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/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/bit_spinlock.h>
35 #include <linux/version.h>
36 #include <linux/pagevec.h>
40 #include "transaction.h"
41 #include "btrfs_inode.h"
43 #include "ordered-data.h"
44 #include "compression.h"
45 #include "extent_io.h"
46 #include "extent_map.h"
48 struct compressed_bio
{
49 /* number of bios pending for this compressed extent */
50 atomic_t pending_bios
;
52 /* the pages with the compressed data on them */
53 struct page
**compressed_pages
;
55 /* inode that owns this data */
58 /* starting offset in the inode for our pages */
61 /* number of bytes in the inode we're working on */
64 /* number of bytes on disk */
65 unsigned long compressed_len
;
67 /* number of compressed pages in the array */
68 unsigned long nr_pages
;
74 /* for reads, this is the bio we are copying the data into */
78 * the start of a variable length array of checksums only
84 static inline int compressed_bio_size(struct btrfs_root
*root
,
85 unsigned long disk_size
)
87 u16 csum_size
= btrfs_super_csum_size(&root
->fs_info
->super_copy
);
88 return sizeof(struct compressed_bio
) +
89 ((disk_size
+ root
->sectorsize
- 1) / root
->sectorsize
) *
93 static struct bio
*compressed_bio_alloc(struct block_device
*bdev
,
94 u64 first_byte
, gfp_t gfp_flags
)
99 nr_vecs
= bio_get_nr_vecs(bdev
);
100 bio
= bio_alloc(gfp_flags
, nr_vecs
);
102 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
103 while (!bio
&& (nr_vecs
/= 2))
104 bio
= bio_alloc(gfp_flags
, nr_vecs
);
110 bio
->bi_sector
= first_byte
>> 9;
115 static int check_compressed_csum(struct inode
*inode
,
116 struct compressed_bio
*cb
,
120 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
125 u32
*cb_sum
= &cb
->sums
;
127 if (btrfs_test_flag(inode
, NODATASUM
))
130 for (i
= 0; i
< cb
->nr_pages
; i
++) {
131 page
= cb
->compressed_pages
[i
];
134 kaddr
= kmap_atomic(page
, KM_USER0
);
135 csum
= btrfs_csum_data(root
, kaddr
, csum
, PAGE_CACHE_SIZE
);
136 btrfs_csum_final(csum
, (char *)&csum
);
137 kunmap_atomic(kaddr
, KM_USER0
);
139 if (csum
!= *cb_sum
) {
140 printk(KERN_INFO
"btrfs csum failed ino %lu "
141 "extent %llu csum %u "
142 "wanted %u mirror %d\n", inode
->i_ino
,
143 (unsigned long long)disk_start
,
144 csum
, *cb_sum
, cb
->mirror_num
);
156 /* when we finish reading compressed pages from the disk, we
157 * decompress them and then run the bio end_io routines on the
158 * decompressed pages (in the inode address space).
160 * This allows the checksumming and other IO error handling routines
163 * The compressed pages are freed here, and it must be run
166 static void end_compressed_bio_read(struct bio
*bio
, int err
)
168 struct extent_io_tree
*tree
;
169 struct compressed_bio
*cb
= bio
->bi_private
;
178 /* if there are more bios still pending for this compressed
181 if (!atomic_dec_and_test(&cb
->pending_bios
))
185 ret
= check_compressed_csum(inode
, cb
, (u64
)bio
->bi_sector
<< 9);
189 /* ok, we're the last bio for this extent, lets start
192 tree
= &BTRFS_I(inode
)->io_tree
;
193 ret
= btrfs_zlib_decompress_biovec(cb
->compressed_pages
,
195 cb
->orig_bio
->bi_io_vec
,
196 cb
->orig_bio
->bi_vcnt
,
202 /* release the compressed pages */
204 for (index
= 0; index
< cb
->nr_pages
; index
++) {
205 page
= cb
->compressed_pages
[index
];
206 page
->mapping
= NULL
;
207 page_cache_release(page
);
210 /* do io completion on the original bio */
212 bio_io_error(cb
->orig_bio
);
215 struct bio_vec
*bvec
= cb
->orig_bio
->bi_io_vec
;
218 * we have verified the checksum already, set page
219 * checked so the end_io handlers know about it
221 while (bio_index
< cb
->orig_bio
->bi_vcnt
) {
222 SetPageChecked(bvec
->bv_page
);
226 bio_endio(cb
->orig_bio
, 0);
229 /* finally free the cb struct */
230 kfree(cb
->compressed_pages
);
237 * Clear the writeback bits on all of the file
238 * pages for a compressed write
240 static noinline
int end_compressed_writeback(struct inode
*inode
, u64 start
,
241 unsigned long ram_size
)
243 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
244 unsigned long end_index
= (start
+ ram_size
- 1) >> PAGE_CACHE_SHIFT
;
245 struct page
*pages
[16];
246 unsigned long nr_pages
= end_index
- index
+ 1;
250 while (nr_pages
> 0) {
251 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
253 nr_pages
, ARRAY_SIZE(pages
)), pages
);
259 for (i
= 0; i
< ret
; i
++) {
260 end_page_writeback(pages
[i
]);
261 page_cache_release(pages
[i
]);
266 /* the inode may be gone now */
271 * do the cleanup once all the compressed pages hit the disk.
272 * This will clear writeback on the file pages and free the compressed
275 * This also calls the writeback end hooks for the file pages so that
276 * metadata and checksums can be updated in the file.
278 static void end_compressed_bio_write(struct bio
*bio
, int err
)
280 struct extent_io_tree
*tree
;
281 struct compressed_bio
*cb
= bio
->bi_private
;
289 /* if there are more bios still pending for this compressed
292 if (!atomic_dec_and_test(&cb
->pending_bios
))
295 /* ok, we're the last bio for this extent, step one is to
296 * call back into the FS and do all the end_io operations
299 tree
= &BTRFS_I(inode
)->io_tree
;
300 cb
->compressed_pages
[0]->mapping
= cb
->inode
->i_mapping
;
301 tree
->ops
->writepage_end_io_hook(cb
->compressed_pages
[0],
303 cb
->start
+ cb
->len
- 1,
305 cb
->compressed_pages
[0]->mapping
= NULL
;
307 end_compressed_writeback(inode
, cb
->start
, cb
->len
);
308 /* note, our inode could be gone now */
311 * release the compressed pages, these came from alloc_page and
312 * are not attached to the inode at all
315 for (index
= 0; index
< cb
->nr_pages
; index
++) {
316 page
= cb
->compressed_pages
[index
];
317 page
->mapping
= NULL
;
318 page_cache_release(page
);
321 /* finally free the cb struct */
322 kfree(cb
->compressed_pages
);
329 * worker function to build and submit bios for previously compressed pages.
330 * The corresponding pages in the inode should be marked for writeback
331 * and the compressed pages should have a reference on them for dropping
332 * when the IO is complete.
334 * This also checksums the file bytes and gets things ready for
337 int btrfs_submit_compressed_write(struct inode
*inode
, u64 start
,
338 unsigned long len
, u64 disk_start
,
339 unsigned long compressed_len
,
340 struct page
**compressed_pages
,
341 unsigned long nr_pages
)
343 struct bio
*bio
= NULL
;
344 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
345 struct compressed_bio
*cb
;
346 unsigned long bytes_left
;
347 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
350 u64 first_byte
= disk_start
;
351 struct block_device
*bdev
;
354 WARN_ON(start
& ((u64
)PAGE_CACHE_SIZE
- 1));
355 cb
= kmalloc(compressed_bio_size(root
, compressed_len
), GFP_NOFS
);
356 atomic_set(&cb
->pending_bios
, 0);
362 cb
->compressed_pages
= compressed_pages
;
363 cb
->compressed_len
= compressed_len
;
365 cb
->nr_pages
= nr_pages
;
367 bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
369 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
370 bio
->bi_private
= cb
;
371 bio
->bi_end_io
= end_compressed_bio_write
;
372 atomic_inc(&cb
->pending_bios
);
374 /* create and submit bios for the compressed pages */
375 bytes_left
= compressed_len
;
376 for (page_index
= 0; page_index
< cb
->nr_pages
; page_index
++) {
377 page
= compressed_pages
[page_index
];
378 page
->mapping
= inode
->i_mapping
;
380 ret
= io_tree
->ops
->merge_bio_hook(page
, 0,
386 page
->mapping
= NULL
;
387 if (ret
|| bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0) <
392 * inc the count before we submit the bio so
393 * we know the end IO handler won't happen before
394 * we inc the count. Otherwise, the cb might get
395 * freed before we're done setting it up
397 atomic_inc(&cb
->pending_bios
);
398 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
401 ret
= btrfs_csum_one_bio(root
, inode
, bio
, start
, 1);
404 ret
= btrfs_map_bio(root
, WRITE
, bio
, 0, 1);
409 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
410 bio
->bi_private
= cb
;
411 bio
->bi_end_io
= end_compressed_bio_write
;
412 bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0);
414 if (bytes_left
< PAGE_CACHE_SIZE
) {
415 printk("bytes left %lu compress len %lu nr %lu\n",
416 bytes_left
, cb
->compressed_len
, cb
->nr_pages
);
418 bytes_left
-= PAGE_CACHE_SIZE
;
419 first_byte
+= PAGE_CACHE_SIZE
;
424 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
427 ret
= btrfs_csum_one_bio(root
, inode
, bio
, start
, 1);
430 ret
= btrfs_map_bio(root
, WRITE
, bio
, 0, 1);
437 static noinline
int add_ra_bio_pages(struct inode
*inode
,
439 struct compressed_bio
*cb
)
441 unsigned long end_index
;
442 unsigned long page_index
;
444 u64 isize
= i_size_read(inode
);
447 unsigned long nr_pages
= 0;
448 struct extent_map
*em
;
449 struct address_space
*mapping
= inode
->i_mapping
;
451 struct extent_map_tree
*em_tree
;
452 struct extent_io_tree
*tree
;
456 page
= cb
->orig_bio
->bi_io_vec
[cb
->orig_bio
->bi_vcnt
- 1].bv_page
;
457 last_offset
= (page_offset(page
) + PAGE_CACHE_SIZE
);
458 em_tree
= &BTRFS_I(inode
)->extent_tree
;
459 tree
= &BTRFS_I(inode
)->io_tree
;
464 end_index
= (i_size_read(inode
) - 1) >> PAGE_CACHE_SHIFT
;
466 pagevec_init(&pvec
, 0);
467 while (last_offset
< compressed_end
) {
468 page_index
= last_offset
>> PAGE_CACHE_SHIFT
;
470 if (page_index
> end_index
)
474 page
= radix_tree_lookup(&mapping
->page_tree
, page_index
);
483 page
= alloc_page(mapping_gfp_mask(mapping
) | GFP_NOFS
);
487 page
->index
= page_index
;
489 * what we want to do here is call add_to_page_cache_lru,
490 * but that isn't exported, so we reproduce it here
492 if (add_to_page_cache(page
, mapping
,
493 page
->index
, GFP_NOFS
)) {
494 page_cache_release(page
);
498 /* open coding of lru_cache_add, also not exported */
499 page_cache_get(page
);
500 if (!pagevec_add(&pvec
, page
))
501 __pagevec_lru_add_file(&pvec
);
503 end
= last_offset
+ PAGE_CACHE_SIZE
- 1;
505 * at this point, we have a locked page in the page cache
506 * for these bytes in the file. But, we have to make
507 * sure they map to this compressed extent on disk.
509 set_page_extent_mapped(page
);
510 lock_extent(tree
, last_offset
, end
, GFP_NOFS
);
511 spin_lock(&em_tree
->lock
);
512 em
= lookup_extent_mapping(em_tree
, last_offset
,
514 spin_unlock(&em_tree
->lock
);
516 if (!em
|| last_offset
< em
->start
||
517 (last_offset
+ PAGE_CACHE_SIZE
> extent_map_end(em
)) ||
518 (em
->block_start
>> 9) != cb
->orig_bio
->bi_sector
) {
520 unlock_extent(tree
, last_offset
, end
, GFP_NOFS
);
522 page_cache_release(page
);
527 if (page
->index
== end_index
) {
529 size_t zero_offset
= isize
& (PAGE_CACHE_SIZE
- 1);
533 zeros
= PAGE_CACHE_SIZE
- zero_offset
;
534 userpage
= kmap_atomic(page
, KM_USER0
);
535 memset(userpage
+ zero_offset
, 0, zeros
);
536 flush_dcache_page(page
);
537 kunmap_atomic(userpage
, KM_USER0
);
541 ret
= bio_add_page(cb
->orig_bio
, page
,
544 if (ret
== PAGE_CACHE_SIZE
) {
546 page_cache_release(page
);
548 unlock_extent(tree
, last_offset
, end
, GFP_NOFS
);
550 page_cache_release(page
);
554 last_offset
+= PAGE_CACHE_SIZE
;
556 if (pagevec_count(&pvec
))
557 __pagevec_lru_add_file(&pvec
);
562 * for a compressed read, the bio we get passed has all the inode pages
563 * in it. We don't actually do IO on those pages but allocate new ones
564 * to hold the compressed pages on disk.
566 * bio->bi_sector points to the compressed extent on disk
567 * bio->bi_io_vec points to all of the inode pages
568 * bio->bi_vcnt is a count of pages
570 * After the compressed pages are read, we copy the bytes into the
571 * bio we were passed and then call the bio end_io calls
573 int btrfs_submit_compressed_read(struct inode
*inode
, struct bio
*bio
,
574 int mirror_num
, unsigned long bio_flags
)
576 struct extent_io_tree
*tree
;
577 struct extent_map_tree
*em_tree
;
578 struct compressed_bio
*cb
;
579 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
580 unsigned long uncompressed_len
= bio
->bi_vcnt
* PAGE_CACHE_SIZE
;
581 unsigned long compressed_len
;
582 unsigned long nr_pages
;
583 unsigned long page_index
;
585 struct block_device
*bdev
;
586 struct bio
*comp_bio
;
587 u64 cur_disk_byte
= (u64
)bio
->bi_sector
<< 9;
590 struct extent_map
*em
;
594 tree
= &BTRFS_I(inode
)->io_tree
;
595 em_tree
= &BTRFS_I(inode
)->extent_tree
;
597 /* we need the actual starting offset of this extent in the file */
598 spin_lock(&em_tree
->lock
);
599 em
= lookup_extent_mapping(em_tree
,
600 page_offset(bio
->bi_io_vec
->bv_page
),
602 spin_unlock(&em_tree
->lock
);
604 compressed_len
= em
->block_len
;
605 cb
= kmalloc(compressed_bio_size(root
, compressed_len
), GFP_NOFS
);
606 atomic_set(&cb
->pending_bios
, 0);
609 cb
->mirror_num
= mirror_num
;
612 cb
->start
= em
->orig_start
;
614 em_start
= em
->start
;
619 cb
->len
= uncompressed_len
;
620 cb
->compressed_len
= compressed_len
;
623 nr_pages
= (compressed_len
+ PAGE_CACHE_SIZE
- 1) /
625 cb
->compressed_pages
= kmalloc(sizeof(struct page
*) * nr_pages
,
627 bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
629 for (page_index
= 0; page_index
< nr_pages
; page_index
++) {
630 cb
->compressed_pages
[page_index
] = alloc_page(GFP_NOFS
|
633 cb
->nr_pages
= nr_pages
;
635 add_ra_bio_pages(inode
, em_start
+ em_len
, cb
);
637 /* include any pages we added in add_ra-bio_pages */
638 uncompressed_len
= bio
->bi_vcnt
* PAGE_CACHE_SIZE
;
639 cb
->len
= uncompressed_len
;
641 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
, GFP_NOFS
);
642 comp_bio
->bi_private
= cb
;
643 comp_bio
->bi_end_io
= end_compressed_bio_read
;
644 atomic_inc(&cb
->pending_bios
);
646 for (page_index
= 0; page_index
< nr_pages
; page_index
++) {
647 page
= cb
->compressed_pages
[page_index
];
648 page
->mapping
= inode
->i_mapping
;
649 page
->index
= em_start
>> PAGE_CACHE_SHIFT
;
651 if (comp_bio
->bi_size
)
652 ret
= tree
->ops
->merge_bio_hook(page
, 0,
658 page
->mapping
= NULL
;
659 if (ret
|| bio_add_page(comp_bio
, page
, PAGE_CACHE_SIZE
, 0) <
663 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
, 0);
667 * inc the count before we submit the bio so
668 * we know the end IO handler won't happen before
669 * we inc the count. Otherwise, the cb might get
670 * freed before we're done setting it up
672 atomic_inc(&cb
->pending_bios
);
674 if (!btrfs_test_flag(inode
, NODATASUM
)) {
675 btrfs_lookup_bio_sums(root
, inode
, comp_bio
,
678 sums
+= (comp_bio
->bi_size
+ root
->sectorsize
- 1) /
681 ret
= btrfs_map_bio(root
, READ
, comp_bio
,
687 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
,
689 comp_bio
->bi_private
= cb
;
690 comp_bio
->bi_end_io
= end_compressed_bio_read
;
692 bio_add_page(comp_bio
, page
, PAGE_CACHE_SIZE
, 0);
694 cur_disk_byte
+= PAGE_CACHE_SIZE
;
698 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
, 0);
701 if (!btrfs_test_flag(inode
, NODATASUM
))
702 btrfs_lookup_bio_sums(root
, inode
, comp_bio
, sums
);
704 ret
= btrfs_map_bio(root
, READ
, comp_bio
, mirror_num
, 0);