct btrfs_fs_info
*iotree_fs_info(void *private_data
)
10790 struct inode
*inode
= private_data
;
10791 return btrfs_sb(inode
->i_sb
);
10794 static void btrfs_check_extent_io_range(void *private_data
, const char *caller
,
10795 u64 start
, u64 end
)
10797 struct inode
*inode
= private_data
;
10800 isize
= i_size_read(inode
);
10801 if (end
>= PAGE_SIZE
&& (end
% 2) == 0 && end
!= isize
- 1) {
10802 btrfs_debug_rl(BTRFS_I(inode
)->root
->fs_info
,
10803 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10804 caller
, btrfs_ino(BTRFS_I(inode
)), isize
, start
, end
);
10808 void btrfs_set_range_writeback(void *private_data
, u64 start
, u64 end
)
10810 struct inode
*inode
= private_data
;
10811 unsigned long index
= start
>> PAGE_SHIFT
;
10812 unsigned long end_index
= end
>> PAGE_SHIFT
;
10815 while (index
<= end_index
) {
10816 page
= find_get_page(inode
->i_mapping
, index
);
10817 ASSERT(page
); /* Pages should be in the extent_io_tree */
10818 set_page_writeback(page
);
10824 static const struct inode_operations btrfs_dir_inode_operations
= {
10825 .getattr
= btrfs_getattr
,
10826 .lookup
= btrfs_lookup
,
10827 .create
= btrfs_create
,
10828 .unlink
= btrfs_unlink
,
10829 .link
= btrfs_link
,
10830 .mkdir
= btrfs_mkdir
,
10831 .rmdir
= btrfs_rmdir
,
10832 .rename
= btrfs_rename2
,
10833 .symlink
= btrfs_symlink
,
10834 .setattr
= btrfs_setattr
,
10835 .mknod
= btrfs_mknod
,
10836 .listxattr
= btrfs_listxattr
,
10837 .permission
= btrfs_permission
,
10838 .get_acl
= btrfs_get_acl
,
10839 .set_acl
= btrfs_set_acl
,
10840 .update_time
= btrfs_update_time
,
10841 .tmpfile
= btrfs_tmpfile
,
10843 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
10844 .lookup
= btrfs_lookup
,
10845 .permission
= btrfs_permission
,
10846 .update_time
= btrfs_update_time
,
10849 static const struct file_operations btrfs_dir_file_operations
= {
10850 .llseek
= generic_file_llseek
,
10851 .read
= generic_read_dir
,
10852 .iterate_shared
= btrfs_real_readdir
,
10853 .open
= btrfs_opendir
,
10854 .unlocked_ioctl
= btrfs_ioctl
,
10855 #ifdef CONFIG_COMPAT
10856 .compat_ioctl
= btrfs_compat_ioctl
,
10858 .release
= btrfs_release_file
,
10859 .fsync
= btrfs_sync_file
,
10862 static const struct extent_io_ops btrfs_extent_io_ops
= {
10863 /* mandatory callbacks */
10864 .submit_bio_hook
= btrfs_submit_bio_hook
,
10865 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
10866 .merge_bio_hook
= btrfs_merge_bio_hook
,
10867 .readpage_io_failed_hook
= btrfs_readpage_io_failed_hook
,
10868 .tree_fs_info
= iotree_fs_info
,
10869 .set_range_writeback
= btrfs_set_range_writeback
,
10871 /* optional callbacks */
10872 .fill_delalloc
= run_delalloc_range
,
10873 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
10874 .writepage_start_hook
= btrfs_writepage_start_hook
,
10875 .set_bit_hook
= btrfs_set_bit_hook
,
10876 .clear_bit_hook
= btrfs_clear_bit_hook
,
10877 .merge_extent_hook
= btrfs_merge_extent_hook
,
10878 .split_extent_hook
= btrfs_split_extent_hook
,
10879 .check_extent_io_range
= btrfs_check_extent_io_range
,
10883 * btrfs doesn't support the bmap operation because swapfiles
10884 * use bmap to make a mapping of extents in the file. They assume
10885 * these extents won't change over the life of the file and they
10886 * use the bmap result to do IO directly to the drive.
10888 * the btrfs bmap call would return logical addresses that aren't
10889 * suitable for IO and they also will change frequently as COW
10890 * operations happen. So, swapfile + btrfs == corruption.
10892 * For now we're avoiding this by dropping bmap.
10894 static const struct address_space_operations btrfs_aops
= {
10895 .readpage
= btrfs_readpage
,
10896 .writepage
= btrfs_writepage
,
10897 .writepages
= btrfs_writepages
,
10898 .readpages
= btrfs_readpages
,
10899 .direct_IO
= btrfs_direct_IO
,
10900 .invalidatepage
= btrfs_invalidatepage
,
10901 .releasepage
= btrfs_releasepage
,
10902 .set_page_dirty
= btrfs_set_page_dirty
,
10903 .error_remove_page
= generic_error_remove_page
,
10906 static const struct address_space_operations btrfs_symlink_aops
= {
10907 .readpage
= btrfs_readpage
,
10908 .writepage
= btrfs_writepage
,
10909 .invalidatepage
= btrfs_invalidatepage
,
10910 .releasepage
= btrfs_releasepage
,
10913 static const struct inode_operations btrfs_file_inode_operations
= {
10914 .getattr
= btrfs_getattr
,
10915 .setattr
= btrfs_setattr
,
10916 .listxattr
= btrfs_listxattr
,
10917 .permission
= btrfs_permission
,
10918 .fiemap
= btrfs_fiemap
,
10919 .get_acl
= btrfs_get_acl
,
10920 .set_acl
= btrfs_set_acl
,
10921 .update_time
= btrfs_update_time
,
10923 static const struct inode_operations btrfs_special_inode_operations
= {
10924 .getattr
= btrfs_getattr
,
10925 .setattr
= btrfs_setattr
,
10926 .permission
= btrfs_permission
,
10927 .listxattr
= btrfs_listxattr
,
10928 .get_acl
= btrfs_get_acl
,
10929 .set_acl
= btrfs_set_acl
,
10930 .update_time
= btrfs_update_time
,
10932 static const struct inode_operations btrfs_symlink_inode_operations
= {
10933 .get_link
= page_get_link
,
10934 .getattr
= btrfs_getattr
,
10935 .setattr
= btrfs_setattr
,
10936 .permission
= btrfs_permission
,
10937 .listxattr
= btrfs_listxattr
,
10938 .update_time
= btrfs_update_time
,
10941 const struct dentry_operations btrfs_dentry_operations
= {
10942 .d_delete
= btrfs_dentry_delete
,
10943 .d_release
= btrfs_dentry_release
,