1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/kernel.h>
8 #include <linux/buffer_head.h>
9 #include <linux/file.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/writeback.h>
18 #include <linux/compat.h>
19 #include <linux/xattr.h>
20 #include <linux/posix_acl.h>
21 #include <linux/falloc.h>
22 #include <linux/slab.h>
23 #include <linux/ratelimit.h>
24 #include <linux/btrfs.h>
25 #include <linux/blkdev.h>
26 #include <linux/posix_acl_xattr.h>
27 #include <linux/uio.h>
28 #include <linux/magic.h>
29 #include <linux/iversion.h>
30 #include <linux/swap.h>
31 #include <asm/unaligned.h>
34 #include "transaction.h"
35 #include "btrfs_inode.h"
36 #include "print-tree.h"
37 #include "ordered-data.h"
41 #include "compression.h"
43 #include "free-space-cache.h"
44 #include "inode-map.h"
50 struct btrfs_iget_args
{
51 struct btrfs_key
*location
;
52 struct btrfs_root
*root
;
55 struct btrfs_dio_data
{
57 u64 unsubmitted_oe_range_start
;
58 u64 unsubmitted_oe_range_end
;
62 static const struct inode_operations btrfs_dir_inode_operations
;
63 static const struct inode_operations btrfs_symlink_inode_operations
;
64 static const struct inode_operations btrfs_dir_ro_inode_operations
;
65 static const struct inode_operations btrfs_special_inode_operations
;
66 static const struct inode_operations btrfs_file_inode_operations
;
67 static const struct address_space_operations btrfs_aops
;
68 static const struct file_operations btrfs_dir_file_operations
;
69 static const struct extent_io_ops btrfs_extent_io_ops
;
71 static struct kmem_cache
*btrfs_inode_cachep
;
72 struct kmem_cache
*btrfs_trans_handle_cachep
;
73 struct kmem_cache
*btrfs_path_cachep
;
74 struct kmem_cache
*btrfs_free_space_cachep
;
77 static const unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
78 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
79 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
80 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
81 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
82 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
83 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
84 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
87 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
);
88 static int btrfs_truncate(struct inode
*inode
, bool skip_writeback
);
89 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
);
90 static noinline
int cow_file_range(struct inode
*inode
,
91 struct page
*locked_page
,
92 u64 start
, u64 end
, u64 delalloc_end
,
93 int *page_started
, unsigned long *nr_written
,
94 int unlock
, struct btrfs_dedupe_hash
*hash
);
95 static struct extent_map
*create_io_em(struct inode
*inode
, u64 start
, u64 len
,
96 u64 orig_start
, u64 block_start
,
97 u64 block_len
, u64 orig_block_len
,
98 u64 ram_bytes
, int compress_type
,
101 static void __endio_write_update_ordered(struct inode
*inode
,
102 const u64 offset
, const u64 bytes
,
103 const bool uptodate
);
106 * Cleanup all submitted ordered extents in specified range to handle errors
107 * from the btrfs_run_delalloc_range() callback.
109 * NOTE: caller must ensure that when an error happens, it can not call
110 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
111 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
112 * to be released, which we want to happen only when finishing the ordered
113 * extent (btrfs_finish_ordered_io()).
115 static inline void btrfs_cleanup_ordered_extents(struct inode
*inode
,
116 struct page
*locked_page
,
117 u64 offset
, u64 bytes
)
119 unsigned long index
= offset
>> PAGE_SHIFT
;
120 unsigned long end_index
= (offset
+ bytes
- 1) >> PAGE_SHIFT
;
121 u64 page_start
= page_offset(locked_page
);
122 u64 page_end
= page_start
+ PAGE_SIZE
- 1;
126 while (index
<= end_index
) {
127 page
= find_get_page(inode
->i_mapping
, index
);
131 ClearPagePrivate2(page
);
136 * In case this page belongs to the delalloc range being instantiated
137 * then skip it, since the first page of a range is going to be
138 * properly cleaned up by the caller of run_delalloc_range
140 if (page_start
>= offset
&& page_end
<= (offset
+ bytes
- 1)) {
145 return __endio_write_update_ordered(inode
, offset
, bytes
, false);
148 static int btrfs_dirty_inode(struct inode
*inode
);
150 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
151 void btrfs_test_inode_set_ops(struct inode
*inode
)
153 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
157 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
158 struct inode
*inode
, struct inode
*dir
,
159 const struct qstr
*qstr
)
163 err
= btrfs_init_acl(trans
, inode
, dir
);
165 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
170 * this does all the hard work for inserting an inline extent into
171 * the btree. The caller should have done a btrfs_drop_extents so that
172 * no overlapping inline items exist in the btree
174 static int insert_inline_extent(struct btrfs_trans_handle
*trans
,
175 struct btrfs_path
*path
, int extent_inserted
,
176 struct btrfs_root
*root
, struct inode
*inode
,
177 u64 start
, size_t size
, size_t compressed_size
,
179 struct page
**compressed_pages
)
181 struct extent_buffer
*leaf
;
182 struct page
*page
= NULL
;
185 struct btrfs_file_extent_item
*ei
;
187 size_t cur_size
= size
;
188 unsigned long offset
;
190 if (compressed_size
&& compressed_pages
)
191 cur_size
= compressed_size
;
193 inode_add_bytes(inode
, size
);
195 if (!extent_inserted
) {
196 struct btrfs_key key
;
199 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
201 key
.type
= BTRFS_EXTENT_DATA_KEY
;
203 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
204 path
->leave_spinning
= 1;
205 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
210 leaf
= path
->nodes
[0];
211 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
212 struct btrfs_file_extent_item
);
213 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
214 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
215 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
216 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
217 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
218 ptr
= btrfs_file_extent_inline_start(ei
);
220 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
223 while (compressed_size
> 0) {
224 cpage
= compressed_pages
[i
];
225 cur_size
= min_t(unsigned long, compressed_size
,
228 kaddr
= kmap_atomic(cpage
);
229 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
230 kunmap_atomic(kaddr
);
234 compressed_size
-= cur_size
;
236 btrfs_set_file_extent_compression(leaf
, ei
,
239 page
= find_get_page(inode
->i_mapping
,
240 start
>> PAGE_SHIFT
);
241 btrfs_set_file_extent_compression(leaf
, ei
, 0);
242 kaddr
= kmap_atomic(page
);
243 offset
= offset_in_page(start
);
244 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
245 kunmap_atomic(kaddr
);
248 btrfs_mark_buffer_dirty(leaf
);
249 btrfs_release_path(path
);
252 * we're an inline extent, so nobody can
253 * extend the file past i_size without locking
254 * a page we already have locked.
256 * We must do any isize and inode updates
257 * before we unlock the pages. Otherwise we
258 * could end up racing with unlink.
260 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
261 ret
= btrfs_update_inode(trans
, root
, inode
);
269 * conditionally insert an inline extent into the file. This
270 * does the checks required to make sure the data is small enough
271 * to fit as an inline extent.
273 static noinline
int cow_file_range_inline(struct inode
*inode
, u64 start
,
274 u64 end
, size_t compressed_size
,
276 struct page
**compressed_pages
)
278 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
279 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
280 struct btrfs_trans_handle
*trans
;
281 u64 isize
= i_size_read(inode
);
282 u64 actual_end
= min(end
+ 1, isize
);
283 u64 inline_len
= actual_end
- start
;
284 u64 aligned_end
= ALIGN(end
, fs_info
->sectorsize
);
285 u64 data_len
= inline_len
;
287 struct btrfs_path
*path
;
288 int extent_inserted
= 0;
289 u32 extent_item_size
;
292 data_len
= compressed_size
;
295 actual_end
> fs_info
->sectorsize
||
296 data_len
> BTRFS_MAX_INLINE_DATA_SIZE(fs_info
) ||
298 (actual_end
& (fs_info
->sectorsize
- 1)) == 0) ||
300 data_len
> fs_info
->max_inline
) {
304 path
= btrfs_alloc_path();
308 trans
= btrfs_join_transaction(root
);
310 btrfs_free_path(path
);
311 return PTR_ERR(trans
);
313 trans
->block_rsv
= &BTRFS_I(inode
)->block_rsv
;
315 if (compressed_size
&& compressed_pages
)
316 extent_item_size
= btrfs_file_extent_calc_inline_size(
319 extent_item_size
= btrfs_file_extent_calc_inline_size(
322 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
,
323 start
, aligned_end
, NULL
,
324 1, 1, extent_item_size
, &extent_inserted
);
326 btrfs_abort_transaction(trans
, ret
);
330 if (isize
> actual_end
)
331 inline_len
= min_t(u64
, isize
, actual_end
);
332 ret
= insert_inline_extent(trans
, path
, extent_inserted
,
334 inline_len
, compressed_size
,
335 compress_type
, compressed_pages
);
336 if (ret
&& ret
!= -ENOSPC
) {
337 btrfs_abort_transaction(trans
, ret
);
339 } else if (ret
== -ENOSPC
) {
344 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
345 btrfs_drop_extent_cache(BTRFS_I(inode
), start
, aligned_end
- 1, 0);
348 * Don't forget to free the reserved space, as for inlined extent
349 * it won't count as data extent, free them directly here.
350 * And at reserve time, it's always aligned to page size, so
351 * just free one page here.
353 btrfs_qgroup_free_data(inode
, NULL
, 0, PAGE_SIZE
);
354 btrfs_free_path(path
);
355 btrfs_end_transaction(trans
);
359 struct async_extent
{
364 unsigned long nr_pages
;
366 struct list_head list
;
371 struct btrfs_fs_info
*fs_info
;
372 struct page
*locked_page
;
375 unsigned int write_flags
;
376 struct list_head extents
;
377 struct btrfs_work work
;
380 static noinline
int add_async_extent(struct async_cow
*cow
,
381 u64 start
, u64 ram_size
,
384 unsigned long nr_pages
,
387 struct async_extent
*async_extent
;
389 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
390 BUG_ON(!async_extent
); /* -ENOMEM */
391 async_extent
->start
= start
;
392 async_extent
->ram_size
= ram_size
;
393 async_extent
->compressed_size
= compressed_size
;
394 async_extent
->pages
= pages
;
395 async_extent
->nr_pages
= nr_pages
;
396 async_extent
->compress_type
= compress_type
;
397 list_add_tail(&async_extent
->list
, &cow
->extents
);
401 static inline int inode_need_compress(struct inode
*inode
, u64 start
, u64 end
)
403 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
406 if (btrfs_test_opt(fs_info
, FORCE_COMPRESS
))
409 if (BTRFS_I(inode
)->defrag_compress
)
411 /* bad compression ratios */
412 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
414 if (btrfs_test_opt(fs_info
, COMPRESS
) ||
415 BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
||
416 BTRFS_I(inode
)->prop_compress
)
417 return btrfs_compress_heuristic(inode
, start
, end
);
421 static inline void inode_should_defrag(struct btrfs_inode
*inode
,
422 u64 start
, u64 end
, u64 num_bytes
, u64 small_write
)
424 /* If this is a small write inside eof, kick off a defrag */
425 if (num_bytes
< small_write
&&
426 (start
> 0 || end
+ 1 < inode
->disk_i_size
))
427 btrfs_add_inode_defrag(NULL
, inode
);
431 * we create compressed extents in two phases. The first
432 * phase compresses a range of pages that have already been
433 * locked (both pages and state bits are locked).
435 * This is done inside an ordered work queue, and the compression
436 * is spread across many cpus. The actual IO submission is step
437 * two, and the ordered work queue takes care of making sure that
438 * happens in the same order things were put onto the queue by
439 * writepages and friends.
441 * If this code finds it can't get good compression, it puts an
442 * entry onto the work queue to write the uncompressed bytes. This
443 * makes sure that both compressed inodes and uncompressed inodes
444 * are written in the same order that the flusher thread sent them
447 static noinline
void compress_file_range(struct inode
*inode
,
448 struct page
*locked_page
,
450 struct async_cow
*async_cow
,
453 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
454 u64 blocksize
= fs_info
->sectorsize
;
457 struct page
**pages
= NULL
;
458 unsigned long nr_pages
;
459 unsigned long total_compressed
= 0;
460 unsigned long total_in
= 0;
463 int compress_type
= fs_info
->compress_type
;
466 inode_should_defrag(BTRFS_I(inode
), start
, end
, end
- start
+ 1,
469 actual_end
= min_t(u64
, i_size_read(inode
), end
+ 1);
472 nr_pages
= (end
>> PAGE_SHIFT
) - (start
>> PAGE_SHIFT
) + 1;
473 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED
% PAGE_SIZE
) != 0);
474 nr_pages
= min_t(unsigned long, nr_pages
,
475 BTRFS_MAX_COMPRESSED
/ PAGE_SIZE
);
478 * we don't want to send crud past the end of i_size through
479 * compression, that's just a waste of CPU time. So, if the
480 * end of the file is before the start of our current
481 * requested range of bytes, we bail out to the uncompressed
482 * cleanup code that can deal with all of this.
484 * It isn't really the fastest way to fix things, but this is a
485 * very uncommon corner.
487 if (actual_end
<= start
)
488 goto cleanup_and_bail_uncompressed
;
490 total_compressed
= actual_end
- start
;
493 * skip compression for a small file range(<=blocksize) that
494 * isn't an inline extent, since it doesn't save disk space at all.
496 if (total_compressed
<= blocksize
&&
497 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
498 goto cleanup_and_bail_uncompressed
;
500 total_compressed
= min_t(unsigned long, total_compressed
,
501 BTRFS_MAX_UNCOMPRESSED
);
506 * we do compression for mount -o compress and when the
507 * inode has not been flagged as nocompress. This flag can
508 * change at any time if we discover bad compression ratios.
510 if (inode_need_compress(inode
, start
, end
)) {
512 pages
= kcalloc(nr_pages
, sizeof(struct page
*), GFP_NOFS
);
514 /* just bail out to the uncompressed code */
519 if (BTRFS_I(inode
)->defrag_compress
)
520 compress_type
= BTRFS_I(inode
)->defrag_compress
;
521 else if (BTRFS_I(inode
)->prop_compress
)
522 compress_type
= BTRFS_I(inode
)->prop_compress
;
525 * we need to call clear_page_dirty_for_io on each
526 * page in the range. Otherwise applications with the file
527 * mmap'd can wander in and change the page contents while
528 * we are compressing them.
530 * If the compression fails for any reason, we set the pages
531 * dirty again later on.
533 * Note that the remaining part is redirtied, the start pointer
534 * has moved, the end is the original one.
537 extent_range_clear_dirty_for_io(inode
, start
, end
);
541 /* Compression level is applied here and only here */
542 ret
= btrfs_compress_pages(
543 compress_type
| (fs_info
->compress_level
<< 4),
544 inode
->i_mapping
, start
,
551 unsigned long offset
= offset_in_page(total_compressed
);
552 struct page
*page
= pages
[nr_pages
- 1];
555 /* zero the tail end of the last page, we might be
556 * sending it down to disk
559 kaddr
= kmap_atomic(page
);
560 memset(kaddr
+ offset
, 0,
562 kunmap_atomic(kaddr
);
569 /* lets try to make an inline extent */
570 if (ret
|| total_in
< actual_end
) {
571 /* we didn't compress the entire range, try
572 * to make an uncompressed inline extent.
574 ret
= cow_file_range_inline(inode
, start
, end
, 0,
575 BTRFS_COMPRESS_NONE
, NULL
);
577 /* try making a compressed inline extent */
578 ret
= cow_file_range_inline(inode
, start
, end
,
580 compress_type
, pages
);
583 unsigned long clear_flags
= EXTENT_DELALLOC
|
584 EXTENT_DELALLOC_NEW
| EXTENT_DEFRAG
|
585 EXTENT_DO_ACCOUNTING
;
586 unsigned long page_error_op
;
588 page_error_op
= ret
< 0 ? PAGE_SET_ERROR
: 0;
591 * inline extent creation worked or returned error,
592 * we don't need to create any more async work items.
593 * Unlock and free up our temp pages.
595 * We use DO_ACCOUNTING here because we need the
596 * delalloc_release_metadata to be done _after_ we drop
597 * our outstanding extent for clearing delalloc for this
600 extent_clear_unlock_delalloc(inode
, start
, end
, end
,
613 * we aren't doing an inline extent round the compressed size
614 * up to a block size boundary so the allocator does sane
617 total_compressed
= ALIGN(total_compressed
, blocksize
);
620 * one last check to make sure the compression is really a
621 * win, compare the page count read with the blocks on disk,
622 * compression must free at least one sector size
624 total_in
= ALIGN(total_in
, PAGE_SIZE
);
625 if (total_compressed
+ blocksize
<= total_in
) {
629 * The async work queues will take care of doing actual
630 * allocation on disk for these compressed pages, and
631 * will submit them to the elevator.
633 add_async_extent(async_cow
, start
, total_in
,
634 total_compressed
, pages
, nr_pages
,
637 if (start
+ total_in
< end
) {
648 * the compression code ran but failed to make things smaller,
649 * free any pages it allocated and our page pointer array
651 for (i
= 0; i
< nr_pages
; i
++) {
652 WARN_ON(pages
[i
]->mapping
);
657 total_compressed
= 0;
660 /* flag the file so we don't compress in the future */
661 if (!btrfs_test_opt(fs_info
, FORCE_COMPRESS
) &&
662 !(BTRFS_I(inode
)->prop_compress
)) {
663 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
666 cleanup_and_bail_uncompressed
:
668 * No compression, but we still need to write the pages in the file
669 * we've been given so far. redirty the locked page if it corresponds
670 * to our extent and set things up for the async work queue to run
671 * cow_file_range to do the normal delalloc dance.
673 if (page_offset(locked_page
) >= start
&&
674 page_offset(locked_page
) <= end
)
675 __set_page_dirty_nobuffers(locked_page
);
676 /* unlocked later on in the async handlers */
679 extent_range_redirty_for_io(inode
, start
, end
);
680 add_async_extent(async_cow
, start
, end
- start
+ 1, 0, NULL
, 0,
681 BTRFS_COMPRESS_NONE
);
687 for (i
= 0; i
< nr_pages
; i
++) {
688 WARN_ON(pages
[i
]->mapping
);
694 static void free_async_extent_pages(struct async_extent
*async_extent
)
698 if (!async_extent
->pages
)
701 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
702 WARN_ON(async_extent
->pages
[i
]->mapping
);
703 put_page(async_extent
->pages
[i
]);
705 kfree(async_extent
->pages
);
706 async_extent
->nr_pages
= 0;
707 async_extent
->pages
= NULL
;
711 * phase two of compressed writeback. This is the ordered portion
712 * of the code, which only gets called in the order the work was
713 * queued. We walk all the async extents created by compress_file_range
714 * and send them down to the disk.
716 static noinline
void submit_compressed_extents(struct async_cow
*async_cow
)
718 struct inode
*inode
= async_cow
->inode
;
719 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
720 struct async_extent
*async_extent
;
722 struct btrfs_key ins
;
723 struct extent_map
*em
;
724 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
725 struct extent_io_tree
*io_tree
;
729 while (!list_empty(&async_cow
->extents
)) {
730 async_extent
= list_entry(async_cow
->extents
.next
,
731 struct async_extent
, list
);
732 list_del(&async_extent
->list
);
734 io_tree
= &BTRFS_I(inode
)->io_tree
;
737 /* did the compression code fall back to uncompressed IO? */
738 if (!async_extent
->pages
) {
739 int page_started
= 0;
740 unsigned long nr_written
= 0;
742 lock_extent(io_tree
, async_extent
->start
,
743 async_extent
->start
+
744 async_extent
->ram_size
- 1);
746 /* allocate blocks */
747 ret
= cow_file_range(inode
, async_cow
->locked_page
,
749 async_extent
->start
+
750 async_extent
->ram_size
- 1,
751 async_extent
->start
+
752 async_extent
->ram_size
- 1,
753 &page_started
, &nr_written
, 0,
759 * if page_started, cow_file_range inserted an
760 * inline extent and took care of all the unlocking
761 * and IO for us. Otherwise, we need to submit
762 * all those pages down to the drive.
764 if (!page_started
&& !ret
)
765 extent_write_locked_range(inode
,
767 async_extent
->start
+
768 async_extent
->ram_size
- 1,
771 unlock_page(async_cow
->locked_page
);
777 lock_extent(io_tree
, async_extent
->start
,
778 async_extent
->start
+ async_extent
->ram_size
- 1);
780 ret
= btrfs_reserve_extent(root
, async_extent
->ram_size
,
781 async_extent
->compressed_size
,
782 async_extent
->compressed_size
,
783 0, alloc_hint
, &ins
, 1, 1);
785 free_async_extent_pages(async_extent
);
787 if (ret
== -ENOSPC
) {
788 unlock_extent(io_tree
, async_extent
->start
,
789 async_extent
->start
+
790 async_extent
->ram_size
- 1);
793 * we need to redirty the pages if we decide to
794 * fallback to uncompressed IO, otherwise we
795 * will not submit these pages down to lower
798 extent_range_redirty_for_io(inode
,
800 async_extent
->start
+
801 async_extent
->ram_size
- 1);
808 * here we're doing allocation and writeback of the
811 em
= create_io_em(inode
, async_extent
->start
,
812 async_extent
->ram_size
, /* len */
813 async_extent
->start
, /* orig_start */
814 ins
.objectid
, /* block_start */
815 ins
.offset
, /* block_len */
816 ins
.offset
, /* orig_block_len */
817 async_extent
->ram_size
, /* ram_bytes */
818 async_extent
->compress_type
,
819 BTRFS_ORDERED_COMPRESSED
);
821 /* ret value is not necessary due to void function */
822 goto out_free_reserve
;
825 ret
= btrfs_add_ordered_extent_compress(inode
,
828 async_extent
->ram_size
,
830 BTRFS_ORDERED_COMPRESSED
,
831 async_extent
->compress_type
);
833 btrfs_drop_extent_cache(BTRFS_I(inode
),
835 async_extent
->start
+
836 async_extent
->ram_size
- 1, 0);
837 goto out_free_reserve
;
839 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
842 * clear dirty, set writeback and unlock the pages.
844 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
845 async_extent
->start
+
846 async_extent
->ram_size
- 1,
847 async_extent
->start
+
848 async_extent
->ram_size
- 1,
849 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
,
850 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
852 if (btrfs_submit_compressed_write(inode
,
854 async_extent
->ram_size
,
856 ins
.offset
, async_extent
->pages
,
857 async_extent
->nr_pages
,
858 async_cow
->write_flags
)) {
859 struct page
*p
= async_extent
->pages
[0];
860 const u64 start
= async_extent
->start
;
861 const u64 end
= start
+ async_extent
->ram_size
- 1;
863 p
->mapping
= inode
->i_mapping
;
864 btrfs_writepage_endio_finish_ordered(p
, start
, end
, 0);
867 extent_clear_unlock_delalloc(inode
, start
, end
, end
,
871 free_async_extent_pages(async_extent
);
873 alloc_hint
= ins
.objectid
+ ins
.offset
;
879 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
880 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 1);
882 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
883 async_extent
->start
+
884 async_extent
->ram_size
- 1,
885 async_extent
->start
+
886 async_extent
->ram_size
- 1,
887 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
|
888 EXTENT_DELALLOC_NEW
|
889 EXTENT_DEFRAG
| EXTENT_DO_ACCOUNTING
,
890 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
891 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
|
893 free_async_extent_pages(async_extent
);
898 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
901 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
902 struct extent_map
*em
;
905 read_lock(&em_tree
->lock
);
906 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
909 * if block start isn't an actual block number then find the
910 * first block in this inode and use that as a hint. If that
911 * block is also bogus then just don't worry about it.
913 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
915 em
= search_extent_mapping(em_tree
, 0, 0);
916 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
917 alloc_hint
= em
->block_start
;
921 alloc_hint
= em
->block_start
;
925 read_unlock(&em_tree
->lock
);
931 * when extent_io.c finds a delayed allocation range in the file,
932 * the call backs end up in this code. The basic idea is to
933 * allocate extents on disk for the range, and create ordered data structs
934 * in ram to track those extents.
936 * locked_page is the page that writepage had locked already. We use
937 * it to make sure we don't do extra locks or unlocks.
939 * *page_started is set to one if we unlock locked_page and do everything
940 * required to start IO on it. It may be clean and already done with
943 static noinline
int cow_file_range(struct inode
*inode
,
944 struct page
*locked_page
,
945 u64 start
, u64 end
, u64 delalloc_end
,
946 int *page_started
, unsigned long *nr_written
,
947 int unlock
, struct btrfs_dedupe_hash
*hash
)
949 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
950 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
953 unsigned long ram_size
;
954 u64 cur_alloc_size
= 0;
955 u64 blocksize
= fs_info
->sectorsize
;
956 struct btrfs_key ins
;
957 struct extent_map
*em
;
959 unsigned long page_ops
;
960 bool extent_reserved
= false;
963 if (btrfs_is_free_space_inode(BTRFS_I(inode
))) {
969 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
970 num_bytes
= max(blocksize
, num_bytes
);
971 ASSERT(num_bytes
<= btrfs_super_total_bytes(fs_info
->super_copy
));
973 inode_should_defrag(BTRFS_I(inode
), start
, end
, num_bytes
, SZ_64K
);
976 /* lets try to make an inline extent */
977 ret
= cow_file_range_inline(inode
, start
, end
, 0,
978 BTRFS_COMPRESS_NONE
, NULL
);
981 * We use DO_ACCOUNTING here because we need the
982 * delalloc_release_metadata to be run _after_ we drop
983 * our outstanding extent for clearing delalloc for this
986 extent_clear_unlock_delalloc(inode
, start
, end
,
988 EXTENT_LOCKED
| EXTENT_DELALLOC
|
989 EXTENT_DELALLOC_NEW
| EXTENT_DEFRAG
|
990 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
991 PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
993 *nr_written
= *nr_written
+
994 (end
- start
+ PAGE_SIZE
) / PAGE_SIZE
;
997 } else if (ret
< 0) {
1002 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
1003 btrfs_drop_extent_cache(BTRFS_I(inode
), start
,
1004 start
+ num_bytes
- 1, 0);
1006 while (num_bytes
> 0) {
1007 cur_alloc_size
= num_bytes
;
1008 ret
= btrfs_reserve_extent(root
, cur_alloc_size
, cur_alloc_size
,
1009 fs_info
->sectorsize
, 0, alloc_hint
,
1013 cur_alloc_size
= ins
.offset
;
1014 extent_reserved
= true;
1016 ram_size
= ins
.offset
;
1017 em
= create_io_em(inode
, start
, ins
.offset
, /* len */
1018 start
, /* orig_start */
1019 ins
.objectid
, /* block_start */
1020 ins
.offset
, /* block_len */
1021 ins
.offset
, /* orig_block_len */
1022 ram_size
, /* ram_bytes */
1023 BTRFS_COMPRESS_NONE
, /* compress_type */
1024 BTRFS_ORDERED_REGULAR
/* type */);
1029 free_extent_map(em
);
1031 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
1032 ram_size
, cur_alloc_size
, 0);
1034 goto out_drop_extent_cache
;
1036 if (root
->root_key
.objectid
==
1037 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1038 ret
= btrfs_reloc_clone_csums(inode
, start
,
1041 * Only drop cache here, and process as normal.
1043 * We must not allow extent_clear_unlock_delalloc()
1044 * at out_unlock label to free meta of this ordered
1045 * extent, as its meta should be freed by
1046 * btrfs_finish_ordered_io().
1048 * So we must continue until @start is increased to
1049 * skip current ordered extent.
1052 btrfs_drop_extent_cache(BTRFS_I(inode
), start
,
1053 start
+ ram_size
- 1, 0);
1056 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
1058 /* we're not doing compressed IO, don't unlock the first
1059 * page (which the caller expects to stay locked), don't
1060 * clear any dirty bits and don't set any writeback bits
1062 * Do set the Private2 bit so we know this page was properly
1063 * setup for writepage
1065 page_ops
= unlock
? PAGE_UNLOCK
: 0;
1066 page_ops
|= PAGE_SET_PRIVATE2
;
1068 extent_clear_unlock_delalloc(inode
, start
,
1069 start
+ ram_size
- 1,
1070 delalloc_end
, locked_page
,
1071 EXTENT_LOCKED
| EXTENT_DELALLOC
,
1073 if (num_bytes
< cur_alloc_size
)
1076 num_bytes
-= cur_alloc_size
;
1077 alloc_hint
= ins
.objectid
+ ins
.offset
;
1078 start
+= cur_alloc_size
;
1079 extent_reserved
= false;
1082 * btrfs_reloc_clone_csums() error, since start is increased
1083 * extent_clear_unlock_delalloc() at out_unlock label won't
1084 * free metadata of current ordered extent, we're OK to exit.
1092 out_drop_extent_cache
:
1093 btrfs_drop_extent_cache(BTRFS_I(inode
), start
, start
+ ram_size
- 1, 0);
1095 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
1096 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 1);
1098 clear_bits
= EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DELALLOC_NEW
|
1099 EXTENT_DEFRAG
| EXTENT_CLEAR_META_RESV
;
1100 page_ops
= PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
1103 * If we reserved an extent for our delalloc range (or a subrange) and
1104 * failed to create the respective ordered extent, then it means that
1105 * when we reserved the extent we decremented the extent's size from
1106 * the data space_info's bytes_may_use counter and incremented the
1107 * space_info's bytes_reserved counter by the same amount. We must make
1108 * sure extent_clear_unlock_delalloc() does not try to decrement again
1109 * the data space_info's bytes_may_use counter, therefore we do not pass
1110 * it the flag EXTENT_CLEAR_DATA_RESV.
1112 if (extent_reserved
) {
1113 extent_clear_unlock_delalloc(inode
, start
,
1114 start
+ cur_alloc_size
,
1115 start
+ cur_alloc_size
,
1119 start
+= cur_alloc_size
;
1123 extent_clear_unlock_delalloc(inode
, start
, end
, delalloc_end
,
1125 clear_bits
| EXTENT_CLEAR_DATA_RESV
,
1131 * work queue call back to started compression on a file and pages
1133 static noinline
void async_cow_start(struct btrfs_work
*work
)
1135 struct async_cow
*async_cow
;
1137 async_cow
= container_of(work
, struct async_cow
, work
);
1139 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
1140 async_cow
->start
, async_cow
->end
, async_cow
,
1142 if (num_added
== 0) {
1143 btrfs_add_delayed_iput(async_cow
->inode
);
1144 async_cow
->inode
= NULL
;
1149 * work queue call back to submit previously compressed pages
1151 static noinline
void async_cow_submit(struct btrfs_work
*work
)
1153 struct btrfs_fs_info
*fs_info
;
1154 struct async_cow
*async_cow
;
1155 unsigned long nr_pages
;
1157 async_cow
= container_of(work
, struct async_cow
, work
);
1159 fs_info
= async_cow
->fs_info
;
1160 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_SIZE
) >>
1163 /* atomic_sub_return implies a barrier */
1164 if (atomic_sub_return(nr_pages
, &fs_info
->async_delalloc_pages
) <
1166 cond_wake_up_nomb(&fs_info
->async_submit_wait
);
1169 * ->inode could be NULL if async_cow_start has failed to compress,
1170 * in which case we don't have anything to submit, yet we need to
1171 * always adjust ->async_delalloc_pages as its paired with the init
1172 * happening in cow_file_range_async
1174 if (async_cow
->inode
)
1175 submit_compressed_extents(async_cow
);
1178 static noinline
void async_cow_free(struct btrfs_work
*work
)
1180 struct async_cow
*async_cow
;
1181 async_cow
= container_of(work
, struct async_cow
, work
);
1182 if (async_cow
->inode
)
1183 btrfs_add_delayed_iput(async_cow
->inode
);
1187 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1188 u64 start
, u64 end
, int *page_started
,
1189 unsigned long *nr_written
,
1190 unsigned int write_flags
)
1192 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1193 struct async_cow
*async_cow
;
1194 unsigned long nr_pages
;
1197 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1199 while (start
< end
) {
1200 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1201 BUG_ON(!async_cow
); /* -ENOMEM */
1203 * igrab is called higher up in the call chain, take only the
1204 * lightweight reference for the callback lifetime
1207 async_cow
->inode
= inode
;
1208 async_cow
->fs_info
= fs_info
;
1209 async_cow
->locked_page
= locked_page
;
1210 async_cow
->start
= start
;
1211 async_cow
->write_flags
= write_flags
;
1213 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
&&
1214 !btrfs_test_opt(fs_info
, FORCE_COMPRESS
))
1217 cur_end
= min(end
, start
+ SZ_512K
- 1);
1219 async_cow
->end
= cur_end
;
1220 INIT_LIST_HEAD(&async_cow
->extents
);
1222 btrfs_init_work(&async_cow
->work
,
1223 btrfs_delalloc_helper
,
1224 async_cow_start
, async_cow_submit
,
1227 nr_pages
= (cur_end
- start
+ PAGE_SIZE
) >>
1229 atomic_add(nr_pages
, &fs_info
->async_delalloc_pages
);
1231 btrfs_queue_work(fs_info
->delalloc_workers
, &async_cow
->work
);
1233 *nr_written
+= nr_pages
;
1234 start
= cur_end
+ 1;
1240 static noinline
int csum_exist_in_range(struct btrfs_fs_info
*fs_info
,
1241 u64 bytenr
, u64 num_bytes
)
1244 struct btrfs_ordered_sum
*sums
;
1247 ret
= btrfs_lookup_csums_range(fs_info
->csum_root
, bytenr
,
1248 bytenr
+ num_bytes
- 1, &list
, 0);
1249 if (ret
== 0 && list_empty(&list
))
1252 while (!list_empty(&list
)) {
1253 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1254 list_del(&sums
->list
);
1263 * when nowcow writeback call back. This checks for snapshots or COW copies
1264 * of the extents that exist in the file, and COWs the file as required.
1266 * If no cow copies or snapshots exist, we write directly to the existing
1269 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1270 struct page
*locked_page
,
1271 u64 start
, u64 end
, int *page_started
, int force
,
1272 unsigned long *nr_written
)
1274 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1275 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1276 struct extent_buffer
*leaf
;
1277 struct btrfs_path
*path
;
1278 struct btrfs_file_extent_item
*fi
;
1279 struct btrfs_key found_key
;
1280 struct extent_map
*em
;
1295 u64 ino
= btrfs_ino(BTRFS_I(inode
));
1297 path
= btrfs_alloc_path();
1299 extent_clear_unlock_delalloc(inode
, start
, end
, end
,
1301 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1302 EXTENT_DO_ACCOUNTING
|
1303 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1305 PAGE_SET_WRITEBACK
|
1306 PAGE_END_WRITEBACK
);
1310 nolock
= btrfs_is_free_space_inode(BTRFS_I(inode
));
1312 cow_start
= (u64
)-1;
1315 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, ino
,
1319 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1320 leaf
= path
->nodes
[0];
1321 btrfs_item_key_to_cpu(leaf
, &found_key
,
1322 path
->slots
[0] - 1);
1323 if (found_key
.objectid
== ino
&&
1324 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1329 leaf
= path
->nodes
[0];
1330 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1331 ret
= btrfs_next_leaf(root
, path
);
1333 if (cow_start
!= (u64
)-1)
1334 cur_offset
= cow_start
;
1339 leaf
= path
->nodes
[0];
1345 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1347 if (found_key
.objectid
> ino
)
1349 if (WARN_ON_ONCE(found_key
.objectid
< ino
) ||
1350 found_key
.type
< BTRFS_EXTENT_DATA_KEY
) {
1354 if (found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1355 found_key
.offset
> end
)
1358 if (found_key
.offset
> cur_offset
) {
1359 extent_end
= found_key
.offset
;
1364 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1365 struct btrfs_file_extent_item
);
1366 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1368 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
1369 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1370 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1371 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1372 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1373 extent_end
= found_key
.offset
+
1374 btrfs_file_extent_num_bytes(leaf
, fi
);
1376 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1377 if (extent_end
<= start
) {
1381 if (disk_bytenr
== 0)
1383 if (btrfs_file_extent_compression(leaf
, fi
) ||
1384 btrfs_file_extent_encryption(leaf
, fi
) ||
1385 btrfs_file_extent_other_encoding(leaf
, fi
))
1388 * Do the same check as in btrfs_cross_ref_exist but
1389 * without the unnecessary search.
1392 btrfs_file_extent_generation(leaf
, fi
) <=
1393 btrfs_root_last_snapshot(&root
->root_item
))
1395 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1397 if (btrfs_extent_readonly(fs_info
, disk_bytenr
))
1399 ret
= btrfs_cross_ref_exist(root
, ino
,
1401 extent_offset
, disk_bytenr
);
1404 * ret could be -EIO if the above fails to read
1408 if (cow_start
!= (u64
)-1)
1409 cur_offset
= cow_start
;
1413 WARN_ON_ONCE(nolock
);
1416 disk_bytenr
+= extent_offset
;
1417 disk_bytenr
+= cur_offset
- found_key
.offset
;
1418 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1420 * if there are pending snapshots for this root,
1421 * we fall into common COW way.
1423 if (!nolock
&& atomic_read(&root
->snapshot_force_cow
))
1426 * force cow if csum exists in the range.
1427 * this ensure that csum for a given extent are
1428 * either valid or do not exist.
1430 ret
= csum_exist_in_range(fs_info
, disk_bytenr
,
1434 * ret could be -EIO if the above fails to read
1438 if (cow_start
!= (u64
)-1)
1439 cur_offset
= cow_start
;
1442 WARN_ON_ONCE(nolock
);
1445 if (!btrfs_inc_nocow_writers(fs_info
, disk_bytenr
))
1448 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1449 extent_end
= found_key
.offset
+
1450 btrfs_file_extent_ram_bytes(leaf
, fi
);
1451 extent_end
= ALIGN(extent_end
,
1452 fs_info
->sectorsize
);
1457 if (extent_end
<= start
) {
1460 btrfs_dec_nocow_writers(fs_info
, disk_bytenr
);
1464 if (cow_start
== (u64
)-1)
1465 cow_start
= cur_offset
;
1466 cur_offset
= extent_end
;
1467 if (cur_offset
> end
)
1473 btrfs_release_path(path
);
1474 if (cow_start
!= (u64
)-1) {
1475 ret
= cow_file_range(inode
, locked_page
,
1476 cow_start
, found_key
.offset
- 1,
1477 end
, page_started
, nr_written
, 1,
1481 btrfs_dec_nocow_writers(fs_info
,
1485 cow_start
= (u64
)-1;
1488 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1489 u64 orig_start
= found_key
.offset
- extent_offset
;
1491 em
= create_io_em(inode
, cur_offset
, num_bytes
,
1493 disk_bytenr
, /* block_start */
1494 num_bytes
, /* block_len */
1495 disk_num_bytes
, /* orig_block_len */
1496 ram_bytes
, BTRFS_COMPRESS_NONE
,
1497 BTRFS_ORDERED_PREALLOC
);
1500 btrfs_dec_nocow_writers(fs_info
,
1505 free_extent_map(em
);
1508 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1509 type
= BTRFS_ORDERED_PREALLOC
;
1511 type
= BTRFS_ORDERED_NOCOW
;
1514 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1515 num_bytes
, num_bytes
, type
);
1517 btrfs_dec_nocow_writers(fs_info
, disk_bytenr
);
1518 BUG_ON(ret
); /* -ENOMEM */
1520 if (root
->root_key
.objectid
==
1521 BTRFS_DATA_RELOC_TREE_OBJECTID
)
1523 * Error handled later, as we must prevent
1524 * extent_clear_unlock_delalloc() in error handler
1525 * from freeing metadata of created ordered extent.
1527 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1530 extent_clear_unlock_delalloc(inode
, cur_offset
,
1531 cur_offset
+ num_bytes
- 1, end
,
1532 locked_page
, EXTENT_LOCKED
|
1534 EXTENT_CLEAR_DATA_RESV
,
1535 PAGE_UNLOCK
| PAGE_SET_PRIVATE2
);
1537 cur_offset
= extent_end
;
1540 * btrfs_reloc_clone_csums() error, now we're OK to call error
1541 * handler, as metadata for created ordered extent will only
1542 * be freed by btrfs_finish_ordered_io().
1546 if (cur_offset
> end
)
1549 btrfs_release_path(path
);
1551 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1552 cow_start
= cur_offset
;
1554 if (cow_start
!= (u64
)-1) {
1556 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
, end
,
1557 page_started
, nr_written
, 1, NULL
);
1563 if (ret
&& cur_offset
< end
)
1564 extent_clear_unlock_delalloc(inode
, cur_offset
, end
, end
,
1565 locked_page
, EXTENT_LOCKED
|
1566 EXTENT_DELALLOC
| EXTENT_DEFRAG
|
1567 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
1569 PAGE_SET_WRITEBACK
|
1570 PAGE_END_WRITEBACK
);
1571 btrfs_free_path(path
);
1575 static inline int need_force_cow(struct inode
*inode
, u64 start
, u64 end
)
1578 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
1579 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
))
1583 * @defrag_bytes is a hint value, no spinlock held here,
1584 * if is not zero, it means the file is defragging.
1585 * Force cow if given extent needs to be defragged.
1587 if (BTRFS_I(inode
)->defrag_bytes
&&
1588 test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1589 EXTENT_DEFRAG
, 0, NULL
))
1596 * Function to process delayed allocation (create CoW) for ranges which are
1597 * being touched for the first time.
1599 int btrfs_run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1600 u64 start
, u64 end
, int *page_started
, unsigned long *nr_written
,
1601 struct writeback_control
*wbc
)
1604 int force_cow
= need_force_cow(inode
, start
, end
);
1605 unsigned int write_flags
= wbc_to_write_flags(wbc
);
1607 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
&& !force_cow
) {
1608 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1609 page_started
, 1, nr_written
);
1610 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
&& !force_cow
) {
1611 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1612 page_started
, 0, nr_written
);
1613 } else if (!inode_need_compress(inode
, start
, end
)) {
1614 ret
= cow_file_range(inode
, locked_page
, start
, end
, end
,
1615 page_started
, nr_written
, 1, NULL
);
1617 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1618 &BTRFS_I(inode
)->runtime_flags
);
1619 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1620 page_started
, nr_written
,
1624 btrfs_cleanup_ordered_extents(inode
, locked_page
, start
,
1629 void btrfs_split_delalloc_extent(struct inode
*inode
,
1630 struct extent_state
*orig
, u64 split
)
1634 /* not delalloc, ignore it */
1635 if (!(orig
->state
& EXTENT_DELALLOC
))
1638 size
= orig
->end
- orig
->start
+ 1;
1639 if (size
> BTRFS_MAX_EXTENT_SIZE
) {
1644 * See the explanation in btrfs_merge_delalloc_extent, the same
1645 * applies here, just in reverse.
1647 new_size
= orig
->end
- split
+ 1;
1648 num_extents
= count_max_extents(new_size
);
1649 new_size
= split
- orig
->start
;
1650 num_extents
+= count_max_extents(new_size
);
1651 if (count_max_extents(size
) >= num_extents
)
1655 spin_lock(&BTRFS_I(inode
)->lock
);
1656 btrfs_mod_outstanding_extents(BTRFS_I(inode
), 1);
1657 spin_unlock(&BTRFS_I(inode
)->lock
);
1661 * Handle merged delayed allocation extents so we can keep track of new extents
1662 * that are just merged onto old extents, such as when we are doing sequential
1663 * writes, so we can properly account for the metadata space we'll need.
1665 void btrfs_merge_delalloc_extent(struct inode
*inode
, struct extent_state
*new,
1666 struct extent_state
*other
)
1668 u64 new_size
, old_size
;
1671 /* not delalloc, ignore it */
1672 if (!(other
->state
& EXTENT_DELALLOC
))
1675 if (new->start
> other
->start
)
1676 new_size
= new->end
- other
->start
+ 1;
1678 new_size
= other
->end
- new->start
+ 1;
1680 /* we're not bigger than the max, unreserve the space and go */
1681 if (new_size
<= BTRFS_MAX_EXTENT_SIZE
) {
1682 spin_lock(&BTRFS_I(inode
)->lock
);
1683 btrfs_mod_outstanding_extents(BTRFS_I(inode
), -1);
1684 spin_unlock(&BTRFS_I(inode
)->lock
);
1689 * We have to add up either side to figure out how many extents were
1690 * accounted for before we merged into one big extent. If the number of
1691 * extents we accounted for is <= the amount we need for the new range
1692 * then we can return, otherwise drop. Think of it like this
1696 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1697 * need 2 outstanding extents, on one side we have 1 and the other side
1698 * we have 1 so they are == and we can return. But in this case
1700 * [MAX_SIZE+4k][MAX_SIZE+4k]
1702 * Each range on their own accounts for 2 extents, but merged together
1703 * they are only 3 extents worth of accounting, so we need to drop in
1706 old_size
= other
->end
- other
->start
+ 1;
1707 num_extents
= count_max_extents(old_size
);
1708 old_size
= new->end
- new->start
+ 1;
1709 num_extents
+= count_max_extents(old_size
);
1710 if (count_max_extents(new_size
) >= num_extents
)
1713 spin_lock(&BTRFS_I(inode
)->lock
);
1714 btrfs_mod_outstanding_extents(BTRFS_I(inode
), -1);
1715 spin_unlock(&BTRFS_I(inode
)->lock
);
1718 static void btrfs_add_delalloc_inodes(struct btrfs_root
*root
,
1719 struct inode
*inode
)
1721 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1723 spin_lock(&root
->delalloc_lock
);
1724 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1725 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1726 &root
->delalloc_inodes
);
1727 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1728 &BTRFS_I(inode
)->runtime_flags
);
1729 root
->nr_delalloc_inodes
++;
1730 if (root
->nr_delalloc_inodes
== 1) {
1731 spin_lock(&fs_info
->delalloc_root_lock
);
1732 BUG_ON(!list_empty(&root
->delalloc_root
));
1733 list_add_tail(&root
->delalloc_root
,
1734 &fs_info
->delalloc_roots
);
1735 spin_unlock(&fs_info
->delalloc_root_lock
);
1738 spin_unlock(&root
->delalloc_lock
);
1742 void __btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1743 struct btrfs_inode
*inode
)
1745 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1747 if (!list_empty(&inode
->delalloc_inodes
)) {
1748 list_del_init(&inode
->delalloc_inodes
);
1749 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1750 &inode
->runtime_flags
);
1751 root
->nr_delalloc_inodes
--;
1752 if (!root
->nr_delalloc_inodes
) {
1753 ASSERT(list_empty(&root
->delalloc_inodes
));
1754 spin_lock(&fs_info
->delalloc_root_lock
);
1755 BUG_ON(list_empty(&root
->delalloc_root
));
1756 list_del_init(&root
->delalloc_root
);
1757 spin_unlock(&fs_info
->delalloc_root_lock
);
1762 static void btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1763 struct btrfs_inode
*inode
)
1765 spin_lock(&root
->delalloc_lock
);
1766 __btrfs_del_delalloc_inode(root
, inode
);
1767 spin_unlock(&root
->delalloc_lock
);
1771 * Properly track delayed allocation bytes in the inode and to maintain the
1772 * list of inodes that have pending delalloc work to be done.
1774 void btrfs_set_delalloc_extent(struct inode
*inode
, struct extent_state
*state
,
1777 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1779 if ((*bits
& EXTENT_DEFRAG
) && !(*bits
& EXTENT_DELALLOC
))
1782 * set_bit and clear bit hooks normally require _irqsave/restore
1783 * but in this case, we are only testing for the DELALLOC
1784 * bit, which is only set or cleared with irqs on
1786 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1787 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1788 u64 len
= state
->end
+ 1 - state
->start
;
1789 u32 num_extents
= count_max_extents(len
);
1790 bool do_list
= !btrfs_is_free_space_inode(BTRFS_I(inode
));
1792 spin_lock(&BTRFS_I(inode
)->lock
);
1793 btrfs_mod_outstanding_extents(BTRFS_I(inode
), num_extents
);
1794 spin_unlock(&BTRFS_I(inode
)->lock
);
1796 /* For sanity tests */
1797 if (btrfs_is_testing(fs_info
))
1800 percpu_counter_add_batch(&fs_info
->delalloc_bytes
, len
,
1801 fs_info
->delalloc_batch
);
1802 spin_lock(&BTRFS_I(inode
)->lock
);
1803 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1804 if (*bits
& EXTENT_DEFRAG
)
1805 BTRFS_I(inode
)->defrag_bytes
+= len
;
1806 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1807 &BTRFS_I(inode
)->runtime_flags
))
1808 btrfs_add_delalloc_inodes(root
, inode
);
1809 spin_unlock(&BTRFS_I(inode
)->lock
);
1812 if (!(state
->state
& EXTENT_DELALLOC_NEW
) &&
1813 (*bits
& EXTENT_DELALLOC_NEW
)) {
1814 spin_lock(&BTRFS_I(inode
)->lock
);
1815 BTRFS_I(inode
)->new_delalloc_bytes
+= state
->end
+ 1 -
1817 spin_unlock(&BTRFS_I(inode
)->lock
);
1822 * Once a range is no longer delalloc this function ensures that proper
1823 * accounting happens.
1825 void btrfs_clear_delalloc_extent(struct inode
*vfs_inode
,
1826 struct extent_state
*state
, unsigned *bits
)
1828 struct btrfs_inode
*inode
= BTRFS_I(vfs_inode
);
1829 struct btrfs_fs_info
*fs_info
= btrfs_sb(vfs_inode
->i_sb
);
1830 u64 len
= state
->end
+ 1 - state
->start
;
1831 u32 num_extents
= count_max_extents(len
);
1833 if ((state
->state
& EXTENT_DEFRAG
) && (*bits
& EXTENT_DEFRAG
)) {
1834 spin_lock(&inode
->lock
);
1835 inode
->defrag_bytes
-= len
;
1836 spin_unlock(&inode
->lock
);
1840 * set_bit and clear bit hooks normally require _irqsave/restore
1841 * but in this case, we are only testing for the DELALLOC
1842 * bit, which is only set or cleared with irqs on
1844 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1845 struct btrfs_root
*root
= inode
->root
;
1846 bool do_list
= !btrfs_is_free_space_inode(inode
);
1848 spin_lock(&inode
->lock
);
1849 btrfs_mod_outstanding_extents(inode
, -num_extents
);
1850 spin_unlock(&inode
->lock
);
1853 * We don't reserve metadata space for space cache inodes so we
1854 * don't need to call delalloc_release_metadata if there is an
1857 if (*bits
& EXTENT_CLEAR_META_RESV
&&
1858 root
!= fs_info
->tree_root
)
1859 btrfs_delalloc_release_metadata(inode
, len
, false);
1861 /* For sanity tests. */
1862 if (btrfs_is_testing(fs_info
))
1865 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1866 do_list
&& !(state
->state
& EXTENT_NORESERVE
) &&
1867 (*bits
& EXTENT_CLEAR_DATA_RESV
))
1868 btrfs_free_reserved_data_space_noquota(
1872 percpu_counter_add_batch(&fs_info
->delalloc_bytes
, -len
,
1873 fs_info
->delalloc_batch
);
1874 spin_lock(&inode
->lock
);
1875 inode
->delalloc_bytes
-= len
;
1876 if (do_list
&& inode
->delalloc_bytes
== 0 &&
1877 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1878 &inode
->runtime_flags
))
1879 btrfs_del_delalloc_inode(root
, inode
);
1880 spin_unlock(&inode
->lock
);
1883 if ((state
->state
& EXTENT_DELALLOC_NEW
) &&
1884 (*bits
& EXTENT_DELALLOC_NEW
)) {
1885 spin_lock(&inode
->lock
);
1886 ASSERT(inode
->new_delalloc_bytes
>= len
);
1887 inode
->new_delalloc_bytes
-= len
;
1888 spin_unlock(&inode
->lock
);
1893 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
1894 * in a chunk's stripe. This function ensures that bios do not span a
1897 * @page - The page we are about to add to the bio
1898 * @size - size we want to add to the bio
1899 * @bio - bio we want to ensure is smaller than a stripe
1900 * @bio_flags - flags of the bio
1902 * return 1 if page cannot be added to the bio
1903 * return 0 if page can be added to the bio
1904 * return error otherwise
1906 int btrfs_bio_fits_in_stripe(struct page
*page
, size_t size
, struct bio
*bio
,
1907 unsigned long bio_flags
)
1909 struct inode
*inode
= page
->mapping
->host
;
1910 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1911 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
1916 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1919 length
= bio
->bi_iter
.bi_size
;
1920 map_length
= length
;
1921 ret
= btrfs_map_block(fs_info
, btrfs_op(bio
), logical
, &map_length
,
1925 if (map_length
< length
+ size
)
1931 * in order to insert checksums into the metadata in large chunks,
1932 * we wait until bio submission time. All the pages in the bio are
1933 * checksummed and sums are attached onto the ordered extent record.
1935 * At IO completion time the cums attached on the ordered extent record
1936 * are inserted into the btree
1938 static blk_status_t
btrfs_submit_bio_start(void *private_data
, struct bio
*bio
,
1941 struct inode
*inode
= private_data
;
1942 blk_status_t ret
= 0;
1944 ret
= btrfs_csum_one_bio(inode
, bio
, 0, 0);
1945 BUG_ON(ret
); /* -ENOMEM */
1950 * extent_io.c submission hook. This does the right thing for csum calculation
1951 * on write, or reading the csums from the tree before a read.
1953 * Rules about async/sync submit,
1954 * a) read: sync submit
1956 * b) write without checksum: sync submit
1958 * c) write with checksum:
1959 * c-1) if bio is issued by fsync: sync submit
1960 * (sync_writers != 0)
1962 * c-2) if root is reloc root: sync submit
1963 * (only in case of buffered IO)
1965 * c-3) otherwise: async submit
1967 static blk_status_t
btrfs_submit_bio_hook(void *private_data
, struct bio
*bio
,
1968 int mirror_num
, unsigned long bio_flags
,
1971 struct inode
*inode
= private_data
;
1972 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1973 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1974 enum btrfs_wq_endio_type metadata
= BTRFS_WQ_ENDIO_DATA
;
1975 blk_status_t ret
= 0;
1977 int async
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
1979 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1981 if (btrfs_is_free_space_inode(BTRFS_I(inode
)))
1982 metadata
= BTRFS_WQ_ENDIO_FREE_SPACE
;
1984 if (bio_op(bio
) != REQ_OP_WRITE
) {
1985 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, metadata
);
1989 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1990 ret
= btrfs_submit_compressed_read(inode
, bio
,
1994 } else if (!skip_sum
) {
1995 ret
= btrfs_lookup_bio_sums(inode
, bio
, NULL
);
2000 } else if (async
&& !skip_sum
) {
2001 /* csum items have already been cloned */
2002 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2004 /* we're doing a write, do the async checksumming */
2005 ret
= btrfs_wq_submit_bio(fs_info
, bio
, mirror_num
, bio_flags
,
2007 btrfs_submit_bio_start
);
2009 } else if (!skip_sum
) {
2010 ret
= btrfs_csum_one_bio(inode
, bio
, 0, 0);
2016 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
2020 bio
->bi_status
= ret
;
2027 * given a list of ordered sums record them in the inode. This happens
2028 * at IO completion time based on sums calculated at bio submission time.
2030 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
2031 struct inode
*inode
, struct list_head
*list
)
2033 struct btrfs_ordered_sum
*sum
;
2036 list_for_each_entry(sum
, list
, list
) {
2037 trans
->adding_csums
= true;
2038 ret
= btrfs_csum_file_blocks(trans
,
2039 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
2040 trans
->adding_csums
= false;
2047 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
2048 unsigned int extra_bits
,
2049 struct extent_state
**cached_state
, int dedupe
)
2051 WARN_ON(PAGE_ALIGNED(end
));
2052 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
2053 extra_bits
, cached_state
);
2056 /* see btrfs_writepage_start_hook for details on why this is required */
2057 struct btrfs_writepage_fixup
{
2059 struct btrfs_work work
;
2062 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
2064 struct btrfs_writepage_fixup
*fixup
;
2065 struct btrfs_ordered_extent
*ordered
;
2066 struct extent_state
*cached_state
= NULL
;
2067 struct extent_changeset
*data_reserved
= NULL
;
2069 struct inode
*inode
;
2074 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
2078 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
2079 ClearPageChecked(page
);
2083 inode
= page
->mapping
->host
;
2084 page_start
= page_offset(page
);
2085 page_end
= page_offset(page
) + PAGE_SIZE
- 1;
2087 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2090 /* already ordered? We're done */
2091 if (PagePrivate2(page
))
2094 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), page_start
,
2097 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
2098 page_end
, &cached_state
);
2100 btrfs_start_ordered_extent(inode
, ordered
, 1);
2101 btrfs_put_ordered_extent(ordered
);
2105 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
, page_start
,
2108 mapping_set_error(page
->mapping
, ret
);
2109 end_extent_writepage(page
, ret
, page_start
, page_end
);
2110 ClearPageChecked(page
);
2114 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
, 0,
2117 mapping_set_error(page
->mapping
, ret
);
2118 end_extent_writepage(page
, ret
, page_start
, page_end
);
2119 ClearPageChecked(page
);
2123 ClearPageChecked(page
);
2124 set_page_dirty(page
);
2125 btrfs_delalloc_release_extents(BTRFS_I(inode
), PAGE_SIZE
, false);
2127 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2133 extent_changeset_free(data_reserved
);
2137 * There are a few paths in the higher layers of the kernel that directly
2138 * set the page dirty bit without asking the filesystem if it is a
2139 * good idea. This causes problems because we want to make sure COW
2140 * properly happens and the data=ordered rules are followed.
2142 * In our case any range that doesn't have the ORDERED bit set
2143 * hasn't been properly setup for IO. We kick off an async process
2144 * to fix it up. The async helper will wait for ordered extents, set
2145 * the delalloc bit and make it safe to write the page.
2147 int btrfs_writepage_cow_fixup(struct page
*page
, u64 start
, u64 end
)
2149 struct inode
*inode
= page
->mapping
->host
;
2150 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2151 struct btrfs_writepage_fixup
*fixup
;
2153 /* this page is properly in the ordered list */
2154 if (TestClearPagePrivate2(page
))
2157 if (PageChecked(page
))
2160 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
2164 SetPageChecked(page
);
2166 btrfs_init_work(&fixup
->work
, btrfs_fixup_helper
,
2167 btrfs_writepage_fixup_worker
, NULL
, NULL
);
2169 btrfs_queue_work(fs_info
->fixup_workers
, &fixup
->work
);
2173 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
2174 struct inode
*inode
, u64 file_pos
,
2175 u64 disk_bytenr
, u64 disk_num_bytes
,
2176 u64 num_bytes
, u64 ram_bytes
,
2177 u8 compression
, u8 encryption
,
2178 u16 other_encoding
, int extent_type
)
2180 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2181 struct btrfs_file_extent_item
*fi
;
2182 struct btrfs_path
*path
;
2183 struct extent_buffer
*leaf
;
2184 struct btrfs_key ins
;
2186 int extent_inserted
= 0;
2189 path
= btrfs_alloc_path();
2194 * we may be replacing one extent in the tree with another.
2195 * The new extent is pinned in the extent map, and we don't want
2196 * to drop it from the cache until it is completely in the btree.
2198 * So, tell btrfs_drop_extents to leave this extent in the cache.
2199 * the caller is expected to unpin it and allow it to be merged
2202 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, file_pos
,
2203 file_pos
+ num_bytes
, NULL
, 0,
2204 1, sizeof(*fi
), &extent_inserted
);
2208 if (!extent_inserted
) {
2209 ins
.objectid
= btrfs_ino(BTRFS_I(inode
));
2210 ins
.offset
= file_pos
;
2211 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
2213 path
->leave_spinning
= 1;
2214 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
2219 leaf
= path
->nodes
[0];
2220 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2221 struct btrfs_file_extent_item
);
2222 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
2223 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
2224 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
2225 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
2226 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2227 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2228 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
2229 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
2230 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
2231 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
2233 btrfs_mark_buffer_dirty(leaf
);
2234 btrfs_release_path(path
);
2236 inode_add_bytes(inode
, num_bytes
);
2238 ins
.objectid
= disk_bytenr
;
2239 ins
.offset
= disk_num_bytes
;
2240 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2243 * Release the reserved range from inode dirty range map, as it is
2244 * already moved into delayed_ref_head
2246 ret
= btrfs_qgroup_release_data(inode
, file_pos
, ram_bytes
);
2250 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
2251 btrfs_ino(BTRFS_I(inode
)),
2252 file_pos
, qg_released
, &ins
);
2254 btrfs_free_path(path
);
2259 /* snapshot-aware defrag */
2260 struct sa_defrag_extent_backref
{
2261 struct rb_node node
;
2262 struct old_sa_defrag_extent
*old
;
2271 struct old_sa_defrag_extent
{
2272 struct list_head list
;
2273 struct new_sa_defrag_extent
*new;
2282 struct new_sa_defrag_extent
{
2283 struct rb_root root
;
2284 struct list_head head
;
2285 struct btrfs_path
*path
;
2286 struct inode
*inode
;
2294 static int backref_comp(struct sa_defrag_extent_backref
*b1
,
2295 struct sa_defrag_extent_backref
*b2
)
2297 if (b1
->root_id
< b2
->root_id
)
2299 else if (b1
->root_id
> b2
->root_id
)
2302 if (b1
->inum
< b2
->inum
)
2304 else if (b1
->inum
> b2
->inum
)
2307 if (b1
->file_pos
< b2
->file_pos
)
2309 else if (b1
->file_pos
> b2
->file_pos
)
2313 * [------------------------------] ===> (a range of space)
2314 * |<--->| |<---->| =============> (fs/file tree A)
2315 * |<---------------------------->| ===> (fs/file tree B)
2317 * A range of space can refer to two file extents in one tree while
2318 * refer to only one file extent in another tree.
2320 * So we may process a disk offset more than one time(two extents in A)
2321 * and locate at the same extent(one extent in B), then insert two same
2322 * backrefs(both refer to the extent in B).
2327 static void backref_insert(struct rb_root
*root
,
2328 struct sa_defrag_extent_backref
*backref
)
2330 struct rb_node
**p
= &root
->rb_node
;
2331 struct rb_node
*parent
= NULL
;
2332 struct sa_defrag_extent_backref
*entry
;
2337 entry
= rb_entry(parent
, struct sa_defrag_extent_backref
, node
);
2339 ret
= backref_comp(backref
, entry
);
2343 p
= &(*p
)->rb_right
;
2346 rb_link_node(&backref
->node
, parent
, p
);
2347 rb_insert_color(&backref
->node
, root
);
2351 * Note the backref might has changed, and in this case we just return 0.
2353 static noinline
int record_one_backref(u64 inum
, u64 offset
, u64 root_id
,
2356 struct btrfs_file_extent_item
*extent
;
2357 struct old_sa_defrag_extent
*old
= ctx
;
2358 struct new_sa_defrag_extent
*new = old
->new;
2359 struct btrfs_path
*path
= new->path
;
2360 struct btrfs_key key
;
2361 struct btrfs_root
*root
;
2362 struct sa_defrag_extent_backref
*backref
;
2363 struct extent_buffer
*leaf
;
2364 struct inode
*inode
= new->inode
;
2365 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2371 if (BTRFS_I(inode
)->root
->root_key
.objectid
== root_id
&&
2372 inum
== btrfs_ino(BTRFS_I(inode
)))
2375 key
.objectid
= root_id
;
2376 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2377 key
.offset
= (u64
)-1;
2379 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2381 if (PTR_ERR(root
) == -ENOENT
)
2384 btrfs_debug(fs_info
, "inum=%llu, offset=%llu, root_id=%llu",
2385 inum
, offset
, root_id
);
2386 return PTR_ERR(root
);
2389 key
.objectid
= inum
;
2390 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2391 if (offset
> (u64
)-1 << 32)
2394 key
.offset
= offset
;
2396 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2397 if (WARN_ON(ret
< 0))
2404 leaf
= path
->nodes
[0];
2405 slot
= path
->slots
[0];
2407 if (slot
>= btrfs_header_nritems(leaf
)) {
2408 ret
= btrfs_next_leaf(root
, path
);
2411 } else if (ret
> 0) {
2420 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2422 if (key
.objectid
> inum
)
2425 if (key
.objectid
< inum
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2428 extent
= btrfs_item_ptr(leaf
, slot
,
2429 struct btrfs_file_extent_item
);
2431 if (btrfs_file_extent_disk_bytenr(leaf
, extent
) != old
->bytenr
)
2435 * 'offset' refers to the exact key.offset,
2436 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2437 * (key.offset - extent_offset).
2439 if (key
.offset
!= offset
)
2442 extent_offset
= btrfs_file_extent_offset(leaf
, extent
);
2443 num_bytes
= btrfs_file_extent_num_bytes(leaf
, extent
);
2445 if (extent_offset
>= old
->extent_offset
+ old
->offset
+
2446 old
->len
|| extent_offset
+ num_bytes
<=
2447 old
->extent_offset
+ old
->offset
)
2452 backref
= kmalloc(sizeof(*backref
), GFP_NOFS
);
2458 backref
->root_id
= root_id
;
2459 backref
->inum
= inum
;
2460 backref
->file_pos
= offset
;
2461 backref
->num_bytes
= num_bytes
;
2462 backref
->extent_offset
= extent_offset
;
2463 backref
->generation
= btrfs_file_extent_generation(leaf
, extent
);
2465 backref_insert(&new->root
, backref
);
2468 btrfs_release_path(path
);
2473 static noinline
bool record_extent_backrefs(struct btrfs_path
*path
,
2474 struct new_sa_defrag_extent
*new)
2476 struct btrfs_fs_info
*fs_info
= btrfs_sb(new->inode
->i_sb
);
2477 struct old_sa_defrag_extent
*old
, *tmp
;
2482 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2483 ret
= iterate_inodes_from_logical(old
->bytenr
+
2484 old
->extent_offset
, fs_info
,
2485 path
, record_one_backref
,
2487 if (ret
< 0 && ret
!= -ENOENT
)
2490 /* no backref to be processed for this extent */
2492 list_del(&old
->list
);
2497 if (list_empty(&new->head
))
2503 static int relink_is_mergable(struct extent_buffer
*leaf
,
2504 struct btrfs_file_extent_item
*fi
,
2505 struct new_sa_defrag_extent
*new)
2507 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) != new->bytenr
)
2510 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2513 if (btrfs_file_extent_compression(leaf
, fi
) != new->compress_type
)
2516 if (btrfs_file_extent_encryption(leaf
, fi
) ||
2517 btrfs_file_extent_other_encoding(leaf
, fi
))
2524 * Note the backref might has changed, and in this case we just return 0.
2526 static noinline
int relink_extent_backref(struct btrfs_path
*path
,
2527 struct sa_defrag_extent_backref
*prev
,
2528 struct sa_defrag_extent_backref
*backref
)
2530 struct btrfs_file_extent_item
*extent
;
2531 struct btrfs_file_extent_item
*item
;
2532 struct btrfs_ordered_extent
*ordered
;
2533 struct btrfs_trans_handle
*trans
;
2534 struct btrfs_root
*root
;
2535 struct btrfs_key key
;
2536 struct extent_buffer
*leaf
;
2537 struct old_sa_defrag_extent
*old
= backref
->old
;
2538 struct new_sa_defrag_extent
*new = old
->new;
2539 struct btrfs_fs_info
*fs_info
= btrfs_sb(new->inode
->i_sb
);
2540 struct inode
*inode
;
2541 struct extent_state
*cached
= NULL
;
2550 if (prev
&& prev
->root_id
== backref
->root_id
&&
2551 prev
->inum
== backref
->inum
&&
2552 prev
->file_pos
+ prev
->num_bytes
== backref
->file_pos
)
2555 /* step 1: get root */
2556 key
.objectid
= backref
->root_id
;
2557 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2558 key
.offset
= (u64
)-1;
2560 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
2562 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2564 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2565 if (PTR_ERR(root
) == -ENOENT
)
2567 return PTR_ERR(root
);
2570 if (btrfs_root_readonly(root
)) {
2571 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2575 /* step 2: get inode */
2576 key
.objectid
= backref
->inum
;
2577 key
.type
= BTRFS_INODE_ITEM_KEY
;
2580 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
2581 if (IS_ERR(inode
)) {
2582 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2586 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2588 /* step 3: relink backref */
2589 lock_start
= backref
->file_pos
;
2590 lock_end
= backref
->file_pos
+ backref
->num_bytes
- 1;
2591 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2594 ordered
= btrfs_lookup_first_ordered_extent(inode
, lock_end
);
2596 btrfs_put_ordered_extent(ordered
);
2600 trans
= btrfs_join_transaction(root
);
2601 if (IS_ERR(trans
)) {
2602 ret
= PTR_ERR(trans
);
2606 key
.objectid
= backref
->inum
;
2607 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2608 key
.offset
= backref
->file_pos
;
2610 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2613 } else if (ret
> 0) {
2618 extent
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2619 struct btrfs_file_extent_item
);
2621 if (btrfs_file_extent_generation(path
->nodes
[0], extent
) !=
2622 backref
->generation
)
2625 btrfs_release_path(path
);
2627 start
= backref
->file_pos
;
2628 if (backref
->extent_offset
< old
->extent_offset
+ old
->offset
)
2629 start
+= old
->extent_offset
+ old
->offset
-
2630 backref
->extent_offset
;
2632 len
= min(backref
->extent_offset
+ backref
->num_bytes
,
2633 old
->extent_offset
+ old
->offset
+ old
->len
);
2634 len
-= max(backref
->extent_offset
, old
->extent_offset
+ old
->offset
);
2636 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
2641 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
2642 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2645 path
->leave_spinning
= 1;
2647 struct btrfs_file_extent_item
*fi
;
2649 struct btrfs_key found_key
;
2651 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2656 leaf
= path
->nodes
[0];
2657 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2659 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2660 struct btrfs_file_extent_item
);
2661 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
2663 if (extent_len
+ found_key
.offset
== start
&&
2664 relink_is_mergable(leaf
, fi
, new)) {
2665 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2667 btrfs_mark_buffer_dirty(leaf
);
2668 inode_add_bytes(inode
, len
);
2674 btrfs_release_path(path
);
2679 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2682 btrfs_abort_transaction(trans
, ret
);
2686 leaf
= path
->nodes
[0];
2687 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2688 struct btrfs_file_extent_item
);
2689 btrfs_set_file_extent_disk_bytenr(leaf
, item
, new->bytenr
);
2690 btrfs_set_file_extent_disk_num_bytes(leaf
, item
, new->disk_len
);
2691 btrfs_set_file_extent_offset(leaf
, item
, start
- new->file_pos
);
2692 btrfs_set_file_extent_num_bytes(leaf
, item
, len
);
2693 btrfs_set_file_extent_ram_bytes(leaf
, item
, new->len
);
2694 btrfs_set_file_extent_generation(leaf
, item
, trans
->transid
);
2695 btrfs_set_file_extent_type(leaf
, item
, BTRFS_FILE_EXTENT_REG
);
2696 btrfs_set_file_extent_compression(leaf
, item
, new->compress_type
);
2697 btrfs_set_file_extent_encryption(leaf
, item
, 0);
2698 btrfs_set_file_extent_other_encoding(leaf
, item
, 0);
2700 btrfs_mark_buffer_dirty(leaf
);
2701 inode_add_bytes(inode
, len
);
2702 btrfs_release_path(path
);
2704 ret
= btrfs_inc_extent_ref(trans
, root
, new->bytenr
,
2706 backref
->root_id
, backref
->inum
,
2707 new->file_pos
); /* start - extent_offset */
2709 btrfs_abort_transaction(trans
, ret
);
2715 btrfs_release_path(path
);
2716 path
->leave_spinning
= 0;
2717 btrfs_end_transaction(trans
);
2719 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2725 static void free_sa_defrag_extent(struct new_sa_defrag_extent
*new)
2727 struct old_sa_defrag_extent
*old
, *tmp
;
2732 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2738 static void relink_file_extents(struct new_sa_defrag_extent
*new)
2740 struct btrfs_fs_info
*fs_info
= btrfs_sb(new->inode
->i_sb
);
2741 struct btrfs_path
*path
;
2742 struct sa_defrag_extent_backref
*backref
;
2743 struct sa_defrag_extent_backref
*prev
= NULL
;
2744 struct rb_node
*node
;
2747 path
= btrfs_alloc_path();
2751 if (!record_extent_backrefs(path
, new)) {
2752 btrfs_free_path(path
);
2755 btrfs_release_path(path
);
2758 node
= rb_first(&new->root
);
2761 rb_erase(node
, &new->root
);
2763 backref
= rb_entry(node
, struct sa_defrag_extent_backref
, node
);
2765 ret
= relink_extent_backref(path
, prev
, backref
);
2778 btrfs_free_path(path
);
2780 free_sa_defrag_extent(new);
2782 atomic_dec(&fs_info
->defrag_running
);
2783 wake_up(&fs_info
->transaction_wait
);
2786 static struct new_sa_defrag_extent
*
2787 record_old_file_extents(struct inode
*inode
,
2788 struct btrfs_ordered_extent
*ordered
)
2790 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2791 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2792 struct btrfs_path
*path
;
2793 struct btrfs_key key
;
2794 struct old_sa_defrag_extent
*old
;
2795 struct new_sa_defrag_extent
*new;
2798 new = kmalloc(sizeof(*new), GFP_NOFS
);
2803 new->file_pos
= ordered
->file_offset
;
2804 new->len
= ordered
->len
;
2805 new->bytenr
= ordered
->start
;
2806 new->disk_len
= ordered
->disk_len
;
2807 new->compress_type
= ordered
->compress_type
;
2808 new->root
= RB_ROOT
;
2809 INIT_LIST_HEAD(&new->head
);
2811 path
= btrfs_alloc_path();
2815 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
2816 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2817 key
.offset
= new->file_pos
;
2819 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2822 if (ret
> 0 && path
->slots
[0] > 0)
2825 /* find out all the old extents for the file range */
2827 struct btrfs_file_extent_item
*extent
;
2828 struct extent_buffer
*l
;
2837 slot
= path
->slots
[0];
2839 if (slot
>= btrfs_header_nritems(l
)) {
2840 ret
= btrfs_next_leaf(root
, path
);
2848 btrfs_item_key_to_cpu(l
, &key
, slot
);
2850 if (key
.objectid
!= btrfs_ino(BTRFS_I(inode
)))
2852 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2854 if (key
.offset
>= new->file_pos
+ new->len
)
2857 extent
= btrfs_item_ptr(l
, slot
, struct btrfs_file_extent_item
);
2859 num_bytes
= btrfs_file_extent_num_bytes(l
, extent
);
2860 if (key
.offset
+ num_bytes
< new->file_pos
)
2863 disk_bytenr
= btrfs_file_extent_disk_bytenr(l
, extent
);
2867 extent_offset
= btrfs_file_extent_offset(l
, extent
);
2869 old
= kmalloc(sizeof(*old
), GFP_NOFS
);
2873 offset
= max(new->file_pos
, key
.offset
);
2874 end
= min(new->file_pos
+ new->len
, key
.offset
+ num_bytes
);
2876 old
->bytenr
= disk_bytenr
;
2877 old
->extent_offset
= extent_offset
;
2878 old
->offset
= offset
- key
.offset
;
2879 old
->len
= end
- offset
;
2882 list_add_tail(&old
->list
, &new->head
);
2888 btrfs_free_path(path
);
2889 atomic_inc(&fs_info
->defrag_running
);
2894 btrfs_free_path(path
);
2896 free_sa_defrag_extent(new);
2900 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info
*fs_info
,
2903 struct btrfs_block_group_cache
*cache
;
2905 cache
= btrfs_lookup_block_group(fs_info
, start
);
2908 spin_lock(&cache
->lock
);
2909 cache
->delalloc_bytes
-= len
;
2910 spin_unlock(&cache
->lock
);
2912 btrfs_put_block_group(cache
);
2915 /* as ordered data IO finishes, this gets called so we can finish
2916 * an ordered extent if the range of bytes in the file it covers are
2919 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
2921 struct inode
*inode
= ordered_extent
->inode
;
2922 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2923 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2924 struct btrfs_trans_handle
*trans
= NULL
;
2925 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2926 struct extent_state
*cached_state
= NULL
;
2927 struct new_sa_defrag_extent
*new = NULL
;
2928 int compress_type
= 0;
2930 u64 logical_len
= ordered_extent
->len
;
2932 bool truncated
= false;
2933 bool range_locked
= false;
2934 bool clear_new_delalloc_bytes
= false;
2935 bool clear_reserved_extent
= true;
2937 if (!test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
2938 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
) &&
2939 !test_bit(BTRFS_ORDERED_DIRECT
, &ordered_extent
->flags
))
2940 clear_new_delalloc_bytes
= true;
2942 nolock
= btrfs_is_free_space_inode(BTRFS_I(inode
));
2944 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
2949 btrfs_free_io_failure_record(BTRFS_I(inode
),
2950 ordered_extent
->file_offset
,
2951 ordered_extent
->file_offset
+
2952 ordered_extent
->len
- 1);
2954 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered_extent
->flags
)) {
2956 logical_len
= ordered_extent
->truncated_len
;
2957 /* Truncated the entire extent, don't bother adding */
2962 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
2963 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
2966 * For mwrite(mmap + memset to write) case, we still reserve
2967 * space for NOCOW range.
2968 * As NOCOW won't cause a new delayed ref, just free the space
2970 btrfs_qgroup_free_data(inode
, NULL
, ordered_extent
->file_offset
,
2971 ordered_extent
->len
);
2972 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2974 trans
= btrfs_join_transaction_nolock(root
);
2976 trans
= btrfs_join_transaction(root
);
2977 if (IS_ERR(trans
)) {
2978 ret
= PTR_ERR(trans
);
2982 trans
->block_rsv
= &BTRFS_I(inode
)->block_rsv
;
2983 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2984 if (ret
) /* -ENOMEM or corruption */
2985 btrfs_abort_transaction(trans
, ret
);
2989 range_locked
= true;
2990 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
2991 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2994 ret
= test_range_bit(io_tree
, ordered_extent
->file_offset
,
2995 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2996 EXTENT_DEFRAG
, 0, cached_state
);
2998 u64 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
2999 if (0 && last_snapshot
>= BTRFS_I(inode
)->generation
)
3000 /* the inode is shared */
3001 new = record_old_file_extents(inode
, ordered_extent
);
3003 clear_extent_bit(io_tree
, ordered_extent
->file_offset
,
3004 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
3005 EXTENT_DEFRAG
, 0, 0, &cached_state
);
3009 trans
= btrfs_join_transaction_nolock(root
);
3011 trans
= btrfs_join_transaction(root
);
3012 if (IS_ERR(trans
)) {
3013 ret
= PTR_ERR(trans
);
3018 trans
->block_rsv
= &BTRFS_I(inode
)->block_rsv
;
3020 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
3021 compress_type
= ordered_extent
->compress_type
;
3022 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
3023 BUG_ON(compress_type
);
3024 btrfs_qgroup_free_data(inode
, NULL
, ordered_extent
->file_offset
,
3025 ordered_extent
->len
);
3026 ret
= btrfs_mark_extent_written(trans
, BTRFS_I(inode
),
3027 ordered_extent
->file_offset
,
3028 ordered_extent
->file_offset
+
3031 BUG_ON(root
== fs_info
->tree_root
);
3032 ret
= insert_reserved_file_extent(trans
, inode
,
3033 ordered_extent
->file_offset
,
3034 ordered_extent
->start
,
3035 ordered_extent
->disk_len
,
3036 logical_len
, logical_len
,
3037 compress_type
, 0, 0,
3038 BTRFS_FILE_EXTENT_REG
);
3040 clear_reserved_extent
= false;
3041 btrfs_release_delalloc_bytes(fs_info
,
3042 ordered_extent
->start
,
3043 ordered_extent
->disk_len
);
3046 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
3047 ordered_extent
->file_offset
, ordered_extent
->len
,
3050 btrfs_abort_transaction(trans
, ret
);
3054 ret
= add_pending_csums(trans
, inode
, &ordered_extent
->list
);
3056 btrfs_abort_transaction(trans
, ret
);
3060 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
3061 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
3062 if (ret
) { /* -ENOMEM or corruption */
3063 btrfs_abort_transaction(trans
, ret
);
3068 if (range_locked
|| clear_new_delalloc_bytes
) {
3069 unsigned int clear_bits
= 0;
3072 clear_bits
|= EXTENT_LOCKED
;
3073 if (clear_new_delalloc_bytes
)
3074 clear_bits
|= EXTENT_DELALLOC_NEW
;
3075 clear_extent_bit(&BTRFS_I(inode
)->io_tree
,
3076 ordered_extent
->file_offset
,
3077 ordered_extent
->file_offset
+
3078 ordered_extent
->len
- 1,
3080 (clear_bits
& EXTENT_LOCKED
) ? 1 : 0,
3085 btrfs_end_transaction(trans
);
3087 if (ret
|| truncated
) {
3091 start
= ordered_extent
->file_offset
+ logical_len
;
3093 start
= ordered_extent
->file_offset
;
3094 end
= ordered_extent
->file_offset
+ ordered_extent
->len
- 1;
3095 clear_extent_uptodate(io_tree
, start
, end
, NULL
);
3097 /* Drop the cache for the part of the extent we didn't write. */
3098 btrfs_drop_extent_cache(BTRFS_I(inode
), start
, end
, 0);
3101 * If the ordered extent had an IOERR or something else went
3102 * wrong we need to return the space for this ordered extent
3103 * back to the allocator. We only free the extent in the
3104 * truncated case if we didn't write out the extent at all.
3106 * If we made it past insert_reserved_file_extent before we
3107 * errored out then we don't need to do this as the accounting
3108 * has already been done.
3110 if ((ret
|| !logical_len
) &&
3111 clear_reserved_extent
&&
3112 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
3113 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
))
3114 btrfs_free_reserved_extent(fs_info
,
3115 ordered_extent
->start
,
3116 ordered_extent
->disk_len
, 1);
3121 * This needs to be done to make sure anybody waiting knows we are done
3122 * updating everything for this ordered extent.
3124 btrfs_remove_ordered_extent(inode
, ordered_extent
);
3126 /* for snapshot-aware defrag */
3129 free_sa_defrag_extent(new);
3130 atomic_dec(&fs_info
->defrag_running
);
3132 relink_file_extents(new);
3137 btrfs_put_ordered_extent(ordered_extent
);
3138 /* once for the tree */
3139 btrfs_put_ordered_extent(ordered_extent
);
3144 static void finish_ordered_fn(struct btrfs_work
*work
)
3146 struct btrfs_ordered_extent
*ordered_extent
;
3147 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
3148 btrfs_finish_ordered_io(ordered_extent
);
3151 void btrfs_writepage_endio_finish_ordered(struct page
*page
, u64 start
,
3152 u64 end
, int uptodate
)
3154 struct inode
*inode
= page
->mapping
->host
;
3155 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3156 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
3157 struct btrfs_workqueue
*wq
;
3158 btrfs_work_func_t func
;
3160 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
3162 ClearPagePrivate2(page
);
3163 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
3164 end
- start
+ 1, uptodate
))
3167 if (btrfs_is_free_space_inode(BTRFS_I(inode
))) {
3168 wq
= fs_info
->endio_freespace_worker
;
3169 func
= btrfs_freespace_write_helper
;
3171 wq
= fs_info
->endio_write_workers
;
3172 func
= btrfs_endio_write_helper
;
3175 btrfs_init_work(&ordered_extent
->work
, func
, finish_ordered_fn
, NULL
,
3177 btrfs_queue_work(wq
, &ordered_extent
->work
);
3180 static int __readpage_endio_check(struct inode
*inode
,
3181 struct btrfs_io_bio
*io_bio
,
3182 int icsum
, struct page
*page
,
3183 int pgoff
, u64 start
, size_t len
)
3189 csum_expected
= *(((u32
*)io_bio
->csum
) + icsum
);
3191 kaddr
= kmap_atomic(page
);
3192 csum
= btrfs_csum_data(kaddr
+ pgoff
, csum
, len
);
3193 btrfs_csum_final(csum
, (u8
*)&csum
);
3194 if (csum
!= csum_expected
)
3197 kunmap_atomic(kaddr
);
3200 btrfs_print_data_csum_error(BTRFS_I(inode
), start
, csum
, csum_expected
,
3201 io_bio
->mirror_num
);
3202 memset(kaddr
+ pgoff
, 1, len
);
3203 flush_dcache_page(page
);
3204 kunmap_atomic(kaddr
);
3209 * when reads are done, we need to check csums to verify the data is correct
3210 * if there's a match, we allow the bio to finish. If not, the code in
3211 * extent_io.c will try to find good copies for us.
3213 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
3214 u64 phy_offset
, struct page
*page
,
3215 u64 start
, u64 end
, int mirror
)
3217 size_t offset
= start
- page_offset(page
);
3218 struct inode
*inode
= page
->mapping
->host
;
3219 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3220 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3222 if (PageChecked(page
)) {
3223 ClearPageChecked(page
);
3227 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
3230 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
3231 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
3232 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
);
3236 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
3237 return __readpage_endio_check(inode
, io_bio
, phy_offset
, page
, offset
,
3238 start
, (size_t)(end
- start
+ 1));
3242 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3244 * @inode: The inode we want to perform iput on
3246 * This function uses the generic vfs_inode::i_count to track whether we should
3247 * just decrement it (in case it's > 1) or if this is the last iput then link
3248 * the inode to the delayed iput machinery. Delayed iputs are processed at
3249 * transaction commit time/superblock commit/cleaner kthread.
3251 void btrfs_add_delayed_iput(struct inode
*inode
)
3253 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3254 struct btrfs_inode
*binode
= BTRFS_I(inode
);
3256 if (atomic_add_unless(&inode
->i_count
, -1, 1))
3259 atomic_inc(&fs_info
->nr_delayed_iputs
);
3260 spin_lock(&fs_info
->delayed_iput_lock
);
3261 ASSERT(list_empty(&binode
->delayed_iput
));
3262 list_add_tail(&binode
->delayed_iput
, &fs_info
->delayed_iputs
);
3263 spin_unlock(&fs_info
->delayed_iput_lock
);
3264 if (!test_bit(BTRFS_FS_CLEANER_RUNNING
, &fs_info
->flags
))
3265 wake_up_process(fs_info
->cleaner_kthread
);
3268 void btrfs_run_delayed_iputs(struct btrfs_fs_info
*fs_info
)
3271 spin_lock(&fs_info
->delayed_iput_lock
);
3272 while (!list_empty(&fs_info
->delayed_iputs
)) {
3273 struct btrfs_inode
*inode
;
3275 inode
= list_first_entry(&fs_info
->delayed_iputs
,
3276 struct btrfs_inode
, delayed_iput
);
3277 list_del_init(&inode
->delayed_iput
);
3278 spin_unlock(&fs_info
->delayed_iput_lock
);
3279 iput(&inode
->vfs_inode
);
3280 if (atomic_dec_and_test(&fs_info
->nr_delayed_iputs
))
3281 wake_up(&fs_info
->delayed_iputs_wait
);
3282 spin_lock(&fs_info
->delayed_iput_lock
);
3284 spin_unlock(&fs_info
->delayed_iput_lock
);
3288 * btrfs_wait_on_delayed_iputs - wait on the delayed iputs to be done running
3289 * @fs_info - the fs_info for this fs
3290 * @return - EINTR if we were killed, 0 if nothing's pending
3292 * This will wait on any delayed iputs that are currently running with KILLABLE
3293 * set. Once they are all done running we will return, unless we are killed in
3294 * which case we return EINTR. This helps in user operations like fallocate etc
3295 * that might get blocked on the iputs.
3297 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info
*fs_info
)
3299 int ret
= wait_event_killable(fs_info
->delayed_iputs_wait
,
3300 atomic_read(&fs_info
->nr_delayed_iputs
) == 0);
3307 * This creates an orphan entry for the given inode in case something goes wrong
3308 * in the middle of an unlink.
3310 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
,
3311 struct btrfs_inode
*inode
)
3315 ret
= btrfs_insert_orphan_item(trans
, inode
->root
, btrfs_ino(inode
));
3316 if (ret
&& ret
!= -EEXIST
) {
3317 btrfs_abort_transaction(trans
, ret
);
3325 * We have done the delete so we can go ahead and remove the orphan item for
3326 * this particular inode.
3328 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3329 struct btrfs_inode
*inode
)
3331 return btrfs_del_orphan_item(trans
, inode
->root
, btrfs_ino(inode
));
3335 * this cleans up any orphans that may be left on the list from the last use
3338 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3340 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3341 struct btrfs_path
*path
;
3342 struct extent_buffer
*leaf
;
3343 struct btrfs_key key
, found_key
;
3344 struct btrfs_trans_handle
*trans
;
3345 struct inode
*inode
;
3346 u64 last_objectid
= 0;
3347 int ret
= 0, nr_unlink
= 0;
3349 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
3352 path
= btrfs_alloc_path();
3357 path
->reada
= READA_BACK
;
3359 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3360 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
3361 key
.offset
= (u64
)-1;
3364 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3369 * if ret == 0 means we found what we were searching for, which
3370 * is weird, but possible, so only screw with path if we didn't
3371 * find the key and see if we have stuff that matches
3375 if (path
->slots
[0] == 0)
3380 /* pull out the item */
3381 leaf
= path
->nodes
[0];
3382 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3384 /* make sure the item matches what we want */
3385 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3387 if (found_key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
3390 /* release the path since we're done with it */
3391 btrfs_release_path(path
);
3394 * this is where we are basically btrfs_lookup, without the
3395 * crossing root thing. we store the inode number in the
3396 * offset of the orphan item.
3399 if (found_key
.offset
== last_objectid
) {
3401 "Error removing orphan entry, stopping orphan cleanup");
3406 last_objectid
= found_key
.offset
;
3408 found_key
.objectid
= found_key
.offset
;
3409 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3410 found_key
.offset
= 0;
3411 inode
= btrfs_iget(fs_info
->sb
, &found_key
, root
, NULL
);
3412 ret
= PTR_ERR_OR_ZERO(inode
);
3413 if (ret
&& ret
!= -ENOENT
)
3416 if (ret
== -ENOENT
&& root
== fs_info
->tree_root
) {
3417 struct btrfs_root
*dead_root
;
3418 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3419 int is_dead_root
= 0;
3422 * this is an orphan in the tree root. Currently these
3423 * could come from 2 sources:
3424 * a) a snapshot deletion in progress
3425 * b) a free space cache inode
3426 * We need to distinguish those two, as the snapshot
3427 * orphan must not get deleted.
3428 * find_dead_roots already ran before us, so if this
3429 * is a snapshot deletion, we should find the root
3430 * in the dead_roots list
3432 spin_lock(&fs_info
->trans_lock
);
3433 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
3435 if (dead_root
->root_key
.objectid
==
3436 found_key
.objectid
) {
3441 spin_unlock(&fs_info
->trans_lock
);
3443 /* prevent this orphan from being found again */
3444 key
.offset
= found_key
.objectid
- 1;
3451 * If we have an inode with links, there are a couple of
3452 * possibilities. Old kernels (before v3.12) used to create an
3453 * orphan item for truncate indicating that there were possibly
3454 * extent items past i_size that needed to be deleted. In v3.12,
3455 * truncate was changed to update i_size in sync with the extent
3456 * items, but the (useless) orphan item was still created. Since
3457 * v4.18, we don't create the orphan item for truncate at all.
3459 * So, this item could mean that we need to do a truncate, but
3460 * only if this filesystem was last used on a pre-v3.12 kernel
3461 * and was not cleanly unmounted. The odds of that are quite
3462 * slim, and it's a pain to do the truncate now, so just delete
3465 * It's also possible that this orphan item was supposed to be
3466 * deleted but wasn't. The inode number may have been reused,
3467 * but either way, we can delete the orphan item.
3469 if (ret
== -ENOENT
|| inode
->i_nlink
) {
3472 trans
= btrfs_start_transaction(root
, 1);
3473 if (IS_ERR(trans
)) {
3474 ret
= PTR_ERR(trans
);
3477 btrfs_debug(fs_info
, "auto deleting %Lu",
3478 found_key
.objectid
);
3479 ret
= btrfs_del_orphan_item(trans
, root
,
3480 found_key
.objectid
);
3481 btrfs_end_transaction(trans
);
3489 /* this will do delete_inode and everything for us */
3492 /* release the path since we're done with it */
3493 btrfs_release_path(path
);
3495 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
3497 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
)) {
3498 trans
= btrfs_join_transaction(root
);
3500 btrfs_end_transaction(trans
);
3504 btrfs_debug(fs_info
, "unlinked %d orphans", nr_unlink
);
3508 btrfs_err(fs_info
, "could not do orphan cleanup %d", ret
);
3509 btrfs_free_path(path
);
3514 * very simple check to peek ahead in the leaf looking for xattrs. If we
3515 * don't find any xattrs, we know there can't be any acls.
3517 * slot is the slot the inode is in, objectid is the objectid of the inode
3519 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3520 int slot
, u64 objectid
,
3521 int *first_xattr_slot
)
3523 u32 nritems
= btrfs_header_nritems(leaf
);
3524 struct btrfs_key found_key
;
3525 static u64 xattr_access
= 0;
3526 static u64 xattr_default
= 0;
3529 if (!xattr_access
) {
3530 xattr_access
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS
,
3531 strlen(XATTR_NAME_POSIX_ACL_ACCESS
));
3532 xattr_default
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT
,
3533 strlen(XATTR_NAME_POSIX_ACL_DEFAULT
));
3537 *first_xattr_slot
= -1;
3538 while (slot
< nritems
) {
3539 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3541 /* we found a different objectid, there must not be acls */
3542 if (found_key
.objectid
!= objectid
)
3545 /* we found an xattr, assume we've got an acl */
3546 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
3547 if (*first_xattr_slot
== -1)
3548 *first_xattr_slot
= slot
;
3549 if (found_key
.offset
== xattr_access
||
3550 found_key
.offset
== xattr_default
)
3555 * we found a key greater than an xattr key, there can't
3556 * be any acls later on
3558 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3565 * it goes inode, inode backrefs, xattrs, extents,
3566 * so if there are a ton of hard links to an inode there can
3567 * be a lot of backrefs. Don't waste time searching too hard,
3568 * this is just an optimization
3573 /* we hit the end of the leaf before we found an xattr or
3574 * something larger than an xattr. We have to assume the inode
3577 if (*first_xattr_slot
== -1)
3578 *first_xattr_slot
= slot
;
3583 * read an inode from the btree into the in-memory inode
3585 static int btrfs_read_locked_inode(struct inode
*inode
,
3586 struct btrfs_path
*in_path
)
3588 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3589 struct btrfs_path
*path
= in_path
;
3590 struct extent_buffer
*leaf
;
3591 struct btrfs_inode_item
*inode_item
;
3592 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3593 struct btrfs_key location
;
3598 bool filled
= false;
3599 int first_xattr_slot
;
3601 ret
= btrfs_fill_inode(inode
, &rdev
);
3606 path
= btrfs_alloc_path();
3611 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3613 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3615 if (path
!= in_path
)
3616 btrfs_free_path(path
);
3620 leaf
= path
->nodes
[0];
3625 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3626 struct btrfs_inode_item
);
3627 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3628 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3629 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3630 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3631 btrfs_i_size_write(BTRFS_I(inode
), btrfs_inode_size(leaf
, inode_item
));
3633 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->atime
);
3634 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->atime
);
3636 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->mtime
);
3637 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->mtime
);
3639 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->ctime
);
3640 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->ctime
);
3642 BTRFS_I(inode
)->i_otime
.tv_sec
=
3643 btrfs_timespec_sec(leaf
, &inode_item
->otime
);
3644 BTRFS_I(inode
)->i_otime
.tv_nsec
=
3645 btrfs_timespec_nsec(leaf
, &inode_item
->otime
);
3647 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3648 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3649 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3651 inode_set_iversion_queried(inode
,
3652 btrfs_inode_sequence(leaf
, inode_item
));
3653 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3655 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3657 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3658 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
3662 * If we were modified in the current generation and evicted from memory
3663 * and then re-read we need to do a full sync since we don't have any
3664 * idea about which extents were modified before we were evicted from
3667 * This is required for both inode re-read from disk and delayed inode
3668 * in delayed_nodes_tree.
3670 if (BTRFS_I(inode
)->last_trans
== fs_info
->generation
)
3671 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3672 &BTRFS_I(inode
)->runtime_flags
);
3675 * We don't persist the id of the transaction where an unlink operation
3676 * against the inode was last made. So here we assume the inode might
3677 * have been evicted, and therefore the exact value of last_unlink_trans
3678 * lost, and set it to last_trans to avoid metadata inconsistencies
3679 * between the inode and its parent if the inode is fsync'ed and the log
3680 * replayed. For example, in the scenario:
3683 * ln mydir/foo mydir/bar
3686 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3687 * xfs_io -c fsync mydir/foo
3689 * mount fs, triggers fsync log replay
3691 * We must make sure that when we fsync our inode foo we also log its
3692 * parent inode, otherwise after log replay the parent still has the
3693 * dentry with the "bar" name but our inode foo has a link count of 1
3694 * and doesn't have an inode ref with the name "bar" anymore.
3696 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3697 * but it guarantees correctness at the expense of occasional full
3698 * transaction commits on fsync if our inode is a directory, or if our
3699 * inode is not a directory, logging its parent unnecessarily.
3701 BTRFS_I(inode
)->last_unlink_trans
= BTRFS_I(inode
)->last_trans
;
3703 * Similar reasoning for last_link_trans, needs to be set otherwise
3704 * for a case like the following:
3709 * echo 2 > /proc/sys/vm/drop_caches
3713 * Would result in link bar and directory A not existing after the power
3716 BTRFS_I(inode
)->last_link_trans
= BTRFS_I(inode
)->last_trans
;
3719 if (inode
->i_nlink
!= 1 ||
3720 path
->slots
[0] >= btrfs_header_nritems(leaf
))
3723 btrfs_item_key_to_cpu(leaf
, &location
, path
->slots
[0]);
3724 if (location
.objectid
!= btrfs_ino(BTRFS_I(inode
)))
3727 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3728 if (location
.type
== BTRFS_INODE_REF_KEY
) {
3729 struct btrfs_inode_ref
*ref
;
3731 ref
= (struct btrfs_inode_ref
*)ptr
;
3732 BTRFS_I(inode
)->dir_index
= btrfs_inode_ref_index(leaf
, ref
);
3733 } else if (location
.type
== BTRFS_INODE_EXTREF_KEY
) {
3734 struct btrfs_inode_extref
*extref
;
3736 extref
= (struct btrfs_inode_extref
*)ptr
;
3737 BTRFS_I(inode
)->dir_index
= btrfs_inode_extref_index(leaf
,
3742 * try to precache a NULL acl entry for files that don't have
3743 * any xattrs or acls
3745 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
3746 btrfs_ino(BTRFS_I(inode
)), &first_xattr_slot
);
3747 if (first_xattr_slot
!= -1) {
3748 path
->slots
[0] = first_xattr_slot
;
3749 ret
= btrfs_load_inode_props(inode
, path
);
3752 "error loading props for ino %llu (root %llu): %d",
3753 btrfs_ino(BTRFS_I(inode
)),
3754 root
->root_key
.objectid
, ret
);
3756 if (path
!= in_path
)
3757 btrfs_free_path(path
);
3760 cache_no_acl(inode
);
3762 switch (inode
->i_mode
& S_IFMT
) {
3764 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3765 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3766 inode
->i_fop
= &btrfs_file_operations
;
3767 inode
->i_op
= &btrfs_file_inode_operations
;
3770 inode
->i_fop
= &btrfs_dir_file_operations
;
3771 inode
->i_op
= &btrfs_dir_inode_operations
;
3774 inode
->i_op
= &btrfs_symlink_inode_operations
;
3775 inode_nohighmem(inode
);
3776 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3779 inode
->i_op
= &btrfs_special_inode_operations
;
3780 init_special_inode(inode
, inode
->i_mode
, rdev
);
3784 btrfs_sync_inode_flags_to_i_flags(inode
);
3789 * given a leaf and an inode, copy the inode fields into the leaf
3791 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3792 struct extent_buffer
*leaf
,
3793 struct btrfs_inode_item
*item
,
3794 struct inode
*inode
)
3796 struct btrfs_map_token token
;
3798 btrfs_init_map_token(&token
);
3800 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3801 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3802 btrfs_set_token_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
,
3804 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3805 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3807 btrfs_set_token_timespec_sec(leaf
, &item
->atime
,
3808 inode
->i_atime
.tv_sec
, &token
);
3809 btrfs_set_token_timespec_nsec(leaf
, &item
->atime
,
3810 inode
->i_atime
.tv_nsec
, &token
);
3812 btrfs_set_token_timespec_sec(leaf
, &item
->mtime
,
3813 inode
->i_mtime
.tv_sec
, &token
);
3814 btrfs_set_token_timespec_nsec(leaf
, &item
->mtime
,
3815 inode
->i_mtime
.tv_nsec
, &token
);
3817 btrfs_set_token_timespec_sec(leaf
, &item
->ctime
,
3818 inode
->i_ctime
.tv_sec
, &token
);
3819 btrfs_set_token_timespec_nsec(leaf
, &item
->ctime
,
3820 inode
->i_ctime
.tv_nsec
, &token
);
3822 btrfs_set_token_timespec_sec(leaf
, &item
->otime
,
3823 BTRFS_I(inode
)->i_otime
.tv_sec
, &token
);
3824 btrfs_set_token_timespec_nsec(leaf
, &item
->otime
,
3825 BTRFS_I(inode
)->i_otime
.tv_nsec
, &token
);
3827 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3829 btrfs_set_token_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
,
3831 btrfs_set_token_inode_sequence(leaf
, item
, inode_peek_iversion(inode
),
3833 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3834 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3835 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3836 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3840 * copy everything in the in-memory inode into the btree.
3842 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
3843 struct btrfs_root
*root
, struct inode
*inode
)
3845 struct btrfs_inode_item
*inode_item
;
3846 struct btrfs_path
*path
;
3847 struct extent_buffer
*leaf
;
3850 path
= btrfs_alloc_path();
3854 path
->leave_spinning
= 1;
3855 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
3863 leaf
= path
->nodes
[0];
3864 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3865 struct btrfs_inode_item
);
3867 fill_inode_item(trans
, leaf
, inode_item
, inode
);
3868 btrfs_mark_buffer_dirty(leaf
);
3869 btrfs_set_inode_last_trans(trans
, inode
);
3872 btrfs_free_path(path
);
3877 * copy everything in the in-memory inode into the btree.
3879 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
3880 struct btrfs_root
*root
, struct inode
*inode
)
3882 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3886 * If the inode is a free space inode, we can deadlock during commit
3887 * if we put it into the delayed code.
3889 * The data relocation inode should also be directly updated
3892 if (!btrfs_is_free_space_inode(BTRFS_I(inode
))
3893 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
3894 && !test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
)) {
3895 btrfs_update_root_times(trans
, root
);
3897 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
3899 btrfs_set_inode_last_trans(trans
, inode
);
3903 return btrfs_update_inode_item(trans
, root
, inode
);
3906 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
3907 struct btrfs_root
*root
,
3908 struct inode
*inode
)
3912 ret
= btrfs_update_inode(trans
, root
, inode
);
3914 return btrfs_update_inode_item(trans
, root
, inode
);
3919 * unlink helper that gets used here in inode.c and in the tree logging
3920 * recovery code. It remove a link in a directory with a given name, and
3921 * also drops the back refs in the inode to the directory
3923 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3924 struct btrfs_root
*root
,
3925 struct btrfs_inode
*dir
,
3926 struct btrfs_inode
*inode
,
3927 const char *name
, int name_len
)
3929 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3930 struct btrfs_path
*path
;
3932 struct extent_buffer
*leaf
;
3933 struct btrfs_dir_item
*di
;
3934 struct btrfs_key key
;
3936 u64 ino
= btrfs_ino(inode
);
3937 u64 dir_ino
= btrfs_ino(dir
);
3939 path
= btrfs_alloc_path();
3945 path
->leave_spinning
= 1;
3946 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3947 name
, name_len
, -1);
3948 if (IS_ERR_OR_NULL(di
)) {
3949 ret
= di
? PTR_ERR(di
) : -ENOENT
;
3952 leaf
= path
->nodes
[0];
3953 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3954 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3957 btrfs_release_path(path
);
3960 * If we don't have dir index, we have to get it by looking up
3961 * the inode ref, since we get the inode ref, remove it directly,
3962 * it is unnecessary to do delayed deletion.
3964 * But if we have dir index, needn't search inode ref to get it.
3965 * Since the inode ref is close to the inode item, it is better
3966 * that we delay to delete it, and just do this deletion when
3967 * we update the inode item.
3969 if (inode
->dir_index
) {
3970 ret
= btrfs_delayed_delete_inode_ref(inode
);
3972 index
= inode
->dir_index
;
3977 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
3981 "failed to delete reference to %.*s, inode %llu parent %llu",
3982 name_len
, name
, ino
, dir_ino
);
3983 btrfs_abort_transaction(trans
, ret
);
3987 ret
= btrfs_delete_delayed_dir_index(trans
, dir
, index
);
3989 btrfs_abort_transaction(trans
, ret
);
3993 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
, inode
,
3995 if (ret
!= 0 && ret
!= -ENOENT
) {
3996 btrfs_abort_transaction(trans
, ret
);
4000 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
, dir
,
4005 btrfs_abort_transaction(trans
, ret
);
4007 btrfs_free_path(path
);
4011 btrfs_i_size_write(dir
, dir
->vfs_inode
.i_size
- name_len
* 2);
4012 inode_inc_iversion(&inode
->vfs_inode
);
4013 inode_inc_iversion(&dir
->vfs_inode
);
4014 inode
->vfs_inode
.i_ctime
= dir
->vfs_inode
.i_mtime
=
4015 dir
->vfs_inode
.i_ctime
= current_time(&inode
->vfs_inode
);
4016 ret
= btrfs_update_inode(trans
, root
, &dir
->vfs_inode
);
4021 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4022 struct btrfs_root
*root
,
4023 struct btrfs_inode
*dir
, struct btrfs_inode
*inode
,
4024 const char *name
, int name_len
)
4027 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
4029 drop_nlink(&inode
->vfs_inode
);
4030 ret
= btrfs_update_inode(trans
, root
, &inode
->vfs_inode
);
4036 * helper to start transaction for unlink and rmdir.
4038 * unlink and rmdir are special in btrfs, they do not always free space, so
4039 * if we cannot make our reservations the normal way try and see if there is
4040 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4041 * allow the unlink to occur.
4043 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
)
4045 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4048 * 1 for the possible orphan item
4049 * 1 for the dir item
4050 * 1 for the dir index
4051 * 1 for the inode ref
4054 return btrfs_start_transaction_fallback_global_rsv(root
, 5, 5);
4057 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
4059 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4060 struct btrfs_trans_handle
*trans
;
4061 struct inode
*inode
= d_inode(dentry
);
4064 trans
= __unlink_start_trans(dir
);
4066 return PTR_ERR(trans
);
4068 btrfs_record_unlink_dir(trans
, BTRFS_I(dir
), BTRFS_I(d_inode(dentry
)),
4071 ret
= btrfs_unlink_inode(trans
, root
, BTRFS_I(dir
),
4072 BTRFS_I(d_inode(dentry
)), dentry
->d_name
.name
,
4073 dentry
->d_name
.len
);
4077 if (inode
->i_nlink
== 0) {
4078 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
4084 btrfs_end_transaction(trans
);
4085 btrfs_btree_balance_dirty(root
->fs_info
);
4089 static int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
4090 struct inode
*dir
, u64 objectid
,
4091 const char *name
, int name_len
)
4093 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4094 struct btrfs_path
*path
;
4095 struct extent_buffer
*leaf
;
4096 struct btrfs_dir_item
*di
;
4097 struct btrfs_key key
;
4100 u64 dir_ino
= btrfs_ino(BTRFS_I(dir
));
4102 path
= btrfs_alloc_path();
4106 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4107 name
, name_len
, -1);
4108 if (IS_ERR_OR_NULL(di
)) {
4109 ret
= di
? PTR_ERR(di
) : -ENOENT
;
4113 leaf
= path
->nodes
[0];
4114 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4115 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
4116 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4118 btrfs_abort_transaction(trans
, ret
);
4121 btrfs_release_path(path
);
4123 ret
= btrfs_del_root_ref(trans
, objectid
, root
->root_key
.objectid
,
4124 dir_ino
, &index
, name
, name_len
);
4126 if (ret
!= -ENOENT
) {
4127 btrfs_abort_transaction(trans
, ret
);
4130 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
4132 if (IS_ERR_OR_NULL(di
)) {
4137 btrfs_abort_transaction(trans
, ret
);
4141 leaf
= path
->nodes
[0];
4142 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4145 btrfs_release_path(path
);
4147 ret
= btrfs_delete_delayed_dir_index(trans
, BTRFS_I(dir
), index
);
4149 btrfs_abort_transaction(trans
, ret
);
4153 btrfs_i_size_write(BTRFS_I(dir
), dir
->i_size
- name_len
* 2);
4154 inode_inc_iversion(dir
);
4155 dir
->i_mtime
= dir
->i_ctime
= current_time(dir
);
4156 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
4158 btrfs_abort_transaction(trans
, ret
);
4160 btrfs_free_path(path
);
4165 * Helper to check if the subvolume references other subvolumes or if it's
4168 static noinline
int may_destroy_subvol(struct btrfs_root
*root
)
4170 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4171 struct btrfs_path
*path
;
4172 struct btrfs_dir_item
*di
;
4173 struct btrfs_key key
;
4177 path
= btrfs_alloc_path();
4181 /* Make sure this root isn't set as the default subvol */
4182 dir_id
= btrfs_super_root_dir(fs_info
->super_copy
);
4183 di
= btrfs_lookup_dir_item(NULL
, fs_info
->tree_root
, path
,
4184 dir_id
, "default", 7, 0);
4185 if (di
&& !IS_ERR(di
)) {
4186 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &key
);
4187 if (key
.objectid
== root
->root_key
.objectid
) {
4190 "deleting default subvolume %llu is not allowed",
4194 btrfs_release_path(path
);
4197 key
.objectid
= root
->root_key
.objectid
;
4198 key
.type
= BTRFS_ROOT_REF_KEY
;
4199 key
.offset
= (u64
)-1;
4201 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
4207 if (path
->slots
[0] > 0) {
4209 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
4210 if (key
.objectid
== root
->root_key
.objectid
&&
4211 key
.type
== BTRFS_ROOT_REF_KEY
)
4215 btrfs_free_path(path
);
4219 /* Delete all dentries for inodes belonging to the root */
4220 static void btrfs_prune_dentries(struct btrfs_root
*root
)
4222 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4223 struct rb_node
*node
;
4224 struct rb_node
*prev
;
4225 struct btrfs_inode
*entry
;
4226 struct inode
*inode
;
4229 if (!test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
4230 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
4232 spin_lock(&root
->inode_lock
);
4234 node
= root
->inode_tree
.rb_node
;
4238 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4240 if (objectid
< btrfs_ino(entry
))
4241 node
= node
->rb_left
;
4242 else if (objectid
> btrfs_ino(entry
))
4243 node
= node
->rb_right
;
4249 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
4250 if (objectid
<= btrfs_ino(entry
)) {
4254 prev
= rb_next(prev
);
4258 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4259 objectid
= btrfs_ino(entry
) + 1;
4260 inode
= igrab(&entry
->vfs_inode
);
4262 spin_unlock(&root
->inode_lock
);
4263 if (atomic_read(&inode
->i_count
) > 1)
4264 d_prune_aliases(inode
);
4266 * btrfs_drop_inode will have it removed from the inode
4267 * cache when its usage count hits zero.
4271 spin_lock(&root
->inode_lock
);
4275 if (cond_resched_lock(&root
->inode_lock
))
4278 node
= rb_next(node
);
4280 spin_unlock(&root
->inode_lock
);
4283 int btrfs_delete_subvolume(struct inode
*dir
, struct dentry
*dentry
)
4285 struct btrfs_fs_info
*fs_info
= btrfs_sb(dentry
->d_sb
);
4286 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4287 struct inode
*inode
= d_inode(dentry
);
4288 struct btrfs_root
*dest
= BTRFS_I(inode
)->root
;
4289 struct btrfs_trans_handle
*trans
;
4290 struct btrfs_block_rsv block_rsv
;
4296 * Don't allow to delete a subvolume with send in progress. This is
4297 * inside the inode lock so the error handling that has to drop the bit
4298 * again is not run concurrently.
4300 spin_lock(&dest
->root_item_lock
);
4301 if (dest
->send_in_progress
) {
4302 spin_unlock(&dest
->root_item_lock
);
4304 "attempt to delete subvolume %llu during send",
4305 dest
->root_key
.objectid
);
4308 root_flags
= btrfs_root_flags(&dest
->root_item
);
4309 btrfs_set_root_flags(&dest
->root_item
,
4310 root_flags
| BTRFS_ROOT_SUBVOL_DEAD
);
4311 spin_unlock(&dest
->root_item_lock
);
4313 down_write(&fs_info
->subvol_sem
);
4315 err
= may_destroy_subvol(dest
);
4319 btrfs_init_block_rsv(&block_rsv
, BTRFS_BLOCK_RSV_TEMP
);
4321 * One for dir inode,
4322 * two for dir entries,
4323 * two for root ref/backref.
4325 err
= btrfs_subvolume_reserve_metadata(root
, &block_rsv
, 5, true);
4329 trans
= btrfs_start_transaction(root
, 0);
4330 if (IS_ERR(trans
)) {
4331 err
= PTR_ERR(trans
);
4334 trans
->block_rsv
= &block_rsv
;
4335 trans
->bytes_reserved
= block_rsv
.size
;
4337 btrfs_record_snapshot_destroy(trans
, BTRFS_I(dir
));
4339 ret
= btrfs_unlink_subvol(trans
, dir
, dest
->root_key
.objectid
,
4340 dentry
->d_name
.name
, dentry
->d_name
.len
);
4343 btrfs_abort_transaction(trans
, ret
);
4347 btrfs_record_root_in_trans(trans
, dest
);
4349 memset(&dest
->root_item
.drop_progress
, 0,
4350 sizeof(dest
->root_item
.drop_progress
));
4351 dest
->root_item
.drop_level
= 0;
4352 btrfs_set_root_refs(&dest
->root_item
, 0);
4354 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &dest
->state
)) {
4355 ret
= btrfs_insert_orphan_item(trans
,
4357 dest
->root_key
.objectid
);
4359 btrfs_abort_transaction(trans
, ret
);
4365 ret
= btrfs_uuid_tree_remove(trans
, dest
->root_item
.uuid
,
4366 BTRFS_UUID_KEY_SUBVOL
,
4367 dest
->root_key
.objectid
);
4368 if (ret
&& ret
!= -ENOENT
) {
4369 btrfs_abort_transaction(trans
, ret
);
4373 if (!btrfs_is_empty_uuid(dest
->root_item
.received_uuid
)) {
4374 ret
= btrfs_uuid_tree_remove(trans
,
4375 dest
->root_item
.received_uuid
,
4376 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
4377 dest
->root_key
.objectid
);
4378 if (ret
&& ret
!= -ENOENT
) {
4379 btrfs_abort_transaction(trans
, ret
);
4386 trans
->block_rsv
= NULL
;
4387 trans
->bytes_reserved
= 0;
4388 ret
= btrfs_end_transaction(trans
);
4391 inode
->i_flags
|= S_DEAD
;
4393 btrfs_subvolume_release_metadata(fs_info
, &block_rsv
);
4395 up_write(&fs_info
->subvol_sem
);
4397 spin_lock(&dest
->root_item_lock
);
4398 root_flags
= btrfs_root_flags(&dest
->root_item
);
4399 btrfs_set_root_flags(&dest
->root_item
,
4400 root_flags
& ~BTRFS_ROOT_SUBVOL_DEAD
);
4401 spin_unlock(&dest
->root_item_lock
);
4403 d_invalidate(dentry
);
4404 btrfs_prune_dentries(dest
);
4405 ASSERT(dest
->send_in_progress
== 0);
4408 if (dest
->ino_cache_inode
) {
4409 iput(dest
->ino_cache_inode
);
4410 dest
->ino_cache_inode
= NULL
;
4417 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
4419 struct inode
*inode
= d_inode(dentry
);
4421 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4422 struct btrfs_trans_handle
*trans
;
4423 u64 last_unlink_trans
;
4425 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
4427 if (btrfs_ino(BTRFS_I(inode
)) == BTRFS_FIRST_FREE_OBJECTID
)
4428 return btrfs_delete_subvolume(dir
, dentry
);
4430 trans
= __unlink_start_trans(dir
);
4432 return PTR_ERR(trans
);
4434 if (unlikely(btrfs_ino(BTRFS_I(inode
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
4435 err
= btrfs_unlink_subvol(trans
, dir
,
4436 BTRFS_I(inode
)->location
.objectid
,
4437 dentry
->d_name
.name
,
4438 dentry
->d_name
.len
);
4442 err
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
4446 last_unlink_trans
= BTRFS_I(inode
)->last_unlink_trans
;
4448 /* now the directory is empty */
4449 err
= btrfs_unlink_inode(trans
, root
, BTRFS_I(dir
),
4450 BTRFS_I(d_inode(dentry
)), dentry
->d_name
.name
,
4451 dentry
->d_name
.len
);
4453 btrfs_i_size_write(BTRFS_I(inode
), 0);
4455 * Propagate the last_unlink_trans value of the deleted dir to
4456 * its parent directory. This is to prevent an unrecoverable
4457 * log tree in the case we do something like this:
4459 * 2) create snapshot under dir foo
4460 * 3) delete the snapshot
4463 * 6) fsync foo or some file inside foo
4465 if (last_unlink_trans
>= trans
->transid
)
4466 BTRFS_I(dir
)->last_unlink_trans
= last_unlink_trans
;
4469 btrfs_end_transaction(trans
);
4470 btrfs_btree_balance_dirty(root
->fs_info
);
4476 * Return this if we need to call truncate_block for the last bit of the
4479 #define NEED_TRUNCATE_BLOCK 1
4482 * this can truncate away extent items, csum items and directory items.
4483 * It starts at a high offset and removes keys until it can't find
4484 * any higher than new_size
4486 * csum items that cross the new i_size are truncated to the new size
4489 * min_type is the minimum key type to truncate down to. If set to 0, this
4490 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4492 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
4493 struct btrfs_root
*root
,
4494 struct inode
*inode
,
4495 u64 new_size
, u32 min_type
)
4497 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4498 struct btrfs_path
*path
;
4499 struct extent_buffer
*leaf
;
4500 struct btrfs_file_extent_item
*fi
;
4501 struct btrfs_key key
;
4502 struct btrfs_key found_key
;
4503 u64 extent_start
= 0;
4504 u64 extent_num_bytes
= 0;
4505 u64 extent_offset
= 0;
4507 u64 last_size
= new_size
;
4508 u32 found_type
= (u8
)-1;
4511 int pending_del_nr
= 0;
4512 int pending_del_slot
= 0;
4513 int extent_type
= -1;
4515 u64 ino
= btrfs_ino(BTRFS_I(inode
));
4516 u64 bytes_deleted
= 0;
4517 bool be_nice
= false;
4518 bool should_throttle
= false;
4520 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
4523 * for non-free space inodes and ref cows, we want to back off from
4526 if (!btrfs_is_free_space_inode(BTRFS_I(inode
)) &&
4527 test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4530 path
= btrfs_alloc_path();
4533 path
->reada
= READA_BACK
;
4536 * We want to drop from the next block forward in case this new size is
4537 * not block aligned since we will be keeping the last block of the
4538 * extent just the way it is.
4540 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4541 root
== fs_info
->tree_root
)
4542 btrfs_drop_extent_cache(BTRFS_I(inode
), ALIGN(new_size
,
4543 fs_info
->sectorsize
),
4547 * This function is also used to drop the items in the log tree before
4548 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4549 * it is used to drop the logged items. So we shouldn't kill the delayed
4552 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
4553 btrfs_kill_delayed_inode_items(BTRFS_I(inode
));
4556 key
.offset
= (u64
)-1;
4561 * with a 16K leaf size and 128MB extents, you can actually queue
4562 * up a huge file in a single leaf. Most of the time that
4563 * bytes_deleted is > 0, it will be huge by the time we get here
4565 if (be_nice
&& bytes_deleted
> SZ_32M
&&
4566 btrfs_should_end_transaction(trans
)) {
4571 path
->leave_spinning
= 1;
4572 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
4578 /* there are no items in the tree for us to truncate, we're
4581 if (path
->slots
[0] == 0)
4588 leaf
= path
->nodes
[0];
4589 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4590 found_type
= found_key
.type
;
4592 if (found_key
.objectid
!= ino
)
4595 if (found_type
< min_type
)
4598 item_end
= found_key
.offset
;
4599 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
4600 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4601 struct btrfs_file_extent_item
);
4602 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4603 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4605 btrfs_file_extent_num_bytes(leaf
, fi
);
4607 trace_btrfs_truncate_show_fi_regular(
4608 BTRFS_I(inode
), leaf
, fi
,
4610 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4611 item_end
+= btrfs_file_extent_ram_bytes(leaf
,
4614 trace_btrfs_truncate_show_fi_inline(
4615 BTRFS_I(inode
), leaf
, fi
, path
->slots
[0],
4620 if (found_type
> min_type
) {
4623 if (item_end
< new_size
)
4625 if (found_key
.offset
>= new_size
)
4631 /* FIXME, shrink the extent if the ref count is only 1 */
4632 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
4635 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4637 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4639 u64 orig_num_bytes
=
4640 btrfs_file_extent_num_bytes(leaf
, fi
);
4641 extent_num_bytes
= ALIGN(new_size
-
4643 fs_info
->sectorsize
);
4644 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4646 num_dec
= (orig_num_bytes
-
4648 if (test_bit(BTRFS_ROOT_REF_COWS
,
4651 inode_sub_bytes(inode
, num_dec
);
4652 btrfs_mark_buffer_dirty(leaf
);
4655 btrfs_file_extent_disk_num_bytes(leaf
,
4657 extent_offset
= found_key
.offset
-
4658 btrfs_file_extent_offset(leaf
, fi
);
4660 /* FIXME blocksize != 4096 */
4661 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
4662 if (extent_start
!= 0) {
4664 if (test_bit(BTRFS_ROOT_REF_COWS
,
4666 inode_sub_bytes(inode
, num_dec
);
4669 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4671 * we can't truncate inline items that have had
4675 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
4676 btrfs_file_extent_other_encoding(leaf
, fi
) == 0 &&
4677 btrfs_file_extent_compression(leaf
, fi
) == 0) {
4678 u32 size
= (u32
)(new_size
- found_key
.offset
);
4680 btrfs_set_file_extent_ram_bytes(leaf
, fi
, size
);
4681 size
= btrfs_file_extent_calc_inline_size(size
);
4682 btrfs_truncate_item(root
->fs_info
, path
, size
, 1);
4683 } else if (!del_item
) {
4685 * We have to bail so the last_size is set to
4686 * just before this extent.
4688 ret
= NEED_TRUNCATE_BLOCK
;
4692 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4693 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4697 last_size
= found_key
.offset
;
4699 last_size
= new_size
;
4701 if (!pending_del_nr
) {
4702 /* no pending yet, add ourselves */
4703 pending_del_slot
= path
->slots
[0];
4705 } else if (pending_del_nr
&&
4706 path
->slots
[0] + 1 == pending_del_slot
) {
4707 /* hop on the pending chunk */
4709 pending_del_slot
= path
->slots
[0];
4716 should_throttle
= false;
4719 (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4720 root
== fs_info
->tree_root
)) {
4721 btrfs_set_path_blocking(path
);
4722 bytes_deleted
+= extent_num_bytes
;
4723 ret
= btrfs_free_extent(trans
, root
, extent_start
,
4724 extent_num_bytes
, 0,
4725 btrfs_header_owner(leaf
),
4726 ino
, extent_offset
);
4728 btrfs_abort_transaction(trans
, ret
);
4732 if (btrfs_should_throttle_delayed_refs(trans
))
4733 should_throttle
= true;
4737 if (found_type
== BTRFS_INODE_ITEM_KEY
)
4740 if (path
->slots
[0] == 0 ||
4741 path
->slots
[0] != pending_del_slot
||
4743 if (pending_del_nr
) {
4744 ret
= btrfs_del_items(trans
, root
, path
,
4748 btrfs_abort_transaction(trans
, ret
);
4753 btrfs_release_path(path
);
4756 * We can generate a lot of delayed refs, so we need to
4757 * throttle every once and a while and make sure we're
4758 * adding enough space to keep up with the work we are
4759 * generating. Since we hold a transaction here we
4760 * can't flush, and we don't want to FLUSH_LIMIT because
4761 * we could have generated too many delayed refs to
4762 * actually allocate, so just bail if we're short and
4763 * let the normal reservation dance happen higher up.
4765 if (should_throttle
) {
4766 ret
= btrfs_delayed_refs_rsv_refill(fs_info
,
4767 BTRFS_RESERVE_NO_FLUSH
);
4779 if (ret
>= 0 && pending_del_nr
) {
4782 err
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
4785 btrfs_abort_transaction(trans
, err
);
4789 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4790 ASSERT(last_size
>= new_size
);
4791 if (!ret
&& last_size
> new_size
)
4792 last_size
= new_size
;
4793 btrfs_ordered_update_i_size(inode
, last_size
, NULL
);
4796 btrfs_free_path(path
);
4801 * btrfs_truncate_block - read, zero a chunk and write a block
4802 * @inode - inode that we're zeroing
4803 * @from - the offset to start zeroing
4804 * @len - the length to zero, 0 to zero the entire range respective to the
4806 * @front - zero up to the offset instead of from the offset on
4808 * This will find the block for the "from" offset and cow the block and zero the
4809 * part we want to zero. This is used with truncate and hole punching.
4811 int btrfs_truncate_block(struct inode
*inode
, loff_t from
, loff_t len
,
4814 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4815 struct address_space
*mapping
= inode
->i_mapping
;
4816 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4817 struct btrfs_ordered_extent
*ordered
;
4818 struct extent_state
*cached_state
= NULL
;
4819 struct extent_changeset
*data_reserved
= NULL
;
4821 u32 blocksize
= fs_info
->sectorsize
;
4822 pgoff_t index
= from
>> PAGE_SHIFT
;
4823 unsigned offset
= from
& (blocksize
- 1);
4825 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4830 if (IS_ALIGNED(offset
, blocksize
) &&
4831 (!len
|| IS_ALIGNED(len
, blocksize
)))
4834 block_start
= round_down(from
, blocksize
);
4835 block_end
= block_start
+ blocksize
- 1;
4837 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
,
4838 block_start
, blocksize
);
4843 page
= find_or_create_page(mapping
, index
, mask
);
4845 btrfs_delalloc_release_space(inode
, data_reserved
,
4846 block_start
, blocksize
, true);
4847 btrfs_delalloc_release_extents(BTRFS_I(inode
), blocksize
, true);
4852 if (!PageUptodate(page
)) {
4853 ret
= btrfs_readpage(NULL
, page
);
4855 if (page
->mapping
!= mapping
) {
4860 if (!PageUptodate(page
)) {
4865 wait_on_page_writeback(page
);
4867 lock_extent_bits(io_tree
, block_start
, block_end
, &cached_state
);
4868 set_page_extent_mapped(page
);
4870 ordered
= btrfs_lookup_ordered_extent(inode
, block_start
);
4872 unlock_extent_cached(io_tree
, block_start
, block_end
,
4876 btrfs_start_ordered_extent(inode
, ordered
, 1);
4877 btrfs_put_ordered_extent(ordered
);
4881 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, block_start
, block_end
,
4882 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4883 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4884 0, 0, &cached_state
);
4886 ret
= btrfs_set_extent_delalloc(inode
, block_start
, block_end
, 0,
4889 unlock_extent_cached(io_tree
, block_start
, block_end
,
4894 if (offset
!= blocksize
) {
4896 len
= blocksize
- offset
;
4899 memset(kaddr
+ (block_start
- page_offset(page
)),
4902 memset(kaddr
+ (block_start
- page_offset(page
)) + offset
,
4904 flush_dcache_page(page
);
4907 ClearPageChecked(page
);
4908 set_page_dirty(page
);
4909 unlock_extent_cached(io_tree
, block_start
, block_end
, &cached_state
);
4913 btrfs_delalloc_release_space(inode
, data_reserved
, block_start
,
4915 btrfs_delalloc_release_extents(BTRFS_I(inode
), blocksize
, (ret
!= 0));
4919 extent_changeset_free(data_reserved
);
4923 static int maybe_insert_hole(struct btrfs_root
*root
, struct inode
*inode
,
4924 u64 offset
, u64 len
)
4926 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4927 struct btrfs_trans_handle
*trans
;
4931 * Still need to make sure the inode looks like it's been updated so
4932 * that any holes get logged if we fsync.
4934 if (btrfs_fs_incompat(fs_info
, NO_HOLES
)) {
4935 BTRFS_I(inode
)->last_trans
= fs_info
->generation
;
4936 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
4937 BTRFS_I(inode
)->last_log_commit
= root
->last_log_commit
;
4942 * 1 - for the one we're dropping
4943 * 1 - for the one we're adding
4944 * 1 - for updating the inode.
4946 trans
= btrfs_start_transaction(root
, 3);
4948 return PTR_ERR(trans
);
4950 ret
= btrfs_drop_extents(trans
, root
, inode
, offset
, offset
+ len
, 1);
4952 btrfs_abort_transaction(trans
, ret
);
4953 btrfs_end_transaction(trans
);
4957 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(BTRFS_I(inode
)),
4958 offset
, 0, 0, len
, 0, len
, 0, 0, 0);
4960 btrfs_abort_transaction(trans
, ret
);
4962 btrfs_update_inode(trans
, root
, inode
);
4963 btrfs_end_transaction(trans
);
4968 * This function puts in dummy file extents for the area we're creating a hole
4969 * for. So if we are truncating this file to a larger size we need to insert
4970 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4971 * the range between oldsize and size
4973 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
4975 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4976 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4977 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4978 struct extent_map
*em
= NULL
;
4979 struct extent_state
*cached_state
= NULL
;
4980 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4981 u64 hole_start
= ALIGN(oldsize
, fs_info
->sectorsize
);
4982 u64 block_end
= ALIGN(size
, fs_info
->sectorsize
);
4989 * If our size started in the middle of a block we need to zero out the
4990 * rest of the block before we expand the i_size, otherwise we could
4991 * expose stale data.
4993 err
= btrfs_truncate_block(inode
, oldsize
, 0, 0);
4997 if (size
<= hole_start
)
5001 struct btrfs_ordered_extent
*ordered
;
5003 lock_extent_bits(io_tree
, hole_start
, block_end
- 1,
5005 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), hole_start
,
5006 block_end
- hole_start
);
5009 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
5011 btrfs_start_ordered_extent(inode
, ordered
, 1);
5012 btrfs_put_ordered_extent(ordered
);
5015 cur_offset
= hole_start
;
5017 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0, cur_offset
,
5018 block_end
- cur_offset
, 0);
5024 last_byte
= min(extent_map_end(em
), block_end
);
5025 last_byte
= ALIGN(last_byte
, fs_info
->sectorsize
);
5026 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
5027 struct extent_map
*hole_em
;
5028 hole_size
= last_byte
- cur_offset
;
5030 err
= maybe_insert_hole(root
, inode
, cur_offset
,
5034 btrfs_drop_extent_cache(BTRFS_I(inode
), cur_offset
,
5035 cur_offset
+ hole_size
- 1, 0);
5036 hole_em
= alloc_extent_map();
5038 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
5039 &BTRFS_I(inode
)->runtime_flags
);
5042 hole_em
->start
= cur_offset
;
5043 hole_em
->len
= hole_size
;
5044 hole_em
->orig_start
= cur_offset
;
5046 hole_em
->block_start
= EXTENT_MAP_HOLE
;
5047 hole_em
->block_len
= 0;
5048 hole_em
->orig_block_len
= 0;
5049 hole_em
->ram_bytes
= hole_size
;
5050 hole_em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
5051 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
5052 hole_em
->generation
= fs_info
->generation
;
5055 write_lock(&em_tree
->lock
);
5056 err
= add_extent_mapping(em_tree
, hole_em
, 1);
5057 write_unlock(&em_tree
->lock
);
5060 btrfs_drop_extent_cache(BTRFS_I(inode
),
5065 free_extent_map(hole_em
);
5068 free_extent_map(em
);
5070 cur_offset
= last_byte
;
5071 if (cur_offset
>= block_end
)
5074 free_extent_map(em
);
5075 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
);
5079 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
5081 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5082 struct btrfs_trans_handle
*trans
;
5083 loff_t oldsize
= i_size_read(inode
);
5084 loff_t newsize
= attr
->ia_size
;
5085 int mask
= attr
->ia_valid
;
5089 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5090 * special case where we need to update the times despite not having
5091 * these flags set. For all other operations the VFS set these flags
5092 * explicitly if it wants a timestamp update.
5094 if (newsize
!= oldsize
) {
5095 inode_inc_iversion(inode
);
5096 if (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))
5097 inode
->i_ctime
= inode
->i_mtime
=
5098 current_time(inode
);
5101 if (newsize
> oldsize
) {
5103 * Don't do an expanding truncate while snapshotting is ongoing.
5104 * This is to ensure the snapshot captures a fully consistent
5105 * state of this file - if the snapshot captures this expanding
5106 * truncation, it must capture all writes that happened before
5109 btrfs_wait_for_snapshot_creation(root
);
5110 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
5112 btrfs_end_write_no_snapshotting(root
);
5116 trans
= btrfs_start_transaction(root
, 1);
5117 if (IS_ERR(trans
)) {
5118 btrfs_end_write_no_snapshotting(root
);
5119 return PTR_ERR(trans
);
5122 i_size_write(inode
, newsize
);
5123 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
5124 pagecache_isize_extended(inode
, oldsize
, newsize
);
5125 ret
= btrfs_update_inode(trans
, root
, inode
);
5126 btrfs_end_write_no_snapshotting(root
);
5127 btrfs_end_transaction(trans
);
5131 * We're truncating a file that used to have good data down to
5132 * zero. Make sure it gets into the ordered flush list so that
5133 * any new writes get down to disk quickly.
5136 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
5137 &BTRFS_I(inode
)->runtime_flags
);
5139 truncate_setsize(inode
, newsize
);
5141 /* Disable nonlocked read DIO to avoid the endless truncate */
5142 btrfs_inode_block_unlocked_dio(BTRFS_I(inode
));
5143 inode_dio_wait(inode
);
5144 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode
));
5146 ret
= btrfs_truncate(inode
, newsize
== oldsize
);
5147 if (ret
&& inode
->i_nlink
) {
5151 * Truncate failed, so fix up the in-memory size. We
5152 * adjusted disk_i_size down as we removed extents, so
5153 * wait for disk_i_size to be stable and then update the
5154 * in-memory size to match.
5156 err
= btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5159 i_size_write(inode
, BTRFS_I(inode
)->disk_i_size
);
5166 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5168 struct inode
*inode
= d_inode(dentry
);
5169 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5172 if (btrfs_root_readonly(root
))
5175 err
= setattr_prepare(dentry
, attr
);
5179 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
5180 err
= btrfs_setsize(inode
, attr
);
5185 if (attr
->ia_valid
) {
5186 setattr_copy(inode
, attr
);
5187 inode_inc_iversion(inode
);
5188 err
= btrfs_dirty_inode(inode
);
5190 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
5191 err
= posix_acl_chmod(inode
, inode
->i_mode
);
5198 * While truncating the inode pages during eviction, we get the VFS calling
5199 * btrfs_invalidatepage() against each page of the inode. This is slow because
5200 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5201 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5202 * extent_state structures over and over, wasting lots of time.
5204 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5205 * those expensive operations on a per page basis and do only the ordered io
5206 * finishing, while we release here the extent_map and extent_state structures,
5207 * without the excessive merging and splitting.
5209 static void evict_inode_truncate_pages(struct inode
*inode
)
5211 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5212 struct extent_map_tree
*map_tree
= &BTRFS_I(inode
)->extent_tree
;
5213 struct rb_node
*node
;
5215 ASSERT(inode
->i_state
& I_FREEING
);
5216 truncate_inode_pages_final(&inode
->i_data
);
5218 write_lock(&map_tree
->lock
);
5219 while (!RB_EMPTY_ROOT(&map_tree
->map
.rb_root
)) {
5220 struct extent_map
*em
;
5222 node
= rb_first_cached(&map_tree
->map
);
5223 em
= rb_entry(node
, struct extent_map
, rb_node
);
5224 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5225 clear_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
5226 remove_extent_mapping(map_tree
, em
);
5227 free_extent_map(em
);
5228 if (need_resched()) {
5229 write_unlock(&map_tree
->lock
);
5231 write_lock(&map_tree
->lock
);
5234 write_unlock(&map_tree
->lock
);
5237 * Keep looping until we have no more ranges in the io tree.
5238 * We can have ongoing bios started by readpages (called from readahead)
5239 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5240 * still in progress (unlocked the pages in the bio but did not yet
5241 * unlocked the ranges in the io tree). Therefore this means some
5242 * ranges can still be locked and eviction started because before
5243 * submitting those bios, which are executed by a separate task (work
5244 * queue kthread), inode references (inode->i_count) were not taken
5245 * (which would be dropped in the end io callback of each bio).
5246 * Therefore here we effectively end up waiting for those bios and
5247 * anyone else holding locked ranges without having bumped the inode's
5248 * reference count - if we don't do it, when they access the inode's
5249 * io_tree to unlock a range it may be too late, leading to an
5250 * use-after-free issue.
5252 spin_lock(&io_tree
->lock
);
5253 while (!RB_EMPTY_ROOT(&io_tree
->state
)) {
5254 struct extent_state
*state
;
5255 struct extent_state
*cached_state
= NULL
;
5258 unsigned state_flags
;
5260 node
= rb_first(&io_tree
->state
);
5261 state
= rb_entry(node
, struct extent_state
, rb_node
);
5262 start
= state
->start
;
5264 state_flags
= state
->state
;
5265 spin_unlock(&io_tree
->lock
);
5267 lock_extent_bits(io_tree
, start
, end
, &cached_state
);
5270 * If still has DELALLOC flag, the extent didn't reach disk,
5271 * and its reserved space won't be freed by delayed_ref.
5272 * So we need to free its reserved space here.
5273 * (Refer to comment in btrfs_invalidatepage, case 2)
5275 * Note, end is the bytenr of last byte, so we need + 1 here.
5277 if (state_flags
& EXTENT_DELALLOC
)
5278 btrfs_qgroup_free_data(inode
, NULL
, start
, end
- start
+ 1);
5280 clear_extent_bit(io_tree
, start
, end
,
5281 EXTENT_LOCKED
| EXTENT_DIRTY
|
5282 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
5283 EXTENT_DEFRAG
, 1, 1, &cached_state
);
5286 spin_lock(&io_tree
->lock
);
5288 spin_unlock(&io_tree
->lock
);
5291 static struct btrfs_trans_handle
*evict_refill_and_join(struct btrfs_root
*root
,
5292 struct btrfs_block_rsv
*rsv
)
5294 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5295 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5296 u64 delayed_refs_extra
= btrfs_calc_trans_metadata_size(fs_info
, 1);
5300 struct btrfs_trans_handle
*trans
;
5303 ret
= btrfs_block_rsv_refill(root
, rsv
,
5304 rsv
->size
+ delayed_refs_extra
,
5305 BTRFS_RESERVE_FLUSH_LIMIT
);
5307 if (ret
&& ++failures
> 2) {
5309 "could not allocate space for a delete; will truncate on mount");
5310 return ERR_PTR(-ENOSPC
);
5314 * Evict can generate a large amount of delayed refs without
5315 * having a way to add space back since we exhaust our temporary
5316 * block rsv. We aren't allowed to do FLUSH_ALL in this case
5317 * because we could deadlock with so many things in the flushing
5318 * code, so we have to try and hold some extra space to
5319 * compensate for our delayed ref generation. If we can't get
5320 * that space then we need see if we can steal our minimum from
5321 * the global reserve. We will be ratelimited by the amount of
5322 * space we have for the delayed refs rsv, so we'll end up
5323 * committing and trying again.
5325 trans
= btrfs_join_transaction(root
);
5326 if (IS_ERR(trans
) || !ret
) {
5327 if (!IS_ERR(trans
)) {
5328 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
5329 trans
->bytes_reserved
= delayed_refs_extra
;
5330 btrfs_block_rsv_migrate(rsv
, trans
->block_rsv
,
5331 delayed_refs_extra
, 1);
5337 * Try to steal from the global reserve if there is space for
5340 if (!btrfs_check_space_for_delayed_refs(fs_info
) &&
5341 !btrfs_block_rsv_migrate(global_rsv
, rsv
, rsv
->size
, 0))
5344 /* If not, commit and try again. */
5345 ret
= btrfs_commit_transaction(trans
);
5347 return ERR_PTR(ret
);
5351 void btrfs_evict_inode(struct inode
*inode
)
5353 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5354 struct btrfs_trans_handle
*trans
;
5355 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5356 struct btrfs_block_rsv
*rsv
;
5359 trace_btrfs_inode_evict(inode
);
5366 evict_inode_truncate_pages(inode
);
5368 if (inode
->i_nlink
&&
5369 ((btrfs_root_refs(&root
->root_item
) != 0 &&
5370 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
) ||
5371 btrfs_is_free_space_inode(BTRFS_I(inode
))))
5374 if (is_bad_inode(inode
))
5377 btrfs_free_io_failure_record(BTRFS_I(inode
), 0, (u64
)-1);
5379 if (test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
))
5382 if (inode
->i_nlink
> 0) {
5383 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0 &&
5384 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
);
5388 ret
= btrfs_commit_inode_delayed_inode(BTRFS_I(inode
));
5392 rsv
= btrfs_alloc_block_rsv(fs_info
, BTRFS_BLOCK_RSV_TEMP
);
5395 rsv
->size
= btrfs_calc_trunc_metadata_size(fs_info
, 1);
5398 btrfs_i_size_write(BTRFS_I(inode
), 0);
5401 trans
= evict_refill_and_join(root
, rsv
);
5405 trans
->block_rsv
= rsv
;
5407 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
5408 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
5409 btrfs_end_transaction(trans
);
5410 btrfs_btree_balance_dirty(fs_info
);
5411 if (ret
&& ret
!= -ENOSPC
&& ret
!= -EAGAIN
)
5418 * Errors here aren't a big deal, it just means we leave orphan items in
5419 * the tree. They will be cleaned up on the next mount. If the inode
5420 * number gets reused, cleanup deletes the orphan item without doing
5421 * anything, and unlink reuses the existing orphan item.
5423 * If it turns out that we are dropping too many of these, we might want
5424 * to add a mechanism for retrying these after a commit.
5426 trans
= evict_refill_and_join(root
, rsv
);
5427 if (!IS_ERR(trans
)) {
5428 trans
->block_rsv
= rsv
;
5429 btrfs_orphan_del(trans
, BTRFS_I(inode
));
5430 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
5431 btrfs_end_transaction(trans
);
5434 if (!(root
== fs_info
->tree_root
||
5435 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
5436 btrfs_return_ino(root
, btrfs_ino(BTRFS_I(inode
)));
5439 btrfs_free_block_rsv(fs_info
, rsv
);
5442 * If we didn't successfully delete, the orphan item will still be in
5443 * the tree and we'll retry on the next mount. Again, we might also want
5444 * to retry these periodically in the future.
5446 btrfs_remove_delayed_node(BTRFS_I(inode
));
5451 * this returns the key found in the dir entry in the location pointer.
5452 * If no dir entries were found, returns -ENOENT.
5453 * If found a corrupted location in dir entry, returns -EUCLEAN.
5455 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
5456 struct btrfs_key
*location
)
5458 const char *name
= dentry
->d_name
.name
;
5459 int namelen
= dentry
->d_name
.len
;
5460 struct btrfs_dir_item
*di
;
5461 struct btrfs_path
*path
;
5462 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5465 path
= btrfs_alloc_path();
5469 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(BTRFS_I(dir
)),
5471 if (IS_ERR_OR_NULL(di
)) {
5472 ret
= di
? PTR_ERR(di
) : -ENOENT
;
5476 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
5477 if (location
->type
!= BTRFS_INODE_ITEM_KEY
&&
5478 location
->type
!= BTRFS_ROOT_ITEM_KEY
) {
5480 btrfs_warn(root
->fs_info
,
5481 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5482 __func__
, name
, btrfs_ino(BTRFS_I(dir
)),
5483 location
->objectid
, location
->type
, location
->offset
);
5486 btrfs_free_path(path
);
5491 * when we hit a tree root in a directory, the btrfs part of the inode
5492 * needs to be changed to reflect the root directory of the tree root. This
5493 * is kind of like crossing a mount point.
5495 static int fixup_tree_root_location(struct btrfs_fs_info
*fs_info
,
5497 struct dentry
*dentry
,
5498 struct btrfs_key
*location
,
5499 struct btrfs_root
**sub_root
)
5501 struct btrfs_path
*path
;
5502 struct btrfs_root
*new_root
;
5503 struct btrfs_root_ref
*ref
;
5504 struct extent_buffer
*leaf
;
5505 struct btrfs_key key
;
5509 path
= btrfs_alloc_path();
5516 key
.objectid
= BTRFS_I(dir
)->root
->root_key
.objectid
;
5517 key
.type
= BTRFS_ROOT_REF_KEY
;
5518 key
.offset
= location
->objectid
;
5520 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
5527 leaf
= path
->nodes
[0];
5528 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
5529 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(BTRFS_I(dir
)) ||
5530 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
5533 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
5534 (unsigned long)(ref
+ 1),
5535 dentry
->d_name
.len
);
5539 btrfs_release_path(path
);
5541 new_root
= btrfs_read_fs_root_no_name(fs_info
, location
);
5542 if (IS_ERR(new_root
)) {
5543 err
= PTR_ERR(new_root
);
5547 *sub_root
= new_root
;
5548 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
5549 location
->type
= BTRFS_INODE_ITEM_KEY
;
5550 location
->offset
= 0;
5553 btrfs_free_path(path
);
5557 static void inode_tree_add(struct inode
*inode
)
5559 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5560 struct btrfs_inode
*entry
;
5562 struct rb_node
*parent
;
5563 struct rb_node
*new = &BTRFS_I(inode
)->rb_node
;
5564 u64 ino
= btrfs_ino(BTRFS_I(inode
));
5566 if (inode_unhashed(inode
))
5569 spin_lock(&root
->inode_lock
);
5570 p
= &root
->inode_tree
.rb_node
;
5573 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
5575 if (ino
< btrfs_ino(entry
))
5576 p
= &parent
->rb_left
;
5577 else if (ino
> btrfs_ino(entry
))
5578 p
= &parent
->rb_right
;
5580 WARN_ON(!(entry
->vfs_inode
.i_state
&
5581 (I_WILL_FREE
| I_FREEING
)));
5582 rb_replace_node(parent
, new, &root
->inode_tree
);
5583 RB_CLEAR_NODE(parent
);
5584 spin_unlock(&root
->inode_lock
);
5588 rb_link_node(new, parent
, p
);
5589 rb_insert_color(new, &root
->inode_tree
);
5590 spin_unlock(&root
->inode_lock
);
5593 static void inode_tree_del(struct inode
*inode
)
5595 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5596 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5599 spin_lock(&root
->inode_lock
);
5600 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
5601 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
5602 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
5603 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5605 spin_unlock(&root
->inode_lock
);
5607 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0) {
5608 synchronize_srcu(&fs_info
->subvol_srcu
);
5609 spin_lock(&root
->inode_lock
);
5610 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5611 spin_unlock(&root
->inode_lock
);
5613 btrfs_add_dead_root(root
);
5618 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
5620 struct btrfs_iget_args
*args
= p
;
5621 inode
->i_ino
= args
->location
->objectid
;
5622 memcpy(&BTRFS_I(inode
)->location
, args
->location
,
5623 sizeof(*args
->location
));
5624 BTRFS_I(inode
)->root
= args
->root
;
5628 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
5630 struct btrfs_iget_args
*args
= opaque
;
5631 return args
->location
->objectid
== BTRFS_I(inode
)->location
.objectid
&&
5632 args
->root
== BTRFS_I(inode
)->root
;
5635 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
5636 struct btrfs_key
*location
,
5637 struct btrfs_root
*root
)
5639 struct inode
*inode
;
5640 struct btrfs_iget_args args
;
5641 unsigned long hashval
= btrfs_inode_hash(location
->objectid
, root
);
5643 args
.location
= location
;
5646 inode
= iget5_locked(s
, hashval
, btrfs_find_actor
,
5647 btrfs_init_locked_inode
,
5652 /* Get an inode object given its location and corresponding root.
5653 * Returns in *is_new if the inode was read from disk
5655 struct inode
*btrfs_iget_path(struct super_block
*s
, struct btrfs_key
*location
,
5656 struct btrfs_root
*root
, int *new,
5657 struct btrfs_path
*path
)
5659 struct inode
*inode
;
5661 inode
= btrfs_iget_locked(s
, location
, root
);
5663 return ERR_PTR(-ENOMEM
);
5665 if (inode
->i_state
& I_NEW
) {
5668 ret
= btrfs_read_locked_inode(inode
, path
);
5670 inode_tree_add(inode
);
5671 unlock_new_inode(inode
);
5677 * ret > 0 can come from btrfs_search_slot called by
5678 * btrfs_read_locked_inode, this means the inode item
5683 inode
= ERR_PTR(ret
);
5690 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
5691 struct btrfs_root
*root
, int *new)
5693 return btrfs_iget_path(s
, location
, root
, new, NULL
);
5696 static struct inode
*new_simple_dir(struct super_block
*s
,
5697 struct btrfs_key
*key
,
5698 struct btrfs_root
*root
)
5700 struct inode
*inode
= new_inode(s
);
5703 return ERR_PTR(-ENOMEM
);
5705 BTRFS_I(inode
)->root
= root
;
5706 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
5707 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
5709 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
5710 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
5711 inode
->i_opflags
&= ~IOP_XATTR
;
5712 inode
->i_fop
= &simple_dir_operations
;
5713 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
5714 inode
->i_mtime
= current_time(inode
);
5715 inode
->i_atime
= inode
->i_mtime
;
5716 inode
->i_ctime
= inode
->i_mtime
;
5717 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
5722 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
5724 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
5725 struct inode
*inode
;
5726 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5727 struct btrfs_root
*sub_root
= root
;
5728 struct btrfs_key location
;
5732 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
5733 return ERR_PTR(-ENAMETOOLONG
);
5735 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
5737 return ERR_PTR(ret
);
5739 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
5740 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
5744 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
5745 ret
= fixup_tree_root_location(fs_info
, dir
, dentry
,
5746 &location
, &sub_root
);
5749 inode
= ERR_PTR(ret
);
5751 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
5753 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
5755 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5757 if (!IS_ERR(inode
) && root
!= sub_root
) {
5758 down_read(&fs_info
->cleanup_work_sem
);
5759 if (!sb_rdonly(inode
->i_sb
))
5760 ret
= btrfs_orphan_cleanup(sub_root
);
5761 up_read(&fs_info
->cleanup_work_sem
);
5764 inode
= ERR_PTR(ret
);
5771 static int btrfs_dentry_delete(const struct dentry
*dentry
)
5773 struct btrfs_root
*root
;
5774 struct inode
*inode
= d_inode(dentry
);
5776 if (!inode
&& !IS_ROOT(dentry
))
5777 inode
= d_inode(dentry
->d_parent
);
5780 root
= BTRFS_I(inode
)->root
;
5781 if (btrfs_root_refs(&root
->root_item
) == 0)
5784 if (btrfs_ino(BTRFS_I(inode
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5790 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5793 struct inode
*inode
= btrfs_lookup_dentry(dir
, dentry
);
5795 if (inode
== ERR_PTR(-ENOENT
))
5797 return d_splice_alias(inode
, dentry
);
5800 unsigned char btrfs_filetype_table
[] = {
5801 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
5805 * All this infrastructure exists because dir_emit can fault, and we are holding
5806 * the tree lock when doing readdir. For now just allocate a buffer and copy
5807 * our information into that, and then dir_emit from the buffer. This is
5808 * similar to what NFS does, only we don't keep the buffer around in pagecache
5809 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5810 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5813 static int btrfs_opendir(struct inode
*inode
, struct file
*file
)
5815 struct btrfs_file_private
*private;
5817 private = kzalloc(sizeof(struct btrfs_file_private
), GFP_KERNEL
);
5820 private->filldir_buf
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
5821 if (!private->filldir_buf
) {
5825 file
->private_data
= private;
5836 static int btrfs_filldir(void *addr
, int entries
, struct dir_context
*ctx
)
5839 struct dir_entry
*entry
= addr
;
5840 char *name
= (char *)(entry
+ 1);
5842 ctx
->pos
= get_unaligned(&entry
->offset
);
5843 if (!dir_emit(ctx
, name
, get_unaligned(&entry
->name_len
),
5844 get_unaligned(&entry
->ino
),
5845 get_unaligned(&entry
->type
)))
5847 addr
+= sizeof(struct dir_entry
) +
5848 get_unaligned(&entry
->name_len
);
5854 static int btrfs_real_readdir(struct file
*file
, struct dir_context
*ctx
)
5856 struct inode
*inode
= file_inode(file
);
5857 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5858 struct btrfs_file_private
*private = file
->private_data
;
5859 struct btrfs_dir_item
*di
;
5860 struct btrfs_key key
;
5861 struct btrfs_key found_key
;
5862 struct btrfs_path
*path
;
5864 struct list_head ins_list
;
5865 struct list_head del_list
;
5867 struct extent_buffer
*leaf
;
5874 struct btrfs_key location
;
5876 if (!dir_emit_dots(file
, ctx
))
5879 path
= btrfs_alloc_path();
5883 addr
= private->filldir_buf
;
5884 path
->reada
= READA_FORWARD
;
5886 INIT_LIST_HEAD(&ins_list
);
5887 INIT_LIST_HEAD(&del_list
);
5888 put
= btrfs_readdir_get_delayed_items(inode
, &ins_list
, &del_list
);
5891 key
.type
= BTRFS_DIR_INDEX_KEY
;
5892 key
.offset
= ctx
->pos
;
5893 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
5895 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5900 struct dir_entry
*entry
;
5902 leaf
= path
->nodes
[0];
5903 slot
= path
->slots
[0];
5904 if (slot
>= btrfs_header_nritems(leaf
)) {
5905 ret
= btrfs_next_leaf(root
, path
);
5913 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5915 if (found_key
.objectid
!= key
.objectid
)
5917 if (found_key
.type
!= BTRFS_DIR_INDEX_KEY
)
5919 if (found_key
.offset
< ctx
->pos
)
5921 if (btrfs_should_delete_dir_index(&del_list
, found_key
.offset
))
5923 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
5924 name_len
= btrfs_dir_name_len(leaf
, di
);
5925 if ((total_len
+ sizeof(struct dir_entry
) + name_len
) >=
5927 btrfs_release_path(path
);
5928 ret
= btrfs_filldir(private->filldir_buf
, entries
, ctx
);
5931 addr
= private->filldir_buf
;
5938 put_unaligned(name_len
, &entry
->name_len
);
5939 name_ptr
= (char *)(entry
+ 1);
5940 read_extent_buffer(leaf
, name_ptr
, (unsigned long)(di
+ 1),
5942 put_unaligned(btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)],
5944 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
5945 put_unaligned(location
.objectid
, &entry
->ino
);
5946 put_unaligned(found_key
.offset
, &entry
->offset
);
5948 addr
+= sizeof(struct dir_entry
) + name_len
;
5949 total_len
+= sizeof(struct dir_entry
) + name_len
;
5953 btrfs_release_path(path
);
5955 ret
= btrfs_filldir(private->filldir_buf
, entries
, ctx
);
5959 ret
= btrfs_readdir_delayed_dir_index(ctx
, &ins_list
);
5964 * Stop new entries from being returned after we return the last
5967 * New directory entries are assigned a strictly increasing
5968 * offset. This means that new entries created during readdir
5969 * are *guaranteed* to be seen in the future by that readdir.
5970 * This has broken buggy programs which operate on names as
5971 * they're returned by readdir. Until we re-use freed offsets
5972 * we have this hack to stop new entries from being returned
5973 * under the assumption that they'll never reach this huge
5976 * This is being careful not to overflow 32bit loff_t unless the
5977 * last entry requires it because doing so has broken 32bit apps
5980 if (ctx
->pos
>= INT_MAX
)
5981 ctx
->pos
= LLONG_MAX
;
5988 btrfs_readdir_put_delayed_items(inode
, &ins_list
, &del_list
);
5989 btrfs_free_path(path
);
5994 * This is somewhat expensive, updating the tree every time the
5995 * inode changes. But, it is most likely to find the inode in cache.
5996 * FIXME, needs more benchmarking...there are no reasons other than performance
5997 * to keep or drop this code.
5999 static int btrfs_dirty_inode(struct inode
*inode
)
6001 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
6002 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6003 struct btrfs_trans_handle
*trans
;
6006 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
6009 trans
= btrfs_join_transaction(root
);
6011 return PTR_ERR(trans
);
6013 ret
= btrfs_update_inode(trans
, root
, inode
);
6014 if (ret
&& ret
== -ENOSPC
) {
6015 /* whoops, lets try again with the full transaction */
6016 btrfs_end_transaction(trans
);
6017 trans
= btrfs_start_transaction(root
, 1);
6019 return PTR_ERR(trans
);
6021 ret
= btrfs_update_inode(trans
, root
, inode
);
6023 btrfs_end_transaction(trans
);
6024 if (BTRFS_I(inode
)->delayed_node
)
6025 btrfs_balance_delayed_items(fs_info
);
6031 * This is a copy of file_update_time. We need this so we can return error on
6032 * ENOSPC for updating the inode in the case of file write and mmap writes.
6034 static int btrfs_update_time(struct inode
*inode
, struct timespec64
*now
,
6037 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6038 bool dirty
= flags
& ~S_VERSION
;
6040 if (btrfs_root_readonly(root
))
6043 if (flags
& S_VERSION
)
6044 dirty
|= inode_maybe_inc_iversion(inode
, dirty
);
6045 if (flags
& S_CTIME
)
6046 inode
->i_ctime
= *now
;
6047 if (flags
& S_MTIME
)
6048 inode
->i_mtime
= *now
;
6049 if (flags
& S_ATIME
)
6050 inode
->i_atime
= *now
;
6051 return dirty
? btrfs_dirty_inode(inode
) : 0;
6055 * find the highest existing sequence number in a directory
6056 * and then set the in-memory index_cnt variable to reflect
6057 * free sequence numbers
6059 static int btrfs_set_inode_index_count(struct btrfs_inode
*inode
)
6061 struct btrfs_root
*root
= inode
->root
;
6062 struct btrfs_key key
, found_key
;
6063 struct btrfs_path
*path
;
6064 struct extent_buffer
*leaf
;
6067 key
.objectid
= btrfs_ino(inode
);
6068 key
.type
= BTRFS_DIR_INDEX_KEY
;
6069 key
.offset
= (u64
)-1;
6071 path
= btrfs_alloc_path();
6075 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6078 /* FIXME: we should be able to handle this */
6084 * MAGIC NUMBER EXPLANATION:
6085 * since we search a directory based on f_pos we have to start at 2
6086 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6087 * else has to start at 2
6089 if (path
->slots
[0] == 0) {
6090 inode
->index_cnt
= 2;
6096 leaf
= path
->nodes
[0];
6097 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6099 if (found_key
.objectid
!= btrfs_ino(inode
) ||
6100 found_key
.type
!= BTRFS_DIR_INDEX_KEY
) {
6101 inode
->index_cnt
= 2;
6105 inode
->index_cnt
= found_key
.offset
+ 1;
6107 btrfs_free_path(path
);
6112 * helper to find a free sequence number in a given directory. This current
6113 * code is very simple, later versions will do smarter things in the btree
6115 int btrfs_set_inode_index(struct btrfs_inode
*dir
, u64
*index
)
6119 if (dir
->index_cnt
== (u64
)-1) {
6120 ret
= btrfs_inode_delayed_dir_index_count(dir
);
6122 ret
= btrfs_set_inode_index_count(dir
);
6128 *index
= dir
->index_cnt
;
6134 static int btrfs_insert_inode_locked(struct inode
*inode
)
6136 struct btrfs_iget_args args
;
6137 args
.location
= &BTRFS_I(inode
)->location
;
6138 args
.root
= BTRFS_I(inode
)->root
;
6140 return insert_inode_locked4(inode
,
6141 btrfs_inode_hash(inode
->i_ino
, BTRFS_I(inode
)->root
),
6142 btrfs_find_actor
, &args
);
6146 * Inherit flags from the parent inode.
6148 * Currently only the compression flags and the cow flags are inherited.
6150 static void btrfs_inherit_iflags(struct inode
*inode
, struct inode
*dir
)
6157 flags
= BTRFS_I(dir
)->flags
;
6159 if (flags
& BTRFS_INODE_NOCOMPRESS
) {
6160 BTRFS_I(inode
)->flags
&= ~BTRFS_INODE_COMPRESS
;
6161 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
6162 } else if (flags
& BTRFS_INODE_COMPRESS
) {
6163 BTRFS_I(inode
)->flags
&= ~BTRFS_INODE_NOCOMPRESS
;
6164 BTRFS_I(inode
)->flags
|= BTRFS_INODE_COMPRESS
;
6167 if (flags
& BTRFS_INODE_NODATACOW
) {
6168 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
6169 if (S_ISREG(inode
->i_mode
))
6170 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6173 btrfs_sync_inode_flags_to_i_flags(inode
);
6176 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
6177 struct btrfs_root
*root
,
6179 const char *name
, int name_len
,
6180 u64 ref_objectid
, u64 objectid
,
6181 umode_t mode
, u64
*index
)
6183 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6184 struct inode
*inode
;
6185 struct btrfs_inode_item
*inode_item
;
6186 struct btrfs_key
*location
;
6187 struct btrfs_path
*path
;
6188 struct btrfs_inode_ref
*ref
;
6189 struct btrfs_key key
[2];
6191 int nitems
= name
? 2 : 1;
6195 path
= btrfs_alloc_path();
6197 return ERR_PTR(-ENOMEM
);
6199 inode
= new_inode(fs_info
->sb
);
6201 btrfs_free_path(path
);
6202 return ERR_PTR(-ENOMEM
);
6206 * O_TMPFILE, set link count to 0, so that after this point,
6207 * we fill in an inode item with the correct link count.
6210 set_nlink(inode
, 0);
6213 * we have to initialize this early, so we can reclaim the inode
6214 * number if we fail afterwards in this function.
6216 inode
->i_ino
= objectid
;
6219 trace_btrfs_inode_request(dir
);
6221 ret
= btrfs_set_inode_index(BTRFS_I(dir
), index
);
6223 btrfs_free_path(path
);
6225 return ERR_PTR(ret
);
6231 * index_cnt is ignored for everything but a dir,
6232 * btrfs_set_inode_index_count has an explanation for the magic
6235 BTRFS_I(inode
)->index_cnt
= 2;
6236 BTRFS_I(inode
)->dir_index
= *index
;
6237 BTRFS_I(inode
)->root
= root
;
6238 BTRFS_I(inode
)->generation
= trans
->transid
;
6239 inode
->i_generation
= BTRFS_I(inode
)->generation
;
6242 * We could have gotten an inode number from somebody who was fsynced
6243 * and then removed in this same transaction, so let's just set full
6244 * sync since it will be a full sync anyway and this will blow away the
6245 * old info in the log.
6247 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
6249 key
[0].objectid
= objectid
;
6250 key
[0].type
= BTRFS_INODE_ITEM_KEY
;
6253 sizes
[0] = sizeof(struct btrfs_inode_item
);
6257 * Start new inodes with an inode_ref. This is slightly more
6258 * efficient for small numbers of hard links since they will
6259 * be packed into one item. Extended refs will kick in if we
6260 * add more hard links than can fit in the ref item.
6262 key
[1].objectid
= objectid
;
6263 key
[1].type
= BTRFS_INODE_REF_KEY
;
6264 key
[1].offset
= ref_objectid
;
6266 sizes
[1] = name_len
+ sizeof(*ref
);
6269 location
= &BTRFS_I(inode
)->location
;
6270 location
->objectid
= objectid
;
6271 location
->offset
= 0;
6272 location
->type
= BTRFS_INODE_ITEM_KEY
;
6274 ret
= btrfs_insert_inode_locked(inode
);
6280 path
->leave_spinning
= 1;
6281 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, nitems
);
6285 inode_init_owner(inode
, dir
, mode
);
6286 inode_set_bytes(inode
, 0);
6288 inode
->i_mtime
= current_time(inode
);
6289 inode
->i_atime
= inode
->i_mtime
;
6290 inode
->i_ctime
= inode
->i_mtime
;
6291 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
6293 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
6294 struct btrfs_inode_item
);
6295 memzero_extent_buffer(path
->nodes
[0], (unsigned long)inode_item
,
6296 sizeof(*inode_item
));
6297 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
6300 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
6301 struct btrfs_inode_ref
);
6302 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
6303 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
6304 ptr
= (unsigned long)(ref
+ 1);
6305 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
6308 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6309 btrfs_free_path(path
);
6311 btrfs_inherit_iflags(inode
, dir
);
6313 if (S_ISREG(mode
)) {
6314 if (btrfs_test_opt(fs_info
, NODATASUM
))
6315 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6316 if (btrfs_test_opt(fs_info
, NODATACOW
))
6317 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
6318 BTRFS_INODE_NODATASUM
;
6321 inode_tree_add(inode
);
6323 trace_btrfs_inode_new(inode
);
6324 btrfs_set_inode_last_trans(trans
, inode
);
6326 btrfs_update_root_times(trans
, root
);
6328 ret
= btrfs_inode_inherit_props(trans
, inode
, dir
);
6331 "error inheriting props for ino %llu (root %llu): %d",
6332 btrfs_ino(BTRFS_I(inode
)), root
->root_key
.objectid
, ret
);
6337 discard_new_inode(inode
);
6340 BTRFS_I(dir
)->index_cnt
--;
6341 btrfs_free_path(path
);
6342 return ERR_PTR(ret
);
6345 static inline u8
btrfs_inode_type(struct inode
*inode
)
6347 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
6351 * utility function to add 'inode' into 'parent_inode' with
6352 * a give name and a given sequence number.
6353 * if 'add_backref' is true, also insert a backref from the
6354 * inode to the parent directory.
6356 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
6357 struct btrfs_inode
*parent_inode
, struct btrfs_inode
*inode
,
6358 const char *name
, int name_len
, int add_backref
, u64 index
)
6361 struct btrfs_key key
;
6362 struct btrfs_root
*root
= parent_inode
->root
;
6363 u64 ino
= btrfs_ino(inode
);
6364 u64 parent_ino
= btrfs_ino(parent_inode
);
6366 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6367 memcpy(&key
, &inode
->root
->root_key
, sizeof(key
));
6370 key
.type
= BTRFS_INODE_ITEM_KEY
;
6374 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6375 ret
= btrfs_add_root_ref(trans
, key
.objectid
,
6376 root
->root_key
.objectid
, parent_ino
,
6377 index
, name
, name_len
);
6378 } else if (add_backref
) {
6379 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
6383 /* Nothing to clean up yet */
6387 ret
= btrfs_insert_dir_item(trans
, name
, name_len
, parent_inode
, &key
,
6388 btrfs_inode_type(&inode
->vfs_inode
), index
);
6389 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
6392 btrfs_abort_transaction(trans
, ret
);
6396 btrfs_i_size_write(parent_inode
, parent_inode
->vfs_inode
.i_size
+
6398 inode_inc_iversion(&parent_inode
->vfs_inode
);
6399 parent_inode
->vfs_inode
.i_mtime
= parent_inode
->vfs_inode
.i_ctime
=
6400 current_time(&parent_inode
->vfs_inode
);
6401 ret
= btrfs_update_inode(trans
, root
, &parent_inode
->vfs_inode
);
6403 btrfs_abort_transaction(trans
, ret
);
6407 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6410 err
= btrfs_del_root_ref(trans
, key
.objectid
,
6411 root
->root_key
.objectid
, parent_ino
,
6412 &local_index
, name
, name_len
);
6414 btrfs_abort_transaction(trans
, err
);
6415 } else if (add_backref
) {
6419 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
6420 ino
, parent_ino
, &local_index
);
6422 btrfs_abort_transaction(trans
, err
);
6425 /* Return the original error code */
6429 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
6430 struct btrfs_inode
*dir
, struct dentry
*dentry
,
6431 struct btrfs_inode
*inode
, int backref
, u64 index
)
6433 int err
= btrfs_add_link(trans
, dir
, inode
,
6434 dentry
->d_name
.name
, dentry
->d_name
.len
,
6441 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
6442 umode_t mode
, dev_t rdev
)
6444 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6445 struct btrfs_trans_handle
*trans
;
6446 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6447 struct inode
*inode
= NULL
;
6453 * 2 for inode item and ref
6455 * 1 for xattr if selinux is on
6457 trans
= btrfs_start_transaction(root
, 5);
6459 return PTR_ERR(trans
);
6461 err
= btrfs_find_free_ino(root
, &objectid
);
6465 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6466 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)), objectid
,
6468 if (IS_ERR(inode
)) {
6469 err
= PTR_ERR(inode
);
6475 * If the active LSM wants to access the inode during
6476 * d_instantiate it needs these. Smack checks to see
6477 * if the filesystem supports xattrs by looking at the
6480 inode
->i_op
= &btrfs_special_inode_operations
;
6481 init_special_inode(inode
, inode
->i_mode
, rdev
);
6483 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6487 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
, BTRFS_I(inode
),
6492 btrfs_update_inode(trans
, root
, inode
);
6493 d_instantiate_new(dentry
, inode
);
6496 btrfs_end_transaction(trans
);
6497 btrfs_btree_balance_dirty(fs_info
);
6499 inode_dec_link_count(inode
);
6500 discard_new_inode(inode
);
6505 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
6506 umode_t mode
, bool excl
)
6508 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6509 struct btrfs_trans_handle
*trans
;
6510 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6511 struct inode
*inode
= NULL
;
6517 * 2 for inode item and ref
6519 * 1 for xattr if selinux is on
6521 trans
= btrfs_start_transaction(root
, 5);
6523 return PTR_ERR(trans
);
6525 err
= btrfs_find_free_ino(root
, &objectid
);
6529 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6530 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)), objectid
,
6532 if (IS_ERR(inode
)) {
6533 err
= PTR_ERR(inode
);
6538 * If the active LSM wants to access the inode during
6539 * d_instantiate it needs these. Smack checks to see
6540 * if the filesystem supports xattrs by looking at the
6543 inode
->i_fop
= &btrfs_file_operations
;
6544 inode
->i_op
= &btrfs_file_inode_operations
;
6545 inode
->i_mapping
->a_ops
= &btrfs_aops
;
6547 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6551 err
= btrfs_update_inode(trans
, root
, inode
);
6555 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
, BTRFS_I(inode
),
6560 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
6561 d_instantiate_new(dentry
, inode
);
6564 btrfs_end_transaction(trans
);
6566 inode_dec_link_count(inode
);
6567 discard_new_inode(inode
);
6569 btrfs_btree_balance_dirty(fs_info
);
6573 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
6574 struct dentry
*dentry
)
6576 struct btrfs_trans_handle
*trans
= NULL
;
6577 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6578 struct inode
*inode
= d_inode(old_dentry
);
6579 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
6584 /* do not allow sys_link's with other subvols of the same device */
6585 if (root
->root_key
.objectid
!= BTRFS_I(inode
)->root
->root_key
.objectid
)
6588 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
6591 err
= btrfs_set_inode_index(BTRFS_I(dir
), &index
);
6596 * 2 items for inode and inode ref
6597 * 2 items for dir items
6598 * 1 item for parent inode
6599 * 1 item for orphan item deletion if O_TMPFILE
6601 trans
= btrfs_start_transaction(root
, inode
->i_nlink
? 5 : 6);
6602 if (IS_ERR(trans
)) {
6603 err
= PTR_ERR(trans
);
6608 /* There are several dir indexes for this inode, clear the cache. */
6609 BTRFS_I(inode
)->dir_index
= 0ULL;
6611 inode_inc_iversion(inode
);
6612 inode
->i_ctime
= current_time(inode
);
6614 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
6616 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
, BTRFS_I(inode
),
6622 struct dentry
*parent
= dentry
->d_parent
;
6625 err
= btrfs_update_inode(trans
, root
, inode
);
6628 if (inode
->i_nlink
== 1) {
6630 * If new hard link count is 1, it's a file created
6631 * with open(2) O_TMPFILE flag.
6633 err
= btrfs_orphan_del(trans
, BTRFS_I(inode
));
6637 BTRFS_I(inode
)->last_link_trans
= trans
->transid
;
6638 d_instantiate(dentry
, inode
);
6639 ret
= btrfs_log_new_name(trans
, BTRFS_I(inode
), NULL
, parent
,
6641 if (ret
== BTRFS_NEED_TRANS_COMMIT
) {
6642 err
= btrfs_commit_transaction(trans
);
6649 btrfs_end_transaction(trans
);
6651 inode_dec_link_count(inode
);
6654 btrfs_btree_balance_dirty(fs_info
);
6658 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
6660 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6661 struct inode
*inode
= NULL
;
6662 struct btrfs_trans_handle
*trans
;
6663 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6669 * 2 items for inode and ref
6670 * 2 items for dir items
6671 * 1 for xattr if selinux is on
6673 trans
= btrfs_start_transaction(root
, 5);
6675 return PTR_ERR(trans
);
6677 err
= btrfs_find_free_ino(root
, &objectid
);
6681 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6682 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)), objectid
,
6683 S_IFDIR
| mode
, &index
);
6684 if (IS_ERR(inode
)) {
6685 err
= PTR_ERR(inode
);
6690 /* these must be set before we unlock the inode */
6691 inode
->i_op
= &btrfs_dir_inode_operations
;
6692 inode
->i_fop
= &btrfs_dir_file_operations
;
6694 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6698 btrfs_i_size_write(BTRFS_I(inode
), 0);
6699 err
= btrfs_update_inode(trans
, root
, inode
);
6703 err
= btrfs_add_link(trans
, BTRFS_I(dir
), BTRFS_I(inode
),
6704 dentry
->d_name
.name
,
6705 dentry
->d_name
.len
, 0, index
);
6709 d_instantiate_new(dentry
, inode
);
6712 btrfs_end_transaction(trans
);
6714 inode_dec_link_count(inode
);
6715 discard_new_inode(inode
);
6717 btrfs_btree_balance_dirty(fs_info
);
6721 static noinline
int uncompress_inline(struct btrfs_path
*path
,
6723 size_t pg_offset
, u64 extent_offset
,
6724 struct btrfs_file_extent_item
*item
)
6727 struct extent_buffer
*leaf
= path
->nodes
[0];
6730 unsigned long inline_size
;
6734 WARN_ON(pg_offset
!= 0);
6735 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6736 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6737 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
6738 btrfs_item_nr(path
->slots
[0]));
6739 tmp
= kmalloc(inline_size
, GFP_NOFS
);
6742 ptr
= btrfs_file_extent_inline_start(item
);
6744 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
6746 max_size
= min_t(unsigned long, PAGE_SIZE
, max_size
);
6747 ret
= btrfs_decompress(compress_type
, tmp
, page
,
6748 extent_offset
, inline_size
, max_size
);
6751 * decompression code contains a memset to fill in any space between the end
6752 * of the uncompressed data and the end of max_size in case the decompressed
6753 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6754 * the end of an inline extent and the beginning of the next block, so we
6755 * cover that region here.
6758 if (max_size
+ pg_offset
< PAGE_SIZE
) {
6759 char *map
= kmap(page
);
6760 memset(map
+ pg_offset
+ max_size
, 0, PAGE_SIZE
- max_size
- pg_offset
);
6768 * a bit scary, this does extent mapping from logical file offset to the disk.
6769 * the ugly parts come from merging extents from the disk with the in-ram
6770 * representation. This gets more complex because of the data=ordered code,
6771 * where the in-ram extents might be locked pending data=ordered completion.
6773 * This also copies inline extents directly into the page.
6775 struct extent_map
*btrfs_get_extent(struct btrfs_inode
*inode
,
6777 size_t pg_offset
, u64 start
, u64 len
,
6780 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
6783 u64 extent_start
= 0;
6785 u64 objectid
= btrfs_ino(inode
);
6787 struct btrfs_path
*path
= NULL
;
6788 struct btrfs_root
*root
= inode
->root
;
6789 struct btrfs_file_extent_item
*item
;
6790 struct extent_buffer
*leaf
;
6791 struct btrfs_key found_key
;
6792 struct extent_map
*em
= NULL
;
6793 struct extent_map_tree
*em_tree
= &inode
->extent_tree
;
6794 struct extent_io_tree
*io_tree
= &inode
->io_tree
;
6795 const bool new_inline
= !page
|| create
;
6797 read_lock(&em_tree
->lock
);
6798 em
= lookup_extent_mapping(em_tree
, start
, len
);
6800 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
6801 read_unlock(&em_tree
->lock
);
6804 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
6805 free_extent_map(em
);
6806 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
6807 free_extent_map(em
);
6811 em
= alloc_extent_map();
6816 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
6817 em
->start
= EXTENT_MAP_HOLE
;
6818 em
->orig_start
= EXTENT_MAP_HOLE
;
6820 em
->block_len
= (u64
)-1;
6822 path
= btrfs_alloc_path();
6828 /* Chances are we'll be called again, so go ahead and do readahead */
6829 path
->reada
= READA_FORWARD
;
6832 * Unless we're going to uncompress the inline extent, no sleep would
6835 path
->leave_spinning
= 1;
6837 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, objectid
, start
, 0);
6841 } else if (ret
> 0) {
6842 if (path
->slots
[0] == 0)
6847 leaf
= path
->nodes
[0];
6848 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6849 struct btrfs_file_extent_item
);
6850 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6851 if (found_key
.objectid
!= objectid
||
6852 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
6854 * If we backup past the first extent we want to move forward
6855 * and see if there is an extent in front of us, otherwise we'll
6856 * say there is a hole for our whole search range which can
6863 extent_type
= btrfs_file_extent_type(leaf
, item
);
6864 extent_start
= found_key
.offset
;
6865 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
6866 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6867 extent_end
= extent_start
+
6868 btrfs_file_extent_num_bytes(leaf
, item
);
6870 trace_btrfs_get_extent_show_fi_regular(inode
, leaf
, item
,
6872 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
6875 size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6876 extent_end
= ALIGN(extent_start
+ size
,
6877 fs_info
->sectorsize
);
6879 trace_btrfs_get_extent_show_fi_inline(inode
, leaf
, item
,
6884 if (start
>= extent_end
) {
6886 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
6887 ret
= btrfs_next_leaf(root
, path
);
6891 } else if (ret
> 0) {
6894 leaf
= path
->nodes
[0];
6896 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6897 if (found_key
.objectid
!= objectid
||
6898 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6900 if (start
+ len
<= found_key
.offset
)
6902 if (start
> found_key
.offset
)
6905 /* New extent overlaps with existing one */
6907 em
->orig_start
= start
;
6908 em
->len
= found_key
.offset
- start
;
6909 em
->block_start
= EXTENT_MAP_HOLE
;
6913 btrfs_extent_item_to_extent_map(inode
, path
, item
,
6916 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
6917 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6919 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
6923 size_t extent_offset
;
6929 size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6930 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
6931 copy_size
= min_t(u64
, PAGE_SIZE
- pg_offset
,
6932 size
- extent_offset
);
6933 em
->start
= extent_start
+ extent_offset
;
6934 em
->len
= ALIGN(copy_size
, fs_info
->sectorsize
);
6935 em
->orig_block_len
= em
->len
;
6936 em
->orig_start
= em
->start
;
6937 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
6939 btrfs_set_path_blocking(path
);
6940 if (!PageUptodate(page
)) {
6941 if (btrfs_file_extent_compression(leaf
, item
) !=
6942 BTRFS_COMPRESS_NONE
) {
6943 ret
= uncompress_inline(path
, page
, pg_offset
,
6944 extent_offset
, item
);
6951 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6953 if (pg_offset
+ copy_size
< PAGE_SIZE
) {
6954 memset(map
+ pg_offset
+ copy_size
, 0,
6955 PAGE_SIZE
- pg_offset
-
6960 flush_dcache_page(page
);
6962 set_extent_uptodate(io_tree
, em
->start
,
6963 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
6968 em
->orig_start
= start
;
6970 em
->block_start
= EXTENT_MAP_HOLE
;
6972 btrfs_release_path(path
);
6973 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
6975 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6976 em
->start
, em
->len
, start
, len
);
6982 write_lock(&em_tree
->lock
);
6983 err
= btrfs_add_extent_mapping(fs_info
, em_tree
, &em
, start
, len
);
6984 write_unlock(&em_tree
->lock
);
6986 btrfs_free_path(path
);
6988 trace_btrfs_get_extent(root
, inode
, em
);
6991 free_extent_map(em
);
6992 return ERR_PTR(err
);
6994 BUG_ON(!em
); /* Error is always set */
6998 struct extent_map
*btrfs_get_extent_fiemap(struct btrfs_inode
*inode
,
7001 struct extent_map
*em
;
7002 struct extent_map
*hole_em
= NULL
;
7003 u64 delalloc_start
= start
;
7009 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
7013 * If our em maps to:
7015 * - a pre-alloc extent,
7016 * there might actually be delalloc bytes behind it.
7018 if (em
->block_start
!= EXTENT_MAP_HOLE
&&
7019 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7024 /* check to see if we've wrapped (len == -1 or similar) */
7033 /* ok, we didn't find anything, lets look for delalloc */
7034 delalloc_len
= count_range_bits(&inode
->io_tree
, &delalloc_start
,
7035 end
, len
, EXTENT_DELALLOC
, 1);
7036 delalloc_end
= delalloc_start
+ delalloc_len
;
7037 if (delalloc_end
< delalloc_start
)
7038 delalloc_end
= (u64
)-1;
7041 * We didn't find anything useful, return the original results from
7044 if (delalloc_start
> end
|| delalloc_end
<= start
) {
7051 * Adjust the delalloc_start to make sure it doesn't go backwards from
7052 * the start they passed in
7054 delalloc_start
= max(start
, delalloc_start
);
7055 delalloc_len
= delalloc_end
- delalloc_start
;
7057 if (delalloc_len
> 0) {
7060 const u64 hole_end
= extent_map_end(hole_em
);
7062 em
= alloc_extent_map();
7071 * When btrfs_get_extent can't find anything it returns one
7074 * Make sure what it found really fits our range, and adjust to
7075 * make sure it is based on the start from the caller
7077 if (hole_end
<= start
|| hole_em
->start
> end
) {
7078 free_extent_map(hole_em
);
7081 hole_start
= max(hole_em
->start
, start
);
7082 hole_len
= hole_end
- hole_start
;
7085 if (hole_em
&& delalloc_start
> hole_start
) {
7087 * Our hole starts before our delalloc, so we have to
7088 * return just the parts of the hole that go until the
7091 em
->len
= min(hole_len
, delalloc_start
- hole_start
);
7092 em
->start
= hole_start
;
7093 em
->orig_start
= hole_start
;
7095 * Don't adjust block start at all, it is fixed at
7098 em
->block_start
= hole_em
->block_start
;
7099 em
->block_len
= hole_len
;
7100 if (test_bit(EXTENT_FLAG_PREALLOC
, &hole_em
->flags
))
7101 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
7104 * Hole is out of passed range or it starts after
7107 em
->start
= delalloc_start
;
7108 em
->len
= delalloc_len
;
7109 em
->orig_start
= delalloc_start
;
7110 em
->block_start
= EXTENT_MAP_DELALLOC
;
7111 em
->block_len
= delalloc_len
;
7118 free_extent_map(hole_em
);
7120 free_extent_map(em
);
7121 return ERR_PTR(err
);
7126 static struct extent_map
*btrfs_create_dio_extent(struct inode
*inode
,
7129 const u64 orig_start
,
7130 const u64 block_start
,
7131 const u64 block_len
,
7132 const u64 orig_block_len
,
7133 const u64 ram_bytes
,
7136 struct extent_map
*em
= NULL
;
7139 if (type
!= BTRFS_ORDERED_NOCOW
) {
7140 em
= create_io_em(inode
, start
, len
, orig_start
,
7141 block_start
, block_len
, orig_block_len
,
7143 BTRFS_COMPRESS_NONE
, /* compress_type */
7148 ret
= btrfs_add_ordered_extent_dio(inode
, start
, block_start
,
7149 len
, block_len
, type
);
7152 free_extent_map(em
);
7153 btrfs_drop_extent_cache(BTRFS_I(inode
), start
,
7154 start
+ len
- 1, 0);
7163 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
7166 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7167 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7168 struct extent_map
*em
;
7169 struct btrfs_key ins
;
7173 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
7174 ret
= btrfs_reserve_extent(root
, len
, len
, fs_info
->sectorsize
,
7175 0, alloc_hint
, &ins
, 1, 1);
7177 return ERR_PTR(ret
);
7179 em
= btrfs_create_dio_extent(inode
, start
, ins
.offset
, start
,
7180 ins
.objectid
, ins
.offset
, ins
.offset
,
7181 ins
.offset
, BTRFS_ORDERED_REGULAR
);
7182 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
7184 btrfs_free_reserved_extent(fs_info
, ins
.objectid
,
7191 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7192 * block must be cow'd
7194 noinline
int can_nocow_extent(struct inode
*inode
, u64 offset
, u64
*len
,
7195 u64
*orig_start
, u64
*orig_block_len
,
7198 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7199 struct btrfs_path
*path
;
7201 struct extent_buffer
*leaf
;
7202 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7203 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7204 struct btrfs_file_extent_item
*fi
;
7205 struct btrfs_key key
;
7212 bool nocow
= (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
);
7214 path
= btrfs_alloc_path();
7218 ret
= btrfs_lookup_file_extent(NULL
, root
, path
,
7219 btrfs_ino(BTRFS_I(inode
)), offset
, 0);
7223 slot
= path
->slots
[0];
7226 /* can't find the item, must cow */
7233 leaf
= path
->nodes
[0];
7234 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
7235 if (key
.objectid
!= btrfs_ino(BTRFS_I(inode
)) ||
7236 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7237 /* not our file or wrong item type, must cow */
7241 if (key
.offset
> offset
) {
7242 /* Wrong offset, must cow */
7246 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
7247 found_type
= btrfs_file_extent_type(leaf
, fi
);
7248 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
7249 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
7250 /* not a regular extent, must cow */
7254 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_REG
)
7257 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
7258 if (extent_end
<= offset
)
7261 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7262 if (disk_bytenr
== 0)
7265 if (btrfs_file_extent_compression(leaf
, fi
) ||
7266 btrfs_file_extent_encryption(leaf
, fi
) ||
7267 btrfs_file_extent_other_encoding(leaf
, fi
))
7271 * Do the same check as in btrfs_cross_ref_exist but without the
7272 * unnecessary search.
7274 if (btrfs_file_extent_generation(leaf
, fi
) <=
7275 btrfs_root_last_snapshot(&root
->root_item
))
7278 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
7281 *orig_start
= key
.offset
- backref_offset
;
7282 *orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7283 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
7286 if (btrfs_extent_readonly(fs_info
, disk_bytenr
))
7289 num_bytes
= min(offset
+ *len
, extent_end
) - offset
;
7290 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7293 range_end
= round_up(offset
+ num_bytes
,
7294 root
->fs_info
->sectorsize
) - 1;
7295 ret
= test_range_bit(io_tree
, offset
, range_end
,
7296 EXTENT_DELALLOC
, 0, NULL
);
7303 btrfs_release_path(path
);
7306 * look for other files referencing this extent, if we
7307 * find any we must cow
7310 ret
= btrfs_cross_ref_exist(root
, btrfs_ino(BTRFS_I(inode
)),
7311 key
.offset
- backref_offset
, disk_bytenr
);
7318 * adjust disk_bytenr and num_bytes to cover just the bytes
7319 * in this extent we are about to write. If there
7320 * are any csums in that range we have to cow in order
7321 * to keep the csums correct
7323 disk_bytenr
+= backref_offset
;
7324 disk_bytenr
+= offset
- key
.offset
;
7325 if (csum_exist_in_range(fs_info
, disk_bytenr
, num_bytes
))
7328 * all of the above have passed, it is safe to overwrite this extent
7334 btrfs_free_path(path
);
7338 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
7339 struct extent_state
**cached_state
, int writing
)
7341 struct btrfs_ordered_extent
*ordered
;
7345 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7348 * We're concerned with the entire range that we're going to be
7349 * doing DIO to, so we need to make sure there's no ordered
7350 * extents in this range.
7352 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), lockstart
,
7353 lockend
- lockstart
+ 1);
7356 * We need to make sure there are no buffered pages in this
7357 * range either, we could have raced between the invalidate in
7358 * generic_file_direct_write and locking the extent. The
7359 * invalidate needs to happen so that reads after a write do not
7363 (!writing
|| !filemap_range_has_page(inode
->i_mapping
,
7364 lockstart
, lockend
)))
7367 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7372 * If we are doing a DIO read and the ordered extent we
7373 * found is for a buffered write, we can not wait for it
7374 * to complete and retry, because if we do so we can
7375 * deadlock with concurrent buffered writes on page
7376 * locks. This happens only if our DIO read covers more
7377 * than one extent map, if at this point has already
7378 * created an ordered extent for a previous extent map
7379 * and locked its range in the inode's io tree, and a
7380 * concurrent write against that previous extent map's
7381 * range and this range started (we unlock the ranges
7382 * in the io tree only when the bios complete and
7383 * buffered writes always lock pages before attempting
7384 * to lock range in the io tree).
7387 test_bit(BTRFS_ORDERED_DIRECT
, &ordered
->flags
))
7388 btrfs_start_ordered_extent(inode
, ordered
, 1);
7391 btrfs_put_ordered_extent(ordered
);
7394 * We could trigger writeback for this range (and wait
7395 * for it to complete) and then invalidate the pages for
7396 * this range (through invalidate_inode_pages2_range()),
7397 * but that can lead us to a deadlock with a concurrent
7398 * call to readpages() (a buffered read or a defrag call
7399 * triggered a readahead) on a page lock due to an
7400 * ordered dio extent we created before but did not have
7401 * yet a corresponding bio submitted (whence it can not
7402 * complete), which makes readpages() wait for that
7403 * ordered extent to complete while holding a lock on
7418 /* The callers of this must take lock_extent() */
7419 static struct extent_map
*create_io_em(struct inode
*inode
, u64 start
, u64 len
,
7420 u64 orig_start
, u64 block_start
,
7421 u64 block_len
, u64 orig_block_len
,
7422 u64 ram_bytes
, int compress_type
,
7425 struct extent_map_tree
*em_tree
;
7426 struct extent_map
*em
;
7427 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7430 ASSERT(type
== BTRFS_ORDERED_PREALLOC
||
7431 type
== BTRFS_ORDERED_COMPRESSED
||
7432 type
== BTRFS_ORDERED_NOCOW
||
7433 type
== BTRFS_ORDERED_REGULAR
);
7435 em_tree
= &BTRFS_I(inode
)->extent_tree
;
7436 em
= alloc_extent_map();
7438 return ERR_PTR(-ENOMEM
);
7441 em
->orig_start
= orig_start
;
7443 em
->block_len
= block_len
;
7444 em
->block_start
= block_start
;
7445 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
7446 em
->orig_block_len
= orig_block_len
;
7447 em
->ram_bytes
= ram_bytes
;
7448 em
->generation
= -1;
7449 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
7450 if (type
== BTRFS_ORDERED_PREALLOC
) {
7451 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
7452 } else if (type
== BTRFS_ORDERED_COMPRESSED
) {
7453 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
7454 em
->compress_type
= compress_type
;
7458 btrfs_drop_extent_cache(BTRFS_I(inode
), em
->start
,
7459 em
->start
+ em
->len
- 1, 0);
7460 write_lock(&em_tree
->lock
);
7461 ret
= add_extent_mapping(em_tree
, em
, 1);
7462 write_unlock(&em_tree
->lock
);
7464 * The caller has taken lock_extent(), who could race with us
7467 } while (ret
== -EEXIST
);
7470 free_extent_map(em
);
7471 return ERR_PTR(ret
);
7474 /* em got 2 refs now, callers needs to do free_extent_map once. */
7479 static int btrfs_get_blocks_direct_read(struct extent_map
*em
,
7480 struct buffer_head
*bh_result
,
7481 struct inode
*inode
,
7484 if (em
->block_start
== EXTENT_MAP_HOLE
||
7485 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7488 len
= min(len
, em
->len
- (start
- em
->start
));
7490 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
7492 bh_result
->b_size
= len
;
7493 bh_result
->b_bdev
= em
->bdev
;
7494 set_buffer_mapped(bh_result
);
7499 static int btrfs_get_blocks_direct_write(struct extent_map
**map
,
7500 struct buffer_head
*bh_result
,
7501 struct inode
*inode
,
7502 struct btrfs_dio_data
*dio_data
,
7505 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7506 struct extent_map
*em
= *map
;
7510 * We don't allocate a new extent in the following cases
7512 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7514 * 2) The extent is marked as PREALLOC. We're good to go here and can
7515 * just use the extent.
7518 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
7519 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
7520 em
->block_start
!= EXTENT_MAP_HOLE
)) {
7522 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
7524 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7525 type
= BTRFS_ORDERED_PREALLOC
;
7527 type
= BTRFS_ORDERED_NOCOW
;
7528 len
= min(len
, em
->len
- (start
- em
->start
));
7529 block_start
= em
->block_start
+ (start
- em
->start
);
7531 if (can_nocow_extent(inode
, start
, &len
, &orig_start
,
7532 &orig_block_len
, &ram_bytes
) == 1 &&
7533 btrfs_inc_nocow_writers(fs_info
, block_start
)) {
7534 struct extent_map
*em2
;
7536 em2
= btrfs_create_dio_extent(inode
, start
, len
,
7537 orig_start
, block_start
,
7538 len
, orig_block_len
,
7540 btrfs_dec_nocow_writers(fs_info
, block_start
);
7541 if (type
== BTRFS_ORDERED_PREALLOC
) {
7542 free_extent_map(em
);
7546 if (em2
&& IS_ERR(em2
)) {
7551 * For inode marked NODATACOW or extent marked PREALLOC,
7552 * use the existing or preallocated extent, so does not
7553 * need to adjust btrfs_space_info's bytes_may_use.
7555 btrfs_free_reserved_data_space_noquota(inode
, start
,
7561 /* this will cow the extent */
7562 len
= bh_result
->b_size
;
7563 free_extent_map(em
);
7564 *map
= em
= btrfs_new_extent_direct(inode
, start
, len
);
7570 len
= min(len
, em
->len
- (start
- em
->start
));
7573 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
7575 bh_result
->b_size
= len
;
7576 bh_result
->b_bdev
= em
->bdev
;
7577 set_buffer_mapped(bh_result
);
7579 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7580 set_buffer_new(bh_result
);
7583 * Need to update the i_size under the extent lock so buffered
7584 * readers will get the updated i_size when we unlock.
7586 if (!dio_data
->overwrite
&& start
+ len
> i_size_read(inode
))
7587 i_size_write(inode
, start
+ len
);
7589 WARN_ON(dio_data
->reserve
< len
);
7590 dio_data
->reserve
-= len
;
7591 dio_data
->unsubmitted_oe_range_end
= start
+ len
;
7592 current
->journal_info
= dio_data
;
7597 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
7598 struct buffer_head
*bh_result
, int create
)
7600 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7601 struct extent_map
*em
;
7602 struct extent_state
*cached_state
= NULL
;
7603 struct btrfs_dio_data
*dio_data
= NULL
;
7604 u64 start
= iblock
<< inode
->i_blkbits
;
7605 u64 lockstart
, lockend
;
7606 u64 len
= bh_result
->b_size
;
7607 int unlock_bits
= EXTENT_LOCKED
;
7611 unlock_bits
|= EXTENT_DIRTY
;
7613 len
= min_t(u64
, len
, fs_info
->sectorsize
);
7616 lockend
= start
+ len
- 1;
7618 if (current
->journal_info
) {
7620 * Need to pull our outstanding extents and set journal_info to NULL so
7621 * that anything that needs to check if there's a transaction doesn't get
7624 dio_data
= current
->journal_info
;
7625 current
->journal_info
= NULL
;
7629 * If this errors out it's because we couldn't invalidate pagecache for
7630 * this range and we need to fallback to buffered.
7632 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
,
7638 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0, start
, len
, 0);
7645 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7646 * io. INLINE is special, and we could probably kludge it in here, but
7647 * it's still buffered so for safety lets just fall back to the generic
7650 * For COMPRESSED we _have_ to read the entire extent in so we can
7651 * decompress it, so there will be buffering required no matter what we
7652 * do, so go ahead and fallback to buffered.
7654 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7655 * to buffered IO. Don't blame me, this is the price we pay for using
7658 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
7659 em
->block_start
== EXTENT_MAP_INLINE
) {
7660 free_extent_map(em
);
7666 ret
= btrfs_get_blocks_direct_write(&em
, bh_result
, inode
,
7667 dio_data
, start
, len
);
7671 /* clear and unlock the entire range */
7672 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7673 unlock_bits
, 1, 0, &cached_state
);
7675 ret
= btrfs_get_blocks_direct_read(em
, bh_result
, inode
,
7677 /* Can be negative only if we read from a hole */
7680 free_extent_map(em
);
7684 * We need to unlock only the end area that we aren't using.
7685 * The rest is going to be unlocked by the endio routine.
7687 lockstart
= start
+ bh_result
->b_size
;
7688 if (lockstart
< lockend
) {
7689 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
7690 lockend
, unlock_bits
, 1, 0,
7693 free_extent_state(cached_state
);
7697 free_extent_map(em
);
7702 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7703 unlock_bits
, 1, 0, &cached_state
);
7706 current
->journal_info
= dio_data
;
7710 static inline blk_status_t
submit_dio_repair_bio(struct inode
*inode
,
7714 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7717 BUG_ON(bio_op(bio
) == REQ_OP_WRITE
);
7719 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, BTRFS_WQ_ENDIO_DIO_REPAIR
);
7723 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
7728 static int btrfs_check_dio_repairable(struct inode
*inode
,
7729 struct bio
*failed_bio
,
7730 struct io_failure_record
*failrec
,
7733 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7736 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
, failrec
->len
);
7737 if (num_copies
== 1) {
7739 * we only have a single copy of the data, so don't bother with
7740 * all the retry and error correction code that follows. no
7741 * matter what the error is, it is very likely to persist.
7743 btrfs_debug(fs_info
,
7744 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7745 num_copies
, failrec
->this_mirror
, failed_mirror
);
7749 failrec
->failed_mirror
= failed_mirror
;
7750 failrec
->this_mirror
++;
7751 if (failrec
->this_mirror
== failed_mirror
)
7752 failrec
->this_mirror
++;
7754 if (failrec
->this_mirror
> num_copies
) {
7755 btrfs_debug(fs_info
,
7756 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7757 num_copies
, failrec
->this_mirror
, failed_mirror
);
7764 static blk_status_t
dio_read_error(struct inode
*inode
, struct bio
*failed_bio
,
7765 struct page
*page
, unsigned int pgoff
,
7766 u64 start
, u64 end
, int failed_mirror
,
7767 bio_end_io_t
*repair_endio
, void *repair_arg
)
7769 struct io_failure_record
*failrec
;
7770 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7771 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
7774 unsigned int read_mode
= 0;
7777 blk_status_t status
;
7778 struct bio_vec bvec
;
7780 BUG_ON(bio_op(failed_bio
) == REQ_OP_WRITE
);
7782 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
7784 return errno_to_blk_status(ret
);
7786 ret
= btrfs_check_dio_repairable(inode
, failed_bio
, failrec
,
7789 free_io_failure(failure_tree
, io_tree
, failrec
);
7790 return BLK_STS_IOERR
;
7793 segs
= bio_segments(failed_bio
);
7794 bio_get_first_bvec(failed_bio
, &bvec
);
7796 (bvec
.bv_len
> btrfs_inode_sectorsize(inode
)))
7797 read_mode
|= REQ_FAILFAST_DEV
;
7799 isector
= start
- btrfs_io_bio(failed_bio
)->logical
;
7800 isector
>>= inode
->i_sb
->s_blocksize_bits
;
7801 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
7802 pgoff
, isector
, repair_endio
, repair_arg
);
7803 bio
->bi_opf
= REQ_OP_READ
| read_mode
;
7805 btrfs_debug(BTRFS_I(inode
)->root
->fs_info
,
7806 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7807 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
7809 status
= submit_dio_repair_bio(inode
, bio
, failrec
->this_mirror
);
7811 free_io_failure(failure_tree
, io_tree
, failrec
);
7818 struct btrfs_retry_complete
{
7819 struct completion done
;
7820 struct inode
*inode
;
7825 static void btrfs_retry_endio_nocsum(struct bio
*bio
)
7827 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7828 struct inode
*inode
= done
->inode
;
7829 struct bio_vec
*bvec
;
7830 struct extent_io_tree
*io_tree
, *failure_tree
;
7832 struct bvec_iter_all iter_all
;
7837 ASSERT(bio
->bi_vcnt
== 1);
7838 io_tree
= &BTRFS_I(inode
)->io_tree
;
7839 failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
7840 ASSERT(bio_first_bvec_all(bio
)->bv_len
== btrfs_inode_sectorsize(inode
));
7843 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
7844 bio_for_each_segment_all(bvec
, bio
, i
, iter_all
)
7845 clean_io_failure(BTRFS_I(inode
)->root
->fs_info
, failure_tree
,
7846 io_tree
, done
->start
, bvec
->bv_page
,
7847 btrfs_ino(BTRFS_I(inode
)), 0);
7849 complete(&done
->done
);
7853 static blk_status_t
__btrfs_correct_data_nocsum(struct inode
*inode
,
7854 struct btrfs_io_bio
*io_bio
)
7856 struct btrfs_fs_info
*fs_info
;
7857 struct bio_vec bvec
;
7858 struct bvec_iter iter
;
7859 struct btrfs_retry_complete done
;
7865 blk_status_t err
= BLK_STS_OK
;
7867 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
7868 sectorsize
= fs_info
->sectorsize
;
7870 start
= io_bio
->logical
;
7872 io_bio
->bio
.bi_iter
= io_bio
->iter
;
7874 bio_for_each_segment(bvec
, &io_bio
->bio
, iter
) {
7875 nr_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, bvec
.bv_len
);
7876 pgoff
= bvec
.bv_offset
;
7878 next_block_or_try_again
:
7881 init_completion(&done
.done
);
7883 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
.bv_page
,
7884 pgoff
, start
, start
+ sectorsize
- 1,
7886 btrfs_retry_endio_nocsum
, &done
);
7892 wait_for_completion_io(&done
.done
);
7894 if (!done
.uptodate
) {
7895 /* We might have another mirror, so try again */
7896 goto next_block_or_try_again
;
7900 start
+= sectorsize
;
7904 pgoff
+= sectorsize
;
7905 ASSERT(pgoff
< PAGE_SIZE
);
7906 goto next_block_or_try_again
;
7913 static void btrfs_retry_endio(struct bio
*bio
)
7915 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7916 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
7917 struct extent_io_tree
*io_tree
, *failure_tree
;
7918 struct inode
*inode
= done
->inode
;
7919 struct bio_vec
*bvec
;
7923 struct bvec_iter_all iter_all
;
7930 ASSERT(bio
->bi_vcnt
== 1);
7931 ASSERT(bio_first_bvec_all(bio
)->bv_len
== btrfs_inode_sectorsize(done
->inode
));
7933 io_tree
= &BTRFS_I(inode
)->io_tree
;
7934 failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
7936 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
7937 bio_for_each_segment_all(bvec
, bio
, i
, iter_all
) {
7938 ret
= __readpage_endio_check(inode
, io_bio
, i
, bvec
->bv_page
,
7939 bvec
->bv_offset
, done
->start
,
7942 clean_io_failure(BTRFS_I(inode
)->root
->fs_info
,
7943 failure_tree
, io_tree
, done
->start
,
7945 btrfs_ino(BTRFS_I(inode
)),
7951 done
->uptodate
= uptodate
;
7953 complete(&done
->done
);
7957 static blk_status_t
__btrfs_subio_endio_read(struct inode
*inode
,
7958 struct btrfs_io_bio
*io_bio
, blk_status_t err
)
7960 struct btrfs_fs_info
*fs_info
;
7961 struct bio_vec bvec
;
7962 struct bvec_iter iter
;
7963 struct btrfs_retry_complete done
;
7970 bool uptodate
= (err
== 0);
7972 blk_status_t status
;
7974 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
7975 sectorsize
= fs_info
->sectorsize
;
7978 start
= io_bio
->logical
;
7980 io_bio
->bio
.bi_iter
= io_bio
->iter
;
7982 bio_for_each_segment(bvec
, &io_bio
->bio
, iter
) {
7983 nr_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, bvec
.bv_len
);
7985 pgoff
= bvec
.bv_offset
;
7988 csum_pos
= BTRFS_BYTES_TO_BLKS(fs_info
, offset
);
7989 ret
= __readpage_endio_check(inode
, io_bio
, csum_pos
,
7990 bvec
.bv_page
, pgoff
, start
, sectorsize
);
7997 init_completion(&done
.done
);
7999 status
= dio_read_error(inode
, &io_bio
->bio
, bvec
.bv_page
,
8000 pgoff
, start
, start
+ sectorsize
- 1,
8001 io_bio
->mirror_num
, btrfs_retry_endio
,
8008 wait_for_completion_io(&done
.done
);
8010 if (!done
.uptodate
) {
8011 /* We might have another mirror, so try again */
8015 offset
+= sectorsize
;
8016 start
+= sectorsize
;
8022 pgoff
+= sectorsize
;
8023 ASSERT(pgoff
< PAGE_SIZE
);
8031 static blk_status_t
btrfs_subio_endio_read(struct inode
*inode
,
8032 struct btrfs_io_bio
*io_bio
, blk_status_t err
)
8034 bool skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
8038 return __btrfs_correct_data_nocsum(inode
, io_bio
);
8042 return __btrfs_subio_endio_read(inode
, io_bio
, err
);
8046 static void btrfs_endio_direct_read(struct bio
*bio
)
8048 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8049 struct inode
*inode
= dip
->inode
;
8050 struct bio
*dio_bio
;
8051 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8052 blk_status_t err
= bio
->bi_status
;
8054 if (dip
->flags
& BTRFS_DIO_ORIG_BIO_SUBMITTED
)
8055 err
= btrfs_subio_endio_read(inode
, io_bio
, err
);
8057 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
8058 dip
->logical_offset
+ dip
->bytes
- 1);
8059 dio_bio
= dip
->dio_bio
;
8063 dio_bio
->bi_status
= err
;
8064 dio_end_io(dio_bio
);
8065 btrfs_io_bio_free_csum(io_bio
);
8069 static void __endio_write_update_ordered(struct inode
*inode
,
8070 const u64 offset
, const u64 bytes
,
8071 const bool uptodate
)
8073 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8074 struct btrfs_ordered_extent
*ordered
= NULL
;
8075 struct btrfs_workqueue
*wq
;
8076 btrfs_work_func_t func
;
8077 u64 ordered_offset
= offset
;
8078 u64 ordered_bytes
= bytes
;
8081 if (btrfs_is_free_space_inode(BTRFS_I(inode
))) {
8082 wq
= fs_info
->endio_freespace_worker
;
8083 func
= btrfs_freespace_write_helper
;
8085 wq
= fs_info
->endio_write_workers
;
8086 func
= btrfs_endio_write_helper
;
8089 while (ordered_offset
< offset
+ bytes
) {
8090 last_offset
= ordered_offset
;
8091 if (btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
8095 btrfs_init_work(&ordered
->work
, func
,
8098 btrfs_queue_work(wq
, &ordered
->work
);
8101 * If btrfs_dec_test_ordered_pending does not find any ordered
8102 * extent in the range, we can exit.
8104 if (ordered_offset
== last_offset
)
8107 * Our bio might span multiple ordered extents. In this case
8108 * we keep going until we have accounted the whole dio.
8110 if (ordered_offset
< offset
+ bytes
) {
8111 ordered_bytes
= offset
+ bytes
- ordered_offset
;
8117 static void btrfs_endio_direct_write(struct bio
*bio
)
8119 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8120 struct bio
*dio_bio
= dip
->dio_bio
;
8122 __endio_write_update_ordered(dip
->inode
, dip
->logical_offset
,
8123 dip
->bytes
, !bio
->bi_status
);
8127 dio_bio
->bi_status
= bio
->bi_status
;
8128 dio_end_io(dio_bio
);
8132 static blk_status_t
btrfs_submit_bio_start_direct_io(void *private_data
,
8133 struct bio
*bio
, u64 offset
)
8135 struct inode
*inode
= private_data
;
8137 ret
= btrfs_csum_one_bio(inode
, bio
, offset
, 1);
8138 BUG_ON(ret
); /* -ENOMEM */
8142 static void btrfs_end_dio_bio(struct bio
*bio
)
8144 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8145 blk_status_t err
= bio
->bi_status
;
8148 btrfs_warn(BTRFS_I(dip
->inode
)->root
->fs_info
,
8149 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8150 btrfs_ino(BTRFS_I(dip
->inode
)), bio_op(bio
),
8152 (unsigned long long)bio
->bi_iter
.bi_sector
,
8153 bio
->bi_iter
.bi_size
, err
);
8155 if (dip
->subio_endio
)
8156 err
= dip
->subio_endio(dip
->inode
, btrfs_io_bio(bio
), err
);
8160 * We want to perceive the errors flag being set before
8161 * decrementing the reference count. We don't need a barrier
8162 * since atomic operations with a return value are fully
8163 * ordered as per atomic_t.txt
8168 /* if there are more bios still pending for this dio, just exit */
8169 if (!atomic_dec_and_test(&dip
->pending_bios
))
8173 bio_io_error(dip
->orig_bio
);
8175 dip
->dio_bio
->bi_status
= BLK_STS_OK
;
8176 bio_endio(dip
->orig_bio
);
8182 static inline blk_status_t
btrfs_lookup_and_bind_dio_csum(struct inode
*inode
,
8183 struct btrfs_dio_private
*dip
,
8187 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8188 struct btrfs_io_bio
*orig_io_bio
= btrfs_io_bio(dip
->orig_bio
);
8192 * We load all the csum data we need when we submit
8193 * the first bio to reduce the csum tree search and
8196 if (dip
->logical_offset
== file_offset
) {
8197 ret
= btrfs_lookup_bio_sums_dio(inode
, dip
->orig_bio
,
8203 if (bio
== dip
->orig_bio
)
8206 file_offset
-= dip
->logical_offset
;
8207 file_offset
>>= inode
->i_sb
->s_blocksize_bits
;
8208 io_bio
->csum
= (u8
*)(((u32
*)orig_io_bio
->csum
) + file_offset
);
8213 static inline blk_status_t
btrfs_submit_dio_bio(struct bio
*bio
,
8214 struct inode
*inode
, u64 file_offset
, int async_submit
)
8216 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8217 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8218 bool write
= bio_op(bio
) == REQ_OP_WRITE
;
8221 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8223 async_submit
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
8226 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, BTRFS_WQ_ENDIO_DATA
);
8231 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
8234 if (write
&& async_submit
) {
8235 ret
= btrfs_wq_submit_bio(fs_info
, bio
, 0, 0,
8237 btrfs_submit_bio_start_direct_io
);
8241 * If we aren't doing async submit, calculate the csum of the
8244 ret
= btrfs_csum_one_bio(inode
, bio
, file_offset
, 1);
8248 ret
= btrfs_lookup_and_bind_dio_csum(inode
, dip
, bio
,
8254 ret
= btrfs_map_bio(fs_info
, bio
, 0, 0);
8259 static int btrfs_submit_direct_hook(struct btrfs_dio_private
*dip
)
8261 struct inode
*inode
= dip
->inode
;
8262 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8264 struct bio
*orig_bio
= dip
->orig_bio
;
8265 u64 start_sector
= orig_bio
->bi_iter
.bi_sector
;
8266 u64 file_offset
= dip
->logical_offset
;
8268 int async_submit
= 0;
8270 int clone_offset
= 0;
8273 blk_status_t status
;
8275 map_length
= orig_bio
->bi_iter
.bi_size
;
8276 submit_len
= map_length
;
8277 ret
= btrfs_map_block(fs_info
, btrfs_op(orig_bio
), start_sector
<< 9,
8278 &map_length
, NULL
, 0);
8282 if (map_length
>= submit_len
) {
8284 dip
->flags
|= BTRFS_DIO_ORIG_BIO_SUBMITTED
;
8288 /* async crcs make it difficult to collect full stripe writes. */
8289 if (btrfs_data_alloc_profile(fs_info
) & BTRFS_BLOCK_GROUP_RAID56_MASK
)
8295 ASSERT(map_length
<= INT_MAX
);
8296 atomic_inc(&dip
->pending_bios
);
8298 clone_len
= min_t(int, submit_len
, map_length
);
8301 * This will never fail as it's passing GPF_NOFS and
8302 * the allocation is backed by btrfs_bioset.
8304 bio
= btrfs_bio_clone_partial(orig_bio
, clone_offset
,
8306 bio
->bi_private
= dip
;
8307 bio
->bi_end_io
= btrfs_end_dio_bio
;
8308 btrfs_io_bio(bio
)->logical
= file_offset
;
8310 ASSERT(submit_len
>= clone_len
);
8311 submit_len
-= clone_len
;
8312 if (submit_len
== 0)
8316 * Increase the count before we submit the bio so we know
8317 * the end IO handler won't happen before we increase the
8318 * count. Otherwise, the dip might get freed before we're
8319 * done setting it up.
8321 atomic_inc(&dip
->pending_bios
);
8323 status
= btrfs_submit_dio_bio(bio
, inode
, file_offset
,
8327 atomic_dec(&dip
->pending_bios
);
8331 clone_offset
+= clone_len
;
8332 start_sector
+= clone_len
>> 9;
8333 file_offset
+= clone_len
;
8335 map_length
= submit_len
;
8336 ret
= btrfs_map_block(fs_info
, btrfs_op(orig_bio
),
8337 start_sector
<< 9, &map_length
, NULL
, 0);
8340 } while (submit_len
> 0);
8343 status
= btrfs_submit_dio_bio(bio
, inode
, file_offset
, async_submit
);
8351 * Before atomic variable goto zero, we must make sure dip->errors is
8352 * perceived to be set. This ordering is ensured by the fact that an
8353 * atomic operations with a return value are fully ordered as per
8356 if (atomic_dec_and_test(&dip
->pending_bios
))
8357 bio_io_error(dip
->orig_bio
);
8359 /* bio_end_io() will handle error, so we needn't return it */
8363 static void btrfs_submit_direct(struct bio
*dio_bio
, struct inode
*inode
,
8366 struct btrfs_dio_private
*dip
= NULL
;
8367 struct bio
*bio
= NULL
;
8368 struct btrfs_io_bio
*io_bio
;
8369 bool write
= (bio_op(dio_bio
) == REQ_OP_WRITE
);
8372 bio
= btrfs_bio_clone(dio_bio
);
8374 dip
= kzalloc(sizeof(*dip
), GFP_NOFS
);
8380 dip
->private = dio_bio
->bi_private
;
8382 dip
->logical_offset
= file_offset
;
8383 dip
->bytes
= dio_bio
->bi_iter
.bi_size
;
8384 dip
->disk_bytenr
= (u64
)dio_bio
->bi_iter
.bi_sector
<< 9;
8385 bio
->bi_private
= dip
;
8386 dip
->orig_bio
= bio
;
8387 dip
->dio_bio
= dio_bio
;
8388 atomic_set(&dip
->pending_bios
, 0);
8389 io_bio
= btrfs_io_bio(bio
);
8390 io_bio
->logical
= file_offset
;
8393 bio
->bi_end_io
= btrfs_endio_direct_write
;
8395 bio
->bi_end_io
= btrfs_endio_direct_read
;
8396 dip
->subio_endio
= btrfs_subio_endio_read
;
8400 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8401 * even if we fail to submit a bio, because in such case we do the
8402 * corresponding error handling below and it must not be done a second
8403 * time by btrfs_direct_IO().
8406 struct btrfs_dio_data
*dio_data
= current
->journal_info
;
8408 dio_data
->unsubmitted_oe_range_end
= dip
->logical_offset
+
8410 dio_data
->unsubmitted_oe_range_start
=
8411 dio_data
->unsubmitted_oe_range_end
;
8414 ret
= btrfs_submit_direct_hook(dip
);
8418 btrfs_io_bio_free_csum(io_bio
);
8422 * If we arrived here it means either we failed to submit the dip
8423 * or we either failed to clone the dio_bio or failed to allocate the
8424 * dip. If we cloned the dio_bio and allocated the dip, we can just
8425 * call bio_endio against our io_bio so that we get proper resource
8426 * cleanup if we fail to submit the dip, otherwise, we must do the
8427 * same as btrfs_endio_direct_[write|read] because we can't call these
8428 * callbacks - they require an allocated dip and a clone of dio_bio.
8433 * The end io callbacks free our dip, do the final put on bio
8434 * and all the cleanup and final put for dio_bio (through
8441 __endio_write_update_ordered(inode
,
8443 dio_bio
->bi_iter
.bi_size
,
8446 unlock_extent(&BTRFS_I(inode
)->io_tree
, file_offset
,
8447 file_offset
+ dio_bio
->bi_iter
.bi_size
- 1);
8449 dio_bio
->bi_status
= BLK_STS_IOERR
;
8451 * Releases and cleans up our dio_bio, no need to bio_put()
8452 * nor bio_endio()/bio_io_error() against dio_bio.
8454 dio_end_io(dio_bio
);
8461 static ssize_t
check_direct_IO(struct btrfs_fs_info
*fs_info
,
8462 const struct iov_iter
*iter
, loff_t offset
)
8466 unsigned int blocksize_mask
= fs_info
->sectorsize
- 1;
8467 ssize_t retval
= -EINVAL
;
8469 if (offset
& blocksize_mask
)
8472 if (iov_iter_alignment(iter
) & blocksize_mask
)
8475 /* If this is a write we don't need to check anymore */
8476 if (iov_iter_rw(iter
) != READ
|| !iter_is_iovec(iter
))
8479 * Check to make sure we don't have duplicate iov_base's in this
8480 * iovec, if so return EINVAL, otherwise we'll get csum errors
8481 * when reading back.
8483 for (seg
= 0; seg
< iter
->nr_segs
; seg
++) {
8484 for (i
= seg
+ 1; i
< iter
->nr_segs
; i
++) {
8485 if (iter
->iov
[seg
].iov_base
== iter
->iov
[i
].iov_base
)
8494 static ssize_t
btrfs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
8496 struct file
*file
= iocb
->ki_filp
;
8497 struct inode
*inode
= file
->f_mapping
->host
;
8498 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8499 struct btrfs_dio_data dio_data
= { 0 };
8500 struct extent_changeset
*data_reserved
= NULL
;
8501 loff_t offset
= iocb
->ki_pos
;
8505 bool relock
= false;
8508 if (check_direct_IO(fs_info
, iter
, offset
))
8511 inode_dio_begin(inode
);
8514 * The generic stuff only does filemap_write_and_wait_range, which
8515 * isn't enough if we've written compressed pages to this area, so
8516 * we need to flush the dirty pages again to make absolutely sure
8517 * that any outstanding dirty pages are on disk.
8519 count
= iov_iter_count(iter
);
8520 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
8521 &BTRFS_I(inode
)->runtime_flags
))
8522 filemap_fdatawrite_range(inode
->i_mapping
, offset
,
8523 offset
+ count
- 1);
8525 if (iov_iter_rw(iter
) == WRITE
) {
8527 * If the write DIO is beyond the EOF, we need update
8528 * the isize, but it is protected by i_mutex. So we can
8529 * not unlock the i_mutex at this case.
8531 if (offset
+ count
<= inode
->i_size
) {
8532 dio_data
.overwrite
= 1;
8533 inode_unlock(inode
);
8535 } else if (iocb
->ki_flags
& IOCB_NOWAIT
) {
8539 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
,
8545 * We need to know how many extents we reserved so that we can
8546 * do the accounting properly if we go over the number we
8547 * originally calculated. Abuse current->journal_info for this.
8549 dio_data
.reserve
= round_up(count
,
8550 fs_info
->sectorsize
);
8551 dio_data
.unsubmitted_oe_range_start
= (u64
)offset
;
8552 dio_data
.unsubmitted_oe_range_end
= (u64
)offset
;
8553 current
->journal_info
= &dio_data
;
8554 down_read(&BTRFS_I(inode
)->dio_sem
);
8555 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK
,
8556 &BTRFS_I(inode
)->runtime_flags
)) {
8557 inode_dio_end(inode
);
8558 flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
8562 ret
= __blockdev_direct_IO(iocb
, inode
,
8563 fs_info
->fs_devices
->latest_bdev
,
8564 iter
, btrfs_get_blocks_direct
, NULL
,
8565 btrfs_submit_direct
, flags
);
8566 if (iov_iter_rw(iter
) == WRITE
) {
8567 up_read(&BTRFS_I(inode
)->dio_sem
);
8568 current
->journal_info
= NULL
;
8569 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
8570 if (dio_data
.reserve
)
8571 btrfs_delalloc_release_space(inode
, data_reserved
,
8572 offset
, dio_data
.reserve
, true);
8574 * On error we might have left some ordered extents
8575 * without submitting corresponding bios for them, so
8576 * cleanup them up to avoid other tasks getting them
8577 * and waiting for them to complete forever.
8579 if (dio_data
.unsubmitted_oe_range_start
<
8580 dio_data
.unsubmitted_oe_range_end
)
8581 __endio_write_update_ordered(inode
,
8582 dio_data
.unsubmitted_oe_range_start
,
8583 dio_data
.unsubmitted_oe_range_end
-
8584 dio_data
.unsubmitted_oe_range_start
,
8586 } else if (ret
>= 0 && (size_t)ret
< count
)
8587 btrfs_delalloc_release_space(inode
, data_reserved
,
8588 offset
, count
- (size_t)ret
, true);
8589 btrfs_delalloc_release_extents(BTRFS_I(inode
), count
, false);
8593 inode_dio_end(inode
);
8597 extent_changeset_free(data_reserved
);
8601 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8603 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
8604 __u64 start
, __u64 len
)
8608 ret
= fiemap_check_flags(fieinfo
, BTRFS_FIEMAP_FLAGS
);
8612 return extent_fiemap(inode
, fieinfo
, start
, len
);
8615 int btrfs_readpage(struct file
*file
, struct page
*page
)
8617 struct extent_io_tree
*tree
;
8618 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8619 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
8622 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
8624 struct inode
*inode
= page
->mapping
->host
;
8627 if (current
->flags
& PF_MEMALLOC
) {
8628 redirty_page_for_writepage(wbc
, page
);
8634 * If we are under memory pressure we will call this directly from the
8635 * VM, we need to make sure we have the inode referenced for the ordered
8636 * extent. If not just return like we didn't do anything.
8638 if (!igrab(inode
)) {
8639 redirty_page_for_writepage(wbc
, page
);
8640 return AOP_WRITEPAGE_ACTIVATE
;
8642 ret
= extent_write_full_page(page
, wbc
);
8643 btrfs_add_delayed_iput(inode
);
8647 static int btrfs_writepages(struct address_space
*mapping
,
8648 struct writeback_control
*wbc
)
8650 return extent_writepages(mapping
, wbc
);
8654 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
8655 struct list_head
*pages
, unsigned nr_pages
)
8657 return extent_readpages(mapping
, pages
, nr_pages
);
8660 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8662 int ret
= try_release_extent_mapping(page
, gfp_flags
);
8664 ClearPagePrivate(page
);
8665 set_page_private(page
, 0);
8671 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8673 if (PageWriteback(page
) || PageDirty(page
))
8675 return __btrfs_releasepage(page
, gfp_flags
);
8678 static void btrfs_invalidatepage(struct page
*page
, unsigned int offset
,
8679 unsigned int length
)
8681 struct inode
*inode
= page
->mapping
->host
;
8682 struct extent_io_tree
*tree
;
8683 struct btrfs_ordered_extent
*ordered
;
8684 struct extent_state
*cached_state
= NULL
;
8685 u64 page_start
= page_offset(page
);
8686 u64 page_end
= page_start
+ PAGE_SIZE
- 1;
8689 int inode_evicting
= inode
->i_state
& I_FREEING
;
8692 * we have the page locked, so new writeback can't start,
8693 * and the dirty bit won't be cleared while we are here.
8695 * Wait for IO on this page so that we can safely clear
8696 * the PagePrivate2 bit and do ordered accounting
8698 wait_on_page_writeback(page
);
8700 tree
= &BTRFS_I(inode
)->io_tree
;
8702 btrfs_releasepage(page
, GFP_NOFS
);
8706 if (!inode_evicting
)
8707 lock_extent_bits(tree
, page_start
, page_end
, &cached_state
);
8710 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), start
,
8711 page_end
- start
+ 1);
8713 end
= min(page_end
, ordered
->file_offset
+ ordered
->len
- 1);
8715 * IO on this page will never be started, so we need
8716 * to account for any ordered extents now
8718 if (!inode_evicting
)
8719 clear_extent_bit(tree
, start
, end
,
8720 EXTENT_DIRTY
| EXTENT_DELALLOC
|
8721 EXTENT_DELALLOC_NEW
|
8722 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
8723 EXTENT_DEFRAG
, 1, 0, &cached_state
);
8725 * whoever cleared the private bit is responsible
8726 * for the finish_ordered_io
8728 if (TestClearPagePrivate2(page
)) {
8729 struct btrfs_ordered_inode_tree
*tree
;
8732 tree
= &BTRFS_I(inode
)->ordered_tree
;
8734 spin_lock_irq(&tree
->lock
);
8735 set_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
);
8736 new_len
= start
- ordered
->file_offset
;
8737 if (new_len
< ordered
->truncated_len
)
8738 ordered
->truncated_len
= new_len
;
8739 spin_unlock_irq(&tree
->lock
);
8741 if (btrfs_dec_test_ordered_pending(inode
, &ordered
,
8743 end
- start
+ 1, 1))
8744 btrfs_finish_ordered_io(ordered
);
8746 btrfs_put_ordered_extent(ordered
);
8747 if (!inode_evicting
) {
8748 cached_state
= NULL
;
8749 lock_extent_bits(tree
, start
, end
,
8754 if (start
< page_end
)
8759 * Qgroup reserved space handler
8760 * Page here will be either
8761 * 1) Already written to disk
8762 * In this case, its reserved space is released from data rsv map
8763 * and will be freed by delayed_ref handler finally.
8764 * So even we call qgroup_free_data(), it won't decrease reserved
8766 * 2) Not written to disk
8767 * This means the reserved space should be freed here. However,
8768 * if a truncate invalidates the page (by clearing PageDirty)
8769 * and the page is accounted for while allocating extent
8770 * in btrfs_check_data_free_space() we let delayed_ref to
8771 * free the entire extent.
8773 if (PageDirty(page
))
8774 btrfs_qgroup_free_data(inode
, NULL
, page_start
, PAGE_SIZE
);
8775 if (!inode_evicting
) {
8776 clear_extent_bit(tree
, page_start
, page_end
,
8777 EXTENT_LOCKED
| EXTENT_DIRTY
|
8778 EXTENT_DELALLOC
| EXTENT_DELALLOC_NEW
|
8779 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
, 1, 1,
8782 __btrfs_releasepage(page
, GFP_NOFS
);
8785 ClearPageChecked(page
);
8786 if (PagePrivate(page
)) {
8787 ClearPagePrivate(page
);
8788 set_page_private(page
, 0);
8794 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8795 * called from a page fault handler when a page is first dirtied. Hence we must
8796 * be careful to check for EOF conditions here. We set the page up correctly
8797 * for a written page which means we get ENOSPC checking when writing into
8798 * holes and correct delalloc and unwritten extent mapping on filesystems that
8799 * support these features.
8801 * We are not allowed to take the i_mutex here so we have to play games to
8802 * protect against truncate races as the page could now be beyond EOF. Because
8803 * truncate_setsize() writes the inode size before removing pages, once we have
8804 * the page lock we can determine safely if the page is beyond EOF. If it is not
8805 * beyond EOF, then the page is guaranteed safe against truncation until we
8808 vm_fault_t
btrfs_page_mkwrite(struct vm_fault
*vmf
)
8810 struct page
*page
= vmf
->page
;
8811 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
8812 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8813 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
8814 struct btrfs_ordered_extent
*ordered
;
8815 struct extent_state
*cached_state
= NULL
;
8816 struct extent_changeset
*data_reserved
= NULL
;
8818 unsigned long zero_start
;
8828 reserved_space
= PAGE_SIZE
;
8830 sb_start_pagefault(inode
->i_sb
);
8831 page_start
= page_offset(page
);
8832 page_end
= page_start
+ PAGE_SIZE
- 1;
8836 * Reserving delalloc space after obtaining the page lock can lead to
8837 * deadlock. For example, if a dirty page is locked by this function
8838 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8839 * dirty page write out, then the btrfs_writepage() function could
8840 * end up waiting indefinitely to get a lock on the page currently
8841 * being processed by btrfs_page_mkwrite() function.
8843 ret2
= btrfs_delalloc_reserve_space(inode
, &data_reserved
, page_start
,
8846 ret2
= file_update_time(vmf
->vma
->vm_file
);
8850 ret
= vmf_error(ret2
);
8856 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
8859 size
= i_size_read(inode
);
8861 if ((page
->mapping
!= inode
->i_mapping
) ||
8862 (page_start
>= size
)) {
8863 /* page got truncated out from underneath us */
8866 wait_on_page_writeback(page
);
8868 lock_extent_bits(io_tree
, page_start
, page_end
, &cached_state
);
8869 set_page_extent_mapped(page
);
8872 * we can't set the delalloc bits if there are pending ordered
8873 * extents. Drop our locks and wait for them to finish
8875 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), page_start
,
8878 unlock_extent_cached(io_tree
, page_start
, page_end
,
8881 btrfs_start_ordered_extent(inode
, ordered
, 1);
8882 btrfs_put_ordered_extent(ordered
);
8886 if (page
->index
== ((size
- 1) >> PAGE_SHIFT
)) {
8887 reserved_space
= round_up(size
- page_start
,
8888 fs_info
->sectorsize
);
8889 if (reserved_space
< PAGE_SIZE
) {
8890 end
= page_start
+ reserved_space
- 1;
8891 btrfs_delalloc_release_space(inode
, data_reserved
,
8892 page_start
, PAGE_SIZE
- reserved_space
,
8898 * page_mkwrite gets called when the page is firstly dirtied after it's
8899 * faulted in, but write(2) could also dirty a page and set delalloc
8900 * bits, thus in this case for space account reason, we still need to
8901 * clear any delalloc bits within this page range since we have to
8902 * reserve data&meta space before lock_page() (see above comments).
8904 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, end
,
8905 EXTENT_DIRTY
| EXTENT_DELALLOC
|
8906 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
8907 0, 0, &cached_state
);
8909 ret2
= btrfs_set_extent_delalloc(inode
, page_start
, end
, 0,
8912 unlock_extent_cached(io_tree
, page_start
, page_end
,
8914 ret
= VM_FAULT_SIGBUS
;
8919 /* page is wholly or partially inside EOF */
8920 if (page_start
+ PAGE_SIZE
> size
)
8921 zero_start
= offset_in_page(size
);
8923 zero_start
= PAGE_SIZE
;
8925 if (zero_start
!= PAGE_SIZE
) {
8927 memset(kaddr
+ zero_start
, 0, PAGE_SIZE
- zero_start
);
8928 flush_dcache_page(page
);
8931 ClearPageChecked(page
);
8932 set_page_dirty(page
);
8933 SetPageUptodate(page
);
8935 BTRFS_I(inode
)->last_trans
= fs_info
->generation
;
8936 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
8937 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
8939 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
);
8942 btrfs_delalloc_release_extents(BTRFS_I(inode
), PAGE_SIZE
, true);
8943 sb_end_pagefault(inode
->i_sb
);
8944 extent_changeset_free(data_reserved
);
8945 return VM_FAULT_LOCKED
;
8951 btrfs_delalloc_release_extents(BTRFS_I(inode
), PAGE_SIZE
, (ret
!= 0));
8952 btrfs_delalloc_release_space(inode
, data_reserved
, page_start
,
8953 reserved_space
, (ret
!= 0));
8955 sb_end_pagefault(inode
->i_sb
);
8956 extent_changeset_free(data_reserved
);
8960 static int btrfs_truncate(struct inode
*inode
, bool skip_writeback
)
8962 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8963 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8964 struct btrfs_block_rsv
*rsv
;
8966 struct btrfs_trans_handle
*trans
;
8967 u64 mask
= fs_info
->sectorsize
- 1;
8968 u64 min_size
= btrfs_calc_trunc_metadata_size(fs_info
, 1);
8970 if (!skip_writeback
) {
8971 ret
= btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
),
8978 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8979 * things going on here:
8981 * 1) We need to reserve space to update our inode.
8983 * 2) We need to have something to cache all the space that is going to
8984 * be free'd up by the truncate operation, but also have some slack
8985 * space reserved in case it uses space during the truncate (thank you
8986 * very much snapshotting).
8988 * And we need these to be separate. The fact is we can use a lot of
8989 * space doing the truncate, and we have no earthly idea how much space
8990 * we will use, so we need the truncate reservation to be separate so it
8991 * doesn't end up using space reserved for updating the inode. We also
8992 * need to be able to stop the transaction and start a new one, which
8993 * means we need to be able to update the inode several times, and we
8994 * have no idea of knowing how many times that will be, so we can't just
8995 * reserve 1 item for the entirety of the operation, so that has to be
8996 * done separately as well.
8998 * So that leaves us with
9000 * 1) rsv - for the truncate reservation, which we will steal from the
9001 * transaction reservation.
9002 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
9003 * updating the inode.
9005 rsv
= btrfs_alloc_block_rsv(fs_info
, BTRFS_BLOCK_RSV_TEMP
);
9008 rsv
->size
= min_size
;
9012 * 1 for the truncate slack space
9013 * 1 for updating the inode.
9015 trans
= btrfs_start_transaction(root
, 2);
9016 if (IS_ERR(trans
)) {
9017 ret
= PTR_ERR(trans
);
9021 /* Migrate the slack space for the truncate to our reserve */
9022 ret
= btrfs_block_rsv_migrate(&fs_info
->trans_block_rsv
, rsv
,
9027 * So if we truncate and then write and fsync we normally would just
9028 * write the extents that changed, which is a problem if we need to
9029 * first truncate that entire inode. So set this flag so we write out
9030 * all of the extents in the inode to the sync log so we're completely
9033 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
9034 trans
->block_rsv
= rsv
;
9037 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
9039 BTRFS_EXTENT_DATA_KEY
);
9040 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
9041 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
)
9044 ret
= btrfs_update_inode(trans
, root
, inode
);
9048 btrfs_end_transaction(trans
);
9049 btrfs_btree_balance_dirty(fs_info
);
9051 trans
= btrfs_start_transaction(root
, 2);
9052 if (IS_ERR(trans
)) {
9053 ret
= PTR_ERR(trans
);
9058 btrfs_block_rsv_release(fs_info
, rsv
, -1);
9059 ret
= btrfs_block_rsv_migrate(&fs_info
->trans_block_rsv
,
9060 rsv
, min_size
, false);
9061 BUG_ON(ret
); /* shouldn't happen */
9062 trans
->block_rsv
= rsv
;
9066 * We can't call btrfs_truncate_block inside a trans handle as we could
9067 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9068 * we've truncated everything except the last little bit, and can do
9069 * btrfs_truncate_block and then update the disk_i_size.
9071 if (ret
== NEED_TRUNCATE_BLOCK
) {
9072 btrfs_end_transaction(trans
);
9073 btrfs_btree_balance_dirty(fs_info
);
9075 ret
= btrfs_truncate_block(inode
, inode
->i_size
, 0, 0);
9078 trans
= btrfs_start_transaction(root
, 1);
9079 if (IS_ERR(trans
)) {
9080 ret
= PTR_ERR(trans
);
9083 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
9089 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
9090 ret2
= btrfs_update_inode(trans
, root
, inode
);
9094 ret2
= btrfs_end_transaction(trans
);
9097 btrfs_btree_balance_dirty(fs_info
);
9100 btrfs_free_block_rsv(fs_info
, rsv
);
9106 * create a new subvolume directory/inode (helper for the ioctl).
9108 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
9109 struct btrfs_root
*new_root
,
9110 struct btrfs_root
*parent_root
,
9113 struct inode
*inode
;
9117 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
9118 new_dirid
, new_dirid
,
9119 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
9122 return PTR_ERR(inode
);
9123 inode
->i_op
= &btrfs_dir_inode_operations
;
9124 inode
->i_fop
= &btrfs_dir_file_operations
;
9126 set_nlink(inode
, 1);
9127 btrfs_i_size_write(BTRFS_I(inode
), 0);
9128 unlock_new_inode(inode
);
9130 err
= btrfs_subvol_inherit_props(trans
, new_root
, parent_root
);
9132 btrfs_err(new_root
->fs_info
,
9133 "error inheriting subvolume %llu properties: %d",
9134 new_root
->root_key
.objectid
, err
);
9136 err
= btrfs_update_inode(trans
, new_root
, inode
);
9142 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
9144 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
9145 struct btrfs_inode
*ei
;
9146 struct inode
*inode
;
9148 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_KERNEL
);
9155 ei
->last_sub_trans
= 0;
9156 ei
->logged_trans
= 0;
9157 ei
->delalloc_bytes
= 0;
9158 ei
->new_delalloc_bytes
= 0;
9159 ei
->defrag_bytes
= 0;
9160 ei
->disk_i_size
= 0;
9163 ei
->index_cnt
= (u64
)-1;
9165 ei
->last_unlink_trans
= 0;
9166 ei
->last_link_trans
= 0;
9167 ei
->last_log_commit
= 0;
9169 spin_lock_init(&ei
->lock
);
9170 ei
->outstanding_extents
= 0;
9171 if (sb
->s_magic
!= BTRFS_TEST_MAGIC
)
9172 btrfs_init_metadata_block_rsv(fs_info
, &ei
->block_rsv
,
9173 BTRFS_BLOCK_RSV_DELALLOC
);
9174 ei
->runtime_flags
= 0;
9175 ei
->prop_compress
= BTRFS_COMPRESS_NONE
;
9176 ei
->defrag_compress
= BTRFS_COMPRESS_NONE
;
9178 ei
->delayed_node
= NULL
;
9180 ei
->i_otime
.tv_sec
= 0;
9181 ei
->i_otime
.tv_nsec
= 0;
9183 inode
= &ei
->vfs_inode
;
9184 extent_map_tree_init(&ei
->extent_tree
);
9185 extent_io_tree_init(&ei
->io_tree
, inode
);
9186 extent_io_tree_init(&ei
->io_failure_tree
, inode
);
9187 ei
->io_tree
.track_uptodate
= 1;
9188 ei
->io_failure_tree
.track_uptodate
= 1;
9189 atomic_set(&ei
->sync_writers
, 0);
9190 mutex_init(&ei
->log_mutex
);
9191 mutex_init(&ei
->delalloc_mutex
);
9192 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
9193 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
9194 INIT_LIST_HEAD(&ei
->delayed_iput
);
9195 RB_CLEAR_NODE(&ei
->rb_node
);
9196 init_rwsem(&ei
->dio_sem
);
9201 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9202 void btrfs_test_destroy_inode(struct inode
*inode
)
9204 btrfs_drop_extent_cache(BTRFS_I(inode
), 0, (u64
)-1, 0);
9205 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9209 static void btrfs_i_callback(struct rcu_head
*head
)
9211 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
9212 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9215 void btrfs_destroy_inode(struct inode
*inode
)
9217 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
9218 struct btrfs_ordered_extent
*ordered
;
9219 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9221 WARN_ON(!hlist_empty(&inode
->i_dentry
));
9222 WARN_ON(inode
->i_data
.nrpages
);
9223 WARN_ON(BTRFS_I(inode
)->block_rsv
.reserved
);
9224 WARN_ON(BTRFS_I(inode
)->block_rsv
.size
);
9225 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
9226 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
9227 WARN_ON(BTRFS_I(inode
)->new_delalloc_bytes
);
9228 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
9229 WARN_ON(BTRFS_I(inode
)->defrag_bytes
);
9232 * This can happen where we create an inode, but somebody else also
9233 * created the same inode and we need to destroy the one we already
9240 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
9245 "found ordered extent %llu %llu on inode cleanup",
9246 ordered
->file_offset
, ordered
->len
);
9247 btrfs_remove_ordered_extent(inode
, ordered
);
9248 btrfs_put_ordered_extent(ordered
);
9249 btrfs_put_ordered_extent(ordered
);
9252 btrfs_qgroup_check_reserved_leak(inode
);
9253 inode_tree_del(inode
);
9254 btrfs_drop_extent_cache(BTRFS_I(inode
), 0, (u64
)-1, 0);
9256 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
9259 int btrfs_drop_inode(struct inode
*inode
)
9261 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9266 /* the snap/subvol tree is on deleting */
9267 if (btrfs_root_refs(&root
->root_item
) == 0)
9270 return generic_drop_inode(inode
);
9273 static void init_once(void *foo
)
9275 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
9277 inode_init_once(&ei
->vfs_inode
);
9280 void __cold
btrfs_destroy_cachep(void)
9283 * Make sure all delayed rcu free inodes are flushed before we
9287 kmem_cache_destroy(btrfs_inode_cachep
);
9288 kmem_cache_destroy(btrfs_trans_handle_cachep
);
9289 kmem_cache_destroy(btrfs_path_cachep
);
9290 kmem_cache_destroy(btrfs_free_space_cachep
);
9293 int __init
btrfs_init_cachep(void)
9295 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
9296 sizeof(struct btrfs_inode
), 0,
9297 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
| SLAB_ACCOUNT
,
9299 if (!btrfs_inode_cachep
)
9302 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
9303 sizeof(struct btrfs_trans_handle
), 0,
9304 SLAB_TEMPORARY
| SLAB_MEM_SPREAD
, NULL
);
9305 if (!btrfs_trans_handle_cachep
)
9308 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
9309 sizeof(struct btrfs_path
), 0,
9310 SLAB_MEM_SPREAD
, NULL
);
9311 if (!btrfs_path_cachep
)
9314 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
9315 sizeof(struct btrfs_free_space
), 0,
9316 SLAB_MEM_SPREAD
, NULL
);
9317 if (!btrfs_free_space_cachep
)
9322 btrfs_destroy_cachep();
9326 static int btrfs_getattr(const struct path
*path
, struct kstat
*stat
,
9327 u32 request_mask
, unsigned int flags
)
9330 struct inode
*inode
= d_inode(path
->dentry
);
9331 u32 blocksize
= inode
->i_sb
->s_blocksize
;
9332 u32 bi_flags
= BTRFS_I(inode
)->flags
;
9334 stat
->result_mask
|= STATX_BTIME
;
9335 stat
->btime
.tv_sec
= BTRFS_I(inode
)->i_otime
.tv_sec
;
9336 stat
->btime
.tv_nsec
= BTRFS_I(inode
)->i_otime
.tv_nsec
;
9337 if (bi_flags
& BTRFS_INODE_APPEND
)
9338 stat
->attributes
|= STATX_ATTR_APPEND
;
9339 if (bi_flags
& BTRFS_INODE_COMPRESS
)
9340 stat
->attributes
|= STATX_ATTR_COMPRESSED
;
9341 if (bi_flags
& BTRFS_INODE_IMMUTABLE
)
9342 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
9343 if (bi_flags
& BTRFS_INODE_NODUMP
)
9344 stat
->attributes
|= STATX_ATTR_NODUMP
;
9346 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
9347 STATX_ATTR_COMPRESSED
|
9348 STATX_ATTR_IMMUTABLE
|
9351 generic_fillattr(inode
, stat
);
9352 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
9354 spin_lock(&BTRFS_I(inode
)->lock
);
9355 delalloc_bytes
= BTRFS_I(inode
)->new_delalloc_bytes
;
9356 spin_unlock(&BTRFS_I(inode
)->lock
);
9357 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
9358 ALIGN(delalloc_bytes
, blocksize
)) >> 9;
9362 static int btrfs_rename_exchange(struct inode
*old_dir
,
9363 struct dentry
*old_dentry
,
9364 struct inode
*new_dir
,
9365 struct dentry
*new_dentry
)
9367 struct btrfs_fs_info
*fs_info
= btrfs_sb(old_dir
->i_sb
);
9368 struct btrfs_trans_handle
*trans
;
9369 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9370 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9371 struct inode
*new_inode
= new_dentry
->d_inode
;
9372 struct inode
*old_inode
= old_dentry
->d_inode
;
9373 struct timespec64 ctime
= current_time(old_inode
);
9374 struct dentry
*parent
;
9375 u64 old_ino
= btrfs_ino(BTRFS_I(old_inode
));
9376 u64 new_ino
= btrfs_ino(BTRFS_I(new_inode
));
9381 bool root_log_pinned
= false;
9382 bool dest_log_pinned
= false;
9383 struct btrfs_log_ctx ctx_root
;
9384 struct btrfs_log_ctx ctx_dest
;
9385 bool sync_log_root
= false;
9386 bool sync_log_dest
= false;
9387 bool commit_transaction
= false;
9389 /* we only allow rename subvolume link between subvolumes */
9390 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9393 btrfs_init_log_ctx(&ctx_root
, old_inode
);
9394 btrfs_init_log_ctx(&ctx_dest
, new_inode
);
9396 /* close the race window with snapshot create/destroy ioctl */
9397 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9398 down_read(&fs_info
->subvol_sem
);
9399 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9400 down_read(&fs_info
->subvol_sem
);
9403 * We want to reserve the absolute worst case amount of items. So if
9404 * both inodes are subvols and we need to unlink them then that would
9405 * require 4 item modifications, but if they are both normal inodes it
9406 * would require 5 item modifications, so we'll assume their normal
9407 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9408 * should cover the worst case number of items we'll modify.
9410 trans
= btrfs_start_transaction(root
, 12);
9411 if (IS_ERR(trans
)) {
9412 ret
= PTR_ERR(trans
);
9417 * We need to find a free sequence number both in the source and
9418 * in the destination directory for the exchange.
9420 ret
= btrfs_set_inode_index(BTRFS_I(new_dir
), &old_idx
);
9423 ret
= btrfs_set_inode_index(BTRFS_I(old_dir
), &new_idx
);
9427 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9428 BTRFS_I(new_inode
)->dir_index
= 0ULL;
9430 /* Reference for the source. */
9431 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9432 /* force full log commit if subvolume involved. */
9433 btrfs_set_log_full_commit(fs_info
, trans
);
9435 btrfs_pin_log_trans(root
);
9436 root_log_pinned
= true;
9437 ret
= btrfs_insert_inode_ref(trans
, dest
,
9438 new_dentry
->d_name
.name
,
9439 new_dentry
->d_name
.len
,
9441 btrfs_ino(BTRFS_I(new_dir
)),
9447 /* And now for the dest. */
9448 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9449 /* force full log commit if subvolume involved. */
9450 btrfs_set_log_full_commit(fs_info
, trans
);
9452 btrfs_pin_log_trans(dest
);
9453 dest_log_pinned
= true;
9454 ret
= btrfs_insert_inode_ref(trans
, root
,
9455 old_dentry
->d_name
.name
,
9456 old_dentry
->d_name
.len
,
9458 btrfs_ino(BTRFS_I(old_dir
)),
9464 /* Update inode version and ctime/mtime. */
9465 inode_inc_iversion(old_dir
);
9466 inode_inc_iversion(new_dir
);
9467 inode_inc_iversion(old_inode
);
9468 inode_inc_iversion(new_inode
);
9469 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
9470 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
9471 old_inode
->i_ctime
= ctime
;
9472 new_inode
->i_ctime
= ctime
;
9474 if (old_dentry
->d_parent
!= new_dentry
->d_parent
) {
9475 btrfs_record_unlink_dir(trans
, BTRFS_I(old_dir
),
9476 BTRFS_I(old_inode
), 1);
9477 btrfs_record_unlink_dir(trans
, BTRFS_I(new_dir
),
9478 BTRFS_I(new_inode
), 1);
9481 /* src is a subvolume */
9482 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9483 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
9484 ret
= btrfs_unlink_subvol(trans
, old_dir
, root_objectid
,
9485 old_dentry
->d_name
.name
,
9486 old_dentry
->d_name
.len
);
9487 } else { /* src is an inode */
9488 ret
= __btrfs_unlink_inode(trans
, root
, BTRFS_I(old_dir
),
9489 BTRFS_I(old_dentry
->d_inode
),
9490 old_dentry
->d_name
.name
,
9491 old_dentry
->d_name
.len
);
9493 ret
= btrfs_update_inode(trans
, root
, old_inode
);
9496 btrfs_abort_transaction(trans
, ret
);
9500 /* dest is a subvolume */
9501 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9502 root_objectid
= BTRFS_I(new_inode
)->root
->root_key
.objectid
;
9503 ret
= btrfs_unlink_subvol(trans
, new_dir
, root_objectid
,
9504 new_dentry
->d_name
.name
,
9505 new_dentry
->d_name
.len
);
9506 } else { /* dest is an inode */
9507 ret
= __btrfs_unlink_inode(trans
, dest
, BTRFS_I(new_dir
),
9508 BTRFS_I(new_dentry
->d_inode
),
9509 new_dentry
->d_name
.name
,
9510 new_dentry
->d_name
.len
);
9512 ret
= btrfs_update_inode(trans
, dest
, new_inode
);
9515 btrfs_abort_transaction(trans
, ret
);
9519 ret
= btrfs_add_link(trans
, BTRFS_I(new_dir
), BTRFS_I(old_inode
),
9520 new_dentry
->d_name
.name
,
9521 new_dentry
->d_name
.len
, 0, old_idx
);
9523 btrfs_abort_transaction(trans
, ret
);
9527 ret
= btrfs_add_link(trans
, BTRFS_I(old_dir
), BTRFS_I(new_inode
),
9528 old_dentry
->d_name
.name
,
9529 old_dentry
->d_name
.len
, 0, new_idx
);
9531 btrfs_abort_transaction(trans
, ret
);
9535 if (old_inode
->i_nlink
== 1)
9536 BTRFS_I(old_inode
)->dir_index
= old_idx
;
9537 if (new_inode
->i_nlink
== 1)
9538 BTRFS_I(new_inode
)->dir_index
= new_idx
;
9540 if (root_log_pinned
) {
9541 parent
= new_dentry
->d_parent
;
9542 ret
= btrfs_log_new_name(trans
, BTRFS_I(old_inode
),
9543 BTRFS_I(old_dir
), parent
,
9545 if (ret
== BTRFS_NEED_LOG_SYNC
)
9546 sync_log_root
= true;
9547 else if (ret
== BTRFS_NEED_TRANS_COMMIT
)
9548 commit_transaction
= true;
9550 btrfs_end_log_trans(root
);
9551 root_log_pinned
= false;
9553 if (dest_log_pinned
) {
9554 if (!commit_transaction
) {
9555 parent
= old_dentry
->d_parent
;
9556 ret
= btrfs_log_new_name(trans
, BTRFS_I(new_inode
),
9557 BTRFS_I(new_dir
), parent
,
9559 if (ret
== BTRFS_NEED_LOG_SYNC
)
9560 sync_log_dest
= true;
9561 else if (ret
== BTRFS_NEED_TRANS_COMMIT
)
9562 commit_transaction
= true;
9565 btrfs_end_log_trans(dest
);
9566 dest_log_pinned
= false;
9570 * If we have pinned a log and an error happened, we unpin tasks
9571 * trying to sync the log and force them to fallback to a transaction
9572 * commit if the log currently contains any of the inodes involved in
9573 * this rename operation (to ensure we do not persist a log with an
9574 * inconsistent state for any of these inodes or leading to any
9575 * inconsistencies when replayed). If the transaction was aborted, the
9576 * abortion reason is propagated to userspace when attempting to commit
9577 * the transaction. If the log does not contain any of these inodes, we
9578 * allow the tasks to sync it.
9580 if (ret
&& (root_log_pinned
|| dest_log_pinned
)) {
9581 if (btrfs_inode_in_log(BTRFS_I(old_dir
), fs_info
->generation
) ||
9582 btrfs_inode_in_log(BTRFS_I(new_dir
), fs_info
->generation
) ||
9583 btrfs_inode_in_log(BTRFS_I(old_inode
), fs_info
->generation
) ||
9585 btrfs_inode_in_log(BTRFS_I(new_inode
), fs_info
->generation
)))
9586 btrfs_set_log_full_commit(fs_info
, trans
);
9588 if (root_log_pinned
) {
9589 btrfs_end_log_trans(root
);
9590 root_log_pinned
= false;
9592 if (dest_log_pinned
) {
9593 btrfs_end_log_trans(dest
);
9594 dest_log_pinned
= false;
9597 if (!ret
&& sync_log_root
&& !commit_transaction
) {
9598 ret
= btrfs_sync_log(trans
, BTRFS_I(old_inode
)->root
,
9601 commit_transaction
= true;
9603 if (!ret
&& sync_log_dest
&& !commit_transaction
) {
9604 ret
= btrfs_sync_log(trans
, BTRFS_I(new_inode
)->root
,
9607 commit_transaction
= true;
9609 if (commit_transaction
) {
9610 ret
= btrfs_commit_transaction(trans
);
9614 ret2
= btrfs_end_transaction(trans
);
9615 ret
= ret
? ret
: ret2
;
9618 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9619 up_read(&fs_info
->subvol_sem
);
9620 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9621 up_read(&fs_info
->subvol_sem
);
9626 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle
*trans
,
9627 struct btrfs_root
*root
,
9629 struct dentry
*dentry
)
9632 struct inode
*inode
;
9636 ret
= btrfs_find_free_ino(root
, &objectid
);
9640 inode
= btrfs_new_inode(trans
, root
, dir
,
9641 dentry
->d_name
.name
,
9643 btrfs_ino(BTRFS_I(dir
)),
9645 S_IFCHR
| WHITEOUT_MODE
,
9648 if (IS_ERR(inode
)) {
9649 ret
= PTR_ERR(inode
);
9653 inode
->i_op
= &btrfs_special_inode_operations
;
9654 init_special_inode(inode
, inode
->i_mode
,
9657 ret
= btrfs_init_inode_security(trans
, inode
, dir
,
9662 ret
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
,
9663 BTRFS_I(inode
), 0, index
);
9667 ret
= btrfs_update_inode(trans
, root
, inode
);
9669 unlock_new_inode(inode
);
9671 inode_dec_link_count(inode
);
9677 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
9678 struct inode
*new_dir
, struct dentry
*new_dentry
,
9681 struct btrfs_fs_info
*fs_info
= btrfs_sb(old_dir
->i_sb
);
9682 struct btrfs_trans_handle
*trans
;
9683 unsigned int trans_num_items
;
9684 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9685 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9686 struct inode
*new_inode
= d_inode(new_dentry
);
9687 struct inode
*old_inode
= d_inode(old_dentry
);
9691 u64 old_ino
= btrfs_ino(BTRFS_I(old_inode
));
9692 bool log_pinned
= false;
9693 struct btrfs_log_ctx ctx
;
9694 bool sync_log
= false;
9695 bool commit_transaction
= false;
9697 if (btrfs_ino(BTRFS_I(new_dir
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
9700 /* we only allow rename subvolume link between subvolumes */
9701 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9704 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
9705 (new_inode
&& btrfs_ino(BTRFS_I(new_inode
)) == BTRFS_FIRST_FREE_OBJECTID
))
9708 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
9709 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
9713 /* check for collisions, even if the name isn't there */
9714 ret
= btrfs_check_dir_item_collision(dest
, new_dir
->i_ino
,
9715 new_dentry
->d_name
.name
,
9716 new_dentry
->d_name
.len
);
9719 if (ret
== -EEXIST
) {
9721 * eexist without a new_inode */
9722 if (WARN_ON(!new_inode
)) {
9726 /* maybe -EOVERFLOW */
9733 * we're using rename to replace one file with another. Start IO on it
9734 * now so we don't add too much work to the end of the transaction
9736 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
)
9737 filemap_flush(old_inode
->i_mapping
);
9739 /* close the racy window with snapshot create/destroy ioctl */
9740 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9741 down_read(&fs_info
->subvol_sem
);
9743 * We want to reserve the absolute worst case amount of items. So if
9744 * both inodes are subvols and we need to unlink them then that would
9745 * require 4 item modifications, but if they are both normal inodes it
9746 * would require 5 item modifications, so we'll assume they are normal
9747 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9748 * should cover the worst case number of items we'll modify.
9749 * If our rename has the whiteout flag, we need more 5 units for the
9750 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9751 * when selinux is enabled).
9753 trans_num_items
= 11;
9754 if (flags
& RENAME_WHITEOUT
)
9755 trans_num_items
+= 5;
9756 trans
= btrfs_start_transaction(root
, trans_num_items
);
9757 if (IS_ERR(trans
)) {
9758 ret
= PTR_ERR(trans
);
9763 btrfs_record_root_in_trans(trans
, dest
);
9765 ret
= btrfs_set_inode_index(BTRFS_I(new_dir
), &index
);
9769 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9770 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9771 /* force full log commit if subvolume involved. */
9772 btrfs_set_log_full_commit(fs_info
, trans
);
9774 btrfs_pin_log_trans(root
);
9776 ret
= btrfs_insert_inode_ref(trans
, dest
,
9777 new_dentry
->d_name
.name
,
9778 new_dentry
->d_name
.len
,
9780 btrfs_ino(BTRFS_I(new_dir
)), index
);
9785 inode_inc_iversion(old_dir
);
9786 inode_inc_iversion(new_dir
);
9787 inode_inc_iversion(old_inode
);
9788 old_dir
->i_ctime
= old_dir
->i_mtime
=
9789 new_dir
->i_ctime
= new_dir
->i_mtime
=
9790 old_inode
->i_ctime
= current_time(old_dir
);
9792 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
9793 btrfs_record_unlink_dir(trans
, BTRFS_I(old_dir
),
9794 BTRFS_I(old_inode
), 1);
9796 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9797 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
9798 ret
= btrfs_unlink_subvol(trans
, old_dir
, root_objectid
,
9799 old_dentry
->d_name
.name
,
9800 old_dentry
->d_name
.len
);
9802 ret
= __btrfs_unlink_inode(trans
, root
, BTRFS_I(old_dir
),
9803 BTRFS_I(d_inode(old_dentry
)),
9804 old_dentry
->d_name
.name
,
9805 old_dentry
->d_name
.len
);
9807 ret
= btrfs_update_inode(trans
, root
, old_inode
);
9810 btrfs_abort_transaction(trans
, ret
);
9815 inode_inc_iversion(new_inode
);
9816 new_inode
->i_ctime
= current_time(new_inode
);
9817 if (unlikely(btrfs_ino(BTRFS_I(new_inode
)) ==
9818 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
9819 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
9820 ret
= btrfs_unlink_subvol(trans
, new_dir
, root_objectid
,
9821 new_dentry
->d_name
.name
,
9822 new_dentry
->d_name
.len
);
9823 BUG_ON(new_inode
->i_nlink
== 0);
9825 ret
= btrfs_unlink_inode(trans
, dest
, BTRFS_I(new_dir
),
9826 BTRFS_I(d_inode(new_dentry
)),
9827 new_dentry
->d_name
.name
,
9828 new_dentry
->d_name
.len
);
9830 if (!ret
&& new_inode
->i_nlink
== 0)
9831 ret
= btrfs_orphan_add(trans
,
9832 BTRFS_I(d_inode(new_dentry
)));
9834 btrfs_abort_transaction(trans
, ret
);
9839 ret
= btrfs_add_link(trans
, BTRFS_I(new_dir
), BTRFS_I(old_inode
),
9840 new_dentry
->d_name
.name
,
9841 new_dentry
->d_name
.len
, 0, index
);
9843 btrfs_abort_transaction(trans
, ret
);
9847 if (old_inode
->i_nlink
== 1)
9848 BTRFS_I(old_inode
)->dir_index
= index
;
9851 struct dentry
*parent
= new_dentry
->d_parent
;
9853 btrfs_init_log_ctx(&ctx
, old_inode
);
9854 ret
= btrfs_log_new_name(trans
, BTRFS_I(old_inode
),
9855 BTRFS_I(old_dir
), parent
,
9857 if (ret
== BTRFS_NEED_LOG_SYNC
)
9859 else if (ret
== BTRFS_NEED_TRANS_COMMIT
)
9860 commit_transaction
= true;
9862 btrfs_end_log_trans(root
);
9866 if (flags
& RENAME_WHITEOUT
) {
9867 ret
= btrfs_whiteout_for_rename(trans
, root
, old_dir
,
9871 btrfs_abort_transaction(trans
, ret
);
9877 * If we have pinned the log and an error happened, we unpin tasks
9878 * trying to sync the log and force them to fallback to a transaction
9879 * commit if the log currently contains any of the inodes involved in
9880 * this rename operation (to ensure we do not persist a log with an
9881 * inconsistent state for any of these inodes or leading to any
9882 * inconsistencies when replayed). If the transaction was aborted, the
9883 * abortion reason is propagated to userspace when attempting to commit
9884 * the transaction. If the log does not contain any of these inodes, we
9885 * allow the tasks to sync it.
9887 if (ret
&& log_pinned
) {
9888 if (btrfs_inode_in_log(BTRFS_I(old_dir
), fs_info
->generation
) ||
9889 btrfs_inode_in_log(BTRFS_I(new_dir
), fs_info
->generation
) ||
9890 btrfs_inode_in_log(BTRFS_I(old_inode
), fs_info
->generation
) ||
9892 btrfs_inode_in_log(BTRFS_I(new_inode
), fs_info
->generation
)))
9893 btrfs_set_log_full_commit(fs_info
, trans
);
9895 btrfs_end_log_trans(root
);
9898 if (!ret
&& sync_log
) {
9899 ret
= btrfs_sync_log(trans
, BTRFS_I(old_inode
)->root
, &ctx
);
9901 commit_transaction
= true;
9903 if (commit_transaction
) {
9904 ret
= btrfs_commit_transaction(trans
);
9908 ret2
= btrfs_end_transaction(trans
);
9909 ret
= ret
? ret
: ret2
;
9912 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9913 up_read(&fs_info
->subvol_sem
);
9918 static int btrfs_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
9919 struct inode
*new_dir
, struct dentry
*new_dentry
,
9922 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
9925 if (flags
& RENAME_EXCHANGE
)
9926 return btrfs_rename_exchange(old_dir
, old_dentry
, new_dir
,
9929 return btrfs_rename(old_dir
, old_dentry
, new_dir
, new_dentry
, flags
);
9932 struct btrfs_delalloc_work
{
9933 struct inode
*inode
;
9934 struct completion completion
;
9935 struct list_head list
;
9936 struct btrfs_work work
;
9939 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
9941 struct btrfs_delalloc_work
*delalloc_work
;
9942 struct inode
*inode
;
9944 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
9946 inode
= delalloc_work
->inode
;
9947 filemap_flush(inode
->i_mapping
);
9948 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
9949 &BTRFS_I(inode
)->runtime_flags
))
9950 filemap_flush(inode
->i_mapping
);
9953 complete(&delalloc_work
->completion
);
9956 static struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
)
9958 struct btrfs_delalloc_work
*work
;
9960 work
= kmalloc(sizeof(*work
), GFP_NOFS
);
9964 init_completion(&work
->completion
);
9965 INIT_LIST_HEAD(&work
->list
);
9966 work
->inode
= inode
;
9967 btrfs_init_work(&work
->work
, btrfs_flush_delalloc_helper
,
9968 btrfs_run_delalloc_work
, NULL
, NULL
);
9974 * some fairly slow code that needs optimization. This walks the list
9975 * of all the inodes with pending delalloc and forces them to disk.
9977 static int start_delalloc_inodes(struct btrfs_root
*root
, int nr
, bool snapshot
)
9979 struct btrfs_inode
*binode
;
9980 struct inode
*inode
;
9981 struct btrfs_delalloc_work
*work
, *next
;
9982 struct list_head works
;
9983 struct list_head splice
;
9986 INIT_LIST_HEAD(&works
);
9987 INIT_LIST_HEAD(&splice
);
9989 mutex_lock(&root
->delalloc_mutex
);
9990 spin_lock(&root
->delalloc_lock
);
9991 list_splice_init(&root
->delalloc_inodes
, &splice
);
9992 while (!list_empty(&splice
)) {
9993 binode
= list_entry(splice
.next
, struct btrfs_inode
,
9996 list_move_tail(&binode
->delalloc_inodes
,
9997 &root
->delalloc_inodes
);
9998 inode
= igrab(&binode
->vfs_inode
);
10000 cond_resched_lock(&root
->delalloc_lock
);
10003 spin_unlock(&root
->delalloc_lock
);
10006 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH
,
10007 &binode
->runtime_flags
);
10008 work
= btrfs_alloc_delalloc_work(inode
);
10014 list_add_tail(&work
->list
, &works
);
10015 btrfs_queue_work(root
->fs_info
->flush_workers
,
10018 if (nr
!= -1 && ret
>= nr
)
10021 spin_lock(&root
->delalloc_lock
);
10023 spin_unlock(&root
->delalloc_lock
);
10026 list_for_each_entry_safe(work
, next
, &works
, list
) {
10027 list_del_init(&work
->list
);
10028 wait_for_completion(&work
->completion
);
10032 if (!list_empty(&splice
)) {
10033 spin_lock(&root
->delalloc_lock
);
10034 list_splice_tail(&splice
, &root
->delalloc_inodes
);
10035 spin_unlock(&root
->delalloc_lock
);
10037 mutex_unlock(&root
->delalloc_mutex
);
10041 int btrfs_start_delalloc_snapshot(struct btrfs_root
*root
)
10043 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
10046 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
10049 ret
= start_delalloc_inodes(root
, -1, true);
10055 int btrfs_start_delalloc_roots(struct btrfs_fs_info
*fs_info
, int nr
)
10057 struct btrfs_root
*root
;
10058 struct list_head splice
;
10061 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
10064 INIT_LIST_HEAD(&splice
);
10066 mutex_lock(&fs_info
->delalloc_root_mutex
);
10067 spin_lock(&fs_info
->delalloc_root_lock
);
10068 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
10069 while (!list_empty(&splice
) && nr
) {
10070 root
= list_first_entry(&splice
, struct btrfs_root
,
10072 root
= btrfs_grab_fs_root(root
);
10074 list_move_tail(&root
->delalloc_root
,
10075 &fs_info
->delalloc_roots
);
10076 spin_unlock(&fs_info
->delalloc_root_lock
);
10078 ret
= start_delalloc_inodes(root
, nr
, false);
10079 btrfs_put_fs_root(root
);
10087 spin_lock(&fs_info
->delalloc_root_lock
);
10089 spin_unlock(&fs_info
->delalloc_root_lock
);
10093 if (!list_empty(&splice
)) {
10094 spin_lock(&fs_info
->delalloc_root_lock
);
10095 list_splice_tail(&splice
, &fs_info
->delalloc_roots
);
10096 spin_unlock(&fs_info
->delalloc_root_lock
);
10098 mutex_unlock(&fs_info
->delalloc_root_mutex
);
10102 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
10103 const char *symname
)
10105 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
10106 struct btrfs_trans_handle
*trans
;
10107 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
10108 struct btrfs_path
*path
;
10109 struct btrfs_key key
;
10110 struct inode
*inode
= NULL
;
10117 struct btrfs_file_extent_item
*ei
;
10118 struct extent_buffer
*leaf
;
10120 name_len
= strlen(symname
);
10121 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(fs_info
))
10122 return -ENAMETOOLONG
;
10125 * 2 items for inode item and ref
10126 * 2 items for dir items
10127 * 1 item for updating parent inode item
10128 * 1 item for the inline extent item
10129 * 1 item for xattr if selinux is on
10131 trans
= btrfs_start_transaction(root
, 7);
10133 return PTR_ERR(trans
);
10135 err
= btrfs_find_free_ino(root
, &objectid
);
10139 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
10140 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)),
10141 objectid
, S_IFLNK
|S_IRWXUGO
, &index
);
10142 if (IS_ERR(inode
)) {
10143 err
= PTR_ERR(inode
);
10149 * If the active LSM wants to access the inode during
10150 * d_instantiate it needs these. Smack checks to see
10151 * if the filesystem supports xattrs by looking at the
10154 inode
->i_fop
= &btrfs_file_operations
;
10155 inode
->i_op
= &btrfs_file_inode_operations
;
10156 inode
->i_mapping
->a_ops
= &btrfs_aops
;
10157 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
10159 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
10163 path
= btrfs_alloc_path();
10168 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
10170 key
.type
= BTRFS_EXTENT_DATA_KEY
;
10171 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
10172 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
10175 btrfs_free_path(path
);
10178 leaf
= path
->nodes
[0];
10179 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
10180 struct btrfs_file_extent_item
);
10181 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
10182 btrfs_set_file_extent_type(leaf
, ei
,
10183 BTRFS_FILE_EXTENT_INLINE
);
10184 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
10185 btrfs_set_file_extent_compression(leaf
, ei
, 0);
10186 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
10187 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
10189 ptr
= btrfs_file_extent_inline_start(ei
);
10190 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
10191 btrfs_mark_buffer_dirty(leaf
);
10192 btrfs_free_path(path
);
10194 inode
->i_op
= &btrfs_symlink_inode_operations
;
10195 inode_nohighmem(inode
);
10196 inode
->i_mapping
->a_ops
= &btrfs_aops
;
10197 inode_set_bytes(inode
, name_len
);
10198 btrfs_i_size_write(BTRFS_I(inode
), name_len
);
10199 err
= btrfs_update_inode(trans
, root
, inode
);
10201 * Last step, add directory indexes for our symlink inode. This is the
10202 * last step to avoid extra cleanup of these indexes if an error happens
10206 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
,
10207 BTRFS_I(inode
), 0, index
);
10211 d_instantiate_new(dentry
, inode
);
10214 btrfs_end_transaction(trans
);
10215 if (err
&& inode
) {
10216 inode_dec_link_count(inode
);
10217 discard_new_inode(inode
);
10219 btrfs_btree_balance_dirty(fs_info
);
10223 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
10224 u64 start
, u64 num_bytes
, u64 min_size
,
10225 loff_t actual_len
, u64
*alloc_hint
,
10226 struct btrfs_trans_handle
*trans
)
10228 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
10229 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
10230 struct extent_map
*em
;
10231 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
10232 struct btrfs_key ins
;
10233 u64 cur_offset
= start
;
10236 u64 last_alloc
= (u64
)-1;
10238 bool own_trans
= true;
10239 u64 end
= start
+ num_bytes
- 1;
10243 while (num_bytes
> 0) {
10245 trans
= btrfs_start_transaction(root
, 3);
10246 if (IS_ERR(trans
)) {
10247 ret
= PTR_ERR(trans
);
10252 cur_bytes
= min_t(u64
, num_bytes
, SZ_256M
);
10253 cur_bytes
= max(cur_bytes
, min_size
);
10255 * If we are severely fragmented we could end up with really
10256 * small allocations, so if the allocator is returning small
10257 * chunks lets make its job easier by only searching for those
10260 cur_bytes
= min(cur_bytes
, last_alloc
);
10261 ret
= btrfs_reserve_extent(root
, cur_bytes
, cur_bytes
,
10262 min_size
, 0, *alloc_hint
, &ins
, 1, 0);
10265 btrfs_end_transaction(trans
);
10268 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
10270 last_alloc
= ins
.offset
;
10271 ret
= insert_reserved_file_extent(trans
, inode
,
10272 cur_offset
, ins
.objectid
,
10273 ins
.offset
, ins
.offset
,
10274 ins
.offset
, 0, 0, 0,
10275 BTRFS_FILE_EXTENT_PREALLOC
);
10277 btrfs_free_reserved_extent(fs_info
, ins
.objectid
,
10279 btrfs_abort_transaction(trans
, ret
);
10281 btrfs_end_transaction(trans
);
10285 btrfs_drop_extent_cache(BTRFS_I(inode
), cur_offset
,
10286 cur_offset
+ ins
.offset
-1, 0);
10288 em
= alloc_extent_map();
10290 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
10291 &BTRFS_I(inode
)->runtime_flags
);
10295 em
->start
= cur_offset
;
10296 em
->orig_start
= cur_offset
;
10297 em
->len
= ins
.offset
;
10298 em
->block_start
= ins
.objectid
;
10299 em
->block_len
= ins
.offset
;
10300 em
->orig_block_len
= ins
.offset
;
10301 em
->ram_bytes
= ins
.offset
;
10302 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
10303 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
10304 em
->generation
= trans
->transid
;
10307 write_lock(&em_tree
->lock
);
10308 ret
= add_extent_mapping(em_tree
, em
, 1);
10309 write_unlock(&em_tree
->lock
);
10310 if (ret
!= -EEXIST
)
10312 btrfs_drop_extent_cache(BTRFS_I(inode
), cur_offset
,
10313 cur_offset
+ ins
.offset
- 1,
10316 free_extent_map(em
);
10318 num_bytes
-= ins
.offset
;
10319 cur_offset
+= ins
.offset
;
10320 *alloc_hint
= ins
.objectid
+ ins
.offset
;
10322 inode_inc_iversion(inode
);
10323 inode
->i_ctime
= current_time(inode
);
10324 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
10325 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
10326 (actual_len
> inode
->i_size
) &&
10327 (cur_offset
> inode
->i_size
)) {
10328 if (cur_offset
> actual_len
)
10329 i_size
= actual_len
;
10331 i_size
= cur_offset
;
10332 i_size_write(inode
, i_size
);
10333 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
10336 ret
= btrfs_update_inode(trans
, root
, inode
);
10339 btrfs_abort_transaction(trans
, ret
);
10341 btrfs_end_transaction(trans
);
10346 btrfs_end_transaction(trans
);
10348 if (cur_offset
< end
)
10349 btrfs_free_reserved_data_space(inode
, NULL
, cur_offset
,
10350 end
- cur_offset
+ 1);
10354 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
10355 u64 start
, u64 num_bytes
, u64 min_size
,
10356 loff_t actual_len
, u64
*alloc_hint
)
10358 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
10359 min_size
, actual_len
, alloc_hint
,
10363 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
10364 struct btrfs_trans_handle
*trans
, int mode
,
10365 u64 start
, u64 num_bytes
, u64 min_size
,
10366 loff_t actual_len
, u64
*alloc_hint
)
10368 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
10369 min_size
, actual_len
, alloc_hint
, trans
);
10372 static int btrfs_set_page_dirty(struct page
*page
)
10374 return __set_page_dirty_nobuffers(page
);
10377 static int btrfs_permission(struct inode
*inode
, int mask
)
10379 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
10380 umode_t mode
= inode
->i_mode
;
10382 if (mask
& MAY_WRITE
&&
10383 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
10384 if (btrfs_root_readonly(root
))
10386 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
10389 return generic_permission(inode
, mask
);
10392 static int btrfs_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
10394 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
10395 struct btrfs_trans_handle
*trans
;
10396 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
10397 struct inode
*inode
= NULL
;
10403 * 5 units required for adding orphan entry
10405 trans
= btrfs_start_transaction(root
, 5);
10407 return PTR_ERR(trans
);
10409 ret
= btrfs_find_free_ino(root
, &objectid
);
10413 inode
= btrfs_new_inode(trans
, root
, dir
, NULL
, 0,
10414 btrfs_ino(BTRFS_I(dir
)), objectid
, mode
, &index
);
10415 if (IS_ERR(inode
)) {
10416 ret
= PTR_ERR(inode
);
10421 inode
->i_fop
= &btrfs_file_operations
;
10422 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 ret
= btrfs_init_inode_security(trans
, inode
, dir
, NULL
);
10431 ret
= btrfs_update_inode(trans
, root
, inode
);
10434 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
10439 * We set number of links to 0 in btrfs_new_inode(), and here we set
10440 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10443 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10445 set_nlink(inode
, 1);
10446 d_tmpfile(dentry
, inode
);
10447 unlock_new_inode(inode
);
10448 mark_inode_dirty(inode
);
10450 btrfs_end_transaction(trans
);
10452 discard_new_inode(inode
);
10453 btrfs_btree_balance_dirty(fs_info
);
10457 void btrfs_set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
10459 struct inode
*inode
= tree
->private_data
;
10460 unsigned long index
= start
>> PAGE_SHIFT
;
10461 unsigned long end_index
= end
>> PAGE_SHIFT
;
10464 while (index
<= end_index
) {
10465 page
= find_get_page(inode
->i_mapping
, index
);
10466 ASSERT(page
); /* Pages should be in the extent_io_tree */
10467 set_page_writeback(page
);
10475 * Add an entry indicating a block group or device which is pinned by a
10476 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10477 * negative errno on failure.
10479 static int btrfs_add_swapfile_pin(struct inode
*inode
, void *ptr
,
10480 bool is_block_group
)
10482 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
10483 struct btrfs_swapfile_pin
*sp
, *entry
;
10484 struct rb_node
**p
;
10485 struct rb_node
*parent
= NULL
;
10487 sp
= kmalloc(sizeof(*sp
), GFP_NOFS
);
10492 sp
->is_block_group
= is_block_group
;
10494 spin_lock(&fs_info
->swapfile_pins_lock
);
10495 p
= &fs_info
->swapfile_pins
.rb_node
;
10498 entry
= rb_entry(parent
, struct btrfs_swapfile_pin
, node
);
10499 if (sp
->ptr
< entry
->ptr
||
10500 (sp
->ptr
== entry
->ptr
&& sp
->inode
< entry
->inode
)) {
10501 p
= &(*p
)->rb_left
;
10502 } else if (sp
->ptr
> entry
->ptr
||
10503 (sp
->ptr
== entry
->ptr
&& sp
->inode
> entry
->inode
)) {
10504 p
= &(*p
)->rb_right
;
10506 spin_unlock(&fs_info
->swapfile_pins_lock
);
10511 rb_link_node(&sp
->node
, parent
, p
);
10512 rb_insert_color(&sp
->node
, &fs_info
->swapfile_pins
);
10513 spin_unlock(&fs_info
->swapfile_pins_lock
);
10517 /* Free all of the entries pinned by this swapfile. */
10518 static void btrfs_free_swapfile_pins(struct inode
*inode
)
10520 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
10521 struct btrfs_swapfile_pin
*sp
;
10522 struct rb_node
*node
, *next
;
10524 spin_lock(&fs_info
->swapfile_pins_lock
);
10525 node
= rb_first(&fs_info
->swapfile_pins
);
10527 next
= rb_next(node
);
10528 sp
= rb_entry(node
, struct btrfs_swapfile_pin
, node
);
10529 if (sp
->inode
== inode
) {
10530 rb_erase(&sp
->node
, &fs_info
->swapfile_pins
);
10531 if (sp
->is_block_group
)
10532 btrfs_put_block_group(sp
->ptr
);
10537 spin_unlock(&fs_info
->swapfile_pins_lock
);
10540 struct btrfs_swap_info
{
10546 unsigned long nr_pages
;
10550 static int btrfs_add_swap_extent(struct swap_info_struct
*sis
,
10551 struct btrfs_swap_info
*bsi
)
10553 unsigned long nr_pages
;
10554 u64 first_ppage
, first_ppage_reported
, next_ppage
;
10557 first_ppage
= ALIGN(bsi
->block_start
, PAGE_SIZE
) >> PAGE_SHIFT
;
10558 next_ppage
= ALIGN_DOWN(bsi
->block_start
+ bsi
->block_len
,
10559 PAGE_SIZE
) >> PAGE_SHIFT
;
10561 if (first_ppage
>= next_ppage
)
10563 nr_pages
= next_ppage
- first_ppage
;
10565 first_ppage_reported
= first_ppage
;
10566 if (bsi
->start
== 0)
10567 first_ppage_reported
++;
10568 if (bsi
->lowest_ppage
> first_ppage_reported
)
10569 bsi
->lowest_ppage
= first_ppage_reported
;
10570 if (bsi
->highest_ppage
< (next_ppage
- 1))
10571 bsi
->highest_ppage
= next_ppage
- 1;
10573 ret
= add_swap_extent(sis
, bsi
->nr_pages
, nr_pages
, first_ppage
);
10576 bsi
->nr_extents
+= ret
;
10577 bsi
->nr_pages
+= nr_pages
;
10581 static void btrfs_swap_deactivate(struct file
*file
)
10583 struct inode
*inode
= file_inode(file
);
10585 btrfs_free_swapfile_pins(inode
);
10586 atomic_dec(&BTRFS_I(inode
)->root
->nr_swapfiles
);
10589 static int btrfs_swap_activate(struct swap_info_struct
*sis
, struct file
*file
,
10592 struct inode
*inode
= file_inode(file
);
10593 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
10594 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
10595 struct extent_state
*cached_state
= NULL
;
10596 struct extent_map
*em
= NULL
;
10597 struct btrfs_device
*device
= NULL
;
10598 struct btrfs_swap_info bsi
= {
10599 .lowest_ppage
= (sector_t
)-1ULL,
10606 * If the swap file was just created, make sure delalloc is done. If the
10607 * file changes again after this, the user is doing something stupid and
10608 * we don't really care.
10610 ret
= btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
10615 * The inode is locked, so these flags won't change after we check them.
10617 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
) {
10618 btrfs_warn(fs_info
, "swapfile must not be compressed");
10621 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
)) {
10622 btrfs_warn(fs_info
, "swapfile must not be copy-on-write");
10625 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
10626 btrfs_warn(fs_info
, "swapfile must not be checksummed");
10631 * Balance or device remove/replace/resize can move stuff around from
10632 * under us. The EXCL_OP flag makes sure they aren't running/won't run
10633 * concurrently while we are mapping the swap extents, and
10634 * fs_info->swapfile_pins prevents them from running while the swap file
10635 * is active and moving the extents. Note that this also prevents a
10636 * concurrent device add which isn't actually necessary, but it's not
10637 * really worth the trouble to allow it.
10639 if (test_and_set_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
)) {
10640 btrfs_warn(fs_info
,
10641 "cannot activate swapfile while exclusive operation is running");
10645 * Snapshots can create extents which require COW even if NODATACOW is
10646 * set. We use this counter to prevent snapshots. We must increment it
10647 * before walking the extents because we don't want a concurrent
10648 * snapshot to run after we've already checked the extents.
10650 atomic_inc(&BTRFS_I(inode
)->root
->nr_swapfiles
);
10652 isize
= ALIGN_DOWN(inode
->i_size
, fs_info
->sectorsize
);
10654 lock_extent_bits(io_tree
, 0, isize
- 1, &cached_state
);
10656 while (start
< isize
) {
10657 u64 logical_block_start
, physical_block_start
;
10658 struct btrfs_block_group_cache
*bg
;
10659 u64 len
= isize
- start
;
10661 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0, start
, len
, 0);
10667 if (em
->block_start
== EXTENT_MAP_HOLE
) {
10668 btrfs_warn(fs_info
, "swapfile must not have holes");
10672 if (em
->block_start
== EXTENT_MAP_INLINE
) {
10674 * It's unlikely we'll ever actually find ourselves
10675 * here, as a file small enough to fit inline won't be
10676 * big enough to store more than the swap header, but in
10677 * case something changes in the future, let's catch it
10678 * here rather than later.
10680 btrfs_warn(fs_info
, "swapfile must not be inline");
10684 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
10685 btrfs_warn(fs_info
, "swapfile must not be compressed");
10690 logical_block_start
= em
->block_start
+ (start
- em
->start
);
10691 len
= min(len
, em
->len
- (start
- em
->start
));
10692 free_extent_map(em
);
10695 ret
= can_nocow_extent(inode
, start
, &len
, NULL
, NULL
, NULL
);
10701 btrfs_warn(fs_info
,
10702 "swapfile must not be copy-on-write");
10707 em
= btrfs_get_chunk_map(fs_info
, logical_block_start
, len
);
10713 if (em
->map_lookup
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
10714 btrfs_warn(fs_info
,
10715 "swapfile must have single data profile");
10720 if (device
== NULL
) {
10721 device
= em
->map_lookup
->stripes
[0].dev
;
10722 ret
= btrfs_add_swapfile_pin(inode
, device
, false);
10727 } else if (device
!= em
->map_lookup
->stripes
[0].dev
) {
10728 btrfs_warn(fs_info
, "swapfile must be on one device");
10733 physical_block_start
= (em
->map_lookup
->stripes
[0].physical
+
10734 (logical_block_start
- em
->start
));
10735 len
= min(len
, em
->len
- (logical_block_start
- em
->start
));
10736 free_extent_map(em
);
10739 bg
= btrfs_lookup_block_group(fs_info
, logical_block_start
);
10741 btrfs_warn(fs_info
,
10742 "could not find block group containing swapfile");
10747 ret
= btrfs_add_swapfile_pin(inode
, bg
, true);
10749 btrfs_put_block_group(bg
);
10756 if (bsi
.block_len
&&
10757 bsi
.block_start
+ bsi
.block_len
== physical_block_start
) {
10758 bsi
.block_len
+= len
;
10760 if (bsi
.block_len
) {
10761 ret
= btrfs_add_swap_extent(sis
, &bsi
);
10766 bsi
.block_start
= physical_block_start
;
10767 bsi
.block_len
= len
;
10774 ret
= btrfs_add_swap_extent(sis
, &bsi
);
10777 if (!IS_ERR_OR_NULL(em
))
10778 free_extent_map(em
);
10780 unlock_extent_cached(io_tree
, 0, isize
- 1, &cached_state
);
10783 btrfs_swap_deactivate(file
);
10785 clear_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
);
10791 sis
->bdev
= device
->bdev
;
10792 *span
= bsi
.highest_ppage
- bsi
.lowest_ppage
+ 1;
10793 sis
->max
= bsi
.nr_pages
;
10794 sis
->pages
= bsi
.nr_pages
- 1;
10795 sis
->highest_bit
= bsi
.nr_pages
- 1;
10796 return bsi
.nr_extents
;
10799 static void btrfs_swap_deactivate(struct file
*file
)
10803 static int btrfs_swap_activate(struct swap_info_struct
*sis
, struct file
*file
,
10806 return -EOPNOTSUPP
;
10810 static const struct inode_operations btrfs_dir_inode_operations
= {
10811 .getattr
= btrfs_getattr
,
10812 .lookup
= btrfs_lookup
,
10813 .create
= btrfs_create
,
10814 .unlink
= btrfs_unlink
,
10815 .link
= btrfs_link
,
10816 .mkdir
= btrfs_mkdir
,
10817 .rmdir
= btrfs_rmdir
,
10818 .rename
= btrfs_rename2
,
10819 .symlink
= btrfs_symlink
,
10820 .setattr
= btrfs_setattr
,
10821 .mknod
= btrfs_mknod
,
10822 .listxattr
= btrfs_listxattr
,
10823 .permission
= btrfs_permission
,
10824 .get_acl
= btrfs_get_acl
,
10825 .set_acl
= btrfs_set_acl
,
10826 .update_time
= btrfs_update_time
,
10827 .tmpfile
= btrfs_tmpfile
,
10829 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
10830 .lookup
= btrfs_lookup
,
10831 .permission
= btrfs_permission
,
10832 .update_time
= btrfs_update_time
,
10835 static const struct file_operations btrfs_dir_file_operations
= {
10836 .llseek
= generic_file_llseek
,
10837 .read
= generic_read_dir
,
10838 .iterate_shared
= btrfs_real_readdir
,
10839 .open
= btrfs_opendir
,
10840 .unlocked_ioctl
= btrfs_ioctl
,
10841 #ifdef CONFIG_COMPAT
10842 .compat_ioctl
= btrfs_compat_ioctl
,
10844 .release
= btrfs_release_file
,
10845 .fsync
= btrfs_sync_file
,
10848 static const struct extent_io_ops btrfs_extent_io_ops
= {
10849 /* mandatory callbacks */
10850 .submit_bio_hook
= btrfs_submit_bio_hook
,
10851 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
10855 * btrfs doesn't support the bmap operation because swapfiles
10856 * use bmap to make a mapping of extents in the file. They assume
10857 * these extents won't change over the life of the file and they
10858 * use the bmap result to do IO directly to the drive.
10860 * the btrfs bmap call would return logical addresses that aren't
10861 * suitable for IO and they also will change frequently as COW
10862 * operations happen. So, swapfile + btrfs == corruption.
10864 * For now we're avoiding this by dropping bmap.
10866 static const struct address_space_operations btrfs_aops
= {
10867 .readpage
= btrfs_readpage
,
10868 .writepage
= btrfs_writepage
,
10869 .writepages
= btrfs_writepages
,
10870 .readpages
= btrfs_readpages
,
10871 .direct_IO
= btrfs_direct_IO
,
10872 .invalidatepage
= btrfs_invalidatepage
,
10873 .releasepage
= btrfs_releasepage
,
10874 .set_page_dirty
= btrfs_set_page_dirty
,
10875 .error_remove_page
= generic_error_remove_page
,
10876 .swap_activate
= btrfs_swap_activate
,
10877 .swap_deactivate
= btrfs_swap_deactivate
,
10880 static const struct inode_operations btrfs_file_inode_operations
= {
10881 .getattr
= btrfs_getattr
,
10882 .setattr
= btrfs_setattr
,
10883 .listxattr
= btrfs_listxattr
,
10884 .permission
= btrfs_permission
,
10885 .fiemap
= btrfs_fiemap
,
10886 .get_acl
= btrfs_get_acl
,
10887 .set_acl
= btrfs_set_acl
,
10888 .update_time
= btrfs_update_time
,
10890 static const struct inode_operations btrfs_special_inode_operations
= {
10891 .getattr
= btrfs_getattr
,
10892 .setattr
= btrfs_setattr
,
10893 .permission
= btrfs_permission
,
10894 .listxattr
= btrfs_listxattr
,
10895 .get_acl
= btrfs_get_acl
,
10896 .set_acl
= btrfs_set_acl
,
10897 .update_time
= btrfs_update_time
,
10899 static const struct inode_operations btrfs_symlink_inode_operations
= {
10900 .get_link
= page_get_link
,
10901 .getattr
= btrfs_getattr
,
10902 .setattr
= btrfs_setattr
,
10903 .permission
= btrfs_permission
,
10904 .listxattr
= btrfs_listxattr
,
10905 .update_time
= btrfs_update_time
,
10908 const struct dentry_operations btrfs_dentry_operations
= {
10909 .d_delete
= btrfs_dentry_delete
,