blob: d3220d31d3cbf0e653898d15816f5895045c06d5 [file] [log] [blame]
Chris Masonc8b97812008-10-29 14:49:59 -04001/*
2 * Copyright (C) 2008 Oracle. All rights reserved.
3 *
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.
7 *
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.
12 *
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.
17 */
18
19#include <linux/kernel.h>
20#include <linux/bio.h>
21#include <linux/buffer_head.h>
22#include <linux/file.h>
23#include <linux/fs.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>
Chris Masonc8b97812008-10-29 14:49:59 -040029#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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090034#include <linux/slab.h>
Chris Masonc8b97812008-10-29 14:49:59 -040035#include "ctree.h"
36#include "disk-io.h"
37#include "transaction.h"
38#include "btrfs_inode.h"
39#include "volumes.h"
40#include "ordered-data.h"
Chris Masonc8b97812008-10-29 14:49:59 -040041#include "compression.h"
42#include "extent_io.h"
43#include "extent_map.h"
44
45struct compressed_bio {
46 /* number of bios pending for this compressed extent */
47 atomic_t pending_bios;
48
49 /* the pages with the compressed data on them */
50 struct page **compressed_pages;
51
52 /* inode that owns this data */
53 struct inode *inode;
54
55 /* starting offset in the inode for our pages */
56 u64 start;
57
58 /* number of bytes in the inode we're working on */
59 unsigned long len;
60
61 /* number of bytes on disk */
62 unsigned long compressed_len;
63
Li Zefan261507a02010-12-17 14:21:50 +080064 /* the compression algorithm for this bio */
65 int compress_type;
66
Chris Masonc8b97812008-10-29 14:49:59 -040067 /* number of compressed pages in the array */
68 unsigned long nr_pages;
69
70 /* IO errors */
71 int errors;
Chris Masond20f7042008-12-08 16:58:54 -050072 int mirror_num;
Chris Masonc8b97812008-10-29 14:49:59 -040073
74 /* for reads, this is the bio we are copying the data into */
75 struct bio *orig_bio;
Chris Masond20f7042008-12-08 16:58:54 -050076
77 /*
78 * the start of a variable length array of checksums only
79 * used by reads
80 */
81 u32 sums;
Chris Masonc8b97812008-10-29 14:49:59 -040082};
83
Eric Sandeen48a3b632013-04-25 20:41:01 +000084static int btrfs_decompress_biovec(int type, struct page **pages_in,
85 u64 disk_start, struct bio_vec *bvec,
86 int vcnt, size_t srclen);
87
Chris Masond20f7042008-12-08 16:58:54 -050088static inline int compressed_bio_size(struct btrfs_root *root,
89 unsigned long disk_size)
90{
David Sterba6c417612011-04-13 15:41:04 +020091 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
92
Chris Masond20f7042008-12-08 16:58:54 -050093 return sizeof(struct compressed_bio) +
David Sterbaed6078f2014-06-05 01:59:57 +020094 (DIV_ROUND_UP(disk_size, root->sectorsize)) * csum_size;
Chris Masond20f7042008-12-08 16:58:54 -050095}
96
Chris Masonc8b97812008-10-29 14:49:59 -040097static struct bio *compressed_bio_alloc(struct block_device *bdev,
98 u64 first_byte, gfp_t gfp_flags)
99{
Chris Masonc8b97812008-10-29 14:49:59 -0400100 int nr_vecs;
101
102 nr_vecs = bio_get_nr_vecs(bdev);
Miao Xie88f794e2010-11-22 03:02:55 +0000103 return btrfs_bio_alloc(bdev, first_byte >> 9, nr_vecs, gfp_flags);
Chris Masonc8b97812008-10-29 14:49:59 -0400104}
105
Chris Masond20f7042008-12-08 16:58:54 -0500106static int check_compressed_csum(struct inode *inode,
107 struct compressed_bio *cb,
108 u64 disk_start)
109{
110 int ret;
Chris Masond20f7042008-12-08 16:58:54 -0500111 struct page *page;
112 unsigned long i;
113 char *kaddr;
114 u32 csum;
115 u32 *cb_sum = &cb->sums;
116
Christoph Hellwig6cbff002009-04-17 10:37:41 +0200117 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
Chris Masond20f7042008-12-08 16:58:54 -0500118 return 0;
119
120 for (i = 0; i < cb->nr_pages; i++) {
121 page = cb->compressed_pages[i];
122 csum = ~(u32)0;
123
Cong Wang7ac687d2011-11-25 23:14:28 +0800124 kaddr = kmap_atomic(page);
Liu Bob0496682013-03-14 14:57:45 +0000125 csum = btrfs_csum_data(kaddr, csum, PAGE_CACHE_SIZE);
Chris Masond20f7042008-12-08 16:58:54 -0500126 btrfs_csum_final(csum, (char *)&csum);
Cong Wang7ac687d2011-11-25 23:14:28 +0800127 kunmap_atomic(kaddr);
Chris Masond20f7042008-12-08 16:58:54 -0500128
129 if (csum != *cb_sum) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500130 btrfs_info(BTRFS_I(inode)->root->fs_info,
131 "csum failed ino %llu extent %llu csum %u wanted %u mirror %d",
132 btrfs_ino(inode), disk_start, csum, *cb_sum,
133 cb->mirror_num);
Chris Masond20f7042008-12-08 16:58:54 -0500134 ret = -EIO;
135 goto fail;
136 }
137 cb_sum++;
138
139 }
140 ret = 0;
141fail:
142 return ret;
143}
144
Chris Masonc8b97812008-10-29 14:49:59 -0400145/* when we finish reading compressed pages from the disk, we
146 * decompress them and then run the bio end_io routines on the
147 * decompressed pages (in the inode address space).
148 *
149 * This allows the checksumming and other IO error handling routines
150 * to work normally
151 *
152 * The compressed pages are freed here, and it must be run
153 * in process context
154 */
155static void end_compressed_bio_read(struct bio *bio, int err)
156{
Chris Masonc8b97812008-10-29 14:49:59 -0400157 struct compressed_bio *cb = bio->bi_private;
158 struct inode *inode;
159 struct page *page;
160 unsigned long index;
161 int ret;
162
163 if (err)
164 cb->errors = 1;
165
166 /* if there are more bios still pending for this compressed
167 * extent, just exit
168 */
169 if (!atomic_dec_and_test(&cb->pending_bios))
170 goto out;
171
Chris Masond20f7042008-12-08 16:58:54 -0500172 inode = cb->inode;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700173 ret = check_compressed_csum(inode, cb,
174 (u64)bio->bi_iter.bi_sector << 9);
Chris Masond20f7042008-12-08 16:58:54 -0500175 if (ret)
176 goto csum_failed;
177
Chris Masonc8b97812008-10-29 14:49:59 -0400178 /* ok, we're the last bio for this extent, lets start
179 * the decompression.
180 */
Li Zefan261507a02010-12-17 14:21:50 +0800181 ret = btrfs_decompress_biovec(cb->compress_type,
182 cb->compressed_pages,
183 cb->start,
184 cb->orig_bio->bi_io_vec,
185 cb->orig_bio->bi_vcnt,
186 cb->compressed_len);
Chris Masond20f7042008-12-08 16:58:54 -0500187csum_failed:
Chris Masonc8b97812008-10-29 14:49:59 -0400188 if (ret)
189 cb->errors = 1;
190
191 /* release the compressed pages */
192 index = 0;
193 for (index = 0; index < cb->nr_pages; index++) {
194 page = cb->compressed_pages[index];
195 page->mapping = NULL;
196 page_cache_release(page);
197 }
198
199 /* do io completion on the original bio */
Chris Mason771ed682008-11-06 22:02:51 -0500200 if (cb->errors) {
Chris Masonc8b97812008-10-29 14:49:59 -0400201 bio_io_error(cb->orig_bio);
Chris Masond20f7042008-12-08 16:58:54 -0500202 } else {
Kent Overstreet2c30c712013-11-07 12:20:26 -0800203 int i;
204 struct bio_vec *bvec;
Chris Masond20f7042008-12-08 16:58:54 -0500205
206 /*
207 * we have verified the checksum already, set page
208 * checked so the end_io handlers know about it
209 */
Kent Overstreet2c30c712013-11-07 12:20:26 -0800210 bio_for_each_segment_all(bvec, cb->orig_bio, i)
Chris Masond20f7042008-12-08 16:58:54 -0500211 SetPageChecked(bvec->bv_page);
Kent Overstreet2c30c712013-11-07 12:20:26 -0800212
Chris Masonc8b97812008-10-29 14:49:59 -0400213 bio_endio(cb->orig_bio, 0);
Chris Masond20f7042008-12-08 16:58:54 -0500214 }
Chris Masonc8b97812008-10-29 14:49:59 -0400215
216 /* finally free the cb struct */
217 kfree(cb->compressed_pages);
218 kfree(cb);
219out:
220 bio_put(bio);
221}
222
223/*
224 * Clear the writeback bits on all of the file
225 * pages for a compressed write
226 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100227static noinline void end_compressed_writeback(struct inode *inode, u64 start,
228 unsigned long ram_size)
Chris Masonc8b97812008-10-29 14:49:59 -0400229{
230 unsigned long index = start >> PAGE_CACHE_SHIFT;
231 unsigned long end_index = (start + ram_size - 1) >> PAGE_CACHE_SHIFT;
232 struct page *pages[16];
233 unsigned long nr_pages = end_index - index + 1;
234 int i;
235 int ret;
236
Chris Masond3977122009-01-05 21:25:51 -0500237 while (nr_pages > 0) {
Chris Masonc8b97812008-10-29 14:49:59 -0400238 ret = find_get_pages_contig(inode->i_mapping, index,
Chris Mason5b050f02008-11-11 09:34:41 -0500239 min_t(unsigned long,
240 nr_pages, ARRAY_SIZE(pages)), pages);
Chris Masonc8b97812008-10-29 14:49:59 -0400241 if (ret == 0) {
242 nr_pages -= 1;
243 index += 1;
244 continue;
245 }
246 for (i = 0; i < ret; i++) {
247 end_page_writeback(pages[i]);
248 page_cache_release(pages[i]);
249 }
250 nr_pages -= ret;
251 index += ret;
252 }
253 /* the inode may be gone now */
Chris Masonc8b97812008-10-29 14:49:59 -0400254}
255
256/*
257 * do the cleanup once all the compressed pages hit the disk.
258 * This will clear writeback on the file pages and free the compressed
259 * pages.
260 *
261 * This also calls the writeback end hooks for the file pages so that
262 * metadata and checksums can be updated in the file.
263 */
264static void end_compressed_bio_write(struct bio *bio, int err)
265{
266 struct extent_io_tree *tree;
267 struct compressed_bio *cb = bio->bi_private;
268 struct inode *inode;
269 struct page *page;
270 unsigned long index;
271
272 if (err)
273 cb->errors = 1;
274
275 /* if there are more bios still pending for this compressed
276 * extent, just exit
277 */
278 if (!atomic_dec_and_test(&cb->pending_bios))
279 goto out;
280
281 /* ok, we're the last bio for this extent, step one is to
282 * call back into the FS and do all the end_io operations
283 */
284 inode = cb->inode;
285 tree = &BTRFS_I(inode)->io_tree;
Chris Mason70b99e62008-10-31 12:46:39 -0400286 cb->compressed_pages[0]->mapping = cb->inode->i_mapping;
Chris Masonc8b97812008-10-29 14:49:59 -0400287 tree->ops->writepage_end_io_hook(cb->compressed_pages[0],
288 cb->start,
289 cb->start + cb->len - 1,
290 NULL, 1);
Chris Mason70b99e62008-10-31 12:46:39 -0400291 cb->compressed_pages[0]->mapping = NULL;
Chris Masonc8b97812008-10-29 14:49:59 -0400292
293 end_compressed_writeback(inode, cb->start, cb->len);
294 /* note, our inode could be gone now */
295
296 /*
297 * release the compressed pages, these came from alloc_page and
298 * are not attached to the inode at all
299 */
300 index = 0;
301 for (index = 0; index < cb->nr_pages; index++) {
302 page = cb->compressed_pages[index];
303 page->mapping = NULL;
304 page_cache_release(page);
305 }
306
307 /* finally free the cb struct */
308 kfree(cb->compressed_pages);
309 kfree(cb);
310out:
311 bio_put(bio);
312}
313
314/*
315 * worker function to build and submit bios for previously compressed pages.
316 * The corresponding pages in the inode should be marked for writeback
317 * and the compressed pages should have a reference on them for dropping
318 * when the IO is complete.
319 *
320 * This also checksums the file bytes and gets things ready for
321 * the end io hooks.
322 */
323int btrfs_submit_compressed_write(struct inode *inode, u64 start,
324 unsigned long len, u64 disk_start,
325 unsigned long compressed_len,
326 struct page **compressed_pages,
327 unsigned long nr_pages)
328{
329 struct bio *bio = NULL;
330 struct btrfs_root *root = BTRFS_I(inode)->root;
331 struct compressed_bio *cb;
332 unsigned long bytes_left;
333 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
David Sterba306e16c2011-04-19 14:29:38 +0200334 int pg_index = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400335 struct page *page;
336 u64 first_byte = disk_start;
337 struct block_device *bdev;
338 int ret;
Li Zefane55179b2011-07-14 03:16:47 +0000339 int skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Chris Masonc8b97812008-10-29 14:49:59 -0400340
341 WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1));
Chris Masond20f7042008-12-08 16:58:54 -0500342 cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
Yoshinori Sanodac97e512011-02-15 12:01:42 +0000343 if (!cb)
344 return -ENOMEM;
Chris Masonc8b97812008-10-29 14:49:59 -0400345 atomic_set(&cb->pending_bios, 0);
346 cb->errors = 0;
347 cb->inode = inode;
348 cb->start = start;
349 cb->len = len;
Chris Masond20f7042008-12-08 16:58:54 -0500350 cb->mirror_num = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400351 cb->compressed_pages = compressed_pages;
352 cb->compressed_len = compressed_len;
353 cb->orig_bio = NULL;
354 cb->nr_pages = nr_pages;
355
356 bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
357
Chris Masonc8b97812008-10-29 14:49:59 -0400358 bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
Dulshani Gunawardhana67871252013-10-31 10:33:04 +0530359 if (!bio) {
Yoshinori Sanodac97e512011-02-15 12:01:42 +0000360 kfree(cb);
361 return -ENOMEM;
362 }
Chris Masonc8b97812008-10-29 14:49:59 -0400363 bio->bi_private = cb;
364 bio->bi_end_io = end_compressed_bio_write;
365 atomic_inc(&cb->pending_bios);
366
367 /* create and submit bios for the compressed pages */
368 bytes_left = compressed_len;
David Sterba306e16c2011-04-19 14:29:38 +0200369 for (pg_index = 0; pg_index < cb->nr_pages; pg_index++) {
370 page = compressed_pages[pg_index];
Chris Masonc8b97812008-10-29 14:49:59 -0400371 page->mapping = inode->i_mapping;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700372 if (bio->bi_iter.bi_size)
David Woodhouse64a16702009-07-15 23:29:37 +0100373 ret = io_tree->ops->merge_bio_hook(WRITE, page, 0,
Chris Masonc8b97812008-10-29 14:49:59 -0400374 PAGE_CACHE_SIZE,
375 bio, 0);
376 else
377 ret = 0;
378
Chris Mason70b99e62008-10-31 12:46:39 -0400379 page->mapping = NULL;
Chris Masonc8b97812008-10-29 14:49:59 -0400380 if (ret || bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) <
381 PAGE_CACHE_SIZE) {
382 bio_get(bio);
383
Chris Masonaf09abf2008-11-07 12:35:44 -0500384 /*
385 * inc the count before we submit the bio so
386 * we know the end IO handler won't happen before
387 * we inc the count. Otherwise, the cb might get
388 * freed before we're done setting it up
389 */
390 atomic_inc(&cb->pending_bios);
David Sterbabfebd8b2014-07-30 00:25:45 +0200391 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
392 BTRFS_WQ_ENDIO_DATA);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100393 BUG_ON(ret); /* -ENOMEM */
Chris Masonc8b97812008-10-29 14:49:59 -0400394
Li Zefane55179b2011-07-14 03:16:47 +0000395 if (!skip_sum) {
396 ret = btrfs_csum_one_bio(root, inode, bio,
397 start, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100398 BUG_ON(ret); /* -ENOMEM */
Li Zefane55179b2011-07-14 03:16:47 +0000399 }
Chris Masond20f7042008-12-08 16:58:54 -0500400
Chris Masonc8b97812008-10-29 14:49:59 -0400401 ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100402 BUG_ON(ret); /* -ENOMEM */
Chris Masonc8b97812008-10-29 14:49:59 -0400403
404 bio_put(bio);
405
406 bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
Tsutomu Itohe627ee72012-04-12 16:03:56 -0400407 BUG_ON(!bio);
Chris Masonc8b97812008-10-29 14:49:59 -0400408 bio->bi_private = cb;
409 bio->bi_end_io = end_compressed_bio_write;
410 bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
411 }
Chris Masoncfbc2462008-10-30 13:22:14 -0400412 if (bytes_left < PAGE_CACHE_SIZE) {
Frank Holtonefe120a2013-12-20 11:37:06 -0500413 btrfs_info(BTRFS_I(inode)->root->fs_info,
414 "bytes left %lu compress len %lu nr %lu",
Chris Masoncfbc2462008-10-30 13:22:14 -0400415 bytes_left, cb->compressed_len, cb->nr_pages);
416 }
Chris Masonc8b97812008-10-29 14:49:59 -0400417 bytes_left -= PAGE_CACHE_SIZE;
418 first_byte += PAGE_CACHE_SIZE;
Chris Mason771ed682008-11-06 22:02:51 -0500419 cond_resched();
Chris Masonc8b97812008-10-29 14:49:59 -0400420 }
421 bio_get(bio);
422
David Sterbabfebd8b2014-07-30 00:25:45 +0200423 ret = btrfs_bio_wq_end_io(root->fs_info, bio, BTRFS_WQ_ENDIO_DATA);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100424 BUG_ON(ret); /* -ENOMEM */
Chris Masonc8b97812008-10-29 14:49:59 -0400425
Li Zefane55179b2011-07-14 03:16:47 +0000426 if (!skip_sum) {
427 ret = btrfs_csum_one_bio(root, inode, bio, start, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100428 BUG_ON(ret); /* -ENOMEM */
Li Zefane55179b2011-07-14 03:16:47 +0000429 }
Chris Masond20f7042008-12-08 16:58:54 -0500430
Chris Masonc8b97812008-10-29 14:49:59 -0400431 ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100432 BUG_ON(ret); /* -ENOMEM */
Chris Masonc8b97812008-10-29 14:49:59 -0400433
434 bio_put(bio);
435 return 0;
436}
437
Chris Mason771ed682008-11-06 22:02:51 -0500438static noinline int add_ra_bio_pages(struct inode *inode,
439 u64 compressed_end,
440 struct compressed_bio *cb)
441{
442 unsigned long end_index;
David Sterba306e16c2011-04-19 14:29:38 +0200443 unsigned long pg_index;
Chris Mason771ed682008-11-06 22:02:51 -0500444 u64 last_offset;
445 u64 isize = i_size_read(inode);
446 int ret;
447 struct page *page;
448 unsigned long nr_pages = 0;
449 struct extent_map *em;
450 struct address_space *mapping = inode->i_mapping;
Chris Mason771ed682008-11-06 22:02:51 -0500451 struct extent_map_tree *em_tree;
452 struct extent_io_tree *tree;
453 u64 end;
454 int misses = 0;
455
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;
460
461 if (isize == 0)
462 return 0;
463
464 end_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
465
Chris Masond3977122009-01-05 21:25:51 -0500466 while (last_offset < compressed_end) {
David Sterba306e16c2011-04-19 14:29:38 +0200467 pg_index = last_offset >> PAGE_CACHE_SHIFT;
Chris Mason771ed682008-11-06 22:02:51 -0500468
David Sterba306e16c2011-04-19 14:29:38 +0200469 if (pg_index > end_index)
Chris Mason771ed682008-11-06 22:02:51 -0500470 break;
471
472 rcu_read_lock();
David Sterba306e16c2011-04-19 14:29:38 +0200473 page = radix_tree_lookup(&mapping->page_tree, pg_index);
Chris Mason771ed682008-11-06 22:02:51 -0500474 rcu_read_unlock();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700475 if (page && !radix_tree_exceptional_entry(page)) {
Chris Mason771ed682008-11-06 22:02:51 -0500476 misses++;
477 if (misses > 4)
478 break;
479 goto next;
480 }
481
Nick Piggin28ecb6092010-03-17 13:31:04 +0000482 page = __page_cache_alloc(mapping_gfp_mask(mapping) &
483 ~__GFP_FS);
Chris Mason771ed682008-11-06 22:02:51 -0500484 if (!page)
485 break;
486
David Sterba306e16c2011-04-19 14:29:38 +0200487 if (add_to_page_cache_lru(page, mapping, pg_index,
Nick Piggin28ecb6092010-03-17 13:31:04 +0000488 GFP_NOFS)) {
Chris Mason771ed682008-11-06 22:02:51 -0500489 page_cache_release(page);
490 goto next;
491 }
492
Chris Mason771ed682008-11-06 22:02:51 -0500493 end = last_offset + PAGE_CACHE_SIZE - 1;
494 /*
495 * at this point, we have a locked page in the page cache
496 * for these bytes in the file. But, we have to make
497 * sure they map to this compressed extent on disk.
498 */
499 set_page_extent_mapped(page);
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100500 lock_extent(tree, last_offset, end);
Chris Mason890871b2009-09-02 16:24:52 -0400501 read_lock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500502 em = lookup_extent_mapping(em_tree, last_offset,
503 PAGE_CACHE_SIZE);
Chris Mason890871b2009-09-02 16:24:52 -0400504 read_unlock(&em_tree->lock);
Chris Mason771ed682008-11-06 22:02:51 -0500505
506 if (!em || last_offset < em->start ||
507 (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) ||
Kent Overstreet4f024f32013-10-11 15:44:27 -0700508 (em->block_start >> 9) != cb->orig_bio->bi_iter.bi_sector) {
Chris Mason771ed682008-11-06 22:02:51 -0500509 free_extent_map(em);
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100510 unlock_extent(tree, last_offset, end);
Chris Mason771ed682008-11-06 22:02:51 -0500511 unlock_page(page);
512 page_cache_release(page);
513 break;
514 }
515 free_extent_map(em);
516
517 if (page->index == end_index) {
518 char *userpage;
519 size_t zero_offset = isize & (PAGE_CACHE_SIZE - 1);
520
521 if (zero_offset) {
522 int zeros;
523 zeros = PAGE_CACHE_SIZE - zero_offset;
Cong Wang7ac687d2011-11-25 23:14:28 +0800524 userpage = kmap_atomic(page);
Chris Mason771ed682008-11-06 22:02:51 -0500525 memset(userpage + zero_offset, 0, zeros);
526 flush_dcache_page(page);
Cong Wang7ac687d2011-11-25 23:14:28 +0800527 kunmap_atomic(userpage);
Chris Mason771ed682008-11-06 22:02:51 -0500528 }
529 }
530
531 ret = bio_add_page(cb->orig_bio, page,
532 PAGE_CACHE_SIZE, 0);
533
534 if (ret == PAGE_CACHE_SIZE) {
535 nr_pages++;
536 page_cache_release(page);
537 } else {
Jeff Mahoneyd0082372012-03-01 14:57:19 +0100538 unlock_extent(tree, last_offset, end);
Chris Mason771ed682008-11-06 22:02:51 -0500539 unlock_page(page);
540 page_cache_release(page);
541 break;
542 }
543next:
544 last_offset += PAGE_CACHE_SIZE;
545 }
Chris Mason771ed682008-11-06 22:02:51 -0500546 return 0;
547}
548
Chris Masonc8b97812008-10-29 14:49:59 -0400549/*
550 * for a compressed read, the bio we get passed has all the inode pages
551 * in it. We don't actually do IO on those pages but allocate new ones
552 * to hold the compressed pages on disk.
553 *
Kent Overstreet4f024f32013-10-11 15:44:27 -0700554 * bio->bi_iter.bi_sector points to the compressed extent on disk
Chris Masonc8b97812008-10-29 14:49:59 -0400555 * bio->bi_io_vec points to all of the inode pages
556 * bio->bi_vcnt is a count of pages
557 *
558 * After the compressed pages are read, we copy the bytes into the
559 * bio we were passed and then call the bio end_io calls
560 */
561int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
562 int mirror_num, unsigned long bio_flags)
563{
564 struct extent_io_tree *tree;
565 struct extent_map_tree *em_tree;
566 struct compressed_bio *cb;
567 struct btrfs_root *root = BTRFS_I(inode)->root;
568 unsigned long uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
569 unsigned long compressed_len;
570 unsigned long nr_pages;
David Sterba306e16c2011-04-19 14:29:38 +0200571 unsigned long pg_index;
Chris Masonc8b97812008-10-29 14:49:59 -0400572 struct page *page;
573 struct block_device *bdev;
574 struct bio *comp_bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700575 u64 cur_disk_byte = (u64)bio->bi_iter.bi_sector << 9;
Chris Masone04ca622008-11-10 11:44:58 -0500576 u64 em_len;
577 u64 em_start;
Chris Masonc8b97812008-10-29 14:49:59 -0400578 struct extent_map *em;
liubo6b82ce82011-01-26 06:21:39 +0000579 int ret = -ENOMEM;
Josef Bacik15e30042012-10-05 13:39:50 -0400580 int faili = 0;
Chris Masond20f7042008-12-08 16:58:54 -0500581 u32 *sums;
Chris Masonc8b97812008-10-29 14:49:59 -0400582
583 tree = &BTRFS_I(inode)->io_tree;
584 em_tree = &BTRFS_I(inode)->extent_tree;
585
586 /* we need the actual starting offset of this extent in the file */
Chris Mason890871b2009-09-02 16:24:52 -0400587 read_lock(&em_tree->lock);
Chris Masonc8b97812008-10-29 14:49:59 -0400588 em = lookup_extent_mapping(em_tree,
589 page_offset(bio->bi_io_vec->bv_page),
590 PAGE_CACHE_SIZE);
Chris Mason890871b2009-09-02 16:24:52 -0400591 read_unlock(&em_tree->lock);
Tsutomu Itoh285190d2012-02-16 16:23:58 +0900592 if (!em)
593 return -EIO;
Chris Masonc8b97812008-10-29 14:49:59 -0400594
Chris Masond20f7042008-12-08 16:58:54 -0500595 compressed_len = em->block_len;
596 cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
liubo6b82ce82011-01-26 06:21:39 +0000597 if (!cb)
598 goto out;
599
Chris Masonc8b97812008-10-29 14:49:59 -0400600 atomic_set(&cb->pending_bios, 0);
601 cb->errors = 0;
602 cb->inode = inode;
Chris Masond20f7042008-12-08 16:58:54 -0500603 cb->mirror_num = mirror_num;
604 sums = &cb->sums;
Chris Masonc8b97812008-10-29 14:49:59 -0400605
Yan Zhengff5b7ee2008-11-10 07:34:43 -0500606 cb->start = em->orig_start;
Chris Masone04ca622008-11-10 11:44:58 -0500607 em_len = em->len;
608 em_start = em->start;
Chris Masond20f7042008-12-08 16:58:54 -0500609
Chris Masonc8b97812008-10-29 14:49:59 -0400610 free_extent_map(em);
Chris Masone04ca622008-11-10 11:44:58 -0500611 em = NULL;
Chris Masonc8b97812008-10-29 14:49:59 -0400612
613 cb->len = uncompressed_len;
614 cb->compressed_len = compressed_len;
Li Zefan261507a02010-12-17 14:21:50 +0800615 cb->compress_type = extent_compress_type(bio_flags);
Chris Masonc8b97812008-10-29 14:49:59 -0400616 cb->orig_bio = bio;
617
David Sterbaed6078f2014-06-05 01:59:57 +0200618 nr_pages = DIV_ROUND_UP(compressed_len, PAGE_CACHE_SIZE);
liubo6b82ce82011-01-26 06:21:39 +0000619 cb->compressed_pages = kzalloc(sizeof(struct page *) * nr_pages,
Chris Masonc8b97812008-10-29 14:49:59 -0400620 GFP_NOFS);
liubo6b82ce82011-01-26 06:21:39 +0000621 if (!cb->compressed_pages)
622 goto fail1;
623
Chris Masonc8b97812008-10-29 14:49:59 -0400624 bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
625
David Sterba306e16c2011-04-19 14:29:38 +0200626 for (pg_index = 0; pg_index < nr_pages; pg_index++) {
627 cb->compressed_pages[pg_index] = alloc_page(GFP_NOFS |
Chris Masonc8b97812008-10-29 14:49:59 -0400628 __GFP_HIGHMEM);
Josef Bacik15e30042012-10-05 13:39:50 -0400629 if (!cb->compressed_pages[pg_index]) {
630 faili = pg_index - 1;
631 ret = -ENOMEM;
liubo6b82ce82011-01-26 06:21:39 +0000632 goto fail2;
Josef Bacik15e30042012-10-05 13:39:50 -0400633 }
Chris Masonc8b97812008-10-29 14:49:59 -0400634 }
Josef Bacik15e30042012-10-05 13:39:50 -0400635 faili = nr_pages - 1;
Chris Masonc8b97812008-10-29 14:49:59 -0400636 cb->nr_pages = nr_pages;
637
Mark Fasheh4b384312013-08-06 11:42:50 -0700638 /* In the parent-locked case, we only locked the range we are
639 * interested in. In all other cases, we can opportunistically
640 * cache decompressed data that goes beyond the requested range. */
641 if (!(bio_flags & EXTENT_BIO_PARENT_LOCKED))
642 add_ra_bio_pages(inode, em_start + em_len, cb);
Chris Mason771ed682008-11-06 22:02:51 -0500643
Chris Mason771ed682008-11-06 22:02:51 -0500644 /* include any pages we added in add_ra-bio_pages */
645 uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
646 cb->len = uncompressed_len;
647
Chris Masonc8b97812008-10-29 14:49:59 -0400648 comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS);
liubo6b82ce82011-01-26 06:21:39 +0000649 if (!comp_bio)
650 goto fail2;
Chris Masonc8b97812008-10-29 14:49:59 -0400651 comp_bio->bi_private = cb;
652 comp_bio->bi_end_io = end_compressed_bio_read;
653 atomic_inc(&cb->pending_bios);
654
David Sterba306e16c2011-04-19 14:29:38 +0200655 for (pg_index = 0; pg_index < nr_pages; pg_index++) {
656 page = cb->compressed_pages[pg_index];
Chris Masonc8b97812008-10-29 14:49:59 -0400657 page->mapping = inode->i_mapping;
Chris Masond20f7042008-12-08 16:58:54 -0500658 page->index = em_start >> PAGE_CACHE_SHIFT;
659
Kent Overstreet4f024f32013-10-11 15:44:27 -0700660 if (comp_bio->bi_iter.bi_size)
David Woodhouse64a16702009-07-15 23:29:37 +0100661 ret = tree->ops->merge_bio_hook(READ, page, 0,
Chris Masonc8b97812008-10-29 14:49:59 -0400662 PAGE_CACHE_SIZE,
663 comp_bio, 0);
664 else
665 ret = 0;
666
Chris Mason70b99e62008-10-31 12:46:39 -0400667 page->mapping = NULL;
Chris Masonc8b97812008-10-29 14:49:59 -0400668 if (ret || bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0) <
669 PAGE_CACHE_SIZE) {
670 bio_get(comp_bio);
671
David Sterbabfebd8b2014-07-30 00:25:45 +0200672 ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio,
673 BTRFS_WQ_ENDIO_DATA);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100674 BUG_ON(ret); /* -ENOMEM */
Chris Masonc8b97812008-10-29 14:49:59 -0400675
Chris Masonaf09abf2008-11-07 12:35:44 -0500676 /*
677 * inc the count before we submit the bio so
678 * we know the end IO handler won't happen before
679 * we inc the count. Otherwise, the cb might get
680 * freed before we're done setting it up
681 */
682 atomic_inc(&cb->pending_bios);
683
Christoph Hellwig6cbff002009-04-17 10:37:41 +0200684 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
Tsutomu Itohc2db1072011-03-01 06:48:31 +0000685 ret = btrfs_lookup_bio_sums(root, inode,
686 comp_bio, sums);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100687 BUG_ON(ret); /* -ENOMEM */
Chris Masond20f7042008-12-08 16:58:54 -0500688 }
David Sterbaed6078f2014-06-05 01:59:57 +0200689 sums += DIV_ROUND_UP(comp_bio->bi_iter.bi_size,
690 root->sectorsize);
Chris Masond20f7042008-12-08 16:58:54 -0500691
692 ret = btrfs_map_bio(root, READ, comp_bio,
693 mirror_num, 0);
Stefan Behrens61891922012-11-05 18:51:52 +0100694 if (ret)
695 bio_endio(comp_bio, ret);
Chris Masonc8b97812008-10-29 14:49:59 -0400696
697 bio_put(comp_bio);
698
699 comp_bio = compressed_bio_alloc(bdev, cur_disk_byte,
700 GFP_NOFS);
Tsutomu Itohe627ee72012-04-12 16:03:56 -0400701 BUG_ON(!comp_bio);
Chris Mason771ed682008-11-06 22:02:51 -0500702 comp_bio->bi_private = cb;
703 comp_bio->bi_end_io = end_compressed_bio_read;
704
705 bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0);
Chris Masonc8b97812008-10-29 14:49:59 -0400706 }
707 cur_disk_byte += PAGE_CACHE_SIZE;
708 }
709 bio_get(comp_bio);
710
David Sterbabfebd8b2014-07-30 00:25:45 +0200711 ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio,
712 BTRFS_WQ_ENDIO_DATA);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100713 BUG_ON(ret); /* -ENOMEM */
Chris Masonc8b97812008-10-29 14:49:59 -0400714
Tsutomu Itohc2db1072011-03-01 06:48:31 +0000715 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
716 ret = btrfs_lookup_bio_sums(root, inode, comp_bio, sums);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100717 BUG_ON(ret); /* -ENOMEM */
Tsutomu Itohc2db1072011-03-01 06:48:31 +0000718 }
Chris Masond20f7042008-12-08 16:58:54 -0500719
720 ret = btrfs_map_bio(root, READ, comp_bio, mirror_num, 0);
Stefan Behrens61891922012-11-05 18:51:52 +0100721 if (ret)
722 bio_endio(comp_bio, ret);
Chris Masonc8b97812008-10-29 14:49:59 -0400723
724 bio_put(comp_bio);
725 return 0;
liubo6b82ce82011-01-26 06:21:39 +0000726
727fail2:
Josef Bacik15e30042012-10-05 13:39:50 -0400728 while (faili >= 0) {
729 __free_page(cb->compressed_pages[faili]);
730 faili--;
731 }
liubo6b82ce82011-01-26 06:21:39 +0000732
733 kfree(cb->compressed_pages);
734fail1:
735 kfree(cb);
736out:
737 free_extent_map(em);
738 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400739}
Li Zefan261507a02010-12-17 14:21:50 +0800740
741static struct list_head comp_idle_workspace[BTRFS_COMPRESS_TYPES];
742static spinlock_t comp_workspace_lock[BTRFS_COMPRESS_TYPES];
743static int comp_num_workspace[BTRFS_COMPRESS_TYPES];
744static atomic_t comp_alloc_workspace[BTRFS_COMPRESS_TYPES];
745static wait_queue_head_t comp_workspace_wait[BTRFS_COMPRESS_TYPES];
746
Eric Sandeen48a3b632013-04-25 20:41:01 +0000747static struct btrfs_compress_op *btrfs_compress_op[] = {
Li Zefan261507a02010-12-17 14:21:50 +0800748 &btrfs_zlib_compress,
Li Zefana6fa6fa2010-10-25 15:12:26 +0800749 &btrfs_lzo_compress,
Li Zefan261507a02010-12-17 14:21:50 +0800750};
751
Jeff Mahoney143bede2012-03-01 14:56:26 +0100752void __init btrfs_init_compress(void)
Li Zefan261507a02010-12-17 14:21:50 +0800753{
754 int i;
755
756 for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) {
757 INIT_LIST_HEAD(&comp_idle_workspace[i]);
758 spin_lock_init(&comp_workspace_lock[i]);
759 atomic_set(&comp_alloc_workspace[i], 0);
760 init_waitqueue_head(&comp_workspace_wait[i]);
761 }
Li Zefan261507a02010-12-17 14:21:50 +0800762}
763
764/*
765 * this finds an available workspace or allocates a new one
766 * ERR_PTR is returned if things go bad.
767 */
768static struct list_head *find_workspace(int type)
769{
770 struct list_head *workspace;
771 int cpus = num_online_cpus();
772 int idx = type - 1;
773
774 struct list_head *idle_workspace = &comp_idle_workspace[idx];
775 spinlock_t *workspace_lock = &comp_workspace_lock[idx];
776 atomic_t *alloc_workspace = &comp_alloc_workspace[idx];
777 wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx];
778 int *num_workspace = &comp_num_workspace[idx];
779again:
780 spin_lock(workspace_lock);
781 if (!list_empty(idle_workspace)) {
782 workspace = idle_workspace->next;
783 list_del(workspace);
784 (*num_workspace)--;
785 spin_unlock(workspace_lock);
786 return workspace;
787
788 }
789 if (atomic_read(alloc_workspace) > cpus) {
790 DEFINE_WAIT(wait);
791
792 spin_unlock(workspace_lock);
793 prepare_to_wait(workspace_wait, &wait, TASK_UNINTERRUPTIBLE);
794 if (atomic_read(alloc_workspace) > cpus && !*num_workspace)
795 schedule();
796 finish_wait(workspace_wait, &wait);
797 goto again;
798 }
799 atomic_inc(alloc_workspace);
800 spin_unlock(workspace_lock);
801
802 workspace = btrfs_compress_op[idx]->alloc_workspace();
803 if (IS_ERR(workspace)) {
804 atomic_dec(alloc_workspace);
805 wake_up(workspace_wait);
806 }
807 return workspace;
808}
809
810/*
811 * put a workspace struct back on the list or free it if we have enough
812 * idle ones sitting around
813 */
814static void free_workspace(int type, struct list_head *workspace)
815{
816 int idx = type - 1;
817 struct list_head *idle_workspace = &comp_idle_workspace[idx];
818 spinlock_t *workspace_lock = &comp_workspace_lock[idx];
819 atomic_t *alloc_workspace = &comp_alloc_workspace[idx];
820 wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx];
821 int *num_workspace = &comp_num_workspace[idx];
822
823 spin_lock(workspace_lock);
824 if (*num_workspace < num_online_cpus()) {
Sergey Senozhatskyc39aa702014-06-25 00:00:44 +0900825 list_add(workspace, idle_workspace);
Li Zefan261507a02010-12-17 14:21:50 +0800826 (*num_workspace)++;
827 spin_unlock(workspace_lock);
828 goto wake;
829 }
830 spin_unlock(workspace_lock);
831
832 btrfs_compress_op[idx]->free_workspace(workspace);
833 atomic_dec(alloc_workspace);
834wake:
Josef Bacik66657b32012-08-01 15:36:24 -0400835 smp_mb();
Li Zefan261507a02010-12-17 14:21:50 +0800836 if (waitqueue_active(workspace_wait))
837 wake_up(workspace_wait);
838}
839
840/*
841 * cleanup function for module exit
842 */
843static void free_workspaces(void)
844{
845 struct list_head *workspace;
846 int i;
847
848 for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) {
849 while (!list_empty(&comp_idle_workspace[i])) {
850 workspace = comp_idle_workspace[i].next;
851 list_del(workspace);
852 btrfs_compress_op[i]->free_workspace(workspace);
853 atomic_dec(&comp_alloc_workspace[i]);
854 }
855 }
856}
857
858/*
859 * given an address space and start/len, compress the bytes.
860 *
861 * pages are allocated to hold the compressed result and stored
862 * in 'pages'
863 *
864 * out_pages is used to return the number of pages allocated. There
865 * may be pages allocated even if we return an error
866 *
867 * total_in is used to return the number of bytes actually read. It
868 * may be smaller then len if we had to exit early because we
869 * ran out of room in the pages array or because we cross the
870 * max_out threshold.
871 *
872 * total_out is used to return the total number of compressed bytes
873 *
874 * max_out tells us the max number of bytes that we're allowed to
875 * stuff into pages
876 */
877int btrfs_compress_pages(int type, struct address_space *mapping,
878 u64 start, unsigned long len,
879 struct page **pages,
880 unsigned long nr_dest_pages,
881 unsigned long *out_pages,
882 unsigned long *total_in,
883 unsigned long *total_out,
884 unsigned long max_out)
885{
886 struct list_head *workspace;
887 int ret;
888
889 workspace = find_workspace(type);
890 if (IS_ERR(workspace))
Zach Brown774bcb32014-05-09 17:15:09 -0400891 return PTR_ERR(workspace);
Li Zefan261507a02010-12-17 14:21:50 +0800892
893 ret = btrfs_compress_op[type-1]->compress_pages(workspace, mapping,
894 start, len, pages,
895 nr_dest_pages, out_pages,
896 total_in, total_out,
897 max_out);
898 free_workspace(type, workspace);
899 return ret;
900}
901
902/*
903 * pages_in is an array of pages with compressed data.
904 *
905 * disk_start is the starting logical offset of this array in the file
906 *
907 * bvec is a bio_vec of pages from the file that we want to decompress into
908 *
909 * vcnt is the count of pages in the biovec
910 *
911 * srclen is the number of bytes in pages_in
912 *
913 * The basic idea is that we have a bio that was created by readpages.
914 * The pages in the bio are for the uncompressed data, and they may not
915 * be contiguous. They all correspond to the range of bytes covered by
916 * the compressed extent.
917 */
Eric Sandeen48a3b632013-04-25 20:41:01 +0000918static int btrfs_decompress_biovec(int type, struct page **pages_in,
919 u64 disk_start, struct bio_vec *bvec,
920 int vcnt, size_t srclen)
Li Zefan261507a02010-12-17 14:21:50 +0800921{
922 struct list_head *workspace;
923 int ret;
924
925 workspace = find_workspace(type);
926 if (IS_ERR(workspace))
Zach Brown774bcb32014-05-09 17:15:09 -0400927 return PTR_ERR(workspace);
Li Zefan261507a02010-12-17 14:21:50 +0800928
929 ret = btrfs_compress_op[type-1]->decompress_biovec(workspace, pages_in,
930 disk_start,
931 bvec, vcnt, srclen);
932 free_workspace(type, workspace);
933 return ret;
934}
935
936/*
937 * a less complex decompression routine. Our compressed data fits in a
938 * single page, and we want to read a single page out of it.
939 * start_byte tells us the offset into the compressed data we're interested in
940 */
941int btrfs_decompress(int type, unsigned char *data_in, struct page *dest_page,
942 unsigned long start_byte, size_t srclen, size_t destlen)
943{
944 struct list_head *workspace;
945 int ret;
946
947 workspace = find_workspace(type);
948 if (IS_ERR(workspace))
Zach Brown774bcb32014-05-09 17:15:09 -0400949 return PTR_ERR(workspace);
Li Zefan261507a02010-12-17 14:21:50 +0800950
951 ret = btrfs_compress_op[type-1]->decompress(workspace, data_in,
952 dest_page, start_byte,
953 srclen, destlen);
954
955 free_workspace(type, workspace);
956 return ret;
957}
958
Alexey Charkov8e4eef72011-02-02 21:15:35 +0000959void btrfs_exit_compress(void)
Li Zefan261507a02010-12-17 14:21:50 +0800960{
961 free_workspaces();
962}
Li Zefan3a39c182010-11-08 15:22:19 +0800963
964/*
965 * Copy uncompressed data from working buffer to pages.
966 *
967 * buf_start is the byte offset we're of the start of our workspace buffer.
968 *
969 * total_out is the last byte of the buffer
970 */
971int btrfs_decompress_buf2page(char *buf, unsigned long buf_start,
972 unsigned long total_out, u64 disk_start,
973 struct bio_vec *bvec, int vcnt,
David Sterba306e16c2011-04-19 14:29:38 +0200974 unsigned long *pg_index,
Li Zefan3a39c182010-11-08 15:22:19 +0800975 unsigned long *pg_offset)
976{
977 unsigned long buf_offset;
978 unsigned long current_buf_start;
979 unsigned long start_byte;
980 unsigned long working_bytes = total_out - buf_start;
981 unsigned long bytes;
982 char *kaddr;
David Sterba306e16c2011-04-19 14:29:38 +0200983 struct page *page_out = bvec[*pg_index].bv_page;
Li Zefan3a39c182010-11-08 15:22:19 +0800984
985 /*
986 * start byte is the first byte of the page we're currently
987 * copying into relative to the start of the compressed data.
988 */
989 start_byte = page_offset(page_out) - disk_start;
990
991 /* we haven't yet hit data corresponding to this page */
992 if (total_out <= start_byte)
993 return 1;
994
995 /*
996 * the start of the data we care about is offset into
997 * the middle of our working buffer
998 */
999 if (total_out > start_byte && buf_start < start_byte) {
1000 buf_offset = start_byte - buf_start;
1001 working_bytes -= buf_offset;
1002 } else {
1003 buf_offset = 0;
1004 }
1005 current_buf_start = buf_start;
1006
1007 /* copy bytes from the working buffer into the pages */
1008 while (working_bytes > 0) {
1009 bytes = min(PAGE_CACHE_SIZE - *pg_offset,
1010 PAGE_CACHE_SIZE - buf_offset);
1011 bytes = min(bytes, working_bytes);
Cong Wang7ac687d2011-11-25 23:14:28 +08001012 kaddr = kmap_atomic(page_out);
Li Zefan3a39c182010-11-08 15:22:19 +08001013 memcpy(kaddr + *pg_offset, buf + buf_offset, bytes);
Filipe David Borba Mananaa2aa75e2014-02-08 15:47:46 +00001014 if (*pg_index == (vcnt - 1) && *pg_offset == 0)
1015 memset(kaddr + bytes, 0, PAGE_CACHE_SIZE - bytes);
Cong Wang7ac687d2011-11-25 23:14:28 +08001016 kunmap_atomic(kaddr);
Li Zefan3a39c182010-11-08 15:22:19 +08001017 flush_dcache_page(page_out);
1018
1019 *pg_offset += bytes;
1020 buf_offset += bytes;
1021 working_bytes -= bytes;
1022 current_buf_start += bytes;
1023
1024 /* check if we need to pick another page */
1025 if (*pg_offset == PAGE_CACHE_SIZE) {
David Sterba306e16c2011-04-19 14:29:38 +02001026 (*pg_index)++;
1027 if (*pg_index >= vcnt)
Li Zefan3a39c182010-11-08 15:22:19 +08001028 return 0;
1029
David Sterba306e16c2011-04-19 14:29:38 +02001030 page_out = bvec[*pg_index].bv_page;
Li Zefan3a39c182010-11-08 15:22:19 +08001031 *pg_offset = 0;
1032 start_byte = page_offset(page_out) - disk_start;
1033
1034 /*
1035 * make sure our new page is covered by this
1036 * working buffer
1037 */
1038 if (total_out <= start_byte)
1039 return 1;
1040
1041 /*
1042 * the next page in the biovec might not be adjacent
1043 * to the last page, but it might still be found
1044 * inside this working buffer. bump our offset pointer
1045 */
1046 if (total_out > start_byte &&
1047 current_buf_start < start_byte) {
1048 buf_offset = start_byte - buf_start;
1049 working_bytes = total_out - start_byte;
1050 current_buf_start = buf_start + buf_offset;
1051 }
1052 }
1053 }
1054
1055 return 1;
1056}