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>
35 #include <linux/sched/mm.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 static int btrfs_decompress_bio(struct compressed_bio
*cb
);
48 static inline int compressed_bio_size(struct btrfs_fs_info
*fs_info
,
49 unsigned long disk_size
)
51 u16 csum_size
= btrfs_super_csum_size(fs_info
->super_copy
);
53 return sizeof(struct compressed_bio
) +
54 (DIV_ROUND_UP(disk_size
, fs_info
->sectorsize
)) * csum_size
;
57 static int check_compressed_csum(struct btrfs_inode
*inode
,
58 struct compressed_bio
*cb
,
66 u32
*cb_sum
= &cb
->sums
;
68 if (inode
->flags
& BTRFS_INODE_NODATASUM
)
71 for (i
= 0; i
< cb
->nr_pages
; i
++) {
72 page
= cb
->compressed_pages
[i
];
75 kaddr
= kmap_atomic(page
);
76 csum
= btrfs_csum_data(kaddr
, csum
, PAGE_SIZE
);
77 btrfs_csum_final(csum
, (u8
*)&csum
);
80 if (csum
!= *cb_sum
) {
81 btrfs_print_data_csum_error(inode
, disk_start
, csum
,
82 *cb_sum
, cb
->mirror_num
);
94 /* when we finish reading compressed pages from the disk, we
95 * decompress them and then run the bio end_io routines on the
96 * decompressed pages (in the inode address space).
98 * This allows the checksumming and other IO error handling routines
101 * The compressed pages are freed here, and it must be run
104 static void end_compressed_bio_read(struct bio
*bio
)
106 struct compressed_bio
*cb
= bio
->bi_private
;
110 unsigned int mirror
= btrfs_io_bio(bio
)->mirror_num
;
116 /* if there are more bios still pending for this compressed
119 if (!refcount_dec_and_test(&cb
->pending_bios
))
123 * Record the correct mirror_num in cb->orig_bio so that
124 * read-repair can work properly.
126 ASSERT(btrfs_io_bio(cb
->orig_bio
));
127 btrfs_io_bio(cb
->orig_bio
)->mirror_num
= mirror
;
128 cb
->mirror_num
= mirror
;
131 * Some IO in this cb have failed, just skip checksum as there
132 * is no way it could be correct.
138 ret
= check_compressed_csum(BTRFS_I(inode
), cb
,
139 (u64
)bio
->bi_iter
.bi_sector
<< 9);
143 /* ok, we're the last bio for this extent, lets start
146 ret
= btrfs_decompress_bio(cb
);
152 /* release the compressed pages */
154 for (index
= 0; index
< cb
->nr_pages
; index
++) {
155 page
= cb
->compressed_pages
[index
];
156 page
->mapping
= NULL
;
160 /* do io completion on the original bio */
162 bio_io_error(cb
->orig_bio
);
165 struct bio_vec
*bvec
;
168 * we have verified the checksum already, set page
169 * checked so the end_io handlers know about it
171 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
172 bio_for_each_segment_all(bvec
, cb
->orig_bio
, i
)
173 SetPageChecked(bvec
->bv_page
);
175 bio_endio(cb
->orig_bio
);
178 /* finally free the cb struct */
179 kfree(cb
->compressed_pages
);
186 * Clear the writeback bits on all of the file
187 * pages for a compressed write
189 static noinline
void end_compressed_writeback(struct inode
*inode
,
190 const struct compressed_bio
*cb
)
192 unsigned long index
= cb
->start
>> PAGE_SHIFT
;
193 unsigned long end_index
= (cb
->start
+ cb
->len
- 1) >> PAGE_SHIFT
;
194 struct page
*pages
[16];
195 unsigned long nr_pages
= end_index
- index
+ 1;
200 mapping_set_error(inode
->i_mapping
, -EIO
);
202 while (nr_pages
> 0) {
203 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
205 nr_pages
, ARRAY_SIZE(pages
)), pages
);
211 for (i
= 0; i
< ret
; i
++) {
213 SetPageError(pages
[i
]);
214 end_page_writeback(pages
[i
]);
220 /* the inode may be gone now */
224 * do the cleanup once all the compressed pages hit the disk.
225 * This will clear writeback on the file pages and free the compressed
228 * This also calls the writeback end hooks for the file pages so that
229 * metadata and checksums can be updated in the file.
231 static void end_compressed_bio_write(struct bio
*bio
)
233 struct extent_io_tree
*tree
;
234 struct compressed_bio
*cb
= bio
->bi_private
;
242 /* if there are more bios still pending for this compressed
245 if (!refcount_dec_and_test(&cb
->pending_bios
))
248 /* ok, we're the last bio for this extent, step one is to
249 * call back into the FS and do all the end_io operations
252 tree
= &BTRFS_I(inode
)->io_tree
;
253 cb
->compressed_pages
[0]->mapping
= cb
->inode
->i_mapping
;
254 tree
->ops
->writepage_end_io_hook(cb
->compressed_pages
[0],
256 cb
->start
+ cb
->len
- 1,
259 BLK_STS_OK
: BLK_STS_NOTSUPP
);
260 cb
->compressed_pages
[0]->mapping
= NULL
;
262 end_compressed_writeback(inode
, cb
);
263 /* note, our inode could be gone now */
266 * release the compressed pages, these came from alloc_page and
267 * are not attached to the inode at all
270 for (index
= 0; index
< cb
->nr_pages
; index
++) {
271 page
= cb
->compressed_pages
[index
];
272 page
->mapping
= NULL
;
276 /* finally free the cb struct */
277 kfree(cb
->compressed_pages
);
284 * worker function to build and submit bios for previously compressed pages.
285 * The corresponding pages in the inode should be marked for writeback
286 * and the compressed pages should have a reference on them for dropping
287 * when the IO is complete.
289 * This also checksums the file bytes and gets things ready for
292 blk_status_t
btrfs_submit_compressed_write(struct inode
*inode
, u64 start
,
293 unsigned long len
, u64 disk_start
,
294 unsigned long compressed_len
,
295 struct page
**compressed_pages
,
296 unsigned long nr_pages
)
298 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
299 struct bio
*bio
= NULL
;
300 struct compressed_bio
*cb
;
301 unsigned long bytes_left
;
302 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
305 u64 first_byte
= disk_start
;
306 struct block_device
*bdev
;
308 int skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
310 WARN_ON(start
& ((u64
)PAGE_SIZE
- 1));
311 cb
= kmalloc(compressed_bio_size(fs_info
, compressed_len
), GFP_NOFS
);
313 return BLK_STS_RESOURCE
;
314 refcount_set(&cb
->pending_bios
, 0);
320 cb
->compressed_pages
= compressed_pages
;
321 cb
->compressed_len
= compressed_len
;
323 cb
->nr_pages
= nr_pages
;
325 bdev
= fs_info
->fs_devices
->latest_bdev
;
327 bio
= btrfs_bio_alloc(bdev
, first_byte
);
328 bio_set_op_attrs(bio
, REQ_OP_WRITE
, 0);
329 bio
->bi_private
= cb
;
330 bio
->bi_end_io
= end_compressed_bio_write
;
331 refcount_set(&cb
->pending_bios
, 1);
333 /* create and submit bios for the compressed pages */
334 bytes_left
= compressed_len
;
335 for (pg_index
= 0; pg_index
< cb
->nr_pages
; pg_index
++) {
338 page
= compressed_pages
[pg_index
];
339 page
->mapping
= inode
->i_mapping
;
340 if (bio
->bi_iter
.bi_size
)
341 submit
= io_tree
->ops
->merge_bio_hook(page
, 0,
345 page
->mapping
= NULL
;
346 if (submit
|| bio_add_page(bio
, page
, PAGE_SIZE
, 0) <
351 * inc the count before we submit the bio so
352 * we know the end IO handler won't happen before
353 * we inc the count. Otherwise, the cb might get
354 * freed before we're done setting it up
356 refcount_inc(&cb
->pending_bios
);
357 ret
= btrfs_bio_wq_end_io(fs_info
, bio
,
358 BTRFS_WQ_ENDIO_DATA
);
359 BUG_ON(ret
); /* -ENOMEM */
362 ret
= btrfs_csum_one_bio(inode
, bio
, start
, 1);
363 BUG_ON(ret
); /* -ENOMEM */
366 ret
= btrfs_map_bio(fs_info
, bio
, 0, 1);
368 bio
->bi_status
= ret
;
374 bio
= btrfs_bio_alloc(bdev
, first_byte
);
375 bio_set_op_attrs(bio
, REQ_OP_WRITE
, 0);
376 bio
->bi_private
= cb
;
377 bio
->bi_end_io
= end_compressed_bio_write
;
378 bio_add_page(bio
, page
, PAGE_SIZE
, 0);
380 if (bytes_left
< PAGE_SIZE
) {
382 "bytes left %lu compress len %lu nr %lu",
383 bytes_left
, cb
->compressed_len
, cb
->nr_pages
);
385 bytes_left
-= PAGE_SIZE
;
386 first_byte
+= PAGE_SIZE
;
391 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, BTRFS_WQ_ENDIO_DATA
);
392 BUG_ON(ret
); /* -ENOMEM */
395 ret
= btrfs_csum_one_bio(inode
, bio
, start
, 1);
396 BUG_ON(ret
); /* -ENOMEM */
399 ret
= btrfs_map_bio(fs_info
, bio
, 0, 1);
401 bio
->bi_status
= ret
;
409 static u64
bio_end_offset(struct bio
*bio
)
411 struct bio_vec
*last
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
413 return page_offset(last
->bv_page
) + last
->bv_len
+ last
->bv_offset
;
416 static noinline
int add_ra_bio_pages(struct inode
*inode
,
418 struct compressed_bio
*cb
)
420 unsigned long end_index
;
421 unsigned long pg_index
;
423 u64 isize
= i_size_read(inode
);
426 unsigned long nr_pages
= 0;
427 struct extent_map
*em
;
428 struct address_space
*mapping
= inode
->i_mapping
;
429 struct extent_map_tree
*em_tree
;
430 struct extent_io_tree
*tree
;
434 last_offset
= bio_end_offset(cb
->orig_bio
);
435 em_tree
= &BTRFS_I(inode
)->extent_tree
;
436 tree
= &BTRFS_I(inode
)->io_tree
;
441 end_index
= (i_size_read(inode
) - 1) >> PAGE_SHIFT
;
443 while (last_offset
< compressed_end
) {
444 pg_index
= last_offset
>> PAGE_SHIFT
;
446 if (pg_index
> end_index
)
450 page
= radix_tree_lookup(&mapping
->page_tree
, pg_index
);
452 if (page
&& !radix_tree_exceptional_entry(page
)) {
459 page
= __page_cache_alloc(mapping_gfp_constraint(mapping
,
464 if (add_to_page_cache_lru(page
, mapping
, pg_index
, GFP_NOFS
)) {
469 end
= last_offset
+ PAGE_SIZE
- 1;
471 * at this point, we have a locked page in the page cache
472 * for these bytes in the file. But, we have to make
473 * sure they map to this compressed extent on disk.
475 set_page_extent_mapped(page
);
476 lock_extent(tree
, last_offset
, end
);
477 read_lock(&em_tree
->lock
);
478 em
= lookup_extent_mapping(em_tree
, last_offset
,
480 read_unlock(&em_tree
->lock
);
482 if (!em
|| last_offset
< em
->start
||
483 (last_offset
+ PAGE_SIZE
> extent_map_end(em
)) ||
484 (em
->block_start
>> 9) != cb
->orig_bio
->bi_iter
.bi_sector
) {
486 unlock_extent(tree
, last_offset
, end
);
493 if (page
->index
== end_index
) {
495 size_t zero_offset
= isize
& (PAGE_SIZE
- 1);
499 zeros
= PAGE_SIZE
- zero_offset
;
500 userpage
= kmap_atomic(page
);
501 memset(userpage
+ zero_offset
, 0, zeros
);
502 flush_dcache_page(page
);
503 kunmap_atomic(userpage
);
507 ret
= bio_add_page(cb
->orig_bio
, page
,
510 if (ret
== PAGE_SIZE
) {
514 unlock_extent(tree
, last_offset
, end
);
520 last_offset
+= PAGE_SIZE
;
526 * for a compressed read, the bio we get passed has all the inode pages
527 * in it. We don't actually do IO on those pages but allocate new ones
528 * to hold the compressed pages on disk.
530 * bio->bi_iter.bi_sector points to the compressed extent on disk
531 * bio->bi_io_vec points to all of the inode pages
533 * After the compressed pages are read, we copy the bytes into the
534 * bio we were passed and then call the bio end_io calls
536 blk_status_t
btrfs_submit_compressed_read(struct inode
*inode
, struct bio
*bio
,
537 int mirror_num
, unsigned long bio_flags
)
539 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
540 struct extent_io_tree
*tree
;
541 struct extent_map_tree
*em_tree
;
542 struct compressed_bio
*cb
;
543 unsigned long compressed_len
;
544 unsigned long nr_pages
;
545 unsigned long pg_index
;
547 struct block_device
*bdev
;
548 struct bio
*comp_bio
;
549 u64 cur_disk_byte
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
552 struct extent_map
*em
;
553 blk_status_t ret
= BLK_STS_RESOURCE
;
557 tree
= &BTRFS_I(inode
)->io_tree
;
558 em_tree
= &BTRFS_I(inode
)->extent_tree
;
560 /* we need the actual starting offset of this extent in the file */
561 read_lock(&em_tree
->lock
);
562 em
= lookup_extent_mapping(em_tree
,
563 page_offset(bio
->bi_io_vec
->bv_page
),
565 read_unlock(&em_tree
->lock
);
567 return BLK_STS_IOERR
;
569 compressed_len
= em
->block_len
;
570 cb
= kmalloc(compressed_bio_size(fs_info
, compressed_len
), GFP_NOFS
);
574 refcount_set(&cb
->pending_bios
, 0);
577 cb
->mirror_num
= mirror_num
;
580 cb
->start
= em
->orig_start
;
582 em_start
= em
->start
;
587 cb
->len
= bio
->bi_iter
.bi_size
;
588 cb
->compressed_len
= compressed_len
;
589 cb
->compress_type
= extent_compress_type(bio_flags
);
592 nr_pages
= DIV_ROUND_UP(compressed_len
, PAGE_SIZE
);
593 cb
->compressed_pages
= kcalloc(nr_pages
, sizeof(struct page
*),
595 if (!cb
->compressed_pages
)
598 bdev
= fs_info
->fs_devices
->latest_bdev
;
600 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
601 cb
->compressed_pages
[pg_index
] = alloc_page(GFP_NOFS
|
603 if (!cb
->compressed_pages
[pg_index
]) {
604 faili
= pg_index
- 1;
605 ret
= BLK_STS_RESOURCE
;
609 faili
= nr_pages
- 1;
610 cb
->nr_pages
= nr_pages
;
612 add_ra_bio_pages(inode
, em_start
+ em_len
, cb
);
614 /* include any pages we added in add_ra-bio_pages */
615 cb
->len
= bio
->bi_iter
.bi_size
;
617 comp_bio
= btrfs_bio_alloc(bdev
, cur_disk_byte
);
618 bio_set_op_attrs (comp_bio
, REQ_OP_READ
, 0);
619 comp_bio
->bi_private
= cb
;
620 comp_bio
->bi_end_io
= end_compressed_bio_read
;
621 refcount_set(&cb
->pending_bios
, 1);
623 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
626 page
= cb
->compressed_pages
[pg_index
];
627 page
->mapping
= inode
->i_mapping
;
628 page
->index
= em_start
>> PAGE_SHIFT
;
630 if (comp_bio
->bi_iter
.bi_size
)
631 submit
= tree
->ops
->merge_bio_hook(page
, 0,
635 page
->mapping
= NULL
;
636 if (submit
|| bio_add_page(comp_bio
, page
, PAGE_SIZE
, 0) <
640 ret
= btrfs_bio_wq_end_io(fs_info
, comp_bio
,
641 BTRFS_WQ_ENDIO_DATA
);
642 BUG_ON(ret
); /* -ENOMEM */
645 * inc the count before we submit the bio so
646 * we know the end IO handler won't happen before
647 * we inc the count. Otherwise, the cb might get
648 * freed before we're done setting it up
650 refcount_inc(&cb
->pending_bios
);
652 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
653 ret
= btrfs_lookup_bio_sums(inode
, comp_bio
,
655 BUG_ON(ret
); /* -ENOMEM */
657 sums
+= DIV_ROUND_UP(comp_bio
->bi_iter
.bi_size
,
658 fs_info
->sectorsize
);
660 ret
= btrfs_map_bio(fs_info
, comp_bio
, mirror_num
, 0);
662 comp_bio
->bi_status
= ret
;
668 comp_bio
= btrfs_bio_alloc(bdev
, cur_disk_byte
);
669 bio_set_op_attrs(comp_bio
, REQ_OP_READ
, 0);
670 comp_bio
->bi_private
= cb
;
671 comp_bio
->bi_end_io
= end_compressed_bio_read
;
673 bio_add_page(comp_bio
, page
, PAGE_SIZE
, 0);
675 cur_disk_byte
+= PAGE_SIZE
;
679 ret
= btrfs_bio_wq_end_io(fs_info
, comp_bio
, BTRFS_WQ_ENDIO_DATA
);
680 BUG_ON(ret
); /* -ENOMEM */
682 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
683 ret
= btrfs_lookup_bio_sums(inode
, comp_bio
, sums
);
684 BUG_ON(ret
); /* -ENOMEM */
687 ret
= btrfs_map_bio(fs_info
, comp_bio
, mirror_num
, 0);
689 comp_bio
->bi_status
= ret
;
698 __free_page(cb
->compressed_pages
[faili
]);
702 kfree(cb
->compressed_pages
);
711 struct list_head idle_ws
;
713 /* Number of free workspaces */
715 /* Total number of allocated workspaces */
717 /* Waiters for a free workspace */
718 wait_queue_head_t ws_wait
;
719 } btrfs_comp_ws
[BTRFS_COMPRESS_TYPES
];
721 static const struct btrfs_compress_op
* const btrfs_compress_op
[] = {
722 &btrfs_zlib_compress
,
724 &btrfs_zstd_compress
,
727 void __init
btrfs_init_compress(void)
731 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
732 struct list_head
*workspace
;
734 INIT_LIST_HEAD(&btrfs_comp_ws
[i
].idle_ws
);
735 spin_lock_init(&btrfs_comp_ws
[i
].ws_lock
);
736 atomic_set(&btrfs_comp_ws
[i
].total_ws
, 0);
737 init_waitqueue_head(&btrfs_comp_ws
[i
].ws_wait
);
740 * Preallocate one workspace for each compression type so
741 * we can guarantee forward progress in the worst case
743 workspace
= btrfs_compress_op
[i
]->alloc_workspace();
744 if (IS_ERR(workspace
)) {
745 pr_warn("BTRFS: cannot preallocate compression workspace, will try later\n");
747 atomic_set(&btrfs_comp_ws
[i
].total_ws
, 1);
748 btrfs_comp_ws
[i
].free_ws
= 1;
749 list_add(workspace
, &btrfs_comp_ws
[i
].idle_ws
);
755 * This finds an available workspace or allocates a new one.
756 * If it's not possible to allocate a new one, waits until there's one.
757 * Preallocation makes a forward progress guarantees and we do not return
760 static struct list_head
*find_workspace(int type
)
762 struct list_head
*workspace
;
763 int cpus
= num_online_cpus();
767 struct list_head
*idle_ws
= &btrfs_comp_ws
[idx
].idle_ws
;
768 spinlock_t
*ws_lock
= &btrfs_comp_ws
[idx
].ws_lock
;
769 atomic_t
*total_ws
= &btrfs_comp_ws
[idx
].total_ws
;
770 wait_queue_head_t
*ws_wait
= &btrfs_comp_ws
[idx
].ws_wait
;
771 int *free_ws
= &btrfs_comp_ws
[idx
].free_ws
;
774 if (!list_empty(idle_ws
)) {
775 workspace
= idle_ws
->next
;
778 spin_unlock(ws_lock
);
782 if (atomic_read(total_ws
) > cpus
) {
785 spin_unlock(ws_lock
);
786 prepare_to_wait(ws_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
787 if (atomic_read(total_ws
) > cpus
&& !*free_ws
)
789 finish_wait(ws_wait
, &wait
);
792 atomic_inc(total_ws
);
793 spin_unlock(ws_lock
);
796 * Allocation helpers call vmalloc that can't use GFP_NOFS, so we have
797 * to turn it off here because we might get called from the restricted
798 * context of btrfs_compress_bio/btrfs_compress_pages
800 nofs_flag
= memalloc_nofs_save();
801 workspace
= btrfs_compress_op
[idx
]->alloc_workspace();
802 memalloc_nofs_restore(nofs_flag
);
804 if (IS_ERR(workspace
)) {
805 atomic_dec(total_ws
);
809 * Do not return the error but go back to waiting. There's a
810 * workspace preallocated for each type and the compression
811 * time is bounded so we get to a workspace eventually. This
812 * makes our caller's life easier.
814 * To prevent silent and low-probability deadlocks (when the
815 * initial preallocation fails), check if there are any
818 if (atomic_read(total_ws
) == 0) {
819 static DEFINE_RATELIMIT_STATE(_rs
,
820 /* once per minute */ 60 * HZ
,
823 if (__ratelimit(&_rs
)) {
824 pr_warn("BTRFS: no compression workspaces, low memory, retrying\n");
833 * put a workspace struct back on the list or free it if we have enough
834 * idle ones sitting around
836 static void free_workspace(int type
, struct list_head
*workspace
)
839 struct list_head
*idle_ws
= &btrfs_comp_ws
[idx
].idle_ws
;
840 spinlock_t
*ws_lock
= &btrfs_comp_ws
[idx
].ws_lock
;
841 atomic_t
*total_ws
= &btrfs_comp_ws
[idx
].total_ws
;
842 wait_queue_head_t
*ws_wait
= &btrfs_comp_ws
[idx
].ws_wait
;
843 int *free_ws
= &btrfs_comp_ws
[idx
].free_ws
;
846 if (*free_ws
<= num_online_cpus()) {
847 list_add(workspace
, idle_ws
);
849 spin_unlock(ws_lock
);
852 spin_unlock(ws_lock
);
854 btrfs_compress_op
[idx
]->free_workspace(workspace
);
855 atomic_dec(total_ws
);
858 * Make sure counter is updated before we wake up waiters.
861 if (waitqueue_active(ws_wait
))
866 * cleanup function for module exit
868 static void free_workspaces(void)
870 struct list_head
*workspace
;
873 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
874 while (!list_empty(&btrfs_comp_ws
[i
].idle_ws
)) {
875 workspace
= btrfs_comp_ws
[i
].idle_ws
.next
;
877 btrfs_compress_op
[i
]->free_workspace(workspace
);
878 atomic_dec(&btrfs_comp_ws
[i
].total_ws
);
884 * Given an address space and start and length, compress the bytes into @pages
885 * that are allocated on demand.
887 * @type_level is encoded algorithm and level, where level 0 means whatever
888 * default the algorithm chooses and is opaque here;
889 * - compression algo are 0-3
890 * - the level are bits 4-7
892 * @out_pages is an in/out parameter, holds maximum number of pages to allocate
893 * and returns number of actually allocated pages
895 * @total_in is used to return the number of bytes actually read. It
896 * may be smaller than the input length if we had to exit early because we
897 * ran out of room in the pages array or because we cross the
900 * @total_out is an in/out parameter, must be set to the input length and will
901 * be also used to return the total number of compressed bytes
903 * @max_out tells us the max number of bytes that we're allowed to
906 int btrfs_compress_pages(unsigned int type_level
, struct address_space
*mapping
,
907 u64 start
, struct page
**pages
,
908 unsigned long *out_pages
,
909 unsigned long *total_in
,
910 unsigned long *total_out
)
912 struct list_head
*workspace
;
914 int type
= type_level
& 0xF;
916 workspace
= find_workspace(type
);
918 btrfs_compress_op
[type
- 1]->set_level(workspace
, type_level
);
919 ret
= btrfs_compress_op
[type
-1]->compress_pages(workspace
, mapping
,
922 total_in
, total_out
);
923 free_workspace(type
, workspace
);
928 * pages_in is an array of pages with compressed data.
930 * disk_start is the starting logical offset of this array in the file
932 * orig_bio contains the pages from the file that we want to decompress into
934 * srclen is the number of bytes in pages_in
936 * The basic idea is that we have a bio that was created by readpages.
937 * The pages in the bio are for the uncompressed data, and they may not
938 * be contiguous. They all correspond to the range of bytes covered by
939 * the compressed extent.
941 static int btrfs_decompress_bio(struct compressed_bio
*cb
)
943 struct list_head
*workspace
;
945 int type
= cb
->compress_type
;
947 workspace
= find_workspace(type
);
948 ret
= btrfs_compress_op
[type
- 1]->decompress_bio(workspace
, cb
);
949 free_workspace(type
, workspace
);
955 * a less complex decompression routine. Our compressed data fits in a
956 * single page, and we want to read a single page out of it.
957 * start_byte tells us the offset into the compressed data we're interested in
959 int btrfs_decompress(int type
, unsigned char *data_in
, struct page
*dest_page
,
960 unsigned long start_byte
, size_t srclen
, size_t destlen
)
962 struct list_head
*workspace
;
965 workspace
= find_workspace(type
);
967 ret
= btrfs_compress_op
[type
-1]->decompress(workspace
, data_in
,
968 dest_page
, start_byte
,
971 free_workspace(type
, workspace
);
975 void btrfs_exit_compress(void)
981 * Copy uncompressed data from working buffer to pages.
983 * buf_start is the byte offset we're of the start of our workspace buffer.
985 * total_out is the last byte of the buffer
987 int btrfs_decompress_buf2page(const char *buf
, unsigned long buf_start
,
988 unsigned long total_out
, u64 disk_start
,
991 unsigned long buf_offset
;
992 unsigned long current_buf_start
;
993 unsigned long start_byte
;
994 unsigned long prev_start_byte
;
995 unsigned long working_bytes
= total_out
- buf_start
;
998 struct bio_vec bvec
= bio_iter_iovec(bio
, bio
->bi_iter
);
1001 * start byte is the first byte of the page we're currently
1002 * copying into relative to the start of the compressed data.
1004 start_byte
= page_offset(bvec
.bv_page
) - disk_start
;
1006 /* we haven't yet hit data corresponding to this page */
1007 if (total_out
<= start_byte
)
1011 * the start of the data we care about is offset into
1012 * the middle of our working buffer
1014 if (total_out
> start_byte
&& buf_start
< start_byte
) {
1015 buf_offset
= start_byte
- buf_start
;
1016 working_bytes
-= buf_offset
;
1020 current_buf_start
= buf_start
;
1022 /* copy bytes from the working buffer into the pages */
1023 while (working_bytes
> 0) {
1024 bytes
= min_t(unsigned long, bvec
.bv_len
,
1025 PAGE_SIZE
- buf_offset
);
1026 bytes
= min(bytes
, working_bytes
);
1028 kaddr
= kmap_atomic(bvec
.bv_page
);
1029 memcpy(kaddr
+ bvec
.bv_offset
, buf
+ buf_offset
, bytes
);
1030 kunmap_atomic(kaddr
);
1031 flush_dcache_page(bvec
.bv_page
);
1033 buf_offset
+= bytes
;
1034 working_bytes
-= bytes
;
1035 current_buf_start
+= bytes
;
1037 /* check if we need to pick another page */
1038 bio_advance(bio
, bytes
);
1039 if (!bio
->bi_iter
.bi_size
)
1041 bvec
= bio_iter_iovec(bio
, bio
->bi_iter
);
1042 prev_start_byte
= start_byte
;
1043 start_byte
= page_offset(bvec
.bv_page
) - disk_start
;
1046 * We need to make sure we're only adjusting
1047 * our offset into compression working buffer when
1048 * we're switching pages. Otherwise we can incorrectly
1049 * keep copying when we were actually done.
1051 if (start_byte
!= prev_start_byte
) {
1053 * make sure our new page is covered by this
1056 if (total_out
<= start_byte
)
1060 * the next page in the biovec might not be adjacent
1061 * to the last page, but it might still be found
1062 * inside this working buffer. bump our offset pointer
1064 if (total_out
> start_byte
&&
1065 current_buf_start
< start_byte
) {
1066 buf_offset
= start_byte
- buf_start
;
1067 working_bytes
= total_out
- start_byte
;
1068 current_buf_start
= buf_start
+ buf_offset
;
1077 * Compression heuristic.
1079 * For now is's a naive and optimistic 'return true', we'll extend the logic to
1080 * quickly (compared to direct compression) detect data characteristics
1081 * (compressible/uncompressible) to avoid wasting CPU time on uncompressible
1084 * The following types of analysis can be performed:
1085 * - detect mostly zero data
1086 * - detect data with low "byte set" size (text, etc)
1087 * - detect data with low/high "core byte" set
1089 * Return non-zero if the compression should be done, 0 otherwise.
1091 int btrfs_compress_heuristic(struct inode
*inode
, u64 start
, u64 end
)
1093 u64 index
= start
>> PAGE_SHIFT
;
1094 u64 end_index
= end
>> PAGE_SHIFT
;
1098 while (index
<= end_index
) {
1099 page
= find_get_page(inode
->i_mapping
, index
);
1109 unsigned int btrfs_compress_str2level(const char *str
)
1111 if (strncmp(str
, "zlib", 4) != 0)
1114 /* Accepted form: zlib:1 up to zlib:9 and nothing left after the number */
1115 if (str
[4] == ':' && '1' <= str
[5] && str
[5] <= '9' && str
[6] == 0)
1116 return str
[5] - '0';