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>
39 #include "transaction.h"
40 #include "btrfs_inode.h"
42 #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;
73 /* for reads, this is the bio we are copying the data into */
77 static struct bio *compressed_bio_alloc(struct block_device *bdev,
78 u64 first_byte, gfp_t gfp_flags)
83 nr_vecs = bio_get_nr_vecs(bdev);
84 bio = bio_alloc(gfp_flags, nr_vecs);
86 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
87 while (!bio && (nr_vecs /= 2))
88 bio = bio_alloc(gfp_flags, nr_vecs);
94 bio->bi_sector = first_byte >> 9;
99 /* when we finish reading compressed pages from the disk, we
100 * decompress them and then run the bio end_io routines on the
101 * decompressed pages (in the inode address space).
103 * This allows the checksumming and other IO error handling routines
106 * The compressed pages are freed here, and it must be run
109 static void end_compressed_bio_read(struct bio *bio, int err)
111 struct extent_io_tree *tree;
112 struct compressed_bio *cb = bio->bi_private;
121 /* if there are more bios still pending for this compressed
124 if (!atomic_dec_and_test(&cb->pending_bios))
127 /* ok, we're the last bio for this extent, lets start
131 tree = &BTRFS_I(inode)->io_tree;
132 ret = btrfs_zlib_decompress_biovec(cb->compressed_pages,
134 cb->orig_bio->bi_io_vec,
135 cb->orig_bio->bi_vcnt,
140 /* release the compressed pages */
142 for (index = 0; index < cb->nr_pages; index++) {
143 page = cb->compressed_pages[index];
144 page->mapping = NULL;
145 page_cache_release(page);
148 /* do io completion on the original bio */
150 bio_io_error(cb->orig_bio);
152 bio_endio(cb->orig_bio, 0);
154 /* finally free the cb struct */
155 kfree(cb->compressed_pages);
162 * Clear the writeback bits on all of the file
163 * pages for a compressed write
165 static noinline int end_compressed_writeback(struct inode *inode, u64 start,
166 unsigned long ram_size)
168 unsigned long index = start >> PAGE_CACHE_SHIFT;
169 unsigned long end_index = (start + ram_size - 1) >> PAGE_CACHE_SHIFT;
170 struct page *pages[16];
171 unsigned long nr_pages = end_index - index + 1;
175 while(nr_pages > 0) {
176 ret = find_get_pages_contig(inode->i_mapping, index,
178 nr_pages, ARRAY_SIZE(pages)), pages);
184 for (i = 0; i < ret; i++) {
185 end_page_writeback(pages[i]);
186 page_cache_release(pages[i]);
191 /* the inode may be gone now */
196 * do the cleanup once all the compressed pages hit the disk.
197 * This will clear writeback on the file pages and free the compressed
200 * This also calls the writeback end hooks for the file pages so that
201 * metadata and checksums can be updated in the file.
203 static void end_compressed_bio_write(struct bio *bio, int err)
205 struct extent_io_tree *tree;
206 struct compressed_bio *cb = bio->bi_private;
214 /* if there are more bios still pending for this compressed
217 if (!atomic_dec_and_test(&cb->pending_bios))
220 /* ok, we're the last bio for this extent, step one is to
221 * call back into the FS and do all the end_io operations
224 tree = &BTRFS_I(inode)->io_tree;
225 cb->compressed_pages[0]->mapping = cb->inode->i_mapping;
226 tree->ops->writepage_end_io_hook(cb->compressed_pages[0],
228 cb->start + cb->len - 1,
230 cb->compressed_pages[0]->mapping = NULL;
232 end_compressed_writeback(inode, cb->start, cb->len);
233 /* note, our inode could be gone now */
236 * release the compressed pages, these came from alloc_page and
237 * are not attached to the inode at all
240 for (index = 0; index < cb->nr_pages; index++) {
241 page = cb->compressed_pages[index];
242 page->mapping = NULL;
243 page_cache_release(page);
246 /* finally free the cb struct */
247 kfree(cb->compressed_pages);
254 * worker function to build and submit bios for previously compressed pages.
255 * The corresponding pages in the inode should be marked for writeback
256 * and the compressed pages should have a reference on them for dropping
257 * when the IO is complete.
259 * This also checksums the file bytes and gets things ready for
262 int btrfs_submit_compressed_write(struct inode *inode, u64 start,
263 unsigned long len, u64 disk_start,
264 unsigned long compressed_len,
265 struct page **compressed_pages,
266 unsigned long nr_pages)
268 struct bio *bio = NULL;
269 struct btrfs_root *root = BTRFS_I(inode)->root;
270 struct compressed_bio *cb;
271 unsigned long bytes_left;
272 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
275 u64 first_byte = disk_start;
276 struct block_device *bdev;
279 WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1));
280 cb = kmalloc(sizeof(*cb), GFP_NOFS);
281 atomic_set(&cb->pending_bios, 0);
286 cb->compressed_pages = compressed_pages;
287 cb->compressed_len = compressed_len;
289 cb->nr_pages = nr_pages;
291 bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
293 ret = btrfs_csum_file_bytes(root, inode, start, len);
296 bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
297 bio->bi_private = cb;
298 bio->bi_end_io = end_compressed_bio_write;
299 atomic_inc(&cb->pending_bios);
301 /* create and submit bios for the compressed pages */
302 bytes_left = compressed_len;
303 for (page_index = 0; page_index < cb->nr_pages; page_index++) {
304 page = compressed_pages[page_index];
305 page->mapping = inode->i_mapping;
307 ret = io_tree->ops->merge_bio_hook(page, 0,
313 page->mapping = NULL;
314 if (ret || bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) <
319 * inc the count before we submit the bio so
320 * we know the end IO handler won't happen before
321 * we inc the count. Otherwise, the cb might get
322 * freed before we're done setting it up
324 atomic_inc(&cb->pending_bios);
325 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
328 ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
333 bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
334 bio->bi_private = cb;
335 bio->bi_end_io = end_compressed_bio_write;
336 bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
338 if (bytes_left < PAGE_CACHE_SIZE) {
339 printk("bytes left %lu compress len %lu nr %lu\n",
340 bytes_left, cb->compressed_len, cb->nr_pages);
342 bytes_left -= PAGE_CACHE_SIZE;
343 first_byte += PAGE_CACHE_SIZE;
348 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
351 ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
358 static noinline int add_ra_bio_pages(struct inode *inode,
360 struct compressed_bio *cb)
362 unsigned long end_index;
363 unsigned long page_index;
365 u64 isize = i_size_read(inode);
368 unsigned long nr_pages = 0;
369 struct extent_map *em;
370 struct address_space *mapping = inode->i_mapping;
372 struct extent_map_tree *em_tree;
373 struct extent_io_tree *tree;
377 page = cb->orig_bio->bi_io_vec[cb->orig_bio->bi_vcnt - 1].bv_page;
378 last_offset = (page_offset(page) + PAGE_CACHE_SIZE);
379 em_tree = &BTRFS_I(inode)->extent_tree;
380 tree = &BTRFS_I(inode)->io_tree;
385 end_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
387 pagevec_init(&pvec, 0);
388 while(last_offset < compressed_end) {
389 page_index = last_offset >> PAGE_CACHE_SHIFT;
391 if (page_index > end_index)
395 page = radix_tree_lookup(&mapping->page_tree, page_index);
404 page = alloc_page(mapping_gfp_mask(mapping) | GFP_NOFS);
408 page->index = page_index;
410 * what we want to do here is call add_to_page_cache_lru,
411 * but that isn't exported, so we reproduce it here
413 if (add_to_page_cache(page, mapping,
414 page->index, GFP_NOFS)) {
415 page_cache_release(page);
419 /* open coding of lru_cache_add, also not exported */
420 page_cache_get(page);
421 if (!pagevec_add(&pvec, page))
422 __pagevec_lru_add(&pvec);
424 end = last_offset + PAGE_CACHE_SIZE - 1;
426 * at this point, we have a locked page in the page cache
427 * for these bytes in the file. But, we have to make
428 * sure they map to this compressed extent on disk.
430 set_page_extent_mapped(page);
431 lock_extent(tree, last_offset, end, GFP_NOFS);
432 spin_lock(&em_tree->lock);
433 em = lookup_extent_mapping(em_tree, last_offset,
435 spin_unlock(&em_tree->lock);
437 if (!em || last_offset < em->start ||
438 (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) ||
439 (em->block_start >> 9) != cb->orig_bio->bi_sector) {
441 unlock_extent(tree, last_offset, end, GFP_NOFS);
443 page_cache_release(page);
448 if (page->index == end_index) {
450 size_t zero_offset = isize & (PAGE_CACHE_SIZE - 1);
454 zeros = PAGE_CACHE_SIZE - zero_offset;
455 userpage = kmap_atomic(page, KM_USER0);
456 memset(userpage + zero_offset, 0, zeros);
457 flush_dcache_page(page);
458 kunmap_atomic(userpage, KM_USER0);
462 ret = bio_add_page(cb->orig_bio, page,
465 if (ret == PAGE_CACHE_SIZE) {
467 page_cache_release(page);
469 unlock_extent(tree, last_offset, end, GFP_NOFS);
471 page_cache_release(page);
475 last_offset += PAGE_CACHE_SIZE;
477 if (pagevec_count(&pvec))
478 __pagevec_lru_add(&pvec);
483 * for a compressed read, the bio we get passed has all the inode pages
484 * in it. We don't actually do IO on those pages but allocate new ones
485 * to hold the compressed pages on disk.
487 * bio->bi_sector points to the compressed extent on disk
488 * bio->bi_io_vec points to all of the inode pages
489 * bio->bi_vcnt is a count of pages
491 * After the compressed pages are read, we copy the bytes into the
492 * bio we were passed and then call the bio end_io calls
494 int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
495 int mirror_num, unsigned long bio_flags)
497 struct extent_io_tree *tree;
498 struct extent_map_tree *em_tree;
499 struct compressed_bio *cb;
500 struct btrfs_root *root = BTRFS_I(inode)->root;
501 unsigned long uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
502 unsigned long compressed_len;
503 unsigned long nr_pages;
504 unsigned long page_index;
506 struct block_device *bdev;
507 struct bio *comp_bio;
508 u64 cur_disk_byte = (u64)bio->bi_sector << 9;
511 struct extent_map *em;
514 tree = &BTRFS_I(inode)->io_tree;
515 em_tree = &BTRFS_I(inode)->extent_tree;
517 /* we need the actual starting offset of this extent in the file */
518 spin_lock(&em_tree->lock);
519 em = lookup_extent_mapping(em_tree,
520 page_offset(bio->bi_io_vec->bv_page),
522 spin_unlock(&em_tree->lock);
524 cb = kmalloc(sizeof(*cb), GFP_NOFS);
525 atomic_set(&cb->pending_bios, 0);
529 cb->start = em->orig_start;
530 compressed_len = em->block_len;
532 em_start = em->start;
536 cb->len = uncompressed_len;
537 cb->compressed_len = compressed_len;
540 nr_pages = (compressed_len + PAGE_CACHE_SIZE - 1) /
542 cb->compressed_pages = kmalloc(sizeof(struct page *) * nr_pages,
544 bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
546 for (page_index = 0; page_index < nr_pages; page_index++) {
547 cb->compressed_pages[page_index] = alloc_page(GFP_NOFS |
550 cb->nr_pages = nr_pages;
552 add_ra_bio_pages(inode, em_start + em_len, cb);
554 if (!btrfs_test_opt(root, NODATASUM) &&
555 !btrfs_test_flag(inode, NODATASUM)) {
556 btrfs_lookup_bio_sums(root, inode, cb->orig_bio);
559 /* include any pages we added in add_ra-bio_pages */
560 uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
561 cb->len = uncompressed_len;
563 comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS);
564 comp_bio->bi_private = cb;
565 comp_bio->bi_end_io = end_compressed_bio_read;
566 atomic_inc(&cb->pending_bios);
568 for (page_index = 0; page_index < nr_pages; page_index++) {
569 page = cb->compressed_pages[page_index];
570 page->mapping = inode->i_mapping;
571 if (comp_bio->bi_size)
572 ret = tree->ops->merge_bio_hook(page, 0,
578 page->mapping = NULL;
579 if (ret || bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0) <
583 ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
587 * inc the count before we submit the bio so
588 * we know the end IO handler won't happen before
589 * we inc the count. Otherwise, the cb might get
590 * freed before we're done setting it up
592 atomic_inc(&cb->pending_bios);
594 ret = btrfs_map_bio(root, READ, comp_bio, 0, 0);
599 comp_bio = compressed_bio_alloc(bdev, cur_disk_byte,
601 comp_bio->bi_private = cb;
602 comp_bio->bi_end_io = end_compressed_bio_read;
604 bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0);
606 cur_disk_byte += PAGE_CACHE_SIZE;
610 ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
613 ret = btrfs_map_bio(root, READ, comp_bio, 0, 0);