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>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "ordered-data.h"
41 #include "compression.h"
42 #include "extent_io.h"
43 #include "extent_map.h"
45 struct compressed_bio
{
46 /* number of bios pending for this compressed extent */
47 atomic_t pending_bios
;
49 /* the pages with the compressed data on them */
50 struct page
**compressed_pages
;
52 /* inode that owns this data */
55 /* starting offset in the inode for our pages */
58 /* number of bytes in the inode we're working on */
61 /* number of bytes on disk */
62 unsigned long compressed_len
;
64 /* number of compressed pages in the array */
65 unsigned long nr_pages
;
71 /* for reads, this is the bio we are copying the data into */
75 * the start of a variable length array of checksums only
81 static inline int compressed_bio_size(struct btrfs_root
*root
,
82 unsigned long disk_size
)
84 u16 csum_size
= btrfs_super_csum_size(&root
->fs_info
->super_copy
);
85 return sizeof(struct compressed_bio
) +
86 ((disk_size
+ root
->sectorsize
- 1) / root
->sectorsize
) *
90 static struct bio
*compressed_bio_alloc(struct block_device
*bdev
,
91 u64 first_byte
, gfp_t gfp_flags
)
96 nr_vecs
= bio_get_nr_vecs(bdev
);
97 bio
= bio_alloc(gfp_flags
, nr_vecs
);
99 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
100 while (!bio
&& (nr_vecs
/= 2))
101 bio
= bio_alloc(gfp_flags
, nr_vecs
);
107 bio
->bi_sector
= first_byte
>> 9;
112 static int check_compressed_csum(struct inode
*inode
,
113 struct compressed_bio
*cb
,
117 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
122 u32
*cb_sum
= &cb
->sums
;
124 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
127 for (i
= 0; i
< cb
->nr_pages
; i
++) {
128 page
= cb
->compressed_pages
[i
];
131 kaddr
= kmap_atomic(page
, KM_USER0
);
132 csum
= btrfs_csum_data(root
, kaddr
, csum
, PAGE_CACHE_SIZE
);
133 btrfs_csum_final(csum
, (char *)&csum
);
134 kunmap_atomic(kaddr
, KM_USER0
);
136 if (csum
!= *cb_sum
) {
137 printk(KERN_INFO
"btrfs csum failed ino %lu "
138 "extent %llu csum %u "
139 "wanted %u mirror %d\n", inode
->i_ino
,
140 (unsigned long long)disk_start
,
141 csum
, *cb_sum
, cb
->mirror_num
);
153 /* when we finish reading compressed pages from the disk, we
154 * decompress them and then run the bio end_io routines on the
155 * decompressed pages (in the inode address space).
157 * This allows the checksumming and other IO error handling routines
160 * The compressed pages are freed here, and it must be run
163 static void end_compressed_bio_read(struct bio
*bio
, int err
)
165 struct extent_io_tree
*tree
;
166 struct compressed_bio
*cb
= bio
->bi_private
;
175 /* if there are more bios still pending for this compressed
178 if (!atomic_dec_and_test(&cb
->pending_bios
))
182 ret
= check_compressed_csum(inode
, cb
, (u64
)bio
->bi_sector
<< 9);
186 /* ok, we're the last bio for this extent, lets start
189 tree
= &BTRFS_I(inode
)->io_tree
;
190 ret
= btrfs_zlib_decompress_biovec(cb
->compressed_pages
,
192 cb
->orig_bio
->bi_io_vec
,
193 cb
->orig_bio
->bi_vcnt
,
199 /* release the compressed pages */
201 for (index
= 0; index
< cb
->nr_pages
; index
++) {
202 page
= cb
->compressed_pages
[index
];
203 page
->mapping
= NULL
;
204 page_cache_release(page
);
207 /* do io completion on the original bio */
209 bio_io_error(cb
->orig_bio
);
212 struct bio_vec
*bvec
= cb
->orig_bio
->bi_io_vec
;
215 * we have verified the checksum already, set page
216 * checked so the end_io handlers know about it
218 while (bio_index
< cb
->orig_bio
->bi_vcnt
) {
219 SetPageChecked(bvec
->bv_page
);
223 bio_endio(cb
->orig_bio
, 0);
226 /* finally free the cb struct */
227 kfree(cb
->compressed_pages
);
234 * Clear the writeback bits on all of the file
235 * pages for a compressed write
237 static noinline
int end_compressed_writeback(struct inode
*inode
, u64 start
,
238 unsigned long ram_size
)
240 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
241 unsigned long end_index
= (start
+ ram_size
- 1) >> PAGE_CACHE_SHIFT
;
242 struct page
*pages
[16];
243 unsigned long nr_pages
= end_index
- index
+ 1;
247 while (nr_pages
> 0) {
248 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
250 nr_pages
, ARRAY_SIZE(pages
)), pages
);
256 for (i
= 0; i
< ret
; i
++) {
257 end_page_writeback(pages
[i
]);
258 page_cache_release(pages
[i
]);
263 /* the inode may be gone now */
268 * do the cleanup once all the compressed pages hit the disk.
269 * This will clear writeback on the file pages and free the compressed
272 * This also calls the writeback end hooks for the file pages so that
273 * metadata and checksums can be updated in the file.
275 static void end_compressed_bio_write(struct bio
*bio
, int err
)
277 struct extent_io_tree
*tree
;
278 struct compressed_bio
*cb
= bio
->bi_private
;
286 /* if there are more bios still pending for this compressed
289 if (!atomic_dec_and_test(&cb
->pending_bios
))
292 /* ok, we're the last bio for this extent, step one is to
293 * call back into the FS and do all the end_io operations
296 tree
= &BTRFS_I(inode
)->io_tree
;
297 cb
->compressed_pages
[0]->mapping
= cb
->inode
->i_mapping
;
298 tree
->ops
->writepage_end_io_hook(cb
->compressed_pages
[0],
300 cb
->start
+ cb
->len
- 1,
302 cb
->compressed_pages
[0]->mapping
= NULL
;
304 end_compressed_writeback(inode
, cb
->start
, cb
->len
);
305 /* note, our inode could be gone now */
308 * release the compressed pages, these came from alloc_page and
309 * are not attached to the inode at all
312 for (index
= 0; index
< cb
->nr_pages
; index
++) {
313 page
= cb
->compressed_pages
[index
];
314 page
->mapping
= NULL
;
315 page_cache_release(page
);
318 /* finally free the cb struct */
319 kfree(cb
->compressed_pages
);
326 * worker function to build and submit bios for previously compressed pages.
327 * The corresponding pages in the inode should be marked for writeback
328 * and the compressed pages should have a reference on them for dropping
329 * when the IO is complete.
331 * This also checksums the file bytes and gets things ready for
334 int btrfs_submit_compressed_write(struct inode
*inode
, u64 start
,
335 unsigned long len
, u64 disk_start
,
336 unsigned long compressed_len
,
337 struct page
**compressed_pages
,
338 unsigned long nr_pages
)
340 struct bio
*bio
= NULL
;
341 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
342 struct compressed_bio
*cb
;
343 unsigned long bytes_left
;
344 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
347 u64 first_byte
= disk_start
;
348 struct block_device
*bdev
;
351 WARN_ON(start
& ((u64
)PAGE_CACHE_SIZE
- 1));
352 cb
= kmalloc(compressed_bio_size(root
, compressed_len
), GFP_NOFS
);
353 atomic_set(&cb
->pending_bios
, 0);
359 cb
->compressed_pages
= compressed_pages
;
360 cb
->compressed_len
= compressed_len
;
362 cb
->nr_pages
= nr_pages
;
364 bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
366 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
367 bio
->bi_private
= cb
;
368 bio
->bi_end_io
= end_compressed_bio_write
;
369 atomic_inc(&cb
->pending_bios
);
371 /* create and submit bios for the compressed pages */
372 bytes_left
= compressed_len
;
373 for (page_index
= 0; page_index
< cb
->nr_pages
; page_index
++) {
374 page
= compressed_pages
[page_index
];
375 page
->mapping
= inode
->i_mapping
;
377 ret
= io_tree
->ops
->merge_bio_hook(page
, 0,
383 page
->mapping
= NULL
;
384 if (ret
|| bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0) <
389 * inc the count before we submit the bio so
390 * we know the end IO handler won't happen before
391 * we inc the count. Otherwise, the cb might get
392 * freed before we're done setting it up
394 atomic_inc(&cb
->pending_bios
);
395 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
398 ret
= btrfs_csum_one_bio(root
, inode
, bio
, start
, 1);
401 ret
= btrfs_map_bio(root
, WRITE
, bio
, 0, 1);
406 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
407 bio
->bi_private
= cb
;
408 bio
->bi_end_io
= end_compressed_bio_write
;
409 bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0);
411 if (bytes_left
< PAGE_CACHE_SIZE
) {
412 printk("bytes left %lu compress len %lu nr %lu\n",
413 bytes_left
, cb
->compressed_len
, cb
->nr_pages
);
415 bytes_left
-= PAGE_CACHE_SIZE
;
416 first_byte
+= PAGE_CACHE_SIZE
;
421 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
424 ret
= btrfs_csum_one_bio(root
, inode
, bio
, start
, 1);
427 ret
= btrfs_map_bio(root
, WRITE
, bio
, 0, 1);
434 static noinline
int add_ra_bio_pages(struct inode
*inode
,
436 struct compressed_bio
*cb
)
438 unsigned long end_index
;
439 unsigned long page_index
;
441 u64 isize
= i_size_read(inode
);
444 unsigned long nr_pages
= 0;
445 struct extent_map
*em
;
446 struct address_space
*mapping
= inode
->i_mapping
;
447 struct extent_map_tree
*em_tree
;
448 struct extent_io_tree
*tree
;
452 page
= cb
->orig_bio
->bi_io_vec
[cb
->orig_bio
->bi_vcnt
- 1].bv_page
;
453 last_offset
= (page_offset(page
) + PAGE_CACHE_SIZE
);
454 em_tree
= &BTRFS_I(inode
)->extent_tree
;
455 tree
= &BTRFS_I(inode
)->io_tree
;
460 end_index
= (i_size_read(inode
) - 1) >> PAGE_CACHE_SHIFT
;
462 while (last_offset
< compressed_end
) {
463 page_index
= last_offset
>> PAGE_CACHE_SHIFT
;
465 if (page_index
> end_index
)
469 page
= radix_tree_lookup(&mapping
->page_tree
, page_index
);
478 page
= __page_cache_alloc(mapping_gfp_mask(mapping
) &
483 if (add_to_page_cache_lru(page
, mapping
, page_index
,
485 page_cache_release(page
);
489 end
= last_offset
+ PAGE_CACHE_SIZE
- 1;
491 * at this point, we have a locked page in the page cache
492 * for these bytes in the file. But, we have to make
493 * sure they map to this compressed extent on disk.
495 set_page_extent_mapped(page
);
496 lock_extent(tree
, last_offset
, end
, GFP_NOFS
);
497 read_lock(&em_tree
->lock
);
498 em
= lookup_extent_mapping(em_tree
, last_offset
,
500 read_unlock(&em_tree
->lock
);
502 if (!em
|| last_offset
< em
->start
||
503 (last_offset
+ PAGE_CACHE_SIZE
> extent_map_end(em
)) ||
504 (em
->block_start
>> 9) != cb
->orig_bio
->bi_sector
) {
506 unlock_extent(tree
, last_offset
, end
, GFP_NOFS
);
508 page_cache_release(page
);
513 if (page
->index
== end_index
) {
515 size_t zero_offset
= isize
& (PAGE_CACHE_SIZE
- 1);
519 zeros
= PAGE_CACHE_SIZE
- zero_offset
;
520 userpage
= kmap_atomic(page
, KM_USER0
);
521 memset(userpage
+ zero_offset
, 0, zeros
);
522 flush_dcache_page(page
);
523 kunmap_atomic(userpage
, KM_USER0
);
527 ret
= bio_add_page(cb
->orig_bio
, page
,
530 if (ret
== PAGE_CACHE_SIZE
) {
532 page_cache_release(page
);
534 unlock_extent(tree
, last_offset
, end
, GFP_NOFS
);
536 page_cache_release(page
);
540 last_offset
+= PAGE_CACHE_SIZE
;
546 * for a compressed read, the bio we get passed has all the inode pages
547 * in it. We don't actually do IO on those pages but allocate new ones
548 * to hold the compressed pages on disk.
550 * bio->bi_sector points to the compressed extent on disk
551 * bio->bi_io_vec points to all of the inode pages
552 * bio->bi_vcnt is a count of pages
554 * After the compressed pages are read, we copy the bytes into the
555 * bio we were passed and then call the bio end_io calls
557 int btrfs_submit_compressed_read(struct inode
*inode
, struct bio
*bio
,
558 int mirror_num
, unsigned long bio_flags
)
560 struct extent_io_tree
*tree
;
561 struct extent_map_tree
*em_tree
;
562 struct compressed_bio
*cb
;
563 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
564 unsigned long uncompressed_len
= bio
->bi_vcnt
* PAGE_CACHE_SIZE
;
565 unsigned long compressed_len
;
566 unsigned long nr_pages
;
567 unsigned long page_index
;
569 struct block_device
*bdev
;
570 struct bio
*comp_bio
;
571 u64 cur_disk_byte
= (u64
)bio
->bi_sector
<< 9;
574 struct extent_map
*em
;
578 tree
= &BTRFS_I(inode
)->io_tree
;
579 em_tree
= &BTRFS_I(inode
)->extent_tree
;
581 /* we need the actual starting offset of this extent in the file */
582 read_lock(&em_tree
->lock
);
583 em
= lookup_extent_mapping(em_tree
,
584 page_offset(bio
->bi_io_vec
->bv_page
),
586 read_unlock(&em_tree
->lock
);
588 compressed_len
= em
->block_len
;
589 cb
= kmalloc(compressed_bio_size(root
, compressed_len
), GFP_NOFS
);
590 atomic_set(&cb
->pending_bios
, 0);
593 cb
->mirror_num
= mirror_num
;
596 cb
->start
= em
->orig_start
;
598 em_start
= em
->start
;
603 cb
->len
= uncompressed_len
;
604 cb
->compressed_len
= compressed_len
;
607 nr_pages
= (compressed_len
+ PAGE_CACHE_SIZE
- 1) /
609 cb
->compressed_pages
= kmalloc(sizeof(struct page
*) * nr_pages
,
611 bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
613 for (page_index
= 0; page_index
< nr_pages
; page_index
++) {
614 cb
->compressed_pages
[page_index
] = alloc_page(GFP_NOFS
|
617 cb
->nr_pages
= nr_pages
;
619 add_ra_bio_pages(inode
, em_start
+ em_len
, cb
);
621 /* include any pages we added in add_ra-bio_pages */
622 uncompressed_len
= bio
->bi_vcnt
* PAGE_CACHE_SIZE
;
623 cb
->len
= uncompressed_len
;
625 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
, GFP_NOFS
);
626 comp_bio
->bi_private
= cb
;
627 comp_bio
->bi_end_io
= end_compressed_bio_read
;
628 atomic_inc(&cb
->pending_bios
);
630 for (page_index
= 0; page_index
< nr_pages
; page_index
++) {
631 page
= cb
->compressed_pages
[page_index
];
632 page
->mapping
= inode
->i_mapping
;
633 page
->index
= em_start
>> PAGE_CACHE_SHIFT
;
635 if (comp_bio
->bi_size
)
636 ret
= tree
->ops
->merge_bio_hook(page
, 0,
642 page
->mapping
= NULL
;
643 if (ret
|| bio_add_page(comp_bio
, page
, PAGE_CACHE_SIZE
, 0) <
647 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
, 0);
651 * inc the count before we submit the bio so
652 * we know the end IO handler won't happen before
653 * we inc the count. Otherwise, the cb might get
654 * freed before we're done setting it up
656 atomic_inc(&cb
->pending_bios
);
658 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
659 btrfs_lookup_bio_sums(root
, inode
, comp_bio
,
662 sums
+= (comp_bio
->bi_size
+ root
->sectorsize
- 1) /
665 ret
= btrfs_map_bio(root
, READ
, comp_bio
,
671 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
,
673 comp_bio
->bi_private
= cb
;
674 comp_bio
->bi_end_io
= end_compressed_bio_read
;
676 bio_add_page(comp_bio
, page
, PAGE_CACHE_SIZE
, 0);
678 cur_disk_byte
+= PAGE_CACHE_SIZE
;
682 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
, 0);
685 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
))
686 btrfs_lookup_bio_sums(root
, inode
, comp_bio
, sums
);
688 ret
= btrfs_map_bio(root
, READ
, comp_bio
, mirror_num
, 0);
This page took 0.046058 seconds and 6 git commands to generate.