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_CACHE_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
;
193 page_cache_release(page
);
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_CACHE_SHIFT
;
228 unsigned long end_index
= (cb
->start
+ cb
->len
- 1) >> PAGE_CACHE_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
]);
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
)
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
;
307 page_cache_release(page
);
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_CACHE_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_CACHE_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);
405 BUG_ON(ret
); /* -ENOMEM */
409 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
411 bio
->bi_private
= cb
;
412 bio
->bi_end_io
= end_compressed_bio_write
;
413 bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0);
415 if (bytes_left
< PAGE_CACHE_SIZE
) {
416 btrfs_info(BTRFS_I(inode
)->root
->fs_info
,
417 "bytes left %lu compress len %lu nr %lu",
418 bytes_left
, cb
->compressed_len
, cb
->nr_pages
);
420 bytes_left
-= PAGE_CACHE_SIZE
;
421 first_byte
+= PAGE_CACHE_SIZE
;
426 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, BTRFS_WQ_ENDIO_DATA
);
427 BUG_ON(ret
); /* -ENOMEM */
430 ret
= btrfs_csum_one_bio(root
, inode
, bio
, start
, 1);
431 BUG_ON(ret
); /* -ENOMEM */
434 ret
= btrfs_map_bio(root
, WRITE
, bio
, 0, 1);
435 BUG_ON(ret
); /* -ENOMEM */
441 static noinline
int add_ra_bio_pages(struct inode
*inode
,
443 struct compressed_bio
*cb
)
445 unsigned long end_index
;
446 unsigned long pg_index
;
448 u64 isize
= i_size_read(inode
);
451 unsigned long nr_pages
= 0;
452 struct extent_map
*em
;
453 struct address_space
*mapping
= inode
->i_mapping
;
454 struct extent_map_tree
*em_tree
;
455 struct extent_io_tree
*tree
;
459 page
= cb
->orig_bio
->bi_io_vec
[cb
->orig_bio
->bi_vcnt
- 1].bv_page
;
460 last_offset
= (page_offset(page
) + PAGE_CACHE_SIZE
);
461 em_tree
= &BTRFS_I(inode
)->extent_tree
;
462 tree
= &BTRFS_I(inode
)->io_tree
;
467 end_index
= (i_size_read(inode
) - 1) >> PAGE_CACHE_SHIFT
;
469 while (last_offset
< compressed_end
) {
470 pg_index
= last_offset
>> PAGE_CACHE_SHIFT
;
472 if (pg_index
> end_index
)
476 page
= radix_tree_lookup(&mapping
->page_tree
, pg_index
);
478 if (page
&& !radix_tree_exceptional_entry(page
)) {
485 page
= __page_cache_alloc(mapping_gfp_constraint(mapping
,
490 if (add_to_page_cache_lru(page
, mapping
, pg_index
, GFP_NOFS
)) {
491 page_cache_release(page
);
495 end
= last_offset
+ PAGE_CACHE_SIZE
- 1;
497 * at this point, we have a locked page in the page cache
498 * for these bytes in the file. But, we have to make
499 * sure they map to this compressed extent on disk.
501 set_page_extent_mapped(page
);
502 lock_extent(tree
, last_offset
, end
);
503 read_lock(&em_tree
->lock
);
504 em
= lookup_extent_mapping(em_tree
, last_offset
,
506 read_unlock(&em_tree
->lock
);
508 if (!em
|| last_offset
< em
->start
||
509 (last_offset
+ PAGE_CACHE_SIZE
> extent_map_end(em
)) ||
510 (em
->block_start
>> 9) != cb
->orig_bio
->bi_iter
.bi_sector
) {
512 unlock_extent(tree
, last_offset
, end
);
514 page_cache_release(page
);
519 if (page
->index
== end_index
) {
521 size_t zero_offset
= isize
& (PAGE_CACHE_SIZE
- 1);
525 zeros
= PAGE_CACHE_SIZE
- zero_offset
;
526 userpage
= kmap_atomic(page
);
527 memset(userpage
+ zero_offset
, 0, zeros
);
528 flush_dcache_page(page
);
529 kunmap_atomic(userpage
);
533 ret
= bio_add_page(cb
->orig_bio
, page
,
536 if (ret
== PAGE_CACHE_SIZE
) {
538 page_cache_release(page
);
540 unlock_extent(tree
, last_offset
, end
);
542 page_cache_release(page
);
546 last_offset
+= PAGE_CACHE_SIZE
;
552 * for a compressed read, the bio we get passed has all the inode pages
553 * in it. We don't actually do IO on those pages but allocate new ones
554 * to hold the compressed pages on disk.
556 * bio->bi_iter.bi_sector points to the compressed extent on disk
557 * bio->bi_io_vec points to all of the inode pages
558 * bio->bi_vcnt is a count of pages
560 * After the compressed pages are read, we copy the bytes into the
561 * bio we were passed and then call the bio end_io calls
563 int btrfs_submit_compressed_read(struct inode
*inode
, struct bio
*bio
,
564 int mirror_num
, unsigned long bio_flags
)
566 struct extent_io_tree
*tree
;
567 struct extent_map_tree
*em_tree
;
568 struct compressed_bio
*cb
;
569 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
570 unsigned long uncompressed_len
= bio
->bi_vcnt
* PAGE_CACHE_SIZE
;
571 unsigned long compressed_len
;
572 unsigned long nr_pages
;
573 unsigned long pg_index
;
575 struct block_device
*bdev
;
576 struct bio
*comp_bio
;
577 u64 cur_disk_byte
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
580 struct extent_map
*em
;
585 tree
= &BTRFS_I(inode
)->io_tree
;
586 em_tree
= &BTRFS_I(inode
)->extent_tree
;
588 /* we need the actual starting offset of this extent in the file */
589 read_lock(&em_tree
->lock
);
590 em
= lookup_extent_mapping(em_tree
,
591 page_offset(bio
->bi_io_vec
->bv_page
),
593 read_unlock(&em_tree
->lock
);
597 compressed_len
= em
->block_len
;
598 cb
= kmalloc(compressed_bio_size(root
, compressed_len
), GFP_NOFS
);
602 atomic_set(&cb
->pending_bios
, 0);
605 cb
->mirror_num
= mirror_num
;
608 cb
->start
= em
->orig_start
;
610 em_start
= em
->start
;
615 cb
->len
= uncompressed_len
;
616 cb
->compressed_len
= compressed_len
;
617 cb
->compress_type
= extent_compress_type(bio_flags
);
620 nr_pages
= DIV_ROUND_UP(compressed_len
, PAGE_CACHE_SIZE
);
621 cb
->compressed_pages
= kcalloc(nr_pages
, sizeof(struct page
*),
623 if (!cb
->compressed_pages
)
626 bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
628 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
629 cb
->compressed_pages
[pg_index
] = alloc_page(GFP_NOFS
|
631 if (!cb
->compressed_pages
[pg_index
]) {
632 faili
= pg_index
- 1;
637 faili
= nr_pages
- 1;
638 cb
->nr_pages
= nr_pages
;
640 /* In the parent-locked case, we only locked the range we are
641 * interested in. In all other cases, we can opportunistically
642 * cache decompressed data that goes beyond the requested range. */
643 if (!(bio_flags
& EXTENT_BIO_PARENT_LOCKED
))
644 add_ra_bio_pages(inode
, em_start
+ em_len
, cb
);
646 /* include any pages we added in add_ra-bio_pages */
647 uncompressed_len
= bio
->bi_vcnt
* PAGE_CACHE_SIZE
;
648 cb
->len
= uncompressed_len
;
650 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
, GFP_NOFS
);
653 comp_bio
->bi_private
= cb
;
654 comp_bio
->bi_end_io
= end_compressed_bio_read
;
655 atomic_inc(&cb
->pending_bios
);
657 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
658 page
= cb
->compressed_pages
[pg_index
];
659 page
->mapping
= inode
->i_mapping
;
660 page
->index
= em_start
>> PAGE_CACHE_SHIFT
;
662 if (comp_bio
->bi_iter
.bi_size
)
663 ret
= tree
->ops
->merge_bio_hook(READ
, page
, 0,
669 page
->mapping
= NULL
;
670 if (ret
|| bio_add_page(comp_bio
, page
, PAGE_CACHE_SIZE
, 0) <
674 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
,
675 BTRFS_WQ_ENDIO_DATA
);
676 BUG_ON(ret
); /* -ENOMEM */
679 * inc the count before we submit the bio so
680 * we know the end IO handler won't happen before
681 * we inc the count. Otherwise, the cb might get
682 * freed before we're done setting it up
684 atomic_inc(&cb
->pending_bios
);
686 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
687 ret
= btrfs_lookup_bio_sums(root
, inode
,
689 BUG_ON(ret
); /* -ENOMEM */
691 sums
+= DIV_ROUND_UP(comp_bio
->bi_iter
.bi_size
,
694 ret
= btrfs_map_bio(root
, READ
, comp_bio
,
703 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
,
706 comp_bio
->bi_private
= cb
;
707 comp_bio
->bi_end_io
= end_compressed_bio_read
;
709 bio_add_page(comp_bio
, page
, PAGE_CACHE_SIZE
, 0);
711 cur_disk_byte
+= PAGE_CACHE_SIZE
;
715 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
,
716 BTRFS_WQ_ENDIO_DATA
);
717 BUG_ON(ret
); /* -ENOMEM */
719 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
720 ret
= btrfs_lookup_bio_sums(root
, inode
, comp_bio
, sums
);
721 BUG_ON(ret
); /* -ENOMEM */
724 ret
= btrfs_map_bio(root
, READ
, comp_bio
, mirror_num
, 0);
735 __free_page(cb
->compressed_pages
[faili
]);
739 kfree(cb
->compressed_pages
);
747 static struct list_head comp_idle_workspace
[BTRFS_COMPRESS_TYPES
];
748 static spinlock_t comp_workspace_lock
[BTRFS_COMPRESS_TYPES
];
749 static int comp_num_workspace
[BTRFS_COMPRESS_TYPES
];
750 static atomic_t comp_alloc_workspace
[BTRFS_COMPRESS_TYPES
];
751 static wait_queue_head_t comp_workspace_wait
[BTRFS_COMPRESS_TYPES
];
753 static const struct btrfs_compress_op
* const btrfs_compress_op
[] = {
754 &btrfs_zlib_compress
,
758 void __init
btrfs_init_compress(void)
762 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
763 INIT_LIST_HEAD(&comp_idle_workspace
[i
]);
764 spin_lock_init(&comp_workspace_lock
[i
]);
765 atomic_set(&comp_alloc_workspace
[i
], 0);
766 init_waitqueue_head(&comp_workspace_wait
[i
]);
771 * this finds an available workspace or allocates a new one
772 * ERR_PTR is returned if things go bad.
774 static struct list_head
*find_workspace(int type
)
776 struct list_head
*workspace
;
777 int cpus
= num_online_cpus();
780 struct list_head
*idle_workspace
= &comp_idle_workspace
[idx
];
781 spinlock_t
*workspace_lock
= &comp_workspace_lock
[idx
];
782 atomic_t
*alloc_workspace
= &comp_alloc_workspace
[idx
];
783 wait_queue_head_t
*workspace_wait
= &comp_workspace_wait
[idx
];
784 int *num_workspace
= &comp_num_workspace
[idx
];
786 spin_lock(workspace_lock
);
787 if (!list_empty(idle_workspace
)) {
788 workspace
= idle_workspace
->next
;
791 spin_unlock(workspace_lock
);
795 if (atomic_read(alloc_workspace
) > cpus
) {
798 spin_unlock(workspace_lock
);
799 prepare_to_wait(workspace_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
800 if (atomic_read(alloc_workspace
) > cpus
&& !*num_workspace
)
802 finish_wait(workspace_wait
, &wait
);
805 atomic_inc(alloc_workspace
);
806 spin_unlock(workspace_lock
);
808 workspace
= btrfs_compress_op
[idx
]->alloc_workspace();
809 if (IS_ERR(workspace
)) {
810 atomic_dec(alloc_workspace
);
811 wake_up(workspace_wait
);
817 * put a workspace struct back on the list or free it if we have enough
818 * idle ones sitting around
820 static void free_workspace(int type
, struct list_head
*workspace
)
823 struct list_head
*idle_workspace
= &comp_idle_workspace
[idx
];
824 spinlock_t
*workspace_lock
= &comp_workspace_lock
[idx
];
825 atomic_t
*alloc_workspace
= &comp_alloc_workspace
[idx
];
826 wait_queue_head_t
*workspace_wait
= &comp_workspace_wait
[idx
];
827 int *num_workspace
= &comp_num_workspace
[idx
];
829 spin_lock(workspace_lock
);
830 if (*num_workspace
< num_online_cpus()) {
831 list_add(workspace
, idle_workspace
);
833 spin_unlock(workspace_lock
);
836 spin_unlock(workspace_lock
);
838 btrfs_compress_op
[idx
]->free_workspace(workspace
);
839 atomic_dec(alloc_workspace
);
842 if (waitqueue_active(workspace_wait
))
843 wake_up(workspace_wait
);
847 * cleanup function for module exit
849 static void free_workspaces(void)
851 struct list_head
*workspace
;
854 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
855 while (!list_empty(&comp_idle_workspace
[i
])) {
856 workspace
= comp_idle_workspace
[i
].next
;
858 btrfs_compress_op
[i
]->free_workspace(workspace
);
859 atomic_dec(&comp_alloc_workspace
[i
]);
865 * given an address space and start/len, compress the bytes.
867 * pages are allocated to hold the compressed result and stored
870 * out_pages is used to return the number of pages allocated. There
871 * may be pages allocated even if we return an error
873 * total_in is used to return the number of bytes actually read. It
874 * may be smaller then len if we had to exit early because we
875 * ran out of room in the pages array or because we cross the
878 * total_out is used to return the total number of compressed bytes
880 * max_out tells us the max number of bytes that we're allowed to
883 int btrfs_compress_pages(int type
, struct address_space
*mapping
,
884 u64 start
, unsigned long len
,
886 unsigned long nr_dest_pages
,
887 unsigned long *out_pages
,
888 unsigned long *total_in
,
889 unsigned long *total_out
,
890 unsigned long max_out
)
892 struct list_head
*workspace
;
895 workspace
= find_workspace(type
);
896 if (IS_ERR(workspace
))
897 return PTR_ERR(workspace
);
899 ret
= btrfs_compress_op
[type
-1]->compress_pages(workspace
, mapping
,
901 nr_dest_pages
, out_pages
,
904 free_workspace(type
, workspace
);
909 * pages_in is an array of pages with compressed data.
911 * disk_start is the starting logical offset of this array in the file
913 * bvec is a bio_vec of pages from the file that we want to decompress into
915 * vcnt is the count of pages in the biovec
917 * srclen is the number of bytes in pages_in
919 * The basic idea is that we have a bio that was created by readpages.
920 * The pages in the bio are for the uncompressed data, and they may not
921 * be contiguous. They all correspond to the range of bytes covered by
922 * the compressed extent.
924 static int btrfs_decompress_biovec(int type
, struct page
**pages_in
,
925 u64 disk_start
, struct bio_vec
*bvec
,
926 int vcnt
, size_t srclen
)
928 struct list_head
*workspace
;
931 workspace
= find_workspace(type
);
932 if (IS_ERR(workspace
))
933 return PTR_ERR(workspace
);
935 ret
= btrfs_compress_op
[type
-1]->decompress_biovec(workspace
, pages_in
,
938 free_workspace(type
, workspace
);
943 * a less complex decompression routine. Our compressed data fits in a
944 * single page, and we want to read a single page out of it.
945 * start_byte tells us the offset into the compressed data we're interested in
947 int btrfs_decompress(int type
, unsigned char *data_in
, struct page
*dest_page
,
948 unsigned long start_byte
, size_t srclen
, size_t destlen
)
950 struct list_head
*workspace
;
953 workspace
= find_workspace(type
);
954 if (IS_ERR(workspace
))
955 return PTR_ERR(workspace
);
957 ret
= btrfs_compress_op
[type
-1]->decompress(workspace
, data_in
,
958 dest_page
, start_byte
,
961 free_workspace(type
, workspace
);
965 void btrfs_exit_compress(void)
971 * Copy uncompressed data from working buffer to pages.
973 * buf_start is the byte offset we're of the start of our workspace buffer.
975 * total_out is the last byte of the buffer
977 int btrfs_decompress_buf2page(char *buf
, unsigned long buf_start
,
978 unsigned long total_out
, u64 disk_start
,
979 struct bio_vec
*bvec
, int vcnt
,
980 unsigned long *pg_index
,
981 unsigned long *pg_offset
)
983 unsigned long buf_offset
;
984 unsigned long current_buf_start
;
985 unsigned long start_byte
;
986 unsigned long working_bytes
= total_out
- buf_start
;
989 struct page
*page_out
= bvec
[*pg_index
].bv_page
;
992 * start byte is the first byte of the page we're currently
993 * copying into relative to the start of the compressed data.
995 start_byte
= page_offset(page_out
) - disk_start
;
997 /* we haven't yet hit data corresponding to this page */
998 if (total_out
<= start_byte
)
1002 * the start of the data we care about is offset into
1003 * the middle of our working buffer
1005 if (total_out
> start_byte
&& buf_start
< start_byte
) {
1006 buf_offset
= start_byte
- buf_start
;
1007 working_bytes
-= buf_offset
;
1011 current_buf_start
= buf_start
;
1013 /* copy bytes from the working buffer into the pages */
1014 while (working_bytes
> 0) {
1015 bytes
= min(PAGE_CACHE_SIZE
- *pg_offset
,
1016 PAGE_CACHE_SIZE
- buf_offset
);
1017 bytes
= min(bytes
, working_bytes
);
1018 kaddr
= kmap_atomic(page_out
);
1019 memcpy(kaddr
+ *pg_offset
, buf
+ buf_offset
, bytes
);
1020 kunmap_atomic(kaddr
);
1021 flush_dcache_page(page_out
);
1023 *pg_offset
+= bytes
;
1024 buf_offset
+= bytes
;
1025 working_bytes
-= bytes
;
1026 current_buf_start
+= bytes
;
1028 /* check if we need to pick another page */
1029 if (*pg_offset
== PAGE_CACHE_SIZE
) {
1031 if (*pg_index
>= vcnt
)
1034 page_out
= bvec
[*pg_index
].bv_page
;
1036 start_byte
= page_offset(page_out
) - disk_start
;
1039 * make sure our new page is covered by this
1042 if (total_out
<= start_byte
)
1046 * the next page in the biovec might not be adjacent
1047 * to the last page, but it might still be found
1048 * inside this working buffer. bump our offset pointer
1050 if (total_out
> start_byte
&&
1051 current_buf_start
< start_byte
) {
1052 buf_offset
= start_byte
- buf_start
;
1053 working_bytes
= total_out
- start_byte
;
1054 current_buf_start
= buf_start
+ buf_offset
;
1063 * When uncompressing data, we need to make sure and zero any parts of
1064 * the biovec that were not filled in by the decompression code. pg_index
1065 * and pg_offset indicate the last page and the last offset of that page
1066 * that have been filled in. This will zero everything remaining in the
1069 void btrfs_clear_biovec_end(struct bio_vec
*bvec
, int vcnt
,
1070 unsigned long pg_index
,
1071 unsigned long pg_offset
)
1073 while (pg_index
< vcnt
) {
1074 struct page
*page
= bvec
[pg_index
].bv_page
;
1075 unsigned long off
= bvec
[pg_index
].bv_offset
;
1076 unsigned long len
= bvec
[pg_index
].bv_len
;
1078 if (pg_offset
< off
)
1080 if (pg_offset
< off
+ len
) {
1081 unsigned long bytes
= off
+ len
- pg_offset
;
1084 kaddr
= kmap_atomic(page
);
1085 memset(kaddr
+ pg_offset
, 0, bytes
);
1086 kunmap_atomic(kaddr
);