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 <linux/sched/mm.h>
32 #include <asm/unaligned.h>
36 #include "transaction.h"
37 #include "btrfs_inode.h"
38 #include "print-tree.h"
39 #include "ordered-data.h"
43 #include "compression.h"
45 #include "free-space-cache.h"
46 #include "inode-map.h"
50 #include "delalloc-space.h"
51 #include "block-group.h"
53 struct btrfs_iget_args
{
54 struct btrfs_key
*location
;
55 struct btrfs_root
*root
;
58 struct btrfs_dio_data
{
60 u64 unsubmitted_oe_range_start
;
61 u64 unsubmitted_oe_range_end
;
65 static const struct inode_operations btrfs_dir_inode_operations
;
66 static const struct inode_operations btrfs_symlink_inode_operations
;
67 static const struct inode_operations btrfs_dir_ro_inode_operations
;
68 static const struct inode_operations btrfs_special_inode_operations
;
69 static const struct inode_operations btrfs_file_inode_operations
;
70 static const struct address_space_operations btrfs_aops
;
71 static const struct file_operations btrfs_dir_file_operations
;
72 static const struct extent_io_ops btrfs_extent_io_ops
;
74 static struct kmem_cache
*btrfs_inode_cachep
;
75 struct kmem_cache
*btrfs_trans_handle_cachep
;
76 struct kmem_cache
*btrfs_path_cachep
;
77 struct kmem_cache
*btrfs_free_space_cachep
;
78 struct kmem_cache
*btrfs_free_space_bitmap_cachep
;
80 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
);
81 static int btrfs_truncate(struct inode
*inode
, bool skip_writeback
);
82 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
);
83 static noinline
int cow_file_range(struct inode
*inode
,
84 struct page
*locked_page
,
85 u64 start
, u64 end
, int *page_started
,
86 unsigned long *nr_written
, int unlock
);
87 static struct extent_map
*create_io_em(struct inode
*inode
, u64 start
, u64 len
,
88 u64 orig_start
, u64 block_start
,
89 u64 block_len
, u64 orig_block_len
,
90 u64 ram_bytes
, int compress_type
,
93 static void __endio_write_update_ordered(struct inode
*inode
,
94 const u64 offset
, const u64 bytes
,
98 * Cleanup all submitted ordered extents in specified range to handle errors
99 * from the btrfs_run_delalloc_range() callback.
101 * NOTE: caller must ensure that when an error happens, it can not call
102 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
103 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
104 * to be released, which we want to happen only when finishing the ordered
105 * extent (btrfs_finish_ordered_io()).
107 static inline void btrfs_cleanup_ordered_extents(struct inode
*inode
,
108 struct page
*locked_page
,
109 u64 offset
, u64 bytes
)
111 unsigned long index
= offset
>> PAGE_SHIFT
;
112 unsigned long end_index
= (offset
+ bytes
- 1) >> PAGE_SHIFT
;
113 u64 page_start
= page_offset(locked_page
);
114 u64 page_end
= page_start
+ PAGE_SIZE
- 1;
118 while (index
<= end_index
) {
119 page
= find_get_page(inode
->i_mapping
, index
);
123 ClearPagePrivate2(page
);
128 * In case this page belongs to the delalloc range being instantiated
129 * then skip it, since the first page of a range is going to be
130 * properly cleaned up by the caller of run_delalloc_range
132 if (page_start
>= offset
&& page_end
<= (offset
+ bytes
- 1)) {
137 return __endio_write_update_ordered(inode
, offset
, bytes
, false);
140 static int btrfs_dirty_inode(struct inode
*inode
);
142 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
143 void btrfs_test_inode_set_ops(struct inode
*inode
)
145 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
149 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
150 struct inode
*inode
, struct inode
*dir
,
151 const struct qstr
*qstr
)
155 err
= btrfs_init_acl(trans
, inode
, dir
);
157 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
162 * this does all the hard work for inserting an inline extent into
163 * the btree. The caller should have done a btrfs_drop_extents so that
164 * no overlapping inline items exist in the btree
166 static int insert_inline_extent(struct btrfs_trans_handle
*trans
,
167 struct btrfs_path
*path
, int extent_inserted
,
168 struct btrfs_root
*root
, struct inode
*inode
,
169 u64 start
, size_t size
, size_t compressed_size
,
171 struct page
**compressed_pages
)
173 struct extent_buffer
*leaf
;
174 struct page
*page
= NULL
;
177 struct btrfs_file_extent_item
*ei
;
179 size_t cur_size
= size
;
180 unsigned long offset
;
182 ASSERT((compressed_size
> 0 && compressed_pages
) ||
183 (compressed_size
== 0 && !compressed_pages
));
185 if (compressed_size
&& compressed_pages
)
186 cur_size
= compressed_size
;
188 inode_add_bytes(inode
, size
);
190 if (!extent_inserted
) {
191 struct btrfs_key key
;
194 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
196 key
.type
= BTRFS_EXTENT_DATA_KEY
;
198 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
199 path
->leave_spinning
= 1;
200 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
205 leaf
= path
->nodes
[0];
206 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
207 struct btrfs_file_extent_item
);
208 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
209 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
210 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
211 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
212 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
213 ptr
= btrfs_file_extent_inline_start(ei
);
215 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
218 while (compressed_size
> 0) {
219 cpage
= compressed_pages
[i
];
220 cur_size
= min_t(unsigned long, compressed_size
,
223 kaddr
= kmap_atomic(cpage
);
224 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
225 kunmap_atomic(kaddr
);
229 compressed_size
-= cur_size
;
231 btrfs_set_file_extent_compression(leaf
, ei
,
234 page
= find_get_page(inode
->i_mapping
,
235 start
>> PAGE_SHIFT
);
236 btrfs_set_file_extent_compression(leaf
, ei
, 0);
237 kaddr
= kmap_atomic(page
);
238 offset
= offset_in_page(start
);
239 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
240 kunmap_atomic(kaddr
);
243 btrfs_mark_buffer_dirty(leaf
);
244 btrfs_release_path(path
);
247 * we're an inline extent, so nobody can
248 * extend the file past i_size without locking
249 * a page we already have locked.
251 * We must do any isize and inode updates
252 * before we unlock the pages. Otherwise we
253 * could end up racing with unlink.
255 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
256 ret
= btrfs_update_inode(trans
, root
, inode
);
264 * conditionally insert an inline extent into the file. This
265 * does the checks required to make sure the data is small enough
266 * to fit as an inline extent.
268 static noinline
int cow_file_range_inline(struct inode
*inode
, u64 start
,
269 u64 end
, size_t compressed_size
,
271 struct page
**compressed_pages
)
273 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
274 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
275 struct btrfs_trans_handle
*trans
;
276 u64 isize
= i_size_read(inode
);
277 u64 actual_end
= min(end
+ 1, isize
);
278 u64 inline_len
= actual_end
- start
;
279 u64 aligned_end
= ALIGN(end
, fs_info
->sectorsize
);
280 u64 data_len
= inline_len
;
282 struct btrfs_path
*path
;
283 int extent_inserted
= 0;
284 u32 extent_item_size
;
287 data_len
= compressed_size
;
290 actual_end
> fs_info
->sectorsize
||
291 data_len
> BTRFS_MAX_INLINE_DATA_SIZE(fs_info
) ||
293 (actual_end
& (fs_info
->sectorsize
- 1)) == 0) ||
295 data_len
> fs_info
->max_inline
) {
299 path
= btrfs_alloc_path();
303 trans
= btrfs_join_transaction(root
);
305 btrfs_free_path(path
);
306 return PTR_ERR(trans
);
308 trans
->block_rsv
= &BTRFS_I(inode
)->block_rsv
;
310 if (compressed_size
&& compressed_pages
)
311 extent_item_size
= btrfs_file_extent_calc_inline_size(
314 extent_item_size
= btrfs_file_extent_calc_inline_size(
317 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
,
318 start
, aligned_end
, NULL
,
319 1, 1, extent_item_size
, &extent_inserted
);
321 btrfs_abort_transaction(trans
, ret
);
325 if (isize
> actual_end
)
326 inline_len
= min_t(u64
, isize
, actual_end
);
327 ret
= insert_inline_extent(trans
, path
, extent_inserted
,
329 inline_len
, compressed_size
,
330 compress_type
, compressed_pages
);
331 if (ret
&& ret
!= -ENOSPC
) {
332 btrfs_abort_transaction(trans
, ret
);
334 } else if (ret
== -ENOSPC
) {
339 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
340 btrfs_drop_extent_cache(BTRFS_I(inode
), start
, aligned_end
- 1, 0);
343 * Don't forget to free the reserved space, as for inlined extent
344 * it won't count as data extent, free them directly here.
345 * And at reserve time, it's always aligned to page size, so
346 * just free one page here.
348 btrfs_qgroup_free_data(inode
, NULL
, 0, PAGE_SIZE
);
349 btrfs_free_path(path
);
350 btrfs_end_transaction(trans
);
354 struct async_extent
{
359 unsigned long nr_pages
;
361 struct list_head list
;
366 struct page
*locked_page
;
369 unsigned int write_flags
;
370 struct list_head extents
;
371 struct btrfs_work work
;
376 /* Number of chunks in flight; must be first in the structure */
378 struct async_chunk chunks
[];
381 static noinline
int add_async_extent(struct async_chunk
*cow
,
382 u64 start
, u64 ram_size
,
385 unsigned long nr_pages
,
388 struct async_extent
*async_extent
;
390 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
391 BUG_ON(!async_extent
); /* -ENOMEM */
392 async_extent
->start
= start
;
393 async_extent
->ram_size
= ram_size
;
394 async_extent
->compressed_size
= compressed_size
;
395 async_extent
->pages
= pages
;
396 async_extent
->nr_pages
= nr_pages
;
397 async_extent
->compress_type
= compress_type
;
398 list_add_tail(&async_extent
->list
, &cow
->extents
);
403 * Check if the inode has flags compatible with compression
405 static inline bool inode_can_compress(struct inode
*inode
)
407 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
||
408 BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
414 * Check if the inode needs to be submitted to compression, based on mount
415 * options, defragmentation, properties or heuristics.
417 static inline int inode_need_compress(struct inode
*inode
, u64 start
, u64 end
)
419 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
421 if (!inode_can_compress(inode
)) {
422 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG
),
423 KERN_ERR
"BTRFS: unexpected compression for ino %llu\n",
424 btrfs_ino(BTRFS_I(inode
)));
428 if (btrfs_test_opt(fs_info
, FORCE_COMPRESS
))
431 if (BTRFS_I(inode
)->defrag_compress
)
433 /* bad compression ratios */
434 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
436 if (btrfs_test_opt(fs_info
, COMPRESS
) ||
437 BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
||
438 BTRFS_I(inode
)->prop_compress
)
439 return btrfs_compress_heuristic(inode
, start
, end
);
443 static inline void inode_should_defrag(struct btrfs_inode
*inode
,
444 u64 start
, u64 end
, u64 num_bytes
, u64 small_write
)
446 /* If this is a small write inside eof, kick off a defrag */
447 if (num_bytes
< small_write
&&
448 (start
> 0 || end
+ 1 < inode
->disk_i_size
))
449 btrfs_add_inode_defrag(NULL
, inode
);
453 * we create compressed extents in two phases. The first
454 * phase compresses a range of pages that have already been
455 * locked (both pages and state bits are locked).
457 * This is done inside an ordered work queue, and the compression
458 * is spread across many cpus. The actual IO submission is step
459 * two, and the ordered work queue takes care of making sure that
460 * happens in the same order things were put onto the queue by
461 * writepages and friends.
463 * If this code finds it can't get good compression, it puts an
464 * entry onto the work queue to write the uncompressed bytes. This
465 * makes sure that both compressed inodes and uncompressed inodes
466 * are written in the same order that the flusher thread sent them
469 static noinline
int compress_file_range(struct async_chunk
*async_chunk
)
471 struct inode
*inode
= async_chunk
->inode
;
472 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
473 u64 blocksize
= fs_info
->sectorsize
;
474 u64 start
= async_chunk
->start
;
475 u64 end
= async_chunk
->end
;
478 struct page
**pages
= NULL
;
479 unsigned long nr_pages
;
480 unsigned long total_compressed
= 0;
481 unsigned long total_in
= 0;
484 int compress_type
= fs_info
->compress_type
;
485 int compressed_extents
= 0;
488 inode_should_defrag(BTRFS_I(inode
), start
, end
, end
- start
+ 1,
491 actual_end
= min_t(u64
, i_size_read(inode
), end
+ 1);
494 nr_pages
= (end
>> PAGE_SHIFT
) - (start
>> PAGE_SHIFT
) + 1;
495 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED
% PAGE_SIZE
) != 0);
496 nr_pages
= min_t(unsigned long, nr_pages
,
497 BTRFS_MAX_COMPRESSED
/ PAGE_SIZE
);
500 * we don't want to send crud past the end of i_size through
501 * compression, that's just a waste of CPU time. So, if the
502 * end of the file is before the start of our current
503 * requested range of bytes, we bail out to the uncompressed
504 * cleanup code that can deal with all of this.
506 * It isn't really the fastest way to fix things, but this is a
507 * very uncommon corner.
509 if (actual_end
<= start
)
510 goto cleanup_and_bail_uncompressed
;
512 total_compressed
= actual_end
- start
;
515 * skip compression for a small file range(<=blocksize) that
516 * isn't an inline extent, since it doesn't save disk space at all.
518 if (total_compressed
<= blocksize
&&
519 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
520 goto cleanup_and_bail_uncompressed
;
522 total_compressed
= min_t(unsigned long, total_compressed
,
523 BTRFS_MAX_UNCOMPRESSED
);
528 * we do compression for mount -o compress and when the
529 * inode has not been flagged as nocompress. This flag can
530 * change at any time if we discover bad compression ratios.
532 if (inode_need_compress(inode
, start
, end
)) {
534 pages
= kcalloc(nr_pages
, sizeof(struct page
*), GFP_NOFS
);
536 /* just bail out to the uncompressed code */
541 if (BTRFS_I(inode
)->defrag_compress
)
542 compress_type
= BTRFS_I(inode
)->defrag_compress
;
543 else if (BTRFS_I(inode
)->prop_compress
)
544 compress_type
= BTRFS_I(inode
)->prop_compress
;
547 * we need to call clear_page_dirty_for_io on each
548 * page in the range. Otherwise applications with the file
549 * mmap'd can wander in and change the page contents while
550 * we are compressing them.
552 * If the compression fails for any reason, we set the pages
553 * dirty again later on.
555 * Note that the remaining part is redirtied, the start pointer
556 * has moved, the end is the original one.
559 extent_range_clear_dirty_for_io(inode
, start
, end
);
563 /* Compression level is applied here and only here */
564 ret
= btrfs_compress_pages(
565 compress_type
| (fs_info
->compress_level
<< 4),
566 inode
->i_mapping
, start
,
573 unsigned long offset
= offset_in_page(total_compressed
);
574 struct page
*page
= pages
[nr_pages
- 1];
577 /* zero the tail end of the last page, we might be
578 * sending it down to disk
581 kaddr
= kmap_atomic(page
);
582 memset(kaddr
+ offset
, 0,
584 kunmap_atomic(kaddr
);
591 /* lets try to make an inline extent */
592 if (ret
|| total_in
< actual_end
) {
593 /* we didn't compress the entire range, try
594 * to make an uncompressed inline extent.
596 ret
= cow_file_range_inline(inode
, start
, end
, 0,
597 BTRFS_COMPRESS_NONE
, NULL
);
599 /* try making a compressed inline extent */
600 ret
= cow_file_range_inline(inode
, start
, end
,
602 compress_type
, pages
);
605 unsigned long clear_flags
= EXTENT_DELALLOC
|
606 EXTENT_DELALLOC_NEW
| EXTENT_DEFRAG
|
607 EXTENT_DO_ACCOUNTING
;
608 unsigned long page_error_op
;
610 page_error_op
= ret
< 0 ? PAGE_SET_ERROR
: 0;
613 * inline extent creation worked or returned error,
614 * we don't need to create any more async work items.
615 * Unlock and free up our temp pages.
617 * We use DO_ACCOUNTING here because we need the
618 * delalloc_release_metadata to be done _after_ we drop
619 * our outstanding extent for clearing delalloc for this
622 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
,
630 for (i
= 0; i
< nr_pages
; i
++) {
631 WARN_ON(pages
[i
]->mapping
);
642 * we aren't doing an inline extent round the compressed size
643 * up to a block size boundary so the allocator does sane
646 total_compressed
= ALIGN(total_compressed
, blocksize
);
649 * one last check to make sure the compression is really a
650 * win, compare the page count read with the blocks on disk,
651 * compression must free at least one sector size
653 total_in
= ALIGN(total_in
, PAGE_SIZE
);
654 if (total_compressed
+ blocksize
<= total_in
) {
655 compressed_extents
++;
658 * The async work queues will take care of doing actual
659 * allocation on disk for these compressed pages, and
660 * will submit them to the elevator.
662 add_async_extent(async_chunk
, start
, total_in
,
663 total_compressed
, pages
, nr_pages
,
666 if (start
+ total_in
< end
) {
672 return compressed_extents
;
677 * the compression code ran but failed to make things smaller,
678 * free any pages it allocated and our page pointer array
680 for (i
= 0; i
< nr_pages
; i
++) {
681 WARN_ON(pages
[i
]->mapping
);
686 total_compressed
= 0;
689 /* flag the file so we don't compress in the future */
690 if (!btrfs_test_opt(fs_info
, FORCE_COMPRESS
) &&
691 !(BTRFS_I(inode
)->prop_compress
)) {
692 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
695 cleanup_and_bail_uncompressed
:
697 * No compression, but we still need to write the pages in the file
698 * we've been given so far. redirty the locked page if it corresponds
699 * to our extent and set things up for the async work queue to run
700 * cow_file_range to do the normal delalloc dance.
702 if (page_offset(async_chunk
->locked_page
) >= start
&&
703 page_offset(async_chunk
->locked_page
) <= end
)
704 __set_page_dirty_nobuffers(async_chunk
->locked_page
);
705 /* unlocked later on in the async handlers */
708 extent_range_redirty_for_io(inode
, start
, end
);
709 add_async_extent(async_chunk
, start
, end
- start
+ 1, 0, NULL
, 0,
710 BTRFS_COMPRESS_NONE
);
711 compressed_extents
++;
713 return compressed_extents
;
716 static void free_async_extent_pages(struct async_extent
*async_extent
)
720 if (!async_extent
->pages
)
723 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
724 WARN_ON(async_extent
->pages
[i
]->mapping
);
725 put_page(async_extent
->pages
[i
]);
727 kfree(async_extent
->pages
);
728 async_extent
->nr_pages
= 0;
729 async_extent
->pages
= NULL
;
733 * phase two of compressed writeback. This is the ordered portion
734 * of the code, which only gets called in the order the work was
735 * queued. We walk all the async extents created by compress_file_range
736 * and send them down to the disk.
738 static noinline
void submit_compressed_extents(struct async_chunk
*async_chunk
)
740 struct inode
*inode
= async_chunk
->inode
;
741 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
742 struct async_extent
*async_extent
;
744 struct btrfs_key ins
;
745 struct extent_map
*em
;
746 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
747 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
751 while (!list_empty(&async_chunk
->extents
)) {
752 async_extent
= list_entry(async_chunk
->extents
.next
,
753 struct async_extent
, list
);
754 list_del(&async_extent
->list
);
757 lock_extent(io_tree
, async_extent
->start
,
758 async_extent
->start
+ async_extent
->ram_size
- 1);
759 /* did the compression code fall back to uncompressed IO? */
760 if (!async_extent
->pages
) {
761 int page_started
= 0;
762 unsigned long nr_written
= 0;
764 /* allocate blocks */
765 ret
= cow_file_range(inode
, async_chunk
->locked_page
,
767 async_extent
->start
+
768 async_extent
->ram_size
- 1,
769 &page_started
, &nr_written
, 0);
774 * if page_started, cow_file_range inserted an
775 * inline extent and took care of all the unlocking
776 * and IO for us. Otherwise, we need to submit
777 * all those pages down to the drive.
779 if (!page_started
&& !ret
)
780 extent_write_locked_range(inode
,
782 async_extent
->start
+
783 async_extent
->ram_size
- 1,
786 unlock_page(async_chunk
->locked_page
);
792 ret
= btrfs_reserve_extent(root
, async_extent
->ram_size
,
793 async_extent
->compressed_size
,
794 async_extent
->compressed_size
,
795 0, alloc_hint
, &ins
, 1, 1);
797 free_async_extent_pages(async_extent
);
799 if (ret
== -ENOSPC
) {
800 unlock_extent(io_tree
, async_extent
->start
,
801 async_extent
->start
+
802 async_extent
->ram_size
- 1);
805 * we need to redirty the pages if we decide to
806 * fallback to uncompressed IO, otherwise we
807 * will not submit these pages down to lower
810 extent_range_redirty_for_io(inode
,
812 async_extent
->start
+
813 async_extent
->ram_size
- 1);
820 * here we're doing allocation and writeback of the
823 em
= create_io_em(inode
, async_extent
->start
,
824 async_extent
->ram_size
, /* len */
825 async_extent
->start
, /* orig_start */
826 ins
.objectid
, /* block_start */
827 ins
.offset
, /* block_len */
828 ins
.offset
, /* orig_block_len */
829 async_extent
->ram_size
, /* ram_bytes */
830 async_extent
->compress_type
,
831 BTRFS_ORDERED_COMPRESSED
);
833 /* ret value is not necessary due to void function */
834 goto out_free_reserve
;
837 ret
= btrfs_add_ordered_extent_compress(inode
,
840 async_extent
->ram_size
,
842 BTRFS_ORDERED_COMPRESSED
,
843 async_extent
->compress_type
);
845 btrfs_drop_extent_cache(BTRFS_I(inode
),
847 async_extent
->start
+
848 async_extent
->ram_size
- 1, 0);
849 goto out_free_reserve
;
851 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
854 * clear dirty, set writeback and unlock the pages.
856 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
857 async_extent
->start
+
858 async_extent
->ram_size
- 1,
859 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
,
860 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
862 if (btrfs_submit_compressed_write(inode
,
864 async_extent
->ram_size
,
866 ins
.offset
, async_extent
->pages
,
867 async_extent
->nr_pages
,
868 async_chunk
->write_flags
)) {
869 struct page
*p
= async_extent
->pages
[0];
870 const u64 start
= async_extent
->start
;
871 const u64 end
= start
+ async_extent
->ram_size
- 1;
873 p
->mapping
= inode
->i_mapping
;
874 btrfs_writepage_endio_finish_ordered(p
, start
, end
, 0);
877 extent_clear_unlock_delalloc(inode
, start
, end
,
881 free_async_extent_pages(async_extent
);
883 alloc_hint
= ins
.objectid
+ ins
.offset
;
889 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
890 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 1);
892 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
893 async_extent
->start
+
894 async_extent
->ram_size
- 1,
895 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
|
896 EXTENT_DELALLOC_NEW
|
897 EXTENT_DEFRAG
| EXTENT_DO_ACCOUNTING
,
898 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
899 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
|
901 free_async_extent_pages(async_extent
);
906 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
909 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
910 struct extent_map
*em
;
913 read_lock(&em_tree
->lock
);
914 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
917 * if block start isn't an actual block number then find the
918 * first block in this inode and use that as a hint. If that
919 * block is also bogus then just don't worry about it.
921 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
923 em
= search_extent_mapping(em_tree
, 0, 0);
924 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
925 alloc_hint
= em
->block_start
;
929 alloc_hint
= em
->block_start
;
933 read_unlock(&em_tree
->lock
);
939 * when extent_io.c finds a delayed allocation range in the file,
940 * the call backs end up in this code. The basic idea is to
941 * allocate extents on disk for the range, and create ordered data structs
942 * in ram to track those extents.
944 * locked_page is the page that writepage had locked already. We use
945 * it to make sure we don't do extra locks or unlocks.
947 * *page_started is set to one if we unlock locked_page and do everything
948 * required to start IO on it. It may be clean and already done with
951 static noinline
int cow_file_range(struct inode
*inode
,
952 struct page
*locked_page
,
953 u64 start
, u64 end
, int *page_started
,
954 unsigned long *nr_written
, int unlock
)
956 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
957 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
960 unsigned long ram_size
;
961 u64 cur_alloc_size
= 0;
962 u64 blocksize
= fs_info
->sectorsize
;
963 struct btrfs_key ins
;
964 struct extent_map
*em
;
966 unsigned long page_ops
;
967 bool extent_reserved
= false;
970 if (btrfs_is_free_space_inode(BTRFS_I(inode
))) {
976 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
977 num_bytes
= max(blocksize
, num_bytes
);
978 ASSERT(num_bytes
<= btrfs_super_total_bytes(fs_info
->super_copy
));
980 inode_should_defrag(BTRFS_I(inode
), start
, end
, num_bytes
, SZ_64K
);
983 /* lets try to make an inline extent */
984 ret
= cow_file_range_inline(inode
, start
, end
, 0,
985 BTRFS_COMPRESS_NONE
, NULL
);
988 * We use DO_ACCOUNTING here because we need the
989 * delalloc_release_metadata to be run _after_ we drop
990 * our outstanding extent for clearing delalloc for this
993 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
,
994 EXTENT_LOCKED
| EXTENT_DELALLOC
|
995 EXTENT_DELALLOC_NEW
| EXTENT_DEFRAG
|
996 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
997 PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
999 *nr_written
= *nr_written
+
1000 (end
- start
+ PAGE_SIZE
) / PAGE_SIZE
;
1003 } else if (ret
< 0) {
1008 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
1009 btrfs_drop_extent_cache(BTRFS_I(inode
), start
,
1010 start
+ num_bytes
- 1, 0);
1012 while (num_bytes
> 0) {
1013 cur_alloc_size
= num_bytes
;
1014 ret
= btrfs_reserve_extent(root
, cur_alloc_size
, cur_alloc_size
,
1015 fs_info
->sectorsize
, 0, alloc_hint
,
1019 cur_alloc_size
= ins
.offset
;
1020 extent_reserved
= true;
1022 ram_size
= ins
.offset
;
1023 em
= create_io_em(inode
, start
, ins
.offset
, /* len */
1024 start
, /* orig_start */
1025 ins
.objectid
, /* block_start */
1026 ins
.offset
, /* block_len */
1027 ins
.offset
, /* orig_block_len */
1028 ram_size
, /* ram_bytes */
1029 BTRFS_COMPRESS_NONE
, /* compress_type */
1030 BTRFS_ORDERED_REGULAR
/* type */);
1035 free_extent_map(em
);
1037 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
1038 ram_size
, cur_alloc_size
, 0);
1040 goto out_drop_extent_cache
;
1042 if (root
->root_key
.objectid
==
1043 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1044 ret
= btrfs_reloc_clone_csums(inode
, start
,
1047 * Only drop cache here, and process as normal.
1049 * We must not allow extent_clear_unlock_delalloc()
1050 * at out_unlock label to free meta of this ordered
1051 * extent, as its meta should be freed by
1052 * btrfs_finish_ordered_io().
1054 * So we must continue until @start is increased to
1055 * skip current ordered extent.
1058 btrfs_drop_extent_cache(BTRFS_I(inode
), start
,
1059 start
+ ram_size
- 1, 0);
1062 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
1064 /* we're not doing compressed IO, don't unlock the first
1065 * page (which the caller expects to stay locked), don't
1066 * clear any dirty bits and don't set any writeback bits
1068 * Do set the Private2 bit so we know this page was properly
1069 * setup for writepage
1071 page_ops
= unlock
? PAGE_UNLOCK
: 0;
1072 page_ops
|= PAGE_SET_PRIVATE2
;
1074 extent_clear_unlock_delalloc(inode
, start
,
1075 start
+ ram_size
- 1,
1077 EXTENT_LOCKED
| EXTENT_DELALLOC
,
1079 if (num_bytes
< cur_alloc_size
)
1082 num_bytes
-= cur_alloc_size
;
1083 alloc_hint
= ins
.objectid
+ ins
.offset
;
1084 start
+= cur_alloc_size
;
1085 extent_reserved
= false;
1088 * btrfs_reloc_clone_csums() error, since start is increased
1089 * extent_clear_unlock_delalloc() at out_unlock label won't
1090 * free metadata of current ordered extent, we're OK to exit.
1098 out_drop_extent_cache
:
1099 btrfs_drop_extent_cache(BTRFS_I(inode
), start
, start
+ ram_size
- 1, 0);
1101 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
1102 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 1);
1104 clear_bits
= EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DELALLOC_NEW
|
1105 EXTENT_DEFRAG
| EXTENT_CLEAR_META_RESV
;
1106 page_ops
= PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
1109 * If we reserved an extent for our delalloc range (or a subrange) and
1110 * failed to create the respective ordered extent, then it means that
1111 * when we reserved the extent we decremented the extent's size from
1112 * the data space_info's bytes_may_use counter and incremented the
1113 * space_info's bytes_reserved counter by the same amount. We must make
1114 * sure extent_clear_unlock_delalloc() does not try to decrement again
1115 * the data space_info's bytes_may_use counter, therefore we do not pass
1116 * it the flag EXTENT_CLEAR_DATA_RESV.
1118 if (extent_reserved
) {
1119 extent_clear_unlock_delalloc(inode
, start
,
1120 start
+ cur_alloc_size
,
1124 start
+= cur_alloc_size
;
1128 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1129 clear_bits
| EXTENT_CLEAR_DATA_RESV
,
1135 * work queue call back to started compression on a file and pages
1137 static noinline
void async_cow_start(struct btrfs_work
*work
)
1139 struct async_chunk
*async_chunk
;
1140 int compressed_extents
;
1142 async_chunk
= container_of(work
, struct async_chunk
, work
);
1144 compressed_extents
= compress_file_range(async_chunk
);
1145 if (compressed_extents
== 0) {
1146 btrfs_add_delayed_iput(async_chunk
->inode
);
1147 async_chunk
->inode
= NULL
;
1152 * work queue call back to submit previously compressed pages
1154 static noinline
void async_cow_submit(struct btrfs_work
*work
)
1156 struct async_chunk
*async_chunk
= container_of(work
, struct async_chunk
,
1158 struct btrfs_fs_info
*fs_info
= btrfs_work_owner(work
);
1159 unsigned long nr_pages
;
1161 nr_pages
= (async_chunk
->end
- async_chunk
->start
+ PAGE_SIZE
) >>
1164 /* atomic_sub_return implies a barrier */
1165 if (atomic_sub_return(nr_pages
, &fs_info
->async_delalloc_pages
) <
1167 cond_wake_up_nomb(&fs_info
->async_submit_wait
);
1170 * ->inode could be NULL if async_chunk_start has failed to compress,
1171 * in which case we don't have anything to submit, yet we need to
1172 * always adjust ->async_delalloc_pages as its paired with the init
1173 * happening in cow_file_range_async
1175 if (async_chunk
->inode
)
1176 submit_compressed_extents(async_chunk
);
1179 static noinline
void async_cow_free(struct btrfs_work
*work
)
1181 struct async_chunk
*async_chunk
;
1183 async_chunk
= container_of(work
, struct async_chunk
, work
);
1184 if (async_chunk
->inode
)
1185 btrfs_add_delayed_iput(async_chunk
->inode
);
1187 * Since the pointer to 'pending' is at the beginning of the array of
1188 * async_chunk's, freeing it ensures the whole array has been freed.
1190 if (atomic_dec_and_test(async_chunk
->pending
))
1191 kvfree(async_chunk
->pending
);
1194 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1195 u64 start
, u64 end
, int *page_started
,
1196 unsigned long *nr_written
,
1197 unsigned int write_flags
)
1199 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1200 struct async_cow
*ctx
;
1201 struct async_chunk
*async_chunk
;
1202 unsigned long nr_pages
;
1204 u64 num_chunks
= DIV_ROUND_UP(end
- start
, SZ_512K
);
1206 bool should_compress
;
1209 unlock_extent(&BTRFS_I(inode
)->io_tree
, start
, end
);
1211 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
&&
1212 !btrfs_test_opt(fs_info
, FORCE_COMPRESS
)) {
1214 should_compress
= false;
1216 should_compress
= true;
1219 nofs_flag
= memalloc_nofs_save();
1220 ctx
= kvmalloc(struct_size(ctx
, chunks
, num_chunks
), GFP_KERNEL
);
1221 memalloc_nofs_restore(nofs_flag
);
1224 unsigned clear_bits
= EXTENT_LOCKED
| EXTENT_DELALLOC
|
1225 EXTENT_DELALLOC_NEW
| EXTENT_DEFRAG
|
1226 EXTENT_DO_ACCOUNTING
;
1227 unsigned long page_ops
= PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
1228 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
|
1231 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1232 clear_bits
, page_ops
);
1236 async_chunk
= ctx
->chunks
;
1237 atomic_set(&ctx
->num_chunks
, num_chunks
);
1239 for (i
= 0; i
< num_chunks
; i
++) {
1240 if (should_compress
)
1241 cur_end
= min(end
, start
+ SZ_512K
- 1);
1246 * igrab is called higher up in the call chain, take only the
1247 * lightweight reference for the callback lifetime
1250 async_chunk
[i
].pending
= &ctx
->num_chunks
;
1251 async_chunk
[i
].inode
= inode
;
1252 async_chunk
[i
].start
= start
;
1253 async_chunk
[i
].end
= cur_end
;
1254 async_chunk
[i
].locked_page
= locked_page
;
1255 async_chunk
[i
].write_flags
= write_flags
;
1256 INIT_LIST_HEAD(&async_chunk
[i
].extents
);
1258 btrfs_init_work(&async_chunk
[i
].work
,
1259 btrfs_delalloc_helper
,
1260 async_cow_start
, async_cow_submit
,
1263 nr_pages
= DIV_ROUND_UP(cur_end
- start
, PAGE_SIZE
);
1264 atomic_add(nr_pages
, &fs_info
->async_delalloc_pages
);
1266 btrfs_queue_work(fs_info
->delalloc_workers
, &async_chunk
[i
].work
);
1268 *nr_written
+= nr_pages
;
1269 start
= cur_end
+ 1;
1275 static noinline
int csum_exist_in_range(struct btrfs_fs_info
*fs_info
,
1276 u64 bytenr
, u64 num_bytes
)
1279 struct btrfs_ordered_sum
*sums
;
1282 ret
= btrfs_lookup_csums_range(fs_info
->csum_root
, bytenr
,
1283 bytenr
+ num_bytes
- 1, &list
, 0);
1284 if (ret
== 0 && list_empty(&list
))
1287 while (!list_empty(&list
)) {
1288 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1289 list_del(&sums
->list
);
1298 * when nowcow writeback call back. This checks for snapshots or COW copies
1299 * of the extents that exist in the file, and COWs the file as required.
1301 * If no cow copies or snapshots exist, we write directly to the existing
1304 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1305 struct page
*locked_page
,
1306 const u64 start
, const u64 end
,
1307 int *page_started
, int force
,
1308 unsigned long *nr_written
)
1310 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1311 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1312 struct btrfs_path
*path
;
1313 u64 cow_start
= (u64
)-1;
1314 u64 cur_offset
= start
;
1316 bool check_prev
= true;
1317 const bool freespace_inode
= btrfs_is_free_space_inode(BTRFS_I(inode
));
1318 u64 ino
= btrfs_ino(BTRFS_I(inode
));
1320 u64 disk_bytenr
= 0;
1322 path
= btrfs_alloc_path();
1324 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1325 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1326 EXTENT_DO_ACCOUNTING
|
1327 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1329 PAGE_SET_WRITEBACK
|
1330 PAGE_END_WRITEBACK
);
1335 struct btrfs_key found_key
;
1336 struct btrfs_file_extent_item
*fi
;
1337 struct extent_buffer
*leaf
;
1347 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, ino
,
1353 * If there is no extent for our range when doing the initial
1354 * search, then go back to the previous slot as it will be the
1355 * one containing the search offset
1357 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1358 leaf
= path
->nodes
[0];
1359 btrfs_item_key_to_cpu(leaf
, &found_key
,
1360 path
->slots
[0] - 1);
1361 if (found_key
.objectid
== ino
&&
1362 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1367 /* Go to next leaf if we have exhausted the current one */
1368 leaf
= path
->nodes
[0];
1369 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1370 ret
= btrfs_next_leaf(root
, path
);
1372 if (cow_start
!= (u64
)-1)
1373 cur_offset
= cow_start
;
1378 leaf
= path
->nodes
[0];
1381 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1383 /* Didn't find anything for our INO */
1384 if (found_key
.objectid
> ino
)
1387 * Keep searching until we find an EXTENT_ITEM or there are no
1388 * more extents for this inode
1390 if (WARN_ON_ONCE(found_key
.objectid
< ino
) ||
1391 found_key
.type
< BTRFS_EXTENT_DATA_KEY
) {
1396 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1397 if (found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1398 found_key
.offset
> end
)
1402 * If the found extent starts after requested offset, then
1403 * adjust extent_end to be right before this extent begins
1405 if (found_key
.offset
> cur_offset
) {
1406 extent_end
= found_key
.offset
;
1412 * Found extent which begins before our range and potentially
1415 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1416 struct btrfs_file_extent_item
);
1417 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1419 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
1420 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1421 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1422 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1423 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1424 extent_end
= found_key
.offset
+
1425 btrfs_file_extent_num_bytes(leaf
, fi
);
1427 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1429 * If extent we got ends before our range starts, skip
1432 if (extent_end
<= start
) {
1437 if (disk_bytenr
== 0)
1439 /* Skip compressed/encrypted/encoded extents */
1440 if (btrfs_file_extent_compression(leaf
, fi
) ||
1441 btrfs_file_extent_encryption(leaf
, fi
) ||
1442 btrfs_file_extent_other_encoding(leaf
, fi
))
1445 * If extent is created before the last volume's snapshot
1446 * this implies the extent is shared, hence we can't do
1447 * nocow. This is the same check as in
1448 * btrfs_cross_ref_exist but without calling
1449 * btrfs_search_slot.
1451 if (!freespace_inode
&&
1452 btrfs_file_extent_generation(leaf
, fi
) <=
1453 btrfs_root_last_snapshot(&root
->root_item
))
1455 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1457 /* If extent is RO, we must COW it */
1458 if (btrfs_extent_readonly(fs_info
, disk_bytenr
))
1460 ret
= btrfs_cross_ref_exist(root
, ino
,
1462 extent_offset
, disk_bytenr
);
1465 * ret could be -EIO if the above fails to read
1469 if (cow_start
!= (u64
)-1)
1470 cur_offset
= cow_start
;
1474 WARN_ON_ONCE(freespace_inode
);
1477 disk_bytenr
+= extent_offset
;
1478 disk_bytenr
+= cur_offset
- found_key
.offset
;
1479 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1481 * If there are pending snapshots for this root, we
1482 * fall into common COW way
1484 if (!freespace_inode
&& atomic_read(&root
->snapshot_force_cow
))
1487 * force cow if csum exists in the range.
1488 * this ensure that csum for a given extent are
1489 * either valid or do not exist.
1491 ret
= csum_exist_in_range(fs_info
, disk_bytenr
,
1495 * ret could be -EIO if the above fails to read
1499 if (cow_start
!= (u64
)-1)
1500 cur_offset
= cow_start
;
1503 WARN_ON_ONCE(freespace_inode
);
1506 if (!btrfs_inc_nocow_writers(fs_info
, disk_bytenr
))
1509 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1510 extent_end
= found_key
.offset
+ ram_bytes
;
1511 extent_end
= ALIGN(extent_end
, fs_info
->sectorsize
);
1512 /* Skip extents outside of our requested range */
1513 if (extent_end
<= start
) {
1518 /* If this triggers then we have a memory corruption */
1523 * If nocow is false then record the beginning of the range
1524 * that needs to be COWed
1527 if (cow_start
== (u64
)-1)
1528 cow_start
= cur_offset
;
1529 cur_offset
= extent_end
;
1530 if (cur_offset
> end
)
1536 btrfs_release_path(path
);
1539 * COW range from cow_start to found_key.offset - 1. As the key
1540 * will contain the beginning of the first extent that can be
1541 * NOCOW, following one which needs to be COW'ed
1543 if (cow_start
!= (u64
)-1) {
1544 ret
= cow_file_range(inode
, locked_page
,
1545 cow_start
, found_key
.offset
- 1,
1546 page_started
, nr_written
, 1);
1549 btrfs_dec_nocow_writers(fs_info
,
1553 cow_start
= (u64
)-1;
1556 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1557 u64 orig_start
= found_key
.offset
- extent_offset
;
1558 struct extent_map
*em
;
1560 em
= create_io_em(inode
, cur_offset
, num_bytes
,
1562 disk_bytenr
, /* block_start */
1563 num_bytes
, /* block_len */
1564 disk_num_bytes
, /* orig_block_len */
1565 ram_bytes
, BTRFS_COMPRESS_NONE
,
1566 BTRFS_ORDERED_PREALLOC
);
1569 btrfs_dec_nocow_writers(fs_info
,
1574 free_extent_map(em
);
1575 ret
= btrfs_add_ordered_extent(inode
, cur_offset
,
1576 disk_bytenr
, num_bytes
,
1578 BTRFS_ORDERED_PREALLOC
);
1580 btrfs_drop_extent_cache(BTRFS_I(inode
),
1582 cur_offset
+ num_bytes
- 1,
1587 ret
= btrfs_add_ordered_extent(inode
, cur_offset
,
1588 disk_bytenr
, num_bytes
,
1590 BTRFS_ORDERED_NOCOW
);
1596 btrfs_dec_nocow_writers(fs_info
, disk_bytenr
);
1599 if (root
->root_key
.objectid
==
1600 BTRFS_DATA_RELOC_TREE_OBJECTID
)
1602 * Error handled later, as we must prevent
1603 * extent_clear_unlock_delalloc() in error handler
1604 * from freeing metadata of created ordered extent.
1606 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1609 extent_clear_unlock_delalloc(inode
, cur_offset
,
1610 cur_offset
+ num_bytes
- 1,
1611 locked_page
, EXTENT_LOCKED
|
1613 EXTENT_CLEAR_DATA_RESV
,
1614 PAGE_UNLOCK
| PAGE_SET_PRIVATE2
);
1616 cur_offset
= extent_end
;
1619 * btrfs_reloc_clone_csums() error, now we're OK to call error
1620 * handler, as metadata for created ordered extent will only
1621 * be freed by btrfs_finish_ordered_io().
1625 if (cur_offset
> end
)
1628 btrfs_release_path(path
);
1630 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1631 cow_start
= cur_offset
;
1633 if (cow_start
!= (u64
)-1) {
1635 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1636 page_started
, nr_written
, 1);
1643 btrfs_dec_nocow_writers(fs_info
, disk_bytenr
);
1645 if (ret
&& cur_offset
< end
)
1646 extent_clear_unlock_delalloc(inode
, cur_offset
, end
,
1647 locked_page
, EXTENT_LOCKED
|
1648 EXTENT_DELALLOC
| EXTENT_DEFRAG
|
1649 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
1651 PAGE_SET_WRITEBACK
|
1652 PAGE_END_WRITEBACK
);
1653 btrfs_free_path(path
);
1657 static inline int need_force_cow(struct inode
*inode
, u64 start
, u64 end
)
1660 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
1661 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
))
1665 * @defrag_bytes is a hint value, no spinlock held here,
1666 * if is not zero, it means the file is defragging.
1667 * Force cow if given extent needs to be defragged.
1669 if (BTRFS_I(inode
)->defrag_bytes
&&
1670 test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1671 EXTENT_DEFRAG
, 0, NULL
))
1678 * Function to process delayed allocation (create CoW) for ranges which are
1679 * being touched for the first time.
1681 int btrfs_run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1682 u64 start
, u64 end
, int *page_started
, unsigned long *nr_written
,
1683 struct writeback_control
*wbc
)
1686 int force_cow
= need_force_cow(inode
, start
, end
);
1687 unsigned int write_flags
= wbc_to_write_flags(wbc
);
1689 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
&& !force_cow
) {
1690 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1691 page_started
, 1, nr_written
);
1692 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
&& !force_cow
) {
1693 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1694 page_started
, 0, nr_written
);
1695 } else if (!inode_can_compress(inode
) ||
1696 !inode_need_compress(inode
, start
, end
)) {
1697 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1698 page_started
, nr_written
, 1);
1700 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1701 &BTRFS_I(inode
)->runtime_flags
);
1702 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1703 page_started
, nr_written
,
1707 btrfs_cleanup_ordered_extents(inode
, locked_page
, start
,
1712 void btrfs_split_delalloc_extent(struct inode
*inode
,
1713 struct extent_state
*orig
, u64 split
)
1717 /* not delalloc, ignore it */
1718 if (!(orig
->state
& EXTENT_DELALLOC
))
1721 size
= orig
->end
- orig
->start
+ 1;
1722 if (size
> BTRFS_MAX_EXTENT_SIZE
) {
1727 * See the explanation in btrfs_merge_delalloc_extent, the same
1728 * applies here, just in reverse.
1730 new_size
= orig
->end
- split
+ 1;
1731 num_extents
= count_max_extents(new_size
);
1732 new_size
= split
- orig
->start
;
1733 num_extents
+= count_max_extents(new_size
);
1734 if (count_max_extents(size
) >= num_extents
)
1738 spin_lock(&BTRFS_I(inode
)->lock
);
1739 btrfs_mod_outstanding_extents(BTRFS_I(inode
), 1);
1740 spin_unlock(&BTRFS_I(inode
)->lock
);
1744 * Handle merged delayed allocation extents so we can keep track of new extents
1745 * that are just merged onto old extents, such as when we are doing sequential
1746 * writes, so we can properly account for the metadata space we'll need.
1748 void btrfs_merge_delalloc_extent(struct inode
*inode
, struct extent_state
*new,
1749 struct extent_state
*other
)
1751 u64 new_size
, old_size
;
1754 /* not delalloc, ignore it */
1755 if (!(other
->state
& EXTENT_DELALLOC
))
1758 if (new->start
> other
->start
)
1759 new_size
= new->end
- other
->start
+ 1;
1761 new_size
= other
->end
- new->start
+ 1;
1763 /* we're not bigger than the max, unreserve the space and go */
1764 if (new_size
<= BTRFS_MAX_EXTENT_SIZE
) {
1765 spin_lock(&BTRFS_I(inode
)->lock
);
1766 btrfs_mod_outstanding_extents(BTRFS_I(inode
), -1);
1767 spin_unlock(&BTRFS_I(inode
)->lock
);
1772 * We have to add up either side to figure out how many extents were
1773 * accounted for before we merged into one big extent. If the number of
1774 * extents we accounted for is <= the amount we need for the new range
1775 * then we can return, otherwise drop. Think of it like this
1779 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1780 * need 2 outstanding extents, on one side we have 1 and the other side
1781 * we have 1 so they are == and we can return. But in this case
1783 * [MAX_SIZE+4k][MAX_SIZE+4k]
1785 * Each range on their own accounts for 2 extents, but merged together
1786 * they are only 3 extents worth of accounting, so we need to drop in
1789 old_size
= other
->end
- other
->start
+ 1;
1790 num_extents
= count_max_extents(old_size
);
1791 old_size
= new->end
- new->start
+ 1;
1792 num_extents
+= count_max_extents(old_size
);
1793 if (count_max_extents(new_size
) >= num_extents
)
1796 spin_lock(&BTRFS_I(inode
)->lock
);
1797 btrfs_mod_outstanding_extents(BTRFS_I(inode
), -1);
1798 spin_unlock(&BTRFS_I(inode
)->lock
);
1801 static void btrfs_add_delalloc_inodes(struct btrfs_root
*root
,
1802 struct inode
*inode
)
1804 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1806 spin_lock(&root
->delalloc_lock
);
1807 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1808 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1809 &root
->delalloc_inodes
);
1810 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1811 &BTRFS_I(inode
)->runtime_flags
);
1812 root
->nr_delalloc_inodes
++;
1813 if (root
->nr_delalloc_inodes
== 1) {
1814 spin_lock(&fs_info
->delalloc_root_lock
);
1815 BUG_ON(!list_empty(&root
->delalloc_root
));
1816 list_add_tail(&root
->delalloc_root
,
1817 &fs_info
->delalloc_roots
);
1818 spin_unlock(&fs_info
->delalloc_root_lock
);
1821 spin_unlock(&root
->delalloc_lock
);
1825 void __btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1826 struct btrfs_inode
*inode
)
1828 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1830 if (!list_empty(&inode
->delalloc_inodes
)) {
1831 list_del_init(&inode
->delalloc_inodes
);
1832 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1833 &inode
->runtime_flags
);
1834 root
->nr_delalloc_inodes
--;
1835 if (!root
->nr_delalloc_inodes
) {
1836 ASSERT(list_empty(&root
->delalloc_inodes
));
1837 spin_lock(&fs_info
->delalloc_root_lock
);
1838 BUG_ON(list_empty(&root
->delalloc_root
));
1839 list_del_init(&root
->delalloc_root
);
1840 spin_unlock(&fs_info
->delalloc_root_lock
);
1845 static void btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1846 struct btrfs_inode
*inode
)
1848 spin_lock(&root
->delalloc_lock
);
1849 __btrfs_del_delalloc_inode(root
, inode
);
1850 spin_unlock(&root
->delalloc_lock
);
1854 * Properly track delayed allocation bytes in the inode and to maintain the
1855 * list of inodes that have pending delalloc work to be done.
1857 void btrfs_set_delalloc_extent(struct inode
*inode
, struct extent_state
*state
,
1860 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1862 if ((*bits
& EXTENT_DEFRAG
) && !(*bits
& EXTENT_DELALLOC
))
1865 * set_bit and clear bit hooks normally require _irqsave/restore
1866 * but in this case, we are only testing for the DELALLOC
1867 * bit, which is only set or cleared with irqs on
1869 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1870 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1871 u64 len
= state
->end
+ 1 - state
->start
;
1872 u32 num_extents
= count_max_extents(len
);
1873 bool do_list
= !btrfs_is_free_space_inode(BTRFS_I(inode
));
1875 spin_lock(&BTRFS_I(inode
)->lock
);
1876 btrfs_mod_outstanding_extents(BTRFS_I(inode
), num_extents
);
1877 spin_unlock(&BTRFS_I(inode
)->lock
);
1879 /* For sanity tests */
1880 if (btrfs_is_testing(fs_info
))
1883 percpu_counter_add_batch(&fs_info
->delalloc_bytes
, len
,
1884 fs_info
->delalloc_batch
);
1885 spin_lock(&BTRFS_I(inode
)->lock
);
1886 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1887 if (*bits
& EXTENT_DEFRAG
)
1888 BTRFS_I(inode
)->defrag_bytes
+= len
;
1889 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1890 &BTRFS_I(inode
)->runtime_flags
))
1891 btrfs_add_delalloc_inodes(root
, inode
);
1892 spin_unlock(&BTRFS_I(inode
)->lock
);
1895 if (!(state
->state
& EXTENT_DELALLOC_NEW
) &&
1896 (*bits
& EXTENT_DELALLOC_NEW
)) {
1897 spin_lock(&BTRFS_I(inode
)->lock
);
1898 BTRFS_I(inode
)->new_delalloc_bytes
+= state
->end
+ 1 -
1900 spin_unlock(&BTRFS_I(inode
)->lock
);
1905 * Once a range is no longer delalloc this function ensures that proper
1906 * accounting happens.
1908 void btrfs_clear_delalloc_extent(struct inode
*vfs_inode
,
1909 struct extent_state
*state
, unsigned *bits
)
1911 struct btrfs_inode
*inode
= BTRFS_I(vfs_inode
);
1912 struct btrfs_fs_info
*fs_info
= btrfs_sb(vfs_inode
->i_sb
);
1913 u64 len
= state
->end
+ 1 - state
->start
;
1914 u32 num_extents
= count_max_extents(len
);
1916 if ((state
->state
& EXTENT_DEFRAG
) && (*bits
& EXTENT_DEFRAG
)) {
1917 spin_lock(&inode
->lock
);
1918 inode
->defrag_bytes
-= len
;
1919 spin_unlock(&inode
->lock
);
1923 * set_bit and clear bit hooks normally require _irqsave/restore
1924 * but in this case, we are only testing for the DELALLOC
1925 * bit, which is only set or cleared with irqs on
1927 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1928 struct btrfs_root
*root
= inode
->root
;
1929 bool do_list
= !btrfs_is_free_space_inode(inode
);
1931 spin_lock(&inode
->lock
);
1932 btrfs_mod_outstanding_extents(inode
, -num_extents
);
1933 spin_unlock(&inode
->lock
);
1936 * We don't reserve metadata space for space cache inodes so we
1937 * don't need to call delalloc_release_metadata if there is an
1940 if (*bits
& EXTENT_CLEAR_META_RESV
&&
1941 root
!= fs_info
->tree_root
)
1942 btrfs_delalloc_release_metadata(inode
, len
, false);
1944 /* For sanity tests. */
1945 if (btrfs_is_testing(fs_info
))
1948 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1949 do_list
&& !(state
->state
& EXTENT_NORESERVE
) &&
1950 (*bits
& EXTENT_CLEAR_DATA_RESV
))
1951 btrfs_free_reserved_data_space_noquota(
1955 percpu_counter_add_batch(&fs_info
->delalloc_bytes
, -len
,
1956 fs_info
->delalloc_batch
);
1957 spin_lock(&inode
->lock
);
1958 inode
->delalloc_bytes
-= len
;
1959 if (do_list
&& inode
->delalloc_bytes
== 0 &&
1960 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1961 &inode
->runtime_flags
))
1962 btrfs_del_delalloc_inode(root
, inode
);
1963 spin_unlock(&inode
->lock
);
1966 if ((state
->state
& EXTENT_DELALLOC_NEW
) &&
1967 (*bits
& EXTENT_DELALLOC_NEW
)) {
1968 spin_lock(&inode
->lock
);
1969 ASSERT(inode
->new_delalloc_bytes
>= len
);
1970 inode
->new_delalloc_bytes
-= len
;
1971 spin_unlock(&inode
->lock
);
1976 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
1977 * in a chunk's stripe. This function ensures that bios do not span a
1980 * @page - The page we are about to add to the bio
1981 * @size - size we want to add to the bio
1982 * @bio - bio we want to ensure is smaller than a stripe
1983 * @bio_flags - flags of the bio
1985 * return 1 if page cannot be added to the bio
1986 * return 0 if page can be added to the bio
1987 * return error otherwise
1989 int btrfs_bio_fits_in_stripe(struct page
*page
, size_t size
, struct bio
*bio
,
1990 unsigned long bio_flags
)
1992 struct inode
*inode
= page
->mapping
->host
;
1993 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
1994 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
1998 struct btrfs_io_geometry geom
;
2000 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
2003 length
= bio
->bi_iter
.bi_size
;
2004 map_length
= length
;
2005 ret
= btrfs_get_io_geometry(fs_info
, btrfs_op(bio
), logical
, map_length
,
2010 if (geom
.len
< length
+ size
)
2016 * in order to insert checksums into the metadata in large chunks,
2017 * we wait until bio submission time. All the pages in the bio are
2018 * checksummed and sums are attached onto the ordered extent record.
2020 * At IO completion time the cums attached on the ordered extent record
2021 * are inserted into the btree
2023 static blk_status_t
btrfs_submit_bio_start(void *private_data
, struct bio
*bio
,
2026 struct inode
*inode
= private_data
;
2027 blk_status_t ret
= 0;
2029 ret
= btrfs_csum_one_bio(inode
, bio
, 0, 0);
2030 BUG_ON(ret
); /* -ENOMEM */
2035 * extent_io.c submission hook. This does the right thing for csum calculation
2036 * on write, or reading the csums from the tree before a read.
2038 * Rules about async/sync submit,
2039 * a) read: sync submit
2041 * b) write without checksum: sync submit
2043 * c) write with checksum:
2044 * c-1) if bio is issued by fsync: sync submit
2045 * (sync_writers != 0)
2047 * c-2) if root is reloc root: sync submit
2048 * (only in case of buffered IO)
2050 * c-3) otherwise: async submit
2052 static blk_status_t
btrfs_submit_bio_hook(struct inode
*inode
, struct bio
*bio
,
2054 unsigned long bio_flags
)
2057 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2058 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2059 enum btrfs_wq_endio_type metadata
= BTRFS_WQ_ENDIO_DATA
;
2060 blk_status_t ret
= 0;
2062 int async
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
2064 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
2066 if (btrfs_is_free_space_inode(BTRFS_I(inode
)))
2067 metadata
= BTRFS_WQ_ENDIO_FREE_SPACE
;
2069 if (bio_op(bio
) != REQ_OP_WRITE
) {
2070 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, metadata
);
2074 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
2075 ret
= btrfs_submit_compressed_read(inode
, bio
,
2079 } else if (!skip_sum
) {
2080 ret
= btrfs_lookup_bio_sums(inode
, bio
, NULL
);
2085 } else if (async
&& !skip_sum
) {
2086 /* csum items have already been cloned */
2087 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2089 /* we're doing a write, do the async checksumming */
2090 ret
= btrfs_wq_submit_bio(fs_info
, bio
, mirror_num
, bio_flags
,
2091 0, inode
, btrfs_submit_bio_start
);
2093 } else if (!skip_sum
) {
2094 ret
= btrfs_csum_one_bio(inode
, bio
, 0, 0);
2100 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
2104 bio
->bi_status
= ret
;
2111 * given a list of ordered sums record them in the inode. This happens
2112 * at IO completion time based on sums calculated at bio submission time.
2114 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
2115 struct inode
*inode
, struct list_head
*list
)
2117 struct btrfs_ordered_sum
*sum
;
2120 list_for_each_entry(sum
, list
, list
) {
2121 trans
->adding_csums
= true;
2122 ret
= btrfs_csum_file_blocks(trans
,
2123 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
2124 trans
->adding_csums
= false;
2131 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
2132 unsigned int extra_bits
,
2133 struct extent_state
**cached_state
)
2135 WARN_ON(PAGE_ALIGNED(end
));
2136 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
2137 extra_bits
, cached_state
);
2140 /* see btrfs_writepage_start_hook for details on why this is required */
2141 struct btrfs_writepage_fixup
{
2143 struct btrfs_work work
;
2146 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
2148 struct btrfs_writepage_fixup
*fixup
;
2149 struct btrfs_ordered_extent
*ordered
;
2150 struct extent_state
*cached_state
= NULL
;
2151 struct extent_changeset
*data_reserved
= NULL
;
2153 struct inode
*inode
;
2158 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
2162 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
2163 ClearPageChecked(page
);
2167 inode
= page
->mapping
->host
;
2168 page_start
= page_offset(page
);
2169 page_end
= page_offset(page
) + PAGE_SIZE
- 1;
2171 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2174 /* already ordered? We're done */
2175 if (PagePrivate2(page
))
2178 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), page_start
,
2181 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
2182 page_end
, &cached_state
);
2184 btrfs_start_ordered_extent(inode
, ordered
, 1);
2185 btrfs_put_ordered_extent(ordered
);
2189 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
, page_start
,
2192 mapping_set_error(page
->mapping
, ret
);
2193 end_extent_writepage(page
, ret
, page_start
, page_end
);
2194 ClearPageChecked(page
);
2198 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
, 0,
2201 mapping_set_error(page
->mapping
, ret
);
2202 end_extent_writepage(page
, ret
, page_start
, page_end
);
2203 ClearPageChecked(page
);
2207 ClearPageChecked(page
);
2208 set_page_dirty(page
);
2209 btrfs_delalloc_release_extents(BTRFS_I(inode
), PAGE_SIZE
, false);
2211 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2217 extent_changeset_free(data_reserved
);
2221 * There are a few paths in the higher layers of the kernel that directly
2222 * set the page dirty bit without asking the filesystem if it is a
2223 * good idea. This causes problems because we want to make sure COW
2224 * properly happens and the data=ordered rules are followed.
2226 * In our case any range that doesn't have the ORDERED bit set
2227 * hasn't been properly setup for IO. We kick off an async process
2228 * to fix it up. The async helper will wait for ordered extents, set
2229 * the delalloc bit and make it safe to write the page.
2231 int btrfs_writepage_cow_fixup(struct page
*page
, u64 start
, u64 end
)
2233 struct inode
*inode
= page
->mapping
->host
;
2234 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2235 struct btrfs_writepage_fixup
*fixup
;
2237 /* this page is properly in the ordered list */
2238 if (TestClearPagePrivate2(page
))
2241 if (PageChecked(page
))
2244 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
2248 SetPageChecked(page
);
2250 btrfs_init_work(&fixup
->work
, btrfs_fixup_helper
,
2251 btrfs_writepage_fixup_worker
, NULL
, NULL
);
2253 btrfs_queue_work(fs_info
->fixup_workers
, &fixup
->work
);
2257 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
2258 struct inode
*inode
, u64 file_pos
,
2259 u64 disk_bytenr
, u64 disk_num_bytes
,
2260 u64 num_bytes
, u64 ram_bytes
,
2261 u8 compression
, u8 encryption
,
2262 u16 other_encoding
, int extent_type
)
2264 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2265 struct btrfs_file_extent_item
*fi
;
2266 struct btrfs_path
*path
;
2267 struct extent_buffer
*leaf
;
2268 struct btrfs_key ins
;
2270 int extent_inserted
= 0;
2273 path
= btrfs_alloc_path();
2278 * we may be replacing one extent in the tree with another.
2279 * The new extent is pinned in the extent map, and we don't want
2280 * to drop it from the cache until it is completely in the btree.
2282 * So, tell btrfs_drop_extents to leave this extent in the cache.
2283 * the caller is expected to unpin it and allow it to be merged
2286 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, file_pos
,
2287 file_pos
+ num_bytes
, NULL
, 0,
2288 1, sizeof(*fi
), &extent_inserted
);
2292 if (!extent_inserted
) {
2293 ins
.objectid
= btrfs_ino(BTRFS_I(inode
));
2294 ins
.offset
= file_pos
;
2295 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
2297 path
->leave_spinning
= 1;
2298 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
2303 leaf
= path
->nodes
[0];
2304 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2305 struct btrfs_file_extent_item
);
2306 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
2307 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
2308 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
2309 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
2310 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2311 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2312 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
2313 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
2314 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
2315 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
2317 btrfs_mark_buffer_dirty(leaf
);
2318 btrfs_release_path(path
);
2320 inode_add_bytes(inode
, num_bytes
);
2322 ins
.objectid
= disk_bytenr
;
2323 ins
.offset
= disk_num_bytes
;
2324 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2327 * Release the reserved range from inode dirty range map, as it is
2328 * already moved into delayed_ref_head
2330 ret
= btrfs_qgroup_release_data(inode
, file_pos
, ram_bytes
);
2334 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
2335 btrfs_ino(BTRFS_I(inode
)),
2336 file_pos
, qg_released
, &ins
);
2338 btrfs_free_path(path
);
2343 /* snapshot-aware defrag */
2344 struct sa_defrag_extent_backref
{
2345 struct rb_node node
;
2346 struct old_sa_defrag_extent
*old
;
2355 struct old_sa_defrag_extent
{
2356 struct list_head list
;
2357 struct new_sa_defrag_extent
*new;
2366 struct new_sa_defrag_extent
{
2367 struct rb_root root
;
2368 struct list_head head
;
2369 struct btrfs_path
*path
;
2370 struct inode
*inode
;
2378 static int backref_comp(struct sa_defrag_extent_backref
*b1
,
2379 struct sa_defrag_extent_backref
*b2
)
2381 if (b1
->root_id
< b2
->root_id
)
2383 else if (b1
->root_id
> b2
->root_id
)
2386 if (b1
->inum
< b2
->inum
)
2388 else if (b1
->inum
> b2
->inum
)
2391 if (b1
->file_pos
< b2
->file_pos
)
2393 else if (b1
->file_pos
> b2
->file_pos
)
2397 * [------------------------------] ===> (a range of space)
2398 * |<--->| |<---->| =============> (fs/file tree A)
2399 * |<---------------------------->| ===> (fs/file tree B)
2401 * A range of space can refer to two file extents in one tree while
2402 * refer to only one file extent in another tree.
2404 * So we may process a disk offset more than one time(two extents in A)
2405 * and locate at the same extent(one extent in B), then insert two same
2406 * backrefs(both refer to the extent in B).
2411 static void backref_insert(struct rb_root
*root
,
2412 struct sa_defrag_extent_backref
*backref
)
2414 struct rb_node
**p
= &root
->rb_node
;
2415 struct rb_node
*parent
= NULL
;
2416 struct sa_defrag_extent_backref
*entry
;
2421 entry
= rb_entry(parent
, struct sa_defrag_extent_backref
, node
);
2423 ret
= backref_comp(backref
, entry
);
2427 p
= &(*p
)->rb_right
;
2430 rb_link_node(&backref
->node
, parent
, p
);
2431 rb_insert_color(&backref
->node
, root
);
2435 * Note the backref might has changed, and in this case we just return 0.
2437 static noinline
int record_one_backref(u64 inum
, u64 offset
, u64 root_id
,
2440 struct btrfs_file_extent_item
*extent
;
2441 struct old_sa_defrag_extent
*old
= ctx
;
2442 struct new_sa_defrag_extent
*new = old
->new;
2443 struct btrfs_path
*path
= new->path
;
2444 struct btrfs_key key
;
2445 struct btrfs_root
*root
;
2446 struct sa_defrag_extent_backref
*backref
;
2447 struct extent_buffer
*leaf
;
2448 struct inode
*inode
= new->inode
;
2449 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2455 if (BTRFS_I(inode
)->root
->root_key
.objectid
== root_id
&&
2456 inum
== btrfs_ino(BTRFS_I(inode
)))
2459 key
.objectid
= root_id
;
2460 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2461 key
.offset
= (u64
)-1;
2463 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2465 if (PTR_ERR(root
) == -ENOENT
)
2468 btrfs_debug(fs_info
, "inum=%llu, offset=%llu, root_id=%llu",
2469 inum
, offset
, root_id
);
2470 return PTR_ERR(root
);
2473 key
.objectid
= inum
;
2474 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2475 if (offset
> (u64
)-1 << 32)
2478 key
.offset
= offset
;
2480 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2481 if (WARN_ON(ret
< 0))
2488 leaf
= path
->nodes
[0];
2489 slot
= path
->slots
[0];
2491 if (slot
>= btrfs_header_nritems(leaf
)) {
2492 ret
= btrfs_next_leaf(root
, path
);
2495 } else if (ret
> 0) {
2504 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2506 if (key
.objectid
> inum
)
2509 if (key
.objectid
< inum
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2512 extent
= btrfs_item_ptr(leaf
, slot
,
2513 struct btrfs_file_extent_item
);
2515 if (btrfs_file_extent_disk_bytenr(leaf
, extent
) != old
->bytenr
)
2519 * 'offset' refers to the exact key.offset,
2520 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2521 * (key.offset - extent_offset).
2523 if (key
.offset
!= offset
)
2526 extent_offset
= btrfs_file_extent_offset(leaf
, extent
);
2527 num_bytes
= btrfs_file_extent_num_bytes(leaf
, extent
);
2529 if (extent_offset
>= old
->extent_offset
+ old
->offset
+
2530 old
->len
|| extent_offset
+ num_bytes
<=
2531 old
->extent_offset
+ old
->offset
)
2536 backref
= kmalloc(sizeof(*backref
), GFP_NOFS
);
2542 backref
->root_id
= root_id
;
2543 backref
->inum
= inum
;
2544 backref
->file_pos
= offset
;
2545 backref
->num_bytes
= num_bytes
;
2546 backref
->extent_offset
= extent_offset
;
2547 backref
->generation
= btrfs_file_extent_generation(leaf
, extent
);
2549 backref_insert(&new->root
, backref
);
2552 btrfs_release_path(path
);
2557 static noinline
bool record_extent_backrefs(struct btrfs_path
*path
,
2558 struct new_sa_defrag_extent
*new)
2560 struct btrfs_fs_info
*fs_info
= btrfs_sb(new->inode
->i_sb
);
2561 struct old_sa_defrag_extent
*old
, *tmp
;
2566 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2567 ret
= iterate_inodes_from_logical(old
->bytenr
+
2568 old
->extent_offset
, fs_info
,
2569 path
, record_one_backref
,
2571 if (ret
< 0 && ret
!= -ENOENT
)
2574 /* no backref to be processed for this extent */
2576 list_del(&old
->list
);
2581 if (list_empty(&new->head
))
2587 static int relink_is_mergable(struct extent_buffer
*leaf
,
2588 struct btrfs_file_extent_item
*fi
,
2589 struct new_sa_defrag_extent
*new)
2591 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) != new->bytenr
)
2594 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2597 if (btrfs_file_extent_compression(leaf
, fi
) != new->compress_type
)
2600 if (btrfs_file_extent_encryption(leaf
, fi
) ||
2601 btrfs_file_extent_other_encoding(leaf
, fi
))
2608 * Note the backref might has changed, and in this case we just return 0.
2610 static noinline
int relink_extent_backref(struct btrfs_path
*path
,
2611 struct sa_defrag_extent_backref
*prev
,
2612 struct sa_defrag_extent_backref
*backref
)
2614 struct btrfs_file_extent_item
*extent
;
2615 struct btrfs_file_extent_item
*item
;
2616 struct btrfs_ordered_extent
*ordered
;
2617 struct btrfs_trans_handle
*trans
;
2618 struct btrfs_ref ref
= { 0 };
2619 struct btrfs_root
*root
;
2620 struct btrfs_key key
;
2621 struct extent_buffer
*leaf
;
2622 struct old_sa_defrag_extent
*old
= backref
->old
;
2623 struct new_sa_defrag_extent
*new = old
->new;
2624 struct btrfs_fs_info
*fs_info
= btrfs_sb(new->inode
->i_sb
);
2625 struct inode
*inode
;
2626 struct extent_state
*cached
= NULL
;
2635 if (prev
&& prev
->root_id
== backref
->root_id
&&
2636 prev
->inum
== backref
->inum
&&
2637 prev
->file_pos
+ prev
->num_bytes
== backref
->file_pos
)
2640 /* step 1: get root */
2641 key
.objectid
= backref
->root_id
;
2642 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2643 key
.offset
= (u64
)-1;
2645 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
2647 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2649 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2650 if (PTR_ERR(root
) == -ENOENT
)
2652 return PTR_ERR(root
);
2655 if (btrfs_root_readonly(root
)) {
2656 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2660 /* step 2: get inode */
2661 key
.objectid
= backref
->inum
;
2662 key
.type
= BTRFS_INODE_ITEM_KEY
;
2665 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
2666 if (IS_ERR(inode
)) {
2667 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2671 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2673 /* step 3: relink backref */
2674 lock_start
= backref
->file_pos
;
2675 lock_end
= backref
->file_pos
+ backref
->num_bytes
- 1;
2676 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2679 ordered
= btrfs_lookup_first_ordered_extent(inode
, lock_end
);
2681 btrfs_put_ordered_extent(ordered
);
2685 trans
= btrfs_join_transaction(root
);
2686 if (IS_ERR(trans
)) {
2687 ret
= PTR_ERR(trans
);
2691 key
.objectid
= backref
->inum
;
2692 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2693 key
.offset
= backref
->file_pos
;
2695 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2698 } else if (ret
> 0) {
2703 extent
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2704 struct btrfs_file_extent_item
);
2706 if (btrfs_file_extent_generation(path
->nodes
[0], extent
) !=
2707 backref
->generation
)
2710 btrfs_release_path(path
);
2712 start
= backref
->file_pos
;
2713 if (backref
->extent_offset
< old
->extent_offset
+ old
->offset
)
2714 start
+= old
->extent_offset
+ old
->offset
-
2715 backref
->extent_offset
;
2717 len
= min(backref
->extent_offset
+ backref
->num_bytes
,
2718 old
->extent_offset
+ old
->offset
+ old
->len
);
2719 len
-= max(backref
->extent_offset
, old
->extent_offset
+ old
->offset
);
2721 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
2726 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
2727 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2730 path
->leave_spinning
= 1;
2732 struct btrfs_file_extent_item
*fi
;
2734 struct btrfs_key found_key
;
2736 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2741 leaf
= path
->nodes
[0];
2742 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2744 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2745 struct btrfs_file_extent_item
);
2746 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
2748 if (extent_len
+ found_key
.offset
== start
&&
2749 relink_is_mergable(leaf
, fi
, new)) {
2750 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2752 btrfs_mark_buffer_dirty(leaf
);
2753 inode_add_bytes(inode
, len
);
2759 btrfs_release_path(path
);
2764 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2767 btrfs_abort_transaction(trans
, ret
);
2771 leaf
= path
->nodes
[0];
2772 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2773 struct btrfs_file_extent_item
);
2774 btrfs_set_file_extent_disk_bytenr(leaf
, item
, new->bytenr
);
2775 btrfs_set_file_extent_disk_num_bytes(leaf
, item
, new->disk_len
);
2776 btrfs_set_file_extent_offset(leaf
, item
, start
- new->file_pos
);
2777 btrfs_set_file_extent_num_bytes(leaf
, item
, len
);
2778 btrfs_set_file_extent_ram_bytes(leaf
, item
, new->len
);
2779 btrfs_set_file_extent_generation(leaf
, item
, trans
->transid
);
2780 btrfs_set_file_extent_type(leaf
, item
, BTRFS_FILE_EXTENT_REG
);
2781 btrfs_set_file_extent_compression(leaf
, item
, new->compress_type
);
2782 btrfs_set_file_extent_encryption(leaf
, item
, 0);
2783 btrfs_set_file_extent_other_encoding(leaf
, item
, 0);
2785 btrfs_mark_buffer_dirty(leaf
);
2786 inode_add_bytes(inode
, len
);
2787 btrfs_release_path(path
);
2789 btrfs_init_generic_ref(&ref
, BTRFS_ADD_DELAYED_REF
, new->bytenr
,
2791 btrfs_init_data_ref(&ref
, backref
->root_id
, backref
->inum
,
2792 new->file_pos
); /* start - extent_offset */
2793 ret
= btrfs_inc_extent_ref(trans
, &ref
);
2795 btrfs_abort_transaction(trans
, ret
);
2801 btrfs_release_path(path
);
2802 path
->leave_spinning
= 0;
2803 btrfs_end_transaction(trans
);
2805 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2811 static void free_sa_defrag_extent(struct new_sa_defrag_extent
*new)
2813 struct old_sa_defrag_extent
*old
, *tmp
;
2818 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2824 static void relink_file_extents(struct new_sa_defrag_extent
*new)
2826 struct btrfs_fs_info
*fs_info
= btrfs_sb(new->inode
->i_sb
);
2827 struct btrfs_path
*path
;
2828 struct sa_defrag_extent_backref
*backref
;
2829 struct sa_defrag_extent_backref
*prev
= NULL
;
2830 struct rb_node
*node
;
2833 path
= btrfs_alloc_path();
2837 if (!record_extent_backrefs(path
, new)) {
2838 btrfs_free_path(path
);
2841 btrfs_release_path(path
);
2844 node
= rb_first(&new->root
);
2847 rb_erase(node
, &new->root
);
2849 backref
= rb_entry(node
, struct sa_defrag_extent_backref
, node
);
2851 ret
= relink_extent_backref(path
, prev
, backref
);
2864 btrfs_free_path(path
);
2866 free_sa_defrag_extent(new);
2868 atomic_dec(&fs_info
->defrag_running
);
2869 wake_up(&fs_info
->transaction_wait
);
2872 static struct new_sa_defrag_extent
*
2873 record_old_file_extents(struct inode
*inode
,
2874 struct btrfs_ordered_extent
*ordered
)
2876 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2877 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2878 struct btrfs_path
*path
;
2879 struct btrfs_key key
;
2880 struct old_sa_defrag_extent
*old
;
2881 struct new_sa_defrag_extent
*new;
2884 new = kmalloc(sizeof(*new), GFP_NOFS
);
2889 new->file_pos
= ordered
->file_offset
;
2890 new->len
= ordered
->len
;
2891 new->bytenr
= ordered
->start
;
2892 new->disk_len
= ordered
->disk_len
;
2893 new->compress_type
= ordered
->compress_type
;
2894 new->root
= RB_ROOT
;
2895 INIT_LIST_HEAD(&new->head
);
2897 path
= btrfs_alloc_path();
2901 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
2902 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2903 key
.offset
= new->file_pos
;
2905 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2908 if (ret
> 0 && path
->slots
[0] > 0)
2911 /* find out all the old extents for the file range */
2913 struct btrfs_file_extent_item
*extent
;
2914 struct extent_buffer
*l
;
2923 slot
= path
->slots
[0];
2925 if (slot
>= btrfs_header_nritems(l
)) {
2926 ret
= btrfs_next_leaf(root
, path
);
2934 btrfs_item_key_to_cpu(l
, &key
, slot
);
2936 if (key
.objectid
!= btrfs_ino(BTRFS_I(inode
)))
2938 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2940 if (key
.offset
>= new->file_pos
+ new->len
)
2943 extent
= btrfs_item_ptr(l
, slot
, struct btrfs_file_extent_item
);
2945 num_bytes
= btrfs_file_extent_num_bytes(l
, extent
);
2946 if (key
.offset
+ num_bytes
< new->file_pos
)
2949 disk_bytenr
= btrfs_file_extent_disk_bytenr(l
, extent
);
2953 extent_offset
= btrfs_file_extent_offset(l
, extent
);
2955 old
= kmalloc(sizeof(*old
), GFP_NOFS
);
2959 offset
= max(new->file_pos
, key
.offset
);
2960 end
= min(new->file_pos
+ new->len
, key
.offset
+ num_bytes
);
2962 old
->bytenr
= disk_bytenr
;
2963 old
->extent_offset
= extent_offset
;
2964 old
->offset
= offset
- key
.offset
;
2965 old
->len
= end
- offset
;
2968 list_add_tail(&old
->list
, &new->head
);
2974 btrfs_free_path(path
);
2975 atomic_inc(&fs_info
->defrag_running
);
2980 btrfs_free_path(path
);
2982 free_sa_defrag_extent(new);
2986 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info
*fs_info
,
2989 struct btrfs_block_group_cache
*cache
;
2991 cache
= btrfs_lookup_block_group(fs_info
, start
);
2994 spin_lock(&cache
->lock
);
2995 cache
->delalloc_bytes
-= len
;
2996 spin_unlock(&cache
->lock
);
2998 btrfs_put_block_group(cache
);
3001 /* as ordered data IO finishes, this gets called so we can finish
3002 * an ordered extent if the range of bytes in the file it covers are
3005 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
3007 struct inode
*inode
= ordered_extent
->inode
;
3008 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3009 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3010 struct btrfs_trans_handle
*trans
= NULL
;
3011 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3012 struct extent_state
*cached_state
= NULL
;
3013 struct new_sa_defrag_extent
*new = NULL
;
3014 int compress_type
= 0;
3016 u64 logical_len
= ordered_extent
->len
;
3018 bool truncated
= false;
3019 bool range_locked
= false;
3020 bool clear_new_delalloc_bytes
= false;
3021 bool clear_reserved_extent
= true;
3023 if (!test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
3024 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
) &&
3025 !test_bit(BTRFS_ORDERED_DIRECT
, &ordered_extent
->flags
))
3026 clear_new_delalloc_bytes
= true;
3028 nolock
= btrfs_is_free_space_inode(BTRFS_I(inode
));
3030 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
3035 btrfs_free_io_failure_record(BTRFS_I(inode
),
3036 ordered_extent
->file_offset
,
3037 ordered_extent
->file_offset
+
3038 ordered_extent
->len
- 1);
3040 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered_extent
->flags
)) {
3042 logical_len
= ordered_extent
->truncated_len
;
3043 /* Truncated the entire extent, don't bother adding */
3048 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
3049 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
3052 * For mwrite(mmap + memset to write) case, we still reserve
3053 * space for NOCOW range.
3054 * As NOCOW won't cause a new delayed ref, just free the space
3056 btrfs_qgroup_free_data(inode
, NULL
, ordered_extent
->file_offset
,
3057 ordered_extent
->len
);
3058 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
3060 trans
= btrfs_join_transaction_nolock(root
);
3062 trans
= btrfs_join_transaction(root
);
3063 if (IS_ERR(trans
)) {
3064 ret
= PTR_ERR(trans
);
3068 trans
->block_rsv
= &BTRFS_I(inode
)->block_rsv
;
3069 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
3070 if (ret
) /* -ENOMEM or corruption */
3071 btrfs_abort_transaction(trans
, ret
);
3075 range_locked
= true;
3076 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
3077 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
3080 ret
= test_range_bit(io_tree
, ordered_extent
->file_offset
,
3081 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
3082 EXTENT_DEFRAG
, 0, cached_state
);
3084 u64 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
3085 if (0 && last_snapshot
>= BTRFS_I(inode
)->generation
)
3086 /* the inode is shared */
3087 new = record_old_file_extents(inode
, ordered_extent
);
3089 clear_extent_bit(io_tree
, ordered_extent
->file_offset
,
3090 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
3091 EXTENT_DEFRAG
, 0, 0, &cached_state
);
3095 trans
= btrfs_join_transaction_nolock(root
);
3097 trans
= btrfs_join_transaction(root
);
3098 if (IS_ERR(trans
)) {
3099 ret
= PTR_ERR(trans
);
3104 trans
->block_rsv
= &BTRFS_I(inode
)->block_rsv
;
3106 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
3107 compress_type
= ordered_extent
->compress_type
;
3108 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
3109 BUG_ON(compress_type
);
3110 btrfs_qgroup_free_data(inode
, NULL
, ordered_extent
->file_offset
,
3111 ordered_extent
->len
);
3112 ret
= btrfs_mark_extent_written(trans
, BTRFS_I(inode
),
3113 ordered_extent
->file_offset
,
3114 ordered_extent
->file_offset
+
3117 BUG_ON(root
== fs_info
->tree_root
);
3118 ret
= insert_reserved_file_extent(trans
, inode
,
3119 ordered_extent
->file_offset
,
3120 ordered_extent
->start
,
3121 ordered_extent
->disk_len
,
3122 logical_len
, logical_len
,
3123 compress_type
, 0, 0,
3124 BTRFS_FILE_EXTENT_REG
);
3126 clear_reserved_extent
= false;
3127 btrfs_release_delalloc_bytes(fs_info
,
3128 ordered_extent
->start
,
3129 ordered_extent
->disk_len
);
3132 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
3133 ordered_extent
->file_offset
, ordered_extent
->len
,
3136 btrfs_abort_transaction(trans
, ret
);
3140 ret
= add_pending_csums(trans
, inode
, &ordered_extent
->list
);
3142 btrfs_abort_transaction(trans
, ret
);
3146 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
3147 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
3148 if (ret
) { /* -ENOMEM or corruption */
3149 btrfs_abort_transaction(trans
, ret
);
3154 if (range_locked
|| clear_new_delalloc_bytes
) {
3155 unsigned int clear_bits
= 0;
3158 clear_bits
|= EXTENT_LOCKED
;
3159 if (clear_new_delalloc_bytes
)
3160 clear_bits
|= EXTENT_DELALLOC_NEW
;
3161 clear_extent_bit(&BTRFS_I(inode
)->io_tree
,
3162 ordered_extent
->file_offset
,
3163 ordered_extent
->file_offset
+
3164 ordered_extent
->len
- 1,
3166 (clear_bits
& EXTENT_LOCKED
) ? 1 : 0,
3171 btrfs_end_transaction(trans
);
3173 if (ret
|| truncated
) {
3177 start
= ordered_extent
->file_offset
+ logical_len
;
3179 start
= ordered_extent
->file_offset
;
3180 end
= ordered_extent
->file_offset
+ ordered_extent
->len
- 1;
3181 clear_extent_uptodate(io_tree
, start
, end
, NULL
);
3183 /* Drop the cache for the part of the extent we didn't write. */
3184 btrfs_drop_extent_cache(BTRFS_I(inode
), start
, end
, 0);
3187 * If the ordered extent had an IOERR or something else went
3188 * wrong we need to return the space for this ordered extent
3189 * back to the allocator. We only free the extent in the
3190 * truncated case if we didn't write out the extent at all.
3192 * If we made it past insert_reserved_file_extent before we
3193 * errored out then we don't need to do this as the accounting
3194 * has already been done.
3196 if ((ret
|| !logical_len
) &&
3197 clear_reserved_extent
&&
3198 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
3199 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
))
3200 btrfs_free_reserved_extent(fs_info
,
3201 ordered_extent
->start
,
3202 ordered_extent
->disk_len
, 1);
3207 * This needs to be done to make sure anybody waiting knows we are done
3208 * updating everything for this ordered extent.
3210 btrfs_remove_ordered_extent(inode
, ordered_extent
);
3212 /* for snapshot-aware defrag */
3215 free_sa_defrag_extent(new);
3216 atomic_dec(&fs_info
->defrag_running
);
3218 relink_file_extents(new);
3223 btrfs_put_ordered_extent(ordered_extent
);
3224 /* once for the tree */
3225 btrfs_put_ordered_extent(ordered_extent
);
3230 static void finish_ordered_fn(struct btrfs_work
*work
)
3232 struct btrfs_ordered_extent
*ordered_extent
;
3233 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
3234 btrfs_finish_ordered_io(ordered_extent
);
3237 void btrfs_writepage_endio_finish_ordered(struct page
*page
, u64 start
,
3238 u64 end
, int uptodate
)
3240 struct inode
*inode
= page
->mapping
->host
;
3241 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3242 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
3243 struct btrfs_workqueue
*wq
;
3244 btrfs_work_func_t func
;
3246 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
3248 ClearPagePrivate2(page
);
3249 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
3250 end
- start
+ 1, uptodate
))
3253 if (btrfs_is_free_space_inode(BTRFS_I(inode
))) {
3254 wq
= fs_info
->endio_freespace_worker
;
3255 func
= btrfs_freespace_write_helper
;
3257 wq
= fs_info
->endio_write_workers
;
3258 func
= btrfs_endio_write_helper
;
3261 btrfs_init_work(&ordered_extent
->work
, func
, finish_ordered_fn
, NULL
,
3263 btrfs_queue_work(wq
, &ordered_extent
->work
);
3266 static int __readpage_endio_check(struct inode
*inode
,
3267 struct btrfs_io_bio
*io_bio
,
3268 int icsum
, struct page
*page
,
3269 int pgoff
, u64 start
, size_t len
)
3271 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3272 SHASH_DESC_ON_STACK(shash
, fs_info
->csum_shash
);
3274 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
3276 u8 csum
[BTRFS_CSUM_SIZE
];
3278 csum_expected
= ((u8
*)io_bio
->csum
) + icsum
* csum_size
;
3280 kaddr
= kmap_atomic(page
);
3281 shash
->tfm
= fs_info
->csum_shash
;
3283 crypto_shash_init(shash
);
3284 crypto_shash_update(shash
, kaddr
+ pgoff
, len
);
3285 crypto_shash_final(shash
, csum
);
3287 if (memcmp(csum
, csum_expected
, csum_size
))
3290 kunmap_atomic(kaddr
);
3293 btrfs_print_data_csum_error(BTRFS_I(inode
), start
, csum
, csum_expected
,
3294 io_bio
->mirror_num
);
3295 memset(kaddr
+ pgoff
, 1, len
);
3296 flush_dcache_page(page
);
3297 kunmap_atomic(kaddr
);
3302 * when reads are done, we need to check csums to verify the data is correct
3303 * if there's a match, we allow the bio to finish. If not, the code in
3304 * extent_io.c will try to find good copies for us.
3306 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
3307 u64 phy_offset
, struct page
*page
,
3308 u64 start
, u64 end
, int mirror
)
3310 size_t offset
= start
- page_offset(page
);
3311 struct inode
*inode
= page
->mapping
->host
;
3312 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3313 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3315 if (PageChecked(page
)) {
3316 ClearPageChecked(page
);
3320 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
3323 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
3324 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
3325 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
);
3329 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
3330 return __readpage_endio_check(inode
, io_bio
, phy_offset
, page
, offset
,
3331 start
, (size_t)(end
- start
+ 1));
3335 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3337 * @inode: The inode we want to perform iput on
3339 * This function uses the generic vfs_inode::i_count to track whether we should
3340 * just decrement it (in case it's > 1) or if this is the last iput then link
3341 * the inode to the delayed iput machinery. Delayed iputs are processed at
3342 * transaction commit time/superblock commit/cleaner kthread.
3344 void btrfs_add_delayed_iput(struct inode
*inode
)
3346 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3347 struct btrfs_inode
*binode
= BTRFS_I(inode
);
3349 if (atomic_add_unless(&inode
->i_count
, -1, 1))
3352 atomic_inc(&fs_info
->nr_delayed_iputs
);
3353 spin_lock(&fs_info
->delayed_iput_lock
);
3354 ASSERT(list_empty(&binode
->delayed_iput
));
3355 list_add_tail(&binode
->delayed_iput
, &fs_info
->delayed_iputs
);
3356 spin_unlock(&fs_info
->delayed_iput_lock
);
3357 if (!test_bit(BTRFS_FS_CLEANER_RUNNING
, &fs_info
->flags
))
3358 wake_up_process(fs_info
->cleaner_kthread
);
3361 static void run_delayed_iput_locked(struct btrfs_fs_info
*fs_info
,
3362 struct btrfs_inode
*inode
)
3364 list_del_init(&inode
->delayed_iput
);
3365 spin_unlock(&fs_info
->delayed_iput_lock
);
3366 iput(&inode
->vfs_inode
);
3367 if (atomic_dec_and_test(&fs_info
->nr_delayed_iputs
))
3368 wake_up(&fs_info
->delayed_iputs_wait
);
3369 spin_lock(&fs_info
->delayed_iput_lock
);
3372 static void btrfs_run_delayed_iput(struct btrfs_fs_info
*fs_info
,
3373 struct btrfs_inode
*inode
)
3375 if (!list_empty(&inode
->delayed_iput
)) {
3376 spin_lock(&fs_info
->delayed_iput_lock
);
3377 if (!list_empty(&inode
->delayed_iput
))
3378 run_delayed_iput_locked(fs_info
, inode
);
3379 spin_unlock(&fs_info
->delayed_iput_lock
);
3383 void btrfs_run_delayed_iputs(struct btrfs_fs_info
*fs_info
)
3386 spin_lock(&fs_info
->delayed_iput_lock
);
3387 while (!list_empty(&fs_info
->delayed_iputs
)) {
3388 struct btrfs_inode
*inode
;
3390 inode
= list_first_entry(&fs_info
->delayed_iputs
,
3391 struct btrfs_inode
, delayed_iput
);
3392 run_delayed_iput_locked(fs_info
, inode
);
3394 spin_unlock(&fs_info
->delayed_iput_lock
);
3398 * btrfs_wait_on_delayed_iputs - wait on the delayed iputs to be done running
3399 * @fs_info - the fs_info for this fs
3400 * @return - EINTR if we were killed, 0 if nothing's pending
3402 * This will wait on any delayed iputs that are currently running with KILLABLE
3403 * set. Once they are all done running we will return, unless we are killed in
3404 * which case we return EINTR. This helps in user operations like fallocate etc
3405 * that might get blocked on the iputs.
3407 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info
*fs_info
)
3409 int ret
= wait_event_killable(fs_info
->delayed_iputs_wait
,
3410 atomic_read(&fs_info
->nr_delayed_iputs
) == 0);
3417 * This creates an orphan entry for the given inode in case something goes wrong
3418 * in the middle of an unlink.
3420 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
,
3421 struct btrfs_inode
*inode
)
3425 ret
= btrfs_insert_orphan_item(trans
, inode
->root
, btrfs_ino(inode
));
3426 if (ret
&& ret
!= -EEXIST
) {
3427 btrfs_abort_transaction(trans
, ret
);
3435 * We have done the delete so we can go ahead and remove the orphan item for
3436 * this particular inode.
3438 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3439 struct btrfs_inode
*inode
)
3441 return btrfs_del_orphan_item(trans
, inode
->root
, btrfs_ino(inode
));
3445 * this cleans up any orphans that may be left on the list from the last use
3448 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3450 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3451 struct btrfs_path
*path
;
3452 struct extent_buffer
*leaf
;
3453 struct btrfs_key key
, found_key
;
3454 struct btrfs_trans_handle
*trans
;
3455 struct inode
*inode
;
3456 u64 last_objectid
= 0;
3457 int ret
= 0, nr_unlink
= 0;
3459 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
3462 path
= btrfs_alloc_path();
3467 path
->reada
= READA_BACK
;
3469 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3470 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
3471 key
.offset
= (u64
)-1;
3474 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3479 * if ret == 0 means we found what we were searching for, which
3480 * is weird, but possible, so only screw with path if we didn't
3481 * find the key and see if we have stuff that matches
3485 if (path
->slots
[0] == 0)
3490 /* pull out the item */
3491 leaf
= path
->nodes
[0];
3492 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3494 /* make sure the item matches what we want */
3495 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3497 if (found_key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
3500 /* release the path since we're done with it */
3501 btrfs_release_path(path
);
3504 * this is where we are basically btrfs_lookup, without the
3505 * crossing root thing. we store the inode number in the
3506 * offset of the orphan item.
3509 if (found_key
.offset
== last_objectid
) {
3511 "Error removing orphan entry, stopping orphan cleanup");
3516 last_objectid
= found_key
.offset
;
3518 found_key
.objectid
= found_key
.offset
;
3519 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3520 found_key
.offset
= 0;
3521 inode
= btrfs_iget(fs_info
->sb
, &found_key
, root
, NULL
);
3522 ret
= PTR_ERR_OR_ZERO(inode
);
3523 if (ret
&& ret
!= -ENOENT
)
3526 if (ret
== -ENOENT
&& root
== fs_info
->tree_root
) {
3527 struct btrfs_root
*dead_root
;
3528 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3529 int is_dead_root
= 0;
3532 * this is an orphan in the tree root. Currently these
3533 * could come from 2 sources:
3534 * a) a snapshot deletion in progress
3535 * b) a free space cache inode
3536 * We need to distinguish those two, as the snapshot
3537 * orphan must not get deleted.
3538 * find_dead_roots already ran before us, so if this
3539 * is a snapshot deletion, we should find the root
3540 * in the dead_roots list
3542 spin_lock(&fs_info
->trans_lock
);
3543 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
3545 if (dead_root
->root_key
.objectid
==
3546 found_key
.objectid
) {
3551 spin_unlock(&fs_info
->trans_lock
);
3553 /* prevent this orphan from being found again */
3554 key
.offset
= found_key
.objectid
- 1;
3561 * If we have an inode with links, there are a couple of
3562 * possibilities. Old kernels (before v3.12) used to create an
3563 * orphan item for truncate indicating that there were possibly
3564 * extent items past i_size that needed to be deleted. In v3.12,
3565 * truncate was changed to update i_size in sync with the extent
3566 * items, but the (useless) orphan item was still created. Since
3567 * v4.18, we don't create the orphan item for truncate at all.
3569 * So, this item could mean that we need to do a truncate, but
3570 * only if this filesystem was last used on a pre-v3.12 kernel
3571 * and was not cleanly unmounted. The odds of that are quite
3572 * slim, and it's a pain to do the truncate now, so just delete
3575 * It's also possible that this orphan item was supposed to be
3576 * deleted but wasn't. The inode number may have been reused,
3577 * but either way, we can delete the orphan item.
3579 if (ret
== -ENOENT
|| inode
->i_nlink
) {
3582 trans
= btrfs_start_transaction(root
, 1);
3583 if (IS_ERR(trans
)) {
3584 ret
= PTR_ERR(trans
);
3587 btrfs_debug(fs_info
, "auto deleting %Lu",
3588 found_key
.objectid
);
3589 ret
= btrfs_del_orphan_item(trans
, root
,
3590 found_key
.objectid
);
3591 btrfs_end_transaction(trans
);
3599 /* this will do delete_inode and everything for us */
3602 /* release the path since we're done with it */
3603 btrfs_release_path(path
);
3605 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
3607 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
)) {
3608 trans
= btrfs_join_transaction(root
);
3610 btrfs_end_transaction(trans
);
3614 btrfs_debug(fs_info
, "unlinked %d orphans", nr_unlink
);
3618 btrfs_err(fs_info
, "could not do orphan cleanup %d", ret
);
3619 btrfs_free_path(path
);
3624 * very simple check to peek ahead in the leaf looking for xattrs. If we
3625 * don't find any xattrs, we know there can't be any acls.
3627 * slot is the slot the inode is in, objectid is the objectid of the inode
3629 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3630 int slot
, u64 objectid
,
3631 int *first_xattr_slot
)
3633 u32 nritems
= btrfs_header_nritems(leaf
);
3634 struct btrfs_key found_key
;
3635 static u64 xattr_access
= 0;
3636 static u64 xattr_default
= 0;
3639 if (!xattr_access
) {
3640 xattr_access
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS
,
3641 strlen(XATTR_NAME_POSIX_ACL_ACCESS
));
3642 xattr_default
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT
,
3643 strlen(XATTR_NAME_POSIX_ACL_DEFAULT
));
3647 *first_xattr_slot
= -1;
3648 while (slot
< nritems
) {
3649 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3651 /* we found a different objectid, there must not be acls */
3652 if (found_key
.objectid
!= objectid
)
3655 /* we found an xattr, assume we've got an acl */
3656 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
3657 if (*first_xattr_slot
== -1)
3658 *first_xattr_slot
= slot
;
3659 if (found_key
.offset
== xattr_access
||
3660 found_key
.offset
== xattr_default
)
3665 * we found a key greater than an xattr key, there can't
3666 * be any acls later on
3668 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3675 * it goes inode, inode backrefs, xattrs, extents,
3676 * so if there are a ton of hard links to an inode there can
3677 * be a lot of backrefs. Don't waste time searching too hard,
3678 * this is just an optimization
3683 /* we hit the end of the leaf before we found an xattr or
3684 * something larger than an xattr. We have to assume the inode
3687 if (*first_xattr_slot
== -1)
3688 *first_xattr_slot
= slot
;
3693 * read an inode from the btree into the in-memory inode
3695 static int btrfs_read_locked_inode(struct inode
*inode
,
3696 struct btrfs_path
*in_path
)
3698 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3699 struct btrfs_path
*path
= in_path
;
3700 struct extent_buffer
*leaf
;
3701 struct btrfs_inode_item
*inode_item
;
3702 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3703 struct btrfs_key location
;
3708 bool filled
= false;
3709 int first_xattr_slot
;
3711 ret
= btrfs_fill_inode(inode
, &rdev
);
3716 path
= btrfs_alloc_path();
3721 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3723 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3725 if (path
!= in_path
)
3726 btrfs_free_path(path
);
3730 leaf
= path
->nodes
[0];
3735 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3736 struct btrfs_inode_item
);
3737 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3738 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3739 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3740 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3741 btrfs_i_size_write(BTRFS_I(inode
), btrfs_inode_size(leaf
, inode_item
));
3743 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->atime
);
3744 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->atime
);
3746 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->mtime
);
3747 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->mtime
);
3749 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->ctime
);
3750 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->ctime
);
3752 BTRFS_I(inode
)->i_otime
.tv_sec
=
3753 btrfs_timespec_sec(leaf
, &inode_item
->otime
);
3754 BTRFS_I(inode
)->i_otime
.tv_nsec
=
3755 btrfs_timespec_nsec(leaf
, &inode_item
->otime
);
3757 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3758 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3759 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3761 inode_set_iversion_queried(inode
,
3762 btrfs_inode_sequence(leaf
, inode_item
));
3763 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3765 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3767 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3768 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
3772 * If we were modified in the current generation and evicted from memory
3773 * and then re-read we need to do a full sync since we don't have any
3774 * idea about which extents were modified before we were evicted from
3777 * This is required for both inode re-read from disk and delayed inode
3778 * in delayed_nodes_tree.
3780 if (BTRFS_I(inode
)->last_trans
== fs_info
->generation
)
3781 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3782 &BTRFS_I(inode
)->runtime_flags
);
3785 * We don't persist the id of the transaction where an unlink operation
3786 * against the inode was last made. So here we assume the inode might
3787 * have been evicted, and therefore the exact value of last_unlink_trans
3788 * lost, and set it to last_trans to avoid metadata inconsistencies
3789 * between the inode and its parent if the inode is fsync'ed and the log
3790 * replayed. For example, in the scenario:
3793 * ln mydir/foo mydir/bar
3796 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3797 * xfs_io -c fsync mydir/foo
3799 * mount fs, triggers fsync log replay
3801 * We must make sure that when we fsync our inode foo we also log its
3802 * parent inode, otherwise after log replay the parent still has the
3803 * dentry with the "bar" name but our inode foo has a link count of 1
3804 * and doesn't have an inode ref with the name "bar" anymore.
3806 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3807 * but it guarantees correctness at the expense of occasional full
3808 * transaction commits on fsync if our inode is a directory, or if our
3809 * inode is not a directory, logging its parent unnecessarily.
3811 BTRFS_I(inode
)->last_unlink_trans
= BTRFS_I(inode
)->last_trans
;
3814 if (inode
->i_nlink
!= 1 ||
3815 path
->slots
[0] >= btrfs_header_nritems(leaf
))
3818 btrfs_item_key_to_cpu(leaf
, &location
, path
->slots
[0]);
3819 if (location
.objectid
!= btrfs_ino(BTRFS_I(inode
)))
3822 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3823 if (location
.type
== BTRFS_INODE_REF_KEY
) {
3824 struct btrfs_inode_ref
*ref
;
3826 ref
= (struct btrfs_inode_ref
*)ptr
;
3827 BTRFS_I(inode
)->dir_index
= btrfs_inode_ref_index(leaf
, ref
);
3828 } else if (location
.type
== BTRFS_INODE_EXTREF_KEY
) {
3829 struct btrfs_inode_extref
*extref
;
3831 extref
= (struct btrfs_inode_extref
*)ptr
;
3832 BTRFS_I(inode
)->dir_index
= btrfs_inode_extref_index(leaf
,
3837 * try to precache a NULL acl entry for files that don't have
3838 * any xattrs or acls
3840 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
3841 btrfs_ino(BTRFS_I(inode
)), &first_xattr_slot
);
3842 if (first_xattr_slot
!= -1) {
3843 path
->slots
[0] = first_xattr_slot
;
3844 ret
= btrfs_load_inode_props(inode
, path
);
3847 "error loading props for ino %llu (root %llu): %d",
3848 btrfs_ino(BTRFS_I(inode
)),
3849 root
->root_key
.objectid
, ret
);
3851 if (path
!= in_path
)
3852 btrfs_free_path(path
);
3855 cache_no_acl(inode
);
3857 switch (inode
->i_mode
& S_IFMT
) {
3859 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3860 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3861 inode
->i_fop
= &btrfs_file_operations
;
3862 inode
->i_op
= &btrfs_file_inode_operations
;
3865 inode
->i_fop
= &btrfs_dir_file_operations
;
3866 inode
->i_op
= &btrfs_dir_inode_operations
;
3869 inode
->i_op
= &btrfs_symlink_inode_operations
;
3870 inode_nohighmem(inode
);
3871 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3874 inode
->i_op
= &btrfs_special_inode_operations
;
3875 init_special_inode(inode
, inode
->i_mode
, rdev
);
3879 btrfs_sync_inode_flags_to_i_flags(inode
);
3884 * given a leaf and an inode, copy the inode fields into the leaf
3886 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3887 struct extent_buffer
*leaf
,
3888 struct btrfs_inode_item
*item
,
3889 struct inode
*inode
)
3891 struct btrfs_map_token token
;
3893 btrfs_init_map_token(&token
, leaf
);
3895 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3896 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3897 btrfs_set_token_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
,
3899 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3900 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3902 btrfs_set_token_timespec_sec(leaf
, &item
->atime
,
3903 inode
->i_atime
.tv_sec
, &token
);
3904 btrfs_set_token_timespec_nsec(leaf
, &item
->atime
,
3905 inode
->i_atime
.tv_nsec
, &token
);
3907 btrfs_set_token_timespec_sec(leaf
, &item
->mtime
,
3908 inode
->i_mtime
.tv_sec
, &token
);
3909 btrfs_set_token_timespec_nsec(leaf
, &item
->mtime
,
3910 inode
->i_mtime
.tv_nsec
, &token
);
3912 btrfs_set_token_timespec_sec(leaf
, &item
->ctime
,
3913 inode
->i_ctime
.tv_sec
, &token
);
3914 btrfs_set_token_timespec_nsec(leaf
, &item
->ctime
,
3915 inode
->i_ctime
.tv_nsec
, &token
);
3917 btrfs_set_token_timespec_sec(leaf
, &item
->otime
,
3918 BTRFS_I(inode
)->i_otime
.tv_sec
, &token
);
3919 btrfs_set_token_timespec_nsec(leaf
, &item
->otime
,
3920 BTRFS_I(inode
)->i_otime
.tv_nsec
, &token
);
3922 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3924 btrfs_set_token_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
,
3926 btrfs_set_token_inode_sequence(leaf
, item
, inode_peek_iversion(inode
),
3928 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3929 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3930 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3931 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3935 * copy everything in the in-memory inode into the btree.
3937 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
3938 struct btrfs_root
*root
, struct inode
*inode
)
3940 struct btrfs_inode_item
*inode_item
;
3941 struct btrfs_path
*path
;
3942 struct extent_buffer
*leaf
;
3945 path
= btrfs_alloc_path();
3949 path
->leave_spinning
= 1;
3950 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
3958 leaf
= path
->nodes
[0];
3959 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3960 struct btrfs_inode_item
);
3962 fill_inode_item(trans
, leaf
, inode_item
, inode
);
3963 btrfs_mark_buffer_dirty(leaf
);
3964 btrfs_set_inode_last_trans(trans
, inode
);
3967 btrfs_free_path(path
);
3972 * copy everything in the in-memory inode into the btree.
3974 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
3975 struct btrfs_root
*root
, struct inode
*inode
)
3977 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3981 * If the inode is a free space inode, we can deadlock during commit
3982 * if we put it into the delayed code.
3984 * The data relocation inode should also be directly updated
3987 if (!btrfs_is_free_space_inode(BTRFS_I(inode
))
3988 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
3989 && !test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
)) {
3990 btrfs_update_root_times(trans
, root
);
3992 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
3994 btrfs_set_inode_last_trans(trans
, inode
);
3998 return btrfs_update_inode_item(trans
, root
, inode
);
4001 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
4002 struct btrfs_root
*root
,
4003 struct inode
*inode
)
4007 ret
= btrfs_update_inode(trans
, root
, inode
);
4009 return btrfs_update_inode_item(trans
, root
, inode
);
4014 * unlink helper that gets used here in inode.c and in the tree logging
4015 * recovery code. It remove a link in a directory with a given name, and
4016 * also drops the back refs in the inode to the directory
4018 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4019 struct btrfs_root
*root
,
4020 struct btrfs_inode
*dir
,
4021 struct btrfs_inode
*inode
,
4022 const char *name
, int name_len
)
4024 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4025 struct btrfs_path
*path
;
4027 struct btrfs_dir_item
*di
;
4029 u64 ino
= btrfs_ino(inode
);
4030 u64 dir_ino
= btrfs_ino(dir
);
4032 path
= btrfs_alloc_path();
4038 path
->leave_spinning
= 1;
4039 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4040 name
, name_len
, -1);
4041 if (IS_ERR_OR_NULL(di
)) {
4042 ret
= di
? PTR_ERR(di
) : -ENOENT
;
4045 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4048 btrfs_release_path(path
);
4051 * If we don't have dir index, we have to get it by looking up
4052 * the inode ref, since we get the inode ref, remove it directly,
4053 * it is unnecessary to do delayed deletion.
4055 * But if we have dir index, needn't search inode ref to get it.
4056 * Since the inode ref is close to the inode item, it is better
4057 * that we delay to delete it, and just do this deletion when
4058 * we update the inode item.
4060 if (inode
->dir_index
) {
4061 ret
= btrfs_delayed_delete_inode_ref(inode
);
4063 index
= inode
->dir_index
;
4068 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
4072 "failed to delete reference to %.*s, inode %llu parent %llu",
4073 name_len
, name
, ino
, dir_ino
);
4074 btrfs_abort_transaction(trans
, ret
);
4078 ret
= btrfs_delete_delayed_dir_index(trans
, dir
, index
);
4080 btrfs_abort_transaction(trans
, ret
);
4084 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
, inode
,
4086 if (ret
!= 0 && ret
!= -ENOENT
) {
4087 btrfs_abort_transaction(trans
, ret
);
4091 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
, dir
,
4096 btrfs_abort_transaction(trans
, ret
);
4099 * If we have a pending delayed iput we could end up with the final iput
4100 * being run in btrfs-cleaner context. If we have enough of these built
4101 * up we can end up burning a lot of time in btrfs-cleaner without any
4102 * way to throttle the unlinks. Since we're currently holding a ref on
4103 * the inode we can run the delayed iput here without any issues as the
4104 * final iput won't be done until after we drop the ref we're currently
4107 btrfs_run_delayed_iput(fs_info
, inode
);
4109 btrfs_free_path(path
);
4113 btrfs_i_size_write(dir
, dir
->vfs_inode
.i_size
- name_len
* 2);
4114 inode_inc_iversion(&inode
->vfs_inode
);
4115 inode_inc_iversion(&dir
->vfs_inode
);
4116 inode
->vfs_inode
.i_ctime
= dir
->vfs_inode
.i_mtime
=
4117 dir
->vfs_inode
.i_ctime
= current_time(&inode
->vfs_inode
);
4118 ret
= btrfs_update_inode(trans
, root
, &dir
->vfs_inode
);
4123 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4124 struct btrfs_root
*root
,
4125 struct btrfs_inode
*dir
, struct btrfs_inode
*inode
,
4126 const char *name
, int name_len
)
4129 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
4131 drop_nlink(&inode
->vfs_inode
);
4132 ret
= btrfs_update_inode(trans
, root
, &inode
->vfs_inode
);
4138 * helper to start transaction for unlink and rmdir.
4140 * unlink and rmdir are special in btrfs, they do not always free space, so
4141 * if we cannot make our reservations the normal way try and see if there is
4142 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4143 * allow the unlink to occur.
4145 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
)
4147 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4150 * 1 for the possible orphan item
4151 * 1 for the dir item
4152 * 1 for the dir index
4153 * 1 for the inode ref
4156 return btrfs_start_transaction_fallback_global_rsv(root
, 5, 5);
4159 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
4161 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4162 struct btrfs_trans_handle
*trans
;
4163 struct inode
*inode
= d_inode(dentry
);
4166 trans
= __unlink_start_trans(dir
);
4168 return PTR_ERR(trans
);
4170 btrfs_record_unlink_dir(trans
, BTRFS_I(dir
), BTRFS_I(d_inode(dentry
)),
4173 ret
= btrfs_unlink_inode(trans
, root
, BTRFS_I(dir
),
4174 BTRFS_I(d_inode(dentry
)), dentry
->d_name
.name
,
4175 dentry
->d_name
.len
);
4179 if (inode
->i_nlink
== 0) {
4180 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
4186 btrfs_end_transaction(trans
);
4187 btrfs_btree_balance_dirty(root
->fs_info
);
4191 static int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
4192 struct inode
*dir
, u64 objectid
,
4193 const char *name
, int name_len
)
4195 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4196 struct btrfs_path
*path
;
4197 struct extent_buffer
*leaf
;
4198 struct btrfs_dir_item
*di
;
4199 struct btrfs_key key
;
4202 u64 dir_ino
= btrfs_ino(BTRFS_I(dir
));
4204 path
= btrfs_alloc_path();
4208 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4209 name
, name_len
, -1);
4210 if (IS_ERR_OR_NULL(di
)) {
4211 ret
= di
? PTR_ERR(di
) : -ENOENT
;
4215 leaf
= path
->nodes
[0];
4216 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4217 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
4218 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4220 btrfs_abort_transaction(trans
, ret
);
4223 btrfs_release_path(path
);
4225 ret
= btrfs_del_root_ref(trans
, objectid
, root
->root_key
.objectid
,
4226 dir_ino
, &index
, name
, name_len
);
4228 if (ret
!= -ENOENT
) {
4229 btrfs_abort_transaction(trans
, ret
);
4232 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
4234 if (IS_ERR_OR_NULL(di
)) {
4239 btrfs_abort_transaction(trans
, ret
);
4243 leaf
= path
->nodes
[0];
4244 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4247 btrfs_release_path(path
);
4249 ret
= btrfs_delete_delayed_dir_index(trans
, BTRFS_I(dir
), index
);
4251 btrfs_abort_transaction(trans
, ret
);
4255 btrfs_i_size_write(BTRFS_I(dir
), dir
->i_size
- name_len
* 2);
4256 inode_inc_iversion(dir
);
4257 dir
->i_mtime
= dir
->i_ctime
= current_time(dir
);
4258 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
4260 btrfs_abort_transaction(trans
, ret
);
4262 btrfs_free_path(path
);
4267 * Helper to check if the subvolume references other subvolumes or if it's
4270 static noinline
int may_destroy_subvol(struct btrfs_root
*root
)
4272 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4273 struct btrfs_path
*path
;
4274 struct btrfs_dir_item
*di
;
4275 struct btrfs_key key
;
4279 path
= btrfs_alloc_path();
4283 /* Make sure this root isn't set as the default subvol */
4284 dir_id
= btrfs_super_root_dir(fs_info
->super_copy
);
4285 di
= btrfs_lookup_dir_item(NULL
, fs_info
->tree_root
, path
,
4286 dir_id
, "default", 7, 0);
4287 if (di
&& !IS_ERR(di
)) {
4288 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &key
);
4289 if (key
.objectid
== root
->root_key
.objectid
) {
4292 "deleting default subvolume %llu is not allowed",
4296 btrfs_release_path(path
);
4299 key
.objectid
= root
->root_key
.objectid
;
4300 key
.type
= BTRFS_ROOT_REF_KEY
;
4301 key
.offset
= (u64
)-1;
4303 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
4309 if (path
->slots
[0] > 0) {
4311 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
4312 if (key
.objectid
== root
->root_key
.objectid
&&
4313 key
.type
== BTRFS_ROOT_REF_KEY
)
4317 btrfs_free_path(path
);
4321 /* Delete all dentries for inodes belonging to the root */
4322 static void btrfs_prune_dentries(struct btrfs_root
*root
)
4324 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4325 struct rb_node
*node
;
4326 struct rb_node
*prev
;
4327 struct btrfs_inode
*entry
;
4328 struct inode
*inode
;
4331 if (!test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
4332 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
4334 spin_lock(&root
->inode_lock
);
4336 node
= root
->inode_tree
.rb_node
;
4340 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4342 if (objectid
< btrfs_ino(entry
))
4343 node
= node
->rb_left
;
4344 else if (objectid
> btrfs_ino(entry
))
4345 node
= node
->rb_right
;
4351 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
4352 if (objectid
<= btrfs_ino(entry
)) {
4356 prev
= rb_next(prev
);
4360 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4361 objectid
= btrfs_ino(entry
) + 1;
4362 inode
= igrab(&entry
->vfs_inode
);
4364 spin_unlock(&root
->inode_lock
);
4365 if (atomic_read(&inode
->i_count
) > 1)
4366 d_prune_aliases(inode
);
4368 * btrfs_drop_inode will have it removed from the inode
4369 * cache when its usage count hits zero.
4373 spin_lock(&root
->inode_lock
);
4377 if (cond_resched_lock(&root
->inode_lock
))
4380 node
= rb_next(node
);
4382 spin_unlock(&root
->inode_lock
);
4385 int btrfs_delete_subvolume(struct inode
*dir
, struct dentry
*dentry
)
4387 struct btrfs_fs_info
*fs_info
= btrfs_sb(dentry
->d_sb
);
4388 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4389 struct inode
*inode
= d_inode(dentry
);
4390 struct btrfs_root
*dest
= BTRFS_I(inode
)->root
;
4391 struct btrfs_trans_handle
*trans
;
4392 struct btrfs_block_rsv block_rsv
;
4398 * Don't allow to delete a subvolume with send in progress. This is
4399 * inside the inode lock so the error handling that has to drop the bit
4400 * again is not run concurrently.
4402 spin_lock(&dest
->root_item_lock
);
4403 if (dest
->send_in_progress
) {
4404 spin_unlock(&dest
->root_item_lock
);
4406 "attempt to delete subvolume %llu during send",
4407 dest
->root_key
.objectid
);
4410 root_flags
= btrfs_root_flags(&dest
->root_item
);
4411 btrfs_set_root_flags(&dest
->root_item
,
4412 root_flags
| BTRFS_ROOT_SUBVOL_DEAD
);
4413 spin_unlock(&dest
->root_item_lock
);
4415 down_write(&fs_info
->subvol_sem
);
4417 err
= may_destroy_subvol(dest
);
4421 btrfs_init_block_rsv(&block_rsv
, BTRFS_BLOCK_RSV_TEMP
);
4423 * One for dir inode,
4424 * two for dir entries,
4425 * two for root ref/backref.
4427 err
= btrfs_subvolume_reserve_metadata(root
, &block_rsv
, 5, true);
4431 trans
= btrfs_start_transaction(root
, 0);
4432 if (IS_ERR(trans
)) {
4433 err
= PTR_ERR(trans
);
4436 trans
->block_rsv
= &block_rsv
;
4437 trans
->bytes_reserved
= block_rsv
.size
;
4439 btrfs_record_snapshot_destroy(trans
, BTRFS_I(dir
));
4441 ret
= btrfs_unlink_subvol(trans
, dir
, dest
->root_key
.objectid
,
4442 dentry
->d_name
.name
, dentry
->d_name
.len
);
4445 btrfs_abort_transaction(trans
, ret
);
4449 btrfs_record_root_in_trans(trans
, dest
);
4451 memset(&dest
->root_item
.drop_progress
, 0,
4452 sizeof(dest
->root_item
.drop_progress
));
4453 dest
->root_item
.drop_level
= 0;
4454 btrfs_set_root_refs(&dest
->root_item
, 0);
4456 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &dest
->state
)) {
4457 ret
= btrfs_insert_orphan_item(trans
,
4459 dest
->root_key
.objectid
);
4461 btrfs_abort_transaction(trans
, ret
);
4467 ret
= btrfs_uuid_tree_remove(trans
, dest
->root_item
.uuid
,
4468 BTRFS_UUID_KEY_SUBVOL
,
4469 dest
->root_key
.objectid
);
4470 if (ret
&& ret
!= -ENOENT
) {
4471 btrfs_abort_transaction(trans
, ret
);
4475 if (!btrfs_is_empty_uuid(dest
->root_item
.received_uuid
)) {
4476 ret
= btrfs_uuid_tree_remove(trans
,
4477 dest
->root_item
.received_uuid
,
4478 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
4479 dest
->root_key
.objectid
);
4480 if (ret
&& ret
!= -ENOENT
) {
4481 btrfs_abort_transaction(trans
, ret
);
4488 trans
->block_rsv
= NULL
;
4489 trans
->bytes_reserved
= 0;
4490 ret
= btrfs_end_transaction(trans
);
4493 inode
->i_flags
|= S_DEAD
;
4495 btrfs_subvolume_release_metadata(fs_info
, &block_rsv
);
4497 up_write(&fs_info
->subvol_sem
);
4499 spin_lock(&dest
->root_item_lock
);
4500 root_flags
= btrfs_root_flags(&dest
->root_item
);
4501 btrfs_set_root_flags(&dest
->root_item
,
4502 root_flags
& ~BTRFS_ROOT_SUBVOL_DEAD
);
4503 spin_unlock(&dest
->root_item_lock
);
4505 d_invalidate(dentry
);
4506 btrfs_prune_dentries(dest
);
4507 ASSERT(dest
->send_in_progress
== 0);
4510 if (dest
->ino_cache_inode
) {
4511 iput(dest
->ino_cache_inode
);
4512 dest
->ino_cache_inode
= NULL
;
4519 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
4521 struct inode
*inode
= d_inode(dentry
);
4523 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4524 struct btrfs_trans_handle
*trans
;
4525 u64 last_unlink_trans
;
4527 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
4529 if (btrfs_ino(BTRFS_I(inode
)) == BTRFS_FIRST_FREE_OBJECTID
)
4530 return btrfs_delete_subvolume(dir
, dentry
);
4532 trans
= __unlink_start_trans(dir
);
4534 return PTR_ERR(trans
);
4536 if (unlikely(btrfs_ino(BTRFS_I(inode
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
4537 err
= btrfs_unlink_subvol(trans
, dir
,
4538 BTRFS_I(inode
)->location
.objectid
,
4539 dentry
->d_name
.name
,
4540 dentry
->d_name
.len
);
4544 err
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
4548 last_unlink_trans
= BTRFS_I(inode
)->last_unlink_trans
;
4550 /* now the directory is empty */
4551 err
= btrfs_unlink_inode(trans
, root
, BTRFS_I(dir
),
4552 BTRFS_I(d_inode(dentry
)), dentry
->d_name
.name
,
4553 dentry
->d_name
.len
);
4555 btrfs_i_size_write(BTRFS_I(inode
), 0);
4557 * Propagate the last_unlink_trans value of the deleted dir to
4558 * its parent directory. This is to prevent an unrecoverable
4559 * log tree in the case we do something like this:
4561 * 2) create snapshot under dir foo
4562 * 3) delete the snapshot
4565 * 6) fsync foo or some file inside foo
4567 if (last_unlink_trans
>= trans
->transid
)
4568 BTRFS_I(dir
)->last_unlink_trans
= last_unlink_trans
;
4571 btrfs_end_transaction(trans
);
4572 btrfs_btree_balance_dirty(root
->fs_info
);
4578 * Return this if we need to call truncate_block for the last bit of the
4581 #define NEED_TRUNCATE_BLOCK 1
4584 * this can truncate away extent items, csum items and directory items.
4585 * It starts at a high offset and removes keys until it can't find
4586 * any higher than new_size
4588 * csum items that cross the new i_size are truncated to the new size
4591 * min_type is the minimum key type to truncate down to. If set to 0, this
4592 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4594 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
4595 struct btrfs_root
*root
,
4596 struct inode
*inode
,
4597 u64 new_size
, u32 min_type
)
4599 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4600 struct btrfs_path
*path
;
4601 struct extent_buffer
*leaf
;
4602 struct btrfs_file_extent_item
*fi
;
4603 struct btrfs_key key
;
4604 struct btrfs_key found_key
;
4605 u64 extent_start
= 0;
4606 u64 extent_num_bytes
= 0;
4607 u64 extent_offset
= 0;
4609 u64 last_size
= new_size
;
4610 u32 found_type
= (u8
)-1;
4613 int pending_del_nr
= 0;
4614 int pending_del_slot
= 0;
4615 int extent_type
= -1;
4617 u64 ino
= btrfs_ino(BTRFS_I(inode
));
4618 u64 bytes_deleted
= 0;
4619 bool be_nice
= false;
4620 bool should_throttle
= false;
4622 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
4625 * for non-free space inodes and ref cows, we want to back off from
4628 if (!btrfs_is_free_space_inode(BTRFS_I(inode
)) &&
4629 test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4632 path
= btrfs_alloc_path();
4635 path
->reada
= READA_BACK
;
4638 * We want to drop from the next block forward in case this new size is
4639 * not block aligned since we will be keeping the last block of the
4640 * extent just the way it is.
4642 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4643 root
== fs_info
->tree_root
)
4644 btrfs_drop_extent_cache(BTRFS_I(inode
), ALIGN(new_size
,
4645 fs_info
->sectorsize
),
4649 * This function is also used to drop the items in the log tree before
4650 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4651 * it is used to drop the logged items. So we shouldn't kill the delayed
4654 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
4655 btrfs_kill_delayed_inode_items(BTRFS_I(inode
));
4658 key
.offset
= (u64
)-1;
4663 * with a 16K leaf size and 128MB extents, you can actually queue
4664 * up a huge file in a single leaf. Most of the time that
4665 * bytes_deleted is > 0, it will be huge by the time we get here
4667 if (be_nice
&& bytes_deleted
> SZ_32M
&&
4668 btrfs_should_end_transaction(trans
)) {
4673 path
->leave_spinning
= 1;
4674 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
4680 /* there are no items in the tree for us to truncate, we're
4683 if (path
->slots
[0] == 0)
4690 leaf
= path
->nodes
[0];
4691 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4692 found_type
= found_key
.type
;
4694 if (found_key
.objectid
!= ino
)
4697 if (found_type
< min_type
)
4700 item_end
= found_key
.offset
;
4701 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
4702 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4703 struct btrfs_file_extent_item
);
4704 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4705 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4707 btrfs_file_extent_num_bytes(leaf
, fi
);
4709 trace_btrfs_truncate_show_fi_regular(
4710 BTRFS_I(inode
), leaf
, fi
,
4712 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4713 item_end
+= btrfs_file_extent_ram_bytes(leaf
,
4716 trace_btrfs_truncate_show_fi_inline(
4717 BTRFS_I(inode
), leaf
, fi
, path
->slots
[0],
4722 if (found_type
> min_type
) {
4725 if (item_end
< new_size
)
4727 if (found_key
.offset
>= new_size
)
4733 /* FIXME, shrink the extent if the ref count is only 1 */
4734 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
4737 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4739 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4741 u64 orig_num_bytes
=
4742 btrfs_file_extent_num_bytes(leaf
, fi
);
4743 extent_num_bytes
= ALIGN(new_size
-
4745 fs_info
->sectorsize
);
4746 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4748 num_dec
= (orig_num_bytes
-
4750 if (test_bit(BTRFS_ROOT_REF_COWS
,
4753 inode_sub_bytes(inode
, num_dec
);
4754 btrfs_mark_buffer_dirty(leaf
);
4757 btrfs_file_extent_disk_num_bytes(leaf
,
4759 extent_offset
= found_key
.offset
-
4760 btrfs_file_extent_offset(leaf
, fi
);
4762 /* FIXME blocksize != 4096 */
4763 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
4764 if (extent_start
!= 0) {
4766 if (test_bit(BTRFS_ROOT_REF_COWS
,
4768 inode_sub_bytes(inode
, num_dec
);
4771 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4773 * we can't truncate inline items that have had
4777 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
4778 btrfs_file_extent_other_encoding(leaf
, fi
) == 0 &&
4779 btrfs_file_extent_compression(leaf
, fi
) == 0) {
4780 u32 size
= (u32
)(new_size
- found_key
.offset
);
4782 btrfs_set_file_extent_ram_bytes(leaf
, fi
, size
);
4783 size
= btrfs_file_extent_calc_inline_size(size
);
4784 btrfs_truncate_item(path
, size
, 1);
4785 } else if (!del_item
) {
4787 * We have to bail so the last_size is set to
4788 * just before this extent.
4790 ret
= NEED_TRUNCATE_BLOCK
;
4794 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4795 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4799 last_size
= found_key
.offset
;
4801 last_size
= new_size
;
4803 if (!pending_del_nr
) {
4804 /* no pending yet, add ourselves */
4805 pending_del_slot
= path
->slots
[0];
4807 } else if (pending_del_nr
&&
4808 path
->slots
[0] + 1 == pending_del_slot
) {
4809 /* hop on the pending chunk */
4811 pending_del_slot
= path
->slots
[0];
4818 should_throttle
= false;
4821 (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4822 root
== fs_info
->tree_root
)) {
4823 struct btrfs_ref ref
= { 0 };
4825 btrfs_set_path_blocking(path
);
4826 bytes_deleted
+= extent_num_bytes
;
4828 btrfs_init_generic_ref(&ref
, BTRFS_DROP_DELAYED_REF
,
4829 extent_start
, extent_num_bytes
, 0);
4830 ref
.real_root
= root
->root_key
.objectid
;
4831 btrfs_init_data_ref(&ref
, btrfs_header_owner(leaf
),
4832 ino
, extent_offset
);
4833 ret
= btrfs_free_extent(trans
, &ref
);
4835 btrfs_abort_transaction(trans
, ret
);
4839 if (btrfs_should_throttle_delayed_refs(trans
))
4840 should_throttle
= true;
4844 if (found_type
== BTRFS_INODE_ITEM_KEY
)
4847 if (path
->slots
[0] == 0 ||
4848 path
->slots
[0] != pending_del_slot
||
4850 if (pending_del_nr
) {
4851 ret
= btrfs_del_items(trans
, root
, path
,
4855 btrfs_abort_transaction(trans
, ret
);
4860 btrfs_release_path(path
);
4863 * We can generate a lot of delayed refs, so we need to
4864 * throttle every once and a while and make sure we're
4865 * adding enough space to keep up with the work we are
4866 * generating. Since we hold a transaction here we
4867 * can't flush, and we don't want to FLUSH_LIMIT because
4868 * we could have generated too many delayed refs to
4869 * actually allocate, so just bail if we're short and
4870 * let the normal reservation dance happen higher up.
4872 if (should_throttle
) {
4873 ret
= btrfs_delayed_refs_rsv_refill(fs_info
,
4874 BTRFS_RESERVE_NO_FLUSH
);
4886 if (ret
>= 0 && pending_del_nr
) {
4889 err
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
4892 btrfs_abort_transaction(trans
, err
);
4896 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4897 ASSERT(last_size
>= new_size
);
4898 if (!ret
&& last_size
> new_size
)
4899 last_size
= new_size
;
4900 btrfs_ordered_update_i_size(inode
, last_size
, NULL
);
4903 btrfs_free_path(path
);
4908 * btrfs_truncate_block - read, zero a chunk and write a block
4909 * @inode - inode that we're zeroing
4910 * @from - the offset to start zeroing
4911 * @len - the length to zero, 0 to zero the entire range respective to the
4913 * @front - zero up to the offset instead of from the offset on
4915 * This will find the block for the "from" offset and cow the block and zero the
4916 * part we want to zero. This is used with truncate and hole punching.
4918 int btrfs_truncate_block(struct inode
*inode
, loff_t from
, loff_t len
,
4921 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
4922 struct address_space
*mapping
= inode
->i_mapping
;
4923 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4924 struct btrfs_ordered_extent
*ordered
;
4925 struct extent_state
*cached_state
= NULL
;
4926 struct extent_changeset
*data_reserved
= NULL
;
4928 u32 blocksize
= fs_info
->sectorsize
;
4929 pgoff_t index
= from
>> PAGE_SHIFT
;
4930 unsigned offset
= from
& (blocksize
- 1);
4932 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4937 if (IS_ALIGNED(offset
, blocksize
) &&
4938 (!len
|| IS_ALIGNED(len
, blocksize
)))
4941 block_start
= round_down(from
, blocksize
);
4942 block_end
= block_start
+ blocksize
- 1;
4944 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
,
4945 block_start
, blocksize
);
4950 page
= find_or_create_page(mapping
, index
, mask
);
4952 btrfs_delalloc_release_space(inode
, data_reserved
,
4953 block_start
, blocksize
, true);
4954 btrfs_delalloc_release_extents(BTRFS_I(inode
), blocksize
, true);
4959 if (!PageUptodate(page
)) {
4960 ret
= btrfs_readpage(NULL
, page
);
4962 if (page
->mapping
!= mapping
) {
4967 if (!PageUptodate(page
)) {
4972 wait_on_page_writeback(page
);
4974 lock_extent_bits(io_tree
, block_start
, block_end
, &cached_state
);
4975 set_page_extent_mapped(page
);
4977 ordered
= btrfs_lookup_ordered_extent(inode
, block_start
);
4979 unlock_extent_cached(io_tree
, block_start
, block_end
,
4983 btrfs_start_ordered_extent(inode
, ordered
, 1);
4984 btrfs_put_ordered_extent(ordered
);
4988 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, block_start
, block_end
,
4989 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4990 0, 0, &cached_state
);
4992 ret
= btrfs_set_extent_delalloc(inode
, block_start
, block_end
, 0,
4995 unlock_extent_cached(io_tree
, block_start
, block_end
,
5000 if (offset
!= blocksize
) {
5002 len
= blocksize
- offset
;
5005 memset(kaddr
+ (block_start
- page_offset(page
)),
5008 memset(kaddr
+ (block_start
- page_offset(page
)) + offset
,
5010 flush_dcache_page(page
);
5013 ClearPageChecked(page
);
5014 set_page_dirty(page
);
5015 unlock_extent_cached(io_tree
, block_start
, block_end
, &cached_state
);
5019 btrfs_delalloc_release_space(inode
, data_reserved
, block_start
,
5021 btrfs_delalloc_release_extents(BTRFS_I(inode
), blocksize
, (ret
!= 0));
5025 extent_changeset_free(data_reserved
);
5029 static int maybe_insert_hole(struct btrfs_root
*root
, struct inode
*inode
,
5030 u64 offset
, u64 len
)
5032 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5033 struct btrfs_trans_handle
*trans
;
5037 * Still need to make sure the inode looks like it's been updated so
5038 * that any holes get logged if we fsync.
5040 if (btrfs_fs_incompat(fs_info
, NO_HOLES
)) {
5041 BTRFS_I(inode
)->last_trans
= fs_info
->generation
;
5042 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
5043 BTRFS_I(inode
)->last_log_commit
= root
->last_log_commit
;
5048 * 1 - for the one we're dropping
5049 * 1 - for the one we're adding
5050 * 1 - for updating the inode.
5052 trans
= btrfs_start_transaction(root
, 3);
5054 return PTR_ERR(trans
);
5056 ret
= btrfs_drop_extents(trans
, root
, inode
, offset
, offset
+ len
, 1);
5058 btrfs_abort_transaction(trans
, ret
);
5059 btrfs_end_transaction(trans
);
5063 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(BTRFS_I(inode
)),
5064 offset
, 0, 0, len
, 0, len
, 0, 0, 0);
5066 btrfs_abort_transaction(trans
, ret
);
5068 btrfs_update_inode(trans
, root
, inode
);
5069 btrfs_end_transaction(trans
);
5074 * This function puts in dummy file extents for the area we're creating a hole
5075 * for. So if we are truncating this file to a larger size we need to insert
5076 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5077 * the range between oldsize and size
5079 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
5081 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5082 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5083 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5084 struct extent_map
*em
= NULL
;
5085 struct extent_state
*cached_state
= NULL
;
5086 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5087 u64 hole_start
= ALIGN(oldsize
, fs_info
->sectorsize
);
5088 u64 block_end
= ALIGN(size
, fs_info
->sectorsize
);
5095 * If our size started in the middle of a block we need to zero out the
5096 * rest of the block before we expand the i_size, otherwise we could
5097 * expose stale data.
5099 err
= btrfs_truncate_block(inode
, oldsize
, 0, 0);
5103 if (size
<= hole_start
)
5106 btrfs_lock_and_flush_ordered_range(io_tree
, BTRFS_I(inode
), hole_start
,
5107 block_end
- 1, &cached_state
);
5108 cur_offset
= hole_start
;
5110 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0, cur_offset
,
5111 block_end
- cur_offset
, 0);
5117 last_byte
= min(extent_map_end(em
), block_end
);
5118 last_byte
= ALIGN(last_byte
, fs_info
->sectorsize
);
5119 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
5120 struct extent_map
*hole_em
;
5121 hole_size
= last_byte
- cur_offset
;
5123 err
= maybe_insert_hole(root
, inode
, cur_offset
,
5127 btrfs_drop_extent_cache(BTRFS_I(inode
), cur_offset
,
5128 cur_offset
+ hole_size
- 1, 0);
5129 hole_em
= alloc_extent_map();
5131 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
5132 &BTRFS_I(inode
)->runtime_flags
);
5135 hole_em
->start
= cur_offset
;
5136 hole_em
->len
= hole_size
;
5137 hole_em
->orig_start
= cur_offset
;
5139 hole_em
->block_start
= EXTENT_MAP_HOLE
;
5140 hole_em
->block_len
= 0;
5141 hole_em
->orig_block_len
= 0;
5142 hole_em
->ram_bytes
= hole_size
;
5143 hole_em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
5144 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
5145 hole_em
->generation
= fs_info
->generation
;
5148 write_lock(&em_tree
->lock
);
5149 err
= add_extent_mapping(em_tree
, hole_em
, 1);
5150 write_unlock(&em_tree
->lock
);
5153 btrfs_drop_extent_cache(BTRFS_I(inode
),
5158 free_extent_map(hole_em
);
5161 free_extent_map(em
);
5163 cur_offset
= last_byte
;
5164 if (cur_offset
>= block_end
)
5167 free_extent_map(em
);
5168 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
);
5172 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
5174 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5175 struct btrfs_trans_handle
*trans
;
5176 loff_t oldsize
= i_size_read(inode
);
5177 loff_t newsize
= attr
->ia_size
;
5178 int mask
= attr
->ia_valid
;
5182 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5183 * special case where we need to update the times despite not having
5184 * these flags set. For all other operations the VFS set these flags
5185 * explicitly if it wants a timestamp update.
5187 if (newsize
!= oldsize
) {
5188 inode_inc_iversion(inode
);
5189 if (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))
5190 inode
->i_ctime
= inode
->i_mtime
=
5191 current_time(inode
);
5194 if (newsize
> oldsize
) {
5196 * Don't do an expanding truncate while snapshotting is ongoing.
5197 * This is to ensure the snapshot captures a fully consistent
5198 * state of this file - if the snapshot captures this expanding
5199 * truncation, it must capture all writes that happened before
5202 btrfs_wait_for_snapshot_creation(root
);
5203 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
5205 btrfs_end_write_no_snapshotting(root
);
5209 trans
= btrfs_start_transaction(root
, 1);
5210 if (IS_ERR(trans
)) {
5211 btrfs_end_write_no_snapshotting(root
);
5212 return PTR_ERR(trans
);
5215 i_size_write(inode
, newsize
);
5216 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
5217 pagecache_isize_extended(inode
, oldsize
, newsize
);
5218 ret
= btrfs_update_inode(trans
, root
, inode
);
5219 btrfs_end_write_no_snapshotting(root
);
5220 btrfs_end_transaction(trans
);
5224 * We're truncating a file that used to have good data down to
5225 * zero. Make sure it gets into the ordered flush list so that
5226 * any new writes get down to disk quickly.
5229 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
5230 &BTRFS_I(inode
)->runtime_flags
);
5232 truncate_setsize(inode
, newsize
);
5234 /* Disable nonlocked read DIO to avoid the endless truncate */
5235 btrfs_inode_block_unlocked_dio(BTRFS_I(inode
));
5236 inode_dio_wait(inode
);
5237 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode
));
5239 ret
= btrfs_truncate(inode
, newsize
== oldsize
);
5240 if (ret
&& inode
->i_nlink
) {
5244 * Truncate failed, so fix up the in-memory size. We
5245 * adjusted disk_i_size down as we removed extents, so
5246 * wait for disk_i_size to be stable and then update the
5247 * in-memory size to match.
5249 err
= btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5252 i_size_write(inode
, BTRFS_I(inode
)->disk_i_size
);
5259 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5261 struct inode
*inode
= d_inode(dentry
);
5262 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5265 if (btrfs_root_readonly(root
))
5268 err
= setattr_prepare(dentry
, attr
);
5272 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
5273 err
= btrfs_setsize(inode
, attr
);
5278 if (attr
->ia_valid
) {
5279 setattr_copy(inode
, attr
);
5280 inode_inc_iversion(inode
);
5281 err
= btrfs_dirty_inode(inode
);
5283 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
5284 err
= posix_acl_chmod(inode
, inode
->i_mode
);
5291 * While truncating the inode pages during eviction, we get the VFS calling
5292 * btrfs_invalidatepage() against each page of the inode. This is slow because
5293 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5294 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5295 * extent_state structures over and over, wasting lots of time.
5297 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5298 * those expensive operations on a per page basis and do only the ordered io
5299 * finishing, while we release here the extent_map and extent_state structures,
5300 * without the excessive merging and splitting.
5302 static void evict_inode_truncate_pages(struct inode
*inode
)
5304 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5305 struct extent_map_tree
*map_tree
= &BTRFS_I(inode
)->extent_tree
;
5306 struct rb_node
*node
;
5308 ASSERT(inode
->i_state
& I_FREEING
);
5309 truncate_inode_pages_final(&inode
->i_data
);
5311 write_lock(&map_tree
->lock
);
5312 while (!RB_EMPTY_ROOT(&map_tree
->map
.rb_root
)) {
5313 struct extent_map
*em
;
5315 node
= rb_first_cached(&map_tree
->map
);
5316 em
= rb_entry(node
, struct extent_map
, rb_node
);
5317 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5318 clear_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
5319 remove_extent_mapping(map_tree
, em
);
5320 free_extent_map(em
);
5321 if (need_resched()) {
5322 write_unlock(&map_tree
->lock
);
5324 write_lock(&map_tree
->lock
);
5327 write_unlock(&map_tree
->lock
);
5330 * Keep looping until we have no more ranges in the io tree.
5331 * We can have ongoing bios started by readpages (called from readahead)
5332 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5333 * still in progress (unlocked the pages in the bio but did not yet
5334 * unlocked the ranges in the io tree). Therefore this means some
5335 * ranges can still be locked and eviction started because before
5336 * submitting those bios, which are executed by a separate task (work
5337 * queue kthread), inode references (inode->i_count) were not taken
5338 * (which would be dropped in the end io callback of each bio).
5339 * Therefore here we effectively end up waiting for those bios and
5340 * anyone else holding locked ranges without having bumped the inode's
5341 * reference count - if we don't do it, when they access the inode's
5342 * io_tree to unlock a range it may be too late, leading to an
5343 * use-after-free issue.
5345 spin_lock(&io_tree
->lock
);
5346 while (!RB_EMPTY_ROOT(&io_tree
->state
)) {
5347 struct extent_state
*state
;
5348 struct extent_state
*cached_state
= NULL
;
5351 unsigned state_flags
;
5353 node
= rb_first(&io_tree
->state
);
5354 state
= rb_entry(node
, struct extent_state
, rb_node
);
5355 start
= state
->start
;
5357 state_flags
= state
->state
;
5358 spin_unlock(&io_tree
->lock
);
5360 lock_extent_bits(io_tree
, start
, end
, &cached_state
);
5363 * If still has DELALLOC flag, the extent didn't reach disk,
5364 * and its reserved space won't be freed by delayed_ref.
5365 * So we need to free its reserved space here.
5366 * (Refer to comment in btrfs_invalidatepage, case 2)
5368 * Note, end is the bytenr of last byte, so we need + 1 here.
5370 if (state_flags
& EXTENT_DELALLOC
)
5371 btrfs_qgroup_free_data(inode
, NULL
, start
, end
- start
+ 1);
5373 clear_extent_bit(io_tree
, start
, end
,
5374 EXTENT_LOCKED
| EXTENT_DELALLOC
|
5375 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
, 1, 1,
5379 spin_lock(&io_tree
->lock
);
5381 spin_unlock(&io_tree
->lock
);
5384 static struct btrfs_trans_handle
*evict_refill_and_join(struct btrfs_root
*root
,
5385 struct btrfs_block_rsv
*rsv
)
5387 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5388 struct btrfs_block_rsv
*global_rsv
= &fs_info
->global_block_rsv
;
5389 struct btrfs_trans_handle
*trans
;
5390 u64 delayed_refs_extra
= btrfs_calc_insert_metadata_size(fs_info
, 1);
5394 * Eviction should be taking place at some place safe because of our
5395 * delayed iputs. However the normal flushing code will run delayed
5396 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5398 * We reserve the delayed_refs_extra here again because we can't use
5399 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5400 * above. We reserve our extra bit here because we generate a ton of
5401 * delayed refs activity by truncating.
5403 * If we cannot make our reservation we'll attempt to steal from the
5404 * global reserve, because we really want to be able to free up space.
5406 ret
= btrfs_block_rsv_refill(root
, rsv
, rsv
->size
+ delayed_refs_extra
,
5407 BTRFS_RESERVE_FLUSH_EVICT
);
5410 * Try to steal from the global reserve if there is space for
5413 if (btrfs_check_space_for_delayed_refs(fs_info
) ||
5414 btrfs_block_rsv_migrate(global_rsv
, rsv
, rsv
->size
, 0)) {
5416 "could not allocate space for delete; will truncate on mount");
5417 return ERR_PTR(-ENOSPC
);
5419 delayed_refs_extra
= 0;
5422 trans
= btrfs_join_transaction(root
);
5426 if (delayed_refs_extra
) {
5427 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
5428 trans
->bytes_reserved
= delayed_refs_extra
;
5429 btrfs_block_rsv_migrate(rsv
, trans
->block_rsv
,
5430 delayed_refs_extra
, 1);
5435 void btrfs_evict_inode(struct inode
*inode
)
5437 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5438 struct btrfs_trans_handle
*trans
;
5439 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5440 struct btrfs_block_rsv
*rsv
;
5443 trace_btrfs_inode_evict(inode
);
5450 evict_inode_truncate_pages(inode
);
5452 if (inode
->i_nlink
&&
5453 ((btrfs_root_refs(&root
->root_item
) != 0 &&
5454 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
) ||
5455 btrfs_is_free_space_inode(BTRFS_I(inode
))))
5458 if (is_bad_inode(inode
))
5461 btrfs_free_io_failure_record(BTRFS_I(inode
), 0, (u64
)-1);
5463 if (test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
))
5466 if (inode
->i_nlink
> 0) {
5467 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0 &&
5468 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
);
5472 ret
= btrfs_commit_inode_delayed_inode(BTRFS_I(inode
));
5476 rsv
= btrfs_alloc_block_rsv(fs_info
, BTRFS_BLOCK_RSV_TEMP
);
5479 rsv
->size
= btrfs_calc_metadata_size(fs_info
, 1);
5482 btrfs_i_size_write(BTRFS_I(inode
), 0);
5485 trans
= evict_refill_and_join(root
, rsv
);
5489 trans
->block_rsv
= rsv
;
5491 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
5492 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
5493 btrfs_end_transaction(trans
);
5494 btrfs_btree_balance_dirty(fs_info
);
5495 if (ret
&& ret
!= -ENOSPC
&& ret
!= -EAGAIN
)
5502 * Errors here aren't a big deal, it just means we leave orphan items in
5503 * the tree. They will be cleaned up on the next mount. If the inode
5504 * number gets reused, cleanup deletes the orphan item without doing
5505 * anything, and unlink reuses the existing orphan item.
5507 * If it turns out that we are dropping too many of these, we might want
5508 * to add a mechanism for retrying these after a commit.
5510 trans
= evict_refill_and_join(root
, rsv
);
5511 if (!IS_ERR(trans
)) {
5512 trans
->block_rsv
= rsv
;
5513 btrfs_orphan_del(trans
, BTRFS_I(inode
));
5514 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
5515 btrfs_end_transaction(trans
);
5518 if (!(root
== fs_info
->tree_root
||
5519 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
5520 btrfs_return_ino(root
, btrfs_ino(BTRFS_I(inode
)));
5523 btrfs_free_block_rsv(fs_info
, rsv
);
5526 * If we didn't successfully delete, the orphan item will still be in
5527 * the tree and we'll retry on the next mount. Again, we might also want
5528 * to retry these periodically in the future.
5530 btrfs_remove_delayed_node(BTRFS_I(inode
));
5535 * Return the key found in the dir entry in the location pointer, fill @type
5536 * with BTRFS_FT_*, and return 0.
5538 * If no dir entries were found, returns -ENOENT.
5539 * If found a corrupted location in dir entry, returns -EUCLEAN.
5541 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
5542 struct btrfs_key
*location
, u8
*type
)
5544 const char *name
= dentry
->d_name
.name
;
5545 int namelen
= dentry
->d_name
.len
;
5546 struct btrfs_dir_item
*di
;
5547 struct btrfs_path
*path
;
5548 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5551 path
= btrfs_alloc_path();
5555 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(BTRFS_I(dir
)),
5557 if (IS_ERR_OR_NULL(di
)) {
5558 ret
= di
? PTR_ERR(di
) : -ENOENT
;
5562 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
5563 if (location
->type
!= BTRFS_INODE_ITEM_KEY
&&
5564 location
->type
!= BTRFS_ROOT_ITEM_KEY
) {
5566 btrfs_warn(root
->fs_info
,
5567 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5568 __func__
, name
, btrfs_ino(BTRFS_I(dir
)),
5569 location
->objectid
, location
->type
, location
->offset
);
5572 *type
= btrfs_dir_type(path
->nodes
[0], di
);
5574 btrfs_free_path(path
);
5579 * when we hit a tree root in a directory, the btrfs part of the inode
5580 * needs to be changed to reflect the root directory of the tree root. This
5581 * is kind of like crossing a mount point.
5583 static int fixup_tree_root_location(struct btrfs_fs_info
*fs_info
,
5585 struct dentry
*dentry
,
5586 struct btrfs_key
*location
,
5587 struct btrfs_root
**sub_root
)
5589 struct btrfs_path
*path
;
5590 struct btrfs_root
*new_root
;
5591 struct btrfs_root_ref
*ref
;
5592 struct extent_buffer
*leaf
;
5593 struct btrfs_key key
;
5597 path
= btrfs_alloc_path();
5604 key
.objectid
= BTRFS_I(dir
)->root
->root_key
.objectid
;
5605 key
.type
= BTRFS_ROOT_REF_KEY
;
5606 key
.offset
= location
->objectid
;
5608 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
5615 leaf
= path
->nodes
[0];
5616 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
5617 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(BTRFS_I(dir
)) ||
5618 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
5621 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
5622 (unsigned long)(ref
+ 1),
5623 dentry
->d_name
.len
);
5627 btrfs_release_path(path
);
5629 new_root
= btrfs_read_fs_root_no_name(fs_info
, location
);
5630 if (IS_ERR(new_root
)) {
5631 err
= PTR_ERR(new_root
);
5635 *sub_root
= new_root
;
5636 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
5637 location
->type
= BTRFS_INODE_ITEM_KEY
;
5638 location
->offset
= 0;
5641 btrfs_free_path(path
);
5645 static void inode_tree_add(struct inode
*inode
)
5647 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5648 struct btrfs_inode
*entry
;
5650 struct rb_node
*parent
;
5651 struct rb_node
*new = &BTRFS_I(inode
)->rb_node
;
5652 u64 ino
= btrfs_ino(BTRFS_I(inode
));
5654 if (inode_unhashed(inode
))
5657 spin_lock(&root
->inode_lock
);
5658 p
= &root
->inode_tree
.rb_node
;
5661 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
5663 if (ino
< btrfs_ino(entry
))
5664 p
= &parent
->rb_left
;
5665 else if (ino
> btrfs_ino(entry
))
5666 p
= &parent
->rb_right
;
5668 WARN_ON(!(entry
->vfs_inode
.i_state
&
5669 (I_WILL_FREE
| I_FREEING
)));
5670 rb_replace_node(parent
, new, &root
->inode_tree
);
5671 RB_CLEAR_NODE(parent
);
5672 spin_unlock(&root
->inode_lock
);
5676 rb_link_node(new, parent
, p
);
5677 rb_insert_color(new, &root
->inode_tree
);
5678 spin_unlock(&root
->inode_lock
);
5681 static void inode_tree_del(struct inode
*inode
)
5683 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5684 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5687 spin_lock(&root
->inode_lock
);
5688 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
5689 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
5690 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
5691 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5693 spin_unlock(&root
->inode_lock
);
5695 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0) {
5696 synchronize_srcu(&fs_info
->subvol_srcu
);
5697 spin_lock(&root
->inode_lock
);
5698 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5699 spin_unlock(&root
->inode_lock
);
5701 btrfs_add_dead_root(root
);
5706 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
5708 struct btrfs_iget_args
*args
= p
;
5709 inode
->i_ino
= args
->location
->objectid
;
5710 memcpy(&BTRFS_I(inode
)->location
, args
->location
,
5711 sizeof(*args
->location
));
5712 BTRFS_I(inode
)->root
= args
->root
;
5716 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
5718 struct btrfs_iget_args
*args
= opaque
;
5719 return args
->location
->objectid
== BTRFS_I(inode
)->location
.objectid
&&
5720 args
->root
== BTRFS_I(inode
)->root
;
5723 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
5724 struct btrfs_key
*location
,
5725 struct btrfs_root
*root
)
5727 struct inode
*inode
;
5728 struct btrfs_iget_args args
;
5729 unsigned long hashval
= btrfs_inode_hash(location
->objectid
, root
);
5731 args
.location
= location
;
5734 inode
= iget5_locked(s
, hashval
, btrfs_find_actor
,
5735 btrfs_init_locked_inode
,
5740 /* Get an inode object given its location and corresponding root.
5741 * Returns in *is_new if the inode was read from disk
5743 struct inode
*btrfs_iget_path(struct super_block
*s
, struct btrfs_key
*location
,
5744 struct btrfs_root
*root
, int *new,
5745 struct btrfs_path
*path
)
5747 struct inode
*inode
;
5749 inode
= btrfs_iget_locked(s
, location
, root
);
5751 return ERR_PTR(-ENOMEM
);
5753 if (inode
->i_state
& I_NEW
) {
5756 ret
= btrfs_read_locked_inode(inode
, path
);
5758 inode_tree_add(inode
);
5759 unlock_new_inode(inode
);
5765 * ret > 0 can come from btrfs_search_slot called by
5766 * btrfs_read_locked_inode, this means the inode item
5771 inode
= ERR_PTR(ret
);
5778 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
5779 struct btrfs_root
*root
, int *new)
5781 return btrfs_iget_path(s
, location
, root
, new, NULL
);
5784 static struct inode
*new_simple_dir(struct super_block
*s
,
5785 struct btrfs_key
*key
,
5786 struct btrfs_root
*root
)
5788 struct inode
*inode
= new_inode(s
);
5791 return ERR_PTR(-ENOMEM
);
5793 BTRFS_I(inode
)->root
= root
;
5794 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
5795 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
5797 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
5798 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
5799 inode
->i_opflags
&= ~IOP_XATTR
;
5800 inode
->i_fop
= &simple_dir_operations
;
5801 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
5802 inode
->i_mtime
= current_time(inode
);
5803 inode
->i_atime
= inode
->i_mtime
;
5804 inode
->i_ctime
= inode
->i_mtime
;
5805 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
5810 static inline u8
btrfs_inode_type(struct inode
*inode
)
5813 * Compile-time asserts that generic FT_* types still match
5816 BUILD_BUG_ON(BTRFS_FT_UNKNOWN
!= FT_UNKNOWN
);
5817 BUILD_BUG_ON(BTRFS_FT_REG_FILE
!= FT_REG_FILE
);
5818 BUILD_BUG_ON(BTRFS_FT_DIR
!= FT_DIR
);
5819 BUILD_BUG_ON(BTRFS_FT_CHRDEV
!= FT_CHRDEV
);
5820 BUILD_BUG_ON(BTRFS_FT_BLKDEV
!= FT_BLKDEV
);
5821 BUILD_BUG_ON(BTRFS_FT_FIFO
!= FT_FIFO
);
5822 BUILD_BUG_ON(BTRFS_FT_SOCK
!= FT_SOCK
);
5823 BUILD_BUG_ON(BTRFS_FT_SYMLINK
!= FT_SYMLINK
);
5825 return fs_umode_to_ftype(inode
->i_mode
);
5828 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
5830 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
5831 struct inode
*inode
;
5832 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5833 struct btrfs_root
*sub_root
= root
;
5834 struct btrfs_key location
;
5839 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
5840 return ERR_PTR(-ENAMETOOLONG
);
5842 ret
= btrfs_inode_by_name(dir
, dentry
, &location
, &di_type
);
5844 return ERR_PTR(ret
);
5846 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
5847 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
5851 /* Do extra check against inode mode with di_type */
5852 if (btrfs_inode_type(inode
) != di_type
) {
5854 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5855 inode
->i_mode
, btrfs_inode_type(inode
),
5858 return ERR_PTR(-EUCLEAN
);
5863 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
5864 ret
= fixup_tree_root_location(fs_info
, dir
, dentry
,
5865 &location
, &sub_root
);
5868 inode
= ERR_PTR(ret
);
5870 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
5872 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
5874 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
5876 if (!IS_ERR(inode
) && root
!= sub_root
) {
5877 down_read(&fs_info
->cleanup_work_sem
);
5878 if (!sb_rdonly(inode
->i_sb
))
5879 ret
= btrfs_orphan_cleanup(sub_root
);
5880 up_read(&fs_info
->cleanup_work_sem
);
5883 inode
= ERR_PTR(ret
);
5890 static int btrfs_dentry_delete(const struct dentry
*dentry
)
5892 struct btrfs_root
*root
;
5893 struct inode
*inode
= d_inode(dentry
);
5895 if (!inode
&& !IS_ROOT(dentry
))
5896 inode
= d_inode(dentry
->d_parent
);
5899 root
= BTRFS_I(inode
)->root
;
5900 if (btrfs_root_refs(&root
->root_item
) == 0)
5903 if (btrfs_ino(BTRFS_I(inode
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5909 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5912 struct inode
*inode
= btrfs_lookup_dentry(dir
, dentry
);
5914 if (inode
== ERR_PTR(-ENOENT
))
5916 return d_splice_alias(inode
, dentry
);
5920 * All this infrastructure exists because dir_emit can fault, and we are holding
5921 * the tree lock when doing readdir. For now just allocate a buffer and copy
5922 * our information into that, and then dir_emit from the buffer. This is
5923 * similar to what NFS does, only we don't keep the buffer around in pagecache
5924 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5925 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5928 static int btrfs_opendir(struct inode
*inode
, struct file
*file
)
5930 struct btrfs_file_private
*private;
5932 private = kzalloc(sizeof(struct btrfs_file_private
), GFP_KERNEL
);
5935 private->filldir_buf
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
5936 if (!private->filldir_buf
) {
5940 file
->private_data
= private;
5951 static int btrfs_filldir(void *addr
, int entries
, struct dir_context
*ctx
)
5954 struct dir_entry
*entry
= addr
;
5955 char *name
= (char *)(entry
+ 1);
5957 ctx
->pos
= get_unaligned(&entry
->offset
);
5958 if (!dir_emit(ctx
, name
, get_unaligned(&entry
->name_len
),
5959 get_unaligned(&entry
->ino
),
5960 get_unaligned(&entry
->type
)))
5962 addr
+= sizeof(struct dir_entry
) +
5963 get_unaligned(&entry
->name_len
);
5969 static int btrfs_real_readdir(struct file
*file
, struct dir_context
*ctx
)
5971 struct inode
*inode
= file_inode(file
);
5972 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5973 struct btrfs_file_private
*private = file
->private_data
;
5974 struct btrfs_dir_item
*di
;
5975 struct btrfs_key key
;
5976 struct btrfs_key found_key
;
5977 struct btrfs_path
*path
;
5979 struct list_head ins_list
;
5980 struct list_head del_list
;
5982 struct extent_buffer
*leaf
;
5989 struct btrfs_key location
;
5991 if (!dir_emit_dots(file
, ctx
))
5994 path
= btrfs_alloc_path();
5998 addr
= private->filldir_buf
;
5999 path
->reada
= READA_FORWARD
;
6001 INIT_LIST_HEAD(&ins_list
);
6002 INIT_LIST_HEAD(&del_list
);
6003 put
= btrfs_readdir_get_delayed_items(inode
, &ins_list
, &del_list
);
6006 key
.type
= BTRFS_DIR_INDEX_KEY
;
6007 key
.offset
= ctx
->pos
;
6008 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
6010 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6015 struct dir_entry
*entry
;
6017 leaf
= path
->nodes
[0];
6018 slot
= path
->slots
[0];
6019 if (slot
>= btrfs_header_nritems(leaf
)) {
6020 ret
= btrfs_next_leaf(root
, path
);
6028 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6030 if (found_key
.objectid
!= key
.objectid
)
6032 if (found_key
.type
!= BTRFS_DIR_INDEX_KEY
)
6034 if (found_key
.offset
< ctx
->pos
)
6036 if (btrfs_should_delete_dir_index(&del_list
, found_key
.offset
))
6038 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
6039 name_len
= btrfs_dir_name_len(leaf
, di
);
6040 if ((total_len
+ sizeof(struct dir_entry
) + name_len
) >=
6042 btrfs_release_path(path
);
6043 ret
= btrfs_filldir(private->filldir_buf
, entries
, ctx
);
6046 addr
= private->filldir_buf
;
6053 put_unaligned(name_len
, &entry
->name_len
);
6054 name_ptr
= (char *)(entry
+ 1);
6055 read_extent_buffer(leaf
, name_ptr
, (unsigned long)(di
+ 1),
6057 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf
, di
)),
6059 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
6060 put_unaligned(location
.objectid
, &entry
->ino
);
6061 put_unaligned(found_key
.offset
, &entry
->offset
);
6063 addr
+= sizeof(struct dir_entry
) + name_len
;
6064 total_len
+= sizeof(struct dir_entry
) + name_len
;
6068 btrfs_release_path(path
);
6070 ret
= btrfs_filldir(private->filldir_buf
, entries
, ctx
);
6074 ret
= btrfs_readdir_delayed_dir_index(ctx
, &ins_list
);
6079 * Stop new entries from being returned after we return the last
6082 * New directory entries are assigned a strictly increasing
6083 * offset. This means that new entries created during readdir
6084 * are *guaranteed* to be seen in the future by that readdir.
6085 * This has broken buggy programs which operate on names as
6086 * they're returned by readdir. Until we re-use freed offsets
6087 * we have this hack to stop new entries from being returned
6088 * under the assumption that they'll never reach this huge
6091 * This is being careful not to overflow 32bit loff_t unless the
6092 * last entry requires it because doing so has broken 32bit apps
6095 if (ctx
->pos
>= INT_MAX
)
6096 ctx
->pos
= LLONG_MAX
;
6103 btrfs_readdir_put_delayed_items(inode
, &ins_list
, &del_list
);
6104 btrfs_free_path(path
);
6109 * This is somewhat expensive, updating the tree every time the
6110 * inode changes. But, it is most likely to find the inode in cache.
6111 * FIXME, needs more benchmarking...there are no reasons other than performance
6112 * to keep or drop this code.
6114 static int btrfs_dirty_inode(struct inode
*inode
)
6116 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
6117 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6118 struct btrfs_trans_handle
*trans
;
6121 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
6124 trans
= btrfs_join_transaction(root
);
6126 return PTR_ERR(trans
);
6128 ret
= btrfs_update_inode(trans
, root
, inode
);
6129 if (ret
&& ret
== -ENOSPC
) {
6130 /* whoops, lets try again with the full transaction */
6131 btrfs_end_transaction(trans
);
6132 trans
= btrfs_start_transaction(root
, 1);
6134 return PTR_ERR(trans
);
6136 ret
= btrfs_update_inode(trans
, root
, inode
);
6138 btrfs_end_transaction(trans
);
6139 if (BTRFS_I(inode
)->delayed_node
)
6140 btrfs_balance_delayed_items(fs_info
);
6146 * This is a copy of file_update_time. We need this so we can return error on
6147 * ENOSPC for updating the inode in the case of file write and mmap writes.
6149 static int btrfs_update_time(struct inode
*inode
, struct timespec64
*now
,
6152 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6153 bool dirty
= flags
& ~S_VERSION
;
6155 if (btrfs_root_readonly(root
))
6158 if (flags
& S_VERSION
)
6159 dirty
|= inode_maybe_inc_iversion(inode
, dirty
);
6160 if (flags
& S_CTIME
)
6161 inode
->i_ctime
= *now
;
6162 if (flags
& S_MTIME
)
6163 inode
->i_mtime
= *now
;
6164 if (flags
& S_ATIME
)
6165 inode
->i_atime
= *now
;
6166 return dirty
? btrfs_dirty_inode(inode
) : 0;
6170 * find the highest existing sequence number in a directory
6171 * and then set the in-memory index_cnt variable to reflect
6172 * free sequence numbers
6174 static int btrfs_set_inode_index_count(struct btrfs_inode
*inode
)
6176 struct btrfs_root
*root
= inode
->root
;
6177 struct btrfs_key key
, found_key
;
6178 struct btrfs_path
*path
;
6179 struct extent_buffer
*leaf
;
6182 key
.objectid
= btrfs_ino(inode
);
6183 key
.type
= BTRFS_DIR_INDEX_KEY
;
6184 key
.offset
= (u64
)-1;
6186 path
= btrfs_alloc_path();
6190 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6193 /* FIXME: we should be able to handle this */
6199 * MAGIC NUMBER EXPLANATION:
6200 * since we search a directory based on f_pos we have to start at 2
6201 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6202 * else has to start at 2
6204 if (path
->slots
[0] == 0) {
6205 inode
->index_cnt
= 2;
6211 leaf
= path
->nodes
[0];
6212 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6214 if (found_key
.objectid
!= btrfs_ino(inode
) ||
6215 found_key
.type
!= BTRFS_DIR_INDEX_KEY
) {
6216 inode
->index_cnt
= 2;
6220 inode
->index_cnt
= found_key
.offset
+ 1;
6222 btrfs_free_path(path
);
6227 * helper to find a free sequence number in a given directory. This current
6228 * code is very simple, later versions will do smarter things in the btree
6230 int btrfs_set_inode_index(struct btrfs_inode
*dir
, u64
*index
)
6234 if (dir
->index_cnt
== (u64
)-1) {
6235 ret
= btrfs_inode_delayed_dir_index_count(dir
);
6237 ret
= btrfs_set_inode_index_count(dir
);
6243 *index
= dir
->index_cnt
;
6249 static int btrfs_insert_inode_locked(struct inode
*inode
)
6251 struct btrfs_iget_args args
;
6252 args
.location
= &BTRFS_I(inode
)->location
;
6253 args
.root
= BTRFS_I(inode
)->root
;
6255 return insert_inode_locked4(inode
,
6256 btrfs_inode_hash(inode
->i_ino
, BTRFS_I(inode
)->root
),
6257 btrfs_find_actor
, &args
);
6261 * Inherit flags from the parent inode.
6263 * Currently only the compression flags and the cow flags are inherited.
6265 static void btrfs_inherit_iflags(struct inode
*inode
, struct inode
*dir
)
6272 flags
= BTRFS_I(dir
)->flags
;
6274 if (flags
& BTRFS_INODE_NOCOMPRESS
) {
6275 BTRFS_I(inode
)->flags
&= ~BTRFS_INODE_COMPRESS
;
6276 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
6277 } else if (flags
& BTRFS_INODE_COMPRESS
) {
6278 BTRFS_I(inode
)->flags
&= ~BTRFS_INODE_NOCOMPRESS
;
6279 BTRFS_I(inode
)->flags
|= BTRFS_INODE_COMPRESS
;
6282 if (flags
& BTRFS_INODE_NODATACOW
) {
6283 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
6284 if (S_ISREG(inode
->i_mode
))
6285 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6288 btrfs_sync_inode_flags_to_i_flags(inode
);
6291 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
6292 struct btrfs_root
*root
,
6294 const char *name
, int name_len
,
6295 u64 ref_objectid
, u64 objectid
,
6296 umode_t mode
, u64
*index
)
6298 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6299 struct inode
*inode
;
6300 struct btrfs_inode_item
*inode_item
;
6301 struct btrfs_key
*location
;
6302 struct btrfs_path
*path
;
6303 struct btrfs_inode_ref
*ref
;
6304 struct btrfs_key key
[2];
6306 int nitems
= name
? 2 : 1;
6310 path
= btrfs_alloc_path();
6312 return ERR_PTR(-ENOMEM
);
6314 inode
= new_inode(fs_info
->sb
);
6316 btrfs_free_path(path
);
6317 return ERR_PTR(-ENOMEM
);
6321 * O_TMPFILE, set link count to 0, so that after this point,
6322 * we fill in an inode item with the correct link count.
6325 set_nlink(inode
, 0);
6328 * we have to initialize this early, so we can reclaim the inode
6329 * number if we fail afterwards in this function.
6331 inode
->i_ino
= objectid
;
6334 trace_btrfs_inode_request(dir
);
6336 ret
= btrfs_set_inode_index(BTRFS_I(dir
), index
);
6338 btrfs_free_path(path
);
6340 return ERR_PTR(ret
);
6346 * index_cnt is ignored for everything but a dir,
6347 * btrfs_set_inode_index_count has an explanation for the magic
6350 BTRFS_I(inode
)->index_cnt
= 2;
6351 BTRFS_I(inode
)->dir_index
= *index
;
6352 BTRFS_I(inode
)->root
= root
;
6353 BTRFS_I(inode
)->generation
= trans
->transid
;
6354 inode
->i_generation
= BTRFS_I(inode
)->generation
;
6357 * We could have gotten an inode number from somebody who was fsynced
6358 * and then removed in this same transaction, so let's just set full
6359 * sync since it will be a full sync anyway and this will blow away the
6360 * old info in the log.
6362 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
6364 key
[0].objectid
= objectid
;
6365 key
[0].type
= BTRFS_INODE_ITEM_KEY
;
6368 sizes
[0] = sizeof(struct btrfs_inode_item
);
6372 * Start new inodes with an inode_ref. This is slightly more
6373 * efficient for small numbers of hard links since they will
6374 * be packed into one item. Extended refs will kick in if we
6375 * add more hard links than can fit in the ref item.
6377 key
[1].objectid
= objectid
;
6378 key
[1].type
= BTRFS_INODE_REF_KEY
;
6379 key
[1].offset
= ref_objectid
;
6381 sizes
[1] = name_len
+ sizeof(*ref
);
6384 location
= &BTRFS_I(inode
)->location
;
6385 location
->objectid
= objectid
;
6386 location
->offset
= 0;
6387 location
->type
= BTRFS_INODE_ITEM_KEY
;
6389 ret
= btrfs_insert_inode_locked(inode
);
6395 path
->leave_spinning
= 1;
6396 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, nitems
);
6400 inode_init_owner(inode
, dir
, mode
);
6401 inode_set_bytes(inode
, 0);
6403 inode
->i_mtime
= current_time(inode
);
6404 inode
->i_atime
= inode
->i_mtime
;
6405 inode
->i_ctime
= inode
->i_mtime
;
6406 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
6408 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
6409 struct btrfs_inode_item
);
6410 memzero_extent_buffer(path
->nodes
[0], (unsigned long)inode_item
,
6411 sizeof(*inode_item
));
6412 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
6415 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
6416 struct btrfs_inode_ref
);
6417 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
6418 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
6419 ptr
= (unsigned long)(ref
+ 1);
6420 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
6423 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6424 btrfs_free_path(path
);
6426 btrfs_inherit_iflags(inode
, dir
);
6428 if (S_ISREG(mode
)) {
6429 if (btrfs_test_opt(fs_info
, NODATASUM
))
6430 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6431 if (btrfs_test_opt(fs_info
, NODATACOW
))
6432 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
6433 BTRFS_INODE_NODATASUM
;
6436 inode_tree_add(inode
);
6438 trace_btrfs_inode_new(inode
);
6439 btrfs_set_inode_last_trans(trans
, inode
);
6441 btrfs_update_root_times(trans
, root
);
6443 ret
= btrfs_inode_inherit_props(trans
, inode
, dir
);
6446 "error inheriting props for ino %llu (root %llu): %d",
6447 btrfs_ino(BTRFS_I(inode
)), root
->root_key
.objectid
, ret
);
6452 discard_new_inode(inode
);
6455 BTRFS_I(dir
)->index_cnt
--;
6456 btrfs_free_path(path
);
6457 return ERR_PTR(ret
);
6461 * utility function to add 'inode' into 'parent_inode' with
6462 * a give name and a given sequence number.
6463 * if 'add_backref' is true, also insert a backref from the
6464 * inode to the parent directory.
6466 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
6467 struct btrfs_inode
*parent_inode
, struct btrfs_inode
*inode
,
6468 const char *name
, int name_len
, int add_backref
, u64 index
)
6471 struct btrfs_key key
;
6472 struct btrfs_root
*root
= parent_inode
->root
;
6473 u64 ino
= btrfs_ino(inode
);
6474 u64 parent_ino
= btrfs_ino(parent_inode
);
6476 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6477 memcpy(&key
, &inode
->root
->root_key
, sizeof(key
));
6480 key
.type
= BTRFS_INODE_ITEM_KEY
;
6484 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6485 ret
= btrfs_add_root_ref(trans
, key
.objectid
,
6486 root
->root_key
.objectid
, parent_ino
,
6487 index
, name
, name_len
);
6488 } else if (add_backref
) {
6489 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
6493 /* Nothing to clean up yet */
6497 ret
= btrfs_insert_dir_item(trans
, name
, name_len
, parent_inode
, &key
,
6498 btrfs_inode_type(&inode
->vfs_inode
), index
);
6499 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
6502 btrfs_abort_transaction(trans
, ret
);
6506 btrfs_i_size_write(parent_inode
, parent_inode
->vfs_inode
.i_size
+
6508 inode_inc_iversion(&parent_inode
->vfs_inode
);
6510 * If we are replaying a log tree, we do not want to update the mtime
6511 * and ctime of the parent directory with the current time, since the
6512 * log replay procedure is responsible for setting them to their correct
6513 * values (the ones it had when the fsync was done).
6515 if (!test_bit(BTRFS_FS_LOG_RECOVERING
, &root
->fs_info
->flags
)) {
6516 struct timespec64 now
= current_time(&parent_inode
->vfs_inode
);
6518 parent_inode
->vfs_inode
.i_mtime
= now
;
6519 parent_inode
->vfs_inode
.i_ctime
= now
;
6521 ret
= btrfs_update_inode(trans
, root
, &parent_inode
->vfs_inode
);
6523 btrfs_abort_transaction(trans
, ret
);
6527 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6530 err
= btrfs_del_root_ref(trans
, key
.objectid
,
6531 root
->root_key
.objectid
, parent_ino
,
6532 &local_index
, name
, name_len
);
6534 btrfs_abort_transaction(trans
, err
);
6535 } else if (add_backref
) {
6539 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
6540 ino
, parent_ino
, &local_index
);
6542 btrfs_abort_transaction(trans
, err
);
6545 /* Return the original error code */
6549 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
6550 struct btrfs_inode
*dir
, struct dentry
*dentry
,
6551 struct btrfs_inode
*inode
, int backref
, u64 index
)
6553 int err
= btrfs_add_link(trans
, dir
, inode
,
6554 dentry
->d_name
.name
, dentry
->d_name
.len
,
6561 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
6562 umode_t mode
, dev_t rdev
)
6564 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6565 struct btrfs_trans_handle
*trans
;
6566 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6567 struct inode
*inode
= NULL
;
6573 * 2 for inode item and ref
6575 * 1 for xattr if selinux is on
6577 trans
= btrfs_start_transaction(root
, 5);
6579 return PTR_ERR(trans
);
6581 err
= btrfs_find_free_ino(root
, &objectid
);
6585 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6586 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)), objectid
,
6588 if (IS_ERR(inode
)) {
6589 err
= PTR_ERR(inode
);
6595 * If the active LSM wants to access the inode during
6596 * d_instantiate it needs these. Smack checks to see
6597 * if the filesystem supports xattrs by looking at the
6600 inode
->i_op
= &btrfs_special_inode_operations
;
6601 init_special_inode(inode
, inode
->i_mode
, rdev
);
6603 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6607 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
, BTRFS_I(inode
),
6612 btrfs_update_inode(trans
, root
, inode
);
6613 d_instantiate_new(dentry
, inode
);
6616 btrfs_end_transaction(trans
);
6617 btrfs_btree_balance_dirty(fs_info
);
6619 inode_dec_link_count(inode
);
6620 discard_new_inode(inode
);
6625 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
6626 umode_t mode
, bool excl
)
6628 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6629 struct btrfs_trans_handle
*trans
;
6630 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6631 struct inode
*inode
= NULL
;
6637 * 2 for inode item and ref
6639 * 1 for xattr if selinux is on
6641 trans
= btrfs_start_transaction(root
, 5);
6643 return PTR_ERR(trans
);
6645 err
= btrfs_find_free_ino(root
, &objectid
);
6649 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6650 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)), objectid
,
6652 if (IS_ERR(inode
)) {
6653 err
= PTR_ERR(inode
);
6658 * If the active LSM wants to access the inode during
6659 * d_instantiate it needs these. Smack checks to see
6660 * if the filesystem supports xattrs by looking at the
6663 inode
->i_fop
= &btrfs_file_operations
;
6664 inode
->i_op
= &btrfs_file_inode_operations
;
6665 inode
->i_mapping
->a_ops
= &btrfs_aops
;
6667 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6671 err
= btrfs_update_inode(trans
, root
, inode
);
6675 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
, BTRFS_I(inode
),
6680 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
6681 d_instantiate_new(dentry
, inode
);
6684 btrfs_end_transaction(trans
);
6686 inode_dec_link_count(inode
);
6687 discard_new_inode(inode
);
6689 btrfs_btree_balance_dirty(fs_info
);
6693 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
6694 struct dentry
*dentry
)
6696 struct btrfs_trans_handle
*trans
= NULL
;
6697 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6698 struct inode
*inode
= d_inode(old_dentry
);
6699 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
6704 /* do not allow sys_link's with other subvols of the same device */
6705 if (root
->root_key
.objectid
!= BTRFS_I(inode
)->root
->root_key
.objectid
)
6708 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
6711 err
= btrfs_set_inode_index(BTRFS_I(dir
), &index
);
6716 * 2 items for inode and inode ref
6717 * 2 items for dir items
6718 * 1 item for parent inode
6719 * 1 item for orphan item deletion if O_TMPFILE
6721 trans
= btrfs_start_transaction(root
, inode
->i_nlink
? 5 : 6);
6722 if (IS_ERR(trans
)) {
6723 err
= PTR_ERR(trans
);
6728 /* There are several dir indexes for this inode, clear the cache. */
6729 BTRFS_I(inode
)->dir_index
= 0ULL;
6731 inode_inc_iversion(inode
);
6732 inode
->i_ctime
= current_time(inode
);
6734 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
6736 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
, BTRFS_I(inode
),
6742 struct dentry
*parent
= dentry
->d_parent
;
6745 err
= btrfs_update_inode(trans
, root
, inode
);
6748 if (inode
->i_nlink
== 1) {
6750 * If new hard link count is 1, it's a file created
6751 * with open(2) O_TMPFILE flag.
6753 err
= btrfs_orphan_del(trans
, BTRFS_I(inode
));
6757 d_instantiate(dentry
, inode
);
6758 ret
= btrfs_log_new_name(trans
, BTRFS_I(inode
), NULL
, parent
,
6760 if (ret
== BTRFS_NEED_TRANS_COMMIT
) {
6761 err
= btrfs_commit_transaction(trans
);
6768 btrfs_end_transaction(trans
);
6770 inode_dec_link_count(inode
);
6773 btrfs_btree_balance_dirty(fs_info
);
6777 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
6779 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6780 struct inode
*inode
= NULL
;
6781 struct btrfs_trans_handle
*trans
;
6782 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6788 * 2 items for inode and ref
6789 * 2 items for dir items
6790 * 1 for xattr if selinux is on
6792 trans
= btrfs_start_transaction(root
, 5);
6794 return PTR_ERR(trans
);
6796 err
= btrfs_find_free_ino(root
, &objectid
);
6800 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6801 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)), objectid
,
6802 S_IFDIR
| mode
, &index
);
6803 if (IS_ERR(inode
)) {
6804 err
= PTR_ERR(inode
);
6809 /* these must be set before we unlock the inode */
6810 inode
->i_op
= &btrfs_dir_inode_operations
;
6811 inode
->i_fop
= &btrfs_dir_file_operations
;
6813 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6817 btrfs_i_size_write(BTRFS_I(inode
), 0);
6818 err
= btrfs_update_inode(trans
, root
, inode
);
6822 err
= btrfs_add_link(trans
, BTRFS_I(dir
), BTRFS_I(inode
),
6823 dentry
->d_name
.name
,
6824 dentry
->d_name
.len
, 0, index
);
6828 d_instantiate_new(dentry
, inode
);
6831 btrfs_end_transaction(trans
);
6833 inode_dec_link_count(inode
);
6834 discard_new_inode(inode
);
6836 btrfs_btree_balance_dirty(fs_info
);
6840 static noinline
int uncompress_inline(struct btrfs_path
*path
,
6842 size_t pg_offset
, u64 extent_offset
,
6843 struct btrfs_file_extent_item
*item
)
6846 struct extent_buffer
*leaf
= path
->nodes
[0];
6849 unsigned long inline_size
;
6853 WARN_ON(pg_offset
!= 0);
6854 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6855 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6856 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
6857 btrfs_item_nr(path
->slots
[0]));
6858 tmp
= kmalloc(inline_size
, GFP_NOFS
);
6861 ptr
= btrfs_file_extent_inline_start(item
);
6863 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
6865 max_size
= min_t(unsigned long, PAGE_SIZE
, max_size
);
6866 ret
= btrfs_decompress(compress_type
, tmp
, page
,
6867 extent_offset
, inline_size
, max_size
);
6870 * decompression code contains a memset to fill in any space between the end
6871 * of the uncompressed data and the end of max_size in case the decompressed
6872 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6873 * the end of an inline extent and the beginning of the next block, so we
6874 * cover that region here.
6877 if (max_size
+ pg_offset
< PAGE_SIZE
) {
6878 char *map
= kmap(page
);
6879 memset(map
+ pg_offset
+ max_size
, 0, PAGE_SIZE
- max_size
- pg_offset
);
6887 * a bit scary, this does extent mapping from logical file offset to the disk.
6888 * the ugly parts come from merging extents from the disk with the in-ram
6889 * representation. This gets more complex because of the data=ordered code,
6890 * where the in-ram extents might be locked pending data=ordered completion.
6892 * This also copies inline extents directly into the page.
6894 struct extent_map
*btrfs_get_extent(struct btrfs_inode
*inode
,
6896 size_t pg_offset
, u64 start
, u64 len
,
6899 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
6902 u64 extent_start
= 0;
6904 u64 objectid
= btrfs_ino(inode
);
6905 int extent_type
= -1;
6906 struct btrfs_path
*path
= NULL
;
6907 struct btrfs_root
*root
= inode
->root
;
6908 struct btrfs_file_extent_item
*item
;
6909 struct extent_buffer
*leaf
;
6910 struct btrfs_key found_key
;
6911 struct extent_map
*em
= NULL
;
6912 struct extent_map_tree
*em_tree
= &inode
->extent_tree
;
6913 struct extent_io_tree
*io_tree
= &inode
->io_tree
;
6914 const bool new_inline
= !page
|| create
;
6916 read_lock(&em_tree
->lock
);
6917 em
= lookup_extent_mapping(em_tree
, start
, len
);
6919 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
6920 read_unlock(&em_tree
->lock
);
6923 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
6924 free_extent_map(em
);
6925 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
6926 free_extent_map(em
);
6930 em
= alloc_extent_map();
6935 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
6936 em
->start
= EXTENT_MAP_HOLE
;
6937 em
->orig_start
= EXTENT_MAP_HOLE
;
6939 em
->block_len
= (u64
)-1;
6941 path
= btrfs_alloc_path();
6947 /* Chances are we'll be called again, so go ahead and do readahead */
6948 path
->reada
= READA_FORWARD
;
6951 * Unless we're going to uncompress the inline extent, no sleep would
6954 path
->leave_spinning
= 1;
6956 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, objectid
, start
, 0);
6960 } else if (ret
> 0) {
6961 if (path
->slots
[0] == 0)
6966 leaf
= path
->nodes
[0];
6967 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6968 struct btrfs_file_extent_item
);
6969 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6970 if (found_key
.objectid
!= objectid
||
6971 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
6973 * If we backup past the first extent we want to move forward
6974 * and see if there is an extent in front of us, otherwise we'll
6975 * say there is a hole for our whole search range which can
6982 extent_type
= btrfs_file_extent_type(leaf
, item
);
6983 extent_start
= found_key
.offset
;
6984 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
6985 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6986 /* Only regular file could have regular/prealloc extent */
6987 if (!S_ISREG(inode
->vfs_inode
.i_mode
)) {
6990 "regular/prealloc extent found for non-regular inode %llu",
6994 extent_end
= extent_start
+
6995 btrfs_file_extent_num_bytes(leaf
, item
);
6997 trace_btrfs_get_extent_show_fi_regular(inode
, leaf
, item
,
6999 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
7002 size
= btrfs_file_extent_ram_bytes(leaf
, item
);
7003 extent_end
= ALIGN(extent_start
+ size
,
7004 fs_info
->sectorsize
);
7006 trace_btrfs_get_extent_show_fi_inline(inode
, leaf
, item
,
7011 if (start
>= extent_end
) {
7013 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
7014 ret
= btrfs_next_leaf(root
, path
);
7018 } else if (ret
> 0) {
7021 leaf
= path
->nodes
[0];
7023 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7024 if (found_key
.objectid
!= objectid
||
7025 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
7027 if (start
+ len
<= found_key
.offset
)
7029 if (start
> found_key
.offset
)
7032 /* New extent overlaps with existing one */
7034 em
->orig_start
= start
;
7035 em
->len
= found_key
.offset
- start
;
7036 em
->block_start
= EXTENT_MAP_HOLE
;
7040 btrfs_extent_item_to_extent_map(inode
, path
, item
,
7043 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
7044 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7046 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
7050 size_t extent_offset
;
7056 size
= btrfs_file_extent_ram_bytes(leaf
, item
);
7057 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
7058 copy_size
= min_t(u64
, PAGE_SIZE
- pg_offset
,
7059 size
- extent_offset
);
7060 em
->start
= extent_start
+ extent_offset
;
7061 em
->len
= ALIGN(copy_size
, fs_info
->sectorsize
);
7062 em
->orig_block_len
= em
->len
;
7063 em
->orig_start
= em
->start
;
7064 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
7066 btrfs_set_path_blocking(path
);
7067 if (!PageUptodate(page
)) {
7068 if (btrfs_file_extent_compression(leaf
, item
) !=
7069 BTRFS_COMPRESS_NONE
) {
7070 ret
= uncompress_inline(path
, page
, pg_offset
,
7071 extent_offset
, item
);
7078 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
7080 if (pg_offset
+ copy_size
< PAGE_SIZE
) {
7081 memset(map
+ pg_offset
+ copy_size
, 0,
7082 PAGE_SIZE
- pg_offset
-
7087 flush_dcache_page(page
);
7089 set_extent_uptodate(io_tree
, em
->start
,
7090 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
7095 em
->orig_start
= start
;
7097 em
->block_start
= EXTENT_MAP_HOLE
;
7099 btrfs_release_path(path
);
7100 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
7102 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7103 em
->start
, em
->len
, start
, len
);
7109 write_lock(&em_tree
->lock
);
7110 err
= btrfs_add_extent_mapping(fs_info
, em_tree
, &em
, start
, len
);
7111 write_unlock(&em_tree
->lock
);
7113 btrfs_free_path(path
);
7115 trace_btrfs_get_extent(root
, inode
, em
);
7118 free_extent_map(em
);
7119 return ERR_PTR(err
);
7121 BUG_ON(!em
); /* Error is always set */
7125 struct extent_map
*btrfs_get_extent_fiemap(struct btrfs_inode
*inode
,
7128 struct extent_map
*em
;
7129 struct extent_map
*hole_em
= NULL
;
7130 u64 delalloc_start
= start
;
7136 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
7140 * If our em maps to:
7142 * - a pre-alloc extent,
7143 * there might actually be delalloc bytes behind it.
7145 if (em
->block_start
!= EXTENT_MAP_HOLE
&&
7146 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7151 /* check to see if we've wrapped (len == -1 or similar) */
7160 /* ok, we didn't find anything, lets look for delalloc */
7161 delalloc_len
= count_range_bits(&inode
->io_tree
, &delalloc_start
,
7162 end
, len
, EXTENT_DELALLOC
, 1);
7163 delalloc_end
= delalloc_start
+ delalloc_len
;
7164 if (delalloc_end
< delalloc_start
)
7165 delalloc_end
= (u64
)-1;
7168 * We didn't find anything useful, return the original results from
7171 if (delalloc_start
> end
|| delalloc_end
<= start
) {
7178 * Adjust the delalloc_start to make sure it doesn't go backwards from
7179 * the start they passed in
7181 delalloc_start
= max(start
, delalloc_start
);
7182 delalloc_len
= delalloc_end
- delalloc_start
;
7184 if (delalloc_len
> 0) {
7187 const u64 hole_end
= extent_map_end(hole_em
);
7189 em
= alloc_extent_map();
7198 * When btrfs_get_extent can't find anything it returns one
7201 * Make sure what it found really fits our range, and adjust to
7202 * make sure it is based on the start from the caller
7204 if (hole_end
<= start
|| hole_em
->start
> end
) {
7205 free_extent_map(hole_em
);
7208 hole_start
= max(hole_em
->start
, start
);
7209 hole_len
= hole_end
- hole_start
;
7212 if (hole_em
&& delalloc_start
> hole_start
) {
7214 * Our hole starts before our delalloc, so we have to
7215 * return just the parts of the hole that go until the
7218 em
->len
= min(hole_len
, delalloc_start
- hole_start
);
7219 em
->start
= hole_start
;
7220 em
->orig_start
= hole_start
;
7222 * Don't adjust block start at all, it is fixed at
7225 em
->block_start
= hole_em
->block_start
;
7226 em
->block_len
= hole_len
;
7227 if (test_bit(EXTENT_FLAG_PREALLOC
, &hole_em
->flags
))
7228 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
7231 * Hole is out of passed range or it starts after
7234 em
->start
= delalloc_start
;
7235 em
->len
= delalloc_len
;
7236 em
->orig_start
= delalloc_start
;
7237 em
->block_start
= EXTENT_MAP_DELALLOC
;
7238 em
->block_len
= delalloc_len
;
7245 free_extent_map(hole_em
);
7247 free_extent_map(em
);
7248 return ERR_PTR(err
);
7253 static struct extent_map
*btrfs_create_dio_extent(struct inode
*inode
,
7256 const u64 orig_start
,
7257 const u64 block_start
,
7258 const u64 block_len
,
7259 const u64 orig_block_len
,
7260 const u64 ram_bytes
,
7263 struct extent_map
*em
= NULL
;
7266 if (type
!= BTRFS_ORDERED_NOCOW
) {
7267 em
= create_io_em(inode
, start
, len
, orig_start
,
7268 block_start
, block_len
, orig_block_len
,
7270 BTRFS_COMPRESS_NONE
, /* compress_type */
7275 ret
= btrfs_add_ordered_extent_dio(inode
, start
, block_start
,
7276 len
, block_len
, type
);
7279 free_extent_map(em
);
7280 btrfs_drop_extent_cache(BTRFS_I(inode
), start
,
7281 start
+ len
- 1, 0);
7290 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
7293 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7294 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7295 struct extent_map
*em
;
7296 struct btrfs_key ins
;
7300 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
7301 ret
= btrfs_reserve_extent(root
, len
, len
, fs_info
->sectorsize
,
7302 0, alloc_hint
, &ins
, 1, 1);
7304 return ERR_PTR(ret
);
7306 em
= btrfs_create_dio_extent(inode
, start
, ins
.offset
, start
,
7307 ins
.objectid
, ins
.offset
, ins
.offset
,
7308 ins
.offset
, BTRFS_ORDERED_REGULAR
);
7309 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
7311 btrfs_free_reserved_extent(fs_info
, ins
.objectid
,
7318 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7319 * block must be cow'd
7321 noinline
int can_nocow_extent(struct inode
*inode
, u64 offset
, u64
*len
,
7322 u64
*orig_start
, u64
*orig_block_len
,
7325 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7326 struct btrfs_path
*path
;
7328 struct extent_buffer
*leaf
;
7329 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7330 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7331 struct btrfs_file_extent_item
*fi
;
7332 struct btrfs_key key
;
7339 bool nocow
= (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
);
7341 path
= btrfs_alloc_path();
7345 ret
= btrfs_lookup_file_extent(NULL
, root
, path
,
7346 btrfs_ino(BTRFS_I(inode
)), offset
, 0);
7350 slot
= path
->slots
[0];
7353 /* can't find the item, must cow */
7360 leaf
= path
->nodes
[0];
7361 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
7362 if (key
.objectid
!= btrfs_ino(BTRFS_I(inode
)) ||
7363 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7364 /* not our file or wrong item type, must cow */
7368 if (key
.offset
> offset
) {
7369 /* Wrong offset, must cow */
7373 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
7374 found_type
= btrfs_file_extent_type(leaf
, fi
);
7375 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
7376 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
7377 /* not a regular extent, must cow */
7381 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_REG
)
7384 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
7385 if (extent_end
<= offset
)
7388 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7389 if (disk_bytenr
== 0)
7392 if (btrfs_file_extent_compression(leaf
, fi
) ||
7393 btrfs_file_extent_encryption(leaf
, fi
) ||
7394 btrfs_file_extent_other_encoding(leaf
, fi
))
7398 * Do the same check as in btrfs_cross_ref_exist but without the
7399 * unnecessary search.
7401 if (btrfs_file_extent_generation(leaf
, fi
) <=
7402 btrfs_root_last_snapshot(&root
->root_item
))
7405 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
7408 *orig_start
= key
.offset
- backref_offset
;
7409 *orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7410 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
7413 if (btrfs_extent_readonly(fs_info
, disk_bytenr
))
7416 num_bytes
= min(offset
+ *len
, extent_end
) - offset
;
7417 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7420 range_end
= round_up(offset
+ num_bytes
,
7421 root
->fs_info
->sectorsize
) - 1;
7422 ret
= test_range_bit(io_tree
, offset
, range_end
,
7423 EXTENT_DELALLOC
, 0, NULL
);
7430 btrfs_release_path(path
);
7433 * look for other files referencing this extent, if we
7434 * find any we must cow
7437 ret
= btrfs_cross_ref_exist(root
, btrfs_ino(BTRFS_I(inode
)),
7438 key
.offset
- backref_offset
, disk_bytenr
);
7445 * adjust disk_bytenr and num_bytes to cover just the bytes
7446 * in this extent we are about to write. If there
7447 * are any csums in that range we have to cow in order
7448 * to keep the csums correct
7450 disk_bytenr
+= backref_offset
;
7451 disk_bytenr
+= offset
- key
.offset
;
7452 if (csum_exist_in_range(fs_info
, disk_bytenr
, num_bytes
))
7455 * all of the above have passed, it is safe to overwrite this extent
7461 btrfs_free_path(path
);
7465 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
7466 struct extent_state
**cached_state
, int writing
)
7468 struct btrfs_ordered_extent
*ordered
;
7472 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7475 * We're concerned with the entire range that we're going to be
7476 * doing DIO to, so we need to make sure there's no ordered
7477 * extents in this range.
7479 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), lockstart
,
7480 lockend
- lockstart
+ 1);
7483 * We need to make sure there are no buffered pages in this
7484 * range either, we could have raced between the invalidate in
7485 * generic_file_direct_write and locking the extent. The
7486 * invalidate needs to happen so that reads after a write do not
7490 (!writing
|| !filemap_range_has_page(inode
->i_mapping
,
7491 lockstart
, lockend
)))
7494 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7499 * If we are doing a DIO read and the ordered extent we
7500 * found is for a buffered write, we can not wait for it
7501 * to complete and retry, because if we do so we can
7502 * deadlock with concurrent buffered writes on page
7503 * locks. This happens only if our DIO read covers more
7504 * than one extent map, if at this point has already
7505 * created an ordered extent for a previous extent map
7506 * and locked its range in the inode's io tree, and a
7507 * concurrent write against that previous extent map's
7508 * range and this range started (we unlock the ranges
7509 * in the io tree only when the bios complete and
7510 * buffered writes always lock pages before attempting
7511 * to lock range in the io tree).
7514 test_bit(BTRFS_ORDERED_DIRECT
, &ordered
->flags
))
7515 btrfs_start_ordered_extent(inode
, ordered
, 1);
7518 btrfs_put_ordered_extent(ordered
);
7521 * We could trigger writeback for this range (and wait
7522 * for it to complete) and then invalidate the pages for
7523 * this range (through invalidate_inode_pages2_range()),
7524 * but that can lead us to a deadlock with a concurrent
7525 * call to readpages() (a buffered read or a defrag call
7526 * triggered a readahead) on a page lock due to an
7527 * ordered dio extent we created before but did not have
7528 * yet a corresponding bio submitted (whence it can not
7529 * complete), which makes readpages() wait for that
7530 * ordered extent to complete while holding a lock on
7545 /* The callers of this must take lock_extent() */
7546 static struct extent_map
*create_io_em(struct inode
*inode
, u64 start
, u64 len
,
7547 u64 orig_start
, u64 block_start
,
7548 u64 block_len
, u64 orig_block_len
,
7549 u64 ram_bytes
, int compress_type
,
7552 struct extent_map_tree
*em_tree
;
7553 struct extent_map
*em
;
7554 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7557 ASSERT(type
== BTRFS_ORDERED_PREALLOC
||
7558 type
== BTRFS_ORDERED_COMPRESSED
||
7559 type
== BTRFS_ORDERED_NOCOW
||
7560 type
== BTRFS_ORDERED_REGULAR
);
7562 em_tree
= &BTRFS_I(inode
)->extent_tree
;
7563 em
= alloc_extent_map();
7565 return ERR_PTR(-ENOMEM
);
7568 em
->orig_start
= orig_start
;
7570 em
->block_len
= block_len
;
7571 em
->block_start
= block_start
;
7572 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
7573 em
->orig_block_len
= orig_block_len
;
7574 em
->ram_bytes
= ram_bytes
;
7575 em
->generation
= -1;
7576 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
7577 if (type
== BTRFS_ORDERED_PREALLOC
) {
7578 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
7579 } else if (type
== BTRFS_ORDERED_COMPRESSED
) {
7580 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
7581 em
->compress_type
= compress_type
;
7585 btrfs_drop_extent_cache(BTRFS_I(inode
), em
->start
,
7586 em
->start
+ em
->len
- 1, 0);
7587 write_lock(&em_tree
->lock
);
7588 ret
= add_extent_mapping(em_tree
, em
, 1);
7589 write_unlock(&em_tree
->lock
);
7591 * The caller has taken lock_extent(), who could race with us
7594 } while (ret
== -EEXIST
);
7597 free_extent_map(em
);
7598 return ERR_PTR(ret
);
7601 /* em got 2 refs now, callers needs to do free_extent_map once. */
7606 static int btrfs_get_blocks_direct_read(struct extent_map
*em
,
7607 struct buffer_head
*bh_result
,
7608 struct inode
*inode
,
7611 if (em
->block_start
== EXTENT_MAP_HOLE
||
7612 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7615 len
= min(len
, em
->len
- (start
- em
->start
));
7617 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
7619 bh_result
->b_size
= len
;
7620 bh_result
->b_bdev
= em
->bdev
;
7621 set_buffer_mapped(bh_result
);
7626 static int btrfs_get_blocks_direct_write(struct extent_map
**map
,
7627 struct buffer_head
*bh_result
,
7628 struct inode
*inode
,
7629 struct btrfs_dio_data
*dio_data
,
7632 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7633 struct extent_map
*em
= *map
;
7637 * We don't allocate a new extent in the following cases
7639 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7641 * 2) The extent is marked as PREALLOC. We're good to go here and can
7642 * just use the extent.
7645 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
7646 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
7647 em
->block_start
!= EXTENT_MAP_HOLE
)) {
7649 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
7651 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7652 type
= BTRFS_ORDERED_PREALLOC
;
7654 type
= BTRFS_ORDERED_NOCOW
;
7655 len
= min(len
, em
->len
- (start
- em
->start
));
7656 block_start
= em
->block_start
+ (start
- em
->start
);
7658 if (can_nocow_extent(inode
, start
, &len
, &orig_start
,
7659 &orig_block_len
, &ram_bytes
) == 1 &&
7660 btrfs_inc_nocow_writers(fs_info
, block_start
)) {
7661 struct extent_map
*em2
;
7663 em2
= btrfs_create_dio_extent(inode
, start
, len
,
7664 orig_start
, block_start
,
7665 len
, orig_block_len
,
7667 btrfs_dec_nocow_writers(fs_info
, block_start
);
7668 if (type
== BTRFS_ORDERED_PREALLOC
) {
7669 free_extent_map(em
);
7673 if (em2
&& IS_ERR(em2
)) {
7678 * For inode marked NODATACOW or extent marked PREALLOC,
7679 * use the existing or preallocated extent, so does not
7680 * need to adjust btrfs_space_info's bytes_may_use.
7682 btrfs_free_reserved_data_space_noquota(inode
, start
,
7688 /* this will cow the extent */
7689 len
= bh_result
->b_size
;
7690 free_extent_map(em
);
7691 *map
= em
= btrfs_new_extent_direct(inode
, start
, len
);
7697 len
= min(len
, em
->len
- (start
- em
->start
));
7700 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
7702 bh_result
->b_size
= len
;
7703 bh_result
->b_bdev
= em
->bdev
;
7704 set_buffer_mapped(bh_result
);
7706 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7707 set_buffer_new(bh_result
);
7710 * Need to update the i_size under the extent lock so buffered
7711 * readers will get the updated i_size when we unlock.
7713 if (!dio_data
->overwrite
&& start
+ len
> i_size_read(inode
))
7714 i_size_write(inode
, start
+ len
);
7716 WARN_ON(dio_data
->reserve
< len
);
7717 dio_data
->reserve
-= len
;
7718 dio_data
->unsubmitted_oe_range_end
= start
+ len
;
7719 current
->journal_info
= dio_data
;
7724 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
7725 struct buffer_head
*bh_result
, int create
)
7727 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7728 struct extent_map
*em
;
7729 struct extent_state
*cached_state
= NULL
;
7730 struct btrfs_dio_data
*dio_data
= NULL
;
7731 u64 start
= iblock
<< inode
->i_blkbits
;
7732 u64 lockstart
, lockend
;
7733 u64 len
= bh_result
->b_size
;
7737 len
= min_t(u64
, len
, fs_info
->sectorsize
);
7740 lockend
= start
+ len
- 1;
7742 if (current
->journal_info
) {
7744 * Need to pull our outstanding extents and set journal_info to NULL so
7745 * that anything that needs to check if there's a transaction doesn't get
7748 dio_data
= current
->journal_info
;
7749 current
->journal_info
= NULL
;
7753 * If this errors out it's because we couldn't invalidate pagecache for
7754 * this range and we need to fallback to buffered.
7756 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
,
7762 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0, start
, len
, 0);
7769 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7770 * io. INLINE is special, and we could probably kludge it in here, but
7771 * it's still buffered so for safety lets just fall back to the generic
7774 * For COMPRESSED we _have_ to read the entire extent in so we can
7775 * decompress it, so there will be buffering required no matter what we
7776 * do, so go ahead and fallback to buffered.
7778 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7779 * to buffered IO. Don't blame me, this is the price we pay for using
7782 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
7783 em
->block_start
== EXTENT_MAP_INLINE
) {
7784 free_extent_map(em
);
7790 ret
= btrfs_get_blocks_direct_write(&em
, bh_result
, inode
,
7791 dio_data
, start
, len
);
7795 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
,
7796 lockend
, &cached_state
);
7798 ret
= btrfs_get_blocks_direct_read(em
, bh_result
, inode
,
7800 /* Can be negative only if we read from a hole */
7803 free_extent_map(em
);
7807 * We need to unlock only the end area that we aren't using.
7808 * The rest is going to be unlocked by the endio routine.
7810 lockstart
= start
+ bh_result
->b_size
;
7811 if (lockstart
< lockend
) {
7812 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
,
7813 lockstart
, lockend
, &cached_state
);
7815 free_extent_state(cached_state
);
7819 free_extent_map(em
);
7824 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7828 current
->journal_info
= dio_data
;
7832 static inline blk_status_t
submit_dio_repair_bio(struct inode
*inode
,
7836 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7839 BUG_ON(bio_op(bio
) == REQ_OP_WRITE
);
7841 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, BTRFS_WQ_ENDIO_DIO_REPAIR
);
7845 ret
= btrfs_map_bio(fs_info
, bio
, mirror_num
, 0);
7850 static int btrfs_check_dio_repairable(struct inode
*inode
,
7851 struct bio
*failed_bio
,
7852 struct io_failure_record
*failrec
,
7855 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7858 num_copies
= btrfs_num_copies(fs_info
, failrec
->logical
, failrec
->len
);
7859 if (num_copies
== 1) {
7861 * we only have a single copy of the data, so don't bother with
7862 * all the retry and error correction code that follows. no
7863 * matter what the error is, it is very likely to persist.
7865 btrfs_debug(fs_info
,
7866 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7867 num_copies
, failrec
->this_mirror
, failed_mirror
);
7871 failrec
->failed_mirror
= failed_mirror
;
7872 failrec
->this_mirror
++;
7873 if (failrec
->this_mirror
== failed_mirror
)
7874 failrec
->this_mirror
++;
7876 if (failrec
->this_mirror
> num_copies
) {
7877 btrfs_debug(fs_info
,
7878 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7879 num_copies
, failrec
->this_mirror
, failed_mirror
);
7886 static blk_status_t
dio_read_error(struct inode
*inode
, struct bio
*failed_bio
,
7887 struct page
*page
, unsigned int pgoff
,
7888 u64 start
, u64 end
, int failed_mirror
,
7889 bio_end_io_t
*repair_endio
, void *repair_arg
)
7891 struct io_failure_record
*failrec
;
7892 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7893 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
7896 unsigned int read_mode
= 0;
7899 blk_status_t status
;
7900 struct bio_vec bvec
;
7902 BUG_ON(bio_op(failed_bio
) == REQ_OP_WRITE
);
7904 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
7906 return errno_to_blk_status(ret
);
7908 ret
= btrfs_check_dio_repairable(inode
, failed_bio
, failrec
,
7911 free_io_failure(failure_tree
, io_tree
, failrec
);
7912 return BLK_STS_IOERR
;
7915 segs
= bio_segments(failed_bio
);
7916 bio_get_first_bvec(failed_bio
, &bvec
);
7918 (bvec
.bv_len
> btrfs_inode_sectorsize(inode
)))
7919 read_mode
|= REQ_FAILFAST_DEV
;
7921 isector
= start
- btrfs_io_bio(failed_bio
)->logical
;
7922 isector
>>= inode
->i_sb
->s_blocksize_bits
;
7923 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
7924 pgoff
, isector
, repair_endio
, repair_arg
);
7925 bio
->bi_opf
= REQ_OP_READ
| read_mode
;
7927 btrfs_debug(BTRFS_I(inode
)->root
->fs_info
,
7928 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7929 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
7931 status
= submit_dio_repair_bio(inode
, bio
, failrec
->this_mirror
);
7933 free_io_failure(failure_tree
, io_tree
, failrec
);
7940 struct btrfs_retry_complete
{
7941 struct completion done
;
7942 struct inode
*inode
;
7947 static void btrfs_retry_endio_nocsum(struct bio
*bio
)
7949 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7950 struct inode
*inode
= done
->inode
;
7951 struct bio_vec
*bvec
;
7952 struct extent_io_tree
*io_tree
, *failure_tree
;
7953 struct bvec_iter_all iter_all
;
7958 ASSERT(bio
->bi_vcnt
== 1);
7959 io_tree
= &BTRFS_I(inode
)->io_tree
;
7960 failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
7961 ASSERT(bio_first_bvec_all(bio
)->bv_len
== btrfs_inode_sectorsize(inode
));
7964 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
7965 bio_for_each_segment_all(bvec
, bio
, iter_all
)
7966 clean_io_failure(BTRFS_I(inode
)->root
->fs_info
, failure_tree
,
7967 io_tree
, done
->start
, bvec
->bv_page
,
7968 btrfs_ino(BTRFS_I(inode
)), 0);
7970 complete(&done
->done
);
7974 static blk_status_t
__btrfs_correct_data_nocsum(struct inode
*inode
,
7975 struct btrfs_io_bio
*io_bio
)
7977 struct btrfs_fs_info
*fs_info
;
7978 struct bio_vec bvec
;
7979 struct bvec_iter iter
;
7980 struct btrfs_retry_complete done
;
7986 blk_status_t err
= BLK_STS_OK
;
7988 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
7989 sectorsize
= fs_info
->sectorsize
;
7991 start
= io_bio
->logical
;
7993 io_bio
->bio
.bi_iter
= io_bio
->iter
;
7995 bio_for_each_segment(bvec
, &io_bio
->bio
, iter
) {
7996 nr_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, bvec
.bv_len
);
7997 pgoff
= bvec
.bv_offset
;
7999 next_block_or_try_again
:
8002 init_completion(&done
.done
);
8004 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
.bv_page
,
8005 pgoff
, start
, start
+ sectorsize
- 1,
8007 btrfs_retry_endio_nocsum
, &done
);
8013 wait_for_completion_io(&done
.done
);
8015 if (!done
.uptodate
) {
8016 /* We might have another mirror, so try again */
8017 goto next_block_or_try_again
;
8021 start
+= sectorsize
;
8025 pgoff
+= sectorsize
;
8026 ASSERT(pgoff
< PAGE_SIZE
);
8027 goto next_block_or_try_again
;
8034 static void btrfs_retry_endio(struct bio
*bio
)
8036 struct btrfs_retry_complete
*done
= bio
->bi_private
;
8037 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8038 struct extent_io_tree
*io_tree
, *failure_tree
;
8039 struct inode
*inode
= done
->inode
;
8040 struct bio_vec
*bvec
;
8044 struct bvec_iter_all iter_all
;
8051 ASSERT(bio
->bi_vcnt
== 1);
8052 ASSERT(bio_first_bvec_all(bio
)->bv_len
== btrfs_inode_sectorsize(done
->inode
));
8054 io_tree
= &BTRFS_I(inode
)->io_tree
;
8055 failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
8057 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
8058 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
8059 ret
= __readpage_endio_check(inode
, io_bio
, i
, bvec
->bv_page
,
8060 bvec
->bv_offset
, done
->start
,
8063 clean_io_failure(BTRFS_I(inode
)->root
->fs_info
,
8064 failure_tree
, io_tree
, done
->start
,
8066 btrfs_ino(BTRFS_I(inode
)),
8073 done
->uptodate
= uptodate
;
8075 complete(&done
->done
);
8079 static blk_status_t
__btrfs_subio_endio_read(struct inode
*inode
,
8080 struct btrfs_io_bio
*io_bio
, blk_status_t err
)
8082 struct btrfs_fs_info
*fs_info
;
8083 struct bio_vec bvec
;
8084 struct bvec_iter iter
;
8085 struct btrfs_retry_complete done
;
8092 bool uptodate
= (err
== 0);
8094 blk_status_t status
;
8096 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
8097 sectorsize
= fs_info
->sectorsize
;
8100 start
= io_bio
->logical
;
8102 io_bio
->bio
.bi_iter
= io_bio
->iter
;
8104 bio_for_each_segment(bvec
, &io_bio
->bio
, iter
) {
8105 nr_sectors
= BTRFS_BYTES_TO_BLKS(fs_info
, bvec
.bv_len
);
8107 pgoff
= bvec
.bv_offset
;
8110 csum_pos
= BTRFS_BYTES_TO_BLKS(fs_info
, offset
);
8111 ret
= __readpage_endio_check(inode
, io_bio
, csum_pos
,
8112 bvec
.bv_page
, pgoff
, start
, sectorsize
);
8119 init_completion(&done
.done
);
8121 status
= dio_read_error(inode
, &io_bio
->bio
, bvec
.bv_page
,
8122 pgoff
, start
, start
+ sectorsize
- 1,
8123 io_bio
->mirror_num
, btrfs_retry_endio
,
8130 wait_for_completion_io(&done
.done
);
8132 if (!done
.uptodate
) {
8133 /* We might have another mirror, so try again */
8137 offset
+= sectorsize
;
8138 start
+= sectorsize
;
8144 pgoff
+= sectorsize
;
8145 ASSERT(pgoff
< PAGE_SIZE
);
8153 static blk_status_t
btrfs_subio_endio_read(struct inode
*inode
,
8154 struct btrfs_io_bio
*io_bio
, blk_status_t err
)
8156 bool skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
8160 return __btrfs_correct_data_nocsum(inode
, io_bio
);
8164 return __btrfs_subio_endio_read(inode
, io_bio
, err
);
8168 static void btrfs_endio_direct_read(struct bio
*bio
)
8170 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8171 struct inode
*inode
= dip
->inode
;
8172 struct bio
*dio_bio
;
8173 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8174 blk_status_t err
= bio
->bi_status
;
8176 if (dip
->flags
& BTRFS_DIO_ORIG_BIO_SUBMITTED
)
8177 err
= btrfs_subio_endio_read(inode
, io_bio
, err
);
8179 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
8180 dip
->logical_offset
+ dip
->bytes
- 1);
8181 dio_bio
= dip
->dio_bio
;
8185 dio_bio
->bi_status
= err
;
8186 dio_end_io(dio_bio
);
8187 btrfs_io_bio_free_csum(io_bio
);
8191 static void __endio_write_update_ordered(struct inode
*inode
,
8192 const u64 offset
, const u64 bytes
,
8193 const bool uptodate
)
8195 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8196 struct btrfs_ordered_extent
*ordered
= NULL
;
8197 struct btrfs_workqueue
*wq
;
8198 btrfs_work_func_t func
;
8199 u64 ordered_offset
= offset
;
8200 u64 ordered_bytes
= bytes
;
8203 if (btrfs_is_free_space_inode(BTRFS_I(inode
))) {
8204 wq
= fs_info
->endio_freespace_worker
;
8205 func
= btrfs_freespace_write_helper
;
8207 wq
= fs_info
->endio_write_workers
;
8208 func
= btrfs_endio_write_helper
;
8211 while (ordered_offset
< offset
+ bytes
) {
8212 last_offset
= ordered_offset
;
8213 if (btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
8217 btrfs_init_work(&ordered
->work
, func
,
8220 btrfs_queue_work(wq
, &ordered
->work
);
8223 * If btrfs_dec_test_ordered_pending does not find any ordered
8224 * extent in the range, we can exit.
8226 if (ordered_offset
== last_offset
)
8229 * Our bio might span multiple ordered extents. In this case
8230 * we keep going until we have accounted the whole dio.
8232 if (ordered_offset
< offset
+ bytes
) {
8233 ordered_bytes
= offset
+ bytes
- ordered_offset
;
8239 static void btrfs_endio_direct_write(struct bio
*bio
)
8241 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8242 struct bio
*dio_bio
= dip
->dio_bio
;
8244 __endio_write_update_ordered(dip
->inode
, dip
->logical_offset
,
8245 dip
->bytes
, !bio
->bi_status
);
8249 dio_bio
->bi_status
= bio
->bi_status
;
8250 dio_end_io(dio_bio
);
8254 static blk_status_t
btrfs_submit_bio_start_direct_io(void *private_data
,
8255 struct bio
*bio
, u64 offset
)
8257 struct inode
*inode
= private_data
;
8259 ret
= btrfs_csum_one_bio(inode
, bio
, offset
, 1);
8260 BUG_ON(ret
); /* -ENOMEM */
8264 static void btrfs_end_dio_bio(struct bio
*bio
)
8266 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8267 blk_status_t err
= bio
->bi_status
;
8270 btrfs_warn(BTRFS_I(dip
->inode
)->root
->fs_info
,
8271 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8272 btrfs_ino(BTRFS_I(dip
->inode
)), bio_op(bio
),
8274 (unsigned long long)bio
->bi_iter
.bi_sector
,
8275 bio
->bi_iter
.bi_size
, err
);
8277 if (dip
->subio_endio
)
8278 err
= dip
->subio_endio(dip
->inode
, btrfs_io_bio(bio
), err
);
8282 * We want to perceive the errors flag being set before
8283 * decrementing the reference count. We don't need a barrier
8284 * since atomic operations with a return value are fully
8285 * ordered as per atomic_t.txt
8290 /* if there are more bios still pending for this dio, just exit */
8291 if (!atomic_dec_and_test(&dip
->pending_bios
))
8295 bio_io_error(dip
->orig_bio
);
8297 dip
->dio_bio
->bi_status
= BLK_STS_OK
;
8298 bio_endio(dip
->orig_bio
);
8304 static inline blk_status_t
btrfs_lookup_and_bind_dio_csum(struct inode
*inode
,
8305 struct btrfs_dio_private
*dip
,
8309 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8310 struct btrfs_io_bio
*orig_io_bio
= btrfs_io_bio(dip
->orig_bio
);
8314 * We load all the csum data we need when we submit
8315 * the first bio to reduce the csum tree search and
8318 if (dip
->logical_offset
== file_offset
) {
8319 ret
= btrfs_lookup_bio_sums_dio(inode
, dip
->orig_bio
,
8325 if (bio
== dip
->orig_bio
)
8328 file_offset
-= dip
->logical_offset
;
8329 file_offset
>>= inode
->i_sb
->s_blocksize_bits
;
8330 io_bio
->csum
= (u8
*)(((u32
*)orig_io_bio
->csum
) + file_offset
);
8335 static inline blk_status_t
btrfs_submit_dio_bio(struct bio
*bio
,
8336 struct inode
*inode
, u64 file_offset
, int async_submit
)
8338 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8339 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8340 bool write
= bio_op(bio
) == REQ_OP_WRITE
;
8343 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8345 async_submit
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
8348 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, BTRFS_WQ_ENDIO_DATA
);
8353 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
8356 if (write
&& async_submit
) {
8357 ret
= btrfs_wq_submit_bio(fs_info
, bio
, 0, 0,
8359 btrfs_submit_bio_start_direct_io
);
8363 * If we aren't doing async submit, calculate the csum of the
8366 ret
= btrfs_csum_one_bio(inode
, bio
, file_offset
, 1);
8370 ret
= btrfs_lookup_and_bind_dio_csum(inode
, dip
, bio
,
8376 ret
= btrfs_map_bio(fs_info
, bio
, 0, 0);
8381 static int btrfs_submit_direct_hook(struct btrfs_dio_private
*dip
)
8383 struct inode
*inode
= dip
->inode
;
8384 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8386 struct bio
*orig_bio
= dip
->orig_bio
;
8387 u64 start_sector
= orig_bio
->bi_iter
.bi_sector
;
8388 u64 file_offset
= dip
->logical_offset
;
8389 int async_submit
= 0;
8391 int clone_offset
= 0;
8394 blk_status_t status
;
8395 struct btrfs_io_geometry geom
;
8397 submit_len
= orig_bio
->bi_iter
.bi_size
;
8398 ret
= btrfs_get_io_geometry(fs_info
, btrfs_op(orig_bio
),
8399 start_sector
<< 9, submit_len
, &geom
);
8403 if (geom
.len
>= submit_len
) {
8405 dip
->flags
|= BTRFS_DIO_ORIG_BIO_SUBMITTED
;
8409 /* async crcs make it difficult to collect full stripe writes. */
8410 if (btrfs_data_alloc_profile(fs_info
) & BTRFS_BLOCK_GROUP_RAID56_MASK
)
8416 ASSERT(geom
.len
<= INT_MAX
);
8417 atomic_inc(&dip
->pending_bios
);
8419 clone_len
= min_t(int, submit_len
, geom
.len
);
8422 * This will never fail as it's passing GPF_NOFS and
8423 * the allocation is backed by btrfs_bioset.
8425 bio
= btrfs_bio_clone_partial(orig_bio
, clone_offset
,
8427 bio
->bi_private
= dip
;
8428 bio
->bi_end_io
= btrfs_end_dio_bio
;
8429 btrfs_io_bio(bio
)->logical
= file_offset
;
8431 ASSERT(submit_len
>= clone_len
);
8432 submit_len
-= clone_len
;
8433 if (submit_len
== 0)
8437 * Increase the count before we submit the bio so we know
8438 * the end IO handler won't happen before we increase the
8439 * count. Otherwise, the dip might get freed before we're
8440 * done setting it up.
8442 atomic_inc(&dip
->pending_bios
);
8444 status
= btrfs_submit_dio_bio(bio
, inode
, file_offset
,
8448 atomic_dec(&dip
->pending_bios
);
8452 clone_offset
+= clone_len
;
8453 start_sector
+= clone_len
>> 9;
8454 file_offset
+= clone_len
;
8456 ret
= btrfs_get_io_geometry(fs_info
, btrfs_op(orig_bio
),
8457 start_sector
<< 9, submit_len
, &geom
);
8460 } while (submit_len
> 0);
8463 status
= btrfs_submit_dio_bio(bio
, inode
, file_offset
, async_submit
);
8471 * Before atomic variable goto zero, we must make sure dip->errors is
8472 * perceived to be set. This ordering is ensured by the fact that an
8473 * atomic operations with a return value are fully ordered as per
8476 if (atomic_dec_and_test(&dip
->pending_bios
))
8477 bio_io_error(dip
->orig_bio
);
8479 /* bio_end_io() will handle error, so we needn't return it */
8483 static void btrfs_submit_direct(struct bio
*dio_bio
, struct inode
*inode
,
8486 struct btrfs_dio_private
*dip
= NULL
;
8487 struct bio
*bio
= NULL
;
8488 struct btrfs_io_bio
*io_bio
;
8489 bool write
= (bio_op(dio_bio
) == REQ_OP_WRITE
);
8492 bio
= btrfs_bio_clone(dio_bio
);
8494 dip
= kzalloc(sizeof(*dip
), GFP_NOFS
);
8500 dip
->private = dio_bio
->bi_private
;
8502 dip
->logical_offset
= file_offset
;
8503 dip
->bytes
= dio_bio
->bi_iter
.bi_size
;
8504 dip
->disk_bytenr
= (u64
)dio_bio
->bi_iter
.bi_sector
<< 9;
8505 bio
->bi_private
= dip
;
8506 dip
->orig_bio
= bio
;
8507 dip
->dio_bio
= dio_bio
;
8508 atomic_set(&dip
->pending_bios
, 0);
8509 io_bio
= btrfs_io_bio(bio
);
8510 io_bio
->logical
= file_offset
;
8513 bio
->bi_end_io
= btrfs_endio_direct_write
;
8515 bio
->bi_end_io
= btrfs_endio_direct_read
;
8516 dip
->subio_endio
= btrfs_subio_endio_read
;
8520 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8521 * even if we fail to submit a bio, because in such case we do the
8522 * corresponding error handling below and it must not be done a second
8523 * time by btrfs_direct_IO().
8526 struct btrfs_dio_data
*dio_data
= current
->journal_info
;
8528 dio_data
->unsubmitted_oe_range_end
= dip
->logical_offset
+
8530 dio_data
->unsubmitted_oe_range_start
=
8531 dio_data
->unsubmitted_oe_range_end
;
8534 ret
= btrfs_submit_direct_hook(dip
);
8538 btrfs_io_bio_free_csum(io_bio
);
8542 * If we arrived here it means either we failed to submit the dip
8543 * or we either failed to clone the dio_bio or failed to allocate the
8544 * dip. If we cloned the dio_bio and allocated the dip, we can just
8545 * call bio_endio against our io_bio so that we get proper resource
8546 * cleanup if we fail to submit the dip, otherwise, we must do the
8547 * same as btrfs_endio_direct_[write|read] because we can't call these
8548 * callbacks - they require an allocated dip and a clone of dio_bio.
8553 * The end io callbacks free our dip, do the final put on bio
8554 * and all the cleanup and final put for dio_bio (through
8561 __endio_write_update_ordered(inode
,
8563 dio_bio
->bi_iter
.bi_size
,
8566 unlock_extent(&BTRFS_I(inode
)->io_tree
, file_offset
,
8567 file_offset
+ dio_bio
->bi_iter
.bi_size
- 1);
8569 dio_bio
->bi_status
= BLK_STS_IOERR
;
8571 * Releases and cleans up our dio_bio, no need to bio_put()
8572 * nor bio_endio()/bio_io_error() against dio_bio.
8574 dio_end_io(dio_bio
);
8581 static ssize_t
check_direct_IO(struct btrfs_fs_info
*fs_info
,
8582 const struct iov_iter
*iter
, loff_t offset
)
8586 unsigned int blocksize_mask
= fs_info
->sectorsize
- 1;
8587 ssize_t retval
= -EINVAL
;
8589 if (offset
& blocksize_mask
)
8592 if (iov_iter_alignment(iter
) & blocksize_mask
)
8595 /* If this is a write we don't need to check anymore */
8596 if (iov_iter_rw(iter
) != READ
|| !iter_is_iovec(iter
))
8599 * Check to make sure we don't have duplicate iov_base's in this
8600 * iovec, if so return EINVAL, otherwise we'll get csum errors
8601 * when reading back.
8603 for (seg
= 0; seg
< iter
->nr_segs
; seg
++) {
8604 for (i
= seg
+ 1; i
< iter
->nr_segs
; i
++) {
8605 if (iter
->iov
[seg
].iov_base
== iter
->iov
[i
].iov_base
)
8614 static ssize_t
btrfs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
8616 struct file
*file
= iocb
->ki_filp
;
8617 struct inode
*inode
= file
->f_mapping
->host
;
8618 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8619 struct btrfs_dio_data dio_data
= { 0 };
8620 struct extent_changeset
*data_reserved
= NULL
;
8621 loff_t offset
= iocb
->ki_pos
;
8625 bool relock
= false;
8628 if (check_direct_IO(fs_info
, iter
, offset
))
8631 inode_dio_begin(inode
);
8634 * The generic stuff only does filemap_write_and_wait_range, which
8635 * isn't enough if we've written compressed pages to this area, so
8636 * we need to flush the dirty pages again to make absolutely sure
8637 * that any outstanding dirty pages are on disk.
8639 count
= iov_iter_count(iter
);
8640 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
8641 &BTRFS_I(inode
)->runtime_flags
))
8642 filemap_fdatawrite_range(inode
->i_mapping
, offset
,
8643 offset
+ count
- 1);
8645 if (iov_iter_rw(iter
) == WRITE
) {
8647 * If the write DIO is beyond the EOF, we need update
8648 * the isize, but it is protected by i_mutex. So we can
8649 * not unlock the i_mutex at this case.
8651 if (offset
+ count
<= inode
->i_size
) {
8652 dio_data
.overwrite
= 1;
8653 inode_unlock(inode
);
8655 } else if (iocb
->ki_flags
& IOCB_NOWAIT
) {
8659 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
,
8665 * We need to know how many extents we reserved so that we can
8666 * do the accounting properly if we go over the number we
8667 * originally calculated. Abuse current->journal_info for this.
8669 dio_data
.reserve
= round_up(count
,
8670 fs_info
->sectorsize
);
8671 dio_data
.unsubmitted_oe_range_start
= (u64
)offset
;
8672 dio_data
.unsubmitted_oe_range_end
= (u64
)offset
;
8673 current
->journal_info
= &dio_data
;
8674 down_read(&BTRFS_I(inode
)->dio_sem
);
8675 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK
,
8676 &BTRFS_I(inode
)->runtime_flags
)) {
8677 inode_dio_end(inode
);
8678 flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
8682 ret
= __blockdev_direct_IO(iocb
, inode
,
8683 fs_info
->fs_devices
->latest_bdev
,
8684 iter
, btrfs_get_blocks_direct
, NULL
,
8685 btrfs_submit_direct
, flags
);
8686 if (iov_iter_rw(iter
) == WRITE
) {
8687 up_read(&BTRFS_I(inode
)->dio_sem
);
8688 current
->journal_info
= NULL
;
8689 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
8690 if (dio_data
.reserve
)
8691 btrfs_delalloc_release_space(inode
, data_reserved
,
8692 offset
, dio_data
.reserve
, true);
8694 * On error we might have left some ordered extents
8695 * without submitting corresponding bios for them, so
8696 * cleanup them up to avoid other tasks getting them
8697 * and waiting for them to complete forever.
8699 if (dio_data
.unsubmitted_oe_range_start
<
8700 dio_data
.unsubmitted_oe_range_end
)
8701 __endio_write_update_ordered(inode
,
8702 dio_data
.unsubmitted_oe_range_start
,
8703 dio_data
.unsubmitted_oe_range_end
-
8704 dio_data
.unsubmitted_oe_range_start
,
8706 } else if (ret
>= 0 && (size_t)ret
< count
)
8707 btrfs_delalloc_release_space(inode
, data_reserved
,
8708 offset
, count
- (size_t)ret
, true);
8709 btrfs_delalloc_release_extents(BTRFS_I(inode
), count
, false);
8713 inode_dio_end(inode
);
8717 extent_changeset_free(data_reserved
);
8721 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8723 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
8724 __u64 start
, __u64 len
)
8728 ret
= fiemap_check_flags(fieinfo
, BTRFS_FIEMAP_FLAGS
);
8732 return extent_fiemap(inode
, fieinfo
, start
, len
);
8735 int btrfs_readpage(struct file
*file
, struct page
*page
)
8737 struct extent_io_tree
*tree
;
8738 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8739 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
8742 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
8744 struct inode
*inode
= page
->mapping
->host
;
8747 if (current
->flags
& PF_MEMALLOC
) {
8748 redirty_page_for_writepage(wbc
, page
);
8754 * If we are under memory pressure we will call this directly from the
8755 * VM, we need to make sure we have the inode referenced for the ordered
8756 * extent. If not just return like we didn't do anything.
8758 if (!igrab(inode
)) {
8759 redirty_page_for_writepage(wbc
, page
);
8760 return AOP_WRITEPAGE_ACTIVATE
;
8762 ret
= extent_write_full_page(page
, wbc
);
8763 btrfs_add_delayed_iput(inode
);
8767 static int btrfs_writepages(struct address_space
*mapping
,
8768 struct writeback_control
*wbc
)
8770 return extent_writepages(mapping
, wbc
);
8774 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
8775 struct list_head
*pages
, unsigned nr_pages
)
8777 return extent_readpages(mapping
, pages
, nr_pages
);
8780 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8782 int ret
= try_release_extent_mapping(page
, gfp_flags
);
8784 ClearPagePrivate(page
);
8785 set_page_private(page
, 0);
8791 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8793 if (PageWriteback(page
) || PageDirty(page
))
8795 return __btrfs_releasepage(page
, gfp_flags
);
8798 static void btrfs_invalidatepage(struct page
*page
, unsigned int offset
,
8799 unsigned int length
)
8801 struct inode
*inode
= page
->mapping
->host
;
8802 struct extent_io_tree
*tree
;
8803 struct btrfs_ordered_extent
*ordered
;
8804 struct extent_state
*cached_state
= NULL
;
8805 u64 page_start
= page_offset(page
);
8806 u64 page_end
= page_start
+ PAGE_SIZE
- 1;
8809 int inode_evicting
= inode
->i_state
& I_FREEING
;
8812 * we have the page locked, so new writeback can't start,
8813 * and the dirty bit won't be cleared while we are here.
8815 * Wait for IO on this page so that we can safely clear
8816 * the PagePrivate2 bit and do ordered accounting
8818 wait_on_page_writeback(page
);
8820 tree
= &BTRFS_I(inode
)->io_tree
;
8822 btrfs_releasepage(page
, GFP_NOFS
);
8826 if (!inode_evicting
)
8827 lock_extent_bits(tree
, page_start
, page_end
, &cached_state
);
8830 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), start
,
8831 page_end
- start
+ 1);
8833 end
= min(page_end
, ordered
->file_offset
+ ordered
->len
- 1);
8835 * IO on this page will never be started, so we need
8836 * to account for any ordered extents now
8838 if (!inode_evicting
)
8839 clear_extent_bit(tree
, start
, end
,
8840 EXTENT_DELALLOC
| EXTENT_DELALLOC_NEW
|
8841 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
8842 EXTENT_DEFRAG
, 1, 0, &cached_state
);
8844 * whoever cleared the private bit is responsible
8845 * for the finish_ordered_io
8847 if (TestClearPagePrivate2(page
)) {
8848 struct btrfs_ordered_inode_tree
*tree
;
8851 tree
= &BTRFS_I(inode
)->ordered_tree
;
8853 spin_lock_irq(&tree
->lock
);
8854 set_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
);
8855 new_len
= start
- ordered
->file_offset
;
8856 if (new_len
< ordered
->truncated_len
)
8857 ordered
->truncated_len
= new_len
;
8858 spin_unlock_irq(&tree
->lock
);
8860 if (btrfs_dec_test_ordered_pending(inode
, &ordered
,
8862 end
- start
+ 1, 1))
8863 btrfs_finish_ordered_io(ordered
);
8865 btrfs_put_ordered_extent(ordered
);
8866 if (!inode_evicting
) {
8867 cached_state
= NULL
;
8868 lock_extent_bits(tree
, start
, end
,
8873 if (start
< page_end
)
8878 * Qgroup reserved space handler
8879 * Page here will be either
8880 * 1) Already written to disk
8881 * In this case, its reserved space is released from data rsv map
8882 * and will be freed by delayed_ref handler finally.
8883 * So even we call qgroup_free_data(), it won't decrease reserved
8885 * 2) Not written to disk
8886 * This means the reserved space should be freed here. However,
8887 * if a truncate invalidates the page (by clearing PageDirty)
8888 * and the page is accounted for while allocating extent
8889 * in btrfs_check_data_free_space() we let delayed_ref to
8890 * free the entire extent.
8892 if (PageDirty(page
))
8893 btrfs_qgroup_free_data(inode
, NULL
, page_start
, PAGE_SIZE
);
8894 if (!inode_evicting
) {
8895 clear_extent_bit(tree
, page_start
, page_end
, EXTENT_LOCKED
|
8896 EXTENT_DELALLOC
| EXTENT_DELALLOC_NEW
|
8897 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
, 1, 1,
8900 __btrfs_releasepage(page
, GFP_NOFS
);
8903 ClearPageChecked(page
);
8904 if (PagePrivate(page
)) {
8905 ClearPagePrivate(page
);
8906 set_page_private(page
, 0);
8912 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8913 * called from a page fault handler when a page is first dirtied. Hence we must
8914 * be careful to check for EOF conditions here. We set the page up correctly
8915 * for a written page which means we get ENOSPC checking when writing into
8916 * holes and correct delalloc and unwritten extent mapping on filesystems that
8917 * support these features.
8919 * We are not allowed to take the i_mutex here so we have to play games to
8920 * protect against truncate races as the page could now be beyond EOF. Because
8921 * truncate_setsize() writes the inode size before removing pages, once we have
8922 * the page lock we can determine safely if the page is beyond EOF. If it is not
8923 * beyond EOF, then the page is guaranteed safe against truncation until we
8926 vm_fault_t
btrfs_page_mkwrite(struct vm_fault
*vmf
)
8928 struct page
*page
= vmf
->page
;
8929 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
8930 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8931 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
8932 struct btrfs_ordered_extent
*ordered
;
8933 struct extent_state
*cached_state
= NULL
;
8934 struct extent_changeset
*data_reserved
= NULL
;
8936 unsigned long zero_start
;
8946 reserved_space
= PAGE_SIZE
;
8948 sb_start_pagefault(inode
->i_sb
);
8949 page_start
= page_offset(page
);
8950 page_end
= page_start
+ PAGE_SIZE
- 1;
8954 * Reserving delalloc space after obtaining the page lock can lead to
8955 * deadlock. For example, if a dirty page is locked by this function
8956 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8957 * dirty page write out, then the btrfs_writepage() function could
8958 * end up waiting indefinitely to get a lock on the page currently
8959 * being processed by btrfs_page_mkwrite() function.
8961 ret2
= btrfs_delalloc_reserve_space(inode
, &data_reserved
, page_start
,
8964 ret2
= file_update_time(vmf
->vma
->vm_file
);
8968 ret
= vmf_error(ret2
);
8974 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
8977 size
= i_size_read(inode
);
8979 if ((page
->mapping
!= inode
->i_mapping
) ||
8980 (page_start
>= size
)) {
8981 /* page got truncated out from underneath us */
8984 wait_on_page_writeback(page
);
8986 lock_extent_bits(io_tree
, page_start
, page_end
, &cached_state
);
8987 set_page_extent_mapped(page
);
8990 * we can't set the delalloc bits if there are pending ordered
8991 * extents. Drop our locks and wait for them to finish
8993 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), page_start
,
8996 unlock_extent_cached(io_tree
, page_start
, page_end
,
8999 btrfs_start_ordered_extent(inode
, ordered
, 1);
9000 btrfs_put_ordered_extent(ordered
);
9004 if (page
->index
== ((size
- 1) >> PAGE_SHIFT
)) {
9005 reserved_space
= round_up(size
- page_start
,
9006 fs_info
->sectorsize
);
9007 if (reserved_space
< PAGE_SIZE
) {
9008 end
= page_start
+ reserved_space
- 1;
9009 btrfs_delalloc_release_space(inode
, data_reserved
,
9010 page_start
, PAGE_SIZE
- reserved_space
,
9016 * page_mkwrite gets called when the page is firstly dirtied after it's
9017 * faulted in, but write(2) could also dirty a page and set delalloc
9018 * bits, thus in this case for space account reason, we still need to
9019 * clear any delalloc bits within this page range since we have to
9020 * reserve data&meta space before lock_page() (see above comments).
9022 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, end
,
9023 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
9024 EXTENT_DEFRAG
, 0, 0, &cached_state
);
9026 ret2
= btrfs_set_extent_delalloc(inode
, page_start
, end
, 0,
9029 unlock_extent_cached(io_tree
, page_start
, page_end
,
9031 ret
= VM_FAULT_SIGBUS
;
9036 /* page is wholly or partially inside EOF */
9037 if (page_start
+ PAGE_SIZE
> size
)
9038 zero_start
= offset_in_page(size
);
9040 zero_start
= PAGE_SIZE
;
9042 if (zero_start
!= PAGE_SIZE
) {
9044 memset(kaddr
+ zero_start
, 0, PAGE_SIZE
- zero_start
);
9045 flush_dcache_page(page
);
9048 ClearPageChecked(page
);
9049 set_page_dirty(page
);
9050 SetPageUptodate(page
);
9052 BTRFS_I(inode
)->last_trans
= fs_info
->generation
;
9053 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
9054 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
9056 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
);
9059 btrfs_delalloc_release_extents(BTRFS_I(inode
), PAGE_SIZE
, true);
9060 sb_end_pagefault(inode
->i_sb
);
9061 extent_changeset_free(data_reserved
);
9062 return VM_FAULT_LOCKED
;
9068 btrfs_delalloc_release_extents(BTRFS_I(inode
), PAGE_SIZE
, (ret
!= 0));
9069 btrfs_delalloc_release_space(inode
, data_reserved
, page_start
,
9070 reserved_space
, (ret
!= 0));
9072 sb_end_pagefault(inode
->i_sb
);
9073 extent_changeset_free(data_reserved
);
9077 static int btrfs_truncate(struct inode
*inode
, bool skip_writeback
)
9079 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
9080 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9081 struct btrfs_block_rsv
*rsv
;
9083 struct btrfs_trans_handle
*trans
;
9084 u64 mask
= fs_info
->sectorsize
- 1;
9085 u64 min_size
= btrfs_calc_metadata_size(fs_info
, 1);
9087 if (!skip_writeback
) {
9088 ret
= btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
),
9095 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
9096 * things going on here:
9098 * 1) We need to reserve space to update our inode.
9100 * 2) We need to have something to cache all the space that is going to
9101 * be free'd up by the truncate operation, but also have some slack
9102 * space reserved in case it uses space during the truncate (thank you
9103 * very much snapshotting).
9105 * And we need these to be separate. The fact is we can use a lot of
9106 * space doing the truncate, and we have no earthly idea how much space
9107 * we will use, so we need the truncate reservation to be separate so it
9108 * doesn't end up using space reserved for updating the inode. We also
9109 * need to be able to stop the transaction and start a new one, which
9110 * means we need to be able to update the inode several times, and we
9111 * have no idea of knowing how many times that will be, so we can't just
9112 * reserve 1 item for the entirety of the operation, so that has to be
9113 * done separately as well.
9115 * So that leaves us with
9117 * 1) rsv - for the truncate reservation, which we will steal from the
9118 * transaction reservation.
9119 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
9120 * updating the inode.
9122 rsv
= btrfs_alloc_block_rsv(fs_info
, BTRFS_BLOCK_RSV_TEMP
);
9125 rsv
->size
= min_size
;
9129 * 1 for the truncate slack space
9130 * 1 for updating the inode.
9132 trans
= btrfs_start_transaction(root
, 2);
9133 if (IS_ERR(trans
)) {
9134 ret
= PTR_ERR(trans
);
9138 /* Migrate the slack space for the truncate to our reserve */
9139 ret
= btrfs_block_rsv_migrate(&fs_info
->trans_block_rsv
, rsv
,
9144 * So if we truncate and then write and fsync we normally would just
9145 * write the extents that changed, which is a problem if we need to
9146 * first truncate that entire inode. So set this flag so we write out
9147 * all of the extents in the inode to the sync log so we're completely
9150 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
9151 trans
->block_rsv
= rsv
;
9154 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
9156 BTRFS_EXTENT_DATA_KEY
);
9157 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
9158 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
)
9161 ret
= btrfs_update_inode(trans
, root
, inode
);
9165 btrfs_end_transaction(trans
);
9166 btrfs_btree_balance_dirty(fs_info
);
9168 trans
= btrfs_start_transaction(root
, 2);
9169 if (IS_ERR(trans
)) {
9170 ret
= PTR_ERR(trans
);
9175 btrfs_block_rsv_release(fs_info
, rsv
, -1);
9176 ret
= btrfs_block_rsv_migrate(&fs_info
->trans_block_rsv
,
9177 rsv
, min_size
, false);
9178 BUG_ON(ret
); /* shouldn't happen */
9179 trans
->block_rsv
= rsv
;
9183 * We can't call btrfs_truncate_block inside a trans handle as we could
9184 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9185 * we've truncated everything except the last little bit, and can do
9186 * btrfs_truncate_block and then update the disk_i_size.
9188 if (ret
== NEED_TRUNCATE_BLOCK
) {
9189 btrfs_end_transaction(trans
);
9190 btrfs_btree_balance_dirty(fs_info
);
9192 ret
= btrfs_truncate_block(inode
, inode
->i_size
, 0, 0);
9195 trans
= btrfs_start_transaction(root
, 1);
9196 if (IS_ERR(trans
)) {
9197 ret
= PTR_ERR(trans
);
9200 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
9206 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
9207 ret2
= btrfs_update_inode(trans
, root
, inode
);
9211 ret2
= btrfs_end_transaction(trans
);
9214 btrfs_btree_balance_dirty(fs_info
);
9217 btrfs_free_block_rsv(fs_info
, rsv
);
9223 * create a new subvolume directory/inode (helper for the ioctl).
9225 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
9226 struct btrfs_root
*new_root
,
9227 struct btrfs_root
*parent_root
,
9230 struct inode
*inode
;
9234 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
9235 new_dirid
, new_dirid
,
9236 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
9239 return PTR_ERR(inode
);
9240 inode
->i_op
= &btrfs_dir_inode_operations
;
9241 inode
->i_fop
= &btrfs_dir_file_operations
;
9243 set_nlink(inode
, 1);
9244 btrfs_i_size_write(BTRFS_I(inode
), 0);
9245 unlock_new_inode(inode
);
9247 err
= btrfs_subvol_inherit_props(trans
, new_root
, parent_root
);
9249 btrfs_err(new_root
->fs_info
,
9250 "error inheriting subvolume %llu properties: %d",
9251 new_root
->root_key
.objectid
, err
);
9253 err
= btrfs_update_inode(trans
, new_root
, inode
);
9259 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
9261 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
9262 struct btrfs_inode
*ei
;
9263 struct inode
*inode
;
9265 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_KERNEL
);
9272 ei
->last_sub_trans
= 0;
9273 ei
->logged_trans
= 0;
9274 ei
->delalloc_bytes
= 0;
9275 ei
->new_delalloc_bytes
= 0;
9276 ei
->defrag_bytes
= 0;
9277 ei
->disk_i_size
= 0;
9280 ei
->index_cnt
= (u64
)-1;
9282 ei
->last_unlink_trans
= 0;
9283 ei
->last_log_commit
= 0;
9285 spin_lock_init(&ei
->lock
);
9286 ei
->outstanding_extents
= 0;
9287 if (sb
->s_magic
!= BTRFS_TEST_MAGIC
)
9288 btrfs_init_metadata_block_rsv(fs_info
, &ei
->block_rsv
,
9289 BTRFS_BLOCK_RSV_DELALLOC
);
9290 ei
->runtime_flags
= 0;
9291 ei
->prop_compress
= BTRFS_COMPRESS_NONE
;
9292 ei
->defrag_compress
= BTRFS_COMPRESS_NONE
;
9294 ei
->delayed_node
= NULL
;
9296 ei
->i_otime
.tv_sec
= 0;
9297 ei
->i_otime
.tv_nsec
= 0;
9299 inode
= &ei
->vfs_inode
;
9300 extent_map_tree_init(&ei
->extent_tree
);
9301 extent_io_tree_init(fs_info
, &ei
->io_tree
, IO_TREE_INODE_IO
, inode
);
9302 extent_io_tree_init(fs_info
, &ei
->io_failure_tree
,
9303 IO_TREE_INODE_IO_FAILURE
, inode
);
9304 ei
->io_tree
.track_uptodate
= true;
9305 ei
->io_failure_tree
.track_uptodate
= true;
9306 atomic_set(&ei
->sync_writers
, 0);
9307 mutex_init(&ei
->log_mutex
);
9308 mutex_init(&ei
->delalloc_mutex
);
9309 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
9310 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
9311 INIT_LIST_HEAD(&ei
->delayed_iput
);
9312 RB_CLEAR_NODE(&ei
->rb_node
);
9313 init_rwsem(&ei
->dio_sem
);
9318 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9319 void btrfs_test_destroy_inode(struct inode
*inode
)
9321 btrfs_drop_extent_cache(BTRFS_I(inode
), 0, (u64
)-1, 0);
9322 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9326 void btrfs_free_inode(struct inode
*inode
)
9328 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9331 void btrfs_destroy_inode(struct inode
*inode
)
9333 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
9334 struct btrfs_ordered_extent
*ordered
;
9335 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9337 WARN_ON(!hlist_empty(&inode
->i_dentry
));
9338 WARN_ON(inode
->i_data
.nrpages
);
9339 WARN_ON(BTRFS_I(inode
)->block_rsv
.reserved
);
9340 WARN_ON(BTRFS_I(inode
)->block_rsv
.size
);
9341 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
9342 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
9343 WARN_ON(BTRFS_I(inode
)->new_delalloc_bytes
);
9344 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
9345 WARN_ON(BTRFS_I(inode
)->defrag_bytes
);
9348 * This can happen where we create an inode, but somebody else also
9349 * created the same inode and we need to destroy the one we already
9356 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
9361 "found ordered extent %llu %llu on inode cleanup",
9362 ordered
->file_offset
, ordered
->len
);
9363 btrfs_remove_ordered_extent(inode
, ordered
);
9364 btrfs_put_ordered_extent(ordered
);
9365 btrfs_put_ordered_extent(ordered
);
9368 btrfs_qgroup_check_reserved_leak(inode
);
9369 inode_tree_del(inode
);
9370 btrfs_drop_extent_cache(BTRFS_I(inode
), 0, (u64
)-1, 0);
9373 int btrfs_drop_inode(struct inode
*inode
)
9375 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9380 /* the snap/subvol tree is on deleting */
9381 if (btrfs_root_refs(&root
->root_item
) == 0)
9384 return generic_drop_inode(inode
);
9387 static void init_once(void *foo
)
9389 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
9391 inode_init_once(&ei
->vfs_inode
);
9394 void __cold
btrfs_destroy_cachep(void)
9397 * Make sure all delayed rcu free inodes are flushed before we
9401 kmem_cache_destroy(btrfs_inode_cachep
);
9402 kmem_cache_destroy(btrfs_trans_handle_cachep
);
9403 kmem_cache_destroy(btrfs_path_cachep
);
9404 kmem_cache_destroy(btrfs_free_space_cachep
);
9405 kmem_cache_destroy(btrfs_free_space_bitmap_cachep
);
9408 int __init
btrfs_init_cachep(void)
9410 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
9411 sizeof(struct btrfs_inode
), 0,
9412 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
| SLAB_ACCOUNT
,
9414 if (!btrfs_inode_cachep
)
9417 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
9418 sizeof(struct btrfs_trans_handle
), 0,
9419 SLAB_TEMPORARY
| SLAB_MEM_SPREAD
, NULL
);
9420 if (!btrfs_trans_handle_cachep
)
9423 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
9424 sizeof(struct btrfs_path
), 0,
9425 SLAB_MEM_SPREAD
, NULL
);
9426 if (!btrfs_path_cachep
)
9429 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
9430 sizeof(struct btrfs_free_space
), 0,
9431 SLAB_MEM_SPREAD
, NULL
);
9432 if (!btrfs_free_space_cachep
)
9435 btrfs_free_space_bitmap_cachep
= kmem_cache_create("btrfs_free_space_bitmap",
9436 PAGE_SIZE
, PAGE_SIZE
,
9437 SLAB_RED_ZONE
, NULL
);
9438 if (!btrfs_free_space_bitmap_cachep
)
9443 btrfs_destroy_cachep();
9447 static int btrfs_getattr(const struct path
*path
, struct kstat
*stat
,
9448 u32 request_mask
, unsigned int flags
)
9451 struct inode
*inode
= d_inode(path
->dentry
);
9452 u32 blocksize
= inode
->i_sb
->s_blocksize
;
9453 u32 bi_flags
= BTRFS_I(inode
)->flags
;
9455 stat
->result_mask
|= STATX_BTIME
;
9456 stat
->btime
.tv_sec
= BTRFS_I(inode
)->i_otime
.tv_sec
;
9457 stat
->btime
.tv_nsec
= BTRFS_I(inode
)->i_otime
.tv_nsec
;
9458 if (bi_flags
& BTRFS_INODE_APPEND
)
9459 stat
->attributes
|= STATX_ATTR_APPEND
;
9460 if (bi_flags
& BTRFS_INODE_COMPRESS
)
9461 stat
->attributes
|= STATX_ATTR_COMPRESSED
;
9462 if (bi_flags
& BTRFS_INODE_IMMUTABLE
)
9463 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
9464 if (bi_flags
& BTRFS_INODE_NODUMP
)
9465 stat
->attributes
|= STATX_ATTR_NODUMP
;
9467 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
9468 STATX_ATTR_COMPRESSED
|
9469 STATX_ATTR_IMMUTABLE
|
9472 generic_fillattr(inode
, stat
);
9473 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
9475 spin_lock(&BTRFS_I(inode
)->lock
);
9476 delalloc_bytes
= BTRFS_I(inode
)->new_delalloc_bytes
;
9477 spin_unlock(&BTRFS_I(inode
)->lock
);
9478 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
9479 ALIGN(delalloc_bytes
, blocksize
)) >> 9;
9483 static int btrfs_rename_exchange(struct inode
*old_dir
,
9484 struct dentry
*old_dentry
,
9485 struct inode
*new_dir
,
9486 struct dentry
*new_dentry
)
9488 struct btrfs_fs_info
*fs_info
= btrfs_sb(old_dir
->i_sb
);
9489 struct btrfs_trans_handle
*trans
;
9490 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9491 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9492 struct inode
*new_inode
= new_dentry
->d_inode
;
9493 struct inode
*old_inode
= old_dentry
->d_inode
;
9494 struct timespec64 ctime
= current_time(old_inode
);
9495 struct dentry
*parent
;
9496 u64 old_ino
= btrfs_ino(BTRFS_I(old_inode
));
9497 u64 new_ino
= btrfs_ino(BTRFS_I(new_inode
));
9502 bool root_log_pinned
= false;
9503 bool dest_log_pinned
= false;
9504 struct btrfs_log_ctx ctx_root
;
9505 struct btrfs_log_ctx ctx_dest
;
9506 bool sync_log_root
= false;
9507 bool sync_log_dest
= false;
9508 bool commit_transaction
= false;
9510 /* we only allow rename subvolume link between subvolumes */
9511 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9514 btrfs_init_log_ctx(&ctx_root
, old_inode
);
9515 btrfs_init_log_ctx(&ctx_dest
, new_inode
);
9517 /* close the race window with snapshot create/destroy ioctl */
9518 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9519 down_read(&fs_info
->subvol_sem
);
9520 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9521 down_read(&fs_info
->subvol_sem
);
9524 * We want to reserve the absolute worst case amount of items. So if
9525 * both inodes are subvols and we need to unlink them then that would
9526 * require 4 item modifications, but if they are both normal inodes it
9527 * would require 5 item modifications, so we'll assume their normal
9528 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9529 * should cover the worst case number of items we'll modify.
9531 trans
= btrfs_start_transaction(root
, 12);
9532 if (IS_ERR(trans
)) {
9533 ret
= PTR_ERR(trans
);
9538 * We need to find a free sequence number both in the source and
9539 * in the destination directory for the exchange.
9541 ret
= btrfs_set_inode_index(BTRFS_I(new_dir
), &old_idx
);
9544 ret
= btrfs_set_inode_index(BTRFS_I(old_dir
), &new_idx
);
9548 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9549 BTRFS_I(new_inode
)->dir_index
= 0ULL;
9551 /* Reference for the source. */
9552 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9553 /* force full log commit if subvolume involved. */
9554 btrfs_set_log_full_commit(trans
);
9556 btrfs_pin_log_trans(root
);
9557 root_log_pinned
= true;
9558 ret
= btrfs_insert_inode_ref(trans
, dest
,
9559 new_dentry
->d_name
.name
,
9560 new_dentry
->d_name
.len
,
9562 btrfs_ino(BTRFS_I(new_dir
)),
9568 /* And now for the dest. */
9569 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9570 /* force full log commit if subvolume involved. */
9571 btrfs_set_log_full_commit(trans
);
9573 btrfs_pin_log_trans(dest
);
9574 dest_log_pinned
= true;
9575 ret
= btrfs_insert_inode_ref(trans
, root
,
9576 old_dentry
->d_name
.name
,
9577 old_dentry
->d_name
.len
,
9579 btrfs_ino(BTRFS_I(old_dir
)),
9585 /* Update inode version and ctime/mtime. */
9586 inode_inc_iversion(old_dir
);
9587 inode_inc_iversion(new_dir
);
9588 inode_inc_iversion(old_inode
);
9589 inode_inc_iversion(new_inode
);
9590 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
9591 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
9592 old_inode
->i_ctime
= ctime
;
9593 new_inode
->i_ctime
= ctime
;
9595 if (old_dentry
->d_parent
!= new_dentry
->d_parent
) {
9596 btrfs_record_unlink_dir(trans
, BTRFS_I(old_dir
),
9597 BTRFS_I(old_inode
), 1);
9598 btrfs_record_unlink_dir(trans
, BTRFS_I(new_dir
),
9599 BTRFS_I(new_inode
), 1);
9602 /* src is a subvolume */
9603 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9604 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
9605 ret
= btrfs_unlink_subvol(trans
, old_dir
, root_objectid
,
9606 old_dentry
->d_name
.name
,
9607 old_dentry
->d_name
.len
);
9608 } else { /* src is an inode */
9609 ret
= __btrfs_unlink_inode(trans
, root
, BTRFS_I(old_dir
),
9610 BTRFS_I(old_dentry
->d_inode
),
9611 old_dentry
->d_name
.name
,
9612 old_dentry
->d_name
.len
);
9614 ret
= btrfs_update_inode(trans
, root
, old_inode
);
9617 btrfs_abort_transaction(trans
, ret
);
9621 /* dest is a subvolume */
9622 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9623 root_objectid
= BTRFS_I(new_inode
)->root
->root_key
.objectid
;
9624 ret
= btrfs_unlink_subvol(trans
, new_dir
, root_objectid
,
9625 new_dentry
->d_name
.name
,
9626 new_dentry
->d_name
.len
);
9627 } else { /* dest is an inode */
9628 ret
= __btrfs_unlink_inode(trans
, dest
, BTRFS_I(new_dir
),
9629 BTRFS_I(new_dentry
->d_inode
),
9630 new_dentry
->d_name
.name
,
9631 new_dentry
->d_name
.len
);
9633 ret
= btrfs_update_inode(trans
, dest
, new_inode
);
9636 btrfs_abort_transaction(trans
, ret
);
9640 ret
= btrfs_add_link(trans
, BTRFS_I(new_dir
), BTRFS_I(old_inode
),
9641 new_dentry
->d_name
.name
,
9642 new_dentry
->d_name
.len
, 0, old_idx
);
9644 btrfs_abort_transaction(trans
, ret
);
9648 ret
= btrfs_add_link(trans
, BTRFS_I(old_dir
), BTRFS_I(new_inode
),
9649 old_dentry
->d_name
.name
,
9650 old_dentry
->d_name
.len
, 0, new_idx
);
9652 btrfs_abort_transaction(trans
, ret
);
9656 if (old_inode
->i_nlink
== 1)
9657 BTRFS_I(old_inode
)->dir_index
= old_idx
;
9658 if (new_inode
->i_nlink
== 1)
9659 BTRFS_I(new_inode
)->dir_index
= new_idx
;
9661 if (root_log_pinned
) {
9662 parent
= new_dentry
->d_parent
;
9663 ret
= btrfs_log_new_name(trans
, BTRFS_I(old_inode
),
9664 BTRFS_I(old_dir
), parent
,
9666 if (ret
== BTRFS_NEED_LOG_SYNC
)
9667 sync_log_root
= true;
9668 else if (ret
== BTRFS_NEED_TRANS_COMMIT
)
9669 commit_transaction
= true;
9671 btrfs_end_log_trans(root
);
9672 root_log_pinned
= false;
9674 if (dest_log_pinned
) {
9675 if (!commit_transaction
) {
9676 parent
= old_dentry
->d_parent
;
9677 ret
= btrfs_log_new_name(trans
, BTRFS_I(new_inode
),
9678 BTRFS_I(new_dir
), parent
,
9680 if (ret
== BTRFS_NEED_LOG_SYNC
)
9681 sync_log_dest
= true;
9682 else if (ret
== BTRFS_NEED_TRANS_COMMIT
)
9683 commit_transaction
= true;
9686 btrfs_end_log_trans(dest
);
9687 dest_log_pinned
= false;
9691 * If we have pinned a log and an error happened, we unpin tasks
9692 * trying to sync the log and force them to fallback to a transaction
9693 * commit if the log currently contains any of the inodes involved in
9694 * this rename operation (to ensure we do not persist a log with an
9695 * inconsistent state for any of these inodes or leading to any
9696 * inconsistencies when replayed). If the transaction was aborted, the
9697 * abortion reason is propagated to userspace when attempting to commit
9698 * the transaction. If the log does not contain any of these inodes, we
9699 * allow the tasks to sync it.
9701 if (ret
&& (root_log_pinned
|| dest_log_pinned
)) {
9702 if (btrfs_inode_in_log(BTRFS_I(old_dir
), fs_info
->generation
) ||
9703 btrfs_inode_in_log(BTRFS_I(new_dir
), fs_info
->generation
) ||
9704 btrfs_inode_in_log(BTRFS_I(old_inode
), fs_info
->generation
) ||
9706 btrfs_inode_in_log(BTRFS_I(new_inode
), fs_info
->generation
)))
9707 btrfs_set_log_full_commit(trans
);
9709 if (root_log_pinned
) {
9710 btrfs_end_log_trans(root
);
9711 root_log_pinned
= false;
9713 if (dest_log_pinned
) {
9714 btrfs_end_log_trans(dest
);
9715 dest_log_pinned
= false;
9718 if (!ret
&& sync_log_root
&& !commit_transaction
) {
9719 ret
= btrfs_sync_log(trans
, BTRFS_I(old_inode
)->root
,
9722 commit_transaction
= true;
9724 if (!ret
&& sync_log_dest
&& !commit_transaction
) {
9725 ret
= btrfs_sync_log(trans
, BTRFS_I(new_inode
)->root
,
9728 commit_transaction
= true;
9730 if (commit_transaction
) {
9731 ret
= btrfs_commit_transaction(trans
);
9735 ret2
= btrfs_end_transaction(trans
);
9736 ret
= ret
? ret
: ret2
;
9739 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9740 up_read(&fs_info
->subvol_sem
);
9741 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9742 up_read(&fs_info
->subvol_sem
);
9747 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle
*trans
,
9748 struct btrfs_root
*root
,
9750 struct dentry
*dentry
)
9753 struct inode
*inode
;
9757 ret
= btrfs_find_free_ino(root
, &objectid
);
9761 inode
= btrfs_new_inode(trans
, root
, dir
,
9762 dentry
->d_name
.name
,
9764 btrfs_ino(BTRFS_I(dir
)),
9766 S_IFCHR
| WHITEOUT_MODE
,
9769 if (IS_ERR(inode
)) {
9770 ret
= PTR_ERR(inode
);
9774 inode
->i_op
= &btrfs_special_inode_operations
;
9775 init_special_inode(inode
, inode
->i_mode
,
9778 ret
= btrfs_init_inode_security(trans
, inode
, dir
,
9783 ret
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
,
9784 BTRFS_I(inode
), 0, index
);
9788 ret
= btrfs_update_inode(trans
, root
, inode
);
9790 unlock_new_inode(inode
);
9792 inode_dec_link_count(inode
);
9798 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
9799 struct inode
*new_dir
, struct dentry
*new_dentry
,
9802 struct btrfs_fs_info
*fs_info
= btrfs_sb(old_dir
->i_sb
);
9803 struct btrfs_trans_handle
*trans
;
9804 unsigned int trans_num_items
;
9805 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9806 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9807 struct inode
*new_inode
= d_inode(new_dentry
);
9808 struct inode
*old_inode
= d_inode(old_dentry
);
9812 u64 old_ino
= btrfs_ino(BTRFS_I(old_inode
));
9813 bool log_pinned
= false;
9814 struct btrfs_log_ctx ctx
;
9815 bool sync_log
= false;
9816 bool commit_transaction
= false;
9818 if (btrfs_ino(BTRFS_I(new_dir
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
9821 /* we only allow rename subvolume link between subvolumes */
9822 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9825 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
9826 (new_inode
&& btrfs_ino(BTRFS_I(new_inode
)) == BTRFS_FIRST_FREE_OBJECTID
))
9829 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
9830 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
9834 /* check for collisions, even if the name isn't there */
9835 ret
= btrfs_check_dir_item_collision(dest
, new_dir
->i_ino
,
9836 new_dentry
->d_name
.name
,
9837 new_dentry
->d_name
.len
);
9840 if (ret
== -EEXIST
) {
9842 * eexist without a new_inode */
9843 if (WARN_ON(!new_inode
)) {
9847 /* maybe -EOVERFLOW */
9854 * we're using rename to replace one file with another. Start IO on it
9855 * now so we don't add too much work to the end of the transaction
9857 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
)
9858 filemap_flush(old_inode
->i_mapping
);
9860 /* close the racy window with snapshot create/destroy ioctl */
9861 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9862 down_read(&fs_info
->subvol_sem
);
9864 * We want to reserve the absolute worst case amount of items. So if
9865 * both inodes are subvols and we need to unlink them then that would
9866 * require 4 item modifications, but if they are both normal inodes it
9867 * would require 5 item modifications, so we'll assume they are normal
9868 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9869 * should cover the worst case number of items we'll modify.
9870 * If our rename has the whiteout flag, we need more 5 units for the
9871 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9872 * when selinux is enabled).
9874 trans_num_items
= 11;
9875 if (flags
& RENAME_WHITEOUT
)
9876 trans_num_items
+= 5;
9877 trans
= btrfs_start_transaction(root
, trans_num_items
);
9878 if (IS_ERR(trans
)) {
9879 ret
= PTR_ERR(trans
);
9884 btrfs_record_root_in_trans(trans
, dest
);
9886 ret
= btrfs_set_inode_index(BTRFS_I(new_dir
), &index
);
9890 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9891 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9892 /* force full log commit if subvolume involved. */
9893 btrfs_set_log_full_commit(trans
);
9895 btrfs_pin_log_trans(root
);
9897 ret
= btrfs_insert_inode_ref(trans
, dest
,
9898 new_dentry
->d_name
.name
,
9899 new_dentry
->d_name
.len
,
9901 btrfs_ino(BTRFS_I(new_dir
)), index
);
9906 inode_inc_iversion(old_dir
);
9907 inode_inc_iversion(new_dir
);
9908 inode_inc_iversion(old_inode
);
9909 old_dir
->i_ctime
= old_dir
->i_mtime
=
9910 new_dir
->i_ctime
= new_dir
->i_mtime
=
9911 old_inode
->i_ctime
= current_time(old_dir
);
9913 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
9914 btrfs_record_unlink_dir(trans
, BTRFS_I(old_dir
),
9915 BTRFS_I(old_inode
), 1);
9917 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9918 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
9919 ret
= btrfs_unlink_subvol(trans
, old_dir
, root_objectid
,
9920 old_dentry
->d_name
.name
,
9921 old_dentry
->d_name
.len
);
9923 ret
= __btrfs_unlink_inode(trans
, root
, BTRFS_I(old_dir
),
9924 BTRFS_I(d_inode(old_dentry
)),
9925 old_dentry
->d_name
.name
,
9926 old_dentry
->d_name
.len
);
9928 ret
= btrfs_update_inode(trans
, root
, old_inode
);
9931 btrfs_abort_transaction(trans
, ret
);
9936 inode_inc_iversion(new_inode
);
9937 new_inode
->i_ctime
= current_time(new_inode
);
9938 if (unlikely(btrfs_ino(BTRFS_I(new_inode
)) ==
9939 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
9940 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
9941 ret
= btrfs_unlink_subvol(trans
, new_dir
, root_objectid
,
9942 new_dentry
->d_name
.name
,
9943 new_dentry
->d_name
.len
);
9944 BUG_ON(new_inode
->i_nlink
== 0);
9946 ret
= btrfs_unlink_inode(trans
, dest
, BTRFS_I(new_dir
),
9947 BTRFS_I(d_inode(new_dentry
)),
9948 new_dentry
->d_name
.name
,
9949 new_dentry
->d_name
.len
);
9951 if (!ret
&& new_inode
->i_nlink
== 0)
9952 ret
= btrfs_orphan_add(trans
,
9953 BTRFS_I(d_inode(new_dentry
)));
9955 btrfs_abort_transaction(trans
, ret
);
9960 ret
= btrfs_add_link(trans
, BTRFS_I(new_dir
), BTRFS_I(old_inode
),
9961 new_dentry
->d_name
.name
,
9962 new_dentry
->d_name
.len
, 0, index
);
9964 btrfs_abort_transaction(trans
, ret
);
9968 if (old_inode
->i_nlink
== 1)
9969 BTRFS_I(old_inode
)->dir_index
= index
;
9972 struct dentry
*parent
= new_dentry
->d_parent
;
9974 btrfs_init_log_ctx(&ctx
, old_inode
);
9975 ret
= btrfs_log_new_name(trans
, BTRFS_I(old_inode
),
9976 BTRFS_I(old_dir
), parent
,
9978 if (ret
== BTRFS_NEED_LOG_SYNC
)
9980 else if (ret
== BTRFS_NEED_TRANS_COMMIT
)
9981 commit_transaction
= true;
9983 btrfs_end_log_trans(root
);
9987 if (flags
& RENAME_WHITEOUT
) {
9988 ret
= btrfs_whiteout_for_rename(trans
, root
, old_dir
,
9992 btrfs_abort_transaction(trans
, ret
);
9998 * If we have pinned the log and an error happened, we unpin tasks
9999 * trying to sync the log and force them to fallback to a transaction
10000 * commit if the log currently contains any of the inodes involved in
10001 * this rename operation (to ensure we do not persist a log with an
10002 * inconsistent state for any of these inodes or leading to any
10003 * inconsistencies when replayed). If the transaction was aborted, the
10004 * abortion reason is propagated to userspace when attempting to commit
10005 * the transaction. If the log does not contain any of these inodes, we
10006 * allow the tasks to sync it.
10008 if (ret
&& log_pinned
) {
10009 if (btrfs_inode_in_log(BTRFS_I(old_dir
), fs_info
->generation
) ||
10010 btrfs_inode_in_log(BTRFS_I(new_dir
), fs_info
->generation
) ||
10011 btrfs_inode_in_log(BTRFS_I(old_inode
), fs_info
->generation
) ||
10013 btrfs_inode_in_log(BTRFS_I(new_inode
), fs_info
->generation
)))
10014 btrfs_set_log_full_commit(trans
);
10016 btrfs_end_log_trans(root
);
10017 log_pinned
= false;
10019 if (!ret
&& sync_log
) {
10020 ret
= btrfs_sync_log(trans
, BTRFS_I(old_inode
)->root
, &ctx
);
10022 commit_transaction
= true;
10024 if (commit_transaction
) {
10025 ret
= btrfs_commit_transaction(trans
);
10029 ret2
= btrfs_end_transaction(trans
);
10030 ret
= ret
? ret
: ret2
;
10033 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
10034 up_read(&fs_info
->subvol_sem
);
10039 static int btrfs_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
10040 struct inode
*new_dir
, struct dentry
*new_dentry
,
10041 unsigned int flags
)
10043 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
10046 if (flags
& RENAME_EXCHANGE
)
10047 return btrfs_rename_exchange(old_dir
, old_dentry
, new_dir
,
10050 return btrfs_rename(old_dir
, old_dentry
, new_dir
, new_dentry
, flags
);
10053 struct btrfs_delalloc_work
{
10054 struct inode
*inode
;
10055 struct completion completion
;
10056 struct list_head list
;
10057 struct btrfs_work work
;
10060 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
10062 struct btrfs_delalloc_work
*delalloc_work
;
10063 struct inode
*inode
;
10065 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
10067 inode
= delalloc_work
->inode
;
10068 filemap_flush(inode
->i_mapping
);
10069 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
10070 &BTRFS_I(inode
)->runtime_flags
))
10071 filemap_flush(inode
->i_mapping
);
10074 complete(&delalloc_work
->completion
);
10077 static struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
)
10079 struct btrfs_delalloc_work
*work
;
10081 work
= kmalloc(sizeof(*work
), GFP_NOFS
);
10085 init_completion(&work
->completion
);
10086 INIT_LIST_HEAD(&work
->list
);
10087 work
->inode
= inode
;
10088 btrfs_init_work(&work
->work
, btrfs_flush_delalloc_helper
,
10089 btrfs_run_delalloc_work
, NULL
, NULL
);
10095 * some fairly slow code that needs optimization. This walks the list
10096 * of all the inodes with pending delalloc and forces them to disk.
10098 static int start_delalloc_inodes(struct btrfs_root
*root
, int nr
, bool snapshot
)
10100 struct btrfs_inode
*binode
;
10101 struct inode
*inode
;
10102 struct btrfs_delalloc_work
*work
, *next
;
10103 struct list_head works
;
10104 struct list_head splice
;
10107 INIT_LIST_HEAD(&works
);
10108 INIT_LIST_HEAD(&splice
);
10110 mutex_lock(&root
->delalloc_mutex
);
10111 spin_lock(&root
->delalloc_lock
);
10112 list_splice_init(&root
->delalloc_inodes
, &splice
);
10113 while (!list_empty(&splice
)) {
10114 binode
= list_entry(splice
.next
, struct btrfs_inode
,
10117 list_move_tail(&binode
->delalloc_inodes
,
10118 &root
->delalloc_inodes
);
10119 inode
= igrab(&binode
->vfs_inode
);
10121 cond_resched_lock(&root
->delalloc_lock
);
10124 spin_unlock(&root
->delalloc_lock
);
10127 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH
,
10128 &binode
->runtime_flags
);
10129 work
= btrfs_alloc_delalloc_work(inode
);
10135 list_add_tail(&work
->list
, &works
);
10136 btrfs_queue_work(root
->fs_info
->flush_workers
,
10139 if (nr
!= -1 && ret
>= nr
)
10142 spin_lock(&root
->delalloc_lock
);
10144 spin_unlock(&root
->delalloc_lock
);
10147 list_for_each_entry_safe(work
, next
, &works
, list
) {
10148 list_del_init(&work
->list
);
10149 wait_for_completion(&work
->completion
);
10153 if (!list_empty(&splice
)) {
10154 spin_lock(&root
->delalloc_lock
);
10155 list_splice_tail(&splice
, &root
->delalloc_inodes
);
10156 spin_unlock(&root
->delalloc_lock
);
10158 mutex_unlock(&root
->delalloc_mutex
);
10162 int btrfs_start_delalloc_snapshot(struct btrfs_root
*root
)
10164 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
10167 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
10170 ret
= start_delalloc_inodes(root
, -1, true);
10176 int btrfs_start_delalloc_roots(struct btrfs_fs_info
*fs_info
, int nr
)
10178 struct btrfs_root
*root
;
10179 struct list_head splice
;
10182 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
10185 INIT_LIST_HEAD(&splice
);
10187 mutex_lock(&fs_info
->delalloc_root_mutex
);
10188 spin_lock(&fs_info
->delalloc_root_lock
);
10189 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
10190 while (!list_empty(&splice
) && nr
) {
10191 root
= list_first_entry(&splice
, struct btrfs_root
,
10193 root
= btrfs_grab_fs_root(root
);
10195 list_move_tail(&root
->delalloc_root
,
10196 &fs_info
->delalloc_roots
);
10197 spin_unlock(&fs_info
->delalloc_root_lock
);
10199 ret
= start_delalloc_inodes(root
, nr
, false);
10200 btrfs_put_fs_root(root
);
10208 spin_lock(&fs_info
->delalloc_root_lock
);
10210 spin_unlock(&fs_info
->delalloc_root_lock
);
10214 if (!list_empty(&splice
)) {
10215 spin_lock(&fs_info
->delalloc_root_lock
);
10216 list_splice_tail(&splice
, &fs_info
->delalloc_roots
);
10217 spin_unlock(&fs_info
->delalloc_root_lock
);
10219 mutex_unlock(&fs_info
->delalloc_root_mutex
);
10223 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
10224 const char *symname
)
10226 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
10227 struct btrfs_trans_handle
*trans
;
10228 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
10229 struct btrfs_path
*path
;
10230 struct btrfs_key key
;
10231 struct inode
*inode
= NULL
;
10238 struct btrfs_file_extent_item
*ei
;
10239 struct extent_buffer
*leaf
;
10241 name_len
= strlen(symname
);
10242 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(fs_info
))
10243 return -ENAMETOOLONG
;
10246 * 2 items for inode item and ref
10247 * 2 items for dir items
10248 * 1 item for updating parent inode item
10249 * 1 item for the inline extent item
10250 * 1 item for xattr if selinux is on
10252 trans
= btrfs_start_transaction(root
, 7);
10254 return PTR_ERR(trans
);
10256 err
= btrfs_find_free_ino(root
, &objectid
);
10260 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
10261 dentry
->d_name
.len
, btrfs_ino(BTRFS_I(dir
)),
10262 objectid
, S_IFLNK
|S_IRWXUGO
, &index
);
10263 if (IS_ERR(inode
)) {
10264 err
= PTR_ERR(inode
);
10270 * If the active LSM wants to access the inode during
10271 * d_instantiate it needs these. Smack checks to see
10272 * if the filesystem supports xattrs by looking at the
10275 inode
->i_fop
= &btrfs_file_operations
;
10276 inode
->i_op
= &btrfs_file_inode_operations
;
10277 inode
->i_mapping
->a_ops
= &btrfs_aops
;
10278 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
10280 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
10284 path
= btrfs_alloc_path();
10289 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
10291 key
.type
= BTRFS_EXTENT_DATA_KEY
;
10292 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
10293 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
10296 btrfs_free_path(path
);
10299 leaf
= path
->nodes
[0];
10300 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
10301 struct btrfs_file_extent_item
);
10302 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
10303 btrfs_set_file_extent_type(leaf
, ei
,
10304 BTRFS_FILE_EXTENT_INLINE
);
10305 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
10306 btrfs_set_file_extent_compression(leaf
, ei
, 0);
10307 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
10308 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
10310 ptr
= btrfs_file_extent_inline_start(ei
);
10311 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
10312 btrfs_mark_buffer_dirty(leaf
);
10313 btrfs_free_path(path
);
10315 inode
->i_op
= &btrfs_symlink_inode_operations
;
10316 inode_nohighmem(inode
);
10317 inode_set_bytes(inode
, name_len
);
10318 btrfs_i_size_write(BTRFS_I(inode
), name_len
);
10319 err
= btrfs_update_inode(trans
, root
, inode
);
10321 * Last step, add directory indexes for our symlink inode. This is the
10322 * last step to avoid extra cleanup of these indexes if an error happens
10326 err
= btrfs_add_nondir(trans
, BTRFS_I(dir
), dentry
,
10327 BTRFS_I(inode
), 0, index
);
10331 d_instantiate_new(dentry
, inode
);
10334 btrfs_end_transaction(trans
);
10335 if (err
&& inode
) {
10336 inode_dec_link_count(inode
);
10337 discard_new_inode(inode
);
10339 btrfs_btree_balance_dirty(fs_info
);
10343 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
10344 u64 start
, u64 num_bytes
, u64 min_size
,
10345 loff_t actual_len
, u64
*alloc_hint
,
10346 struct btrfs_trans_handle
*trans
)
10348 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
10349 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
10350 struct extent_map
*em
;
10351 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
10352 struct btrfs_key ins
;
10353 u64 cur_offset
= start
;
10356 u64 last_alloc
= (u64
)-1;
10358 bool own_trans
= true;
10359 u64 end
= start
+ num_bytes
- 1;
10363 while (num_bytes
> 0) {
10365 trans
= btrfs_start_transaction(root
, 3);
10366 if (IS_ERR(trans
)) {
10367 ret
= PTR_ERR(trans
);
10372 cur_bytes
= min_t(u64
, num_bytes
, SZ_256M
);
10373 cur_bytes
= max(cur_bytes
, min_size
);
10375 * If we are severely fragmented we could end up with really
10376 * small allocations, so if the allocator is returning small
10377 * chunks lets make its job easier by only searching for those
10380 cur_bytes
= min(cur_bytes
, last_alloc
);
10381 ret
= btrfs_reserve_extent(root
, cur_bytes
, cur_bytes
,
10382 min_size
, 0, *alloc_hint
, &ins
, 1, 0);
10385 btrfs_end_transaction(trans
);
10388 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
10390 last_alloc
= ins
.offset
;
10391 ret
= insert_reserved_file_extent(trans
, inode
,
10392 cur_offset
, ins
.objectid
,
10393 ins
.offset
, ins
.offset
,
10394 ins
.offset
, 0, 0, 0,
10395 BTRFS_FILE_EXTENT_PREALLOC
);
10397 btrfs_free_reserved_extent(fs_info
, ins
.objectid
,
10399 btrfs_abort_transaction(trans
, ret
);
10401 btrfs_end_transaction(trans
);
10405 btrfs_drop_extent_cache(BTRFS_I(inode
), cur_offset
,
10406 cur_offset
+ ins
.offset
-1, 0);
10408 em
= alloc_extent_map();
10410 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
10411 &BTRFS_I(inode
)->runtime_flags
);
10415 em
->start
= cur_offset
;
10416 em
->orig_start
= cur_offset
;
10417 em
->len
= ins
.offset
;
10418 em
->block_start
= ins
.objectid
;
10419 em
->block_len
= ins
.offset
;
10420 em
->orig_block_len
= ins
.offset
;
10421 em
->ram_bytes
= ins
.offset
;
10422 em
->bdev
= fs_info
->fs_devices
->latest_bdev
;
10423 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
10424 em
->generation
= trans
->transid
;
10427 write_lock(&em_tree
->lock
);
10428 ret
= add_extent_mapping(em_tree
, em
, 1);
10429 write_unlock(&em_tree
->lock
);
10430 if (ret
!= -EEXIST
)
10432 btrfs_drop_extent_cache(BTRFS_I(inode
), cur_offset
,
10433 cur_offset
+ ins
.offset
- 1,
10436 free_extent_map(em
);
10438 num_bytes
-= ins
.offset
;
10439 cur_offset
+= ins
.offset
;
10440 *alloc_hint
= ins
.objectid
+ ins
.offset
;
10442 inode_inc_iversion(inode
);
10443 inode
->i_ctime
= current_time(inode
);
10444 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
10445 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
10446 (actual_len
> inode
->i_size
) &&
10447 (cur_offset
> inode
->i_size
)) {
10448 if (cur_offset
> actual_len
)
10449 i_size
= actual_len
;
10451 i_size
= cur_offset
;
10452 i_size_write(inode
, i_size
);
10453 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
10456 ret
= btrfs_update_inode(trans
, root
, inode
);
10459 btrfs_abort_transaction(trans
, ret
);
10461 btrfs_end_transaction(trans
);
10466 btrfs_end_transaction(trans
);
10468 if (cur_offset
< end
)
10469 btrfs_free_reserved_data_space(inode
, NULL
, cur_offset
,
10470 end
- cur_offset
+ 1);
10474 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
10475 u64 start
, u64 num_bytes
, u64 min_size
,
10476 loff_t actual_len
, u64
*alloc_hint
)
10478 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
10479 min_size
, actual_len
, alloc_hint
,
10483 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
10484 struct btrfs_trans_handle
*trans
, int mode
,
10485 u64 start
, u64 num_bytes
, u64 min_size
,
10486 loff_t actual_len
, u64
*alloc_hint
)
10488 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
10489 min_size
, actual_len
, alloc_hint
, trans
);
10492 static int btrfs_set_page_dirty(struct page
*page
)
10494 return __set_page_dirty_nobuffers(page
);
10497 static int btrfs_permission(struct inode
*inode
, int mask
)
10499 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
10500 umode_t mode
= inode
->i_mode
;
10502 if (mask
& MAY_WRITE
&&
10503 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
10504 if (btrfs_root_readonly(root
))
10506 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
10509 return generic_permission(inode
, mask
);
10512 static int btrfs_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
10514 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
10515 struct btrfs_trans_handle
*trans
;
10516 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
10517 struct inode
*inode
= NULL
;
10523 * 5 units required for adding orphan entry
10525 trans
= btrfs_start_transaction(root
, 5);
10527 return PTR_ERR(trans
);
10529 ret
= btrfs_find_free_ino(root
, &objectid
);
10533 inode
= btrfs_new_inode(trans
, root
, dir
, NULL
, 0,
10534 btrfs_ino(BTRFS_I(dir
)), objectid
, mode
, &index
);
10535 if (IS_ERR(inode
)) {
10536 ret
= PTR_ERR(inode
);
10541 inode
->i_fop
= &btrfs_file_operations
;
10542 inode
->i_op
= &btrfs_file_inode_operations
;
10544 inode
->i_mapping
->a_ops
= &btrfs_aops
;
10545 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
10547 ret
= btrfs_init_inode_security(trans
, inode
, dir
, NULL
);
10551 ret
= btrfs_update_inode(trans
, root
, inode
);
10554 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
10559 * We set number of links to 0 in btrfs_new_inode(), and here we set
10560 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10563 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10565 set_nlink(inode
, 1);
10566 d_tmpfile(dentry
, inode
);
10567 unlock_new_inode(inode
);
10568 mark_inode_dirty(inode
);
10570 btrfs_end_transaction(trans
);
10572 discard_new_inode(inode
);
10573 btrfs_btree_balance_dirty(fs_info
);
10577 void btrfs_set_range_writeback(struct extent_io_tree
*tree
, u64 start
, u64 end
)
10579 struct inode
*inode
= tree
->private_data
;
10580 unsigned long index
= start
>> PAGE_SHIFT
;
10581 unsigned long end_index
= end
>> PAGE_SHIFT
;
10584 while (index
<= end_index
) {
10585 page
= find_get_page(inode
->i_mapping
, index
);
10586 ASSERT(page
); /* Pages should be in the extent_io_tree */
10587 set_page_writeback(page
);
10595 * Add an entry indicating a block group or device which is pinned by a
10596 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10597 * negative errno on failure.
10599 static int btrfs_add_swapfile_pin(struct inode
*inode
, void *ptr
,
10600 bool is_block_group
)
10602 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
10603 struct btrfs_swapfile_pin
*sp
, *entry
;
10604 struct rb_node
**p
;
10605 struct rb_node
*parent
= NULL
;
10607 sp
= kmalloc(sizeof(*sp
), GFP_NOFS
);
10612 sp
->is_block_group
= is_block_group
;
10614 spin_lock(&fs_info
->swapfile_pins_lock
);
10615 p
= &fs_info
->swapfile_pins
.rb_node
;
10618 entry
= rb_entry(parent
, struct btrfs_swapfile_pin
, node
);
10619 if (sp
->ptr
< entry
->ptr
||
10620 (sp
->ptr
== entry
->ptr
&& sp
->inode
< entry
->inode
)) {
10621 p
= &(*p
)->rb_left
;
10622 } else if (sp
->ptr
> entry
->ptr
||
10623 (sp
->ptr
== entry
->ptr
&& sp
->inode
> entry
->inode
)) {
10624 p
= &(*p
)->rb_right
;
10626 spin_unlock(&fs_info
->swapfile_pins_lock
);
10631 rb_link_node(&sp
->node
, parent
, p
);
10632 rb_insert_color(&sp
->node
, &fs_info
->swapfile_pins
);
10633 spin_unlock(&fs_info
->swapfile_pins_lock
);
10637 /* Free all of the entries pinned by this swapfile. */
10638 static void btrfs_free_swapfile_pins(struct inode
*inode
)
10640 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
10641 struct btrfs_swapfile_pin
*sp
;
10642 struct rb_node
*node
, *next
;
10644 spin_lock(&fs_info
->swapfile_pins_lock
);
10645 node
= rb_first(&fs_info
->swapfile_pins
);
10647 next
= rb_next(node
);
10648 sp
= rb_entry(node
, struct btrfs_swapfile_pin
, node
);
10649 if (sp
->inode
== inode
) {
10650 rb_erase(&sp
->node
, &fs_info
->swapfile_pins
);
10651 if (sp
->is_block_group
)
10652 btrfs_put_block_group(sp
->ptr
);
10657 spin_unlock(&fs_info
->swapfile_pins_lock
);
10660 struct btrfs_swap_info
{
10666 unsigned long nr_pages
;
10670 static int btrfs_add_swap_extent(struct swap_info_struct
*sis
,
10671 struct btrfs_swap_info
*bsi
)
10673 unsigned long nr_pages
;
10674 u64 first_ppage
, first_ppage_reported
, next_ppage
;
10677 first_ppage
= ALIGN(bsi
->block_start
, PAGE_SIZE
) >> PAGE_SHIFT
;
10678 next_ppage
= ALIGN_DOWN(bsi
->block_start
+ bsi
->block_len
,
10679 PAGE_SIZE
) >> PAGE_SHIFT
;
10681 if (first_ppage
>= next_ppage
)
10683 nr_pages
= next_ppage
- first_ppage
;
10685 first_ppage_reported
= first_ppage
;
10686 if (bsi
->start
== 0)
10687 first_ppage_reported
++;
10688 if (bsi
->lowest_ppage
> first_ppage_reported
)
10689 bsi
->lowest_ppage
= first_ppage_reported
;
10690 if (bsi
->highest_ppage
< (next_ppage
- 1))
10691 bsi
->highest_ppage
= next_ppage
- 1;
10693 ret
= add_swap_extent(sis
, bsi
->nr_pages
, nr_pages
, first_ppage
);
10696 bsi
->nr_extents
+= ret
;
10697 bsi
->nr_pages
+= nr_pages
;
10701 static void btrfs_swap_deactivate(struct file
*file
)
10703 struct inode
*inode
= file_inode(file
);
10705 btrfs_free_swapfile_pins(inode
);
10706 atomic_dec(&BTRFS_I(inode
)->root
->nr_swapfiles
);
10709 static int btrfs_swap_activate(struct swap_info_struct
*sis
, struct file
*file
,
10712 struct inode
*inode
= file_inode(file
);
10713 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
10714 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
10715 struct extent_state
*cached_state
= NULL
;
10716 struct extent_map
*em
= NULL
;
10717 struct btrfs_device
*device
= NULL
;
10718 struct btrfs_swap_info bsi
= {
10719 .lowest_ppage
= (sector_t
)-1ULL,
10726 * If the swap file was just created, make sure delalloc is done. If the
10727 * file changes again after this, the user is doing something stupid and
10728 * we don't really care.
10730 ret
= btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
10735 * The inode is locked, so these flags won't change after we check them.
10737 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
) {
10738 btrfs_warn(fs_info
, "swapfile must not be compressed");
10741 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
)) {
10742 btrfs_warn(fs_info
, "swapfile must not be copy-on-write");
10745 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
10746 btrfs_warn(fs_info
, "swapfile must not be checksummed");
10751 * Balance or device remove/replace/resize can move stuff around from
10752 * under us. The EXCL_OP flag makes sure they aren't running/won't run
10753 * concurrently while we are mapping the swap extents, and
10754 * fs_info->swapfile_pins prevents them from running while the swap file
10755 * is active and moving the extents. Note that this also prevents a
10756 * concurrent device add which isn't actually necessary, but it's not
10757 * really worth the trouble to allow it.
10759 if (test_and_set_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
)) {
10760 btrfs_warn(fs_info
,
10761 "cannot activate swapfile while exclusive operation is running");
10765 * Snapshots can create extents which require COW even if NODATACOW is
10766 * set. We use this counter to prevent snapshots. We must increment it
10767 * before walking the extents because we don't want a concurrent
10768 * snapshot to run after we've already checked the extents.
10770 atomic_inc(&BTRFS_I(inode
)->root
->nr_swapfiles
);
10772 isize
= ALIGN_DOWN(inode
->i_size
, fs_info
->sectorsize
);
10774 lock_extent_bits(io_tree
, 0, isize
- 1, &cached_state
);
10776 while (start
< isize
) {
10777 u64 logical_block_start
, physical_block_start
;
10778 struct btrfs_block_group_cache
*bg
;
10779 u64 len
= isize
- start
;
10781 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0, start
, len
, 0);
10787 if (em
->block_start
== EXTENT_MAP_HOLE
) {
10788 btrfs_warn(fs_info
, "swapfile must not have holes");
10792 if (em
->block_start
== EXTENT_MAP_INLINE
) {
10794 * It's unlikely we'll ever actually find ourselves
10795 * here, as a file small enough to fit inline won't be
10796 * big enough to store more than the swap header, but in
10797 * case something changes in the future, let's catch it
10798 * here rather than later.
10800 btrfs_warn(fs_info
, "swapfile must not be inline");
10804 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
10805 btrfs_warn(fs_info
, "swapfile must not be compressed");
10810 logical_block_start
= em
->block_start
+ (start
- em
->start
);
10811 len
= min(len
, em
->len
- (start
- em
->start
));
10812 free_extent_map(em
);
10815 ret
= can_nocow_extent(inode
, start
, &len
, NULL
, NULL
, NULL
);
10821 btrfs_warn(fs_info
,
10822 "swapfile must not be copy-on-write");
10827 em
= btrfs_get_chunk_map(fs_info
, logical_block_start
, len
);
10833 if (em
->map_lookup
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
10834 btrfs_warn(fs_info
,
10835 "swapfile must have single data profile");
10840 if (device
== NULL
) {
10841 device
= em
->map_lookup
->stripes
[0].dev
;
10842 ret
= btrfs_add_swapfile_pin(inode
, device
, false);
10847 } else if (device
!= em
->map_lookup
->stripes
[0].dev
) {
10848 btrfs_warn(fs_info
, "swapfile must be on one device");
10853 physical_block_start
= (em
->map_lookup
->stripes
[0].physical
+
10854 (logical_block_start
- em
->start
));
10855 len
= min(len
, em
->len
- (logical_block_start
- em
->start
));
10856 free_extent_map(em
);
10859 bg
= btrfs_lookup_block_group(fs_info
, logical_block_start
);
10861 btrfs_warn(fs_info
,
10862 "could not find block group containing swapfile");
10867 ret
= btrfs_add_swapfile_pin(inode
, bg
, true);
10869 btrfs_put_block_group(bg
);
10876 if (bsi
.block_len
&&
10877 bsi
.block_start
+ bsi
.block_len
== physical_block_start
) {
10878 bsi
.block_len
+= len
;
10880 if (bsi
.block_len
) {
10881 ret
= btrfs_add_swap_extent(sis
, &bsi
);
10886 bsi
.block_start
= physical_block_start
;
10887 bsi
.block_len
= len
;
10894 ret
= btrfs_add_swap_extent(sis
, &bsi
);
10897 if (!IS_ERR_OR_NULL(em
))
10898 free_extent_map(em
);
10900 unlock_extent_cached(io_tree
, 0, isize
- 1, &cached_state
);
10903 btrfs_swap_deactivate(file
);
10905 clear_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
);
10911 sis
->bdev
= device
->bdev
;
10912 *span
= bsi
.highest_ppage
- bsi
.lowest_ppage
+ 1;
10913 sis
->max
= bsi
.nr_pages
;
10914 sis
->pages
= bsi
.nr_pages
- 1;
10915 sis
->highest_bit
= bsi
.nr_pages
- 1;
10916 return bsi
.nr_extents
;
10919 static void btrfs_swap_deactivate(struct file
*file
)
10923 static int btrfs_swap_activate(struct swap_info_struct
*sis
, struct file
*file
,
10926 return -EOPNOTSUPP
;
10930 static const struct inode_operations btrfs_dir_inode_operations
= {
10931 .getattr
= btrfs_getattr
,
10932 .lookup
= btrfs_lookup
,
10933 .create
= btrfs_create
,
10934 .unlink
= btrfs_unlink
,
10935 .link
= btrfs_link
,
10936 .mkdir
= btrfs_mkdir
,
10937 .rmdir
= btrfs_rmdir
,
10938 .rename
= btrfs_rename2
,
10939 .symlink
= btrfs_symlink
,
10940 .setattr
= btrfs_setattr
,
10941 .mknod
= btrfs_mknod
,
10942 .listxattr
= btrfs_listxattr
,
10943 .permission
= btrfs_permission
,
10944 .get_acl
= btrfs_get_acl
,
10945 .set_acl
= btrfs_set_acl
,
10946 .update_time
= btrfs_update_time
,
10947 .tmpfile
= btrfs_tmpfile
,
10949 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
10950 .lookup
= btrfs_lookup
,
10951 .permission
= btrfs_permission
,
10952 .update_time
= btrfs_update_time
,
10955 static const struct file_operations btrfs_dir_file_operations
= {
10956 .llseek
= generic_file_llseek
,
10957 .read
= generic_read_dir
,
10958 .iterate_shared
= btrfs_real_readdir
,
10959 .open
= btrfs_opendir
,
10960 .unlocked_ioctl
= btrfs_ioctl
,
10961 #ifdef CONFIG_COMPAT
10962 .compat_ioctl
= btrfs_compat_ioctl
,
10964 .release
= btrfs_release_file
,
10965 .fsync
= btrfs_sync_file
,
10968 static const struct extent_io_ops btrfs_extent_io_ops
= {
10969 /* mandatory callbacks */
10970 .submit_bio_hook
= btrfs_submit_bio_hook
,
10971 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
10975 * btrfs doesn't support the bmap operation because swapfiles
10976 * use bmap to make a mapping of extents in the file. They assume
10977 * these extents won't change over the life of the file and they
10978 * use the bmap result to do IO directly to the drive.
10980 * the btrfs bmap call would return logical addresses that aren't
10981 * suitable for IO and they also will change frequently as COW
10982 * operations happen. So, swapfile + btrfs == corruption.
10984 * For now we're avoiding this by dropping bmap.
10986 static const struct address_space_operations btrfs_aops
= {
10987 .readpage
= btrfs_readpage
,
10988 .writepage
= btrfs_writepage
,
10989 .writepages
= btrfs_writepages
,
10990 .readpages
= btrfs_readpages
,
10991 .direct_IO
= btrfs_direct_IO
,
10992 .invalidatepage
= btrfs_invalidatepage
,
10993 .releasepage
= btrfs_releasepage
,
10994 .set_page_dirty
= btrfs_set_page_dirty
,
10995 .error_remove_page
= generic_error_remove_page
,
10996 .swap_activate
= btrfs_swap_activate
,
10997 .swap_deactivate
= btrfs_swap_deactivate
,
11000 static const struct inode_operations btrfs_file_inode_operations
= {
11001 .getattr
= btrfs_getattr
,
11002 .setattr
= btrfs_setattr
,
11003 .listxattr
= btrfs_listxattr
,
11004 .permission
= btrfs_permission
,
11005 .fiemap
= btrfs_fiemap
,
11006 .get_acl
= btrfs_get_acl
,
11007 .set_acl
= btrfs_set_acl
,
11008 .update_time
= btrfs_update_time
,
11010 static const struct inode_operations btrfs_special_inode_operations
= {
11011 .getattr
= btrfs_getattr
,
11012 .setattr
= btrfs_setattr
,
11013 .permission
= btrfs_permission
,
11014 .listxattr
= btrfs_listxattr
,
11015 .get_acl
= btrfs_get_acl
,
11016 .set_acl
= btrfs_set_acl
,
11017 .update_time
= btrfs_update_time
,
11019 static const struct inode_operations btrfs_symlink_inode_operations
= {
11020 .get_link
= page_get_link
,
11021 .getattr
= btrfs_getattr
,
11022 .setattr
= btrfs_setattr
,
11023 .permission
= btrfs_permission
,
11024 .listxattr
= btrfs_listxattr
,
11025 .update_time
= btrfs_update_time
,
11028 const struct dentry_operations btrfs_dentry_operations
= {
11029 .d_delete
= btrfs_dentry_delete
,