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)
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 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 ret = btrfs_zlib_decompress_biovec(cb->compressed_pages,
191 cb->orig_bio->bi_io_vec,
192 cb->orig_bio->bi_vcnt,
198 /* release the compressed pages */
200 for (index = 0; index < cb->nr_pages; index++) {
201 page = cb->compressed_pages[index];
202 page->mapping = NULL;
203 page_cache_release(page);
206 /* do io completion on the original bio */
208 bio_io_error(cb->orig_bio);
211 struct bio_vec *bvec = cb->orig_bio->bi_io_vec;
214 * we have verified the checksum already, set page
215 * checked so the end_io handlers know about it
217 while (bio_index < cb->orig_bio->bi_vcnt) {
218 SetPageChecked(bvec->bv_page);
222 bio_endio(cb->orig_bio, 0);
225 /* finally free the cb struct */
226 kfree(cb->compressed_pages);
233 * Clear the writeback bits on all of the file
234 * pages for a compressed write
236 static noinline int end_compressed_writeback(struct inode *inode, u64 start,
237 unsigned long ram_size)
239 unsigned long index = start >> PAGE_CACHE_SHIFT;
240 unsigned long end_index = (start + ram_size - 1) >> PAGE_CACHE_SHIFT;
241 struct page *pages[16];
242 unsigned long nr_pages = end_index - index + 1;
246 while (nr_pages > 0) {
247 ret = find_get_pages_contig(inode->i_mapping, index,
249 nr_pages, ARRAY_SIZE(pages)), pages);
255 for (i = 0; i < ret; i++) {
256 end_page_writeback(pages[i]);
257 page_cache_release(pages[i]);
262 /* the inode may be gone now */
267 * do the cleanup once all the compressed pages hit the disk.
268 * This will clear writeback on the file pages and free the compressed
271 * This also calls the writeback end hooks for the file pages so that
272 * metadata and checksums can be updated in the file.
274 static void end_compressed_bio_write(struct bio *bio, int err)
276 struct extent_io_tree *tree;
277 struct compressed_bio *cb = bio->bi_private;
285 /* if there are more bios still pending for this compressed
288 if (!atomic_dec_and_test(&cb->pending_bios))
291 /* ok, we're the last bio for this extent, step one is to
292 * call back into the FS and do all the end_io operations
295 tree = &BTRFS_I(inode)->io_tree;
296 cb->compressed_pages[0]->mapping = cb->inode->i_mapping;
297 tree->ops->writepage_end_io_hook(cb->compressed_pages[0],
299 cb->start + cb->len - 1,
301 cb->compressed_pages[0]->mapping = NULL;
303 end_compressed_writeback(inode, cb->start, cb->len);
304 /* note, our inode could be gone now */
307 * release the compressed pages, these came from alloc_page and
308 * are not attached to the inode at all
311 for (index = 0; index < cb->nr_pages; index++) {
312 page = cb->compressed_pages[index];
313 page->mapping = NULL;
314 page_cache_release(page);
317 /* finally free the cb struct */
318 kfree(cb->compressed_pages);
325 * worker function to build and submit bios for previously compressed pages.
326 * The corresponding pages in the inode should be marked for writeback
327 * and the compressed pages should have a reference on them for dropping
328 * when the IO is complete.
330 * This also checksums the file bytes and gets things ready for
333 int btrfs_submit_compressed_write(struct inode *inode, u64 start,
334 unsigned long len, u64 disk_start,
335 unsigned long compressed_len,
336 struct page **compressed_pages,
337 unsigned long nr_pages)
339 struct bio *bio = NULL;
340 struct btrfs_root *root = BTRFS_I(inode)->root;
341 struct compressed_bio *cb;
342 unsigned long bytes_left;
343 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
346 u64 first_byte = disk_start;
347 struct block_device *bdev;
350 WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1));
351 cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
352 atomic_set(&cb->pending_bios, 0);
358 cb->compressed_pages = compressed_pages;
359 cb->compressed_len = compressed_len;
361 cb->nr_pages = nr_pages;
363 bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
365 bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
366 bio->bi_private = cb;
367 bio->bi_end_io = end_compressed_bio_write;
368 atomic_inc(&cb->pending_bios);
370 /* create and submit bios for the compressed pages */
371 bytes_left = compressed_len;
372 for (page_index = 0; page_index < cb->nr_pages; page_index++) {
373 page = compressed_pages[page_index];
374 page->mapping = inode->i_mapping;
376 ret = io_tree->ops->merge_bio_hook(page, 0,
382 page->mapping = NULL;
383 if (ret || bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) <
388 * inc the count before we submit the bio so
389 * we know the end IO handler won't happen before
390 * we inc the count. Otherwise, the cb might get
391 * freed before we're done setting it up
393 atomic_inc(&cb->pending_bios);
394 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
397 ret = btrfs_csum_one_bio(root, inode, bio, start, 1);
400 ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
405 bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
406 bio->bi_private = cb;
407 bio->bi_end_io = end_compressed_bio_write;
408 bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
410 if (bytes_left < PAGE_CACHE_SIZE) {
411 printk("bytes left %lu compress len %lu nr %lu\n",
412 bytes_left, cb->compressed_len, cb->nr_pages);
414 bytes_left -= PAGE_CACHE_SIZE;
415 first_byte += PAGE_CACHE_SIZE;
420 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
423 ret = btrfs_csum_one_bio(root, inode, bio, start, 1);
426 ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
433 static noinline int add_ra_bio_pages(struct inode *inode,
435 struct compressed_bio *cb)
437 unsigned long end_index;
438 unsigned long page_index;
440 u64 isize = i_size_read(inode);
443 unsigned long nr_pages = 0;
444 struct extent_map *em;
445 struct address_space *mapping = inode->i_mapping;
446 struct extent_map_tree *em_tree;
447 struct extent_io_tree *tree;
451 page = cb->orig_bio->bi_io_vec[cb->orig_bio->bi_vcnt - 1].bv_page;
452 last_offset = (page_offset(page) + PAGE_CACHE_SIZE);
453 em_tree = &BTRFS_I(inode)->extent_tree;
454 tree = &BTRFS_I(inode)->io_tree;
459 end_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
461 while (last_offset < compressed_end) {
462 page_index = last_offset >> PAGE_CACHE_SHIFT;
464 if (page_index > end_index)
468 page = radix_tree_lookup(&mapping->page_tree, page_index);
477 page = __page_cache_alloc(mapping_gfp_mask(mapping) &
482 if (add_to_page_cache_lru(page, mapping, page_index,
484 page_cache_release(page);
488 end = last_offset + PAGE_CACHE_SIZE - 1;
490 * at this point, we have a locked page in the page cache
491 * for these bytes in the file. But, we have to make
492 * sure they map to this compressed extent on disk.
494 set_page_extent_mapped(page);
495 lock_extent(tree, last_offset, end, GFP_NOFS);
496 read_lock(&em_tree->lock);
497 em = lookup_extent_mapping(em_tree, last_offset,
499 read_unlock(&em_tree->lock);
501 if (!em || last_offset < em->start ||
502 (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) ||
503 (em->block_start >> 9) != cb->orig_bio->bi_sector) {
505 unlock_extent(tree, last_offset, end, GFP_NOFS);
507 page_cache_release(page);
512 if (page->index == end_index) {
514 size_t zero_offset = isize & (PAGE_CACHE_SIZE - 1);
518 zeros = PAGE_CACHE_SIZE - zero_offset;
519 userpage = kmap_atomic(page, KM_USER0);
520 memset(userpage + zero_offset, 0, zeros);
521 flush_dcache_page(page);
522 kunmap_atomic(userpage, KM_USER0);
526 ret = bio_add_page(cb->orig_bio, page,
529 if (ret == PAGE_CACHE_SIZE) {
531 page_cache_release(page);
533 unlock_extent(tree, last_offset, end, GFP_NOFS);
535 page_cache_release(page);
539 last_offset += PAGE_CACHE_SIZE;
545 * for a compressed read, the bio we get passed has all the inode pages
546 * in it. We don't actually do IO on those pages but allocate new ones
547 * to hold the compressed pages on disk.
549 * bio->bi_sector points to the compressed extent on disk
550 * bio->bi_io_vec points to all of the inode pages
551 * bio->bi_vcnt is a count of pages
553 * After the compressed pages are read, we copy the bytes into the
554 * bio we were passed and then call the bio end_io calls
556 int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
557 int mirror_num, unsigned long bio_flags)
559 struct extent_io_tree *tree;
560 struct extent_map_tree *em_tree;
561 struct compressed_bio *cb;
562 struct btrfs_root *root = BTRFS_I(inode)->root;
563 unsigned long uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
564 unsigned long compressed_len;
565 unsigned long nr_pages;
566 unsigned long page_index;
568 struct block_device *bdev;
569 struct bio *comp_bio;
570 u64 cur_disk_byte = (u64)bio->bi_sector << 9;
573 struct extent_map *em;
577 tree = &BTRFS_I(inode)->io_tree;
578 em_tree = &BTRFS_I(inode)->extent_tree;
580 /* we need the actual starting offset of this extent in the file */
581 read_lock(&em_tree->lock);
582 em = lookup_extent_mapping(em_tree,
583 page_offset(bio->bi_io_vec->bv_page),
585 read_unlock(&em_tree->lock);
587 compressed_len = em->block_len;
588 cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
589 atomic_set(&cb->pending_bios, 0);
592 cb->mirror_num = mirror_num;
595 cb->start = em->orig_start;
597 em_start = em->start;
602 cb->len = uncompressed_len;
603 cb->compressed_len = compressed_len;
606 nr_pages = (compressed_len + PAGE_CACHE_SIZE - 1) /
608 cb->compressed_pages = kmalloc(sizeof(struct page *) * nr_pages,
610 bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
612 for (page_index = 0; page_index < nr_pages; page_index++) {
613 cb->compressed_pages[page_index] = alloc_page(GFP_NOFS |
616 cb->nr_pages = nr_pages;
618 add_ra_bio_pages(inode, em_start + em_len, cb);
620 /* include any pages we added in add_ra-bio_pages */
621 uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
622 cb->len = uncompressed_len;
624 comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS);
625 comp_bio->bi_private = cb;
626 comp_bio->bi_end_io = end_compressed_bio_read;
627 atomic_inc(&cb->pending_bios);
629 for (page_index = 0; page_index < nr_pages; page_index++) {
630 page = cb->compressed_pages[page_index];
631 page->mapping = inode->i_mapping;
632 page->index = em_start >> PAGE_CACHE_SHIFT;
634 if (comp_bio->bi_size)
635 ret = tree->ops->merge_bio_hook(page, 0,
641 page->mapping = NULL;
642 if (ret || bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0) <
646 ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
650 * inc the count before we submit the bio so
651 * we know the end IO handler won't happen before
652 * we inc the count. Otherwise, the cb might get
653 * freed before we're done setting it up
655 atomic_inc(&cb->pending_bios);
657 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
658 btrfs_lookup_bio_sums(root, inode, comp_bio,
661 sums += (comp_bio->bi_size + root->sectorsize - 1) /
664 ret = btrfs_map_bio(root, READ, comp_bio,
670 comp_bio = compressed_bio_alloc(bdev, cur_disk_byte,
672 comp_bio->bi_private = cb;
673 comp_bio->bi_end_io = end_compressed_bio_read;
675 bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0);
677 cur_disk_byte += PAGE_CACHE_SIZE;
681 ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
684 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM))
685 btrfs_lookup_bio_sums(root, inode, comp_bio, sums);
687 ret = btrfs_map_bio(root, READ, comp_bio, mirror_num, 0);