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 bio_for_each_segment_all(bvec
, cb
->orig_bio
, i
)
156 SetPageChecked(bvec
->bv_page
);
158 bio_endio(cb
->orig_bio
);
161 /* finally free the cb struct */
162 kfree(cb
->compressed_pages
);
169 * Clear the writeback bits on all of the file
170 * pages for a compressed write
172 static noinline
void end_compressed_writeback(struct inode
*inode
,
173 const struct compressed_bio
*cb
)
175 unsigned long index
= cb
->start
>> PAGE_SHIFT
;
176 unsigned long end_index
= (cb
->start
+ cb
->len
- 1) >> PAGE_SHIFT
;
177 struct page
*pages
[16];
178 unsigned long nr_pages
= end_index
- index
+ 1;
183 mapping_set_error(inode
->i_mapping
, -EIO
);
185 while (nr_pages
> 0) {
186 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
188 nr_pages
, ARRAY_SIZE(pages
)), pages
);
194 for (i
= 0; i
< ret
; i
++) {
196 SetPageError(pages
[i
]);
197 end_page_writeback(pages
[i
]);
203 /* the inode may be gone now */
207 * do the cleanup once all the compressed pages hit the disk.
208 * This will clear writeback on the file pages and free the compressed
211 * This also calls the writeback end hooks for the file pages so that
212 * metadata and checksums can be updated in the file.
214 static void end_compressed_bio_write(struct bio
*bio
)
216 struct extent_io_tree
*tree
;
217 struct compressed_bio
*cb
= bio
->bi_private
;
225 /* if there are more bios still pending for this compressed
228 if (!refcount_dec_and_test(&cb
->pending_bios
))
231 /* ok, we're the last bio for this extent, step one is to
232 * call back into the FS and do all the end_io operations
235 tree
= &BTRFS_I(inode
)->io_tree
;
236 cb
->compressed_pages
[0]->mapping
= cb
->inode
->i_mapping
;
237 tree
->ops
->writepage_end_io_hook(cb
->compressed_pages
[0],
239 cb
->start
+ cb
->len
- 1,
241 bio
->bi_error
? 0 : 1);
242 cb
->compressed_pages
[0]->mapping
= NULL
;
244 end_compressed_writeback(inode
, cb
);
245 /* note, our inode could be gone now */
248 * release the compressed pages, these came from alloc_page and
249 * are not attached to the inode at all
252 for (index
= 0; index
< cb
->nr_pages
; index
++) {
253 page
= cb
->compressed_pages
[index
];
254 page
->mapping
= NULL
;
258 /* finally free the cb struct */
259 kfree(cb
->compressed_pages
);
266 * worker function to build and submit bios for previously compressed pages.
267 * The corresponding pages in the inode should be marked for writeback
268 * and the compressed pages should have a reference on them for dropping
269 * when the IO is complete.
271 * This also checksums the file bytes and gets things ready for
274 int btrfs_submit_compressed_write(struct inode
*inode
, u64 start
,
275 unsigned long len
, u64 disk_start
,
276 unsigned long compressed_len
,
277 struct page
**compressed_pages
,
278 unsigned long nr_pages
)
280 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
281 struct bio
*bio
= NULL
;
282 struct compressed_bio
*cb
;
283 unsigned long bytes_left
;
284 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
287 u64 first_byte
= disk_start
;
288 struct block_device
*bdev
;
290 int skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
292 WARN_ON(start
& ((u64
)PAGE_SIZE
- 1));
293 cb
= kmalloc(compressed_bio_size(fs_info
, compressed_len
), GFP_NOFS
);
296 refcount_set(&cb
->pending_bios
, 0);
302 cb
->compressed_pages
= compressed_pages
;
303 cb
->compressed_len
= compressed_len
;
305 cb
->nr_pages
= nr_pages
;
307 bdev
= fs_info
->fs_devices
->latest_bdev
;
309 bio
= btrfs_bio_alloc(bdev
, first_byte
);
310 bio_set_op_attrs(bio
, REQ_OP_WRITE
, 0);
311 bio
->bi_private
= cb
;
312 bio
->bi_end_io
= end_compressed_bio_write
;
313 refcount_set(&cb
->pending_bios
, 1);
315 /* create and submit bios for the compressed pages */
316 bytes_left
= compressed_len
;
317 for (pg_index
= 0; pg_index
< cb
->nr_pages
; pg_index
++) {
318 page
= compressed_pages
[pg_index
];
319 page
->mapping
= inode
->i_mapping
;
320 if (bio
->bi_iter
.bi_size
)
321 ret
= io_tree
->ops
->merge_bio_hook(page
, 0,
327 page
->mapping
= NULL
;
328 if (ret
|| bio_add_page(bio
, page
, PAGE_SIZE
, 0) <
333 * inc the count before we submit the bio so
334 * we know the end IO handler won't happen before
335 * we inc the count. Otherwise, the cb might get
336 * freed before we're done setting it up
338 refcount_inc(&cb
->pending_bios
);
339 ret
= btrfs_bio_wq_end_io(fs_info
, bio
,
340 BTRFS_WQ_ENDIO_DATA
);
341 BUG_ON(ret
); /* -ENOMEM */
344 ret
= btrfs_csum_one_bio(inode
, bio
, start
, 1);
345 BUG_ON(ret
); /* -ENOMEM */
348 ret
= btrfs_map_bio(fs_info
, bio
, 0, 1);
356 bio
= btrfs_bio_alloc(bdev
, first_byte
);
357 bio_set_op_attrs(bio
, REQ_OP_WRITE
, 0);
358 bio
->bi_private
= cb
;
359 bio
->bi_end_io
= end_compressed_bio_write
;
360 bio_add_page(bio
, page
, PAGE_SIZE
, 0);
362 if (bytes_left
< PAGE_SIZE
) {
364 "bytes left %lu compress len %lu nr %lu",
365 bytes_left
, cb
->compressed_len
, cb
->nr_pages
);
367 bytes_left
-= PAGE_SIZE
;
368 first_byte
+= PAGE_SIZE
;
373 ret
= btrfs_bio_wq_end_io(fs_info
, bio
, BTRFS_WQ_ENDIO_DATA
);
374 BUG_ON(ret
); /* -ENOMEM */
377 ret
= btrfs_csum_one_bio(inode
, bio
, start
, 1);
378 BUG_ON(ret
); /* -ENOMEM */
381 ret
= btrfs_map_bio(fs_info
, bio
, 0, 1);
391 static u64
bio_end_offset(struct bio
*bio
)
393 struct bio_vec
*last
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
395 return page_offset(last
->bv_page
) + last
->bv_len
+ last
->bv_offset
;
398 static noinline
int add_ra_bio_pages(struct inode
*inode
,
400 struct compressed_bio
*cb
)
402 unsigned long end_index
;
403 unsigned long pg_index
;
405 u64 isize
= i_size_read(inode
);
408 unsigned long nr_pages
= 0;
409 struct extent_map
*em
;
410 struct address_space
*mapping
= inode
->i_mapping
;
411 struct extent_map_tree
*em_tree
;
412 struct extent_io_tree
*tree
;
416 last_offset
= bio_end_offset(cb
->orig_bio
);
417 em_tree
= &BTRFS_I(inode
)->extent_tree
;
418 tree
= &BTRFS_I(inode
)->io_tree
;
423 end_index
= (i_size_read(inode
) - 1) >> PAGE_SHIFT
;
425 while (last_offset
< compressed_end
) {
426 pg_index
= last_offset
>> PAGE_SHIFT
;
428 if (pg_index
> end_index
)
432 page
= radix_tree_lookup(&mapping
->page_tree
, pg_index
);
434 if (page
&& !radix_tree_exceptional_entry(page
)) {
441 page
= __page_cache_alloc(mapping_gfp_constraint(mapping
,
446 if (add_to_page_cache_lru(page
, mapping
, pg_index
, GFP_NOFS
)) {
451 end
= last_offset
+ PAGE_SIZE
- 1;
453 * at this point, we have a locked page in the page cache
454 * for these bytes in the file. But, we have to make
455 * sure they map to this compressed extent on disk.
457 set_page_extent_mapped(page
);
458 lock_extent(tree
, last_offset
, end
);
459 read_lock(&em_tree
->lock
);
460 em
= lookup_extent_mapping(em_tree
, last_offset
,
462 read_unlock(&em_tree
->lock
);
464 if (!em
|| last_offset
< em
->start
||
465 (last_offset
+ PAGE_SIZE
> extent_map_end(em
)) ||
466 (em
->block_start
>> 9) != cb
->orig_bio
->bi_iter
.bi_sector
) {
468 unlock_extent(tree
, last_offset
, end
);
475 if (page
->index
== end_index
) {
477 size_t zero_offset
= isize
& (PAGE_SIZE
- 1);
481 zeros
= PAGE_SIZE
- zero_offset
;
482 userpage
= kmap_atomic(page
);
483 memset(userpage
+ zero_offset
, 0, zeros
);
484 flush_dcache_page(page
);
485 kunmap_atomic(userpage
);
489 ret
= bio_add_page(cb
->orig_bio
, page
,
492 if (ret
== PAGE_SIZE
) {
496 unlock_extent(tree
, last_offset
, end
);
502 last_offset
+= PAGE_SIZE
;
508 * for a compressed read, the bio we get passed has all the inode pages
509 * in it. We don't actually do IO on those pages but allocate new ones
510 * to hold the compressed pages on disk.
512 * bio->bi_iter.bi_sector points to the compressed extent on disk
513 * bio->bi_io_vec points to all of the inode pages
515 * After the compressed pages are read, we copy the bytes into the
516 * bio we were passed and then call the bio end_io calls
518 int btrfs_submit_compressed_read(struct inode
*inode
, struct bio
*bio
,
519 int mirror_num
, unsigned long bio_flags
)
521 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
522 struct extent_io_tree
*tree
;
523 struct extent_map_tree
*em_tree
;
524 struct compressed_bio
*cb
;
525 unsigned long compressed_len
;
526 unsigned long nr_pages
;
527 unsigned long pg_index
;
529 struct block_device
*bdev
;
530 struct bio
*comp_bio
;
531 u64 cur_disk_byte
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
534 struct extent_map
*em
;
539 tree
= &BTRFS_I(inode
)->io_tree
;
540 em_tree
= &BTRFS_I(inode
)->extent_tree
;
542 /* we need the actual starting offset of this extent in the file */
543 read_lock(&em_tree
->lock
);
544 em
= lookup_extent_mapping(em_tree
,
545 page_offset(bio
->bi_io_vec
->bv_page
),
547 read_unlock(&em_tree
->lock
);
551 compressed_len
= em
->block_len
;
552 cb
= kmalloc(compressed_bio_size(fs_info
, compressed_len
), GFP_NOFS
);
556 refcount_set(&cb
->pending_bios
, 0);
559 cb
->mirror_num
= mirror_num
;
562 cb
->start
= em
->orig_start
;
564 em_start
= em
->start
;
569 cb
->len
= bio
->bi_iter
.bi_size
;
570 cb
->compressed_len
= compressed_len
;
571 cb
->compress_type
= extent_compress_type(bio_flags
);
574 nr_pages
= DIV_ROUND_UP(compressed_len
, PAGE_SIZE
);
575 cb
->compressed_pages
= kcalloc(nr_pages
, sizeof(struct page
*),
577 if (!cb
->compressed_pages
)
580 bdev
= fs_info
->fs_devices
->latest_bdev
;
582 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
583 cb
->compressed_pages
[pg_index
] = alloc_page(GFP_NOFS
|
585 if (!cb
->compressed_pages
[pg_index
]) {
586 faili
= pg_index
- 1;
591 faili
= nr_pages
- 1;
592 cb
->nr_pages
= nr_pages
;
594 add_ra_bio_pages(inode
, em_start
+ em_len
, cb
);
596 /* include any pages we added in add_ra-bio_pages */
597 cb
->len
= bio
->bi_iter
.bi_size
;
599 comp_bio
= btrfs_bio_alloc(bdev
, cur_disk_byte
);
600 bio_set_op_attrs (comp_bio
, REQ_OP_READ
, 0);
601 comp_bio
->bi_private
= cb
;
602 comp_bio
->bi_end_io
= end_compressed_bio_read
;
603 refcount_set(&cb
->pending_bios
, 1);
605 for (pg_index
= 0; pg_index
< nr_pages
; pg_index
++) {
606 page
= cb
->compressed_pages
[pg_index
];
607 page
->mapping
= inode
->i_mapping
;
608 page
->index
= em_start
>> PAGE_SHIFT
;
610 if (comp_bio
->bi_iter
.bi_size
)
611 ret
= tree
->ops
->merge_bio_hook(page
, 0,
617 page
->mapping
= NULL
;
618 if (ret
|| bio_add_page(comp_bio
, page
, PAGE_SIZE
, 0) <
622 ret
= btrfs_bio_wq_end_io(fs_info
, comp_bio
,
623 BTRFS_WQ_ENDIO_DATA
);
624 BUG_ON(ret
); /* -ENOMEM */
627 * inc the count before we submit the bio so
628 * we know the end IO handler won't happen before
629 * we inc the count. Otherwise, the cb might get
630 * freed before we're done setting it up
632 refcount_inc(&cb
->pending_bios
);
634 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
635 ret
= btrfs_lookup_bio_sums(inode
, comp_bio
,
637 BUG_ON(ret
); /* -ENOMEM */
639 sums
+= DIV_ROUND_UP(comp_bio
->bi_iter
.bi_size
,
640 fs_info
->sectorsize
);
642 ret
= btrfs_map_bio(fs_info
, comp_bio
, mirror_num
, 0);
644 comp_bio
->bi_error
= ret
;
650 comp_bio
= btrfs_bio_alloc(bdev
, cur_disk_byte
);
651 bio_set_op_attrs(comp_bio
, REQ_OP_READ
, 0);
652 comp_bio
->bi_private
= cb
;
653 comp_bio
->bi_end_io
= end_compressed_bio_read
;
655 bio_add_page(comp_bio
, page
, PAGE_SIZE
, 0);
657 cur_disk_byte
+= PAGE_SIZE
;
661 ret
= btrfs_bio_wq_end_io(fs_info
, comp_bio
, BTRFS_WQ_ENDIO_DATA
);
662 BUG_ON(ret
); /* -ENOMEM */
664 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
665 ret
= btrfs_lookup_bio_sums(inode
, comp_bio
, sums
);
666 BUG_ON(ret
); /* -ENOMEM */
669 ret
= btrfs_map_bio(fs_info
, comp_bio
, mirror_num
, 0);
671 comp_bio
->bi_error
= ret
;
680 __free_page(cb
->compressed_pages
[faili
]);
684 kfree(cb
->compressed_pages
);
693 struct list_head idle_ws
;
695 /* Number of free workspaces */
697 /* Total number of allocated workspaces */
699 /* Waiters for a free workspace */
700 wait_queue_head_t ws_wait
;
701 } btrfs_comp_ws
[BTRFS_COMPRESS_TYPES
];
703 static const struct btrfs_compress_op
* const btrfs_compress_op
[] = {
704 &btrfs_zlib_compress
,
708 void __init
btrfs_init_compress(void)
712 for (i
= 0; i
< BTRFS_COMPRESS_TYPES
; i
++) {
713 struct list_head
*workspace
;
715 INIT_LIST_HEAD(&btrfs_comp_ws
[i
].idle_ws
);
716 spin_lock_init(&btrfs_comp_ws
[i
].ws_lock
);
717 atomic_set(&btrfs_comp_ws
[i
].total_ws
, 0);
718 init_waitqueue_head(&btrfs_comp_ws
[i
].ws_wait
);
721 * Preallocate one workspace for each compression type so
722 * we can guarantee forward progress in the worst case
724 workspace
= btrfs_compress_op
[i
]->alloc_workspace();
725 if (IS_ERR(workspace
)) {
726 pr_warn("BTRFS: cannot preallocate compression workspace, will try later\n");
728 atomic_set(&btrfs_comp_ws
[i
].total_ws
, 1);
729 btrfs_comp_ws
[i
].free_ws
= 1;
730 list_add(workspace
, &btrfs_comp_ws
[i
].idle_ws
);
736 * This finds an available workspace or allocates a new one.
737 * If it's not possible to allocate a new one, waits until there's one.
738 * Preallocation makes a forward progress guarantees and we do not return
741 static struct list_head
*find_workspace(int type
)
743 struct list_head
*workspace
;
744 int cpus
= num_online_cpus();
748 struct list_head
*idle_ws
= &btrfs_comp_ws
[idx
].idle_ws
;
749 spinlock_t
*ws_lock
= &btrfs_comp_ws
[idx
].ws_lock
;
750 atomic_t
*total_ws
= &btrfs_comp_ws
[idx
].total_ws
;
751 wait_queue_head_t
*ws_wait
= &btrfs_comp_ws
[idx
].ws_wait
;
752 int *free_ws
= &btrfs_comp_ws
[idx
].free_ws
;
755 if (!list_empty(idle_ws
)) {
756 workspace
= idle_ws
->next
;
759 spin_unlock(ws_lock
);
763 if (atomic_read(total_ws
) > cpus
) {
766 spin_unlock(ws_lock
);
767 prepare_to_wait(ws_wait
, &wait
, TASK_UNINTERRUPTIBLE
);
768 if (atomic_read(total_ws
) > cpus
&& !*free_ws
)
770 finish_wait(ws_wait
, &wait
);
773 atomic_inc(total_ws
);
774 spin_unlock(ws_lock
);
777 * Allocation helpers call vmalloc that can't use GFP_NOFS, so we have
778 * to turn it off here because we might get called from the restricted
779 * context of btrfs_compress_bio/btrfs_compress_pages
781 nofs_flag
= memalloc_nofs_save();
782 workspace
= btrfs_compress_op
[idx
]->alloc_workspace();
783 memalloc_nofs_restore(nofs_flag
);
785 if (IS_ERR(workspace
)) {
786 atomic_dec(total_ws
);
790 * Do not return the error but go back to waiting. There's a
791 * workspace preallocated for each type and the compression
792 * time is bounded so we get to a workspace eventually. This
793 * makes our caller's life easier.
795 * To prevent silent and low-probability deadlocks (when the
796 * initial preallocation fails), check if there are any
799 if (atomic_read(total_ws
) == 0) {
800 static DEFINE_RATELIMIT_STATE(_rs
,
801 /* once per minute */ 60 * HZ
,
804 if (__ratelimit(&_rs
)) {
805 pr_warn("BTRFS: no compression workspaces, low memory, retrying\n");
814 * put a workspace struct back on the list or free it if we have enough
815 * idle ones sitting around
817 static void free_workspace(int type
, struct list_head
*workspace
)
820 struct list_head
*idle_ws
= &btrfs_comp_ws
[idx
].idle_ws
;
821 spinlock_t
*ws_lock
= &btrfs_comp_ws
[idx
].ws_lock
;
822 atomic_t
*total_ws
= &btrfs_comp_ws
[idx
].total_ws
;
823 wait_queue_head_t
*ws_wait
= &btrfs_comp_ws
[idx
].ws_wait
;
824 int *free_ws
= &btrfs_comp_ws
[idx
].free_ws
;
827 if (*free_ws
< num_online_cpus()) {
828 list_add(workspace
, idle_ws
);
830 spin_unlock(ws_lock
);
833 spin_unlock(ws_lock
);
835 btrfs_compress_op
[idx
]->free_workspace(workspace
);
836 atomic_dec(total_ws
);
839 * Make sure counter is updated before we wake up waiters.
842 if (waitqueue_active(ws_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(&btrfs_comp_ws
[i
].idle_ws
)) {
856 workspace
= btrfs_comp_ws
[i
].idle_ws
.next
;
858 btrfs_compress_op
[i
]->free_workspace(workspace
);
859 atomic_dec(&btrfs_comp_ws
[i
].total_ws
);
865 * Given an address space and start and length, compress the bytes into @pages
866 * that are allocated on demand.
868 * @out_pages is an in/out parameter, holds maximum number of pages to allocate
869 * and returns number of actually allocated pages
871 * @total_in is used to return the number of bytes actually read. It
872 * may be smaller than the input length if we had to exit early because we
873 * ran out of room in the pages array or because we cross the
876 * @total_out is an in/out parameter, must be set to the input length and will
877 * be also used to return the total number of compressed bytes
879 * @max_out tells us the max number of bytes that we're allowed to
882 int btrfs_compress_pages(int type
, struct address_space
*mapping
,
883 u64 start
, struct page
**pages
,
884 unsigned long *out_pages
,
885 unsigned long *total_in
,
886 unsigned long *total_out
)
888 struct list_head
*workspace
;
891 workspace
= find_workspace(type
);
893 ret
= btrfs_compress_op
[type
-1]->compress_pages(workspace
, mapping
,
896 total_in
, total_out
);
897 free_workspace(type
, workspace
);
902 * pages_in is an array of pages with compressed data.
904 * disk_start is the starting logical offset of this array in the file
906 * orig_bio contains the pages from the file that we want to decompress into
908 * srclen is the number of bytes in pages_in
910 * The basic idea is that we have a bio that was created by readpages.
911 * The pages in the bio are for the uncompressed data, and they may not
912 * be contiguous. They all correspond to the range of bytes covered by
913 * the compressed extent.
915 static int btrfs_decompress_bio(struct compressed_bio
*cb
)
917 struct list_head
*workspace
;
919 int type
= cb
->compress_type
;
921 workspace
= find_workspace(type
);
922 ret
= btrfs_compress_op
[type
- 1]->decompress_bio(workspace
, cb
);
923 free_workspace(type
, workspace
);
929 * a less complex decompression routine. Our compressed data fits in a
930 * single page, and we want to read a single page out of it.
931 * start_byte tells us the offset into the compressed data we're interested in
933 int btrfs_decompress(int type
, unsigned char *data_in
, struct page
*dest_page
,
934 unsigned long start_byte
, size_t srclen
, size_t destlen
)
936 struct list_head
*workspace
;
939 workspace
= find_workspace(type
);
941 ret
= btrfs_compress_op
[type
-1]->decompress(workspace
, data_in
,
942 dest_page
, start_byte
,
945 free_workspace(type
, workspace
);
949 void btrfs_exit_compress(void)
955 * Copy uncompressed data from working buffer to pages.
957 * buf_start is the byte offset we're of the start of our workspace buffer.
959 * total_out is the last byte of the buffer
961 int btrfs_decompress_buf2page(const char *buf
, unsigned long buf_start
,
962 unsigned long total_out
, u64 disk_start
,
965 unsigned long buf_offset
;
966 unsigned long current_buf_start
;
967 unsigned long start_byte
;
968 unsigned long prev_start_byte
;
969 unsigned long working_bytes
= total_out
- buf_start
;
972 struct bio_vec bvec
= bio_iter_iovec(bio
, bio
->bi_iter
);
975 * start byte is the first byte of the page we're currently
976 * copying into relative to the start of the compressed data.
978 start_byte
= page_offset(bvec
.bv_page
) - disk_start
;
980 /* we haven't yet hit data corresponding to this page */
981 if (total_out
<= start_byte
)
985 * the start of the data we care about is offset into
986 * the middle of our working buffer
988 if (total_out
> start_byte
&& buf_start
< start_byte
) {
989 buf_offset
= start_byte
- buf_start
;
990 working_bytes
-= buf_offset
;
994 current_buf_start
= buf_start
;
996 /* copy bytes from the working buffer into the pages */
997 while (working_bytes
> 0) {
998 bytes
= min_t(unsigned long, bvec
.bv_len
,
999 PAGE_SIZE
- buf_offset
);
1000 bytes
= min(bytes
, working_bytes
);
1002 kaddr
= kmap_atomic(bvec
.bv_page
);
1003 memcpy(kaddr
+ bvec
.bv_offset
, buf
+ buf_offset
, bytes
);
1004 kunmap_atomic(kaddr
);
1005 flush_dcache_page(bvec
.bv_page
);
1007 buf_offset
+= bytes
;
1008 working_bytes
-= bytes
;
1009 current_buf_start
+= bytes
;
1011 /* check if we need to pick another page */
1012 bio_advance(bio
, bytes
);
1013 if (!bio
->bi_iter
.bi_size
)
1015 bvec
= bio_iter_iovec(bio
, bio
->bi_iter
);
1016 prev_start_byte
= start_byte
;
1017 start_byte
= page_offset(bvec
.bv_page
) - disk_start
;
1020 * We need to make sure we're only adjusting
1021 * our offset into compression working buffer when
1022 * we're switching pages. Otherwise we can incorrectly
1023 * keep copying when we were actually done.
1025 if (start_byte
!= prev_start_byte
) {
1027 * make sure our new page is covered by this
1030 if (total_out
<= start_byte
)
1034 * the next page in the biovec might not be adjacent
1035 * to the last page, but it might still be found
1036 * inside this working buffer. bump our offset pointer
1038 if (total_out
> start_byte
&&
1039 current_buf_start
< start_byte
) {
1040 buf_offset
= start_byte
- buf_start
;
1041 working_bytes
= total_out
- start_byte
;
1042 current_buf_start
= buf_start
+ buf_offset
;