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
);
89 return sizeof(struct compressed_bio
) +
90 ((disk_size
+ root
->sectorsize
- 1) / root
->sectorsize
) *
94 static struct bio
*compressed_bio_alloc(struct block_device
*bdev
,
95 u64 first_byte
, gfp_t gfp_flags
)
99 nr_vecs
= bio_get_nr_vecs(bdev
);
100 return btrfs_bio_alloc(bdev
, first_byte
>> 9, nr_vecs
, gfp_flags
);
103 static int check_compressed_csum(struct inode
*inode
,
104 struct compressed_bio
*cb
,
108 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
113 u32
*cb_sum
= &cb
->sums
;
115 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
118 for (i
= 0; i
< cb
->nr_pages
; i
++) {
119 page
= cb
->compressed_pages
[i
];
122 kaddr
= kmap_atomic(page
, KM_USER0
);
123 csum
= btrfs_csum_data(root
, kaddr
, csum
, PAGE_CACHE_SIZE
);
124 btrfs_csum_final(csum
, (char *)&csum
);
125 kunmap_atomic(kaddr
, KM_USER0
);
127 if (csum
!= *cb_sum
) {
128 printk(KERN_INFO
"btrfs csum failed ino %lu "
129 "extent %llu csum %u "
130 "wanted %u mirror %d\n", inode
->i_ino
,
131 (unsigned long long)disk_start
,
132 csum
, *cb_sum
, cb
->mirror_num
);
144 /* when we finish reading compressed pages from the disk, we
145 * decompress them and then run the bio end_io routines on the
146 * decompressed pages (in the inode address space).
148 * This allows the checksumming and other IO error handling routines
151 * The compressed pages are freed here, and it must be run
154 static void end_compressed_bio_read(struct bio
*bio
, int err
)
156 struct compressed_bio
*cb
= bio
->bi_private
;
165 /* if there are more bios still pending for this compressed
168 if (!atomic_dec_and_test(&cb
->pending_bios
))
172 ret
= check_compressed_csum(inode
, cb
, (u64
)bio
->bi_sector
<< 9);
176 /* ok, we're the last bio for this extent, lets start
179 ret
= btrfs_decompress_biovec(cb
->compress_type
,
180 cb
->compressed_pages
,
182 cb
->orig_bio
->bi_io_vec
,
183 cb
->orig_bio
->bi_vcnt
,
189 /* release the compressed pages */
191 for (index
= 0; index
< cb
->nr_pages
; index
++) {
192 page
= cb
->compressed_pages
[index
];
193 page
->mapping
= NULL
;
194 page_cache_release(page
);
197 /* do io completion on the original bio */
199 bio_io_error(cb
->orig_bio
);
202 struct bio_vec
*bvec
= cb
->orig_bio
->bi_io_vec
;
205 * we have verified the checksum already, set page
206 * checked so the end_io handlers know about it
208 while (bio_index
< cb
->orig_bio
->bi_vcnt
) {
209 SetPageChecked(bvec
->bv_page
);
213 bio_endio(cb
->orig_bio
, 0);
216 /* finally free the cb struct */
217 kfree(cb
->compressed_pages
);
224 * Clear the writeback bits on all of the file
225 * pages for a compressed write
227 static noinline
int end_compressed_writeback(struct inode
*inode
, u64 start
,
228 unsigned long ram_size
)
230 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
231 unsigned long end_index
= (start
+ ram_size
- 1) >> PAGE_CACHE_SHIFT
;
232 struct page
*pages
[16];
233 unsigned long nr_pages
= end_index
- index
+ 1;
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
++) {
247 end_page_writeback(pages
[i
]);
248 page_cache_release(pages
[i
]);
253 /* 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
, int err
)
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 cb
->compressed_pages
[0]->mapping
= NULL
;
294 end_compressed_writeback(inode
, cb
->start
, cb
->len
);
295 /* note, our inode could be gone now */
298 * release the compressed pages, these came from alloc_page and
299 * are not attached to the inode at all
302 for (index
= 0; index
< cb
->nr_pages
; index
++) {
303 page
= cb
->compressed_pages
[index
];
304 page
->mapping
= NULL
;
305 page_cache_release(page
);
308 /* finally free the cb struct */
309 kfree(cb
->compressed_pages
);
316 * worker function to build and submit bios for previously compressed pages.
317 * The corresponding pages in the inode should be marked for writeback
318 * and the compressed pages should have a reference on them for dropping
319 * when the IO is complete.
321 * This also checksums the file bytes and gets things ready for
324 int btrfs_submit_compressed_write(struct inode
*inode
, u64 start
,
325 unsigned long len
, u64 disk_start
,
326 unsigned long compressed_len
,
327 struct page
**compressed_pages
,
328 unsigned long nr_pages
)
330 struct bio
*bio
= NULL
;
331 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
332 struct compressed_bio
*cb
;
333 unsigned long bytes_left
;
334 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
337 u64 first_byte
= disk_start
;
338 struct block_device
*bdev
;
341 WARN_ON(start
& ((u64
)PAGE_CACHE_SIZE
- 1));
342 cb
= kmalloc(compressed_bio_size(root
, compressed_len
), GFP_NOFS
);
345 atomic_set(&cb
->pending_bios
, 0);
351 cb
->compressed_pages
= compressed_pages
;
352 cb
->compressed_len
= compressed_len
;
354 cb
->nr_pages
= nr_pages
;
356 bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
358 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
363 bio
->bi_private
= cb
;
364 bio
->bi_end_io
= end_compressed_bio_write
;
365 atomic_inc(&cb
->pending_bios
);
367 /* create and submit bios for the compressed pages */
368 bytes_left
= compressed_len
;
369 for (page_index
= 0; page_index
< cb
->nr_pages
; page_index
++) {
370 page
= compressed_pages
[page_index
];
371 page
->mapping
= inode
->i_mapping
;
373 ret
= io_tree
->ops
->merge_bio_hook(page
, 0,
379 page
->mapping
= NULL
;
380 if (ret
|| bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0) <
385 * inc the count before we submit the bio so
386 * we know the end IO handler won't happen before
387 * we inc the count. Otherwise, the cb might get
388 * freed before we're done setting it up
390 atomic_inc(&cb
->pending_bios
);
391 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
394 ret
= btrfs_csum_one_bio(root
, inode
, bio
, start
, 1);
397 ret
= btrfs_map_bio(root
, WRITE
, bio
, 0, 1);
402 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
403 bio
->bi_private
= cb
;
404 bio
->bi_end_io
= end_compressed_bio_write
;
405 bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0);
407 if (bytes_left
< PAGE_CACHE_SIZE
) {
408 printk("bytes left %lu compress len %lu nr %lu\n",
409 bytes_left
, cb
->compressed_len
, cb
->nr_pages
);
411 bytes_left
-= PAGE_CACHE_SIZE
;
412 first_byte
+= PAGE_CACHE_SIZE
;
417 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
420 ret
= btrfs_csum_one_bio(root
, inode
, bio
, start
, 1);
423 ret
= btrfs_map_bio(root
, WRITE
, bio
, 0, 1);
430 static noinline
int add_ra_bio_pages(struct inode
*inode
,
432 struct compressed_bio
*cb
)
434 unsigned long end_index
;
435 unsigned long page_index
;
437 u64 isize
= i_size_read(inode
);
440 unsigned long nr_pages
= 0;
441 struct extent_map
*em
;
442 struct address_space
*mapping
= inode
->i_mapping
;
443 struct extent_map_tree
*em_tree
;
444 struct extent_io_tree
*tree
;
448 page
= cb
->orig_bio
->bi_io_vec
[cb
->orig_bio
->bi_vcnt
- 1].bv_page
;
449 last_offset
= (page_offset(page
) + PAGE_CACHE_SIZE
);
450 em_tree
= &BTRFS_I(inode
)->extent_tree
;
451 tree
= &BTRFS_I(inode
)->io_tree
;
456 end_index
= (i_size_read(inode
) - 1) >> PAGE_CACHE_SHIFT
;
458 while (last_offset
< compressed_end
) {
459 page_index
= last_offset
>> PAGE_CACHE_SHIFT
;
461 if (page_index
> end_index
)
465 page
= radix_tree_lookup(&mapping
->page_tree
, page_index
);
474 page
= __page_cache_alloc(mapping_gfp_mask(mapping
) &
479 if (add_to_page_cache_lru(page
, mapping
, page_index
,
481 page_cache_release(page
);
485 end
= last_offset
+ PAGE_CACHE_SIZE
- 1;
487 * at this point, we have a locked page in the page cache
488 * for these bytes in the file. But, we have to make
489 * sure they map to this compressed extent on disk.
491 set_page_extent_mapped(page
);
492 lock_extent(tree
, last_offset
, end
, GFP_NOFS
);
493 read_lock(&em_tree
->lock
);
494 em
= lookup_extent_mapping(em_tree
, last_offset
,
496 read_unlock(&em_tree
->lock
);
498 if (!em
|| last_offset
< em
->start
||
499 (last_offset
+ PAGE_CACHE_SIZE
> extent_map_end(em
)) ||
500 (em
->block_start
>> 9) != cb
->orig_bio
->bi_sector
) {
502 unlock_extent(tree
, last_offset
, end
, GFP_NOFS
);
504 page_cache_release(page
);
509 if (page
->index
== end_index
) {
511 size_t zero_offset
= isize
& (PAGE_CACHE_SIZE
- 1);
515 zeros
= PAGE_CACHE_SIZE
- zero_offset
;
516 userpage
= kmap_atomic(page
, KM_USER0
);
517 memset(userpage
+ zero_offset
, 0, zeros
);
518 flush_dcache_page(page
);
519 kunmap_atomic(userpage
, KM_USER0
);
523 ret
= bio_add_page(cb
->orig_bio
, page
,
526 if (ret
== PAGE_CACHE_SIZE
) {
528 page_cache_release(page
);
530 unlock_extent(tree
, last_offset
, end
, GFP_NOFS
);
532 page_cache_release(page
);
536 last_offset
+= PAGE_CACHE_SIZE
;
542 * for a compressed read, the bio we get passed has all the inode pages
543 * in it. We don't actually do IO on those pages but allocate new ones
544 * to hold the compressed pages on disk.
546 * bio->bi_sector points to the compressed extent on disk
547 * bio->bi_io_vec points to all of the inode pages
548 * bio->bi_vcnt is a count of pages
550 * After the compressed pages are read, we copy the bytes into the
551 * bio we were passed and then call the bio end_io calls
553 int btrfs_submit_compressed_read(struct inode
*inode
, struct bio
*bio
,
554 int mirror_num
, unsigned long bio_flags
)
556 struct extent_io_tree
*tree
;
557 struct extent_map_tree
*em_tree
;
558 struct compressed_bio
*cb
;
559 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
560 unsigned long uncompressed_len
= bio
->bi_vcnt
* PAGE_CACHE_SIZE
;
561 unsigned long compressed_len
;
562 unsigned long nr_pages
;
563 unsigned long page_index
;
565 struct block_device
*bdev
;
566 struct bio
*comp_bio
;
567 u64 cur_disk_byte
= (u64
)bio
->bi_sector
<< 9;
570 struct extent_map
*em
;
574 tree
= &BTRFS_I(inode
)->io_tree
;
575 em_tree
= &BTRFS_I(inode
)->extent_tree
;
577 /* we need the actual starting offset of this extent in the file */
578 read_lock(&em_tree
->lock
);
579 em
= lookup_extent_mapping(em_tree
,
580 page_offset(bio
->bi_io_vec
->bv_page
),
582 read_unlock(&em_tree
->lock
);
584 compressed_len
= em
->block_len
;
585 cb
= kmalloc(compressed_bio_size(root
, compressed_len
), GFP_NOFS
);
589 atomic_set(&cb
->pending_bios
, 0);
592 cb
->mirror_num
= mirror_num
;
595 cb
->start
= em
->orig_start
;
597 em_start
= em
->start
;
602 cb
->len
= uncompressed_len
;
603 cb
->compressed_len
= compressed_len
;
604 cb
->compress_type
= extent_compress_type(bio_flags
);
607 nr_pages
= (compressed_len
+ PAGE_CACHE_SIZE
- 1) /
609 cb
->compressed_pages
= kzalloc(sizeof(struct page
*) * nr_pages
,
611 if (!cb
->compressed_pages
)
614 bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
616 for (page_index
= 0; page_index
< nr_pages
; page_index
++) {
617 cb
->compressed_pages
[page_index
] = alloc_page(GFP_NOFS
|
619 if (!cb
->compressed_pages
[page_index
])
622 cb
->nr_pages
= nr_pages
;
624 add_ra_bio_pages(inode
, em_start
+ em_len
, cb
);
626 /* include any pages we added in add_ra-bio_pages */
627 uncompressed_len
= bio
->bi_vcnt
* PAGE_CACHE_SIZE
;
628 cb
->len
= uncompressed_len
;
630 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
, GFP_NOFS
);
633 comp_bio
->bi_private
= cb
;
634 comp_bio
->bi_end_io
= end_compressed_bio_read
;
635 atomic_inc(&cb
->pending_bios
);
637 for (page_index
= 0; page_index
< nr_pages
; page_index
++) {
638 page
= cb
->compressed_pages
[page_index
];
639 page
->mapping
= inode
->i_mapping
;
640 page
->index
= em_start
>> PAGE_CACHE_SHIFT
;
642 if (comp_bio
->bi_size
)
643 ret
= tree
->ops
->merge_bio_hook(page
, 0,
649 page
->mapping
= NULL
;
650 if (ret
|| bio_add_page(comp_bio
, page
, PAGE_CACHE_SIZE
, 0) <
654 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
, 0);
658 * inc the count before we submit the bio so
659 * we know the end IO handler won't happen before
660 * we inc the count. Otherwise, the cb might get
661 * freed before we're done setting it up
663 atomic_inc(&cb
->pending_bios
);
665 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
666 btrfs_lookup_bio_sums(root
, inode
, comp_bio
,
669 sums
+= (comp_bio
->bi_size
+ root
->sectorsize
- 1) /
672 ret
= btrfs_map_bio(root
, READ
, comp_bio
,
678 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
,
680 comp_bio
->bi_private
= cb
;
681 comp_bio
->bi_end_io
= end_compressed_bio_read
;
683 bio_add_page(comp_bio
, page
, PAGE_CACHE_SIZE
, 0);
685 cur_disk_byte
+= PAGE_CACHE_SIZE
;
689 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
, 0);
692 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
))
693 btrfs_lookup_bio_sums(root
, inode
, comp_bio
, sums
);
695 ret
= btrfs_map_bio(root
, READ
, comp_bio
, mirror_num
, 0);
702 for (page_index
= 0; page_index
< nr_pages
; page_index
++)
703 free_page((unsigned long)cb
->compressed_pages
[page_index
]);
705 kfree(cb
->compressed_pages
);
713 static struct list_head comp_idle_workspace
[BTRFS_COMPRESS_TYPES
];
714 static spinlock_t comp_workspace_lock
[BTRFS_COMPRESS_TYPES
];
715 static int comp_num_workspace
[BTRFS_COMPRESS_TYPES
];
716 static atomic_t comp_alloc_workspace
[BTRFS_COMPRESS_TYPES
];
717 static wait_queue_head_t comp_workspace_wait
[BTRFS_COMPRESS_TYPES
];
719 struct btrfs_compress_op
*btrfs_compress_op
[] = {
720 &btrfs_zlib_compress
,
724 int __init
btrfs_init_compress(void)
728 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
729 INIT_LIST_HEAD(&comp_idle_workspace
[i
]);
730 spin_lock_init(&comp_workspace_lock
[i
]);
731 atomic_set(&comp_alloc_workspace
[i
], 0);
732 init_waitqueue_head(&comp_workspace_wait
[i
]);
738 * this finds an available workspace or allocates a new one
739 * ERR_PTR is returned if things go bad.
741 static struct list_head
*find_workspace(int type
)
743 struct list_head
*workspace
;
744 int cpus
= num_online_cpus();
747 struct list_head
*idle_workspace
= &comp_idle_workspace
[idx
];
748 spinlock_t
*workspace_lock
= &comp_workspace_lock
[idx
];
749 atomic_t
*alloc_workspace
= &comp_alloc_workspace
[idx
];
750 wait_queue_head_t
*workspace_wait
= &comp_workspace_wait
[idx
];
751 int *num_workspace
= &comp_num_workspace
[idx
];
753 spin_lock(workspace_lock
);
754 if (!list_empty(idle_workspace
)) {
755 workspace
= idle_workspace
->next
;
758 spin_unlock(workspace_lock
);
762 if (atomic_read(alloc_workspace
) > cpus
) {
765 spin_unlock(workspace_lock
);
766 prepare_to_wait(workspace_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
767 if (atomic_read(alloc_workspace
) > cpus
&& !*num_workspace
)
769 finish_wait(workspace_wait
, &wait
);
772 atomic_inc(alloc_workspace
);
773 spin_unlock(workspace_lock
);
775 workspace
= btrfs_compress_op
[idx
]->alloc_workspace();
776 if (IS_ERR(workspace
)) {
777 atomic_dec(alloc_workspace
);
778 wake_up(workspace_wait
);
784 * put a workspace struct back on the list or free it if we have enough
785 * idle ones sitting around
787 static void free_workspace(int type
, struct list_head
*workspace
)
790 struct list_head
*idle_workspace
= &comp_idle_workspace
[idx
];
791 spinlock_t
*workspace_lock
= &comp_workspace_lock
[idx
];
792 atomic_t
*alloc_workspace
= &comp_alloc_workspace
[idx
];
793 wait_queue_head_t
*workspace_wait
= &comp_workspace_wait
[idx
];
794 int *num_workspace
= &comp_num_workspace
[idx
];
796 spin_lock(workspace_lock
);
797 if (*num_workspace
< num_online_cpus()) {
798 list_add_tail(workspace
, idle_workspace
);
800 spin_unlock(workspace_lock
);
803 spin_unlock(workspace_lock
);
805 btrfs_compress_op
[idx
]->free_workspace(workspace
);
806 atomic_dec(alloc_workspace
);
808 if (waitqueue_active(workspace_wait
))
809 wake_up(workspace_wait
);
813 * cleanup function for module exit
815 static void free_workspaces(void)
817 struct list_head
*workspace
;
820 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
821 while (!list_empty(&comp_idle_workspace
[i
])) {
822 workspace
= comp_idle_workspace
[i
].next
;
824 btrfs_compress_op
[i
]->free_workspace(workspace
);
825 atomic_dec(&comp_alloc_workspace
[i
]);
831 * given an address space and start/len, compress the bytes.
833 * pages are allocated to hold the compressed result and stored
836 * out_pages is used to return the number of pages allocated. There
837 * may be pages allocated even if we return an error
839 * total_in is used to return the number of bytes actually read. It
840 * may be smaller then len if we had to exit early because we
841 * ran out of room in the pages array or because we cross the
844 * total_out is used to return the total number of compressed bytes
846 * max_out tells us the max number of bytes that we're allowed to
849 int btrfs_compress_pages(int type
, struct address_space
*mapping
,
850 u64 start
, unsigned long len
,
852 unsigned long nr_dest_pages
,
853 unsigned long *out_pages
,
854 unsigned long *total_in
,
855 unsigned long *total_out
,
856 unsigned long max_out
)
858 struct list_head
*workspace
;
861 workspace
= find_workspace(type
);
862 if (IS_ERR(workspace
))
865 ret
= btrfs_compress_op
[type
-1]->compress_pages(workspace
, mapping
,
867 nr_dest_pages
, out_pages
,
870 free_workspace(type
, workspace
);
875 * pages_in is an array of pages with compressed data.
877 * disk_start is the starting logical offset of this array in the file
879 * bvec is a bio_vec of pages from the file that we want to decompress into
881 * vcnt is the count of pages in the biovec
883 * srclen is the number of bytes in pages_in
885 * The basic idea is that we have a bio that was created by readpages.
886 * The pages in the bio are for the uncompressed data, and they may not
887 * be contiguous. They all correspond to the range of bytes covered by
888 * the compressed extent.
890 int btrfs_decompress_biovec(int type
, struct page
**pages_in
, u64 disk_start
,
891 struct bio_vec
*bvec
, int vcnt
, size_t srclen
)
893 struct list_head
*workspace
;
896 workspace
= find_workspace(type
);
897 if (IS_ERR(workspace
))
900 ret
= btrfs_compress_op
[type
-1]->decompress_biovec(workspace
, pages_in
,
903 free_workspace(type
, workspace
);
908 * a less complex decompression routine. Our compressed data fits in a
909 * single page, and we want to read a single page out of it.
910 * start_byte tells us the offset into the compressed data we're interested in
912 int btrfs_decompress(int type
, unsigned char *data_in
, struct page
*dest_page
,
913 unsigned long start_byte
, size_t srclen
, size_t destlen
)
915 struct list_head
*workspace
;
918 workspace
= find_workspace(type
);
919 if (IS_ERR(workspace
))
922 ret
= btrfs_compress_op
[type
-1]->decompress(workspace
, data_in
,
923 dest_page
, start_byte
,
926 free_workspace(type
, workspace
);
930 void btrfs_exit_compress(void)
936 * Copy uncompressed data from working buffer to pages.
938 * buf_start is the byte offset we're of the start of our workspace buffer.
940 * total_out is the last byte of the buffer
942 int btrfs_decompress_buf2page(char *buf
, unsigned long buf_start
,
943 unsigned long total_out
, u64 disk_start
,
944 struct bio_vec
*bvec
, int vcnt
,
945 unsigned long *page_index
,
946 unsigned long *pg_offset
)
948 unsigned long buf_offset
;
949 unsigned long current_buf_start
;
950 unsigned long start_byte
;
951 unsigned long working_bytes
= total_out
- buf_start
;
954 struct page
*page_out
= bvec
[*page_index
].bv_page
;
957 * start byte is the first byte of the page we're currently
958 * copying into relative to the start of the compressed data.
960 start_byte
= page_offset(page_out
) - disk_start
;
962 /* we haven't yet hit data corresponding to this page */
963 if (total_out
<= start_byte
)
967 * the start of the data we care about is offset into
968 * the middle of our working buffer
970 if (total_out
> start_byte
&& buf_start
< start_byte
) {
971 buf_offset
= start_byte
- buf_start
;
972 working_bytes
-= buf_offset
;
976 current_buf_start
= buf_start
;
978 /* copy bytes from the working buffer into the pages */
979 while (working_bytes
> 0) {
980 bytes
= min(PAGE_CACHE_SIZE
- *pg_offset
,
981 PAGE_CACHE_SIZE
- buf_offset
);
982 bytes
= min(bytes
, working_bytes
);
983 kaddr
= kmap_atomic(page_out
, KM_USER0
);
984 memcpy(kaddr
+ *pg_offset
, buf
+ buf_offset
, bytes
);
985 kunmap_atomic(kaddr
, KM_USER0
);
986 flush_dcache_page(page_out
);
990 working_bytes
-= bytes
;
991 current_buf_start
+= bytes
;
993 /* check if we need to pick another page */
994 if (*pg_offset
== PAGE_CACHE_SIZE
) {
996 if (*page_index
>= vcnt
)
999 page_out
= bvec
[*page_index
].bv_page
;
1001 start_byte
= page_offset(page_out
) - disk_start
;
1004 * make sure our new page is covered by this
1007 if (total_out
<= start_byte
)
1011 * the next page in the biovec might not be adjacent
1012 * to the last page, but it might still be found
1013 * inside this working buffer. bump our offset pointer
1015 if (total_out
> start_byte
&&
1016 current_buf_start
< start_byte
) {
1017 buf_offset
= start_byte
- buf_start
;
1018 working_bytes
= total_out
- start_byte
;
1019 current_buf_start
= buf_start
+ buf_offset
;