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 /* the compression algorithm for this bio */
68 /* number of compressed pages in the array */
69 unsigned long nr_pages;
75 /* for reads, this is the bio we are copying the data into */
79 * the start of a variable length array of checksums only
85 static int btrfs_decompress_biovec(int type, struct page **pages_in,
86 u64 disk_start, struct bio_vec *bvec,
87 int vcnt, size_t srclen);
89 static inline int compressed_bio_size(struct btrfs_root *root,
90 unsigned long disk_size)
92 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
94 return sizeof(struct compressed_bio) +
95 ((disk_size + root->sectorsize - 1) / root->sectorsize) *
99 static struct bio *compressed_bio_alloc(struct block_device *bdev,
100 u64 first_byte, gfp_t gfp_flags)
104 nr_vecs = bio_get_nr_vecs(bdev);
105 return btrfs_bio_alloc(bdev, first_byte >> 9, nr_vecs, gfp_flags);
108 static int check_compressed_csum(struct inode *inode,
109 struct compressed_bio *cb,
117 u32 *cb_sum = &cb->sums;
119 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
122 for (i = 0; i < cb->nr_pages; i++) {
123 page = cb->compressed_pages[i];
126 kaddr = kmap_atomic(page);
127 csum = btrfs_csum_data(kaddr, csum, PAGE_CACHE_SIZE);
128 btrfs_csum_final(csum, (char *)&csum);
129 kunmap_atomic(kaddr);
131 if (csum != *cb_sum) {
132 printk(KERN_INFO "btrfs csum failed ino %llu "
133 "extent %llu csum %u "
134 "wanted %u mirror %d\n",
135 (unsigned long long)btrfs_ino(inode),
136 (unsigned long long)disk_start,
137 csum, *cb_sum, cb->mirror_num);
149 /* when we finish reading compressed pages from the disk, we
150 * decompress them and then run the bio end_io routines on the
151 * decompressed pages (in the inode address space).
153 * This allows the checksumming and other IO error handling routines
156 * The compressed pages are freed here, and it must be run
159 static void end_compressed_bio_read(struct bio *bio, int err)
161 struct compressed_bio *cb = bio->bi_private;
170 /* if there are more bios still pending for this compressed
173 if (!atomic_dec_and_test(&cb->pending_bios))
177 ret = check_compressed_csum(inode, cb, (u64)bio->bi_sector << 9);
181 /* ok, we're the last bio for this extent, lets start
184 ret = btrfs_decompress_biovec(cb->compress_type,
185 cb->compressed_pages,
187 cb->orig_bio->bi_io_vec,
188 cb->orig_bio->bi_vcnt,
194 /* release the compressed pages */
196 for (index = 0; index < cb->nr_pages; index++) {
197 page = cb->compressed_pages[index];
198 page->mapping = NULL;
199 page_cache_release(page);
202 /* do io completion on the original bio */
204 bio_io_error(cb->orig_bio);
207 struct bio_vec *bvec = cb->orig_bio->bi_io_vec;
210 * we have verified the checksum already, set page
211 * checked so the end_io handlers know about it
213 while (bio_index < cb->orig_bio->bi_vcnt) {
214 SetPageChecked(bvec->bv_page);
218 bio_endio(cb->orig_bio, 0);
221 /* finally free the cb struct */
222 kfree(cb->compressed_pages);
229 * Clear the writeback bits on all of the file
230 * pages for a compressed write
232 static noinline void end_compressed_writeback(struct inode *inode, u64 start,
233 unsigned long ram_size)
235 unsigned long index = start >> PAGE_CACHE_SHIFT;
236 unsigned long end_index = (start + ram_size - 1) >> PAGE_CACHE_SHIFT;
237 struct page *pages[16];
238 unsigned long nr_pages = end_index - index + 1;
242 while (nr_pages > 0) {
243 ret = find_get_pages_contig(inode->i_mapping, index,
245 nr_pages, ARRAY_SIZE(pages)), pages);
251 for (i = 0; i < ret; i++) {
252 end_page_writeback(pages[i]);
253 page_cache_release(pages[i]);
258 /* the inode may be gone now */
262 * do the cleanup once all the compressed pages hit the disk.
263 * This will clear writeback on the file pages and free the compressed
266 * This also calls the writeback end hooks for the file pages so that
267 * metadata and checksums can be updated in the file.
269 static void end_compressed_bio_write(struct bio *bio, int err)
271 struct extent_io_tree *tree;
272 struct compressed_bio *cb = bio->bi_private;
280 /* if there are more bios still pending for this compressed
283 if (!atomic_dec_and_test(&cb->pending_bios))
286 /* ok, we're the last bio for this extent, step one is to
287 * call back into the FS and do all the end_io operations
290 tree = &BTRFS_I(inode)->io_tree;
291 cb->compressed_pages[0]->mapping = cb->inode->i_mapping;
292 tree->ops->writepage_end_io_hook(cb->compressed_pages[0],
294 cb->start + cb->len - 1,
296 cb->compressed_pages[0]->mapping = NULL;
298 end_compressed_writeback(inode, cb->start, cb->len);
299 /* note, our inode could be gone now */
302 * release the compressed pages, these came from alloc_page and
303 * are not attached to the inode at all
306 for (index = 0; index < cb->nr_pages; index++) {
307 page = cb->compressed_pages[index];
308 page->mapping = NULL;
309 page_cache_release(page);
312 /* finally free the cb struct */
313 kfree(cb->compressed_pages);
320 * worker function to build and submit bios for previously compressed pages.
321 * The corresponding pages in the inode should be marked for writeback
322 * and the compressed pages should have a reference on them for dropping
323 * when the IO is complete.
325 * This also checksums the file bytes and gets things ready for
328 int btrfs_submit_compressed_write(struct inode *inode, u64 start,
329 unsigned long len, u64 disk_start,
330 unsigned long compressed_len,
331 struct page **compressed_pages,
332 unsigned long nr_pages)
334 struct bio *bio = NULL;
335 struct btrfs_root *root = BTRFS_I(inode)->root;
336 struct compressed_bio *cb;
337 unsigned long bytes_left;
338 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
341 u64 first_byte = disk_start;
342 struct block_device *bdev;
344 int skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
346 WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1));
347 cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
350 atomic_set(&cb->pending_bios, 0);
356 cb->compressed_pages = compressed_pages;
357 cb->compressed_len = compressed_len;
359 cb->nr_pages = nr_pages;
361 bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
363 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 (pg_index = 0; pg_index < cb->nr_pages; pg_index++) {
375 page = compressed_pages[pg_index];
376 page->mapping = inode->i_mapping;
378 ret = io_tree->ops->merge_bio_hook(WRITE, 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);
397 BUG_ON(ret); /* -ENOMEM */
400 ret = btrfs_csum_one_bio(root, inode, bio,
402 BUG_ON(ret); /* -ENOMEM */
405 ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
406 BUG_ON(ret); /* -ENOMEM */
410 bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
412 bio->bi_private = cb;
413 bio->bi_end_io = end_compressed_bio_write;
414 bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
416 if (bytes_left < PAGE_CACHE_SIZE) {
417 printk("bytes left %lu compress len %lu nr %lu\n",
418 bytes_left, cb->compressed_len, cb->nr_pages);
420 bytes_left -= PAGE_CACHE_SIZE;
421 first_byte += PAGE_CACHE_SIZE;
426 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
427 BUG_ON(ret); /* -ENOMEM */
430 ret = btrfs_csum_one_bio(root, inode, bio, start, 1);
431 BUG_ON(ret); /* -ENOMEM */
434 ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
435 BUG_ON(ret); /* -ENOMEM */
441 static noinline int add_ra_bio_pages(struct inode *inode,
443 struct compressed_bio *cb)
445 unsigned long end_index;
446 unsigned long pg_index;
448 u64 isize = i_size_read(inode);
451 unsigned long nr_pages = 0;
452 struct extent_map *em;
453 struct address_space *mapping = inode->i_mapping;
454 struct extent_map_tree *em_tree;
455 struct extent_io_tree *tree;
459 page = cb->orig_bio->bi_io_vec[cb->orig_bio->bi_vcnt - 1].bv_page;
460 last_offset = (page_offset(page) + PAGE_CACHE_SIZE);
461 em_tree = &BTRFS_I(inode)->extent_tree;
462 tree = &BTRFS_I(inode)->io_tree;
467 end_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
469 while (last_offset < compressed_end) {
470 pg_index = last_offset >> PAGE_CACHE_SHIFT;
472 if (pg_index > end_index)
476 page = radix_tree_lookup(&mapping->page_tree, pg_index);
485 page = __page_cache_alloc(mapping_gfp_mask(mapping) &
490 if (add_to_page_cache_lru(page, mapping, pg_index,
492 page_cache_release(page);
496 end = last_offset + PAGE_CACHE_SIZE - 1;
498 * at this point, we have a locked page in the page cache
499 * for these bytes in the file. But, we have to make
500 * sure they map to this compressed extent on disk.
502 set_page_extent_mapped(page);
503 lock_extent(tree, last_offset, end);
504 read_lock(&em_tree->lock);
505 em = lookup_extent_mapping(em_tree, last_offset,
507 read_unlock(&em_tree->lock);
509 if (!em || last_offset < em->start ||
510 (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) ||
511 (em->block_start >> 9) != cb->orig_bio->bi_sector) {
513 unlock_extent(tree, last_offset, end);
515 page_cache_release(page);
520 if (page->index == end_index) {
522 size_t zero_offset = isize & (PAGE_CACHE_SIZE - 1);
526 zeros = PAGE_CACHE_SIZE - zero_offset;
527 userpage = kmap_atomic(page);
528 memset(userpage + zero_offset, 0, zeros);
529 flush_dcache_page(page);
530 kunmap_atomic(userpage);
534 ret = bio_add_page(cb->orig_bio, page,
537 if (ret == PAGE_CACHE_SIZE) {
539 page_cache_release(page);
541 unlock_extent(tree, last_offset, end);
543 page_cache_release(page);
547 last_offset += PAGE_CACHE_SIZE;
553 * for a compressed read, the bio we get passed has all the inode pages
554 * in it. We don't actually do IO on those pages but allocate new ones
555 * to hold the compressed pages on disk.
557 * bio->bi_sector points to the compressed extent on disk
558 * bio->bi_io_vec points to all of the inode pages
559 * bio->bi_vcnt is a count of pages
561 * After the compressed pages are read, we copy the bytes into the
562 * bio we were passed and then call the bio end_io calls
564 int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
565 int mirror_num, unsigned long bio_flags)
567 struct extent_io_tree *tree;
568 struct extent_map_tree *em_tree;
569 struct compressed_bio *cb;
570 struct btrfs_root *root = BTRFS_I(inode)->root;
571 unsigned long uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
572 unsigned long compressed_len;
573 unsigned long nr_pages;
574 unsigned long pg_index;
576 struct block_device *bdev;
577 struct bio *comp_bio;
578 u64 cur_disk_byte = (u64)bio->bi_sector << 9;
581 struct extent_map *em;
586 tree = &BTRFS_I(inode)->io_tree;
587 em_tree = &BTRFS_I(inode)->extent_tree;
589 /* we need the actual starting offset of this extent in the file */
590 read_lock(&em_tree->lock);
591 em = lookup_extent_mapping(em_tree,
592 page_offset(bio->bi_io_vec->bv_page),
594 read_unlock(&em_tree->lock);
598 compressed_len = em->block_len;
599 cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
603 atomic_set(&cb->pending_bios, 0);
606 cb->mirror_num = mirror_num;
609 cb->start = em->orig_start;
611 em_start = em->start;
616 cb->len = uncompressed_len;
617 cb->compressed_len = compressed_len;
618 cb->compress_type = extent_compress_type(bio_flags);
621 nr_pages = (compressed_len + PAGE_CACHE_SIZE - 1) /
623 cb->compressed_pages = kzalloc(sizeof(struct page *) * nr_pages,
625 if (!cb->compressed_pages)
628 bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
630 for (pg_index = 0; pg_index < nr_pages; pg_index++) {
631 cb->compressed_pages[pg_index] = alloc_page(GFP_NOFS |
633 if (!cb->compressed_pages[pg_index]) {
634 faili = pg_index - 1;
639 faili = nr_pages - 1;
640 cb->nr_pages = nr_pages;
642 add_ra_bio_pages(inode, em_start + em_len, cb);
644 /* include any pages we added in add_ra-bio_pages */
645 uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
646 cb->len = uncompressed_len;
648 comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS);
651 comp_bio->bi_private = cb;
652 comp_bio->bi_end_io = end_compressed_bio_read;
653 atomic_inc(&cb->pending_bios);
655 for (pg_index = 0; pg_index < nr_pages; pg_index++) {
656 page = cb->compressed_pages[pg_index];
657 page->mapping = inode->i_mapping;
658 page->index = em_start >> PAGE_CACHE_SHIFT;
660 if (comp_bio->bi_size)
661 ret = tree->ops->merge_bio_hook(READ, page, 0,
667 page->mapping = NULL;
668 if (ret || bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0) <
672 ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
673 BUG_ON(ret); /* -ENOMEM */
676 * inc the count before we submit the bio so
677 * we know the end IO handler won't happen before
678 * we inc the count. Otherwise, the cb might get
679 * freed before we're done setting it up
681 atomic_inc(&cb->pending_bios);
683 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
684 ret = btrfs_lookup_bio_sums(root, inode,
686 BUG_ON(ret); /* -ENOMEM */
688 sums += (comp_bio->bi_size + root->sectorsize - 1) /
691 ret = btrfs_map_bio(root, READ, comp_bio,
694 bio_endio(comp_bio, ret);
698 comp_bio = compressed_bio_alloc(bdev, cur_disk_byte,
701 comp_bio->bi_private = cb;
702 comp_bio->bi_end_io = end_compressed_bio_read;
704 bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0);
706 cur_disk_byte += PAGE_CACHE_SIZE;
710 ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
711 BUG_ON(ret); /* -ENOMEM */
713 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
714 ret = btrfs_lookup_bio_sums(root, inode, comp_bio, sums);
715 BUG_ON(ret); /* -ENOMEM */
718 ret = btrfs_map_bio(root, READ, comp_bio, mirror_num, 0);
720 bio_endio(comp_bio, ret);
727 __free_page(cb->compressed_pages[faili]);
731 kfree(cb->compressed_pages);
739 static struct list_head comp_idle_workspace[BTRFS_COMPRESS_TYPES];
740 static spinlock_t comp_workspace_lock[BTRFS_COMPRESS_TYPES];
741 static int comp_num_workspace[BTRFS_COMPRESS_TYPES];
742 static atomic_t comp_alloc_workspace[BTRFS_COMPRESS_TYPES];
743 static wait_queue_head_t comp_workspace_wait[BTRFS_COMPRESS_TYPES];
745 static struct btrfs_compress_op *btrfs_compress_op[] = {
746 &btrfs_zlib_compress,
750 void __init btrfs_init_compress(void)
754 for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) {
755 INIT_LIST_HEAD(&comp_idle_workspace[i]);
756 spin_lock_init(&comp_workspace_lock[i]);
757 atomic_set(&comp_alloc_workspace[i], 0);
758 init_waitqueue_head(&comp_workspace_wait[i]);
763 * this finds an available workspace or allocates a new one
764 * ERR_PTR is returned if things go bad.
766 static struct list_head *find_workspace(int type)
768 struct list_head *workspace;
769 int cpus = num_online_cpus();
772 struct list_head *idle_workspace = &comp_idle_workspace[idx];
773 spinlock_t *workspace_lock = &comp_workspace_lock[idx];
774 atomic_t *alloc_workspace = &comp_alloc_workspace[idx];
775 wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx];
776 int *num_workspace = &comp_num_workspace[idx];
778 spin_lock(workspace_lock);
779 if (!list_empty(idle_workspace)) {
780 workspace = idle_workspace->next;
783 spin_unlock(workspace_lock);
787 if (atomic_read(alloc_workspace) > cpus) {
790 spin_unlock(workspace_lock);
791 prepare_to_wait(workspace_wait, &wait, TASK_UNINTERRUPTIBLE);
792 if (atomic_read(alloc_workspace) > cpus && !*num_workspace)
794 finish_wait(workspace_wait, &wait);
797 atomic_inc(alloc_workspace);
798 spin_unlock(workspace_lock);
800 workspace = btrfs_compress_op[idx]->alloc_workspace();
801 if (IS_ERR(workspace)) {
802 atomic_dec(alloc_workspace);
803 wake_up(workspace_wait);
809 * put a workspace struct back on the list or free it if we have enough
810 * idle ones sitting around
812 static void free_workspace(int type, struct list_head *workspace)
815 struct list_head *idle_workspace = &comp_idle_workspace[idx];
816 spinlock_t *workspace_lock = &comp_workspace_lock[idx];
817 atomic_t *alloc_workspace = &comp_alloc_workspace[idx];
818 wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx];
819 int *num_workspace = &comp_num_workspace[idx];
821 spin_lock(workspace_lock);
822 if (*num_workspace < num_online_cpus()) {
823 list_add_tail(workspace, idle_workspace);
825 spin_unlock(workspace_lock);
828 spin_unlock(workspace_lock);
830 btrfs_compress_op[idx]->free_workspace(workspace);
831 atomic_dec(alloc_workspace);
834 if (waitqueue_active(workspace_wait))
835 wake_up(workspace_wait);
839 * cleanup function for module exit
841 static void free_workspaces(void)
843 struct list_head *workspace;
846 for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) {
847 while (!list_empty(&comp_idle_workspace[i])) {
848 workspace = comp_idle_workspace[i].next;
850 btrfs_compress_op[i]->free_workspace(workspace);
851 atomic_dec(&comp_alloc_workspace[i]);
857 * given an address space and start/len, compress the bytes.
859 * pages are allocated to hold the compressed result and stored
862 * out_pages is used to return the number of pages allocated. There
863 * may be pages allocated even if we return an error
865 * total_in is used to return the number of bytes actually read. It
866 * may be smaller then len if we had to exit early because we
867 * ran out of room in the pages array or because we cross the
870 * total_out is used to return the total number of compressed bytes
872 * max_out tells us the max number of bytes that we're allowed to
875 int btrfs_compress_pages(int type, struct address_space *mapping,
876 u64 start, unsigned long len,
878 unsigned long nr_dest_pages,
879 unsigned long *out_pages,
880 unsigned long *total_in,
881 unsigned long *total_out,
882 unsigned long max_out)
884 struct list_head *workspace;
887 workspace = find_workspace(type);
888 if (IS_ERR(workspace))
891 ret = btrfs_compress_op[type-1]->compress_pages(workspace, mapping,
893 nr_dest_pages, out_pages,
896 free_workspace(type, workspace);
901 * pages_in is an array of pages with compressed data.
903 * disk_start is the starting logical offset of this array in the file
905 * bvec is a bio_vec of pages from the file that we want to decompress into
907 * vcnt is the count of pages in the biovec
909 * srclen is the number of bytes in pages_in
911 * The basic idea is that we have a bio that was created by readpages.
912 * The pages in the bio are for the uncompressed data, and they may not
913 * be contiguous. They all correspond to the range of bytes covered by
914 * the compressed extent.
916 static int btrfs_decompress_biovec(int type, struct page **pages_in,
917 u64 disk_start, struct bio_vec *bvec,
918 int vcnt, size_t srclen)
920 struct list_head *workspace;
923 workspace = find_workspace(type);
924 if (IS_ERR(workspace))
927 ret = btrfs_compress_op[type-1]->decompress_biovec(workspace, pages_in,
930 free_workspace(type, workspace);
935 * a less complex decompression routine. Our compressed data fits in a
936 * single page, and we want to read a single page out of it.
937 * start_byte tells us the offset into the compressed data we're interested in
939 int btrfs_decompress(int type, unsigned char *data_in, struct page *dest_page,
940 unsigned long start_byte, size_t srclen, size_t destlen)
942 struct list_head *workspace;
945 workspace = find_workspace(type);
946 if (IS_ERR(workspace))
949 ret = btrfs_compress_op[type-1]->decompress(workspace, data_in,
950 dest_page, start_byte,
953 free_workspace(type, workspace);
957 void btrfs_exit_compress(void)
963 * Copy uncompressed data from working buffer to pages.
965 * buf_start is the byte offset we're of the start of our workspace buffer.
967 * total_out is the last byte of the buffer
969 int btrfs_decompress_buf2page(char *buf, unsigned long buf_start,
970 unsigned long total_out, u64 disk_start,
971 struct bio_vec *bvec, int vcnt,
972 unsigned long *pg_index,
973 unsigned long *pg_offset)
975 unsigned long buf_offset;
976 unsigned long current_buf_start;
977 unsigned long start_byte;
978 unsigned long working_bytes = total_out - buf_start;
981 struct page *page_out = bvec[*pg_index].bv_page;
984 * start byte is the first byte of the page we're currently
985 * copying into relative to the start of the compressed data.
987 start_byte = page_offset(page_out) - disk_start;
989 /* we haven't yet hit data corresponding to this page */
990 if (total_out <= start_byte)
994 * the start of the data we care about is offset into
995 * the middle of our working buffer
997 if (total_out > start_byte && buf_start < start_byte) {
998 buf_offset = start_byte - buf_start;
999 working_bytes -= buf_offset;
1003 current_buf_start = buf_start;
1005 /* copy bytes from the working buffer into the pages */
1006 while (working_bytes > 0) {
1007 bytes = min(PAGE_CACHE_SIZE - *pg_offset,
1008 PAGE_CACHE_SIZE - buf_offset);
1009 bytes = min(bytes, working_bytes);
1010 kaddr = kmap_atomic(page_out);
1011 memcpy(kaddr + *pg_offset, buf + buf_offset, bytes);
1012 kunmap_atomic(kaddr);
1013 flush_dcache_page(page_out);
1015 *pg_offset += bytes;
1016 buf_offset += bytes;
1017 working_bytes -= bytes;
1018 current_buf_start += bytes;
1020 /* check if we need to pick another page */
1021 if (*pg_offset == PAGE_CACHE_SIZE) {
1023 if (*pg_index >= vcnt)
1026 page_out = bvec[*pg_index].bv_page;
1028 start_byte = page_offset(page_out) - disk_start;
1031 * make sure our new page is covered by this
1034 if (total_out <= start_byte)
1038 * the next page in the biovec might not be adjacent
1039 * to the last page, but it might still be found
1040 * inside this working buffer. bump our offset pointer
1042 if (total_out > start_byte &&
1043 current_buf_start < start_byte) {
1044 buf_offset = start_byte - buf_start;
1045 working_bytes = total_out - start_byte;
1046 current_buf_start = buf_start + buf_offset;