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 refcount_t pending_bios
;
49 /* the pages with the compressed data on them */
50 struct page
**compressed_pages
;
52 /* inode that owns this data */
55 /* starting offset in the inode for our pages */
58 /* number of bytes in the inode we're working on */
61 /* number of bytes on disk */
62 unsigned long compressed_len
;
64 /* the compression algorithm for this bio */
67 /* number of compressed pages in the array */
68 unsigned long nr_pages
;
74 /* for reads, this is the bio we are copying the data into */
78 * the start of a variable length array of checksums only
84 static int btrfs_decompress_bio(int type
, struct page
**pages_in
,
85 u64 disk_start
, struct bio
*orig_bio
,
88 static inline int compressed_bio_size(struct btrfs_fs_info
*fs_info
,
89 unsigned long disk_size
)
91 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
93 return sizeof(struct compressed_bio
) +
94 (DIV_ROUND_UP(disk_size
, fs_info
->sectorsize
)) * csum_size
;
97 static struct bio
*compressed_bio_alloc(struct block_device
*bdev
,
98 u64 first_byte
, gfp_t gfp_flags
)
100 return btrfs_bio_alloc(bdev
, first_byte
>> 9, BIO_MAX_PAGES
, gfp_flags
);
103 static int check_compressed_csum(struct btrfs_inode
*inode
,
104 struct compressed_bio
*cb
,
112 u32
*cb_sum
= &cb
->sums
;
114 if (inode
->flags
& BTRFS_INODE_NODATASUM
)
117 for (i
= 0; i
< cb
->nr_pages
; i
++) {
118 page
= cb
->compressed_pages
[i
];
121 kaddr
= kmap_atomic(page
);
122 csum
= btrfs_csum_data(kaddr
, csum
, PAGE_SIZE
);
123 btrfs_csum_final(csum
, (u8
*)&csum
);
124 kunmap_atomic(kaddr
);
126 if (csum
!= *cb_sum
) {
127 btrfs_print_data_csum_error(inode
, disk_start
, csum
,
128 *cb_sum
, cb
->mirror_num
);
140 /* when we finish reading compressed pages from the disk, we
141 * decompress them and then run the bio end_io routines on the
142 * decompressed pages (in the inode address space).
144 * This allows the checksumming and other IO error handling routines
147 * The compressed pages are freed here, and it must be run
150 static void end_compressed_bio_read(struct bio
*bio
)
152 struct compressed_bio
*cb
= bio
->bi_private
;
161 /* if there are more bios still pending for this compressed
164 if (!refcount_dec_and_test(&cb
->pending_bios
))
168 ret
= check_compressed_csum(BTRFS_I(inode
), cb
,
169 (u64
)bio
->bi_iter
.bi_sector
<< 9);
173 /* ok, we're the last bio for this extent, lets start
176 ret
= btrfs_decompress_bio(cb
->compress_type
,
177 cb
->compressed_pages
,
185 /* release the compressed pages */
187 for (index
= 0; index
< cb
->nr_pages
; index
++) {
188 page
= cb
->compressed_pages
[index
];
189 page
->mapping
= NULL
;
193 /* do io completion on the original bio */
195 bio_io_error(cb
->orig_bio
);
198 struct bio_vec
*bvec
;
201 * we have verified the checksum already, set page
202 * checked so the end_io handlers know about it
204 bio_for_each_segment_all(bvec
, cb
->orig_bio
, i
)
205 SetPageChecked(bvec
->bv_page
);
207 bio_endio(cb
->orig_bio
);
210 /* finally free the cb struct */
211 kfree(cb
->compressed_pages
);
218 * Clear the writeback bits on all of the file
219 * pages for a compressed write
221 static noinline
void end_compressed_writeback(struct inode
*inode
,
222 const struct compressed_bio
*cb
)
224 unsigned long index
= cb
->start
>> PAGE_SHIFT
;
225 unsigned long end_index
= (cb
->start
+ cb
->len
- 1) >> PAGE_SHIFT
;
226 struct page
*pages
[16];
227 unsigned long nr_pages
= end_index
- index
+ 1;
232 mapping_set_error(inode
->i_mapping
, -EIO
);
234 while (nr_pages
> 0) {
235 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
237 nr_pages
, ARRAY_SIZE(pages
)), pages
);
243 for (i
= 0; i
< ret
; i
++) {
245 SetPageError(pages
[i
]);
246 end_page_writeback(pages
[i
]);
252 /* the inode may be gone now */
256 * do the cleanup once all the compressed pages hit the disk.
257 * This will clear writeback on the file pages and free the compressed
260 * This also calls the writeback end hooks for the file pages so that
261 * metadata and checksums can be updated in the file.
263 static void end_compressed_bio_write(struct bio
*bio
)
265 struct extent_io_tree
*tree
;
266 struct compressed_bio
*cb
= bio
->bi_private
;
274 /* if there are more bios still pending for this compressed
277 if (!refcount_dec_and_test(&cb
->pending_bios
))
280 /* ok, we're the last bio for this extent, step one is to
281 * call back into the FS and do all the end_io operations
284 tree
= &BTRFS_I(inode
)->io_tree
;
285 cb
->compressed_pages
[0]->mapping
= cb
->inode
->i_mapping
;
286 tree
->ops
->writepage_end_io_hook(cb
->compressed_pages
[0],
288 cb
->start
+ cb
->len
- 1,
290 bio
->bi_error
? 0 : 1);
291 cb
->compressed_pages
[0]->mapping
= NULL
;
293 end_compressed_writeback(inode
, cb
);
294 /* note, our inode could be gone now */
297 * release the compressed pages, these came from alloc_page and
298 * are not attached to the inode at all
301 for (index
= 0; index
< cb
->nr_pages
; index
++) {
302 page
= cb
->compressed_pages
[index
];
303 page
->mapping
= NULL
;
307 /* finally free the cb struct */
308 kfree(cb
->compressed_pages
);
315 * worker function to build and submit bios for previously compressed pages.
316 * The corresponding pages in the inode should be marked for writeback
317 * and the compressed pages should have a reference on them for dropping
318 * when the IO is complete.
320 * This also checksums the file bytes and gets things ready for
323 int btrfs_submit_compressed_write(struct inode
*inode
, u64 start
,
324 unsigned long len
, u64 disk_start
,
325 unsigned long compressed_len
,
326 struct page
**compressed_pages
,
327 unsigned long nr_pages
)
329 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
330 struct bio
*bio
= NULL
;
331 struct compressed_bio
*cb
;
332 unsigned long bytes_left
;
333 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
336 u64 first_byte
= disk_start
;
337 struct block_device
*bdev
;
339 int skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
341 WARN_ON(start
& ((u64
)PAGE_SIZE
- 1));
342 cb
= kmalloc(compressed_bio_size(fs_info
, compressed_len
), GFP_NOFS
);
345 refcount_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
= fs_info
->fs_devices
->latest_bdev
;
358 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
363 bio_set_op_attrs(bio
, REQ_OP_WRITE
, 0);
364 bio
->bi_private
= cb
;
365 bio
->bi_end_io
= end_compressed_bio_write
;
366 refcount_set(&cb
->pending_bios
, 1);
368 /* create and submit bios for the compressed pages */
369 bytes_left
= compressed_len
;
370 for (pg_index
= 0; pg_index
< cb
->nr_pages
; pg_index
++) {
371 page
= compressed_pages
[pg_index
];
372 page
->mapping
= inode
->i_mapping
;
373 if (bio
->bi_iter
.bi_size
)
374 ret
= io_tree
->ops
->merge_bio_hook(page
, 0,
380 page
->mapping
= NULL
;
381 if (ret
|| bio_add_page(bio
, page
, PAGE_SIZE
, 0) <
386 * inc the count before we submit the bio so
387 * we know the end IO handler won't happen before
388 * we inc the count. Otherwise, the cb might get
389 * freed before we're done setting it up
391 refcount_inc(&cb
->pending_bios
);
392 ret
= btrfs_bio_wq_end_io(fs_info
, bio
,
393 BTRFS_WQ_ENDIO_DATA
);
394 BUG_ON(ret
); /* -ENOMEM */
397 ret
= btrfs_csum_one_bio(inode
, bio
, start
, 1);
398 BUG_ON(ret
); /* -ENOMEM */
401 ret
= btrfs_map_bio(fs_info
, bio
, 0, 1);
409 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
411 bio_set_op_attrs(bio
, REQ_OP_WRITE
, 0);
412 bio
->bi_private
= cb
;
413 bio
->bi_end_io
= end_compressed_bio_write
;
414 bio_add_page(bio
, page
, PAGE_SIZE
, 0);
416 if (bytes_left
< PAGE_SIZE
) {
418 "bytes left %lu compress len %lu nr %lu",
419 bytes_left
, cb
->compressed_len
, cb
->nr_pages
);
421 bytes_left
-= PAGE_SIZE
;
422 first_byte
+= PAGE_SIZE
;
427 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, BTRFS_WQ_ENDIO_DATA
);
428 BUG_ON(ret
); /* -ENOMEM */
431 ret
= btrfs_csum_one_bio(inode
, bio
, start
, 1);
432 BUG_ON(ret
); /* -ENOMEM */
435 ret
= btrfs_map_bio(fs_info
, bio
, 0, 1);
445 static u64
bio_end_offset(struct bio
*bio
)
447 struct bio_vec
*last
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
449 return page_offset(last
->bv_page
) + last
->bv_len
+ last
->bv_offset
;
452 static noinline
int add_ra_bio_pages(struct inode
*inode
,
454 struct compressed_bio
*cb
)
456 unsigned long end_index
;
457 unsigned long pg_index
;
459 u64 isize
= i_size_read(inode
);
462 unsigned long nr_pages
= 0;
463 struct extent_map
*em
;
464 struct address_space
*mapping
= inode
->i_mapping
;
465 struct extent_map_tree
*em_tree
;
466 struct extent_io_tree
*tree
;
470 last_offset
= bio_end_offset(cb
->orig_bio
);
471 em_tree
= &BTRFS_I(inode
)->extent_tree
;
472 tree
= &BTRFS_I(inode
)->io_tree
;
477 end_index
= (i_size_read(inode
) - 1) >> PAGE_SHIFT
;
479 while (last_offset
< compressed_end
) {
480 pg_index
= last_offset
>> PAGE_SHIFT
;
482 if (pg_index
> end_index
)
486 page
= radix_tree_lookup(&mapping
->page_tree
, pg_index
);
488 if (page
&& !radix_tree_exceptional_entry(page
)) {
495 page
= __page_cache_alloc(mapping_gfp_constraint(mapping
,
500 if (add_to_page_cache_lru(page
, mapping
, pg_index
, GFP_NOFS
)) {
505 end
= last_offset
+ PAGE_SIZE
- 1;
507 * at this point, we have a locked page in the page cache
508 * for these bytes in the file. But, we have to make
509 * sure they map to this compressed extent on disk.
511 set_page_extent_mapped(page
);
512 lock_extent(tree
, last_offset
, end
);
513 read_lock(&em_tree
->lock
);
514 em
= lookup_extent_mapping(em_tree
, last_offset
,
516 read_unlock(&em_tree
->lock
);
518 if (!em
|| last_offset
< em
->start
||
519 (last_offset
+ PAGE_SIZE
> extent_map_end(em
)) ||
520 (em
->block_start
>> 9) != cb
->orig_bio
->bi_iter
.bi_sector
) {
522 unlock_extent(tree
, last_offset
, end
);
529 if (page
->index
== end_index
) {
531 size_t zero_offset
= isize
& (PAGE_SIZE
- 1);
535 zeros
= PAGE_SIZE
- zero_offset
;
536 userpage
= kmap_atomic(page
);
537 memset(userpage
+ zero_offset
, 0, zeros
);
538 flush_dcache_page(page
);
539 kunmap_atomic(userpage
);
543 ret
= bio_add_page(cb
->orig_bio
, page
,
546 if (ret
== PAGE_SIZE
) {
550 unlock_extent(tree
, last_offset
, end
);
556 last_offset
+= PAGE_SIZE
;
562 * for a compressed read, the bio we get passed has all the inode pages
563 * in it. We don't actually do IO on those pages but allocate new ones
564 * to hold the compressed pages on disk.
566 * bio->bi_iter.bi_sector points to the compressed extent on disk
567 * bio->bi_io_vec points to all of the inode pages
569 * After the compressed pages are read, we copy the bytes into the
570 * bio we were passed and then call the bio end_io calls
572 int btrfs_submit_compressed_read(struct inode
*inode
, struct bio
*bio
,
573 int mirror_num
, unsigned long bio_flags
)
575 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
576 struct extent_io_tree
*tree
;
577 struct extent_map_tree
*em_tree
;
578 struct compressed_bio
*cb
;
579 unsigned long compressed_len
;
580 unsigned long nr_pages
;
581 unsigned long pg_index
;
583 struct block_device
*bdev
;
584 struct bio
*comp_bio
;
585 u64 cur_disk_byte
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
588 struct extent_map
*em
;
593 tree
= &BTRFS_I(inode
)->io_tree
;
594 em_tree
= &BTRFS_I(inode
)->extent_tree
;
596 /* we need the actual starting offset of this extent in the file */
597 read_lock(&em_tree
->lock
);
598 em
= lookup_extent_mapping(em_tree
,
599 page_offset(bio
->bi_io_vec
->bv_page
),
601 read_unlock(&em_tree
->lock
);
605 compressed_len
= em
->block_len
;
606 cb
= kmalloc(compressed_bio_size(fs_info
, compressed_len
), GFP_NOFS
);
610 refcount_set(&cb
->pending_bios
, 0);
613 cb
->mirror_num
= mirror_num
;
616 cb
->start
= em
->orig_start
;
618 em_start
= em
->start
;
623 cb
->len
= bio
->bi_iter
.bi_size
;
624 cb
->compressed_len
= compressed_len
;
625 cb
->compress_type
= extent_compress_type(bio_flags
);
628 nr_pages
= DIV_ROUND_UP(compressed_len
, PAGE_SIZE
);
629 cb
->compressed_pages
= kcalloc(nr_pages
, sizeof(struct page
*),
631 if (!cb
->compressed_pages
)
634 bdev
= fs_info
->fs_devices
->latest_bdev
;
636 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
637 cb
->compressed_pages
[pg_index
] = alloc_page(GFP_NOFS
|
639 if (!cb
->compressed_pages
[pg_index
]) {
640 faili
= pg_index
- 1;
645 faili
= nr_pages
- 1;
646 cb
->nr_pages
= nr_pages
;
648 add_ra_bio_pages(inode
, em_start
+ em_len
, cb
);
650 /* include any pages we added in add_ra-bio_pages */
651 cb
->len
= bio
->bi_iter
.bi_size
;
653 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
, GFP_NOFS
);
656 bio_set_op_attrs (comp_bio
, REQ_OP_READ
, 0);
657 comp_bio
->bi_private
= cb
;
658 comp_bio
->bi_end_io
= end_compressed_bio_read
;
659 refcount_set(&cb
->pending_bios
, 1);
661 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
662 page
= cb
->compressed_pages
[pg_index
];
663 page
->mapping
= inode
->i_mapping
;
664 page
->index
= em_start
>> PAGE_SHIFT
;
666 if (comp_bio
->bi_iter
.bi_size
)
667 ret
= tree
->ops
->merge_bio_hook(page
, 0,
673 page
->mapping
= NULL
;
674 if (ret
|| bio_add_page(comp_bio
, page
, PAGE_SIZE
, 0) <
678 ret
= btrfs_bio_wq_end_io(fs_info
, comp_bio
,
679 BTRFS_WQ_ENDIO_DATA
);
680 BUG_ON(ret
); /* -ENOMEM */
683 * inc the count before we submit the bio so
684 * we know the end IO handler won't happen before
685 * we inc the count. Otherwise, the cb might get
686 * freed before we're done setting it up
688 refcount_inc(&cb
->pending_bios
);
690 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
691 ret
= btrfs_lookup_bio_sums(inode
, comp_bio
,
693 BUG_ON(ret
); /* -ENOMEM */
695 sums
+= DIV_ROUND_UP(comp_bio
->bi_iter
.bi_size
,
696 fs_info
->sectorsize
);
698 ret
= btrfs_map_bio(fs_info
, comp_bio
, mirror_num
, 0);
700 comp_bio
->bi_error
= ret
;
706 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
,
709 bio_set_op_attrs(comp_bio
, REQ_OP_READ
, 0);
710 comp_bio
->bi_private
= cb
;
711 comp_bio
->bi_end_io
= end_compressed_bio_read
;
713 bio_add_page(comp_bio
, page
, PAGE_SIZE
, 0);
715 cur_disk_byte
+= PAGE_SIZE
;
719 ret
= btrfs_bio_wq_end_io(fs_info
, comp_bio
, BTRFS_WQ_ENDIO_DATA
);
720 BUG_ON(ret
); /* -ENOMEM */
722 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
723 ret
= btrfs_lookup_bio_sums(inode
, comp_bio
, sums
);
724 BUG_ON(ret
); /* -ENOMEM */
727 ret
= btrfs_map_bio(fs_info
, comp_bio
, mirror_num
, 0);
729 comp_bio
->bi_error
= ret
;
738 __free_page(cb
->compressed_pages
[faili
]);
742 kfree(cb
->compressed_pages
);
751 struct list_head idle_ws
;
753 /* Number of free workspaces */
755 /* Total number of allocated workspaces */
757 /* Waiters for a free workspace */
758 wait_queue_head_t ws_wait
;
759 } btrfs_comp_ws
[BTRFS_COMPRESS_TYPES
];
761 static const struct btrfs_compress_op
* const btrfs_compress_op
[] = {
762 &btrfs_zlib_compress
,
766 void __init
btrfs_init_compress(void)
770 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
771 struct list_head
*workspace
;
773 INIT_LIST_HEAD(&btrfs_comp_ws
[i
].idle_ws
);
774 spin_lock_init(&btrfs_comp_ws
[i
].ws_lock
);
775 atomic_set(&btrfs_comp_ws
[i
].total_ws
, 0);
776 init_waitqueue_head(&btrfs_comp_ws
[i
].ws_wait
);
779 * Preallocate one workspace for each compression type so
780 * we can guarantee forward progress in the worst case
782 workspace
= btrfs_compress_op
[i
]->alloc_workspace();
783 if (IS_ERR(workspace
)) {
784 pr_warn("BTRFS: cannot preallocate compression workspace, will try later\n");
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
)) {
854 pr_warn("BTRFS: no compression workspaces, low memory, retrying\n");
863 * put a workspace struct back on the list or free it if we have enough
864 * idle ones sitting around
866 static void free_workspace(int type
, struct list_head
*workspace
)
869 struct list_head
*idle_ws
= &btrfs_comp_ws
[idx
].idle_ws
;
870 spinlock_t
*ws_lock
= &btrfs_comp_ws
[idx
].ws_lock
;
871 atomic_t
*total_ws
= &btrfs_comp_ws
[idx
].total_ws
;
872 wait_queue_head_t
*ws_wait
= &btrfs_comp_ws
[idx
].ws_wait
;
873 int *free_ws
= &btrfs_comp_ws
[idx
].free_ws
;
876 if (*free_ws
< num_online_cpus()) {
877 list_add(workspace
, idle_ws
);
879 spin_unlock(ws_lock
);
882 spin_unlock(ws_lock
);
884 btrfs_compress_op
[idx
]->free_workspace(workspace
);
885 atomic_dec(total_ws
);
888 * Make sure counter is updated before we wake up waiters.
891 if (waitqueue_active(ws_wait
))
896 * cleanup function for module exit
898 static void free_workspaces(void)
900 struct list_head
*workspace
;
903 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
904 while (!list_empty(&btrfs_comp_ws
[i
].idle_ws
)) {
905 workspace
= btrfs_comp_ws
[i
].idle_ws
.next
;
907 btrfs_compress_op
[i
]->free_workspace(workspace
);
908 atomic_dec(&btrfs_comp_ws
[i
].total_ws
);
914 * Given an address space and start and length, compress the bytes into @pages
915 * that are allocated on demand.
917 * @out_pages is an in/out parameter, holds maximum number of pages to allocate
918 * and returns number of actually allocated pages
920 * @total_in is used to return the number of bytes actually read. It
921 * may be smaller than the input length if we had to exit early because we
922 * ran out of room in the pages array or because we cross the
925 * @total_out is an in/out parameter, must be set to the input length and will
926 * be also used to return the total number of compressed bytes
928 * @max_out tells us the max number of bytes that we're allowed to
931 int btrfs_compress_pages(int type
, struct address_space
*mapping
,
932 u64 start
, struct page
**pages
,
933 unsigned long *out_pages
,
934 unsigned long *total_in
,
935 unsigned long *total_out
)
937 struct list_head
*workspace
;
940 workspace
= find_workspace(type
);
942 ret
= btrfs_compress_op
[type
-1]->compress_pages(workspace
, mapping
,
945 total_in
, total_out
);
946 free_workspace(type
, workspace
);
951 * pages_in is an array of pages with compressed data.
953 * disk_start is the starting logical offset of this array in the file
955 * orig_bio contains the pages from the file that we want to decompress into
957 * srclen is the number of bytes in pages_in
959 * The basic idea is that we have a bio that was created by readpages.
960 * The pages in the bio are for the uncompressed data, and they may not
961 * be contiguous. They all correspond to the range of bytes covered by
962 * the compressed extent.
964 static int btrfs_decompress_bio(int type
, struct page
**pages_in
,
965 u64 disk_start
, struct bio
*orig_bio
,
968 struct list_head
*workspace
;
971 workspace
= find_workspace(type
);
973 ret
= btrfs_compress_op
[type
-1]->decompress_bio(workspace
, pages_in
,
974 disk_start
, orig_bio
,
976 free_workspace(type
, workspace
);
981 * a less complex decompression routine. Our compressed data fits in a
982 * single page, and we want to read a single page out of it.
983 * start_byte tells us the offset into the compressed data we're interested in
985 int btrfs_decompress(int type
, unsigned char *data_in
, struct page
*dest_page
,
986 unsigned long start_byte
, size_t srclen
, size_t destlen
)
988 struct list_head
*workspace
;
991 workspace
= find_workspace(type
);
993 ret
= btrfs_compress_op
[type
-1]->decompress(workspace
, data_in
,
994 dest_page
, start_byte
,
997 free_workspace(type
, workspace
);
1001 void btrfs_exit_compress(void)
1007 * Copy uncompressed data from working buffer to pages.
1009 * buf_start is the byte offset we're of the start of our workspace buffer.
1011 * total_out is the last byte of the buffer
1013 int btrfs_decompress_buf2page(const char *buf
, unsigned long buf_start
,
1014 unsigned long total_out
, u64 disk_start
,
1017 unsigned long buf_offset
;
1018 unsigned long current_buf_start
;
1019 unsigned long start_byte
;
1020 unsigned long prev_start_byte
;
1021 unsigned long working_bytes
= total_out
- buf_start
;
1022 unsigned long bytes
;
1024 struct bio_vec bvec
= bio_iter_iovec(bio
, bio
->bi_iter
);
1027 * start byte is the first byte of the page we're currently
1028 * copying into relative to the start of the compressed data.
1030 start_byte
= page_offset(bvec
.bv_page
) - disk_start
;
1032 /* we haven't yet hit data corresponding to this page */
1033 if (total_out
<= start_byte
)
1037 * the start of the data we care about is offset into
1038 * the middle of our working buffer
1040 if (total_out
> start_byte
&& buf_start
< start_byte
) {
1041 buf_offset
= start_byte
- buf_start
;
1042 working_bytes
-= buf_offset
;
1046 current_buf_start
= buf_start
;
1048 /* copy bytes from the working buffer into the pages */
1049 while (working_bytes
> 0) {
1050 bytes
= min_t(unsigned long, bvec
.bv_len
,
1051 PAGE_SIZE
- buf_offset
);
1052 bytes
= min(bytes
, working_bytes
);
1054 kaddr
= kmap_atomic(bvec
.bv_page
);
1055 memcpy(kaddr
+ bvec
.bv_offset
, buf
+ buf_offset
, bytes
);
1056 kunmap_atomic(kaddr
);
1057 flush_dcache_page(bvec
.bv_page
);
1059 buf_offset
+= bytes
;
1060 working_bytes
-= bytes
;
1061 current_buf_start
+= bytes
;
1063 /* check if we need to pick another page */
1064 bio_advance(bio
, bytes
);
1065 if (!bio
->bi_iter
.bi_size
)
1067 bvec
= bio_iter_iovec(bio
, bio
->bi_iter
);
1068 prev_start_byte
= start_byte
;
1069 start_byte
= page_offset(bvec
.bv_page
) - disk_start
;
1072 * We need to make sure we're only adjusting
1073 * our offset into compression working buffer when
1074 * we're switching pages. Otherwise we can incorrectly
1075 * keep copying when we were actually done.
1077 if (start_byte
!= prev_start_byte
) {
1079 * make sure our new page is covered by this
1082 if (total_out
<= start_byte
)
1086 * the next page in the biovec might not be adjacent
1087 * to the last page, but it might still be found
1088 * inside this working buffer. bump our offset pointer
1090 if (total_out
> start_byte
&&
1091 current_buf_start
< start_byte
) {
1092 buf_offset
= start_byte
- buf_start
;
1093 working_bytes
= total_out
- start_byte
;
1094 current_buf_start
= buf_start
+ buf_offset
;