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/pagevec.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 struct compressed_bio
{
47 /* number of bios pending for this compressed extent */
48 atomic_t pending_bios
;
50 /* the pages with the compressed data on them */
51 struct page
**compressed_pages
;
53 /* inode that owns this data */
56 /* starting offset in the inode for our pages */
59 /* number of bytes in the inode we're working on */
62 /* number of bytes on disk */
63 unsigned long compressed_len
;
65 /* number of compressed pages in the array */
66 unsigned long nr_pages
;
72 /* for reads, this is the bio we are copying the data into */
76 * the start of a variable length array of checksums only
82 static inline int compressed_bio_size(struct btrfs_root
*root
,
83 unsigned long disk_size
)
85 u16 csum_size
= btrfs_super_csum_size(&root
->fs_info
->super_copy
);
86 return sizeof(struct compressed_bio
) +
87 ((disk_size
+ root
->sectorsize
- 1) / root
->sectorsize
) *
91 static struct bio
*compressed_bio_alloc(struct block_device
*bdev
,
92 u64 first_byte
, gfp_t gfp_flags
)
97 nr_vecs
= bio_get_nr_vecs(bdev
);
98 bio
= bio_alloc(gfp_flags
, nr_vecs
);
100 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
101 while (!bio
&& (nr_vecs
/= 2))
102 bio
= bio_alloc(gfp_flags
, nr_vecs
);
108 bio
->bi_sector
= first_byte
>> 9;
113 static int check_compressed_csum(struct inode
*inode
,
114 struct compressed_bio
*cb
,
118 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
123 u32
*cb_sum
= &cb
->sums
;
125 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
128 for (i
= 0; i
< cb
->nr_pages
; i
++) {
129 page
= cb
->compressed_pages
[i
];
132 kaddr
= kmap_atomic(page
, KM_USER0
);
133 csum
= btrfs_csum_data(root
, kaddr
, csum
, PAGE_CACHE_SIZE
);
134 btrfs_csum_final(csum
, (char *)&csum
);
135 kunmap_atomic(kaddr
, KM_USER0
);
137 if (csum
!= *cb_sum
) {
138 printk(KERN_INFO
"btrfs csum failed ino %lu "
139 "extent %llu csum %u "
140 "wanted %u mirror %d\n", inode
->i_ino
,
141 (unsigned long long)disk_start
,
142 csum
, *cb_sum
, cb
->mirror_num
);
154 /* when we finish reading compressed pages from the disk, we
155 * decompress them and then run the bio end_io routines on the
156 * decompressed pages (in the inode address space).
158 * This allows the checksumming and other IO error handling routines
161 * The compressed pages are freed here, and it must be run
164 static void end_compressed_bio_read(struct bio
*bio
, int err
)
166 struct extent_io_tree
*tree
;
167 struct compressed_bio
*cb
= bio
->bi_private
;
176 /* if there are more bios still pending for this compressed
179 if (!atomic_dec_and_test(&cb
->pending_bios
))
183 ret
= check_compressed_csum(inode
, cb
, (u64
)bio
->bi_sector
<< 9);
187 /* ok, we're the last bio for this extent, lets start
190 tree
= &BTRFS_I(inode
)->io_tree
;
191 ret
= btrfs_zlib_decompress_biovec(cb
->compressed_pages
,
193 cb
->orig_bio
->bi_io_vec
,
194 cb
->orig_bio
->bi_vcnt
,
200 /* release the compressed pages */
202 for (index
= 0; index
< cb
->nr_pages
; index
++) {
203 page
= cb
->compressed_pages
[index
];
204 page
->mapping
= NULL
;
205 page_cache_release(page
);
208 /* do io completion on the original bio */
210 bio_io_error(cb
->orig_bio
);
213 struct bio_vec
*bvec
= cb
->orig_bio
->bi_io_vec
;
216 * we have verified the checksum already, set page
217 * checked so the end_io handlers know about it
219 while (bio_index
< cb
->orig_bio
->bi_vcnt
) {
220 SetPageChecked(bvec
->bv_page
);
224 bio_endio(cb
->orig_bio
, 0);
227 /* finally free the cb struct */
228 kfree(cb
->compressed_pages
);
235 * Clear the writeback bits on all of the file
236 * pages for a compressed write
238 static noinline
int end_compressed_writeback(struct inode
*inode
, u64 start
,
239 unsigned long ram_size
)
241 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
242 unsigned long end_index
= (start
+ ram_size
- 1) >> PAGE_CACHE_SHIFT
;
243 struct page
*pages
[16];
244 unsigned long nr_pages
= end_index
- index
+ 1;
248 while (nr_pages
> 0) {
249 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
251 nr_pages
, ARRAY_SIZE(pages
)), pages
);
257 for (i
= 0; i
< ret
; i
++) {
258 end_page_writeback(pages
[i
]);
259 page_cache_release(pages
[i
]);
264 /* the inode may be gone now */
269 * do the cleanup once all the compressed pages hit the disk.
270 * This will clear writeback on the file pages and free the compressed
273 * This also calls the writeback end hooks for the file pages so that
274 * metadata and checksums can be updated in the file.
276 static void end_compressed_bio_write(struct bio
*bio
, int err
)
278 struct extent_io_tree
*tree
;
279 struct compressed_bio
*cb
= bio
->bi_private
;
287 /* if there are more bios still pending for this compressed
290 if (!atomic_dec_and_test(&cb
->pending_bios
))
293 /* ok, we're the last bio for this extent, step one is to
294 * call back into the FS and do all the end_io operations
297 tree
= &BTRFS_I(inode
)->io_tree
;
298 cb
->compressed_pages
[0]->mapping
= cb
->inode
->i_mapping
;
299 tree
->ops
->writepage_end_io_hook(cb
->compressed_pages
[0],
301 cb
->start
+ cb
->len
- 1,
303 cb
->compressed_pages
[0]->mapping
= NULL
;
305 end_compressed_writeback(inode
, cb
->start
, cb
->len
);
306 /* note, our inode could be gone now */
309 * release the compressed pages, these came from alloc_page and
310 * are not attached to the inode at all
313 for (index
= 0; index
< cb
->nr_pages
; index
++) {
314 page
= cb
->compressed_pages
[index
];
315 page
->mapping
= NULL
;
316 page_cache_release(page
);
319 /* finally free the cb struct */
320 kfree(cb
->compressed_pages
);
327 * worker function to build and submit bios for previously compressed pages.
328 * The corresponding pages in the inode should be marked for writeback
329 * and the compressed pages should have a reference on them for dropping
330 * when the IO is complete.
332 * This also checksums the file bytes and gets things ready for
335 int btrfs_submit_compressed_write(struct inode
*inode
, u64 start
,
336 unsigned long len
, u64 disk_start
,
337 unsigned long compressed_len
,
338 struct page
**compressed_pages
,
339 unsigned long nr_pages
)
341 struct bio
*bio
= NULL
;
342 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
343 struct compressed_bio
*cb
;
344 unsigned long bytes_left
;
345 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
348 u64 first_byte
= disk_start
;
349 struct block_device
*bdev
;
352 WARN_ON(start
& ((u64
)PAGE_CACHE_SIZE
- 1));
353 cb
= kmalloc(compressed_bio_size(root
, compressed_len
), GFP_NOFS
);
354 atomic_set(&cb
->pending_bios
, 0);
360 cb
->compressed_pages
= compressed_pages
;
361 cb
->compressed_len
= compressed_len
;
363 cb
->nr_pages
= nr_pages
;
365 bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
367 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
368 bio
->bi_private
= cb
;
369 bio
->bi_end_io
= end_compressed_bio_write
;
370 atomic_inc(&cb
->pending_bios
);
372 /* create and submit bios for the compressed pages */
373 bytes_left
= compressed_len
;
374 for (page_index
= 0; page_index
< cb
->nr_pages
; page_index
++) {
375 page
= compressed_pages
[page_index
];
376 page
->mapping
= inode
->i_mapping
;
378 ret
= io_tree
->ops
->merge_bio_hook(page
, 0,
384 page
->mapping
= NULL
;
385 if (ret
|| bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0) <
390 * inc the count before we submit the bio so
391 * we know the end IO handler won't happen before
392 * we inc the count. Otherwise, the cb might get
393 * freed before we're done setting it up
395 atomic_inc(&cb
->pending_bios
);
396 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
399 ret
= btrfs_csum_one_bio(root
, inode
, bio
, start
, 1);
402 ret
= btrfs_map_bio(root
, WRITE
, bio
, 0, 1);
407 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
408 bio
->bi_private
= cb
;
409 bio
->bi_end_io
= end_compressed_bio_write
;
410 bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0);
412 if (bytes_left
< PAGE_CACHE_SIZE
) {
413 printk("bytes left %lu compress len %lu nr %lu\n",
414 bytes_left
, cb
->compressed_len
, cb
->nr_pages
);
416 bytes_left
-= PAGE_CACHE_SIZE
;
417 first_byte
+= PAGE_CACHE_SIZE
;
422 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
425 ret
= btrfs_csum_one_bio(root
, inode
, bio
, start
, 1);
428 ret
= btrfs_map_bio(root
, WRITE
, bio
, 0, 1);
435 static noinline
int add_ra_bio_pages(struct inode
*inode
,
437 struct compressed_bio
*cb
)
439 unsigned long end_index
;
440 unsigned long page_index
;
442 u64 isize
= i_size_read(inode
);
445 unsigned long nr_pages
= 0;
446 struct extent_map
*em
;
447 struct address_space
*mapping
= inode
->i_mapping
;
449 struct extent_map_tree
*em_tree
;
450 struct extent_io_tree
*tree
;
454 page
= cb
->orig_bio
->bi_io_vec
[cb
->orig_bio
->bi_vcnt
- 1].bv_page
;
455 last_offset
= (page_offset(page
) + PAGE_CACHE_SIZE
);
456 em_tree
= &BTRFS_I(inode
)->extent_tree
;
457 tree
= &BTRFS_I(inode
)->io_tree
;
462 end_index
= (i_size_read(inode
) - 1) >> PAGE_CACHE_SHIFT
;
464 pagevec_init(&pvec
, 0);
465 while (last_offset
< compressed_end
) {
466 page_index
= last_offset
>> PAGE_CACHE_SHIFT
;
468 if (page_index
> end_index
)
472 page
= radix_tree_lookup(&mapping
->page_tree
, page_index
);
481 page
= alloc_page(mapping_gfp_mask(mapping
) & ~__GFP_FS
);
485 page
->index
= page_index
;
487 * what we want to do here is call add_to_page_cache_lru,
488 * but that isn't exported, so we reproduce it here
490 if (add_to_page_cache(page
, mapping
,
491 page
->index
, GFP_NOFS
)) {
492 page_cache_release(page
);
496 /* open coding of lru_cache_add, also not exported */
497 page_cache_get(page
);
498 if (!pagevec_add(&pvec
, page
))
499 __pagevec_lru_add_file(&pvec
);
501 end
= last_offset
+ PAGE_CACHE_SIZE
- 1;
503 * at this point, we have a locked page in the page cache
504 * for these bytes in the file. But, we have to make
505 * sure they map to this compressed extent on disk.
507 set_page_extent_mapped(page
);
508 lock_extent(tree
, last_offset
, end
, GFP_NOFS
);
509 read_lock(&em_tree
->lock
);
510 em
= lookup_extent_mapping(em_tree
, last_offset
,
512 read_unlock(&em_tree
->lock
);
514 if (!em
|| last_offset
< em
->start
||
515 (last_offset
+ PAGE_CACHE_SIZE
> extent_map_end(em
)) ||
516 (em
->block_start
>> 9) != cb
->orig_bio
->bi_sector
) {
518 unlock_extent(tree
, last_offset
, end
, GFP_NOFS
);
520 page_cache_release(page
);
525 if (page
->index
== end_index
) {
527 size_t zero_offset
= isize
& (PAGE_CACHE_SIZE
- 1);
531 zeros
= PAGE_CACHE_SIZE
- zero_offset
;
532 userpage
= kmap_atomic(page
, KM_USER0
);
533 memset(userpage
+ zero_offset
, 0, zeros
);
534 flush_dcache_page(page
);
535 kunmap_atomic(userpage
, KM_USER0
);
539 ret
= bio_add_page(cb
->orig_bio
, page
,
542 if (ret
== PAGE_CACHE_SIZE
) {
544 page_cache_release(page
);
546 unlock_extent(tree
, last_offset
, end
, GFP_NOFS
);
548 page_cache_release(page
);
552 last_offset
+= PAGE_CACHE_SIZE
;
554 if (pagevec_count(&pvec
))
555 __pagevec_lru_add_file(&pvec
);
560 * for a compressed read, the bio we get passed has all the inode pages
561 * in it. We don't actually do IO on those pages but allocate new ones
562 * to hold the compressed pages on disk.
564 * bio->bi_sector points to the compressed extent on disk
565 * bio->bi_io_vec points to all of the inode pages
566 * bio->bi_vcnt is a count of pages
568 * After the compressed pages are read, we copy the bytes into the
569 * bio we were passed and then call the bio end_io calls
571 int btrfs_submit_compressed_read(struct inode
*inode
, struct bio
*bio
,
572 int mirror_num
, unsigned long bio_flags
)
574 struct extent_io_tree
*tree
;
575 struct extent_map_tree
*em_tree
;
576 struct compressed_bio
*cb
;
577 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
578 unsigned long uncompressed_len
= bio
->bi_vcnt
* PAGE_CACHE_SIZE
;
579 unsigned long compressed_len
;
580 unsigned long nr_pages
;
581 unsigned long page_index
;
583 struct block_device
*bdev
;
584 struct bio
*comp_bio
;
585 u64 cur_disk_byte
= (u64
)bio
->bi_sector
<< 9;
588 struct extent_map
*em
;
592 tree
= &BTRFS_I(inode
)->io_tree
;
593 em_tree
= &BTRFS_I(inode
)->extent_tree
;
595 /* we need the actual starting offset of this extent in the file */
596 read_lock(&em_tree
->lock
);
597 em
= lookup_extent_mapping(em_tree
,
598 page_offset(bio
->bi_io_vec
->bv_page
),
600 read_unlock(&em_tree
->lock
);
602 compressed_len
= em
->block_len
;
603 cb
= kmalloc(compressed_bio_size(root
, compressed_len
), GFP_NOFS
);
604 atomic_set(&cb
->pending_bios
, 0);
607 cb
->mirror_num
= mirror_num
;
610 cb
->start
= em
->orig_start
;
612 em_start
= em
->start
;
617 cb
->len
= uncompressed_len
;
618 cb
->compressed_len
= compressed_len
;
621 nr_pages
= (compressed_len
+ PAGE_CACHE_SIZE
- 1) /
623 cb
->compressed_pages
= kmalloc(sizeof(struct page
*) * nr_pages
,
625 bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
627 for (page_index
= 0; page_index
< nr_pages
; page_index
++) {
628 cb
->compressed_pages
[page_index
] = alloc_page(GFP_NOFS
|
631 cb
->nr_pages
= nr_pages
;
633 add_ra_bio_pages(inode
, em_start
+ em_len
, cb
);
635 /* include any pages we added in add_ra-bio_pages */
636 uncompressed_len
= bio
->bi_vcnt
* PAGE_CACHE_SIZE
;
637 cb
->len
= uncompressed_len
;
639 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
, GFP_NOFS
);
640 comp_bio
->bi_private
= cb
;
641 comp_bio
->bi_end_io
= end_compressed_bio_read
;
642 atomic_inc(&cb
->pending_bios
);
644 for (page_index
= 0; page_index
< nr_pages
; page_index
++) {
645 page
= cb
->compressed_pages
[page_index
];
646 page
->mapping
= inode
->i_mapping
;
647 page
->index
= em_start
>> PAGE_CACHE_SHIFT
;
649 if (comp_bio
->bi_size
)
650 ret
= tree
->ops
->merge_bio_hook(page
, 0,
656 page
->mapping
= NULL
;
657 if (ret
|| bio_add_page(comp_bio
, page
, PAGE_CACHE_SIZE
, 0) <
661 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
, 0);
665 * inc the count before we submit the bio so
666 * we know the end IO handler won't happen before
667 * we inc the count. Otherwise, the cb might get
668 * freed before we're done setting it up
670 atomic_inc(&cb
->pending_bios
);
672 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
673 btrfs_lookup_bio_sums(root
, inode
, comp_bio
,
676 sums
+= (comp_bio
->bi_size
+ root
->sectorsize
- 1) /
679 ret
= btrfs_map_bio(root
, READ
, comp_bio
,
685 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
,
687 comp_bio
->bi_private
= cb
;
688 comp_bio
->bi_end_io
= end_compressed_bio_read
;
690 bio_add_page(comp_bio
, page
, PAGE_CACHE_SIZE
, 0);
692 cur_disk_byte
+= PAGE_CACHE_SIZE
;
696 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
, 0);
699 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
))
700 btrfs_lookup_bio_sums(root
, inode
, comp_bio
, sums
);
702 ret
= btrfs_map_bio(root
, READ
, comp_bio
, mirror_num
, 0);