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>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "ordered-data.h"
42 #include "compression.h"
43 #include "extent_io.h"
44 #include "extent_map.h"
46 struct compressed_bio
{
47 /* number of bios pending for this compressed extent */
48 atomic_t pending_bios
;
50 /* the pages with the compressed data on them */
51 struct page
**compressed_pages
;
53 /* inode that owns this data */
56 /* starting offset in the inode for our pages */
59 /* number of bytes in the inode we're working on */
62 /* number of bytes on disk */
63 unsigned long compressed_len
;
65 /* the compression algorithm for this bio */
68 /* number of compressed pages in the array */
69 unsigned long nr_pages
;
75 /* for reads, this is the bio we are copying the data into */
79 * the start of a variable length array of checksums only
85 static inline int compressed_bio_size(struct btrfs_root
*root
,
86 unsigned long disk_size
)
88 u16 csum_size
= btrfs_super_csum_size(root
->fs_info
->super_copy
);
90 return sizeof(struct compressed_bio
) +
91 ((disk_size
+ root
->sectorsize
- 1) / root
->sectorsize
) *
95 static struct bio
*compressed_bio_alloc(struct block_device
*bdev
,
96 u64 first_byte
, gfp_t gfp_flags
)
100 nr_vecs
= bio_get_nr_vecs(bdev
);
101 return btrfs_bio_alloc(bdev
, first_byte
>> 9, nr_vecs
, gfp_flags
);
104 static int check_compressed_csum(struct inode
*inode
,
105 struct compressed_bio
*cb
,
109 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
114 u32
*cb_sum
= &cb
->sums
;
116 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
119 for (i
= 0; i
< cb
->nr_pages
; i
++) {
120 page
= cb
->compressed_pages
[i
];
123 kaddr
= kmap_atomic(page
);
124 csum
= btrfs_csum_data(root
, kaddr
, csum
, PAGE_CACHE_SIZE
);
125 btrfs_csum_final(csum
, (char *)&csum
);
126 kunmap_atomic(kaddr
);
128 if (csum
!= *cb_sum
) {
129 printk(KERN_INFO
"btrfs csum failed ino %llu "
130 "extent %llu csum %u "
131 "wanted %u mirror %d\n",
132 (unsigned long long)btrfs_ino(inode
),
133 (unsigned long long)disk_start
,
134 csum
, *cb_sum
, cb
->mirror_num
);
146 /* when we finish reading compressed pages from the disk, we
147 * decompress them and then run the bio end_io routines on the
148 * decompressed pages (in the inode address space).
150 * This allows the checksumming and other IO error handling routines
153 * The compressed pages are freed here, and it must be run
156 static void end_compressed_bio_read(struct bio
*bio
, int err
)
158 struct compressed_bio
*cb
= bio
->bi_private
;
167 /* if there are more bios still pending for this compressed
170 if (!atomic_dec_and_test(&cb
->pending_bios
))
174 ret
= check_compressed_csum(inode
, cb
, (u64
)bio
->bi_sector
<< 9);
178 /* ok, we're the last bio for this extent, lets start
181 ret
= btrfs_decompress_biovec(cb
->compress_type
,
182 cb
->compressed_pages
,
184 cb
->orig_bio
->bi_io_vec
,
185 cb
->orig_bio
->bi_vcnt
,
191 /* release the compressed pages */
193 for (index
= 0; index
< cb
->nr_pages
; index
++) {
194 page
= cb
->compressed_pages
[index
];
195 page
->mapping
= NULL
;
196 page_cache_release(page
);
199 /* do io completion on the original bio */
201 bio_io_error(cb
->orig_bio
);
204 struct bio_vec
*bvec
= cb
->orig_bio
->bi_io_vec
;
207 * we have verified the checksum already, set page
208 * checked so the end_io handlers know about it
210 while (bio_index
< cb
->orig_bio
->bi_vcnt
) {
211 SetPageChecked(bvec
->bv_page
);
215 bio_endio(cb
->orig_bio
, 0);
218 /* finally free the cb struct */
219 kfree(cb
->compressed_pages
);
226 * Clear the writeback bits on all of the file
227 * pages for a compressed write
229 static noinline
void end_compressed_writeback(struct inode
*inode
, u64 start
,
230 unsigned long ram_size
)
232 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
233 unsigned long end_index
= (start
+ ram_size
- 1) >> PAGE_CACHE_SHIFT
;
234 struct page
*pages
[16];
235 unsigned long nr_pages
= end_index
- index
+ 1;
239 while (nr_pages
> 0) {
240 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
242 nr_pages
, ARRAY_SIZE(pages
)), pages
);
248 for (i
= 0; i
< ret
; i
++) {
249 end_page_writeback(pages
[i
]);
250 page_cache_release(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
, int err
)
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 cb
->compressed_pages
[0]->mapping
= NULL
;
295 end_compressed_writeback(inode
, cb
->start
, cb
->len
);
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
;
306 page_cache_release(page
);
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 bio
*bio
= NULL
;
332 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
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_CACHE_SIZE
- 1));
344 cb
= kmalloc(compressed_bio_size(root
, 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
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
360 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
365 bio
->bi_private
= cb
;
366 bio
->bi_end_io
= end_compressed_bio_write
;
367 atomic_inc(&cb
->pending_bios
);
369 /* create and submit bios for the compressed pages */
370 bytes_left
= compressed_len
;
371 for (pg_index
= 0; pg_index
< cb
->nr_pages
; pg_index
++) {
372 page
= compressed_pages
[pg_index
];
373 page
->mapping
= inode
->i_mapping
;
375 ret
= io_tree
->ops
->merge_bio_hook(page
, 0,
381 page
->mapping
= NULL
;
382 if (ret
|| bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0) <
387 * inc the count before we submit the bio so
388 * we know the end IO handler won't happen before
389 * we inc the count. Otherwise, the cb might get
390 * freed before we're done setting it up
392 atomic_inc(&cb
->pending_bios
);
393 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
394 BUG_ON(ret
); /* -ENOMEM */
397 ret
= btrfs_csum_one_bio(root
, inode
, bio
,
399 BUG_ON(ret
); /* -ENOMEM */
402 ret
= btrfs_map_bio(root
, WRITE
, bio
, 0, 1);
403 BUG_ON(ret
); /* -ENOMEM */
407 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
409 bio
->bi_private
= cb
;
410 bio
->bi_end_io
= end_compressed_bio_write
;
411 bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0);
413 if (bytes_left
< PAGE_CACHE_SIZE
) {
414 printk("bytes left %lu compress len %lu nr %lu\n",
415 bytes_left
, cb
->compressed_len
, cb
->nr_pages
);
417 bytes_left
-= PAGE_CACHE_SIZE
;
418 first_byte
+= PAGE_CACHE_SIZE
;
423 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
424 BUG_ON(ret
); /* -ENOMEM */
427 ret
= btrfs_csum_one_bio(root
, inode
, bio
, start
, 1);
428 BUG_ON(ret
); /* -ENOMEM */
431 ret
= btrfs_map_bio(root
, WRITE
, bio
, 0, 1);
432 BUG_ON(ret
); /* -ENOMEM */
438 static noinline
int add_ra_bio_pages(struct inode
*inode
,
440 struct compressed_bio
*cb
)
442 unsigned long end_index
;
443 unsigned long pg_index
;
445 u64 isize
= i_size_read(inode
);
448 unsigned long nr_pages
= 0;
449 struct extent_map
*em
;
450 struct address_space
*mapping
= inode
->i_mapping
;
451 struct extent_map_tree
*em_tree
;
452 struct extent_io_tree
*tree
;
456 page
= cb
->orig_bio
->bi_io_vec
[cb
->orig_bio
->bi_vcnt
- 1].bv_page
;
457 last_offset
= (page_offset(page
) + PAGE_CACHE_SIZE
);
458 em_tree
= &BTRFS_I(inode
)->extent_tree
;
459 tree
= &BTRFS_I(inode
)->io_tree
;
464 end_index
= (i_size_read(inode
) - 1) >> PAGE_CACHE_SHIFT
;
466 while (last_offset
< compressed_end
) {
467 pg_index
= last_offset
>> PAGE_CACHE_SHIFT
;
469 if (pg_index
> end_index
)
473 page
= radix_tree_lookup(&mapping
->page_tree
, pg_index
);
482 page
= __page_cache_alloc(mapping_gfp_mask(mapping
) &
487 if (add_to_page_cache_lru(page
, mapping
, pg_index
,
489 page_cache_release(page
);
493 end
= last_offset
+ PAGE_CACHE_SIZE
- 1;
495 * at this point, we have a locked page in the page cache
496 * for these bytes in the file. But, we have to make
497 * sure they map to this compressed extent on disk.
499 set_page_extent_mapped(page
);
500 lock_extent(tree
, last_offset
, end
);
501 read_lock(&em_tree
->lock
);
502 em
= lookup_extent_mapping(em_tree
, last_offset
,
504 read_unlock(&em_tree
->lock
);
506 if (!em
|| last_offset
< em
->start
||
507 (last_offset
+ PAGE_CACHE_SIZE
> extent_map_end(em
)) ||
508 (em
->block_start
>> 9) != cb
->orig_bio
->bi_sector
) {
510 unlock_extent(tree
, last_offset
, end
);
512 page_cache_release(page
);
517 if (page
->index
== end_index
) {
519 size_t zero_offset
= isize
& (PAGE_CACHE_SIZE
- 1);
523 zeros
= PAGE_CACHE_SIZE
- zero_offset
;
524 userpage
= kmap_atomic(page
);
525 memset(userpage
+ zero_offset
, 0, zeros
);
526 flush_dcache_page(page
);
527 kunmap_atomic(userpage
);
531 ret
= bio_add_page(cb
->orig_bio
, page
,
534 if (ret
== PAGE_CACHE_SIZE
) {
536 page_cache_release(page
);
538 unlock_extent(tree
, last_offset
, end
);
540 page_cache_release(page
);
544 last_offset
+= PAGE_CACHE_SIZE
;
550 * for a compressed read, the bio we get passed has all the inode pages
551 * in it. We don't actually do IO on those pages but allocate new ones
552 * to hold the compressed pages on disk.
554 * bio->bi_sector points to the compressed extent on disk
555 * bio->bi_io_vec points to all of the inode pages
556 * bio->bi_vcnt is a count of pages
558 * After the compressed pages are read, we copy the bytes into the
559 * bio we were passed and then call the bio end_io calls
561 int btrfs_submit_compressed_read(struct inode
*inode
, struct bio
*bio
,
562 int mirror_num
, unsigned long bio_flags
)
564 struct extent_io_tree
*tree
;
565 struct extent_map_tree
*em_tree
;
566 struct compressed_bio
*cb
;
567 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
568 unsigned long uncompressed_len
= bio
->bi_vcnt
* PAGE_CACHE_SIZE
;
569 unsigned long compressed_len
;
570 unsigned long nr_pages
;
571 unsigned long pg_index
;
573 struct block_device
*bdev
;
574 struct bio
*comp_bio
;
575 u64 cur_disk_byte
= (u64
)bio
->bi_sector
<< 9;
578 struct extent_map
*em
;
583 tree
= &BTRFS_I(inode
)->io_tree
;
584 em_tree
= &BTRFS_I(inode
)->extent_tree
;
586 /* we need the actual starting offset of this extent in the file */
587 read_lock(&em_tree
->lock
);
588 em
= lookup_extent_mapping(em_tree
,
589 page_offset(bio
->bi_io_vec
->bv_page
),
591 read_unlock(&em_tree
->lock
);
595 compressed_len
= em
->block_len
;
596 cb
= kmalloc(compressed_bio_size(root
, compressed_len
), GFP_NOFS
);
600 atomic_set(&cb
->pending_bios
, 0);
603 cb
->mirror_num
= mirror_num
;
606 cb
->start
= em
->orig_start
;
608 em_start
= em
->start
;
613 cb
->len
= uncompressed_len
;
614 cb
->compressed_len
= compressed_len
;
615 cb
->compress_type
= extent_compress_type(bio_flags
);
618 nr_pages
= (compressed_len
+ PAGE_CACHE_SIZE
- 1) /
620 cb
->compressed_pages
= kzalloc(sizeof(struct page
*) * nr_pages
,
622 if (!cb
->compressed_pages
)
625 bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
627 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
628 cb
->compressed_pages
[pg_index
] = alloc_page(GFP_NOFS
|
630 if (!cb
->compressed_pages
[pg_index
]) {
631 faili
= pg_index
- 1;
636 faili
= nr_pages
- 1;
637 cb
->nr_pages
= nr_pages
;
639 add_ra_bio_pages(inode
, em_start
+ em_len
, cb
);
641 /* include any pages we added in add_ra-bio_pages */
642 uncompressed_len
= bio
->bi_vcnt
* PAGE_CACHE_SIZE
;
643 cb
->len
= uncompressed_len
;
645 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
, GFP_NOFS
);
648 comp_bio
->bi_private
= cb
;
649 comp_bio
->bi_end_io
= end_compressed_bio_read
;
650 atomic_inc(&cb
->pending_bios
);
652 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
653 page
= cb
->compressed_pages
[pg_index
];
654 page
->mapping
= inode
->i_mapping
;
655 page
->index
= em_start
>> PAGE_CACHE_SHIFT
;
657 if (comp_bio
->bi_size
)
658 ret
= tree
->ops
->merge_bio_hook(page
, 0,
664 page
->mapping
= NULL
;
665 if (ret
|| bio_add_page(comp_bio
, page
, PAGE_CACHE_SIZE
, 0) <
669 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
, 0);
670 BUG_ON(ret
); /* -ENOMEM */
673 * inc the count before we submit the bio so
674 * we know the end IO handler won't happen before
675 * we inc the count. Otherwise, the cb might get
676 * freed before we're done setting it up
678 atomic_inc(&cb
->pending_bios
);
680 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
681 ret
= btrfs_lookup_bio_sums(root
, inode
,
683 BUG_ON(ret
); /* -ENOMEM */
685 sums
+= (comp_bio
->bi_size
+ root
->sectorsize
- 1) /
688 ret
= btrfs_map_bio(root
, READ
, comp_bio
,
690 BUG_ON(ret
); /* -ENOMEM */
694 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
,
697 comp_bio
->bi_private
= cb
;
698 comp_bio
->bi_end_io
= end_compressed_bio_read
;
700 bio_add_page(comp_bio
, page
, PAGE_CACHE_SIZE
, 0);
702 cur_disk_byte
+= PAGE_CACHE_SIZE
;
706 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
, 0);
707 BUG_ON(ret
); /* -ENOMEM */
709 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
710 ret
= btrfs_lookup_bio_sums(root
, inode
, comp_bio
, sums
);
711 BUG_ON(ret
); /* -ENOMEM */
714 ret
= btrfs_map_bio(root
, READ
, comp_bio
, mirror_num
, 0);
715 BUG_ON(ret
); /* -ENOMEM */
722 __free_page(cb
->compressed_pages
[faili
]);
726 kfree(cb
->compressed_pages
);
734 static struct list_head comp_idle_workspace
[BTRFS_COMPRESS_TYPES
];
735 static spinlock_t comp_workspace_lock
[BTRFS_COMPRESS_TYPES
];
736 static int comp_num_workspace
[BTRFS_COMPRESS_TYPES
];
737 static atomic_t comp_alloc_workspace
[BTRFS_COMPRESS_TYPES
];
738 static wait_queue_head_t comp_workspace_wait
[BTRFS_COMPRESS_TYPES
];
740 struct btrfs_compress_op
*btrfs_compress_op
[] = {
741 &btrfs_zlib_compress
,
745 void __init
btrfs_init_compress(void)
749 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
750 INIT_LIST_HEAD(&comp_idle_workspace
[i
]);
751 spin_lock_init(&comp_workspace_lock
[i
]);
752 atomic_set(&comp_alloc_workspace
[i
], 0);
753 init_waitqueue_head(&comp_workspace_wait
[i
]);
758 * this finds an available workspace or allocates a new one
759 * ERR_PTR is returned if things go bad.
761 static struct list_head
*find_workspace(int type
)
763 struct list_head
*workspace
;
764 int cpus
= num_online_cpus();
767 struct list_head
*idle_workspace
= &comp_idle_workspace
[idx
];
768 spinlock_t
*workspace_lock
= &comp_workspace_lock
[idx
];
769 atomic_t
*alloc_workspace
= &comp_alloc_workspace
[idx
];
770 wait_queue_head_t
*workspace_wait
= &comp_workspace_wait
[idx
];
771 int *num_workspace
= &comp_num_workspace
[idx
];
773 spin_lock(workspace_lock
);
774 if (!list_empty(idle_workspace
)) {
775 workspace
= idle_workspace
->next
;
778 spin_unlock(workspace_lock
);
782 if (atomic_read(alloc_workspace
) > cpus
) {
785 spin_unlock(workspace_lock
);
786 prepare_to_wait(workspace_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
787 if (atomic_read(alloc_workspace
) > cpus
&& !*num_workspace
)
789 finish_wait(workspace_wait
, &wait
);
792 atomic_inc(alloc_workspace
);
793 spin_unlock(workspace_lock
);
795 workspace
= btrfs_compress_op
[idx
]->alloc_workspace();
796 if (IS_ERR(workspace
)) {
797 atomic_dec(alloc_workspace
);
798 wake_up(workspace_wait
);
804 * put a workspace struct back on the list or free it if we have enough
805 * idle ones sitting around
807 static void free_workspace(int type
, struct list_head
*workspace
)
810 struct list_head
*idle_workspace
= &comp_idle_workspace
[idx
];
811 spinlock_t
*workspace_lock
= &comp_workspace_lock
[idx
];
812 atomic_t
*alloc_workspace
= &comp_alloc_workspace
[idx
];
813 wait_queue_head_t
*workspace_wait
= &comp_workspace_wait
[idx
];
814 int *num_workspace
= &comp_num_workspace
[idx
];
816 spin_lock(workspace_lock
);
817 if (*num_workspace
< num_online_cpus()) {
818 list_add_tail(workspace
, idle_workspace
);
820 spin_unlock(workspace_lock
);
823 spin_unlock(workspace_lock
);
825 btrfs_compress_op
[idx
]->free_workspace(workspace
);
826 atomic_dec(alloc_workspace
);
829 if (waitqueue_active(workspace_wait
))
830 wake_up(workspace_wait
);
834 * cleanup function for module exit
836 static void free_workspaces(void)
838 struct list_head
*workspace
;
841 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
842 while (!list_empty(&comp_idle_workspace
[i
])) {
843 workspace
= comp_idle_workspace
[i
].next
;
845 btrfs_compress_op
[i
]->free_workspace(workspace
);
846 atomic_dec(&comp_alloc_workspace
[i
]);
852 * given an address space and start/len, compress the bytes.
854 * pages are allocated to hold the compressed result and stored
857 * out_pages is used to return the number of pages allocated. There
858 * may be pages allocated even if we return an error
860 * total_in is used to return the number of bytes actually read. It
861 * may be smaller then len if we had to exit early because we
862 * ran out of room in the pages array or because we cross the
865 * total_out is used to return the total number of compressed bytes
867 * max_out tells us the max number of bytes that we're allowed to
870 int btrfs_compress_pages(int type
, struct address_space
*mapping
,
871 u64 start
, unsigned long len
,
873 unsigned long nr_dest_pages
,
874 unsigned long *out_pages
,
875 unsigned long *total_in
,
876 unsigned long *total_out
,
877 unsigned long max_out
)
879 struct list_head
*workspace
;
882 workspace
= find_workspace(type
);
883 if (IS_ERR(workspace
))
886 ret
= btrfs_compress_op
[type
-1]->compress_pages(workspace
, mapping
,
888 nr_dest_pages
, out_pages
,
891 free_workspace(type
, workspace
);
896 * pages_in is an array of pages with compressed data.
898 * disk_start is the starting logical offset of this array in the file
900 * bvec is a bio_vec of pages from the file that we want to decompress into
902 * vcnt is the count of pages in the biovec
904 * srclen is the number of bytes in pages_in
906 * The basic idea is that we have a bio that was created by readpages.
907 * The pages in the bio are for the uncompressed data, and they may not
908 * be contiguous. They all correspond to the range of bytes covered by
909 * the compressed extent.
911 int btrfs_decompress_biovec(int type
, struct page
**pages_in
, u64 disk_start
,
912 struct bio_vec
*bvec
, int vcnt
, size_t srclen
)
914 struct list_head
*workspace
;
917 workspace
= find_workspace(type
);
918 if (IS_ERR(workspace
))
921 ret
= btrfs_compress_op
[type
-1]->decompress_biovec(workspace
, pages_in
,
924 free_workspace(type
, workspace
);
929 * a less complex decompression routine. Our compressed data fits in a
930 * single page, and we want to read a single page out of it.
931 * start_byte tells us the offset into the compressed data we're interested in
933 int btrfs_decompress(int type
, unsigned char *data_in
, struct page
*dest_page
,
934 unsigned long start_byte
, size_t srclen
, size_t destlen
)
936 struct list_head
*workspace
;
939 workspace
= find_workspace(type
);
940 if (IS_ERR(workspace
))
943 ret
= btrfs_compress_op
[type
-1]->decompress(workspace
, data_in
,
944 dest_page
, start_byte
,
947 free_workspace(type
, workspace
);
951 void btrfs_exit_compress(void)
957 * Copy uncompressed data from working buffer to pages.
959 * buf_start is the byte offset we're of the start of our workspace buffer.
961 * total_out is the last byte of the buffer
963 int btrfs_decompress_buf2page(char *buf
, unsigned long buf_start
,
964 unsigned long total_out
, u64 disk_start
,
965 struct bio_vec
*bvec
, int vcnt
,
966 unsigned long *pg_index
,
967 unsigned long *pg_offset
)
969 unsigned long buf_offset
;
970 unsigned long current_buf_start
;
971 unsigned long start_byte
;
972 unsigned long working_bytes
= total_out
- buf_start
;
975 struct page
*page_out
= bvec
[*pg_index
].bv_page
;
978 * start byte is the first byte of the page we're currently
979 * copying into relative to the start of the compressed data.
981 start_byte
= page_offset(page_out
) - disk_start
;
983 /* we haven't yet hit data corresponding to this page */
984 if (total_out
<= start_byte
)
988 * the start of the data we care about is offset into
989 * the middle of our working buffer
991 if (total_out
> start_byte
&& buf_start
< start_byte
) {
992 buf_offset
= start_byte
- buf_start
;
993 working_bytes
-= buf_offset
;
997 current_buf_start
= buf_start
;
999 /* copy bytes from the working buffer into the pages */
1000 while (working_bytes
> 0) {
1001 bytes
= min(PAGE_CACHE_SIZE
- *pg_offset
,
1002 PAGE_CACHE_SIZE
- buf_offset
);
1003 bytes
= min(bytes
, working_bytes
);
1004 kaddr
= kmap_atomic(page_out
);
1005 memcpy(kaddr
+ *pg_offset
, buf
+ buf_offset
, bytes
);
1006 kunmap_atomic(kaddr
);
1007 flush_dcache_page(page_out
);
1009 *pg_offset
+= bytes
;
1010 buf_offset
+= bytes
;
1011 working_bytes
-= bytes
;
1012 current_buf_start
+= bytes
;
1014 /* check if we need to pick another page */
1015 if (*pg_offset
== PAGE_CACHE_SIZE
) {
1017 if (*pg_index
>= vcnt
)
1020 page_out
= bvec
[*pg_index
].bv_page
;
1022 start_byte
= page_offset(page_out
) - disk_start
;
1025 * make sure our new page is covered by this
1028 if (total_out
<= start_byte
)
1032 * the next page in the biovec might not be adjacent
1033 * to the last page, but it might still be found
1034 * inside this working buffer. bump our offset pointer
1036 if (total_out
> start_byte
&&
1037 current_buf_start
< start_byte
) {
1038 buf_offset
= start_byte
- buf_start
;
1039 working_bytes
= total_out
- start_byte
;
1040 current_buf_start
= buf_start
+ buf_offset
;