2 * Copyright (C) 2008 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/bit_spinlock.h>
34 #include <linux/slab.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "ordered-data.h"
41 #include "compression.h"
42 #include "extent_io.h"
43 #include "extent_map.h"
45 struct compressed_bio
{
46 /* number of bios pending for this compressed extent */
47 atomic_t pending_bios
;
49 /* the pages with the compressed data on them */
50 struct page
**compressed_pages
;
52 /* inode that owns this data */
55 /* starting offset in the inode for our pages */
58 /* number of bytes in the inode we're working on */
61 /* number of bytes on disk */
62 unsigned long compressed_len
;
64 /* the compression algorithm for this bio */
67 /* number of compressed pages in the array */
68 unsigned long nr_pages
;
74 /* for reads, this is the bio we are copying the data into */
78 * the start of a variable length array of checksums only
84 static int btrfs_decompress_bio(int type
, struct page
**pages_in
,
85 u64 disk_start
, struct bio
*orig_bio
,
88 static inline int compressed_bio_size(struct btrfs_fs_info
*fs_info
,
89 unsigned long disk_size
)
91 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
93 return sizeof(struct compressed_bio
) +
94 (DIV_ROUND_UP(disk_size
, fs_info
->sectorsize
)) * csum_size
;
97 static struct bio
*compressed_bio_alloc(struct block_device
*bdev
,
98 u64 first_byte
, gfp_t gfp_flags
)
100 return btrfs_bio_alloc(bdev
, first_byte
>> 9, BIO_MAX_PAGES
, gfp_flags
);
103 static int check_compressed_csum(struct inode
*inode
,
104 struct compressed_bio
*cb
,
112 u32
*cb_sum
= &cb
->sums
;
114 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
117 for (i
= 0; i
< cb
->nr_pages
; i
++) {
118 page
= cb
->compressed_pages
[i
];
121 kaddr
= kmap_atomic(page
);
122 csum
= btrfs_csum_data(kaddr
, csum
, PAGE_SIZE
);
123 btrfs_csum_final(csum
, (u8
*)&csum
);
124 kunmap_atomic(kaddr
);
126 if (csum
!= *cb_sum
) {
127 btrfs_info(BTRFS_I(inode
)->root
->fs_info
,
128 "csum failed ino %llu extent %llu csum %u wanted %u mirror %d",
129 btrfs_ino(inode
), disk_start
, csum
, *cb_sum
,
142 /* when we finish reading compressed pages from the disk, we
143 * decompress them and then run the bio end_io routines on the
144 * decompressed pages (in the inode address space).
146 * This allows the checksumming and other IO error handling routines
149 * The compressed pages are freed here, and it must be run
152 static void end_compressed_bio_read(struct bio
*bio
)
154 struct compressed_bio
*cb
= bio
->bi_private
;
163 /* if there are more bios still pending for this compressed
166 if (!atomic_dec_and_test(&cb
->pending_bios
))
170 ret
= check_compressed_csum(inode
, cb
,
171 (u64
)bio
->bi_iter
.bi_sector
<< 9);
175 /* ok, we're the last bio for this extent, lets start
178 ret
= btrfs_decompress_bio(cb
->compress_type
,
179 cb
->compressed_pages
,
187 /* release the compressed pages */
189 for (index
= 0; index
< cb
->nr_pages
; index
++) {
190 page
= cb
->compressed_pages
[index
];
191 page
->mapping
= NULL
;
195 /* do io completion on the original bio */
197 bio_io_error(cb
->orig_bio
);
200 struct bio_vec
*bvec
;
203 * we have verified the checksum already, set page
204 * checked so the end_io handlers know about it
206 bio_for_each_segment_all(bvec
, cb
->orig_bio
, i
)
207 SetPageChecked(bvec
->bv_page
);
209 bio_endio(cb
->orig_bio
);
212 /* finally free the cb struct */
213 kfree(cb
->compressed_pages
);
220 * Clear the writeback bits on all of the file
221 * pages for a compressed write
223 static noinline
void end_compressed_writeback(struct inode
*inode
,
224 const struct compressed_bio
*cb
)
226 unsigned long index
= cb
->start
>> PAGE_SHIFT
;
227 unsigned long end_index
= (cb
->start
+ cb
->len
- 1) >> PAGE_SHIFT
;
228 struct page
*pages
[16];
229 unsigned long nr_pages
= end_index
- index
+ 1;
234 mapping_set_error(inode
->i_mapping
, -EIO
);
236 while (nr_pages
> 0) {
237 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
239 nr_pages
, ARRAY_SIZE(pages
)), pages
);
245 for (i
= 0; i
< ret
; i
++) {
247 SetPageError(pages
[i
]);
248 end_page_writeback(pages
[i
]);
254 /* the inode may be gone now */
258 * do the cleanup once all the compressed pages hit the disk.
259 * This will clear writeback on the file pages and free the compressed
262 * This also calls the writeback end hooks for the file pages so that
263 * metadata and checksums can be updated in the file.
265 static void end_compressed_bio_write(struct bio
*bio
)
267 struct extent_io_tree
*tree
;
268 struct compressed_bio
*cb
= bio
->bi_private
;
276 /* if there are more bios still pending for this compressed
279 if (!atomic_dec_and_test(&cb
->pending_bios
))
282 /* ok, we're the last bio for this extent, step one is to
283 * call back into the FS and do all the end_io operations
286 tree
= &BTRFS_I(inode
)->io_tree
;
287 cb
->compressed_pages
[0]->mapping
= cb
->inode
->i_mapping
;
288 tree
->ops
->writepage_end_io_hook(cb
->compressed_pages
[0],
290 cb
->start
+ cb
->len
- 1,
292 bio
->bi_error
? 0 : 1);
293 cb
->compressed_pages
[0]->mapping
= NULL
;
295 end_compressed_writeback(inode
, cb
);
296 /* note, our inode could be gone now */
299 * release the compressed pages, these came from alloc_page and
300 * are not attached to the inode at all
303 for (index
= 0; index
< cb
->nr_pages
; index
++) {
304 page
= cb
->compressed_pages
[index
];
305 page
->mapping
= NULL
;
309 /* finally free the cb struct */
310 kfree(cb
->compressed_pages
);
317 * worker function to build and submit bios for previously compressed pages.
318 * The corresponding pages in the inode should be marked for writeback
319 * and the compressed pages should have a reference on them for dropping
320 * when the IO is complete.
322 * This also checksums the file bytes and gets things ready for
325 int btrfs_submit_compressed_write(struct inode
*inode
, u64 start
,
326 unsigned long len
, u64 disk_start
,
327 unsigned long compressed_len
,
328 struct page
**compressed_pages
,
329 unsigned long nr_pages
)
331 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
332 struct bio
*bio
= NULL
;
333 struct compressed_bio
*cb
;
334 unsigned long bytes_left
;
335 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
338 u64 first_byte
= disk_start
;
339 struct block_device
*bdev
;
341 int skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
343 WARN_ON(start
& ((u64
)PAGE_SIZE
- 1));
344 cb
= kmalloc(compressed_bio_size(fs_info
, compressed_len
), GFP_NOFS
);
347 atomic_set(&cb
->pending_bios
, 0);
353 cb
->compressed_pages
= compressed_pages
;
354 cb
->compressed_len
= compressed_len
;
356 cb
->nr_pages
= nr_pages
;
358 bdev
= fs_info
->fs_devices
->latest_bdev
;
360 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
365 bio_set_op_attrs(bio
, REQ_OP_WRITE
, 0);
366 bio
->bi_private
= cb
;
367 bio
->bi_end_io
= end_compressed_bio_write
;
368 atomic_inc(&cb
->pending_bios
);
370 /* create and submit bios for the compressed pages */
371 bytes_left
= compressed_len
;
372 for (pg_index
= 0; pg_index
< cb
->nr_pages
; pg_index
++) {
373 page
= compressed_pages
[pg_index
];
374 page
->mapping
= inode
->i_mapping
;
375 if (bio
->bi_iter
.bi_size
)
376 ret
= io_tree
->ops
->merge_bio_hook(page
, 0,
382 page
->mapping
= NULL
;
383 if (ret
|| bio_add_page(bio
, page
, PAGE_SIZE
, 0) <
388 * inc the count before we submit the bio so
389 * we know the end IO handler won't happen before
390 * we inc the count. Otherwise, the cb might get
391 * freed before we're done setting it up
393 atomic_inc(&cb
->pending_bios
);
394 ret
= btrfs_bio_wq_end_io(fs_info
, bio
,
395 BTRFS_WQ_ENDIO_DATA
);
396 BUG_ON(ret
); /* -ENOMEM */
399 ret
= btrfs_csum_one_bio(inode
, bio
, start
, 1);
400 BUG_ON(ret
); /* -ENOMEM */
403 ret
= btrfs_map_bio(fs_info
, bio
, 0, 1);
411 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
413 bio_set_op_attrs(bio
, REQ_OP_WRITE
, 0);
414 bio
->bi_private
= cb
;
415 bio
->bi_end_io
= end_compressed_bio_write
;
416 bio_add_page(bio
, page
, PAGE_SIZE
, 0);
418 if (bytes_left
< PAGE_SIZE
) {
420 "bytes left %lu compress len %lu nr %lu",
421 bytes_left
, cb
->compressed_len
, cb
->nr_pages
);
423 bytes_left
-= PAGE_SIZE
;
424 first_byte
+= PAGE_SIZE
;
429 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, BTRFS_WQ_ENDIO_DATA
);
430 BUG_ON(ret
); /* -ENOMEM */
433 ret
= btrfs_csum_one_bio(inode
, bio
, start
, 1);
434 BUG_ON(ret
); /* -ENOMEM */
437 ret
= btrfs_map_bio(fs_info
, bio
, 0, 1);
447 static u64
bio_end_offset(struct bio
*bio
)
449 struct bio_vec
*last
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
451 return page_offset(last
->bv_page
) + last
->bv_len
+ last
->bv_offset
;
454 static noinline
int add_ra_bio_pages(struct inode
*inode
,
456 struct compressed_bio
*cb
)
458 unsigned long end_index
;
459 unsigned long pg_index
;
461 u64 isize
= i_size_read(inode
);
464 unsigned long nr_pages
= 0;
465 struct extent_map
*em
;
466 struct address_space
*mapping
= inode
->i_mapping
;
467 struct extent_map_tree
*em_tree
;
468 struct extent_io_tree
*tree
;
472 last_offset
= bio_end_offset(cb
->orig_bio
);
473 em_tree
= &BTRFS_I(inode
)->extent_tree
;
474 tree
= &BTRFS_I(inode
)->io_tree
;
479 end_index
= (i_size_read(inode
) - 1) >> PAGE_SHIFT
;
481 while (last_offset
< compressed_end
) {
482 pg_index
= last_offset
>> PAGE_SHIFT
;
484 if (pg_index
> end_index
)
488 page
= radix_tree_lookup(&mapping
->page_tree
, pg_index
);
490 if (page
&& !radix_tree_exceptional_entry(page
)) {
497 page
= __page_cache_alloc(mapping_gfp_constraint(mapping
,
502 if (add_to_page_cache_lru(page
, mapping
, pg_index
, GFP_NOFS
)) {
507 end
= last_offset
+ PAGE_SIZE
- 1;
509 * at this point, we have a locked page in the page cache
510 * for these bytes in the file. But, we have to make
511 * sure they map to this compressed extent on disk.
513 set_page_extent_mapped(page
);
514 lock_extent(tree
, last_offset
, end
);
515 read_lock(&em_tree
->lock
);
516 em
= lookup_extent_mapping(em_tree
, last_offset
,
518 read_unlock(&em_tree
->lock
);
520 if (!em
|| last_offset
< em
->start
||
521 (last_offset
+ PAGE_SIZE
> extent_map_end(em
)) ||
522 (em
->block_start
>> 9) != cb
->orig_bio
->bi_iter
.bi_sector
) {
524 unlock_extent(tree
, last_offset
, end
);
531 if (page
->index
== end_index
) {
533 size_t zero_offset
= isize
& (PAGE_SIZE
- 1);
537 zeros
= PAGE_SIZE
- zero_offset
;
538 userpage
= kmap_atomic(page
);
539 memset(userpage
+ zero_offset
, 0, zeros
);
540 flush_dcache_page(page
);
541 kunmap_atomic(userpage
);
545 ret
= bio_add_page(cb
->orig_bio
, page
,
548 if (ret
== PAGE_SIZE
) {
552 unlock_extent(tree
, last_offset
, end
);
558 last_offset
+= PAGE_SIZE
;
564 * for a compressed read, the bio we get passed has all the inode pages
565 * in it. We don't actually do IO on those pages but allocate new ones
566 * to hold the compressed pages on disk.
568 * bio->bi_iter.bi_sector points to the compressed extent on disk
569 * bio->bi_io_vec points to all of the inode pages
571 * After the compressed pages are read, we copy the bytes into the
572 * bio we were passed and then call the bio end_io calls
574 int btrfs_submit_compressed_read(struct inode
*inode
, struct bio
*bio
,
575 int mirror_num
, unsigned long bio_flags
)
577 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
578 struct extent_io_tree
*tree
;
579 struct extent_map_tree
*em_tree
;
580 struct compressed_bio
*cb
;
581 unsigned long compressed_len
;
582 unsigned long nr_pages
;
583 unsigned long pg_index
;
585 struct block_device
*bdev
;
586 struct bio
*comp_bio
;
587 u64 cur_disk_byte
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
590 struct extent_map
*em
;
595 tree
= &BTRFS_I(inode
)->io_tree
;
596 em_tree
= &BTRFS_I(inode
)->extent_tree
;
598 /* we need the actual starting offset of this extent in the file */
599 read_lock(&em_tree
->lock
);
600 em
= lookup_extent_mapping(em_tree
,
601 page_offset(bio
->bi_io_vec
->bv_page
),
603 read_unlock(&em_tree
->lock
);
607 compressed_len
= em
->block_len
;
608 cb
= kmalloc(compressed_bio_size(fs_info
, compressed_len
), GFP_NOFS
);
612 atomic_set(&cb
->pending_bios
, 0);
615 cb
->mirror_num
= mirror_num
;
618 cb
->start
= em
->orig_start
;
620 em_start
= em
->start
;
625 cb
->len
= bio
->bi_iter
.bi_size
;
626 cb
->compressed_len
= compressed_len
;
627 cb
->compress_type
= extent_compress_type(bio_flags
);
630 nr_pages
= DIV_ROUND_UP(compressed_len
, PAGE_SIZE
);
631 cb
->compressed_pages
= kcalloc(nr_pages
, sizeof(struct page
*),
633 if (!cb
->compressed_pages
)
636 bdev
= fs_info
->fs_devices
->latest_bdev
;
638 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
639 cb
->compressed_pages
[pg_index
] = alloc_page(GFP_NOFS
|
641 if (!cb
->compressed_pages
[pg_index
]) {
642 faili
= pg_index
- 1;
647 faili
= nr_pages
- 1;
648 cb
->nr_pages
= nr_pages
;
650 add_ra_bio_pages(inode
, em_start
+ em_len
, cb
);
652 /* include any pages we added in add_ra-bio_pages */
653 cb
->len
= bio
->bi_iter
.bi_size
;
655 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
, GFP_NOFS
);
658 bio_set_op_attrs (comp_bio
, REQ_OP_READ
, 0);
659 comp_bio
->bi_private
= cb
;
660 comp_bio
->bi_end_io
= end_compressed_bio_read
;
661 atomic_inc(&cb
->pending_bios
);
663 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
664 page
= cb
->compressed_pages
[pg_index
];
665 page
->mapping
= inode
->i_mapping
;
666 page
->index
= em_start
>> PAGE_SHIFT
;
668 if (comp_bio
->bi_iter
.bi_size
)
669 ret
= tree
->ops
->merge_bio_hook(page
, 0,
675 page
->mapping
= NULL
;
676 if (ret
|| bio_add_page(comp_bio
, page
, PAGE_SIZE
, 0) <
680 ret
= btrfs_bio_wq_end_io(fs_info
, comp_bio
,
681 BTRFS_WQ_ENDIO_DATA
);
682 BUG_ON(ret
); /* -ENOMEM */
685 * inc the count before we submit the bio so
686 * we know the end IO handler won't happen before
687 * we inc the count. Otherwise, the cb might get
688 * freed before we're done setting it up
690 atomic_inc(&cb
->pending_bios
);
692 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
693 ret
= btrfs_lookup_bio_sums(inode
, comp_bio
,
695 BUG_ON(ret
); /* -ENOMEM */
697 sums
+= DIV_ROUND_UP(comp_bio
->bi_iter
.bi_size
,
698 fs_info
->sectorsize
);
700 ret
= btrfs_map_bio(fs_info
, comp_bio
, mirror_num
, 0);
702 comp_bio
->bi_error
= ret
;
708 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
,
711 bio_set_op_attrs(comp_bio
, REQ_OP_READ
, 0);
712 comp_bio
->bi_private
= cb
;
713 comp_bio
->bi_end_io
= end_compressed_bio_read
;
715 bio_add_page(comp_bio
, page
, PAGE_SIZE
, 0);
717 cur_disk_byte
+= PAGE_SIZE
;
721 ret
= btrfs_bio_wq_end_io(fs_info
, comp_bio
, BTRFS_WQ_ENDIO_DATA
);
722 BUG_ON(ret
); /* -ENOMEM */
724 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
725 ret
= btrfs_lookup_bio_sums(inode
, comp_bio
, sums
);
726 BUG_ON(ret
); /* -ENOMEM */
729 ret
= btrfs_map_bio(fs_info
, comp_bio
, mirror_num
, 0);
731 comp_bio
->bi_error
= ret
;
740 __free_page(cb
->compressed_pages
[faili
]);
744 kfree(cb
->compressed_pages
);
753 struct list_head idle_ws
;
755 /* Number of free workspaces */
757 /* Total number of allocated workspaces */
759 /* Waiters for a free workspace */
760 wait_queue_head_t ws_wait
;
761 } btrfs_comp_ws
[BTRFS_COMPRESS_TYPES
];
763 static const struct btrfs_compress_op
* const btrfs_compress_op
[] = {
764 &btrfs_zlib_compress
,
768 void __init
btrfs_init_compress(void)
772 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
773 struct list_head
*workspace
;
775 INIT_LIST_HEAD(&btrfs_comp_ws
[i
].idle_ws
);
776 spin_lock_init(&btrfs_comp_ws
[i
].ws_lock
);
777 atomic_set(&btrfs_comp_ws
[i
].total_ws
, 0);
778 init_waitqueue_head(&btrfs_comp_ws
[i
].ws_wait
);
781 * Preallocate one workspace for each compression type so
782 * we can guarantee forward progress in the worst case
784 workspace
= btrfs_compress_op
[i
]->alloc_workspace();
785 if (IS_ERR(workspace
)) {
786 pr_warn("BTRFS: cannot preallocate compression workspace, will try later\n");
788 atomic_set(&btrfs_comp_ws
[i
].total_ws
, 1);
789 btrfs_comp_ws
[i
].free_ws
= 1;
790 list_add(workspace
, &btrfs_comp_ws
[i
].idle_ws
);
796 * This finds an available workspace or allocates a new one.
797 * If it's not possible to allocate a new one, waits until there's one.
798 * Preallocation makes a forward progress guarantees and we do not return
801 static struct list_head
*find_workspace(int type
)
803 struct list_head
*workspace
;
804 int cpus
= num_online_cpus();
807 struct list_head
*idle_ws
= &btrfs_comp_ws
[idx
].idle_ws
;
808 spinlock_t
*ws_lock
= &btrfs_comp_ws
[idx
].ws_lock
;
809 atomic_t
*total_ws
= &btrfs_comp_ws
[idx
].total_ws
;
810 wait_queue_head_t
*ws_wait
= &btrfs_comp_ws
[idx
].ws_wait
;
811 int *free_ws
= &btrfs_comp_ws
[idx
].free_ws
;
814 if (!list_empty(idle_ws
)) {
815 workspace
= idle_ws
->next
;
818 spin_unlock(ws_lock
);
822 if (atomic_read(total_ws
) > cpus
) {
825 spin_unlock(ws_lock
);
826 prepare_to_wait(ws_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
827 if (atomic_read(total_ws
) > cpus
&& !*free_ws
)
829 finish_wait(ws_wait
, &wait
);
832 atomic_inc(total_ws
);
833 spin_unlock(ws_lock
);
835 workspace
= btrfs_compress_op
[idx
]->alloc_workspace();
836 if (IS_ERR(workspace
)) {
837 atomic_dec(total_ws
);
841 * Do not return the error but go back to waiting. There's a
842 * workspace preallocated for each type and the compression
843 * time is bounded so we get to a workspace eventually. This
844 * makes our caller's life easier.
846 * To prevent silent and low-probability deadlocks (when the
847 * initial preallocation fails), check if there are any
850 if (atomic_read(total_ws
) == 0) {
851 static DEFINE_RATELIMIT_STATE(_rs
,
852 /* once per minute */ 60 * HZ
,
855 if (__ratelimit(&_rs
)) {
856 pr_warn("BTRFS: no compression workspaces, low memory, retrying\n");
865 * put a workspace struct back on the list or free it if we have enough
866 * idle ones sitting around
868 static void free_workspace(int type
, struct list_head
*workspace
)
871 struct list_head
*idle_ws
= &btrfs_comp_ws
[idx
].idle_ws
;
872 spinlock_t
*ws_lock
= &btrfs_comp_ws
[idx
].ws_lock
;
873 atomic_t
*total_ws
= &btrfs_comp_ws
[idx
].total_ws
;
874 wait_queue_head_t
*ws_wait
= &btrfs_comp_ws
[idx
].ws_wait
;
875 int *free_ws
= &btrfs_comp_ws
[idx
].free_ws
;
878 if (*free_ws
< num_online_cpus()) {
879 list_add(workspace
, idle_ws
);
881 spin_unlock(ws_lock
);
884 spin_unlock(ws_lock
);
886 btrfs_compress_op
[idx
]->free_workspace(workspace
);
887 atomic_dec(total_ws
);
890 * Make sure counter is updated before we wake up waiters.
893 if (waitqueue_active(ws_wait
))
898 * cleanup function for module exit
900 static void free_workspaces(void)
902 struct list_head
*workspace
;
905 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
906 while (!list_empty(&btrfs_comp_ws
[i
].idle_ws
)) {
907 workspace
= btrfs_comp_ws
[i
].idle_ws
.next
;
909 btrfs_compress_op
[i
]->free_workspace(workspace
);
910 atomic_dec(&btrfs_comp_ws
[i
].total_ws
);
916 * given an address space and start/len, compress the bytes.
918 * pages are allocated to hold the compressed result and stored
921 * out_pages is used to return the number of pages allocated. There
922 * may be pages allocated even if we return an error
924 * total_in is used to return the number of bytes actually read. It
925 * may be smaller then len if we had to exit early because we
926 * ran out of room in the pages array or because we cross the
929 * total_out is used to return the total number of compressed bytes
931 * max_out tells us the max number of bytes that we're allowed to
934 int btrfs_compress_pages(int type
, struct address_space
*mapping
,
935 u64 start
, unsigned long len
,
937 unsigned long nr_dest_pages
,
938 unsigned long *out_pages
,
939 unsigned long *total_in
,
940 unsigned long *total_out
,
941 unsigned long max_out
)
943 struct list_head
*workspace
;
946 workspace
= find_workspace(type
);
948 ret
= btrfs_compress_op
[type
-1]->compress_pages(workspace
, mapping
,
950 nr_dest_pages
, out_pages
,
953 free_workspace(type
, workspace
);
958 * pages_in is an array of pages with compressed data.
960 * disk_start is the starting logical offset of this array in the file
962 * orig_bio contains the pages from the file that we want to decompress into
964 * srclen is the number of bytes in pages_in
966 * The basic idea is that we have a bio that was created by readpages.
967 * The pages in the bio are for the uncompressed data, and they may not
968 * be contiguous. They all correspond to the range of bytes covered by
969 * the compressed extent.
971 static int btrfs_decompress_bio(int type
, struct page
**pages_in
,
972 u64 disk_start
, struct bio
*orig_bio
,
975 struct list_head
*workspace
;
978 workspace
= find_workspace(type
);
980 ret
= btrfs_compress_op
[type
-1]->decompress_bio(workspace
, pages_in
,
981 disk_start
, orig_bio
,
983 free_workspace(type
, workspace
);
988 * a less complex decompression routine. Our compressed data fits in a
989 * single page, and we want to read a single page out of it.
990 * start_byte tells us the offset into the compressed data we're interested in
992 int btrfs_decompress(int type
, unsigned char *data_in
, struct page
*dest_page
,
993 unsigned long start_byte
, size_t srclen
, size_t destlen
)
995 struct list_head
*workspace
;
998 workspace
= find_workspace(type
);
1000 ret
= btrfs_compress_op
[type
-1]->decompress(workspace
, data_in
,
1001 dest_page
, start_byte
,
1004 free_workspace(type
, workspace
);
1008 void btrfs_exit_compress(void)
1014 * Copy uncompressed data from working buffer to pages.
1016 * buf_start is the byte offset we're of the start of our workspace buffer.
1018 * total_out is the last byte of the buffer
1020 int btrfs_decompress_buf2page(char *buf
, unsigned long buf_start
,
1021 unsigned long total_out
, u64 disk_start
,
1024 unsigned long buf_offset
;
1025 unsigned long current_buf_start
;
1026 unsigned long start_byte
;
1027 unsigned long working_bytes
= total_out
- buf_start
;
1028 unsigned long bytes
;
1030 struct bio_vec bvec
= bio_iter_iovec(bio
, bio
->bi_iter
);
1033 * start byte is the first byte of the page we're currently
1034 * copying into relative to the start of the compressed data.
1036 start_byte
= page_offset(bvec
.bv_page
) - disk_start
;
1038 /* we haven't yet hit data corresponding to this page */
1039 if (total_out
<= start_byte
)
1043 * the start of the data we care about is offset into
1044 * the middle of our working buffer
1046 if (total_out
> start_byte
&& buf_start
< start_byte
) {
1047 buf_offset
= start_byte
- buf_start
;
1048 working_bytes
-= buf_offset
;
1052 current_buf_start
= buf_start
;
1054 /* copy bytes from the working buffer into the pages */
1055 while (working_bytes
> 0) {
1056 bytes
= min_t(unsigned long, bvec
.bv_len
,
1057 PAGE_SIZE
- buf_offset
);
1058 bytes
= min(bytes
, working_bytes
);
1060 kaddr
= kmap_atomic(bvec
.bv_page
);
1061 memcpy(kaddr
+ bvec
.bv_offset
, buf
+ buf_offset
, bytes
);
1062 kunmap_atomic(kaddr
);
1063 flush_dcache_page(bvec
.bv_page
);
1065 buf_offset
+= bytes
;
1066 working_bytes
-= bytes
;
1067 current_buf_start
+= bytes
;
1069 /* check if we need to pick another page */
1070 bio_advance(bio
, bytes
);
1071 if (!bio
->bi_iter
.bi_size
)
1073 bvec
= bio_iter_iovec(bio
, bio
->bi_iter
);
1075 start_byte
= page_offset(bvec
.bv_page
) - disk_start
;
1078 * make sure our new page is covered by this
1081 if (total_out
<= start_byte
)
1085 * the next page in the biovec might not be adjacent
1086 * to the last page, but it might still be found
1087 * inside this working buffer. bump our offset pointer
1089 if (total_out
> start_byte
&&
1090 current_buf_start
< start_byte
) {
1091 buf_offset
= start_byte
- buf_start
;
1092 working_bytes
= total_out
- start_byte
;
1093 current_buf_start
= buf_start
+ buf_offset
;