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 <linux/magic.h>
46 #include <asm/unaligned.h>
49 #include "transaction.h"
50 #include "btrfs_inode.h"
51 #include "print-tree.h"
52 #include "ordered-data.h"
56 #include "compression.h"
58 #include "free-space-cache.h"
59 #include "inode-map.h"
66 struct btrfs_iget_args
{
67 struct btrfs_key
*location
;
68 struct btrfs_root
*root
;
71 struct btrfs_dio_data
{
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
= &BTRFS_I(inode
)->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_drop_extent_cache(BTRFS_I(inode
), start
, aligned_end
- 1, 0);
355 * Don't forget to free the reserved space, as for inlined extent
356 * it won't count as data extent, free them directly here.
357 * And at reserve time, it's always aligned to page size, so
358 * just free one page here.
360 btrfs_qgroup_free_data(inode
, NULL
, 0, PAGE_SIZE
);
361 btrfs_free_path(path
);
362 btrfs_end_transaction(trans
);
366 struct async_extent
{
371 unsigned long nr_pages
;
373 struct list_head list
;
378 struct btrfs_root
*root
;
379 struct page
*locked_page
;
382 unsigned int write_flags
;
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
;
462 u64 blocksize
= fs_info
->sectorsize
;
464 u64 isize
= i_size_read(inode
);
466 struct page
**pages
= NULL
;
467 unsigned long nr_pages
;
468 unsigned long total_compressed
= 0;
469 unsigned long total_in
= 0;
472 int compress_type
= fs_info
->compress_type
;
475 inode_should_defrag(BTRFS_I(inode
), start
, end
, end
- start
+ 1,
478 actual_end
= min_t(u64
, isize
, end
+ 1);
481 nr_pages
= (end
>> PAGE_SHIFT
) - (start
>> PAGE_SHIFT
) + 1;
482 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED
% PAGE_SIZE
) != 0);
483 nr_pages
= min_t(unsigned long, nr_pages
,
484 BTRFS_MAX_COMPRESSED
/ PAGE_SIZE
);
487 * we don't want to send crud past the end of i_size through
488 * compression, that's just a waste of CPU time. So, if the
489 * end of the file is before the start of our current
490 * requested range of bytes, we bail out to the uncompressed
491 * cleanup code that can deal with all of this.
493 * It isn't really the fastest way to fix things, but this is a
494 * very uncommon corner.
496 if (actual_end
<= start
)
497 goto cleanup_and_bail_uncompressed
;
499 total_compressed
= actual_end
- start
;
502 * skip compression for a small file range(<=blocksize) that
503 * isn't an inline extent, since it doesn't save disk space at all.
505 if (total_compressed
<= blocksize
&&
506 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
507 goto cleanup_and_bail_uncompressed
;
509 total_compressed
= min_t(unsigned long, total_compressed
,
510 BTRFS_MAX_UNCOMPRESSED
);
515 * we do compression for mount -o compress and when the
516 * inode has not been flagged as nocompress. This flag can
517 * change at any time if we discover bad compression ratios.
519 if (inode_need_compress(inode
, start
, end
)) {
521 pages
= kcalloc(nr_pages
, sizeof(struct page
*), GFP_NOFS
);
523 /* just bail out to the uncompressed code */
527 if (BTRFS_I(inode
)->defrag_compress
)
528 compress_type
= BTRFS_I(inode
)->defrag_compress
;
529 else if (BTRFS_I(inode
)->prop_compress
)
530 compress_type
= BTRFS_I(inode
)->prop_compress
;
533 * we need to call clear_page_dirty_for_io on each
534 * page in the range. Otherwise applications with the file
535 * mmap'd can wander in and change the page contents while
536 * we are compressing them.
538 * If the compression fails for any reason, we set the pages
539 * dirty again later on.
541 extent_range_clear_dirty_for_io(inode
, start
, end
);
544 /* Compression level is applied here and only here */
545 ret
= btrfs_compress_pages(
546 compress_type
| (fs_info
->compress_level
<< 4),
547 inode
->i_mapping
, start
,
554 unsigned long offset
= total_compressed
&
556 struct page
*page
= pages
[nr_pages
- 1];
559 /* zero the tail end of the last page, we might be
560 * sending it down to disk
563 kaddr
= kmap_atomic(page
);
564 memset(kaddr
+ offset
, 0,
566 kunmap_atomic(kaddr
);
573 /* lets try to make an inline extent */
574 if (ret
|| total_in
< actual_end
) {
575 /* we didn't compress the entire range, try
576 * to make an uncompressed inline extent.
578 ret
= cow_file_range_inline(root
, inode
, start
, end
,
579 0, BTRFS_COMPRESS_NONE
, NULL
);
581 /* try making a compressed inline extent */
582 ret
= cow_file_range_inline(root
, inode
, start
, end
,
584 compress_type
, pages
);
587 unsigned long clear_flags
= EXTENT_DELALLOC
|
588 EXTENT_DELALLOC_NEW
| EXTENT_DEFRAG
|
589 EXTENT_DO_ACCOUNTING
;
590 unsigned long page_error_op
;
592 page_error_op
= ret
< 0 ? PAGE_SET_ERROR
: 0;
595 * inline extent creation worked or returned error,
596 * we don't need to create any more async work items.
597 * Unlock and free up our temp pages.
599 * We use DO_ACCOUNTING here because we need the
600 * delalloc_release_metadata to be done _after_ we drop
601 * our outstanding extent for clearing delalloc for this
604 extent_clear_unlock_delalloc(inode
, start
, end
, end
,
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
) {
633 * The async work queues will take care of doing actual
634 * allocation on disk for these compressed pages, and
635 * will submit them to the elevator.
637 add_async_extent(async_cow
, start
, total_in
,
638 total_compressed
, pages
, nr_pages
,
641 if (start
+ total_in
< end
) {
652 * the compression code ran but failed to make things smaller,
653 * free any pages it allocated and our page pointer array
655 for (i
= 0; i
< nr_pages
; i
++) {
656 WARN_ON(pages
[i
]->mapping
);
661 total_compressed
= 0;
664 /* flag the file so we don't compress in the future */
665 if (!btrfs_test_opt(fs_info
, FORCE_COMPRESS
) &&
666 !(BTRFS_I(inode
)->prop_compress
)) {
667 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
670 cleanup_and_bail_uncompressed
:
672 * No compression, but we still need to write the pages in the file
673 * we've been given so far. redirty the locked page if it corresponds
674 * to our extent and set things up for the async work queue to run
675 * cow_file_range to do the normal delalloc dance.
677 if (page_offset(locked_page
) >= start
&&
678 page_offset(locked_page
) <= end
)
679 __set_page_dirty_nobuffers(locked_page
);
680 /* unlocked later on in the async handlers */
683 extent_range_redirty_for_io(inode
, start
, end
);
684 add_async_extent(async_cow
, start
, end
- start
+ 1, 0, NULL
, 0,
685 BTRFS_COMPRESS_NONE
);
691 for (i
= 0; i
< nr_pages
; i
++) {
692 WARN_ON(pages
[i
]->mapping
);
698 static void free_async_extent_pages(struct async_extent
*async_extent
)
702 if (!async_extent
->pages
)
705 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
706 WARN_ON(async_extent
->pages
[i
]->mapping
);
707 put_page(async_extent
->pages
[i
]);
709 kfree(async_extent
->pages
);
710 async_extent
->nr_pages
= 0;
711 async_extent
->pages
= NULL
;
715 * phase two of compressed writeback. This is the ordered portion
716 * of the code, which only gets called in the order the work was
717 * queued. We walk all the async extents created by compress_file_range
718 * and send them down to the disk.
720 static noinline
void submit_compressed_extents(struct inode
*inode
,
721 struct async_cow
*async_cow
)
723 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
724 struct async_extent
*async_extent
;
726 struct btrfs_key ins
;
727 struct extent_map
*em
;
728 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
729 struct extent_io_tree
*io_tree
;
733 while (!list_empty(&async_cow
->extents
)) {
734 async_extent
= list_entry(async_cow
->extents
.next
,
735 struct async_extent
, list
);
736 list_del(&async_extent
->list
);
738 io_tree
= &BTRFS_I(inode
)->io_tree
;
741 /* did the compression code fall back to uncompressed IO? */
742 if (!async_extent
->pages
) {
743 int page_started
= 0;
744 unsigned long nr_written
= 0;
746 lock_extent(io_tree
, async_extent
->start
,
747 async_extent
->start
+
748 async_extent
->ram_size
- 1);
750 /* allocate blocks */
751 ret
= cow_file_range(inode
, async_cow
->locked_page
,
753 async_extent
->start
+
754 async_extent
->ram_size
- 1,
755 async_extent
->start
+
756 async_extent
->ram_size
- 1,
757 &page_started
, &nr_written
, 0,
763 * if page_started, cow_file_range inserted an
764 * inline extent and took care of all the unlocking
765 * and IO for us. Otherwise, we need to submit
766 * all those pages down to the drive.
768 if (!page_started
&& !ret
)
769 extent_write_locked_range(io_tree
,
770 inode
, async_extent
->start
,
771 async_extent
->start
+
772 async_extent
->ram_size
- 1,
776 unlock_page(async_cow
->locked_page
);
782 lock_extent(io_tree
, async_extent
->start
,
783 async_extent
->start
+ async_extent
->ram_size
- 1);
785 ret
= btrfs_reserve_extent(root
, async_extent
->ram_size
,
786 async_extent
->compressed_size
,
787 async_extent
->compressed_size
,
788 0, alloc_hint
, &ins
, 1, 1);
790 free_async_extent_pages(async_extent
);
792 if (ret
== -ENOSPC
) {
793 unlock_extent(io_tree
, async_extent
->start
,
794 async_extent
->start
+
795 async_extent
->ram_size
- 1);
798 * we need to redirty the pages if we decide to
799 * fallback to uncompressed IO, otherwise we
800 * will not submit these pages down to lower
803 extent_range_redirty_for_io(inode
,
805 async_extent
->start
+
806 async_extent
->ram_size
- 1);
813 * here we're doing allocation and writeback of the
816 em
= create_io_em(inode
, async_extent
->start
,
817 async_extent
->ram_size
, /* len */
818 async_extent
->start
, /* orig_start */
819 ins
.objectid
, /* block_start */
820 ins
.offset
, /* block_len */
821 ins
.offset
, /* orig_block_len */
822 async_extent
->ram_size
, /* ram_bytes */
823 async_extent
->compress_type
,
824 BTRFS_ORDERED_COMPRESSED
);
826 /* ret value is not necessary due to void function */
827 goto out_free_reserve
;
830 ret
= btrfs_add_ordered_extent_compress(inode
,
833 async_extent
->ram_size
,
835 BTRFS_ORDERED_COMPRESSED
,
836 async_extent
->compress_type
);
838 btrfs_drop_extent_cache(BTRFS_I(inode
),
840 async_extent
->start
+
841 async_extent
->ram_size
- 1, 0);
842 goto out_free_reserve
;
844 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
847 * clear dirty, set writeback and unlock the pages.
849 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
850 async_extent
->start
+
851 async_extent
->ram_size
- 1,
852 async_extent
->start
+
853 async_extent
->ram_size
- 1,
854 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
,
855 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
857 if (btrfs_submit_compressed_write(inode
,
859 async_extent
->ram_size
,
861 ins
.offset
, async_extent
->pages
,
862 async_extent
->nr_pages
,
863 async_cow
->write_flags
)) {
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
);
988 * We use DO_ACCOUNTING here because we need the
989 * delalloc_release_metadata to be run _after_ we drop
990 * our outstanding extent for clearing delalloc for this
993 extent_clear_unlock_delalloc(inode
, start
, end
,
995 EXTENT_LOCKED
| EXTENT_DELALLOC
|
996 EXTENT_DELALLOC_NEW
| EXTENT_DEFRAG
|
997 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
998 PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
1000 *nr_written
= *nr_written
+
1001 (end
- start
+ PAGE_SIZE
) / PAGE_SIZE
;
1004 } else if (ret
< 0) {
1009 BUG_ON(disk_num_bytes
>
1010 btrfs_super_total_bytes(fs_info
->super_copy
));
1012 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
1013 btrfs_drop_extent_cache(BTRFS_I(inode
), start
,
1014 start
+ num_bytes
- 1, 0);
1016 while (disk_num_bytes
> 0) {
1017 cur_alloc_size
= disk_num_bytes
;
1018 ret
= btrfs_reserve_extent(root
, cur_alloc_size
, cur_alloc_size
,
1019 fs_info
->sectorsize
, 0, alloc_hint
,
1023 cur_alloc_size
= ins
.offset
;
1024 extent_reserved
= true;
1026 ram_size
= ins
.offset
;
1027 em
= create_io_em(inode
, start
, ins
.offset
, /* len */
1028 start
, /* orig_start */
1029 ins
.objectid
, /* block_start */
1030 ins
.offset
, /* block_len */
1031 ins
.offset
, /* orig_block_len */
1032 ram_size
, /* ram_bytes */
1033 BTRFS_COMPRESS_NONE
, /* compress_type */
1034 BTRFS_ORDERED_REGULAR
/* type */);
1037 free_extent_map(em
);
1039 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
1040 ram_size
, cur_alloc_size
, 0);
1042 goto out_drop_extent_cache
;
1044 if (root
->root_key
.objectid
==
1045 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1046 ret
= btrfs_reloc_clone_csums(inode
, start
,
1049 * Only drop cache here, and process as normal.
1051 * We must not allow extent_clear_unlock_delalloc()
1052 * at out_unlock label to free meta of this ordered
1053 * extent, as its meta should be freed by
1054 * btrfs_finish_ordered_io().
1056 * So we must continue until @start is increased to
1057 * skip current ordered extent.
1060 btrfs_drop_extent_cache(BTRFS_I(inode
), start
,
1061 start
+ ram_size
- 1, 0);
1064 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
1066 /* we're not doing compressed IO, don't unlock the first
1067 * page (which the caller expects to stay locked), don't
1068 * clear any dirty bits and don't set any writeback bits
1070 * Do set the Private2 bit so we know this page was properly
1071 * setup for writepage
1073 page_ops
= unlock
? PAGE_UNLOCK
: 0;
1074 page_ops
|= PAGE_SET_PRIVATE2
;
1076 extent_clear_unlock_delalloc(inode
, start
,
1077 start
+ ram_size
- 1,
1078 delalloc_end
, locked_page
,
1079 EXTENT_LOCKED
| EXTENT_DELALLOC
,
1081 if (disk_num_bytes
< cur_alloc_size
)
1084 disk_num_bytes
-= cur_alloc_size
;
1085 num_bytes
-= cur_alloc_size
;
1086 alloc_hint
= ins
.objectid
+ ins
.offset
;
1087 start
+= cur_alloc_size
;
1088 extent_reserved
= false;
1091 * btrfs_reloc_clone_csums() error, since start is increased
1092 * extent_clear_unlock_delalloc() at out_unlock label won't
1093 * free metadata of current ordered extent, we're OK to exit.
1101 out_drop_extent_cache
:
1102 btrfs_drop_extent_cache(BTRFS_I(inode
), start
, start
+ ram_size
- 1, 0);
1104 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
1105 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 1);
1107 clear_bits
= EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DELALLOC_NEW
|
1108 EXTENT_DEFRAG
| EXTENT_CLEAR_META_RESV
;
1109 page_ops
= PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
1112 * If we reserved an extent for our delalloc range (or a subrange) and
1113 * failed to create the respective ordered extent, then it means that
1114 * when we reserved the extent we decremented the extent's size from
1115 * the data space_info's bytes_may_use counter and incremented the
1116 * space_info's bytes_reserved counter by the same amount. We must make
1117 * sure extent_clear_unlock_delalloc() does not try to decrement again
1118 * the data space_info's bytes_may_use counter, therefore we do not pass
1119 * it the flag EXTENT_CLEAR_DATA_RESV.
1121 if (extent_reserved
) {
1122 extent_clear_unlock_delalloc(inode
, start
,
1123 start
+ cur_alloc_size
,
1124 start
+ cur_alloc_size
,
1128 start
+= cur_alloc_size
;
1132 extent_clear_unlock_delalloc(inode
, start
, end
, delalloc_end
,
1134 clear_bits
| EXTENT_CLEAR_DATA_RESV
,
1140 * work queue call back to started compression on a file and pages
1142 static noinline
void async_cow_start(struct btrfs_work
*work
)
1144 struct async_cow
*async_cow
;
1146 async_cow
= container_of(work
, struct async_cow
, work
);
1148 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
1149 async_cow
->start
, async_cow
->end
, async_cow
,
1151 if (num_added
== 0) {
1152 btrfs_add_delayed_iput(async_cow
->inode
);
1153 async_cow
->inode
= NULL
;
1158 * work queue call back to submit previously compressed pages
1160 static noinline
void async_cow_submit(struct btrfs_work
*work
)
1162 struct btrfs_fs_info
*fs_info
;
1163 struct async_cow
*async_cow
;
1164 struct btrfs_root
*root
;
1165 unsigned long nr_pages
;
1167 async_cow
= container_of(work
, struct async_cow
, work
);
1169 root
= async_cow
->root
;
1170 fs_info
= root
->fs_info
;
1171 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_SIZE
) >>
1175 * atomic_sub_return implies a barrier for waitqueue_active
1177 if (atomic_sub_return(nr_pages
, &fs_info
->async_delalloc_pages
) <
1179 waitqueue_active(&fs_info
->async_submit_wait
))
1180 wake_up(&fs_info
->async_submit_wait
);
1182 if (async_cow
->inode
)
1183 submit_compressed_extents(async_cow
->inode
, async_cow
);
1186 static noinline
void async_cow_free(struct btrfs_work
*work
)
1188 struct async_cow
*async_cow
;
1189 async_cow
= container_of(work
, struct async_cow
, work
);
1190 if (async_cow
->inode
)
1191 btrfs_add_delayed_iput(async_cow
->inode
);
1195 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1196 u64 start
, u64 end
, int *page_started
,
1197 unsigned long *nr_written
,
1198 unsigned int write_flags
)
1200 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1201 struct async_cow
*async_cow
;
1202 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1203 unsigned long nr_pages
;
1206 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1207 1, 0, NULL
, GFP_NOFS
);
1208 while (start
< end
) {
1209 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1210 BUG_ON(!async_cow
); /* -ENOMEM */
1211 async_cow
->inode
= igrab(inode
);
1212 async_cow
->root
= root
;
1213 async_cow
->locked_page
= locked_page
;
1214 async_cow
->start
= start
;
1215 async_cow
->write_flags
= write_flags
;
1217 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
&&
1218 !btrfs_test_opt(fs_info
, FORCE_COMPRESS
))
1221 cur_end
= min(end
, start
+ SZ_512K
- 1);
1223 async_cow
->end
= cur_end
;
1224 INIT_LIST_HEAD(&async_cow
->extents
);
1226 btrfs_init_work(&async_cow
->work
,
1227 btrfs_delalloc_helper
,
1228 async_cow_start
, async_cow_submit
,
1231 nr_pages
= (cur_end
- start
+ PAGE_SIZE
) >>
1233 atomic_add(nr_pages
, &fs_info
->async_delalloc_pages
);
1235 btrfs_queue_work(fs_info
->delalloc_workers
, &async_cow
->work
);
1237 *nr_written
+= nr_pages
;
1238 start
= cur_end
+ 1;
1244 static noinline
int csum_exist_in_range(struct btrfs_fs_info
*fs_info
,
1245 u64 bytenr
, u64 num_bytes
)
1248 struct btrfs_ordered_sum
*sums
;
1251 ret
= btrfs_lookup_csums_range(fs_info
->csum_root
, bytenr
,
1252 bytenr
+ num_bytes
- 1, &list
, 0);
1253 if (ret
== 0 && list_empty(&list
))
1256 while (!list_empty(&list
)) {
1257 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1258 list_del(&sums
->list
);
1267 * when nowcow writeback call back. This checks for snapshots or COW copies
1268 * of the extents that exist in the file, and COWs the file as required.
1270 * If no cow copies or snapshots exist, we write directly to the existing
1273 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1274 struct page
*locked_page
,
1275 u64 start
, u64 end
, int *page_started
, int force
,
1276 unsigned long *nr_written
)
1278 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1279 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1280 struct extent_buffer
*leaf
;
1281 struct btrfs_path
*path
;
1282 struct btrfs_file_extent_item
*fi
;
1283 struct btrfs_key found_key
;
1284 struct extent_map
*em
;
1299 u64 ino
= btrfs_ino(BTRFS_I(inode
));
1301 path
= btrfs_alloc_path();
1303 extent_clear_unlock_delalloc(inode
, start
, end
, end
,
1305 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1306 EXTENT_DO_ACCOUNTING
|
1307 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1309 PAGE_SET_WRITEBACK
|
1310 PAGE_END_WRITEBACK
);
1314 nolock
= btrfs_is_free_space_inode(BTRFS_I(inode
));
1316 cow_start
= (u64
)-1;
1319 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, ino
,
1323 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1324 leaf
= path
->nodes
[0];
1325 btrfs_item_key_to_cpu(leaf
, &found_key
,
1326 path
->slots
[0] - 1);
1327 if (found_key
.objectid
== ino
&&
1328 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1333 leaf
= path
->nodes
[0];
1334 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1335 ret
= btrfs_next_leaf(root
, path
);
1337 if (cow_start
!= (u64
)-1)
1338 cur_offset
= cow_start
;
1343 leaf
= path
->nodes
[0];
1349 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1351 if (found_key
.objectid
> ino
)
1353 if (WARN_ON_ONCE(found_key
.objectid
< ino
) ||
1354 found_key
.type
< BTRFS_EXTENT_DATA_KEY
) {
1358 if (found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1359 found_key
.offset
> end
)
1362 if (found_key
.offset
> cur_offset
) {
1363 extent_end
= found_key
.offset
;
1368 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1369 struct btrfs_file_extent_item
);
1370 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1372 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
1373 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1374 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1375 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1376 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1377 extent_end
= found_key
.offset
+
1378 btrfs_file_extent_num_bytes(leaf
, fi
);
1380 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1381 if (extent_end
<= start
) {
1385 if (disk_bytenr
== 0)
1387 if (btrfs_file_extent_compression(leaf
, fi
) ||
1388 btrfs_file_extent_encryption(leaf
, fi
) ||
1389 btrfs_file_extent_other_encoding(leaf
, fi
))
1391 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1393 if (btrfs_extent_readonly(fs_info
, disk_bytenr
))
1395 ret
= btrfs_cross_ref_exist(root
, ino
,
1397 extent_offset
, disk_bytenr
);
1400 * ret could be -EIO if the above fails to read
1404 if (cow_start
!= (u64
)-1)
1405 cur_offset
= cow_start
;
1409 WARN_ON_ONCE(nolock
);
1412 disk_bytenr
+= extent_offset
;
1413 disk_bytenr
+= cur_offset
- found_key
.offset
;
1414 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1416 * if there are pending snapshots for this root,
1417 * we fall into common COW way.
1420 err
= btrfs_start_write_no_snapshotting(root
);
1425 * force cow if csum exists in the range.
1426 * this ensure that csum for a given extent are
1427 * either valid or do not exist.
1429 ret
= csum_exist_in_range(fs_info
, disk_bytenr
,
1433 btrfs_end_write_no_snapshotting(root
);
1436 * ret could be -EIO if the above fails to read
1440 if (cow_start
!= (u64
)-1)
1441 cur_offset
= cow_start
;
1444 WARN_ON_ONCE(nolock
);
1447 if (!btrfs_inc_nocow_writers(fs_info
, disk_bytenr
)) {
1449 btrfs_end_write_no_snapshotting(root
);
1453 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1454 extent_end
= found_key
.offset
+
1455 btrfs_file_extent_inline_len(leaf
,
1456 path
->slots
[0], fi
);
1457 extent_end
= ALIGN(extent_end
,
1458 fs_info
->sectorsize
);
1463 if (extent_end
<= start
) {
1465 if (!nolock
&& nocow
)
1466 btrfs_end_write_no_snapshotting(root
);
1468 btrfs_dec_nocow_writers(fs_info
, disk_bytenr
);
1472 if (cow_start
== (u64
)-1)
1473 cow_start
= cur_offset
;
1474 cur_offset
= extent_end
;
1475 if (cur_offset
> end
)
1481 btrfs_release_path(path
);
1482 if (cow_start
!= (u64
)-1) {
1483 ret
= cow_file_range(inode
, locked_page
,
1484 cow_start
, found_key
.offset
- 1,
1485 end
, page_started
, nr_written
, 1,
1488 if (!nolock
&& nocow
)
1489 btrfs_end_write_no_snapshotting(root
);
1491 btrfs_dec_nocow_writers(fs_info
,
1495 cow_start
= (u64
)-1;
1498 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1499 u64 orig_start
= found_key
.offset
- extent_offset
;
1501 em
= create_io_em(inode
, cur_offset
, num_bytes
,
1503 disk_bytenr
, /* block_start */
1504 num_bytes
, /* block_len */
1505 disk_num_bytes
, /* orig_block_len */
1506 ram_bytes
, BTRFS_COMPRESS_NONE
,
1507 BTRFS_ORDERED_PREALLOC
);
1509 if (!nolock
&& nocow
)
1510 btrfs_end_write_no_snapshotting(root
);
1512 btrfs_dec_nocow_writers(fs_info
,
1517 free_extent_map(em
);
1520 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1521 type
= BTRFS_ORDERED_PREALLOC
;
1523 type
= BTRFS_ORDERED_NOCOW
;
1526 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1527 num_bytes
, num_bytes
, type
);
1529 btrfs_dec_nocow_writers(fs_info
, disk_bytenr
);
1530 BUG_ON(ret
); /* -ENOMEM */
1532 if (root
->root_key
.objectid
==
1533 BTRFS_DATA_RELOC_TREE_OBJECTID
)
1535 * Error handled later, as we must prevent
1536 * extent_clear_unlock_delalloc() in error handler
1537 * from freeing metadata of created ordered extent.
1539 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1542 extent_clear_unlock_delalloc(inode
, cur_offset
,
1543 cur_offset
+ num_bytes
- 1, end
,
1544 locked_page
, EXTENT_LOCKED
|
1546 EXTENT_CLEAR_DATA_RESV
,
1547 PAGE_UNLOCK
| PAGE_SET_PRIVATE2
);
1549 if (!nolock
&& nocow
)
1550 btrfs_end_write_no_snapshotting(root
);
1551 cur_offset
= extent_end
;
1554 * btrfs_reloc_clone_csums() error, now we're OK to call error
1555 * handler, as metadata for created ordered extent will only
1556 * be freed by btrfs_finish_ordered_io().
1560 if (cur_offset
> end
)
1563 btrfs_release_path(path
);
1565 if (cur_offset
<= end
&& cow_start
== (u64
)-1) {
1566 cow_start
= cur_offset
;
1570 if (cow_start
!= (u64
)-1) {
1571 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
, end
,
1572 page_started
, nr_written
, 1, NULL
);
1578 if (ret
&& cur_offset
< end
)
1579 extent_clear_unlock_delalloc(inode
, cur_offset
, end
, end
,
1580 locked_page
, EXTENT_LOCKED
|
1581 EXTENT_DELALLOC
| EXTENT_DEFRAG
|
1582 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
1584 PAGE_SET_WRITEBACK
|
1585 PAGE_END_WRITEBACK
);
1586 btrfs_free_path(path
);
1590 static inline int need_force_cow(struct inode
*inode
, u64 start
, u64 end
)
1593 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
1594 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
))
1598 * @defrag_bytes is a hint value, no spinlock held here,
1599 * if is not zero, it means the file is defragging.
1600 * Force cow if given extent needs to be defragged.
1602 if (BTRFS_I(inode
)->defrag_bytes
&&
1603 test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1604 EXTENT_DEFRAG
, 0, NULL
))
1611 * extent_io.c call back to do delayed allocation processing
1613 static int run_delalloc_range(void *private_data
, struct page
*locked_page
,
1614 u64 start
, u64 end
, int *page_started
,
1615 unsigned long *nr_written
,
1616 struct writeback_control
*wbc
)
1618 struct inode
*inode
= private_data
;
1620 int force_cow
= need_force_cow(inode
, start
, end
);
1621 unsigned int write_flags
= wbc_to_write_flags(wbc
);
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
,
1640 btrfs_cleanup_ordered_extents(inode
, start
, end
- start
+ 1);
1644 static void btrfs_split_extent_hook(void *private_data
,
1645 struct extent_state
*orig
, u64 split
)
1647 struct inode
*inode
= private_data
;
1650 /* not delalloc, ignore it */
1651 if (!(orig
->state
& EXTENT_DELALLOC
))
1654 size
= orig
->end
- orig
->start
+ 1;
1655 if (size
> BTRFS_MAX_EXTENT_SIZE
) {
1660 * See the explanation in btrfs_merge_extent_hook, the same
1661 * applies here, just in reverse.
1663 new_size
= orig
->end
- split
+ 1;
1664 num_extents
= count_max_extents(new_size
);
1665 new_size
= split
- orig
->start
;
1666 num_extents
+= count_max_extents(new_size
);
1667 if (count_max_extents(size
) >= num_extents
)
1671 spin_lock(&BTRFS_I(inode
)->lock
);
1672 btrfs_mod_outstanding_extents(BTRFS_I(inode
), 1);
1673 spin_unlock(&BTRFS_I(inode
)->lock
);
1677 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1678 * extents so we can keep track of new extents that are just merged onto old
1679 * extents, such as when we are doing sequential writes, so we can properly
1680 * account for the metadata space we'll need.
1682 static void btrfs_merge_extent_hook(void *private_data
,
1683 struct extent_state
*new,
1684 struct extent_state
*other
)
1686 struct inode
*inode
= private_data
;
1687 u64 new_size
, old_size
;
1690 /* not delalloc, ignore it */
1691 if (!(other
->state
& EXTENT_DELALLOC
))
1694 if (new->start
> other
->start
)
1695 new_size
= new->end
- other
->start
+ 1;
1697 new_size
= other
->end
- new->start
+ 1;
1699 /* we're not bigger than the max, unreserve the space and go */
1700 if (new_size
<= BTRFS_MAX_EXTENT_SIZE
) {
1701 spin_lock(&BTRFS_I(inode
)->lock
);
1702 btrfs_mod_outstanding_extents(BTRFS_I(inode
), -1);
1703 spin_unlock(&BTRFS_I(inode
)->lock
);
1708 * We have to add up either side to figure out how many extents were
1709 * accounted for before we merged into one big extent. If the number of
1710 * extents we accounted for is <= the amount we need for the new range
1711 * then we can return, otherwise drop. Think of it like this
1715 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1716 * need 2 outstanding extents, on one side we have 1 and the other side
1717 * we have 1 so they are == and we can return. But in this case
1719 * [MAX_SIZE+4k][MAX_SIZE+4k]
1721 * Each range on their own accounts for 2 extents, but merged together
1722 * they are only 3 extents worth of accounting, so we need to drop in
1725 old_size
= other
->end
- other
->start
+ 1;
1726 num_extents
= count_max_extents(old_size
);
1727 old_size
= new->end
- new->start
+ 1;
1728 num_extents
+= count_max_extents(old_size
);
1729 if (count_max_extents(new_size
) >= num_extents
)
1732 spin_lock(&BTRFS_I(inode
)->lock
);
1733 btrfs_mod_outstanding_extents(BTRFS_I(inode
), -1);
1734 spin_unlock(&BTRFS_I(inode
)->lock
);
1737 static void btrfs_add_delalloc_inodes(struct btrfs_root
*root
,
1738 struct inode
*inode
)
1740 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1742 spin_lock(&root
->delalloc_lock
);
1743 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1744 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1745 &root
->delalloc_inodes
);
1746 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1747 &BTRFS_I(inode
)->runtime_flags
);
1748 root
->nr_delalloc_inodes
++;
1749 if (root
->nr_delalloc_inodes
== 1) {
1750 spin_lock(&fs_info
->delalloc_root_lock
);
1751 BUG_ON(!list_empty(&root
->delalloc_root
));
1752 list_add_tail(&root
->delalloc_root
,
1753 &fs_info
->delalloc_roots
);
1754 spin_unlock(&fs_info
->delalloc_root_lock
);
1757 spin_unlock(&root
->delalloc_lock
);
1761 void __btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1762 struct btrfs_inode
*inode
)
1764 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
1766 if (!list_empty(&inode
->delalloc_inodes
)) {
1767 list_del_init(&inode
->delalloc_inodes
);
1768 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1769 &inode
->runtime_flags
);
1770 root
->nr_delalloc_inodes
--;
1771 if (!root
->nr_delalloc_inodes
) {
1772 spin_lock(&fs_info
->delalloc_root_lock
);
1773 BUG_ON(list_empty(&root
->delalloc_root
));
1774 list_del_init(&root
->delalloc_root
);
1775 spin_unlock(&fs_info
->delalloc_root_lock
);
1780 static void btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1781 struct btrfs_inode
*inode
)
1783 spin_lock(&root
->delalloc_lock
);
1784 __btrfs_del_delalloc_inode(root
, inode
);
1785 spin_unlock(&root
->delalloc_lock
);
1789 * extent_io.c set_bit_hook, used to track delayed allocation
1790 * bytes in this file, and to maintain the list of inodes that
1791 * have pending delalloc work to be done.
1793 static void btrfs_set_bit_hook(void *private_data
,
1794 struct extent_state
*state
, unsigned *bits
)
1796 struct inode
*inode
= private_data
;
1798 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1800 if ((*bits
& EXTENT_DEFRAG
) && !(*bits
& EXTENT_DELALLOC
))
1803 * set_bit and clear bit hooks normally require _irqsave/restore
1804 * but in this case, we are only testing for the DELALLOC
1805 * bit, which is only set or cleared with irqs on
1807 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1808 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1809 u64 len
= state
->end
+ 1 - state
->start
;
1810 u32 num_extents
= count_max_extents(len
);
1811 bool do_list
= !btrfs_is_free_space_inode(BTRFS_I(inode
));
1813 spin_lock(&BTRFS_I(inode
)->lock
);
1814 btrfs_mod_outstanding_extents(BTRFS_I(inode
), num_extents
);
1815 spin_unlock(&BTRFS_I(inode
)->lock
);
1817 /* For sanity tests */
1818 if (btrfs_is_testing(fs_info
))
1821 percpu_counter_add_batch(&fs_info
->delalloc_bytes
, len
,
1822 fs_info
->delalloc_batch
);
1823 spin_lock(&BTRFS_I(inode
)->lock
);
1824 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1825 if (*bits
& EXTENT_DEFRAG
)
1826 BTRFS_I(inode
)->defrag_bytes
+= len
;
1827 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1828 &BTRFS_I(inode
)->runtime_flags
))
1829 btrfs_add_delalloc_inodes(root
, inode
);
1830 spin_unlock(&BTRFS_I(inode
)->lock
);
1833 if (!(state
->state
& EXTENT_DELALLOC_NEW
) &&
1834 (*bits
& EXTENT_DELALLOC_NEW
)) {
1835 spin_lock(&BTRFS_I(inode
)->lock
);
1836 BTRFS_I(inode
)->new_delalloc_bytes
+= state
->end
+ 1 -
1838 spin_unlock(&BTRFS_I(inode
)->lock
);
1843 * extent_io.c clear_bit_hook, see set_bit_hook for why
1845 static void btrfs_clear_bit_hook(void *private_data
,
1846 struct extent_state
*state
,
1849 struct btrfs_inode
*inode
= BTRFS_I((struct inode
*)private_data
);
1850 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
1851 u64 len
= state
->end
+ 1 - state
->start
;
1852 u32 num_extents
= count_max_extents(len
);
1854 if ((state
->state
& EXTENT_DEFRAG
) && (*bits
& EXTENT_DEFRAG
)) {
1855 spin_lock(&inode
->lock
);
1856 inode
->defrag_bytes
-= len
;
1857 spin_unlock(&inode
->lock
);
1861 * set_bit and clear bit hooks normally require _irqsave/restore
1862 * but in this case, we are only testing for the DELALLOC
1863 * bit, which is only set or cleared with irqs on
1865 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1866 struct btrfs_root
*root
= inode
->root
;
1867 bool do_list
= !btrfs_is_free_space_inode(inode
);
1869 spin_lock(&inode
->lock
);
1870 btrfs_mod_outstanding_extents(inode
, -num_extents
);
1871 spin_unlock(&inode
->lock
);
1874 * We don't reserve metadata space for space cache inodes so we
1875 * don't need to call dellalloc_release_metadata if there is an
1878 if (*bits
& EXTENT_CLEAR_META_RESV
&&
1879 root
!= fs_info
->tree_root
)
1880 btrfs_delalloc_release_metadata(inode
, len
);
1882 /* For sanity tests. */
1883 if (btrfs_is_testing(fs_info
))
1886 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1887 do_list
&& !(state
->state
& EXTENT_NORESERVE
) &&
1888 (*bits
& EXTENT_CLEAR_DATA_RESV
))
1889 btrfs_free_reserved_data_space_noquota(
1893 percpu_counter_add_batch(&fs_info
->delalloc_bytes
, -len
,
1894 fs_info
->delalloc_batch
);
1895 spin_lock(&inode
->lock
);
1896 inode
->delalloc_bytes
-= len
;
1897 if (do_list
&& inode
->delalloc_bytes
== 0 &&
1898 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1899 &inode
->runtime_flags
))
1900 btrfs_del_delalloc_inode(root
, inode
);
1901 spin_unlock(&inode
->lock
);
1904 if ((state
->state
& EXTENT_DELALLOC_NEW
) &&
1905 (*bits
& EXTENT_DELALLOC_NEW
)) {
1906 spin_lock(&inode
->lock
);
1907 ASSERT(inode
->new_delalloc_bytes
>= len
);
1908 inode
->new_delalloc_bytes
-= len
;
1909 spin_unlock(&inode
->lock
);
1914 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1915 * we don't create bios that span stripes or chunks
1917 * return 1 if page cannot be merged to bio
1918 * return 0 if page can be merged to bio
1919 * return error otherwise
1921 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1922 size_t size
, struct bio
*bio
,
1923 unsigned long bio_flags
)
1925 struct inode
*inode
= page
->mapping
->host
;
1926 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1927 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
1932 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1935 length
= bio
->bi_iter
.bi_size
;
1936 map_length
= length
;
1937 ret
= btrfs_map_block(fs_info
, btrfs_op(bio
), logical
, &map_length
,
1941 if (map_length
< length
+ size
)
1947 * in order to insert checksums into the metadata in large chunks,
1948 * we wait until bio submission time. All the pages in the bio are
1949 * checksummed and sums are attached onto the ordered extent record.
1951 * At IO completion time the cums attached on the ordered extent record
1952 * are inserted into the btree
1954 static blk_status_t
__btrfs_submit_bio_start(void *private_data
, struct bio
*bio
,
1955 int mirror_num
, unsigned long bio_flags
,
1958 struct inode
*inode
= private_data
;
1959 blk_status_t ret
= 0;
1961 ret
= btrfs_csum_one_bio(inode
, bio
, 0, 0);
1962 BUG_ON(ret
); /* -ENOMEM */
1967 * in order to insert checksums into the metadata in large chunks,
1968 * we wait until bio submission time. All the pages in the bio are
1969 * checksummed and sums are attached onto the ordered extent record.
1971 * At IO completion time the cums attached on the ordered extent record
1972 * are inserted into the btree
1974 static blk_status_t
__btrfs_submit_bio_done(void *private_data
, struct bio
*bio
,
1975 int mirror_num
, unsigned long bio_flags
,
1978 struct inode
*inode
= private_data
;
1979 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1982 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 1);
1984 bio
->bi_status
= ret
;
1991 * extent_io.c submission hook. This does the right thing for csum calculation
1992 * on write, or reading the csums from the tree before a read
1994 static blk_status_t
btrfs_submit_bio_hook(void *private_data
, struct bio
*bio
,
1995 int mirror_num
, unsigned long bio_flags
,
1998 struct inode
*inode
= private_data
;
1999 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2000 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2001 enum btrfs_wq_endio_type metadata
= BTRFS_WQ_ENDIO_DATA
;
2002 blk_status_t ret
= 0;
2004 int async
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
2006 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
2008 if (btrfs_is_free_space_inode(BTRFS_I(inode
)))
2009 metadata
= BTRFS_WQ_ENDIO_FREE_SPACE
;
2011 if (bio_op(bio
) != REQ_OP_WRITE
) {
2012 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, metadata
);
2016 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
2017 ret
= btrfs_submit_compressed_read(inode
, bio
,
2021 } else if (!skip_sum
) {
2022 ret
= btrfs_lookup_bio_sums(inode
, bio
, NULL
);
2027 } else if (async
&& !skip_sum
) {
2028 /* csum items have already been cloned */
2029 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2031 /* we're doing a write, do the async checksumming */
2032 ret
= btrfs_wq_submit_bio(fs_info
, bio
, mirror_num
, bio_flags
,
2034 __btrfs_submit_bio_start
,
2035 __btrfs_submit_bio_done
);
2037 } else if (!skip_sum
) {
2038 ret
= btrfs_csum_one_bio(inode
, bio
, 0, 0);
2044 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
2048 bio
->bi_status
= ret
;
2055 * given a list of ordered sums record them in the inode. This happens
2056 * at IO completion time based on sums calculated at bio submission time.
2058 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
2059 struct inode
*inode
, struct list_head
*list
)
2061 struct btrfs_ordered_sum
*sum
;
2063 list_for_each_entry(sum
, list
, list
) {
2064 trans
->adding_csums
= 1;
2065 btrfs_csum_file_blocks(trans
,
2066 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
2067 trans
->adding_csums
= 0;
2072 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
2073 unsigned int extra_bits
,
2074 struct extent_state
**cached_state
, int dedupe
)
2076 WARN_ON((end
& (PAGE_SIZE
- 1)) == 0);
2077 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
2078 extra_bits
, cached_state
);
2081 /* see btrfs_writepage_start_hook for details on why this is required */
2082 struct btrfs_writepage_fixup
{
2084 struct btrfs_work work
;
2087 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
2089 struct btrfs_writepage_fixup
*fixup
;
2090 struct btrfs_ordered_extent
*ordered
;
2091 struct extent_state
*cached_state
= NULL
;
2092 struct extent_changeset
*data_reserved
= NULL
;
2094 struct inode
*inode
;
2099 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
2103 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
2104 ClearPageChecked(page
);
2108 inode
= page
->mapping
->host
;
2109 page_start
= page_offset(page
);
2110 page_end
= page_offset(page
) + PAGE_SIZE
- 1;
2112 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2115 /* already ordered? We're done */
2116 if (PagePrivate2(page
))
2119 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), page_start
,
2122 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
2123 page_end
, &cached_state
, GFP_NOFS
);
2125 btrfs_start_ordered_extent(inode
, ordered
, 1);
2126 btrfs_put_ordered_extent(ordered
);
2130 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
, page_start
,
2133 mapping_set_error(page
->mapping
, ret
);
2134 end_extent_writepage(page
, ret
, page_start
, page_end
);
2135 ClearPageChecked(page
);
2139 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
, 0,
2142 mapping_set_error(page
->mapping
, ret
);
2143 end_extent_writepage(page
, ret
, page_start
, page_end
);
2144 ClearPageChecked(page
);
2148 ClearPageChecked(page
);
2149 set_page_dirty(page
);
2150 btrfs_delalloc_release_extents(BTRFS_I(inode
), PAGE_SIZE
);
2152 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2153 &cached_state
, GFP_NOFS
);
2158 extent_changeset_free(data_reserved
);
2162 * There are a few paths in the higher layers of the kernel that directly
2163 * set the page dirty bit without asking the filesystem if it is a
2164 * good idea. This causes problems because we want to make sure COW
2165 * properly happens and the data=ordered rules are followed.
2167 * In our case any range that doesn't have the ORDERED bit set
2168 * hasn't been properly setup for IO. We kick off an async process
2169 * to fix it up. The async helper will wait for ordered extents, set
2170 * the delalloc bit and make it safe to write the page.
2172 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
2174 struct inode
*inode
= page
->mapping
->host
;
2175 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2176 struct btrfs_writepage_fixup
*fixup
;
2178 /* this page is properly in the ordered list */
2179 if (TestClearPagePrivate2(page
))
2182 if (PageChecked(page
))
2185 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
2189 SetPageChecked(page
);
2191 btrfs_init_work(&fixup
->work
, btrfs_fixup_helper
,
2192 btrfs_writepage_fixup_worker
, NULL
, NULL
);
2194 btrfs_queue_work(fs_info
->fixup_workers
, &fixup
->work
);
2198 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
2199 struct inode
*inode
, u64 file_pos
,
2200 u64 disk_bytenr
, u64 disk_num_bytes
,
2201 u64 num_bytes
, u64 ram_bytes
,
2202 u8 compression
, u8 encryption
,
2203 u16 other_encoding
, int extent_type
)
2205 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2206 struct btrfs_file_extent_item
*fi
;
2207 struct btrfs_path
*path
;
2208 struct extent_buffer
*leaf
;
2209 struct btrfs_key ins
;
2211 int extent_inserted
= 0;
2214 path
= btrfs_alloc_path();
2219 * we may be replacing one extent in the tree with another.
2220 * The new extent is pinned in the extent map, and we don't want
2221 * to drop it from the cache until it is completely in the btree.
2223 * So, tell btrfs_drop_extents to leave this extent in the cache.
2224 * the caller is expected to unpin it and allow it to be merged
2227 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, file_pos
,
2228 file_pos
+ num_bytes
, NULL
, 0,
2229 1, sizeof(*fi
), &extent_inserted
);
2233 if (!extent_inserted
) {
2234 ins
.objectid
= btrfs_ino(BTRFS_I(inode
));
2235 ins
.offset
= file_pos
;
2236 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
2238 path
->leave_spinning
= 1;
2239 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
2244 leaf
= path
->nodes
[0];
2245 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2246 struct btrfs_file_extent_item
);
2247 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
2248 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
2249 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
2250 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
2251 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2252 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2253 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
2254 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
2255 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
2256 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
2258 btrfs_mark_buffer_dirty(leaf
);
2259 btrfs_release_path(path
);
2261 inode_add_bytes(inode
, num_bytes
);
2263 ins
.objectid
= disk_bytenr
;
2264 ins
.offset
= disk_num_bytes
;
2265 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2268 * Release the reserved range from inode dirty range map, as it is
2269 * already moved into delayed_ref_head
2271 ret
= btrfs_qgroup_release_data(inode
, file_pos
, ram_bytes
);
2275 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
2276 btrfs_ino(BTRFS_I(inode
)),
2277 file_pos
, qg_released
, &ins
);
2279 btrfs_free_path(path
);
2284 /* snapshot-aware defrag */
2285 struct sa_defrag_extent_backref
{
2286 struct rb_node node
;
2287 struct old_sa_defrag_extent
*old
;
2296 struct old_sa_defrag_extent
{
2297 struct list_head list
;
2298 struct new_sa_defrag_extent
*new;
2307 struct new_sa_defrag_extent
{
2308 struct rb_root root
;
2309 struct list_head head
;
2310 struct btrfs_path
*path
;
2311 struct inode
*inode
;
2319 static int backref_comp(struct sa_defrag_extent_backref
*b1
,
2320 struct sa_defrag_extent_backref
*b2
)
2322 if (b1
->root_id
< b2
->root_id
)
2324 else if (b1
->root_id
> b2
->root_id
)
2327 if (b1
->inum
< b2
->inum
)
2329 else if (b1
->inum
> b2
->inum
)
2332 if (b1
->file_pos
< b2
->file_pos
)
2334 else if (b1
->file_pos
> b2
->file_pos
)
2338 * [------------------------------] ===> (a range of space)
2339 * |<--->| |<---->| =============> (fs/file tree A)
2340 * |<---------------------------->| ===> (fs/file tree B)
2342 * A range of space can refer to two file extents in one tree while
2343 * refer to only one file extent in another tree.
2345 * So we may process a disk offset more than one time(two extents in A)
2346 * and locate at the same extent(one extent in B), then insert two same
2347 * backrefs(both refer to the extent in B).
2352 static void backref_insert(struct rb_root
*root
,
2353 struct sa_defrag_extent_backref
*backref
)
2355 struct rb_node
**p
= &root
->rb_node
;
2356 struct rb_node
*parent
= NULL
;
2357 struct sa_defrag_extent_backref
*entry
;
2362 entry
= rb_entry(parent
, struct sa_defrag_extent_backref
, node
);
2364 ret
= backref_comp(backref
, entry
);
2368 p
= &(*p
)->rb_right
;
2371 rb_link_node(&backref
->node
, parent
, p
);
2372 rb_insert_color(&backref
->node
, root
);
2376 * Note the backref might has changed, and in this case we just return 0.
2378 static noinline
int record_one_backref(u64 inum
, u64 offset
, u64 root_id
,
2381 struct btrfs_file_extent_item
*extent
;
2382 struct old_sa_defrag_extent
*old
= ctx
;
2383 struct new_sa_defrag_extent
*new = old
->new;
2384 struct btrfs_path
*path
= new->path
;
2385 struct btrfs_key key
;
2386 struct btrfs_root
*root
;
2387 struct sa_defrag_extent_backref
*backref
;
2388 struct extent_buffer
*leaf
;
2389 struct inode
*inode
= new->inode
;
2390 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2396 if (BTRFS_I(inode
)->root
->root_key
.objectid
== root_id
&&
2397 inum
== btrfs_ino(BTRFS_I(inode
)))
2400 key
.objectid
= root_id
;
2401 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2402 key
.offset
= (u64
)-1;
2404 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2406 if (PTR_ERR(root
) == -ENOENT
)
2409 btrfs_debug(fs_info
, "inum=%llu, offset=%llu, root_id=%llu",
2410 inum
, offset
, root_id
);
2411 return PTR_ERR(root
);
2414 key
.objectid
= inum
;
2415 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2416 if (offset
> (u64
)-1 << 32)
2419 key
.offset
= offset
;
2421 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2422 if (WARN_ON(ret
< 0))
2429 leaf
= path
->nodes
[0];
2430 slot
= path
->slots
[0];
2432 if (slot
>= btrfs_header_nritems(leaf
)) {
2433 ret
= btrfs_next_leaf(root
, path
);
2436 } else if (ret
> 0) {
2445 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2447 if (key
.objectid
> inum
)
2450 if (key
.objectid
< inum
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2453 extent
= btrfs_item_ptr(leaf
, slot
,
2454 struct btrfs_file_extent_item
);
2456 if (btrfs_file_extent_disk_bytenr(leaf
, extent
) != old
->bytenr
)
2460 * 'offset' refers to the exact key.offset,
2461 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2462 * (key.offset - extent_offset).
2464 if (key
.offset
!= offset
)
2467 extent_offset
= btrfs_file_extent_offset(leaf
, extent
);
2468 num_bytes
= btrfs_file_extent_num_bytes(leaf
, extent
);
2470 if (extent_offset
>= old
->extent_offset
+ old
->offset
+
2471 old
->len
|| extent_offset
+ num_bytes
<=
2472 old
->extent_offset
+ old
->offset
)
2477 backref
= kmalloc(sizeof(*backref
), GFP_NOFS
);
2483 backref
->root_id
= root_id
;
2484 backref
->inum
= inum
;
2485 backref
->file_pos
= offset
;
2486 backref
->num_bytes
= num_bytes
;
2487 backref
->extent_offset
= extent_offset
;
2488 backref
->generation
= btrfs_file_extent_generation(leaf
, extent
);
2490 backref_insert(&new->root
, backref
);
2493 btrfs_release_path(path
);
2498 static noinline
bool record_extent_backrefs(struct btrfs_path
*path
,
2499 struct new_sa_defrag_extent
*new)
2501 struct btrfs_fs_info
*fs_info
= btrfs_sb(new->inode
->i_sb
);
2502 struct old_sa_defrag_extent
*old
, *tmp
;
2507 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2508 ret
= iterate_inodes_from_logical(old
->bytenr
+
2509 old
->extent_offset
, fs_info
,
2510 path
, record_one_backref
,
2512 if (ret
< 0 && ret
!= -ENOENT
)
2515 /* no backref to be processed for this extent */
2517 list_del(&old
->list
);
2522 if (list_empty(&new->head
))
2528 static int relink_is_mergable(struct extent_buffer
*leaf
,
2529 struct btrfs_file_extent_item
*fi
,
2530 struct new_sa_defrag_extent
*new)
2532 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) != new->bytenr
)
2535 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2538 if (btrfs_file_extent_compression(leaf
, fi
) != new->compress_type
)
2541 if (btrfs_file_extent_encryption(leaf
, fi
) ||
2542 btrfs_file_extent_other_encoding(leaf
, fi
))
2549 * Note the backref might has changed, and in this case we just return 0.
2551 static noinline
int relink_extent_backref(struct btrfs_path
*path
,
2552 struct sa_defrag_extent_backref
*prev
,
2553 struct sa_defrag_extent_backref
*backref
)
2555 struct btrfs_file_extent_item
*extent
;
2556 struct btrfs_file_extent_item
*item
;
2557 struct btrfs_ordered_extent
*ordered
;
2558 struct btrfs_trans_handle
*trans
;
2559 struct btrfs_root
*root
;
2560 struct btrfs_key key
;
2561 struct extent_buffer
*leaf
;
2562 struct old_sa_defrag_extent
*old
= backref
->old
;
2563 struct new_sa_defrag_extent
*new = old
->new;
2564 struct btrfs_fs_info
*fs_info
= btrfs_sb(new->inode
->i_sb
);
2565 struct inode
*inode
;
2566 struct extent_state
*cached
= NULL
;
2575 if (prev
&& prev
->root_id
== backref
->root_id
&&
2576 prev
->inum
== backref
->inum
&&
2577 prev
->file_pos
+ prev
->num_bytes
== backref
->file_pos
)
2580 /* step 1: get root */
2581 key
.objectid
= backref
->root_id
;
2582 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2583 key
.offset
= (u64
)-1;
2585 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
2587 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2589 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2590 if (PTR_ERR(root
) == -ENOENT
)
2592 return PTR_ERR(root
);
2595 if (btrfs_root_readonly(root
)) {
2596 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2600 /* step 2: get inode */
2601 key
.objectid
= backref
->inum
;
2602 key
.type
= BTRFS_INODE_ITEM_KEY
;
2605 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
2606 if (IS_ERR(inode
)) {
2607 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2611 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2613 /* step 3: relink backref */
2614 lock_start
= backref
->file_pos
;
2615 lock_end
= backref
->file_pos
+ backref
->num_bytes
- 1;
2616 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2619 ordered
= btrfs_lookup_first_ordered_extent(inode
, lock_end
);
2621 btrfs_put_ordered_extent(ordered
);
2625 trans
= btrfs_join_transaction(root
);
2626 if (IS_ERR(trans
)) {
2627 ret
= PTR_ERR(trans
);
2631 key
.objectid
= backref
->inum
;
2632 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2633 key
.offset
= backref
->file_pos
;
2635 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2638 } else if (ret
> 0) {
2643 extent
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2644 struct btrfs_file_extent_item
);
2646 if (btrfs_file_extent_generation(path
->nodes
[0], extent
) !=
2647 backref
->generation
)
2650 btrfs_release_path(path
);
2652 start
= backref
->file_pos
;
2653 if (backref
->extent_offset
< old
->extent_offset
+ old
->offset
)
2654 start
+= old
->extent_offset
+ old
->offset
-
2655 backref
->extent_offset
;
2657 len
= min(backref
->extent_offset
+ backref
->num_bytes
,
2658 old
->extent_offset
+ old
->offset
+ old
->len
);
2659 len
-= max(backref
->extent_offset
, old
->extent_offset
+ old
->offset
);
2661 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
2666 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
2667 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2670 path
->leave_spinning
= 1;
2672 struct btrfs_file_extent_item
*fi
;
2674 struct btrfs_key found_key
;
2676 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2681 leaf
= path
->nodes
[0];
2682 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2684 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2685 struct btrfs_file_extent_item
);
2686 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
2688 if (extent_len
+ found_key
.offset
== start
&&
2689 relink_is_mergable(leaf
, fi
, new)) {
2690 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2692 btrfs_mark_buffer_dirty(leaf
);
2693 inode_add_bytes(inode
, len
);
2699 btrfs_release_path(path
);
2704 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2707 btrfs_abort_transaction(trans
, ret
);
2711 leaf
= path
->nodes
[0];
2712 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2713 struct btrfs_file_extent_item
);
2714 btrfs_set_file_extent_disk_bytenr(leaf
, item
, new->bytenr
);
2715 btrfs_set_file_extent_disk_num_bytes(leaf
, item
, new->disk_len
);
2716 btrfs_set_file_extent_offset(leaf
, item
, start
- new->file_pos
);
2717 btrfs_set_file_extent_num_bytes(leaf
, item
, len
);
2718 btrfs_set_file_extent_ram_bytes(leaf
, item
, new->len
);
2719 btrfs_set_file_extent_generation(leaf
, item
, trans
->transid
);
2720 btrfs_set_file_extent_type(leaf
, item
, BTRFS_FILE_EXTENT_REG
);
2721 btrfs_set_file_extent_compression(leaf
, item
, new->compress_type
);
2722 btrfs_set_file_extent_encryption(leaf
, item
, 0);
2723 btrfs_set_file_extent_other_encoding(leaf
, item
, 0);
2725 btrfs_mark_buffer_dirty(leaf
);
2726 inode_add_bytes(inode
, len
);
2727 btrfs_release_path(path
);
2729 ret
= btrfs_inc_extent_ref(trans
, root
, new->bytenr
,
2731 backref
->root_id
, backref
->inum
,
2732 new->file_pos
); /* start - extent_offset */
2734 btrfs_abort_transaction(trans
, ret
);
2740 btrfs_release_path(path
);
2741 path
->leave_spinning
= 0;
2742 btrfs_end_transaction(trans
);
2744 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2750 static void free_sa_defrag_extent(struct new_sa_defrag_extent
*new)
2752 struct old_sa_defrag_extent
*old
, *tmp
;
2757 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2763 static void relink_file_extents(struct new_sa_defrag_extent
*new)
2765 struct btrfs_fs_info
*fs_info
= btrfs_sb(new->inode
->i_sb
);
2766 struct btrfs_path
*path
;
2767 struct sa_defrag_extent_backref
*backref
;
2768 struct sa_defrag_extent_backref
*prev
= NULL
;
2769 struct inode
*inode
;
2770 struct btrfs_root
*root
;
2771 struct rb_node
*node
;
2775 root
= BTRFS_I(inode
)->root
;
2777 path
= btrfs_alloc_path();
2781 if (!record_extent_backrefs(path
, new)) {
2782 btrfs_free_path(path
);
2785 btrfs_release_path(path
);
2788 node
= rb_first(&new->root
);
2791 rb_erase(node
, &new->root
);
2793 backref
= rb_entry(node
, struct sa_defrag_extent_backref
, node
);
2795 ret
= relink_extent_backref(path
, prev
, backref
);
2808 btrfs_free_path(path
);
2810 free_sa_defrag_extent(new);
2812 atomic_dec(&fs_info
->defrag_running
);
2813 wake_up(&fs_info
->transaction_wait
);
2816 static struct new_sa_defrag_extent
*
2817 record_old_file_extents(struct inode
*inode
,
2818 struct btrfs_ordered_extent
*ordered
)
2820 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2821 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2822 struct btrfs_path
*path
;
2823 struct btrfs_key key
;
2824 struct old_sa_defrag_extent
*old
;
2825 struct new_sa_defrag_extent
*new;
2828 new = kmalloc(sizeof(*new), GFP_NOFS
);
2833 new->file_pos
= ordered
->file_offset
;
2834 new->len
= ordered
->len
;
2835 new->bytenr
= ordered
->start
;
2836 new->disk_len
= ordered
->disk_len
;
2837 new->compress_type
= ordered
->compress_type
;
2838 new->root
= RB_ROOT
;
2839 INIT_LIST_HEAD(&new->head
);
2841 path
= btrfs_alloc_path();
2845 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
2846 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2847 key
.offset
= new->file_pos
;
2849 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2852 if (ret
> 0 && path
->slots
[0] > 0)
2855 /* find out all the old extents for the file range */
2857 struct btrfs_file_extent_item
*extent
;
2858 struct extent_buffer
*l
;
2867 slot
= path
->slots
[0];
2869 if (slot
>= btrfs_header_nritems(l
)) {
2870 ret
= btrfs_next_leaf(root
, path
);
2878 btrfs_item_key_to_cpu(l
, &key
, slot
);
2880 if (key
.objectid
!= btrfs_ino(BTRFS_I(inode
)))
2882 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2884 if (key
.offset
>= new->file_pos
+ new->len
)
2887 extent
= btrfs_item_ptr(l
, slot
, struct btrfs_file_extent_item
);
2889 num_bytes
= btrfs_file_extent_num_bytes(l
, extent
);
2890 if (key
.offset
+ num_bytes
< new->file_pos
)
2893 disk_bytenr
= btrfs_file_extent_disk_bytenr(l
, extent
);
2897 extent_offset
= btrfs_file_extent_offset(l
, extent
);
2899 old
= kmalloc(sizeof(*old
), GFP_NOFS
);
2903 offset
= max(new->file_pos
, key
.offset
);
2904 end
= min(new->file_pos
+ new->len
, key
.offset
+ num_bytes
);
2906 old
->bytenr
= disk_bytenr
;
2907 old
->extent_offset
= extent_offset
;
2908 old
->offset
= offset
- key
.offset
;
2909 old
->len
= end
- offset
;
2912 list_add_tail(&old
->list
, &new->head
);
2918 btrfs_free_path(path
);
2919 atomic_inc(&fs_info
->defrag_running
);
2924 btrfs_free_path(path
);
2926 free_sa_defrag_extent(new);
2930 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info
*fs_info
,
2933 struct btrfs_block_group_cache
*cache
;
2935 cache
= btrfs_lookup_block_group(fs_info
, start
);
2938 spin_lock(&cache
->lock
);
2939 cache
->delalloc_bytes
-= len
;
2940 spin_unlock(&cache
->lock
);
2942 btrfs_put_block_group(cache
);
2945 /* as ordered data IO finishes, this gets called so we can finish
2946 * an ordered extent if the range of bytes in the file it covers are
2949 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
2951 struct inode
*inode
= ordered_extent
->inode
;
2952 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2953 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2954 struct btrfs_trans_handle
*trans
= NULL
;
2955 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2956 struct extent_state
*cached_state
= NULL
;
2957 struct new_sa_defrag_extent
*new = NULL
;
2958 int compress_type
= 0;
2960 u64 logical_len
= ordered_extent
->len
;
2962 bool truncated
= false;
2963 bool range_locked
= false;
2964 bool clear_new_delalloc_bytes
= false;
2966 if (!test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
2967 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
) &&
2968 !test_bit(BTRFS_ORDERED_DIRECT
, &ordered_extent
->flags
))
2969 clear_new_delalloc_bytes
= true;
2971 nolock
= btrfs_is_free_space_inode(BTRFS_I(inode
));
2973 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
2978 btrfs_free_io_failure_record(BTRFS_I(inode
),
2979 ordered_extent
->file_offset
,
2980 ordered_extent
->file_offset
+
2981 ordered_extent
->len
- 1);
2983 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered_extent
->flags
)) {
2985 logical_len
= ordered_extent
->truncated_len
;
2986 /* Truncated the entire extent, don't bother adding */
2991 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
2992 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
2995 * For mwrite(mmap + memset to write) case, we still reserve
2996 * space for NOCOW range.
2997 * As NOCOW won't cause a new delayed ref, just free the space
2999 btrfs_qgroup_free_data(inode
, NULL
, ordered_extent
->file_offset
,
3000 ordered_extent
->len
);
3001 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
3003 trans
= btrfs_join_transaction_nolock(root
);
3005 trans
= btrfs_join_transaction(root
);
3006 if (IS_ERR(trans
)) {
3007 ret
= PTR_ERR(trans
);
3011 trans
->block_rsv
= &BTRFS_I(inode
)->block_rsv
;
3012 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
3013 if (ret
) /* -ENOMEM or corruption */
3014 btrfs_abort_transaction(trans
, ret
);
3018 range_locked
= true;
3019 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
3020 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
3023 ret
= test_range_bit(io_tree
, ordered_extent
->file_offset
,
3024 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
3025 EXTENT_DEFRAG
, 0, cached_state
);
3027 u64 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
3028 if (0 && last_snapshot
>= BTRFS_I(inode
)->generation
)
3029 /* the inode is shared */
3030 new = record_old_file_extents(inode
, ordered_extent
);
3032 clear_extent_bit(io_tree
, ordered_extent
->file_offset
,
3033 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
3034 EXTENT_DEFRAG
, 0, 0, &cached_state
, GFP_NOFS
);
3038 trans
= btrfs_join_transaction_nolock(root
);
3040 trans
= btrfs_join_transaction(root
);
3041 if (IS_ERR(trans
)) {
3042 ret
= PTR_ERR(trans
);
3047 trans
->block_rsv
= &BTRFS_I(inode
)->block_rsv
;
3049 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
3050 compress_type
= ordered_extent
->compress_type
;
3051 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
3052 BUG_ON(compress_type
);
3053 btrfs_qgroup_free_data(inode
, NULL
, ordered_extent
->file_offset
,
3054 ordered_extent
->len
);
3055 ret
= btrfs_mark_extent_written(trans
, BTRFS_I(inode
),
3056 ordered_extent
->file_offset
,
3057 ordered_extent
->file_offset
+
3060 BUG_ON(root
== fs_info
->tree_root
);
3061 ret
= insert_reserved_file_extent(trans
, inode
,
3062 ordered_extent
->file_offset
,
3063 ordered_extent
->start
,
3064 ordered_extent
->disk_len
,
3065 logical_len
, logical_len
,
3066 compress_type
, 0, 0,
3067 BTRFS_FILE_EXTENT_REG
);
3069 btrfs_release_delalloc_bytes(fs_info
,
3070 ordered_extent
->start
,
3071 ordered_extent
->disk_len
);
3073 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
3074 ordered_extent
->file_offset
, ordered_extent
->len
,
3077 btrfs_abort_transaction(trans
, ret
);
3081 add_pending_csums(trans
, inode
, &ordered_extent
->list
);
3083 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
3084 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
3085 if (ret
) { /* -ENOMEM or corruption */
3086 btrfs_abort_transaction(trans
, ret
);
3091 if (range_locked
|| clear_new_delalloc_bytes
) {
3092 unsigned int clear_bits
= 0;
3095 clear_bits
|= EXTENT_LOCKED
;
3096 if (clear_new_delalloc_bytes
)
3097 clear_bits
|= EXTENT_DELALLOC_NEW
;
3098 clear_extent_bit(&BTRFS_I(inode
)->io_tree
,
3099 ordered_extent
->file_offset
,
3100 ordered_extent
->file_offset
+
3101 ordered_extent
->len
- 1,
3103 (clear_bits
& EXTENT_LOCKED
) ? 1 : 0,
3104 0, &cached_state
, GFP_NOFS
);
3108 btrfs_end_transaction(trans
);
3110 if (ret
|| truncated
) {
3114 start
= ordered_extent
->file_offset
+ logical_len
;
3116 start
= ordered_extent
->file_offset
;
3117 end
= ordered_extent
->file_offset
+ ordered_extent
->len
- 1;
3118 clear_extent_uptodate(io_tree
, start
, end
, NULL
, GFP_NOFS
);
3120 /* Drop the cache for the part of the extent we didn't write. */
3121 btrfs_drop_extent_cache(BTRFS_I(inode
), start
, end
, 0);
3124 * If the ordered extent had an IOERR or something else went
3125 * wrong we need to return the space for this ordered extent
3126 * back to the allocator. We only free the extent in the
3127 * truncated case if we didn't write out the extent at all.
3129 if ((ret
|| !logical_len
) &&
3130 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
3131 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
))
3132 btrfs_free_reserved_extent(fs_info
,
3133 ordered_extent
->start
,
3134 ordered_extent
->disk_len
, 1);
3139 * This needs to be done to make sure anybody waiting knows we are done
3140 * updating everything for this ordered extent.
3142 btrfs_remove_ordered_extent(inode
, ordered_extent
);
3144 /* for snapshot-aware defrag */
3147 free_sa_defrag_extent(new);
3148 atomic_dec(&fs_info
->defrag_running
);
3150 relink_file_extents(new);
3155 btrfs_put_ordered_extent(ordered_extent
);
3156 /* once for the tree */
3157 btrfs_put_ordered_extent(ordered_extent
);
3162 static void finish_ordered_fn(struct btrfs_work
*work
)
3164 struct btrfs_ordered_extent
*ordered_extent
;
3165 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
3166 btrfs_finish_ordered_io(ordered_extent
);
3169 static void btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
3170 struct extent_state
*state
, int uptodate
)
3172 struct inode
*inode
= page
->mapping
->host
;
3173 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3174 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
3175 struct btrfs_workqueue
*wq
;
3176 btrfs_work_func_t func
;
3178 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
3180 ClearPagePrivate2(page
);
3181 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
3182 end
- start
+ 1, uptodate
))
3185 if (btrfs_is_free_space_inode(BTRFS_I(inode
))) {
3186 wq
= fs_info
->endio_freespace_worker
;
3187 func
= btrfs_freespace_write_helper
;
3189 wq
= fs_info
->endio_write_workers
;
3190 func
= btrfs_endio_write_helper
;
3193 btrfs_init_work(&ordered_extent
->work
, func
, finish_ordered_fn
, NULL
,
3195 btrfs_queue_work(wq
, &ordered_extent
->work
);
3198 static int __readpage_endio_check(struct inode
*inode
,
3199 struct btrfs_io_bio
*io_bio
,
3200 int icsum
, struct page
*page
,
3201 int pgoff
, u64 start
, size_t len
)
3207 csum_expected
= *(((u32
*)io_bio
->csum
) + icsum
);
3209 kaddr
= kmap_atomic(page
);
3210 csum
= btrfs_csum_data(kaddr
+ pgoff
, csum
, len
);
3211 btrfs_csum_final(csum
, (u8
*)&csum
);
3212 if (csum
!= csum_expected
)
3215 kunmap_atomic(kaddr
);
3218 btrfs_print_data_csum_error(BTRFS_I(inode
), start
, csum
, csum_expected
,
3219 io_bio
->mirror_num
);
3220 memset(kaddr
+ pgoff
, 1, len
);
3221 flush_dcache_page(page
);
3222 kunmap_atomic(kaddr
);
3227 * when reads are done, we need to check csums to verify the data is correct
3228 * if there's a match, we allow the bio to finish. If not, the code in
3229 * extent_io.c will try to find good copies for us.
3231 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
3232 u64 phy_offset
, struct page
*page
,
3233 u64 start
, u64 end
, int mirror
)
3235 size_t offset
= start
- page_offset(page
);
3236 struct inode
*inode
= page
->mapping
->host
;
3237 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3238 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3240 if (PageChecked(page
)) {
3241 ClearPageChecked(page
);
3245 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
3248 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
3249 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
3250 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
);
3254 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
3255 return __readpage_endio_check(inode
, io_bio
, phy_offset
, page
, offset
,
3256 start
, (size_t)(end
- start
+ 1));
3259 void btrfs_add_delayed_iput(struct inode
*inode
)
3261 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3262 struct btrfs_inode
*binode
= BTRFS_I(inode
);
3264 if (atomic_add_unless(&inode
->i_count
, -1, 1))
3267 spin_lock(&fs_info
->delayed_iput_lock
);
3268 if (binode
->delayed_iput_count
== 0) {
3269 ASSERT(list_empty(&binode
->delayed_iput
));
3270 list_add_tail(&binode
->delayed_iput
, &fs_info
->delayed_iputs
);
3272 binode
->delayed_iput_count
++;
3274 spin_unlock(&fs_info
->delayed_iput_lock
);
3277 void btrfs_run_delayed_iputs(struct btrfs_fs_info
*fs_info
)
3280 spin_lock(&fs_info
->delayed_iput_lock
);
3281 while (!list_empty(&fs_info
->delayed_iputs
)) {
3282 struct btrfs_inode
*inode
;
3284 inode
= list_first_entry(&fs_info
->delayed_iputs
,
3285 struct btrfs_inode
, delayed_iput
);
3286 if (inode
->delayed_iput_count
) {
3287 inode
->delayed_iput_count
--;
3288 list_move_tail(&inode
->delayed_iput
,
3289 &fs_info
->delayed_iputs
);
3291 list_del_init(&inode
->delayed_iput
);
3293 spin_unlock(&fs_info
->delayed_iput_lock
);
3294 iput(&inode
->vfs_inode
);
3295 spin_lock(&fs_info
->delayed_iput_lock
);
3297 spin_unlock(&fs_info
->delayed_iput_lock
);
3301 * This is called in transaction commit time. If there are no orphan
3302 * files in the subvolume, it removes orphan item and frees block_rsv
3305 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
3306 struct btrfs_root
*root
)
3308 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3309 struct btrfs_block_rsv
*block_rsv
;
3312 if (atomic_read(&root
->orphan_inodes
) ||
3313 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
3316 spin_lock(&root
->orphan_lock
);
3317 if (atomic_read(&root
->orphan_inodes
)) {
3318 spin_unlock(&root
->orphan_lock
);
3322 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
3323 spin_unlock(&root
->orphan_lock
);
3327 block_rsv
= root
->orphan_block_rsv
;
3328 root
->orphan_block_rsv
= NULL
;
3329 spin_unlock(&root
->orphan_lock
);
3331 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
) &&
3332 btrfs_root_refs(&root
->root_item
) > 0) {
3333 ret
= btrfs_del_orphan_item(trans
, fs_info
->tree_root
,
3334 root
->root_key
.objectid
);
3336 btrfs_abort_transaction(trans
, ret
);
3338 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
,
3343 WARN_ON(block_rsv
->size
> 0);
3344 btrfs_free_block_rsv(fs_info
, block_rsv
);
3349 * This creates an orphan entry for the given inode in case something goes
3350 * wrong in the middle of an unlink/truncate.
3352 * NOTE: caller of this function should reserve 5 units of metadata for
3355 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
,
3356 struct btrfs_inode
*inode
)
3358 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
3359 struct btrfs_root
*root
= inode
->root
;
3360 struct btrfs_block_rsv
*block_rsv
= NULL
;
3365 if (!root
->orphan_block_rsv
) {
3366 block_rsv
= btrfs_alloc_block_rsv(fs_info
,
3367 BTRFS_BLOCK_RSV_TEMP
);
3372 spin_lock(&root
->orphan_lock
);
3373 if (!root
->orphan_block_rsv
) {
3374 root
->orphan_block_rsv
= block_rsv
;
3375 } else if (block_rsv
) {
3376 btrfs_free_block_rsv(fs_info
, block_rsv
);
3380 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3381 &inode
->runtime_flags
)) {
3384 * For proper ENOSPC handling, we should do orphan
3385 * cleanup when mounting. But this introduces backward
3386 * compatibility issue.
3388 if (!xchg(&root
->orphan_item_inserted
, 1))
3394 atomic_inc(&root
->orphan_inodes
);
3397 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3398 &inode
->runtime_flags
))
3400 spin_unlock(&root
->orphan_lock
);
3402 /* grab metadata reservation from transaction handle */
3404 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
3408 * dec doesn't need spin_lock as ->orphan_block_rsv
3409 * would be released only if ->orphan_inodes is
3412 atomic_dec(&root
->orphan_inodes
);
3413 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3414 &inode
->runtime_flags
);
3416 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3417 &inode
->runtime_flags
);
3422 /* insert an orphan item to track this unlinked/truncated file */
3424 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
3427 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3428 &inode
->runtime_flags
);
3429 btrfs_orphan_release_metadata(inode
);
3432 * btrfs_orphan_commit_root may race with us and set
3433 * ->orphan_block_rsv to zero, in order to avoid that,
3434 * decrease ->orphan_inodes after everything is done.
3436 atomic_dec(&root
->orphan_inodes
);
3437 if (ret
!= -EEXIST
) {
3438 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3439 &inode
->runtime_flags
);
3440 btrfs_abort_transaction(trans
, ret
);
3447 /* insert an orphan item to track subvolume contains orphan files */
3449 ret
= btrfs_insert_orphan_item(trans
, fs_info
->tree_root
,
3450 root
->root_key
.objectid
);
3451 if (ret
&& ret
!= -EEXIST
) {
3452 btrfs_abort_transaction(trans
, ret
);
3460 * We have done the truncate/delete so we can go ahead and remove the orphan
3461 * item for this particular inode.
3463 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3464 struct btrfs_inode
*inode
)
3466 struct btrfs_root
*root
= inode
->root
;
3467 int delete_item
= 0;
3470 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3471 &inode
->runtime_flags
))
3474 if (delete_item
&& trans
)
3475 ret
= btrfs_del_orphan_item(trans
, root
, btrfs_ino(inode
));
3477 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3478 &inode
->runtime_flags
))
3479 btrfs_orphan_release_metadata(inode
);
3482 * btrfs_orphan_commit_root may race with us and set ->orphan_block_rsv
3483 * to zero, in order to avoid that, decrease ->orphan_inodes after
3484 * everything is done.
3487 atomic_dec(&root
->orphan_inodes
);
3493 * this cleans up any orphans that may be left on the list from the last use
3496 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3498 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3499 struct btrfs_path
*path
;
3500 struct extent_buffer
*leaf
;
3501 struct btrfs_key key
, found_key
;
3502 struct btrfs_trans_handle
*trans
;
3503 struct inode
*inode
;
3504 u64 last_objectid
= 0;
3505 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
3507 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
3510 path
= btrfs_alloc_path();
3515 path
->reada
= READA_BACK
;
3517 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3518 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
3519 key
.offset
= (u64
)-1;
3522 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3527 * if ret == 0 means we found what we were searching for, which
3528 * is weird, but possible, so only screw with path if we didn't
3529 * find the key and see if we have stuff that matches
3533 if (path
->slots
[0] == 0)
3538 /* pull out the item */
3539 leaf
= path
->nodes
[0];
3540 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3542 /* make sure the item matches what we want */
3543 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3545 if (found_key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
3548 /* release the path since we're done with it */
3549 btrfs_release_path(path
);
3552 * this is where we are basically btrfs_lookup, without the
3553 * crossing root thing. we store the inode number in the
3554 * offset of the orphan item.
3557 if (found_key
.offset
== last_objectid
) {
3559 "Error removing orphan entry, stopping orphan cleanup");
3564 last_objectid
= found_key
.offset
;
3566 found_key
.objectid
= found_key
.offset
;
3567 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3568 found_key
.offset
= 0;
3569 inode
= btrfs_iget(fs_info
->sb
, &found_key
, root
, NULL
);
3570 ret
= PTR_ERR_OR_ZERO(inode
);
3571 if (ret
&& ret
!= -ENOENT
)
3574 if (ret
== -ENOENT
&& root
== fs_info
->tree_root
) {
3575 struct btrfs_root
*dead_root
;
3576 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3577 int is_dead_root
= 0;
3580 * this is an orphan in the tree root. Currently these
3581 * could come from 2 sources:
3582 * a) a snapshot deletion in progress
3583 * b) a free space cache inode
3584 * We need to distinguish those two, as the snapshot
3585 * orphan must not get deleted.
3586 * find_dead_roots already ran before us, so if this
3587 * is a snapshot deletion, we should find the root
3588 * in the dead_roots list
3590 spin_lock(&fs_info
->trans_lock
);
3591 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
3593 if (dead_root
->root_key
.objectid
==
3594 found_key
.objectid
) {
3599 spin_unlock(&fs_info
->trans_lock
);
3601 /* prevent this orphan from being found again */
3602 key
.offset
= found_key
.objectid
- 1;
3607 * Inode is already gone but the orphan item is still there,
3608 * kill the orphan item.
3610 if (ret
== -ENOENT
) {
3611 trans
= btrfs_start_transaction(root
, 1);
3612 if (IS_ERR(trans
)) {
3613 ret
= PTR_ERR(trans
);
3616 btrfs_debug(fs_info
, "auto deleting %Lu",
3617 found_key
.objectid
);
3618 ret
= btrfs_del_orphan_item(trans
, root
,
3619 found_key
.objectid
);
3620 btrfs_end_transaction(trans
);
3627 * add this inode to the orphan list so btrfs_orphan_del does
3628 * the proper thing when we hit it
3630 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3631 &BTRFS_I(inode
)->runtime_flags
);
3632 atomic_inc(&root
->orphan_inodes
);
3634 /* if we have links, this was a truncate, lets do that */
3635 if (inode
->i_nlink
) {
3636 if (WARN_ON(!S_ISREG(inode
->i_mode
))) {
3642 /* 1 for the orphan item deletion. */
3643 trans
= btrfs_start_transaction(root
, 1);
3644 if (IS_ERR(trans
)) {
3646 ret
= PTR_ERR(trans
);
3649 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
3650 btrfs_end_transaction(trans
);
3656 ret
= btrfs_truncate(inode
);
3658 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
3663 /* this will do delete_inode and everything for us */
3668 /* release the path since we're done with it */
3669 btrfs_release_path(path
);
3671 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
3673 if (root
->orphan_block_rsv
)
3674 btrfs_block_rsv_release(fs_info
, root
->orphan_block_rsv
,
3677 if (root
->orphan_block_rsv
||
3678 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
)) {
3679 trans
= btrfs_join_transaction(root
);
3681 btrfs_end_transaction(trans
);
3685 btrfs_debug(fs_info
, "unlinked %d orphans", nr_unlink
);
3687 btrfs_debug(fs_info
, "truncated %d orphans", nr_truncate
);
3691 btrfs_err(fs_info
, "could not do orphan cleanup %d", ret
);
3692 btrfs_free_path(path
);
3697 * very simple check to peek ahead in the leaf looking for xattrs. If we
3698 * don't find any xattrs, we know there can't be any acls.
3700 * slot is the slot the inode is in, objectid is the objectid of the inode
3702 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3703 int slot
, u64 objectid
,
3704 int *first_xattr_slot
)
3706 u32 nritems
= btrfs_header_nritems(leaf
);
3707 struct btrfs_key found_key
;
3708 static u64 xattr_access
= 0;
3709 static u64 xattr_default
= 0;
3712 if (!xattr_access
) {
3713 xattr_access
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS
,
3714 strlen(XATTR_NAME_POSIX_ACL_ACCESS
));
3715 xattr_default
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT
,
3716 strlen(XATTR_NAME_POSIX_ACL_DEFAULT
));
3720 *first_xattr_slot
= -1;
3721 while (slot
< nritems
) {
3722 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3724 /* we found a different objectid, there must not be acls */
3725 if (found_key
.objectid
!= objectid
)
3728 /* we found an xattr, assume we've got an acl */
3729 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
3730 if (*first_xattr_slot
== -1)
3731 *first_xattr_slot
= slot
;
3732 if (found_key
.offset
== xattr_access
||
3733 found_key
.offset
== xattr_default
)
3738 * we found a key greater than an xattr key, there can't
3739 * be any acls later on
3741 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3748 * it goes inode, inode backrefs, xattrs, extents,
3749 * so if there are a ton of hard links to an inode there can
3750 * be a lot of backrefs. Don't waste time searching too hard,
3751 * this is just an optimization
3756 /* we hit the end of the leaf before we found an xattr or
3757 * something larger than an xattr. We have to assume the inode
3760 if (*first_xattr_slot
== -1)
3761 *first_xattr_slot
= slot
;
3766 * read an inode from the btree into the in-memory inode
3768 static int btrfs_read_locked_inode(struct inode
*inode
)
3770 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3771 struct btrfs_path
*path
;
3772 struct extent_buffer
*leaf
;
3773 struct btrfs_inode_item
*inode_item
;
3774 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3775 struct btrfs_key location
;
3780 bool filled
= false;
3781 int first_xattr_slot
;
3783 ret
= btrfs_fill_inode(inode
, &rdev
);
3787 path
= btrfs_alloc_path();
3793 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3795 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3802 leaf
= path
->nodes
[0];
3807 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3808 struct btrfs_inode_item
);
3809 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3810 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3811 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3812 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3813 btrfs_i_size_write(BTRFS_I(inode
), btrfs_inode_size(leaf
, inode_item
));
3815 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->atime
);
3816 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->atime
);
3818 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->mtime
);
3819 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->mtime
);
3821 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->ctime
);
3822 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->ctime
);
3824 BTRFS_I(inode
)->i_otime
.tv_sec
=
3825 btrfs_timespec_sec(leaf
, &inode_item
->otime
);
3826 BTRFS_I(inode
)->i_otime
.tv_nsec
=
3827 btrfs_timespec_nsec(leaf
, &inode_item
->otime
);
3829 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3830 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3831 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3833 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
3834 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3836 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3838 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3839 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
3843 * If we were modified in the current generation and evicted from memory
3844 * and then re-read we need to do a full sync since we don't have any
3845 * idea about which extents were modified before we were evicted from
3848 * This is required for both inode re-read from disk and delayed inode
3849 * in delayed_nodes_tree.
3851 if (BTRFS_I(inode
)->last_trans
== fs_info
->generation
)
3852 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3853 &BTRFS_I(inode
)->runtime_flags
);
3856 * We don't persist the id of the transaction where an unlink operation
3857 * against the inode was last made. So here we assume the inode might
3858 * have been evicted, and therefore the exact value of last_unlink_trans
3859 * lost, and set it to last_trans to avoid metadata inconsistencies
3860 * between the inode and its parent if the inode is fsync'ed and the log
3861 * replayed. For example, in the scenario:
3864 * ln mydir/foo mydir/bar
3867 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3868 * xfs_io -c fsync mydir/foo
3870 * mount fs, triggers fsync log replay
3872 * We must make sure that when we fsync our inode foo we also log its
3873 * parent inode, otherwise after log replay the parent still has the
3874 * dentry with the "bar" name but our inode foo has a link count of 1
3875 * and doesn't have an inode ref with the name "bar" anymore.
3877 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3878 * but it guarantees correctness at the expense of occasional full
3879 * transaction commits on fsync if our inode is a directory, or if our
3880 * inode is not a directory, logging its parent unnecessarily.
3882 BTRFS_I(inode
)->last_unlink_trans
= BTRFS_I(inode
)->last_trans
;
3885 if (inode
->i_nlink
!= 1 ||
3886 path
->slots
[0] >= btrfs_header_nritems(leaf
))
3889 btrfs_item_key_to_cpu(leaf
, &location
, path
->slots
[0]);
3890 if (location
.objectid
!= btrfs_ino(BTRFS_I(inode
)))
3893 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3894 if (location
.type
== BTRFS_INODE_REF_KEY
) {
3895 struct btrfs_inode_ref
*ref
;
3897 ref
= (struct btrfs_inode_ref
*)ptr
;
3898 BTRFS_I(inode
)->dir_index
= btrfs_inode_ref_index(leaf
, ref
);
3899 } else if (location
.type
== BTRFS_INODE_EXTREF_KEY
) {
3900 struct btrfs_inode_extref
*extref
;
3902 extref
= (struct btrfs_inode_extref
*)ptr
;
3903 BTRFS_I(inode
)->dir_index
= btrfs_inode_extref_index(leaf
,
3908 * try to precache a NULL acl entry for files that don't have
3909 * any xattrs or acls
3911 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
3912 btrfs_ino(BTRFS_I(inode
)), &first_xattr_slot
);
3913 if (first_xattr_slot
!= -1) {
3914 path
->slots
[0] = first_xattr_slot
;
3915 ret
= btrfs_load_inode_props(inode
, path
);
3918 "error loading props for ino %llu (root %llu): %d",
3919 btrfs_ino(BTRFS_I(inode
)),
3920 root
->root_key
.objectid
, ret
);
3922 btrfs_free_path(path
);
3925 cache_no_acl(inode
);
3927 switch (inode
->i_mode
& S_IFMT
) {
3929 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3930 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3931 inode
->i_fop
= &btrfs_file_operations
;
3932 inode
->i_op
= &btrfs_file_inode_operations
;
3935 inode
->i_fop
= &btrfs_dir_file_operations
;
3936 inode
->i_op
= &btrfs_dir_inode_operations
;
3939 inode
->i_op
= &btrfs_symlink_inode_operations
;
3940 inode_nohighmem(inode
);
3941 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
3944 inode
->i_op
= &btrfs_special_inode_operations
;
3945 init_special_inode(inode
, inode
->i_mode
, rdev
);
3949 btrfs_update_iflags(inode
);
3953 btrfs_free_path(path
);
3954 make_bad_inode(inode
);
3959 * given a leaf and an inode, copy the inode fields into the leaf
3961 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3962 struct extent_buffer
*leaf
,
3963 struct btrfs_inode_item
*item
,
3964 struct inode
*inode
)
3966 struct btrfs_map_token token
;
3968 btrfs_init_map_token(&token
);
3970 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3971 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3972 btrfs_set_token_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
,
3974 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3975 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3977 btrfs_set_token_timespec_sec(leaf
, &item
->atime
,
3978 inode
->i_atime
.tv_sec
, &token
);
3979 btrfs_set_token_timespec_nsec(leaf
, &item
->atime
,
3980 inode
->i_atime
.tv_nsec
, &token
);
3982 btrfs_set_token_timespec_sec(leaf
, &item
->mtime
,
3983 inode
->i_mtime
.tv_sec
, &token
);
3984 btrfs_set_token_timespec_nsec(leaf
, &item
->mtime
,
3985 inode
->i_mtime
.tv_nsec
, &token
);
3987 btrfs_set_token_timespec_sec(leaf
, &item
->ctime
,
3988 inode
->i_ctime
.tv_sec
, &token
);
3989 btrfs_set_token_timespec_nsec(leaf
, &item
->ctime
,
3990 inode
->i_ctime
.tv_nsec
, &token
);
3992 btrfs_set_token_timespec_sec(leaf
, &item
->otime
,
3993 BTRFS_I(inode
)->i_otime
.tv_sec
, &token
);
3994 btrfs_set_token_timespec_nsec(leaf
, &item
->otime
,
3995 BTRFS_I(inode
)->i_otime
.tv_nsec
, &token
);
3997 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3999 btrfs_set_token_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
,
4001 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
4002 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
4003 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
4004 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
4005 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
4009 * copy everything in the in-memory inode into the btree.
4011 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
4012 struct btrfs_root
*root
, struct inode
*inode
)
4014 struct btrfs_inode_item
*inode_item
;
4015 struct btrfs_path
*path
;
4016 struct extent_buffer
*leaf
;
4019 path
= btrfs_alloc_path();
4023 path
->leave_spinning
= 1;
4024 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
4032 leaf
= path
->nodes
[0];
4033 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4034 struct btrfs_inode_item
);
4036 fill_inode_item(trans
, leaf
, inode_item
, inode
);
4037 btrfs_mark_buffer_dirty(leaf
);
4038 btrfs_set_inode_last_trans(trans
, inode
);
4041 btrfs_free_path(path
);
4046 * copy everything in the in-memory inode into the btree.
4048 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
4049 struct btrfs_root
*root
, struct inode
*inode
)
4051 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4055 * If the inode is a free space inode, we can deadlock during commit
4056 * if we put it into the delayed code.
4058 * The data relocation inode should also be directly updated
4061 if (!btrfs_is_free_space_inode(BTRFS_I(inode
))
4062 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
4063 && !test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
)) {
4064 btrfs_update_root_times(trans
, root
);
4066 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
4068 btrfs_set_inode_last_trans(trans
, inode
);
4072 return btrfs_update_inode_item(trans
, root
, inode
);
4075 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
4076 struct btrfs_root
*root
,
4077 struct inode
*inode
)
4081 ret
= btrfs_update_inode(trans
, root
, inode
);
4083 return btrfs_update_inode_item(trans
, root
, inode
);
4088 * unlink helper that gets used here in inode.c and in the tree logging
4089 * recovery code. It remove a link in a directory with a given name, and
4090 * also drops the back refs in the inode to the directory
4092 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4093 struct btrfs_root
*root
,
4094 struct btrfs_inode
*dir
,
4095 struct btrfs_inode
*inode
,
4096 const char *name
, int name_len
)
4098 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4099 struct btrfs_path
*path
;
4101 struct extent_buffer
*leaf
;
4102 struct btrfs_dir_item
*di
;
4103 struct btrfs_key key
;
4105 u64 ino
= btrfs_ino(inode
);
4106 u64 dir_ino
= btrfs_ino(dir
);
4108 path
= btrfs_alloc_path();
4114 path
->leave_spinning
= 1;
4115 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4116 name
, name_len
, -1);
4125 leaf
= path
->nodes
[0];
4126 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4127 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4130 btrfs_release_path(path
);
4133 * If we don't have dir index, we have to get it by looking up
4134 * the inode ref, since we get the inode ref, remove it directly,
4135 * it is unnecessary to do delayed deletion.
4137 * But if we have dir index, needn't search inode ref to get it.
4138 * Since the inode ref is close to the inode item, it is better
4139 * that we delay to delete it, and just do this deletion when
4140 * we update the inode item.
4142 if (inode
->dir_index
) {
4143 ret
= btrfs_delayed_delete_inode_ref(inode
);
4145 index
= inode
->dir_index
;
4150 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
4154 "failed to delete reference to %.*s, inode %llu parent %llu",
4155 name_len
, name
, ino
, dir_ino
);
4156 btrfs_abort_transaction(trans
, ret
);
4160 ret
= btrfs_delete_delayed_dir_index(trans
, fs_info
, dir
, index
);
4162 btrfs_abort_transaction(trans
, ret
);
4166 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
, inode
,
4168 if (ret
!= 0 && ret
!= -ENOENT
) {
4169 btrfs_abort_transaction(trans
, ret
);
4173 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
, dir
,
4178 btrfs_abort_transaction(trans
, ret
);
4180 btrfs_free_path(path
);
4184 btrfs_i_size_write(dir
, dir
->vfs_inode
.i_size
- name_len
* 2);
4185 inode_inc_iversion(&inode
->vfs_inode
);
4186 inode_inc_iversion(&dir
->vfs_inode
);
4187 inode
->vfs_inode
.i_ctime
= dir
->vfs_inode
.i_mtime
=
4188 dir
->vfs_inode
.i_ctime
= current_time(&inode
->vfs_inode
);
4189 ret
= btrfs_update_inode(trans
, root
, &dir
->vfs_inode
);
4194 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4195 struct btrfs_root
*root
,
4196 struct btrfs_inode
*dir
, struct btrfs_inode
*inode
,
4197 const char *name
, int name_len
)
4200 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
4202 drop_nlink(&inode
->vfs_inode
);
4203 ret
= btrfs_update_inode(trans
, root
, &inode
->vfs_inode
);
4209 * helper to start transaction for unlink and rmdir.
4211 * unlink and rmdir are special in btrfs, they do not always free space, so
4212 * if we cannot make our reservations the normal way try and see if there is
4213 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4214 * allow the unlink to occur.
4216 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
)
4218 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4221 * 1 for the possible orphan item
4222 * 1 for the dir item
4223 * 1 for the dir index
4224 * 1 for the inode ref
4227 return btrfs_start_transaction_fallback_global_rsv(root
, 5, 5);
4230 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
4232 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4233 struct btrfs_trans_handle
*trans
;
4234 struct inode
*inode
= d_inode(dentry
);
4237 trans
= __unlink_start_trans(dir
);
4239 return PTR_ERR(trans
);
4241 btrfs_record_unlink_dir(trans
, BTRFS_I(dir
), BTRFS_I(d_inode(dentry
)),
4244 ret
= btrfs_unlink_inode(trans
, root
, BTRFS_I(dir
),
4245 BTRFS_I(d_inode(dentry
)), dentry
->d_name
.name
,
4246 dentry
->d_name
.len
);
4250 if (inode
->i_nlink
== 0) {
4251 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
4257 btrfs_end_transaction(trans
);
4258 btrfs_btree_balance_dirty(root
->fs_info
);
4262 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
4263 struct btrfs_root
*root
,
4264 struct inode
*dir
, u64 objectid
,
4265 const char *name
, int name_len
)
4267 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4268 struct btrfs_path
*path
;
4269 struct extent_buffer
*leaf
;
4270 struct btrfs_dir_item
*di
;
4271 struct btrfs_key key
;
4274 u64 dir_ino
= btrfs_ino(BTRFS_I(dir
));
4276 path
= btrfs_alloc_path();
4280 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4281 name
, name_len
, -1);
4282 if (IS_ERR_OR_NULL(di
)) {
4290 leaf
= path
->nodes
[0];
4291 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4292 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
4293 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4295 btrfs_abort_transaction(trans
, ret
);
4298 btrfs_release_path(path
);
4300 ret
= btrfs_del_root_ref(trans
, fs_info
, objectid
,
4301 root
->root_key
.objectid
, dir_ino
,
4302 &index
, name
, name_len
);
4304 if (ret
!= -ENOENT
) {
4305 btrfs_abort_transaction(trans
, ret
);
4308 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
4310 if (IS_ERR_OR_NULL(di
)) {
4315 btrfs_abort_transaction(trans
, ret
);
4319 leaf
= path
->nodes
[0];
4320 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4321 btrfs_release_path(path
);
4324 btrfs_release_path(path
);
4326 ret
= btrfs_delete_delayed_dir_index(trans
, fs_info
, BTRFS_I(dir
), index
);
4328 btrfs_abort_transaction(trans
, ret
);
4332 btrfs_i_size_write(BTRFS_I(dir
), dir
->i_size
- name_len
* 2);
4333 inode_inc_iversion(dir
);
4334 dir
->i_mtime
= dir
->i_ctime
= current_time(dir
);
4335 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
4337 btrfs_abort_transaction(trans
, ret
);
4339 btrfs_free_path(path
);
4343 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
4345 struct inode
*inode
= d_inode(dentry
);
4347 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4348 struct btrfs_trans_handle
*trans
;
4349 u64 last_unlink_trans
;
4351 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
4353 if (btrfs_ino(BTRFS_I(inode
)) == BTRFS_FIRST_FREE_OBJECTID
)
4356 trans
= __unlink_start_trans(dir
);
4358 return PTR_ERR(trans
);
4360 if (unlikely(btrfs_ino(BTRFS_I(inode
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
4361 err
= btrfs_unlink_subvol(trans
, root
, dir
,
4362 BTRFS_I(inode
)->location
.objectid
,
4363 dentry
->d_name
.name
,
4364 dentry
->d_name
.len
);
4368 err
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
4372 last_unlink_trans
= BTRFS_I(inode
)->last_unlink_trans
;
4374 /* now the directory is empty */
4375 err
= btrfs_unlink_inode(trans
, root
, BTRFS_I(dir
),
4376 BTRFS_I(d_inode(dentry
)), dentry
->d_name
.name
,
4377 dentry
->d_name
.len
);
4379 btrfs_i_size_write(BTRFS_I(inode
), 0);
4381 * Propagate the last_unlink_trans value of the deleted dir to
4382 * its parent directory. This is to prevent an unrecoverable
4383 * log tree in the case we do something like this:
4385 * 2) create snapshot under dir foo
4386 * 3) delete the snapshot
4389 * 6) fsync foo or some file inside foo
4391 if (last_unlink_trans
>= trans
->transid
)
4392 BTRFS_I(dir
)->last_unlink_trans
= last_unlink_trans
;
4395 btrfs_end_transaction(trans
);
4396 btrfs_btree_balance_dirty(root
->fs_info
);
4401 static int truncate_space_check(struct btrfs_trans_handle
*trans
,
4402 struct btrfs_root
*root
,
4405 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4409 * This is only used to apply pressure to the enospc system, we don't
4410 * intend to use this reservation at all.
4412 bytes_deleted
= btrfs_csum_bytes_to_leaves(fs_info
, bytes_deleted
);
4413 bytes_deleted
*= fs_info
->nodesize
;
4414 ret
= btrfs_block_rsv_add(root
, &fs_info
->trans_block_rsv
,
4415 bytes_deleted
, BTRFS_RESERVE_NO_FLUSH
);
4417 trace_btrfs_space_reservation(fs_info
, "transaction",
4420 trans
->bytes_reserved
+= bytes_deleted
;
4427 * Return this if we need to call truncate_block for the last bit of the
4430 #define NEED_TRUNCATE_BLOCK 1
4433 * this can truncate away extent items, csum items and directory items.
4434 * It starts at a high offset and removes keys until it can't find
4435 * any higher than new_size
4437 * csum items that cross the new i_size are truncated to the new size
4440 * min_type is the minimum key type to truncate down to. If set to 0, this
4441 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4443 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
4444 struct btrfs_root
*root
,
4445 struct inode
*inode
,
4446 u64 new_size
, u32 min_type
)
4448 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4449 struct btrfs_path
*path
;
4450 struct extent_buffer
*leaf
;
4451 struct btrfs_file_extent_item
*fi
;
4452 struct btrfs_key key
;
4453 struct btrfs_key found_key
;
4454 u64 extent_start
= 0;
4455 u64 extent_num_bytes
= 0;
4456 u64 extent_offset
= 0;
4458 u64 last_size
= new_size
;
4459 u32 found_type
= (u8
)-1;
4462 int pending_del_nr
= 0;
4463 int pending_del_slot
= 0;
4464 int extent_type
= -1;
4467 u64 ino
= btrfs_ino(BTRFS_I(inode
));
4468 u64 bytes_deleted
= 0;
4469 bool be_nice
= false;
4470 bool should_throttle
= false;
4471 bool should_end
= false;
4473 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
4476 * for non-free space inodes and ref cows, we want to back off from
4479 if (!btrfs_is_free_space_inode(BTRFS_I(inode
)) &&
4480 test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4483 path
= btrfs_alloc_path();
4486 path
->reada
= READA_BACK
;
4489 * We want to drop from the next block forward in case this new size is
4490 * not block aligned since we will be keeping the last block of the
4491 * extent just the way it is.
4493 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4494 root
== fs_info
->tree_root
)
4495 btrfs_drop_extent_cache(BTRFS_I(inode
), ALIGN(new_size
,
4496 fs_info
->sectorsize
),
4500 * This function is also used to drop the items in the log tree before
4501 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4502 * it is used to drop the loged items. So we shouldn't kill the delayed
4505 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
4506 btrfs_kill_delayed_inode_items(BTRFS_I(inode
));
4509 key
.offset
= (u64
)-1;
4514 * with a 16K leaf size and 128MB extents, you can actually queue
4515 * up a huge file in a single leaf. Most of the time that
4516 * bytes_deleted is > 0, it will be huge by the time we get here
4518 if (be_nice
&& bytes_deleted
> SZ_32M
) {
4519 if (btrfs_should_end_transaction(trans
)) {
4526 path
->leave_spinning
= 1;
4527 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
4534 /* there are no items in the tree for us to truncate, we're
4537 if (path
->slots
[0] == 0)
4544 leaf
= path
->nodes
[0];
4545 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4546 found_type
= found_key
.type
;
4548 if (found_key
.objectid
!= ino
)
4551 if (found_type
< min_type
)
4554 item_end
= found_key
.offset
;
4555 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
4556 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4557 struct btrfs_file_extent_item
);
4558 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4559 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4561 btrfs_file_extent_num_bytes(leaf
, fi
);
4563 trace_btrfs_truncate_show_fi_regular(
4564 BTRFS_I(inode
), leaf
, fi
,
4566 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4567 item_end
+= btrfs_file_extent_inline_len(leaf
,
4568 path
->slots
[0], fi
);
4570 trace_btrfs_truncate_show_fi_inline(
4571 BTRFS_I(inode
), leaf
, fi
, path
->slots
[0],
4576 if (found_type
> min_type
) {
4579 if (item_end
< new_size
)
4581 if (found_key
.offset
>= new_size
)
4587 /* FIXME, shrink the extent if the ref count is only 1 */
4588 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
4591 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4593 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4595 u64 orig_num_bytes
=
4596 btrfs_file_extent_num_bytes(leaf
, fi
);
4597 extent_num_bytes
= ALIGN(new_size
-
4599 fs_info
->sectorsize
);
4600 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4602 num_dec
= (orig_num_bytes
-
4604 if (test_bit(BTRFS_ROOT_REF_COWS
,
4607 inode_sub_bytes(inode
, num_dec
);
4608 btrfs_mark_buffer_dirty(leaf
);
4611 btrfs_file_extent_disk_num_bytes(leaf
,
4613 extent_offset
= found_key
.offset
-
4614 btrfs_file_extent_offset(leaf
, fi
);
4616 /* FIXME blocksize != 4096 */
4617 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
4618 if (extent_start
!= 0) {
4620 if (test_bit(BTRFS_ROOT_REF_COWS
,
4622 inode_sub_bytes(inode
, num_dec
);
4625 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4627 * we can't truncate inline items that have had
4631 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
4632 btrfs_file_extent_other_encoding(leaf
, fi
) == 0 &&
4633 btrfs_file_extent_compression(leaf
, fi
) == 0) {
4634 u32 size
= (u32
)(new_size
- found_key
.offset
);
4636 btrfs_set_file_extent_ram_bytes(leaf
, fi
, size
);
4637 size
= btrfs_file_extent_calc_inline_size(size
);
4638 btrfs_truncate_item(root
->fs_info
, path
, size
, 1);
4639 } else if (!del_item
) {
4641 * We have to bail so the last_size is set to
4642 * just before this extent.
4644 err
= NEED_TRUNCATE_BLOCK
;
4648 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4649 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4653 last_size
= found_key
.offset
;
4655 last_size
= new_size
;
4657 if (!pending_del_nr
) {
4658 /* no pending yet, add ourselves */
4659 pending_del_slot
= path
->slots
[0];
4661 } else if (pending_del_nr
&&
4662 path
->slots
[0] + 1 == pending_del_slot
) {
4663 /* hop on the pending chunk */
4665 pending_del_slot
= path
->slots
[0];
4672 should_throttle
= false;
4675 (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4676 root
== fs_info
->tree_root
)) {
4677 btrfs_set_path_blocking(path
);
4678 bytes_deleted
+= extent_num_bytes
;
4679 ret
= btrfs_free_extent(trans
, root
, extent_start
,
4680 extent_num_bytes
, 0,
4681 btrfs_header_owner(leaf
),
4682 ino
, extent_offset
);
4684 if (btrfs_should_throttle_delayed_refs(trans
, fs_info
))
4685 btrfs_async_run_delayed_refs(fs_info
,
4686 trans
->delayed_ref_updates
* 2,
4689 if (truncate_space_check(trans
, root
,
4690 extent_num_bytes
)) {
4693 if (btrfs_should_throttle_delayed_refs(trans
,
4695 should_throttle
= true;
4699 if (found_type
== BTRFS_INODE_ITEM_KEY
)
4702 if (path
->slots
[0] == 0 ||
4703 path
->slots
[0] != pending_del_slot
||
4704 should_throttle
|| should_end
) {
4705 if (pending_del_nr
) {
4706 ret
= btrfs_del_items(trans
, root
, path
,
4710 btrfs_abort_transaction(trans
, ret
);
4715 btrfs_release_path(path
);
4716 if (should_throttle
) {
4717 unsigned long updates
= trans
->delayed_ref_updates
;
4719 trans
->delayed_ref_updates
= 0;
4720 ret
= btrfs_run_delayed_refs(trans
,
4728 * if we failed to refill our space rsv, bail out
4729 * and let the transaction restart
4741 if (pending_del_nr
) {
4742 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
4745 btrfs_abort_transaction(trans
, ret
);
4748 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4749 ASSERT(last_size
>= new_size
);
4750 if (!err
&& last_size
> new_size
)
4751 last_size
= new_size
;
4752 btrfs_ordered_update_i_size(inode
, last_size
, NULL
);
4755 btrfs_free_path(path
);
4757 if (be_nice
&& bytes_deleted
> SZ_32M
) {
4758 unsigned long updates
= trans
->delayed_ref_updates
;
4760 trans
->delayed_ref_updates
= 0;
4761 ret
= btrfs_run_delayed_refs(trans
, fs_info
,
4771 * btrfs_truncate_block - read, zero a chunk and write a block
4772 * @inode - inode that we're zeroing
4773 * @from - the offset to start zeroing
4774 * @len - the length to zero, 0 to zero the entire range respective to the
4776 * @front - zero up to the offset instead of from the offset on
4778 * This will find the block for the "from" offset and cow the block and zero the
4779 * part we want to zero. This is used with truncate and hole punching.
4781 int btrfs_truncate_block(struct inode
*inode
, loff_t from
, loff_t len
,
4784 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4785 struct address_space
*mapping
= inode
->i_mapping
;
4786 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4787 struct btrfs_ordered_extent
*ordered
;
4788 struct extent_state
*cached_state
= NULL
;
4789 struct extent_changeset
*data_reserved
= NULL
;
4791 u32 blocksize
= fs_info
->sectorsize
;
4792 pgoff_t index
= from
>> PAGE_SHIFT
;
4793 unsigned offset
= from
& (blocksize
- 1);
4795 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4800 if ((offset
& (blocksize
- 1)) == 0 &&
4801 (!len
|| ((len
& (blocksize
- 1)) == 0)))
4804 block_start
= round_down(from
, blocksize
);
4805 block_end
= block_start
+ blocksize
- 1;
4807 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
,
4808 block_start
, blocksize
);
4813 page
= find_or_create_page(mapping
, index
, mask
);
4815 btrfs_delalloc_release_space(inode
, data_reserved
,
4816 block_start
, blocksize
);
4817 btrfs_delalloc_release_extents(BTRFS_I(inode
), blocksize
);
4822 if (!PageUptodate(page
)) {
4823 ret
= btrfs_readpage(NULL
, page
);
4825 if (page
->mapping
!= mapping
) {
4830 if (!PageUptodate(page
)) {
4835 wait_on_page_writeback(page
);
4837 lock_extent_bits(io_tree
, block_start
, block_end
, &cached_state
);
4838 set_page_extent_mapped(page
);
4840 ordered
= btrfs_lookup_ordered_extent(inode
, block_start
);
4842 unlock_extent_cached(io_tree
, block_start
, block_end
,
4843 &cached_state
, GFP_NOFS
);
4846 btrfs_start_ordered_extent(inode
, ordered
, 1);
4847 btrfs_put_ordered_extent(ordered
);
4851 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, block_start
, block_end
,
4852 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4853 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4854 0, 0, &cached_state
, GFP_NOFS
);
4856 ret
= btrfs_set_extent_delalloc(inode
, block_start
, block_end
, 0,
4859 unlock_extent_cached(io_tree
, block_start
, block_end
,
4860 &cached_state
, GFP_NOFS
);
4864 if (offset
!= blocksize
) {
4866 len
= blocksize
- offset
;
4869 memset(kaddr
+ (block_start
- page_offset(page
)),
4872 memset(kaddr
+ (block_start
- page_offset(page
)) + offset
,
4874 flush_dcache_page(page
);
4877 ClearPageChecked(page
);
4878 set_page_dirty(page
);
4879 unlock_extent_cached(io_tree
, block_start
, block_end
, &cached_state
,
4884 btrfs_delalloc_release_space(inode
, data_reserved
, block_start
,
4886 btrfs_delalloc_release_extents(BTRFS_I(inode
), blocksize
);
4890 extent_changeset_free(data_reserved
);
4894 static int maybe_insert_hole(struct btrfs_root
*root
, struct inode
*inode
,
4895 u64 offset
, u64 len
)
4897 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4898 struct btrfs_trans_handle
*trans
;
4902 * Still need to make sure the inode looks like it's been updated so
4903 * that any holes get logged if we fsync.
4905 if (btrfs_fs_incompat(fs_info
, NO_HOLES
)) {
4906 BTRFS_I(inode
)->last_trans
= fs_info
->generation
;
4907 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
4908 BTRFS_I(inode
)->last_log_commit
= root
->last_log_commit
;
4913 * 1 - for the one we're dropping
4914 * 1 - for the one we're adding
4915 * 1 - for updating the inode.
4917 trans
= btrfs_start_transaction(root
, 3);
4919 return PTR_ERR(trans
);
4921 ret
= btrfs_drop_extents(trans
, root
, inode
, offset
, offset
+ len
, 1);
4923 btrfs_abort_transaction(trans
, ret
);
4924 btrfs_end_transaction(trans
);
4928 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(BTRFS_I(inode
)),
4929 offset
, 0, 0, len
, 0, len
, 0, 0, 0);
4931 btrfs_abort_transaction(trans
, ret
);
4933 btrfs_update_inode(trans
, root
, inode
);
4934 btrfs_end_transaction(trans
);
4939 * This function puts in dummy file extents for the area we're creating a hole
4940 * for. So if we are truncating this file to a larger size we need to insert
4941 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4942 * the range between oldsize and size
4944 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
4946 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4947 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4948 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4949 struct extent_map
*em
= NULL
;
4950 struct extent_state
*cached_state
= NULL
;
4951 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4952 u64 hole_start
= ALIGN(oldsize
, fs_info
->sectorsize
);
4953 u64 block_end
= ALIGN(size
, fs_info
->sectorsize
);
4960 * If our size started in the middle of a block we need to zero out the
4961 * rest of the block before we expand the i_size, otherwise we could
4962 * expose stale data.
4964 err
= btrfs_truncate_block(inode
, oldsize
, 0, 0);
4968 if (size
<= hole_start
)
4972 struct btrfs_ordered_extent
*ordered
;
4974 lock_extent_bits(io_tree
, hole_start
, block_end
- 1,
4976 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), hole_start
,
4977 block_end
- hole_start
);
4980 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
4981 &cached_state
, GFP_NOFS
);
4982 btrfs_start_ordered_extent(inode
, ordered
, 1);
4983 btrfs_put_ordered_extent(ordered
);
4986 cur_offset
= hole_start
;
4988 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0, cur_offset
,
4989 block_end
- cur_offset
, 0);
4995 last_byte
= min(extent_map_end(em
), block_end
);
4996 last_byte
= ALIGN(last_byte
, fs_info
->sectorsize
);
4997 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
4998 struct extent_map
*hole_em
;
4999 hole_size
= last_byte
- cur_offset
;
5001 err
= maybe_insert_hole(root
, inode
, cur_offset
,
5005 btrfs_drop_extent_cache(BTRFS_I(inode
), cur_offset
,
5006 cur_offset
+ hole_size
- 1, 0);
5007 hole_em
= alloc_extent_map();
5009 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
5010 &BTRFS_I(inode
)->runtime_flags
);
5013 hole_em
->start
= cur_offset
;
5014 hole_em
->len
= hole_size
;
5015 hole_em
->orig_start
= cur_offset
;
5017 hole_em
->block_start
= EXTENT_MAP_HOLE
;
5018 hole_em
->block_len
= 0;
5019 hole_em
->orig_block_len
= 0;
5020 hole_em
->ram_bytes
= hole_size
;
5021 hole_em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
5022 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
5023 hole_em
->generation
= fs_info
->generation
;
5026 write_lock(&em_tree
->lock
);
5027 err
= add_extent_mapping(em_tree
, hole_em
, 1);
5028 write_unlock(&em_tree
->lock
);
5031 btrfs_drop_extent_cache(BTRFS_I(inode
),
5036 free_extent_map(hole_em
);
5039 free_extent_map(em
);
5041 cur_offset
= last_byte
;
5042 if (cur_offset
>= block_end
)
5045 free_extent_map(em
);
5046 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
5051 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
5053 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5054 struct btrfs_trans_handle
*trans
;
5055 loff_t oldsize
= i_size_read(inode
);
5056 loff_t newsize
= attr
->ia_size
;
5057 int mask
= attr
->ia_valid
;
5061 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5062 * special case where we need to update the times despite not having
5063 * these flags set. For all other operations the VFS set these flags
5064 * explicitly if it wants a timestamp update.
5066 if (newsize
!= oldsize
) {
5067 inode_inc_iversion(inode
);
5068 if (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))
5069 inode
->i_ctime
= inode
->i_mtime
=
5070 current_time(inode
);
5073 if (newsize
> oldsize
) {
5075 * Don't do an expanding truncate while snapshotting is ongoing.
5076 * This is to ensure the snapshot captures a fully consistent
5077 * state of this file - if the snapshot captures this expanding
5078 * truncation, it must capture all writes that happened before
5081 btrfs_wait_for_snapshot_creation(root
);
5082 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
5084 btrfs_end_write_no_snapshotting(root
);
5088 trans
= btrfs_start_transaction(root
, 1);
5089 if (IS_ERR(trans
)) {
5090 btrfs_end_write_no_snapshotting(root
);
5091 return PTR_ERR(trans
);
5094 i_size_write(inode
, newsize
);
5095 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
5096 pagecache_isize_extended(inode
, oldsize
, newsize
);
5097 ret
= btrfs_update_inode(trans
, root
, inode
);
5098 btrfs_end_write_no_snapshotting(root
);
5099 btrfs_end_transaction(trans
);
5103 * We're truncating a file that used to have good data down to
5104 * zero. Make sure it gets into the ordered flush list so that
5105 * any new writes get down to disk quickly.
5108 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
5109 &BTRFS_I(inode
)->runtime_flags
);
5112 * 1 for the orphan item we're going to add
5113 * 1 for the orphan item deletion.
5115 trans
= btrfs_start_transaction(root
, 2);
5117 return PTR_ERR(trans
);
5120 * We need to do this in case we fail at _any_ point during the
5121 * actual truncate. Once we do the truncate_setsize we could
5122 * invalidate pages which forces any outstanding ordered io to
5123 * be instantly completed which will give us extents that need
5124 * to be truncated. If we fail to get an orphan inode down we
5125 * could have left over extents that were never meant to live,
5126 * so we need to guarantee from this point on that everything
5127 * will be consistent.
5129 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
5130 btrfs_end_transaction(trans
);
5134 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5135 truncate_setsize(inode
, newsize
);
5137 /* Disable nonlocked read DIO to avoid the end less truncate */
5138 btrfs_inode_block_unlocked_dio(BTRFS_I(inode
));
5139 inode_dio_wait(inode
);
5140 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode
));
5142 ret
= btrfs_truncate(inode
);
5143 if (ret
&& inode
->i_nlink
) {
5146 /* To get a stable disk_i_size */
5147 err
= btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5149 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5154 * failed to truncate, disk_i_size is only adjusted down
5155 * as we remove extents, so it should represent the true
5156 * size of the inode, so reset the in memory size and
5157 * delete our orphan entry.
5159 trans
= btrfs_join_transaction(root
);
5160 if (IS_ERR(trans
)) {
5161 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5164 i_size_write(inode
, BTRFS_I(inode
)->disk_i_size
);
5165 err
= btrfs_orphan_del(trans
, BTRFS_I(inode
));
5167 btrfs_abort_transaction(trans
, err
);
5168 btrfs_end_transaction(trans
);
5175 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5177 struct inode
*inode
= d_inode(dentry
);
5178 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5181 if (btrfs_root_readonly(root
))
5184 err
= setattr_prepare(dentry
, attr
);
5188 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
5189 err
= btrfs_setsize(inode
, attr
);
5194 if (attr
->ia_valid
) {
5195 setattr_copy(inode
, attr
);
5196 inode_inc_iversion(inode
);
5197 err
= btrfs_dirty_inode(inode
);
5199 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
5200 err
= posix_acl_chmod(inode
, inode
->i_mode
);
5207 * While truncating the inode pages during eviction, we get the VFS calling
5208 * btrfs_invalidatepage() against each page of the inode. This is slow because
5209 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5210 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5211 * extent_state structures over and over, wasting lots of time.
5213 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5214 * those expensive operations on a per page basis and do only the ordered io
5215 * finishing, while we release here the extent_map and extent_state structures,
5216 * without the excessive merging and splitting.
5218 static void evict_inode_truncate_pages(struct inode
*inode
)
5220 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5221 struct extent_map_tree
*map_tree
= &BTRFS_I(inode
)->extent_tree
;
5222 struct rb_node
*node
;
5224 ASSERT(inode
->i_state
& I_FREEING
);
5225 truncate_inode_pages_final(&inode
->i_data
);
5227 write_lock(&map_tree
->lock
);
5228 while (!RB_EMPTY_ROOT(&map_tree
->map
)) {
5229 struct extent_map
*em
;
5231 node
= rb_first(&map_tree
->map
);
5232 em
= rb_entry(node
, struct extent_map
, rb_node
);
5233 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5234 clear_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
5235 remove_extent_mapping(map_tree
, em
);
5236 free_extent_map(em
);
5237 if (need_resched()) {
5238 write_unlock(&map_tree
->lock
);
5240 write_lock(&map_tree
->lock
);
5243 write_unlock(&map_tree
->lock
);
5246 * Keep looping until we have no more ranges in the io tree.
5247 * We can have ongoing bios started by readpages (called from readahead)
5248 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5249 * still in progress (unlocked the pages in the bio but did not yet
5250 * unlocked the ranges in the io tree). Therefore this means some
5251 * ranges can still be locked and eviction started because before
5252 * submitting those bios, which are executed by a separate task (work
5253 * queue kthread), inode references (inode->i_count) were not taken
5254 * (which would be dropped in the end io callback of each bio).
5255 * Therefore here we effectively end up waiting for those bios and
5256 * anyone else holding locked ranges without having bumped the inode's
5257 * reference count - if we don't do it, when they access the inode's
5258 * io_tree to unlock a range it may be too late, leading to an
5259 * use-after-free issue.
5261 spin_lock(&io_tree
->lock
);
5262 while (!RB_EMPTY_ROOT(&io_tree
->state
)) {
5263 struct extent_state
*state
;
5264 struct extent_state
*cached_state
= NULL
;
5268 node
= rb_first(&io_tree
->state
);
5269 state
= rb_entry(node
, struct extent_state
, rb_node
);
5270 start
= state
->start
;
5272 spin_unlock(&io_tree
->lock
);
5274 lock_extent_bits(io_tree
, start
, end
, &cached_state
);
5277 * If still has DELALLOC flag, the extent didn't reach disk,
5278 * and its reserved space won't be freed by delayed_ref.
5279 * So we need to free its reserved space here.
5280 * (Refer to comment in btrfs_invalidatepage, case 2)
5282 * Note, end is the bytenr of last byte, so we need + 1 here.
5284 if (state
->state
& EXTENT_DELALLOC
)
5285 btrfs_qgroup_free_data(inode
, NULL
, start
, end
- start
+ 1);
5287 clear_extent_bit(io_tree
, start
, end
,
5288 EXTENT_LOCKED
| EXTENT_DIRTY
|
5289 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
5290 EXTENT_DEFRAG
, 1, 1,
5291 &cached_state
, GFP_NOFS
);
5294 spin_lock(&io_tree
->lock
);
5296 spin_unlock(&io_tree
->lock
);
5299 void btrfs_evict_inode(struct inode
*inode
)
5301 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5302 struct btrfs_trans_handle
*trans
;
5303 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5304 struct btrfs_block_rsv
*rsv
, *global_rsv
;
5305 int steal_from_global
= 0;
5309 trace_btrfs_inode_evict(inode
);
5316 min_size
= btrfs_calc_trunc_metadata_size(fs_info
, 1);
5318 evict_inode_truncate_pages(inode
);
5320 if (inode
->i_nlink
&&
5321 ((btrfs_root_refs(&root
->root_item
) != 0 &&
5322 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
) ||
5323 btrfs_is_free_space_inode(BTRFS_I(inode
))))
5326 if (is_bad_inode(inode
)) {
5327 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5330 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5331 if (!special_file(inode
->i_mode
))
5332 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5334 btrfs_free_io_failure_record(BTRFS_I(inode
), 0, (u64
)-1);
5336 if (test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
)) {
5337 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
5338 &BTRFS_I(inode
)->runtime_flags
));
5342 if (inode
->i_nlink
> 0) {
5343 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0 &&
5344 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
);
5348 ret
= btrfs_commit_inode_delayed_inode(BTRFS_I(inode
));
5350 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5354 rsv
= btrfs_alloc_block_rsv(fs_info
, BTRFS_BLOCK_RSV_TEMP
);
5356 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5359 rsv
->size
= min_size
;
5361 global_rsv
= &fs_info
->global_block_rsv
;
5363 btrfs_i_size_write(BTRFS_I(inode
), 0);
5366 * This is a bit simpler than btrfs_truncate since we've already
5367 * reserved our space for our orphan item in the unlink, so we just
5368 * need to reserve some slack space in case we add bytes and update
5369 * inode item when doing the truncate.
5372 ret
= btrfs_block_rsv_refill(root
, rsv
, min_size
,
5373 BTRFS_RESERVE_FLUSH_LIMIT
);
5376 * Try and steal from the global reserve since we will
5377 * likely not use this space anyway, we want to try as
5378 * hard as possible to get this to work.
5381 steal_from_global
++;
5383 steal_from_global
= 0;
5387 * steal_from_global == 0: we reserved stuff, hooray!
5388 * steal_from_global == 1: we didn't reserve stuff, boo!
5389 * steal_from_global == 2: we've committed, still not a lot of
5390 * room but maybe we'll have room in the global reserve this
5392 * steal_from_global == 3: abandon all hope!
5394 if (steal_from_global
> 2) {
5396 "Could not get space for a delete, will truncate on mount %d",
5398 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5399 btrfs_free_block_rsv(fs_info
, rsv
);
5403 trans
= btrfs_join_transaction(root
);
5404 if (IS_ERR(trans
)) {
5405 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5406 btrfs_free_block_rsv(fs_info
, rsv
);
5411 * We can't just steal from the global reserve, we need to make
5412 * sure there is room to do it, if not we need to commit and try
5415 if (steal_from_global
) {
5416 if (!btrfs_check_space_for_delayed_refs(trans
, fs_info
))
5417 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
,
5424 * Couldn't steal from the global reserve, we have too much
5425 * pending stuff built up, commit the transaction and try it
5429 ret
= btrfs_commit_transaction(trans
);
5431 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5432 btrfs_free_block_rsv(fs_info
, rsv
);
5437 steal_from_global
= 0;
5440 trans
->block_rsv
= rsv
;
5442 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
5443 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
)
5446 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
5447 btrfs_end_transaction(trans
);
5449 btrfs_btree_balance_dirty(fs_info
);
5452 btrfs_free_block_rsv(fs_info
, rsv
);
5455 * Errors here aren't a big deal, it just means we leave orphan items
5456 * in the tree. They will be cleaned up on the next mount.
5459 trans
->block_rsv
= root
->orphan_block_rsv
;
5460 btrfs_orphan_del(trans
, BTRFS_I(inode
));
5462 btrfs_orphan_del(NULL
, BTRFS_I(inode
));
5465 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
5466 if (!(root
== fs_info
->tree_root
||
5467 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
5468 btrfs_return_ino(root
, btrfs_ino(BTRFS_I(inode
)));
5470 btrfs_end_transaction(trans
);
5471 btrfs_btree_balance_dirty(fs_info
);
5473 btrfs_remove_delayed_node(BTRFS_I(inode
));
5478 * this returns the key found in the dir entry in the location pointer.
5479 * If no dir entries were found, location->objectid is 0.
5481 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
5482 struct btrfs_key
*location
)
5484 const char *name
= dentry
->d_name
.name
;
5485 int namelen
= dentry
->d_name
.len
;
5486 struct btrfs_dir_item
*di
;
5487 struct btrfs_path
*path
;
5488 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5491 path
= btrfs_alloc_path();
5495 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(BTRFS_I(dir
)),
5500 if (IS_ERR_OR_NULL(di
))
5503 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
5504 if (location
->type
!= BTRFS_INODE_ITEM_KEY
&&
5505 location
->type
!= BTRFS_ROOT_ITEM_KEY
) {
5506 btrfs_warn(root
->fs_info
,
5507 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5508 __func__
, name
, btrfs_ino(BTRFS_I(dir
)),
5509 location
->objectid
, location
->type
, location
->offset
);
5513 btrfs_free_path(path
);
5516 location
->objectid
= 0;
5521 * when we hit a tree root in a directory, the btrfs part of the inode
5522 * needs to be changed to reflect the root directory of the tree root. This
5523 * is kind of like crossing a mount point.
5525 static int fixup_tree_root_location(struct btrfs_fs_info
*fs_info
,
5527 struct dentry
*dentry
,
5528 struct btrfs_key
*location
,
5529 struct btrfs_root
**sub_root
)
5531 struct btrfs_path
*path
;
5532 struct btrfs_root
*new_root
;
5533 struct btrfs_root_ref
*ref
;
5534 struct extent_buffer
*leaf
;
5535 struct btrfs_key key
;
5539 path
= btrfs_alloc_path();
5546 key
.objectid
= BTRFS_I(dir
)->root
->root_key
.objectid
;
5547 key
.type
= BTRFS_ROOT_REF_KEY
;
5548 key
.offset
= location
->objectid
;
5550 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
5557 leaf
= path
->nodes
[0];
5558 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
5559 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(BTRFS_I(dir
)) ||
5560 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
5563 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
5564 (unsigned long)(ref
+ 1),
5565 dentry
->d_name
.len
);
5569 btrfs_release_path(path
);
5571 new_root
= btrfs_read_fs_root_no_name(fs_info
, location
);
5572 if (IS_ERR(new_root
)) {
5573 err
= PTR_ERR(new_root
);
5577 *sub_root
= new_root
;
5578 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
5579 location
->type
= BTRFS_INODE_ITEM_KEY
;
5580 location
->offset
= 0;
5583 btrfs_free_path(path
);
5587 static void inode_tree_add(struct inode
*inode
)
5589 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5590 struct btrfs_inode
*entry
;
5592 struct rb_node
*parent
;
5593 struct rb_node
*new = &BTRFS_I(inode
)->rb_node
;
5594 u64 ino
= btrfs_ino(BTRFS_I(inode
));
5596 if (inode_unhashed(inode
))
5599 spin_lock(&root
->inode_lock
);
5600 p
= &root
->inode_tree
.rb_node
;
5603 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
5605 if (ino
< btrfs_ino(BTRFS_I(&entry
->vfs_inode
)))
5606 p
= &parent
->rb_left
;
5607 else if (ino
> btrfs_ino(BTRFS_I(&entry
->vfs_inode
)))
5608 p
= &parent
->rb_right
;
5610 WARN_ON(!(entry
->vfs_inode
.i_state
&
5611 (I_WILL_FREE
| I_FREEING
)));
5612 rb_replace_node(parent
, new, &root
->inode_tree
);
5613 RB_CLEAR_NODE(parent
);
5614 spin_unlock(&root
->inode_lock
);
5618 rb_link_node(new, parent
, p
);
5619 rb_insert_color(new, &root
->inode_tree
);
5620 spin_unlock(&root
->inode_lock
);
5623 static void inode_tree_del(struct inode
*inode
)
5625 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5626 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5629 spin_lock(&root
->inode_lock
);
5630 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
5631 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
5632 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
5633 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5635 spin_unlock(&root
->inode_lock
);
5637 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0) {
5638 synchronize_srcu(&fs_info
->subvol_srcu
);
5639 spin_lock(&root
->inode_lock
);
5640 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5641 spin_unlock(&root
->inode_lock
);
5643 btrfs_add_dead_root(root
);
5647 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
5649 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5650 struct rb_node
*node
;
5651 struct rb_node
*prev
;
5652 struct btrfs_inode
*entry
;
5653 struct inode
*inode
;
5656 if (!test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
5657 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
5659 spin_lock(&root
->inode_lock
);
5661 node
= root
->inode_tree
.rb_node
;
5665 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5667 if (objectid
< btrfs_ino(BTRFS_I(&entry
->vfs_inode
)))
5668 node
= node
->rb_left
;
5669 else if (objectid
> btrfs_ino(BTRFS_I(&entry
->vfs_inode
)))
5670 node
= node
->rb_right
;
5676 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
5677 if (objectid
<= btrfs_ino(BTRFS_I(&entry
->vfs_inode
))) {
5681 prev
= rb_next(prev
);
5685 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5686 objectid
= btrfs_ino(BTRFS_I(&entry
->vfs_inode
)) + 1;
5687 inode
= igrab(&entry
->vfs_inode
);
5689 spin_unlock(&root
->inode_lock
);
5690 if (atomic_read(&inode
->i_count
) > 1)
5691 d_prune_aliases(inode
);
5693 * btrfs_drop_inode will have it removed from
5694 * the inode cache when its usage count
5699 spin_lock(&root
->inode_lock
);
5703 if (cond_resched_lock(&root
->inode_lock
))
5706 node
= rb_next(node
);
5708 spin_unlock(&root
->inode_lock
);
5711 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
5713 struct btrfs_iget_args
*args
= p
;
5714 inode
->i_ino
= args
->location
->objectid
;
5715 memcpy(&BTRFS_I(inode
)->location
, args
->location
,
5716 sizeof(*args
->location
));
5717 BTRFS_I(inode
)->root
= args
->root
;
5721 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
5723 struct btrfs_iget_args
*args
= opaque
;
5724 return args
->location
->objectid
== BTRFS_I(inode
)->location
.objectid
&&
5725 args
->root
== BTRFS_I(inode
)->root
;
5728 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
5729 struct btrfs_key
*location
,
5730 struct btrfs_root
*root
)
5732 struct inode
*inode
;
5733 struct btrfs_iget_args args
;
5734 unsigned long hashval
= btrfs_inode_hash(location
->objectid
, root
);
5736 args
.location
= location
;
5739 inode
= iget5_locked(s
, hashval
, btrfs_find_actor
,
5740 btrfs_init_locked_inode
,
5745 /* Get an inode object given its location and corresponding root.
5746 * Returns in *is_new if the inode was read from disk
5748 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
5749 struct btrfs_root
*root
, int *new)
5751 struct inode
*inode
;
5753 inode
= btrfs_iget_locked(s
, location
, root
);
5755 return ERR_PTR(-ENOMEM
);
5757 if (inode
->i_state
& I_NEW
) {
5760 ret
= btrfs_read_locked_inode(inode
);
5761 if (!is_bad_inode(inode
)) {
5762 inode_tree_add(inode
);
5763 unlock_new_inode(inode
);
5767 unlock_new_inode(inode
);
5770 inode
= ERR_PTR(ret
< 0 ? ret
: -ESTALE
);
5777 static struct inode
*new_simple_dir(struct super_block
*s
,
5778 struct btrfs_key
*key
,
5779 struct btrfs_root
*root
)
5781 struct inode
*inode
= new_inode(s
);
5784 return ERR_PTR(-ENOMEM
);
5786 BTRFS_I(inode
)->root
= root
;
5787 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
5788 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
5790 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
5791 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
5792 inode
->i_opflags
&= ~IOP_XATTR
;
5793 inode
->i_fop
= &simple_dir_operations
;
5794 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
5795 inode
->i_mtime
= current_time(inode
);
5796 inode
->i_atime
= inode
->i_mtime
;
5797 inode
->i_ctime
= inode
->i_mtime
;
5798 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
5803 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
5805 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
5806 struct inode
*inode
;
5807 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5808 struct btrfs_root
*sub_root
= root
;
5809 struct btrfs_key location
;
5813 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
5814 return ERR_PTR(-ENAMETOOLONG
);
5816 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
5818 return ERR_PTR(ret
);
5820 if (location
.objectid
== 0)
5821 return ERR_PTR(-ENOENT
);
5823 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
5824 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
5828 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
5829 ret
= fixup_tree_root_location(fs_info
, dir
, dentry
,
5830 &location
, &sub_root
);
5833 inode
= ERR_PTR(ret
);
5835 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
5837 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
5839 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5841 if (!IS_ERR(inode
) && root
!= sub_root
) {
5842 down_read(&fs_info
->cleanup_work_sem
);
5843 if (!sb_rdonly(inode
->i_sb
))
5844 ret
= btrfs_orphan_cleanup(sub_root
);
5845 up_read(&fs_info
->cleanup_work_sem
);
5848 inode
= ERR_PTR(ret
);
5855 static int btrfs_dentry_delete(const struct dentry
*dentry
)
5857 struct btrfs_root
*root
;
5858 struct inode
*inode
= d_inode(dentry
);
5860 if (!inode
&& !IS_ROOT(dentry
))
5861 inode
= d_inode(dentry
->d_parent
);
5864 root
= BTRFS_I(inode
)->root
;
5865 if (btrfs_root_refs(&root
->root_item
) == 0)
5868 if (btrfs_ino(BTRFS_I(inode
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5874 static void btrfs_dentry_release(struct dentry
*dentry
)
5876 kfree(dentry
->d_fsdata
);
5879 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5882 struct inode
*inode
;
5884 inode
= btrfs_lookup_dentry(dir
, dentry
);
5885 if (IS_ERR(inode
)) {
5886 if (PTR_ERR(inode
) == -ENOENT
)
5889 return ERR_CAST(inode
);
5892 return d_splice_alias(inode
, dentry
);
5895 unsigned char btrfs_filetype_table
[] = {
5896 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
5900 * All this infrastructure exists because dir_emit can fault, and we are holding
5901 * the tree lock when doing readdir. For now just allocate a buffer and copy
5902 * our information into that, and then dir_emit from the buffer. This is
5903 * similar to what NFS does, only we don't keep the buffer around in pagecache
5904 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5905 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5908 static int btrfs_opendir(struct inode
*inode
, struct file
*file
)
5910 struct btrfs_file_private
*private;
5912 private = kzalloc(sizeof(struct btrfs_file_private
), GFP_KERNEL
);
5915 private->filldir_buf
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
5916 if (!private->filldir_buf
) {
5920 file
->private_data
= private;
5931 static int btrfs_filldir(void *addr
, int entries
, struct dir_context
*ctx
)
5934 struct dir_entry
*entry
= addr
;
5935 char *name
= (char *)(entry
+ 1);
5937 ctx
->pos
= get_unaligned(&entry
->offset
);
5938 if (!dir_emit(ctx
, name
, get_unaligned(&entry
->name_len
),
5939 get_unaligned(&entry
->ino
),
5940 get_unaligned(&entry
->type
)))
5942 addr
+= sizeof(struct dir_entry
) +
5943 get_unaligned(&entry
->name_len
);
5949 static int btrfs_real_readdir(struct file
*file
, struct dir_context
*ctx
)
5951 struct inode
*inode
= file_inode(file
);
5952 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5953 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5954 struct btrfs_file_private
*private = file
->private_data
;
5955 struct btrfs_dir_item
*di
;
5956 struct btrfs_key key
;
5957 struct btrfs_key found_key
;
5958 struct btrfs_path
*path
;
5960 struct list_head ins_list
;
5961 struct list_head del_list
;
5963 struct extent_buffer
*leaf
;
5970 struct btrfs_key location
;
5972 if (!dir_emit_dots(file
, ctx
))
5975 path
= btrfs_alloc_path();
5979 addr
= private->filldir_buf
;
5980 path
->reada
= READA_FORWARD
;
5982 INIT_LIST_HEAD(&ins_list
);
5983 INIT_LIST_HEAD(&del_list
);
5984 put
= btrfs_readdir_get_delayed_items(inode
, &ins_list
, &del_list
);
5987 key
.type
= BTRFS_DIR_INDEX_KEY
;
5988 key
.offset
= ctx
->pos
;
5989 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
5991 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5996 struct dir_entry
*entry
;
5998 leaf
= path
->nodes
[0];
5999 slot
= path
->slots
[0];
6000 if (slot
>= btrfs_header_nritems(leaf
)) {
6001 ret
= btrfs_next_leaf(root
, path
);
6009 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6011 if (found_key
.objectid
!= key
.objectid
)
6013 if (found_key
.type
!= BTRFS_DIR_INDEX_KEY
)
6015 if (found_key
.offset
< ctx
->pos
)
6017 if (btrfs_should_delete_dir_index(&del_list
, found_key
.offset
))
6019 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
6020 if (verify_dir_item(fs_info
, leaf
, slot
, di
))
6023 name_len
= btrfs_dir_name_len(leaf
, di
);
6024 if ((total_len
+ sizeof(struct dir_entry
) + name_len
) >=
6026 btrfs_release_path(path
);
6027 ret
= btrfs_filldir(private->filldir_buf
, entries
, ctx
);
6030 addr
= private->filldir_buf
;
6037 put_unaligned(name_len
, &entry
->name_len
);
6038 name_ptr
= (char *)(entry
+ 1);
6039 read_extent_buffer(leaf
, name_ptr
, (unsigned long)(di
+ 1),
6041 put_unaligned(btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)],
6043 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
6044 put_unaligned(location
.objectid
, &entry
->ino
);
6045 put_unaligned(found_key
.offset
, &entry
->offset
);
6047 addr
+= sizeof(struct dir_entry
) + name_len
;
6048 total_len
+= sizeof(struct dir_entry
) + name_len
;
6052 btrfs_release_path(path
);
6054 ret
= btrfs_filldir(private->filldir_buf
, entries
, ctx
);
6058 ret
= btrfs_readdir_delayed_dir_index(ctx
, &ins_list
);
6063 * Stop new entries from being returned after we return the last
6066 * New directory entries are assigned a strictly increasing
6067 * offset. This means that new entries created during readdir
6068 * are *guaranteed* to be seen in the future by that readdir.
6069 * This has broken buggy programs which operate on names as
6070 * they're returned by readdir. Until we re-use freed offsets
6071 * we have this hack to stop new entries from being returned
6072 * under the assumption that they'll never reach this huge
6075 * This is being careful not to overflow 32bit loff_t unless the
6076 * last entry requires it because doing so has broken 32bit apps
6079 if (ctx
->pos
>= INT_MAX
)
6080 ctx
->pos
= LLONG_MAX
;
6087 btrfs_readdir_put_delayed_items(inode
, &ins_list
, &del_list
);
6088 btrfs_free_path(path
);
6092 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
6094 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6095 struct btrfs_trans_handle
*trans
;
6097 bool nolock
= false;
6099 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
6102 if (btrfs_fs_closing(root
->fs_info
) &&
6103 btrfs_is_free_space_inode(BTRFS_I(inode
)))
6106 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
6108 trans
= btrfs_join_transaction_nolock(root
);
6110 trans
= btrfs_join_transaction(root
);
6112 return PTR_ERR(trans
);
6113 ret
= btrfs_commit_transaction(trans
);
6119 * This is somewhat expensive, updating the tree every time the
6120 * inode changes. But, it is most likely to find the inode in cache.
6121 * FIXME, needs more benchmarking...there are no reasons other than performance
6122 * to keep or drop this code.
6124 static int btrfs_dirty_inode(struct inode
*inode
)
6126 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
6127 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6128 struct btrfs_trans_handle
*trans
;
6131 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
6134 trans
= btrfs_join_transaction(root
);
6136 return PTR_ERR(trans
);
6138 ret
= btrfs_update_inode(trans
, root
, inode
);
6139 if (ret
&& ret
== -ENOSPC
) {
6140 /* whoops, lets try again with the full transaction */
6141 btrfs_end_transaction(trans
);
6142 trans
= btrfs_start_transaction(root
, 1);
6144 return PTR_ERR(trans
);
6146 ret
= btrfs_update_inode(trans
, root
, inode
);
6148 btrfs_end_transaction(trans
);
6149 if (BTRFS_I(inode
)->delayed_node
)
6150 btrfs_balance_delayed_items(fs_info
);
6156 * This is a copy of file_update_time. We need this so we can return error on
6157 * ENOSPC for updating the inode in the case of file write and mmap writes.
6159 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
6162 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6164 if (btrfs_root_readonly(root
))
6167 if (flags
& S_VERSION
)
6168 inode_inc_iversion(inode
);
6169 if (flags
& S_CTIME
)
6170 inode
->i_ctime
= *now
;
6171 if (flags
& S_MTIME
)
6172 inode
->i_mtime
= *now
;
6173 if (flags
& S_ATIME
)
6174 inode
->i_atime
= *now
;
6175 return btrfs_dirty_inode(inode
);
6179 * find the highest existing sequence number in a directory
6180 * and then set the in-memory index_cnt variable to reflect
6181 * free sequence numbers
6183 static int btrfs_set_inode_index_count(struct btrfs_inode
*inode
)
6185 struct btrfs_root
*root
= inode
->root
;
6186 struct btrfs_key key
, found_key
;
6187 struct btrfs_path
*path
;
6188 struct extent_buffer
*leaf
;
6191 key
.objectid
= btrfs_ino(inode
);
6192 key
.type
= BTRFS_DIR_INDEX_KEY
;
6193 key
.offset
= (u64
)-1;
6195 path
= btrfs_alloc_path();
6199 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6202 /* FIXME: we should be able to handle this */
6208 * MAGIC NUMBER EXPLANATION:
6209 * since we search a directory based on f_pos we have to start at 2
6210 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6211 * else has to start at 2
6213 if (path
->slots
[0] == 0) {
6214 inode
->index_cnt
= 2;
6220 leaf
= path
->nodes
[0];
6221 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6223 if (found_key
.objectid
!= btrfs_ino(inode
) ||
6224 found_key
.type
!= BTRFS_DIR_INDEX_KEY
) {
6225 inode
->index_cnt
= 2;
6229 inode
->index_cnt
= found_key
.offset
+ 1;
6231 btrfs_free_path(path
);
6236 * helper to find a free sequence number in a given directory. This current
6237 * code is very simple, later versions will do smarter things in the btree
6239 int btrfs_set_inode_index(struct btrfs_inode
*dir
, u64
*index
)
6243 if (dir
->index_cnt
== (u64
)-1) {
6244 ret
= btrfs_inode_delayed_dir_index_count(dir
);
6246 ret
= btrfs_set_inode_index_count(dir
);
6252 *index
= dir
->index_cnt
;
6258 static int btrfs_insert_inode_locked(struct inode
*inode
)
6260 struct btrfs_iget_args args
;
6261 args
.location
= &BTRFS_I(inode
)->location
;
6262 args
.root
= BTRFS_I(inode
)->root
;
6264 return insert_inode_locked4(inode
,
6265 btrfs_inode_hash(inode
->i_ino
, BTRFS_I(inode
)->root
),
6266 btrfs_find_actor
, &args
);
6270 * Inherit flags from the parent inode.
6272 * Currently only the compression flags and the cow flags are inherited.
6274 static void btrfs_inherit_iflags(struct inode
*inode
, struct inode
*dir
)
6281 flags
= BTRFS_I(dir
)->flags
;
6283 if (flags
& BTRFS_INODE_NOCOMPRESS
) {
6284 BTRFS_I(inode
)->flags
&= ~BTRFS_INODE_COMPRESS
;
6285 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
6286 } else if (flags
& BTRFS_INODE_COMPRESS
) {
6287 BTRFS_I(inode
)->flags
&= ~BTRFS_INODE_NOCOMPRESS
;
6288 BTRFS_I(inode
)->flags
|= BTRFS_INODE_COMPRESS
;
6291 if (flags
& BTRFS_INODE_NODATACOW
) {
6292 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
6293 if (S_ISREG(inode
->i_mode
))
6294 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6297 btrfs_update_iflags(inode
);
6300 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
6301 struct btrfs_root
*root
,
6303 const char *name
, int name_len
,
6304 u64 ref_objectid
, u64 objectid
,
6305 umode_t mode
, u64
*index
)
6307 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6308 struct inode
*inode
;
6309 struct btrfs_inode_item
*inode_item
;
6310 struct btrfs_key
*location
;
6311 struct btrfs_path
*path
;
6312 struct btrfs_inode_ref
*ref
;
6313 struct btrfs_key key
[2];
6315 int nitems
= name
? 2 : 1;
6319 path
= btrfs_alloc_path();
6321 return ERR_PTR(-ENOMEM
);
6323 inode
= new_inode(fs_info
->sb
);
6325 btrfs_free_path(path
);
6326 return ERR_PTR(-ENOMEM
);
6330 * O_TMPFILE, set link count to 0, so that after this point,
6331 * we fill in an inode item with the correct link count.
6334 set_nlink(inode
, 0);
6337 * we have to initialize this early, so we can reclaim the inode
6338 * number if we fail afterwards in this function.
6340 inode
->i_ino
= objectid
;
6343 trace_btrfs_inode_request(dir
);
6345 ret
= btrfs_set_inode_index(BTRFS_I(dir
), index
);
6347 btrfs_free_path(path
);
6349 return ERR_PTR(ret
);
6355 * index_cnt is ignored for everything but a dir,
6356 * btrfs_get_inode_index_count has an explanation for the magic
6359 BTRFS_I(inode
)->index_cnt
= 2;
6360 BTRFS_I(inode
)->dir_index
= *index
;
6361 BTRFS_I(inode
)->root
= root
;
6362 BTRFS_I(inode
)->generation
= trans
->transid
;
6363 inode
->i_generation
= BTRFS_I(inode
)->generation
;
6366 * We could have gotten an inode number from somebody who was fsynced
6367 * and then removed in this same transaction, so let's just set full
6368 * sync since it will be a full sync anyway and this will blow away the
6369 * old info in the log.
6371 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
6373 key
[0].objectid
= objectid
;
6374 key
[0].type
= BTRFS_INODE_ITEM_KEY
;
6377 sizes
[0] = sizeof(struct btrfs_inode_item
);
6381 * Start new inodes with an inode_ref. This is slightly more
6382 * efficient for small numbers of hard links since they will
6383 * be packed into one item. Extended refs will kick in if we
6384 * add more hard links than can fit in the ref item.
6386 key
[1].objectid
= objectid
;
6387 key
[1].type
= BTRFS_INODE_REF_KEY
;
6388 key
[1].offset
= ref_objectid
;
6390 sizes
[1] = name_len
+ sizeof(*ref
);
6393 location
= &BTRFS_I(inode
)->location
;
6394 location
->objectid
= objectid
;
6395 location
->offset
= 0;
6396 location
->type
= BTRFS_INODE_ITEM_KEY
;
6398 ret
= btrfs_insert_inode_locked(inode
);
6402 path
->leave_spinning
= 1;
6403 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, nitems
);
6407 inode_init_owner(inode
, dir
, mode
);
6408 inode_set_bytes(inode
, 0);
6410 inode
->i_mtime
= current_time(inode
);
6411 inode
->i_atime
= inode
->i_mtime
;
6412 inode
->i_ctime
= inode
->i_mtime
;
6413 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
6415 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
6416 struct btrfs_inode_item
);
6417 memzero_extent_buffer(path
->nodes
[0], (unsigned long)inode_item
,
6418 sizeof(*inode_item
));
6419 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
6422 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
6423 struct btrfs_inode_ref
);
6424 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
6425 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
6426 ptr
= (unsigned long)(ref
+ 1);
6427 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
6430 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6431 btrfs_free_path(path
);
6433 btrfs_inherit_iflags(inode
, dir
);
6435 if (S_ISREG(mode
)) {
6436 if (btrfs_test_opt(fs_info
, NODATASUM
))
6437 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6438 if (btrfs_test_opt(fs_info
, NODATACOW
))
6439 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
6440 BTRFS_INODE_NODATASUM
;
6443 inode_tree_add(inode
);
6445 trace_btrfs_inode_new(inode
);
6446 btrfs_set_inode_last_trans(trans
, inode
);
6448 btrfs_update_root_times(trans
, root
);
6450 ret
= btrfs_inode_inherit_props(trans
, inode
, dir
);
6453 "error inheriting props for ino %llu (root %llu): %d",
6454 btrfs_ino(BTRFS_I(inode
)), root
->root_key
.objectid
, ret
);
6459 unlock_new_inode(inode
);
6462 BTRFS_I(dir
)->index_cnt
--;
6463 btrfs_free_path(path
);
6465 return ERR_PTR(ret
);
6468 static inline u8
btrfs_inode_type(struct inode
*inode
)
6470 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
6474 * utility function to add 'inode' into 'parent_inode' with
6475 * a give name and a given sequence number.
6476 * if 'add_backref' is true, also insert a backref from the
6477 * inode to the parent directory.
6479 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
6480 struct btrfs_inode
*parent_inode
, struct btrfs_inode
*inode
,
6481 const char *name
, int name_len
, int add_backref
, u64 index
)
6483 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6485 struct btrfs_key key
;
6486 struct btrfs_root
*root
= parent_inode
->root
;
6487 u64 ino
= btrfs_ino(inode
);
6488 u64 parent_ino
= btrfs_ino(parent_inode
);
6490 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6491 memcpy(&key
, &inode
->root
->root_key
, sizeof(key
));
6494 key
.type
= BTRFS_INODE_ITEM_KEY
;
6498 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6499 ret
= btrfs_add_root_ref(trans
, fs_info
, key
.objectid
,
6500 root
->root_key
.objectid
, parent_ino
,
6501 index
, name
, name_len
);
6502 } else if (add_backref
) {
6503 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
6507 /* Nothing to clean up yet */
6511 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
6513 btrfs_inode_type(&inode
->vfs_inode
), index
);
6514 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
6517 btrfs_abort_transaction(trans
, ret
);
6521 btrfs_i_size_write(parent_inode
, parent_inode
->vfs_inode
.i_size
+
6523 inode_inc_iversion(&parent_inode
->vfs_inode
);
6524 parent_inode
->vfs_inode
.i_mtime
= parent_inode
->vfs_inode
.i_ctime
=
6525 current_time(&parent_inode
->vfs_inode
);
6526 ret
= btrfs_update_inode(trans
, root
, &parent_inode
->vfs_inode
);
6528 btrfs_abort_transaction(trans
, ret
);
6532 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6535 err
= btrfs_del_root_ref(trans
, fs_info
, key
.objectid
,
6536 root
->root_key
.objectid
, parent_ino
,
6537 &local_index
, name
, name_len
);
6539 } else if (add_backref
) {
6543 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
6544 ino
, parent_ino
, &local_index
);
6549 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
6550 struct btrfs_inode
*dir
, struct dentry
*dentry
,
6551 struct btrfs_inode
*inode
, int backref
, u64 index
)
6553 int err
= btrfs_add_link(trans
, dir
, inode
,
6554 dentry
->d_name
.name
, dentry
->d_name
.len
,
6561 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
6562 umode_t mode
, dev_t rdev
)
6564 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6565 struct btrfs_trans_handle
*trans
;
6566 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6567 struct inode
*inode
= NULL
;
6574 * 2 for inode item and ref
6576 * 1 for xattr if selinux is on
6578 trans
= btrfs_start_transaction(root
, 5);
6580 return PTR_ERR(trans
);
6582 err
= btrfs_find_free_ino(root
, &objectid
);
6586 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6587 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)), objectid
,
6589 if (IS_ERR(inode
)) {
6590 err
= PTR_ERR(inode
);
6595 * If the active LSM wants to access the inode during
6596 * d_instantiate it needs these. Smack checks to see
6597 * if the filesystem supports xattrs by looking at the
6600 inode
->i_op
= &btrfs_special_inode_operations
;
6601 init_special_inode(inode
, inode
->i_mode
, rdev
);
6603 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6605 goto out_unlock_inode
;
6607 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
, BTRFS_I(inode
),
6610 goto out_unlock_inode
;
6612 btrfs_update_inode(trans
, root
, inode
);
6613 d_instantiate_new(dentry
, inode
);
6617 btrfs_end_transaction(trans
);
6618 btrfs_balance_delayed_items(fs_info
);
6619 btrfs_btree_balance_dirty(fs_info
);
6621 inode_dec_link_count(inode
);
6628 unlock_new_inode(inode
);
6633 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
6634 umode_t mode
, bool excl
)
6636 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6637 struct btrfs_trans_handle
*trans
;
6638 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6639 struct inode
*inode
= NULL
;
6640 int drop_inode_on_err
= 0;
6646 * 2 for inode item and ref
6648 * 1 for xattr if selinux is on
6650 trans
= btrfs_start_transaction(root
, 5);
6652 return PTR_ERR(trans
);
6654 err
= btrfs_find_free_ino(root
, &objectid
);
6658 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6659 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)), objectid
,
6661 if (IS_ERR(inode
)) {
6662 err
= PTR_ERR(inode
);
6665 drop_inode_on_err
= 1;
6667 * If the active LSM wants to access the inode during
6668 * d_instantiate it needs these. Smack checks to see
6669 * if the filesystem supports xattrs by looking at the
6672 inode
->i_fop
= &btrfs_file_operations
;
6673 inode
->i_op
= &btrfs_file_inode_operations
;
6674 inode
->i_mapping
->a_ops
= &btrfs_aops
;
6676 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6678 goto out_unlock_inode
;
6680 err
= btrfs_update_inode(trans
, root
, inode
);
6682 goto out_unlock_inode
;
6684 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
, BTRFS_I(inode
),
6687 goto out_unlock_inode
;
6689 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
6690 d_instantiate_new(dentry
, inode
);
6693 btrfs_end_transaction(trans
);
6694 if (err
&& drop_inode_on_err
) {
6695 inode_dec_link_count(inode
);
6698 btrfs_balance_delayed_items(fs_info
);
6699 btrfs_btree_balance_dirty(fs_info
);
6703 unlock_new_inode(inode
);
6708 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
6709 struct dentry
*dentry
)
6711 struct btrfs_trans_handle
*trans
= NULL
;
6712 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6713 struct inode
*inode
= d_inode(old_dentry
);
6714 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
6719 /* do not allow sys_link's with other subvols of the same device */
6720 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
6723 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
6726 err
= btrfs_set_inode_index(BTRFS_I(dir
), &index
);
6731 * 2 items for inode and inode ref
6732 * 2 items for dir items
6733 * 1 item for parent inode
6735 trans
= btrfs_start_transaction(root
, 5);
6736 if (IS_ERR(trans
)) {
6737 err
= PTR_ERR(trans
);
6742 /* There are several dir indexes for this inode, clear the cache. */
6743 BTRFS_I(inode
)->dir_index
= 0ULL;
6745 inode_inc_iversion(inode
);
6746 inode
->i_ctime
= current_time(inode
);
6748 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
6750 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
, BTRFS_I(inode
),
6756 struct dentry
*parent
= dentry
->d_parent
;
6757 err
= btrfs_update_inode(trans
, root
, inode
);
6760 if (inode
->i_nlink
== 1) {
6762 * If new hard link count is 1, it's a file created
6763 * with open(2) O_TMPFILE flag.
6765 err
= btrfs_orphan_del(trans
, BTRFS_I(inode
));
6769 d_instantiate(dentry
, inode
);
6770 btrfs_log_new_name(trans
, BTRFS_I(inode
), NULL
, parent
);
6773 btrfs_balance_delayed_items(fs_info
);
6776 btrfs_end_transaction(trans
);
6778 inode_dec_link_count(inode
);
6781 btrfs_btree_balance_dirty(fs_info
);
6785 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
6787 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6788 struct inode
*inode
= NULL
;
6789 struct btrfs_trans_handle
*trans
;
6790 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6792 int drop_on_err
= 0;
6797 * 2 items for inode and ref
6798 * 2 items for dir items
6799 * 1 for xattr if selinux is on
6801 trans
= btrfs_start_transaction(root
, 5);
6803 return PTR_ERR(trans
);
6805 err
= btrfs_find_free_ino(root
, &objectid
);
6809 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6810 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)), objectid
,
6811 S_IFDIR
| mode
, &index
);
6812 if (IS_ERR(inode
)) {
6813 err
= PTR_ERR(inode
);
6818 /* these must be set before we unlock the inode */
6819 inode
->i_op
= &btrfs_dir_inode_operations
;
6820 inode
->i_fop
= &btrfs_dir_file_operations
;
6822 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6824 goto out_fail_inode
;
6826 btrfs_i_size_write(BTRFS_I(inode
), 0);
6827 err
= btrfs_update_inode(trans
, root
, inode
);
6829 goto out_fail_inode
;
6831 err
= btrfs_add_link(trans
, BTRFS_I(dir
), BTRFS_I(inode
),
6832 dentry
->d_name
.name
,
6833 dentry
->d_name
.len
, 0, index
);
6835 goto out_fail_inode
;
6837 d_instantiate_new(dentry
, inode
);
6841 btrfs_end_transaction(trans
);
6843 inode_dec_link_count(inode
);
6846 btrfs_balance_delayed_items(fs_info
);
6847 btrfs_btree_balance_dirty(fs_info
);
6851 unlock_new_inode(inode
);
6855 /* Find next extent map of a given extent map, caller needs to ensure locks */
6856 static struct extent_map
*next_extent_map(struct extent_map
*em
)
6858 struct rb_node
*next
;
6860 next
= rb_next(&em
->rb_node
);
6863 return container_of(next
, struct extent_map
, rb_node
);
6866 static struct extent_map
*prev_extent_map(struct extent_map
*em
)
6868 struct rb_node
*prev
;
6870 prev
= rb_prev(&em
->rb_node
);
6873 return container_of(prev
, struct extent_map
, rb_node
);
6876 /* helper for btfs_get_extent. Given an existing extent in the tree,
6877 * the existing extent is the nearest extent to map_start,
6878 * and an extent that you want to insert, deal with overlap and insert
6879 * the best fitted new extent into the tree.
6881 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
6882 struct extent_map
*existing
,
6883 struct extent_map
*em
,
6886 struct extent_map
*prev
;
6887 struct extent_map
*next
;
6892 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
6894 if (existing
->start
> map_start
) {
6896 prev
= prev_extent_map(next
);
6899 next
= next_extent_map(prev
);
6902 start
= prev
? extent_map_end(prev
) : em
->start
;
6903 start
= max_t(u64
, start
, em
->start
);
6904 end
= next
? next
->start
: extent_map_end(em
);
6905 end
= min_t(u64
, end
, extent_map_end(em
));
6906 start_diff
= start
- em
->start
;
6908 em
->len
= end
- start
;
6909 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
6910 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
6911 em
->block_start
+= start_diff
;
6912 em
->block_len
-= start_diff
;
6914 return add_extent_mapping(em_tree
, em
, 0);
6917 static noinline
int uncompress_inline(struct btrfs_path
*path
,
6919 size_t pg_offset
, u64 extent_offset
,
6920 struct btrfs_file_extent_item
*item
)
6923 struct extent_buffer
*leaf
= path
->nodes
[0];
6926 unsigned long inline_size
;
6930 WARN_ON(pg_offset
!= 0);
6931 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6932 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6933 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
6934 btrfs_item_nr(path
->slots
[0]));
6935 tmp
= kmalloc(inline_size
, GFP_NOFS
);
6938 ptr
= btrfs_file_extent_inline_start(item
);
6940 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
6942 max_size
= min_t(unsigned long, PAGE_SIZE
, max_size
);
6943 ret
= btrfs_decompress(compress_type
, tmp
, page
,
6944 extent_offset
, inline_size
, max_size
);
6947 * decompression code contains a memset to fill in any space between the end
6948 * of the uncompressed data and the end of max_size in case the decompressed
6949 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6950 * the end of an inline extent and the beginning of the next block, so we
6951 * cover that region here.
6954 if (max_size
+ pg_offset
< PAGE_SIZE
) {
6955 char *map
= kmap(page
);
6956 memset(map
+ pg_offset
+ max_size
, 0, PAGE_SIZE
- max_size
- pg_offset
);
6964 * a bit scary, this does extent mapping from logical file offset to the disk.
6965 * the ugly parts come from merging extents from the disk with the in-ram
6966 * representation. This gets more complex because of the data=ordered code,
6967 * where the in-ram extents might be locked pending data=ordered completion.
6969 * This also copies inline extents directly into the page.
6971 struct extent_map
*btrfs_get_extent(struct btrfs_inode
*inode
,
6973 size_t pg_offset
, u64 start
, u64 len
,
6976 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->vfs_inode
.i_sb
);
6979 u64 extent_start
= 0;
6981 u64 objectid
= btrfs_ino(inode
);
6983 struct btrfs_path
*path
= NULL
;
6984 struct btrfs_root
*root
= inode
->root
;
6985 struct btrfs_file_extent_item
*item
;
6986 struct extent_buffer
*leaf
;
6987 struct btrfs_key found_key
;
6988 struct extent_map
*em
= NULL
;
6989 struct extent_map_tree
*em_tree
= &inode
->extent_tree
;
6990 struct extent_io_tree
*io_tree
= &inode
->io_tree
;
6991 struct btrfs_trans_handle
*trans
= NULL
;
6992 const bool new_inline
= !page
|| create
;
6995 read_lock(&em_tree
->lock
);
6996 em
= lookup_extent_mapping(em_tree
, start
, len
);
6998 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
6999 read_unlock(&em_tree
->lock
);
7002 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
7003 free_extent_map(em
);
7004 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
7005 free_extent_map(em
);
7009 em
= alloc_extent_map();
7014 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
7015 em
->start
= EXTENT_MAP_HOLE
;
7016 em
->orig_start
= EXTENT_MAP_HOLE
;
7018 em
->block_len
= (u64
)-1;
7021 path
= btrfs_alloc_path();
7027 * Chances are we'll be called again, so go ahead and do
7030 path
->reada
= READA_FORWARD
;
7033 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
7034 objectid
, start
, trans
!= NULL
);
7041 if (path
->slots
[0] == 0)
7046 leaf
= path
->nodes
[0];
7047 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
7048 struct btrfs_file_extent_item
);
7049 /* are we inside the extent that was found? */
7050 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7051 found_type
= found_key
.type
;
7052 if (found_key
.objectid
!= objectid
||
7053 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
7055 * If we backup past the first extent we want to move forward
7056 * and see if there is an extent in front of us, otherwise we'll
7057 * say there is a hole for our whole search range which can
7064 found_type
= btrfs_file_extent_type(leaf
, item
);
7065 extent_start
= found_key
.offset
;
7066 if (found_type
== BTRFS_FILE_EXTENT_REG
||
7067 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7068 extent_end
= extent_start
+
7069 btrfs_file_extent_num_bytes(leaf
, item
);
7071 trace_btrfs_get_extent_show_fi_regular(inode
, leaf
, item
,
7073 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
7075 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
7076 extent_end
= ALIGN(extent_start
+ size
,
7077 fs_info
->sectorsize
);
7079 trace_btrfs_get_extent_show_fi_inline(inode
, leaf
, item
,
7084 if (start
>= extent_end
) {
7086 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
7087 ret
= btrfs_next_leaf(root
, path
);
7094 leaf
= path
->nodes
[0];
7096 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7097 if (found_key
.objectid
!= objectid
||
7098 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7100 if (start
+ len
<= found_key
.offset
)
7102 if (start
> found_key
.offset
)
7105 em
->orig_start
= start
;
7106 em
->len
= found_key
.offset
- start
;
7110 btrfs_extent_item_to_extent_map(inode
, path
, item
,
7113 if (found_type
== BTRFS_FILE_EXTENT_REG
||
7114 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7116 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
7120 size_t extent_offset
;
7126 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
7127 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
7128 copy_size
= min_t(u64
, PAGE_SIZE
- pg_offset
,
7129 size
- extent_offset
);
7130 em
->start
= extent_start
+ extent_offset
;
7131 em
->len
= ALIGN(copy_size
, fs_info
->sectorsize
);
7132 em
->orig_block_len
= em
->len
;
7133 em
->orig_start
= em
->start
;
7134 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
7135 if (create
== 0 && !PageUptodate(page
)) {
7136 if (btrfs_file_extent_compression(leaf
, item
) !=
7137 BTRFS_COMPRESS_NONE
) {
7138 ret
= uncompress_inline(path
, page
, pg_offset
,
7139 extent_offset
, item
);
7146 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
7148 if (pg_offset
+ copy_size
< PAGE_SIZE
) {
7149 memset(map
+ pg_offset
+ copy_size
, 0,
7150 PAGE_SIZE
- pg_offset
-
7155 flush_dcache_page(page
);
7156 } else if (create
&& PageUptodate(page
)) {
7160 free_extent_map(em
);
7163 btrfs_release_path(path
);
7164 trans
= btrfs_join_transaction(root
);
7167 return ERR_CAST(trans
);
7171 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
7174 btrfs_mark_buffer_dirty(leaf
);
7176 set_extent_uptodate(io_tree
, em
->start
,
7177 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
7182 em
->orig_start
= start
;
7185 em
->block_start
= EXTENT_MAP_HOLE
;
7186 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
7188 btrfs_release_path(path
);
7189 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
7191 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7192 em
->start
, em
->len
, start
, len
);
7198 write_lock(&em_tree
->lock
);
7199 ret
= add_extent_mapping(em_tree
, em
, 0);
7200 /* it is possible that someone inserted the extent into the tree
7201 * while we had the lock dropped. It is also possible that
7202 * an overlapping map exists in the tree
7204 if (ret
== -EEXIST
) {
7205 struct extent_map
*existing
;
7209 existing
= search_extent_mapping(em_tree
, start
, len
);
7211 * existing will always be non-NULL, since there must be
7212 * extent causing the -EEXIST.
7214 if (start
>= existing
->start
&&
7215 start
< extent_map_end(existing
)) {
7216 free_extent_map(em
);
7221 * The existing extent map is the one nearest to
7222 * the [start, start + len) range which overlaps
7224 err
= merge_extent_mapping(em_tree
, existing
,
7226 free_extent_map(existing
);
7228 free_extent_map(em
);
7233 write_unlock(&em_tree
->lock
);
7236 trace_btrfs_get_extent(root
, inode
, em
);
7238 btrfs_free_path(path
);
7240 ret
= btrfs_end_transaction(trans
);
7245 free_extent_map(em
);
7246 return ERR_PTR(err
);
7248 BUG_ON(!em
); /* Error is always set */
7252 struct extent_map
*btrfs_get_extent_fiemap(struct btrfs_inode
*inode
,
7254 size_t pg_offset
, u64 start
, u64 len
,
7257 struct extent_map
*em
;
7258 struct extent_map
*hole_em
= NULL
;
7259 u64 range_start
= start
;
7265 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
7269 * If our em maps to:
7271 * - a pre-alloc extent,
7272 * there might actually be delalloc bytes behind it.
7274 if (em
->block_start
!= EXTENT_MAP_HOLE
&&
7275 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7280 /* check to see if we've wrapped (len == -1 or similar) */
7289 /* ok, we didn't find anything, lets look for delalloc */
7290 found
= count_range_bits(&inode
->io_tree
, &range_start
,
7291 end
, len
, EXTENT_DELALLOC
, 1);
7292 found_end
= range_start
+ found
;
7293 if (found_end
< range_start
)
7294 found_end
= (u64
)-1;
7297 * we didn't find anything useful, return
7298 * the original results from get_extent()
7300 if (range_start
> end
|| found_end
<= start
) {
7306 /* adjust the range_start to make sure it doesn't
7307 * go backwards from the start they passed in
7309 range_start
= max(start
, range_start
);
7310 found
= found_end
- range_start
;
7313 u64 hole_start
= start
;
7316 em
= alloc_extent_map();
7322 * when btrfs_get_extent can't find anything it
7323 * returns one huge hole
7325 * make sure what it found really fits our range, and
7326 * adjust to make sure it is based on the start from
7330 u64 calc_end
= extent_map_end(hole_em
);
7332 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
7333 free_extent_map(hole_em
);
7336 hole_start
= max(hole_em
->start
, start
);
7337 hole_len
= calc_end
- hole_start
;
7341 if (hole_em
&& range_start
> hole_start
) {
7342 /* our hole starts before our delalloc, so we
7343 * have to return just the parts of the hole
7344 * that go until the delalloc starts
7346 em
->len
= min(hole_len
,
7347 range_start
- hole_start
);
7348 em
->start
= hole_start
;
7349 em
->orig_start
= hole_start
;
7351 * don't adjust block start at all,
7352 * it is fixed at EXTENT_MAP_HOLE
7354 em
->block_start
= hole_em
->block_start
;
7355 em
->block_len
= hole_len
;
7356 if (test_bit(EXTENT_FLAG_PREALLOC
, &hole_em
->flags
))
7357 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
7359 em
->start
= range_start
;
7361 em
->orig_start
= range_start
;
7362 em
->block_start
= EXTENT_MAP_DELALLOC
;
7363 em
->block_len
= found
;
7365 } else if (hole_em
) {
7370 free_extent_map(hole_em
);
7372 free_extent_map(em
);
7373 return ERR_PTR(err
);
7378 static struct extent_map
*btrfs_create_dio_extent(struct inode
*inode
,
7381 const u64 orig_start
,
7382 const u64 block_start
,
7383 const u64 block_len
,
7384 const u64 orig_block_len
,
7385 const u64 ram_bytes
,
7388 struct extent_map
*em
= NULL
;
7391 if (type
!= BTRFS_ORDERED_NOCOW
) {
7392 em
= create_io_em(inode
, start
, len
, orig_start
,
7393 block_start
, block_len
, orig_block_len
,
7395 BTRFS_COMPRESS_NONE
, /* compress_type */
7400 ret
= btrfs_add_ordered_extent_dio(inode
, start
, block_start
,
7401 len
, block_len
, type
);
7404 free_extent_map(em
);
7405 btrfs_drop_extent_cache(BTRFS_I(inode
), start
,
7406 start
+ len
- 1, 0);
7415 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
7418 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7419 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7420 struct extent_map
*em
;
7421 struct btrfs_key ins
;
7425 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
7426 ret
= btrfs_reserve_extent(root
, len
, len
, fs_info
->sectorsize
,
7427 0, alloc_hint
, &ins
, 1, 1);
7429 return ERR_PTR(ret
);
7431 em
= btrfs_create_dio_extent(inode
, start
, ins
.offset
, start
,
7432 ins
.objectid
, ins
.offset
, ins
.offset
,
7433 ins
.offset
, BTRFS_ORDERED_REGULAR
);
7434 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
7436 btrfs_free_reserved_extent(fs_info
, ins
.objectid
,
7443 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7444 * block must be cow'd
7446 noinline
int can_nocow_extent(struct inode
*inode
, u64 offset
, u64
*len
,
7447 u64
*orig_start
, u64
*orig_block_len
,
7450 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7451 struct btrfs_path
*path
;
7453 struct extent_buffer
*leaf
;
7454 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7455 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7456 struct btrfs_file_extent_item
*fi
;
7457 struct btrfs_key key
;
7464 bool nocow
= (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
);
7466 path
= btrfs_alloc_path();
7470 ret
= btrfs_lookup_file_extent(NULL
, root
, path
,
7471 btrfs_ino(BTRFS_I(inode
)), offset
, 0);
7475 slot
= path
->slots
[0];
7478 /* can't find the item, must cow */
7485 leaf
= path
->nodes
[0];
7486 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
7487 if (key
.objectid
!= btrfs_ino(BTRFS_I(inode
)) ||
7488 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7489 /* not our file or wrong item type, must cow */
7493 if (key
.offset
> offset
) {
7494 /* Wrong offset, must cow */
7498 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
7499 found_type
= btrfs_file_extent_type(leaf
, fi
);
7500 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
7501 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
7502 /* not a regular extent, must cow */
7506 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_REG
)
7509 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
7510 if (extent_end
<= offset
)
7513 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7514 if (disk_bytenr
== 0)
7517 if (btrfs_file_extent_compression(leaf
, fi
) ||
7518 btrfs_file_extent_encryption(leaf
, fi
) ||
7519 btrfs_file_extent_other_encoding(leaf
, fi
))
7522 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
7525 *orig_start
= key
.offset
- backref_offset
;
7526 *orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7527 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
7530 if (btrfs_extent_readonly(fs_info
, disk_bytenr
))
7533 num_bytes
= min(offset
+ *len
, extent_end
) - offset
;
7534 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7537 range_end
= round_up(offset
+ num_bytes
,
7538 root
->fs_info
->sectorsize
) - 1;
7539 ret
= test_range_bit(io_tree
, offset
, range_end
,
7540 EXTENT_DELALLOC
, 0, NULL
);
7547 btrfs_release_path(path
);
7550 * look for other files referencing this extent, if we
7551 * find any we must cow
7554 ret
= btrfs_cross_ref_exist(root
, btrfs_ino(BTRFS_I(inode
)),
7555 key
.offset
- backref_offset
, disk_bytenr
);
7562 * adjust disk_bytenr and num_bytes to cover just the bytes
7563 * in this extent we are about to write. If there
7564 * are any csums in that range we have to cow in order
7565 * to keep the csums correct
7567 disk_bytenr
+= backref_offset
;
7568 disk_bytenr
+= offset
- key
.offset
;
7569 if (csum_exist_in_range(fs_info
, disk_bytenr
, num_bytes
))
7572 * all of the above have passed, it is safe to overwrite this extent
7578 btrfs_free_path(path
);
7582 bool btrfs_page_exists_in_range(struct inode
*inode
, loff_t start
, loff_t end
)
7584 struct radix_tree_root
*root
= &inode
->i_mapping
->page_tree
;
7586 void **pagep
= NULL
;
7587 struct page
*page
= NULL
;
7588 unsigned long start_idx
;
7589 unsigned long end_idx
;
7591 start_idx
= start
>> PAGE_SHIFT
;
7594 * end is the last byte in the last page. end == start is legal
7596 end_idx
= end
>> PAGE_SHIFT
;
7600 /* Most of the code in this while loop is lifted from
7601 * find_get_page. It's been modified to begin searching from a
7602 * page and return just the first page found in that range. If the
7603 * found idx is less than or equal to the end idx then we know that
7604 * a page exists. If no pages are found or if those pages are
7605 * outside of the range then we're fine (yay!) */
7606 while (page
== NULL
&&
7607 radix_tree_gang_lookup_slot(root
, &pagep
, NULL
, start_idx
, 1)) {
7608 page
= radix_tree_deref_slot(pagep
);
7609 if (unlikely(!page
))
7612 if (radix_tree_exception(page
)) {
7613 if (radix_tree_deref_retry(page
)) {
7618 * Otherwise, shmem/tmpfs must be storing a swap entry
7619 * here as an exceptional entry: so return it without
7620 * attempting to raise page count.
7623 break; /* TODO: Is this relevant for this use case? */
7626 if (!page_cache_get_speculative(page
)) {
7632 * Has the page moved?
7633 * This is part of the lockless pagecache protocol. See
7634 * include/linux/pagemap.h for details.
7636 if (unlikely(page
!= *pagep
)) {
7643 if (page
->index
<= end_idx
)
7652 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
7653 struct extent_state
**cached_state
, int writing
)
7655 struct btrfs_ordered_extent
*ordered
;
7659 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7662 * We're concerned with the entire range that we're going to be
7663 * doing DIO to, so we need to make sure there's no ordered
7664 * extents in this range.
7666 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), lockstart
,
7667 lockend
- lockstart
+ 1);
7670 * We need to make sure there are no buffered pages in this
7671 * range either, we could have raced between the invalidate in
7672 * generic_file_direct_write and locking the extent. The
7673 * invalidate needs to happen so that reads after a write do not
7678 !btrfs_page_exists_in_range(inode
, lockstart
, lockend
)))
7681 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7682 cached_state
, GFP_NOFS
);
7686 * If we are doing a DIO read and the ordered extent we
7687 * found is for a buffered write, we can not wait for it
7688 * to complete and retry, because if we do so we can
7689 * deadlock with concurrent buffered writes on page
7690 * locks. This happens only if our DIO read covers more
7691 * than one extent map, if at this point has already
7692 * created an ordered extent for a previous extent map
7693 * and locked its range in the inode's io tree, and a
7694 * concurrent write against that previous extent map's
7695 * range and this range started (we unlock the ranges
7696 * in the io tree only when the bios complete and
7697 * buffered writes always lock pages before attempting
7698 * to lock range in the io tree).
7701 test_bit(BTRFS_ORDERED_DIRECT
, &ordered
->flags
))
7702 btrfs_start_ordered_extent(inode
, ordered
, 1);
7705 btrfs_put_ordered_extent(ordered
);
7708 * We could trigger writeback for this range (and wait
7709 * for it to complete) and then invalidate the pages for
7710 * this range (through invalidate_inode_pages2_range()),
7711 * but that can lead us to a deadlock with a concurrent
7712 * call to readpages() (a buffered read or a defrag call
7713 * triggered a readahead) on a page lock due to an
7714 * ordered dio extent we created before but did not have
7715 * yet a corresponding bio submitted (whence it can not
7716 * complete), which makes readpages() wait for that
7717 * ordered extent to complete while holding a lock on
7732 /* The callers of this must take lock_extent() */
7733 static struct extent_map
*create_io_em(struct inode
*inode
, u64 start
, u64 len
,
7734 u64 orig_start
, u64 block_start
,
7735 u64 block_len
, u64 orig_block_len
,
7736 u64 ram_bytes
, int compress_type
,
7739 struct extent_map_tree
*em_tree
;
7740 struct extent_map
*em
;
7741 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7744 ASSERT(type
== BTRFS_ORDERED_PREALLOC
||
7745 type
== BTRFS_ORDERED_COMPRESSED
||
7746 type
== BTRFS_ORDERED_NOCOW
||
7747 type
== BTRFS_ORDERED_REGULAR
);
7749 em_tree
= &BTRFS_I(inode
)->extent_tree
;
7750 em
= alloc_extent_map();
7752 return ERR_PTR(-ENOMEM
);
7755 em
->orig_start
= orig_start
;
7757 em
->block_len
= block_len
;
7758 em
->block_start
= block_start
;
7759 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
7760 em
->orig_block_len
= orig_block_len
;
7761 em
->ram_bytes
= ram_bytes
;
7762 em
->generation
= -1;
7763 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
7764 if (type
== BTRFS_ORDERED_PREALLOC
) {
7765 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
7766 } else if (type
== BTRFS_ORDERED_COMPRESSED
) {
7767 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
7768 em
->compress_type
= compress_type
;
7772 btrfs_drop_extent_cache(BTRFS_I(inode
), em
->start
,
7773 em
->start
+ em
->len
- 1, 0);
7774 write_lock(&em_tree
->lock
);
7775 ret
= add_extent_mapping(em_tree
, em
, 1);
7776 write_unlock(&em_tree
->lock
);
7778 * The caller has taken lock_extent(), who could race with us
7781 } while (ret
== -EEXIST
);
7784 free_extent_map(em
);
7785 return ERR_PTR(ret
);
7788 /* em got 2 refs now, callers needs to do free_extent_map once. */
7792 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
7793 struct buffer_head
*bh_result
, int create
)
7795 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7796 struct extent_map
*em
;
7797 struct extent_state
*cached_state
= NULL
;
7798 struct btrfs_dio_data
*dio_data
= NULL
;
7799 u64 start
= iblock
<< inode
->i_blkbits
;
7800 u64 lockstart
, lockend
;
7801 u64 len
= bh_result
->b_size
;
7802 int unlock_bits
= EXTENT_LOCKED
;
7806 unlock_bits
|= EXTENT_DIRTY
;
7808 len
= min_t(u64
, len
, fs_info
->sectorsize
);
7811 lockend
= start
+ len
- 1;
7813 if (current
->journal_info
) {
7815 * Need to pull our outstanding extents and set journal_info to NULL so
7816 * that anything that needs to check if there's a transaction doesn't get
7819 dio_data
= current
->journal_info
;
7820 current
->journal_info
= NULL
;
7824 * If this errors out it's because we couldn't invalidate pagecache for
7825 * this range and we need to fallback to buffered.
7827 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
,
7833 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0, start
, len
, 0);
7840 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7841 * io. INLINE is special, and we could probably kludge it in here, but
7842 * it's still buffered so for safety lets just fall back to the generic
7845 * For COMPRESSED we _have_ to read the entire extent in so we can
7846 * decompress it, so there will be buffering required no matter what we
7847 * do, so go ahead and fallback to buffered.
7849 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7850 * to buffered IO. Don't blame me, this is the price we pay for using
7853 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
7854 em
->block_start
== EXTENT_MAP_INLINE
) {
7855 free_extent_map(em
);
7860 /* Just a good old fashioned hole, return */
7861 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
7862 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
7863 free_extent_map(em
);
7868 * We don't allocate a new extent in the following cases
7870 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7872 * 2) The extent is marked as PREALLOC. We're good to go here and can
7873 * just use the extent.
7877 len
= min(len
, em
->len
- (start
- em
->start
));
7878 lockstart
= start
+ len
;
7882 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
7883 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
7884 em
->block_start
!= EXTENT_MAP_HOLE
)) {
7886 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
7888 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7889 type
= BTRFS_ORDERED_PREALLOC
;
7891 type
= BTRFS_ORDERED_NOCOW
;
7892 len
= min(len
, em
->len
- (start
- em
->start
));
7893 block_start
= em
->block_start
+ (start
- em
->start
);
7895 if (can_nocow_extent(inode
, start
, &len
, &orig_start
,
7896 &orig_block_len
, &ram_bytes
) == 1 &&
7897 btrfs_inc_nocow_writers(fs_info
, block_start
)) {
7898 struct extent_map
*em2
;
7900 em2
= btrfs_create_dio_extent(inode
, start
, len
,
7901 orig_start
, block_start
,
7902 len
, orig_block_len
,
7904 btrfs_dec_nocow_writers(fs_info
, block_start
);
7905 if (type
== BTRFS_ORDERED_PREALLOC
) {
7906 free_extent_map(em
);
7909 if (em2
&& IS_ERR(em2
)) {
7914 * For inode marked NODATACOW or extent marked PREALLOC,
7915 * use the existing or preallocated extent, so does not
7916 * need to adjust btrfs_space_info's bytes_may_use.
7918 btrfs_free_reserved_data_space_noquota(inode
,
7925 * this will cow the extent, reset the len in case we changed
7928 len
= bh_result
->b_size
;
7929 free_extent_map(em
);
7930 em
= btrfs_new_extent_direct(inode
, start
, len
);
7935 len
= min(len
, em
->len
- (start
- em
->start
));
7937 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
7939 bh_result
->b_size
= len
;
7940 bh_result
->b_bdev
= em
->bdev
;
7941 set_buffer_mapped(bh_result
);
7943 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7944 set_buffer_new(bh_result
);
7947 * Need to update the i_size under the extent lock so buffered
7948 * readers will get the updated i_size when we unlock.
7950 if (!dio_data
->overwrite
&& start
+ len
> i_size_read(inode
))
7951 i_size_write(inode
, start
+ len
);
7953 WARN_ON(dio_data
->reserve
< len
);
7954 dio_data
->reserve
-= len
;
7955 dio_data
->unsubmitted_oe_range_end
= start
+ len
;
7956 current
->journal_info
= dio_data
;
7960 * In the case of write we need to clear and unlock the entire range,
7961 * in the case of read we need to unlock only the end area that we
7962 * aren't using if there is any left over space.
7964 if (lockstart
< lockend
) {
7965 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
7966 lockend
, unlock_bits
, 1, 0,
7967 &cached_state
, GFP_NOFS
);
7969 free_extent_state(cached_state
);
7972 free_extent_map(em
);
7977 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7978 unlock_bits
, 1, 0, &cached_state
, GFP_NOFS
);
7981 current
->journal_info
= dio_data
;
7985 static inline blk_status_t
submit_dio_repair_bio(struct inode
*inode
,
7989 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7992 BUG_ON(bio_op(bio
) == REQ_OP_WRITE
);
7996 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, BTRFS_WQ_ENDIO_DIO_REPAIR
);
8000 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
8006 static int btrfs_check_dio_repairable(struct inode
*inode
,
8007 struct bio
*failed_bio
,
8008 struct io_failure_record
*failrec
,
8011 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8014 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
, failrec
->len
);
8015 if (num_copies
== 1) {
8017 * we only have a single copy of the data, so don't bother with
8018 * all the retry and error correction code that follows. no
8019 * matter what the error is, it is very likely to persist.
8021 btrfs_debug(fs_info
,
8022 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
8023 num_copies
, failrec
->this_mirror
, failed_mirror
);
8027 failrec
->failed_mirror
= failed_mirror
;
8028 failrec
->this_mirror
++;
8029 if (failrec
->this_mirror
== failed_mirror
)
8030 failrec
->this_mirror
++;
8032 if (failrec
->this_mirror
> num_copies
) {
8033 btrfs_debug(fs_info
,
8034 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
8035 num_copies
, failrec
->this_mirror
, failed_mirror
);
8042 static blk_status_t
dio_read_error(struct inode
*inode
, struct bio
*failed_bio
,
8043 struct page
*page
, unsigned int pgoff
,
8044 u64 start
, u64 end
, int failed_mirror
,
8045 bio_end_io_t
*repair_endio
, void *repair_arg
)
8047 struct io_failure_record
*failrec
;
8048 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
8049 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
8052 unsigned int read_mode
= 0;
8055 blk_status_t status
;
8057 BUG_ON(bio_op(failed_bio
) == REQ_OP_WRITE
);
8059 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
8061 return errno_to_blk_status(ret
);
8063 ret
= btrfs_check_dio_repairable(inode
, failed_bio
, failrec
,
8066 free_io_failure(failure_tree
, io_tree
, failrec
);
8067 return BLK_STS_IOERR
;
8070 segs
= bio_segments(failed_bio
);
8072 (failed_bio
->bi_io_vec
->bv_len
> btrfs_inode_sectorsize(inode
)))
8073 read_mode
|= REQ_FAILFAST_DEV
;
8075 isector
= start
- btrfs_io_bio(failed_bio
)->logical
;
8076 isector
>>= inode
->i_sb
->s_blocksize_bits
;
8077 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
8078 pgoff
, isector
, repair_endio
, repair_arg
);
8079 bio_set_op_attrs(bio
, REQ_OP_READ
, read_mode
);
8081 btrfs_debug(BTRFS_I(inode
)->root
->fs_info
,
8082 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
8083 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
8085 status
= submit_dio_repair_bio(inode
, bio
, failrec
->this_mirror
);
8087 free_io_failure(failure_tree
, io_tree
, failrec
);
8094 struct btrfs_retry_complete
{
8095 struct completion done
;
8096 struct inode
*inode
;
8101 static void btrfs_retry_endio_nocsum(struct bio
*bio
)
8103 struct btrfs_retry_complete
*done
= bio
->bi_private
;
8104 struct inode
*inode
= done
->inode
;
8105 struct bio_vec
*bvec
;
8106 struct extent_io_tree
*io_tree
, *failure_tree
;
8112 ASSERT(bio
->bi_vcnt
== 1);
8113 io_tree
= &BTRFS_I(inode
)->io_tree
;
8114 failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
8115 ASSERT(bio
->bi_io_vec
->bv_len
== btrfs_inode_sectorsize(inode
));
8118 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
8119 bio_for_each_segment_all(bvec
, bio
, i
)
8120 clean_io_failure(BTRFS_I(inode
)->root
->fs_info
, failure_tree
,
8121 io_tree
, done
->start
, bvec
->bv_page
,
8122 btrfs_ino(BTRFS_I(inode
)), 0);
8124 complete(&done
->done
);
8128 static blk_status_t
__btrfs_correct_data_nocsum(struct inode
*inode
,
8129 struct btrfs_io_bio
*io_bio
)
8131 struct btrfs_fs_info
*fs_info
;
8132 struct bio_vec bvec
;
8133 struct bvec_iter iter
;
8134 struct btrfs_retry_complete done
;
8140 blk_status_t err
= BLK_STS_OK
;
8142 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
8143 sectorsize
= fs_info
->sectorsize
;
8145 start
= io_bio
->logical
;
8147 io_bio
->bio
.bi_iter
= io_bio
->iter
;
8149 bio_for_each_segment(bvec
, &io_bio
->bio
, iter
) {
8150 nr_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, bvec
.bv_len
);
8151 pgoff
= bvec
.bv_offset
;
8153 next_block_or_try_again
:
8156 init_completion(&done
.done
);
8158 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
.bv_page
,
8159 pgoff
, start
, start
+ sectorsize
- 1,
8161 btrfs_retry_endio_nocsum
, &done
);
8167 wait_for_completion_io(&done
.done
);
8169 if (!done
.uptodate
) {
8170 /* We might have another mirror, so try again */
8171 goto next_block_or_try_again
;
8175 start
+= sectorsize
;
8179 pgoff
+= sectorsize
;
8180 ASSERT(pgoff
< PAGE_SIZE
);
8181 goto next_block_or_try_again
;
8188 static void btrfs_retry_endio(struct bio
*bio
)
8190 struct btrfs_retry_complete
*done
= bio
->bi_private
;
8191 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8192 struct extent_io_tree
*io_tree
, *failure_tree
;
8193 struct inode
*inode
= done
->inode
;
8194 struct bio_vec
*bvec
;
8204 ASSERT(bio
->bi_vcnt
== 1);
8205 ASSERT(bio
->bi_io_vec
->bv_len
== btrfs_inode_sectorsize(done
->inode
));
8207 io_tree
= &BTRFS_I(inode
)->io_tree
;
8208 failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
8210 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
8211 bio_for_each_segment_all(bvec
, bio
, i
) {
8212 ret
= __readpage_endio_check(inode
, io_bio
, i
, bvec
->bv_page
,
8213 bvec
->bv_offset
, done
->start
,
8216 clean_io_failure(BTRFS_I(inode
)->root
->fs_info
,
8217 failure_tree
, io_tree
, done
->start
,
8219 btrfs_ino(BTRFS_I(inode
)),
8225 done
->uptodate
= uptodate
;
8227 complete(&done
->done
);
8231 static blk_status_t
__btrfs_subio_endio_read(struct inode
*inode
,
8232 struct btrfs_io_bio
*io_bio
, blk_status_t err
)
8234 struct btrfs_fs_info
*fs_info
;
8235 struct bio_vec bvec
;
8236 struct bvec_iter iter
;
8237 struct btrfs_retry_complete done
;
8244 bool uptodate
= (err
== 0);
8246 blk_status_t status
;
8248 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
8249 sectorsize
= fs_info
->sectorsize
;
8252 start
= io_bio
->logical
;
8254 io_bio
->bio
.bi_iter
= io_bio
->iter
;
8256 bio_for_each_segment(bvec
, &io_bio
->bio
, iter
) {
8257 nr_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, bvec
.bv_len
);
8259 pgoff
= bvec
.bv_offset
;
8262 csum_pos
= BTRFS_BYTES_TO_BLKS(fs_info
, offset
);
8263 ret
= __readpage_endio_check(inode
, io_bio
, csum_pos
,
8264 bvec
.bv_page
, pgoff
, start
, sectorsize
);
8271 init_completion(&done
.done
);
8273 status
= dio_read_error(inode
, &io_bio
->bio
, bvec
.bv_page
,
8274 pgoff
, start
, start
+ sectorsize
- 1,
8275 io_bio
->mirror_num
, btrfs_retry_endio
,
8282 wait_for_completion_io(&done
.done
);
8284 if (!done
.uptodate
) {
8285 /* We might have another mirror, so try again */
8289 offset
+= sectorsize
;
8290 start
+= sectorsize
;
8296 pgoff
+= sectorsize
;
8297 ASSERT(pgoff
< PAGE_SIZE
);
8305 static blk_status_t
btrfs_subio_endio_read(struct inode
*inode
,
8306 struct btrfs_io_bio
*io_bio
, blk_status_t err
)
8308 bool skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
8312 return __btrfs_correct_data_nocsum(inode
, io_bio
);
8316 return __btrfs_subio_endio_read(inode
, io_bio
, err
);
8320 static void btrfs_endio_direct_read(struct bio
*bio
)
8322 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8323 struct inode
*inode
= dip
->inode
;
8324 struct bio
*dio_bio
;
8325 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8326 blk_status_t err
= bio
->bi_status
;
8328 if (dip
->flags
& BTRFS_DIO_ORIG_BIO_SUBMITTED
)
8329 err
= btrfs_subio_endio_read(inode
, io_bio
, err
);
8331 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
8332 dip
->logical_offset
+ dip
->bytes
- 1);
8333 dio_bio
= dip
->dio_bio
;
8337 dio_bio
->bi_status
= err
;
8338 dio_end_io(dio_bio
);
8341 io_bio
->end_io(io_bio
, blk_status_to_errno(err
));
8345 static void __endio_write_update_ordered(struct inode
*inode
,
8346 const u64 offset
, const u64 bytes
,
8347 const bool uptodate
)
8349 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8350 struct btrfs_ordered_extent
*ordered
= NULL
;
8351 struct btrfs_workqueue
*wq
;
8352 btrfs_work_func_t func
;
8353 u64 ordered_offset
= offset
;
8354 u64 ordered_bytes
= bytes
;
8358 if (btrfs_is_free_space_inode(BTRFS_I(inode
))) {
8359 wq
= fs_info
->endio_freespace_worker
;
8360 func
= btrfs_freespace_write_helper
;
8362 wq
= fs_info
->endio_write_workers
;
8363 func
= btrfs_endio_write_helper
;
8367 last_offset
= ordered_offset
;
8368 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
8375 btrfs_init_work(&ordered
->work
, func
, finish_ordered_fn
, NULL
, NULL
);
8376 btrfs_queue_work(wq
, &ordered
->work
);
8379 * If btrfs_dec_test_ordered_pending does not find any ordered extent
8380 * in the range, we can exit.
8382 if (ordered_offset
== last_offset
)
8385 * our bio might span multiple ordered extents. If we haven't
8386 * completed the accounting for the whole dio, go back and try again
8388 if (ordered_offset
< offset
+ bytes
) {
8389 ordered_bytes
= offset
+ bytes
- ordered_offset
;
8395 static void btrfs_endio_direct_write(struct bio
*bio
)
8397 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8398 struct bio
*dio_bio
= dip
->dio_bio
;
8400 __endio_write_update_ordered(dip
->inode
, dip
->logical_offset
,
8401 dip
->bytes
, !bio
->bi_status
);
8405 dio_bio
->bi_status
= bio
->bi_status
;
8406 dio_end_io(dio_bio
);
8410 static blk_status_t
__btrfs_submit_bio_start_direct_io(void *private_data
,
8411 struct bio
*bio
, int mirror_num
,
8412 unsigned long bio_flags
, u64 offset
)
8414 struct inode
*inode
= private_data
;
8416 ret
= btrfs_csum_one_bio(inode
, bio
, offset
, 1);
8417 BUG_ON(ret
); /* -ENOMEM */
8421 static void btrfs_end_dio_bio(struct bio
*bio
)
8423 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8424 blk_status_t err
= bio
->bi_status
;
8427 btrfs_warn(BTRFS_I(dip
->inode
)->root
->fs_info
,
8428 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8429 btrfs_ino(BTRFS_I(dip
->inode
)), bio_op(bio
),
8431 (unsigned long long)bio
->bi_iter
.bi_sector
,
8432 bio
->bi_iter
.bi_size
, err
);
8434 if (dip
->subio_endio
)
8435 err
= dip
->subio_endio(dip
->inode
, btrfs_io_bio(bio
), err
);
8441 * before atomic variable goto zero, we must make sure
8442 * dip->errors is perceived to be set.
8444 smp_mb__before_atomic();
8447 /* if there are more bios still pending for this dio, just exit */
8448 if (!atomic_dec_and_test(&dip
->pending_bios
))
8452 bio_io_error(dip
->orig_bio
);
8454 dip
->dio_bio
->bi_status
= BLK_STS_OK
;
8455 bio_endio(dip
->orig_bio
);
8461 static inline blk_status_t
btrfs_lookup_and_bind_dio_csum(struct inode
*inode
,
8462 struct btrfs_dio_private
*dip
,
8466 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8467 struct btrfs_io_bio
*orig_io_bio
= btrfs_io_bio(dip
->orig_bio
);
8471 * We load all the csum data we need when we submit
8472 * the first bio to reduce the csum tree search and
8475 if (dip
->logical_offset
== file_offset
) {
8476 ret
= btrfs_lookup_bio_sums_dio(inode
, dip
->orig_bio
,
8482 if (bio
== dip
->orig_bio
)
8485 file_offset
-= dip
->logical_offset
;
8486 file_offset
>>= inode
->i_sb
->s_blocksize_bits
;
8487 io_bio
->csum
= (u8
*)(((u32
*)orig_io_bio
->csum
) + file_offset
);
8492 static inline blk_status_t
8493 __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
, u64 file_offset
,
8496 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8497 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8498 bool write
= bio_op(bio
) == REQ_OP_WRITE
;
8502 async_submit
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
8507 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, BTRFS_WQ_ENDIO_DATA
);
8512 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
8515 if (write
&& async_submit
) {
8516 ret
= btrfs_wq_submit_bio(fs_info
, bio
, 0, 0,
8518 __btrfs_submit_bio_start_direct_io
,
8519 __btrfs_submit_bio_done
);
8523 * If we aren't doing async submit, calculate the csum of the
8526 ret
= btrfs_csum_one_bio(inode
, bio
, file_offset
, 1);
8530 ret
= btrfs_lookup_and_bind_dio_csum(inode
, dip
, bio
,
8536 ret
= btrfs_map_bio(fs_info
, bio
, 0, 0);
8542 static int btrfs_submit_direct_hook(struct btrfs_dio_private
*dip
)
8544 struct inode
*inode
= dip
->inode
;
8545 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8547 struct bio
*orig_bio
= dip
->orig_bio
;
8548 u64 start_sector
= orig_bio
->bi_iter
.bi_sector
;
8549 u64 file_offset
= dip
->logical_offset
;
8551 int async_submit
= 0;
8553 int clone_offset
= 0;
8556 blk_status_t status
;
8558 map_length
= orig_bio
->bi_iter
.bi_size
;
8559 submit_len
= map_length
;
8560 ret
= btrfs_map_block(fs_info
, btrfs_op(orig_bio
), start_sector
<< 9,
8561 &map_length
, NULL
, 0);
8565 if (map_length
>= submit_len
) {
8567 dip
->flags
|= BTRFS_DIO_ORIG_BIO_SUBMITTED
;
8571 /* async crcs make it difficult to collect full stripe writes. */
8572 if (btrfs_data_alloc_profile(fs_info
) & BTRFS_BLOCK_GROUP_RAID56_MASK
)
8578 ASSERT(map_length
<= INT_MAX
);
8579 atomic_inc(&dip
->pending_bios
);
8581 clone_len
= min_t(int, submit_len
, map_length
);
8584 * This will never fail as it's passing GPF_NOFS and
8585 * the allocation is backed by btrfs_bioset.
8587 bio
= btrfs_bio_clone_partial(orig_bio
, clone_offset
,
8589 bio
->bi_private
= dip
;
8590 bio
->bi_end_io
= btrfs_end_dio_bio
;
8591 btrfs_io_bio(bio
)->logical
= file_offset
;
8593 ASSERT(submit_len
>= clone_len
);
8594 submit_len
-= clone_len
;
8595 if (submit_len
== 0)
8599 * Increase the count before we submit the bio so we know
8600 * the end IO handler won't happen before we increase the
8601 * count. Otherwise, the dip might get freed before we're
8602 * done setting it up.
8604 atomic_inc(&dip
->pending_bios
);
8606 status
= __btrfs_submit_dio_bio(bio
, inode
, file_offset
,
8610 atomic_dec(&dip
->pending_bios
);
8614 clone_offset
+= clone_len
;
8615 start_sector
+= clone_len
>> 9;
8616 file_offset
+= clone_len
;
8618 map_length
= submit_len
;
8619 ret
= btrfs_map_block(fs_info
, btrfs_op(orig_bio
),
8620 start_sector
<< 9, &map_length
, NULL
, 0);
8623 } while (submit_len
> 0);
8626 status
= __btrfs_submit_dio_bio(bio
, inode
, file_offset
, async_submit
);
8634 * before atomic variable goto zero, we must
8635 * make sure dip->errors is perceived to be set.
8637 smp_mb__before_atomic();
8638 if (atomic_dec_and_test(&dip
->pending_bios
))
8639 bio_io_error(dip
->orig_bio
);
8641 /* bio_end_io() will handle error, so we needn't return it */
8645 static void btrfs_submit_direct(struct bio
*dio_bio
, struct inode
*inode
,
8648 struct btrfs_dio_private
*dip
= NULL
;
8649 struct bio
*bio
= NULL
;
8650 struct btrfs_io_bio
*io_bio
;
8651 bool write
= (bio_op(dio_bio
) == REQ_OP_WRITE
);
8654 bio
= btrfs_bio_clone(dio_bio
);
8656 dip
= kzalloc(sizeof(*dip
), GFP_NOFS
);
8662 dip
->private = dio_bio
->bi_private
;
8664 dip
->logical_offset
= file_offset
;
8665 dip
->bytes
= dio_bio
->bi_iter
.bi_size
;
8666 dip
->disk_bytenr
= (u64
)dio_bio
->bi_iter
.bi_sector
<< 9;
8667 bio
->bi_private
= dip
;
8668 dip
->orig_bio
= bio
;
8669 dip
->dio_bio
= dio_bio
;
8670 atomic_set(&dip
->pending_bios
, 0);
8671 io_bio
= btrfs_io_bio(bio
);
8672 io_bio
->logical
= file_offset
;
8675 bio
->bi_end_io
= btrfs_endio_direct_write
;
8677 bio
->bi_end_io
= btrfs_endio_direct_read
;
8678 dip
->subio_endio
= btrfs_subio_endio_read
;
8682 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8683 * even if we fail to submit a bio, because in such case we do the
8684 * corresponding error handling below and it must not be done a second
8685 * time by btrfs_direct_IO().
8688 struct btrfs_dio_data
*dio_data
= current
->journal_info
;
8690 dio_data
->unsubmitted_oe_range_end
= dip
->logical_offset
+
8692 dio_data
->unsubmitted_oe_range_start
=
8693 dio_data
->unsubmitted_oe_range_end
;
8696 ret
= btrfs_submit_direct_hook(dip
);
8701 io_bio
->end_io(io_bio
, ret
);
8705 * If we arrived here it means either we failed to submit the dip
8706 * or we either failed to clone the dio_bio or failed to allocate the
8707 * dip. If we cloned the dio_bio and allocated the dip, we can just
8708 * call bio_endio against our io_bio so that we get proper resource
8709 * cleanup if we fail to submit the dip, otherwise, we must do the
8710 * same as btrfs_endio_direct_[write|read] because we can't call these
8711 * callbacks - they require an allocated dip and a clone of dio_bio.
8716 * The end io callbacks free our dip, do the final put on bio
8717 * and all the cleanup and final put for dio_bio (through
8724 __endio_write_update_ordered(inode
,
8726 dio_bio
->bi_iter
.bi_size
,
8729 unlock_extent(&BTRFS_I(inode
)->io_tree
, file_offset
,
8730 file_offset
+ dio_bio
->bi_iter
.bi_size
- 1);
8732 dio_bio
->bi_status
= BLK_STS_IOERR
;
8734 * Releases and cleans up our dio_bio, no need to bio_put()
8735 * nor bio_endio()/bio_io_error() against dio_bio.
8737 dio_end_io(dio_bio
);
8744 static ssize_t
check_direct_IO(struct btrfs_fs_info
*fs_info
,
8745 const struct iov_iter
*iter
, loff_t offset
)
8749 unsigned int blocksize_mask
= fs_info
->sectorsize
- 1;
8750 ssize_t retval
= -EINVAL
;
8752 if (offset
& blocksize_mask
)
8755 if (iov_iter_alignment(iter
) & blocksize_mask
)
8758 /* If this is a write we don't need to check anymore */
8759 if (iov_iter_rw(iter
) != READ
|| !iter_is_iovec(iter
))
8762 * Check to make sure we don't have duplicate iov_base's in this
8763 * iovec, if so return EINVAL, otherwise we'll get csum errors
8764 * when reading back.
8766 for (seg
= 0; seg
< iter
->nr_segs
; seg
++) {
8767 for (i
= seg
+ 1; i
< iter
->nr_segs
; i
++) {
8768 if (iter
->iov
[seg
].iov_base
== iter
->iov
[i
].iov_base
)
8777 static ssize_t
btrfs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
8779 struct file
*file
= iocb
->ki_filp
;
8780 struct inode
*inode
= file
->f_mapping
->host
;
8781 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8782 struct btrfs_dio_data dio_data
= { 0 };
8783 struct extent_changeset
*data_reserved
= NULL
;
8784 loff_t offset
= iocb
->ki_pos
;
8788 bool relock
= false;
8791 if (check_direct_IO(fs_info
, iter
, offset
))
8794 inode_dio_begin(inode
);
8797 * The generic stuff only does filemap_write_and_wait_range, which
8798 * isn't enough if we've written compressed pages to this area, so
8799 * we need to flush the dirty pages again to make absolutely sure
8800 * that any outstanding dirty pages are on disk.
8802 count
= iov_iter_count(iter
);
8803 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
8804 &BTRFS_I(inode
)->runtime_flags
))
8805 filemap_fdatawrite_range(inode
->i_mapping
, offset
,
8806 offset
+ count
- 1);
8808 if (iov_iter_rw(iter
) == WRITE
) {
8810 * If the write DIO is beyond the EOF, we need update
8811 * the isize, but it is protected by i_mutex. So we can
8812 * not unlock the i_mutex at this case.
8814 if (offset
+ count
<= inode
->i_size
) {
8815 dio_data
.overwrite
= 1;
8816 inode_unlock(inode
);
8818 } else if (iocb
->ki_flags
& IOCB_NOWAIT
) {
8822 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
,
8828 * We need to know how many extents we reserved so that we can
8829 * do the accounting properly if we go over the number we
8830 * originally calculated. Abuse current->journal_info for this.
8832 dio_data
.reserve
= round_up(count
,
8833 fs_info
->sectorsize
);
8834 dio_data
.unsubmitted_oe_range_start
= (u64
)offset
;
8835 dio_data
.unsubmitted_oe_range_end
= (u64
)offset
;
8836 current
->journal_info
= &dio_data
;
8837 down_read(&BTRFS_I(inode
)->dio_sem
);
8838 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK
,
8839 &BTRFS_I(inode
)->runtime_flags
)) {
8840 inode_dio_end(inode
);
8841 flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
8845 ret
= __blockdev_direct_IO(iocb
, inode
,
8846 fs_info
->fs_devices
->latest_bdev
,
8847 iter
, btrfs_get_blocks_direct
, NULL
,
8848 btrfs_submit_direct
, flags
);
8849 if (iov_iter_rw(iter
) == WRITE
) {
8850 up_read(&BTRFS_I(inode
)->dio_sem
);
8851 current
->journal_info
= NULL
;
8852 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
8853 if (dio_data
.reserve
)
8854 btrfs_delalloc_release_space(inode
, data_reserved
,
8855 offset
, dio_data
.reserve
);
8857 * On error we might have left some ordered extents
8858 * without submitting corresponding bios for them, so
8859 * cleanup them up to avoid other tasks getting them
8860 * and waiting for them to complete forever.
8862 if (dio_data
.unsubmitted_oe_range_start
<
8863 dio_data
.unsubmitted_oe_range_end
)
8864 __endio_write_update_ordered(inode
,
8865 dio_data
.unsubmitted_oe_range_start
,
8866 dio_data
.unsubmitted_oe_range_end
-
8867 dio_data
.unsubmitted_oe_range_start
,
8869 } else if (ret
>= 0 && (size_t)ret
< count
)
8870 btrfs_delalloc_release_space(inode
, data_reserved
,
8871 offset
, count
- (size_t)ret
);
8872 btrfs_delalloc_release_extents(BTRFS_I(inode
), count
);
8876 inode_dio_end(inode
);
8880 extent_changeset_free(data_reserved
);
8884 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8886 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
8887 __u64 start
, __u64 len
)
8891 ret
= fiemap_check_flags(fieinfo
, BTRFS_FIEMAP_FLAGS
);
8895 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
8898 int btrfs_readpage(struct file
*file
, struct page
*page
)
8900 struct extent_io_tree
*tree
;
8901 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8902 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
8905 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
8907 struct extent_io_tree
*tree
;
8908 struct inode
*inode
= page
->mapping
->host
;
8911 if (current
->flags
& PF_MEMALLOC
) {
8912 redirty_page_for_writepage(wbc
, page
);
8918 * If we are under memory pressure we will call this directly from the
8919 * VM, we need to make sure we have the inode referenced for the ordered
8920 * extent. If not just return like we didn't do anything.
8922 if (!igrab(inode
)) {
8923 redirty_page_for_writepage(wbc
, page
);
8924 return AOP_WRITEPAGE_ACTIVATE
;
8926 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8927 ret
= extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
8928 btrfs_add_delayed_iput(inode
);
8932 static int btrfs_writepages(struct address_space
*mapping
,
8933 struct writeback_control
*wbc
)
8935 struct extent_io_tree
*tree
;
8937 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8938 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
8942 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
8943 struct list_head
*pages
, unsigned nr_pages
)
8945 struct extent_io_tree
*tree
;
8946 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8947 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
8950 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8952 struct extent_io_tree
*tree
;
8953 struct extent_map_tree
*map
;
8956 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8957 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
8958 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
8960 ClearPagePrivate(page
);
8961 set_page_private(page
, 0);
8967 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8969 if (PageWriteback(page
) || PageDirty(page
))
8971 return __btrfs_releasepage(page
, gfp_flags
);
8974 static void btrfs_invalidatepage(struct page
*page
, unsigned int offset
,
8975 unsigned int length
)
8977 struct inode
*inode
= page
->mapping
->host
;
8978 struct extent_io_tree
*tree
;
8979 struct btrfs_ordered_extent
*ordered
;
8980 struct extent_state
*cached_state
= NULL
;
8981 u64 page_start
= page_offset(page
);
8982 u64 page_end
= page_start
+ PAGE_SIZE
- 1;
8985 int inode_evicting
= inode
->i_state
& I_FREEING
;
8988 * we have the page locked, so new writeback can't start,
8989 * and the dirty bit won't be cleared while we are here.
8991 * Wait for IO on this page so that we can safely clear
8992 * the PagePrivate2 bit and do ordered accounting
8994 wait_on_page_writeback(page
);
8996 tree
= &BTRFS_I(inode
)->io_tree
;
8998 btrfs_releasepage(page
, GFP_NOFS
);
9002 if (!inode_evicting
)
9003 lock_extent_bits(tree
, page_start
, page_end
, &cached_state
);
9006 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), start
,
9007 page_end
- start
+ 1);
9009 end
= min(page_end
, ordered
->file_offset
+ ordered
->len
- 1);
9011 * IO on this page will never be started, so we need
9012 * to account for any ordered extents now
9014 if (!inode_evicting
)
9015 clear_extent_bit(tree
, start
, end
,
9016 EXTENT_DIRTY
| EXTENT_DELALLOC
|
9017 EXTENT_DELALLOC_NEW
|
9018 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
9019 EXTENT_DEFRAG
, 1, 0, &cached_state
,
9022 * whoever cleared the private bit is responsible
9023 * for the finish_ordered_io
9025 if (TestClearPagePrivate2(page
)) {
9026 struct btrfs_ordered_inode_tree
*tree
;
9029 tree
= &BTRFS_I(inode
)->ordered_tree
;
9031 spin_lock_irq(&tree
->lock
);
9032 set_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
);
9033 new_len
= start
- ordered
->file_offset
;
9034 if (new_len
< ordered
->truncated_len
)
9035 ordered
->truncated_len
= new_len
;
9036 spin_unlock_irq(&tree
->lock
);
9038 if (btrfs_dec_test_ordered_pending(inode
, &ordered
,
9040 end
- start
+ 1, 1))
9041 btrfs_finish_ordered_io(ordered
);
9043 btrfs_put_ordered_extent(ordered
);
9044 if (!inode_evicting
) {
9045 cached_state
= NULL
;
9046 lock_extent_bits(tree
, start
, end
,
9051 if (start
< page_end
)
9056 * Qgroup reserved space handler
9057 * Page here will be either
9058 * 1) Already written to disk
9059 * In this case, its reserved space is released from data rsv map
9060 * and will be freed by delayed_ref handler finally.
9061 * So even we call qgroup_free_data(), it won't decrease reserved
9063 * 2) Not written to disk
9064 * This means the reserved space should be freed here. However,
9065 * if a truncate invalidates the page (by clearing PageDirty)
9066 * and the page is accounted for while allocating extent
9067 * in btrfs_check_data_free_space() we let delayed_ref to
9068 * free the entire extent.
9070 if (PageDirty(page
))
9071 btrfs_qgroup_free_data(inode
, NULL
, page_start
, PAGE_SIZE
);
9072 if (!inode_evicting
) {
9073 clear_extent_bit(tree
, page_start
, page_end
,
9074 EXTENT_LOCKED
| EXTENT_DIRTY
|
9075 EXTENT_DELALLOC
| EXTENT_DELALLOC_NEW
|
9076 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
, 1, 1,
9077 &cached_state
, GFP_NOFS
);
9079 __btrfs_releasepage(page
, GFP_NOFS
);
9082 ClearPageChecked(page
);
9083 if (PagePrivate(page
)) {
9084 ClearPagePrivate(page
);
9085 set_page_private(page
, 0);
9091 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
9092 * called from a page fault handler when a page is first dirtied. Hence we must
9093 * be careful to check for EOF conditions here. We set the page up correctly
9094 * for a written page which means we get ENOSPC checking when writing into
9095 * holes and correct delalloc and unwritten extent mapping on filesystems that
9096 * support these features.
9098 * We are not allowed to take the i_mutex here so we have to play games to
9099 * protect against truncate races as the page could now be beyond EOF. Because
9100 * vmtruncate() writes the inode size before removing pages, once we have the
9101 * page lock we can determine safely if the page is beyond EOF. If it is not
9102 * beyond EOF, then the page is guaranteed safe against truncation until we
9105 int btrfs_page_mkwrite(struct vm_fault
*vmf
)
9107 struct page
*page
= vmf
->page
;
9108 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
9109 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
9110 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
9111 struct btrfs_ordered_extent
*ordered
;
9112 struct extent_state
*cached_state
= NULL
;
9113 struct extent_changeset
*data_reserved
= NULL
;
9115 unsigned long zero_start
;
9124 reserved_space
= PAGE_SIZE
;
9126 sb_start_pagefault(inode
->i_sb
);
9127 page_start
= page_offset(page
);
9128 page_end
= page_start
+ PAGE_SIZE
- 1;
9132 * Reserving delalloc space after obtaining the page lock can lead to
9133 * deadlock. For example, if a dirty page is locked by this function
9134 * and the call to btrfs_delalloc_reserve_space() ends up triggering
9135 * dirty page write out, then the btrfs_writepage() function could
9136 * end up waiting indefinitely to get a lock on the page currently
9137 * being processed by btrfs_page_mkwrite() function.
9139 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
, page_start
,
9142 ret
= file_update_time(vmf
->vma
->vm_file
);
9148 else /* -ENOSPC, -EIO, etc */
9149 ret
= VM_FAULT_SIGBUS
;
9155 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
9158 size
= i_size_read(inode
);
9160 if ((page
->mapping
!= inode
->i_mapping
) ||
9161 (page_start
>= size
)) {
9162 /* page got truncated out from underneath us */
9165 wait_on_page_writeback(page
);
9167 lock_extent_bits(io_tree
, page_start
, page_end
, &cached_state
);
9168 set_page_extent_mapped(page
);
9171 * we can't set the delalloc bits if there are pending ordered
9172 * extents. Drop our locks and wait for them to finish
9174 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), page_start
,
9177 unlock_extent_cached(io_tree
, page_start
, page_end
,
9178 &cached_state
, GFP_NOFS
);
9180 btrfs_start_ordered_extent(inode
, ordered
, 1);
9181 btrfs_put_ordered_extent(ordered
);
9185 if (page
->index
== ((size
- 1) >> PAGE_SHIFT
)) {
9186 reserved_space
= round_up(size
- page_start
,
9187 fs_info
->sectorsize
);
9188 if (reserved_space
< PAGE_SIZE
) {
9189 end
= page_start
+ reserved_space
- 1;
9190 btrfs_delalloc_release_space(inode
, data_reserved
,
9191 page_start
, PAGE_SIZE
- reserved_space
);
9196 * page_mkwrite gets called when the page is firstly dirtied after it's
9197 * faulted in, but write(2) could also dirty a page and set delalloc
9198 * bits, thus in this case for space account reason, we still need to
9199 * clear any delalloc bits within this page range since we have to
9200 * reserve data&meta space before lock_page() (see above comments).
9202 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, end
,
9203 EXTENT_DIRTY
| EXTENT_DELALLOC
|
9204 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
9205 0, 0, &cached_state
, GFP_NOFS
);
9207 ret
= btrfs_set_extent_delalloc(inode
, page_start
, end
, 0,
9210 unlock_extent_cached(io_tree
, page_start
, page_end
,
9211 &cached_state
, GFP_NOFS
);
9212 ret
= VM_FAULT_SIGBUS
;
9217 /* page is wholly or partially inside EOF */
9218 if (page_start
+ PAGE_SIZE
> size
)
9219 zero_start
= size
& ~PAGE_MASK
;
9221 zero_start
= PAGE_SIZE
;
9223 if (zero_start
!= PAGE_SIZE
) {
9225 memset(kaddr
+ zero_start
, 0, PAGE_SIZE
- zero_start
);
9226 flush_dcache_page(page
);
9229 ClearPageChecked(page
);
9230 set_page_dirty(page
);
9231 SetPageUptodate(page
);
9233 BTRFS_I(inode
)->last_trans
= fs_info
->generation
;
9234 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
9235 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
9237 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
9241 btrfs_delalloc_release_extents(BTRFS_I(inode
), PAGE_SIZE
);
9242 sb_end_pagefault(inode
->i_sb
);
9243 extent_changeset_free(data_reserved
);
9244 return VM_FAULT_LOCKED
;
9248 btrfs_delalloc_release_extents(BTRFS_I(inode
), PAGE_SIZE
);
9249 btrfs_delalloc_release_space(inode
, data_reserved
, page_start
,
9252 sb_end_pagefault(inode
->i_sb
);
9253 extent_changeset_free(data_reserved
);
9257 static int btrfs_truncate(struct inode
*inode
)
9259 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
9260 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9261 struct btrfs_block_rsv
*rsv
;
9264 struct btrfs_trans_handle
*trans
;
9265 u64 mask
= fs_info
->sectorsize
- 1;
9266 u64 min_size
= btrfs_calc_trunc_metadata_size(fs_info
, 1);
9268 ret
= btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
),
9274 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9275 * 3 things going on here
9277 * 1) We need to reserve space for our orphan item and the space to
9278 * delete our orphan item. Lord knows we don't want to have a dangling
9279 * orphan item because we didn't reserve space to remove it.
9281 * 2) We need to reserve space to update our inode.
9283 * 3) We need to have something to cache all the space that is going to
9284 * be free'd up by the truncate operation, but also have some slack
9285 * space reserved in case it uses space during the truncate (thank you
9286 * very much snapshotting).
9288 * And we need these to all be separate. The fact is we can use a lot of
9289 * space doing the truncate, and we have no earthly idea how much space
9290 * we will use, so we need the truncate reservation to be separate so it
9291 * doesn't end up using space reserved for updating the inode or
9292 * removing the orphan item. We also need to be able to stop the
9293 * transaction and start a new one, which means we need to be able to
9294 * update the inode several times, and we have no idea of knowing how
9295 * many times that will be, so we can't just reserve 1 item for the
9296 * entirety of the operation, so that has to be done separately as well.
9297 * Then there is the orphan item, which does indeed need to be held on
9298 * to for the whole operation, and we need nobody to touch this reserved
9299 * space except the orphan code.
9301 * So that leaves us with
9303 * 1) root->orphan_block_rsv - for the orphan deletion.
9304 * 2) rsv - for the truncate reservation, which we will steal from the
9305 * transaction reservation.
9306 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9307 * updating the inode.
9309 rsv
= btrfs_alloc_block_rsv(fs_info
, BTRFS_BLOCK_RSV_TEMP
);
9312 rsv
->size
= min_size
;
9316 * 1 for the truncate slack space
9317 * 1 for updating the inode.
9319 trans
= btrfs_start_transaction(root
, 2);
9320 if (IS_ERR(trans
)) {
9321 err
= PTR_ERR(trans
);
9325 /* Migrate the slack space for the truncate to our reserve */
9326 ret
= btrfs_block_rsv_migrate(&fs_info
->trans_block_rsv
, rsv
,
9331 * So if we truncate and then write and fsync we normally would just
9332 * write the extents that changed, which is a problem if we need to
9333 * first truncate that entire inode. So set this flag so we write out
9334 * all of the extents in the inode to the sync log so we're completely
9337 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
9338 trans
->block_rsv
= rsv
;
9341 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
9343 BTRFS_EXTENT_DATA_KEY
);
9344 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
9345 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
) {
9351 ret
= btrfs_update_inode(trans
, root
, inode
);
9357 btrfs_end_transaction(trans
);
9358 btrfs_btree_balance_dirty(fs_info
);
9360 trans
= btrfs_start_transaction(root
, 2);
9361 if (IS_ERR(trans
)) {
9362 ret
= err
= PTR_ERR(trans
);
9367 btrfs_block_rsv_release(fs_info
, rsv
, -1);
9368 ret
= btrfs_block_rsv_migrate(&fs_info
->trans_block_rsv
,
9370 BUG_ON(ret
); /* shouldn't happen */
9371 trans
->block_rsv
= rsv
;
9375 * We can't call btrfs_truncate_block inside a trans handle as we could
9376 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9377 * we've truncated everything except the last little bit, and can do
9378 * btrfs_truncate_block and then update the disk_i_size.
9380 if (ret
== NEED_TRUNCATE_BLOCK
) {
9381 btrfs_end_transaction(trans
);
9382 btrfs_btree_balance_dirty(fs_info
);
9384 ret
= btrfs_truncate_block(inode
, inode
->i_size
, 0, 0);
9387 trans
= btrfs_start_transaction(root
, 1);
9388 if (IS_ERR(trans
)) {
9389 ret
= PTR_ERR(trans
);
9392 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
9395 if (ret
== 0 && inode
->i_nlink
> 0) {
9396 trans
->block_rsv
= root
->orphan_block_rsv
;
9397 ret
= btrfs_orphan_del(trans
, BTRFS_I(inode
));
9403 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
9404 ret
= btrfs_update_inode(trans
, root
, inode
);
9408 ret
= btrfs_end_transaction(trans
);
9409 btrfs_btree_balance_dirty(fs_info
);
9412 btrfs_free_block_rsv(fs_info
, rsv
);
9421 * create a new subvolume directory/inode (helper for the ioctl).
9423 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
9424 struct btrfs_root
*new_root
,
9425 struct btrfs_root
*parent_root
,
9428 struct inode
*inode
;
9432 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
9433 new_dirid
, new_dirid
,
9434 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
9437 return PTR_ERR(inode
);
9438 inode
->i_op
= &btrfs_dir_inode_operations
;
9439 inode
->i_fop
= &btrfs_dir_file_operations
;
9441 set_nlink(inode
, 1);
9442 btrfs_i_size_write(BTRFS_I(inode
), 0);
9443 unlock_new_inode(inode
);
9445 err
= btrfs_subvol_inherit_props(trans
, new_root
, parent_root
);
9447 btrfs_err(new_root
->fs_info
,
9448 "error inheriting subvolume %llu properties: %d",
9449 new_root
->root_key
.objectid
, err
);
9451 err
= btrfs_update_inode(trans
, new_root
, inode
);
9457 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
9459 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
9460 struct btrfs_inode
*ei
;
9461 struct inode
*inode
;
9463 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
9470 ei
->last_sub_trans
= 0;
9471 ei
->logged_trans
= 0;
9472 ei
->delalloc_bytes
= 0;
9473 ei
->new_delalloc_bytes
= 0;
9474 ei
->defrag_bytes
= 0;
9475 ei
->disk_i_size
= 0;
9478 ei
->index_cnt
= (u64
)-1;
9480 ei
->last_unlink_trans
= 0;
9481 ei
->last_log_commit
= 0;
9482 ei
->delayed_iput_count
= 0;
9484 spin_lock_init(&ei
->lock
);
9485 ei
->outstanding_extents
= 0;
9486 if (sb
->s_magic
!= BTRFS_TEST_MAGIC
)
9487 btrfs_init_metadata_block_rsv(fs_info
, &ei
->block_rsv
,
9488 BTRFS_BLOCK_RSV_DELALLOC
);
9489 ei
->runtime_flags
= 0;
9490 ei
->prop_compress
= BTRFS_COMPRESS_NONE
;
9491 ei
->defrag_compress
= BTRFS_COMPRESS_NONE
;
9493 ei
->delayed_node
= NULL
;
9495 ei
->i_otime
.tv_sec
= 0;
9496 ei
->i_otime
.tv_nsec
= 0;
9498 inode
= &ei
->vfs_inode
;
9499 extent_map_tree_init(&ei
->extent_tree
);
9500 extent_io_tree_init(&ei
->io_tree
, inode
);
9501 extent_io_tree_init(&ei
->io_failure_tree
, inode
);
9502 ei
->io_tree
.track_uptodate
= 1;
9503 ei
->io_failure_tree
.track_uptodate
= 1;
9504 atomic_set(&ei
->sync_writers
, 0);
9505 mutex_init(&ei
->log_mutex
);
9506 mutex_init(&ei
->delalloc_mutex
);
9507 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
9508 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
9509 INIT_LIST_HEAD(&ei
->delayed_iput
);
9510 RB_CLEAR_NODE(&ei
->rb_node
);
9511 init_rwsem(&ei
->dio_sem
);
9516 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9517 void btrfs_test_destroy_inode(struct inode
*inode
)
9519 btrfs_drop_extent_cache(BTRFS_I(inode
), 0, (u64
)-1, 0);
9520 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9524 static void btrfs_i_callback(struct rcu_head
*head
)
9526 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
9527 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9530 void btrfs_destroy_inode(struct inode
*inode
)
9532 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
9533 struct btrfs_ordered_extent
*ordered
;
9534 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9536 WARN_ON(!hlist_empty(&inode
->i_dentry
));
9537 WARN_ON(inode
->i_data
.nrpages
);
9538 WARN_ON(BTRFS_I(inode
)->block_rsv
.reserved
);
9539 WARN_ON(BTRFS_I(inode
)->block_rsv
.size
);
9540 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
9541 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
9542 WARN_ON(BTRFS_I(inode
)->new_delalloc_bytes
);
9543 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
9544 WARN_ON(BTRFS_I(inode
)->defrag_bytes
);
9547 * This can happen where we create an inode, but somebody else also
9548 * created the same inode and we need to destroy the one we already
9554 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
9555 &BTRFS_I(inode
)->runtime_flags
)) {
9556 btrfs_info(fs_info
, "inode %llu still on the orphan list",
9557 btrfs_ino(BTRFS_I(inode
)));
9558 atomic_dec(&root
->orphan_inodes
);
9562 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
9567 "found ordered extent %llu %llu on inode cleanup",
9568 ordered
->file_offset
, ordered
->len
);
9569 btrfs_remove_ordered_extent(inode
, ordered
);
9570 btrfs_put_ordered_extent(ordered
);
9571 btrfs_put_ordered_extent(ordered
);
9574 btrfs_qgroup_check_reserved_leak(inode
);
9575 inode_tree_del(inode
);
9576 btrfs_drop_extent_cache(BTRFS_I(inode
), 0, (u64
)-1, 0);
9578 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
9581 int btrfs_drop_inode(struct inode
*inode
)
9583 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9588 /* the snap/subvol tree is on deleting */
9589 if (btrfs_root_refs(&root
->root_item
) == 0)
9592 return generic_drop_inode(inode
);
9595 static void init_once(void *foo
)
9597 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
9599 inode_init_once(&ei
->vfs_inode
);
9602 void btrfs_destroy_cachep(void)
9605 * Make sure all delayed rcu free inodes are flushed before we
9609 kmem_cache_destroy(btrfs_inode_cachep
);
9610 kmem_cache_destroy(btrfs_trans_handle_cachep
);
9611 kmem_cache_destroy(btrfs_path_cachep
);
9612 kmem_cache_destroy(btrfs_free_space_cachep
);
9615 int btrfs_init_cachep(void)
9617 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
9618 sizeof(struct btrfs_inode
), 0,
9619 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
| SLAB_ACCOUNT
,
9621 if (!btrfs_inode_cachep
)
9624 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
9625 sizeof(struct btrfs_trans_handle
), 0,
9626 SLAB_TEMPORARY
| SLAB_MEM_SPREAD
, NULL
);
9627 if (!btrfs_trans_handle_cachep
)
9630 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
9631 sizeof(struct btrfs_path
), 0,
9632 SLAB_MEM_SPREAD
, NULL
);
9633 if (!btrfs_path_cachep
)
9636 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
9637 sizeof(struct btrfs_free_space
), 0,
9638 SLAB_MEM_SPREAD
, NULL
);
9639 if (!btrfs_free_space_cachep
)
9644 btrfs_destroy_cachep();
9648 static int btrfs_getattr(const struct path
*path
, struct kstat
*stat
,
9649 u32 request_mask
, unsigned int flags
)
9652 struct inode
*inode
= d_inode(path
->dentry
);
9653 u32 blocksize
= inode
->i_sb
->s_blocksize
;
9654 u32 bi_flags
= BTRFS_I(inode
)->flags
;
9656 stat
->result_mask
|= STATX_BTIME
;
9657 stat
->btime
.tv_sec
= BTRFS_I(inode
)->i_otime
.tv_sec
;
9658 stat
->btime
.tv_nsec
= BTRFS_I(inode
)->i_otime
.tv_nsec
;
9659 if (bi_flags
& BTRFS_INODE_APPEND
)
9660 stat
->attributes
|= STATX_ATTR_APPEND
;
9661 if (bi_flags
& BTRFS_INODE_COMPRESS
)
9662 stat
->attributes
|= STATX_ATTR_COMPRESSED
;
9663 if (bi_flags
& BTRFS_INODE_IMMUTABLE
)
9664 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
9665 if (bi_flags
& BTRFS_INODE_NODUMP
)
9666 stat
->attributes
|= STATX_ATTR_NODUMP
;
9668 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
9669 STATX_ATTR_COMPRESSED
|
9670 STATX_ATTR_IMMUTABLE
|
9673 generic_fillattr(inode
, stat
);
9674 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
9676 spin_lock(&BTRFS_I(inode
)->lock
);
9677 delalloc_bytes
= BTRFS_I(inode
)->new_delalloc_bytes
;
9678 spin_unlock(&BTRFS_I(inode
)->lock
);
9679 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
9680 ALIGN(delalloc_bytes
, blocksize
)) >> 9;
9684 static int btrfs_rename_exchange(struct inode
*old_dir
,
9685 struct dentry
*old_dentry
,
9686 struct inode
*new_dir
,
9687 struct dentry
*new_dentry
)
9689 struct btrfs_fs_info
*fs_info
= btrfs_sb(old_dir
->i_sb
);
9690 struct btrfs_trans_handle
*trans
;
9691 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9692 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9693 struct inode
*new_inode
= new_dentry
->d_inode
;
9694 struct inode
*old_inode
= old_dentry
->d_inode
;
9695 struct timespec ctime
= current_time(old_inode
);
9696 struct dentry
*parent
;
9697 u64 old_ino
= btrfs_ino(BTRFS_I(old_inode
));
9698 u64 new_ino
= btrfs_ino(BTRFS_I(new_inode
));
9703 bool root_log_pinned
= false;
9704 bool dest_log_pinned
= false;
9706 /* we only allow rename subvolume link between subvolumes */
9707 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9710 /* close the race window with snapshot create/destroy ioctl */
9711 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9712 down_read(&fs_info
->subvol_sem
);
9713 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9714 down_read(&fs_info
->subvol_sem
);
9717 * We want to reserve the absolute worst case amount of items. So if
9718 * both inodes are subvols and we need to unlink them then that would
9719 * require 4 item modifications, but if they are both normal inodes it
9720 * would require 5 item modifications, so we'll assume their normal
9721 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9722 * should cover the worst case number of items we'll modify.
9724 trans
= btrfs_start_transaction(root
, 12);
9725 if (IS_ERR(trans
)) {
9726 ret
= PTR_ERR(trans
);
9731 * We need to find a free sequence number both in the source and
9732 * in the destination directory for the exchange.
9734 ret
= btrfs_set_inode_index(BTRFS_I(new_dir
), &old_idx
);
9737 ret
= btrfs_set_inode_index(BTRFS_I(old_dir
), &new_idx
);
9741 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9742 BTRFS_I(new_inode
)->dir_index
= 0ULL;
9744 /* Reference for the source. */
9745 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9746 /* force full log commit if subvolume involved. */
9747 btrfs_set_log_full_commit(fs_info
, trans
);
9749 btrfs_pin_log_trans(root
);
9750 root_log_pinned
= true;
9751 ret
= btrfs_insert_inode_ref(trans
, dest
,
9752 new_dentry
->d_name
.name
,
9753 new_dentry
->d_name
.len
,
9755 btrfs_ino(BTRFS_I(new_dir
)),
9761 /* And now for the dest. */
9762 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9763 /* force full log commit if subvolume involved. */
9764 btrfs_set_log_full_commit(fs_info
, trans
);
9766 btrfs_pin_log_trans(dest
);
9767 dest_log_pinned
= true;
9768 ret
= btrfs_insert_inode_ref(trans
, root
,
9769 old_dentry
->d_name
.name
,
9770 old_dentry
->d_name
.len
,
9772 btrfs_ino(BTRFS_I(old_dir
)),
9778 /* Update inode version and ctime/mtime. */
9779 inode_inc_iversion(old_dir
);
9780 inode_inc_iversion(new_dir
);
9781 inode_inc_iversion(old_inode
);
9782 inode_inc_iversion(new_inode
);
9783 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
9784 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
9785 old_inode
->i_ctime
= ctime
;
9786 new_inode
->i_ctime
= ctime
;
9788 if (old_dentry
->d_parent
!= new_dentry
->d_parent
) {
9789 btrfs_record_unlink_dir(trans
, BTRFS_I(old_dir
),
9790 BTRFS_I(old_inode
), 1);
9791 btrfs_record_unlink_dir(trans
, BTRFS_I(new_dir
),
9792 BTRFS_I(new_inode
), 1);
9795 /* src is a subvolume */
9796 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9797 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
9798 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
,
9800 old_dentry
->d_name
.name
,
9801 old_dentry
->d_name
.len
);
9802 } else { /* src is an inode */
9803 ret
= __btrfs_unlink_inode(trans
, root
, BTRFS_I(old_dir
),
9804 BTRFS_I(old_dentry
->d_inode
),
9805 old_dentry
->d_name
.name
,
9806 old_dentry
->d_name
.len
);
9808 ret
= btrfs_update_inode(trans
, root
, old_inode
);
9811 btrfs_abort_transaction(trans
, ret
);
9815 /* dest is a subvolume */
9816 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9817 root_objectid
= BTRFS_I(new_inode
)->root
->root_key
.objectid
;
9818 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
9820 new_dentry
->d_name
.name
,
9821 new_dentry
->d_name
.len
);
9822 } else { /* dest is an inode */
9823 ret
= __btrfs_unlink_inode(trans
, dest
, BTRFS_I(new_dir
),
9824 BTRFS_I(new_dentry
->d_inode
),
9825 new_dentry
->d_name
.name
,
9826 new_dentry
->d_name
.len
);
9828 ret
= btrfs_update_inode(trans
, dest
, new_inode
);
9831 btrfs_abort_transaction(trans
, ret
);
9835 ret
= btrfs_add_link(trans
, BTRFS_I(new_dir
), BTRFS_I(old_inode
),
9836 new_dentry
->d_name
.name
,
9837 new_dentry
->d_name
.len
, 0, old_idx
);
9839 btrfs_abort_transaction(trans
, ret
);
9843 ret
= btrfs_add_link(trans
, BTRFS_I(old_dir
), BTRFS_I(new_inode
),
9844 old_dentry
->d_name
.name
,
9845 old_dentry
->d_name
.len
, 0, new_idx
);
9847 btrfs_abort_transaction(trans
, ret
);
9851 if (old_inode
->i_nlink
== 1)
9852 BTRFS_I(old_inode
)->dir_index
= old_idx
;
9853 if (new_inode
->i_nlink
== 1)
9854 BTRFS_I(new_inode
)->dir_index
= new_idx
;
9856 if (root_log_pinned
) {
9857 parent
= new_dentry
->d_parent
;
9858 btrfs_log_new_name(trans
, BTRFS_I(old_inode
), BTRFS_I(old_dir
),
9860 btrfs_end_log_trans(root
);
9861 root_log_pinned
= false;
9863 if (dest_log_pinned
) {
9864 parent
= old_dentry
->d_parent
;
9865 btrfs_log_new_name(trans
, BTRFS_I(new_inode
), BTRFS_I(new_dir
),
9867 btrfs_end_log_trans(dest
);
9868 dest_log_pinned
= false;
9872 * If we have pinned a log and an error happened, we unpin tasks
9873 * trying to sync the log and force them to fallback to a transaction
9874 * commit if the log currently contains any of the inodes involved in
9875 * this rename operation (to ensure we do not persist a log with an
9876 * inconsistent state for any of these inodes or leading to any
9877 * inconsistencies when replayed). If the transaction was aborted, the
9878 * abortion reason is propagated to userspace when attempting to commit
9879 * the transaction. If the log does not contain any of these inodes, we
9880 * allow the tasks to sync it.
9882 if (ret
&& (root_log_pinned
|| dest_log_pinned
)) {
9883 if (btrfs_inode_in_log(BTRFS_I(old_dir
), fs_info
->generation
) ||
9884 btrfs_inode_in_log(BTRFS_I(new_dir
), fs_info
->generation
) ||
9885 btrfs_inode_in_log(BTRFS_I(old_inode
), fs_info
->generation
) ||
9887 btrfs_inode_in_log(BTRFS_I(new_inode
), fs_info
->generation
)))
9888 btrfs_set_log_full_commit(fs_info
, trans
);
9890 if (root_log_pinned
) {
9891 btrfs_end_log_trans(root
);
9892 root_log_pinned
= false;
9894 if (dest_log_pinned
) {
9895 btrfs_end_log_trans(dest
);
9896 dest_log_pinned
= false;
9899 ret
= btrfs_end_transaction(trans
);
9901 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9902 up_read(&fs_info
->subvol_sem
);
9903 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9904 up_read(&fs_info
->subvol_sem
);
9909 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle
*trans
,
9910 struct btrfs_root
*root
,
9912 struct dentry
*dentry
)
9915 struct inode
*inode
;
9919 ret
= btrfs_find_free_ino(root
, &objectid
);
9923 inode
= btrfs_new_inode(trans
, root
, dir
,
9924 dentry
->d_name
.name
,
9926 btrfs_ino(BTRFS_I(dir
)),
9928 S_IFCHR
| WHITEOUT_MODE
,
9931 if (IS_ERR(inode
)) {
9932 ret
= PTR_ERR(inode
);
9936 inode
->i_op
= &btrfs_special_inode_operations
;
9937 init_special_inode(inode
, inode
->i_mode
,
9940 ret
= btrfs_init_inode_security(trans
, inode
, dir
,
9945 ret
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
,
9946 BTRFS_I(inode
), 0, index
);
9950 ret
= btrfs_update_inode(trans
, root
, inode
);
9952 unlock_new_inode(inode
);
9954 inode_dec_link_count(inode
);
9960 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
9961 struct inode
*new_dir
, struct dentry
*new_dentry
,
9964 struct btrfs_fs_info
*fs_info
= btrfs_sb(old_dir
->i_sb
);
9965 struct btrfs_trans_handle
*trans
;
9966 unsigned int trans_num_items
;
9967 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9968 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9969 struct inode
*new_inode
= d_inode(new_dentry
);
9970 struct inode
*old_inode
= d_inode(old_dentry
);
9974 u64 old_ino
= btrfs_ino(BTRFS_I(old_inode
));
9975 bool log_pinned
= false;
9977 if (btrfs_ino(BTRFS_I(new_dir
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
9980 /* we only allow rename subvolume link between subvolumes */
9981 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9984 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
9985 (new_inode
&& btrfs_ino(BTRFS_I(new_inode
)) == BTRFS_FIRST_FREE_OBJECTID
))
9988 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
9989 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
9993 /* check for collisions, even if the name isn't there */
9994 ret
= btrfs_check_dir_item_collision(dest
, new_dir
->i_ino
,
9995 new_dentry
->d_name
.name
,
9996 new_dentry
->d_name
.len
);
9999 if (ret
== -EEXIST
) {
10000 /* we shouldn't get
10001 * eexist without a new_inode */
10002 if (WARN_ON(!new_inode
)) {
10006 /* maybe -EOVERFLOW */
10013 * we're using rename to replace one file with another. Start IO on it
10014 * now so we don't add too much work to the end of the transaction
10016 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
)
10017 filemap_flush(old_inode
->i_mapping
);
10019 /* close the racy window with snapshot create/destroy ioctl */
10020 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
10021 down_read(&fs_info
->subvol_sem
);
10023 * We want to reserve the absolute worst case amount of items. So if
10024 * both inodes are subvols and we need to unlink them then that would
10025 * require 4 item modifications, but if they are both normal inodes it
10026 * would require 5 item modifications, so we'll assume they are normal
10027 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
10028 * should cover the worst case number of items we'll modify.
10029 * If our rename has the whiteout flag, we need more 5 units for the
10030 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
10031 * when selinux is enabled).
10033 trans_num_items
= 11;
10034 if (flags
& RENAME_WHITEOUT
)
10035 trans_num_items
+= 5;
10036 trans
= btrfs_start_transaction(root
, trans_num_items
);
10037 if (IS_ERR(trans
)) {
10038 ret
= PTR_ERR(trans
);
10043 btrfs_record_root_in_trans(trans
, dest
);
10045 ret
= btrfs_set_inode_index(BTRFS_I(new_dir
), &index
);
10049 BTRFS_I(old_inode
)->dir_index
= 0ULL;
10050 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
10051 /* force full log commit if subvolume involved. */
10052 btrfs_set_log_full_commit(fs_info
, trans
);
10054 btrfs_pin_log_trans(root
);
10056 ret
= btrfs_insert_inode_ref(trans
, dest
,
10057 new_dentry
->d_name
.name
,
10058 new_dentry
->d_name
.len
,
10060 btrfs_ino(BTRFS_I(new_dir
)), index
);
10065 inode_inc_iversion(old_dir
);
10066 inode_inc_iversion(new_dir
);
10067 inode_inc_iversion(old_inode
);
10068 old_dir
->i_ctime
= old_dir
->i_mtime
=
10069 new_dir
->i_ctime
= new_dir
->i_mtime
=
10070 old_inode
->i_ctime
= current_time(old_dir
);
10072 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
10073 btrfs_record_unlink_dir(trans
, BTRFS_I(old_dir
),
10074 BTRFS_I(old_inode
), 1);
10076 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
10077 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
10078 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
10079 old_dentry
->d_name
.name
,
10080 old_dentry
->d_name
.len
);
10082 ret
= __btrfs_unlink_inode(trans
, root
, BTRFS_I(old_dir
),
10083 BTRFS_I(d_inode(old_dentry
)),
10084 old_dentry
->d_name
.name
,
10085 old_dentry
->d_name
.len
);
10087 ret
= btrfs_update_inode(trans
, root
, old_inode
);
10090 btrfs_abort_transaction(trans
, ret
);
10095 inode_inc_iversion(new_inode
);
10096 new_inode
->i_ctime
= current_time(new_inode
);
10097 if (unlikely(btrfs_ino(BTRFS_I(new_inode
)) ==
10098 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
10099 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
10100 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
10102 new_dentry
->d_name
.name
,
10103 new_dentry
->d_name
.len
);
10104 BUG_ON(new_inode
->i_nlink
== 0);
10106 ret
= btrfs_unlink_inode(trans
, dest
, BTRFS_I(new_dir
),
10107 BTRFS_I(d_inode(new_dentry
)),
10108 new_dentry
->d_name
.name
,
10109 new_dentry
->d_name
.len
);
10111 if (!ret
&& new_inode
->i_nlink
== 0)
10112 ret
= btrfs_orphan_add(trans
,
10113 BTRFS_I(d_inode(new_dentry
)));
10115 btrfs_abort_transaction(trans
, ret
);
10120 ret
= btrfs_add_link(trans
, BTRFS_I(new_dir
), BTRFS_I(old_inode
),
10121 new_dentry
->d_name
.name
,
10122 new_dentry
->d_name
.len
, 0, index
);
10124 btrfs_abort_transaction(trans
, ret
);
10128 if (old_inode
->i_nlink
== 1)
10129 BTRFS_I(old_inode
)->dir_index
= index
;
10132 struct dentry
*parent
= new_dentry
->d_parent
;
10134 btrfs_log_new_name(trans
, BTRFS_I(old_inode
), BTRFS_I(old_dir
),
10136 btrfs_end_log_trans(root
);
10137 log_pinned
= false;
10140 if (flags
& RENAME_WHITEOUT
) {
10141 ret
= btrfs_whiteout_for_rename(trans
, root
, old_dir
,
10145 btrfs_abort_transaction(trans
, ret
);
10151 * If we have pinned the log and an error happened, we unpin tasks
10152 * trying to sync the log and force them to fallback to a transaction
10153 * commit if the log currently contains any of the inodes involved in
10154 * this rename operation (to ensure we do not persist a log with an
10155 * inconsistent state for any of these inodes or leading to any
10156 * inconsistencies when replayed). If the transaction was aborted, the
10157 * abortion reason is propagated to userspace when attempting to commit
10158 * the transaction. If the log does not contain any of these inodes, we
10159 * allow the tasks to sync it.
10161 if (ret
&& log_pinned
) {
10162 if (btrfs_inode_in_log(BTRFS_I(old_dir
), fs_info
->generation
) ||
10163 btrfs_inode_in_log(BTRFS_I(new_dir
), fs_info
->generation
) ||
10164 btrfs_inode_in_log(BTRFS_I(old_inode
), fs_info
->generation
) ||
10166 btrfs_inode_in_log(BTRFS_I(new_inode
), fs_info
->generation
)))
10167 btrfs_set_log_full_commit(fs_info
, trans
);
10169 btrfs_end_log_trans(root
);
10170 log_pinned
= false;
10172 btrfs_end_transaction(trans
);
10174 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
10175 up_read(&fs_info
->subvol_sem
);
10180 static int btrfs_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
10181 struct inode
*new_dir
, struct dentry
*new_dentry
,
10182 unsigned int flags
)
10184 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
10187 if (flags
& RENAME_EXCHANGE
)
10188 return btrfs_rename_exchange(old_dir
, old_dentry
, new_dir
,
10191 return btrfs_rename(old_dir
, old_dentry
, new_dir
, new_dentry
, flags
);
10194 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
10196 struct btrfs_delalloc_work
*delalloc_work
;
10197 struct inode
*inode
;
10199 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
10201 inode
= delalloc_work
->inode
;
10202 filemap_flush(inode
->i_mapping
);
10203 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
10204 &BTRFS_I(inode
)->runtime_flags
))
10205 filemap_flush(inode
->i_mapping
);
10207 if (delalloc_work
->delay_iput
)
10208 btrfs_add_delayed_iput(inode
);
10211 complete(&delalloc_work
->completion
);
10214 struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
,
10217 struct btrfs_delalloc_work
*work
;
10219 work
= kmalloc(sizeof(*work
), GFP_NOFS
);
10223 init_completion(&work
->completion
);
10224 INIT_LIST_HEAD(&work
->list
);
10225 work
->inode
= inode
;
10226 work
->delay_iput
= delay_iput
;
10227 WARN_ON_ONCE(!inode
);
10228 btrfs_init_work(&work
->work
, btrfs_flush_delalloc_helper
,
10229 btrfs_run_delalloc_work
, NULL
, NULL
);
10234 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work
*work
)
10236 wait_for_completion(&work
->completion
);
10241 * some fairly slow code that needs optimization. This walks the list
10242 * of all the inodes with pending delalloc and forces them to disk.
10244 static int __start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
,
10247 struct btrfs_inode
*binode
;
10248 struct inode
*inode
;
10249 struct btrfs_delalloc_work
*work
, *next
;
10250 struct list_head works
;
10251 struct list_head splice
;
10254 INIT_LIST_HEAD(&works
);
10255 INIT_LIST_HEAD(&splice
);
10257 mutex_lock(&root
->delalloc_mutex
);
10258 spin_lock(&root
->delalloc_lock
);
10259 list_splice_init(&root
->delalloc_inodes
, &splice
);
10260 while (!list_empty(&splice
)) {
10261 binode
= list_entry(splice
.next
, struct btrfs_inode
,
10264 list_move_tail(&binode
->delalloc_inodes
,
10265 &root
->delalloc_inodes
);
10266 inode
= igrab(&binode
->vfs_inode
);
10268 cond_resched_lock(&root
->delalloc_lock
);
10271 spin_unlock(&root
->delalloc_lock
);
10273 work
= btrfs_alloc_delalloc_work(inode
, delay_iput
);
10276 btrfs_add_delayed_iput(inode
);
10282 list_add_tail(&work
->list
, &works
);
10283 btrfs_queue_work(root
->fs_info
->flush_workers
,
10286 if (nr
!= -1 && ret
>= nr
)
10289 spin_lock(&root
->delalloc_lock
);
10291 spin_unlock(&root
->delalloc_lock
);
10294 list_for_each_entry_safe(work
, next
, &works
, list
) {
10295 list_del_init(&work
->list
);
10296 btrfs_wait_and_free_delalloc_work(work
);
10299 if (!list_empty_careful(&splice
)) {
10300 spin_lock(&root
->delalloc_lock
);
10301 list_splice_tail(&splice
, &root
->delalloc_inodes
);
10302 spin_unlock(&root
->delalloc_lock
);
10304 mutex_unlock(&root
->delalloc_mutex
);
10308 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
10310 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
10313 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
10316 ret
= __start_delalloc_inodes(root
, delay_iput
, -1);
10322 int btrfs_start_delalloc_roots(struct btrfs_fs_info
*fs_info
, int delay_iput
,
10325 struct btrfs_root
*root
;
10326 struct list_head splice
;
10329 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
10332 INIT_LIST_HEAD(&splice
);
10334 mutex_lock(&fs_info
->delalloc_root_mutex
);
10335 spin_lock(&fs_info
->delalloc_root_lock
);
10336 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
10337 while (!list_empty(&splice
) && nr
) {
10338 root
= list_first_entry(&splice
, struct btrfs_root
,
10340 root
= btrfs_grab_fs_root(root
);
10342 list_move_tail(&root
->delalloc_root
,
10343 &fs_info
->delalloc_roots
);
10344 spin_unlock(&fs_info
->delalloc_root_lock
);
10346 ret
= __start_delalloc_inodes(root
, delay_iput
, nr
);
10347 btrfs_put_fs_root(root
);
10355 spin_lock(&fs_info
->delalloc_root_lock
);
10357 spin_unlock(&fs_info
->delalloc_root_lock
);
10361 if (!list_empty_careful(&splice
)) {
10362 spin_lock(&fs_info
->delalloc_root_lock
);
10363 list_splice_tail(&splice
, &fs_info
->delalloc_roots
);
10364 spin_unlock(&fs_info
->delalloc_root_lock
);
10366 mutex_unlock(&fs_info
->delalloc_root_mutex
);
10370 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
10371 const char *symname
)
10373 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
10374 struct btrfs_trans_handle
*trans
;
10375 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
10376 struct btrfs_path
*path
;
10377 struct btrfs_key key
;
10378 struct inode
*inode
= NULL
;
10380 int drop_inode
= 0;
10386 struct btrfs_file_extent_item
*ei
;
10387 struct extent_buffer
*leaf
;
10389 name_len
= strlen(symname
);
10390 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(fs_info
))
10391 return -ENAMETOOLONG
;
10394 * 2 items for inode item and ref
10395 * 2 items for dir items
10396 * 1 item for updating parent inode item
10397 * 1 item for the inline extent item
10398 * 1 item for xattr if selinux is on
10400 trans
= btrfs_start_transaction(root
, 7);
10402 return PTR_ERR(trans
);
10404 err
= btrfs_find_free_ino(root
, &objectid
);
10408 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
10409 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)),
10410 objectid
, S_IFLNK
|S_IRWXUGO
, &index
);
10411 if (IS_ERR(inode
)) {
10412 err
= PTR_ERR(inode
);
10417 * If the active LSM wants to access the inode during
10418 * d_instantiate it needs these. Smack checks to see
10419 * if the filesystem supports xattrs by looking at the
10422 inode
->i_fop
= &btrfs_file_operations
;
10423 inode
->i_op
= &btrfs_file_inode_operations
;
10424 inode
->i_mapping
->a_ops
= &btrfs_aops
;
10425 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
10427 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
10429 goto out_unlock_inode
;
10431 path
= btrfs_alloc_path();
10434 goto out_unlock_inode
;
10436 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
10438 key
.type
= BTRFS_EXTENT_DATA_KEY
;
10439 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
10440 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
10443 btrfs_free_path(path
);
10444 goto out_unlock_inode
;
10446 leaf
= path
->nodes
[0];
10447 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
10448 struct btrfs_file_extent_item
);
10449 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
10450 btrfs_set_file_extent_type(leaf
, ei
,
10451 BTRFS_FILE_EXTENT_INLINE
);
10452 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
10453 btrfs_set_file_extent_compression(leaf
, ei
, 0);
10454 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
10455 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
10457 ptr
= btrfs_file_extent_inline_start(ei
);
10458 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
10459 btrfs_mark_buffer_dirty(leaf
);
10460 btrfs_free_path(path
);
10462 inode
->i_op
= &btrfs_symlink_inode_operations
;
10463 inode_nohighmem(inode
);
10464 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
10465 inode_set_bytes(inode
, name_len
);
10466 btrfs_i_size_write(BTRFS_I(inode
), name_len
);
10467 err
= btrfs_update_inode(trans
, root
, inode
);
10469 * Last step, add directory indexes for our symlink inode. This is the
10470 * last step to avoid extra cleanup of these indexes if an error happens
10474 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
,
10475 BTRFS_I(inode
), 0, index
);
10478 goto out_unlock_inode
;
10481 d_instantiate_new(dentry
, inode
);
10484 btrfs_end_transaction(trans
);
10486 inode_dec_link_count(inode
);
10489 btrfs_btree_balance_dirty(fs_info
);
10494 unlock_new_inode(inode
);
10498 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
10499 u64 start
, u64 num_bytes
, u64 min_size
,
10500 loff_t actual_len
, u64
*alloc_hint
,
10501 struct btrfs_trans_handle
*trans
)
10503 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
10504 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
10505 struct extent_map
*em
;
10506 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
10507 struct btrfs_key ins
;
10508 u64 cur_offset
= start
;
10511 u64 last_alloc
= (u64
)-1;
10513 bool own_trans
= true;
10514 u64 end
= start
+ num_bytes
- 1;
10518 while (num_bytes
> 0) {
10520 trans
= btrfs_start_transaction(root
, 3);
10521 if (IS_ERR(trans
)) {
10522 ret
= PTR_ERR(trans
);
10527 cur_bytes
= min_t(u64
, num_bytes
, SZ_256M
);
10528 cur_bytes
= max(cur_bytes
, min_size
);
10530 * If we are severely fragmented we could end up with really
10531 * small allocations, so if the allocator is returning small
10532 * chunks lets make its job easier by only searching for those
10535 cur_bytes
= min(cur_bytes
, last_alloc
);
10536 ret
= btrfs_reserve_extent(root
, cur_bytes
, cur_bytes
,
10537 min_size
, 0, *alloc_hint
, &ins
, 1, 0);
10540 btrfs_end_transaction(trans
);
10543 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
10545 last_alloc
= ins
.offset
;
10546 ret
= insert_reserved_file_extent(trans
, inode
,
10547 cur_offset
, ins
.objectid
,
10548 ins
.offset
, ins
.offset
,
10549 ins
.offset
, 0, 0, 0,
10550 BTRFS_FILE_EXTENT_PREALLOC
);
10552 btrfs_free_reserved_extent(fs_info
, ins
.objectid
,
10554 btrfs_abort_transaction(trans
, ret
);
10556 btrfs_end_transaction(trans
);
10560 btrfs_drop_extent_cache(BTRFS_I(inode
), cur_offset
,
10561 cur_offset
+ ins
.offset
-1, 0);
10563 em
= alloc_extent_map();
10565 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
10566 &BTRFS_I(inode
)->runtime_flags
);
10570 em
->start
= cur_offset
;
10571 em
->orig_start
= cur_offset
;
10572 em
->len
= ins
.offset
;
10573 em
->block_start
= ins
.objectid
;
10574 em
->block_len
= ins
.offset
;
10575 em
->orig_block_len
= ins
.offset
;
10576 em
->ram_bytes
= ins
.offset
;
10577 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
10578 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
10579 em
->generation
= trans
->transid
;
10582 write_lock(&em_tree
->lock
);
10583 ret
= add_extent_mapping(em_tree
, em
, 1);
10584 write_unlock(&em_tree
->lock
);
10585 if (ret
!= -EEXIST
)
10587 btrfs_drop_extent_cache(BTRFS_I(inode
), cur_offset
,
10588 cur_offset
+ ins
.offset
- 1,
10591 free_extent_map(em
);
10593 num_bytes
-= ins
.offset
;
10594 cur_offset
+= ins
.offset
;
10595 *alloc_hint
= ins
.objectid
+ ins
.offset
;
10597 inode_inc_iversion(inode
);
10598 inode
->i_ctime
= current_time(inode
);
10599 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
10600 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
10601 (actual_len
> inode
->i_size
) &&
10602 (cur_offset
> inode
->i_size
)) {
10603 if (cur_offset
> actual_len
)
10604 i_size
= actual_len
;
10606 i_size
= cur_offset
;
10607 i_size_write(inode
, i_size
);
10608 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
10611 ret
= btrfs_update_inode(trans
, root
, inode
);
10614 btrfs_abort_transaction(trans
, ret
);
10616 btrfs_end_transaction(trans
);
10621 btrfs_end_transaction(trans
);
10623 if (cur_offset
< end
)
10624 btrfs_free_reserved_data_space(inode
, NULL
, cur_offset
,
10625 end
- cur_offset
+ 1);
10629 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
10630 u64 start
, u64 num_bytes
, u64 min_size
,
10631 loff_t actual_len
, u64
*alloc_hint
)
10633 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
10634 min_size
, actual_len
, alloc_hint
,
10638 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
10639 struct btrfs_trans_handle
*trans
, int mode
,
10640 u64 start
, u64 num_bytes
, u64 min_size
,
10641 loff_t actual_len
, u64
*alloc_hint
)
10643 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
10644 min_size
, actual_len
, alloc_hint
, trans
);
10647 static int btrfs_set_page_dirty(struct page
*page
)
10649 return __set_page_dirty_nobuffers(page
);
10652 static int btrfs_permission(struct inode
*inode
, int mask
)
10654 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
10655 umode_t mode
= inode
->i_mode
;
10657 if (mask
& MAY_WRITE
&&
10658 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
10659 if (btrfs_root_readonly(root
))
10661 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
10664 return generic_permission(inode
, mask
);
10667 static int btrfs_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
10669 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
10670 struct btrfs_trans_handle
*trans
;
10671 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
10672 struct inode
*inode
= NULL
;
10678 * 5 units required for adding orphan entry
10680 trans
= btrfs_start_transaction(root
, 5);
10682 return PTR_ERR(trans
);
10684 ret
= btrfs_find_free_ino(root
, &objectid
);
10688 inode
= btrfs_new_inode(trans
, root
, dir
, NULL
, 0,
10689 btrfs_ino(BTRFS_I(dir
)), objectid
, mode
, &index
);
10690 if (IS_ERR(inode
)) {
10691 ret
= PTR_ERR(inode
);
10696 inode
->i_fop
= &btrfs_file_operations
;
10697 inode
->i_op
= &btrfs_file_inode_operations
;
10699 inode
->i_mapping
->a_ops
= &btrfs_aops
;
10700 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
10702 ret
= btrfs_init_inode_security(trans
, inode
, dir
, NULL
);
10706 ret
= btrfs_update_inode(trans
, root
, inode
);
10709 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
10714 * We set number of links to 0 in btrfs_new_inode(), and here we set
10715 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10718 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10720 set_nlink(inode
, 1);
10721 unlock_new_inode(inode
);
10722 d_tmpfile(dentry
, inode
);
10723 mark_inode_dirty(inode
);
10726 btrfs_end_transaction(trans
);
10729 btrfs_balance_delayed_items(fs_info
);
10730 btrfs_btree_balance_dirty(fs_info
);
10734 unlock_new_inode(inode
);
10739 __attribute__((const))
10740 static int btrfs_readpage_io_failed_hook(struct page
*page
, int failed_mirror
)
10745 static struct btrfs_fs_info
*iotree_fs_info(void *private_data
)
10747 struct inode
*inode
= private_data
;
10748 return btrfs_sb(inode
->i_sb
);
10751 static void btrfs_check_extent_io_range(void *private_data
, const char *caller
,
10752 u64 start
, u64 end
)
10754 struct inode
*inode
= private_data
;
10757 isize
= i_size_read(inode
);
10758 if (end
>= PAGE_SIZE
&& (end
% 2) == 0 && end
!= isize
- 1) {
10759 btrfs_debug_rl(BTRFS_I(inode
)->root
->fs_info
,
10760 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10761 caller
, btrfs_ino(BTRFS_I(inode
)), isize
, start
, end
);
10765 void btrfs_set_range_writeback(void *private_data
, u64 start
, u64 end
)
10767 struct inode
*inode
= private_data
;
10768 unsigned long index
= start
>> PAGE_SHIFT
;
10769 unsigned long end_index
= end
>> PAGE_SHIFT
;
10772 while (index
<= end_index
) {
10773 page
= find_get_page(inode
->i_mapping
, index
);
10774 ASSERT(page
); /* Pages should be in the extent_io_tree */
10775 set_page_writeback(page
);
10781 static const struct inode_operations btrfs_dir_inode_operations
= {
10782 .getattr
= btrfs_getattr
,
10783 .lookup
= btrfs_lookup
,
10784 .create
= btrfs_create
,
10785 .unlink
= btrfs_unlink
,
10786 .link
= btrfs_link
,
10787 .mkdir
= btrfs_mkdir
,
10788 .rmdir
= btrfs_rmdir
,
10789 .rename
= btrfs_rename2
,
10790 .symlink
= btrfs_symlink
,
10791 .setattr
= btrfs_setattr
,
10792 .mknod
= btrfs_mknod
,
10793 .listxattr
= btrfs_listxattr
,
10794 .permission
= btrfs_permission
,
10795 .get_acl
= btrfs_get_acl
,
10796 .set_acl
= btrfs_set_acl
,
10797 .update_time
= btrfs_update_time
,
10798 .tmpfile
= btrfs_tmpfile
,
10800 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
10801 .lookup
= btrfs_lookup
,
10802 .permission
= btrfs_permission
,
10803 .update_time
= btrfs_update_time
,
10806 static const struct file_operations btrfs_dir_file_operations
= {
10807 .llseek
= generic_file_llseek
,
10808 .read
= generic_read_dir
,
10809 .iterate_shared
= btrfs_real_readdir
,
10810 .open
= btrfs_opendir
,
10811 .unlocked_ioctl
= btrfs_ioctl
,
10812 #ifdef CONFIG_COMPAT
10813 .compat_ioctl
= btrfs_compat_ioctl
,
10815 .release
= btrfs_release_file
,
10816 .fsync
= btrfs_sync_file
,
10819 static const struct extent_io_ops btrfs_extent_io_ops
= {
10820 /* mandatory callbacks */
10821 .submit_bio_hook
= btrfs_submit_bio_hook
,
10822 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
10823 .merge_bio_hook
= btrfs_merge_bio_hook
,
10824 .readpage_io_failed_hook
= btrfs_readpage_io_failed_hook
,
10825 .tree_fs_info
= iotree_fs_info
,
10826 .set_range_writeback
= btrfs_set_range_writeback
,
10828 /* optional callbacks */
10829 .fill_delalloc
= run_delalloc_range
,
10830 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
10831 .writepage_start_hook
= btrfs_writepage_start_hook
,
10832 .set_bit_hook
= btrfs_set_bit_hook
,
10833 .clear_bit_hook
= btrfs_clear_bit_hook
,
10834 .merge_extent_hook
= btrfs_merge_extent_hook
,
10835 .split_extent_hook
= btrfs_split_extent_hook
,
10836 .check_extent_io_range
= btrfs_check_extent_io_range
,
10840 * btrfs doesn't support the bmap operation because swapfiles
10841 * use bmap to make a mapping of extents in the file. They assume
10842 * these extents won't change over the life of the file and they
10843 * use the bmap result to do IO directly to the drive.
10845 * the btrfs bmap call would return logical addresses that aren't
10846 * suitable for IO and they also will change frequently as COW
10847 * operations happen. So, swapfile + btrfs == corruption.
10849 * For now we're avoiding this by dropping bmap.
10851 static const struct address_space_operations btrfs_aops
= {
10852 .readpage
= btrfs_readpage
,
10853 .writepage
= btrfs_writepage
,
10854 .writepages
= btrfs_writepages
,
10855 .readpages
= btrfs_readpages
,
10856 .direct_IO
= btrfs_direct_IO
,
10857 .invalidatepage
= btrfs_invalidatepage
,
10858 .releasepage
= btrfs_releasepage
,
10859 .set_page_dirty
= btrfs_set_page_dirty
,
10860 .error_remove_page
= generic_error_remove_page
,
10863 static const struct address_space_operations btrfs_symlink_aops
= {
10864 .readpage
= btrfs_readpage
,
10865 .writepage
= btrfs_writepage
,
10866 .invalidatepage
= btrfs_invalidatepage
,
10867 .releasepage
= btrfs_releasepage
,
10870 static const struct inode_operations btrfs_file_inode_operations
= {
10871 .getattr
= btrfs_getattr
,
10872 .setattr
= btrfs_setattr
,
10873 .listxattr
= btrfs_listxattr
,
10874 .permission
= btrfs_permission
,
10875 .fiemap
= btrfs_fiemap
,
10876 .get_acl
= btrfs_get_acl
,
10877 .set_acl
= btrfs_set_acl
,
10878 .update_time
= btrfs_update_time
,
10880 static const struct inode_operations btrfs_special_inode_operations
= {
10881 .getattr
= btrfs_getattr
,
10882 .setattr
= btrfs_setattr
,
10883 .permission
= btrfs_permission
,
10884 .listxattr
= btrfs_listxattr
,
10885 .get_acl
= btrfs_get_acl
,
10886 .set_acl
= btrfs_set_acl
,
10887 .update_time
= btrfs_update_time
,
10889 static const struct inode_operations btrfs_symlink_inode_operations
= {
10890 .get_link
= page_get_link
,
10891 .getattr
= btrfs_getattr
,
10892 .setattr
= btrfs_setattr
,
10893 .permission
= btrfs_permission
,
10894 .listxattr
= btrfs_listxattr
,
10895 .update_time
= btrfs_update_time
,
10898 const struct dentry_operations btrfs_dentry_operations
= {
10899 .d_delete
= btrfs_dentry_delete
,
10900 .d_release
= btrfs_dentry_release
,