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
;
115 /* if there are more bios still pending for this compressed
118 if (!refcount_dec_and_test(&cb
->pending_bios
))
122 ret
= check_compressed_csum(BTRFS_I(inode
), cb
,
123 (u64
)bio
->bi_iter
.bi_sector
<< 9);
127 /* ok, we're the last bio for this extent, lets start
130 ret
= btrfs_decompress_bio(cb
);
136 /* release the compressed pages */
138 for (index
= 0; index
< cb
->nr_pages
; index
++) {
139 page
= cb
->compressed_pages
[index
];
140 page
->mapping
= NULL
;
144 /* do io completion on the original bio */
146 bio_io_error(cb
->orig_bio
);
149 struct bio_vec
*bvec
;
152 * we have verified the checksum already, set page
153 * checked so the end_io handlers know about it
155 ASSERT(!bio_flagged(bio
, BIO_CLONED
));
156 bio_for_each_segment_all(bvec
, cb
->orig_bio
, i
)
157 SetPageChecked(bvec
->bv_page
);
159 bio_endio(cb
->orig_bio
);
162 /* finally free the cb struct */
163 kfree(cb
->compressed_pages
);
170 * Clear the writeback bits on all of the file
171 * pages for a compressed write
173 static noinline
void end_compressed_writeback(struct inode
*inode
,
174 const struct compressed_bio
*cb
)
176 unsigned long index
= cb
->start
>> PAGE_SHIFT
;
177 unsigned long end_index
= (cb
->start
+ cb
->len
- 1) >> PAGE_SHIFT
;
178 struct page
*pages
[16];
179 unsigned long nr_pages
= end_index
- index
+ 1;
184 mapping_set_error(inode
->i_mapping
, -EIO
);
186 while (nr_pages
> 0) {
187 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
189 nr_pages
, ARRAY_SIZE(pages
)), pages
);
195 for (i
= 0; i
< ret
; i
++) {
197 SetPageError(pages
[i
]);
198 end_page_writeback(pages
[i
]);
204 /* the inode may be gone now */
208 * do the cleanup once all the compressed pages hit the disk.
209 * This will clear writeback on the file pages and free the compressed
212 * This also calls the writeback end hooks for the file pages so that
213 * metadata and checksums can be updated in the file.
215 static void end_compressed_bio_write(struct bio
*bio
)
217 struct extent_io_tree
*tree
;
218 struct compressed_bio
*cb
= bio
->bi_private
;
226 /* if there are more bios still pending for this compressed
229 if (!refcount_dec_and_test(&cb
->pending_bios
))
232 /* ok, we're the last bio for this extent, step one is to
233 * call back into the FS and do all the end_io operations
236 tree
= &BTRFS_I(inode
)->io_tree
;
237 cb
->compressed_pages
[0]->mapping
= cb
->inode
->i_mapping
;
238 tree
->ops
->writepage_end_io_hook(cb
->compressed_pages
[0],
240 cb
->start
+ cb
->len
- 1,
242 bio
->bi_status
? 0 : 1);
243 cb
->compressed_pages
[0]->mapping
= NULL
;
245 end_compressed_writeback(inode
, cb
);
246 /* note, our inode could be gone now */
249 * release the compressed pages, these came from alloc_page and
250 * are not attached to the inode at all
253 for (index
= 0; index
< cb
->nr_pages
; index
++) {
254 page
= cb
->compressed_pages
[index
];
255 page
->mapping
= NULL
;
259 /* finally free the cb struct */
260 kfree(cb
->compressed_pages
);
267 * worker function to build and submit bios for previously compressed pages.
268 * The corresponding pages in the inode should be marked for writeback
269 * and the compressed pages should have a reference on them for dropping
270 * when the IO is complete.
272 * This also checksums the file bytes and gets things ready for
275 blk_status_t
btrfs_submit_compressed_write(struct inode
*inode
, u64 start
,
276 unsigned long len
, u64 disk_start
,
277 unsigned long compressed_len
,
278 struct page
**compressed_pages
,
279 unsigned long nr_pages
)
281 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
282 struct bio
*bio
= NULL
;
283 struct compressed_bio
*cb
;
284 unsigned long bytes_left
;
285 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
288 u64 first_byte
= disk_start
;
289 struct block_device
*bdev
;
291 int skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
293 WARN_ON(start
& ((u64
)PAGE_SIZE
- 1));
294 cb
= kmalloc(compressed_bio_size(fs_info
, compressed_len
), GFP_NOFS
);
296 return BLK_STS_RESOURCE
;
297 refcount_set(&cb
->pending_bios
, 0);
303 cb
->compressed_pages
= compressed_pages
;
304 cb
->compressed_len
= compressed_len
;
306 cb
->nr_pages
= nr_pages
;
308 bdev
= fs_info
->fs_devices
->latest_bdev
;
310 bio
= btrfs_bio_alloc(bdev
, first_byte
);
311 bio_set_op_attrs(bio
, REQ_OP_WRITE
, 0);
312 bio
->bi_private
= cb
;
313 bio
->bi_end_io
= end_compressed_bio_write
;
314 refcount_set(&cb
->pending_bios
, 1);
316 /* create and submit bios for the compressed pages */
317 bytes_left
= compressed_len
;
318 for (pg_index
= 0; pg_index
< cb
->nr_pages
; pg_index
++) {
321 page
= compressed_pages
[pg_index
];
322 page
->mapping
= inode
->i_mapping
;
323 if (bio
->bi_iter
.bi_size
)
324 submit
= io_tree
->ops
->merge_bio_hook(page
, 0,
328 page
->mapping
= NULL
;
329 if (submit
|| bio_add_page(bio
, page
, PAGE_SIZE
, 0) <
334 * inc the count before we submit the bio so
335 * we know the end IO handler won't happen before
336 * we inc the count. Otherwise, the cb might get
337 * freed before we're done setting it up
339 refcount_inc(&cb
->pending_bios
);
340 ret
= btrfs_bio_wq_end_io(fs_info
, bio
,
341 BTRFS_WQ_ENDIO_DATA
);
342 BUG_ON(ret
); /* -ENOMEM */
345 ret
= btrfs_csum_one_bio(inode
, bio
, start
, 1);
346 BUG_ON(ret
); /* -ENOMEM */
349 ret
= btrfs_map_bio(fs_info
, bio
, 0, 1);
351 bio
->bi_status
= ret
;
357 bio
= btrfs_bio_alloc(bdev
, first_byte
);
358 bio_set_op_attrs(bio
, REQ_OP_WRITE
, 0);
359 bio
->bi_private
= cb
;
360 bio
->bi_end_io
= end_compressed_bio_write
;
361 bio_add_page(bio
, page
, PAGE_SIZE
, 0);
363 if (bytes_left
< PAGE_SIZE
) {
365 "bytes left %lu compress len %lu nr %lu",
366 bytes_left
, cb
->compressed_len
, cb
->nr_pages
);
368 bytes_left
-= PAGE_SIZE
;
369 first_byte
+= PAGE_SIZE
;
374 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, BTRFS_WQ_ENDIO_DATA
);
375 BUG_ON(ret
); /* -ENOMEM */
378 ret
= btrfs_csum_one_bio(inode
, bio
, start
, 1);
379 BUG_ON(ret
); /* -ENOMEM */
382 ret
= btrfs_map_bio(fs_info
, bio
, 0, 1);
384 bio
->bi_status
= ret
;
392 static u64
bio_end_offset(struct bio
*bio
)
394 struct bio_vec
*last
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
396 return page_offset(last
->bv_page
) + last
->bv_len
+ last
->bv_offset
;
399 static noinline
int add_ra_bio_pages(struct inode
*inode
,
401 struct compressed_bio
*cb
)
403 unsigned long end_index
;
404 unsigned long pg_index
;
406 u64 isize
= i_size_read(inode
);
409 unsigned long nr_pages
= 0;
410 struct extent_map
*em
;
411 struct address_space
*mapping
= inode
->i_mapping
;
412 struct extent_map_tree
*em_tree
;
413 struct extent_io_tree
*tree
;
417 last_offset
= bio_end_offset(cb
->orig_bio
);
418 em_tree
= &BTRFS_I(inode
)->extent_tree
;
419 tree
= &BTRFS_I(inode
)->io_tree
;
424 end_index
= (i_size_read(inode
) - 1) >> PAGE_SHIFT
;
426 while (last_offset
< compressed_end
) {
427 pg_index
= last_offset
>> PAGE_SHIFT
;
429 if (pg_index
> end_index
)
433 page
= radix_tree_lookup(&mapping
->page_tree
, pg_index
);
435 if (page
&& !radix_tree_exceptional_entry(page
)) {
442 page
= __page_cache_alloc(mapping_gfp_constraint(mapping
,
447 if (add_to_page_cache_lru(page
, mapping
, pg_index
, GFP_NOFS
)) {
452 end
= last_offset
+ PAGE_SIZE
- 1;
454 * at this point, we have a locked page in the page cache
455 * for these bytes in the file. But, we have to make
456 * sure they map to this compressed extent on disk.
458 set_page_extent_mapped(page
);
459 lock_extent(tree
, last_offset
, end
);
460 read_lock(&em_tree
->lock
);
461 em
= lookup_extent_mapping(em_tree
, last_offset
,
463 read_unlock(&em_tree
->lock
);
465 if (!em
|| last_offset
< em
->start
||
466 (last_offset
+ PAGE_SIZE
> extent_map_end(em
)) ||
467 (em
->block_start
>> 9) != cb
->orig_bio
->bi_iter
.bi_sector
) {
469 unlock_extent(tree
, last_offset
, end
);
476 if (page
->index
== end_index
) {
478 size_t zero_offset
= isize
& (PAGE_SIZE
- 1);
482 zeros
= PAGE_SIZE
- zero_offset
;
483 userpage
= kmap_atomic(page
);
484 memset(userpage
+ zero_offset
, 0, zeros
);
485 flush_dcache_page(page
);
486 kunmap_atomic(userpage
);
490 ret
= bio_add_page(cb
->orig_bio
, page
,
493 if (ret
== PAGE_SIZE
) {
497 unlock_extent(tree
, last_offset
, end
);
503 last_offset
+= PAGE_SIZE
;
509 * for a compressed read, the bio we get passed has all the inode pages
510 * in it. We don't actually do IO on those pages but allocate new ones
511 * to hold the compressed pages on disk.
513 * bio->bi_iter.bi_sector points to the compressed extent on disk
514 * bio->bi_io_vec points to all of the inode pages
516 * After the compressed pages are read, we copy the bytes into the
517 * bio we were passed and then call the bio end_io calls
519 blk_status_t
btrfs_submit_compressed_read(struct inode
*inode
, struct bio
*bio
,
520 int mirror_num
, unsigned long bio_flags
)
522 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
523 struct extent_io_tree
*tree
;
524 struct extent_map_tree
*em_tree
;
525 struct compressed_bio
*cb
;
526 unsigned long compressed_len
;
527 unsigned long nr_pages
;
528 unsigned long pg_index
;
530 struct block_device
*bdev
;
531 struct bio
*comp_bio
;
532 u64 cur_disk_byte
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
535 struct extent_map
*em
;
536 blk_status_t ret
= BLK_STS_RESOURCE
;
540 tree
= &BTRFS_I(inode
)->io_tree
;
541 em_tree
= &BTRFS_I(inode
)->extent_tree
;
543 /* we need the actual starting offset of this extent in the file */
544 read_lock(&em_tree
->lock
);
545 em
= lookup_extent_mapping(em_tree
,
546 page_offset(bio
->bi_io_vec
->bv_page
),
548 read_unlock(&em_tree
->lock
);
550 return BLK_STS_IOERR
;
552 compressed_len
= em
->block_len
;
553 cb
= kmalloc(compressed_bio_size(fs_info
, compressed_len
), GFP_NOFS
);
557 refcount_set(&cb
->pending_bios
, 0);
560 cb
->mirror_num
= mirror_num
;
563 cb
->start
= em
->orig_start
;
565 em_start
= em
->start
;
570 cb
->len
= bio
->bi_iter
.bi_size
;
571 cb
->compressed_len
= compressed_len
;
572 cb
->compress_type
= extent_compress_type(bio_flags
);
575 nr_pages
= DIV_ROUND_UP(compressed_len
, PAGE_SIZE
);
576 cb
->compressed_pages
= kcalloc(nr_pages
, sizeof(struct page
*),
578 if (!cb
->compressed_pages
)
581 bdev
= fs_info
->fs_devices
->latest_bdev
;
583 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
584 cb
->compressed_pages
[pg_index
] = alloc_page(GFP_NOFS
|
586 if (!cb
->compressed_pages
[pg_index
]) {
587 faili
= pg_index
- 1;
588 ret
= BLK_STS_RESOURCE
;
592 faili
= nr_pages
- 1;
593 cb
->nr_pages
= nr_pages
;
595 add_ra_bio_pages(inode
, em_start
+ em_len
, cb
);
597 /* include any pages we added in add_ra-bio_pages */
598 cb
->len
= bio
->bi_iter
.bi_size
;
600 comp_bio
= btrfs_bio_alloc(bdev
, cur_disk_byte
);
601 bio_set_op_attrs (comp_bio
, REQ_OP_READ
, 0);
602 comp_bio
->bi_private
= cb
;
603 comp_bio
->bi_end_io
= end_compressed_bio_read
;
604 refcount_set(&cb
->pending_bios
, 1);
606 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
609 page
= cb
->compressed_pages
[pg_index
];
610 page
->mapping
= inode
->i_mapping
;
611 page
->index
= em_start
>> PAGE_SHIFT
;
613 if (comp_bio
->bi_iter
.bi_size
)
614 submit
= tree
->ops
->merge_bio_hook(page
, 0,
618 page
->mapping
= NULL
;
619 if (submit
|| bio_add_page(comp_bio
, page
, PAGE_SIZE
, 0) <
623 ret
= btrfs_bio_wq_end_io(fs_info
, comp_bio
,
624 BTRFS_WQ_ENDIO_DATA
);
625 BUG_ON(ret
); /* -ENOMEM */
628 * inc the count before we submit the bio so
629 * we know the end IO handler won't happen before
630 * we inc the count. Otherwise, the cb might get
631 * freed before we're done setting it up
633 refcount_inc(&cb
->pending_bios
);
635 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
636 ret
= btrfs_lookup_bio_sums(inode
, comp_bio
,
638 BUG_ON(ret
); /* -ENOMEM */
640 sums
+= DIV_ROUND_UP(comp_bio
->bi_iter
.bi_size
,
641 fs_info
->sectorsize
);
643 ret
= btrfs_map_bio(fs_info
, comp_bio
, mirror_num
, 0);
645 comp_bio
->bi_status
= ret
;
651 comp_bio
= btrfs_bio_alloc(bdev
, cur_disk_byte
);
652 bio_set_op_attrs(comp_bio
, REQ_OP_READ
, 0);
653 comp_bio
->bi_private
= cb
;
654 comp_bio
->bi_end_io
= end_compressed_bio_read
;
656 bio_add_page(comp_bio
, page
, PAGE_SIZE
, 0);
658 cur_disk_byte
+= PAGE_SIZE
;
662 ret
= btrfs_bio_wq_end_io(fs_info
, comp_bio
, BTRFS_WQ_ENDIO_DATA
);
663 BUG_ON(ret
); /* -ENOMEM */
665 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
666 ret
= btrfs_lookup_bio_sums(inode
, comp_bio
, sums
);
667 BUG_ON(ret
); /* -ENOMEM */
670 ret
= btrfs_map_bio(fs_info
, comp_bio
, mirror_num
, 0);
672 comp_bio
->bi_status
= ret
;
681 __free_page(cb
->compressed_pages
[faili
]);
685 kfree(cb
->compressed_pages
);
694 struct list_head idle_ws
;
696 /* Number of free workspaces */
698 /* Total number of allocated workspaces */
700 /* Waiters for a free workspace */
701 wait_queue_head_t ws_wait
;
702 } btrfs_comp_ws
[BTRFS_COMPRESS_TYPES
];
704 static const struct btrfs_compress_op
* const btrfs_compress_op
[] = {
705 &btrfs_zlib_compress
,
707 &btrfs_zstd_compress
,
710 void __init
btrfs_init_compress(void)
714 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
715 struct list_head
*workspace
;
717 INIT_LIST_HEAD(&btrfs_comp_ws
[i
].idle_ws
);
718 spin_lock_init(&btrfs_comp_ws
[i
].ws_lock
);
719 atomic_set(&btrfs_comp_ws
[i
].total_ws
, 0);
720 init_waitqueue_head(&btrfs_comp_ws
[i
].ws_wait
);
723 * Preallocate one workspace for each compression type so
724 * we can guarantee forward progress in the worst case
726 workspace
= btrfs_compress_op
[i
]->alloc_workspace();
727 if (IS_ERR(workspace
)) {
728 pr_warn("BTRFS: cannot preallocate compression workspace, will try later\n");
730 atomic_set(&btrfs_comp_ws
[i
].total_ws
, 1);
731 btrfs_comp_ws
[i
].free_ws
= 1;
732 list_add(workspace
, &btrfs_comp_ws
[i
].idle_ws
);
738 * This finds an available workspace or allocates a new one.
739 * If it's not possible to allocate a new one, waits until there's one.
740 * Preallocation makes a forward progress guarantees and we do not return
743 static struct list_head
*find_workspace(int type
)
745 struct list_head
*workspace
;
746 int cpus
= num_online_cpus();
750 struct list_head
*idle_ws
= &btrfs_comp_ws
[idx
].idle_ws
;
751 spinlock_t
*ws_lock
= &btrfs_comp_ws
[idx
].ws_lock
;
752 atomic_t
*total_ws
= &btrfs_comp_ws
[idx
].total_ws
;
753 wait_queue_head_t
*ws_wait
= &btrfs_comp_ws
[idx
].ws_wait
;
754 int *free_ws
= &btrfs_comp_ws
[idx
].free_ws
;
757 if (!list_empty(idle_ws
)) {
758 workspace
= idle_ws
->next
;
761 spin_unlock(ws_lock
);
765 if (atomic_read(total_ws
) > cpus
) {
768 spin_unlock(ws_lock
);
769 prepare_to_wait(ws_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
770 if (atomic_read(total_ws
) > cpus
&& !*free_ws
)
772 finish_wait(ws_wait
, &wait
);
775 atomic_inc(total_ws
);
776 spin_unlock(ws_lock
);
779 * Allocation helpers call vmalloc that can't use GFP_NOFS, so we have
780 * to turn it off here because we might get called from the restricted
781 * context of btrfs_compress_bio/btrfs_compress_pages
783 nofs_flag
= memalloc_nofs_save();
784 workspace
= btrfs_compress_op
[idx
]->alloc_workspace();
785 memalloc_nofs_restore(nofs_flag
);
787 if (IS_ERR(workspace
)) {
788 atomic_dec(total_ws
);
792 * Do not return the error but go back to waiting. There's a
793 * workspace preallocated for each type and the compression
794 * time is bounded so we get to a workspace eventually. This
795 * makes our caller's life easier.
797 * To prevent silent and low-probability deadlocks (when the
798 * initial preallocation fails), check if there are any
801 if (atomic_read(total_ws
) == 0) {
802 static DEFINE_RATELIMIT_STATE(_rs
,
803 /* once per minute */ 60 * HZ
,
806 if (__ratelimit(&_rs
)) {
807 pr_warn("BTRFS: no compression workspaces, low memory, retrying\n");
816 * put a workspace struct back on the list or free it if we have enough
817 * idle ones sitting around
819 static void free_workspace(int type
, struct list_head
*workspace
)
822 struct list_head
*idle_ws
= &btrfs_comp_ws
[idx
].idle_ws
;
823 spinlock_t
*ws_lock
= &btrfs_comp_ws
[idx
].ws_lock
;
824 atomic_t
*total_ws
= &btrfs_comp_ws
[idx
].total_ws
;
825 wait_queue_head_t
*ws_wait
= &btrfs_comp_ws
[idx
].ws_wait
;
826 int *free_ws
= &btrfs_comp_ws
[idx
].free_ws
;
829 if (*free_ws
<= num_online_cpus()) {
830 list_add(workspace
, idle_ws
);
832 spin_unlock(ws_lock
);
835 spin_unlock(ws_lock
);
837 btrfs_compress_op
[idx
]->free_workspace(workspace
);
838 atomic_dec(total_ws
);
841 * Make sure counter is updated before we wake up waiters.
844 if (waitqueue_active(ws_wait
))
849 * cleanup function for module exit
851 static void free_workspaces(void)
853 struct list_head
*workspace
;
856 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
857 while (!list_empty(&btrfs_comp_ws
[i
].idle_ws
)) {
858 workspace
= btrfs_comp_ws
[i
].idle_ws
.next
;
860 btrfs_compress_op
[i
]->free_workspace(workspace
);
861 atomic_dec(&btrfs_comp_ws
[i
].total_ws
);
867 * Given an address space and start and length, compress the bytes into @pages
868 * that are allocated on demand.
870 * @out_pages is an in/out parameter, holds maximum number of pages to allocate
871 * and returns number of actually allocated pages
873 * @total_in is used to return the number of bytes actually read. It
874 * may be smaller than the input length 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 an in/out parameter, must be set to the input length and will
879 * be also used to return the total number of compressed bytes
881 * @max_out tells us the max number of bytes that we're allowed to
884 int btrfs_compress_pages(int type
, struct address_space
*mapping
,
885 u64 start
, struct page
**pages
,
886 unsigned long *out_pages
,
887 unsigned long *total_in
,
888 unsigned long *total_out
)
890 struct list_head
*workspace
;
893 workspace
= find_workspace(type
);
895 ret
= btrfs_compress_op
[type
-1]->compress_pages(workspace
, mapping
,
898 total_in
, total_out
);
899 free_workspace(type
, workspace
);
904 * pages_in is an array of pages with compressed data.
906 * disk_start is the starting logical offset of this array in the file
908 * orig_bio contains the pages from the file that we want to decompress into
910 * srclen is the number of bytes in pages_in
912 * The basic idea is that we have a bio that was created by readpages.
913 * The pages in the bio are for the uncompressed data, and they may not
914 * be contiguous. They all correspond to the range of bytes covered by
915 * the compressed extent.
917 static int btrfs_decompress_bio(struct compressed_bio
*cb
)
919 struct list_head
*workspace
;
921 int type
= cb
->compress_type
;
923 workspace
= find_workspace(type
);
924 ret
= btrfs_compress_op
[type
- 1]->decompress_bio(workspace
, cb
);
925 free_workspace(type
, workspace
);
931 * a less complex decompression routine. Our compressed data fits in a
932 * single page, and we want to read a single page out of it.
933 * start_byte tells us the offset into the compressed data we're interested in
935 int btrfs_decompress(int type
, unsigned char *data_in
, struct page
*dest_page
,
936 unsigned long start_byte
, size_t srclen
, size_t destlen
)
938 struct list_head
*workspace
;
941 workspace
= find_workspace(type
);
943 ret
= btrfs_compress_op
[type
-1]->decompress(workspace
, data_in
,
944 dest_page
, start_byte
,
947 free_workspace(type
, workspace
);
951 void btrfs_exit_compress(void)
957 * Copy uncompressed data from working buffer to pages.
959 * buf_start is the byte offset we're of the start of our workspace buffer.
961 * total_out is the last byte of the buffer
963 int btrfs_decompress_buf2page(const char *buf
, unsigned long buf_start
,
964 unsigned long total_out
, u64 disk_start
,
967 unsigned long buf_offset
;
968 unsigned long current_buf_start
;
969 unsigned long start_byte
;
970 unsigned long prev_start_byte
;
971 unsigned long working_bytes
= total_out
- buf_start
;
974 struct bio_vec bvec
= bio_iter_iovec(bio
, bio
->bi_iter
);
977 * start byte is the first byte of the page we're currently
978 * copying into relative to the start of the compressed data.
980 start_byte
= page_offset(bvec
.bv_page
) - disk_start
;
982 /* we haven't yet hit data corresponding to this page */
983 if (total_out
<= start_byte
)
987 * the start of the data we care about is offset into
988 * the middle of our working buffer
990 if (total_out
> start_byte
&& buf_start
< start_byte
) {
991 buf_offset
= start_byte
- buf_start
;
992 working_bytes
-= buf_offset
;
996 current_buf_start
= buf_start
;
998 /* copy bytes from the working buffer into the pages */
999 while (working_bytes
> 0) {
1000 bytes
= min_t(unsigned long, bvec
.bv_len
,
1001 PAGE_SIZE
- buf_offset
);
1002 bytes
= min(bytes
, working_bytes
);
1004 kaddr
= kmap_atomic(bvec
.bv_page
);
1005 memcpy(kaddr
+ bvec
.bv_offset
, buf
+ buf_offset
, bytes
);
1006 kunmap_atomic(kaddr
);
1007 flush_dcache_page(bvec
.bv_page
);
1009 buf_offset
+= bytes
;
1010 working_bytes
-= bytes
;
1011 current_buf_start
+= bytes
;
1013 /* check if we need to pick another page */
1014 bio_advance(bio
, bytes
);
1015 if (!bio
->bi_iter
.bi_size
)
1017 bvec
= bio_iter_iovec(bio
, bio
->bi_iter
);
1018 prev_start_byte
= start_byte
;
1019 start_byte
= page_offset(bvec
.bv_page
) - disk_start
;
1022 * We need to make sure we're only adjusting
1023 * our offset into compression working buffer when
1024 * we're switching pages. Otherwise we can incorrectly
1025 * keep copying when we were actually done.
1027 if (start_byte
!= prev_start_byte
) {
1029 * make sure our new page is covered by this
1032 if (total_out
<= start_byte
)
1036 * the next page in the biovec might not be adjacent
1037 * to the last page, but it might still be found
1038 * inside this working buffer. bump our offset pointer
1040 if (total_out
> start_byte
&&
1041 current_buf_start
< start_byte
) {
1042 buf_offset
= start_byte
- buf_start
;
1043 working_bytes
= total_out
- start_byte
;
1044 current_buf_start
= buf_start
+ buf_offset
;
1053 * Compression heuristic.
1055 * For now is's a naive and optimistic 'return true', we'll extend the logic to
1056 * quickly (compared to direct compression) detect data characteristics
1057 * (compressible/uncompressible) to avoid wasting CPU time on uncompressible
1060 * The following types of analysis can be performed:
1061 * - detect mostly zero data
1062 * - detect data with low "byte set" size (text, etc)
1063 * - detect data with low/high "core byte" set
1065 * Return non-zero if the compression should be done, 0 otherwise.
1067 int btrfs_compress_heuristic(struct inode
*inode
, u64 start
, u64 end
)
1069 u64 index
= start
>> PAGE_SHIFT
;
1070 u64 end_index
= end
>> PAGE_SHIFT
;
1074 while (index
<= end_index
) {
1075 page
= find_get_page(inode
->i_mapping
, index
);