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_biovec(int type
, struct page
**pages_in
,
85 u64 disk_start
, struct bio_vec
*bvec
,
86 int vcnt
, size_t srclen
);
88 static inline int compressed_bio_size(struct btrfs_root
*root
,
89 unsigned long disk_size
)
91 u16 csum_size
= btrfs_super_csum_size(root
->fs_info
->super_copy
);
93 return sizeof(struct compressed_bio
) +
94 (DIV_ROUND_UP(disk_size
, root
->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
, (char *)&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_biovec(cb
->compress_type
,
179 cb
->compressed_pages
,
181 cb
->orig_bio
->bi_io_vec
,
182 cb
->orig_bio
->bi_vcnt
,
188 /* release the compressed pages */
190 for (index
= 0; index
< cb
->nr_pages
; index
++) {
191 page
= cb
->compressed_pages
[index
];
192 page
->mapping
= NULL
;
196 /* do io completion on the original bio */
198 bio_io_error(cb
->orig_bio
);
201 struct bio_vec
*bvec
;
204 * we have verified the checksum already, set page
205 * checked so the end_io handlers know about it
207 bio_for_each_segment_all(bvec
, cb
->orig_bio
, i
)
208 SetPageChecked(bvec
->bv_page
);
210 bio_endio(cb
->orig_bio
);
213 /* finally free the cb struct */
214 kfree(cb
->compressed_pages
);
221 * Clear the writeback bits on all of the file
222 * pages for a compressed write
224 static noinline
void end_compressed_writeback(struct inode
*inode
,
225 const struct compressed_bio
*cb
)
227 unsigned long index
= cb
->start
>> PAGE_SHIFT
;
228 unsigned long end_index
= (cb
->start
+ cb
->len
- 1) >> PAGE_SHIFT
;
229 struct page
*pages
[16];
230 unsigned long nr_pages
= end_index
- index
+ 1;
235 mapping_set_error(inode
->i_mapping
, -EIO
);
237 while (nr_pages
> 0) {
238 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
240 nr_pages
, ARRAY_SIZE(pages
)), pages
);
246 for (i
= 0; i
< ret
; i
++) {
248 SetPageError(pages
[i
]);
249 end_page_writeback(pages
[i
]);
255 /* the inode may be gone now */
259 * do the cleanup once all the compressed pages hit the disk.
260 * This will clear writeback on the file pages and free the compressed
263 * This also calls the writeback end hooks for the file pages so that
264 * metadata and checksums can be updated in the file.
266 static void end_compressed_bio_write(struct bio
*bio
)
268 struct extent_io_tree
*tree
;
269 struct compressed_bio
*cb
= bio
->bi_private
;
277 /* if there are more bios still pending for this compressed
280 if (!atomic_dec_and_test(&cb
->pending_bios
))
283 /* ok, we're the last bio for this extent, step one is to
284 * call back into the FS and do all the end_io operations
287 tree
= &BTRFS_I(inode
)->io_tree
;
288 cb
->compressed_pages
[0]->mapping
= cb
->inode
->i_mapping
;
289 tree
->ops
->writepage_end_io_hook(cb
->compressed_pages
[0],
291 cb
->start
+ cb
->len
- 1,
293 bio
->bi_error
? 0 : 1);
294 cb
->compressed_pages
[0]->mapping
= NULL
;
296 end_compressed_writeback(inode
, cb
);
297 /* note, our inode could be gone now */
300 * release the compressed pages, these came from alloc_page and
301 * are not attached to the inode at all
304 for (index
= 0; index
< cb
->nr_pages
; index
++) {
305 page
= cb
->compressed_pages
[index
];
306 page
->mapping
= NULL
;
310 /* finally free the cb struct */
311 kfree(cb
->compressed_pages
);
318 * worker function to build and submit bios for previously compressed pages.
319 * The corresponding pages in the inode should be marked for writeback
320 * and the compressed pages should have a reference on them for dropping
321 * when the IO is complete.
323 * This also checksums the file bytes and gets things ready for
326 int btrfs_submit_compressed_write(struct inode
*inode
, u64 start
,
327 unsigned long len
, u64 disk_start
,
328 unsigned long compressed_len
,
329 struct page
**compressed_pages
,
330 unsigned long nr_pages
)
332 struct bio
*bio
= NULL
;
333 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
334 struct compressed_bio
*cb
;
335 unsigned long bytes_left
;
336 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
339 u64 first_byte
= disk_start
;
340 struct block_device
*bdev
;
342 int skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
344 WARN_ON(start
& ((u64
)PAGE_SIZE
- 1));
345 cb
= kmalloc(compressed_bio_size(root
, compressed_len
), GFP_NOFS
);
348 atomic_set(&cb
->pending_bios
, 0);
354 cb
->compressed_pages
= compressed_pages
;
355 cb
->compressed_len
= compressed_len
;
357 cb
->nr_pages
= nr_pages
;
359 bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
361 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
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(WRITE
, 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(root
->fs_info
, bio
,
395 BTRFS_WQ_ENDIO_DATA
);
396 BUG_ON(ret
); /* -ENOMEM */
399 ret
= btrfs_csum_one_bio(root
, inode
, bio
,
401 BUG_ON(ret
); /* -ENOMEM */
404 ret
= btrfs_map_bio(root
, WRITE
, bio
, 0, 1);
412 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
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
) {
419 btrfs_info(BTRFS_I(inode
)->root
->fs_info
,
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(root
->fs_info
, bio
, BTRFS_WQ_ENDIO_DATA
);
430 BUG_ON(ret
); /* -ENOMEM */
433 ret
= btrfs_csum_one_bio(root
, inode
, bio
, start
, 1);
434 BUG_ON(ret
); /* -ENOMEM */
437 ret
= btrfs_map_bio(root
, WRITE
, bio
, 0, 1);
447 static noinline
int add_ra_bio_pages(struct inode
*inode
,
449 struct compressed_bio
*cb
)
451 unsigned long end_index
;
452 unsigned long pg_index
;
454 u64 isize
= i_size_read(inode
);
457 unsigned long nr_pages
= 0;
458 struct extent_map
*em
;
459 struct address_space
*mapping
= inode
->i_mapping
;
460 struct extent_map_tree
*em_tree
;
461 struct extent_io_tree
*tree
;
465 page
= cb
->orig_bio
->bi_io_vec
[cb
->orig_bio
->bi_vcnt
- 1].bv_page
;
466 last_offset
= (page_offset(page
) + PAGE_SIZE
);
467 em_tree
= &BTRFS_I(inode
)->extent_tree
;
468 tree
= &BTRFS_I(inode
)->io_tree
;
473 end_index
= (i_size_read(inode
) - 1) >> PAGE_SHIFT
;
475 while (last_offset
< compressed_end
) {
476 pg_index
= last_offset
>> PAGE_SHIFT
;
478 if (pg_index
> end_index
)
482 page
= radix_tree_lookup(&mapping
->page_tree
, pg_index
);
484 if (page
&& !radix_tree_exceptional_entry(page
)) {
491 page
= __page_cache_alloc(mapping_gfp_constraint(mapping
,
496 if (add_to_page_cache_lru(page
, mapping
, pg_index
, GFP_NOFS
)) {
501 end
= last_offset
+ PAGE_SIZE
- 1;
503 * at this point, we have a locked page in the page cache
504 * for these bytes in the file. But, we have to make
505 * sure they map to this compressed extent on disk.
507 set_page_extent_mapped(page
);
508 lock_extent(tree
, last_offset
, end
);
509 read_lock(&em_tree
->lock
);
510 em
= lookup_extent_mapping(em_tree
, last_offset
,
512 read_unlock(&em_tree
->lock
);
514 if (!em
|| last_offset
< em
->start
||
515 (last_offset
+ PAGE_SIZE
> extent_map_end(em
)) ||
516 (em
->block_start
>> 9) != cb
->orig_bio
->bi_iter
.bi_sector
) {
518 unlock_extent(tree
, last_offset
, end
);
525 if (page
->index
== end_index
) {
527 size_t zero_offset
= isize
& (PAGE_SIZE
- 1);
531 zeros
= PAGE_SIZE
- zero_offset
;
532 userpage
= kmap_atomic(page
);
533 memset(userpage
+ zero_offset
, 0, zeros
);
534 flush_dcache_page(page
);
535 kunmap_atomic(userpage
);
539 ret
= bio_add_page(cb
->orig_bio
, page
,
542 if (ret
== PAGE_SIZE
) {
546 unlock_extent(tree
, last_offset
, end
);
552 last_offset
+= PAGE_SIZE
;
558 * for a compressed read, the bio we get passed has all the inode pages
559 * in it. We don't actually do IO on those pages but allocate new ones
560 * to hold the compressed pages on disk.
562 * bio->bi_iter.bi_sector points to the compressed extent on disk
563 * bio->bi_io_vec points to all of the inode pages
564 * bio->bi_vcnt is a count of pages
566 * After the compressed pages are read, we copy the bytes into the
567 * bio we were passed and then call the bio end_io calls
569 int btrfs_submit_compressed_read(struct inode
*inode
, struct bio
*bio
,
570 int mirror_num
, unsigned long bio_flags
)
572 struct extent_io_tree
*tree
;
573 struct extent_map_tree
*em_tree
;
574 struct compressed_bio
*cb
;
575 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
576 unsigned long uncompressed_len
= bio
->bi_vcnt
* PAGE_SIZE
;
577 unsigned long compressed_len
;
578 unsigned long nr_pages
;
579 unsigned long pg_index
;
581 struct block_device
*bdev
;
582 struct bio
*comp_bio
;
583 u64 cur_disk_byte
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
586 struct extent_map
*em
;
591 tree
= &BTRFS_I(inode
)->io_tree
;
592 em_tree
= &BTRFS_I(inode
)->extent_tree
;
594 /* we need the actual starting offset of this extent in the file */
595 read_lock(&em_tree
->lock
);
596 em
= lookup_extent_mapping(em_tree
,
597 page_offset(bio
->bi_io_vec
->bv_page
),
599 read_unlock(&em_tree
->lock
);
603 compressed_len
= em
->block_len
;
604 cb
= kmalloc(compressed_bio_size(root
, compressed_len
), GFP_NOFS
);
608 atomic_set(&cb
->pending_bios
, 0);
611 cb
->mirror_num
= mirror_num
;
614 cb
->start
= em
->orig_start
;
616 em_start
= em
->start
;
621 cb
->len
= uncompressed_len
;
622 cb
->compressed_len
= compressed_len
;
623 cb
->compress_type
= extent_compress_type(bio_flags
);
626 nr_pages
= DIV_ROUND_UP(compressed_len
, PAGE_SIZE
);
627 cb
->compressed_pages
= kcalloc(nr_pages
, sizeof(struct page
*),
629 if (!cb
->compressed_pages
)
632 bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
634 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
635 cb
->compressed_pages
[pg_index
] = alloc_page(GFP_NOFS
|
637 if (!cb
->compressed_pages
[pg_index
]) {
638 faili
= pg_index
- 1;
643 faili
= nr_pages
- 1;
644 cb
->nr_pages
= nr_pages
;
646 add_ra_bio_pages(inode
, em_start
+ em_len
, cb
);
648 /* include any pages we added in add_ra-bio_pages */
649 uncompressed_len
= bio
->bi_vcnt
* PAGE_SIZE
;
650 cb
->len
= uncompressed_len
;
652 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
, GFP_NOFS
);
655 comp_bio
->bi_private
= cb
;
656 comp_bio
->bi_end_io
= end_compressed_bio_read
;
657 atomic_inc(&cb
->pending_bios
);
659 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
660 page
= cb
->compressed_pages
[pg_index
];
661 page
->mapping
= inode
->i_mapping
;
662 page
->index
= em_start
>> PAGE_SHIFT
;
664 if (comp_bio
->bi_iter
.bi_size
)
665 ret
= tree
->ops
->merge_bio_hook(READ
, page
, 0,
671 page
->mapping
= NULL
;
672 if (ret
|| bio_add_page(comp_bio
, page
, PAGE_SIZE
, 0) <
676 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
,
677 BTRFS_WQ_ENDIO_DATA
);
678 BUG_ON(ret
); /* -ENOMEM */
681 * inc the count before we submit the bio so
682 * we know the end IO handler won't happen before
683 * we inc the count. Otherwise, the cb might get
684 * freed before we're done setting it up
686 atomic_inc(&cb
->pending_bios
);
688 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
689 ret
= btrfs_lookup_bio_sums(root
, inode
,
691 BUG_ON(ret
); /* -ENOMEM */
693 sums
+= DIV_ROUND_UP(comp_bio
->bi_iter
.bi_size
,
696 ret
= btrfs_map_bio(root
, READ
, comp_bio
,
705 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
,
708 comp_bio
->bi_private
= cb
;
709 comp_bio
->bi_end_io
= end_compressed_bio_read
;
711 bio_add_page(comp_bio
, page
, PAGE_SIZE
, 0);
713 cur_disk_byte
+= PAGE_SIZE
;
717 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
,
718 BTRFS_WQ_ENDIO_DATA
);
719 BUG_ON(ret
); /* -ENOMEM */
721 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
722 ret
= btrfs_lookup_bio_sums(root
, inode
, comp_bio
, sums
);
723 BUG_ON(ret
); /* -ENOMEM */
726 ret
= btrfs_map_bio(root
, READ
, comp_bio
, mirror_num
, 0);
737 __free_page(cb
->compressed_pages
[faili
]);
741 kfree(cb
->compressed_pages
);
750 struct list_head idle_ws
;
752 /* Number of free workspaces */
754 /* Total number of allocated workspaces */
756 /* Waiters for a free workspace */
757 wait_queue_head_t ws_wait
;
758 } btrfs_comp_ws
[BTRFS_COMPRESS_TYPES
];
760 static const struct btrfs_compress_op
* const btrfs_compress_op
[] = {
761 &btrfs_zlib_compress
,
765 void __init
btrfs_init_compress(void)
769 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
770 struct list_head
*workspace
;
772 INIT_LIST_HEAD(&btrfs_comp_ws
[i
].idle_ws
);
773 spin_lock_init(&btrfs_comp_ws
[i
].ws_lock
);
774 atomic_set(&btrfs_comp_ws
[i
].total_ws
, 0);
775 init_waitqueue_head(&btrfs_comp_ws
[i
].ws_wait
);
778 * Preallocate one workspace for each compression type so
779 * we can guarantee forward progress in the worst case
781 workspace
= btrfs_compress_op
[i
]->alloc_workspace();
782 if (IS_ERR(workspace
)) {
784 "BTRFS: cannot preallocate compression workspace, will try later");
786 atomic_set(&btrfs_comp_ws
[i
].total_ws
, 1);
787 btrfs_comp_ws
[i
].free_ws
= 1;
788 list_add(workspace
, &btrfs_comp_ws
[i
].idle_ws
);
794 * This finds an available workspace or allocates a new one.
795 * If it's not possible to allocate a new one, waits until there's one.
796 * Preallocation makes a forward progress guarantees and we do not return
799 static struct list_head
*find_workspace(int type
)
801 struct list_head
*workspace
;
802 int cpus
= num_online_cpus();
805 struct list_head
*idle_ws
= &btrfs_comp_ws
[idx
].idle_ws
;
806 spinlock_t
*ws_lock
= &btrfs_comp_ws
[idx
].ws_lock
;
807 atomic_t
*total_ws
= &btrfs_comp_ws
[idx
].total_ws
;
808 wait_queue_head_t
*ws_wait
= &btrfs_comp_ws
[idx
].ws_wait
;
809 int *free_ws
= &btrfs_comp_ws
[idx
].free_ws
;
812 if (!list_empty(idle_ws
)) {
813 workspace
= idle_ws
->next
;
816 spin_unlock(ws_lock
);
820 if (atomic_read(total_ws
) > cpus
) {
823 spin_unlock(ws_lock
);
824 prepare_to_wait(ws_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
825 if (atomic_read(total_ws
) > cpus
&& !*free_ws
)
827 finish_wait(ws_wait
, &wait
);
830 atomic_inc(total_ws
);
831 spin_unlock(ws_lock
);
833 workspace
= btrfs_compress_op
[idx
]->alloc_workspace();
834 if (IS_ERR(workspace
)) {
835 atomic_dec(total_ws
);
839 * Do not return the error but go back to waiting. There's a
840 * workspace preallocated for each type and the compression
841 * time is bounded so we get to a workspace eventually. This
842 * makes our caller's life easier.
844 * To prevent silent and low-probability deadlocks (when the
845 * initial preallocation fails), check if there are any
848 if (atomic_read(total_ws
) == 0) {
849 static DEFINE_RATELIMIT_STATE(_rs
,
850 /* once per minute */ 60 * HZ
,
853 if (__ratelimit(&_rs
)) {
855 "no compression workspaces, low memory, retrying");
864 * put a workspace struct back on the list or free it if we have enough
865 * idle ones sitting around
867 static void free_workspace(int type
, struct list_head
*workspace
)
870 struct list_head
*idle_ws
= &btrfs_comp_ws
[idx
].idle_ws
;
871 spinlock_t
*ws_lock
= &btrfs_comp_ws
[idx
].ws_lock
;
872 atomic_t
*total_ws
= &btrfs_comp_ws
[idx
].total_ws
;
873 wait_queue_head_t
*ws_wait
= &btrfs_comp_ws
[idx
].ws_wait
;
874 int *free_ws
= &btrfs_comp_ws
[idx
].free_ws
;
877 if (*free_ws
< num_online_cpus()) {
878 list_add(workspace
, idle_ws
);
880 spin_unlock(ws_lock
);
883 spin_unlock(ws_lock
);
885 btrfs_compress_op
[idx
]->free_workspace(workspace
);
886 atomic_dec(total_ws
);
889 * Make sure counter is updated before we wake up waiters.
892 if (waitqueue_active(ws_wait
))
897 * cleanup function for module exit
899 static void free_workspaces(void)
901 struct list_head
*workspace
;
904 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
905 while (!list_empty(&btrfs_comp_ws
[i
].idle_ws
)) {
906 workspace
= btrfs_comp_ws
[i
].idle_ws
.next
;
908 btrfs_compress_op
[i
]->free_workspace(workspace
);
909 atomic_dec(&btrfs_comp_ws
[i
].total_ws
);
915 * given an address space and start/len, compress the bytes.
917 * pages are allocated to hold the compressed result and stored
920 * out_pages is used to return the number of pages allocated. There
921 * may be pages allocated even if we return an error
923 * total_in is used to return the number of bytes actually read. It
924 * may be smaller then len if we had to exit early because we
925 * ran out of room in the pages array or because we cross the
928 * total_out is used to return the total number of compressed bytes
930 * max_out tells us the max number of bytes that we're allowed to
933 int btrfs_compress_pages(int type
, struct address_space
*mapping
,
934 u64 start
, unsigned long len
,
936 unsigned long nr_dest_pages
,
937 unsigned long *out_pages
,
938 unsigned long *total_in
,
939 unsigned long *total_out
,
940 unsigned long max_out
)
942 struct list_head
*workspace
;
945 workspace
= find_workspace(type
);
947 ret
= btrfs_compress_op
[type
-1]->compress_pages(workspace
, mapping
,
949 nr_dest_pages
, out_pages
,
952 free_workspace(type
, workspace
);
957 * pages_in is an array of pages with compressed data.
959 * disk_start is the starting logical offset of this array in the file
961 * bvec is a bio_vec of pages from the file that we want to decompress into
963 * vcnt is the count of pages in the biovec
965 * srclen is the number of bytes in pages_in
967 * The basic idea is that we have a bio that was created by readpages.
968 * The pages in the bio are for the uncompressed data, and they may not
969 * be contiguous. They all correspond to the range of bytes covered by
970 * the compressed extent.
972 static int btrfs_decompress_biovec(int type
, struct page
**pages_in
,
973 u64 disk_start
, struct bio_vec
*bvec
,
974 int vcnt
, size_t srclen
)
976 struct list_head
*workspace
;
979 workspace
= find_workspace(type
);
981 ret
= btrfs_compress_op
[type
-1]->decompress_biovec(workspace
, pages_in
,
984 free_workspace(type
, workspace
);
989 * a less complex decompression routine. Our compressed data fits in a
990 * single page, and we want to read a single page out of it.
991 * start_byte tells us the offset into the compressed data we're interested in
993 int btrfs_decompress(int type
, unsigned char *data_in
, struct page
*dest_page
,
994 unsigned long start_byte
, size_t srclen
, size_t destlen
)
996 struct list_head
*workspace
;
999 workspace
= find_workspace(type
);
1001 ret
= btrfs_compress_op
[type
-1]->decompress(workspace
, data_in
,
1002 dest_page
, start_byte
,
1005 free_workspace(type
, workspace
);
1009 void btrfs_exit_compress(void)
1015 * Copy uncompressed data from working buffer to pages.
1017 * buf_start is the byte offset we're of the start of our workspace buffer.
1019 * total_out is the last byte of the buffer
1021 int btrfs_decompress_buf2page(char *buf
, unsigned long buf_start
,
1022 unsigned long total_out
, u64 disk_start
,
1023 struct bio_vec
*bvec
, int vcnt
,
1024 unsigned long *pg_index
,
1025 unsigned long *pg_offset
)
1027 unsigned long buf_offset
;
1028 unsigned long current_buf_start
;
1029 unsigned long start_byte
;
1030 unsigned long working_bytes
= total_out
- buf_start
;
1031 unsigned long bytes
;
1033 struct page
*page_out
= bvec
[*pg_index
].bv_page
;
1036 * start byte is the first byte of the page we're currently
1037 * copying into relative to the start of the compressed data.
1039 start_byte
= page_offset(page_out
) - disk_start
;
1041 /* we haven't yet hit data corresponding to this page */
1042 if (total_out
<= start_byte
)
1046 * the start of the data we care about is offset into
1047 * the middle of our working buffer
1049 if (total_out
> start_byte
&& buf_start
< start_byte
) {
1050 buf_offset
= start_byte
- buf_start
;
1051 working_bytes
-= buf_offset
;
1055 current_buf_start
= buf_start
;
1057 /* copy bytes from the working buffer into the pages */
1058 while (working_bytes
> 0) {
1059 bytes
= min(PAGE_SIZE
- *pg_offset
,
1060 PAGE_SIZE
- buf_offset
);
1061 bytes
= min(bytes
, working_bytes
);
1062 kaddr
= kmap_atomic(page_out
);
1063 memcpy(kaddr
+ *pg_offset
, buf
+ buf_offset
, bytes
);
1064 kunmap_atomic(kaddr
);
1065 flush_dcache_page(page_out
);
1067 *pg_offset
+= bytes
;
1068 buf_offset
+= bytes
;
1069 working_bytes
-= bytes
;
1070 current_buf_start
+= bytes
;
1072 /* check if we need to pick another page */
1073 if (*pg_offset
== PAGE_SIZE
) {
1075 if (*pg_index
>= vcnt
)
1078 page_out
= bvec
[*pg_index
].bv_page
;
1080 start_byte
= page_offset(page_out
) - disk_start
;
1083 * make sure our new page is covered by this
1086 if (total_out
<= start_byte
)
1090 * the next page in the biovec might not be adjacent
1091 * to the last page, but it might still be found
1092 * inside this working buffer. bump our offset pointer
1094 if (total_out
> start_byte
&&
1095 current_buf_start
< start_byte
) {
1096 buf_offset
= start_byte
- buf_start
;
1097 working_bytes
= total_out
- start_byte
;
1098 current_buf_start
= buf_start
+ buf_offset
;
1107 * When uncompressing data, we need to make sure and zero any parts of
1108 * the biovec that were not filled in by the decompression code. pg_index
1109 * and pg_offset indicate the last page and the last offset of that page
1110 * that have been filled in. This will zero everything remaining in the
1113 void btrfs_clear_biovec_end(struct bio_vec
*bvec
, int vcnt
,
1114 unsigned long pg_index
,
1115 unsigned long pg_offset
)
1117 while (pg_index
< vcnt
) {
1118 struct page
*page
= bvec
[pg_index
].bv_page
;
1119 unsigned long off
= bvec
[pg_index
].bv_offset
;
1120 unsigned long len
= bvec
[pg_index
].bv_len
;
1122 if (pg_offset
< off
)
1124 if (pg_offset
< off
+ len
) {
1125 unsigned long bytes
= off
+ len
- pg_offset
;
1128 kaddr
= kmap_atomic(page
);
1129 memset(kaddr
+ pg_offset
, 0, bytes
);
1130 kunmap_atomic(kaddr
);