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>
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
)
96 nr_vecs
= bio_get_nr_vecs(bdev
);
97 return btrfs_bio_alloc(bdev
, first_byte
>> 9, nr_vecs
, gfp_flags
);
100 static int check_compressed_csum(struct inode
*inode
,
101 struct compressed_bio
*cb
,
105 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
110 u32
*cb_sum
= &cb
->sums
;
112 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
115 for (i
= 0; i
< cb
->nr_pages
; i
++) {
116 page
= cb
->compressed_pages
[i
];
119 kaddr
= kmap_atomic(page
, KM_USER0
);
120 csum
= btrfs_csum_data(root
, kaddr
, csum
, PAGE_CACHE_SIZE
);
121 btrfs_csum_final(csum
, (char *)&csum
);
122 kunmap_atomic(kaddr
, KM_USER0
);
124 if (csum
!= *cb_sum
) {
125 printk(KERN_INFO
"btrfs csum failed ino %lu "
126 "extent %llu csum %u "
127 "wanted %u mirror %d\n", inode
->i_ino
,
128 (unsigned long long)disk_start
,
129 csum
, *cb_sum
, cb
->mirror_num
);
141 /* when we finish reading compressed pages from the disk, we
142 * decompress them and then run the bio end_io routines on the
143 * decompressed pages (in the inode address space).
145 * This allows the checksumming and other IO error handling routines
148 * The compressed pages are freed here, and it must be run
151 static void end_compressed_bio_read(struct bio
*bio
, int err
)
153 struct compressed_bio
*cb
= bio
->bi_private
;
162 /* if there are more bios still pending for this compressed
165 if (!atomic_dec_and_test(&cb
->pending_bios
))
169 ret
= check_compressed_csum(inode
, cb
, (u64
)bio
->bi_sector
<< 9);
173 /* ok, we're the last bio for this extent, lets start
176 ret
= btrfs_zlib_decompress_biovec(cb
->compressed_pages
,
178 cb
->orig_bio
->bi_io_vec
,
179 cb
->orig_bio
->bi_vcnt
,
185 /* release the compressed pages */
187 for (index
= 0; index
< cb
->nr_pages
; index
++) {
188 page
= cb
->compressed_pages
[index
];
189 page
->mapping
= NULL
;
190 page_cache_release(page
);
193 /* do io completion on the original bio */
195 bio_io_error(cb
->orig_bio
);
198 struct bio_vec
*bvec
= cb
->orig_bio
->bi_io_vec
;
201 * we have verified the checksum already, set page
202 * checked so the end_io handlers know about it
204 while (bio_index
< cb
->orig_bio
->bi_vcnt
) {
205 SetPageChecked(bvec
->bv_page
);
209 bio_endio(cb
->orig_bio
, 0);
212 /* finally free the cb struct */
213 kfree(cb
->compressed_pages
);
220 * Clear the writeback bits on all of the file
221 * pages for a compressed write
223 static noinline
int end_compressed_writeback(struct inode
*inode
, u64 start
,
224 unsigned long ram_size
)
226 unsigned long index
= start
>> PAGE_CACHE_SHIFT
;
227 unsigned long end_index
= (start
+ ram_size
- 1) >> PAGE_CACHE_SHIFT
;
228 struct page
*pages
[16];
229 unsigned long nr_pages
= end_index
- index
+ 1;
233 while (nr_pages
> 0) {
234 ret
= find_get_pages_contig(inode
->i_mapping
, index
,
236 nr_pages
, ARRAY_SIZE(pages
)), pages
);
242 for (i
= 0; i
< ret
; i
++) {
243 end_page_writeback(pages
[i
]);
244 page_cache_release(pages
[i
]);
249 /* the inode may be gone now */
254 * do the cleanup once all the compressed pages hit the disk.
255 * This will clear writeback on the file pages and free the compressed
258 * This also calls the writeback end hooks for the file pages so that
259 * metadata and checksums can be updated in the file.
261 static void end_compressed_bio_write(struct bio
*bio
, int err
)
263 struct extent_io_tree
*tree
;
264 struct compressed_bio
*cb
= bio
->bi_private
;
272 /* if there are more bios still pending for this compressed
275 if (!atomic_dec_and_test(&cb
->pending_bios
))
278 /* ok, we're the last bio for this extent, step one is to
279 * call back into the FS and do all the end_io operations
282 tree
= &BTRFS_I(inode
)->io_tree
;
283 cb
->compressed_pages
[0]->mapping
= cb
->inode
->i_mapping
;
284 tree
->ops
->writepage_end_io_hook(cb
->compressed_pages
[0],
286 cb
->start
+ cb
->len
- 1,
288 cb
->compressed_pages
[0]->mapping
= NULL
;
290 end_compressed_writeback(inode
, cb
->start
, cb
->len
);
291 /* note, our inode could be gone now */
294 * release the compressed pages, these came from alloc_page and
295 * are not attached to the inode at all
298 for (index
= 0; index
< cb
->nr_pages
; index
++) {
299 page
= cb
->compressed_pages
[index
];
300 page
->mapping
= NULL
;
301 page_cache_release(page
);
304 /* finally free the cb struct */
305 kfree(cb
->compressed_pages
);
312 * worker function to build and submit bios for previously compressed pages.
313 * The corresponding pages in the inode should be marked for writeback
314 * and the compressed pages should have a reference on them for dropping
315 * when the IO is complete.
317 * This also checksums the file bytes and gets things ready for
320 int btrfs_submit_compressed_write(struct inode
*inode
, u64 start
,
321 unsigned long len
, u64 disk_start
,
322 unsigned long compressed_len
,
323 struct page
**compressed_pages
,
324 unsigned long nr_pages
)
326 struct bio
*bio
= NULL
;
327 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
328 struct compressed_bio
*cb
;
329 unsigned long bytes_left
;
330 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
333 u64 first_byte
= disk_start
;
334 struct block_device
*bdev
;
337 WARN_ON(start
& ((u64
)PAGE_CACHE_SIZE
- 1));
338 cb
= kmalloc(compressed_bio_size(root
, compressed_len
), GFP_NOFS
);
339 atomic_set(&cb
->pending_bios
, 0);
345 cb
->compressed_pages
= compressed_pages
;
346 cb
->compressed_len
= compressed_len
;
348 cb
->nr_pages
= nr_pages
;
350 bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
352 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
353 bio
->bi_private
= cb
;
354 bio
->bi_end_io
= end_compressed_bio_write
;
355 atomic_inc(&cb
->pending_bios
);
357 /* create and submit bios for the compressed pages */
358 bytes_left
= compressed_len
;
359 for (page_index
= 0; page_index
< cb
->nr_pages
; page_index
++) {
360 page
= compressed_pages
[page_index
];
361 page
->mapping
= inode
->i_mapping
;
363 ret
= io_tree
->ops
->merge_bio_hook(page
, 0,
369 page
->mapping
= NULL
;
370 if (ret
|| bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0) <
375 * inc the count before we submit the bio so
376 * we know the end IO handler won't happen before
377 * we inc the count. Otherwise, the cb might get
378 * freed before we're done setting it up
380 atomic_inc(&cb
->pending_bios
);
381 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
384 ret
= btrfs_csum_one_bio(root
, inode
, bio
, start
, 1);
387 ret
= btrfs_map_bio(root
, WRITE
, bio
, 0, 1);
392 bio
= compressed_bio_alloc(bdev
, first_byte
, GFP_NOFS
);
393 bio
->bi_private
= cb
;
394 bio
->bi_end_io
= end_compressed_bio_write
;
395 bio_add_page(bio
, page
, PAGE_CACHE_SIZE
, 0);
397 if (bytes_left
< PAGE_CACHE_SIZE
) {
398 printk("bytes left %lu compress len %lu nr %lu\n",
399 bytes_left
, cb
->compressed_len
, cb
->nr_pages
);
401 bytes_left
-= PAGE_CACHE_SIZE
;
402 first_byte
+= PAGE_CACHE_SIZE
;
407 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
410 ret
= btrfs_csum_one_bio(root
, inode
, bio
, start
, 1);
413 ret
= btrfs_map_bio(root
, WRITE
, bio
, 0, 1);
420 static noinline
int add_ra_bio_pages(struct inode
*inode
,
422 struct compressed_bio
*cb
)
424 unsigned long end_index
;
425 unsigned long page_index
;
427 u64 isize
= i_size_read(inode
);
430 unsigned long nr_pages
= 0;
431 struct extent_map
*em
;
432 struct address_space
*mapping
= inode
->i_mapping
;
433 struct extent_map_tree
*em_tree
;
434 struct extent_io_tree
*tree
;
438 page
= cb
->orig_bio
->bi_io_vec
[cb
->orig_bio
->bi_vcnt
- 1].bv_page
;
439 last_offset
= (page_offset(page
) + PAGE_CACHE_SIZE
);
440 em_tree
= &BTRFS_I(inode
)->extent_tree
;
441 tree
= &BTRFS_I(inode
)->io_tree
;
446 end_index
= (i_size_read(inode
) - 1) >> PAGE_CACHE_SHIFT
;
448 while (last_offset
< compressed_end
) {
449 page_index
= last_offset
>> PAGE_CACHE_SHIFT
;
451 if (page_index
> end_index
)
455 page
= radix_tree_lookup(&mapping
->page_tree
, page_index
);
464 page
= __page_cache_alloc(mapping_gfp_mask(mapping
) &
469 if (add_to_page_cache_lru(page
, mapping
, page_index
,
471 page_cache_release(page
);
475 end
= last_offset
+ PAGE_CACHE_SIZE
- 1;
477 * at this point, we have a locked page in the page cache
478 * for these bytes in the file. But, we have to make
479 * sure they map to this compressed extent on disk.
481 set_page_extent_mapped(page
);
482 lock_extent(tree
, last_offset
, end
, GFP_NOFS
);
483 read_lock(&em_tree
->lock
);
484 em
= lookup_extent_mapping(em_tree
, last_offset
,
486 read_unlock(&em_tree
->lock
);
488 if (!em
|| last_offset
< em
->start
||
489 (last_offset
+ PAGE_CACHE_SIZE
> extent_map_end(em
)) ||
490 (em
->block_start
>> 9) != cb
->orig_bio
->bi_sector
) {
492 unlock_extent(tree
, last_offset
, end
, GFP_NOFS
);
494 page_cache_release(page
);
499 if (page
->index
== end_index
) {
501 size_t zero_offset
= isize
& (PAGE_CACHE_SIZE
- 1);
505 zeros
= PAGE_CACHE_SIZE
- zero_offset
;
506 userpage
= kmap_atomic(page
, KM_USER0
);
507 memset(userpage
+ zero_offset
, 0, zeros
);
508 flush_dcache_page(page
);
509 kunmap_atomic(userpage
, KM_USER0
);
513 ret
= bio_add_page(cb
->orig_bio
, page
,
516 if (ret
== PAGE_CACHE_SIZE
) {
518 page_cache_release(page
);
520 unlock_extent(tree
, last_offset
, end
, GFP_NOFS
);
522 page_cache_release(page
);
526 last_offset
+= PAGE_CACHE_SIZE
;
532 * for a compressed read, the bio we get passed has all the inode pages
533 * in it. We don't actually do IO on those pages but allocate new ones
534 * to hold the compressed pages on disk.
536 * bio->bi_sector points to the compressed extent on disk
537 * bio->bi_io_vec points to all of the inode pages
538 * bio->bi_vcnt is a count of pages
540 * After the compressed pages are read, we copy the bytes into the
541 * bio we were passed and then call the bio end_io calls
543 int btrfs_submit_compressed_read(struct inode
*inode
, struct bio
*bio
,
544 int mirror_num
, unsigned long bio_flags
)
546 struct extent_io_tree
*tree
;
547 struct extent_map_tree
*em_tree
;
548 struct compressed_bio
*cb
;
549 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
550 unsigned long uncompressed_len
= bio
->bi_vcnt
* PAGE_CACHE_SIZE
;
551 unsigned long compressed_len
;
552 unsigned long nr_pages
;
553 unsigned long page_index
;
555 struct block_device
*bdev
;
556 struct bio
*comp_bio
;
557 u64 cur_disk_byte
= (u64
)bio
->bi_sector
<< 9;
560 struct extent_map
*em
;
564 tree
= &BTRFS_I(inode
)->io_tree
;
565 em_tree
= &BTRFS_I(inode
)->extent_tree
;
567 /* we need the actual starting offset of this extent in the file */
568 read_lock(&em_tree
->lock
);
569 em
= lookup_extent_mapping(em_tree
,
570 page_offset(bio
->bi_io_vec
->bv_page
),
572 read_unlock(&em_tree
->lock
);
574 compressed_len
= em
->block_len
;
575 cb
= kmalloc(compressed_bio_size(root
, compressed_len
), GFP_NOFS
);
576 atomic_set(&cb
->pending_bios
, 0);
579 cb
->mirror_num
= mirror_num
;
582 cb
->start
= em
->orig_start
;
584 em_start
= em
->start
;
589 cb
->len
= uncompressed_len
;
590 cb
->compressed_len
= compressed_len
;
593 nr_pages
= (compressed_len
+ PAGE_CACHE_SIZE
- 1) /
595 cb
->compressed_pages
= kmalloc(sizeof(struct page
*) * nr_pages
,
597 bdev
= BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
;
599 for (page_index
= 0; page_index
< nr_pages
; page_index
++) {
600 cb
->compressed_pages
[page_index
] = alloc_page(GFP_NOFS
|
603 cb
->nr_pages
= nr_pages
;
605 add_ra_bio_pages(inode
, em_start
+ em_len
, cb
);
607 /* include any pages we added in add_ra-bio_pages */
608 uncompressed_len
= bio
->bi_vcnt
* PAGE_CACHE_SIZE
;
609 cb
->len
= uncompressed_len
;
611 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
, GFP_NOFS
);
612 comp_bio
->bi_private
= cb
;
613 comp_bio
->bi_end_io
= end_compressed_bio_read
;
614 atomic_inc(&cb
->pending_bios
);
616 for (page_index
= 0; page_index
< nr_pages
; page_index
++) {
617 page
= cb
->compressed_pages
[page_index
];
618 page
->mapping
= inode
->i_mapping
;
619 page
->index
= em_start
>> PAGE_CACHE_SHIFT
;
621 if (comp_bio
->bi_size
)
622 ret
= tree
->ops
->merge_bio_hook(page
, 0,
628 page
->mapping
= NULL
;
629 if (ret
|| bio_add_page(comp_bio
, page
, PAGE_CACHE_SIZE
, 0) <
633 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
, 0);
637 * inc the count before we submit the bio so
638 * we know the end IO handler won't happen before
639 * we inc the count. Otherwise, the cb might get
640 * freed before we're done setting it up
642 atomic_inc(&cb
->pending_bios
);
644 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
645 btrfs_lookup_bio_sums(root
, inode
, comp_bio
,
648 sums
+= (comp_bio
->bi_size
+ root
->sectorsize
- 1) /
651 ret
= btrfs_map_bio(root
, READ
, comp_bio
,
657 comp_bio
= compressed_bio_alloc(bdev
, cur_disk_byte
,
659 comp_bio
->bi_private
= cb
;
660 comp_bio
->bi_end_io
= end_compressed_bio_read
;
662 bio_add_page(comp_bio
, page
, PAGE_CACHE_SIZE
, 0);
664 cur_disk_byte
+= PAGE_CACHE_SIZE
;
668 ret
= btrfs_bio_wq_end_io(root
->fs_info
, comp_bio
, 0);
671 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
))
672 btrfs_lookup_bio_sums(root
, inode
, comp_bio
, sums
);
674 ret
= btrfs_map_bio(root
, READ
, comp_bio
, mirror_num
, 0);