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>
45 #include <asm/unaligned.h>
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
55 #include "compression.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
65 struct btrfs_iget_args
{
66 struct btrfs_key
*location
;
67 struct btrfs_root
*root
;
70 struct btrfs_dio_data
{
71 u64 outstanding_extents
;
73 u64 unsubmitted_oe_range_start
;
74 u64 unsubmitted_oe_range_end
;
78 static const struct inode_operations btrfs_dir_inode_operations
;
79 static const struct inode_operations btrfs_symlink_inode_operations
;
80 static const struct inode_operations btrfs_dir_ro_inode_operations
;
81 static const struct inode_operations btrfs_special_inode_operations
;
82 static const struct inode_operations btrfs_file_inode_operations
;
83 static const struct address_space_operations btrfs_aops
;
84 static const struct address_space_operations btrfs_symlink_aops
;
85 static const struct file_operations btrfs_dir_file_operations
;
86 static const struct extent_io_ops btrfs_extent_io_ops
;
88 static struct kmem_cache
*btrfs_inode_cachep
;
89 struct kmem_cache
*btrfs_trans_handle_cachep
;
90 struct kmem_cache
*btrfs_path_cachep
;
91 struct kmem_cache
*btrfs_free_space_cachep
;
94 static const unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
95 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
96 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
97 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
98 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
99 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
100 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
101 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
104 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
);
105 static int btrfs_truncate(struct inode
*inode
);
106 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
);
107 static noinline
int cow_file_range(struct inode
*inode
,
108 struct page
*locked_page
,
109 u64 start
, u64 end
, u64 delalloc_end
,
110 int *page_started
, unsigned long *nr_written
,
111 int unlock
, struct btrfs_dedupe_hash
*hash
);
112 static struct extent_map
*create_io_em(struct inode
*inode
, u64 start
, u64 len
,
113 u64 orig_start
, u64 block_start
,
114 u64 block_len
, u64 orig_block_len
,
115 u64 ram_bytes
, int compress_type
,
118 static void __endio_write_update_ordered(struct inode
*inode
,
119 const u64 offset
, const u64 bytes
,
120 const bool uptodate
);
123 * Cleanup all submitted ordered extents in specified range to handle errors
124 * from the fill_dellaloc() callback.
126 * NOTE: caller must ensure that when an error happens, it can not call
127 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
128 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
129 * to be released, which we want to happen only when finishing the ordered
130 * extent (btrfs_finish_ordered_io()). Also note that the caller of the
131 * fill_delalloc() callback already does proper cleanup for the first page of
132 * the range, that is, it invokes the callback writepage_end_io_hook() for the
133 * range of the first page.
135 static inline void btrfs_cleanup_ordered_extents(struct inode
*inode
,
139 unsigned long index
= offset
>> PAGE_SHIFT
;
140 unsigned long end_index
= (offset
+ bytes
- 1) >> PAGE_SHIFT
;
143 while (index
<= end_index
) {
144 page
= find_get_page(inode
->i_mapping
, index
);
148 ClearPagePrivate2(page
);
151 return __endio_write_update_ordered(inode
, offset
+ PAGE_SIZE
,
152 bytes
- PAGE_SIZE
, false);
155 static int btrfs_dirty_inode(struct inode
*inode
);
157 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
158 void btrfs_test_inode_set_ops(struct inode
*inode
)
160 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
164 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
165 struct inode
*inode
, struct inode
*dir
,
166 const struct qstr
*qstr
)
170 err
= btrfs_init_acl(trans
, inode
, dir
);
172 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
177 * this does all the hard work for inserting an inline extent into
178 * the btree. The caller should have done a btrfs_drop_extents so that
179 * no overlapping inline items exist in the btree
181 static int insert_inline_extent(struct btrfs_trans_handle
*trans
,
182 struct btrfs_path
*path
, int extent_inserted
,
183 struct btrfs_root
*root
, struct inode
*inode
,
184 u64 start
, size_t size
, size_t compressed_size
,
186 struct page
**compressed_pages
)
188 struct extent_buffer
*leaf
;
189 struct page
*page
= NULL
;
192 struct btrfs_file_extent_item
*ei
;
194 size_t cur_size
= size
;
195 unsigned long offset
;
197 if (compressed_size
&& compressed_pages
)
198 cur_size
= compressed_size
;
200 inode_add_bytes(inode
, size
);
202 if (!extent_inserted
) {
203 struct btrfs_key key
;
206 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
208 key
.type
= BTRFS_EXTENT_DATA_KEY
;
210 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
211 path
->leave_spinning
= 1;
212 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
217 leaf
= path
->nodes
[0];
218 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
219 struct btrfs_file_extent_item
);
220 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
221 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
222 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
223 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
224 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
225 ptr
= btrfs_file_extent_inline_start(ei
);
227 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
230 while (compressed_size
> 0) {
231 cpage
= compressed_pages
[i
];
232 cur_size
= min_t(unsigned long, compressed_size
,
235 kaddr
= kmap_atomic(cpage
);
236 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
237 kunmap_atomic(kaddr
);
241 compressed_size
-= cur_size
;
243 btrfs_set_file_extent_compression(leaf
, ei
,
246 page
= find_get_page(inode
->i_mapping
,
247 start
>> PAGE_SHIFT
);
248 btrfs_set_file_extent_compression(leaf
, ei
, 0);
249 kaddr
= kmap_atomic(page
);
250 offset
= start
& (PAGE_SIZE
- 1);
251 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
252 kunmap_atomic(kaddr
);
255 btrfs_mark_buffer_dirty(leaf
);
256 btrfs_release_path(path
);
259 * we're an inline extent, so nobody can
260 * extend the file past i_size without locking
261 * a page we already have locked.
263 * We must do any isize and inode updates
264 * before we unlock the pages. Otherwise we
265 * could end up racing with unlink.
267 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
268 ret
= btrfs_update_inode(trans
, root
, inode
);
276 * conditionally insert an inline extent into the file. This
277 * does the checks required to make sure the data is small enough
278 * to fit as an inline extent.
280 static noinline
int cow_file_range_inline(struct btrfs_root
*root
,
281 struct inode
*inode
, u64 start
,
282 u64 end
, size_t compressed_size
,
284 struct page
**compressed_pages
)
286 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
287 struct btrfs_trans_handle
*trans
;
288 u64 isize
= i_size_read(inode
);
289 u64 actual_end
= min(end
+ 1, isize
);
290 u64 inline_len
= actual_end
- start
;
291 u64 aligned_end
= ALIGN(end
, fs_info
->sectorsize
);
292 u64 data_len
= inline_len
;
294 struct btrfs_path
*path
;
295 int extent_inserted
= 0;
296 u32 extent_item_size
;
299 data_len
= compressed_size
;
302 actual_end
> fs_info
->sectorsize
||
303 data_len
> BTRFS_MAX_INLINE_DATA_SIZE(fs_info
) ||
305 (actual_end
& (fs_info
->sectorsize
- 1)) == 0) ||
307 data_len
> fs_info
->max_inline
) {
311 path
= btrfs_alloc_path();
315 trans
= btrfs_join_transaction(root
);
317 btrfs_free_path(path
);
318 return PTR_ERR(trans
);
320 trans
->block_rsv
= &fs_info
->delalloc_block_rsv
;
322 if (compressed_size
&& compressed_pages
)
323 extent_item_size
= btrfs_file_extent_calc_inline_size(
326 extent_item_size
= btrfs_file_extent_calc_inline_size(
329 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
,
330 start
, aligned_end
, NULL
,
331 1, 1, extent_item_size
, &extent_inserted
);
333 btrfs_abort_transaction(trans
, ret
);
337 if (isize
> actual_end
)
338 inline_len
= min_t(u64
, isize
, actual_end
);
339 ret
= insert_inline_extent(trans
, path
, extent_inserted
,
341 inline_len
, compressed_size
,
342 compress_type
, compressed_pages
);
343 if (ret
&& ret
!= -ENOSPC
) {
344 btrfs_abort_transaction(trans
, ret
);
346 } else if (ret
== -ENOSPC
) {
351 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
352 btrfs_delalloc_release_metadata(BTRFS_I(inode
), end
+ 1 - start
);
353 btrfs_drop_extent_cache(BTRFS_I(inode
), start
, aligned_end
- 1, 0);
356 * Don't forget to free the reserved space, as for inlined extent
357 * it won't count as data extent, free them directly here.
358 * And at reserve time, it's always aligned to page size, so
359 * just free one page here.
361 btrfs_qgroup_free_data(inode
, NULL
, 0, PAGE_SIZE
);
362 btrfs_free_path(path
);
363 btrfs_end_transaction(trans
);
367 struct async_extent
{
372 unsigned long nr_pages
;
374 struct list_head list
;
379 struct btrfs_root
*root
;
380 struct page
*locked_page
;
383 struct list_head extents
;
384 struct btrfs_work work
;
387 static noinline
int add_async_extent(struct async_cow
*cow
,
388 u64 start
, u64 ram_size
,
391 unsigned long nr_pages
,
394 struct async_extent
*async_extent
;
396 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
397 BUG_ON(!async_extent
); /* -ENOMEM */
398 async_extent
->start
= start
;
399 async_extent
->ram_size
= ram_size
;
400 async_extent
->compressed_size
= compressed_size
;
401 async_extent
->pages
= pages
;
402 async_extent
->nr_pages
= nr_pages
;
403 async_extent
->compress_type
= compress_type
;
404 list_add_tail(&async_extent
->list
, &cow
->extents
);
408 static inline int inode_need_compress(struct inode
*inode
, u64 start
, u64 end
)
410 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
413 if (btrfs_test_opt(fs_info
, FORCE_COMPRESS
))
416 if (BTRFS_I(inode
)->defrag_compress
)
418 /* bad compression ratios */
419 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
421 if (btrfs_test_opt(fs_info
, COMPRESS
) ||
422 BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
||
423 BTRFS_I(inode
)->prop_compress
)
424 return btrfs_compress_heuristic(inode
, start
, end
);
428 static inline void inode_should_defrag(struct btrfs_inode
*inode
,
429 u64 start
, u64 end
, u64 num_bytes
, u64 small_write
)
431 /* If this is a small write inside eof, kick off a defrag */
432 if (num_bytes
< small_write
&&
433 (start
> 0 || end
+ 1 < inode
->disk_i_size
))
434 btrfs_add_inode_defrag(NULL
, inode
);
438 * we create compressed extents in two phases. The first
439 * phase compresses a range of pages that have already been
440 * locked (both pages and state bits are locked).
442 * This is done inside an ordered work queue, and the compression
443 * is spread across many cpus. The actual IO submission is step
444 * two, and the ordered work queue takes care of making sure that
445 * happens in the same order things were put onto the queue by
446 * writepages and friends.
448 * If this code finds it can't get good compression, it puts an
449 * entry onto the work queue to write the uncompressed bytes. This
450 * makes sure that both compressed inodes and uncompressed inodes
451 * are written in the same order that the flusher thread sent them
454 static noinline
void compress_file_range(struct inode
*inode
,
455 struct page
*locked_page
,
457 struct async_cow
*async_cow
,
460 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
461 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
463 u64 blocksize
= fs_info
->sectorsize
;
465 u64 isize
= i_size_read(inode
);
467 struct page
**pages
= NULL
;
468 unsigned long nr_pages
;
469 unsigned long total_compressed
= 0;
470 unsigned long total_in
= 0;
473 int compress_type
= fs_info
->compress_type
;
476 inode_should_defrag(BTRFS_I(inode
), start
, end
, end
- start
+ 1,
479 actual_end
= min_t(u64
, isize
, end
+ 1);
482 nr_pages
= (end
>> PAGE_SHIFT
) - (start
>> PAGE_SHIFT
) + 1;
483 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED
% PAGE_SIZE
) != 0);
484 nr_pages
= min_t(unsigned long, nr_pages
,
485 BTRFS_MAX_COMPRESSED
/ PAGE_SIZE
);
488 * we don't want to send crud past the end of i_size through
489 * compression, that's just a waste of CPU time. So, if the
490 * end of the file is before the start of our current
491 * requested range of bytes, we bail out to the uncompressed
492 * cleanup code that can deal with all of this.
494 * It isn't really the fastest way to fix things, but this is a
495 * very uncommon corner.
497 if (actual_end
<= start
)
498 goto cleanup_and_bail_uncompressed
;
500 total_compressed
= actual_end
- start
;
503 * skip compression for a small file range(<=blocksize) that
504 * isn't an inline extent, since it doesn't save disk space at all.
506 if (total_compressed
<= blocksize
&&
507 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
508 goto cleanup_and_bail_uncompressed
;
510 total_compressed
= min_t(unsigned long, total_compressed
,
511 BTRFS_MAX_UNCOMPRESSED
);
512 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
513 num_bytes
= max(blocksize
, num_bytes
);
518 * we do compression for mount -o compress and when the
519 * inode has not been flagged as nocompress. This flag can
520 * change at any time if we discover bad compression ratios.
522 if (inode_need_compress(inode
, start
, end
)) {
524 pages
= kcalloc(nr_pages
, sizeof(struct page
*), GFP_NOFS
);
526 /* just bail out to the uncompressed code */
531 if (BTRFS_I(inode
)->defrag_compress
)
532 compress_type
= BTRFS_I(inode
)->defrag_compress
;
533 else if (BTRFS_I(inode
)->prop_compress
)
534 compress_type
= BTRFS_I(inode
)->prop_compress
;
537 * we need to call clear_page_dirty_for_io on each
538 * page in the range. Otherwise applications with the file
539 * mmap'd can wander in and change the page contents while
540 * we are compressing them.
542 * If the compression fails for any reason, we set the pages
543 * dirty again later on.
545 extent_range_clear_dirty_for_io(inode
, start
, end
);
547 ret
= btrfs_compress_pages(compress_type
,
548 inode
->i_mapping
, start
,
555 unsigned long offset
= total_compressed
&
557 struct page
*page
= pages
[nr_pages
- 1];
560 /* zero the tail end of the last page, we might be
561 * sending it down to disk
564 kaddr
= kmap_atomic(page
);
565 memset(kaddr
+ offset
, 0,
567 kunmap_atomic(kaddr
);
574 /* lets try to make an inline extent */
575 if (ret
|| total_in
< (actual_end
- start
)) {
576 /* we didn't compress the entire range, try
577 * to make an uncompressed inline extent.
579 ret
= cow_file_range_inline(root
, inode
, start
, end
,
580 0, BTRFS_COMPRESS_NONE
, NULL
);
582 /* try making a compressed inline extent */
583 ret
= cow_file_range_inline(root
, inode
, start
, end
,
585 compress_type
, pages
);
588 unsigned long clear_flags
= EXTENT_DELALLOC
|
589 EXTENT_DELALLOC_NEW
| EXTENT_DEFRAG
;
590 unsigned long page_error_op
;
592 clear_flags
|= (ret
< 0) ? EXTENT_DO_ACCOUNTING
: 0;
593 page_error_op
= ret
< 0 ? PAGE_SET_ERROR
: 0;
596 * inline extent creation worked or returned error,
597 * we don't need to create any more async work items.
598 * Unlock and free up our temp pages.
600 extent_clear_unlock_delalloc(inode
, start
, end
, end
,
608 btrfs_free_reserved_data_space_noquota(inode
,
617 * we aren't doing an inline extent round the compressed size
618 * up to a block size boundary so the allocator does sane
621 total_compressed
= ALIGN(total_compressed
, blocksize
);
624 * one last check to make sure the compression is really a
625 * win, compare the page count read with the blocks on disk,
626 * compression must free at least one sector size
628 total_in
= ALIGN(total_in
, PAGE_SIZE
);
629 if (total_compressed
+ blocksize
<= total_in
) {
630 num_bytes
= total_in
;
634 * The async work queues will take care of doing actual
635 * allocation on disk for these compressed pages, and
636 * will submit them to the elevator.
638 add_async_extent(async_cow
, start
, num_bytes
,
639 total_compressed
, pages
, nr_pages
,
642 if (start
+ num_bytes
< end
) {
653 * the compression code ran but failed to make things smaller,
654 * free any pages it allocated and our page pointer array
656 for (i
= 0; i
< nr_pages
; i
++) {
657 WARN_ON(pages
[i
]->mapping
);
662 total_compressed
= 0;
665 /* flag the file so we don't compress in the future */
666 if (!btrfs_test_opt(fs_info
, FORCE_COMPRESS
) &&
667 !(BTRFS_I(inode
)->prop_compress
)) {
668 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
671 cleanup_and_bail_uncompressed
:
673 * No compression, but we still need to write the pages in the file
674 * we've been given so far. redirty the locked page if it corresponds
675 * to our extent and set things up for the async work queue to run
676 * cow_file_range to do the normal delalloc dance.
678 if (page_offset(locked_page
) >= start
&&
679 page_offset(locked_page
) <= end
)
680 __set_page_dirty_nobuffers(locked_page
);
681 /* unlocked later on in the async handlers */
684 extent_range_redirty_for_io(inode
, start
, end
);
685 add_async_extent(async_cow
, start
, end
- start
+ 1, 0, NULL
, 0,
686 BTRFS_COMPRESS_NONE
);
692 for (i
= 0; i
< nr_pages
; i
++) {
693 WARN_ON(pages
[i
]->mapping
);
699 static void free_async_extent_pages(struct async_extent
*async_extent
)
703 if (!async_extent
->pages
)
706 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
707 WARN_ON(async_extent
->pages
[i
]->mapping
);
708 put_page(async_extent
->pages
[i
]);
710 kfree(async_extent
->pages
);
711 async_extent
->nr_pages
= 0;
712 async_extent
->pages
= NULL
;
716 * phase two of compressed writeback. This is the ordered portion
717 * of the code, which only gets called in the order the work was
718 * queued. We walk all the async extents created by compress_file_range
719 * and send them down to the disk.
721 static noinline
void submit_compressed_extents(struct inode
*inode
,
722 struct async_cow
*async_cow
)
724 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
725 struct async_extent
*async_extent
;
727 struct btrfs_key ins
;
728 struct extent_map
*em
;
729 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
730 struct extent_io_tree
*io_tree
;
734 while (!list_empty(&async_cow
->extents
)) {
735 async_extent
= list_entry(async_cow
->extents
.next
,
736 struct async_extent
, list
);
737 list_del(&async_extent
->list
);
739 io_tree
= &BTRFS_I(inode
)->io_tree
;
742 /* did the compression code fall back to uncompressed IO? */
743 if (!async_extent
->pages
) {
744 int page_started
= 0;
745 unsigned long nr_written
= 0;
747 lock_extent(io_tree
, async_extent
->start
,
748 async_extent
->start
+
749 async_extent
->ram_size
- 1);
751 /* allocate blocks */
752 ret
= cow_file_range(inode
, async_cow
->locked_page
,
754 async_extent
->start
+
755 async_extent
->ram_size
- 1,
756 async_extent
->start
+
757 async_extent
->ram_size
- 1,
758 &page_started
, &nr_written
, 0,
764 * if page_started, cow_file_range inserted an
765 * inline extent and took care of all the unlocking
766 * and IO for us. Otherwise, we need to submit
767 * all those pages down to the drive.
769 if (!page_started
&& !ret
)
770 extent_write_locked_range(io_tree
,
771 inode
, async_extent
->start
,
772 async_extent
->start
+
773 async_extent
->ram_size
- 1,
777 unlock_page(async_cow
->locked_page
);
783 lock_extent(io_tree
, async_extent
->start
,
784 async_extent
->start
+ async_extent
->ram_size
- 1);
786 ret
= btrfs_reserve_extent(root
, async_extent
->ram_size
,
787 async_extent
->compressed_size
,
788 async_extent
->compressed_size
,
789 0, alloc_hint
, &ins
, 1, 1);
791 free_async_extent_pages(async_extent
);
793 if (ret
== -ENOSPC
) {
794 unlock_extent(io_tree
, async_extent
->start
,
795 async_extent
->start
+
796 async_extent
->ram_size
- 1);
799 * we need to redirty the pages if we decide to
800 * fallback to uncompressed IO, otherwise we
801 * will not submit these pages down to lower
804 extent_range_redirty_for_io(inode
,
806 async_extent
->start
+
807 async_extent
->ram_size
- 1);
814 * here we're doing allocation and writeback of the
817 em
= create_io_em(inode
, async_extent
->start
,
818 async_extent
->ram_size
, /* len */
819 async_extent
->start
, /* orig_start */
820 ins
.objectid
, /* block_start */
821 ins
.offset
, /* block_len */
822 ins
.offset
, /* orig_block_len */
823 async_extent
->ram_size
, /* ram_bytes */
824 async_extent
->compress_type
,
825 BTRFS_ORDERED_COMPRESSED
);
827 /* ret value is not necessary due to void function */
828 goto out_free_reserve
;
831 ret
= btrfs_add_ordered_extent_compress(inode
,
834 async_extent
->ram_size
,
836 BTRFS_ORDERED_COMPRESSED
,
837 async_extent
->compress_type
);
839 btrfs_drop_extent_cache(BTRFS_I(inode
),
841 async_extent
->start
+
842 async_extent
->ram_size
- 1, 0);
843 goto out_free_reserve
;
845 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
848 * clear dirty, set writeback and unlock the pages.
850 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
851 async_extent
->start
+
852 async_extent
->ram_size
- 1,
853 async_extent
->start
+
854 async_extent
->ram_size
- 1,
855 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
,
856 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
858 if (btrfs_submit_compressed_write(inode
,
860 async_extent
->ram_size
,
862 ins
.offset
, async_extent
->pages
,
863 async_extent
->nr_pages
)) {
864 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
865 struct page
*p
= async_extent
->pages
[0];
866 const u64 start
= async_extent
->start
;
867 const u64 end
= start
+ async_extent
->ram_size
- 1;
869 p
->mapping
= inode
->i_mapping
;
870 tree
->ops
->writepage_end_io_hook(p
, start
, end
,
873 extent_clear_unlock_delalloc(inode
, start
, end
, end
,
877 free_async_extent_pages(async_extent
);
879 alloc_hint
= ins
.objectid
+ ins
.offset
;
885 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
886 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 1);
888 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
889 async_extent
->start
+
890 async_extent
->ram_size
- 1,
891 async_extent
->start
+
892 async_extent
->ram_size
- 1,
893 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
|
894 EXTENT_DELALLOC_NEW
|
895 EXTENT_DEFRAG
| EXTENT_DO_ACCOUNTING
,
896 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
897 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
|
899 free_async_extent_pages(async_extent
);
904 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
907 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
908 struct extent_map
*em
;
911 read_lock(&em_tree
->lock
);
912 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
915 * if block start isn't an actual block number then find the
916 * first block in this inode and use that as a hint. If that
917 * block is also bogus then just don't worry about it.
919 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
921 em
= search_extent_mapping(em_tree
, 0, 0);
922 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
923 alloc_hint
= em
->block_start
;
927 alloc_hint
= em
->block_start
;
931 read_unlock(&em_tree
->lock
);
937 * when extent_io.c finds a delayed allocation range in the file,
938 * the call backs end up in this code. The basic idea is to
939 * allocate extents on disk for the range, and create ordered data structs
940 * in ram to track those extents.
942 * locked_page is the page that writepage had locked already. We use
943 * it to make sure we don't do extra locks or unlocks.
945 * *page_started is set to one if we unlock locked_page and do everything
946 * required to start IO on it. It may be clean and already done with
949 static noinline
int cow_file_range(struct inode
*inode
,
950 struct page
*locked_page
,
951 u64 start
, u64 end
, u64 delalloc_end
,
952 int *page_started
, unsigned long *nr_written
,
953 int unlock
, struct btrfs_dedupe_hash
*hash
)
955 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
956 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
959 unsigned long ram_size
;
961 u64 cur_alloc_size
= 0;
962 u64 blocksize
= fs_info
->sectorsize
;
963 struct btrfs_key ins
;
964 struct extent_map
*em
;
966 unsigned long page_ops
;
967 bool extent_reserved
= false;
970 if (btrfs_is_free_space_inode(BTRFS_I(inode
))) {
976 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
977 num_bytes
= max(blocksize
, num_bytes
);
978 disk_num_bytes
= num_bytes
;
980 inode_should_defrag(BTRFS_I(inode
), start
, end
, num_bytes
, SZ_64K
);
983 /* lets try to make an inline extent */
984 ret
= cow_file_range_inline(root
, inode
, start
, end
, 0,
985 BTRFS_COMPRESS_NONE
, NULL
);
987 extent_clear_unlock_delalloc(inode
, start
, end
,
989 EXTENT_LOCKED
| EXTENT_DELALLOC
|
990 EXTENT_DELALLOC_NEW
|
991 EXTENT_DEFRAG
, PAGE_UNLOCK
|
992 PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
994 btrfs_free_reserved_data_space_noquota(inode
, start
,
996 *nr_written
= *nr_written
+
997 (end
- start
+ PAGE_SIZE
) / PAGE_SIZE
;
1000 } else if (ret
< 0) {
1005 BUG_ON(disk_num_bytes
>
1006 btrfs_super_total_bytes(fs_info
->super_copy
));
1008 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
1009 btrfs_drop_extent_cache(BTRFS_I(inode
), start
,
1010 start
+ num_bytes
- 1, 0);
1012 while (disk_num_bytes
> 0) {
1013 cur_alloc_size
= disk_num_bytes
;
1014 ret
= btrfs_reserve_extent(root
, cur_alloc_size
, cur_alloc_size
,
1015 fs_info
->sectorsize
, 0, alloc_hint
,
1019 cur_alloc_size
= ins
.offset
;
1020 extent_reserved
= true;
1022 ram_size
= ins
.offset
;
1023 em
= create_io_em(inode
, start
, ins
.offset
, /* len */
1024 start
, /* orig_start */
1025 ins
.objectid
, /* block_start */
1026 ins
.offset
, /* block_len */
1027 ins
.offset
, /* orig_block_len */
1028 ram_size
, /* ram_bytes */
1029 BTRFS_COMPRESS_NONE
, /* compress_type */
1030 BTRFS_ORDERED_REGULAR
/* type */);
1035 free_extent_map(em
);
1037 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
1038 ram_size
, cur_alloc_size
, 0);
1040 goto out_drop_extent_cache
;
1042 if (root
->root_key
.objectid
==
1043 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1044 ret
= btrfs_reloc_clone_csums(inode
, start
,
1047 * Only drop cache here, and process as normal.
1049 * We must not allow extent_clear_unlock_delalloc()
1050 * at out_unlock label to free meta of this ordered
1051 * extent, as its meta should be freed by
1052 * btrfs_finish_ordered_io().
1054 * So we must continue until @start is increased to
1055 * skip current ordered extent.
1058 btrfs_drop_extent_cache(BTRFS_I(inode
), start
,
1059 start
+ ram_size
- 1, 0);
1062 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
1064 /* we're not doing compressed IO, don't unlock the first
1065 * page (which the caller expects to stay locked), don't
1066 * clear any dirty bits and don't set any writeback bits
1068 * Do set the Private2 bit so we know this page was properly
1069 * setup for writepage
1071 page_ops
= unlock
? PAGE_UNLOCK
: 0;
1072 page_ops
|= PAGE_SET_PRIVATE2
;
1074 extent_clear_unlock_delalloc(inode
, start
,
1075 start
+ ram_size
- 1,
1076 delalloc_end
, locked_page
,
1077 EXTENT_LOCKED
| EXTENT_DELALLOC
,
1079 if (disk_num_bytes
< cur_alloc_size
)
1082 disk_num_bytes
-= cur_alloc_size
;
1083 num_bytes
-= cur_alloc_size
;
1084 alloc_hint
= ins
.objectid
+ ins
.offset
;
1085 start
+= cur_alloc_size
;
1086 extent_reserved
= false;
1089 * btrfs_reloc_clone_csums() error, since start is increased
1090 * extent_clear_unlock_delalloc() at out_unlock label won't
1091 * free metadata of current ordered extent, we're OK to exit.
1099 out_drop_extent_cache
:
1100 btrfs_drop_extent_cache(BTRFS_I(inode
), start
, start
+ ram_size
- 1, 0);
1102 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
1103 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 1);
1105 clear_bits
= EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DELALLOC_NEW
|
1106 EXTENT_DEFRAG
| EXTENT_CLEAR_META_RESV
;
1107 page_ops
= PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
1110 * If we reserved an extent for our delalloc range (or a subrange) and
1111 * failed to create the respective ordered extent, then it means that
1112 * when we reserved the extent we decremented the extent's size from
1113 * the data space_info's bytes_may_use counter and incremented the
1114 * space_info's bytes_reserved counter by the same amount. We must make
1115 * sure extent_clear_unlock_delalloc() does not try to decrement again
1116 * the data space_info's bytes_may_use counter, therefore we do not pass
1117 * it the flag EXTENT_CLEAR_DATA_RESV.
1119 if (extent_reserved
) {
1120 extent_clear_unlock_delalloc(inode
, start
,
1121 start
+ cur_alloc_size
,
1122 start
+ cur_alloc_size
,
1126 start
+= cur_alloc_size
;
1130 extent_clear_unlock_delalloc(inode
, start
, end
, delalloc_end
,
1132 clear_bits
| EXTENT_CLEAR_DATA_RESV
,
1138 * work queue call back to started compression on a file and pages
1140 static noinline
void async_cow_start(struct btrfs_work
*work
)
1142 struct async_cow
*async_cow
;
1144 async_cow
= container_of(work
, struct async_cow
, work
);
1146 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
1147 async_cow
->start
, async_cow
->end
, async_cow
,
1149 if (num_added
== 0) {
1150 btrfs_add_delayed_iput(async_cow
->inode
);
1151 async_cow
->inode
= NULL
;
1156 * work queue call back to submit previously compressed pages
1158 static noinline
void async_cow_submit(struct btrfs_work
*work
)
1160 struct btrfs_fs_info
*fs_info
;
1161 struct async_cow
*async_cow
;
1162 struct btrfs_root
*root
;
1163 unsigned long nr_pages
;
1165 async_cow
= container_of(work
, struct async_cow
, work
);
1167 root
= async_cow
->root
;
1168 fs_info
= root
->fs_info
;
1169 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_SIZE
) >>
1173 * atomic_sub_return implies a barrier for waitqueue_active
1175 if (atomic_sub_return(nr_pages
, &fs_info
->async_delalloc_pages
) <
1177 waitqueue_active(&fs_info
->async_submit_wait
))
1178 wake_up(&fs_info
->async_submit_wait
);
1180 if (async_cow
->inode
)
1181 submit_compressed_extents(async_cow
->inode
, async_cow
);
1184 static noinline
void async_cow_free(struct btrfs_work
*work
)
1186 struct async_cow
*async_cow
;
1187 async_cow
= container_of(work
, struct async_cow
, work
);
1188 if (async_cow
->inode
)
1189 btrfs_add_delayed_iput(async_cow
->inode
);
1193 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1194 u64 start
, u64 end
, int *page_started
,
1195 unsigned long *nr_written
)
1197 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1198 struct async_cow
*async_cow
;
1199 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1200 unsigned long nr_pages
;
1203 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1204 1, 0, NULL
, GFP_NOFS
);
1205 while (start
< end
) {
1206 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1207 BUG_ON(!async_cow
); /* -ENOMEM */
1208 async_cow
->inode
= igrab(inode
);
1209 async_cow
->root
= root
;
1210 async_cow
->locked_page
= locked_page
;
1211 async_cow
->start
= start
;
1213 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
&&
1214 !btrfs_test_opt(fs_info
, FORCE_COMPRESS
))
1217 cur_end
= min(end
, start
+ SZ_512K
- 1);
1219 async_cow
->end
= cur_end
;
1220 INIT_LIST_HEAD(&async_cow
->extents
);
1222 btrfs_init_work(&async_cow
->work
,
1223 btrfs_delalloc_helper
,
1224 async_cow_start
, async_cow_submit
,
1227 nr_pages
= (cur_end
- start
+ PAGE_SIZE
) >>
1229 atomic_add(nr_pages
, &fs_info
->async_delalloc_pages
);
1231 btrfs_queue_work(fs_info
->delalloc_workers
, &async_cow
->work
);
1233 while (atomic_read(&fs_info
->async_submit_draining
) &&
1234 atomic_read(&fs_info
->async_delalloc_pages
)) {
1235 wait_event(fs_info
->async_submit_wait
,
1236 (atomic_read(&fs_info
->async_delalloc_pages
) ==
1240 *nr_written
+= nr_pages
;
1241 start
= cur_end
+ 1;
1247 static noinline
int csum_exist_in_range(struct btrfs_fs_info
*fs_info
,
1248 u64 bytenr
, u64 num_bytes
)
1251 struct btrfs_ordered_sum
*sums
;
1254 ret
= btrfs_lookup_csums_range(fs_info
->csum_root
, bytenr
,
1255 bytenr
+ num_bytes
- 1, &list
, 0);
1256 if (ret
== 0 && list_empty(&list
))
1259 while (!list_empty(&list
)) {
1260 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1261 list_del(&sums
->list
);
1270 * when nowcow writeback call back. This checks for snapshots or COW copies
1271 * of the extents that exist in the file, and COWs the file as required.
1273 * If no cow copies or snapshots exist, we write directly to the existing
1276 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1277 struct page
*locked_page
,
1278 u64 start
, u64 end
, int *page_started
, int force
,
1279 unsigned long *nr_written
)
1281 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1282 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1283 struct extent_buffer
*leaf
;
1284 struct btrfs_path
*path
;
1285 struct btrfs_file_extent_item
*fi
;
1286 struct btrfs_key found_key
;
1287 struct extent_map
*em
;
1302 u64 ino
= btrfs_ino(BTRFS_I(inode
));
1304 path
= btrfs_alloc_path();
1306 extent_clear_unlock_delalloc(inode
, start
, end
, end
,
1308 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1309 EXTENT_DO_ACCOUNTING
|
1310 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1312 PAGE_SET_WRITEBACK
|
1313 PAGE_END_WRITEBACK
);
1317 nolock
= btrfs_is_free_space_inode(BTRFS_I(inode
));
1319 cow_start
= (u64
)-1;
1322 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, ino
,
1326 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1327 leaf
= path
->nodes
[0];
1328 btrfs_item_key_to_cpu(leaf
, &found_key
,
1329 path
->slots
[0] - 1);
1330 if (found_key
.objectid
== ino
&&
1331 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1336 leaf
= path
->nodes
[0];
1337 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1338 ret
= btrfs_next_leaf(root
, path
);
1340 if (cow_start
!= (u64
)-1)
1341 cur_offset
= cow_start
;
1346 leaf
= path
->nodes
[0];
1352 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1354 if (found_key
.objectid
> ino
)
1356 if (WARN_ON_ONCE(found_key
.objectid
< ino
) ||
1357 found_key
.type
< BTRFS_EXTENT_DATA_KEY
) {
1361 if (found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1362 found_key
.offset
> end
)
1365 if (found_key
.offset
> cur_offset
) {
1366 extent_end
= found_key
.offset
;
1371 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1372 struct btrfs_file_extent_item
);
1373 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1375 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
1376 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1377 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1378 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1379 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1380 extent_end
= found_key
.offset
+
1381 btrfs_file_extent_num_bytes(leaf
, fi
);
1383 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1384 if (extent_end
<= start
) {
1388 if (disk_bytenr
== 0)
1390 if (btrfs_file_extent_compression(leaf
, fi
) ||
1391 btrfs_file_extent_encryption(leaf
, fi
) ||
1392 btrfs_file_extent_other_encoding(leaf
, fi
))
1394 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1396 if (btrfs_extent_readonly(fs_info
, disk_bytenr
))
1398 ret
= btrfs_cross_ref_exist(root
, ino
,
1400 extent_offset
, disk_bytenr
);
1403 * ret could be -EIO if the above fails to read
1407 if (cow_start
!= (u64
)-1)
1408 cur_offset
= cow_start
;
1412 WARN_ON_ONCE(nolock
);
1415 disk_bytenr
+= extent_offset
;
1416 disk_bytenr
+= cur_offset
- found_key
.offset
;
1417 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1419 * if there are pending snapshots for this root,
1420 * we fall into common COW way.
1423 err
= btrfs_start_write_no_snapshotting(root
);
1428 * force cow if csum exists in the range.
1429 * this ensure that csum for a given extent are
1430 * either valid or do not exist.
1432 ret
= csum_exist_in_range(fs_info
, disk_bytenr
,
1436 btrfs_end_write_no_snapshotting(root
);
1439 * ret could be -EIO if the above fails to read
1443 if (cow_start
!= (u64
)-1)
1444 cur_offset
= cow_start
;
1447 WARN_ON_ONCE(nolock
);
1450 if (!btrfs_inc_nocow_writers(fs_info
, disk_bytenr
)) {
1452 btrfs_end_write_no_snapshotting(root
);
1456 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1457 extent_end
= found_key
.offset
+
1458 btrfs_file_extent_inline_len(leaf
,
1459 path
->slots
[0], fi
);
1460 extent_end
= ALIGN(extent_end
,
1461 fs_info
->sectorsize
);
1466 if (extent_end
<= start
) {
1468 if (!nolock
&& nocow
)
1469 btrfs_end_write_no_snapshotting(root
);
1471 btrfs_dec_nocow_writers(fs_info
, disk_bytenr
);
1475 if (cow_start
== (u64
)-1)
1476 cow_start
= cur_offset
;
1477 cur_offset
= extent_end
;
1478 if (cur_offset
> end
)
1484 btrfs_release_path(path
);
1485 if (cow_start
!= (u64
)-1) {
1486 ret
= cow_file_range(inode
, locked_page
,
1487 cow_start
, found_key
.offset
- 1,
1488 end
, page_started
, nr_written
, 1,
1491 if (!nolock
&& nocow
)
1492 btrfs_end_write_no_snapshotting(root
);
1494 btrfs_dec_nocow_writers(fs_info
,
1498 cow_start
= (u64
)-1;
1501 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1502 u64 orig_start
= found_key
.offset
- extent_offset
;
1504 em
= create_io_em(inode
, cur_offset
, num_bytes
,
1506 disk_bytenr
, /* block_start */
1507 num_bytes
, /* block_len */
1508 disk_num_bytes
, /* orig_block_len */
1509 ram_bytes
, BTRFS_COMPRESS_NONE
,
1510 BTRFS_ORDERED_PREALLOC
);
1512 if (!nolock
&& nocow
)
1513 btrfs_end_write_no_snapshotting(root
);
1515 btrfs_dec_nocow_writers(fs_info
,
1520 free_extent_map(em
);
1523 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1524 type
= BTRFS_ORDERED_PREALLOC
;
1526 type
= BTRFS_ORDERED_NOCOW
;
1529 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1530 num_bytes
, num_bytes
, type
);
1532 btrfs_dec_nocow_writers(fs_info
, disk_bytenr
);
1533 BUG_ON(ret
); /* -ENOMEM */
1535 if (root
->root_key
.objectid
==
1536 BTRFS_DATA_RELOC_TREE_OBJECTID
)
1538 * Error handled later, as we must prevent
1539 * extent_clear_unlock_delalloc() in error handler
1540 * from freeing metadata of created ordered extent.
1542 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1545 extent_clear_unlock_delalloc(inode
, cur_offset
,
1546 cur_offset
+ num_bytes
- 1, end
,
1547 locked_page
, EXTENT_LOCKED
|
1549 EXTENT_CLEAR_DATA_RESV
,
1550 PAGE_UNLOCK
| PAGE_SET_PRIVATE2
);
1552 if (!nolock
&& nocow
)
1553 btrfs_end_write_no_snapshotting(root
);
1554 cur_offset
= extent_end
;
1557 * btrfs_reloc_clone_csums() error, now we're OK to call error
1558 * handler, as metadata for created ordered extent will only
1559 * be freed by btrfs_finish_ordered_io().
1563 if (cur_offset
> end
)
1566 btrfs_release_path(path
);
1568 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1569 cow_start
= cur_offset
;
1571 if (cow_start
!= (u64
)-1) {
1573 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
, end
,
1574 page_started
, nr_written
, 1, NULL
);
1580 if (ret
&& cur_offset
< end
)
1581 extent_clear_unlock_delalloc(inode
, cur_offset
, end
, end
,
1582 locked_page
, EXTENT_LOCKED
|
1583 EXTENT_DELALLOC
| EXTENT_DEFRAG
|
1584 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
1586 PAGE_SET_WRITEBACK
|
1587 PAGE_END_WRITEBACK
);
1588 btrfs_free_path(path
);
1592 static inline int need_force_cow(struct inode
*inode
, u64 start
, u64 end
)
1595 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
1596 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
))
1600 * @defrag_bytes is a hint value, no spinlock held here,
1601 * if is not zero, it means the file is defragging.
1602 * Force cow if given extent needs to be defragged.
1604 if (BTRFS_I(inode
)->defrag_bytes
&&
1605 test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1606 EXTENT_DEFRAG
, 0, NULL
))
1613 * extent_io.c call back to do delayed allocation processing
1615 static int run_delalloc_range(void *private_data
, struct page
*locked_page
,
1616 u64 start
, u64 end
, int *page_started
,
1617 unsigned long *nr_written
)
1619 struct inode
*inode
= private_data
;
1621 int force_cow
= need_force_cow(inode
, start
, end
);
1623 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
&& !force_cow
) {
1624 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1625 page_started
, 1, nr_written
);
1626 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
&& !force_cow
) {
1627 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1628 page_started
, 0, nr_written
);
1629 } else if (!inode_need_compress(inode
, start
, end
)) {
1630 ret
= cow_file_range(inode
, locked_page
, start
, end
, end
,
1631 page_started
, nr_written
, 1, NULL
);
1633 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1634 &BTRFS_I(inode
)->runtime_flags
);
1635 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1636 page_started
, nr_written
);
1639 btrfs_cleanup_ordered_extents(inode
, start
, end
- start
+ 1);
1643 static void btrfs_split_extent_hook(void *private_data
,
1644 struct extent_state
*orig
, u64 split
)
1646 struct inode
*inode
= private_data
;
1649 /* not delalloc, ignore it */
1650 if (!(orig
->state
& EXTENT_DELALLOC
))
1653 size
= orig
->end
- orig
->start
+ 1;
1654 if (size
> BTRFS_MAX_EXTENT_SIZE
) {
1659 * See the explanation in btrfs_merge_extent_hook, the same
1660 * applies here, just in reverse.
1662 new_size
= orig
->end
- split
+ 1;
1663 num_extents
= count_max_extents(new_size
);
1664 new_size
= split
- orig
->start
;
1665 num_extents
+= count_max_extents(new_size
);
1666 if (count_max_extents(size
) >= num_extents
)
1670 spin_lock(&BTRFS_I(inode
)->lock
);
1671 BTRFS_I(inode
)->outstanding_extents
++;
1672 spin_unlock(&BTRFS_I(inode
)->lock
);
1676 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1677 * extents so we can keep track of new extents that are just merged onto old
1678 * extents, such as when we are doing sequential writes, so we can properly
1679 * account for the metadata space we'll need.
1681 static void btrfs_merge_extent_hook(void *private_data
,
1682 struct extent_state
*new,
1683 struct extent_state
*other
)
1685 struct inode
*inode
= private_data
;
1686 u64 new_size
, old_size
;
1689 /* not delalloc, ignore it */
1690 if (!(other
->state
& EXTENT_DELALLOC
))
1693 if (new->start
> other
->start
)
1694 new_size
= new->end
- other
->start
+ 1;
1696 new_size
= other
->end
- new->start
+ 1;
1698 /* we're not bigger than the max, unreserve the space and go */
1699 if (new_size
<= BTRFS_MAX_EXTENT_SIZE
) {
1700 spin_lock(&BTRFS_I(inode
)->lock
);
1701 BTRFS_I(inode
)->outstanding_extents
--;
1702 spin_unlock(&BTRFS_I(inode
)->lock
);
1707 * We have to add up either side to figure out how many extents were
1708 * accounted for before we merged into one big extent. If the number of
1709 * extents we accounted for is <= the amount we need for the new range
1710 * then we can return, otherwise drop. Think of it like this
1714 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1715 * need 2 outstanding extents, on one side we have 1 and the other side
1716 * we have 1 so they are == and we can return. But in this case
1718 * [MAX_SIZE+4k][MAX_SIZE+4k]
1720 * Each range on their own accounts for 2 extents, but merged together
1721 * they are only 3 extents worth of accounting, so we need to drop in
1724 old_size
= other
->end
- other
->start
+ 1;
1725 num_extents
= count_max_extents(old_size
);
1726 old_size
= new->end
- new->start
+ 1;
1727 num_extents
+= count_max_extents(old_size
);
1728 if (count_max_extents(new_size
) >= num_extents
)
1731 spin_lock(&BTRFS_I(inode
)->lock
);
1732 BTRFS_I(inode
)->outstanding_extents
--;
1733 spin_unlock(&BTRFS_I(inode
)->lock
);
1736 static void btrfs_add_delalloc_inodes(struct btrfs_root
*root
,
1737 struct inode
*inode
)
1739 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1741 spin_lock(&root
->delalloc_lock
);
1742 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1743 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1744 &root
->delalloc_inodes
);
1745 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1746 &BTRFS_I(inode
)->runtime_flags
);
1747 root
->nr_delalloc_inodes
++;
1748 if (root
->nr_delalloc_inodes
== 1) {
1749 spin_lock(&fs_info
->delalloc_root_lock
);
1750 BUG_ON(!list_empty(&root
->delalloc_root
));
1751 list_add_tail(&root
->delalloc_root
,
1752 &fs_info
->delalloc_roots
);
1753 spin_unlock(&fs_info
->delalloc_root_lock
);
1756 spin_unlock(&root
->delalloc_lock
);
1760 void __btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1761 struct btrfs_inode
*inode
)
1763 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
1765 if (!list_empty(&inode
->delalloc_inodes
)) {
1766 list_del_init(&inode
->delalloc_inodes
);
1767 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1768 &inode
->runtime_flags
);
1769 root
->nr_delalloc_inodes
--;
1770 if (!root
->nr_delalloc_inodes
) {
1771 spin_lock(&fs_info
->delalloc_root_lock
);
1772 BUG_ON(list_empty(&root
->delalloc_root
));
1773 list_del_init(&root
->delalloc_root
);
1774 spin_unlock(&fs_info
->delalloc_root_lock
);
1779 static void btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1780 struct btrfs_inode
*inode
)
1782 spin_lock(&root
->delalloc_lock
);
1783 __btrfs_del_delalloc_inode(root
, inode
);
1784 spin_unlock(&root
->delalloc_lock
);
1788 * extent_io.c set_bit_hook, used to track delayed allocation
1789 * bytes in this file, and to maintain the list of inodes that
1790 * have pending delalloc work to be done.
1792 static void btrfs_set_bit_hook(void *private_data
,
1793 struct extent_state
*state
, unsigned *bits
)
1795 struct inode
*inode
= private_data
;
1797 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1799 if ((*bits
& EXTENT_DEFRAG
) && !(*bits
& EXTENT_DELALLOC
))
1802 * set_bit and clear bit hooks normally require _irqsave/restore
1803 * but in this case, we are only testing for the DELALLOC
1804 * bit, which is only set or cleared with irqs on
1806 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1807 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1808 u64 len
= state
->end
+ 1 - state
->start
;
1809 bool do_list
= !btrfs_is_free_space_inode(BTRFS_I(inode
));
1811 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1812 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1814 spin_lock(&BTRFS_I(inode
)->lock
);
1815 BTRFS_I(inode
)->outstanding_extents
++;
1816 spin_unlock(&BTRFS_I(inode
)->lock
);
1819 /* For sanity tests */
1820 if (btrfs_is_testing(fs_info
))
1823 percpu_counter_add_batch(&fs_info
->delalloc_bytes
, len
,
1824 fs_info
->delalloc_batch
);
1825 spin_lock(&BTRFS_I(inode
)->lock
);
1826 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1827 if (*bits
& EXTENT_DEFRAG
)
1828 BTRFS_I(inode
)->defrag_bytes
+= len
;
1829 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1830 &BTRFS_I(inode
)->runtime_flags
))
1831 btrfs_add_delalloc_inodes(root
, inode
);
1832 spin_unlock(&BTRFS_I(inode
)->lock
);
1835 if (!(state
->state
& EXTENT_DELALLOC_NEW
) &&
1836 (*bits
& EXTENT_DELALLOC_NEW
)) {
1837 spin_lock(&BTRFS_I(inode
)->lock
);
1838 BTRFS_I(inode
)->new_delalloc_bytes
+= state
->end
+ 1 -
1840 spin_unlock(&BTRFS_I(inode
)->lock
);
1845 * extent_io.c clear_bit_hook, see set_bit_hook for why
1847 static void btrfs_clear_bit_hook(void *private_data
,
1848 struct extent_state
*state
,
1851 struct btrfs_inode
*inode
= BTRFS_I((struct inode
*)private_data
);
1852 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
1853 u64 len
= state
->end
+ 1 - state
->start
;
1854 u32 num_extents
= count_max_extents(len
);
1856 if ((state
->state
& EXTENT_DEFRAG
) && (*bits
& EXTENT_DEFRAG
)) {
1857 spin_lock(&inode
->lock
);
1858 inode
->defrag_bytes
-= len
;
1859 spin_unlock(&inode
->lock
);
1863 * set_bit and clear bit hooks normally require _irqsave/restore
1864 * but in this case, we are only testing for the DELALLOC
1865 * bit, which is only set or cleared with irqs on
1867 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1868 struct btrfs_root
*root
= inode
->root
;
1869 bool do_list
= !btrfs_is_free_space_inode(inode
);
1871 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1872 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1873 } else if (!(*bits
& EXTENT_CLEAR_META_RESV
)) {
1874 spin_lock(&inode
->lock
);
1875 inode
->outstanding_extents
-= num_extents
;
1876 spin_unlock(&inode
->lock
);
1880 * We don't reserve metadata space for space cache inodes so we
1881 * don't need to call dellalloc_release_metadata if there is an
1884 if (*bits
& EXTENT_CLEAR_META_RESV
&&
1885 root
!= fs_info
->tree_root
)
1886 btrfs_delalloc_release_metadata(inode
, len
);
1888 /* For sanity tests. */
1889 if (btrfs_is_testing(fs_info
))
1892 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1893 do_list
&& !(state
->state
& EXTENT_NORESERVE
) &&
1894 (*bits
& EXTENT_CLEAR_DATA_RESV
))
1895 btrfs_free_reserved_data_space_noquota(
1899 percpu_counter_add_batch(&fs_info
->delalloc_bytes
, -len
,
1900 fs_info
->delalloc_batch
);
1901 spin_lock(&inode
->lock
);
1902 inode
->delalloc_bytes
-= len
;
1903 if (do_list
&& inode
->delalloc_bytes
== 0 &&
1904 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1905 &inode
->runtime_flags
))
1906 btrfs_del_delalloc_inode(root
, inode
);
1907 spin_unlock(&inode
->lock
);
1910 if ((state
->state
& EXTENT_DELALLOC_NEW
) &&
1911 (*bits
& EXTENT_DELALLOC_NEW
)) {
1912 spin_lock(&inode
->lock
);
1913 ASSERT(inode
->new_delalloc_bytes
>= len
);
1914 inode
->new_delalloc_bytes
-= len
;
1915 spin_unlock(&inode
->lock
);
1920 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1921 * we don't create bios that span stripes or chunks
1923 * return 1 if page cannot be merged to bio
1924 * return 0 if page can be merged to bio
1925 * return error otherwise
1927 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1928 size_t size
, struct bio
*bio
,
1929 unsigned long bio_flags
)
1931 struct inode
*inode
= page
->mapping
->host
;
1932 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1933 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
1938 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1941 length
= bio
->bi_iter
.bi_size
;
1942 map_length
= length
;
1943 ret
= btrfs_map_block(fs_info
, btrfs_op(bio
), logical
, &map_length
,
1947 if (map_length
< length
+ size
)
1953 * in order to insert checksums into the metadata in large chunks,
1954 * we wait until bio submission time. All the pages in the bio are
1955 * checksummed and sums are attached onto the ordered extent record.
1957 * At IO completion time the cums attached on the ordered extent record
1958 * are inserted into the btree
1960 static blk_status_t
__btrfs_submit_bio_start(void *private_data
, struct bio
*bio
,
1961 int mirror_num
, unsigned long bio_flags
,
1964 struct inode
*inode
= private_data
;
1965 blk_status_t ret
= 0;
1967 ret
= btrfs_csum_one_bio(inode
, bio
, 0, 0);
1968 BUG_ON(ret
); /* -ENOMEM */
1973 * in order to insert checksums into the metadata in large chunks,
1974 * we wait until bio submission time. All the pages in the bio are
1975 * checksummed and sums are attached onto the ordered extent record.
1977 * At IO completion time the cums attached on the ordered extent record
1978 * are inserted into the btree
1980 static blk_status_t
__btrfs_submit_bio_done(void *private_data
, struct bio
*bio
,
1981 int mirror_num
, unsigned long bio_flags
,
1984 struct inode
*inode
= private_data
;
1985 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1988 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 1);
1990 bio
->bi_status
= ret
;
1997 * extent_io.c submission hook. This does the right thing for csum calculation
1998 * on write, or reading the csums from the tree before a read
2000 static blk_status_t
btrfs_submit_bio_hook(void *private_data
, struct bio
*bio
,
2001 int mirror_num
, unsigned long bio_flags
,
2004 struct inode
*inode
= private_data
;
2005 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2006 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2007 enum btrfs_wq_endio_type metadata
= BTRFS_WQ_ENDIO_DATA
;
2008 blk_status_t ret
= 0;
2010 int async
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
2012 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
2014 if (btrfs_is_free_space_inode(BTRFS_I(inode
)))
2015 metadata
= BTRFS_WQ_ENDIO_FREE_SPACE
;
2017 if (bio_op(bio
) != REQ_OP_WRITE
) {
2018 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, metadata
);
2022 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
2023 ret
= btrfs_submit_compressed_read(inode
, bio
,
2027 } else if (!skip_sum
) {
2028 ret
= btrfs_lookup_bio_sums(inode
, bio
, NULL
);
2033 } else if (async
&& !skip_sum
) {
2034 /* csum items have already been cloned */
2035 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2037 /* we're doing a write, do the async checksumming */
2038 ret
= btrfs_wq_submit_bio(fs_info
, bio
, mirror_num
, bio_flags
,
2040 __btrfs_submit_bio_start
,
2041 __btrfs_submit_bio_done
);
2043 } else if (!skip_sum
) {
2044 ret
= btrfs_csum_one_bio(inode
, bio
, 0, 0);
2050 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
2054 bio
->bi_status
= ret
;
2061 * given a list of ordered sums record them in the inode. This happens
2062 * at IO completion time based on sums calculated at bio submission time.
2064 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
2065 struct inode
*inode
, struct list_head
*list
)
2067 struct btrfs_ordered_sum
*sum
;
2069 list_for_each_entry(sum
, list
, list
) {
2070 trans
->adding_csums
= 1;
2071 btrfs_csum_file_blocks(trans
,
2072 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
2073 trans
->adding_csums
= 0;
2078 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
2079 struct extent_state
**cached_state
, int dedupe
)
2081 WARN_ON((end
& (PAGE_SIZE
- 1)) == 0);
2082 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
2086 /* see btrfs_writepage_start_hook for details on why this is required */
2087 struct btrfs_writepage_fixup
{
2089 struct btrfs_work work
;
2092 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
2094 struct btrfs_writepage_fixup
*fixup
;
2095 struct btrfs_ordered_extent
*ordered
;
2096 struct extent_state
*cached_state
= NULL
;
2097 struct extent_changeset
*data_reserved
= NULL
;
2099 struct inode
*inode
;
2104 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
2108 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
2109 ClearPageChecked(page
);
2113 inode
= page
->mapping
->host
;
2114 page_start
= page_offset(page
);
2115 page_end
= page_offset(page
) + PAGE_SIZE
- 1;
2117 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2120 /* already ordered? We're done */
2121 if (PagePrivate2(page
))
2124 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), page_start
,
2127 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
2128 page_end
, &cached_state
, GFP_NOFS
);
2130 btrfs_start_ordered_extent(inode
, ordered
, 1);
2131 btrfs_put_ordered_extent(ordered
);
2135 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
, page_start
,
2138 mapping_set_error(page
->mapping
, ret
);
2139 end_extent_writepage(page
, ret
, page_start
, page_end
);
2140 ClearPageChecked(page
);
2144 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
2147 mapping_set_error(page
->mapping
, ret
);
2148 end_extent_writepage(page
, ret
, page_start
, page_end
);
2149 ClearPageChecked(page
);
2153 ClearPageChecked(page
);
2154 set_page_dirty(page
);
2156 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2157 &cached_state
, GFP_NOFS
);
2162 extent_changeset_free(data_reserved
);
2166 * There are a few paths in the higher layers of the kernel that directly
2167 * set the page dirty bit without asking the filesystem if it is a
2168 * good idea. This causes problems because we want to make sure COW
2169 * properly happens and the data=ordered rules are followed.
2171 * In our case any range that doesn't have the ORDERED bit set
2172 * hasn't been properly setup for IO. We kick off an async process
2173 * to fix it up. The async helper will wait for ordered extents, set
2174 * the delalloc bit and make it safe to write the page.
2176 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
2178 struct inode
*inode
= page
->mapping
->host
;
2179 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2180 struct btrfs_writepage_fixup
*fixup
;
2182 /* this page is properly in the ordered list */
2183 if (TestClearPagePrivate2(page
))
2186 if (PageChecked(page
))
2189 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
2193 SetPageChecked(page
);
2195 btrfs_init_work(&fixup
->work
, btrfs_fixup_helper
,
2196 btrfs_writepage_fixup_worker
, NULL
, NULL
);
2198 btrfs_queue_work(fs_info
->fixup_workers
, &fixup
->work
);
2202 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
2203 struct inode
*inode
, u64 file_pos
,
2204 u64 disk_bytenr
, u64 disk_num_bytes
,
2205 u64 num_bytes
, u64 ram_bytes
,
2206 u8 compression
, u8 encryption
,
2207 u16 other_encoding
, int extent_type
)
2209 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2210 struct btrfs_file_extent_item
*fi
;
2211 struct btrfs_path
*path
;
2212 struct extent_buffer
*leaf
;
2213 struct btrfs_key ins
;
2215 int extent_inserted
= 0;
2218 path
= btrfs_alloc_path();
2223 * we may be replacing one extent in the tree with another.
2224 * The new extent is pinned in the extent map, and we don't want
2225 * to drop it from the cache until it is completely in the btree.
2227 * So, tell btrfs_drop_extents to leave this extent in the cache.
2228 * the caller is expected to unpin it and allow it to be merged
2231 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, file_pos
,
2232 file_pos
+ num_bytes
, NULL
, 0,
2233 1, sizeof(*fi
), &extent_inserted
);
2237 if (!extent_inserted
) {
2238 ins
.objectid
= btrfs_ino(BTRFS_I(inode
));
2239 ins
.offset
= file_pos
;
2240 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
2242 path
->leave_spinning
= 1;
2243 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
2248 leaf
= path
->nodes
[0];
2249 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2250 struct btrfs_file_extent_item
);
2251 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
2252 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
2253 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
2254 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
2255 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2256 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2257 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
2258 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
2259 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
2260 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
2262 btrfs_mark_buffer_dirty(leaf
);
2263 btrfs_release_path(path
);
2265 inode_add_bytes(inode
, num_bytes
);
2267 ins
.objectid
= disk_bytenr
;
2268 ins
.offset
= disk_num_bytes
;
2269 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2272 * Release the reserved range from inode dirty range map, as it is
2273 * already moved into delayed_ref_head
2275 ret
= btrfs_qgroup_release_data(inode
, file_pos
, ram_bytes
);
2279 ret
= btrfs_alloc_reserved_file_extent(trans
, root
->root_key
.objectid
,
2280 btrfs_ino(BTRFS_I(inode
)), file_pos
, qg_released
, &ins
);
2282 btrfs_free_path(path
);
2287 /* snapshot-aware defrag */
2288 struct sa_defrag_extent_backref
{
2289 struct rb_node node
;
2290 struct old_sa_defrag_extent
*old
;
2299 struct old_sa_defrag_extent
{
2300 struct list_head list
;
2301 struct new_sa_defrag_extent
*new;
2310 struct new_sa_defrag_extent
{
2311 struct rb_root root
;
2312 struct list_head head
;
2313 struct btrfs_path
*path
;
2314 struct inode
*inode
;
2322 static int backref_comp(struct sa_defrag_extent_backref
*b1
,
2323 struct sa_defrag_extent_backref
*b2
)
2325 if (b1
->root_id
< b2
->root_id
)
2327 else if (b1
->root_id
> b2
->root_id
)
2330 if (b1
->inum
< b2
->inum
)
2332 else if (b1
->inum
> b2
->inum
)
2335 if (b1
->file_pos
< b2
->file_pos
)
2337 else if (b1
->file_pos
> b2
->file_pos
)
2341 * [------------------------------] ===> (a range of space)
2342 * |<--->| |<---->| =============> (fs/file tree A)
2343 * |<---------------------------->| ===> (fs/file tree B)
2345 * A range of space can refer to two file extents in one tree while
2346 * refer to only one file extent in another tree.
2348 * So we may process a disk offset more than one time(two extents in A)
2349 * and locate at the same extent(one extent in B), then insert two same
2350 * backrefs(both refer to the extent in B).
2355 static void backref_insert(struct rb_root
*root
,
2356 struct sa_defrag_extent_backref
*backref
)
2358 struct rb_node
**p
= &root
->rb_node
;
2359 struct rb_node
*parent
= NULL
;
2360 struct sa_defrag_extent_backref
*entry
;
2365 entry
= rb_entry(parent
, struct sa_defrag_extent_backref
, node
);
2367 ret
= backref_comp(backref
, entry
);
2371 p
= &(*p
)->rb_right
;
2374 rb_link_node(&backref
->node
, parent
, p
);
2375 rb_insert_color(&backref
->node
, root
);
2379 * Note the backref might has changed, and in this case we just return 0.
2381 static noinline
int record_one_backref(u64 inum
, u64 offset
, u64 root_id
,
2384 struct btrfs_file_extent_item
*extent
;
2385 struct old_sa_defrag_extent
*old
= ctx
;
2386 struct new_sa_defrag_extent
*new = old
->new;
2387 struct btrfs_path
*path
= new->path
;
2388 struct btrfs_key key
;
2389 struct btrfs_root
*root
;
2390 struct sa_defrag_extent_backref
*backref
;
2391 struct extent_buffer
*leaf
;
2392 struct inode
*inode
= new->inode
;
2393 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2399 if (BTRFS_I(inode
)->root
->root_key
.objectid
== root_id
&&
2400 inum
== btrfs_ino(BTRFS_I(inode
)))
2403 key
.objectid
= root_id
;
2404 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2405 key
.offset
= (u64
)-1;
2407 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2409 if (PTR_ERR(root
) == -ENOENT
)
2412 btrfs_debug(fs_info
, "inum=%llu, offset=%llu, root_id=%llu",
2413 inum
, offset
, root_id
);
2414 return PTR_ERR(root
);
2417 key
.objectid
= inum
;
2418 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2419 if (offset
> (u64
)-1 << 32)
2422 key
.offset
= offset
;
2424 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2425 if (WARN_ON(ret
< 0))
2432 leaf
= path
->nodes
[0];
2433 slot
= path
->slots
[0];
2435 if (slot
>= btrfs_header_nritems(leaf
)) {
2436 ret
= btrfs_next_leaf(root
, path
);
2439 } else if (ret
> 0) {
2448 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2450 if (key
.objectid
> inum
)
2453 if (key
.objectid
< inum
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2456 extent
= btrfs_item_ptr(leaf
, slot
,
2457 struct btrfs_file_extent_item
);
2459 if (btrfs_file_extent_disk_bytenr(leaf
, extent
) != old
->bytenr
)
2463 * 'offset' refers to the exact key.offset,
2464 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2465 * (key.offset - extent_offset).
2467 if (key
.offset
!= offset
)
2470 extent_offset
= btrfs_file_extent_offset(leaf
, extent
);
2471 num_bytes
= btrfs_file_extent_num_bytes(leaf
, extent
);
2473 if (extent_offset
>= old
->extent_offset
+ old
->offset
+
2474 old
->len
|| extent_offset
+ num_bytes
<=
2475 old
->extent_offset
+ old
->offset
)
2480 backref
= kmalloc(sizeof(*backref
), GFP_NOFS
);
2486 backref
->root_id
= root_id
;
2487 backref
->inum
= inum
;
2488 backref
->file_pos
= offset
;
2489 backref
->num_bytes
= num_bytes
;
2490 backref
->extent_offset
= extent_offset
;
2491 backref
->generation
= btrfs_file_extent_generation(leaf
, extent
);
2493 backref_insert(&new->root
, backref
);
2496 btrfs_release_path(path
);
2501 static noinline
bool record_extent_backrefs(struct btrfs_path
*path
,
2502 struct new_sa_defrag_extent
*new)
2504 struct btrfs_fs_info
*fs_info
= btrfs_sb(new->inode
->i_sb
);
2505 struct old_sa_defrag_extent
*old
, *tmp
;
2510 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2511 ret
= iterate_inodes_from_logical(old
->bytenr
+
2512 old
->extent_offset
, fs_info
,
2513 path
, record_one_backref
,
2515 if (ret
< 0 && ret
!= -ENOENT
)
2518 /* no backref to be processed for this extent */
2520 list_del(&old
->list
);
2525 if (list_empty(&new->head
))
2531 static int relink_is_mergable(struct extent_buffer
*leaf
,
2532 struct btrfs_file_extent_item
*fi
,
2533 struct new_sa_defrag_extent
*new)
2535 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) != new->bytenr
)
2538 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2541 if (btrfs_file_extent_compression(leaf
, fi
) != new->compress_type
)
2544 if (btrfs_file_extent_encryption(leaf
, fi
) ||
2545 btrfs_file_extent_other_encoding(leaf
, fi
))
2552 * Note the backref might has changed, and in this case we just return 0.
2554 static noinline
int relink_extent_backref(struct btrfs_path
*path
,
2555 struct sa_defrag_extent_backref
*prev
,
2556 struct sa_defrag_extent_backref
*backref
)
2558 struct btrfs_file_extent_item
*extent
;
2559 struct btrfs_file_extent_item
*item
;
2560 struct btrfs_ordered_extent
*ordered
;
2561 struct btrfs_trans_handle
*trans
;
2562 struct btrfs_root
*root
;
2563 struct btrfs_key key
;
2564 struct extent_buffer
*leaf
;
2565 struct old_sa_defrag_extent
*old
= backref
->old
;
2566 struct new_sa_defrag_extent
*new = old
->new;
2567 struct btrfs_fs_info
*fs_info
= btrfs_sb(new->inode
->i_sb
);
2568 struct inode
*inode
;
2569 struct extent_state
*cached
= NULL
;
2578 if (prev
&& prev
->root_id
== backref
->root_id
&&
2579 prev
->inum
== backref
->inum
&&
2580 prev
->file_pos
+ prev
->num_bytes
== backref
->file_pos
)
2583 /* step 1: get root */
2584 key
.objectid
= backref
->root_id
;
2585 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2586 key
.offset
= (u64
)-1;
2588 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
2590 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2592 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2593 if (PTR_ERR(root
) == -ENOENT
)
2595 return PTR_ERR(root
);
2598 if (btrfs_root_readonly(root
)) {
2599 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2603 /* step 2: get inode */
2604 key
.objectid
= backref
->inum
;
2605 key
.type
= BTRFS_INODE_ITEM_KEY
;
2608 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
2609 if (IS_ERR(inode
)) {
2610 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2614 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2616 /* step 3: relink backref */
2617 lock_start
= backref
->file_pos
;
2618 lock_end
= backref
->file_pos
+ backref
->num_bytes
- 1;
2619 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2622 ordered
= btrfs_lookup_first_ordered_extent(inode
, lock_end
);
2624 btrfs_put_ordered_extent(ordered
);
2628 trans
= btrfs_join_transaction(root
);
2629 if (IS_ERR(trans
)) {
2630 ret
= PTR_ERR(trans
);
2634 key
.objectid
= backref
->inum
;
2635 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2636 key
.offset
= backref
->file_pos
;
2638 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2641 } else if (ret
> 0) {
2646 extent
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2647 struct btrfs_file_extent_item
);
2649 if (btrfs_file_extent_generation(path
->nodes
[0], extent
) !=
2650 backref
->generation
)
2653 btrfs_release_path(path
);
2655 start
= backref
->file_pos
;
2656 if (backref
->extent_offset
< old
->extent_offset
+ old
->offset
)
2657 start
+= old
->extent_offset
+ old
->offset
-
2658 backref
->extent_offset
;
2660 len
= min(backref
->extent_offset
+ backref
->num_bytes
,
2661 old
->extent_offset
+ old
->offset
+ old
->len
);
2662 len
-= max(backref
->extent_offset
, old
->extent_offset
+ old
->offset
);
2664 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
2669 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
2670 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2673 path
->leave_spinning
= 1;
2675 struct btrfs_file_extent_item
*fi
;
2677 struct btrfs_key found_key
;
2679 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2684 leaf
= path
->nodes
[0];
2685 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2687 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2688 struct btrfs_file_extent_item
);
2689 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
2691 if (extent_len
+ found_key
.offset
== start
&&
2692 relink_is_mergable(leaf
, fi
, new)) {
2693 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2695 btrfs_mark_buffer_dirty(leaf
);
2696 inode_add_bytes(inode
, len
);
2702 btrfs_release_path(path
);
2707 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2710 btrfs_abort_transaction(trans
, ret
);
2714 leaf
= path
->nodes
[0];
2715 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2716 struct btrfs_file_extent_item
);
2717 btrfs_set_file_extent_disk_bytenr(leaf
, item
, new->bytenr
);
2718 btrfs_set_file_extent_disk_num_bytes(leaf
, item
, new->disk_len
);
2719 btrfs_set_file_extent_offset(leaf
, item
, start
- new->file_pos
);
2720 btrfs_set_file_extent_num_bytes(leaf
, item
, len
);
2721 btrfs_set_file_extent_ram_bytes(leaf
, item
, new->len
);
2722 btrfs_set_file_extent_generation(leaf
, item
, trans
->transid
);
2723 btrfs_set_file_extent_type(leaf
, item
, BTRFS_FILE_EXTENT_REG
);
2724 btrfs_set_file_extent_compression(leaf
, item
, new->compress_type
);
2725 btrfs_set_file_extent_encryption(leaf
, item
, 0);
2726 btrfs_set_file_extent_other_encoding(leaf
, item
, 0);
2728 btrfs_mark_buffer_dirty(leaf
);
2729 inode_add_bytes(inode
, len
);
2730 btrfs_release_path(path
);
2732 ret
= btrfs_inc_extent_ref(trans
, fs_info
, new->bytenr
,
2734 backref
->root_id
, backref
->inum
,
2735 new->file_pos
); /* start - extent_offset */
2737 btrfs_abort_transaction(trans
, ret
);
2743 btrfs_release_path(path
);
2744 path
->leave_spinning
= 0;
2745 btrfs_end_transaction(trans
);
2747 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2753 static void free_sa_defrag_extent(struct new_sa_defrag_extent
*new)
2755 struct old_sa_defrag_extent
*old
, *tmp
;
2760 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2766 static void relink_file_extents(struct new_sa_defrag_extent
*new)
2768 struct btrfs_fs_info
*fs_info
= btrfs_sb(new->inode
->i_sb
);
2769 struct btrfs_path
*path
;
2770 struct sa_defrag_extent_backref
*backref
;
2771 struct sa_defrag_extent_backref
*prev
= NULL
;
2772 struct inode
*inode
;
2773 struct btrfs_root
*root
;
2774 struct rb_node
*node
;
2778 root
= BTRFS_I(inode
)->root
;
2780 path
= btrfs_alloc_path();
2784 if (!record_extent_backrefs(path
, new)) {
2785 btrfs_free_path(path
);
2788 btrfs_release_path(path
);
2791 node
= rb_first(&new->root
);
2794 rb_erase(node
, &new->root
);
2796 backref
= rb_entry(node
, struct sa_defrag_extent_backref
, node
);
2798 ret
= relink_extent_backref(path
, prev
, backref
);
2811 btrfs_free_path(path
);
2813 free_sa_defrag_extent(new);
2815 atomic_dec(&fs_info
->defrag_running
);
2816 wake_up(&fs_info
->transaction_wait
);
2819 static struct new_sa_defrag_extent
*
2820 record_old_file_extents(struct inode
*inode
,
2821 struct btrfs_ordered_extent
*ordered
)
2823 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2824 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2825 struct btrfs_path
*path
;
2826 struct btrfs_key key
;
2827 struct old_sa_defrag_extent
*old
;
2828 struct new_sa_defrag_extent
*new;
2831 new = kmalloc(sizeof(*new), GFP_NOFS
);
2836 new->file_pos
= ordered
->file_offset
;
2837 new->len
= ordered
->len
;
2838 new->bytenr
= ordered
->start
;
2839 new->disk_len
= ordered
->disk_len
;
2840 new->compress_type
= ordered
->compress_type
;
2841 new->root
= RB_ROOT
;
2842 INIT_LIST_HEAD(&new->head
);
2844 path
= btrfs_alloc_path();
2848 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
2849 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2850 key
.offset
= new->file_pos
;
2852 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2855 if (ret
> 0 && path
->slots
[0] > 0)
2858 /* find out all the old extents for the file range */
2860 struct btrfs_file_extent_item
*extent
;
2861 struct extent_buffer
*l
;
2870 slot
= path
->slots
[0];
2872 if (slot
>= btrfs_header_nritems(l
)) {
2873 ret
= btrfs_next_leaf(root
, path
);
2881 btrfs_item_key_to_cpu(l
, &key
, slot
);
2883 if (key
.objectid
!= btrfs_ino(BTRFS_I(inode
)))
2885 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2887 if (key
.offset
>= new->file_pos
+ new->len
)
2890 extent
= btrfs_item_ptr(l
, slot
, struct btrfs_file_extent_item
);
2892 num_bytes
= btrfs_file_extent_num_bytes(l
, extent
);
2893 if (key
.offset
+ num_bytes
< new->file_pos
)
2896 disk_bytenr
= btrfs_file_extent_disk_bytenr(l
, extent
);
2900 extent_offset
= btrfs_file_extent_offset(l
, extent
);
2902 old
= kmalloc(sizeof(*old
), GFP_NOFS
);
2906 offset
= max(new->file_pos
, key
.offset
);
2907 end
= min(new->file_pos
+ new->len
, key
.offset
+ num_bytes
);
2909 old
->bytenr
= disk_bytenr
;
2910 old
->extent_offset
= extent_offset
;
2911 old
->offset
= offset
- key
.offset
;
2912 old
->len
= end
- offset
;
2915 list_add_tail(&old
->list
, &new->head
);
2921 btrfs_free_path(path
);
2922 atomic_inc(&fs_info
->defrag_running
);
2927 btrfs_free_path(path
);
2929 free_sa_defrag_extent(new);
2933 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info
*fs_info
,
2936 struct btrfs_block_group_cache
*cache
;
2938 cache
= btrfs_lookup_block_group(fs_info
, start
);
2941 spin_lock(&cache
->lock
);
2942 cache
->delalloc_bytes
-= len
;
2943 spin_unlock(&cache
->lock
);
2945 btrfs_put_block_group(cache
);
2948 /* as ordered data IO finishes, this gets called so we can finish
2949 * an ordered extent if the range of bytes in the file it covers are
2952 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
2954 struct inode
*inode
= ordered_extent
->inode
;
2955 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2956 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2957 struct btrfs_trans_handle
*trans
= NULL
;
2958 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2959 struct extent_state
*cached_state
= NULL
;
2960 struct new_sa_defrag_extent
*new = NULL
;
2961 int compress_type
= 0;
2963 u64 logical_len
= ordered_extent
->len
;
2965 bool truncated
= false;
2966 bool range_locked
= false;
2967 bool clear_new_delalloc_bytes
= false;
2968 bool clear_reserved_extent
= true;
2970 if (!test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
2971 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
) &&
2972 !test_bit(BTRFS_ORDERED_DIRECT
, &ordered_extent
->flags
))
2973 clear_new_delalloc_bytes
= true;
2975 nolock
= btrfs_is_free_space_inode(BTRFS_I(inode
));
2977 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
2982 btrfs_free_io_failure_record(BTRFS_I(inode
),
2983 ordered_extent
->file_offset
,
2984 ordered_extent
->file_offset
+
2985 ordered_extent
->len
- 1);
2987 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered_extent
->flags
)) {
2989 logical_len
= ordered_extent
->truncated_len
;
2990 /* Truncated the entire extent, don't bother adding */
2995 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
2996 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
2999 * For mwrite(mmap + memset to write) case, we still reserve
3000 * space for NOCOW range.
3001 * As NOCOW won't cause a new delayed ref, just free the space
3003 btrfs_qgroup_free_data(inode
, NULL
, ordered_extent
->file_offset
,
3004 ordered_extent
->len
);
3005 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
3007 trans
= btrfs_join_transaction_nolock(root
);
3009 trans
= btrfs_join_transaction(root
);
3010 if (IS_ERR(trans
)) {
3011 ret
= PTR_ERR(trans
);
3015 trans
->block_rsv
= &fs_info
->delalloc_block_rsv
;
3016 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
3017 if (ret
) /* -ENOMEM or corruption */
3018 btrfs_abort_transaction(trans
, ret
);
3022 range_locked
= true;
3023 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
3024 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
3027 ret
= test_range_bit(io_tree
, ordered_extent
->file_offset
,
3028 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
3029 EXTENT_DEFRAG
, 0, cached_state
);
3031 u64 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
3032 if (0 && last_snapshot
>= BTRFS_I(inode
)->generation
)
3033 /* the inode is shared */
3034 new = record_old_file_extents(inode
, ordered_extent
);
3036 clear_extent_bit(io_tree
, ordered_extent
->file_offset
,
3037 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
3038 EXTENT_DEFRAG
, 0, 0, &cached_state
, GFP_NOFS
);
3042 trans
= btrfs_join_transaction_nolock(root
);
3044 trans
= btrfs_join_transaction(root
);
3045 if (IS_ERR(trans
)) {
3046 ret
= PTR_ERR(trans
);
3051 trans
->block_rsv
= &fs_info
->delalloc_block_rsv
;
3053 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
3054 compress_type
= ordered_extent
->compress_type
;
3055 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
3056 BUG_ON(compress_type
);
3057 btrfs_qgroup_free_data(inode
, NULL
, ordered_extent
->file_offset
,
3058 ordered_extent
->len
);
3059 ret
= btrfs_mark_extent_written(trans
, BTRFS_I(inode
),
3060 ordered_extent
->file_offset
,
3061 ordered_extent
->file_offset
+
3064 BUG_ON(root
== fs_info
->tree_root
);
3065 ret
= insert_reserved_file_extent(trans
, inode
,
3066 ordered_extent
->file_offset
,
3067 ordered_extent
->start
,
3068 ordered_extent
->disk_len
,
3069 logical_len
, logical_len
,
3070 compress_type
, 0, 0,
3071 BTRFS_FILE_EXTENT_REG
);
3073 clear_reserved_extent
= false;
3074 btrfs_release_delalloc_bytes(fs_info
,
3075 ordered_extent
->start
,
3076 ordered_extent
->disk_len
);
3079 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
3080 ordered_extent
->file_offset
, ordered_extent
->len
,
3083 btrfs_abort_transaction(trans
, ret
);
3087 add_pending_csums(trans
, inode
, &ordered_extent
->list
);
3089 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
3090 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
3091 if (ret
) { /* -ENOMEM or corruption */
3092 btrfs_abort_transaction(trans
, ret
);
3097 if (range_locked
|| clear_new_delalloc_bytes
) {
3098 unsigned int clear_bits
= 0;
3101 clear_bits
|= EXTENT_LOCKED
;
3102 if (clear_new_delalloc_bytes
)
3103 clear_bits
|= EXTENT_DELALLOC_NEW
;
3104 clear_extent_bit(&BTRFS_I(inode
)->io_tree
,
3105 ordered_extent
->file_offset
,
3106 ordered_extent
->file_offset
+
3107 ordered_extent
->len
- 1,
3109 (clear_bits
& EXTENT_LOCKED
) ? 1 : 0,
3110 0, &cached_state
, GFP_NOFS
);
3113 if (root
!= fs_info
->tree_root
)
3114 btrfs_delalloc_release_metadata(BTRFS_I(inode
),
3115 ordered_extent
->len
);
3117 btrfs_end_transaction(trans
);
3119 if (ret
|| truncated
) {
3123 start
= ordered_extent
->file_offset
+ logical_len
;
3125 start
= ordered_extent
->file_offset
;
3126 end
= ordered_extent
->file_offset
+ ordered_extent
->len
- 1;
3127 clear_extent_uptodate(io_tree
, start
, end
, NULL
, GFP_NOFS
);
3129 /* Drop the cache for the part of the extent we didn't write. */
3130 btrfs_drop_extent_cache(BTRFS_I(inode
), start
, end
, 0);
3133 * If the ordered extent had an IOERR or something else went
3134 * wrong we need to return the space for this ordered extent
3135 * back to the allocator. We only free the extent in the
3136 * truncated case if we didn't write out the extent at all.
3138 * If we made it past insert_reserved_file_extent before we
3139 * errored out then we don't need to do this as the accounting
3140 * has already been done.
3142 if ((ret
|| !logical_len
) &&
3143 clear_reserved_extent
&&
3144 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
3145 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
))
3146 btrfs_free_reserved_extent(fs_info
,
3147 ordered_extent
->start
,
3148 ordered_extent
->disk_len
, 1);
3153 * This needs to be done to make sure anybody waiting knows we are done
3154 * updating everything for this ordered extent.
3156 btrfs_remove_ordered_extent(inode
, ordered_extent
);
3158 /* for snapshot-aware defrag */
3161 free_sa_defrag_extent(new);
3162 atomic_dec(&fs_info
->defrag_running
);
3164 relink_file_extents(new);
3169 btrfs_put_ordered_extent(ordered_extent
);
3170 /* once for the tree */
3171 btrfs_put_ordered_extent(ordered_extent
);
3176 static void finish_ordered_fn(struct btrfs_work
*work
)
3178 struct btrfs_ordered_extent
*ordered_extent
;
3179 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
3180 btrfs_finish_ordered_io(ordered_extent
);
3183 static void btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
3184 struct extent_state
*state
, int uptodate
)
3186 struct inode
*inode
= page
->mapping
->host
;
3187 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3188 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
3189 struct btrfs_workqueue
*wq
;
3190 btrfs_work_func_t func
;
3192 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
3194 ClearPagePrivate2(page
);
3195 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
3196 end
- start
+ 1, uptodate
))
3199 if (btrfs_is_free_space_inode(BTRFS_I(inode
))) {
3200 wq
= fs_info
->endio_freespace_worker
;
3201 func
= btrfs_freespace_write_helper
;
3203 wq
= fs_info
->endio_write_workers
;
3204 func
= btrfs_endio_write_helper
;
3207 btrfs_init_work(&ordered_extent
->work
, func
, finish_ordered_fn
, NULL
,
3209 btrfs_queue_work(wq
, &ordered_extent
->work
);
3212 static int __readpage_endio_check(struct inode
*inode
,
3213 struct btrfs_io_bio
*io_bio
,
3214 int icsum
, struct page
*page
,
3215 int pgoff
, u64 start
, size_t len
)
3221 csum_expected
= *(((u32
*)io_bio
->csum
) + icsum
);
3223 kaddr
= kmap_atomic(page
);
3224 csum
= btrfs_csum_data(kaddr
+ pgoff
, csum
, len
);
3225 btrfs_csum_final(csum
, (u8
*)&csum
);
3226 if (csum
!= csum_expected
)
3229 kunmap_atomic(kaddr
);
3232 btrfs_print_data_csum_error(BTRFS_I(inode
), start
, csum
, csum_expected
,
3233 io_bio
->mirror_num
);
3234 memset(kaddr
+ pgoff
, 1, len
);
3235 flush_dcache_page(page
);
3236 kunmap_atomic(kaddr
);
3241 * when reads are done, we need to check csums to verify the data is correct
3242 * if there's a match, we allow the bio to finish. If not, the code in
3243 * extent_io.c will try to find good copies for us.
3245 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
3246 u64 phy_offset
, struct page
*page
,
3247 u64 start
, u64 end
, int mirror
)
3249 size_t offset
= start
- page_offset(page
);
3250 struct inode
*inode
= page
->mapping
->host
;
3251 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3252 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3254 if (PageChecked(page
)) {
3255 ClearPageChecked(page
);
3259 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
3262 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
3263 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
3264 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
);
3268 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
3269 return __readpage_endio_check(inode
, io_bio
, phy_offset
, page
, offset
,
3270 start
, (size_t)(end
- start
+ 1));
3273 void btrfs_add_delayed_iput(struct inode
*inode
)
3275 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3276 struct btrfs_inode
*binode
= BTRFS_I(inode
);
3278 if (atomic_add_unless(&inode
->i_count
, -1, 1))
3281 spin_lock(&fs_info
->delayed_iput_lock
);
3282 if (binode
->delayed_iput_count
== 0) {
3283 ASSERT(list_empty(&binode
->delayed_iput
));
3284 list_add_tail(&binode
->delayed_iput
, &fs_info
->delayed_iputs
);
3286 binode
->delayed_iput_count
++;
3288 spin_unlock(&fs_info
->delayed_iput_lock
);
3291 void btrfs_run_delayed_iputs(struct btrfs_fs_info
*fs_info
)
3294 spin_lock(&fs_info
->delayed_iput_lock
);
3295 while (!list_empty(&fs_info
->delayed_iputs
)) {
3296 struct btrfs_inode
*inode
;
3298 inode
= list_first_entry(&fs_info
->delayed_iputs
,
3299 struct btrfs_inode
, delayed_iput
);
3300 if (inode
->delayed_iput_count
) {
3301 inode
->delayed_iput_count
--;
3302 list_move_tail(&inode
->delayed_iput
,
3303 &fs_info
->delayed_iputs
);
3305 list_del_init(&inode
->delayed_iput
);
3307 spin_unlock(&fs_info
->delayed_iput_lock
);
3308 iput(&inode
->vfs_inode
);
3309 spin_lock(&fs_info
->delayed_iput_lock
);
3311 spin_unlock(&fs_info
->delayed_iput_lock
);
3315 * This is called in transaction commit time. If there are no orphan
3316 * files in the subvolume, it removes orphan item and frees block_rsv
3319 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
3320 struct btrfs_root
*root
)
3322 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3323 struct btrfs_block_rsv
*block_rsv
;
3326 if (atomic_read(&root
->orphan_inodes
) ||
3327 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
3330 spin_lock(&root
->orphan_lock
);
3331 if (atomic_read(&root
->orphan_inodes
)) {
3332 spin_unlock(&root
->orphan_lock
);
3336 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
3337 spin_unlock(&root
->orphan_lock
);
3341 block_rsv
= root
->orphan_block_rsv
;
3342 root
->orphan_block_rsv
= NULL
;
3343 spin_unlock(&root
->orphan_lock
);
3345 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
) &&
3346 btrfs_root_refs(&root
->root_item
) > 0) {
3347 ret
= btrfs_del_orphan_item(trans
, fs_info
->tree_root
,
3348 root
->root_key
.objectid
);
3350 btrfs_abort_transaction(trans
, ret
);
3352 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
,
3357 WARN_ON(block_rsv
->size
> 0);
3358 btrfs_free_block_rsv(fs_info
, block_rsv
);
3363 * This creates an orphan entry for the given inode in case something goes
3364 * wrong in the middle of an unlink/truncate.
3366 * NOTE: caller of this function should reserve 5 units of metadata for
3369 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
,
3370 struct btrfs_inode
*inode
)
3372 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
3373 struct btrfs_root
*root
= inode
->root
;
3374 struct btrfs_block_rsv
*block_rsv
= NULL
;
3379 if (!root
->orphan_block_rsv
) {
3380 block_rsv
= btrfs_alloc_block_rsv(fs_info
,
3381 BTRFS_BLOCK_RSV_TEMP
);
3386 spin_lock(&root
->orphan_lock
);
3387 if (!root
->orphan_block_rsv
) {
3388 root
->orphan_block_rsv
= block_rsv
;
3389 } else if (block_rsv
) {
3390 btrfs_free_block_rsv(fs_info
, block_rsv
);
3394 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3395 &inode
->runtime_flags
)) {
3398 * For proper ENOSPC handling, we should do orphan
3399 * cleanup when mounting. But this introduces backward
3400 * compatibility issue.
3402 if (!xchg(&root
->orphan_item_inserted
, 1))
3408 atomic_inc(&root
->orphan_inodes
);
3411 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3412 &inode
->runtime_flags
))
3414 spin_unlock(&root
->orphan_lock
);
3416 /* grab metadata reservation from transaction handle */
3418 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
3422 * dec doesn't need spin_lock as ->orphan_block_rsv
3423 * would be released only if ->orphan_inodes is
3426 atomic_dec(&root
->orphan_inodes
);
3427 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3428 &inode
->runtime_flags
);
3430 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3431 &inode
->runtime_flags
);
3436 /* insert an orphan item to track this unlinked/truncated file */
3438 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
3441 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3442 &inode
->runtime_flags
);
3443 btrfs_orphan_release_metadata(inode
);
3446 * btrfs_orphan_commit_root may race with us and set
3447 * ->orphan_block_rsv to zero, in order to avoid that,
3448 * decrease ->orphan_inodes after everything is done.
3450 atomic_dec(&root
->orphan_inodes
);
3451 if (ret
!= -EEXIST
) {
3452 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3453 &inode
->runtime_flags
);
3454 btrfs_abort_transaction(trans
, ret
);
3461 /* insert an orphan item to track subvolume contains orphan files */
3463 ret
= btrfs_insert_orphan_item(trans
, fs_info
->tree_root
,
3464 root
->root_key
.objectid
);
3465 if (ret
&& ret
!= -EEXIST
) {
3466 btrfs_abort_transaction(trans
, ret
);
3474 * We have done the truncate/delete so we can go ahead and remove the orphan
3475 * item for this particular inode.
3477 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3478 struct btrfs_inode
*inode
)
3480 struct btrfs_root
*root
= inode
->root
;
3481 int delete_item
= 0;
3484 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3485 &inode
->runtime_flags
))
3488 if (delete_item
&& trans
)
3489 ret
= btrfs_del_orphan_item(trans
, root
, btrfs_ino(inode
));
3491 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3492 &inode
->runtime_flags
))
3493 btrfs_orphan_release_metadata(inode
);
3496 * btrfs_orphan_commit_root may race with us and set ->orphan_block_rsv
3497 * to zero, in order to avoid that, decrease ->orphan_inodes after
3498 * everything is done.
3501 atomic_dec(&root
->orphan_inodes
);
3507 * this cleans up any orphans that may be left on the list from the last use
3510 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3512 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3513 struct btrfs_path
*path
;
3514 struct extent_buffer
*leaf
;
3515 struct btrfs_key key
, found_key
;
3516 struct btrfs_trans_handle
*trans
;
3517 struct inode
*inode
;
3518 u64 last_objectid
= 0;
3519 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
3521 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
3524 path
= btrfs_alloc_path();
3529 path
->reada
= READA_BACK
;
3531 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3532 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
3533 key
.offset
= (u64
)-1;
3536 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3541 * if ret == 0 means we found what we were searching for, which
3542 * is weird, but possible, so only screw with path if we didn't
3543 * find the key and see if we have stuff that matches
3547 if (path
->slots
[0] == 0)
3552 /* pull out the item */
3553 leaf
= path
->nodes
[0];
3554 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3556 /* make sure the item matches what we want */
3557 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3559 if (found_key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
3562 /* release the path since we're done with it */
3563 btrfs_release_path(path
);
3566 * this is where we are basically btrfs_lookup, without the
3567 * crossing root thing. we store the inode number in the
3568 * offset of the orphan item.
3571 if (found_key
.offset
== last_objectid
) {
3573 "Error removing orphan entry, stopping orphan cleanup");
3578 last_objectid
= found_key
.offset
;
3580 found_key
.objectid
= found_key
.offset
;
3581 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3582 found_key
.offset
= 0;
3583 inode
= btrfs_iget(fs_info
->sb
, &found_key
, root
, NULL
);
3584 ret
= PTR_ERR_OR_ZERO(inode
);
3585 if (ret
&& ret
!= -ENOENT
)
3588 if (ret
== -ENOENT
&& root
== fs_info
->tree_root
) {
3589 struct btrfs_root
*dead_root
;
3590 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3591 int is_dead_root
= 0;
3594 * this is an orphan in the tree root. Currently these
3595 * could come from 2 sources:
3596 * a) a snapshot deletion in progress
3597 * b) a free space cache inode
3598 * We need to distinguish those two, as the snapshot
3599 * orphan must not get deleted.
3600 * find_dead_roots already ran before us, so if this
3601 * is a snapshot deletion, we should find the root
3602 * in the dead_roots list
3604 spin_lock(&fs_info
->trans_lock
);
3605 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
3607 if (dead_root
->root_key
.objectid
==
3608 found_key
.objectid
) {
3613 spin_unlock(&fs_info
->trans_lock
);
3615 /* prevent this orphan from being found again */
3616 key
.offset
= found_key
.objectid
- 1;
3621 * Inode is already gone but the orphan item is still there,
3622 * kill the orphan item.
3624 if (ret
== -ENOENT
) {
3625 trans
= btrfs_start_transaction(root
, 1);
3626 if (IS_ERR(trans
)) {
3627 ret
= PTR_ERR(trans
);
3630 btrfs_debug(fs_info
, "auto deleting %Lu",
3631 found_key
.objectid
);
3632 ret
= btrfs_del_orphan_item(trans
, root
,
3633 found_key
.objectid
);
3634 btrfs_end_transaction(trans
);
3641 * add this inode to the orphan list so btrfs_orphan_del does
3642 * the proper thing when we hit it
3644 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3645 &BTRFS_I(inode
)->runtime_flags
);
3646 atomic_inc(&root
->orphan_inodes
);
3648 /* if we have links, this was a truncate, lets do that */
3649 if (inode
->i_nlink
) {
3650 if (WARN_ON(!S_ISREG(inode
->i_mode
))) {
3656 /* 1 for the orphan item deletion. */
3657 trans
= btrfs_start_transaction(root
, 1);
3658 if (IS_ERR(trans
)) {
3660 ret
= PTR_ERR(trans
);
3663 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
3664 btrfs_end_transaction(trans
);
3670 ret
= btrfs_truncate(inode
);
3672 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
3677 /* this will do delete_inode and everything for us */
3682 /* release the path since we're done with it */
3683 btrfs_release_path(path
);
3685 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
3687 if (root
->orphan_block_rsv
)
3688 btrfs_block_rsv_release(fs_info
, root
->orphan_block_rsv
,
3691 if (root
->orphan_block_rsv
||
3692 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
)) {
3693 trans
= btrfs_join_transaction(root
);
3695 btrfs_end_transaction(trans
);
3699 btrfs_debug(fs_info
, "unlinked %d orphans", nr_unlink
);
3701 btrfs_debug(fs_info
, "truncated %d orphans", nr_truncate
);
3705 btrfs_err(fs_info
, "could not do orphan cleanup %d", ret
);
3706 btrfs_free_path(path
);
3711 * very simple check to peek ahead in the leaf looking for xattrs. If we
3712 * don't find any xattrs, we know there can't be any acls.
3714 * slot is the slot the inode is in, objectid is the objectid of the inode
3716 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3717 int slot
, u64 objectid
,
3718 int *first_xattr_slot
)
3720 u32 nritems
= btrfs_header_nritems(leaf
);
3721 struct btrfs_key found_key
;
3722 static u64 xattr_access
= 0;
3723 static u64 xattr_default
= 0;
3726 if (!xattr_access
) {
3727 xattr_access
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS
,
3728 strlen(XATTR_NAME_POSIX_ACL_ACCESS
));
3729 xattr_default
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT
,
3730 strlen(XATTR_NAME_POSIX_ACL_DEFAULT
));
3734 *first_xattr_slot
= -1;
3735 while (slot
< nritems
) {
3736 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3738 /* we found a different objectid, there must not be acls */
3739 if (found_key
.objectid
!= objectid
)
3742 /* we found an xattr, assume we've got an acl */
3743 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
3744 if (*first_xattr_slot
== -1)
3745 *first_xattr_slot
= slot
;
3746 if (found_key
.offset
== xattr_access
||
3747 found_key
.offset
== xattr_default
)
3752 * we found a key greater than an xattr key, there can't
3753 * be any acls later on
3755 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3762 * it goes inode, inode backrefs, xattrs, extents,
3763 * so if there are a ton of hard links to an inode there can
3764 * be a lot of backrefs. Don't waste time searching too hard,
3765 * this is just an optimization
3770 /* we hit the end of the leaf before we found an xattr or
3771 * something larger than an xattr. We have to assume the inode
3774 if (*first_xattr_slot
== -1)
3775 *first_xattr_slot
= slot
;
3780 * read an inode from the btree into the in-memory inode
3782 static int btrfs_read_locked_inode(struct inode
*inode
)
3784 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3785 struct btrfs_path
*path
;
3786 struct extent_buffer
*leaf
;
3787 struct btrfs_inode_item
*inode_item
;
3788 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3789 struct btrfs_key location
;
3794 bool filled
= false;
3795 int first_xattr_slot
;
3797 ret
= btrfs_fill_inode(inode
, &rdev
);
3801 path
= btrfs_alloc_path();
3807 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3809 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3816 leaf
= path
->nodes
[0];
3821 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3822 struct btrfs_inode_item
);
3823 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3824 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3825 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3826 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3827 btrfs_i_size_write(BTRFS_I(inode
), btrfs_inode_size(leaf
, inode_item
));
3829 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->atime
);
3830 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->atime
);
3832 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->mtime
);
3833 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->mtime
);
3835 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->ctime
);
3836 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->ctime
);
3838 BTRFS_I(inode
)->i_otime
.tv_sec
=
3839 btrfs_timespec_sec(leaf
, &inode_item
->otime
);
3840 BTRFS_I(inode
)->i_otime
.tv_nsec
=
3841 btrfs_timespec_nsec(leaf
, &inode_item
->otime
);
3843 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3844 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3845 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3847 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
3848 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3850 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3852 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3853 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
3857 * If we were modified in the current generation and evicted from memory
3858 * and then re-read we need to do a full sync since we don't have any
3859 * idea about which extents were modified before we were evicted from
3862 * This is required for both inode re-read from disk and delayed inode
3863 * in delayed_nodes_tree.
3865 if (BTRFS_I(inode
)->last_trans
== fs_info
->generation
)
3866 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3867 &BTRFS_I(inode
)->runtime_flags
);
3870 * We don't persist the id of the transaction where an unlink operation
3871 * against the inode was last made. So here we assume the inode might
3872 * have been evicted, and therefore the exact value of last_unlink_trans
3873 * lost, and set it to last_trans to avoid metadata inconsistencies
3874 * between the inode and its parent if the inode is fsync'ed and the log
3875 * replayed. For example, in the scenario:
3878 * ln mydir/foo mydir/bar
3881 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3882 * xfs_io -c fsync mydir/foo
3884 * mount fs, triggers fsync log replay
3886 * We must make sure that when we fsync our inode foo we also log its
3887 * parent inode, otherwise after log replay the parent still has the
3888 * dentry with the "bar" name but our inode foo has a link count of 1
3889 * and doesn't have an inode ref with the name "bar" anymore.
3891 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3892 * but it guarantees correctness at the expense of occasional full
3893 * transaction commits on fsync if our inode is a directory, or if our
3894 * inode is not a directory, logging its parent unnecessarily.
3896 BTRFS_I(inode
)->last_unlink_trans
= BTRFS_I(inode
)->last_trans
;
3898 * Similar reasoning for last_link_trans, needs to be set otherwise
3899 * for a case like the following:
3904 * echo 2 > /proc/sys/vm/drop_caches
3908 * Would result in link bar and directory A not existing after the power
3911 BTRFS_I(inode
)->last_link_trans
= BTRFS_I(inode
)->last_trans
;
3914 if (inode
->i_nlink
!= 1 ||
3915 path
->slots
[0] >= btrfs_header_nritems(leaf
))
3918 btrfs_item_key_to_cpu(leaf
, &location
, path
->slots
[0]);
3919 if (location
.objectid
!= btrfs_ino(BTRFS_I(inode
)))
3922 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3923 if (location
.type
== BTRFS_INODE_REF_KEY
) {
3924 struct btrfs_inode_ref
*ref
;
3926 ref
= (struct btrfs_inode_ref
*)ptr
;
3927 BTRFS_I(inode
)->dir_index
= btrfs_inode_ref_index(leaf
, ref
);
3928 } else if (location
.type
== BTRFS_INODE_EXTREF_KEY
) {
3929 struct btrfs_inode_extref
*extref
;
3931 extref
= (struct btrfs_inode_extref
*)ptr
;
3932 BTRFS_I(inode
)->dir_index
= btrfs_inode_extref_index(leaf
,
3937 * try to precache a NULL acl entry for files that don't have
3938 * any xattrs or acls
3940 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
3941 btrfs_ino(BTRFS_I(inode
)), &first_xattr_slot
);
3942 if (first_xattr_slot
!= -1) {
3943 path
->slots
[0] = first_xattr_slot
;
3944 ret
= btrfs_load_inode_props(inode
, path
);
3947 "error loading props for ino %llu (root %llu): %d",
3948 btrfs_ino(BTRFS_I(inode
)),
3949 root
->root_key
.objectid
, ret
);
3951 btrfs_free_path(path
);
3954 cache_no_acl(inode
);
3956 switch (inode
->i_mode
& S_IFMT
) {
3958 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3959 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3960 inode
->i_fop
= &btrfs_file_operations
;
3961 inode
->i_op
= &btrfs_file_inode_operations
;
3964 inode
->i_fop
= &btrfs_dir_file_operations
;
3965 inode
->i_op
= &btrfs_dir_inode_operations
;
3968 inode
->i_op
= &btrfs_symlink_inode_operations
;
3969 inode_nohighmem(inode
);
3970 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
3973 inode
->i_op
= &btrfs_special_inode_operations
;
3974 init_special_inode(inode
, inode
->i_mode
, rdev
);
3978 btrfs_update_iflags(inode
);
3982 btrfs_free_path(path
);
3983 make_bad_inode(inode
);
3988 * given a leaf and an inode, copy the inode fields into the leaf
3990 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3991 struct extent_buffer
*leaf
,
3992 struct btrfs_inode_item
*item
,
3993 struct inode
*inode
)
3995 struct btrfs_map_token token
;
3997 btrfs_init_map_token(&token
);
3999 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
4000 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
4001 btrfs_set_token_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
,
4003 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
4004 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
4006 btrfs_set_token_timespec_sec(leaf
, &item
->atime
,
4007 inode
->i_atime
.tv_sec
, &token
);
4008 btrfs_set_token_timespec_nsec(leaf
, &item
->atime
,
4009 inode
->i_atime
.tv_nsec
, &token
);
4011 btrfs_set_token_timespec_sec(leaf
, &item
->mtime
,
4012 inode
->i_mtime
.tv_sec
, &token
);
4013 btrfs_set_token_timespec_nsec(leaf
, &item
->mtime
,
4014 inode
->i_mtime
.tv_nsec
, &token
);
4016 btrfs_set_token_timespec_sec(leaf
, &item
->ctime
,
4017 inode
->i_ctime
.tv_sec
, &token
);
4018 btrfs_set_token_timespec_nsec(leaf
, &item
->ctime
,
4019 inode
->i_ctime
.tv_nsec
, &token
);
4021 btrfs_set_token_timespec_sec(leaf
, &item
->otime
,
4022 BTRFS_I(inode
)->i_otime
.tv_sec
, &token
);
4023 btrfs_set_token_timespec_nsec(leaf
, &item
->otime
,
4024 BTRFS_I(inode
)->i_otime
.tv_nsec
, &token
);
4026 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
4028 btrfs_set_token_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
,
4030 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
4031 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
4032 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
4033 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
4034 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
4038 * copy everything in the in-memory inode into the btree.
4040 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
4041 struct btrfs_root
*root
, struct inode
*inode
)
4043 struct btrfs_inode_item
*inode_item
;
4044 struct btrfs_path
*path
;
4045 struct extent_buffer
*leaf
;
4048 path
= btrfs_alloc_path();
4052 path
->leave_spinning
= 1;
4053 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
4061 leaf
= path
->nodes
[0];
4062 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4063 struct btrfs_inode_item
);
4065 fill_inode_item(trans
, leaf
, inode_item
, inode
);
4066 btrfs_mark_buffer_dirty(leaf
);
4067 btrfs_set_inode_last_trans(trans
, inode
);
4070 btrfs_free_path(path
);
4075 * copy everything in the in-memory inode into the btree.
4077 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
4078 struct btrfs_root
*root
, struct inode
*inode
)
4080 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4084 * If the inode is a free space inode, we can deadlock during commit
4085 * if we put it into the delayed code.
4087 * The data relocation inode should also be directly updated
4090 if (!btrfs_is_free_space_inode(BTRFS_I(inode
))
4091 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
4092 && !test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
)) {
4093 btrfs_update_root_times(trans
, root
);
4095 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
4097 btrfs_set_inode_last_trans(trans
, inode
);
4101 return btrfs_update_inode_item(trans
, root
, inode
);
4104 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
4105 struct btrfs_root
*root
,
4106 struct inode
*inode
)
4110 ret
= btrfs_update_inode(trans
, root
, inode
);
4112 return btrfs_update_inode_item(trans
, root
, inode
);
4117 * unlink helper that gets used here in inode.c and in the tree logging
4118 * recovery code. It remove a link in a directory with a given name, and
4119 * also drops the back refs in the inode to the directory
4121 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4122 struct btrfs_root
*root
,
4123 struct btrfs_inode
*dir
,
4124 struct btrfs_inode
*inode
,
4125 const char *name
, int name_len
)
4127 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4128 struct btrfs_path
*path
;
4130 struct extent_buffer
*leaf
;
4131 struct btrfs_dir_item
*di
;
4132 struct btrfs_key key
;
4134 u64 ino
= btrfs_ino(inode
);
4135 u64 dir_ino
= btrfs_ino(dir
);
4137 path
= btrfs_alloc_path();
4143 path
->leave_spinning
= 1;
4144 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4145 name
, name_len
, -1);
4154 leaf
= path
->nodes
[0];
4155 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4156 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4159 btrfs_release_path(path
);
4162 * If we don't have dir index, we have to get it by looking up
4163 * the inode ref, since we get the inode ref, remove it directly,
4164 * it is unnecessary to do delayed deletion.
4166 * But if we have dir index, needn't search inode ref to get it.
4167 * Since the inode ref is close to the inode item, it is better
4168 * that we delay to delete it, and just do this deletion when
4169 * we update the inode item.
4171 if (inode
->dir_index
) {
4172 ret
= btrfs_delayed_delete_inode_ref(inode
);
4174 index
= inode
->dir_index
;
4179 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
4183 "failed to delete reference to %.*s, inode %llu parent %llu",
4184 name_len
, name
, ino
, dir_ino
);
4185 btrfs_abort_transaction(trans
, ret
);
4189 ret
= btrfs_delete_delayed_dir_index(trans
, fs_info
, dir
, index
);
4191 btrfs_abort_transaction(trans
, ret
);
4195 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
, inode
,
4197 if (ret
!= 0 && ret
!= -ENOENT
) {
4198 btrfs_abort_transaction(trans
, ret
);
4202 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
, dir
,
4207 btrfs_abort_transaction(trans
, ret
);
4209 btrfs_free_path(path
);
4213 btrfs_i_size_write(dir
, dir
->vfs_inode
.i_size
- name_len
* 2);
4214 inode_inc_iversion(&inode
->vfs_inode
);
4215 inode_inc_iversion(&dir
->vfs_inode
);
4216 inode
->vfs_inode
.i_ctime
= dir
->vfs_inode
.i_mtime
=
4217 dir
->vfs_inode
.i_ctime
= current_time(&inode
->vfs_inode
);
4218 ret
= btrfs_update_inode(trans
, root
, &dir
->vfs_inode
);
4223 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4224 struct btrfs_root
*root
,
4225 struct btrfs_inode
*dir
, struct btrfs_inode
*inode
,
4226 const char *name
, int name_len
)
4229 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
4231 drop_nlink(&inode
->vfs_inode
);
4232 ret
= btrfs_update_inode(trans
, root
, &inode
->vfs_inode
);
4238 * helper to start transaction for unlink and rmdir.
4240 * unlink and rmdir are special in btrfs, they do not always free space, so
4241 * if we cannot make our reservations the normal way try and see if there is
4242 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4243 * allow the unlink to occur.
4245 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
)
4247 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4250 * 1 for the possible orphan item
4251 * 1 for the dir item
4252 * 1 for the dir index
4253 * 1 for the inode ref
4256 return btrfs_start_transaction_fallback_global_rsv(root
, 5, 5);
4259 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
4261 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4262 struct btrfs_trans_handle
*trans
;
4263 struct inode
*inode
= d_inode(dentry
);
4266 trans
= __unlink_start_trans(dir
);
4268 return PTR_ERR(trans
);
4270 btrfs_record_unlink_dir(trans
, BTRFS_I(dir
), BTRFS_I(d_inode(dentry
)),
4273 ret
= btrfs_unlink_inode(trans
, root
, BTRFS_I(dir
),
4274 BTRFS_I(d_inode(dentry
)), dentry
->d_name
.name
,
4275 dentry
->d_name
.len
);
4279 if (inode
->i_nlink
== 0) {
4280 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
4286 btrfs_end_transaction(trans
);
4287 btrfs_btree_balance_dirty(root
->fs_info
);
4291 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
4292 struct btrfs_root
*root
,
4293 struct inode
*dir
, u64 objectid
,
4294 const char *name
, int name_len
)
4296 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4297 struct btrfs_path
*path
;
4298 struct extent_buffer
*leaf
;
4299 struct btrfs_dir_item
*di
;
4300 struct btrfs_key key
;
4303 u64 dir_ino
= btrfs_ino(BTRFS_I(dir
));
4305 path
= btrfs_alloc_path();
4309 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4310 name
, name_len
, -1);
4311 if (IS_ERR_OR_NULL(di
)) {
4319 leaf
= path
->nodes
[0];
4320 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4321 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
4322 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4324 btrfs_abort_transaction(trans
, ret
);
4327 btrfs_release_path(path
);
4329 ret
= btrfs_del_root_ref(trans
, fs_info
, objectid
,
4330 root
->root_key
.objectid
, dir_ino
,
4331 &index
, name
, name_len
);
4333 if (ret
!= -ENOENT
) {
4334 btrfs_abort_transaction(trans
, ret
);
4337 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
4339 if (IS_ERR_OR_NULL(di
)) {
4344 btrfs_abort_transaction(trans
, ret
);
4348 leaf
= path
->nodes
[0];
4349 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4350 btrfs_release_path(path
);
4353 btrfs_release_path(path
);
4355 ret
= btrfs_delete_delayed_dir_index(trans
, fs_info
, BTRFS_I(dir
), index
);
4357 btrfs_abort_transaction(trans
, ret
);
4361 btrfs_i_size_write(BTRFS_I(dir
), dir
->i_size
- name_len
* 2);
4362 inode_inc_iversion(dir
);
4363 dir
->i_mtime
= dir
->i_ctime
= current_time(dir
);
4364 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
4366 btrfs_abort_transaction(trans
, ret
);
4368 btrfs_free_path(path
);
4372 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
4374 struct inode
*inode
= d_inode(dentry
);
4376 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4377 struct btrfs_trans_handle
*trans
;
4378 u64 last_unlink_trans
;
4380 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
4382 if (btrfs_ino(BTRFS_I(inode
)) == BTRFS_FIRST_FREE_OBJECTID
)
4385 trans
= __unlink_start_trans(dir
);
4387 return PTR_ERR(trans
);
4389 if (unlikely(btrfs_ino(BTRFS_I(inode
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
4390 err
= btrfs_unlink_subvol(trans
, root
, dir
,
4391 BTRFS_I(inode
)->location
.objectid
,
4392 dentry
->d_name
.name
,
4393 dentry
->d_name
.len
);
4397 err
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
4401 last_unlink_trans
= BTRFS_I(inode
)->last_unlink_trans
;
4403 /* now the directory is empty */
4404 err
= btrfs_unlink_inode(trans
, root
, BTRFS_I(dir
),
4405 BTRFS_I(d_inode(dentry
)), dentry
->d_name
.name
,
4406 dentry
->d_name
.len
);
4408 btrfs_i_size_write(BTRFS_I(inode
), 0);
4410 * Propagate the last_unlink_trans value of the deleted dir to
4411 * its parent directory. This is to prevent an unrecoverable
4412 * log tree in the case we do something like this:
4414 * 2) create snapshot under dir foo
4415 * 3) delete the snapshot
4418 * 6) fsync foo or some file inside foo
4420 if (last_unlink_trans
>= trans
->transid
)
4421 BTRFS_I(dir
)->last_unlink_trans
= last_unlink_trans
;
4424 btrfs_end_transaction(trans
);
4425 btrfs_btree_balance_dirty(root
->fs_info
);
4430 static int truncate_space_check(struct btrfs_trans_handle
*trans
,
4431 struct btrfs_root
*root
,
4434 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4438 * This is only used to apply pressure to the enospc system, we don't
4439 * intend to use this reservation at all.
4441 bytes_deleted
= btrfs_csum_bytes_to_leaves(fs_info
, bytes_deleted
);
4442 bytes_deleted
*= fs_info
->nodesize
;
4443 ret
= btrfs_block_rsv_add(root
, &fs_info
->trans_block_rsv
,
4444 bytes_deleted
, BTRFS_RESERVE_NO_FLUSH
);
4446 trace_btrfs_space_reservation(fs_info
, "transaction",
4449 trans
->bytes_reserved
+= bytes_deleted
;
4455 static int truncate_inline_extent(struct inode
*inode
,
4456 struct btrfs_path
*path
,
4457 struct btrfs_key
*found_key
,
4461 struct extent_buffer
*leaf
= path
->nodes
[0];
4462 int slot
= path
->slots
[0];
4463 struct btrfs_file_extent_item
*fi
;
4464 u32 size
= (u32
)(new_size
- found_key
->offset
);
4465 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4467 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
4469 if (btrfs_file_extent_compression(leaf
, fi
) != BTRFS_COMPRESS_NONE
) {
4470 loff_t offset
= new_size
;
4471 loff_t page_end
= ALIGN(offset
, PAGE_SIZE
);
4474 * Zero out the remaining of the last page of our inline extent,
4475 * instead of directly truncating our inline extent here - that
4476 * would be much more complex (decompressing all the data, then
4477 * compressing the truncated data, which might be bigger than
4478 * the size of the inline extent, resize the extent, etc).
4479 * We release the path because to get the page we might need to
4480 * read the extent item from disk (data not in the page cache).
4482 btrfs_release_path(path
);
4483 return btrfs_truncate_block(inode
, offset
, page_end
- offset
,
4487 btrfs_set_file_extent_ram_bytes(leaf
, fi
, size
);
4488 size
= btrfs_file_extent_calc_inline_size(size
);
4489 btrfs_truncate_item(root
->fs_info
, path
, size
, 1);
4491 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4492 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4498 * this can truncate away extent items, csum items and directory items.
4499 * It starts at a high offset and removes keys until it can't find
4500 * any higher than new_size
4502 * csum items that cross the new i_size are truncated to the new size
4505 * min_type is the minimum key type to truncate down to. If set to 0, this
4506 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4508 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
4509 struct btrfs_root
*root
,
4510 struct inode
*inode
,
4511 u64 new_size
, u32 min_type
)
4513 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4514 struct btrfs_path
*path
;
4515 struct extent_buffer
*leaf
;
4516 struct btrfs_file_extent_item
*fi
;
4517 struct btrfs_key key
;
4518 struct btrfs_key found_key
;
4519 u64 extent_start
= 0;
4520 u64 extent_num_bytes
= 0;
4521 u64 extent_offset
= 0;
4523 u64 last_size
= new_size
;
4524 u32 found_type
= (u8
)-1;
4527 int pending_del_nr
= 0;
4528 int pending_del_slot
= 0;
4529 int extent_type
= -1;
4532 u64 ino
= btrfs_ino(BTRFS_I(inode
));
4533 u64 bytes_deleted
= 0;
4535 bool should_throttle
= 0;
4536 bool should_end
= 0;
4538 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
4541 * for non-free space inodes and ref cows, we want to back off from
4544 if (!btrfs_is_free_space_inode(BTRFS_I(inode
)) &&
4545 test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4548 path
= btrfs_alloc_path();
4551 path
->reada
= READA_BACK
;
4554 * We want to drop from the next block forward in case this new size is
4555 * not block aligned since we will be keeping the last block of the
4556 * extent just the way it is.
4558 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4559 root
== fs_info
->tree_root
)
4560 btrfs_drop_extent_cache(BTRFS_I(inode
), ALIGN(new_size
,
4561 fs_info
->sectorsize
),
4565 * This function is also used to drop the items in the log tree before
4566 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4567 * it is used to drop the loged items. So we shouldn't kill the delayed
4570 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
4571 btrfs_kill_delayed_inode_items(BTRFS_I(inode
));
4574 key
.offset
= (u64
)-1;
4579 * with a 16K leaf size and 128MB extents, you can actually queue
4580 * up a huge file in a single leaf. Most of the time that
4581 * bytes_deleted is > 0, it will be huge by the time we get here
4583 if (be_nice
&& bytes_deleted
> SZ_32M
) {
4584 if (btrfs_should_end_transaction(trans
)) {
4591 path
->leave_spinning
= 1;
4592 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
4599 /* there are no items in the tree for us to truncate, we're
4602 if (path
->slots
[0] == 0)
4609 leaf
= path
->nodes
[0];
4610 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4611 found_type
= found_key
.type
;
4613 if (found_key
.objectid
!= ino
)
4616 if (found_type
< min_type
)
4619 item_end
= found_key
.offset
;
4620 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
4621 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4622 struct btrfs_file_extent_item
);
4623 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4624 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4626 btrfs_file_extent_num_bytes(leaf
, fi
);
4628 trace_btrfs_truncate_show_fi_regular(
4629 BTRFS_I(inode
), leaf
, fi
,
4631 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4632 item_end
+= btrfs_file_extent_inline_len(leaf
,
4633 path
->slots
[0], fi
);
4635 trace_btrfs_truncate_show_fi_inline(
4636 BTRFS_I(inode
), leaf
, fi
, path
->slots
[0],
4641 if (found_type
> min_type
) {
4644 if (item_end
< new_size
)
4646 if (found_key
.offset
>= new_size
)
4652 /* FIXME, shrink the extent if the ref count is only 1 */
4653 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
4657 last_size
= found_key
.offset
;
4659 last_size
= new_size
;
4661 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4663 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4665 u64 orig_num_bytes
=
4666 btrfs_file_extent_num_bytes(leaf
, fi
);
4667 extent_num_bytes
= ALIGN(new_size
-
4669 fs_info
->sectorsize
);
4670 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4672 num_dec
= (orig_num_bytes
-
4674 if (test_bit(BTRFS_ROOT_REF_COWS
,
4677 inode_sub_bytes(inode
, num_dec
);
4678 btrfs_mark_buffer_dirty(leaf
);
4681 btrfs_file_extent_disk_num_bytes(leaf
,
4683 extent_offset
= found_key
.offset
-
4684 btrfs_file_extent_offset(leaf
, fi
);
4686 /* FIXME blocksize != 4096 */
4687 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
4688 if (extent_start
!= 0) {
4690 if (test_bit(BTRFS_ROOT_REF_COWS
,
4692 inode_sub_bytes(inode
, num_dec
);
4695 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4697 * we can't truncate inline items that have had
4701 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
4702 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
4705 * Need to release path in order to truncate a
4706 * compressed extent. So delete any accumulated
4707 * extent items so far.
4709 if (btrfs_file_extent_compression(leaf
, fi
) !=
4710 BTRFS_COMPRESS_NONE
&& pending_del_nr
) {
4711 err
= btrfs_del_items(trans
, root
, path
,
4715 btrfs_abort_transaction(trans
,
4722 err
= truncate_inline_extent(inode
, path
,
4727 btrfs_abort_transaction(trans
, err
);
4730 } else if (test_bit(BTRFS_ROOT_REF_COWS
,
4732 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4737 if (!pending_del_nr
) {
4738 /* no pending yet, add ourselves */
4739 pending_del_slot
= path
->slots
[0];
4741 } else if (pending_del_nr
&&
4742 path
->slots
[0] + 1 == pending_del_slot
) {
4743 /* hop on the pending chunk */
4745 pending_del_slot
= path
->slots
[0];
4752 should_throttle
= 0;
4755 (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4756 root
== fs_info
->tree_root
)) {
4757 btrfs_set_path_blocking(path
);
4758 bytes_deleted
+= extent_num_bytes
;
4759 ret
= btrfs_free_extent(trans
, fs_info
, extent_start
,
4760 extent_num_bytes
, 0,
4761 btrfs_header_owner(leaf
),
4762 ino
, extent_offset
);
4764 btrfs_abort_transaction(trans
, ret
);
4767 if (btrfs_should_throttle_delayed_refs(trans
, fs_info
))
4768 btrfs_async_run_delayed_refs(fs_info
,
4769 trans
->delayed_ref_updates
* 2,
4772 if (truncate_space_check(trans
, root
,
4773 extent_num_bytes
)) {
4776 if (btrfs_should_throttle_delayed_refs(trans
,
4778 should_throttle
= 1;
4782 if (found_type
== BTRFS_INODE_ITEM_KEY
)
4785 if (path
->slots
[0] == 0 ||
4786 path
->slots
[0] != pending_del_slot
||
4787 should_throttle
|| should_end
) {
4788 if (pending_del_nr
) {
4789 ret
= btrfs_del_items(trans
, root
, path
,
4793 btrfs_abort_transaction(trans
, ret
);
4798 btrfs_release_path(path
);
4799 if (should_throttle
) {
4800 unsigned long updates
= trans
->delayed_ref_updates
;
4802 trans
->delayed_ref_updates
= 0;
4803 ret
= btrfs_run_delayed_refs(trans
,
4811 * if we failed to refill our space rsv, bail out
4812 * and let the transaction restart
4824 if (pending_del_nr
) {
4825 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
4828 btrfs_abort_transaction(trans
, ret
);
4831 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4832 ASSERT(last_size
>= new_size
);
4833 if (!err
&& last_size
> new_size
)
4834 last_size
= new_size
;
4835 btrfs_ordered_update_i_size(inode
, last_size
, NULL
);
4838 btrfs_free_path(path
);
4840 if (be_nice
&& bytes_deleted
> SZ_32M
) {
4841 unsigned long updates
= trans
->delayed_ref_updates
;
4843 trans
->delayed_ref_updates
= 0;
4844 ret
= btrfs_run_delayed_refs(trans
, fs_info
,
4854 * btrfs_truncate_block - read, zero a chunk and write a block
4855 * @inode - inode that we're zeroing
4856 * @from - the offset to start zeroing
4857 * @len - the length to zero, 0 to zero the entire range respective to the
4859 * @front - zero up to the offset instead of from the offset on
4861 * This will find the block for the "from" offset and cow the block and zero the
4862 * part we want to zero. This is used with truncate and hole punching.
4864 int btrfs_truncate_block(struct inode
*inode
, loff_t from
, loff_t len
,
4867 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4868 struct address_space
*mapping
= inode
->i_mapping
;
4869 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4870 struct btrfs_ordered_extent
*ordered
;
4871 struct extent_state
*cached_state
= NULL
;
4872 struct extent_changeset
*data_reserved
= NULL
;
4874 u32 blocksize
= fs_info
->sectorsize
;
4875 pgoff_t index
= from
>> PAGE_SHIFT
;
4876 unsigned offset
= from
& (blocksize
- 1);
4878 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4883 if ((offset
& (blocksize
- 1)) == 0 &&
4884 (!len
|| ((len
& (blocksize
- 1)) == 0)))
4887 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
,
4888 round_down(from
, blocksize
), blocksize
);
4893 page
= find_or_create_page(mapping
, index
, mask
);
4895 btrfs_delalloc_release_space(inode
, data_reserved
,
4896 round_down(from
, blocksize
),
4902 block_start
= round_down(from
, blocksize
);
4903 block_end
= block_start
+ blocksize
- 1;
4905 if (!PageUptodate(page
)) {
4906 ret
= btrfs_readpage(NULL
, page
);
4908 if (page
->mapping
!= mapping
) {
4913 if (!PageUptodate(page
)) {
4918 wait_on_page_writeback(page
);
4920 lock_extent_bits(io_tree
, block_start
, block_end
, &cached_state
);
4921 set_page_extent_mapped(page
);
4923 ordered
= btrfs_lookup_ordered_extent(inode
, block_start
);
4925 unlock_extent_cached(io_tree
, block_start
, block_end
,
4926 &cached_state
, GFP_NOFS
);
4929 btrfs_start_ordered_extent(inode
, ordered
, 1);
4930 btrfs_put_ordered_extent(ordered
);
4934 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, block_start
, block_end
,
4935 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4936 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4937 0, 0, &cached_state
, GFP_NOFS
);
4939 ret
= btrfs_set_extent_delalloc(inode
, block_start
, block_end
,
4942 unlock_extent_cached(io_tree
, block_start
, block_end
,
4943 &cached_state
, GFP_NOFS
);
4947 if (offset
!= blocksize
) {
4949 len
= blocksize
- offset
;
4952 memset(kaddr
+ (block_start
- page_offset(page
)),
4955 memset(kaddr
+ (block_start
- page_offset(page
)) + offset
,
4957 flush_dcache_page(page
);
4960 ClearPageChecked(page
);
4961 set_page_dirty(page
);
4962 unlock_extent_cached(io_tree
, block_start
, block_end
, &cached_state
,
4967 btrfs_delalloc_release_space(inode
, data_reserved
, block_start
,
4972 extent_changeset_free(data_reserved
);
4976 static int maybe_insert_hole(struct btrfs_root
*root
, struct inode
*inode
,
4977 u64 offset
, u64 len
)
4979 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4980 struct btrfs_trans_handle
*trans
;
4984 * Still need to make sure the inode looks like it's been updated so
4985 * that any holes get logged if we fsync.
4987 if (btrfs_fs_incompat(fs_info
, NO_HOLES
)) {
4988 BTRFS_I(inode
)->last_trans
= fs_info
->generation
;
4989 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
4990 BTRFS_I(inode
)->last_log_commit
= root
->last_log_commit
;
4995 * 1 - for the one we're dropping
4996 * 1 - for the one we're adding
4997 * 1 - for updating the inode.
4999 trans
= btrfs_start_transaction(root
, 3);
5001 return PTR_ERR(trans
);
5003 ret
= btrfs_drop_extents(trans
, root
, inode
, offset
, offset
+ len
, 1);
5005 btrfs_abort_transaction(trans
, ret
);
5006 btrfs_end_transaction(trans
);
5010 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(BTRFS_I(inode
)),
5011 offset
, 0, 0, len
, 0, len
, 0, 0, 0);
5013 btrfs_abort_transaction(trans
, ret
);
5015 btrfs_update_inode(trans
, root
, inode
);
5016 btrfs_end_transaction(trans
);
5021 * This function puts in dummy file extents for the area we're creating a hole
5022 * for. So if we are truncating this file to a larger size we need to insert
5023 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5024 * the range between oldsize and size
5026 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
5028 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5029 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5030 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5031 struct extent_map
*em
= NULL
;
5032 struct extent_state
*cached_state
= NULL
;
5033 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5034 u64 hole_start
= ALIGN(oldsize
, fs_info
->sectorsize
);
5035 u64 block_end
= ALIGN(size
, fs_info
->sectorsize
);
5042 * If our size started in the middle of a block we need to zero out the
5043 * rest of the block before we expand the i_size, otherwise we could
5044 * expose stale data.
5046 err
= btrfs_truncate_block(inode
, oldsize
, 0, 0);
5050 if (size
<= hole_start
)
5054 struct btrfs_ordered_extent
*ordered
;
5056 lock_extent_bits(io_tree
, hole_start
, block_end
- 1,
5058 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), hole_start
,
5059 block_end
- hole_start
);
5062 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
5063 &cached_state
, GFP_NOFS
);
5064 btrfs_start_ordered_extent(inode
, ordered
, 1);
5065 btrfs_put_ordered_extent(ordered
);
5068 cur_offset
= hole_start
;
5070 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0, cur_offset
,
5071 block_end
- cur_offset
, 0);
5077 last_byte
= min(extent_map_end(em
), block_end
);
5078 last_byte
= ALIGN(last_byte
, fs_info
->sectorsize
);
5079 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
5080 struct extent_map
*hole_em
;
5081 hole_size
= last_byte
- cur_offset
;
5083 err
= maybe_insert_hole(root
, inode
, cur_offset
,
5087 btrfs_drop_extent_cache(BTRFS_I(inode
), cur_offset
,
5088 cur_offset
+ hole_size
- 1, 0);
5089 hole_em
= alloc_extent_map();
5091 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
5092 &BTRFS_I(inode
)->runtime_flags
);
5095 hole_em
->start
= cur_offset
;
5096 hole_em
->len
= hole_size
;
5097 hole_em
->orig_start
= cur_offset
;
5099 hole_em
->block_start
= EXTENT_MAP_HOLE
;
5100 hole_em
->block_len
= 0;
5101 hole_em
->orig_block_len
= 0;
5102 hole_em
->ram_bytes
= hole_size
;
5103 hole_em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
5104 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
5105 hole_em
->generation
= fs_info
->generation
;
5108 write_lock(&em_tree
->lock
);
5109 err
= add_extent_mapping(em_tree
, hole_em
, 1);
5110 write_unlock(&em_tree
->lock
);
5113 btrfs_drop_extent_cache(BTRFS_I(inode
),
5118 free_extent_map(hole_em
);
5121 free_extent_map(em
);
5123 cur_offset
= last_byte
;
5124 if (cur_offset
>= block_end
)
5127 free_extent_map(em
);
5128 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
5133 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
5135 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5136 struct btrfs_trans_handle
*trans
;
5137 loff_t oldsize
= i_size_read(inode
);
5138 loff_t newsize
= attr
->ia_size
;
5139 int mask
= attr
->ia_valid
;
5143 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5144 * special case where we need to update the times despite not having
5145 * these flags set. For all other operations the VFS set these flags
5146 * explicitly if it wants a timestamp update.
5148 if (newsize
!= oldsize
) {
5149 inode_inc_iversion(inode
);
5150 if (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))
5151 inode
->i_ctime
= inode
->i_mtime
=
5152 current_time(inode
);
5155 if (newsize
> oldsize
) {
5157 * Don't do an expanding truncate while snapshotting is ongoing.
5158 * This is to ensure the snapshot captures a fully consistent
5159 * state of this file - if the snapshot captures this expanding
5160 * truncation, it must capture all writes that happened before
5163 btrfs_wait_for_snapshot_creation(root
);
5164 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
5166 btrfs_end_write_no_snapshotting(root
);
5170 trans
= btrfs_start_transaction(root
, 1);
5171 if (IS_ERR(trans
)) {
5172 btrfs_end_write_no_snapshotting(root
);
5173 return PTR_ERR(trans
);
5176 i_size_write(inode
, newsize
);
5177 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
5178 pagecache_isize_extended(inode
, oldsize
, newsize
);
5179 ret
= btrfs_update_inode(trans
, root
, inode
);
5180 btrfs_end_write_no_snapshotting(root
);
5181 btrfs_end_transaction(trans
);
5185 * We're truncating a file that used to have good data down to
5186 * zero. Make sure it gets into the ordered flush list so that
5187 * any new writes get down to disk quickly.
5190 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
5191 &BTRFS_I(inode
)->runtime_flags
);
5194 * 1 for the orphan item we're going to add
5195 * 1 for the orphan item deletion.
5197 trans
= btrfs_start_transaction(root
, 2);
5199 return PTR_ERR(trans
);
5202 * We need to do this in case we fail at _any_ point during the
5203 * actual truncate. Once we do the truncate_setsize we could
5204 * invalidate pages which forces any outstanding ordered io to
5205 * be instantly completed which will give us extents that need
5206 * to be truncated. If we fail to get an orphan inode down we
5207 * could have left over extents that were never meant to live,
5208 * so we need to guarantee from this point on that everything
5209 * will be consistent.
5211 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
5212 btrfs_end_transaction(trans
);
5216 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5217 truncate_setsize(inode
, newsize
);
5219 /* Disable nonlocked read DIO to avoid the end less truncate */
5220 btrfs_inode_block_unlocked_dio(BTRFS_I(inode
));
5221 inode_dio_wait(inode
);
5222 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode
));
5224 ret
= btrfs_truncate(inode
);
5225 if (ret
&& inode
->i_nlink
) {
5228 /* To get a stable disk_i_size */
5229 err
= btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5231 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5236 * failed to truncate, disk_i_size is only adjusted down
5237 * as we remove extents, so it should represent the true
5238 * size of the inode, so reset the in memory size and
5239 * delete our orphan entry.
5241 trans
= btrfs_join_transaction(root
);
5242 if (IS_ERR(trans
)) {
5243 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5246 i_size_write(inode
, BTRFS_I(inode
)->disk_i_size
);
5247 err
= btrfs_orphan_del(trans
, BTRFS_I(inode
));
5249 btrfs_abort_transaction(trans
, err
);
5250 btrfs_end_transaction(trans
);
5257 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5259 struct inode
*inode
= d_inode(dentry
);
5260 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5263 if (btrfs_root_readonly(root
))
5266 err
= setattr_prepare(dentry
, attr
);
5270 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
5271 err
= btrfs_setsize(inode
, attr
);
5276 if (attr
->ia_valid
) {
5277 setattr_copy(inode
, attr
);
5278 inode_inc_iversion(inode
);
5279 err
= btrfs_dirty_inode(inode
);
5281 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
5282 err
= posix_acl_chmod(inode
, inode
->i_mode
);
5289 * While truncating the inode pages during eviction, we get the VFS calling
5290 * btrfs_invalidatepage() against each page of the inode. This is slow because
5291 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5292 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5293 * extent_state structures over and over, wasting lots of time.
5295 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5296 * those expensive operations on a per page basis and do only the ordered io
5297 * finishing, while we release here the extent_map and extent_state structures,
5298 * without the excessive merging and splitting.
5300 static void evict_inode_truncate_pages(struct inode
*inode
)
5302 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5303 struct extent_map_tree
*map_tree
= &BTRFS_I(inode
)->extent_tree
;
5304 struct rb_node
*node
;
5306 ASSERT(inode
->i_state
& I_FREEING
);
5307 truncate_inode_pages_final(&inode
->i_data
);
5309 write_lock(&map_tree
->lock
);
5310 while (!RB_EMPTY_ROOT(&map_tree
->map
)) {
5311 struct extent_map
*em
;
5313 node
= rb_first(&map_tree
->map
);
5314 em
= rb_entry(node
, struct extent_map
, rb_node
);
5315 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5316 clear_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
5317 remove_extent_mapping(map_tree
, em
);
5318 free_extent_map(em
);
5319 if (need_resched()) {
5320 write_unlock(&map_tree
->lock
);
5322 write_lock(&map_tree
->lock
);
5325 write_unlock(&map_tree
->lock
);
5328 * Keep looping until we have no more ranges in the io tree.
5329 * We can have ongoing bios started by readpages (called from readahead)
5330 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5331 * still in progress (unlocked the pages in the bio but did not yet
5332 * unlocked the ranges in the io tree). Therefore this means some
5333 * ranges can still be locked and eviction started because before
5334 * submitting those bios, which are executed by a separate task (work
5335 * queue kthread), inode references (inode->i_count) were not taken
5336 * (which would be dropped in the end io callback of each bio).
5337 * Therefore here we effectively end up waiting for those bios and
5338 * anyone else holding locked ranges without having bumped the inode's
5339 * reference count - if we don't do it, when they access the inode's
5340 * io_tree to unlock a range it may be too late, leading to an
5341 * use-after-free issue.
5343 spin_lock(&io_tree
->lock
);
5344 while (!RB_EMPTY_ROOT(&io_tree
->state
)) {
5345 struct extent_state
*state
;
5346 struct extent_state
*cached_state
= NULL
;
5349 unsigned state_flags
;
5351 node
= rb_first(&io_tree
->state
);
5352 state
= rb_entry(node
, struct extent_state
, rb_node
);
5353 start
= state
->start
;
5355 state_flags
= state
->state
;
5356 spin_unlock(&io_tree
->lock
);
5358 lock_extent_bits(io_tree
, start
, end
, &cached_state
);
5361 * If still has DELALLOC flag, the extent didn't reach disk,
5362 * and its reserved space won't be freed by delayed_ref.
5363 * So we need to free its reserved space here.
5364 * (Refer to comment in btrfs_invalidatepage, case 2)
5366 * Note, end is the bytenr of last byte, so we need + 1 here.
5368 if (state_flags
& EXTENT_DELALLOC
)
5369 btrfs_qgroup_free_data(inode
, NULL
, start
, end
- start
+ 1);
5371 clear_extent_bit(io_tree
, start
, end
,
5372 EXTENT_LOCKED
| EXTENT_DIRTY
|
5373 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
5374 EXTENT_DEFRAG
, 1, 1,
5375 &cached_state
, GFP_NOFS
);
5378 spin_lock(&io_tree
->lock
);
5380 spin_unlock(&io_tree
->lock
);
5383 void btrfs_evict_inode(struct inode
*inode
)
5385 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5386 struct btrfs_trans_handle
*trans
;
5387 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5388 struct btrfs_block_rsv
*rsv
, *global_rsv
;
5389 int steal_from_global
= 0;
5393 trace_btrfs_inode_evict(inode
);
5400 min_size
= btrfs_calc_trunc_metadata_size(fs_info
, 1);
5402 evict_inode_truncate_pages(inode
);
5404 if (inode
->i_nlink
&&
5405 ((btrfs_root_refs(&root
->root_item
) != 0 &&
5406 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
) ||
5407 btrfs_is_free_space_inode(BTRFS_I(inode
))))
5410 if (is_bad_inode(inode
)) {
5411 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5414 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5415 if (!special_file(inode
->i_mode
))
5416 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5418 btrfs_free_io_failure_record(BTRFS_I(inode
), 0, (u64
)-1);
5420 if (test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
)) {
5421 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
5422 &BTRFS_I(inode
)->runtime_flags
));
5426 if (inode
->i_nlink
> 0) {
5427 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0 &&
5428 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
);
5432 ret
= btrfs_commit_inode_delayed_inode(BTRFS_I(inode
));
5434 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5438 rsv
= btrfs_alloc_block_rsv(fs_info
, BTRFS_BLOCK_RSV_TEMP
);
5440 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5443 rsv
->size
= min_size
;
5445 global_rsv
= &fs_info
->global_block_rsv
;
5447 btrfs_i_size_write(BTRFS_I(inode
), 0);
5450 * This is a bit simpler than btrfs_truncate since we've already
5451 * reserved our space for our orphan item in the unlink, so we just
5452 * need to reserve some slack space in case we add bytes and update
5453 * inode item when doing the truncate.
5456 ret
= btrfs_block_rsv_refill(root
, rsv
, min_size
,
5457 BTRFS_RESERVE_FLUSH_LIMIT
);
5460 * Try and steal from the global reserve since we will
5461 * likely not use this space anyway, we want to try as
5462 * hard as possible to get this to work.
5465 steal_from_global
++;
5467 steal_from_global
= 0;
5471 * steal_from_global == 0: we reserved stuff, hooray!
5472 * steal_from_global == 1: we didn't reserve stuff, boo!
5473 * steal_from_global == 2: we've committed, still not a lot of
5474 * room but maybe we'll have room in the global reserve this
5476 * steal_from_global == 3: abandon all hope!
5478 if (steal_from_global
> 2) {
5480 "Could not get space for a delete, will truncate on mount %d",
5482 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5483 btrfs_free_block_rsv(fs_info
, rsv
);
5487 trans
= btrfs_join_transaction(root
);
5488 if (IS_ERR(trans
)) {
5489 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5490 btrfs_free_block_rsv(fs_info
, rsv
);
5495 * We can't just steal from the global reserve, we need to make
5496 * sure there is room to do it, if not we need to commit and try
5499 if (steal_from_global
) {
5500 if (!btrfs_check_space_for_delayed_refs(trans
, fs_info
))
5501 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
,
5508 * Couldn't steal from the global reserve, we have too much
5509 * pending stuff built up, commit the transaction and try it
5513 ret
= btrfs_commit_transaction(trans
);
5515 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5516 btrfs_free_block_rsv(fs_info
, rsv
);
5521 steal_from_global
= 0;
5524 trans
->block_rsv
= rsv
;
5526 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
5528 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
5529 btrfs_end_transaction(trans
);
5530 btrfs_btree_balance_dirty(fs_info
);
5531 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
) {
5532 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5533 btrfs_free_block_rsv(fs_info
, rsv
);
5541 btrfs_free_block_rsv(fs_info
, rsv
);
5544 * Errors here aren't a big deal, it just means we leave orphan items
5545 * in the tree. They will be cleaned up on the next mount.
5547 trans
->block_rsv
= root
->orphan_block_rsv
;
5548 btrfs_orphan_del(trans
, BTRFS_I(inode
));
5550 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
5551 if (!(root
== fs_info
->tree_root
||
5552 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
5553 btrfs_return_ino(root
, btrfs_ino(BTRFS_I(inode
)));
5555 btrfs_end_transaction(trans
);
5556 btrfs_btree_balance_dirty(fs_info
);
5558 btrfs_remove_delayed_node(BTRFS_I(inode
));
5563 * this returns the key found in the dir entry in the location pointer.
5564 * If no dir entries were found, location->objectid is 0.
5566 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
5567 struct btrfs_key
*location
)
5569 const char *name
= dentry
->d_name
.name
;
5570 int namelen
= dentry
->d_name
.len
;
5571 struct btrfs_dir_item
*di
;
5572 struct btrfs_path
*path
;
5573 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5576 path
= btrfs_alloc_path();
5580 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(BTRFS_I(dir
)),
5585 if (IS_ERR_OR_NULL(di
))
5588 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
5589 if (location
->type
!= BTRFS_INODE_ITEM_KEY
&&
5590 location
->type
!= BTRFS_ROOT_ITEM_KEY
) {
5591 btrfs_warn(root
->fs_info
,
5592 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5593 __func__
, name
, btrfs_ino(BTRFS_I(dir
)),
5594 location
->objectid
, location
->type
, location
->offset
);
5598 btrfs_free_path(path
);
5601 location
->objectid
= 0;
5606 * when we hit a tree root in a directory, the btrfs part of the inode
5607 * needs to be changed to reflect the root directory of the tree root. This
5608 * is kind of like crossing a mount point.
5610 static int fixup_tree_root_location(struct btrfs_fs_info
*fs_info
,
5612 struct dentry
*dentry
,
5613 struct btrfs_key
*location
,
5614 struct btrfs_root
**sub_root
)
5616 struct btrfs_path
*path
;
5617 struct btrfs_root
*new_root
;
5618 struct btrfs_root_ref
*ref
;
5619 struct extent_buffer
*leaf
;
5620 struct btrfs_key key
;
5624 path
= btrfs_alloc_path();
5631 key
.objectid
= BTRFS_I(dir
)->root
->root_key
.objectid
;
5632 key
.type
= BTRFS_ROOT_REF_KEY
;
5633 key
.offset
= location
->objectid
;
5635 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
5642 leaf
= path
->nodes
[0];
5643 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
5644 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(BTRFS_I(dir
)) ||
5645 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
5648 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
5649 (unsigned long)(ref
+ 1),
5650 dentry
->d_name
.len
);
5654 btrfs_release_path(path
);
5656 new_root
= btrfs_read_fs_root_no_name(fs_info
, location
);
5657 if (IS_ERR(new_root
)) {
5658 err
= PTR_ERR(new_root
);
5662 *sub_root
= new_root
;
5663 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
5664 location
->type
= BTRFS_INODE_ITEM_KEY
;
5665 location
->offset
= 0;
5668 btrfs_free_path(path
);
5672 static void inode_tree_add(struct inode
*inode
)
5674 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5675 struct btrfs_inode
*entry
;
5677 struct rb_node
*parent
;
5678 struct rb_node
*new = &BTRFS_I(inode
)->rb_node
;
5679 u64 ino
= btrfs_ino(BTRFS_I(inode
));
5681 if (inode_unhashed(inode
))
5684 spin_lock(&root
->inode_lock
);
5685 p
= &root
->inode_tree
.rb_node
;
5688 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
5690 if (ino
< btrfs_ino(BTRFS_I(&entry
->vfs_inode
)))
5691 p
= &parent
->rb_left
;
5692 else if (ino
> btrfs_ino(BTRFS_I(&entry
->vfs_inode
)))
5693 p
= &parent
->rb_right
;
5695 WARN_ON(!(entry
->vfs_inode
.i_state
&
5696 (I_WILL_FREE
| I_FREEING
)));
5697 rb_replace_node(parent
, new, &root
->inode_tree
);
5698 RB_CLEAR_NODE(parent
);
5699 spin_unlock(&root
->inode_lock
);
5703 rb_link_node(new, parent
, p
);
5704 rb_insert_color(new, &root
->inode_tree
);
5705 spin_unlock(&root
->inode_lock
);
5708 static void inode_tree_del(struct inode
*inode
)
5710 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5711 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5714 spin_lock(&root
->inode_lock
);
5715 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
5716 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
5717 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
5718 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5720 spin_unlock(&root
->inode_lock
);
5722 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0) {
5723 synchronize_srcu(&fs_info
->subvol_srcu
);
5724 spin_lock(&root
->inode_lock
);
5725 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5726 spin_unlock(&root
->inode_lock
);
5728 btrfs_add_dead_root(root
);
5732 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
5734 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5735 struct rb_node
*node
;
5736 struct rb_node
*prev
;
5737 struct btrfs_inode
*entry
;
5738 struct inode
*inode
;
5741 if (!test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
5742 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
5744 spin_lock(&root
->inode_lock
);
5746 node
= root
->inode_tree
.rb_node
;
5750 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5752 if (objectid
< btrfs_ino(BTRFS_I(&entry
->vfs_inode
)))
5753 node
= node
->rb_left
;
5754 else if (objectid
> btrfs_ino(BTRFS_I(&entry
->vfs_inode
)))
5755 node
= node
->rb_right
;
5761 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
5762 if (objectid
<= btrfs_ino(BTRFS_I(&entry
->vfs_inode
))) {
5766 prev
= rb_next(prev
);
5770 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5771 objectid
= btrfs_ino(BTRFS_I(&entry
->vfs_inode
)) + 1;
5772 inode
= igrab(&entry
->vfs_inode
);
5774 spin_unlock(&root
->inode_lock
);
5775 if (atomic_read(&inode
->i_count
) > 1)
5776 d_prune_aliases(inode
);
5778 * btrfs_drop_inode will have it removed from
5779 * the inode cache when its usage count
5784 spin_lock(&root
->inode_lock
);
5788 if (cond_resched_lock(&root
->inode_lock
))
5791 node
= rb_next(node
);
5793 spin_unlock(&root
->inode_lock
);
5796 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
5798 struct btrfs_iget_args
*args
= p
;
5799 inode
->i_ino
= args
->location
->objectid
;
5800 memcpy(&BTRFS_I(inode
)->location
, args
->location
,
5801 sizeof(*args
->location
));
5802 BTRFS_I(inode
)->root
= args
->root
;
5806 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
5808 struct btrfs_iget_args
*args
= opaque
;
5809 return args
->location
->objectid
== BTRFS_I(inode
)->location
.objectid
&&
5810 args
->root
== BTRFS_I(inode
)->root
;
5813 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
5814 struct btrfs_key
*location
,
5815 struct btrfs_root
*root
)
5817 struct inode
*inode
;
5818 struct btrfs_iget_args args
;
5819 unsigned long hashval
= btrfs_inode_hash(location
->objectid
, root
);
5821 args
.location
= location
;
5824 inode
= iget5_locked(s
, hashval
, btrfs_find_actor
,
5825 btrfs_init_locked_inode
,
5830 /* Get an inode object given its location and corresponding root.
5831 * Returns in *is_new if the inode was read from disk
5833 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
5834 struct btrfs_root
*root
, int *new)
5836 struct inode
*inode
;
5838 inode
= btrfs_iget_locked(s
, location
, root
);
5840 return ERR_PTR(-ENOMEM
);
5842 if (inode
->i_state
& I_NEW
) {
5845 ret
= btrfs_read_locked_inode(inode
);
5846 if (!is_bad_inode(inode
)) {
5847 inode_tree_add(inode
);
5848 unlock_new_inode(inode
);
5852 unlock_new_inode(inode
);
5855 inode
= ERR_PTR(ret
< 0 ? ret
: -ESTALE
);
5862 static struct inode
*new_simple_dir(struct super_block
*s
,
5863 struct btrfs_key
*key
,
5864 struct btrfs_root
*root
)
5866 struct inode
*inode
= new_inode(s
);
5869 return ERR_PTR(-ENOMEM
);
5871 BTRFS_I(inode
)->root
= root
;
5872 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
5873 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
5875 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
5876 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
5877 inode
->i_opflags
&= ~IOP_XATTR
;
5878 inode
->i_fop
= &simple_dir_operations
;
5879 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
5880 inode
->i_mtime
= current_time(inode
);
5881 inode
->i_atime
= inode
->i_mtime
;
5882 inode
->i_ctime
= inode
->i_mtime
;
5883 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
5888 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
5890 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
5891 struct inode
*inode
;
5892 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5893 struct btrfs_root
*sub_root
= root
;
5894 struct btrfs_key location
;
5898 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
5899 return ERR_PTR(-ENAMETOOLONG
);
5901 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
5903 return ERR_PTR(ret
);
5905 if (location
.objectid
== 0)
5906 return ERR_PTR(-ENOENT
);
5908 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
5909 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
5913 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
5914 ret
= fixup_tree_root_location(fs_info
, dir
, dentry
,
5915 &location
, &sub_root
);
5918 inode
= ERR_PTR(ret
);
5920 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
5922 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
5924 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5926 if (!IS_ERR(inode
) && root
!= sub_root
) {
5927 down_read(&fs_info
->cleanup_work_sem
);
5928 if (!sb_rdonly(inode
->i_sb
))
5929 ret
= btrfs_orphan_cleanup(sub_root
);
5930 up_read(&fs_info
->cleanup_work_sem
);
5933 inode
= ERR_PTR(ret
);
5940 static int btrfs_dentry_delete(const struct dentry
*dentry
)
5942 struct btrfs_root
*root
;
5943 struct inode
*inode
= d_inode(dentry
);
5945 if (!inode
&& !IS_ROOT(dentry
))
5946 inode
= d_inode(dentry
->d_parent
);
5949 root
= BTRFS_I(inode
)->root
;
5950 if (btrfs_root_refs(&root
->root_item
) == 0)
5953 if (btrfs_ino(BTRFS_I(inode
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5959 static void btrfs_dentry_release(struct dentry
*dentry
)
5961 kfree(dentry
->d_fsdata
);
5964 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5967 struct inode
*inode
;
5969 inode
= btrfs_lookup_dentry(dir
, dentry
);
5970 if (IS_ERR(inode
)) {
5971 if (PTR_ERR(inode
) == -ENOENT
)
5974 return ERR_CAST(inode
);
5977 return d_splice_alias(inode
, dentry
);
5980 unsigned char btrfs_filetype_table
[] = {
5981 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
5985 * All this infrastructure exists because dir_emit can fault, and we are holding
5986 * the tree lock when doing readdir. For now just allocate a buffer and copy
5987 * our information into that, and then dir_emit from the buffer. This is
5988 * similar to what NFS does, only we don't keep the buffer around in pagecache
5989 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5990 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5993 static int btrfs_opendir(struct inode
*inode
, struct file
*file
)
5995 struct btrfs_file_private
*private;
5997 private = kzalloc(sizeof(struct btrfs_file_private
), GFP_KERNEL
);
6000 private->filldir_buf
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
6001 if (!private->filldir_buf
) {
6005 file
->private_data
= private;
6016 static int btrfs_filldir(void *addr
, int entries
, struct dir_context
*ctx
)
6019 struct dir_entry
*entry
= addr
;
6020 char *name
= (char *)(entry
+ 1);
6022 ctx
->pos
= get_unaligned(&entry
->offset
);
6023 if (!dir_emit(ctx
, name
, get_unaligned(&entry
->name_len
),
6024 get_unaligned(&entry
->ino
),
6025 get_unaligned(&entry
->type
)))
6027 addr
+= sizeof(struct dir_entry
) +
6028 get_unaligned(&entry
->name_len
);
6034 static int btrfs_real_readdir(struct file
*file
, struct dir_context
*ctx
)
6036 struct inode
*inode
= file_inode(file
);
6037 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
6038 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6039 struct btrfs_file_private
*private = file
->private_data
;
6040 struct btrfs_dir_item
*di
;
6041 struct btrfs_key key
;
6042 struct btrfs_key found_key
;
6043 struct btrfs_path
*path
;
6045 struct list_head ins_list
;
6046 struct list_head del_list
;
6048 struct extent_buffer
*leaf
;
6055 struct btrfs_key location
;
6057 if (!dir_emit_dots(file
, ctx
))
6060 path
= btrfs_alloc_path();
6064 addr
= private->filldir_buf
;
6065 path
->reada
= READA_FORWARD
;
6067 INIT_LIST_HEAD(&ins_list
);
6068 INIT_LIST_HEAD(&del_list
);
6069 put
= btrfs_readdir_get_delayed_items(inode
, &ins_list
, &del_list
);
6072 key
.type
= BTRFS_DIR_INDEX_KEY
;
6073 key
.offset
= ctx
->pos
;
6074 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
6076 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6081 struct dir_entry
*entry
;
6083 leaf
= path
->nodes
[0];
6084 slot
= path
->slots
[0];
6085 if (slot
>= btrfs_header_nritems(leaf
)) {
6086 ret
= btrfs_next_leaf(root
, path
);
6094 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6096 if (found_key
.objectid
!= key
.objectid
)
6098 if (found_key
.type
!= BTRFS_DIR_INDEX_KEY
)
6100 if (found_key
.offset
< ctx
->pos
)
6102 if (btrfs_should_delete_dir_index(&del_list
, found_key
.offset
))
6104 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
6105 if (verify_dir_item(fs_info
, leaf
, slot
, di
))
6108 name_len
= btrfs_dir_name_len(leaf
, di
);
6109 if ((total_len
+ sizeof(struct dir_entry
) + name_len
) >=
6111 btrfs_release_path(path
);
6112 ret
= btrfs_filldir(private->filldir_buf
, entries
, ctx
);
6115 addr
= private->filldir_buf
;
6122 put_unaligned(name_len
, &entry
->name_len
);
6123 name_ptr
= (char *)(entry
+ 1);
6124 read_extent_buffer(leaf
, name_ptr
, (unsigned long)(di
+ 1),
6126 put_unaligned(btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)],
6128 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
6129 put_unaligned(location
.objectid
, &entry
->ino
);
6130 put_unaligned(found_key
.offset
, &entry
->offset
);
6132 addr
+= sizeof(struct dir_entry
) + name_len
;
6133 total_len
+= sizeof(struct dir_entry
) + name_len
;
6137 btrfs_release_path(path
);
6139 ret
= btrfs_filldir(private->filldir_buf
, entries
, ctx
);
6143 ret
= btrfs_readdir_delayed_dir_index(ctx
, &ins_list
);
6148 * Stop new entries from being returned after we return the last
6151 * New directory entries are assigned a strictly increasing
6152 * offset. This means that new entries created during readdir
6153 * are *guaranteed* to be seen in the future by that readdir.
6154 * This has broken buggy programs which operate on names as
6155 * they're returned by readdir. Until we re-use freed offsets
6156 * we have this hack to stop new entries from being returned
6157 * under the assumption that they'll never reach this huge
6160 * This is being careful not to overflow 32bit loff_t unless the
6161 * last entry requires it because doing so has broken 32bit apps
6164 if (ctx
->pos
>= INT_MAX
)
6165 ctx
->pos
= LLONG_MAX
;
6172 btrfs_readdir_put_delayed_items(inode
, &ins_list
, &del_list
);
6173 btrfs_free_path(path
);
6178 * This is somewhat expensive, updating the tree every time the
6179 * inode changes. But, it is most likely to find the inode in cache.
6180 * FIXME, needs more benchmarking...there are no reasons other than performance
6181 * to keep or drop this code.
6183 static int btrfs_dirty_inode(struct inode
*inode
)
6185 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
6186 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6187 struct btrfs_trans_handle
*trans
;
6190 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
6193 trans
= btrfs_join_transaction(root
);
6195 return PTR_ERR(trans
);
6197 ret
= btrfs_update_inode(trans
, root
, inode
);
6198 if (ret
&& ret
== -ENOSPC
) {
6199 /* whoops, lets try again with the full transaction */
6200 btrfs_end_transaction(trans
);
6201 trans
= btrfs_start_transaction(root
, 1);
6203 return PTR_ERR(trans
);
6205 ret
= btrfs_update_inode(trans
, root
, inode
);
6207 btrfs_end_transaction(trans
);
6208 if (BTRFS_I(inode
)->delayed_node
)
6209 btrfs_balance_delayed_items(fs_info
);
6215 * This is a copy of file_update_time. We need this so we can return error on
6216 * ENOSPC for updating the inode in the case of file write and mmap writes.
6218 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
6221 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6223 if (btrfs_root_readonly(root
))
6226 if (flags
& S_VERSION
)
6227 inode_inc_iversion(inode
);
6228 if (flags
& S_CTIME
)
6229 inode
->i_ctime
= *now
;
6230 if (flags
& S_MTIME
)
6231 inode
->i_mtime
= *now
;
6232 if (flags
& S_ATIME
)
6233 inode
->i_atime
= *now
;
6234 return btrfs_dirty_inode(inode
);
6238 * find the highest existing sequence number in a directory
6239 * and then set the in-memory index_cnt variable to reflect
6240 * free sequence numbers
6242 static int btrfs_set_inode_index_count(struct btrfs_inode
*inode
)
6244 struct btrfs_root
*root
= inode
->root
;
6245 struct btrfs_key key
, found_key
;
6246 struct btrfs_path
*path
;
6247 struct extent_buffer
*leaf
;
6250 key
.objectid
= btrfs_ino(inode
);
6251 key
.type
= BTRFS_DIR_INDEX_KEY
;
6252 key
.offset
= (u64
)-1;
6254 path
= btrfs_alloc_path();
6258 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6261 /* FIXME: we should be able to handle this */
6267 * MAGIC NUMBER EXPLANATION:
6268 * since we search a directory based on f_pos we have to start at 2
6269 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6270 * else has to start at 2
6272 if (path
->slots
[0] == 0) {
6273 inode
->index_cnt
= 2;
6279 leaf
= path
->nodes
[0];
6280 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6282 if (found_key
.objectid
!= btrfs_ino(inode
) ||
6283 found_key
.type
!= BTRFS_DIR_INDEX_KEY
) {
6284 inode
->index_cnt
= 2;
6288 inode
->index_cnt
= found_key
.offset
+ 1;
6290 btrfs_free_path(path
);
6295 * helper to find a free sequence number in a given directory. This current
6296 * code is very simple, later versions will do smarter things in the btree
6298 int btrfs_set_inode_index(struct btrfs_inode
*dir
, u64
*index
)
6302 if (dir
->index_cnt
== (u64
)-1) {
6303 ret
= btrfs_inode_delayed_dir_index_count(dir
);
6305 ret
= btrfs_set_inode_index_count(dir
);
6311 *index
= dir
->index_cnt
;
6317 static int btrfs_insert_inode_locked(struct inode
*inode
)
6319 struct btrfs_iget_args args
;
6320 args
.location
= &BTRFS_I(inode
)->location
;
6321 args
.root
= BTRFS_I(inode
)->root
;
6323 return insert_inode_locked4(inode
,
6324 btrfs_inode_hash(inode
->i_ino
, BTRFS_I(inode
)->root
),
6325 btrfs_find_actor
, &args
);
6329 * Inherit flags from the parent inode.
6331 * Currently only the compression flags and the cow flags are inherited.
6333 static void btrfs_inherit_iflags(struct inode
*inode
, struct inode
*dir
)
6340 flags
= BTRFS_I(dir
)->flags
;
6342 if (flags
& BTRFS_INODE_NOCOMPRESS
) {
6343 BTRFS_I(inode
)->flags
&= ~BTRFS_INODE_COMPRESS
;
6344 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
6345 } else if (flags
& BTRFS_INODE_COMPRESS
) {
6346 BTRFS_I(inode
)->flags
&= ~BTRFS_INODE_NOCOMPRESS
;
6347 BTRFS_I(inode
)->flags
|= BTRFS_INODE_COMPRESS
;
6350 if (flags
& BTRFS_INODE_NODATACOW
) {
6351 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
6352 if (S_ISREG(inode
->i_mode
))
6353 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6356 btrfs_update_iflags(inode
);
6359 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
6360 struct btrfs_root
*root
,
6362 const char *name
, int name_len
,
6363 u64 ref_objectid
, u64 objectid
,
6364 umode_t mode
, u64
*index
)
6366 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6367 struct inode
*inode
;
6368 struct btrfs_inode_item
*inode_item
;
6369 struct btrfs_key
*location
;
6370 struct btrfs_path
*path
;
6371 struct btrfs_inode_ref
*ref
;
6372 struct btrfs_key key
[2];
6374 int nitems
= name
? 2 : 1;
6378 path
= btrfs_alloc_path();
6380 return ERR_PTR(-ENOMEM
);
6382 inode
= new_inode(fs_info
->sb
);
6384 btrfs_free_path(path
);
6385 return ERR_PTR(-ENOMEM
);
6389 * O_TMPFILE, set link count to 0, so that after this point,
6390 * we fill in an inode item with the correct link count.
6393 set_nlink(inode
, 0);
6396 * we have to initialize this early, so we can reclaim the inode
6397 * number if we fail afterwards in this function.
6399 inode
->i_ino
= objectid
;
6402 trace_btrfs_inode_request(dir
);
6404 ret
= btrfs_set_inode_index(BTRFS_I(dir
), index
);
6406 btrfs_free_path(path
);
6408 return ERR_PTR(ret
);
6414 * index_cnt is ignored for everything but a dir,
6415 * btrfs_get_inode_index_count has an explanation for the magic
6418 BTRFS_I(inode
)->index_cnt
= 2;
6419 BTRFS_I(inode
)->dir_index
= *index
;
6420 BTRFS_I(inode
)->root
= root
;
6421 BTRFS_I(inode
)->generation
= trans
->transid
;
6422 inode
->i_generation
= BTRFS_I(inode
)->generation
;
6425 * We could have gotten an inode number from somebody who was fsynced
6426 * and then removed in this same transaction, so let's just set full
6427 * sync since it will be a full sync anyway and this will blow away the
6428 * old info in the log.
6430 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
6432 key
[0].objectid
= objectid
;
6433 key
[0].type
= BTRFS_INODE_ITEM_KEY
;
6436 sizes
[0] = sizeof(struct btrfs_inode_item
);
6440 * Start new inodes with an inode_ref. This is slightly more
6441 * efficient for small numbers of hard links since they will
6442 * be packed into one item. Extended refs will kick in if we
6443 * add more hard links than can fit in the ref item.
6445 key
[1].objectid
= objectid
;
6446 key
[1].type
= BTRFS_INODE_REF_KEY
;
6447 key
[1].offset
= ref_objectid
;
6449 sizes
[1] = name_len
+ sizeof(*ref
);
6452 location
= &BTRFS_I(inode
)->location
;
6453 location
->objectid
= objectid
;
6454 location
->offset
= 0;
6455 location
->type
= BTRFS_INODE_ITEM_KEY
;
6457 ret
= btrfs_insert_inode_locked(inode
);
6461 path
->leave_spinning
= 1;
6462 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, nitems
);
6466 inode_init_owner(inode
, dir
, mode
);
6467 inode_set_bytes(inode
, 0);
6469 inode
->i_mtime
= current_time(inode
);
6470 inode
->i_atime
= inode
->i_mtime
;
6471 inode
->i_ctime
= inode
->i_mtime
;
6472 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
6474 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
6475 struct btrfs_inode_item
);
6476 memzero_extent_buffer(path
->nodes
[0], (unsigned long)inode_item
,
6477 sizeof(*inode_item
));
6478 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
6481 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
6482 struct btrfs_inode_ref
);
6483 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
6484 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
6485 ptr
= (unsigned long)(ref
+ 1);
6486 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
6489 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6490 btrfs_free_path(path
);
6492 btrfs_inherit_iflags(inode
, dir
);
6494 if (S_ISREG(mode
)) {
6495 if (btrfs_test_opt(fs_info
, NODATASUM
))
6496 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6497 if (btrfs_test_opt(fs_info
, NODATACOW
))
6498 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
6499 BTRFS_INODE_NODATASUM
;
6502 inode_tree_add(inode
);
6504 trace_btrfs_inode_new(inode
);
6505 btrfs_set_inode_last_trans(trans
, inode
);
6507 btrfs_update_root_times(trans
, root
);
6509 ret
= btrfs_inode_inherit_props(trans
, inode
, dir
);
6512 "error inheriting props for ino %llu (root %llu): %d",
6513 btrfs_ino(BTRFS_I(inode
)), root
->root_key
.objectid
, ret
);
6518 unlock_new_inode(inode
);
6521 BTRFS_I(dir
)->index_cnt
--;
6522 btrfs_free_path(path
);
6524 return ERR_PTR(ret
);
6527 static inline u8
btrfs_inode_type(struct inode
*inode
)
6529 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
6533 * utility function to add 'inode' into 'parent_inode' with
6534 * a give name and a given sequence number.
6535 * if 'add_backref' is true, also insert a backref from the
6536 * inode to the parent directory.
6538 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
6539 struct btrfs_inode
*parent_inode
, struct btrfs_inode
*inode
,
6540 const char *name
, int name_len
, int add_backref
, u64 index
)
6542 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6544 struct btrfs_key key
;
6545 struct btrfs_root
*root
= parent_inode
->root
;
6546 u64 ino
= btrfs_ino(inode
);
6547 u64 parent_ino
= btrfs_ino(parent_inode
);
6549 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6550 memcpy(&key
, &inode
->root
->root_key
, sizeof(key
));
6553 key
.type
= BTRFS_INODE_ITEM_KEY
;
6557 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6558 ret
= btrfs_add_root_ref(trans
, fs_info
, key
.objectid
,
6559 root
->root_key
.objectid
, parent_ino
,
6560 index
, name
, name_len
);
6561 } else if (add_backref
) {
6562 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
6566 /* Nothing to clean up yet */
6570 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
6572 btrfs_inode_type(&inode
->vfs_inode
), index
);
6573 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
6576 btrfs_abort_transaction(trans
, ret
);
6580 btrfs_i_size_write(parent_inode
, parent_inode
->vfs_inode
.i_size
+
6582 inode_inc_iversion(&parent_inode
->vfs_inode
);
6583 parent_inode
->vfs_inode
.i_mtime
= parent_inode
->vfs_inode
.i_ctime
=
6584 current_time(&parent_inode
->vfs_inode
);
6585 ret
= btrfs_update_inode(trans
, root
, &parent_inode
->vfs_inode
);
6587 btrfs_abort_transaction(trans
, ret
);
6591 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6594 err
= btrfs_del_root_ref(trans
, fs_info
, key
.objectid
,
6595 root
->root_key
.objectid
, parent_ino
,
6596 &local_index
, name
, name_len
);
6598 btrfs_abort_transaction(trans
, err
);
6599 } else if (add_backref
) {
6603 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
6604 ino
, parent_ino
, &local_index
);
6606 btrfs_abort_transaction(trans
, err
);
6609 /* Return the original error code */
6613 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
6614 struct btrfs_inode
*dir
, struct dentry
*dentry
,
6615 struct btrfs_inode
*inode
, int backref
, u64 index
)
6617 int err
= btrfs_add_link(trans
, dir
, inode
,
6618 dentry
->d_name
.name
, dentry
->d_name
.len
,
6625 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
6626 umode_t mode
, dev_t rdev
)
6628 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6629 struct btrfs_trans_handle
*trans
;
6630 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6631 struct inode
*inode
= NULL
;
6638 * 2 for inode item and ref
6640 * 1 for xattr if selinux is on
6642 trans
= btrfs_start_transaction(root
, 5);
6644 return PTR_ERR(trans
);
6646 err
= btrfs_find_free_ino(root
, &objectid
);
6650 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6651 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)), objectid
,
6653 if (IS_ERR(inode
)) {
6654 err
= PTR_ERR(inode
);
6659 * If the active LSM wants to access the inode during
6660 * d_instantiate it needs these. Smack checks to see
6661 * if the filesystem supports xattrs by looking at the
6664 inode
->i_op
= &btrfs_special_inode_operations
;
6665 init_special_inode(inode
, inode
->i_mode
, rdev
);
6667 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6669 goto out_unlock_inode
;
6671 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
, BTRFS_I(inode
),
6674 goto out_unlock_inode
;
6676 btrfs_update_inode(trans
, root
, inode
);
6677 d_instantiate_new(dentry
, inode
);
6681 btrfs_end_transaction(trans
);
6682 btrfs_balance_delayed_items(fs_info
);
6683 btrfs_btree_balance_dirty(fs_info
);
6685 inode_dec_link_count(inode
);
6692 unlock_new_inode(inode
);
6697 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
6698 umode_t mode
, bool excl
)
6700 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6701 struct btrfs_trans_handle
*trans
;
6702 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6703 struct inode
*inode
= NULL
;
6704 int drop_inode_on_err
= 0;
6710 * 2 for inode item and ref
6712 * 1 for xattr if selinux is on
6714 trans
= btrfs_start_transaction(root
, 5);
6716 return PTR_ERR(trans
);
6718 err
= btrfs_find_free_ino(root
, &objectid
);
6722 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6723 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)), objectid
,
6725 if (IS_ERR(inode
)) {
6726 err
= PTR_ERR(inode
);
6729 drop_inode_on_err
= 1;
6731 * If the active LSM wants to access the inode during
6732 * d_instantiate it needs these. Smack checks to see
6733 * if the filesystem supports xattrs by looking at the
6736 inode
->i_fop
= &btrfs_file_operations
;
6737 inode
->i_op
= &btrfs_file_inode_operations
;
6738 inode
->i_mapping
->a_ops
= &btrfs_aops
;
6740 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6742 goto out_unlock_inode
;
6744 err
= btrfs_update_inode(trans
, root
, inode
);
6746 goto out_unlock_inode
;
6748 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
, BTRFS_I(inode
),
6751 goto out_unlock_inode
;
6753 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
6754 d_instantiate_new(dentry
, inode
);
6757 btrfs_end_transaction(trans
);
6758 if (err
&& drop_inode_on_err
) {
6759 inode_dec_link_count(inode
);
6762 btrfs_balance_delayed_items(fs_info
);
6763 btrfs_btree_balance_dirty(fs_info
);
6767 unlock_new_inode(inode
);
6772 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
6773 struct dentry
*dentry
)
6775 struct btrfs_trans_handle
*trans
= NULL
;
6776 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6777 struct inode
*inode
= d_inode(old_dentry
);
6778 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
6783 /* do not allow sys_link's with other subvols of the same device */
6784 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
6787 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
6790 err
= btrfs_set_inode_index(BTRFS_I(dir
), &index
);
6795 * 2 items for inode and inode ref
6796 * 2 items for dir items
6797 * 1 item for parent inode
6799 trans
= btrfs_start_transaction(root
, 5);
6800 if (IS_ERR(trans
)) {
6801 err
= PTR_ERR(trans
);
6806 /* There are several dir indexes for this inode, clear the cache. */
6807 BTRFS_I(inode
)->dir_index
= 0ULL;
6809 inode_inc_iversion(inode
);
6810 inode
->i_ctime
= current_time(inode
);
6812 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
6814 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
, BTRFS_I(inode
),
6820 struct dentry
*parent
= dentry
->d_parent
;
6821 err
= btrfs_update_inode(trans
, root
, inode
);
6824 if (inode
->i_nlink
== 1) {
6826 * If new hard link count is 1, it's a file created
6827 * with open(2) O_TMPFILE flag.
6829 err
= btrfs_orphan_del(trans
, BTRFS_I(inode
));
6833 BTRFS_I(inode
)->last_link_trans
= trans
->transid
;
6834 d_instantiate(dentry
, inode
);
6835 btrfs_log_new_name(trans
, BTRFS_I(inode
), NULL
, parent
);
6838 btrfs_balance_delayed_items(fs_info
);
6841 btrfs_end_transaction(trans
);
6843 inode_dec_link_count(inode
);
6846 btrfs_btree_balance_dirty(fs_info
);
6850 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
6852 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6853 struct inode
*inode
= NULL
;
6854 struct btrfs_trans_handle
*trans
;
6855 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6857 int drop_on_err
= 0;
6862 * 2 items for inode and ref
6863 * 2 items for dir items
6864 * 1 for xattr if selinux is on
6866 trans
= btrfs_start_transaction(root
, 5);
6868 return PTR_ERR(trans
);
6870 err
= btrfs_find_free_ino(root
, &objectid
);
6874 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6875 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)), objectid
,
6876 S_IFDIR
| mode
, &index
);
6877 if (IS_ERR(inode
)) {
6878 err
= PTR_ERR(inode
);
6883 /* these must be set before we unlock the inode */
6884 inode
->i_op
= &btrfs_dir_inode_operations
;
6885 inode
->i_fop
= &btrfs_dir_file_operations
;
6887 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6889 goto out_fail_inode
;
6891 btrfs_i_size_write(BTRFS_I(inode
), 0);
6892 err
= btrfs_update_inode(trans
, root
, inode
);
6894 goto out_fail_inode
;
6896 err
= btrfs_add_link(trans
, BTRFS_I(dir
), BTRFS_I(inode
),
6897 dentry
->d_name
.name
,
6898 dentry
->d_name
.len
, 0, index
);
6900 goto out_fail_inode
;
6902 d_instantiate_new(dentry
, inode
);
6906 btrfs_end_transaction(trans
);
6908 inode_dec_link_count(inode
);
6911 btrfs_balance_delayed_items(fs_info
);
6912 btrfs_btree_balance_dirty(fs_info
);
6916 unlock_new_inode(inode
);
6920 /* Find next extent map of a given extent map, caller needs to ensure locks */
6921 static struct extent_map
*next_extent_map(struct extent_map
*em
)
6923 struct rb_node
*next
;
6925 next
= rb_next(&em
->rb_node
);
6928 return container_of(next
, struct extent_map
, rb_node
);
6931 static struct extent_map
*prev_extent_map(struct extent_map
*em
)
6933 struct rb_node
*prev
;
6935 prev
= rb_prev(&em
->rb_node
);
6938 return container_of(prev
, struct extent_map
, rb_node
);
6941 /* helper for btfs_get_extent. Given an existing extent in the tree,
6942 * the existing extent is the nearest extent to map_start,
6943 * and an extent that you want to insert, deal with overlap and insert
6944 * the best fitted new extent into the tree.
6946 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
6947 struct extent_map
*existing
,
6948 struct extent_map
*em
,
6951 struct extent_map
*prev
;
6952 struct extent_map
*next
;
6957 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
6959 if (existing
->start
> map_start
) {
6961 prev
= prev_extent_map(next
);
6964 next
= next_extent_map(prev
);
6967 start
= prev
? extent_map_end(prev
) : em
->start
;
6968 start
= max_t(u64
, start
, em
->start
);
6969 end
= next
? next
->start
: extent_map_end(em
);
6970 end
= min_t(u64
, end
, extent_map_end(em
));
6971 start_diff
= start
- em
->start
;
6973 em
->len
= end
- start
;
6974 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
6975 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
6976 em
->block_start
+= start_diff
;
6977 em
->block_len
-= start_diff
;
6979 return add_extent_mapping(em_tree
, em
, 0);
6982 static noinline
int uncompress_inline(struct btrfs_path
*path
,
6984 size_t pg_offset
, u64 extent_offset
,
6985 struct btrfs_file_extent_item
*item
)
6988 struct extent_buffer
*leaf
= path
->nodes
[0];
6991 unsigned long inline_size
;
6995 WARN_ON(pg_offset
!= 0);
6996 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6997 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6998 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
6999 btrfs_item_nr(path
->slots
[0]));
7000 tmp
= kmalloc(inline_size
, GFP_NOFS
);
7003 ptr
= btrfs_file_extent_inline_start(item
);
7005 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
7007 max_size
= min_t(unsigned long, PAGE_SIZE
, max_size
);
7008 ret
= btrfs_decompress(compress_type
, tmp
, page
,
7009 extent_offset
, inline_size
, max_size
);
7012 * decompression code contains a memset to fill in any space between the end
7013 * of the uncompressed data and the end of max_size in case the decompressed
7014 * data ends up shorter than ram_bytes. That doesn't cover the hole between
7015 * the end of an inline extent and the beginning of the next block, so we
7016 * cover that region here.
7019 if (max_size
+ pg_offset
< PAGE_SIZE
) {
7020 char *map
= kmap(page
);
7021 memset(map
+ pg_offset
+ max_size
, 0, PAGE_SIZE
- max_size
- pg_offset
);
7029 * a bit scary, this does extent mapping from logical file offset to the disk.
7030 * the ugly parts come from merging extents from the disk with the in-ram
7031 * representation. This gets more complex because of the data=ordered code,
7032 * where the in-ram extents might be locked pending data=ordered completion.
7034 * This also copies inline extents directly into the page.
7036 struct extent_map
*btrfs_get_extent(struct btrfs_inode
*inode
,
7038 size_t pg_offset
, u64 start
, u64 len
,
7041 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
7044 u64 extent_start
= 0;
7046 u64 objectid
= btrfs_ino(inode
);
7048 struct btrfs_path
*path
= NULL
;
7049 struct btrfs_root
*root
= inode
->root
;
7050 struct btrfs_file_extent_item
*item
;
7051 struct extent_buffer
*leaf
;
7052 struct btrfs_key found_key
;
7053 struct extent_map
*em
= NULL
;
7054 struct extent_map_tree
*em_tree
= &inode
->extent_tree
;
7055 struct extent_io_tree
*io_tree
= &inode
->io_tree
;
7056 struct btrfs_trans_handle
*trans
= NULL
;
7057 const bool new_inline
= !page
|| create
;
7060 read_lock(&em_tree
->lock
);
7061 em
= lookup_extent_mapping(em_tree
, start
, len
);
7063 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
7064 read_unlock(&em_tree
->lock
);
7067 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
7068 free_extent_map(em
);
7069 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
7070 free_extent_map(em
);
7074 em
= alloc_extent_map();
7079 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
7080 em
->start
= EXTENT_MAP_HOLE
;
7081 em
->orig_start
= EXTENT_MAP_HOLE
;
7083 em
->block_len
= (u64
)-1;
7086 path
= btrfs_alloc_path();
7092 * Chances are we'll be called again, so go ahead and do
7095 path
->reada
= READA_FORWARD
;
7098 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
7099 objectid
, start
, trans
!= NULL
);
7106 if (path
->slots
[0] == 0)
7111 leaf
= path
->nodes
[0];
7112 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7113 struct btrfs_file_extent_item
);
7114 /* are we inside the extent that was found? */
7115 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7116 found_type
= found_key
.type
;
7117 if (found_key
.objectid
!= objectid
||
7118 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
7120 * If we backup past the first extent we want to move forward
7121 * and see if there is an extent in front of us, otherwise we'll
7122 * say there is a hole for our whole search range which can
7129 found_type
= btrfs_file_extent_type(leaf
, item
);
7130 extent_start
= found_key
.offset
;
7131 if (found_type
== BTRFS_FILE_EXTENT_REG
||
7132 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7133 extent_end
= extent_start
+
7134 btrfs_file_extent_num_bytes(leaf
, item
);
7136 trace_btrfs_get_extent_show_fi_regular(inode
, leaf
, item
,
7138 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
7140 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
7141 extent_end
= ALIGN(extent_start
+ size
,
7142 fs_info
->sectorsize
);
7144 trace_btrfs_get_extent_show_fi_inline(inode
, leaf
, item
,
7149 if (start
>= extent_end
) {
7151 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
7152 ret
= btrfs_next_leaf(root
, path
);
7159 leaf
= path
->nodes
[0];
7161 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7162 if (found_key
.objectid
!= objectid
||
7163 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7165 if (start
+ len
<= found_key
.offset
)
7167 if (start
> found_key
.offset
)
7170 em
->orig_start
= start
;
7171 em
->len
= found_key
.offset
- start
;
7175 btrfs_extent_item_to_extent_map(inode
, path
, item
,
7178 if (found_type
== BTRFS_FILE_EXTENT_REG
||
7179 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7181 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
7185 size_t extent_offset
;
7191 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
7192 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
7193 copy_size
= min_t(u64
, PAGE_SIZE
- pg_offset
,
7194 size
- extent_offset
);
7195 em
->start
= extent_start
+ extent_offset
;
7196 em
->len
= ALIGN(copy_size
, fs_info
->sectorsize
);
7197 em
->orig_block_len
= em
->len
;
7198 em
->orig_start
= em
->start
;
7199 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
7200 if (create
== 0 && !PageUptodate(page
)) {
7201 if (btrfs_file_extent_compression(leaf
, item
) !=
7202 BTRFS_COMPRESS_NONE
) {
7203 ret
= uncompress_inline(path
, page
, pg_offset
,
7204 extent_offset
, item
);
7211 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
7213 if (pg_offset
+ copy_size
< PAGE_SIZE
) {
7214 memset(map
+ pg_offset
+ copy_size
, 0,
7215 PAGE_SIZE
- pg_offset
-
7220 flush_dcache_page(page
);
7221 } else if (create
&& PageUptodate(page
)) {
7225 free_extent_map(em
);
7228 btrfs_release_path(path
);
7229 trans
= btrfs_join_transaction(root
);
7232 return ERR_CAST(trans
);
7236 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
7239 btrfs_mark_buffer_dirty(leaf
);
7241 set_extent_uptodate(io_tree
, em
->start
,
7242 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
7247 em
->orig_start
= start
;
7250 em
->block_start
= EXTENT_MAP_HOLE
;
7251 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
7253 btrfs_release_path(path
);
7254 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
7256 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7257 em
->start
, em
->len
, start
, len
);
7263 write_lock(&em_tree
->lock
);
7264 ret
= add_extent_mapping(em_tree
, em
, 0);
7265 /* it is possible that someone inserted the extent into the tree
7266 * while we had the lock dropped. It is also possible that
7267 * an overlapping map exists in the tree
7269 if (ret
== -EEXIST
) {
7270 struct extent_map
*existing
;
7274 existing
= search_extent_mapping(em_tree
, start
, len
);
7276 * existing will always be non-NULL, since there must be
7277 * extent causing the -EEXIST.
7279 if (start
>= existing
->start
&&
7280 start
< extent_map_end(existing
)) {
7281 free_extent_map(em
);
7286 * The existing extent map is the one nearest to
7287 * the [start, start + len) range which overlaps
7289 err
= merge_extent_mapping(em_tree
, existing
,
7291 free_extent_map(existing
);
7293 free_extent_map(em
);
7298 write_unlock(&em_tree
->lock
);
7301 trace_btrfs_get_extent(root
, inode
, em
);
7303 btrfs_free_path(path
);
7305 ret
= btrfs_end_transaction(trans
);
7310 free_extent_map(em
);
7311 return ERR_PTR(err
);
7313 BUG_ON(!em
); /* Error is always set */
7317 struct extent_map
*btrfs_get_extent_fiemap(struct btrfs_inode
*inode
,
7319 size_t pg_offset
, u64 start
, u64 len
,
7322 struct extent_map
*em
;
7323 struct extent_map
*hole_em
= NULL
;
7324 u64 range_start
= start
;
7330 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
7334 * If our em maps to:
7336 * - a pre-alloc extent,
7337 * there might actually be delalloc bytes behind it.
7339 if (em
->block_start
!= EXTENT_MAP_HOLE
&&
7340 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7345 /* check to see if we've wrapped (len == -1 or similar) */
7354 /* ok, we didn't find anything, lets look for delalloc */
7355 found
= count_range_bits(&inode
->io_tree
, &range_start
,
7356 end
, len
, EXTENT_DELALLOC
, 1);
7357 found_end
= range_start
+ found
;
7358 if (found_end
< range_start
)
7359 found_end
= (u64
)-1;
7362 * we didn't find anything useful, return
7363 * the original results from get_extent()
7365 if (range_start
> end
|| found_end
<= start
) {
7371 /* adjust the range_start to make sure it doesn't
7372 * go backwards from the start they passed in
7374 range_start
= max(start
, range_start
);
7375 found
= found_end
- range_start
;
7378 u64 hole_start
= start
;
7381 em
= alloc_extent_map();
7387 * when btrfs_get_extent can't find anything it
7388 * returns one huge hole
7390 * make sure what it found really fits our range, and
7391 * adjust to make sure it is based on the start from
7395 u64 calc_end
= extent_map_end(hole_em
);
7397 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
7398 free_extent_map(hole_em
);
7401 hole_start
= max(hole_em
->start
, start
);
7402 hole_len
= calc_end
- hole_start
;
7406 if (hole_em
&& range_start
> hole_start
) {
7407 /* our hole starts before our delalloc, so we
7408 * have to return just the parts of the hole
7409 * that go until the delalloc starts
7411 em
->len
= min(hole_len
,
7412 range_start
- hole_start
);
7413 em
->start
= hole_start
;
7414 em
->orig_start
= hole_start
;
7416 * don't adjust block start at all,
7417 * it is fixed at EXTENT_MAP_HOLE
7419 em
->block_start
= hole_em
->block_start
;
7420 em
->block_len
= hole_len
;
7421 if (test_bit(EXTENT_FLAG_PREALLOC
, &hole_em
->flags
))
7422 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
7424 em
->start
= range_start
;
7426 em
->orig_start
= range_start
;
7427 em
->block_start
= EXTENT_MAP_DELALLOC
;
7428 em
->block_len
= found
;
7430 } else if (hole_em
) {
7435 free_extent_map(hole_em
);
7437 free_extent_map(em
);
7438 return ERR_PTR(err
);
7443 static struct extent_map
*btrfs_create_dio_extent(struct inode
*inode
,
7446 const u64 orig_start
,
7447 const u64 block_start
,
7448 const u64 block_len
,
7449 const u64 orig_block_len
,
7450 const u64 ram_bytes
,
7453 struct extent_map
*em
= NULL
;
7456 if (type
!= BTRFS_ORDERED_NOCOW
) {
7457 em
= create_io_em(inode
, start
, len
, orig_start
,
7458 block_start
, block_len
, orig_block_len
,
7460 BTRFS_COMPRESS_NONE
, /* compress_type */
7465 ret
= btrfs_add_ordered_extent_dio(inode
, start
, block_start
,
7466 len
, block_len
, type
);
7469 free_extent_map(em
);
7470 btrfs_drop_extent_cache(BTRFS_I(inode
), start
,
7471 start
+ len
- 1, 0);
7480 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
7483 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7484 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7485 struct extent_map
*em
;
7486 struct btrfs_key ins
;
7490 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
7491 ret
= btrfs_reserve_extent(root
, len
, len
, fs_info
->sectorsize
,
7492 0, alloc_hint
, &ins
, 1, 1);
7494 return ERR_PTR(ret
);
7496 em
= btrfs_create_dio_extent(inode
, start
, ins
.offset
, start
,
7497 ins
.objectid
, ins
.offset
, ins
.offset
,
7498 ins
.offset
, BTRFS_ORDERED_REGULAR
);
7499 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
7501 btrfs_free_reserved_extent(fs_info
, ins
.objectid
,
7508 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7509 * block must be cow'd
7511 noinline
int can_nocow_extent(struct inode
*inode
, u64 offset
, u64
*len
,
7512 u64
*orig_start
, u64
*orig_block_len
,
7515 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7516 struct btrfs_path
*path
;
7518 struct extent_buffer
*leaf
;
7519 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7520 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7521 struct btrfs_file_extent_item
*fi
;
7522 struct btrfs_key key
;
7529 bool nocow
= (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
);
7531 path
= btrfs_alloc_path();
7535 ret
= btrfs_lookup_file_extent(NULL
, root
, path
,
7536 btrfs_ino(BTRFS_I(inode
)), offset
, 0);
7540 slot
= path
->slots
[0];
7543 /* can't find the item, must cow */
7550 leaf
= path
->nodes
[0];
7551 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
7552 if (key
.objectid
!= btrfs_ino(BTRFS_I(inode
)) ||
7553 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7554 /* not our file or wrong item type, must cow */
7558 if (key
.offset
> offset
) {
7559 /* Wrong offset, must cow */
7563 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
7564 found_type
= btrfs_file_extent_type(leaf
, fi
);
7565 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
7566 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
7567 /* not a regular extent, must cow */
7571 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_REG
)
7574 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
7575 if (extent_end
<= offset
)
7578 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7579 if (disk_bytenr
== 0)
7582 if (btrfs_file_extent_compression(leaf
, fi
) ||
7583 btrfs_file_extent_encryption(leaf
, fi
) ||
7584 btrfs_file_extent_other_encoding(leaf
, fi
))
7587 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
7590 *orig_start
= key
.offset
- backref_offset
;
7591 *orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7592 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
7595 if (btrfs_extent_readonly(fs_info
, disk_bytenr
))
7598 num_bytes
= min(offset
+ *len
, extent_end
) - offset
;
7599 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7602 range_end
= round_up(offset
+ num_bytes
,
7603 root
->fs_info
->sectorsize
) - 1;
7604 ret
= test_range_bit(io_tree
, offset
, range_end
,
7605 EXTENT_DELALLOC
, 0, NULL
);
7612 btrfs_release_path(path
);
7615 * look for other files referencing this extent, if we
7616 * find any we must cow
7619 ret
= btrfs_cross_ref_exist(root
, btrfs_ino(BTRFS_I(inode
)),
7620 key
.offset
- backref_offset
, disk_bytenr
);
7627 * adjust disk_bytenr and num_bytes to cover just the bytes
7628 * in this extent we are about to write. If there
7629 * are any csums in that range we have to cow in order
7630 * to keep the csums correct
7632 disk_bytenr
+= backref_offset
;
7633 disk_bytenr
+= offset
- key
.offset
;
7634 if (csum_exist_in_range(fs_info
, disk_bytenr
, num_bytes
))
7637 * all of the above have passed, it is safe to overwrite this extent
7643 btrfs_free_path(path
);
7647 bool btrfs_page_exists_in_range(struct inode
*inode
, loff_t start
, loff_t end
)
7649 struct radix_tree_root
*root
= &inode
->i_mapping
->page_tree
;
7651 void **pagep
= NULL
;
7652 struct page
*page
= NULL
;
7653 unsigned long start_idx
;
7654 unsigned long end_idx
;
7656 start_idx
= start
>> PAGE_SHIFT
;
7659 * end is the last byte in the last page. end == start is legal
7661 end_idx
= end
>> PAGE_SHIFT
;
7665 /* Most of the code in this while loop is lifted from
7666 * find_get_page. It's been modified to begin searching from a
7667 * page and return just the first page found in that range. If the
7668 * found idx is less than or equal to the end idx then we know that
7669 * a page exists. If no pages are found or if those pages are
7670 * outside of the range then we're fine (yay!) */
7671 while (page
== NULL
&&
7672 radix_tree_gang_lookup_slot(root
, &pagep
, NULL
, start_idx
, 1)) {
7673 page
= radix_tree_deref_slot(pagep
);
7674 if (unlikely(!page
))
7677 if (radix_tree_exception(page
)) {
7678 if (radix_tree_deref_retry(page
)) {
7683 * Otherwise, shmem/tmpfs must be storing a swap entry
7684 * here as an exceptional entry: so return it without
7685 * attempting to raise page count.
7688 break; /* TODO: Is this relevant for this use case? */
7691 if (!page_cache_get_speculative(page
)) {
7697 * Has the page moved?
7698 * This is part of the lockless pagecache protocol. See
7699 * include/linux/pagemap.h for details.
7701 if (unlikely(page
!= *pagep
)) {
7708 if (page
->index
<= end_idx
)
7717 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
7718 struct extent_state
**cached_state
, int writing
)
7720 struct btrfs_ordered_extent
*ordered
;
7724 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7727 * We're concerned with the entire range that we're going to be
7728 * doing DIO to, so we need to make sure there's no ordered
7729 * extents in this range.
7731 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), lockstart
,
7732 lockend
- lockstart
+ 1);
7735 * We need to make sure there are no buffered pages in this
7736 * range either, we could have raced between the invalidate in
7737 * generic_file_direct_write and locking the extent. The
7738 * invalidate needs to happen so that reads after a write do not
7743 !btrfs_page_exists_in_range(inode
, lockstart
, lockend
)))
7746 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7747 cached_state
, GFP_NOFS
);
7751 * If we are doing a DIO read and the ordered extent we
7752 * found is for a buffered write, we can not wait for it
7753 * to complete and retry, because if we do so we can
7754 * deadlock with concurrent buffered writes on page
7755 * locks. This happens only if our DIO read covers more
7756 * than one extent map, if at this point has already
7757 * created an ordered extent for a previous extent map
7758 * and locked its range in the inode's io tree, and a
7759 * concurrent write against that previous extent map's
7760 * range and this range started (we unlock the ranges
7761 * in the io tree only when the bios complete and
7762 * buffered writes always lock pages before attempting
7763 * to lock range in the io tree).
7766 test_bit(BTRFS_ORDERED_DIRECT
, &ordered
->flags
))
7767 btrfs_start_ordered_extent(inode
, ordered
, 1);
7770 btrfs_put_ordered_extent(ordered
);
7773 * We could trigger writeback for this range (and wait
7774 * for it to complete) and then invalidate the pages for
7775 * this range (through invalidate_inode_pages2_range()),
7776 * but that can lead us to a deadlock with a concurrent
7777 * call to readpages() (a buffered read or a defrag call
7778 * triggered a readahead) on a page lock due to an
7779 * ordered dio extent we created before but did not have
7780 * yet a corresponding bio submitted (whence it can not
7781 * complete), which makes readpages() wait for that
7782 * ordered extent to complete while holding a lock on
7797 /* The callers of this must take lock_extent() */
7798 static struct extent_map
*create_io_em(struct inode
*inode
, u64 start
, u64 len
,
7799 u64 orig_start
, u64 block_start
,
7800 u64 block_len
, u64 orig_block_len
,
7801 u64 ram_bytes
, int compress_type
,
7804 struct extent_map_tree
*em_tree
;
7805 struct extent_map
*em
;
7806 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7809 ASSERT(type
== BTRFS_ORDERED_PREALLOC
||
7810 type
== BTRFS_ORDERED_COMPRESSED
||
7811 type
== BTRFS_ORDERED_NOCOW
||
7812 type
== BTRFS_ORDERED_REGULAR
);
7814 em_tree
= &BTRFS_I(inode
)->extent_tree
;
7815 em
= alloc_extent_map();
7817 return ERR_PTR(-ENOMEM
);
7820 em
->orig_start
= orig_start
;
7822 em
->block_len
= block_len
;
7823 em
->block_start
= block_start
;
7824 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
7825 em
->orig_block_len
= orig_block_len
;
7826 em
->ram_bytes
= ram_bytes
;
7827 em
->generation
= -1;
7828 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
7829 if (type
== BTRFS_ORDERED_PREALLOC
) {
7830 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
7831 } else if (type
== BTRFS_ORDERED_COMPRESSED
) {
7832 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
7833 em
->compress_type
= compress_type
;
7837 btrfs_drop_extent_cache(BTRFS_I(inode
), em
->start
,
7838 em
->start
+ em
->len
- 1, 0);
7839 write_lock(&em_tree
->lock
);
7840 ret
= add_extent_mapping(em_tree
, em
, 1);
7841 write_unlock(&em_tree
->lock
);
7843 * The caller has taken lock_extent(), who could race with us
7846 } while (ret
== -EEXIST
);
7849 free_extent_map(em
);
7850 return ERR_PTR(ret
);
7853 /* em got 2 refs now, callers needs to do free_extent_map once. */
7857 static void adjust_dio_outstanding_extents(struct inode
*inode
,
7858 struct btrfs_dio_data
*dio_data
,
7861 unsigned num_extents
= count_max_extents(len
);
7864 * If we have an outstanding_extents count still set then we're
7865 * within our reservation, otherwise we need to adjust our inode
7866 * counter appropriately.
7868 if (dio_data
->outstanding_extents
>= num_extents
) {
7869 dio_data
->outstanding_extents
-= num_extents
;
7872 * If dio write length has been split due to no large enough
7873 * contiguous space, we need to compensate our inode counter
7876 u64 num_needed
= num_extents
- dio_data
->outstanding_extents
;
7878 spin_lock(&BTRFS_I(inode
)->lock
);
7879 BTRFS_I(inode
)->outstanding_extents
+= num_needed
;
7880 spin_unlock(&BTRFS_I(inode
)->lock
);
7884 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
7885 struct buffer_head
*bh_result
, int create
)
7887 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7888 struct extent_map
*em
;
7889 struct extent_state
*cached_state
= NULL
;
7890 struct btrfs_dio_data
*dio_data
= NULL
;
7891 u64 start
= iblock
<< inode
->i_blkbits
;
7892 u64 lockstart
, lockend
;
7893 u64 len
= bh_result
->b_size
;
7894 int unlock_bits
= EXTENT_LOCKED
;
7898 unlock_bits
|= EXTENT_DIRTY
;
7900 len
= min_t(u64
, len
, fs_info
->sectorsize
);
7903 lockend
= start
+ len
- 1;
7905 if (current
->journal_info
) {
7907 * Need to pull our outstanding extents and set journal_info to NULL so
7908 * that anything that needs to check if there's a transaction doesn't get
7911 dio_data
= current
->journal_info
;
7912 current
->journal_info
= NULL
;
7916 * If this errors out it's because we couldn't invalidate pagecache for
7917 * this range and we need to fallback to buffered.
7919 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
,
7925 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0, start
, len
, 0);
7932 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7933 * io. INLINE is special, and we could probably kludge it in here, but
7934 * it's still buffered so for safety lets just fall back to the generic
7937 * For COMPRESSED we _have_ to read the entire extent in so we can
7938 * decompress it, so there will be buffering required no matter what we
7939 * do, so go ahead and fallback to buffered.
7941 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7942 * to buffered IO. Don't blame me, this is the price we pay for using
7945 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
7946 em
->block_start
== EXTENT_MAP_INLINE
) {
7947 free_extent_map(em
);
7952 /* Just a good old fashioned hole, return */
7953 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
7954 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
7955 free_extent_map(em
);
7960 * We don't allocate a new extent in the following cases
7962 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7964 * 2) The extent is marked as PREALLOC. We're good to go here and can
7965 * just use the extent.
7969 len
= min(len
, em
->len
- (start
- em
->start
));
7970 lockstart
= start
+ len
;
7974 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
7975 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
7976 em
->block_start
!= EXTENT_MAP_HOLE
)) {
7978 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
7980 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7981 type
= BTRFS_ORDERED_PREALLOC
;
7983 type
= BTRFS_ORDERED_NOCOW
;
7984 len
= min(len
, em
->len
- (start
- em
->start
));
7985 block_start
= em
->block_start
+ (start
- em
->start
);
7987 if (can_nocow_extent(inode
, start
, &len
, &orig_start
,
7988 &orig_block_len
, &ram_bytes
) == 1 &&
7989 btrfs_inc_nocow_writers(fs_info
, block_start
)) {
7990 struct extent_map
*em2
;
7992 em2
= btrfs_create_dio_extent(inode
, start
, len
,
7993 orig_start
, block_start
,
7994 len
, orig_block_len
,
7996 btrfs_dec_nocow_writers(fs_info
, block_start
);
7997 if (type
== BTRFS_ORDERED_PREALLOC
) {
7998 free_extent_map(em
);
8001 if (em2
&& IS_ERR(em2
)) {
8006 * For inode marked NODATACOW or extent marked PREALLOC,
8007 * use the existing or preallocated extent, so does not
8008 * need to adjust btrfs_space_info's bytes_may_use.
8010 btrfs_free_reserved_data_space_noquota(inode
,
8017 * this will cow the extent, reset the len in case we changed
8020 len
= bh_result
->b_size
;
8021 free_extent_map(em
);
8022 em
= btrfs_new_extent_direct(inode
, start
, len
);
8027 len
= min(len
, em
->len
- (start
- em
->start
));
8029 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
8031 bh_result
->b_size
= len
;
8032 bh_result
->b_bdev
= em
->bdev
;
8033 set_buffer_mapped(bh_result
);
8035 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
8036 set_buffer_new(bh_result
);
8039 * Need to update the i_size under the extent lock so buffered
8040 * readers will get the updated i_size when we unlock.
8042 if (!dio_data
->overwrite
&& start
+ len
> i_size_read(inode
))
8043 i_size_write(inode
, start
+ len
);
8045 adjust_dio_outstanding_extents(inode
, dio_data
, len
);
8046 WARN_ON(dio_data
->reserve
< len
);
8047 dio_data
->reserve
-= len
;
8048 dio_data
->unsubmitted_oe_range_end
= start
+ len
;
8049 current
->journal_info
= dio_data
;
8053 * In the case of write we need to clear and unlock the entire range,
8054 * in the case of read we need to unlock only the end area that we
8055 * aren't using if there is any left over space.
8057 if (lockstart
< lockend
) {
8058 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
8059 lockend
, unlock_bits
, 1, 0,
8060 &cached_state
, GFP_NOFS
);
8062 free_extent_state(cached_state
);
8065 free_extent_map(em
);
8070 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
8071 unlock_bits
, 1, 0, &cached_state
, GFP_NOFS
);
8074 current
->journal_info
= dio_data
;
8076 * Compensate the delalloc release we do in btrfs_direct_IO() when we
8077 * write less data then expected, so that we don't underflow our inode's
8078 * outstanding extents counter.
8080 if (create
&& dio_data
)
8081 adjust_dio_outstanding_extents(inode
, dio_data
, len
);
8086 static inline blk_status_t
submit_dio_repair_bio(struct inode
*inode
,
8090 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8093 BUG_ON(bio_op(bio
) == REQ_OP_WRITE
);
8097 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, BTRFS_WQ_ENDIO_DIO_REPAIR
);
8101 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
8107 static int btrfs_check_dio_repairable(struct inode
*inode
,
8108 struct bio
*failed_bio
,
8109 struct io_failure_record
*failrec
,
8112 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8115 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
, failrec
->len
);
8116 if (num_copies
== 1) {
8118 * we only have a single copy of the data, so don't bother with
8119 * all the retry and error correction code that follows. no
8120 * matter what the error is, it is very likely to persist.
8122 btrfs_debug(fs_info
,
8123 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
8124 num_copies
, failrec
->this_mirror
, failed_mirror
);
8128 failrec
->failed_mirror
= failed_mirror
;
8129 failrec
->this_mirror
++;
8130 if (failrec
->this_mirror
== failed_mirror
)
8131 failrec
->this_mirror
++;
8133 if (failrec
->this_mirror
> num_copies
) {
8134 btrfs_debug(fs_info
,
8135 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
8136 num_copies
, failrec
->this_mirror
, failed_mirror
);
8143 static blk_status_t
dio_read_error(struct inode
*inode
, struct bio
*failed_bio
,
8144 struct page
*page
, unsigned int pgoff
,
8145 u64 start
, u64 end
, int failed_mirror
,
8146 bio_end_io_t
*repair_endio
, void *repair_arg
)
8148 struct io_failure_record
*failrec
;
8149 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
8150 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
8153 unsigned int read_mode
= 0;
8156 blk_status_t status
;
8158 BUG_ON(bio_op(failed_bio
) == REQ_OP_WRITE
);
8160 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
8162 return errno_to_blk_status(ret
);
8164 ret
= btrfs_check_dio_repairable(inode
, failed_bio
, failrec
,
8167 free_io_failure(failure_tree
, io_tree
, failrec
);
8168 return BLK_STS_IOERR
;
8171 segs
= bio_segments(failed_bio
);
8173 (failed_bio
->bi_io_vec
->bv_len
> btrfs_inode_sectorsize(inode
)))
8174 read_mode
|= REQ_FAILFAST_DEV
;
8176 isector
= start
- btrfs_io_bio(failed_bio
)->logical
;
8177 isector
>>= inode
->i_sb
->s_blocksize_bits
;
8178 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
8179 pgoff
, isector
, repair_endio
, repair_arg
);
8180 bio_set_op_attrs(bio
, REQ_OP_READ
, read_mode
);
8182 btrfs_debug(BTRFS_I(inode
)->root
->fs_info
,
8183 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
8184 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
8186 status
= submit_dio_repair_bio(inode
, bio
, failrec
->this_mirror
);
8188 free_io_failure(failure_tree
, io_tree
, failrec
);
8195 struct btrfs_retry_complete
{
8196 struct completion done
;
8197 struct inode
*inode
;
8202 static void btrfs_retry_endio_nocsum(struct bio
*bio
)
8204 struct btrfs_retry_complete
*done
= bio
->bi_private
;
8205 struct inode
*inode
= done
->inode
;
8206 struct bio_vec
*bvec
;
8207 struct extent_io_tree
*io_tree
, *failure_tree
;
8213 ASSERT(bio
->bi_vcnt
== 1);
8214 io_tree
= &BTRFS_I(inode
)->io_tree
;
8215 failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
8216 ASSERT(bio
->bi_io_vec
->bv_len
== btrfs_inode_sectorsize(inode
));
8219 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
8220 bio_for_each_segment_all(bvec
, bio
, i
)
8221 clean_io_failure(BTRFS_I(inode
)->root
->fs_info
, failure_tree
,
8222 io_tree
, done
->start
, bvec
->bv_page
,
8223 btrfs_ino(BTRFS_I(inode
)), 0);
8225 complete(&done
->done
);
8229 static blk_status_t
__btrfs_correct_data_nocsum(struct inode
*inode
,
8230 struct btrfs_io_bio
*io_bio
)
8232 struct btrfs_fs_info
*fs_info
;
8233 struct bio_vec bvec
;
8234 struct bvec_iter iter
;
8235 struct btrfs_retry_complete done
;
8241 blk_status_t err
= BLK_STS_OK
;
8243 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
8244 sectorsize
= fs_info
->sectorsize
;
8246 start
= io_bio
->logical
;
8248 io_bio
->bio
.bi_iter
= io_bio
->iter
;
8250 bio_for_each_segment(bvec
, &io_bio
->bio
, iter
) {
8251 nr_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, bvec
.bv_len
);
8252 pgoff
= bvec
.bv_offset
;
8254 next_block_or_try_again
:
8257 init_completion(&done
.done
);
8259 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
.bv_page
,
8260 pgoff
, start
, start
+ sectorsize
- 1,
8262 btrfs_retry_endio_nocsum
, &done
);
8268 wait_for_completion_io(&done
.done
);
8270 if (!done
.uptodate
) {
8271 /* We might have another mirror, so try again */
8272 goto next_block_or_try_again
;
8276 start
+= sectorsize
;
8280 pgoff
+= sectorsize
;
8281 ASSERT(pgoff
< PAGE_SIZE
);
8282 goto next_block_or_try_again
;
8289 static void btrfs_retry_endio(struct bio
*bio
)
8291 struct btrfs_retry_complete
*done
= bio
->bi_private
;
8292 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8293 struct extent_io_tree
*io_tree
, *failure_tree
;
8294 struct inode
*inode
= done
->inode
;
8295 struct bio_vec
*bvec
;
8305 ASSERT(bio
->bi_vcnt
== 1);
8306 ASSERT(bio
->bi_io_vec
->bv_len
== btrfs_inode_sectorsize(done
->inode
));
8308 io_tree
= &BTRFS_I(inode
)->io_tree
;
8309 failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
8311 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
8312 bio_for_each_segment_all(bvec
, bio
, i
) {
8313 ret
= __readpage_endio_check(inode
, io_bio
, i
, bvec
->bv_page
,
8314 bvec
->bv_offset
, done
->start
,
8317 clean_io_failure(BTRFS_I(inode
)->root
->fs_info
,
8318 failure_tree
, io_tree
, done
->start
,
8320 btrfs_ino(BTRFS_I(inode
)),
8326 done
->uptodate
= uptodate
;
8328 complete(&done
->done
);
8332 static blk_status_t
__btrfs_subio_endio_read(struct inode
*inode
,
8333 struct btrfs_io_bio
*io_bio
, blk_status_t err
)
8335 struct btrfs_fs_info
*fs_info
;
8336 struct bio_vec bvec
;
8337 struct bvec_iter iter
;
8338 struct btrfs_retry_complete done
;
8345 bool uptodate
= (err
== 0);
8347 blk_status_t status
;
8349 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
8350 sectorsize
= fs_info
->sectorsize
;
8353 start
= io_bio
->logical
;
8355 io_bio
->bio
.bi_iter
= io_bio
->iter
;
8357 bio_for_each_segment(bvec
, &io_bio
->bio
, iter
) {
8358 nr_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, bvec
.bv_len
);
8360 pgoff
= bvec
.bv_offset
;
8363 csum_pos
= BTRFS_BYTES_TO_BLKS(fs_info
, offset
);
8364 ret
= __readpage_endio_check(inode
, io_bio
, csum_pos
,
8365 bvec
.bv_page
, pgoff
, start
, sectorsize
);
8372 init_completion(&done
.done
);
8374 status
= dio_read_error(inode
, &io_bio
->bio
, bvec
.bv_page
,
8375 pgoff
, start
, start
+ sectorsize
- 1,
8376 io_bio
->mirror_num
, btrfs_retry_endio
,
8383 wait_for_completion_io(&done
.done
);
8385 if (!done
.uptodate
) {
8386 /* We might have another mirror, so try again */
8390 offset
+= sectorsize
;
8391 start
+= sectorsize
;
8397 pgoff
+= sectorsize
;
8398 ASSERT(pgoff
< PAGE_SIZE
);
8406 static blk_status_t
btrfs_subio_endio_read(struct inode
*inode
,
8407 struct btrfs_io_bio
*io_bio
, blk_status_t err
)
8409 bool skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
8413 return __btrfs_correct_data_nocsum(inode
, io_bio
);
8417 return __btrfs_subio_endio_read(inode
, io_bio
, err
);
8421 static void btrfs_endio_direct_read(struct bio
*bio
)
8423 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8424 struct inode
*inode
= dip
->inode
;
8425 struct bio
*dio_bio
;
8426 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8427 blk_status_t err
= bio
->bi_status
;
8429 if (dip
->flags
& BTRFS_DIO_ORIG_BIO_SUBMITTED
)
8430 err
= btrfs_subio_endio_read(inode
, io_bio
, err
);
8432 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
8433 dip
->logical_offset
+ dip
->bytes
- 1);
8434 dio_bio
= dip
->dio_bio
;
8438 dio_bio
->bi_status
= err
;
8439 dio_end_io(dio_bio
);
8442 io_bio
->end_io(io_bio
, blk_status_to_errno(err
));
8446 static void __endio_write_update_ordered(struct inode
*inode
,
8447 const u64 offset
, const u64 bytes
,
8448 const bool uptodate
)
8450 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8451 struct btrfs_ordered_extent
*ordered
= NULL
;
8452 struct btrfs_workqueue
*wq
;
8453 btrfs_work_func_t func
;
8454 u64 ordered_offset
= offset
;
8455 u64 ordered_bytes
= bytes
;
8459 if (btrfs_is_free_space_inode(BTRFS_I(inode
))) {
8460 wq
= fs_info
->endio_freespace_worker
;
8461 func
= btrfs_freespace_write_helper
;
8463 wq
= fs_info
->endio_write_workers
;
8464 func
= btrfs_endio_write_helper
;
8468 last_offset
= ordered_offset
;
8469 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
8476 btrfs_init_work(&ordered
->work
, func
, finish_ordered_fn
, NULL
, NULL
);
8477 btrfs_queue_work(wq
, &ordered
->work
);
8480 * If btrfs_dec_test_ordered_pending does not find any ordered extent
8481 * in the range, we can exit.
8483 if (ordered_offset
== last_offset
)
8486 * our bio might span multiple ordered extents. If we haven't
8487 * completed the accounting for the whole dio, go back and try again
8489 if (ordered_offset
< offset
+ bytes
) {
8490 ordered_bytes
= offset
+ bytes
- ordered_offset
;
8496 static void btrfs_endio_direct_write(struct bio
*bio
)
8498 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8499 struct bio
*dio_bio
= dip
->dio_bio
;
8501 __endio_write_update_ordered(dip
->inode
, dip
->logical_offset
,
8502 dip
->bytes
, !bio
->bi_status
);
8506 dio_bio
->bi_status
= bio
->bi_status
;
8507 dio_end_io(dio_bio
);
8511 static blk_status_t
__btrfs_submit_bio_start_direct_io(void *private_data
,
8512 struct bio
*bio
, int mirror_num
,
8513 unsigned long bio_flags
, u64 offset
)
8515 struct inode
*inode
= private_data
;
8517 ret
= btrfs_csum_one_bio(inode
, bio
, offset
, 1);
8518 BUG_ON(ret
); /* -ENOMEM */
8522 static void btrfs_end_dio_bio(struct bio
*bio
)
8524 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8525 blk_status_t err
= bio
->bi_status
;
8528 btrfs_warn(BTRFS_I(dip
->inode
)->root
->fs_info
,
8529 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8530 btrfs_ino(BTRFS_I(dip
->inode
)), bio_op(bio
),
8532 (unsigned long long)bio
->bi_iter
.bi_sector
,
8533 bio
->bi_iter
.bi_size
, err
);
8535 if (dip
->subio_endio
)
8536 err
= dip
->subio_endio(dip
->inode
, btrfs_io_bio(bio
), err
);
8542 * before atomic variable goto zero, we must make sure
8543 * dip->errors is perceived to be set.
8545 smp_mb__before_atomic();
8548 /* if there are more bios still pending for this dio, just exit */
8549 if (!atomic_dec_and_test(&dip
->pending_bios
))
8553 bio_io_error(dip
->orig_bio
);
8555 dip
->dio_bio
->bi_status
= 0;
8556 bio_endio(dip
->orig_bio
);
8562 static inline blk_status_t
btrfs_lookup_and_bind_dio_csum(struct inode
*inode
,
8563 struct btrfs_dio_private
*dip
,
8567 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8568 struct btrfs_io_bio
*orig_io_bio
= btrfs_io_bio(dip
->orig_bio
);
8572 * We load all the csum data we need when we submit
8573 * the first bio to reduce the csum tree search and
8576 if (dip
->logical_offset
== file_offset
) {
8577 ret
= btrfs_lookup_bio_sums_dio(inode
, dip
->orig_bio
,
8583 if (bio
== dip
->orig_bio
)
8586 file_offset
-= dip
->logical_offset
;
8587 file_offset
>>= inode
->i_sb
->s_blocksize_bits
;
8588 io_bio
->csum
= (u8
*)(((u32
*)orig_io_bio
->csum
) + file_offset
);
8593 static inline blk_status_t
8594 __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
, u64 file_offset
,
8597 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8598 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8599 bool write
= bio_op(bio
) == REQ_OP_WRITE
;
8603 async_submit
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
8608 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, BTRFS_WQ_ENDIO_DATA
);
8613 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
8616 if (write
&& async_submit
) {
8617 ret
= btrfs_wq_submit_bio(fs_info
, bio
, 0, 0,
8619 __btrfs_submit_bio_start_direct_io
,
8620 __btrfs_submit_bio_done
);
8624 * If we aren't doing async submit, calculate the csum of the
8627 ret
= btrfs_csum_one_bio(inode
, bio
, file_offset
, 1);
8631 ret
= btrfs_lookup_and_bind_dio_csum(inode
, dip
, bio
,
8637 ret
= btrfs_map_bio(fs_info
, bio
, 0, async_submit
);
8643 static int btrfs_submit_direct_hook(struct btrfs_dio_private
*dip
)
8645 struct inode
*inode
= dip
->inode
;
8646 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8648 struct bio
*orig_bio
= dip
->orig_bio
;
8649 u64 start_sector
= orig_bio
->bi_iter
.bi_sector
;
8650 u64 file_offset
= dip
->logical_offset
;
8652 int async_submit
= 0;
8654 int clone_offset
= 0;
8657 blk_status_t status
;
8659 map_length
= orig_bio
->bi_iter
.bi_size
;
8660 submit_len
= map_length
;
8661 ret
= btrfs_map_block(fs_info
, btrfs_op(orig_bio
), start_sector
<< 9,
8662 &map_length
, NULL
, 0);
8666 if (map_length
>= submit_len
) {
8668 dip
->flags
|= BTRFS_DIO_ORIG_BIO_SUBMITTED
;
8672 /* async crcs make it difficult to collect full stripe writes. */
8673 if (btrfs_data_alloc_profile(fs_info
) & BTRFS_BLOCK_GROUP_RAID56_MASK
)
8679 ASSERT(map_length
<= INT_MAX
);
8680 atomic_inc(&dip
->pending_bios
);
8682 clone_len
= min_t(int, submit_len
, map_length
);
8685 * This will never fail as it's passing GPF_NOFS and
8686 * the allocation is backed by btrfs_bioset.
8688 bio
= btrfs_bio_clone_partial(orig_bio
, clone_offset
,
8690 bio
->bi_private
= dip
;
8691 bio
->bi_end_io
= btrfs_end_dio_bio
;
8692 btrfs_io_bio(bio
)->logical
= file_offset
;
8694 ASSERT(submit_len
>= clone_len
);
8695 submit_len
-= clone_len
;
8696 if (submit_len
== 0)
8700 * Increase the count before we submit the bio so we know
8701 * the end IO handler won't happen before we increase the
8702 * count. Otherwise, the dip might get freed before we're
8703 * done setting it up.
8705 atomic_inc(&dip
->pending_bios
);
8707 status
= __btrfs_submit_dio_bio(bio
, inode
, file_offset
,
8711 atomic_dec(&dip
->pending_bios
);
8715 clone_offset
+= clone_len
;
8716 start_sector
+= clone_len
>> 9;
8717 file_offset
+= clone_len
;
8719 map_length
= submit_len
;
8720 ret
= btrfs_map_block(fs_info
, btrfs_op(orig_bio
),
8721 start_sector
<< 9, &map_length
, NULL
, 0);
8724 } while (submit_len
> 0);
8727 status
= __btrfs_submit_dio_bio(bio
, inode
, file_offset
, async_submit
);
8735 * before atomic variable goto zero, we must
8736 * make sure dip->errors is perceived to be set.
8738 smp_mb__before_atomic();
8739 if (atomic_dec_and_test(&dip
->pending_bios
))
8740 bio_io_error(dip
->orig_bio
);
8742 /* bio_end_io() will handle error, so we needn't return it */
8746 static void btrfs_submit_direct(struct bio
*dio_bio
, struct inode
*inode
,
8749 struct btrfs_dio_private
*dip
= NULL
;
8750 struct bio
*bio
= NULL
;
8751 struct btrfs_io_bio
*io_bio
;
8752 bool write
= (bio_op(dio_bio
) == REQ_OP_WRITE
);
8755 bio
= btrfs_bio_clone(dio_bio
);
8757 dip
= kzalloc(sizeof(*dip
), GFP_NOFS
);
8763 dip
->private = dio_bio
->bi_private
;
8765 dip
->logical_offset
= file_offset
;
8766 dip
->bytes
= dio_bio
->bi_iter
.bi_size
;
8767 dip
->disk_bytenr
= (u64
)dio_bio
->bi_iter
.bi_sector
<< 9;
8768 bio
->bi_private
= dip
;
8769 dip
->orig_bio
= bio
;
8770 dip
->dio_bio
= dio_bio
;
8771 atomic_set(&dip
->pending_bios
, 0);
8772 io_bio
= btrfs_io_bio(bio
);
8773 io_bio
->logical
= file_offset
;
8776 bio
->bi_end_io
= btrfs_endio_direct_write
;
8778 bio
->bi_end_io
= btrfs_endio_direct_read
;
8779 dip
->subio_endio
= btrfs_subio_endio_read
;
8783 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8784 * even if we fail to submit a bio, because in such case we do the
8785 * corresponding error handling below and it must not be done a second
8786 * time by btrfs_direct_IO().
8789 struct btrfs_dio_data
*dio_data
= current
->journal_info
;
8791 dio_data
->unsubmitted_oe_range_end
= dip
->logical_offset
+
8793 dio_data
->unsubmitted_oe_range_start
=
8794 dio_data
->unsubmitted_oe_range_end
;
8797 ret
= btrfs_submit_direct_hook(dip
);
8802 io_bio
->end_io(io_bio
, ret
);
8806 * If we arrived here it means either we failed to submit the dip
8807 * or we either failed to clone the dio_bio or failed to allocate the
8808 * dip. If we cloned the dio_bio and allocated the dip, we can just
8809 * call bio_endio against our io_bio so that we get proper resource
8810 * cleanup if we fail to submit the dip, otherwise, we must do the
8811 * same as btrfs_endio_direct_[write|read] because we can't call these
8812 * callbacks - they require an allocated dip and a clone of dio_bio.
8817 * The end io callbacks free our dip, do the final put on bio
8818 * and all the cleanup and final put for dio_bio (through
8825 __endio_write_update_ordered(inode
,
8827 dio_bio
->bi_iter
.bi_size
,
8830 unlock_extent(&BTRFS_I(inode
)->io_tree
, file_offset
,
8831 file_offset
+ dio_bio
->bi_iter
.bi_size
- 1);
8833 dio_bio
->bi_status
= BLK_STS_IOERR
;
8835 * Releases and cleans up our dio_bio, no need to bio_put()
8836 * nor bio_endio()/bio_io_error() against dio_bio.
8838 dio_end_io(dio_bio
);
8845 static ssize_t
check_direct_IO(struct btrfs_fs_info
*fs_info
,
8847 const struct iov_iter
*iter
, loff_t offset
)
8851 unsigned int blocksize_mask
= fs_info
->sectorsize
- 1;
8852 ssize_t retval
= -EINVAL
;
8854 if (offset
& blocksize_mask
)
8857 if (iov_iter_alignment(iter
) & blocksize_mask
)
8860 /* If this is a write we don't need to check anymore */
8861 if (iov_iter_rw(iter
) != READ
|| !iter_is_iovec(iter
))
8864 * Check to make sure we don't have duplicate iov_base's in this
8865 * iovec, if so return EINVAL, otherwise we'll get csum errors
8866 * when reading back.
8868 for (seg
= 0; seg
< iter
->nr_segs
; seg
++) {
8869 for (i
= seg
+ 1; i
< iter
->nr_segs
; i
++) {
8870 if (iter
->iov
[seg
].iov_base
== iter
->iov
[i
].iov_base
)
8879 static ssize_t
btrfs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
8881 struct file
*file
= iocb
->ki_filp
;
8882 struct inode
*inode
= file
->f_mapping
->host
;
8883 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8884 struct btrfs_dio_data dio_data
= { 0 };
8885 struct extent_changeset
*data_reserved
= NULL
;
8886 loff_t offset
= iocb
->ki_pos
;
8890 bool relock
= false;
8893 if (check_direct_IO(fs_info
, iocb
, iter
, offset
))
8896 inode_dio_begin(inode
);
8899 * The generic stuff only does filemap_write_and_wait_range, which
8900 * isn't enough if we've written compressed pages to this area, so
8901 * we need to flush the dirty pages again to make absolutely sure
8902 * that any outstanding dirty pages are on disk.
8904 count
= iov_iter_count(iter
);
8905 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
8906 &BTRFS_I(inode
)->runtime_flags
))
8907 filemap_fdatawrite_range(inode
->i_mapping
, offset
,
8908 offset
+ count
- 1);
8910 if (iov_iter_rw(iter
) == WRITE
) {
8912 * If the write DIO is beyond the EOF, we need update
8913 * the isize, but it is protected by i_mutex. So we can
8914 * not unlock the i_mutex at this case.
8916 if (offset
+ count
<= inode
->i_size
) {
8917 dio_data
.overwrite
= 1;
8918 inode_unlock(inode
);
8920 } else if (iocb
->ki_flags
& IOCB_NOWAIT
) {
8924 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
,
8928 dio_data
.outstanding_extents
= count_max_extents(count
);
8931 * We need to know how many extents we reserved so that we can
8932 * do the accounting properly if we go over the number we
8933 * originally calculated. Abuse current->journal_info for this.
8935 dio_data
.reserve
= round_up(count
,
8936 fs_info
->sectorsize
);
8937 dio_data
.unsubmitted_oe_range_start
= (u64
)offset
;
8938 dio_data
.unsubmitted_oe_range_end
= (u64
)offset
;
8939 current
->journal_info
= &dio_data
;
8940 down_read(&BTRFS_I(inode
)->dio_sem
);
8941 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK
,
8942 &BTRFS_I(inode
)->runtime_flags
)) {
8943 inode_dio_end(inode
);
8944 flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
8948 ret
= __blockdev_direct_IO(iocb
, inode
,
8949 fs_info
->fs_devices
->latest_bdev
,
8950 iter
, btrfs_get_blocks_direct
, NULL
,
8951 btrfs_submit_direct
, flags
);
8952 if (iov_iter_rw(iter
) == WRITE
) {
8953 up_read(&BTRFS_I(inode
)->dio_sem
);
8954 current
->journal_info
= NULL
;
8955 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
8956 if (dio_data
.reserve
)
8957 btrfs_delalloc_release_space(inode
, data_reserved
,
8958 offset
, dio_data
.reserve
);
8960 * On error we might have left some ordered extents
8961 * without submitting corresponding bios for them, so
8962 * cleanup them up to avoid other tasks getting them
8963 * and waiting for them to complete forever.
8965 if (dio_data
.unsubmitted_oe_range_start
<
8966 dio_data
.unsubmitted_oe_range_end
)
8967 __endio_write_update_ordered(inode
,
8968 dio_data
.unsubmitted_oe_range_start
,
8969 dio_data
.unsubmitted_oe_range_end
-
8970 dio_data
.unsubmitted_oe_range_start
,
8972 } else if (ret
>= 0 && (size_t)ret
< count
)
8973 btrfs_delalloc_release_space(inode
, data_reserved
,
8974 offset
, count
- (size_t)ret
);
8978 inode_dio_end(inode
);
8982 extent_changeset_free(data_reserved
);
8986 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8988 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
8989 __u64 start
, __u64 len
)
8993 ret
= fiemap_check_flags(fieinfo
, BTRFS_FIEMAP_FLAGS
);
8997 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
9000 int btrfs_readpage(struct file
*file
, struct page
*page
)
9002 struct extent_io_tree
*tree
;
9003 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
9004 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
9007 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
9009 struct extent_io_tree
*tree
;
9010 struct inode
*inode
= page
->mapping
->host
;
9013 if (current
->flags
& PF_MEMALLOC
) {
9014 redirty_page_for_writepage(wbc
, page
);
9020 * If we are under memory pressure we will call this directly from the
9021 * VM, we need to make sure we have the inode referenced for the ordered
9022 * extent. If not just return like we didn't do anything.
9024 if (!igrab(inode
)) {
9025 redirty_page_for_writepage(wbc
, page
);
9026 return AOP_WRITEPAGE_ACTIVATE
;
9028 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
9029 ret
= extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
9030 btrfs_add_delayed_iput(inode
);
9034 static int btrfs_writepages(struct address_space
*mapping
,
9035 struct writeback_control
*wbc
)
9037 struct extent_io_tree
*tree
;
9039 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
9040 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
9044 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
9045 struct list_head
*pages
, unsigned nr_pages
)
9047 struct extent_io_tree
*tree
;
9048 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
9049 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
9052 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
9054 struct extent_io_tree
*tree
;
9055 struct extent_map_tree
*map
;
9058 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
9059 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
9060 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
9062 ClearPagePrivate(page
);
9063 set_page_private(page
, 0);
9069 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
9071 if (PageWriteback(page
) || PageDirty(page
))
9073 return __btrfs_releasepage(page
, gfp_flags
);
9076 static void btrfs_invalidatepage(struct page
*page
, unsigned int offset
,
9077 unsigned int length
)
9079 struct inode
*inode
= page
->mapping
->host
;
9080 struct extent_io_tree
*tree
;
9081 struct btrfs_ordered_extent
*ordered
;
9082 struct extent_state
*cached_state
= NULL
;
9083 u64 page_start
= page_offset(page
);
9084 u64 page_end
= page_start
+ PAGE_SIZE
- 1;
9087 int inode_evicting
= inode
->i_state
& I_FREEING
;
9090 * we have the page locked, so new writeback can't start,
9091 * and the dirty bit won't be cleared while we are here.
9093 * Wait for IO on this page so that we can safely clear
9094 * the PagePrivate2 bit and do ordered accounting
9096 wait_on_page_writeback(page
);
9098 tree
= &BTRFS_I(inode
)->io_tree
;
9100 btrfs_releasepage(page
, GFP_NOFS
);
9104 if (!inode_evicting
)
9105 lock_extent_bits(tree
, page_start
, page_end
, &cached_state
);
9108 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), start
,
9109 page_end
- start
+ 1);
9111 end
= min(page_end
, ordered
->file_offset
+ ordered
->len
- 1);
9113 * IO on this page will never be started, so we need
9114 * to account for any ordered extents now
9116 if (!inode_evicting
)
9117 clear_extent_bit(tree
, start
, end
,
9118 EXTENT_DIRTY
| EXTENT_DELALLOC
|
9119 EXTENT_DELALLOC_NEW
|
9120 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
9121 EXTENT_DEFRAG
, 1, 0, &cached_state
,
9124 * whoever cleared the private bit is responsible
9125 * for the finish_ordered_io
9127 if (TestClearPagePrivate2(page
)) {
9128 struct btrfs_ordered_inode_tree
*tree
;
9131 tree
= &BTRFS_I(inode
)->ordered_tree
;
9133 spin_lock_irq(&tree
->lock
);
9134 set_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
);
9135 new_len
= start
- ordered
->file_offset
;
9136 if (new_len
< ordered
->truncated_len
)
9137 ordered
->truncated_len
= new_len
;
9138 spin_unlock_irq(&tree
->lock
);
9140 if (btrfs_dec_test_ordered_pending(inode
, &ordered
,
9142 end
- start
+ 1, 1))
9143 btrfs_finish_ordered_io(ordered
);
9145 btrfs_put_ordered_extent(ordered
);
9146 if (!inode_evicting
) {
9147 cached_state
= NULL
;
9148 lock_extent_bits(tree
, start
, end
,
9153 if (start
< page_end
)
9158 * Qgroup reserved space handler
9159 * Page here will be either
9160 * 1) Already written to disk
9161 * In this case, its reserved space is released from data rsv map
9162 * and will be freed by delayed_ref handler finally.
9163 * So even we call qgroup_free_data(), it won't decrease reserved
9165 * 2) Not written to disk
9166 * This means the reserved space should be freed here. However,
9167 * if a truncate invalidates the page (by clearing PageDirty)
9168 * and the page is accounted for while allocating extent
9169 * in btrfs_check_data_free_space() we let delayed_ref to
9170 * free the entire extent.
9172 if (PageDirty(page
))
9173 btrfs_qgroup_free_data(inode
, NULL
, page_start
, PAGE_SIZE
);
9174 if (!inode_evicting
) {
9175 clear_extent_bit(tree
, page_start
, page_end
,
9176 EXTENT_LOCKED
| EXTENT_DIRTY
|
9177 EXTENT_DELALLOC
| EXTENT_DELALLOC_NEW
|
9178 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
, 1, 1,
9179 &cached_state
, GFP_NOFS
);
9181 __btrfs_releasepage(page
, GFP_NOFS
);
9184 ClearPageChecked(page
);
9185 if (PagePrivate(page
)) {
9186 ClearPagePrivate(page
);
9187 set_page_private(page
, 0);
9193 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
9194 * called from a page fault handler when a page is first dirtied. Hence we must
9195 * be careful to check for EOF conditions here. We set the page up correctly
9196 * for a written page which means we get ENOSPC checking when writing into
9197 * holes and correct delalloc and unwritten extent mapping on filesystems that
9198 * support these features.
9200 * We are not allowed to take the i_mutex here so we have to play games to
9201 * protect against truncate races as the page could now be beyond EOF. Because
9202 * vmtruncate() writes the inode size before removing pages, once we have the
9203 * page lock we can determine safely if the page is beyond EOF. If it is not
9204 * beyond EOF, then the page is guaranteed safe against truncation until we
9207 int btrfs_page_mkwrite(struct vm_fault
*vmf
)
9209 struct page
*page
= vmf
->page
;
9210 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
9211 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
9212 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
9213 struct btrfs_ordered_extent
*ordered
;
9214 struct extent_state
*cached_state
= NULL
;
9215 struct extent_changeset
*data_reserved
= NULL
;
9217 unsigned long zero_start
;
9226 reserved_space
= PAGE_SIZE
;
9228 sb_start_pagefault(inode
->i_sb
);
9229 page_start
= page_offset(page
);
9230 page_end
= page_start
+ PAGE_SIZE
- 1;
9234 * Reserving delalloc space after obtaining the page lock can lead to
9235 * deadlock. For example, if a dirty page is locked by this function
9236 * and the call to btrfs_delalloc_reserve_space() ends up triggering
9237 * dirty page write out, then the btrfs_writepage() function could
9238 * end up waiting indefinitely to get a lock on the page currently
9239 * being processed by btrfs_page_mkwrite() function.
9241 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
, page_start
,
9244 ret
= file_update_time(vmf
->vma
->vm_file
);
9250 else /* -ENOSPC, -EIO, etc */
9251 ret
= VM_FAULT_SIGBUS
;
9257 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
9260 size
= i_size_read(inode
);
9262 if ((page
->mapping
!= inode
->i_mapping
) ||
9263 (page_start
>= size
)) {
9264 /* page got truncated out from underneath us */
9267 wait_on_page_writeback(page
);
9269 lock_extent_bits(io_tree
, page_start
, page_end
, &cached_state
);
9270 set_page_extent_mapped(page
);
9273 * we can't set the delalloc bits if there are pending ordered
9274 * extents. Drop our locks and wait for them to finish
9276 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), page_start
,
9279 unlock_extent_cached(io_tree
, page_start
, page_end
,
9280 &cached_state
, GFP_NOFS
);
9282 btrfs_start_ordered_extent(inode
, ordered
, 1);
9283 btrfs_put_ordered_extent(ordered
);
9287 if (page
->index
== ((size
- 1) >> PAGE_SHIFT
)) {
9288 reserved_space
= round_up(size
- page_start
,
9289 fs_info
->sectorsize
);
9290 if (reserved_space
< PAGE_SIZE
) {
9291 end
= page_start
+ reserved_space
- 1;
9292 spin_lock(&BTRFS_I(inode
)->lock
);
9293 BTRFS_I(inode
)->outstanding_extents
++;
9294 spin_unlock(&BTRFS_I(inode
)->lock
);
9295 btrfs_delalloc_release_space(inode
, data_reserved
,
9296 page_start
, PAGE_SIZE
- reserved_space
);
9301 * page_mkwrite gets called when the page is firstly dirtied after it's
9302 * faulted in, but write(2) could also dirty a page and set delalloc
9303 * bits, thus in this case for space account reason, we still need to
9304 * clear any delalloc bits within this page range since we have to
9305 * reserve data&meta space before lock_page() (see above comments).
9307 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, end
,
9308 EXTENT_DIRTY
| EXTENT_DELALLOC
|
9309 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
9310 0, 0, &cached_state
, GFP_NOFS
);
9312 ret
= btrfs_set_extent_delalloc(inode
, page_start
, end
,
9315 unlock_extent_cached(io_tree
, page_start
, page_end
,
9316 &cached_state
, GFP_NOFS
);
9317 ret
= VM_FAULT_SIGBUS
;
9322 /* page is wholly or partially inside EOF */
9323 if (page_start
+ PAGE_SIZE
> size
)
9324 zero_start
= size
& ~PAGE_MASK
;
9326 zero_start
= PAGE_SIZE
;
9328 if (zero_start
!= PAGE_SIZE
) {
9330 memset(kaddr
+ zero_start
, 0, PAGE_SIZE
- zero_start
);
9331 flush_dcache_page(page
);
9334 ClearPageChecked(page
);
9335 set_page_dirty(page
);
9336 SetPageUptodate(page
);
9338 BTRFS_I(inode
)->last_trans
= fs_info
->generation
;
9339 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
9340 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
9342 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
9346 sb_end_pagefault(inode
->i_sb
);
9347 extent_changeset_free(data_reserved
);
9348 return VM_FAULT_LOCKED
;
9352 btrfs_delalloc_release_space(inode
, data_reserved
, page_start
,
9355 sb_end_pagefault(inode
->i_sb
);
9356 extent_changeset_free(data_reserved
);
9360 static int btrfs_truncate(struct inode
*inode
)
9362 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
9363 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9364 struct btrfs_block_rsv
*rsv
;
9367 struct btrfs_trans_handle
*trans
;
9368 u64 mask
= fs_info
->sectorsize
- 1;
9369 u64 min_size
= btrfs_calc_trunc_metadata_size(fs_info
, 1);
9371 ret
= btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
),
9377 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9378 * 3 things going on here
9380 * 1) We need to reserve space for our orphan item and the space to
9381 * delete our orphan item. Lord knows we don't want to have a dangling
9382 * orphan item because we didn't reserve space to remove it.
9384 * 2) We need to reserve space to update our inode.
9386 * 3) We need to have something to cache all the space that is going to
9387 * be free'd up by the truncate operation, but also have some slack
9388 * space reserved in case it uses space during the truncate (thank you
9389 * very much snapshotting).
9391 * And we need these to all be separate. The fact is we can use a lot of
9392 * space doing the truncate, and we have no earthly idea how much space
9393 * we will use, so we need the truncate reservation to be separate so it
9394 * doesn't end up using space reserved for updating the inode or
9395 * removing the orphan item. We also need to be able to stop the
9396 * transaction and start a new one, which means we need to be able to
9397 * update the inode several times, and we have no idea of knowing how
9398 * many times that will be, so we can't just reserve 1 item for the
9399 * entirety of the operation, so that has to be done separately as well.
9400 * Then there is the orphan item, which does indeed need to be held on
9401 * to for the whole operation, and we need nobody to touch this reserved
9402 * space except the orphan code.
9404 * So that leaves us with
9406 * 1) root->orphan_block_rsv - for the orphan deletion.
9407 * 2) rsv - for the truncate reservation, which we will steal from the
9408 * transaction reservation.
9409 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9410 * updating the inode.
9412 rsv
= btrfs_alloc_block_rsv(fs_info
, BTRFS_BLOCK_RSV_TEMP
);
9415 rsv
->size
= min_size
;
9419 * 1 for the truncate slack space
9420 * 1 for updating the inode.
9422 trans
= btrfs_start_transaction(root
, 2);
9423 if (IS_ERR(trans
)) {
9424 err
= PTR_ERR(trans
);
9428 /* Migrate the slack space for the truncate to our reserve */
9429 ret
= btrfs_block_rsv_migrate(&fs_info
->trans_block_rsv
, rsv
,
9434 * So if we truncate and then write and fsync we normally would just
9435 * write the extents that changed, which is a problem if we need to
9436 * first truncate that entire inode. So set this flag so we write out
9437 * all of the extents in the inode to the sync log so we're completely
9440 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
9441 trans
->block_rsv
= rsv
;
9444 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
9446 BTRFS_EXTENT_DATA_KEY
);
9447 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
) {
9452 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
9453 ret
= btrfs_update_inode(trans
, root
, inode
);
9459 btrfs_end_transaction(trans
);
9460 btrfs_btree_balance_dirty(fs_info
);
9462 trans
= btrfs_start_transaction(root
, 2);
9463 if (IS_ERR(trans
)) {
9464 ret
= err
= PTR_ERR(trans
);
9469 btrfs_block_rsv_release(fs_info
, rsv
, -1);
9470 ret
= btrfs_block_rsv_migrate(&fs_info
->trans_block_rsv
,
9472 BUG_ON(ret
); /* shouldn't happen */
9473 trans
->block_rsv
= rsv
;
9476 if (ret
== 0 && inode
->i_nlink
> 0) {
9477 trans
->block_rsv
= root
->orphan_block_rsv
;
9478 ret
= btrfs_orphan_del(trans
, BTRFS_I(inode
));
9484 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
9485 ret
= btrfs_update_inode(trans
, root
, inode
);
9489 ret
= btrfs_end_transaction(trans
);
9490 btrfs_btree_balance_dirty(fs_info
);
9493 btrfs_free_block_rsv(fs_info
, rsv
);
9502 * create a new subvolume directory/inode (helper for the ioctl).
9504 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
9505 struct btrfs_root
*new_root
,
9506 struct btrfs_root
*parent_root
,
9509 struct inode
*inode
;
9513 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
9514 new_dirid
, new_dirid
,
9515 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
9518 return PTR_ERR(inode
);
9519 inode
->i_op
= &btrfs_dir_inode_operations
;
9520 inode
->i_fop
= &btrfs_dir_file_operations
;
9522 set_nlink(inode
, 1);
9523 btrfs_i_size_write(BTRFS_I(inode
), 0);
9524 unlock_new_inode(inode
);
9526 err
= btrfs_subvol_inherit_props(trans
, new_root
, parent_root
);
9528 btrfs_err(new_root
->fs_info
,
9529 "error inheriting subvolume %llu properties: %d",
9530 new_root
->root_key
.objectid
, err
);
9532 err
= btrfs_update_inode(trans
, new_root
, inode
);
9538 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
9540 struct btrfs_inode
*ei
;
9541 struct inode
*inode
;
9543 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
9550 ei
->last_sub_trans
= 0;
9551 ei
->logged_trans
= 0;
9552 ei
->delalloc_bytes
= 0;
9553 ei
->new_delalloc_bytes
= 0;
9554 ei
->defrag_bytes
= 0;
9555 ei
->disk_i_size
= 0;
9558 ei
->index_cnt
= (u64
)-1;
9560 ei
->last_unlink_trans
= 0;
9561 ei
->last_link_trans
= 0;
9562 ei
->last_log_commit
= 0;
9563 ei
->delayed_iput_count
= 0;
9565 spin_lock_init(&ei
->lock
);
9566 ei
->outstanding_extents
= 0;
9567 ei
->reserved_extents
= 0;
9569 ei
->runtime_flags
= 0;
9570 ei
->prop_compress
= BTRFS_COMPRESS_NONE
;
9571 ei
->defrag_compress
= BTRFS_COMPRESS_NONE
;
9573 ei
->delayed_node
= NULL
;
9575 ei
->i_otime
.tv_sec
= 0;
9576 ei
->i_otime
.tv_nsec
= 0;
9578 inode
= &ei
->vfs_inode
;
9579 extent_map_tree_init(&ei
->extent_tree
);
9580 extent_io_tree_init(&ei
->io_tree
, inode
);
9581 extent_io_tree_init(&ei
->io_failure_tree
, inode
);
9582 ei
->io_tree
.track_uptodate
= 1;
9583 ei
->io_failure_tree
.track_uptodate
= 1;
9584 atomic_set(&ei
->sync_writers
, 0);
9585 mutex_init(&ei
->log_mutex
);
9586 mutex_init(&ei
->delalloc_mutex
);
9587 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
9588 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
9589 INIT_LIST_HEAD(&ei
->delayed_iput
);
9590 RB_CLEAR_NODE(&ei
->rb_node
);
9591 init_rwsem(&ei
->dio_sem
);
9596 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9597 void btrfs_test_destroy_inode(struct inode
*inode
)
9599 btrfs_drop_extent_cache(BTRFS_I(inode
), 0, (u64
)-1, 0);
9600 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9604 static void btrfs_i_callback(struct rcu_head
*head
)
9606 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
9607 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9610 void btrfs_destroy_inode(struct inode
*inode
)
9612 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
9613 struct btrfs_ordered_extent
*ordered
;
9614 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9616 WARN_ON(!hlist_empty(&inode
->i_dentry
));
9617 WARN_ON(inode
->i_data
.nrpages
);
9618 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
9619 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
9620 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
9621 WARN_ON(BTRFS_I(inode
)->new_delalloc_bytes
);
9622 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
9623 WARN_ON(BTRFS_I(inode
)->defrag_bytes
);
9626 * This can happen where we create an inode, but somebody else also
9627 * created the same inode and we need to destroy the one we already
9633 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
9634 &BTRFS_I(inode
)->runtime_flags
)) {
9635 btrfs_info(fs_info
, "inode %llu still on the orphan list",
9636 btrfs_ino(BTRFS_I(inode
)));
9637 atomic_dec(&root
->orphan_inodes
);
9641 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
9646 "found ordered extent %llu %llu on inode cleanup",
9647 ordered
->file_offset
, ordered
->len
);
9648 btrfs_remove_ordered_extent(inode
, ordered
);
9649 btrfs_put_ordered_extent(ordered
);
9650 btrfs_put_ordered_extent(ordered
);
9653 btrfs_qgroup_check_reserved_leak(inode
);
9654 inode_tree_del(inode
);
9655 btrfs_drop_extent_cache(BTRFS_I(inode
), 0, (u64
)-1, 0);
9657 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
9660 int btrfs_drop_inode(struct inode
*inode
)
9662 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9667 /* the snap/subvol tree is on deleting */
9668 if (btrfs_root_refs(&root
->root_item
) == 0)
9671 return generic_drop_inode(inode
);
9674 static void init_once(void *foo
)
9676 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
9678 inode_init_once(&ei
->vfs_inode
);
9681 void btrfs_destroy_cachep(void)
9684 * Make sure all delayed rcu free inodes are flushed before we
9688 kmem_cache_destroy(btrfs_inode_cachep
);
9689 kmem_cache_destroy(btrfs_trans_handle_cachep
);
9690 kmem_cache_destroy(btrfs_path_cachep
);
9691 kmem_cache_destroy(btrfs_free_space_cachep
);
9694 int btrfs_init_cachep(void)
9696 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
9697 sizeof(struct btrfs_inode
), 0,
9698 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
| SLAB_ACCOUNT
,
9700 if (!btrfs_inode_cachep
)
9703 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
9704 sizeof(struct btrfs_trans_handle
), 0,
9705 SLAB_TEMPORARY
| SLAB_MEM_SPREAD
, NULL
);
9706 if (!btrfs_trans_handle_cachep
)
9709 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
9710 sizeof(struct btrfs_path
), 0,
9711 SLAB_MEM_SPREAD
, NULL
);
9712 if (!btrfs_path_cachep
)
9715 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
9716 sizeof(struct btrfs_free_space
), 0,
9717 SLAB_MEM_SPREAD
, NULL
);
9718 if (!btrfs_free_space_cachep
)
9723 btrfs_destroy_cachep();
9727 static int btrfs_getattr(const struct path
*path
, struct kstat
*stat
,
9728 u32 request_mask
, unsigned int flags
)
9731 struct inode
*inode
= d_inode(path
->dentry
);
9732 u32 blocksize
= inode
->i_sb
->s_blocksize
;
9733 u32 bi_flags
= BTRFS_I(inode
)->flags
;
9735 stat
->result_mask
|= STATX_BTIME
;
9736 stat
->btime
.tv_sec
= BTRFS_I(inode
)->i_otime
.tv_sec
;
9737 stat
->btime
.tv_nsec
= BTRFS_I(inode
)->i_otime
.tv_nsec
;
9738 if (bi_flags
& BTRFS_INODE_APPEND
)
9739 stat
->attributes
|= STATX_ATTR_APPEND
;
9740 if (bi_flags
& BTRFS_INODE_COMPRESS
)
9741 stat
->attributes
|= STATX_ATTR_COMPRESSED
;
9742 if (bi_flags
& BTRFS_INODE_IMMUTABLE
)
9743 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
9744 if (bi_flags
& BTRFS_INODE_NODUMP
)
9745 stat
->attributes
|= STATX_ATTR_NODUMP
;
9747 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
9748 STATX_ATTR_COMPRESSED
|
9749 STATX_ATTR_IMMUTABLE
|
9752 generic_fillattr(inode
, stat
);
9753 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
9755 spin_lock(&BTRFS_I(inode
)->lock
);
9756 delalloc_bytes
= BTRFS_I(inode
)->new_delalloc_bytes
;
9757 spin_unlock(&BTRFS_I(inode
)->lock
);
9758 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
9759 ALIGN(delalloc_bytes
, blocksize
)) >> 9;
9763 static int btrfs_rename_exchange(struct inode
*old_dir
,
9764 struct dentry
*old_dentry
,
9765 struct inode
*new_dir
,
9766 struct dentry
*new_dentry
)
9768 struct btrfs_fs_info
*fs_info
= btrfs_sb(old_dir
->i_sb
);
9769 struct btrfs_trans_handle
*trans
;
9770 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9771 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9772 struct inode
*new_inode
= new_dentry
->d_inode
;
9773 struct inode
*old_inode
= old_dentry
->d_inode
;
9774 struct timespec ctime
= current_time(old_inode
);
9775 struct dentry
*parent
;
9776 u64 old_ino
= btrfs_ino(BTRFS_I(old_inode
));
9777 u64 new_ino
= btrfs_ino(BTRFS_I(new_inode
));
9783 bool root_log_pinned
= false;
9784 bool dest_log_pinned
= false;
9786 /* we only allow rename subvolume link between subvolumes */
9787 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9790 /* close the race window with snapshot create/destroy ioctl */
9791 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9792 down_read(&fs_info
->subvol_sem
);
9793 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9794 down_read(&fs_info
->subvol_sem
);
9797 * We want to reserve the absolute worst case amount of items. So if
9798 * both inodes are subvols and we need to unlink them then that would
9799 * require 4 item modifications, but if they are both normal inodes it
9800 * would require 5 item modifications, so we'll assume their normal
9801 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9802 * should cover the worst case number of items we'll modify.
9804 trans
= btrfs_start_transaction(root
, 12);
9805 if (IS_ERR(trans
)) {
9806 ret
= PTR_ERR(trans
);
9811 * We need to find a free sequence number both in the source and
9812 * in the destination directory for the exchange.
9814 ret
= btrfs_set_inode_index(BTRFS_I(new_dir
), &old_idx
);
9817 ret
= btrfs_set_inode_index(BTRFS_I(old_dir
), &new_idx
);
9821 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9822 BTRFS_I(new_inode
)->dir_index
= 0ULL;
9824 /* Reference for the source. */
9825 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9826 /* force full log commit if subvolume involved. */
9827 btrfs_set_log_full_commit(fs_info
, trans
);
9829 btrfs_pin_log_trans(root
);
9830 root_log_pinned
= true;
9831 ret
= btrfs_insert_inode_ref(trans
, dest
,
9832 new_dentry
->d_name
.name
,
9833 new_dentry
->d_name
.len
,
9835 btrfs_ino(BTRFS_I(new_dir
)),
9841 /* And now for the dest. */
9842 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9843 /* force full log commit if subvolume involved. */
9844 btrfs_set_log_full_commit(fs_info
, trans
);
9846 btrfs_pin_log_trans(dest
);
9847 dest_log_pinned
= true;
9848 ret
= btrfs_insert_inode_ref(trans
, root
,
9849 old_dentry
->d_name
.name
,
9850 old_dentry
->d_name
.len
,
9852 btrfs_ino(BTRFS_I(old_dir
)),
9858 /* Update inode version and ctime/mtime. */
9859 inode_inc_iversion(old_dir
);
9860 inode_inc_iversion(new_dir
);
9861 inode_inc_iversion(old_inode
);
9862 inode_inc_iversion(new_inode
);
9863 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
9864 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
9865 old_inode
->i_ctime
= ctime
;
9866 new_inode
->i_ctime
= ctime
;
9868 if (old_dentry
->d_parent
!= new_dentry
->d_parent
) {
9869 btrfs_record_unlink_dir(trans
, BTRFS_I(old_dir
),
9870 BTRFS_I(old_inode
), 1);
9871 btrfs_record_unlink_dir(trans
, BTRFS_I(new_dir
),
9872 BTRFS_I(new_inode
), 1);
9875 /* src is a subvolume */
9876 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9877 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
9878 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
,
9880 old_dentry
->d_name
.name
,
9881 old_dentry
->d_name
.len
);
9882 } else { /* src is an inode */
9883 ret
= __btrfs_unlink_inode(trans
, root
, BTRFS_I(old_dir
),
9884 BTRFS_I(old_dentry
->d_inode
),
9885 old_dentry
->d_name
.name
,
9886 old_dentry
->d_name
.len
);
9888 ret
= btrfs_update_inode(trans
, root
, old_inode
);
9891 btrfs_abort_transaction(trans
, ret
);
9895 /* dest is a subvolume */
9896 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9897 root_objectid
= BTRFS_I(new_inode
)->root
->root_key
.objectid
;
9898 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
9900 new_dentry
->d_name
.name
,
9901 new_dentry
->d_name
.len
);
9902 } else { /* dest is an inode */
9903 ret
= __btrfs_unlink_inode(trans
, dest
, BTRFS_I(new_dir
),
9904 BTRFS_I(new_dentry
->d_inode
),
9905 new_dentry
->d_name
.name
,
9906 new_dentry
->d_name
.len
);
9908 ret
= btrfs_update_inode(trans
, dest
, new_inode
);
9911 btrfs_abort_transaction(trans
, ret
);
9915 ret
= btrfs_add_link(trans
, BTRFS_I(new_dir
), BTRFS_I(old_inode
),
9916 new_dentry
->d_name
.name
,
9917 new_dentry
->d_name
.len
, 0, old_idx
);
9919 btrfs_abort_transaction(trans
, ret
);
9923 ret
= btrfs_add_link(trans
, BTRFS_I(old_dir
), BTRFS_I(new_inode
),
9924 old_dentry
->d_name
.name
,
9925 old_dentry
->d_name
.len
, 0, new_idx
);
9927 btrfs_abort_transaction(trans
, ret
);
9931 if (old_inode
->i_nlink
== 1)
9932 BTRFS_I(old_inode
)->dir_index
= old_idx
;
9933 if (new_inode
->i_nlink
== 1)
9934 BTRFS_I(new_inode
)->dir_index
= new_idx
;
9936 if (root_log_pinned
) {
9937 parent
= new_dentry
->d_parent
;
9938 btrfs_log_new_name(trans
, BTRFS_I(old_inode
), BTRFS_I(old_dir
),
9940 btrfs_end_log_trans(root
);
9941 root_log_pinned
= false;
9943 if (dest_log_pinned
) {
9944 parent
= old_dentry
->d_parent
;
9945 btrfs_log_new_name(trans
, BTRFS_I(new_inode
), BTRFS_I(new_dir
),
9947 btrfs_end_log_trans(dest
);
9948 dest_log_pinned
= false;
9952 * If we have pinned a log and an error happened, we unpin tasks
9953 * trying to sync the log and force them to fallback to a transaction
9954 * commit if the log currently contains any of the inodes involved in
9955 * this rename operation (to ensure we do not persist a log with an
9956 * inconsistent state for any of these inodes or leading to any
9957 * inconsistencies when replayed). If the transaction was aborted, the
9958 * abortion reason is propagated to userspace when attempting to commit
9959 * the transaction. If the log does not contain any of these inodes, we
9960 * allow the tasks to sync it.
9962 if (ret
&& (root_log_pinned
|| dest_log_pinned
)) {
9963 if (btrfs_inode_in_log(BTRFS_I(old_dir
), fs_info
->generation
) ||
9964 btrfs_inode_in_log(BTRFS_I(new_dir
), fs_info
->generation
) ||
9965 btrfs_inode_in_log(BTRFS_I(old_inode
), fs_info
->generation
) ||
9967 btrfs_inode_in_log(BTRFS_I(new_inode
), fs_info
->generation
)))
9968 btrfs_set_log_full_commit(fs_info
, trans
);
9970 if (root_log_pinned
) {
9971 btrfs_end_log_trans(root
);
9972 root_log_pinned
= false;
9974 if (dest_log_pinned
) {
9975 btrfs_end_log_trans(dest
);
9976 dest_log_pinned
= false;
9979 ret2
= btrfs_end_transaction(trans
);
9980 ret
= ret
? ret
: ret2
;
9982 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9983 up_read(&fs_info
->subvol_sem
);
9984 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9985 up_read(&fs_info
->subvol_sem
);
9990 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle
*trans
,
9991 struct btrfs_root
*root
,
9993 struct dentry
*dentry
)
9996 struct inode
*inode
;
10000 ret
= btrfs_find_free_ino(root
, &objectid
);
10004 inode
= btrfs_new_inode(trans
, root
, dir
,
10005 dentry
->d_name
.name
,
10006 dentry
->d_name
.len
,
10007 btrfs_ino(BTRFS_I(dir
)),
10009 S_IFCHR
| WHITEOUT_MODE
,
10012 if (IS_ERR(inode
)) {
10013 ret
= PTR_ERR(inode
);
10017 inode
->i_op
= &btrfs_special_inode_operations
;
10018 init_special_inode(inode
, inode
->i_mode
,
10021 ret
= btrfs_init_inode_security(trans
, inode
, dir
,
10026 ret
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
,
10027 BTRFS_I(inode
), 0, index
);
10031 ret
= btrfs_update_inode(trans
, root
, inode
);
10033 unlock_new_inode(inode
);
10035 inode_dec_link_count(inode
);
10041 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
10042 struct inode
*new_dir
, struct dentry
*new_dentry
,
10043 unsigned int flags
)
10045 struct btrfs_fs_info
*fs_info
= btrfs_sb(old_dir
->i_sb
);
10046 struct btrfs_trans_handle
*trans
;
10047 unsigned int trans_num_items
;
10048 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
10049 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
10050 struct inode
*new_inode
= d_inode(new_dentry
);
10051 struct inode
*old_inode
= d_inode(old_dentry
);
10055 u64 old_ino
= btrfs_ino(BTRFS_I(old_inode
));
10056 bool log_pinned
= false;
10058 if (btrfs_ino(BTRFS_I(new_dir
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
10061 /* we only allow rename subvolume link between subvolumes */
10062 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
10065 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
10066 (new_inode
&& btrfs_ino(BTRFS_I(new_inode
)) == BTRFS_FIRST_FREE_OBJECTID
))
10069 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
10070 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
10074 /* check for collisions, even if the name isn't there */
10075 ret
= btrfs_check_dir_item_collision(dest
, new_dir
->i_ino
,
10076 new_dentry
->d_name
.name
,
10077 new_dentry
->d_name
.len
);
10080 if (ret
== -EEXIST
) {
10081 /* we shouldn't get
10082 * eexist without a new_inode */
10083 if (WARN_ON(!new_inode
)) {
10087 /* maybe -EOVERFLOW */
10094 * we're using rename to replace one file with another. Start IO on it
10095 * now so we don't add too much work to the end of the transaction
10097 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
)
10098 filemap_flush(old_inode
->i_mapping
);
10100 /* close the racy window with snapshot create/destroy ioctl */
10101 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
10102 down_read(&fs_info
->subvol_sem
);
10104 * We want to reserve the absolute worst case amount of items. So if
10105 * both inodes are subvols and we need to unlink them then that would
10106 * require 4 item modifications, but if they are both normal inodes it
10107 * would require 5 item modifications, so we'll assume they are normal
10108 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
10109 * should cover the worst case number of items we'll modify.
10110 * If our rename has the whiteout flag, we need more 5 units for the
10111 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
10112 * when selinux is enabled).
10114 trans_num_items
= 11;
10115 if (flags
& RENAME_WHITEOUT
)
10116 trans_num_items
+= 5;
10117 trans
= btrfs_start_transaction(root
, trans_num_items
);
10118 if (IS_ERR(trans
)) {
10119 ret
= PTR_ERR(trans
);
10124 btrfs_record_root_in_trans(trans
, dest
);
10126 ret
= btrfs_set_inode_index(BTRFS_I(new_dir
), &index
);
10130 BTRFS_I(old_inode
)->dir_index
= 0ULL;
10131 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
10132 /* force full log commit if subvolume involved. */
10133 btrfs_set_log_full_commit(fs_info
, trans
);
10135 btrfs_pin_log_trans(root
);
10137 ret
= btrfs_insert_inode_ref(trans
, dest
,
10138 new_dentry
->d_name
.name
,
10139 new_dentry
->d_name
.len
,
10141 btrfs_ino(BTRFS_I(new_dir
)), index
);
10146 inode_inc_iversion(old_dir
);
10147 inode_inc_iversion(new_dir
);
10148 inode_inc_iversion(old_inode
);
10149 old_dir
->i_ctime
= old_dir
->i_mtime
=
10150 new_dir
->i_ctime
= new_dir
->i_mtime
=
10151 old_inode
->i_ctime
= current_time(old_dir
);
10153 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
10154 btrfs_record_unlink_dir(trans
, BTRFS_I(old_dir
),
10155 BTRFS_I(old_inode
), 1);
10157 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
10158 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
10159 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
10160 old_dentry
->d_name
.name
,
10161 old_dentry
->d_name
.len
);
10163 ret
= __btrfs_unlink_inode(trans
, root
, BTRFS_I(old_dir
),
10164 BTRFS_I(d_inode(old_dentry
)),
10165 old_dentry
->d_name
.name
,
10166 old_dentry
->d_name
.len
);
10168 ret
= btrfs_update_inode(trans
, root
, old_inode
);
10171 btrfs_abort_transaction(trans
, ret
);
10176 inode_inc_iversion(new_inode
);
10177 new_inode
->i_ctime
= current_time(new_inode
);
10178 if (unlikely(btrfs_ino(BTRFS_I(new_inode
)) ==
10179 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
10180 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
10181 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
10183 new_dentry
->d_name
.name
,
10184 new_dentry
->d_name
.len
);
10185 BUG_ON(new_inode
->i_nlink
== 0);
10187 ret
= btrfs_unlink_inode(trans
, dest
, BTRFS_I(new_dir
),
10188 BTRFS_I(d_inode(new_dentry
)),
10189 new_dentry
->d_name
.name
,
10190 new_dentry
->d_name
.len
);
10192 if (!ret
&& new_inode
->i_nlink
== 0)
10193 ret
= btrfs_orphan_add(trans
,
10194 BTRFS_I(d_inode(new_dentry
)));
10196 btrfs_abort_transaction(trans
, ret
);
10201 ret
= btrfs_add_link(trans
, BTRFS_I(new_dir
), BTRFS_I(old_inode
),
10202 new_dentry
->d_name
.name
,
10203 new_dentry
->d_name
.len
, 0, index
);
10205 btrfs_abort_transaction(trans
, ret
);
10209 if (old_inode
->i_nlink
== 1)
10210 BTRFS_I(old_inode
)->dir_index
= index
;
10213 struct dentry
*parent
= new_dentry
->d_parent
;
10215 btrfs_log_new_name(trans
, BTRFS_I(old_inode
), BTRFS_I(old_dir
),
10217 btrfs_end_log_trans(root
);
10218 log_pinned
= false;
10221 if (flags
& RENAME_WHITEOUT
) {
10222 ret
= btrfs_whiteout_for_rename(trans
, root
, old_dir
,
10226 btrfs_abort_transaction(trans
, ret
);
10232 * If we have pinned the log and an error happened, we unpin tasks
10233 * trying to sync the log and force them to fallback to a transaction
10234 * commit if the log currently contains any of the inodes involved in
10235 * this rename operation (to ensure we do not persist a log with an
10236 * inconsistent state for any of these inodes or leading to any
10237 * inconsistencies when replayed). If the transaction was aborted, the
10238 * abortion reason is propagated to userspace when attempting to commit
10239 * the transaction. If the log does not contain any of these inodes, we
10240 * allow the tasks to sync it.
10242 if (ret
&& log_pinned
) {
10243 if (btrfs_inode_in_log(BTRFS_I(old_dir
), fs_info
->generation
) ||
10244 btrfs_inode_in_log(BTRFS_I(new_dir
), fs_info
->generation
) ||
10245 btrfs_inode_in_log(BTRFS_I(old_inode
), fs_info
->generation
) ||
10247 btrfs_inode_in_log(BTRFS_I(new_inode
), fs_info
->generation
)))
10248 btrfs_set_log_full_commit(fs_info
, trans
);
10250 btrfs_end_log_trans(root
);
10251 log_pinned
= false;
10253 btrfs_end_transaction(trans
);
10255 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
10256 up_read(&fs_info
->subvol_sem
);
10261 static int btrfs_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
10262 struct inode
*new_dir
, struct dentry
*new_dentry
,
10263 unsigned int flags
)
10265 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
10268 if (flags
& RENAME_EXCHANGE
)
10269 return btrfs_rename_exchange(old_dir
, old_dentry
, new_dir
,
10272 return btrfs_rename(old_dir
, old_dentry
, new_dir
, new_dentry
, flags
);
10275 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
10277 struct btrfs_delalloc_work
*delalloc_work
;
10278 struct inode
*inode
;
10280 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
10282 inode
= delalloc_work
->inode
;
10283 filemap_flush(inode
->i_mapping
);
10284 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
10285 &BTRFS_I(inode
)->runtime_flags
))
10286 filemap_flush(inode
->i_mapping
);
10288 if (delalloc_work
->delay_iput
)
10289 btrfs_add_delayed_iput(inode
);
10292 complete(&delalloc_work
->completion
);
10295 struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
,
10298 struct btrfs_delalloc_work
*work
;
10300 work
= kmalloc(sizeof(*work
), GFP_NOFS
);
10304 init_completion(&work
->completion
);
10305 INIT_LIST_HEAD(&work
->list
);
10306 work
->inode
= inode
;
10307 work
->delay_iput
= delay_iput
;
10308 WARN_ON_ONCE(!inode
);
10309 btrfs_init_work(&work
->work
, btrfs_flush_delalloc_helper
,
10310 btrfs_run_delalloc_work
, NULL
, NULL
);
10315 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work
*work
)
10317 wait_for_completion(&work
->completion
);
10322 * some fairly slow code that needs optimization. This walks the list
10323 * of all the inodes with pending delalloc and forces them to disk.
10325 static int __start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
,
10328 struct btrfs_inode
*binode
;
10329 struct inode
*inode
;
10330 struct btrfs_delalloc_work
*work
, *next
;
10331 struct list_head works
;
10332 struct list_head splice
;
10335 INIT_LIST_HEAD(&works
);
10336 INIT_LIST_HEAD(&splice
);
10338 mutex_lock(&root
->delalloc_mutex
);
10339 spin_lock(&root
->delalloc_lock
);
10340 list_splice_init(&root
->delalloc_inodes
, &splice
);
10341 while (!list_empty(&splice
)) {
10342 binode
= list_entry(splice
.next
, struct btrfs_inode
,
10345 list_move_tail(&binode
->delalloc_inodes
,
10346 &root
->delalloc_inodes
);
10347 inode
= igrab(&binode
->vfs_inode
);
10349 cond_resched_lock(&root
->delalloc_lock
);
10352 spin_unlock(&root
->delalloc_lock
);
10354 work
= btrfs_alloc_delalloc_work(inode
, delay_iput
);
10357 btrfs_add_delayed_iput(inode
);
10363 list_add_tail(&work
->list
, &works
);
10364 btrfs_queue_work(root
->fs_info
->flush_workers
,
10367 if (nr
!= -1 && ret
>= nr
)
10370 spin_lock(&root
->delalloc_lock
);
10372 spin_unlock(&root
->delalloc_lock
);
10375 list_for_each_entry_safe(work
, next
, &works
, list
) {
10376 list_del_init(&work
->list
);
10377 btrfs_wait_and_free_delalloc_work(work
);
10380 if (!list_empty_careful(&splice
)) {
10381 spin_lock(&root
->delalloc_lock
);
10382 list_splice_tail(&splice
, &root
->delalloc_inodes
);
10383 spin_unlock(&root
->delalloc_lock
);
10385 mutex_unlock(&root
->delalloc_mutex
);
10389 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
10391 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
10394 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
10397 ret
= __start_delalloc_inodes(root
, delay_iput
, -1);
10401 * the filemap_flush will queue IO into the worker threads, but
10402 * we have to make sure the IO is actually started and that
10403 * ordered extents get created before we return
10405 atomic_inc(&fs_info
->async_submit_draining
);
10406 while (atomic_read(&fs_info
->nr_async_submits
) ||
10407 atomic_read(&fs_info
->async_delalloc_pages
)) {
10408 wait_event(fs_info
->async_submit_wait
,
10409 (atomic_read(&fs_info
->nr_async_submits
) == 0 &&
10410 atomic_read(&fs_info
->async_delalloc_pages
) == 0));
10412 atomic_dec(&fs_info
->async_submit_draining
);
10416 int btrfs_start_delalloc_roots(struct btrfs_fs_info
*fs_info
, int delay_iput
,
10419 struct btrfs_root
*root
;
10420 struct list_head splice
;
10423 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
10426 INIT_LIST_HEAD(&splice
);
10428 mutex_lock(&fs_info
->delalloc_root_mutex
);
10429 spin_lock(&fs_info
->delalloc_root_lock
);
10430 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
10431 while (!list_empty(&splice
) && nr
) {
10432 root
= list_first_entry(&splice
, struct btrfs_root
,
10434 root
= btrfs_grab_fs_root(root
);
10436 list_move_tail(&root
->delalloc_root
,
10437 &fs_info
->delalloc_roots
);
10438 spin_unlock(&fs_info
->delalloc_root_lock
);
10440 ret
= __start_delalloc_inodes(root
, delay_iput
, nr
);
10441 btrfs_put_fs_root(root
);
10449 spin_lock(&fs_info
->delalloc_root_lock
);
10451 spin_unlock(&fs_info
->delalloc_root_lock
);
10454 atomic_inc(&fs_info
->async_submit_draining
);
10455 while (atomic_read(&fs_info
->nr_async_submits
) ||
10456 atomic_read(&fs_info
->async_delalloc_pages
)) {
10457 wait_event(fs_info
->async_submit_wait
,
10458 (atomic_read(&fs_info
->nr_async_submits
) == 0 &&
10459 atomic_read(&fs_info
->async_delalloc_pages
) == 0));
10461 atomic_dec(&fs_info
->async_submit_draining
);
10463 if (!list_empty_careful(&splice
)) {
10464 spin_lock(&fs_info
->delalloc_root_lock
);
10465 list_splice_tail(&splice
, &fs_info
->delalloc_roots
);
10466 spin_unlock(&fs_info
->delalloc_root_lock
);
10468 mutex_unlock(&fs_info
->delalloc_root_mutex
);
10472 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
10473 const char *symname
)
10475 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
10476 struct btrfs_trans_handle
*trans
;
10477 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
10478 struct btrfs_path
*path
;
10479 struct btrfs_key key
;
10480 struct inode
*inode
= NULL
;
10482 int drop_inode
= 0;
10488 struct btrfs_file_extent_item
*ei
;
10489 struct extent_buffer
*leaf
;
10491 name_len
= strlen(symname
);
10492 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(fs_info
))
10493 return -ENAMETOOLONG
;
10496 * 2 items for inode item and ref
10497 * 2 items for dir items
10498 * 1 item for updating parent inode item
10499 * 1 item for the inline extent item
10500 * 1 item for xattr if selinux is on
10502 trans
= btrfs_start_transaction(root
, 7);
10504 return PTR_ERR(trans
);
10506 err
= btrfs_find_free_ino(root
, &objectid
);
10510 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
10511 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)),
10512 objectid
, S_IFLNK
|S_IRWXUGO
, &index
);
10513 if (IS_ERR(inode
)) {
10514 err
= PTR_ERR(inode
);
10519 * If the active LSM wants to access the inode during
10520 * d_instantiate it needs these. Smack checks to see
10521 * if the filesystem supports xattrs by looking at the
10524 inode
->i_fop
= &btrfs_file_operations
;
10525 inode
->i_op
= &btrfs_file_inode_operations
;
10526 inode
->i_mapping
->a_ops
= &btrfs_aops
;
10527 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
10529 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
10531 goto out_unlock_inode
;
10533 path
= btrfs_alloc_path();
10536 goto out_unlock_inode
;
10538 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
10540 key
.type
= BTRFS_EXTENT_DATA_KEY
;
10541 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
10542 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
10545 btrfs_free_path(path
);
10546 goto out_unlock_inode
;
10548 leaf
= path
->nodes
[0];
10549 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
10550 struct btrfs_file_extent_item
);
10551 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
10552 btrfs_set_file_extent_type(leaf
, ei
,
10553 BTRFS_FILE_EXTENT_INLINE
);
10554 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
10555 btrfs_set_file_extent_compression(leaf
, ei
, 0);
10556 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
10557 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
10559 ptr
= btrfs_file_extent_inline_start(ei
);
10560 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
10561 btrfs_mark_buffer_dirty(leaf
);
10562 btrfs_free_path(path
);
10564 inode
->i_op
= &btrfs_symlink_inode_operations
;
10565 inode_nohighmem(inode
);
10566 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
10567 inode_set_bytes(inode
, name_len
);
10568 btrfs_i_size_write(BTRFS_I(inode
), name_len
);
10569 err
= btrfs_update_inode(trans
, root
, inode
);
10571 * Last step, add directory indexes for our symlink inode. This is the
10572 * last step to avoid extra cleanup of these indexes if an error happens
10576 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
,
10577 BTRFS_I(inode
), 0, index
);
10580 goto out_unlock_inode
;
10583 d_instantiate_new(dentry
, inode
);
10586 btrfs_end_transaction(trans
);
10588 inode_dec_link_count(inode
);
10591 btrfs_btree_balance_dirty(fs_info
);
10596 unlock_new_inode(inode
);
10600 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
10601 u64 start
, u64 num_bytes
, u64 min_size
,
10602 loff_t actual_len
, u64
*alloc_hint
,
10603 struct btrfs_trans_handle
*trans
)
10605 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
10606 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
10607 struct extent_map
*em
;
10608 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
10609 struct btrfs_key ins
;
10610 u64 cur_offset
= start
;
10613 u64 last_alloc
= (u64
)-1;
10615 bool own_trans
= true;
10616 u64 end
= start
+ num_bytes
- 1;
10620 while (num_bytes
> 0) {
10622 trans
= btrfs_start_transaction(root
, 3);
10623 if (IS_ERR(trans
)) {
10624 ret
= PTR_ERR(trans
);
10629 cur_bytes
= min_t(u64
, num_bytes
, SZ_256M
);
10630 cur_bytes
= max(cur_bytes
, min_size
);
10632 * If we are severely fragmented we could end up with really
10633 * small allocations, so if the allocator is returning small
10634 * chunks lets make its job easier by only searching for those
10637 cur_bytes
= min(cur_bytes
, last_alloc
);
10638 ret
= btrfs_reserve_extent(root
, cur_bytes
, cur_bytes
,
10639 min_size
, 0, *alloc_hint
, &ins
, 1, 0);
10642 btrfs_end_transaction(trans
);
10645 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
10647 last_alloc
= ins
.offset
;
10648 ret
= insert_reserved_file_extent(trans
, inode
,
10649 cur_offset
, ins
.objectid
,
10650 ins
.offset
, ins
.offset
,
10651 ins
.offset
, 0, 0, 0,
10652 BTRFS_FILE_EXTENT_PREALLOC
);
10654 btrfs_free_reserved_extent(fs_info
, ins
.objectid
,
10656 btrfs_abort_transaction(trans
, ret
);
10658 btrfs_end_transaction(trans
);
10662 btrfs_drop_extent_cache(BTRFS_I(inode
), cur_offset
,
10663 cur_offset
+ ins
.offset
-1, 0);
10665 em
= alloc_extent_map();
10667 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
10668 &BTRFS_I(inode
)->runtime_flags
);
10672 em
->start
= cur_offset
;
10673 em
->orig_start
= cur_offset
;
10674 em
->len
= ins
.offset
;
10675 em
->block_start
= ins
.objectid
;
10676 em
->block_len
= ins
.offset
;
10677 em
->orig_block_len
= ins
.offset
;
10678 em
->ram_bytes
= ins
.offset
;
10679 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
10680 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
10681 em
->generation
= trans
->transid
;
10684 write_lock(&em_tree
->lock
);
10685 ret
= add_extent_mapping(em_tree
, em
, 1);
10686 write_unlock(&em_tree
->lock
);
10687 if (ret
!= -EEXIST
)
10689 btrfs_drop_extent_cache(BTRFS_I(inode
), cur_offset
,
10690 cur_offset
+ ins
.offset
- 1,
10693 free_extent_map(em
);
10695 num_bytes
-= ins
.offset
;
10696 cur_offset
+= ins
.offset
;
10697 *alloc_hint
= ins
.objectid
+ ins
.offset
;
10699 inode_inc_iversion(inode
);
10700 inode
->i_ctime
= current_time(inode
);
10701 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
10702 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
10703 (actual_len
> inode
->i_size
) &&
10704 (cur_offset
> inode
->i_size
)) {
10705 if (cur_offset
> actual_len
)
10706 i_size
= actual_len
;
10708 i_size
= cur_offset
;
10709 i_size_write(inode
, i_size
);
10710 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
10713 ret
= btrfs_update_inode(trans
, root
, inode
);
10716 btrfs_abort_transaction(trans
, ret
);
10718 btrfs_end_transaction(trans
);
10723 btrfs_end_transaction(trans
);
10725 if (cur_offset
< end
)
10726 btrfs_free_reserved_data_space(inode
, NULL
, cur_offset
,
10727 end
- cur_offset
+ 1);
10731 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
10732 u64 start
, u64 num_bytes
, u64 min_size
,
10733 loff_t actual_len
, u64
*alloc_hint
)
10735 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
10736 min_size
, actual_len
, alloc_hint
,
10740 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
10741 struct btrfs_trans_handle
*trans
, int mode
,
10742 u64 start
, u64 num_bytes
, u64 min_size
,
10743 loff_t actual_len
, u64
*alloc_hint
)
10745 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
10746 min_size
, actual_len
, alloc_hint
, trans
);
10749 static int btrfs_set_page_dirty(struct page
*page
)
10751 return __set_page_dirty_nobuffers(page
);
10754 static int btrfs_permission(struct inode
*inode
, int mask
)
10756 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
10757 umode_t mode
= inode
->i_mode
;
10759 if (mask
& MAY_WRITE
&&
10760 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
10761 if (btrfs_root_readonly(root
))
10763 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
10766 return generic_permission(inode
, mask
);
10769 static int btrfs_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
10771 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
10772 struct btrfs_trans_handle
*trans
;
10773 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
10774 struct inode
*inode
= NULL
;
10780 * 5 units required for adding orphan entry
10782 trans
= btrfs_start_transaction(root
, 5);
10784 return PTR_ERR(trans
);
10786 ret
= btrfs_find_free_ino(root
, &objectid
);
10790 inode
= btrfs_new_inode(trans
, root
, dir
, NULL
, 0,
10791 btrfs_ino(BTRFS_I(dir
)), objectid
, mode
, &index
);
10792 if (IS_ERR(inode
)) {
10793 ret
= PTR_ERR(inode
);
10798 inode
->i_fop
= &btrfs_file_operations
;
10799 inode
->i_op
= &btrfs_file_inode_operations
;
10801 inode
->i_mapping
->a_ops
= &btrfs_aops
;
10802 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
10804 ret
= btrfs_init_inode_security(trans
, inode
, dir
, NULL
);
10808 ret
= btrfs_update_inode(trans
, root
, inode
);
10811 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
10816 * We set number of links to 0 in btrfs_new_inode(), and here we set
10817 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10820 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10822 set_nlink(inode
, 1);
10823 unlock_new_inode(inode
);
10824 d_tmpfile(dentry
, inode
);
10825 mark_inode_dirty(inode
);
10828 btrfs_end_transaction(trans
);
10831 btrfs_balance_delayed_items(fs_info
);
10832 btrfs_btree_balance_dirty(fs_info
);
10836 unlock_new_inode(inode
);
10841 __attribute__((const))
10842 static int btrfs_readpage_io_failed_hook(struct page
*page
, int failed_mirror
)
10847 static struct btrfs_fs_info
*iotree_fs_info(void *private_data
)
10849 struct inode
*inode
= private_data
;
10850 return btrfs_sb(inode
->i_sb
);
10853 static void btrfs_check_extent_io_range(void *private_data
, const char *caller
,
10854 u64 start
, u64 end
)
10856 struct inode
*inode
= private_data
;
10859 isize
= i_size_read(inode
);
10860 if (end
>= PAGE_SIZE
&& (end
% 2) == 0 && end
!= isize
- 1) {
10861 btrfs_debug_rl(BTRFS_I(inode
)->root
->fs_info
,
10862 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10863 caller
, btrfs_ino(BTRFS_I(inode
)), isize
, start
, end
);
10867 void btrfs_set_range_writeback(void *private_data
, u64 start
, u64 end
)
10869 struct inode
*inode
= private_data
;
10870 unsigned long index
= start
>> PAGE_SHIFT
;
10871 unsigned long end_index
= end
>> PAGE_SHIFT
;
10874 while (index
<= end_index
) {
10875 page
= find_get_page(inode
->i_mapping
, index
);
10876 ASSERT(page
); /* Pages should be in the extent_io_tree */
10877 set_page_writeback(page
);
10883 static const struct inode_operations btrfs_dir_inode_operations
= {
10884 .getattr
= btrfs_getattr
,
10885 .lookup
= btrfs_lookup
,
10886 .create
= btrfs_create
,
10887 .unlink
= btrfs_unlink
,
10888 .link
= btrfs_link
,
10889 .mkdir
= btrfs_mkdir
,
10890 .rmdir
= btrfs_rmdir
,
10891 .rename
= btrfs_rename2
,
10892 .symlink
= btrfs_symlink
,
10893 .setattr
= btrfs_setattr
,
10894 .mknod
= btrfs_mknod
,
10895 .listxattr
= btrfs_listxattr
,
10896 .permission
= btrfs_permission
,
10897 .get_acl
= btrfs_get_acl
,
10898 .set_acl
= btrfs_set_acl
,
10899 .update_time
= btrfs_update_time
,
10900 .tmpfile
= btrfs_tmpfile
,
10902 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
10903 .lookup
= btrfs_lookup
,
10904 .permission
= btrfs_permission
,
10905 .update_time
= btrfs_update_time
,
10908 static const struct file_operations btrfs_dir_file_operations
= {
10909 .llseek
= generic_file_llseek
,
10910 .read
= generic_read_dir
,
10911 .iterate_shared
= btrfs_real_readdir
,
10912 .open
= btrfs_opendir
,
10913 .unlocked_ioctl
= btrfs_ioctl
,
10914 #ifdef CONFIG_COMPAT
10915 .compat_ioctl
= btrfs_compat_ioctl
,
10917 .release
= btrfs_release_file
,
10918 .fsync
= btrfs_sync_file
,
10921 static const struct extent_io_ops btrfs_extent_io_ops
= {
10922 /* mandatory callbacks */
10923 .submit_bio_hook
= btrfs_submit_bio_hook
,
10924 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
10925 .merge_bio_hook
= btrfs_merge_bio_hook
,
10926 .readpage_io_failed_hook
= btrfs_readpage_io_failed_hook
,
10927 .tree_fs_info
= iotree_fs_info
,
10928 .set_range_writeback
= btrfs_set_range_writeback
,
10930 /* optional callbacks */
10931 .fill_delalloc
= run_delalloc_range
,
10932 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
10933 .writepage_start_hook
= btrfs_writepage_start_hook
,
10934 .set_bit_hook
= btrfs_set_bit_hook
,
10935 .clear_bit_hook
= btrfs_clear_bit_hook
,
10936 .merge_extent_hook
= btrfs_merge_extent_hook
,
10937 .split_extent_hook
= btrfs_split_extent_hook
,
10938 .check_extent_io_range
= btrfs_check_extent_io_range
,
10942 * btrfs doesn't support the bmap operation because swapfiles
10943 * use bmap to make a mapping of extents in the file. They assume
10944 * these extents won't change over the life of the file and they
10945 * use the bmap result to do IO directly to the drive.
10947 * the btrfs bmap call would return logical addresses that aren't
10948 * suitable for IO and they also will change frequently as COW
10949 * operations happen. So, swapfile + btrfs == corruption.
10951 * For now we're avoiding this by dropping bmap.
10953 static const struct address_space_operations btrfs_aops
= {
10954 .readpage
= btrfs_readpage
,
10955 .writepage
= btrfs_writepage
,
10956 .writepages
= btrfs_writepages
,
10957 .readpages
= btrfs_readpages
,
10958 .direct_IO
= btrfs_direct_IO
,
10959 .invalidatepage
= btrfs_invalidatepage
,
10960 .releasepage
= btrfs_releasepage
,
10961 .set_page_dirty
= btrfs_set_page_dirty
,
10962 .error_remove_page
= generic_error_remove_page
,
10965 static const struct address_space_operations btrfs_symlink_aops
= {
10966 .readpage
= btrfs_readpage
,
10967 .writepage
= btrfs_writepage
,
10968 .invalidatepage
= btrfs_invalidatepage
,
10969 .releasepage
= btrfs_releasepage
,
10972 static const struct inode_operations btrfs_file_inode_operations
= {
10973 .getattr
= btrfs_getattr
,
10974 .setattr
= btrfs_setattr
,
10975 .listxattr
= btrfs_listxattr
,
10976 .permission
= btrfs_permission
,
10977 .fiemap
= btrfs_fiemap
,
10978 .get_acl
= btrfs_get_acl
,
10979 .set_acl
= btrfs_set_acl
,
10980 .update_time
= btrfs_update_time
,
10982 static const struct inode_operations btrfs_special_inode_operations
= {
10983 .getattr
= btrfs_getattr
,
10984 .setattr
= btrfs_setattr
,
10985 .permission
= btrfs_permission
,
10986 .listxattr
= btrfs_listxattr
,
10987 .get_acl
= btrfs_get_acl
,
10988 .set_acl
= btrfs_set_acl
,
10989 .update_time
= btrfs_update_time
,
10991 static const struct inode_operations btrfs_symlink_inode_operations
= {
10992 .get_link
= page_get_link
,
10993 .getattr
= btrfs_getattr
,
10994 .setattr
= btrfs_setattr
,
10995 .permission
= btrfs_permission
,
10996 .listxattr
= btrfs_listxattr
,
10997 .update_time
= btrfs_update_time
,
11000 const struct dentry_operations btrfs_dentry_operations
= {
11001 .d_delete
= btrfs_dentry_delete
,
11002 .d_release
= btrfs_dentry_release
,