2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/aio.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include <linux/slab.h>
41 #include <linux/ratelimit.h>
42 #include <linux/mount.h>
43 #include <linux/btrfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/posix_acl_xattr.h>
49 #include "transaction.h"
50 #include "btrfs_inode.h"
51 #include "print-tree.h"
52 #include "ordered-data.h"
56 #include "compression.h"
58 #include "free-space-cache.h"
59 #include "inode-map.h"
63 struct btrfs_iget_args
{
65 struct btrfs_root
*root
;
68 static const struct inode_operations btrfs_dir_inode_operations
;
69 static const struct inode_operations btrfs_symlink_inode_operations
;
70 static const struct inode_operations btrfs_dir_ro_inode_operations
;
71 static const struct inode_operations btrfs_special_inode_operations
;
72 static const struct inode_operations btrfs_file_inode_operations
;
73 static const struct address_space_operations btrfs_aops
;
74 static const struct address_space_operations btrfs_symlink_aops
;
75 static const struct file_operations btrfs_dir_file_operations
;
76 static struct extent_io_ops btrfs_extent_io_ops
;
78 static struct kmem_cache
*btrfs_inode_cachep
;
79 static struct kmem_cache
*btrfs_delalloc_work_cachep
;
80 struct kmem_cache
*btrfs_trans_handle_cachep
;
81 struct kmem_cache
*btrfs_transaction_cachep
;
82 struct kmem_cache
*btrfs_path_cachep
;
83 struct kmem_cache
*btrfs_free_space_cachep
;
86 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
87 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
88 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
89 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
90 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
91 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
92 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
93 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
96 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
);
97 static int btrfs_truncate(struct inode
*inode
);
98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
);
99 static noinline
int cow_file_range(struct inode
*inode
,
100 struct page
*locked_page
,
101 u64 start
, u64 end
, int *page_started
,
102 unsigned long *nr_written
, int unlock
);
103 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
104 u64 len
, u64 orig_start
,
105 u64 block_start
, u64 block_len
,
106 u64 orig_block_len
, u64 ram_bytes
,
109 static int btrfs_dirty_inode(struct inode
*inode
);
111 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
112 struct inode
*inode
, struct inode
*dir
,
113 const struct qstr
*qstr
)
117 err
= btrfs_init_acl(trans
, inode
, dir
);
119 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
128 static noinline
int insert_inline_extent(struct btrfs_trans_handle
*trans
,
129 struct btrfs_root
*root
, struct inode
*inode
,
130 u64 start
, size_t size
, size_t compressed_size
,
132 struct page
**compressed_pages
)
134 struct btrfs_key key
;
135 struct btrfs_path
*path
;
136 struct extent_buffer
*leaf
;
137 struct page
*page
= NULL
;
140 struct btrfs_file_extent_item
*ei
;
143 size_t cur_size
= size
;
145 unsigned long offset
;
147 if (compressed_size
&& compressed_pages
)
148 cur_size
= compressed_size
;
150 path
= btrfs_alloc_path();
154 path
->leave_spinning
= 1;
156 key
.objectid
= btrfs_ino(inode
);
158 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
159 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
161 inode_add_bytes(inode
, size
);
162 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
168 leaf
= path
->nodes
[0];
169 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
170 struct btrfs_file_extent_item
);
171 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
172 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
173 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
174 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
175 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
176 ptr
= btrfs_file_extent_inline_start(ei
);
178 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
181 while (compressed_size
> 0) {
182 cpage
= compressed_pages
[i
];
183 cur_size
= min_t(unsigned long, compressed_size
,
186 kaddr
= kmap_atomic(cpage
);
187 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
188 kunmap_atomic(kaddr
);
192 compressed_size
-= cur_size
;
194 btrfs_set_file_extent_compression(leaf
, ei
,
197 page
= find_get_page(inode
->i_mapping
,
198 start
>> PAGE_CACHE_SHIFT
);
199 btrfs_set_file_extent_compression(leaf
, ei
, 0);
200 kaddr
= kmap_atomic(page
);
201 offset
= start
& (PAGE_CACHE_SIZE
- 1);
202 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
203 kunmap_atomic(kaddr
);
204 page_cache_release(page
);
206 btrfs_mark_buffer_dirty(leaf
);
207 btrfs_free_path(path
);
210 * we're an inline extent, so nobody can
211 * extend the file past i_size without locking
212 * a page we already have locked.
214 * We must do any isize and inode updates
215 * before we unlock the pages. Otherwise we
216 * could end up racing with unlink.
218 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
219 ret
= btrfs_update_inode(trans
, root
, inode
);
223 btrfs_free_path(path
);
229 * conditionally insert an inline extent into the file. This
230 * does the checks required to make sure the data is small enough
231 * to fit as an inline extent.
233 static noinline
int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
234 struct btrfs_root
*root
,
235 struct inode
*inode
, u64 start
, u64 end
,
236 size_t compressed_size
, int compress_type
,
237 struct page
**compressed_pages
)
239 u64 isize
= i_size_read(inode
);
240 u64 actual_end
= min(end
+ 1, isize
);
241 u64 inline_len
= actual_end
- start
;
242 u64 aligned_end
= ALIGN(end
, root
->sectorsize
);
243 u64 data_len
= inline_len
;
247 data_len
= compressed_size
;
250 actual_end
>= PAGE_CACHE_SIZE
||
251 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
253 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
255 data_len
> root
->fs_info
->max_inline
) {
259 ret
= btrfs_drop_extents(trans
, root
, inode
, start
, aligned_end
, 1);
263 if (isize
> actual_end
)
264 inline_len
= min_t(u64
, isize
, actual_end
);
265 ret
= insert_inline_extent(trans
, root
, inode
, start
,
266 inline_len
, compressed_size
,
267 compress_type
, compressed_pages
);
268 if (ret
&& ret
!= -ENOSPC
) {
269 btrfs_abort_transaction(trans
, root
, ret
);
271 } else if (ret
== -ENOSPC
) {
275 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
276 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
277 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
281 struct async_extent
{
286 unsigned long nr_pages
;
288 struct list_head list
;
293 struct btrfs_root
*root
;
294 struct page
*locked_page
;
297 struct list_head extents
;
298 struct btrfs_work work
;
301 static noinline
int add_async_extent(struct async_cow
*cow
,
302 u64 start
, u64 ram_size
,
305 unsigned long nr_pages
,
308 struct async_extent
*async_extent
;
310 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
311 BUG_ON(!async_extent
); /* -ENOMEM */
312 async_extent
->start
= start
;
313 async_extent
->ram_size
= ram_size
;
314 async_extent
->compressed_size
= compressed_size
;
315 async_extent
->pages
= pages
;
316 async_extent
->nr_pages
= nr_pages
;
317 async_extent
->compress_type
= compress_type
;
318 list_add_tail(&async_extent
->list
, &cow
->extents
);
323 * we create compressed extents in two phases. The first
324 * phase compresses a range of pages that have already been
325 * locked (both pages and state bits are locked).
327 * This is done inside an ordered work queue, and the compression
328 * is spread across many cpus. The actual IO submission is step
329 * two, and the ordered work queue takes care of making sure that
330 * happens in the same order things were put onto the queue by
331 * writepages and friends.
333 * If this code finds it can't get good compression, it puts an
334 * entry onto the work queue to write the uncompressed bytes. This
335 * makes sure that both compressed inodes and uncompressed inodes
336 * are written in the same order that the flusher thread sent them
339 static noinline
int compress_file_range(struct inode
*inode
,
340 struct page
*locked_page
,
342 struct async_cow
*async_cow
,
345 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
346 struct btrfs_trans_handle
*trans
;
348 u64 blocksize
= root
->sectorsize
;
350 u64 isize
= i_size_read(inode
);
352 struct page
**pages
= NULL
;
353 unsigned long nr_pages
;
354 unsigned long nr_pages_ret
= 0;
355 unsigned long total_compressed
= 0;
356 unsigned long total_in
= 0;
357 unsigned long max_compressed
= 128 * 1024;
358 unsigned long max_uncompressed
= 128 * 1024;
361 int compress_type
= root
->fs_info
->compress_type
;
364 /* if this is a small write inside eof, kick off a defrag */
365 if ((end
- start
+ 1) < 16 * 1024 &&
366 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
367 btrfs_add_inode_defrag(NULL
, inode
);
369 actual_end
= min_t(u64
, isize
, end
+ 1);
372 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
373 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
376 * we don't want to send crud past the end of i_size through
377 * compression, that's just a waste of CPU time. So, if the
378 * end of the file is before the start of our current
379 * requested range of bytes, we bail out to the uncompressed
380 * cleanup code that can deal with all of this.
382 * It isn't really the fastest way to fix things, but this is a
383 * very uncommon corner.
385 if (actual_end
<= start
)
386 goto cleanup_and_bail_uncompressed
;
388 total_compressed
= actual_end
- start
;
390 /* we want to make sure that amount of ram required to uncompress
391 * an extent is reasonable, so we limit the total size in ram
392 * of a compressed extent to 128k. This is a crucial number
393 * because it also controls how easily we can spread reads across
394 * cpus for decompression.
396 * We also want to make sure the amount of IO required to do
397 * a random read is reasonably small, so we limit the size of
398 * a compressed extent to 128k.
400 total_compressed
= min(total_compressed
, max_uncompressed
);
401 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
402 num_bytes
= max(blocksize
, num_bytes
);
407 * we do compression for mount -o compress and when the
408 * inode has not been flagged as nocompress. This flag can
409 * change at any time if we discover bad compression ratios.
411 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
) &&
412 (btrfs_test_opt(root
, COMPRESS
) ||
413 (BTRFS_I(inode
)->force_compress
) ||
414 (BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))) {
416 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
418 /* just bail out to the uncompressed code */
422 if (BTRFS_I(inode
)->force_compress
)
423 compress_type
= BTRFS_I(inode
)->force_compress
;
426 * we need to call clear_page_dirty_for_io on each
427 * page in the range. Otherwise applications with the file
428 * mmap'd can wander in and change the page contents while
429 * we are compressing them.
431 * If the compression fails for any reason, we set the pages
432 * dirty again later on.
434 extent_range_clear_dirty_for_io(inode
, start
, end
);
436 ret
= btrfs_compress_pages(compress_type
,
437 inode
->i_mapping
, start
,
438 total_compressed
, pages
,
439 nr_pages
, &nr_pages_ret
,
445 unsigned long offset
= total_compressed
&
446 (PAGE_CACHE_SIZE
- 1);
447 struct page
*page
= pages
[nr_pages_ret
- 1];
450 /* zero the tail end of the last page, we might be
451 * sending it down to disk
454 kaddr
= kmap_atomic(page
);
455 memset(kaddr
+ offset
, 0,
456 PAGE_CACHE_SIZE
- offset
);
457 kunmap_atomic(kaddr
);
464 trans
= btrfs_join_transaction(root
);
466 ret
= PTR_ERR(trans
);
468 goto cleanup_and_out
;
470 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
472 /* lets try to make an inline extent */
473 if (ret
|| total_in
< (actual_end
- start
)) {
474 /* we didn't compress the entire range, try
475 * to make an uncompressed inline extent.
477 ret
= cow_file_range_inline(trans
, root
, inode
,
478 start
, end
, 0, 0, NULL
);
480 /* try making a compressed inline extent */
481 ret
= cow_file_range_inline(trans
, root
, inode
,
484 compress_type
, pages
);
488 * inline extent creation worked or returned error,
489 * we don't need to create any more async work items.
490 * Unlock and free up our temp pages.
492 extent_clear_unlock_delalloc(inode
,
493 &BTRFS_I(inode
)->io_tree
,
495 EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
496 EXTENT_CLEAR_DELALLOC
|
497 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
);
499 btrfs_end_transaction(trans
, root
);
502 btrfs_end_transaction(trans
, root
);
507 * we aren't doing an inline extent round the compressed size
508 * up to a block size boundary so the allocator does sane
511 total_compressed
= ALIGN(total_compressed
, blocksize
);
514 * one last check to make sure the compression is really a
515 * win, compare the page count read with the blocks on disk
517 total_in
= ALIGN(total_in
, PAGE_CACHE_SIZE
);
518 if (total_compressed
>= total_in
) {
521 num_bytes
= total_in
;
524 if (!will_compress
&& pages
) {
526 * the compression code ran but failed to make things smaller,
527 * free any pages it allocated and our page pointer array
529 for (i
= 0; i
< nr_pages_ret
; i
++) {
530 WARN_ON(pages
[i
]->mapping
);
531 page_cache_release(pages
[i
]);
535 total_compressed
= 0;
538 /* flag the file so we don't compress in the future */
539 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
540 !(BTRFS_I(inode
)->force_compress
)) {
541 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
547 /* the async work queues will take care of doing actual
548 * allocation on disk for these compressed pages,
549 * and will submit them to the elevator.
551 add_async_extent(async_cow
, start
, num_bytes
,
552 total_compressed
, pages
, nr_pages_ret
,
555 if (start
+ num_bytes
< end
) {
562 cleanup_and_bail_uncompressed
:
564 * No compression, but we still need to write the pages in
565 * the file we've been given so far. redirty the locked
566 * page if it corresponds to our extent and set things up
567 * for the async work queue to run cow_file_range to do
568 * the normal delalloc dance
570 if (page_offset(locked_page
) >= start
&&
571 page_offset(locked_page
) <= end
) {
572 __set_page_dirty_nobuffers(locked_page
);
573 /* unlocked later on in the async handlers */
576 extent_range_redirty_for_io(inode
, start
, end
);
577 add_async_extent(async_cow
, start
, end
- start
+ 1,
578 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
586 for (i
= 0; i
< nr_pages_ret
; i
++) {
587 WARN_ON(pages
[i
]->mapping
);
588 page_cache_release(pages
[i
]);
595 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
597 EXTENT_CLEAR_UNLOCK_PAGE
|
599 EXTENT_CLEAR_DELALLOC
|
600 EXTENT_SET_WRITEBACK
|
601 EXTENT_END_WRITEBACK
);
602 if (!trans
|| IS_ERR(trans
))
603 btrfs_error(root
->fs_info
, ret
, "Failed to join transaction");
605 btrfs_abort_transaction(trans
, root
, ret
);
610 * phase two of compressed writeback. This is the ordered portion
611 * of the code, which only gets called in the order the work was
612 * queued. We walk all the async extents created by compress_file_range
613 * and send them down to the disk.
615 static noinline
int submit_compressed_extents(struct inode
*inode
,
616 struct async_cow
*async_cow
)
618 struct async_extent
*async_extent
;
620 struct btrfs_trans_handle
*trans
;
621 struct btrfs_key ins
;
622 struct extent_map
*em
;
623 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
624 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
625 struct extent_io_tree
*io_tree
;
628 if (list_empty(&async_cow
->extents
))
632 while (!list_empty(&async_cow
->extents
)) {
633 async_extent
= list_entry(async_cow
->extents
.next
,
634 struct async_extent
, list
);
635 list_del(&async_extent
->list
);
637 io_tree
= &BTRFS_I(inode
)->io_tree
;
640 /* did the compression code fall back to uncompressed IO? */
641 if (!async_extent
->pages
) {
642 int page_started
= 0;
643 unsigned long nr_written
= 0;
645 lock_extent(io_tree
, async_extent
->start
,
646 async_extent
->start
+
647 async_extent
->ram_size
- 1);
649 /* allocate blocks */
650 ret
= cow_file_range(inode
, async_cow
->locked_page
,
652 async_extent
->start
+
653 async_extent
->ram_size
- 1,
654 &page_started
, &nr_written
, 0);
659 * if page_started, cow_file_range inserted an
660 * inline extent and took care of all the unlocking
661 * and IO for us. Otherwise, we need to submit
662 * all those pages down to the drive.
664 if (!page_started
&& !ret
)
665 extent_write_locked_range(io_tree
,
666 inode
, async_extent
->start
,
667 async_extent
->start
+
668 async_extent
->ram_size
- 1,
672 unlock_page(async_cow
->locked_page
);
678 lock_extent(io_tree
, async_extent
->start
,
679 async_extent
->start
+ async_extent
->ram_size
- 1);
681 trans
= btrfs_join_transaction(root
);
683 ret
= PTR_ERR(trans
);
685 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
686 ret
= btrfs_reserve_extent(trans
, root
,
687 async_extent
->compressed_size
,
688 async_extent
->compressed_size
,
689 0, alloc_hint
, &ins
, 1);
690 if (ret
&& ret
!= -ENOSPC
)
691 btrfs_abort_transaction(trans
, root
, ret
);
692 btrfs_end_transaction(trans
, root
);
698 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
699 WARN_ON(async_extent
->pages
[i
]->mapping
);
700 page_cache_release(async_extent
->pages
[i
]);
702 kfree(async_extent
->pages
);
703 async_extent
->nr_pages
= 0;
704 async_extent
->pages
= NULL
;
706 if (ret
== -ENOSPC
) {
707 unlock_extent(io_tree
, async_extent
->start
,
708 async_extent
->start
+
709 async_extent
->ram_size
- 1);
716 * here we're doing allocation and writeback of the
719 btrfs_drop_extent_cache(inode
, async_extent
->start
,
720 async_extent
->start
+
721 async_extent
->ram_size
- 1, 0);
723 em
= alloc_extent_map();
726 goto out_free_reserve
;
728 em
->start
= async_extent
->start
;
729 em
->len
= async_extent
->ram_size
;
730 em
->orig_start
= em
->start
;
731 em
->mod_start
= em
->start
;
732 em
->mod_len
= em
->len
;
734 em
->block_start
= ins
.objectid
;
735 em
->block_len
= ins
.offset
;
736 em
->orig_block_len
= ins
.offset
;
737 em
->ram_bytes
= async_extent
->ram_size
;
738 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
739 em
->compress_type
= async_extent
->compress_type
;
740 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
741 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
745 write_lock(&em_tree
->lock
);
746 ret
= add_extent_mapping(em_tree
, em
, 1);
747 write_unlock(&em_tree
->lock
);
748 if (ret
!= -EEXIST
) {
752 btrfs_drop_extent_cache(inode
, async_extent
->start
,
753 async_extent
->start
+
754 async_extent
->ram_size
- 1, 0);
758 goto out_free_reserve
;
760 ret
= btrfs_add_ordered_extent_compress(inode
,
763 async_extent
->ram_size
,
765 BTRFS_ORDERED_COMPRESSED
,
766 async_extent
->compress_type
);
768 goto out_free_reserve
;
771 * clear dirty, set writeback and unlock the pages.
773 extent_clear_unlock_delalloc(inode
,
774 &BTRFS_I(inode
)->io_tree
,
776 async_extent
->start
+
777 async_extent
->ram_size
- 1,
778 NULL
, EXTENT_CLEAR_UNLOCK_PAGE
|
779 EXTENT_CLEAR_UNLOCK
|
780 EXTENT_CLEAR_DELALLOC
|
781 EXTENT_CLEAR_DIRTY
| EXTENT_SET_WRITEBACK
);
783 ret
= btrfs_submit_compressed_write(inode
,
785 async_extent
->ram_size
,
787 ins
.offset
, async_extent
->pages
,
788 async_extent
->nr_pages
);
789 alloc_hint
= ins
.objectid
+ ins
.offset
;
799 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
801 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
803 async_extent
->start
+
804 async_extent
->ram_size
- 1,
805 NULL
, EXTENT_CLEAR_UNLOCK_PAGE
|
806 EXTENT_CLEAR_UNLOCK
|
807 EXTENT_CLEAR_DELALLOC
|
809 EXTENT_SET_WRITEBACK
|
810 EXTENT_END_WRITEBACK
);
815 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
818 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
819 struct extent_map
*em
;
822 read_lock(&em_tree
->lock
);
823 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
826 * if block start isn't an actual block number then find the
827 * first block in this inode and use that as a hint. If that
828 * block is also bogus then just don't worry about it.
830 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
832 em
= search_extent_mapping(em_tree
, 0, 0);
833 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
834 alloc_hint
= em
->block_start
;
838 alloc_hint
= em
->block_start
;
842 read_unlock(&em_tree
->lock
);
848 * when extent_io.c finds a delayed allocation range in the file,
849 * the call backs end up in this code. The basic idea is to
850 * allocate extents on disk for the range, and create ordered data structs
851 * in ram to track those extents.
853 * locked_page is the page that writepage had locked already. We use
854 * it to make sure we don't do extra locks or unlocks.
856 * *page_started is set to one if we unlock locked_page and do everything
857 * required to start IO on it. It may be clean and already done with
860 static noinline
int __cow_file_range(struct btrfs_trans_handle
*trans
,
862 struct btrfs_root
*root
,
863 struct page
*locked_page
,
864 u64 start
, u64 end
, int *page_started
,
865 unsigned long *nr_written
,
870 unsigned long ram_size
;
873 u64 blocksize
= root
->sectorsize
;
874 struct btrfs_key ins
;
875 struct extent_map
*em
;
876 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
879 BUG_ON(btrfs_is_free_space_inode(inode
));
881 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
882 num_bytes
= max(blocksize
, num_bytes
);
883 disk_num_bytes
= num_bytes
;
885 /* if this is a small write inside eof, kick off defrag */
886 if (num_bytes
< 64 * 1024 &&
887 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
888 btrfs_add_inode_defrag(trans
, inode
);
891 /* lets try to make an inline extent */
892 ret
= cow_file_range_inline(trans
, root
, inode
,
893 start
, end
, 0, 0, NULL
);
895 extent_clear_unlock_delalloc(inode
,
896 &BTRFS_I(inode
)->io_tree
,
898 EXTENT_CLEAR_UNLOCK_PAGE
|
899 EXTENT_CLEAR_UNLOCK
|
900 EXTENT_CLEAR_DELALLOC
|
902 EXTENT_SET_WRITEBACK
|
903 EXTENT_END_WRITEBACK
);
905 *nr_written
= *nr_written
+
906 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
909 } else if (ret
< 0) {
910 btrfs_abort_transaction(trans
, root
, ret
);
915 BUG_ON(disk_num_bytes
>
916 btrfs_super_total_bytes(root
->fs_info
->super_copy
));
918 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
919 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
921 while (disk_num_bytes
> 0) {
924 cur_alloc_size
= disk_num_bytes
;
925 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
926 root
->sectorsize
, 0, alloc_hint
,
929 btrfs_abort_transaction(trans
, root
, ret
);
933 em
= alloc_extent_map();
939 em
->orig_start
= em
->start
;
940 ram_size
= ins
.offset
;
941 em
->len
= ins
.offset
;
942 em
->mod_start
= em
->start
;
943 em
->mod_len
= em
->len
;
945 em
->block_start
= ins
.objectid
;
946 em
->block_len
= ins
.offset
;
947 em
->orig_block_len
= ins
.offset
;
948 em
->ram_bytes
= ram_size
;
949 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
950 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
954 write_lock(&em_tree
->lock
);
955 ret
= add_extent_mapping(em_tree
, em
, 1);
956 write_unlock(&em_tree
->lock
);
957 if (ret
!= -EEXIST
) {
961 btrfs_drop_extent_cache(inode
, start
,
962 start
+ ram_size
- 1, 0);
967 cur_alloc_size
= ins
.offset
;
968 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
969 ram_size
, cur_alloc_size
, 0);
973 if (root
->root_key
.objectid
==
974 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
975 ret
= btrfs_reloc_clone_csums(inode
, start
,
978 btrfs_abort_transaction(trans
, root
, ret
);
983 if (disk_num_bytes
< cur_alloc_size
)
986 /* we're not doing compressed IO, don't unlock the first
987 * page (which the caller expects to stay locked), don't
988 * clear any dirty bits and don't set any writeback bits
990 * Do set the Private2 bit so we know this page was properly
991 * setup for writepage
993 op
= unlock
? EXTENT_CLEAR_UNLOCK_PAGE
: 0;
994 op
|= EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
997 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
998 start
, start
+ ram_size
- 1,
1000 disk_num_bytes
-= cur_alloc_size
;
1001 num_bytes
-= cur_alloc_size
;
1002 alloc_hint
= ins
.objectid
+ ins
.offset
;
1003 start
+= cur_alloc_size
;
1009 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
1011 extent_clear_unlock_delalloc(inode
,
1012 &BTRFS_I(inode
)->io_tree
,
1013 start
, end
, locked_page
,
1014 EXTENT_CLEAR_UNLOCK_PAGE
|
1015 EXTENT_CLEAR_UNLOCK
|
1016 EXTENT_CLEAR_DELALLOC
|
1017 EXTENT_CLEAR_DIRTY
|
1018 EXTENT_SET_WRITEBACK
|
1019 EXTENT_END_WRITEBACK
);
1024 static noinline
int cow_file_range(struct inode
*inode
,
1025 struct page
*locked_page
,
1026 u64 start
, u64 end
, int *page_started
,
1027 unsigned long *nr_written
,
1030 struct btrfs_trans_handle
*trans
;
1031 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1034 trans
= btrfs_join_transaction(root
);
1035 if (IS_ERR(trans
)) {
1036 extent_clear_unlock_delalloc(inode
,
1037 &BTRFS_I(inode
)->io_tree
,
1038 start
, end
, locked_page
,
1039 EXTENT_CLEAR_UNLOCK_PAGE
|
1040 EXTENT_CLEAR_UNLOCK
|
1041 EXTENT_CLEAR_DELALLOC
|
1042 EXTENT_CLEAR_DIRTY
|
1043 EXTENT_SET_WRITEBACK
|
1044 EXTENT_END_WRITEBACK
);
1045 return PTR_ERR(trans
);
1047 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1049 ret
= __cow_file_range(trans
, inode
, root
, locked_page
, start
, end
,
1050 page_started
, nr_written
, unlock
);
1052 btrfs_end_transaction(trans
, root
);
1058 * work queue call back to started compression on a file and pages
1060 static noinline
void async_cow_start(struct btrfs_work
*work
)
1062 struct async_cow
*async_cow
;
1064 async_cow
= container_of(work
, struct async_cow
, work
);
1066 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
1067 async_cow
->start
, async_cow
->end
, async_cow
,
1069 if (num_added
== 0) {
1070 btrfs_add_delayed_iput(async_cow
->inode
);
1071 async_cow
->inode
= NULL
;
1076 * work queue call back to submit previously compressed pages
1078 static noinline
void async_cow_submit(struct btrfs_work
*work
)
1080 struct async_cow
*async_cow
;
1081 struct btrfs_root
*root
;
1082 unsigned long nr_pages
;
1084 async_cow
= container_of(work
, struct async_cow
, work
);
1086 root
= async_cow
->root
;
1087 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
1090 if (atomic_sub_return(nr_pages
, &root
->fs_info
->async_delalloc_pages
) <
1092 waitqueue_active(&root
->fs_info
->async_submit_wait
))
1093 wake_up(&root
->fs_info
->async_submit_wait
);
1095 if (async_cow
->inode
)
1096 submit_compressed_extents(async_cow
->inode
, async_cow
);
1099 static noinline
void async_cow_free(struct btrfs_work
*work
)
1101 struct async_cow
*async_cow
;
1102 async_cow
= container_of(work
, struct async_cow
, work
);
1103 if (async_cow
->inode
)
1104 btrfs_add_delayed_iput(async_cow
->inode
);
1108 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1109 u64 start
, u64 end
, int *page_started
,
1110 unsigned long *nr_written
)
1112 struct async_cow
*async_cow
;
1113 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1114 unsigned long nr_pages
;
1116 int limit
= 10 * 1024 * 1024;
1118 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1119 1, 0, NULL
, GFP_NOFS
);
1120 while (start
< end
) {
1121 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1122 BUG_ON(!async_cow
); /* -ENOMEM */
1123 async_cow
->inode
= igrab(inode
);
1124 async_cow
->root
= root
;
1125 async_cow
->locked_page
= locked_page
;
1126 async_cow
->start
= start
;
1128 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
1131 cur_end
= min(end
, start
+ 512 * 1024 - 1);
1133 async_cow
->end
= cur_end
;
1134 INIT_LIST_HEAD(&async_cow
->extents
);
1136 async_cow
->work
.func
= async_cow_start
;
1137 async_cow
->work
.ordered_func
= async_cow_submit
;
1138 async_cow
->work
.ordered_free
= async_cow_free
;
1139 async_cow
->work
.flags
= 0;
1141 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
1143 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
1145 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
1148 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
1149 wait_event(root
->fs_info
->async_submit_wait
,
1150 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
1154 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
1155 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
1156 wait_event(root
->fs_info
->async_submit_wait
,
1157 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
1161 *nr_written
+= nr_pages
;
1162 start
= cur_end
+ 1;
1168 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1169 u64 bytenr
, u64 num_bytes
)
1172 struct btrfs_ordered_sum
*sums
;
1175 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1176 bytenr
+ num_bytes
- 1, &list
, 0);
1177 if (ret
== 0 && list_empty(&list
))
1180 while (!list_empty(&list
)) {
1181 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1182 list_del(&sums
->list
);
1189 * when nowcow writeback call back. This checks for snapshots or COW copies
1190 * of the extents that exist in the file, and COWs the file as required.
1192 * If no cow copies or snapshots exist, we write directly to the existing
1195 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1196 struct page
*locked_page
,
1197 u64 start
, u64 end
, int *page_started
, int force
,
1198 unsigned long *nr_written
)
1200 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1201 struct btrfs_trans_handle
*trans
;
1202 struct extent_buffer
*leaf
;
1203 struct btrfs_path
*path
;
1204 struct btrfs_file_extent_item
*fi
;
1205 struct btrfs_key found_key
;
1220 u64 ino
= btrfs_ino(inode
);
1222 path
= btrfs_alloc_path();
1224 extent_clear_unlock_delalloc(inode
,
1225 &BTRFS_I(inode
)->io_tree
,
1226 start
, end
, locked_page
,
1227 EXTENT_CLEAR_UNLOCK_PAGE
|
1228 EXTENT_CLEAR_UNLOCK
|
1229 EXTENT_CLEAR_DELALLOC
|
1230 EXTENT_CLEAR_DIRTY
|
1231 EXTENT_SET_WRITEBACK
|
1232 EXTENT_END_WRITEBACK
);
1236 nolock
= btrfs_is_free_space_inode(inode
);
1239 trans
= btrfs_join_transaction_nolock(root
);
1241 trans
= btrfs_join_transaction(root
);
1243 if (IS_ERR(trans
)) {
1244 extent_clear_unlock_delalloc(inode
,
1245 &BTRFS_I(inode
)->io_tree
,
1246 start
, end
, locked_page
,
1247 EXTENT_CLEAR_UNLOCK_PAGE
|
1248 EXTENT_CLEAR_UNLOCK
|
1249 EXTENT_CLEAR_DELALLOC
|
1250 EXTENT_CLEAR_DIRTY
|
1251 EXTENT_SET_WRITEBACK
|
1252 EXTENT_END_WRITEBACK
);
1253 btrfs_free_path(path
);
1254 return PTR_ERR(trans
);
1257 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1259 cow_start
= (u64
)-1;
1262 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1265 btrfs_abort_transaction(trans
, root
, ret
);
1268 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1269 leaf
= path
->nodes
[0];
1270 btrfs_item_key_to_cpu(leaf
, &found_key
,
1271 path
->slots
[0] - 1);
1272 if (found_key
.objectid
== ino
&&
1273 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1278 leaf
= path
->nodes
[0];
1279 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1280 ret
= btrfs_next_leaf(root
, path
);
1282 btrfs_abort_transaction(trans
, root
, ret
);
1287 leaf
= path
->nodes
[0];
1293 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1295 if (found_key
.objectid
> ino
||
1296 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1297 found_key
.offset
> end
)
1300 if (found_key
.offset
> cur_offset
) {
1301 extent_end
= found_key
.offset
;
1306 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1307 struct btrfs_file_extent_item
);
1308 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1310 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
1311 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1312 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1313 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1314 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1315 extent_end
= found_key
.offset
+
1316 btrfs_file_extent_num_bytes(leaf
, fi
);
1318 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1319 if (extent_end
<= start
) {
1323 if (disk_bytenr
== 0)
1325 if (btrfs_file_extent_compression(leaf
, fi
) ||
1326 btrfs_file_extent_encryption(leaf
, fi
) ||
1327 btrfs_file_extent_other_encoding(leaf
, fi
))
1329 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1331 if (btrfs_extent_readonly(root
, disk_bytenr
))
1333 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1335 extent_offset
, disk_bytenr
))
1337 disk_bytenr
+= extent_offset
;
1338 disk_bytenr
+= cur_offset
- found_key
.offset
;
1339 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1341 * force cow if csum exists in the range.
1342 * this ensure that csum for a given extent are
1343 * either valid or do not exist.
1345 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1348 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1349 extent_end
= found_key
.offset
+
1350 btrfs_file_extent_inline_len(leaf
, fi
);
1351 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1356 if (extent_end
<= start
) {
1361 if (cow_start
== (u64
)-1)
1362 cow_start
= cur_offset
;
1363 cur_offset
= extent_end
;
1364 if (cur_offset
> end
)
1370 btrfs_release_path(path
);
1371 if (cow_start
!= (u64
)-1) {
1372 ret
= __cow_file_range(trans
, inode
, root
, locked_page
,
1373 cow_start
, found_key
.offset
- 1,
1374 page_started
, nr_written
, 1);
1376 btrfs_abort_transaction(trans
, root
, ret
);
1379 cow_start
= (u64
)-1;
1382 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1383 struct extent_map
*em
;
1384 struct extent_map_tree
*em_tree
;
1385 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1386 em
= alloc_extent_map();
1387 BUG_ON(!em
); /* -ENOMEM */
1388 em
->start
= cur_offset
;
1389 em
->orig_start
= found_key
.offset
- extent_offset
;
1390 em
->len
= num_bytes
;
1391 em
->block_len
= num_bytes
;
1392 em
->block_start
= disk_bytenr
;
1393 em
->orig_block_len
= disk_num_bytes
;
1394 em
->ram_bytes
= ram_bytes
;
1395 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1396 em
->mod_start
= em
->start
;
1397 em
->mod_len
= em
->len
;
1398 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1399 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
1400 em
->generation
= -1;
1402 write_lock(&em_tree
->lock
);
1403 ret
= add_extent_mapping(em_tree
, em
, 1);
1404 write_unlock(&em_tree
->lock
);
1405 if (ret
!= -EEXIST
) {
1406 free_extent_map(em
);
1409 btrfs_drop_extent_cache(inode
, em
->start
,
1410 em
->start
+ em
->len
- 1, 0);
1412 type
= BTRFS_ORDERED_PREALLOC
;
1414 type
= BTRFS_ORDERED_NOCOW
;
1417 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1418 num_bytes
, num_bytes
, type
);
1419 BUG_ON(ret
); /* -ENOMEM */
1421 if (root
->root_key
.objectid
==
1422 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1423 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1426 btrfs_abort_transaction(trans
, root
, ret
);
1431 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1432 cur_offset
, cur_offset
+ num_bytes
- 1,
1433 locked_page
, EXTENT_CLEAR_UNLOCK_PAGE
|
1434 EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
1435 EXTENT_SET_PRIVATE2
);
1436 cur_offset
= extent_end
;
1437 if (cur_offset
> end
)
1440 btrfs_release_path(path
);
1442 if (cur_offset
<= end
&& cow_start
== (u64
)-1) {
1443 cow_start
= cur_offset
;
1447 if (cow_start
!= (u64
)-1) {
1448 ret
= __cow_file_range(trans
, inode
, root
, locked_page
,
1450 page_started
, nr_written
, 1);
1452 btrfs_abort_transaction(trans
, root
, ret
);
1458 err
= btrfs_end_transaction(trans
, root
);
1462 if (ret
&& cur_offset
< end
)
1463 extent_clear_unlock_delalloc(inode
,
1464 &BTRFS_I(inode
)->io_tree
,
1465 cur_offset
, end
, locked_page
,
1466 EXTENT_CLEAR_UNLOCK_PAGE
|
1467 EXTENT_CLEAR_UNLOCK
|
1468 EXTENT_CLEAR_DELALLOC
|
1469 EXTENT_CLEAR_DIRTY
|
1470 EXTENT_SET_WRITEBACK
|
1471 EXTENT_END_WRITEBACK
);
1473 btrfs_free_path(path
);
1478 * extent_io.c call back to do delayed allocation processing
1480 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1481 u64 start
, u64 end
, int *page_started
,
1482 unsigned long *nr_written
)
1485 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1487 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) {
1488 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1489 page_started
, 1, nr_written
);
1490 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
) {
1491 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1492 page_started
, 0, nr_written
);
1493 } else if (!btrfs_test_opt(root
, COMPRESS
) &&
1494 !(BTRFS_I(inode
)->force_compress
) &&
1495 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
)) {
1496 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1497 page_started
, nr_written
, 1);
1499 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1500 &BTRFS_I(inode
)->runtime_flags
);
1501 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1502 page_started
, nr_written
);
1507 static void btrfs_split_extent_hook(struct inode
*inode
,
1508 struct extent_state
*orig
, u64 split
)
1510 /* not delalloc, ignore it */
1511 if (!(orig
->state
& EXTENT_DELALLOC
))
1514 spin_lock(&BTRFS_I(inode
)->lock
);
1515 BTRFS_I(inode
)->outstanding_extents
++;
1516 spin_unlock(&BTRFS_I(inode
)->lock
);
1520 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1521 * extents so we can keep track of new extents that are just merged onto old
1522 * extents, such as when we are doing sequential writes, so we can properly
1523 * account for the metadata space we'll need.
1525 static void btrfs_merge_extent_hook(struct inode
*inode
,
1526 struct extent_state
*new,
1527 struct extent_state
*other
)
1529 /* not delalloc, ignore it */
1530 if (!(other
->state
& EXTENT_DELALLOC
))
1533 spin_lock(&BTRFS_I(inode
)->lock
);
1534 BTRFS_I(inode
)->outstanding_extents
--;
1535 spin_unlock(&BTRFS_I(inode
)->lock
);
1538 static void btrfs_add_delalloc_inodes(struct btrfs_root
*root
,
1539 struct inode
*inode
)
1541 spin_lock(&root
->delalloc_lock
);
1542 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1543 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1544 &root
->delalloc_inodes
);
1545 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1546 &BTRFS_I(inode
)->runtime_flags
);
1547 root
->nr_delalloc_inodes
++;
1548 if (root
->nr_delalloc_inodes
== 1) {
1549 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1550 BUG_ON(!list_empty(&root
->delalloc_root
));
1551 list_add_tail(&root
->delalloc_root
,
1552 &root
->fs_info
->delalloc_roots
);
1553 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1556 spin_unlock(&root
->delalloc_lock
);
1559 static void btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1560 struct inode
*inode
)
1562 spin_lock(&root
->delalloc_lock
);
1563 if (!list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1564 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1565 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1566 &BTRFS_I(inode
)->runtime_flags
);
1567 root
->nr_delalloc_inodes
--;
1568 if (!root
->nr_delalloc_inodes
) {
1569 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1570 BUG_ON(list_empty(&root
->delalloc_root
));
1571 list_del_init(&root
->delalloc_root
);
1572 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1575 spin_unlock(&root
->delalloc_lock
);
1579 * extent_io.c set_bit_hook, used to track delayed allocation
1580 * bytes in this file, and to maintain the list of inodes that
1581 * have pending delalloc work to be done.
1583 static void btrfs_set_bit_hook(struct inode
*inode
,
1584 struct extent_state
*state
, unsigned long *bits
)
1588 * set_bit and clear bit hooks normally require _irqsave/restore
1589 * but in this case, we are only testing for the DELALLOC
1590 * bit, which is only set or cleared with irqs on
1592 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1593 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1594 u64 len
= state
->end
+ 1 - state
->start
;
1595 bool do_list
= !btrfs_is_free_space_inode(inode
);
1597 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1598 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1600 spin_lock(&BTRFS_I(inode
)->lock
);
1601 BTRFS_I(inode
)->outstanding_extents
++;
1602 spin_unlock(&BTRFS_I(inode
)->lock
);
1605 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, len
,
1606 root
->fs_info
->delalloc_batch
);
1607 spin_lock(&BTRFS_I(inode
)->lock
);
1608 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1609 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1610 &BTRFS_I(inode
)->runtime_flags
))
1611 btrfs_add_delalloc_inodes(root
, inode
);
1612 spin_unlock(&BTRFS_I(inode
)->lock
);
1617 * extent_io.c clear_bit_hook, see set_bit_hook for why
1619 static void btrfs_clear_bit_hook(struct inode
*inode
,
1620 struct extent_state
*state
,
1621 unsigned long *bits
)
1624 * set_bit and clear bit hooks normally require _irqsave/restore
1625 * but in this case, we are only testing for the DELALLOC
1626 * bit, which is only set or cleared with irqs on
1628 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1629 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1630 u64 len
= state
->end
+ 1 - state
->start
;
1631 bool do_list
= !btrfs_is_free_space_inode(inode
);
1633 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1634 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1635 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1636 spin_lock(&BTRFS_I(inode
)->lock
);
1637 BTRFS_I(inode
)->outstanding_extents
--;
1638 spin_unlock(&BTRFS_I(inode
)->lock
);
1641 if (*bits
& EXTENT_DO_ACCOUNTING
)
1642 btrfs_delalloc_release_metadata(inode
, len
);
1644 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1645 && do_list
&& !(state
->state
& EXTENT_NORESERVE
))
1646 btrfs_free_reserved_data_space(inode
, len
);
1648 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, -len
,
1649 root
->fs_info
->delalloc_batch
);
1650 spin_lock(&BTRFS_I(inode
)->lock
);
1651 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1652 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1653 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1654 &BTRFS_I(inode
)->runtime_flags
))
1655 btrfs_del_delalloc_inode(root
, inode
);
1656 spin_unlock(&BTRFS_I(inode
)->lock
);
1661 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1662 * we don't create bios that span stripes or chunks
1664 int btrfs_merge_bio_hook(int rw
, struct page
*page
, unsigned long offset
,
1665 size_t size
, struct bio
*bio
,
1666 unsigned long bio_flags
)
1668 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1669 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1674 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1677 length
= bio
->bi_size
;
1678 map_length
= length
;
1679 ret
= btrfs_map_block(root
->fs_info
, rw
, logical
,
1680 &map_length
, NULL
, 0);
1681 /* Will always return 0 with map_multi == NULL */
1683 if (map_length
< length
+ size
)
1689 * in order to insert checksums into the metadata in large chunks,
1690 * we wait until bio submission time. All the pages in the bio are
1691 * checksummed and sums are attached onto the ordered extent record.
1693 * At IO completion time the cums attached on the ordered extent record
1694 * are inserted into the btree
1696 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1697 struct bio
*bio
, int mirror_num
,
1698 unsigned long bio_flags
,
1701 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1704 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1705 BUG_ON(ret
); /* -ENOMEM */
1710 * in order to insert checksums into the metadata in large chunks,
1711 * we wait until bio submission time. All the pages in the bio are
1712 * checksummed and sums are attached onto the ordered extent record.
1714 * At IO completion time the cums attached on the ordered extent record
1715 * are inserted into the btree
1717 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1718 int mirror_num
, unsigned long bio_flags
,
1721 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1724 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1726 bio_endio(bio
, ret
);
1731 * extent_io.c submission hook. This does the right thing for csum calculation
1732 * on write, or reading the csums from the tree before a read
1734 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1735 int mirror_num
, unsigned long bio_flags
,
1738 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1742 int async
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
1744 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1746 if (btrfs_is_free_space_inode(inode
))
1749 if (!(rw
& REQ_WRITE
)) {
1750 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, metadata
);
1754 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1755 ret
= btrfs_submit_compressed_read(inode
, bio
,
1759 } else if (!skip_sum
) {
1760 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1765 } else if (async
&& !skip_sum
) {
1766 /* csum items have already been cloned */
1767 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1769 /* we're doing a write, do the async checksumming */
1770 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1771 inode
, rw
, bio
, mirror_num
,
1772 bio_flags
, bio_offset
,
1773 __btrfs_submit_bio_start
,
1774 __btrfs_submit_bio_done
);
1776 } else if (!skip_sum
) {
1777 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1783 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1787 bio_endio(bio
, ret
);
1792 * given a list of ordered sums record them in the inode. This happens
1793 * at IO completion time based on sums calculated at bio submission time.
1795 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1796 struct inode
*inode
, u64 file_offset
,
1797 struct list_head
*list
)
1799 struct btrfs_ordered_sum
*sum
;
1801 list_for_each_entry(sum
, list
, list
) {
1802 trans
->adding_csums
= 1;
1803 btrfs_csum_file_blocks(trans
,
1804 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1805 trans
->adding_csums
= 0;
1810 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1811 struct extent_state
**cached_state
)
1813 WARN_ON((end
& (PAGE_CACHE_SIZE
- 1)) == 0);
1814 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1815 cached_state
, GFP_NOFS
);
1818 /* see btrfs_writepage_start_hook for details on why this is required */
1819 struct btrfs_writepage_fixup
{
1821 struct btrfs_work work
;
1824 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1826 struct btrfs_writepage_fixup
*fixup
;
1827 struct btrfs_ordered_extent
*ordered
;
1828 struct extent_state
*cached_state
= NULL
;
1830 struct inode
*inode
;
1835 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1839 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1840 ClearPageChecked(page
);
1844 inode
= page
->mapping
->host
;
1845 page_start
= page_offset(page
);
1846 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1848 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1851 /* already ordered? We're done */
1852 if (PagePrivate2(page
))
1855 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1857 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1858 page_end
, &cached_state
, GFP_NOFS
);
1860 btrfs_start_ordered_extent(inode
, ordered
, 1);
1861 btrfs_put_ordered_extent(ordered
);
1865 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
1867 mapping_set_error(page
->mapping
, ret
);
1868 end_extent_writepage(page
, ret
, page_start
, page_end
);
1869 ClearPageChecked(page
);
1873 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
1874 ClearPageChecked(page
);
1875 set_page_dirty(page
);
1877 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1878 &cached_state
, GFP_NOFS
);
1881 page_cache_release(page
);
1886 * There are a few paths in the higher layers of the kernel that directly
1887 * set the page dirty bit without asking the filesystem if it is a
1888 * good idea. This causes problems because we want to make sure COW
1889 * properly happens and the data=ordered rules are followed.
1891 * In our case any range that doesn't have the ORDERED bit set
1892 * hasn't been properly setup for IO. We kick off an async process
1893 * to fix it up. The async helper will wait for ordered extents, set
1894 * the delalloc bit and make it safe to write the page.
1896 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1898 struct inode
*inode
= page
->mapping
->host
;
1899 struct btrfs_writepage_fixup
*fixup
;
1900 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1902 /* this page is properly in the ordered list */
1903 if (TestClearPagePrivate2(page
))
1906 if (PageChecked(page
))
1909 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1913 SetPageChecked(page
);
1914 page_cache_get(page
);
1915 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1917 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1921 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1922 struct inode
*inode
, u64 file_pos
,
1923 u64 disk_bytenr
, u64 disk_num_bytes
,
1924 u64 num_bytes
, u64 ram_bytes
,
1925 u8 compression
, u8 encryption
,
1926 u16 other_encoding
, int extent_type
)
1928 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1929 struct btrfs_file_extent_item
*fi
;
1930 struct btrfs_path
*path
;
1931 struct extent_buffer
*leaf
;
1932 struct btrfs_key ins
;
1935 path
= btrfs_alloc_path();
1939 path
->leave_spinning
= 1;
1942 * we may be replacing one extent in the tree with another.
1943 * The new extent is pinned in the extent map, and we don't want
1944 * to drop it from the cache until it is completely in the btree.
1946 * So, tell btrfs_drop_extents to leave this extent in the cache.
1947 * the caller is expected to unpin it and allow it to be merged
1950 ret
= btrfs_drop_extents(trans
, root
, inode
, file_pos
,
1951 file_pos
+ num_bytes
, 0);
1955 ins
.objectid
= btrfs_ino(inode
);
1956 ins
.offset
= file_pos
;
1957 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1958 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1961 leaf
= path
->nodes
[0];
1962 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1963 struct btrfs_file_extent_item
);
1964 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1965 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1966 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1967 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1968 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1969 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1970 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1971 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1972 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1973 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1975 btrfs_mark_buffer_dirty(leaf
);
1976 btrfs_release_path(path
);
1978 inode_add_bytes(inode
, num_bytes
);
1980 ins
.objectid
= disk_bytenr
;
1981 ins
.offset
= disk_num_bytes
;
1982 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1983 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1984 root
->root_key
.objectid
,
1985 btrfs_ino(inode
), file_pos
, &ins
);
1987 btrfs_free_path(path
);
1992 /* snapshot-aware defrag */
1993 struct sa_defrag_extent_backref
{
1994 struct rb_node node
;
1995 struct old_sa_defrag_extent
*old
;
2004 struct old_sa_defrag_extent
{
2005 struct list_head list
;
2006 struct new_sa_defrag_extent
*new;
2015 struct new_sa_defrag_extent
{
2016 struct rb_root root
;
2017 struct list_head head
;
2018 struct btrfs_path
*path
;
2019 struct inode
*inode
;
2027 static int backref_comp(struct sa_defrag_extent_backref
*b1
,
2028 struct sa_defrag_extent_backref
*b2
)
2030 if (b1
->root_id
< b2
->root_id
)
2032 else if (b1
->root_id
> b2
->root_id
)
2035 if (b1
->inum
< b2
->inum
)
2037 else if (b1
->inum
> b2
->inum
)
2040 if (b1
->file_pos
< b2
->file_pos
)
2042 else if (b1
->file_pos
> b2
->file_pos
)
2046 * [------------------------------] ===> (a range of space)
2047 * |<--->| |<---->| =============> (fs/file tree A)
2048 * |<---------------------------->| ===> (fs/file tree B)
2050 * A range of space can refer to two file extents in one tree while
2051 * refer to only one file extent in another tree.
2053 * So we may process a disk offset more than one time(two extents in A)
2054 * and locate at the same extent(one extent in B), then insert two same
2055 * backrefs(both refer to the extent in B).
2060 static void backref_insert(struct rb_root
*root
,
2061 struct sa_defrag_extent_backref
*backref
)
2063 struct rb_node
**p
= &root
->rb_node
;
2064 struct rb_node
*parent
= NULL
;
2065 struct sa_defrag_extent_backref
*entry
;
2070 entry
= rb_entry(parent
, struct sa_defrag_extent_backref
, node
);
2072 ret
= backref_comp(backref
, entry
);
2076 p
= &(*p
)->rb_right
;
2079 rb_link_node(&backref
->node
, parent
, p
);
2080 rb_insert_color(&backref
->node
, root
);
2084 * Note the backref might has changed, and in this case we just return 0.
2086 static noinline
int record_one_backref(u64 inum
, u64 offset
, u64 root_id
,
2089 struct btrfs_file_extent_item
*extent
;
2090 struct btrfs_fs_info
*fs_info
;
2091 struct old_sa_defrag_extent
*old
= ctx
;
2092 struct new_sa_defrag_extent
*new = old
->new;
2093 struct btrfs_path
*path
= new->path
;
2094 struct btrfs_key key
;
2095 struct btrfs_root
*root
;
2096 struct sa_defrag_extent_backref
*backref
;
2097 struct extent_buffer
*leaf
;
2098 struct inode
*inode
= new->inode
;
2104 if (BTRFS_I(inode
)->root
->root_key
.objectid
== root_id
&&
2105 inum
== btrfs_ino(inode
))
2108 key
.objectid
= root_id
;
2109 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2110 key
.offset
= (u64
)-1;
2112 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2113 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2115 if (PTR_ERR(root
) == -ENOENT
)
2118 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2119 inum
, offset
, root_id
);
2120 return PTR_ERR(root
);
2123 key
.objectid
= inum
;
2124 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2125 if (offset
> (u64
)-1 << 32)
2128 key
.offset
= offset
;
2130 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2139 leaf
= path
->nodes
[0];
2140 slot
= path
->slots
[0];
2142 if (slot
>= btrfs_header_nritems(leaf
)) {
2143 ret
= btrfs_next_leaf(root
, path
);
2146 } else if (ret
> 0) {
2155 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2157 if (key
.objectid
> inum
)
2160 if (key
.objectid
< inum
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2163 extent
= btrfs_item_ptr(leaf
, slot
,
2164 struct btrfs_file_extent_item
);
2166 if (btrfs_file_extent_disk_bytenr(leaf
, extent
) != old
->bytenr
)
2169 extent_offset
= btrfs_file_extent_offset(leaf
, extent
);
2170 if (key
.offset
- extent_offset
!= offset
)
2173 num_bytes
= btrfs_file_extent_num_bytes(leaf
, extent
);
2174 if (extent_offset
>= old
->extent_offset
+ old
->offset
+
2175 old
->len
|| extent_offset
+ num_bytes
<=
2176 old
->extent_offset
+ old
->offset
)
2182 backref
= kmalloc(sizeof(*backref
), GFP_NOFS
);
2188 backref
->root_id
= root_id
;
2189 backref
->inum
= inum
;
2190 backref
->file_pos
= offset
+ extent_offset
;
2191 backref
->num_bytes
= num_bytes
;
2192 backref
->extent_offset
= extent_offset
;
2193 backref
->generation
= btrfs_file_extent_generation(leaf
, extent
);
2195 backref_insert(&new->root
, backref
);
2198 btrfs_release_path(path
);
2203 static noinline
bool record_extent_backrefs(struct btrfs_path
*path
,
2204 struct new_sa_defrag_extent
*new)
2206 struct btrfs_fs_info
*fs_info
= BTRFS_I(new->inode
)->root
->fs_info
;
2207 struct old_sa_defrag_extent
*old
, *tmp
;
2212 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2213 ret
= iterate_inodes_from_logical(old
->bytenr
, fs_info
,
2214 path
, record_one_backref
,
2216 BUG_ON(ret
< 0 && ret
!= -ENOENT
);
2218 /* no backref to be processed for this extent */
2220 list_del(&old
->list
);
2225 if (list_empty(&new->head
))
2231 static int relink_is_mergable(struct extent_buffer
*leaf
,
2232 struct btrfs_file_extent_item
*fi
,
2235 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) != disk_bytenr
)
2238 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2241 if (btrfs_file_extent_compression(leaf
, fi
) ||
2242 btrfs_file_extent_encryption(leaf
, fi
) ||
2243 btrfs_file_extent_other_encoding(leaf
, fi
))
2250 * Note the backref might has changed, and in this case we just return 0.
2252 static noinline
int relink_extent_backref(struct btrfs_path
*path
,
2253 struct sa_defrag_extent_backref
*prev
,
2254 struct sa_defrag_extent_backref
*backref
)
2256 struct btrfs_file_extent_item
*extent
;
2257 struct btrfs_file_extent_item
*item
;
2258 struct btrfs_ordered_extent
*ordered
;
2259 struct btrfs_trans_handle
*trans
;
2260 struct btrfs_fs_info
*fs_info
;
2261 struct btrfs_root
*root
;
2262 struct btrfs_key key
;
2263 struct extent_buffer
*leaf
;
2264 struct old_sa_defrag_extent
*old
= backref
->old
;
2265 struct new_sa_defrag_extent
*new = old
->new;
2266 struct inode
*src_inode
= new->inode
;
2267 struct inode
*inode
;
2268 struct extent_state
*cached
= NULL
;
2277 if (prev
&& prev
->root_id
== backref
->root_id
&&
2278 prev
->inum
== backref
->inum
&&
2279 prev
->file_pos
+ prev
->num_bytes
== backref
->file_pos
)
2282 /* step 1: get root */
2283 key
.objectid
= backref
->root_id
;
2284 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2285 key
.offset
= (u64
)-1;
2287 fs_info
= BTRFS_I(src_inode
)->root
->fs_info
;
2288 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
2290 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2292 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2293 if (PTR_ERR(root
) == -ENOENT
)
2295 return PTR_ERR(root
);
2298 /* step 2: get inode */
2299 key
.objectid
= backref
->inum
;
2300 key
.type
= BTRFS_INODE_ITEM_KEY
;
2303 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
2304 if (IS_ERR(inode
)) {
2305 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2309 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2311 /* step 3: relink backref */
2312 lock_start
= backref
->file_pos
;
2313 lock_end
= backref
->file_pos
+ backref
->num_bytes
- 1;
2314 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2317 ordered
= btrfs_lookup_first_ordered_extent(inode
, lock_end
);
2319 btrfs_put_ordered_extent(ordered
);
2323 trans
= btrfs_join_transaction(root
);
2324 if (IS_ERR(trans
)) {
2325 ret
= PTR_ERR(trans
);
2329 key
.objectid
= backref
->inum
;
2330 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2331 key
.offset
= backref
->file_pos
;
2333 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2336 } else if (ret
> 0) {
2341 extent
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2342 struct btrfs_file_extent_item
);
2344 if (btrfs_file_extent_generation(path
->nodes
[0], extent
) !=
2345 backref
->generation
)
2348 btrfs_release_path(path
);
2350 start
= backref
->file_pos
;
2351 if (backref
->extent_offset
< old
->extent_offset
+ old
->offset
)
2352 start
+= old
->extent_offset
+ old
->offset
-
2353 backref
->extent_offset
;
2355 len
= min(backref
->extent_offset
+ backref
->num_bytes
,
2356 old
->extent_offset
+ old
->offset
+ old
->len
);
2357 len
-= max(backref
->extent_offset
, old
->extent_offset
+ old
->offset
);
2359 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
2364 key
.objectid
= btrfs_ino(inode
);
2365 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2368 path
->leave_spinning
= 1;
2370 struct btrfs_file_extent_item
*fi
;
2372 struct btrfs_key found_key
;
2374 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 1, 1);
2379 leaf
= path
->nodes
[0];
2380 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2382 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2383 struct btrfs_file_extent_item
);
2384 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
2386 if (relink_is_mergable(leaf
, fi
, new->bytenr
) &&
2387 extent_len
+ found_key
.offset
== start
) {
2388 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2390 btrfs_mark_buffer_dirty(leaf
);
2391 inode_add_bytes(inode
, len
);
2397 btrfs_release_path(path
);
2402 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2405 btrfs_abort_transaction(trans
, root
, ret
);
2409 leaf
= path
->nodes
[0];
2410 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2411 struct btrfs_file_extent_item
);
2412 btrfs_set_file_extent_disk_bytenr(leaf
, item
, new->bytenr
);
2413 btrfs_set_file_extent_disk_num_bytes(leaf
, item
, new->disk_len
);
2414 btrfs_set_file_extent_offset(leaf
, item
, start
- new->file_pos
);
2415 btrfs_set_file_extent_num_bytes(leaf
, item
, len
);
2416 btrfs_set_file_extent_ram_bytes(leaf
, item
, new->len
);
2417 btrfs_set_file_extent_generation(leaf
, item
, trans
->transid
);
2418 btrfs_set_file_extent_type(leaf
, item
, BTRFS_FILE_EXTENT_REG
);
2419 btrfs_set_file_extent_compression(leaf
, item
, new->compress_type
);
2420 btrfs_set_file_extent_encryption(leaf
, item
, 0);
2421 btrfs_set_file_extent_other_encoding(leaf
, item
, 0);
2423 btrfs_mark_buffer_dirty(leaf
);
2424 inode_add_bytes(inode
, len
);
2425 btrfs_release_path(path
);
2427 ret
= btrfs_inc_extent_ref(trans
, root
, new->bytenr
,
2429 backref
->root_id
, backref
->inum
,
2430 new->file_pos
, 0); /* start - extent_offset */
2432 btrfs_abort_transaction(trans
, root
, ret
);
2438 btrfs_release_path(path
);
2439 path
->leave_spinning
= 0;
2440 btrfs_end_transaction(trans
, root
);
2442 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2448 static void relink_file_extents(struct new_sa_defrag_extent
*new)
2450 struct btrfs_path
*path
;
2451 struct old_sa_defrag_extent
*old
, *tmp
;
2452 struct sa_defrag_extent_backref
*backref
;
2453 struct sa_defrag_extent_backref
*prev
= NULL
;
2454 struct inode
*inode
;
2455 struct btrfs_root
*root
;
2456 struct rb_node
*node
;
2460 root
= BTRFS_I(inode
)->root
;
2462 path
= btrfs_alloc_path();
2466 if (!record_extent_backrefs(path
, new)) {
2467 btrfs_free_path(path
);
2470 btrfs_release_path(path
);
2473 node
= rb_first(&new->root
);
2476 rb_erase(node
, &new->root
);
2478 backref
= rb_entry(node
, struct sa_defrag_extent_backref
, node
);
2480 ret
= relink_extent_backref(path
, prev
, backref
);
2493 btrfs_free_path(path
);
2495 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2496 list_del(&old
->list
);
2500 atomic_dec(&root
->fs_info
->defrag_running
);
2501 wake_up(&root
->fs_info
->transaction_wait
);
2506 static struct new_sa_defrag_extent
*
2507 record_old_file_extents(struct inode
*inode
,
2508 struct btrfs_ordered_extent
*ordered
)
2510 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2511 struct btrfs_path
*path
;
2512 struct btrfs_key key
;
2513 struct old_sa_defrag_extent
*old
, *tmp
;
2514 struct new_sa_defrag_extent
*new;
2517 new = kmalloc(sizeof(*new), GFP_NOFS
);
2522 new->file_pos
= ordered
->file_offset
;
2523 new->len
= ordered
->len
;
2524 new->bytenr
= ordered
->start
;
2525 new->disk_len
= ordered
->disk_len
;
2526 new->compress_type
= ordered
->compress_type
;
2527 new->root
= RB_ROOT
;
2528 INIT_LIST_HEAD(&new->head
);
2530 path
= btrfs_alloc_path();
2534 key
.objectid
= btrfs_ino(inode
);
2535 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2536 key
.offset
= new->file_pos
;
2538 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2541 if (ret
> 0 && path
->slots
[0] > 0)
2544 /* find out all the old extents for the file range */
2546 struct btrfs_file_extent_item
*extent
;
2547 struct extent_buffer
*l
;
2556 slot
= path
->slots
[0];
2558 if (slot
>= btrfs_header_nritems(l
)) {
2559 ret
= btrfs_next_leaf(root
, path
);
2567 btrfs_item_key_to_cpu(l
, &key
, slot
);
2569 if (key
.objectid
!= btrfs_ino(inode
))
2571 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2573 if (key
.offset
>= new->file_pos
+ new->len
)
2576 extent
= btrfs_item_ptr(l
, slot
, struct btrfs_file_extent_item
);
2578 num_bytes
= btrfs_file_extent_num_bytes(l
, extent
);
2579 if (key
.offset
+ num_bytes
< new->file_pos
)
2582 disk_bytenr
= btrfs_file_extent_disk_bytenr(l
, extent
);
2586 extent_offset
= btrfs_file_extent_offset(l
, extent
);
2588 old
= kmalloc(sizeof(*old
), GFP_NOFS
);
2592 offset
= max(new->file_pos
, key
.offset
);
2593 end
= min(new->file_pos
+ new->len
, key
.offset
+ num_bytes
);
2595 old
->bytenr
= disk_bytenr
;
2596 old
->extent_offset
= extent_offset
;
2597 old
->offset
= offset
- key
.offset
;
2598 old
->len
= end
- offset
;
2601 list_add_tail(&old
->list
, &new->head
);
2607 btrfs_free_path(path
);
2608 atomic_inc(&root
->fs_info
->defrag_running
);
2613 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2614 list_del(&old
->list
);
2618 btrfs_free_path(path
);
2625 * helper function for btrfs_finish_ordered_io, this
2626 * just reads in some of the csum leaves to prime them into ram
2627 * before we start the transaction. It limits the amount of btree
2628 * reads required while inside the transaction.
2630 /* as ordered data IO finishes, this gets called so we can finish
2631 * an ordered extent if the range of bytes in the file it covers are
2634 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
2636 struct inode
*inode
= ordered_extent
->inode
;
2637 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2638 struct btrfs_trans_handle
*trans
= NULL
;
2639 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2640 struct extent_state
*cached_state
= NULL
;
2641 struct new_sa_defrag_extent
*new = NULL
;
2642 int compress_type
= 0;
2646 nolock
= btrfs_is_free_space_inode(inode
);
2648 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
2653 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
2654 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
2655 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2657 trans
= btrfs_join_transaction_nolock(root
);
2659 trans
= btrfs_join_transaction(root
);
2660 if (IS_ERR(trans
)) {
2661 ret
= PTR_ERR(trans
);
2665 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2666 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2667 if (ret
) /* -ENOMEM or corruption */
2668 btrfs_abort_transaction(trans
, root
, ret
);
2672 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
2673 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2676 ret
= test_range_bit(io_tree
, ordered_extent
->file_offset
,
2677 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2678 EXTENT_DEFRAG
, 1, cached_state
);
2680 u64 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
2681 if (last_snapshot
>= BTRFS_I(inode
)->generation
)
2682 /* the inode is shared */
2683 new = record_old_file_extents(inode
, ordered_extent
);
2685 clear_extent_bit(io_tree
, ordered_extent
->file_offset
,
2686 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2687 EXTENT_DEFRAG
, 0, 0, &cached_state
, GFP_NOFS
);
2691 trans
= btrfs_join_transaction_nolock(root
);
2693 trans
= btrfs_join_transaction(root
);
2694 if (IS_ERR(trans
)) {
2695 ret
= PTR_ERR(trans
);
2699 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2701 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
2702 compress_type
= ordered_extent
->compress_type
;
2703 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
2704 BUG_ON(compress_type
);
2705 ret
= btrfs_mark_extent_written(trans
, inode
,
2706 ordered_extent
->file_offset
,
2707 ordered_extent
->file_offset
+
2708 ordered_extent
->len
);
2710 BUG_ON(root
== root
->fs_info
->tree_root
);
2711 ret
= insert_reserved_file_extent(trans
, inode
,
2712 ordered_extent
->file_offset
,
2713 ordered_extent
->start
,
2714 ordered_extent
->disk_len
,
2715 ordered_extent
->len
,
2716 ordered_extent
->len
,
2717 compress_type
, 0, 0,
2718 BTRFS_FILE_EXTENT_REG
);
2720 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
2721 ordered_extent
->file_offset
, ordered_extent
->len
,
2724 btrfs_abort_transaction(trans
, root
, ret
);
2728 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
2729 &ordered_extent
->list
);
2731 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2732 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2733 if (ret
) { /* -ENOMEM or corruption */
2734 btrfs_abort_transaction(trans
, root
, ret
);
2739 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
2740 ordered_extent
->file_offset
+
2741 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
2743 if (root
!= root
->fs_info
->tree_root
)
2744 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
2746 btrfs_end_transaction(trans
, root
);
2749 clear_extent_uptodate(io_tree
, ordered_extent
->file_offset
,
2750 ordered_extent
->file_offset
+
2751 ordered_extent
->len
- 1, NULL
, GFP_NOFS
);
2754 * If the ordered extent had an IOERR or something else went
2755 * wrong we need to return the space for this ordered extent
2756 * back to the allocator.
2758 if (!test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
2759 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
))
2760 btrfs_free_reserved_extent(root
, ordered_extent
->start
,
2761 ordered_extent
->disk_len
);
2766 * This needs to be done to make sure anybody waiting knows we are done
2767 * updating everything for this ordered extent.
2769 btrfs_remove_ordered_extent(inode
, ordered_extent
);
2771 /* for snapshot-aware defrag */
2773 relink_file_extents(new);
2776 btrfs_put_ordered_extent(ordered_extent
);
2777 /* once for the tree */
2778 btrfs_put_ordered_extent(ordered_extent
);
2783 static void finish_ordered_fn(struct btrfs_work
*work
)
2785 struct btrfs_ordered_extent
*ordered_extent
;
2786 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
2787 btrfs_finish_ordered_io(ordered_extent
);
2790 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
2791 struct extent_state
*state
, int uptodate
)
2793 struct inode
*inode
= page
->mapping
->host
;
2794 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2795 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
2796 struct btrfs_workers
*workers
;
2798 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
2800 ClearPagePrivate2(page
);
2801 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
2802 end
- start
+ 1, uptodate
))
2805 ordered_extent
->work
.func
= finish_ordered_fn
;
2806 ordered_extent
->work
.flags
= 0;
2808 if (btrfs_is_free_space_inode(inode
))
2809 workers
= &root
->fs_info
->endio_freespace_worker
;
2811 workers
= &root
->fs_info
->endio_write_workers
;
2812 btrfs_queue_worker(workers
, &ordered_extent
->work
);
2818 * when reads are done, we need to check csums to verify the data is correct
2819 * if there's a match, we allow the bio to finish. If not, the code in
2820 * extent_io.c will try to find good copies for us.
2822 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
2823 struct extent_state
*state
, int mirror
)
2825 size_t offset
= start
- page_offset(page
);
2826 struct inode
*inode
= page
->mapping
->host
;
2827 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2829 u64
private = ~(u32
)0;
2831 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2833 static DEFINE_RATELIMIT_STATE(_rs
, DEFAULT_RATELIMIT_INTERVAL
,
2834 DEFAULT_RATELIMIT_BURST
);
2836 if (PageChecked(page
)) {
2837 ClearPageChecked(page
);
2841 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
2844 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
2845 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
2846 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
2851 if (state
&& state
->start
== start
) {
2852 private = state
->private;
2855 ret
= get_state_private(io_tree
, start
, &private);
2857 kaddr
= kmap_atomic(page
);
2861 csum
= btrfs_csum_data(kaddr
+ offset
, csum
, end
- start
+ 1);
2862 btrfs_csum_final(csum
, (char *)&csum
);
2863 if (csum
!= private)
2866 kunmap_atomic(kaddr
);
2871 if (__ratelimit(&_rs
))
2872 btrfs_info(root
->fs_info
, "csum failed ino %llu off %llu csum %u private %llu",
2873 (unsigned long long)btrfs_ino(page
->mapping
->host
),
2874 (unsigned long long)start
, csum
,
2875 (unsigned long long)private);
2876 memset(kaddr
+ offset
, 1, end
- start
+ 1);
2877 flush_dcache_page(page
);
2878 kunmap_atomic(kaddr
);
2884 struct delayed_iput
{
2885 struct list_head list
;
2886 struct inode
*inode
;
2889 /* JDM: If this is fs-wide, why can't we add a pointer to
2890 * btrfs_inode instead and avoid the allocation? */
2891 void btrfs_add_delayed_iput(struct inode
*inode
)
2893 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2894 struct delayed_iput
*delayed
;
2896 if (atomic_add_unless(&inode
->i_count
, -1, 1))
2899 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
2900 delayed
->inode
= inode
;
2902 spin_lock(&fs_info
->delayed_iput_lock
);
2903 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
2904 spin_unlock(&fs_info
->delayed_iput_lock
);
2907 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
2910 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2911 struct delayed_iput
*delayed
;
2914 spin_lock(&fs_info
->delayed_iput_lock
);
2915 empty
= list_empty(&fs_info
->delayed_iputs
);
2916 spin_unlock(&fs_info
->delayed_iput_lock
);
2920 spin_lock(&fs_info
->delayed_iput_lock
);
2921 list_splice_init(&fs_info
->delayed_iputs
, &list
);
2922 spin_unlock(&fs_info
->delayed_iput_lock
);
2924 while (!list_empty(&list
)) {
2925 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
2926 list_del(&delayed
->list
);
2927 iput(delayed
->inode
);
2933 * This is called in transaction commit time. If there are no orphan
2934 * files in the subvolume, it removes orphan item and frees block_rsv
2937 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
2938 struct btrfs_root
*root
)
2940 struct btrfs_block_rsv
*block_rsv
;
2943 if (atomic_read(&root
->orphan_inodes
) ||
2944 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
2947 spin_lock(&root
->orphan_lock
);
2948 if (atomic_read(&root
->orphan_inodes
)) {
2949 spin_unlock(&root
->orphan_lock
);
2953 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
2954 spin_unlock(&root
->orphan_lock
);
2958 block_rsv
= root
->orphan_block_rsv
;
2959 root
->orphan_block_rsv
= NULL
;
2960 spin_unlock(&root
->orphan_lock
);
2962 if (root
->orphan_item_inserted
&&
2963 btrfs_root_refs(&root
->root_item
) > 0) {
2964 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
2965 root
->root_key
.objectid
);
2967 root
->orphan_item_inserted
= 0;
2971 WARN_ON(block_rsv
->size
> 0);
2972 btrfs_free_block_rsv(root
, block_rsv
);
2977 * This creates an orphan entry for the given inode in case something goes
2978 * wrong in the middle of an unlink/truncate.
2980 * NOTE: caller of this function should reserve 5 units of metadata for
2983 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2985 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2986 struct btrfs_block_rsv
*block_rsv
= NULL
;
2991 if (!root
->orphan_block_rsv
) {
2992 block_rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
2997 spin_lock(&root
->orphan_lock
);
2998 if (!root
->orphan_block_rsv
) {
2999 root
->orphan_block_rsv
= block_rsv
;
3000 } else if (block_rsv
) {
3001 btrfs_free_block_rsv(root
, block_rsv
);
3005 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3006 &BTRFS_I(inode
)->runtime_flags
)) {
3009 * For proper ENOSPC handling, we should do orphan
3010 * cleanup when mounting. But this introduces backward
3011 * compatibility issue.
3013 if (!xchg(&root
->orphan_item_inserted
, 1))
3019 atomic_inc(&root
->orphan_inodes
);
3022 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3023 &BTRFS_I(inode
)->runtime_flags
))
3025 spin_unlock(&root
->orphan_lock
);
3027 /* grab metadata reservation from transaction handle */
3029 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
3030 BUG_ON(ret
); /* -ENOSPC in reservation; Logic error? JDM */
3033 /* insert an orphan item to track this unlinked/truncated file */
3035 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
3036 if (ret
&& ret
!= -EEXIST
) {
3037 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3038 &BTRFS_I(inode
)->runtime_flags
);
3039 btrfs_abort_transaction(trans
, root
, ret
);
3045 /* insert an orphan item to track subvolume contains orphan files */
3047 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
3048 root
->root_key
.objectid
);
3049 if (ret
&& ret
!= -EEXIST
) {
3050 btrfs_abort_transaction(trans
, root
, ret
);
3058 * We have done the truncate/delete so we can go ahead and remove the orphan
3059 * item for this particular inode.
3061 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3062 struct inode
*inode
)
3064 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3065 int delete_item
= 0;
3066 int release_rsv
= 0;
3069 spin_lock(&root
->orphan_lock
);
3070 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3071 &BTRFS_I(inode
)->runtime_flags
))
3074 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3075 &BTRFS_I(inode
)->runtime_flags
))
3077 spin_unlock(&root
->orphan_lock
);
3079 if (trans
&& delete_item
) {
3080 ret
= btrfs_del_orphan_item(trans
, root
, btrfs_ino(inode
));
3081 BUG_ON(ret
); /* -ENOMEM or corruption (JDM: Recheck) */
3085 btrfs_orphan_release_metadata(inode
);
3086 atomic_dec(&root
->orphan_inodes
);
3093 * this cleans up any orphans that may be left on the list from the last use
3096 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3098 struct btrfs_path
*path
;
3099 struct extent_buffer
*leaf
;
3100 struct btrfs_key key
, found_key
;
3101 struct btrfs_trans_handle
*trans
;
3102 struct inode
*inode
;
3103 u64 last_objectid
= 0;
3104 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
3106 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
3109 path
= btrfs_alloc_path();
3116 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3117 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
3118 key
.offset
= (u64
)-1;
3121 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3126 * if ret == 0 means we found what we were searching for, which
3127 * is weird, but possible, so only screw with path if we didn't
3128 * find the key and see if we have stuff that matches
3132 if (path
->slots
[0] == 0)
3137 /* pull out the item */
3138 leaf
= path
->nodes
[0];
3139 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3141 /* make sure the item matches what we want */
3142 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3144 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
3147 /* release the path since we're done with it */
3148 btrfs_release_path(path
);
3151 * this is where we are basically btrfs_lookup, without the
3152 * crossing root thing. we store the inode number in the
3153 * offset of the orphan item.
3156 if (found_key
.offset
== last_objectid
) {
3157 btrfs_err(root
->fs_info
,
3158 "Error removing orphan entry, stopping orphan cleanup");
3163 last_objectid
= found_key
.offset
;
3165 found_key
.objectid
= found_key
.offset
;
3166 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3167 found_key
.offset
= 0;
3168 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
3169 ret
= PTR_RET(inode
);
3170 if (ret
&& ret
!= -ESTALE
)
3173 if (ret
== -ESTALE
&& root
== root
->fs_info
->tree_root
) {
3174 struct btrfs_root
*dead_root
;
3175 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3176 int is_dead_root
= 0;
3179 * this is an orphan in the tree root. Currently these
3180 * could come from 2 sources:
3181 * a) a snapshot deletion in progress
3182 * b) a free space cache inode
3183 * We need to distinguish those two, as the snapshot
3184 * orphan must not get deleted.
3185 * find_dead_roots already ran before us, so if this
3186 * is a snapshot deletion, we should find the root
3187 * in the dead_roots list
3189 spin_lock(&fs_info
->trans_lock
);
3190 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
3192 if (dead_root
->root_key
.objectid
==
3193 found_key
.objectid
) {
3198 spin_unlock(&fs_info
->trans_lock
);
3200 /* prevent this orphan from being found again */
3201 key
.offset
= found_key
.objectid
- 1;
3206 * Inode is already gone but the orphan item is still there,
3207 * kill the orphan item.
3209 if (ret
== -ESTALE
) {
3210 trans
= btrfs_start_transaction(root
, 1);
3211 if (IS_ERR(trans
)) {
3212 ret
= PTR_ERR(trans
);
3215 btrfs_debug(root
->fs_info
, "auto deleting %Lu",
3216 found_key
.objectid
);
3217 ret
= btrfs_del_orphan_item(trans
, root
,
3218 found_key
.objectid
);
3219 BUG_ON(ret
); /* -ENOMEM or corruption (JDM: Recheck) */
3220 btrfs_end_transaction(trans
, root
);
3225 * add this inode to the orphan list so btrfs_orphan_del does
3226 * the proper thing when we hit it
3228 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3229 &BTRFS_I(inode
)->runtime_flags
);
3230 atomic_inc(&root
->orphan_inodes
);
3232 /* if we have links, this was a truncate, lets do that */
3233 if (inode
->i_nlink
) {
3234 if (!S_ISREG(inode
->i_mode
)) {
3241 /* 1 for the orphan item deletion. */
3242 trans
= btrfs_start_transaction(root
, 1);
3243 if (IS_ERR(trans
)) {
3245 ret
= PTR_ERR(trans
);
3248 ret
= btrfs_orphan_add(trans
, inode
);
3249 btrfs_end_transaction(trans
, root
);
3255 ret
= btrfs_truncate(inode
);
3257 btrfs_orphan_del(NULL
, inode
);
3262 /* this will do delete_inode and everything for us */
3267 /* release the path since we're done with it */
3268 btrfs_release_path(path
);
3270 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
3272 if (root
->orphan_block_rsv
)
3273 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
3276 if (root
->orphan_block_rsv
|| root
->orphan_item_inserted
) {
3277 trans
= btrfs_join_transaction(root
);
3279 btrfs_end_transaction(trans
, root
);
3283 btrfs_debug(root
->fs_info
, "unlinked %d orphans", nr_unlink
);
3285 btrfs_debug(root
->fs_info
, "truncated %d orphans", nr_truncate
);
3289 btrfs_crit(root
->fs_info
,
3290 "could not do orphan cleanup %d", ret
);
3291 btrfs_free_path(path
);
3296 * very simple check to peek ahead in the leaf looking for xattrs. If we
3297 * don't find any xattrs, we know there can't be any acls.
3299 * slot is the slot the inode is in, objectid is the objectid of the inode
3301 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3302 int slot
, u64 objectid
)
3304 u32 nritems
= btrfs_header_nritems(leaf
);
3305 struct btrfs_key found_key
;
3306 static u64 xattr_access
= 0;
3307 static u64 xattr_default
= 0;
3310 if (!xattr_access
) {
3311 xattr_access
= btrfs_name_hash(POSIX_ACL_XATTR_ACCESS
,
3312 strlen(POSIX_ACL_XATTR_ACCESS
));
3313 xattr_default
= btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT
,
3314 strlen(POSIX_ACL_XATTR_DEFAULT
));
3318 while (slot
< nritems
) {
3319 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3321 /* we found a different objectid, there must not be acls */
3322 if (found_key
.objectid
!= objectid
)
3325 /* we found an xattr, assume we've got an acl */
3326 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
3327 if (found_key
.offset
== xattr_access
||
3328 found_key
.offset
== xattr_default
)
3333 * we found a key greater than an xattr key, there can't
3334 * be any acls later on
3336 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3343 * it goes inode, inode backrefs, xattrs, extents,
3344 * so if there are a ton of hard links to an inode there can
3345 * be a lot of backrefs. Don't waste time searching too hard,
3346 * this is just an optimization
3351 /* we hit the end of the leaf before we found an xattr or
3352 * something larger than an xattr. We have to assume the inode
3359 * read an inode from the btree into the in-memory inode
3361 static void btrfs_read_locked_inode(struct inode
*inode
)
3363 struct btrfs_path
*path
;
3364 struct extent_buffer
*leaf
;
3365 struct btrfs_inode_item
*inode_item
;
3366 struct btrfs_timespec
*tspec
;
3367 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3368 struct btrfs_key location
;
3372 bool filled
= false;
3374 ret
= btrfs_fill_inode(inode
, &rdev
);
3378 path
= btrfs_alloc_path();
3382 path
->leave_spinning
= 1;
3383 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3385 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3389 leaf
= path
->nodes
[0];
3394 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3395 struct btrfs_inode_item
);
3396 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3397 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3398 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3399 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3400 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
3402 tspec
= btrfs_inode_atime(inode_item
);
3403 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
3404 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
3406 tspec
= btrfs_inode_mtime(inode_item
);
3407 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
3408 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
3410 tspec
= btrfs_inode_ctime(inode_item
);
3411 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
3412 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
3414 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3415 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3416 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3419 * If we were modified in the current generation and evicted from memory
3420 * and then re-read we need to do a full sync since we don't have any
3421 * idea about which extents were modified before we were evicted from
3424 if (BTRFS_I(inode
)->last_trans
== root
->fs_info
->generation
)
3425 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3426 &BTRFS_I(inode
)->runtime_flags
);
3428 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
3429 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3431 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3433 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3434 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
3437 * try to precache a NULL acl entry for files that don't have
3438 * any xattrs or acls
3440 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
3443 cache_no_acl(inode
);
3445 btrfs_free_path(path
);
3447 switch (inode
->i_mode
& S_IFMT
) {
3449 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3450 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
3451 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3452 inode
->i_fop
= &btrfs_file_operations
;
3453 inode
->i_op
= &btrfs_file_inode_operations
;
3456 inode
->i_fop
= &btrfs_dir_file_operations
;
3457 if (root
== root
->fs_info
->tree_root
)
3458 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
3460 inode
->i_op
= &btrfs_dir_inode_operations
;
3463 inode
->i_op
= &btrfs_symlink_inode_operations
;
3464 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
3465 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
3468 inode
->i_op
= &btrfs_special_inode_operations
;
3469 init_special_inode(inode
, inode
->i_mode
, rdev
);
3473 btrfs_update_iflags(inode
);
3477 btrfs_free_path(path
);
3478 make_bad_inode(inode
);
3482 * given a leaf and an inode, copy the inode fields into the leaf
3484 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3485 struct extent_buffer
*leaf
,
3486 struct btrfs_inode_item
*item
,
3487 struct inode
*inode
)
3489 struct btrfs_map_token token
;
3491 btrfs_init_map_token(&token
);
3493 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3494 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3495 btrfs_set_token_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
,
3497 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3498 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3500 btrfs_set_token_timespec_sec(leaf
, btrfs_inode_atime(item
),
3501 inode
->i_atime
.tv_sec
, &token
);
3502 btrfs_set_token_timespec_nsec(leaf
, btrfs_inode_atime(item
),
3503 inode
->i_atime
.tv_nsec
, &token
);
3505 btrfs_set_token_timespec_sec(leaf
, btrfs_inode_mtime(item
),
3506 inode
->i_mtime
.tv_sec
, &token
);
3507 btrfs_set_token_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
3508 inode
->i_mtime
.tv_nsec
, &token
);
3510 btrfs_set_token_timespec_sec(leaf
, btrfs_inode_ctime(item
),
3511 inode
->i_ctime
.tv_sec
, &token
);
3512 btrfs_set_token_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
3513 inode
->i_ctime
.tv_nsec
, &token
);
3515 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3517 btrfs_set_token_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
,
3519 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
3520 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3521 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3522 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3523 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3527 * copy everything in the in-memory inode into the btree.
3529 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
3530 struct btrfs_root
*root
, struct inode
*inode
)
3532 struct btrfs_inode_item
*inode_item
;
3533 struct btrfs_path
*path
;
3534 struct extent_buffer
*leaf
;
3537 path
= btrfs_alloc_path();
3541 path
->leave_spinning
= 1;
3542 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
3550 btrfs_unlock_up_safe(path
, 1);
3551 leaf
= path
->nodes
[0];
3552 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3553 struct btrfs_inode_item
);
3555 fill_inode_item(trans
, leaf
, inode_item
, inode
);
3556 btrfs_mark_buffer_dirty(leaf
);
3557 btrfs_set_inode_last_trans(trans
, inode
);
3560 btrfs_free_path(path
);
3565 * copy everything in the in-memory inode into the btree.
3567 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
3568 struct btrfs_root
*root
, struct inode
*inode
)
3573 * If the inode is a free space inode, we can deadlock during commit
3574 * if we put it into the delayed code.
3576 * The data relocation inode should also be directly updated
3579 if (!btrfs_is_free_space_inode(inode
)
3580 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
) {
3581 btrfs_update_root_times(trans
, root
);
3583 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
3585 btrfs_set_inode_last_trans(trans
, inode
);
3589 return btrfs_update_inode_item(trans
, root
, inode
);
3592 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
3593 struct btrfs_root
*root
,
3594 struct inode
*inode
)
3598 ret
= btrfs_update_inode(trans
, root
, inode
);
3600 return btrfs_update_inode_item(trans
, root
, inode
);
3605 * unlink helper that gets used here in inode.c and in the tree logging
3606 * recovery code. It remove a link in a directory with a given name, and
3607 * also drops the back refs in the inode to the directory
3609 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3610 struct btrfs_root
*root
,
3611 struct inode
*dir
, struct inode
*inode
,
3612 const char *name
, int name_len
)
3614 struct btrfs_path
*path
;
3616 struct extent_buffer
*leaf
;
3617 struct btrfs_dir_item
*di
;
3618 struct btrfs_key key
;
3620 u64 ino
= btrfs_ino(inode
);
3621 u64 dir_ino
= btrfs_ino(dir
);
3623 path
= btrfs_alloc_path();
3629 path
->leave_spinning
= 1;
3630 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3631 name
, name_len
, -1);
3640 leaf
= path
->nodes
[0];
3641 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3642 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3645 btrfs_release_path(path
);
3647 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
3650 btrfs_info(root
->fs_info
,
3651 "failed to delete reference to %.*s, inode %llu parent %llu",
3653 (unsigned long long)ino
, (unsigned long long)dir_ino
);
3654 btrfs_abort_transaction(trans
, root
, ret
);
3658 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3660 btrfs_abort_transaction(trans
, root
, ret
);
3664 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
3666 if (ret
!= 0 && ret
!= -ENOENT
) {
3667 btrfs_abort_transaction(trans
, root
, ret
);
3671 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
3676 btrfs_abort_transaction(trans
, root
, ret
);
3678 btrfs_free_path(path
);
3682 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3683 inode_inc_iversion(inode
);
3684 inode_inc_iversion(dir
);
3685 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3686 ret
= btrfs_update_inode(trans
, root
, dir
);
3691 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3692 struct btrfs_root
*root
,
3693 struct inode
*dir
, struct inode
*inode
,
3694 const char *name
, int name_len
)
3697 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
3699 btrfs_drop_nlink(inode
);
3700 ret
= btrfs_update_inode(trans
, root
, inode
);
3706 * helper to start transaction for unlink and rmdir.
3708 * unlink and rmdir are special in btrfs, they do not always free space, so
3709 * if we cannot make our reservations the normal way try and see if there is
3710 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3711 * allow the unlink to occur.
3713 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
)
3715 struct btrfs_trans_handle
*trans
;
3716 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3720 * 1 for the possible orphan item
3721 * 1 for the dir item
3722 * 1 for the dir index
3723 * 1 for the inode ref
3726 trans
= btrfs_start_transaction(root
, 5);
3727 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
3730 if (PTR_ERR(trans
) == -ENOSPC
) {
3731 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 5);
3733 trans
= btrfs_start_transaction(root
, 0);
3736 ret
= btrfs_cond_migrate_bytes(root
->fs_info
,
3737 &root
->fs_info
->trans_block_rsv
,
3740 btrfs_end_transaction(trans
, root
);
3741 return ERR_PTR(ret
);
3743 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3744 trans
->bytes_reserved
= num_bytes
;
3749 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3751 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3752 struct btrfs_trans_handle
*trans
;
3753 struct inode
*inode
= dentry
->d_inode
;
3756 trans
= __unlink_start_trans(dir
);
3758 return PTR_ERR(trans
);
3760 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
3762 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3763 dentry
->d_name
.name
, dentry
->d_name
.len
);
3767 if (inode
->i_nlink
== 0) {
3768 ret
= btrfs_orphan_add(trans
, inode
);
3774 btrfs_end_transaction(trans
, root
);
3775 btrfs_btree_balance_dirty(root
);
3779 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
3780 struct btrfs_root
*root
,
3781 struct inode
*dir
, u64 objectid
,
3782 const char *name
, int name_len
)
3784 struct btrfs_path
*path
;
3785 struct extent_buffer
*leaf
;
3786 struct btrfs_dir_item
*di
;
3787 struct btrfs_key key
;
3790 u64 dir_ino
= btrfs_ino(dir
);
3792 path
= btrfs_alloc_path();
3796 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3797 name
, name_len
, -1);
3798 if (IS_ERR_OR_NULL(di
)) {
3806 leaf
= path
->nodes
[0];
3807 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3808 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
3809 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3811 btrfs_abort_transaction(trans
, root
, ret
);
3814 btrfs_release_path(path
);
3816 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
3817 objectid
, root
->root_key
.objectid
,
3818 dir_ino
, &index
, name
, name_len
);
3820 if (ret
!= -ENOENT
) {
3821 btrfs_abort_transaction(trans
, root
, ret
);
3824 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
3826 if (IS_ERR_OR_NULL(di
)) {
3831 btrfs_abort_transaction(trans
, root
, ret
);
3835 leaf
= path
->nodes
[0];
3836 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3837 btrfs_release_path(path
);
3840 btrfs_release_path(path
);
3842 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3844 btrfs_abort_transaction(trans
, root
, ret
);
3848 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3849 inode_inc_iversion(dir
);
3850 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3851 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
3853 btrfs_abort_transaction(trans
, root
, ret
);
3855 btrfs_free_path(path
);
3859 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3861 struct inode
*inode
= dentry
->d_inode
;
3863 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3864 struct btrfs_trans_handle
*trans
;
3866 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
3868 if (btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
3871 trans
= __unlink_start_trans(dir
);
3873 return PTR_ERR(trans
);
3875 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
3876 err
= btrfs_unlink_subvol(trans
, root
, dir
,
3877 BTRFS_I(inode
)->location
.objectid
,
3878 dentry
->d_name
.name
,
3879 dentry
->d_name
.len
);
3883 err
= btrfs_orphan_add(trans
, inode
);
3887 /* now the directory is empty */
3888 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3889 dentry
->d_name
.name
, dentry
->d_name
.len
);
3891 btrfs_i_size_write(inode
, 0);
3893 btrfs_end_transaction(trans
, root
);
3894 btrfs_btree_balance_dirty(root
);
3900 * this can truncate away extent items, csum items and directory items.
3901 * It starts at a high offset and removes keys until it can't find
3902 * any higher than new_size
3904 * csum items that cross the new i_size are truncated to the new size
3907 * min_type is the minimum key type to truncate down to. If set to 0, this
3908 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3910 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
3911 struct btrfs_root
*root
,
3912 struct inode
*inode
,
3913 u64 new_size
, u32 min_type
)
3915 struct btrfs_path
*path
;
3916 struct extent_buffer
*leaf
;
3917 struct btrfs_file_extent_item
*fi
;
3918 struct btrfs_key key
;
3919 struct btrfs_key found_key
;
3920 u64 extent_start
= 0;
3921 u64 extent_num_bytes
= 0;
3922 u64 extent_offset
= 0;
3924 u32 found_type
= (u8
)-1;
3927 int pending_del_nr
= 0;
3928 int pending_del_slot
= 0;
3929 int extent_type
= -1;
3932 u64 ino
= btrfs_ino(inode
);
3934 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
3936 path
= btrfs_alloc_path();
3942 * We want to drop from the next block forward in case this new size is
3943 * not block aligned since we will be keeping the last block of the
3944 * extent just the way it is.
3946 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
3947 btrfs_drop_extent_cache(inode
, ALIGN(new_size
,
3948 root
->sectorsize
), (u64
)-1, 0);
3951 * This function is also used to drop the items in the log tree before
3952 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3953 * it is used to drop the loged items. So we shouldn't kill the delayed
3956 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
3957 btrfs_kill_delayed_inode_items(inode
);
3960 key
.offset
= (u64
)-1;
3964 path
->leave_spinning
= 1;
3965 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3972 /* there are no items in the tree for us to truncate, we're
3975 if (path
->slots
[0] == 0)
3982 leaf
= path
->nodes
[0];
3983 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3984 found_type
= btrfs_key_type(&found_key
);
3986 if (found_key
.objectid
!= ino
)
3989 if (found_type
< min_type
)
3992 item_end
= found_key
.offset
;
3993 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
3994 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3995 struct btrfs_file_extent_item
);
3996 extent_type
= btrfs_file_extent_type(leaf
, fi
);
3997 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3999 btrfs_file_extent_num_bytes(leaf
, fi
);
4000 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4001 item_end
+= btrfs_file_extent_inline_len(leaf
,
4006 if (found_type
> min_type
) {
4009 if (item_end
< new_size
)
4011 if (found_key
.offset
>= new_size
)
4017 /* FIXME, shrink the extent if the ref count is only 1 */
4018 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
4021 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4023 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4025 u64 orig_num_bytes
=
4026 btrfs_file_extent_num_bytes(leaf
, fi
);
4027 extent_num_bytes
= ALIGN(new_size
-
4030 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4032 num_dec
= (orig_num_bytes
-
4034 if (root
->ref_cows
&& extent_start
!= 0)
4035 inode_sub_bytes(inode
, num_dec
);
4036 btrfs_mark_buffer_dirty(leaf
);
4039 btrfs_file_extent_disk_num_bytes(leaf
,
4041 extent_offset
= found_key
.offset
-
4042 btrfs_file_extent_offset(leaf
, fi
);
4044 /* FIXME blocksize != 4096 */
4045 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
4046 if (extent_start
!= 0) {
4049 inode_sub_bytes(inode
, num_dec
);
4052 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4054 * we can't truncate inline items that have had
4058 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
4059 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
4060 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
4061 u32 size
= new_size
- found_key
.offset
;
4063 if (root
->ref_cows
) {
4064 inode_sub_bytes(inode
, item_end
+ 1 -
4068 btrfs_file_extent_calc_inline_size(size
);
4069 btrfs_truncate_item(root
, path
, size
, 1);
4070 } else if (root
->ref_cows
) {
4071 inode_sub_bytes(inode
, item_end
+ 1 -
4077 if (!pending_del_nr
) {
4078 /* no pending yet, add ourselves */
4079 pending_del_slot
= path
->slots
[0];
4081 } else if (pending_del_nr
&&
4082 path
->slots
[0] + 1 == pending_del_slot
) {
4083 /* hop on the pending chunk */
4085 pending_del_slot
= path
->slots
[0];
4092 if (found_extent
&& (root
->ref_cows
||
4093 root
== root
->fs_info
->tree_root
)) {
4094 btrfs_set_path_blocking(path
);
4095 ret
= btrfs_free_extent(trans
, root
, extent_start
,
4096 extent_num_bytes
, 0,
4097 btrfs_header_owner(leaf
),
4098 ino
, extent_offset
, 0);
4102 if (found_type
== BTRFS_INODE_ITEM_KEY
)
4105 if (path
->slots
[0] == 0 ||
4106 path
->slots
[0] != pending_del_slot
) {
4107 if (pending_del_nr
) {
4108 ret
= btrfs_del_items(trans
, root
, path
,
4112 btrfs_abort_transaction(trans
,
4118 btrfs_release_path(path
);
4125 if (pending_del_nr
) {
4126 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
4129 btrfs_abort_transaction(trans
, root
, ret
);
4132 btrfs_free_path(path
);
4137 * btrfs_truncate_page - read, zero a chunk and write a page
4138 * @inode - inode that we're zeroing
4139 * @from - the offset to start zeroing
4140 * @len - the length to zero, 0 to zero the entire range respective to the
4142 * @front - zero up to the offset instead of from the offset on
4144 * This will find the page for the "from" offset and cow the page and zero the
4145 * part we want to zero. This is used with truncate and hole punching.
4147 int btrfs_truncate_page(struct inode
*inode
, loff_t from
, loff_t len
,
4150 struct address_space
*mapping
= inode
->i_mapping
;
4151 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4152 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4153 struct btrfs_ordered_extent
*ordered
;
4154 struct extent_state
*cached_state
= NULL
;
4156 u32 blocksize
= root
->sectorsize
;
4157 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
4158 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
4160 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4165 if ((offset
& (blocksize
- 1)) == 0 &&
4166 (!len
|| ((len
& (blocksize
- 1)) == 0)))
4168 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
4173 page
= find_or_create_page(mapping
, index
, mask
);
4175 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
4180 page_start
= page_offset(page
);
4181 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4183 if (!PageUptodate(page
)) {
4184 ret
= btrfs_readpage(NULL
, page
);
4186 if (page
->mapping
!= mapping
) {
4188 page_cache_release(page
);
4191 if (!PageUptodate(page
)) {
4196 wait_on_page_writeback(page
);
4198 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
4199 set_page_extent_mapped(page
);
4201 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
4203 unlock_extent_cached(io_tree
, page_start
, page_end
,
4204 &cached_state
, GFP_NOFS
);
4206 page_cache_release(page
);
4207 btrfs_start_ordered_extent(inode
, ordered
, 1);
4208 btrfs_put_ordered_extent(ordered
);
4212 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
4213 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4214 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4215 0, 0, &cached_state
, GFP_NOFS
);
4217 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
4220 unlock_extent_cached(io_tree
, page_start
, page_end
,
4221 &cached_state
, GFP_NOFS
);
4225 if (offset
!= PAGE_CACHE_SIZE
) {
4227 len
= PAGE_CACHE_SIZE
- offset
;
4230 memset(kaddr
, 0, offset
);
4232 memset(kaddr
+ offset
, 0, len
);
4233 flush_dcache_page(page
);
4236 ClearPageChecked(page
);
4237 set_page_dirty(page
);
4238 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
4243 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
4245 page_cache_release(page
);
4251 * This function puts in dummy file extents for the area we're creating a hole
4252 * for. So if we are truncating this file to a larger size we need to insert
4253 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4254 * the range between oldsize and size
4256 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
4258 struct btrfs_trans_handle
*trans
;
4259 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4260 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4261 struct extent_map
*em
= NULL
;
4262 struct extent_state
*cached_state
= NULL
;
4263 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4264 u64 hole_start
= ALIGN(oldsize
, root
->sectorsize
);
4265 u64 block_end
= ALIGN(size
, root
->sectorsize
);
4272 * If our size started in the middle of a page we need to zero out the
4273 * rest of the page before we expand the i_size, otherwise we could
4274 * expose stale data.
4276 err
= btrfs_truncate_page(inode
, oldsize
, 0, 0);
4280 if (size
<= hole_start
)
4284 struct btrfs_ordered_extent
*ordered
;
4285 btrfs_wait_ordered_range(inode
, hole_start
,
4286 block_end
- hole_start
);
4287 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
4289 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
4292 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
4293 &cached_state
, GFP_NOFS
);
4294 btrfs_put_ordered_extent(ordered
);
4297 cur_offset
= hole_start
;
4299 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4300 block_end
- cur_offset
, 0);
4306 last_byte
= min(extent_map_end(em
), block_end
);
4307 last_byte
= ALIGN(last_byte
, root
->sectorsize
);
4308 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
4309 struct extent_map
*hole_em
;
4310 hole_size
= last_byte
- cur_offset
;
4312 trans
= btrfs_start_transaction(root
, 3);
4313 if (IS_ERR(trans
)) {
4314 err
= PTR_ERR(trans
);
4318 err
= btrfs_drop_extents(trans
, root
, inode
,
4320 cur_offset
+ hole_size
, 1);
4322 btrfs_abort_transaction(trans
, root
, err
);
4323 btrfs_end_transaction(trans
, root
);
4327 err
= btrfs_insert_file_extent(trans
, root
,
4328 btrfs_ino(inode
), cur_offset
, 0,
4329 0, hole_size
, 0, hole_size
,
4332 btrfs_abort_transaction(trans
, root
, err
);
4333 btrfs_end_transaction(trans
, root
);
4337 btrfs_drop_extent_cache(inode
, cur_offset
,
4338 cur_offset
+ hole_size
- 1, 0);
4339 hole_em
= alloc_extent_map();
4341 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
4342 &BTRFS_I(inode
)->runtime_flags
);
4345 hole_em
->start
= cur_offset
;
4346 hole_em
->len
= hole_size
;
4347 hole_em
->orig_start
= cur_offset
;
4349 hole_em
->block_start
= EXTENT_MAP_HOLE
;
4350 hole_em
->block_len
= 0;
4351 hole_em
->orig_block_len
= 0;
4352 hole_em
->ram_bytes
= hole_size
;
4353 hole_em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4354 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
4355 hole_em
->generation
= trans
->transid
;
4358 write_lock(&em_tree
->lock
);
4359 err
= add_extent_mapping(em_tree
, hole_em
, 1);
4360 write_unlock(&em_tree
->lock
);
4363 btrfs_drop_extent_cache(inode
, cur_offset
,
4367 free_extent_map(hole_em
);
4369 btrfs_update_inode(trans
, root
, inode
);
4370 btrfs_end_transaction(trans
, root
);
4372 free_extent_map(em
);
4374 cur_offset
= last_byte
;
4375 if (cur_offset
>= block_end
)
4379 free_extent_map(em
);
4380 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
4385 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
4387 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4388 struct btrfs_trans_handle
*trans
;
4389 loff_t oldsize
= i_size_read(inode
);
4390 loff_t newsize
= attr
->ia_size
;
4391 int mask
= attr
->ia_valid
;
4394 if (newsize
== oldsize
)
4398 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4399 * special case where we need to update the times despite not having
4400 * these flags set. For all other operations the VFS set these flags
4401 * explicitly if it wants a timestamp update.
4403 if (newsize
!= oldsize
&& (!(mask
& (ATTR_CTIME
| ATTR_MTIME
))))
4404 inode
->i_ctime
= inode
->i_mtime
= current_fs_time(inode
->i_sb
);
4406 if (newsize
> oldsize
) {
4407 truncate_pagecache(inode
, oldsize
, newsize
);
4408 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
4412 trans
= btrfs_start_transaction(root
, 1);
4414 return PTR_ERR(trans
);
4416 i_size_write(inode
, newsize
);
4417 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
4418 ret
= btrfs_update_inode(trans
, root
, inode
);
4419 btrfs_end_transaction(trans
, root
);
4423 * We're truncating a file that used to have good data down to
4424 * zero. Make sure it gets into the ordered flush list so that
4425 * any new writes get down to disk quickly.
4428 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
4429 &BTRFS_I(inode
)->runtime_flags
);
4432 * 1 for the orphan item we're going to add
4433 * 1 for the orphan item deletion.
4435 trans
= btrfs_start_transaction(root
, 2);
4437 return PTR_ERR(trans
);
4440 * We need to do this in case we fail at _any_ point during the
4441 * actual truncate. Once we do the truncate_setsize we could
4442 * invalidate pages which forces any outstanding ordered io to
4443 * be instantly completed which will give us extents that need
4444 * to be truncated. If we fail to get an orphan inode down we
4445 * could have left over extents that were never meant to live,
4446 * so we need to garuntee from this point on that everything
4447 * will be consistent.
4449 ret
= btrfs_orphan_add(trans
, inode
);
4450 btrfs_end_transaction(trans
, root
);
4454 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4455 truncate_setsize(inode
, newsize
);
4457 /* Disable nonlocked read DIO to avoid the end less truncate */
4458 btrfs_inode_block_unlocked_dio(inode
);
4459 inode_dio_wait(inode
);
4460 btrfs_inode_resume_unlocked_dio(inode
);
4462 ret
= btrfs_truncate(inode
);
4463 if (ret
&& inode
->i_nlink
)
4464 btrfs_orphan_del(NULL
, inode
);
4470 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4472 struct inode
*inode
= dentry
->d_inode
;
4473 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4476 if (btrfs_root_readonly(root
))
4479 err
= inode_change_ok(inode
, attr
);
4483 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
4484 err
= btrfs_setsize(inode
, attr
);
4489 if (attr
->ia_valid
) {
4490 setattr_copy(inode
, attr
);
4491 inode_inc_iversion(inode
);
4492 err
= btrfs_dirty_inode(inode
);
4494 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
4495 err
= btrfs_acl_chmod(inode
);
4501 void btrfs_evict_inode(struct inode
*inode
)
4503 struct btrfs_trans_handle
*trans
;
4504 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4505 struct btrfs_block_rsv
*rsv
, *global_rsv
;
4506 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
4509 trace_btrfs_inode_evict(inode
);
4511 truncate_inode_pages(&inode
->i_data
, 0);
4512 if (inode
->i_nlink
&& (btrfs_root_refs(&root
->root_item
) != 0 ||
4513 btrfs_is_free_space_inode(inode
)))
4516 if (is_bad_inode(inode
)) {
4517 btrfs_orphan_del(NULL
, inode
);
4520 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4521 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
4523 if (root
->fs_info
->log_root_recovering
) {
4524 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
4525 &BTRFS_I(inode
)->runtime_flags
));
4529 if (inode
->i_nlink
> 0) {
4530 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
4534 ret
= btrfs_commit_inode_delayed_inode(inode
);
4536 btrfs_orphan_del(NULL
, inode
);
4540 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
4542 btrfs_orphan_del(NULL
, inode
);
4545 rsv
->size
= min_size
;
4547 global_rsv
= &root
->fs_info
->global_block_rsv
;
4549 btrfs_i_size_write(inode
, 0);
4552 * This is a bit simpler than btrfs_truncate since we've already
4553 * reserved our space for our orphan item in the unlink, so we just
4554 * need to reserve some slack space in case we add bytes and update
4555 * inode item when doing the truncate.
4558 ret
= btrfs_block_rsv_refill(root
, rsv
, min_size
,
4559 BTRFS_RESERVE_FLUSH_LIMIT
);
4562 * Try and steal from the global reserve since we will
4563 * likely not use this space anyway, we want to try as
4564 * hard as possible to get this to work.
4567 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, min_size
);
4570 btrfs_warn(root
->fs_info
,
4571 "Could not get space for a delete, will truncate on mount %d",
4573 btrfs_orphan_del(NULL
, inode
);
4574 btrfs_free_block_rsv(root
, rsv
);
4578 trans
= btrfs_join_transaction(root
);
4579 if (IS_ERR(trans
)) {
4580 btrfs_orphan_del(NULL
, inode
);
4581 btrfs_free_block_rsv(root
, rsv
);
4585 trans
->block_rsv
= rsv
;
4587 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
4591 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
4592 btrfs_end_transaction(trans
, root
);
4594 btrfs_btree_balance_dirty(root
);
4597 btrfs_free_block_rsv(root
, rsv
);
4600 trans
->block_rsv
= root
->orphan_block_rsv
;
4601 ret
= btrfs_orphan_del(trans
, inode
);
4605 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
4606 if (!(root
== root
->fs_info
->tree_root
||
4607 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
4608 btrfs_return_ino(root
, btrfs_ino(inode
));
4610 btrfs_end_transaction(trans
, root
);
4611 btrfs_btree_balance_dirty(root
);
4613 btrfs_remove_delayed_node(inode
);
4619 * this returns the key found in the dir entry in the location pointer.
4620 * If no dir entries were found, location->objectid is 0.
4622 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
4623 struct btrfs_key
*location
)
4625 const char *name
= dentry
->d_name
.name
;
4626 int namelen
= dentry
->d_name
.len
;
4627 struct btrfs_dir_item
*di
;
4628 struct btrfs_path
*path
;
4629 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4632 path
= btrfs_alloc_path();
4636 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
4641 if (IS_ERR_OR_NULL(di
))
4644 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
4646 btrfs_free_path(path
);
4649 location
->objectid
= 0;
4654 * when we hit a tree root in a directory, the btrfs part of the inode
4655 * needs to be changed to reflect the root directory of the tree root. This
4656 * is kind of like crossing a mount point.
4658 static int fixup_tree_root_location(struct btrfs_root
*root
,
4660 struct dentry
*dentry
,
4661 struct btrfs_key
*location
,
4662 struct btrfs_root
**sub_root
)
4664 struct btrfs_path
*path
;
4665 struct btrfs_root
*new_root
;
4666 struct btrfs_root_ref
*ref
;
4667 struct extent_buffer
*leaf
;
4671 path
= btrfs_alloc_path();
4678 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
4679 BTRFS_I(dir
)->root
->root_key
.objectid
,
4680 location
->objectid
);
4687 leaf
= path
->nodes
[0];
4688 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
4689 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
4690 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
4693 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
4694 (unsigned long)(ref
+ 1),
4695 dentry
->d_name
.len
);
4699 btrfs_release_path(path
);
4701 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
4702 if (IS_ERR(new_root
)) {
4703 err
= PTR_ERR(new_root
);
4707 *sub_root
= new_root
;
4708 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
4709 location
->type
= BTRFS_INODE_ITEM_KEY
;
4710 location
->offset
= 0;
4713 btrfs_free_path(path
);
4717 static void inode_tree_add(struct inode
*inode
)
4719 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4720 struct btrfs_inode
*entry
;
4722 struct rb_node
*parent
;
4723 u64 ino
= btrfs_ino(inode
);
4725 if (inode_unhashed(inode
))
4729 spin_lock(&root
->inode_lock
);
4730 p
= &root
->inode_tree
.rb_node
;
4733 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
4735 if (ino
< btrfs_ino(&entry
->vfs_inode
))
4736 p
= &parent
->rb_left
;
4737 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
4738 p
= &parent
->rb_right
;
4740 WARN_ON(!(entry
->vfs_inode
.i_state
&
4741 (I_WILL_FREE
| I_FREEING
)));
4742 rb_erase(parent
, &root
->inode_tree
);
4743 RB_CLEAR_NODE(parent
);
4744 spin_unlock(&root
->inode_lock
);
4748 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
4749 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
4750 spin_unlock(&root
->inode_lock
);
4753 static void inode_tree_del(struct inode
*inode
)
4755 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4758 spin_lock(&root
->inode_lock
);
4759 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
4760 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
4761 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
4762 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
4764 spin_unlock(&root
->inode_lock
);
4767 * Free space cache has inodes in the tree root, but the tree root has a
4768 * root_refs of 0, so this could end up dropping the tree root as a
4769 * snapshot, so we need the extra !root->fs_info->tree_root check to
4770 * make sure we don't drop it.
4772 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0 &&
4773 root
!= root
->fs_info
->tree_root
) {
4774 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
4775 spin_lock(&root
->inode_lock
);
4776 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
4777 spin_unlock(&root
->inode_lock
);
4779 btrfs_add_dead_root(root
);
4783 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
4785 struct rb_node
*node
;
4786 struct rb_node
*prev
;
4787 struct btrfs_inode
*entry
;
4788 struct inode
*inode
;
4791 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
4793 spin_lock(&root
->inode_lock
);
4795 node
= root
->inode_tree
.rb_node
;
4799 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4801 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
4802 node
= node
->rb_left
;
4803 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
4804 node
= node
->rb_right
;
4810 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
4811 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
4815 prev
= rb_next(prev
);
4819 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4820 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
4821 inode
= igrab(&entry
->vfs_inode
);
4823 spin_unlock(&root
->inode_lock
);
4824 if (atomic_read(&inode
->i_count
) > 1)
4825 d_prune_aliases(inode
);
4827 * btrfs_drop_inode will have it removed from
4828 * the inode cache when its usage count
4833 spin_lock(&root
->inode_lock
);
4837 if (cond_resched_lock(&root
->inode_lock
))
4840 node
= rb_next(node
);
4842 spin_unlock(&root
->inode_lock
);
4845 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
4847 struct btrfs_iget_args
*args
= p
;
4848 inode
->i_ino
= args
->ino
;
4849 BTRFS_I(inode
)->root
= args
->root
;
4853 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
4855 struct btrfs_iget_args
*args
= opaque
;
4856 return args
->ino
== btrfs_ino(inode
) &&
4857 args
->root
== BTRFS_I(inode
)->root
;
4860 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
4862 struct btrfs_root
*root
)
4864 struct inode
*inode
;
4865 struct btrfs_iget_args args
;
4866 args
.ino
= objectid
;
4869 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
4870 btrfs_init_locked_inode
,
4875 /* Get an inode object given its location and corresponding root.
4876 * Returns in *is_new if the inode was read from disk
4878 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
4879 struct btrfs_root
*root
, int *new)
4881 struct inode
*inode
;
4883 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
4885 return ERR_PTR(-ENOMEM
);
4887 if (inode
->i_state
& I_NEW
) {
4888 BTRFS_I(inode
)->root
= root
;
4889 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
4890 btrfs_read_locked_inode(inode
);
4891 if (!is_bad_inode(inode
)) {
4892 inode_tree_add(inode
);
4893 unlock_new_inode(inode
);
4897 unlock_new_inode(inode
);
4899 inode
= ERR_PTR(-ESTALE
);
4906 static struct inode
*new_simple_dir(struct super_block
*s
,
4907 struct btrfs_key
*key
,
4908 struct btrfs_root
*root
)
4910 struct inode
*inode
= new_inode(s
);
4913 return ERR_PTR(-ENOMEM
);
4915 BTRFS_I(inode
)->root
= root
;
4916 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
4917 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
4919 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
4920 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
4921 inode
->i_fop
= &simple_dir_operations
;
4922 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
4923 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4928 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
4930 struct inode
*inode
;
4931 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4932 struct btrfs_root
*sub_root
= root
;
4933 struct btrfs_key location
;
4937 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
4938 return ERR_PTR(-ENAMETOOLONG
);
4940 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
4942 return ERR_PTR(ret
);
4944 if (location
.objectid
== 0)
4947 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
4948 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
4952 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
4954 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
4955 ret
= fixup_tree_root_location(root
, dir
, dentry
,
4956 &location
, &sub_root
);
4959 inode
= ERR_PTR(ret
);
4961 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
4963 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
4965 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
4967 if (!IS_ERR(inode
) && root
!= sub_root
) {
4968 down_read(&root
->fs_info
->cleanup_work_sem
);
4969 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
4970 ret
= btrfs_orphan_cleanup(sub_root
);
4971 up_read(&root
->fs_info
->cleanup_work_sem
);
4974 inode
= ERR_PTR(ret
);
4981 static int btrfs_dentry_delete(const struct dentry
*dentry
)
4983 struct btrfs_root
*root
;
4984 struct inode
*inode
= dentry
->d_inode
;
4986 if (!inode
&& !IS_ROOT(dentry
))
4987 inode
= dentry
->d_parent
->d_inode
;
4990 root
= BTRFS_I(inode
)->root
;
4991 if (btrfs_root_refs(&root
->root_item
) == 0)
4994 if (btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5000 static void btrfs_dentry_release(struct dentry
*dentry
)
5002 if (dentry
->d_fsdata
)
5003 kfree(dentry
->d_fsdata
);
5006 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5011 ret
= d_splice_alias(btrfs_lookup_dentry(dir
, dentry
), dentry
);
5015 unsigned char btrfs_filetype_table
[] = {
5016 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
5019 static int btrfs_real_readdir(struct file
*file
, struct dir_context
*ctx
)
5021 struct inode
*inode
= file_inode(file
);
5022 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5023 struct btrfs_item
*item
;
5024 struct btrfs_dir_item
*di
;
5025 struct btrfs_key key
;
5026 struct btrfs_key found_key
;
5027 struct btrfs_path
*path
;
5028 struct list_head ins_list
;
5029 struct list_head del_list
;
5031 struct extent_buffer
*leaf
;
5033 unsigned char d_type
;
5038 int key_type
= BTRFS_DIR_INDEX_KEY
;
5042 int is_curr
= 0; /* ctx->pos points to the current index? */
5044 /* FIXME, use a real flag for deciding about the key type */
5045 if (root
->fs_info
->tree_root
== root
)
5046 key_type
= BTRFS_DIR_ITEM_KEY
;
5048 if (!dir_emit_dots(file
, ctx
))
5051 path
= btrfs_alloc_path();
5057 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5058 INIT_LIST_HEAD(&ins_list
);
5059 INIT_LIST_HEAD(&del_list
);
5060 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
5063 btrfs_set_key_type(&key
, key_type
);
5064 key
.offset
= ctx
->pos
;
5065 key
.objectid
= btrfs_ino(inode
);
5067 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5072 leaf
= path
->nodes
[0];
5073 slot
= path
->slots
[0];
5074 if (slot
>= btrfs_header_nritems(leaf
)) {
5075 ret
= btrfs_next_leaf(root
, path
);
5083 item
= btrfs_item_nr(leaf
, slot
);
5084 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5086 if (found_key
.objectid
!= key
.objectid
)
5088 if (btrfs_key_type(&found_key
) != key_type
)
5090 if (found_key
.offset
< ctx
->pos
)
5092 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
5093 btrfs_should_delete_dir_index(&del_list
,
5097 ctx
->pos
= found_key
.offset
;
5100 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
5102 di_total
= btrfs_item_size(leaf
, item
);
5104 while (di_cur
< di_total
) {
5105 struct btrfs_key location
;
5107 if (verify_dir_item(root
, leaf
, di
))
5110 name_len
= btrfs_dir_name_len(leaf
, di
);
5111 if (name_len
<= sizeof(tmp_name
)) {
5112 name_ptr
= tmp_name
;
5114 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
5120 read_extent_buffer(leaf
, name_ptr
,
5121 (unsigned long)(di
+ 1), name_len
);
5123 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
5124 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
5127 /* is this a reference to our own snapshot? If so
5130 * In contrast to old kernels, we insert the snapshot's
5131 * dir item and dir index after it has been created, so
5132 * we won't find a reference to our own snapshot. We
5133 * still keep the following code for backward
5136 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
5137 location
.objectid
== root
->root_key
.objectid
) {
5141 over
= !dir_emit(ctx
, name_ptr
, name_len
,
5142 location
.objectid
, d_type
);
5145 if (name_ptr
!= tmp_name
)
5150 di_len
= btrfs_dir_name_len(leaf
, di
) +
5151 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
5153 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
5159 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5162 ret
= btrfs_readdir_delayed_dir_index(ctx
, &ins_list
);
5167 /* Reached end of directory/root. Bump pos past the last item. */
5168 if (key_type
== BTRFS_DIR_INDEX_KEY
)
5170 * 32-bit glibc will use getdents64, but then strtol -
5171 * so the last number we can serve is this.
5173 ctx
->pos
= 0x7fffffff;
5179 if (key_type
== BTRFS_DIR_INDEX_KEY
)
5180 btrfs_put_delayed_items(&ins_list
, &del_list
);
5181 btrfs_free_path(path
);
5185 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5187 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5188 struct btrfs_trans_handle
*trans
;
5190 bool nolock
= false;
5192 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5195 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(inode
))
5198 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
5200 trans
= btrfs_join_transaction_nolock(root
);
5202 trans
= btrfs_join_transaction(root
);
5204 return PTR_ERR(trans
);
5205 ret
= btrfs_commit_transaction(trans
, root
);
5211 * This is somewhat expensive, updating the tree every time the
5212 * inode changes. But, it is most likely to find the inode in cache.
5213 * FIXME, needs more benchmarking...there are no reasons other than performance
5214 * to keep or drop this code.
5216 static int btrfs_dirty_inode(struct inode
*inode
)
5218 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5219 struct btrfs_trans_handle
*trans
;
5222 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5225 trans
= btrfs_join_transaction(root
);
5227 return PTR_ERR(trans
);
5229 ret
= btrfs_update_inode(trans
, root
, inode
);
5230 if (ret
&& ret
== -ENOSPC
) {
5231 /* whoops, lets try again with the full transaction */
5232 btrfs_end_transaction(trans
, root
);
5233 trans
= btrfs_start_transaction(root
, 1);
5235 return PTR_ERR(trans
);
5237 ret
= btrfs_update_inode(trans
, root
, inode
);
5239 btrfs_end_transaction(trans
, root
);
5240 if (BTRFS_I(inode
)->delayed_node
)
5241 btrfs_balance_delayed_items(root
);
5247 * This is a copy of file_update_time. We need this so we can return error on
5248 * ENOSPC for updating the inode in the case of file write and mmap writes.
5250 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
5253 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5255 if (btrfs_root_readonly(root
))
5258 if (flags
& S_VERSION
)
5259 inode_inc_iversion(inode
);
5260 if (flags
& S_CTIME
)
5261 inode
->i_ctime
= *now
;
5262 if (flags
& S_MTIME
)
5263 inode
->i_mtime
= *now
;
5264 if (flags
& S_ATIME
)
5265 inode
->i_atime
= *now
;
5266 return btrfs_dirty_inode(inode
);
5270 * find the highest existing sequence number in a directory
5271 * and then set the in-memory index_cnt variable to reflect
5272 * free sequence numbers
5274 static int btrfs_set_inode_index_count(struct inode
*inode
)
5276 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5277 struct btrfs_key key
, found_key
;
5278 struct btrfs_path
*path
;
5279 struct extent_buffer
*leaf
;
5282 key
.objectid
= btrfs_ino(inode
);
5283 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
5284 key
.offset
= (u64
)-1;
5286 path
= btrfs_alloc_path();
5290 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5293 /* FIXME: we should be able to handle this */
5299 * MAGIC NUMBER EXPLANATION:
5300 * since we search a directory based on f_pos we have to start at 2
5301 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5302 * else has to start at 2
5304 if (path
->slots
[0] == 0) {
5305 BTRFS_I(inode
)->index_cnt
= 2;
5311 leaf
= path
->nodes
[0];
5312 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5314 if (found_key
.objectid
!= btrfs_ino(inode
) ||
5315 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
5316 BTRFS_I(inode
)->index_cnt
= 2;
5320 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
5322 btrfs_free_path(path
);
5327 * helper to find a free sequence number in a given directory. This current
5328 * code is very simple, later versions will do smarter things in the btree
5330 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
5334 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
5335 ret
= btrfs_inode_delayed_dir_index_count(dir
);
5337 ret
= btrfs_set_inode_index_count(dir
);
5343 *index
= BTRFS_I(dir
)->index_cnt
;
5344 BTRFS_I(dir
)->index_cnt
++;
5349 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
5350 struct btrfs_root
*root
,
5352 const char *name
, int name_len
,
5353 u64 ref_objectid
, u64 objectid
,
5354 umode_t mode
, u64
*index
)
5356 struct inode
*inode
;
5357 struct btrfs_inode_item
*inode_item
;
5358 struct btrfs_key
*location
;
5359 struct btrfs_path
*path
;
5360 struct btrfs_inode_ref
*ref
;
5361 struct btrfs_key key
[2];
5367 path
= btrfs_alloc_path();
5369 return ERR_PTR(-ENOMEM
);
5371 inode
= new_inode(root
->fs_info
->sb
);
5373 btrfs_free_path(path
);
5374 return ERR_PTR(-ENOMEM
);
5378 * we have to initialize this early, so we can reclaim the inode
5379 * number if we fail afterwards in this function.
5381 inode
->i_ino
= objectid
;
5384 trace_btrfs_inode_request(dir
);
5386 ret
= btrfs_set_inode_index(dir
, index
);
5388 btrfs_free_path(path
);
5390 return ERR_PTR(ret
);
5394 * index_cnt is ignored for everything but a dir,
5395 * btrfs_get_inode_index_count has an explanation for the magic
5398 BTRFS_I(inode
)->index_cnt
= 2;
5399 BTRFS_I(inode
)->root
= root
;
5400 BTRFS_I(inode
)->generation
= trans
->transid
;
5401 inode
->i_generation
= BTRFS_I(inode
)->generation
;
5404 * We could have gotten an inode number from somebody who was fsynced
5405 * and then removed in this same transaction, so let's just set full
5406 * sync since it will be a full sync anyway and this will blow away the
5407 * old info in the log.
5409 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
5416 key
[0].objectid
= objectid
;
5417 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
5421 * Start new inodes with an inode_ref. This is slightly more
5422 * efficient for small numbers of hard links since they will
5423 * be packed into one item. Extended refs will kick in if we
5424 * add more hard links than can fit in the ref item.
5426 key
[1].objectid
= objectid
;
5427 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
5428 key
[1].offset
= ref_objectid
;
5430 sizes
[0] = sizeof(struct btrfs_inode_item
);
5431 sizes
[1] = name_len
+ sizeof(*ref
);
5433 path
->leave_spinning
= 1;
5434 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
5438 inode_init_owner(inode
, dir
, mode
);
5439 inode_set_bytes(inode
, 0);
5440 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
5441 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
5442 struct btrfs_inode_item
);
5443 memset_extent_buffer(path
->nodes
[0], 0, (unsigned long)inode_item
,
5444 sizeof(*inode_item
));
5445 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
5447 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
5448 struct btrfs_inode_ref
);
5449 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
5450 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
5451 ptr
= (unsigned long)(ref
+ 1);
5452 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
5454 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5455 btrfs_free_path(path
);
5457 location
= &BTRFS_I(inode
)->location
;
5458 location
->objectid
= objectid
;
5459 location
->offset
= 0;
5460 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
5462 btrfs_inherit_iflags(inode
, dir
);
5464 if (S_ISREG(mode
)) {
5465 if (btrfs_test_opt(root
, NODATASUM
))
5466 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
5467 if (btrfs_test_opt(root
, NODATACOW
))
5468 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
5469 BTRFS_INODE_NODATASUM
;
5472 insert_inode_hash(inode
);
5473 inode_tree_add(inode
);
5475 trace_btrfs_inode_new(inode
);
5476 btrfs_set_inode_last_trans(trans
, inode
);
5478 btrfs_update_root_times(trans
, root
);
5483 BTRFS_I(dir
)->index_cnt
--;
5484 btrfs_free_path(path
);
5486 return ERR_PTR(ret
);
5489 static inline u8
btrfs_inode_type(struct inode
*inode
)
5491 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
5495 * utility function to add 'inode' into 'parent_inode' with
5496 * a give name and a given sequence number.
5497 * if 'add_backref' is true, also insert a backref from the
5498 * inode to the parent directory.
5500 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
5501 struct inode
*parent_inode
, struct inode
*inode
,
5502 const char *name
, int name_len
, int add_backref
, u64 index
)
5505 struct btrfs_key key
;
5506 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
5507 u64 ino
= btrfs_ino(inode
);
5508 u64 parent_ino
= btrfs_ino(parent_inode
);
5510 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
5511 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
5514 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
5518 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
5519 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
5520 key
.objectid
, root
->root_key
.objectid
,
5521 parent_ino
, index
, name
, name_len
);
5522 } else if (add_backref
) {
5523 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
5527 /* Nothing to clean up yet */
5531 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
5533 btrfs_inode_type(inode
), index
);
5534 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
5537 btrfs_abort_transaction(trans
, root
, ret
);
5541 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
5543 inode_inc_iversion(parent_inode
);
5544 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
5545 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
5547 btrfs_abort_transaction(trans
, root
, ret
);
5551 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
5554 err
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
5555 key
.objectid
, root
->root_key
.objectid
,
5556 parent_ino
, &local_index
, name
, name_len
);
5558 } else if (add_backref
) {
5562 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
5563 ino
, parent_ino
, &local_index
);
5568 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
5569 struct inode
*dir
, struct dentry
*dentry
,
5570 struct inode
*inode
, int backref
, u64 index
)
5572 int err
= btrfs_add_link(trans
, dir
, inode
,
5573 dentry
->d_name
.name
, dentry
->d_name
.len
,
5580 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
5581 umode_t mode
, dev_t rdev
)
5583 struct btrfs_trans_handle
*trans
;
5584 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5585 struct inode
*inode
= NULL
;
5591 if (!new_valid_dev(rdev
))
5595 * 2 for inode item and ref
5597 * 1 for xattr if selinux is on
5599 trans
= btrfs_start_transaction(root
, 5);
5601 return PTR_ERR(trans
);
5603 err
= btrfs_find_free_ino(root
, &objectid
);
5607 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
5608 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
5610 if (IS_ERR(inode
)) {
5611 err
= PTR_ERR(inode
);
5615 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
5622 * If the active LSM wants to access the inode during
5623 * d_instantiate it needs these. Smack checks to see
5624 * if the filesystem supports xattrs by looking at the
5628 inode
->i_op
= &btrfs_special_inode_operations
;
5629 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
5633 init_special_inode(inode
, inode
->i_mode
, rdev
);
5634 btrfs_update_inode(trans
, root
, inode
);
5635 d_instantiate(dentry
, inode
);
5638 btrfs_end_transaction(trans
, root
);
5639 btrfs_btree_balance_dirty(root
);
5641 inode_dec_link_count(inode
);
5647 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
5648 umode_t mode
, bool excl
)
5650 struct btrfs_trans_handle
*trans
;
5651 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5652 struct inode
*inode
= NULL
;
5653 int drop_inode_on_err
= 0;
5659 * 2 for inode item and ref
5661 * 1 for xattr if selinux is on
5663 trans
= btrfs_start_transaction(root
, 5);
5665 return PTR_ERR(trans
);
5667 err
= btrfs_find_free_ino(root
, &objectid
);
5671 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
5672 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
5674 if (IS_ERR(inode
)) {
5675 err
= PTR_ERR(inode
);
5678 drop_inode_on_err
= 1;
5680 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
5684 err
= btrfs_update_inode(trans
, root
, inode
);
5689 * If the active LSM wants to access the inode during
5690 * d_instantiate it needs these. Smack checks to see
5691 * if the filesystem supports xattrs by looking at the
5694 inode
->i_fop
= &btrfs_file_operations
;
5695 inode
->i_op
= &btrfs_file_inode_operations
;
5697 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
5701 inode
->i_mapping
->a_ops
= &btrfs_aops
;
5702 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
5703 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
5704 d_instantiate(dentry
, inode
);
5707 btrfs_end_transaction(trans
, root
);
5708 if (err
&& drop_inode_on_err
) {
5709 inode_dec_link_count(inode
);
5712 btrfs_btree_balance_dirty(root
);
5716 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
5717 struct dentry
*dentry
)
5719 struct btrfs_trans_handle
*trans
;
5720 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5721 struct inode
*inode
= old_dentry
->d_inode
;
5726 /* do not allow sys_link's with other subvols of the same device */
5727 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
5730 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
5733 err
= btrfs_set_inode_index(dir
, &index
);
5738 * 2 items for inode and inode ref
5739 * 2 items for dir items
5740 * 1 item for parent inode
5742 trans
= btrfs_start_transaction(root
, 5);
5743 if (IS_ERR(trans
)) {
5744 err
= PTR_ERR(trans
);
5748 btrfs_inc_nlink(inode
);
5749 inode_inc_iversion(inode
);
5750 inode
->i_ctime
= CURRENT_TIME
;
5752 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
5754 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
5759 struct dentry
*parent
= dentry
->d_parent
;
5760 err
= btrfs_update_inode(trans
, root
, inode
);
5763 d_instantiate(dentry
, inode
);
5764 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
5767 btrfs_end_transaction(trans
, root
);
5770 inode_dec_link_count(inode
);
5773 btrfs_btree_balance_dirty(root
);
5777 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
5779 struct inode
*inode
= NULL
;
5780 struct btrfs_trans_handle
*trans
;
5781 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5783 int drop_on_err
= 0;
5788 * 2 items for inode and ref
5789 * 2 items for dir items
5790 * 1 for xattr if selinux is on
5792 trans
= btrfs_start_transaction(root
, 5);
5794 return PTR_ERR(trans
);
5796 err
= btrfs_find_free_ino(root
, &objectid
);
5800 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
5801 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
5802 S_IFDIR
| mode
, &index
);
5803 if (IS_ERR(inode
)) {
5804 err
= PTR_ERR(inode
);
5810 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
5814 inode
->i_op
= &btrfs_dir_inode_operations
;
5815 inode
->i_fop
= &btrfs_dir_file_operations
;
5817 btrfs_i_size_write(inode
, 0);
5818 err
= btrfs_update_inode(trans
, root
, inode
);
5822 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
5823 dentry
->d_name
.len
, 0, index
);
5827 d_instantiate(dentry
, inode
);
5831 btrfs_end_transaction(trans
, root
);
5834 btrfs_btree_balance_dirty(root
);
5838 /* helper for btfs_get_extent. Given an existing extent in the tree,
5839 * and an extent that you want to insert, deal with overlap and insert
5840 * the new extent into the tree.
5842 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
5843 struct extent_map
*existing
,
5844 struct extent_map
*em
,
5845 u64 map_start
, u64 map_len
)
5849 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
5850 start_diff
= map_start
- em
->start
;
5851 em
->start
= map_start
;
5853 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
5854 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
5855 em
->block_start
+= start_diff
;
5856 em
->block_len
-= start_diff
;
5858 return add_extent_mapping(em_tree
, em
, 0);
5861 static noinline
int uncompress_inline(struct btrfs_path
*path
,
5862 struct inode
*inode
, struct page
*page
,
5863 size_t pg_offset
, u64 extent_offset
,
5864 struct btrfs_file_extent_item
*item
)
5867 struct extent_buffer
*leaf
= path
->nodes
[0];
5870 unsigned long inline_size
;
5874 WARN_ON(pg_offset
!= 0);
5875 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5876 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
5877 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
5878 btrfs_item_nr(leaf
, path
->slots
[0]));
5879 tmp
= kmalloc(inline_size
, GFP_NOFS
);
5882 ptr
= btrfs_file_extent_inline_start(item
);
5884 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
5886 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
5887 ret
= btrfs_decompress(compress_type
, tmp
, page
,
5888 extent_offset
, inline_size
, max_size
);
5890 char *kaddr
= kmap_atomic(page
);
5891 unsigned long copy_size
= min_t(u64
,
5892 PAGE_CACHE_SIZE
- pg_offset
,
5893 max_size
- extent_offset
);
5894 memset(kaddr
+ pg_offset
, 0, copy_size
);
5895 kunmap_atomic(kaddr
);
5902 * a bit scary, this does extent mapping from logical file offset to the disk.
5903 * the ugly parts come from merging extents from the disk with the in-ram
5904 * representation. This gets more complex because of the data=ordered code,
5905 * where the in-ram extents might be locked pending data=ordered completion.
5907 * This also copies inline extents directly into the page.
5910 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
5911 size_t pg_offset
, u64 start
, u64 len
,
5917 u64 extent_start
= 0;
5919 u64 objectid
= btrfs_ino(inode
);
5921 struct btrfs_path
*path
= NULL
;
5922 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5923 struct btrfs_file_extent_item
*item
;
5924 struct extent_buffer
*leaf
;
5925 struct btrfs_key found_key
;
5926 struct extent_map
*em
= NULL
;
5927 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5928 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5929 struct btrfs_trans_handle
*trans
= NULL
;
5933 read_lock(&em_tree
->lock
);
5934 em
= lookup_extent_mapping(em_tree
, start
, len
);
5936 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5937 read_unlock(&em_tree
->lock
);
5940 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
5941 free_extent_map(em
);
5942 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
5943 free_extent_map(em
);
5947 em
= alloc_extent_map();
5952 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5953 em
->start
= EXTENT_MAP_HOLE
;
5954 em
->orig_start
= EXTENT_MAP_HOLE
;
5956 em
->block_len
= (u64
)-1;
5959 path
= btrfs_alloc_path();
5965 * Chances are we'll be called again, so go ahead and do
5971 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
5972 objectid
, start
, trans
!= NULL
);
5979 if (path
->slots
[0] == 0)
5984 leaf
= path
->nodes
[0];
5985 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5986 struct btrfs_file_extent_item
);
5987 /* are we inside the extent that was found? */
5988 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5989 found_type
= btrfs_key_type(&found_key
);
5990 if (found_key
.objectid
!= objectid
||
5991 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
5995 found_type
= btrfs_file_extent_type(leaf
, item
);
5996 extent_start
= found_key
.offset
;
5997 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5998 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5999 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6000 extent_end
= extent_start
+
6001 btrfs_file_extent_num_bytes(leaf
, item
);
6002 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6004 size
= btrfs_file_extent_inline_len(leaf
, item
);
6005 extent_end
= ALIGN(extent_start
+ size
, root
->sectorsize
);
6008 if (start
>= extent_end
) {
6010 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
6011 ret
= btrfs_next_leaf(root
, path
);
6018 leaf
= path
->nodes
[0];
6020 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6021 if (found_key
.objectid
!= objectid
||
6022 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6024 if (start
+ len
<= found_key
.offset
)
6027 em
->orig_start
= start
;
6028 em
->len
= found_key
.offset
- start
;
6032 em
->ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, item
);
6033 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6034 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6035 em
->start
= extent_start
;
6036 em
->len
= extent_end
- extent_start
;
6037 em
->orig_start
= extent_start
-
6038 btrfs_file_extent_offset(leaf
, item
);
6039 em
->orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
6041 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
6043 em
->block_start
= EXTENT_MAP_HOLE
;
6046 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
6047 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
6048 em
->compress_type
= compress_type
;
6049 em
->block_start
= bytenr
;
6050 em
->block_len
= em
->orig_block_len
;
6052 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
6053 em
->block_start
= bytenr
;
6054 em
->block_len
= em
->len
;
6055 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
6056 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
6059 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6063 size_t extent_offset
;
6066 em
->block_start
= EXTENT_MAP_INLINE
;
6067 if (!page
|| create
) {
6068 em
->start
= extent_start
;
6069 em
->len
= extent_end
- extent_start
;
6073 size
= btrfs_file_extent_inline_len(leaf
, item
);
6074 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
6075 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
6076 size
- extent_offset
);
6077 em
->start
= extent_start
+ extent_offset
;
6078 em
->len
= ALIGN(copy_size
, root
->sectorsize
);
6079 em
->orig_block_len
= em
->len
;
6080 em
->orig_start
= em
->start
;
6081 if (compress_type
) {
6082 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
6083 em
->compress_type
= compress_type
;
6085 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
6086 if (create
== 0 && !PageUptodate(page
)) {
6087 if (btrfs_file_extent_compression(leaf
, item
) !=
6088 BTRFS_COMPRESS_NONE
) {
6089 ret
= uncompress_inline(path
, inode
, page
,
6091 extent_offset
, item
);
6092 BUG_ON(ret
); /* -ENOMEM */
6095 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6097 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
6098 memset(map
+ pg_offset
+ copy_size
, 0,
6099 PAGE_CACHE_SIZE
- pg_offset
-
6104 flush_dcache_page(page
);
6105 } else if (create
&& PageUptodate(page
)) {
6109 free_extent_map(em
);
6112 btrfs_release_path(path
);
6113 trans
= btrfs_join_transaction(root
);
6116 return ERR_CAST(trans
);
6120 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6123 btrfs_mark_buffer_dirty(leaf
);
6125 set_extent_uptodate(io_tree
, em
->start
,
6126 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
6129 WARN(1, KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
6133 em
->orig_start
= start
;
6136 em
->block_start
= EXTENT_MAP_HOLE
;
6137 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
6139 btrfs_release_path(path
);
6140 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
6141 btrfs_err(root
->fs_info
, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6142 (unsigned long long)em
->start
,
6143 (unsigned long long)em
->len
,
6144 (unsigned long long)start
,
6145 (unsigned long long)len
);
6151 write_lock(&em_tree
->lock
);
6152 ret
= add_extent_mapping(em_tree
, em
, 0);
6153 /* it is possible that someone inserted the extent into the tree
6154 * while we had the lock dropped. It is also possible that
6155 * an overlapping map exists in the tree
6157 if (ret
== -EEXIST
) {
6158 struct extent_map
*existing
;
6162 existing
= lookup_extent_mapping(em_tree
, start
, len
);
6163 if (existing
&& (existing
->start
> start
||
6164 existing
->start
+ existing
->len
<= start
)) {
6165 free_extent_map(existing
);
6169 existing
= lookup_extent_mapping(em_tree
, em
->start
,
6172 err
= merge_extent_mapping(em_tree
, existing
,
6175 free_extent_map(existing
);
6177 free_extent_map(em
);
6182 free_extent_map(em
);
6186 free_extent_map(em
);
6191 write_unlock(&em_tree
->lock
);
6195 trace_btrfs_get_extent(root
, em
);
6198 btrfs_free_path(path
);
6200 ret
= btrfs_end_transaction(trans
, root
);
6205 free_extent_map(em
);
6206 return ERR_PTR(err
);
6208 BUG_ON(!em
); /* Error is always set */
6212 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
6213 size_t pg_offset
, u64 start
, u64 len
,
6216 struct extent_map
*em
;
6217 struct extent_map
*hole_em
= NULL
;
6218 u64 range_start
= start
;
6224 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
6231 * - a pre-alloc extent,
6232 * there might actually be delalloc bytes behind it.
6234 if (em
->block_start
!= EXTENT_MAP_HOLE
&&
6235 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
6241 /* check to see if we've wrapped (len == -1 or similar) */
6250 /* ok, we didn't find anything, lets look for delalloc */
6251 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
6252 end
, len
, EXTENT_DELALLOC
, 1);
6253 found_end
= range_start
+ found
;
6254 if (found_end
< range_start
)
6255 found_end
= (u64
)-1;
6258 * we didn't find anything useful, return
6259 * the original results from get_extent()
6261 if (range_start
> end
|| found_end
<= start
) {
6267 /* adjust the range_start to make sure it doesn't
6268 * go backwards from the start they passed in
6270 range_start
= max(start
,range_start
);
6271 found
= found_end
- range_start
;
6274 u64 hole_start
= start
;
6277 em
= alloc_extent_map();
6283 * when btrfs_get_extent can't find anything it
6284 * returns one huge hole
6286 * make sure what it found really fits our range, and
6287 * adjust to make sure it is based on the start from
6291 u64 calc_end
= extent_map_end(hole_em
);
6293 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
6294 free_extent_map(hole_em
);
6297 hole_start
= max(hole_em
->start
, start
);
6298 hole_len
= calc_end
- hole_start
;
6302 if (hole_em
&& range_start
> hole_start
) {
6303 /* our hole starts before our delalloc, so we
6304 * have to return just the parts of the hole
6305 * that go until the delalloc starts
6307 em
->len
= min(hole_len
,
6308 range_start
- hole_start
);
6309 em
->start
= hole_start
;
6310 em
->orig_start
= hole_start
;
6312 * don't adjust block start at all,
6313 * it is fixed at EXTENT_MAP_HOLE
6315 em
->block_start
= hole_em
->block_start
;
6316 em
->block_len
= hole_len
;
6317 if (test_bit(EXTENT_FLAG_PREALLOC
, &hole_em
->flags
))
6318 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
6320 em
->start
= range_start
;
6322 em
->orig_start
= range_start
;
6323 em
->block_start
= EXTENT_MAP_DELALLOC
;
6324 em
->block_len
= found
;
6326 } else if (hole_em
) {
6331 free_extent_map(hole_em
);
6333 free_extent_map(em
);
6334 return ERR_PTR(err
);
6339 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
6342 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6343 struct btrfs_trans_handle
*trans
;
6344 struct extent_map
*em
;
6345 struct btrfs_key ins
;
6349 trans
= btrfs_join_transaction(root
);
6351 return ERR_CAST(trans
);
6353 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
6355 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
6356 ret
= btrfs_reserve_extent(trans
, root
, len
, root
->sectorsize
, 0,
6357 alloc_hint
, &ins
, 1);
6363 em
= create_pinned_em(inode
, start
, ins
.offset
, start
, ins
.objectid
,
6364 ins
.offset
, ins
.offset
, ins
.offset
, 0);
6368 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
6369 ins
.offset
, ins
.offset
, 0);
6371 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
6375 btrfs_end_transaction(trans
, root
);
6380 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6381 * block must be cow'd
6383 noinline
int can_nocow_extent(struct btrfs_trans_handle
*trans
,
6384 struct inode
*inode
, u64 offset
, u64
*len
,
6385 u64
*orig_start
, u64
*orig_block_len
,
6388 struct btrfs_path
*path
;
6390 struct extent_buffer
*leaf
;
6391 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6392 struct btrfs_file_extent_item
*fi
;
6393 struct btrfs_key key
;
6400 bool nocow
= (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
);
6401 path
= btrfs_alloc_path();
6405 ret
= btrfs_lookup_file_extent(trans
, root
, path
, btrfs_ino(inode
),
6410 slot
= path
->slots
[0];
6413 /* can't find the item, must cow */
6420 leaf
= path
->nodes
[0];
6421 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
6422 if (key
.objectid
!= btrfs_ino(inode
) ||
6423 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
6424 /* not our file or wrong item type, must cow */
6428 if (key
.offset
> offset
) {
6429 /* Wrong offset, must cow */
6433 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
6434 found_type
= btrfs_file_extent_type(leaf
, fi
);
6435 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
6436 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
6437 /* not a regular extent, must cow */
6441 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_REG
)
6444 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
6445 if (disk_bytenr
== 0)
6448 if (btrfs_file_extent_compression(leaf
, fi
) ||
6449 btrfs_file_extent_encryption(leaf
, fi
) ||
6450 btrfs_file_extent_other_encoding(leaf
, fi
))
6453 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
6456 *orig_start
= key
.offset
- backref_offset
;
6457 *orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
6458 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
6461 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
6463 if (btrfs_extent_readonly(root
, disk_bytenr
))
6467 * look for other files referencing this extent, if we
6468 * find any we must cow
6470 if (btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
6471 key
.offset
- backref_offset
, disk_bytenr
))
6475 * adjust disk_bytenr and num_bytes to cover just the bytes
6476 * in this extent we are about to write. If there
6477 * are any csums in that range we have to cow in order
6478 * to keep the csums correct
6480 disk_bytenr
+= backref_offset
;
6481 disk_bytenr
+= offset
- key
.offset
;
6482 num_bytes
= min(offset
+ *len
, extent_end
) - offset
;
6483 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
6486 * all of the above have passed, it is safe to overwrite this extent
6492 btrfs_free_path(path
);
6496 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
6497 struct extent_state
**cached_state
, int writing
)
6499 struct btrfs_ordered_extent
*ordered
;
6503 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6506 * We're concerned with the entire range that we're going to be
6507 * doing DIO to, so we need to make sure theres no ordered
6508 * extents in this range.
6510 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
6511 lockend
- lockstart
+ 1);
6514 * We need to make sure there are no buffered pages in this
6515 * range either, we could have raced between the invalidate in
6516 * generic_file_direct_write and locking the extent. The
6517 * invalidate needs to happen so that reads after a write do not
6520 if (!ordered
&& (!writing
||
6521 !test_range_bit(&BTRFS_I(inode
)->io_tree
,
6522 lockstart
, lockend
, EXTENT_UPTODATE
, 0,
6526 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6527 cached_state
, GFP_NOFS
);
6530 btrfs_start_ordered_extent(inode
, ordered
, 1);
6531 btrfs_put_ordered_extent(ordered
);
6533 /* Screw you mmap */
6534 ret
= filemap_write_and_wait_range(inode
->i_mapping
,
6541 * If we found a page that couldn't be invalidated just
6542 * fall back to buffered.
6544 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
6545 lockstart
>> PAGE_CACHE_SHIFT
,
6546 lockend
>> PAGE_CACHE_SHIFT
);
6557 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
6558 u64 len
, u64 orig_start
,
6559 u64 block_start
, u64 block_len
,
6560 u64 orig_block_len
, u64 ram_bytes
,
6563 struct extent_map_tree
*em_tree
;
6564 struct extent_map
*em
;
6565 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6568 em_tree
= &BTRFS_I(inode
)->extent_tree
;
6569 em
= alloc_extent_map();
6571 return ERR_PTR(-ENOMEM
);
6574 em
->orig_start
= orig_start
;
6575 em
->mod_start
= start
;
6578 em
->block_len
= block_len
;
6579 em
->block_start
= block_start
;
6580 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6581 em
->orig_block_len
= orig_block_len
;
6582 em
->ram_bytes
= ram_bytes
;
6583 em
->generation
= -1;
6584 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
6585 if (type
== BTRFS_ORDERED_PREALLOC
)
6586 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
6589 btrfs_drop_extent_cache(inode
, em
->start
,
6590 em
->start
+ em
->len
- 1, 0);
6591 write_lock(&em_tree
->lock
);
6592 ret
= add_extent_mapping(em_tree
, em
, 1);
6593 write_unlock(&em_tree
->lock
);
6594 } while (ret
== -EEXIST
);
6597 free_extent_map(em
);
6598 return ERR_PTR(ret
);
6605 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
6606 struct buffer_head
*bh_result
, int create
)
6608 struct extent_map
*em
;
6609 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6610 struct extent_state
*cached_state
= NULL
;
6611 u64 start
= iblock
<< inode
->i_blkbits
;
6612 u64 lockstart
, lockend
;
6613 u64 len
= bh_result
->b_size
;
6614 struct btrfs_trans_handle
*trans
;
6615 int unlock_bits
= EXTENT_LOCKED
;
6619 unlock_bits
|= EXTENT_DELALLOC
| EXTENT_DIRTY
;
6621 len
= min_t(u64
, len
, root
->sectorsize
);
6624 lockend
= start
+ len
- 1;
6627 * If this errors out it's because we couldn't invalidate pagecache for
6628 * this range and we need to fallback to buffered.
6630 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
, create
))
6633 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
6640 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6641 * io. INLINE is special, and we could probably kludge it in here, but
6642 * it's still buffered so for safety lets just fall back to the generic
6645 * For COMPRESSED we _have_ to read the entire extent in so we can
6646 * decompress it, so there will be buffering required no matter what we
6647 * do, so go ahead and fallback to buffered.
6649 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6650 * to buffered IO. Don't blame me, this is the price we pay for using
6653 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
6654 em
->block_start
== EXTENT_MAP_INLINE
) {
6655 free_extent_map(em
);
6660 /* Just a good old fashioned hole, return */
6661 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
6662 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
6663 free_extent_map(em
);
6668 * We don't allocate a new extent in the following cases
6670 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6672 * 2) The extent is marked as PREALLOC. We're good to go here and can
6673 * just use the extent.
6677 len
= min(len
, em
->len
- (start
- em
->start
));
6678 lockstart
= start
+ len
;
6682 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
6683 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
6684 em
->block_start
!= EXTENT_MAP_HOLE
)) {
6687 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
6689 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
6690 type
= BTRFS_ORDERED_PREALLOC
;
6692 type
= BTRFS_ORDERED_NOCOW
;
6693 len
= min(len
, em
->len
- (start
- em
->start
));
6694 block_start
= em
->block_start
+ (start
- em
->start
);
6697 * we're not going to log anything, but we do need
6698 * to make sure the current transaction stays open
6699 * while we look for nocow cross refs
6701 trans
= btrfs_join_transaction(root
);
6705 if (can_nocow_extent(trans
, inode
, start
, &len
, &orig_start
,
6706 &orig_block_len
, &ram_bytes
) == 1) {
6707 if (type
== BTRFS_ORDERED_PREALLOC
) {
6708 free_extent_map(em
);
6709 em
= create_pinned_em(inode
, start
, len
,
6715 btrfs_end_transaction(trans
, root
);
6720 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
6721 block_start
, len
, len
, type
);
6722 btrfs_end_transaction(trans
, root
);
6724 free_extent_map(em
);
6729 btrfs_end_transaction(trans
, root
);
6733 * this will cow the extent, reset the len in case we changed
6736 len
= bh_result
->b_size
;
6737 free_extent_map(em
);
6738 em
= btrfs_new_extent_direct(inode
, start
, len
);
6743 len
= min(len
, em
->len
- (start
- em
->start
));
6745 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
6747 bh_result
->b_size
= len
;
6748 bh_result
->b_bdev
= em
->bdev
;
6749 set_buffer_mapped(bh_result
);
6751 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
6752 set_buffer_new(bh_result
);
6755 * Need to update the i_size under the extent lock so buffered
6756 * readers will get the updated i_size when we unlock.
6758 if (start
+ len
> i_size_read(inode
))
6759 i_size_write(inode
, start
+ len
);
6761 spin_lock(&BTRFS_I(inode
)->lock
);
6762 BTRFS_I(inode
)->outstanding_extents
++;
6763 spin_unlock(&BTRFS_I(inode
)->lock
);
6765 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6766 lockstart
+ len
- 1, EXTENT_DELALLOC
, NULL
,
6767 &cached_state
, GFP_NOFS
);
6772 * In the case of write we need to clear and unlock the entire range,
6773 * in the case of read we need to unlock only the end area that we
6774 * aren't using if there is any left over space.
6776 if (lockstart
< lockend
) {
6777 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6778 lockend
, unlock_bits
, 1, 0,
6779 &cached_state
, GFP_NOFS
);
6781 free_extent_state(cached_state
);
6784 free_extent_map(em
);
6789 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6790 unlock_bits
, 1, 0, &cached_state
, GFP_NOFS
);
6794 struct btrfs_dio_private
{
6795 struct inode
*inode
;
6801 /* number of bios pending for this dio */
6802 atomic_t pending_bios
;
6807 /* orig_bio is our btrfs_io_bio */
6808 struct bio
*orig_bio
;
6810 /* dio_bio came from fs/direct-io.c */
6811 struct bio
*dio_bio
;
6814 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
6816 struct btrfs_dio_private
*dip
= bio
->bi_private
;
6817 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
6818 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6819 struct inode
*inode
= dip
->inode
;
6820 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6821 struct bio
*dio_bio
;
6824 start
= dip
->logical_offset
;
6826 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
6827 struct page
*page
= bvec
->bv_page
;
6830 u64
private = ~(u32
)0;
6831 unsigned long flags
;
6833 if (get_state_private(&BTRFS_I(inode
)->io_tree
,
6836 local_irq_save(flags
);
6837 kaddr
= kmap_atomic(page
);
6838 csum
= btrfs_csum_data(kaddr
+ bvec
->bv_offset
,
6839 csum
, bvec
->bv_len
);
6840 btrfs_csum_final(csum
, (char *)&csum
);
6841 kunmap_atomic(kaddr
);
6842 local_irq_restore(flags
);
6844 flush_dcache_page(bvec
->bv_page
);
6845 if (csum
!= private) {
6847 btrfs_err(root
->fs_info
, "csum failed ino %llu off %llu csum %u private %u",
6848 (unsigned long long)btrfs_ino(inode
),
6849 (unsigned long long)start
,
6850 csum
, (unsigned)private);
6855 start
+= bvec
->bv_len
;
6857 } while (bvec
<= bvec_end
);
6859 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
6860 dip
->logical_offset
+ dip
->bytes
- 1);
6861 dio_bio
= dip
->dio_bio
;
6865 /* If we had a csum failure make sure to clear the uptodate flag */
6867 clear_bit(BIO_UPTODATE
, &dio_bio
->bi_flags
);
6868 dio_end_io(dio_bio
, err
);
6872 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
6874 struct btrfs_dio_private
*dip
= bio
->bi_private
;
6875 struct inode
*inode
= dip
->inode
;
6876 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6877 struct btrfs_ordered_extent
*ordered
= NULL
;
6878 u64 ordered_offset
= dip
->logical_offset
;
6879 u64 ordered_bytes
= dip
->bytes
;
6880 struct bio
*dio_bio
;
6886 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
6888 ordered_bytes
, !err
);
6892 ordered
->work
.func
= finish_ordered_fn
;
6893 ordered
->work
.flags
= 0;
6894 btrfs_queue_worker(&root
->fs_info
->endio_write_workers
,
6898 * our bio might span multiple ordered extents. If we haven't
6899 * completed the accounting for the whole dio, go back and try again
6901 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
6902 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
6908 dio_bio
= dip
->dio_bio
;
6912 /* If we had an error make sure to clear the uptodate flag */
6914 clear_bit(BIO_UPTODATE
, &dio_bio
->bi_flags
);
6915 dio_end_io(dio_bio
, err
);
6919 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
6920 struct bio
*bio
, int mirror_num
,
6921 unsigned long bio_flags
, u64 offset
)
6924 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6925 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
6926 BUG_ON(ret
); /* -ENOMEM */
6930 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
6932 struct btrfs_dio_private
*dip
= bio
->bi_private
;
6935 printk(KERN_ERR
"btrfs direct IO failed ino %llu rw %lu "
6936 "sector %#Lx len %u err no %d\n",
6937 (unsigned long long)btrfs_ino(dip
->inode
), bio
->bi_rw
,
6938 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
6942 * before atomic variable goto zero, we must make sure
6943 * dip->errors is perceived to be set.
6945 smp_mb__before_atomic_dec();
6948 /* if there are more bios still pending for this dio, just exit */
6949 if (!atomic_dec_and_test(&dip
->pending_bios
))
6953 bio_io_error(dip
->orig_bio
);
6955 set_bit(BIO_UPTODATE
, &dip
->dio_bio
->bi_flags
);
6956 bio_endio(dip
->orig_bio
, 0);
6962 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
6963 u64 first_sector
, gfp_t gfp_flags
)
6965 int nr_vecs
= bio_get_nr_vecs(bdev
);
6966 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
6969 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
6970 int rw
, u64 file_offset
, int skip_sum
,
6973 int write
= rw
& REQ_WRITE
;
6974 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6978 async_submit
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
6983 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
6991 if (write
&& async_submit
) {
6992 ret
= btrfs_wq_submit_bio(root
->fs_info
,
6993 inode
, rw
, bio
, 0, 0,
6995 __btrfs_submit_bio_start_direct_io
,
6996 __btrfs_submit_bio_done
);
7000 * If we aren't doing async submit, calculate the csum of the
7003 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
7006 } else if (!skip_sum
) {
7007 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, bio
, file_offset
);
7013 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
7019 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
7022 struct inode
*inode
= dip
->inode
;
7023 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7025 struct bio
*orig_bio
= dip
->orig_bio
;
7026 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
7027 u64 start_sector
= orig_bio
->bi_sector
;
7028 u64 file_offset
= dip
->logical_offset
;
7033 int async_submit
= 0;
7035 map_length
= orig_bio
->bi_size
;
7036 ret
= btrfs_map_block(root
->fs_info
, rw
, start_sector
<< 9,
7037 &map_length
, NULL
, 0);
7042 if (map_length
>= orig_bio
->bi_size
) {
7047 /* async crcs make it difficult to collect full stripe writes. */
7048 if (btrfs_get_alloc_profile(root
, 1) &
7049 (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
))
7054 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
7057 bio
->bi_private
= dip
;
7058 bio
->bi_end_io
= btrfs_end_dio_bio
;
7059 atomic_inc(&dip
->pending_bios
);
7061 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
7062 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
7063 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
7064 bvec
->bv_offset
) < bvec
->bv_len
)) {
7066 * inc the count before we submit the bio so
7067 * we know the end IO handler won't happen before
7068 * we inc the count. Otherwise, the dip might get freed
7069 * before we're done setting it up
7071 atomic_inc(&dip
->pending_bios
);
7072 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
7073 file_offset
, skip_sum
,
7077 atomic_dec(&dip
->pending_bios
);
7081 start_sector
+= submit_len
>> 9;
7082 file_offset
+= submit_len
;
7087 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
7088 start_sector
, GFP_NOFS
);
7091 bio
->bi_private
= dip
;
7092 bio
->bi_end_io
= btrfs_end_dio_bio
;
7094 map_length
= orig_bio
->bi_size
;
7095 ret
= btrfs_map_block(root
->fs_info
, rw
,
7097 &map_length
, NULL
, 0);
7103 submit_len
+= bvec
->bv_len
;
7110 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
7119 * before atomic variable goto zero, we must
7120 * make sure dip->errors is perceived to be set.
7122 smp_mb__before_atomic_dec();
7123 if (atomic_dec_and_test(&dip
->pending_bios
))
7124 bio_io_error(dip
->orig_bio
);
7126 /* bio_end_io() will handle error, so we needn't return it */
7130 static void btrfs_submit_direct(int rw
, struct bio
*dio_bio
,
7131 struct inode
*inode
, loff_t file_offset
)
7133 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7134 struct btrfs_dio_private
*dip
;
7137 int write
= rw
& REQ_WRITE
;
7140 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
7142 io_bio
= btrfs_bio_clone(dio_bio
, GFP_NOFS
);
7149 dip
= kmalloc(sizeof(*dip
), GFP_NOFS
);
7155 dip
->private = dio_bio
->bi_private
;
7157 dip
->logical_offset
= file_offset
;
7158 dip
->bytes
= dio_bio
->bi_size
;
7159 dip
->disk_bytenr
= (u64
)dio_bio
->bi_sector
<< 9;
7160 io_bio
->bi_private
= dip
;
7162 dip
->orig_bio
= io_bio
;
7163 dip
->dio_bio
= dio_bio
;
7164 atomic_set(&dip
->pending_bios
, 0);
7167 io_bio
->bi_end_io
= btrfs_endio_direct_write
;
7169 io_bio
->bi_end_io
= btrfs_endio_direct_read
;
7171 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
7180 * If this is a write, we need to clean up the reserved space and kill
7181 * the ordered extent.
7184 struct btrfs_ordered_extent
*ordered
;
7185 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
7186 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
7187 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
7188 btrfs_free_reserved_extent(root
, ordered
->start
,
7190 btrfs_put_ordered_extent(ordered
);
7191 btrfs_put_ordered_extent(ordered
);
7193 bio_endio(dio_bio
, ret
);
7196 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
7197 const struct iovec
*iov
, loff_t offset
,
7198 unsigned long nr_segs
)
7204 unsigned blocksize_mask
= root
->sectorsize
- 1;
7205 ssize_t retval
= -EINVAL
;
7206 loff_t end
= offset
;
7208 if (offset
& blocksize_mask
)
7211 /* Check the memory alignment. Blocks cannot straddle pages */
7212 for (seg
= 0; seg
< nr_segs
; seg
++) {
7213 addr
= (unsigned long)iov
[seg
].iov_base
;
7214 size
= iov
[seg
].iov_len
;
7216 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
7219 /* If this is a write we don't need to check anymore */
7224 * Check to make sure we don't have duplicate iov_base's in this
7225 * iovec, if so return EINVAL, otherwise we'll get csum errors
7226 * when reading back.
7228 for (i
= seg
+ 1; i
< nr_segs
; i
++) {
7229 if (iov
[seg
].iov_base
== iov
[i
].iov_base
)
7238 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
7239 const struct iovec
*iov
, loff_t offset
,
7240 unsigned long nr_segs
)
7242 struct file
*file
= iocb
->ki_filp
;
7243 struct inode
*inode
= file
->f_mapping
->host
;
7247 bool relock
= false;
7250 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
7254 atomic_inc(&inode
->i_dio_count
);
7255 smp_mb__after_atomic_inc();
7258 * The generic stuff only does filemap_write_and_wait_range, which isn't
7259 * enough if we've written compressed pages to this area, so we need to
7260 * call btrfs_wait_ordered_range to make absolutely sure that any
7261 * outstanding dirty pages are on disk.
7263 count
= iov_length(iov
, nr_segs
);
7264 btrfs_wait_ordered_range(inode
, offset
, count
);
7268 * If the write DIO is beyond the EOF, we need update
7269 * the isize, but it is protected by i_mutex. So we can
7270 * not unlock the i_mutex at this case.
7272 if (offset
+ count
<= inode
->i_size
) {
7273 mutex_unlock(&inode
->i_mutex
);
7276 ret
= btrfs_delalloc_reserve_space(inode
, count
);
7279 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK
,
7280 &BTRFS_I(inode
)->runtime_flags
))) {
7281 inode_dio_done(inode
);
7282 flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
7286 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
7287 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
7288 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
7289 btrfs_submit_direct
, flags
);
7291 if (ret
< 0 && ret
!= -EIOCBQUEUED
)
7292 btrfs_delalloc_release_space(inode
, count
);
7293 else if (ret
>= 0 && (size_t)ret
< count
)
7294 btrfs_delalloc_release_space(inode
,
7295 count
- (size_t)ret
);
7297 btrfs_delalloc_release_metadata(inode
, 0);
7301 inode_dio_done(inode
);
7303 mutex_lock(&inode
->i_mutex
);
7308 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7310 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
7311 __u64 start
, __u64 len
)
7315 ret
= fiemap_check_flags(fieinfo
, BTRFS_FIEMAP_FLAGS
);
7319 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
7322 int btrfs_readpage(struct file
*file
, struct page
*page
)
7324 struct extent_io_tree
*tree
;
7325 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
7326 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
7329 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
7331 struct extent_io_tree
*tree
;
7334 if (current
->flags
& PF_MEMALLOC
) {
7335 redirty_page_for_writepage(wbc
, page
);
7339 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
7340 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
7343 static int btrfs_writepages(struct address_space
*mapping
,
7344 struct writeback_control
*wbc
)
7346 struct extent_io_tree
*tree
;
7348 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
7349 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
7353 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
7354 struct list_head
*pages
, unsigned nr_pages
)
7356 struct extent_io_tree
*tree
;
7357 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
7358 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
7361 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
7363 struct extent_io_tree
*tree
;
7364 struct extent_map_tree
*map
;
7367 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
7368 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
7369 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
7371 ClearPagePrivate(page
);
7372 set_page_private(page
, 0);
7373 page_cache_release(page
);
7378 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
7380 if (PageWriteback(page
) || PageDirty(page
))
7382 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
7385 static void btrfs_invalidatepage(struct page
*page
, unsigned int offset
,
7386 unsigned int length
)
7388 struct inode
*inode
= page
->mapping
->host
;
7389 struct extent_io_tree
*tree
;
7390 struct btrfs_ordered_extent
*ordered
;
7391 struct extent_state
*cached_state
= NULL
;
7392 u64 page_start
= page_offset(page
);
7393 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
7396 * we have the page locked, so new writeback can't start,
7397 * and the dirty bit won't be cleared while we are here.
7399 * Wait for IO on this page so that we can safely clear
7400 * the PagePrivate2 bit and do ordered accounting
7402 wait_on_page_writeback(page
);
7404 tree
= &BTRFS_I(inode
)->io_tree
;
7406 btrfs_releasepage(page
, GFP_NOFS
);
7409 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
);
7410 ordered
= btrfs_lookup_ordered_extent(inode
, page_offset(page
));
7413 * IO on this page will never be started, so we need
7414 * to account for any ordered extents now
7416 clear_extent_bit(tree
, page_start
, page_end
,
7417 EXTENT_DIRTY
| EXTENT_DELALLOC
|
7418 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
7419 EXTENT_DEFRAG
, 1, 0, &cached_state
, GFP_NOFS
);
7421 * whoever cleared the private bit is responsible
7422 * for the finish_ordered_io
7424 if (TestClearPagePrivate2(page
) &&
7425 btrfs_dec_test_ordered_pending(inode
, &ordered
, page_start
,
7426 PAGE_CACHE_SIZE
, 1)) {
7427 btrfs_finish_ordered_io(ordered
);
7429 btrfs_put_ordered_extent(ordered
);
7430 cached_state
= NULL
;
7431 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
);
7433 clear_extent_bit(tree
, page_start
, page_end
,
7434 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
7435 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
, 1, 1,
7436 &cached_state
, GFP_NOFS
);
7437 __btrfs_releasepage(page
, GFP_NOFS
);
7439 ClearPageChecked(page
);
7440 if (PagePrivate(page
)) {
7441 ClearPagePrivate(page
);
7442 set_page_private(page
, 0);
7443 page_cache_release(page
);
7448 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7449 * called from a page fault handler when a page is first dirtied. Hence we must
7450 * be careful to check for EOF conditions here. We set the page up correctly
7451 * for a written page which means we get ENOSPC checking when writing into
7452 * holes and correct delalloc and unwritten extent mapping on filesystems that
7453 * support these features.
7455 * We are not allowed to take the i_mutex here so we have to play games to
7456 * protect against truncate races as the page could now be beyond EOF. Because
7457 * vmtruncate() writes the inode size before removing pages, once we have the
7458 * page lock we can determine safely if the page is beyond EOF. If it is not
7459 * beyond EOF, then the page is guaranteed safe against truncation until we
7462 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
7464 struct page
*page
= vmf
->page
;
7465 struct inode
*inode
= file_inode(vma
->vm_file
);
7466 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7467 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7468 struct btrfs_ordered_extent
*ordered
;
7469 struct extent_state
*cached_state
= NULL
;
7471 unsigned long zero_start
;
7478 sb_start_pagefault(inode
->i_sb
);
7479 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
7481 ret
= file_update_time(vma
->vm_file
);
7487 else /* -ENOSPC, -EIO, etc */
7488 ret
= VM_FAULT_SIGBUS
;
7494 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
7497 size
= i_size_read(inode
);
7498 page_start
= page_offset(page
);
7499 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
7501 if ((page
->mapping
!= inode
->i_mapping
) ||
7502 (page_start
>= size
)) {
7503 /* page got truncated out from underneath us */
7506 wait_on_page_writeback(page
);
7508 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
7509 set_page_extent_mapped(page
);
7512 * we can't set the delalloc bits if there are pending ordered
7513 * extents. Drop our locks and wait for them to finish
7515 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
7517 unlock_extent_cached(io_tree
, page_start
, page_end
,
7518 &cached_state
, GFP_NOFS
);
7520 btrfs_start_ordered_extent(inode
, ordered
, 1);
7521 btrfs_put_ordered_extent(ordered
);
7526 * XXX - page_mkwrite gets called every time the page is dirtied, even
7527 * if it was already dirty, so for space accounting reasons we need to
7528 * clear any delalloc bits for the range we are fixing to save. There
7529 * is probably a better way to do this, but for now keep consistent with
7530 * prepare_pages in the normal write path.
7532 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
7533 EXTENT_DIRTY
| EXTENT_DELALLOC
|
7534 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
7535 0, 0, &cached_state
, GFP_NOFS
);
7537 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
7540 unlock_extent_cached(io_tree
, page_start
, page_end
,
7541 &cached_state
, GFP_NOFS
);
7542 ret
= VM_FAULT_SIGBUS
;
7547 /* page is wholly or partially inside EOF */
7548 if (page_start
+ PAGE_CACHE_SIZE
> size
)
7549 zero_start
= size
& ~PAGE_CACHE_MASK
;
7551 zero_start
= PAGE_CACHE_SIZE
;
7553 if (zero_start
!= PAGE_CACHE_SIZE
) {
7555 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
7556 flush_dcache_page(page
);
7559 ClearPageChecked(page
);
7560 set_page_dirty(page
);
7561 SetPageUptodate(page
);
7563 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
7564 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
7565 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
7567 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
7571 sb_end_pagefault(inode
->i_sb
);
7572 return VM_FAULT_LOCKED
;
7576 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
7578 sb_end_pagefault(inode
->i_sb
);
7582 static int btrfs_truncate(struct inode
*inode
)
7584 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7585 struct btrfs_block_rsv
*rsv
;
7588 struct btrfs_trans_handle
*trans
;
7589 u64 mask
= root
->sectorsize
- 1;
7590 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
7592 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
7593 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
7596 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7597 * 3 things going on here
7599 * 1) We need to reserve space for our orphan item and the space to
7600 * delete our orphan item. Lord knows we don't want to have a dangling
7601 * orphan item because we didn't reserve space to remove it.
7603 * 2) We need to reserve space to update our inode.
7605 * 3) We need to have something to cache all the space that is going to
7606 * be free'd up by the truncate operation, but also have some slack
7607 * space reserved in case it uses space during the truncate (thank you
7608 * very much snapshotting).
7610 * And we need these to all be seperate. The fact is we can use alot of
7611 * space doing the truncate, and we have no earthly idea how much space
7612 * we will use, so we need the truncate reservation to be seperate so it
7613 * doesn't end up using space reserved for updating the inode or
7614 * removing the orphan item. We also need to be able to stop the
7615 * transaction and start a new one, which means we need to be able to
7616 * update the inode several times, and we have no idea of knowing how
7617 * many times that will be, so we can't just reserve 1 item for the
7618 * entirety of the opration, so that has to be done seperately as well.
7619 * Then there is the orphan item, which does indeed need to be held on
7620 * to for the whole operation, and we need nobody to touch this reserved
7621 * space except the orphan code.
7623 * So that leaves us with
7625 * 1) root->orphan_block_rsv - for the orphan deletion.
7626 * 2) rsv - for the truncate reservation, which we will steal from the
7627 * transaction reservation.
7628 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7629 * updating the inode.
7631 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
7634 rsv
->size
= min_size
;
7638 * 1 for the truncate slack space
7639 * 1 for updating the inode.
7641 trans
= btrfs_start_transaction(root
, 2);
7642 if (IS_ERR(trans
)) {
7643 err
= PTR_ERR(trans
);
7647 /* Migrate the slack space for the truncate to our reserve */
7648 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
7653 * setattr is responsible for setting the ordered_data_close flag,
7654 * but that is only tested during the last file release. That
7655 * could happen well after the next commit, leaving a great big
7656 * window where new writes may get lost if someone chooses to write
7657 * to this file after truncating to zero
7659 * The inode doesn't have any dirty data here, and so if we commit
7660 * this is a noop. If someone immediately starts writing to the inode
7661 * it is very likely we'll catch some of their writes in this
7662 * transaction, and the commit will find this file on the ordered
7663 * data list with good things to send down.
7665 * This is a best effort solution, there is still a window where
7666 * using truncate to replace the contents of the file will
7667 * end up with a zero length file after a crash.
7669 if (inode
->i_size
== 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
7670 &BTRFS_I(inode
)->runtime_flags
))
7671 btrfs_add_ordered_operation(trans
, root
, inode
);
7674 * So if we truncate and then write and fsync we normally would just
7675 * write the extents that changed, which is a problem if we need to
7676 * first truncate that entire inode. So set this flag so we write out
7677 * all of the extents in the inode to the sync log so we're completely
7680 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
7681 trans
->block_rsv
= rsv
;
7684 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
7686 BTRFS_EXTENT_DATA_KEY
);
7687 if (ret
!= -ENOSPC
) {
7692 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
7693 ret
= btrfs_update_inode(trans
, root
, inode
);
7699 btrfs_end_transaction(trans
, root
);
7700 btrfs_btree_balance_dirty(root
);
7702 trans
= btrfs_start_transaction(root
, 2);
7703 if (IS_ERR(trans
)) {
7704 ret
= err
= PTR_ERR(trans
);
7709 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
,
7711 BUG_ON(ret
); /* shouldn't happen */
7712 trans
->block_rsv
= rsv
;
7715 if (ret
== 0 && inode
->i_nlink
> 0) {
7716 trans
->block_rsv
= root
->orphan_block_rsv
;
7717 ret
= btrfs_orphan_del(trans
, inode
);
7723 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
7724 ret
= btrfs_update_inode(trans
, root
, inode
);
7728 ret
= btrfs_end_transaction(trans
, root
);
7729 btrfs_btree_balance_dirty(root
);
7733 btrfs_free_block_rsv(root
, rsv
);
7742 * create a new subvolume directory/inode (helper for the ioctl).
7744 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
7745 struct btrfs_root
*new_root
, u64 new_dirid
)
7747 struct inode
*inode
;
7751 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
7752 new_dirid
, new_dirid
,
7753 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
7756 return PTR_ERR(inode
);
7757 inode
->i_op
= &btrfs_dir_inode_operations
;
7758 inode
->i_fop
= &btrfs_dir_file_operations
;
7760 set_nlink(inode
, 1);
7761 btrfs_i_size_write(inode
, 0);
7763 err
= btrfs_update_inode(trans
, new_root
, inode
);
7769 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
7771 struct btrfs_inode
*ei
;
7772 struct inode
*inode
;
7774 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
7781 ei
->last_sub_trans
= 0;
7782 ei
->logged_trans
= 0;
7783 ei
->delalloc_bytes
= 0;
7784 ei
->disk_i_size
= 0;
7787 ei
->index_cnt
= (u64
)-1;
7788 ei
->last_unlink_trans
= 0;
7789 ei
->last_log_commit
= 0;
7791 spin_lock_init(&ei
->lock
);
7792 ei
->outstanding_extents
= 0;
7793 ei
->reserved_extents
= 0;
7795 ei
->runtime_flags
= 0;
7796 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
7798 ei
->delayed_node
= NULL
;
7800 inode
= &ei
->vfs_inode
;
7801 extent_map_tree_init(&ei
->extent_tree
);
7802 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
7803 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
7804 ei
->io_tree
.track_uptodate
= 1;
7805 ei
->io_failure_tree
.track_uptodate
= 1;
7806 atomic_set(&ei
->sync_writers
, 0);
7807 mutex_init(&ei
->log_mutex
);
7808 mutex_init(&ei
->delalloc_mutex
);
7809 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
7810 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
7811 INIT_LIST_HEAD(&ei
->ordered_operations
);
7812 RB_CLEAR_NODE(&ei
->rb_node
);
7817 static void btrfs_i_callback(struct rcu_head
*head
)
7819 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
7820 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
7823 void btrfs_destroy_inode(struct inode
*inode
)
7825 struct btrfs_ordered_extent
*ordered
;
7826 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7828 WARN_ON(!hlist_empty(&inode
->i_dentry
));
7829 WARN_ON(inode
->i_data
.nrpages
);
7830 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
7831 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
7832 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
7833 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
7836 * This can happen where we create an inode, but somebody else also
7837 * created the same inode and we need to destroy the one we already
7844 * Make sure we're properly removed from the ordered operation
7848 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
7849 spin_lock(&root
->fs_info
->ordered_root_lock
);
7850 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
7851 spin_unlock(&root
->fs_info
->ordered_root_lock
);
7854 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
7855 &BTRFS_I(inode
)->runtime_flags
)) {
7856 btrfs_info(root
->fs_info
, "inode %llu still on the orphan list",
7857 (unsigned long long)btrfs_ino(inode
));
7858 atomic_dec(&root
->orphan_inodes
);
7862 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
7866 btrfs_err(root
->fs_info
, "found ordered extent %llu %llu on inode cleanup",
7867 (unsigned long long)ordered
->file_offset
,
7868 (unsigned long long)ordered
->len
);
7869 btrfs_remove_ordered_extent(inode
, ordered
);
7870 btrfs_put_ordered_extent(ordered
);
7871 btrfs_put_ordered_extent(ordered
);
7874 inode_tree_del(inode
);
7875 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
7877 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
7880 int btrfs_drop_inode(struct inode
*inode
)
7882 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7887 /* the snap/subvol tree is on deleting */
7888 if (btrfs_root_refs(&root
->root_item
) == 0 &&
7889 root
!= root
->fs_info
->tree_root
)
7892 return generic_drop_inode(inode
);
7895 static void init_once(void *foo
)
7897 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
7899 inode_init_once(&ei
->vfs_inode
);
7902 void btrfs_destroy_cachep(void)
7905 * Make sure all delayed rcu free inodes are flushed before we
7909 if (btrfs_inode_cachep
)
7910 kmem_cache_destroy(btrfs_inode_cachep
);
7911 if (btrfs_trans_handle_cachep
)
7912 kmem_cache_destroy(btrfs_trans_handle_cachep
);
7913 if (btrfs_transaction_cachep
)
7914 kmem_cache_destroy(btrfs_transaction_cachep
);
7915 if (btrfs_path_cachep
)
7916 kmem_cache_destroy(btrfs_path_cachep
);
7917 if (btrfs_free_space_cachep
)
7918 kmem_cache_destroy(btrfs_free_space_cachep
);
7919 if (btrfs_delalloc_work_cachep
)
7920 kmem_cache_destroy(btrfs_delalloc_work_cachep
);
7923 int btrfs_init_cachep(void)
7925 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
7926 sizeof(struct btrfs_inode
), 0,
7927 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
7928 if (!btrfs_inode_cachep
)
7931 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
7932 sizeof(struct btrfs_trans_handle
), 0,
7933 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7934 if (!btrfs_trans_handle_cachep
)
7937 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction",
7938 sizeof(struct btrfs_transaction
), 0,
7939 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7940 if (!btrfs_transaction_cachep
)
7943 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
7944 sizeof(struct btrfs_path
), 0,
7945 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7946 if (!btrfs_path_cachep
)
7949 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
7950 sizeof(struct btrfs_free_space
), 0,
7951 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7952 if (!btrfs_free_space_cachep
)
7955 btrfs_delalloc_work_cachep
= kmem_cache_create("btrfs_delalloc_work",
7956 sizeof(struct btrfs_delalloc_work
), 0,
7957 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
7959 if (!btrfs_delalloc_work_cachep
)
7964 btrfs_destroy_cachep();
7968 static int btrfs_getattr(struct vfsmount
*mnt
,
7969 struct dentry
*dentry
, struct kstat
*stat
)
7972 struct inode
*inode
= dentry
->d_inode
;
7973 u32 blocksize
= inode
->i_sb
->s_blocksize
;
7975 generic_fillattr(inode
, stat
);
7976 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
7977 stat
->blksize
= PAGE_CACHE_SIZE
;
7979 spin_lock(&BTRFS_I(inode
)->lock
);
7980 delalloc_bytes
= BTRFS_I(inode
)->delalloc_bytes
;
7981 spin_unlock(&BTRFS_I(inode
)->lock
);
7982 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
7983 ALIGN(delalloc_bytes
, blocksize
)) >> 9;
7987 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
7988 struct inode
*new_dir
, struct dentry
*new_dentry
)
7990 struct btrfs_trans_handle
*trans
;
7991 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
7992 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
7993 struct inode
*new_inode
= new_dentry
->d_inode
;
7994 struct inode
*old_inode
= old_dentry
->d_inode
;
7995 struct timespec ctime
= CURRENT_TIME
;
7999 u64 old_ino
= btrfs_ino(old_inode
);
8001 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
8004 /* we only allow rename subvolume link between subvolumes */
8005 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
8008 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
8009 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
8012 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
8013 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
8017 /* check for collisions, even if the name isn't there */
8018 ret
= btrfs_check_dir_item_collision(root
, new_dir
->i_ino
,
8019 new_dentry
->d_name
.name
,
8020 new_dentry
->d_name
.len
);
8023 if (ret
== -EEXIST
) {
8025 * eexist without a new_inode */
8031 /* maybe -EOVERFLOW */
8038 * we're using rename to replace one file with another.
8039 * and the replacement file is large. Start IO on it now so
8040 * we don't add too much work to the end of the transaction
8042 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
8043 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
8044 filemap_flush(old_inode
->i_mapping
);
8046 /* close the racy window with snapshot create/destroy ioctl */
8047 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
8048 down_read(&root
->fs_info
->subvol_sem
);
8050 * We want to reserve the absolute worst case amount of items. So if
8051 * both inodes are subvols and we need to unlink them then that would
8052 * require 4 item modifications, but if they are both normal inodes it
8053 * would require 5 item modifications, so we'll assume their normal
8054 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8055 * should cover the worst case number of items we'll modify.
8057 trans
= btrfs_start_transaction(root
, 11);
8058 if (IS_ERR(trans
)) {
8059 ret
= PTR_ERR(trans
);
8064 btrfs_record_root_in_trans(trans
, dest
);
8066 ret
= btrfs_set_inode_index(new_dir
, &index
);
8070 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
8071 /* force full log commit if subvolume involved. */
8072 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
8074 ret
= btrfs_insert_inode_ref(trans
, dest
,
8075 new_dentry
->d_name
.name
,
8076 new_dentry
->d_name
.len
,
8078 btrfs_ino(new_dir
), index
);
8082 * this is an ugly little race, but the rename is required
8083 * to make sure that if we crash, the inode is either at the
8084 * old name or the new one. pinning the log transaction lets
8085 * us make sure we don't allow a log commit to come in after
8086 * we unlink the name but before we add the new name back in.
8088 btrfs_pin_log_trans(root
);
8091 * make sure the inode gets flushed if it is replacing
8094 if (new_inode
&& new_inode
->i_size
&& S_ISREG(old_inode
->i_mode
))
8095 btrfs_add_ordered_operation(trans
, root
, old_inode
);
8097 inode_inc_iversion(old_dir
);
8098 inode_inc_iversion(new_dir
);
8099 inode_inc_iversion(old_inode
);
8100 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
8101 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
8102 old_inode
->i_ctime
= ctime
;
8104 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
8105 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
8107 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
8108 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
8109 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
8110 old_dentry
->d_name
.name
,
8111 old_dentry
->d_name
.len
);
8113 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
8114 old_dentry
->d_inode
,
8115 old_dentry
->d_name
.name
,
8116 old_dentry
->d_name
.len
);
8118 ret
= btrfs_update_inode(trans
, root
, old_inode
);
8121 btrfs_abort_transaction(trans
, root
, ret
);
8126 inode_inc_iversion(new_inode
);
8127 new_inode
->i_ctime
= CURRENT_TIME
;
8128 if (unlikely(btrfs_ino(new_inode
) ==
8129 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
8130 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
8131 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
8133 new_dentry
->d_name
.name
,
8134 new_dentry
->d_name
.len
);
8135 BUG_ON(new_inode
->i_nlink
== 0);
8137 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
8138 new_dentry
->d_inode
,
8139 new_dentry
->d_name
.name
,
8140 new_dentry
->d_name
.len
);
8142 if (!ret
&& new_inode
->i_nlink
== 0) {
8143 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
8147 btrfs_abort_transaction(trans
, root
, ret
);
8152 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
8153 new_dentry
->d_name
.name
,
8154 new_dentry
->d_name
.len
, 0, index
);
8156 btrfs_abort_transaction(trans
, root
, ret
);
8160 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
8161 struct dentry
*parent
= new_dentry
->d_parent
;
8162 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
8163 btrfs_end_log_trans(root
);
8166 btrfs_end_transaction(trans
, root
);
8168 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
8169 up_read(&root
->fs_info
->subvol_sem
);
8174 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
8176 struct btrfs_delalloc_work
*delalloc_work
;
8178 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
8180 if (delalloc_work
->wait
)
8181 btrfs_wait_ordered_range(delalloc_work
->inode
, 0, (u64
)-1);
8183 filemap_flush(delalloc_work
->inode
->i_mapping
);
8185 if (delalloc_work
->delay_iput
)
8186 btrfs_add_delayed_iput(delalloc_work
->inode
);
8188 iput(delalloc_work
->inode
);
8189 complete(&delalloc_work
->completion
);
8192 struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
,
8193 int wait
, int delay_iput
)
8195 struct btrfs_delalloc_work
*work
;
8197 work
= kmem_cache_zalloc(btrfs_delalloc_work_cachep
, GFP_NOFS
);
8201 init_completion(&work
->completion
);
8202 INIT_LIST_HEAD(&work
->list
);
8203 work
->inode
= inode
;
8205 work
->delay_iput
= delay_iput
;
8206 work
->work
.func
= btrfs_run_delalloc_work
;
8211 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work
*work
)
8213 wait_for_completion(&work
->completion
);
8214 kmem_cache_free(btrfs_delalloc_work_cachep
, work
);
8218 * some fairly slow code that needs optimization. This walks the list
8219 * of all the inodes with pending delalloc and forces them to disk.
8221 static int __start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
8223 struct btrfs_inode
*binode
;
8224 struct inode
*inode
;
8225 struct btrfs_delalloc_work
*work
, *next
;
8226 struct list_head works
;
8227 struct list_head splice
;
8230 INIT_LIST_HEAD(&works
);
8231 INIT_LIST_HEAD(&splice
);
8233 spin_lock(&root
->delalloc_lock
);
8234 list_splice_init(&root
->delalloc_inodes
, &splice
);
8235 while (!list_empty(&splice
)) {
8236 binode
= list_entry(splice
.next
, struct btrfs_inode
,
8239 list_move_tail(&binode
->delalloc_inodes
,
8240 &root
->delalloc_inodes
);
8241 inode
= igrab(&binode
->vfs_inode
);
8243 cond_resched_lock(&root
->delalloc_lock
);
8246 spin_unlock(&root
->delalloc_lock
);
8248 work
= btrfs_alloc_delalloc_work(inode
, 0, delay_iput
);
8249 if (unlikely(!work
)) {
8253 list_add_tail(&work
->list
, &works
);
8254 btrfs_queue_worker(&root
->fs_info
->flush_workers
,
8258 spin_lock(&root
->delalloc_lock
);
8260 spin_unlock(&root
->delalloc_lock
);
8262 list_for_each_entry_safe(work
, next
, &works
, list
) {
8263 list_del_init(&work
->list
);
8264 btrfs_wait_and_free_delalloc_work(work
);
8268 list_for_each_entry_safe(work
, next
, &works
, list
) {
8269 list_del_init(&work
->list
);
8270 btrfs_wait_and_free_delalloc_work(work
);
8273 if (!list_empty_careful(&splice
)) {
8274 spin_lock(&root
->delalloc_lock
);
8275 list_splice_tail(&splice
, &root
->delalloc_inodes
);
8276 spin_unlock(&root
->delalloc_lock
);
8281 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
8285 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
8288 ret
= __start_delalloc_inodes(root
, delay_iput
);
8290 * the filemap_flush will queue IO into the worker threads, but
8291 * we have to make sure the IO is actually started and that
8292 * ordered extents get created before we return
8294 atomic_inc(&root
->fs_info
->async_submit_draining
);
8295 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
8296 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
8297 wait_event(root
->fs_info
->async_submit_wait
,
8298 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
8299 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
8301 atomic_dec(&root
->fs_info
->async_submit_draining
);
8305 int btrfs_start_all_delalloc_inodes(struct btrfs_fs_info
*fs_info
,
8308 struct btrfs_root
*root
;
8309 struct list_head splice
;
8312 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
8315 INIT_LIST_HEAD(&splice
);
8317 spin_lock(&fs_info
->delalloc_root_lock
);
8318 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
8319 while (!list_empty(&splice
)) {
8320 root
= list_first_entry(&splice
, struct btrfs_root
,
8322 root
= btrfs_grab_fs_root(root
);
8324 list_move_tail(&root
->delalloc_root
,
8325 &fs_info
->delalloc_roots
);
8326 spin_unlock(&fs_info
->delalloc_root_lock
);
8328 ret
= __start_delalloc_inodes(root
, delay_iput
);
8329 btrfs_put_fs_root(root
);
8333 spin_lock(&fs_info
->delalloc_root_lock
);
8335 spin_unlock(&fs_info
->delalloc_root_lock
);
8337 atomic_inc(&fs_info
->async_submit_draining
);
8338 while (atomic_read(&fs_info
->nr_async_submits
) ||
8339 atomic_read(&fs_info
->async_delalloc_pages
)) {
8340 wait_event(fs_info
->async_submit_wait
,
8341 (atomic_read(&fs_info
->nr_async_submits
) == 0 &&
8342 atomic_read(&fs_info
->async_delalloc_pages
) == 0));
8344 atomic_dec(&fs_info
->async_submit_draining
);
8347 if (!list_empty_careful(&splice
)) {
8348 spin_lock(&fs_info
->delalloc_root_lock
);
8349 list_splice_tail(&splice
, &fs_info
->delalloc_roots
);
8350 spin_unlock(&fs_info
->delalloc_root_lock
);
8355 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
8356 const char *symname
)
8358 struct btrfs_trans_handle
*trans
;
8359 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
8360 struct btrfs_path
*path
;
8361 struct btrfs_key key
;
8362 struct inode
*inode
= NULL
;
8370 struct btrfs_file_extent_item
*ei
;
8371 struct extent_buffer
*leaf
;
8373 name_len
= strlen(symname
) + 1;
8374 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
8375 return -ENAMETOOLONG
;
8378 * 2 items for inode item and ref
8379 * 2 items for dir items
8380 * 1 item for xattr if selinux is on
8382 trans
= btrfs_start_transaction(root
, 5);
8384 return PTR_ERR(trans
);
8386 err
= btrfs_find_free_ino(root
, &objectid
);
8390 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
8391 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
8392 S_IFLNK
|S_IRWXUGO
, &index
);
8393 if (IS_ERR(inode
)) {
8394 err
= PTR_ERR(inode
);
8398 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
8405 * If the active LSM wants to access the inode during
8406 * d_instantiate it needs these. Smack checks to see
8407 * if the filesystem supports xattrs by looking at the
8410 inode
->i_fop
= &btrfs_file_operations
;
8411 inode
->i_op
= &btrfs_file_inode_operations
;
8413 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
8417 inode
->i_mapping
->a_ops
= &btrfs_aops
;
8418 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
8419 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
8424 path
= btrfs_alloc_path();
8430 key
.objectid
= btrfs_ino(inode
);
8432 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
8433 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
8434 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
8438 btrfs_free_path(path
);
8441 leaf
= path
->nodes
[0];
8442 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
8443 struct btrfs_file_extent_item
);
8444 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
8445 btrfs_set_file_extent_type(leaf
, ei
,
8446 BTRFS_FILE_EXTENT_INLINE
);
8447 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
8448 btrfs_set_file_extent_compression(leaf
, ei
, 0);
8449 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
8450 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
8452 ptr
= btrfs_file_extent_inline_start(ei
);
8453 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
8454 btrfs_mark_buffer_dirty(leaf
);
8455 btrfs_free_path(path
);
8457 inode
->i_op
= &btrfs_symlink_inode_operations
;
8458 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
8459 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
8460 inode_set_bytes(inode
, name_len
);
8461 btrfs_i_size_write(inode
, name_len
- 1);
8462 err
= btrfs_update_inode(trans
, root
, inode
);
8468 d_instantiate(dentry
, inode
);
8469 btrfs_end_transaction(trans
, root
);
8471 inode_dec_link_count(inode
);
8474 btrfs_btree_balance_dirty(root
);
8478 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
8479 u64 start
, u64 num_bytes
, u64 min_size
,
8480 loff_t actual_len
, u64
*alloc_hint
,
8481 struct btrfs_trans_handle
*trans
)
8483 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
8484 struct extent_map
*em
;
8485 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8486 struct btrfs_key ins
;
8487 u64 cur_offset
= start
;
8491 bool own_trans
= true;
8495 while (num_bytes
> 0) {
8497 trans
= btrfs_start_transaction(root
, 3);
8498 if (IS_ERR(trans
)) {
8499 ret
= PTR_ERR(trans
);
8504 cur_bytes
= min(num_bytes
, 256ULL * 1024 * 1024);
8505 cur_bytes
= max(cur_bytes
, min_size
);
8506 ret
= btrfs_reserve_extent(trans
, root
, cur_bytes
,
8507 min_size
, 0, *alloc_hint
, &ins
, 1);
8510 btrfs_end_transaction(trans
, root
);
8514 ret
= insert_reserved_file_extent(trans
, inode
,
8515 cur_offset
, ins
.objectid
,
8516 ins
.offset
, ins
.offset
,
8517 ins
.offset
, 0, 0, 0,
8518 BTRFS_FILE_EXTENT_PREALLOC
);
8520 btrfs_abort_transaction(trans
, root
, ret
);
8522 btrfs_end_transaction(trans
, root
);
8525 btrfs_drop_extent_cache(inode
, cur_offset
,
8526 cur_offset
+ ins
.offset
-1, 0);
8528 em
= alloc_extent_map();
8530 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
8531 &BTRFS_I(inode
)->runtime_flags
);
8535 em
->start
= cur_offset
;
8536 em
->orig_start
= cur_offset
;
8537 em
->len
= ins
.offset
;
8538 em
->block_start
= ins
.objectid
;
8539 em
->block_len
= ins
.offset
;
8540 em
->orig_block_len
= ins
.offset
;
8541 em
->ram_bytes
= ins
.offset
;
8542 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
8543 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
8544 em
->generation
= trans
->transid
;
8547 write_lock(&em_tree
->lock
);
8548 ret
= add_extent_mapping(em_tree
, em
, 1);
8549 write_unlock(&em_tree
->lock
);
8552 btrfs_drop_extent_cache(inode
, cur_offset
,
8553 cur_offset
+ ins
.offset
- 1,
8556 free_extent_map(em
);
8558 num_bytes
-= ins
.offset
;
8559 cur_offset
+= ins
.offset
;
8560 *alloc_hint
= ins
.objectid
+ ins
.offset
;
8562 inode_inc_iversion(inode
);
8563 inode
->i_ctime
= CURRENT_TIME
;
8564 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
8565 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
8566 (actual_len
> inode
->i_size
) &&
8567 (cur_offset
> inode
->i_size
)) {
8568 if (cur_offset
> actual_len
)
8569 i_size
= actual_len
;
8571 i_size
= cur_offset
;
8572 i_size_write(inode
, i_size
);
8573 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
8576 ret
= btrfs_update_inode(trans
, root
, inode
);
8579 btrfs_abort_transaction(trans
, root
, ret
);
8581 btrfs_end_transaction(trans
, root
);
8586 btrfs_end_transaction(trans
, root
);
8591 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
8592 u64 start
, u64 num_bytes
, u64 min_size
,
8593 loff_t actual_len
, u64
*alloc_hint
)
8595 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
8596 min_size
, actual_len
, alloc_hint
,
8600 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
8601 struct btrfs_trans_handle
*trans
, int mode
,
8602 u64 start
, u64 num_bytes
, u64 min_size
,
8603 loff_t actual_len
, u64
*alloc_hint
)
8605 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
8606 min_size
, actual_len
, alloc_hint
, trans
);
8609 static int btrfs_set_page_dirty(struct page
*page
)
8611 return __set_page_dirty_nobuffers(page
);
8614 static int btrfs_permission(struct inode
*inode
, int mask
)
8616 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8617 umode_t mode
= inode
->i_mode
;
8619 if (mask
& MAY_WRITE
&&
8620 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
8621 if (btrfs_root_readonly(root
))
8623 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
8626 return generic_permission(inode
, mask
);
8629 static const struct inode_operations btrfs_dir_inode_operations
= {
8630 .getattr
= btrfs_getattr
,
8631 .lookup
= btrfs_lookup
,
8632 .create
= btrfs_create
,
8633 .unlink
= btrfs_unlink
,
8635 .mkdir
= btrfs_mkdir
,
8636 .rmdir
= btrfs_rmdir
,
8637 .rename
= btrfs_rename
,
8638 .symlink
= btrfs_symlink
,
8639 .setattr
= btrfs_setattr
,
8640 .mknod
= btrfs_mknod
,
8641 .setxattr
= btrfs_setxattr
,
8642 .getxattr
= btrfs_getxattr
,
8643 .listxattr
= btrfs_listxattr
,
8644 .removexattr
= btrfs_removexattr
,
8645 .permission
= btrfs_permission
,
8646 .get_acl
= btrfs_get_acl
,
8648 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
8649 .lookup
= btrfs_lookup
,
8650 .permission
= btrfs_permission
,
8651 .get_acl
= btrfs_get_acl
,
8654 static const struct file_operations btrfs_dir_file_operations
= {
8655 .llseek
= generic_file_llseek
,
8656 .read
= generic_read_dir
,
8657 .iterate
= btrfs_real_readdir
,
8658 .unlocked_ioctl
= btrfs_ioctl
,
8659 #ifdef CONFIG_COMPAT
8660 .compat_ioctl
= btrfs_ioctl
,
8662 .release
= btrfs_release_file
,
8663 .fsync
= btrfs_sync_file
,
8666 static struct extent_io_ops btrfs_extent_io_ops
= {
8667 .fill_delalloc
= run_delalloc_range
,
8668 .submit_bio_hook
= btrfs_submit_bio_hook
,
8669 .merge_bio_hook
= btrfs_merge_bio_hook
,
8670 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
8671 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
8672 .writepage_start_hook
= btrfs_writepage_start_hook
,
8673 .set_bit_hook
= btrfs_set_bit_hook
,
8674 .clear_bit_hook
= btrfs_clear_bit_hook
,
8675 .merge_extent_hook
= btrfs_merge_extent_hook
,
8676 .split_extent_hook
= btrfs_split_extent_hook
,
8680 * btrfs doesn't support the bmap operation because swapfiles
8681 * use bmap to make a mapping of extents in the file. They assume
8682 * these extents won't change over the life of the file and they
8683 * use the bmap result to do IO directly to the drive.
8685 * the btrfs bmap call would return logical addresses that aren't
8686 * suitable for IO and they also will change frequently as COW
8687 * operations happen. So, swapfile + btrfs == corruption.
8689 * For now we're avoiding this by dropping bmap.
8691 static const struct address_space_operations btrfs_aops
= {
8692 .readpage
= btrfs_readpage
,
8693 .writepage
= btrfs_writepage
,
8694 .writepages
= btrfs_writepages
,
8695 .readpages
= btrfs_readpages
,
8696 .direct_IO
= btrfs_direct_IO
,
8697 .invalidatepage
= btrfs_invalidatepage
,
8698 .releasepage
= btrfs_releasepage
,
8699 .set_page_dirty
= btrfs_set_page_dirty
,
8700 .error_remove_page
= generic_error_remove_page
,
8703 static const struct address_space_operations btrfs_symlink_aops
= {
8704 .readpage
= btrfs_readpage
,
8705 .writepage
= btrfs_writepage
,
8706 .invalidatepage
= btrfs_invalidatepage
,
8707 .releasepage
= btrfs_releasepage
,
8710 static const struct inode_operations btrfs_file_inode_operations
= {
8711 .getattr
= btrfs_getattr
,
8712 .setattr
= btrfs_setattr
,
8713 .setxattr
= btrfs_setxattr
,
8714 .getxattr
= btrfs_getxattr
,
8715 .listxattr
= btrfs_listxattr
,
8716 .removexattr
= btrfs_removexattr
,
8717 .permission
= btrfs_permission
,
8718 .fiemap
= btrfs_fiemap
,
8719 .get_acl
= btrfs_get_acl
,
8720 .update_time
= btrfs_update_time
,
8722 static const struct inode_operations btrfs_special_inode_operations
= {
8723 .getattr
= btrfs_getattr
,
8724 .setattr
= btrfs_setattr
,
8725 .permission
= btrfs_permission
,
8726 .setxattr
= btrfs_setxattr
,
8727 .getxattr
= btrfs_getxattr
,
8728 .listxattr
= btrfs_listxattr
,
8729 .removexattr
= btrfs_removexattr
,
8730 .get_acl
= btrfs_get_acl
,
8731 .update_time
= btrfs_update_time
,
8733 static const struct inode_operations btrfs_symlink_inode_operations
= {
8734 .readlink
= generic_readlink
,
8735 .follow_link
= page_follow_link_light
,
8736 .put_link
= page_put_link
,
8737 .getattr
= btrfs_getattr
,
8738 .setattr
= btrfs_setattr
,
8739 .permission
= btrfs_permission
,
8740 .setxattr
= btrfs_setxattr
,
8741 .getxattr
= btrfs_getxattr
,
8742 .listxattr
= btrfs_listxattr
,
8743 .removexattr
= btrfs_removexattr
,
8744 .get_acl
= btrfs_get_acl
,
8745 .update_time
= btrfs_update_time
,
8748 const struct dentry_operations btrfs_dentry_operations
= {
8749 .d_delete
= btrfs_dentry_delete
,
8750 .d_release
= btrfs_dentry_release
,