blob: 86eff48dab786d525d3395d31dcf3f45a11dafc1 [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 Mason4b4e25f2008-11-20 10:22:27 -050035#include "compat.h"
Chris Masonc8b97812008-10-29 14:49:59 -040036#include "ctree.h"
37#include "disk-io.h"
38#include "transaction.h"
39#include "btrfs_inode.h"
40#include "volumes.h"
41#include "ordered-data.h"
Chris Masonc8b97812008-10-29 14:49:59 -040042#include "compression.h"
43#include "extent_io.h"
44#include "extent_map.h"
45
46struct compressed_bio {
47 /* number of bios pending for this compressed extent */
48 atomic_t pending_bios;
49
50 /* the pages with the compressed data on them */
51 struct page **compressed_pages;
52
53 /* inode that owns this data */
54 struct inode *inode;
55
56 /* starting offset in the inode for our pages */
57 u64 start;
58
59 /* number of bytes in the inode we're working on */
60 unsigned long len;
61
62 /* number of bytes on disk */
63 unsigned long compressed_len;
64
Li Zefan261507a02010-12-17 14:21:50 +080065 /* the compression algorithm for this bio */
66 int compress_type;
67
Chris Masonc8b97812008-10-29 14:49:59 -040068 /* number of compressed pages in the array */
69 unsigned long nr_pages;
70
71 /* IO errors */
72 int errors;
Chris Masond20f7042008-12-08 16:58:54 -050073 int mirror_num;
Chris Masonc8b97812008-10-29 14:49:59 -040074
75 /* for reads, this is the bio we are copying the data into */
76 struct bio *orig_bio;
Chris Masond20f7042008-12-08 16:58:54 -050077
78 /*
79 * the start of a variable length array of checksums only
80 * used by reads
81 */
82 u32 sums;
Chris Masonc8b97812008-10-29 14:49:59 -040083};
84
Chris Masond20f7042008-12-08 16:58:54 -050085static inline int compressed_bio_size(struct btrfs_root *root,
86 unsigned long disk_size)
87{
David Sterba6c417612011-04-13 15:41:04 +020088 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
89
Chris Masond20f7042008-12-08 16:58:54 -050090 return sizeof(struct compressed_bio) +
91 ((disk_size + root->sectorsize - 1) / root->sectorsize) *
92 csum_size;
93}
94
Chris Masonc8b97812008-10-29 14:49:59 -040095static struct bio *compressed_bio_alloc(struct block_device *bdev,
96 u64 first_byte, gfp_t gfp_flags)
97{
Chris Masonc8b97812008-10-29 14:49:59 -040098 int nr_vecs;
99
100 nr_vecs = bio_get_nr_vecs(bdev);
Miao Xie88f794e2010-11-22 03:02:55 +0000101 return btrfs_bio_alloc(bdev, first_byte >> 9, nr_vecs, gfp_flags);
Chris Masonc8b97812008-10-29 14:49:59 -0400102}
103
Chris Masond20f7042008-12-08 16:58:54 -0500104static int check_compressed_csum(struct inode *inode,
105 struct compressed_bio *cb,
106 u64 disk_start)
107{
108 int ret;
109 struct btrfs_root *root = BTRFS_I(inode)->root;
110 struct page *page;
111 unsigned long i;
112 char *kaddr;
113 u32 csum;
114 u32 *cb_sum = &cb->sums;
115
Christoph Hellwig6cbff002009-04-17 10:37:41 +0200116 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
Chris Masond20f7042008-12-08 16:58:54 -0500117 return 0;
118
119 for (i = 0; i < cb->nr_pages; i++) {
120 page = cb->compressed_pages[i];
121 csum = ~(u32)0;
122
Cong Wang7ac687d2011-11-25 23:14:28 +0800123 kaddr = kmap_atomic(page);
Chris Masond20f7042008-12-08 16:58:54 -0500124 csum = btrfs_csum_data(root, kaddr, csum, PAGE_CACHE_SIZE);
125 btrfs_csum_final(csum, (char *)&csum);
Cong Wang7ac687d2011-11-25 23:14:28 +0800126 kunmap_atomic(kaddr);
Chris Masond20f7042008-12-08 16:58:54 -0500127
128 if (csum != *cb_sum) {
Li Zefan33345d012011-04-20 10:31:50 +0800129 printk(KERN_INFO "btrfs csum failed ino %llu "
Chris Masond3977122009-01-05 21:25:51 -0500130 "extent %llu csum %u "
Li Zefan33345d012011-04-20 10:31:50 +0800131 "wanted %u mirror %d\n",
132 (unsigned long long)btrfs_ino(inode),
Chris Masond20f7042008-12-08 16:58:54 -0500133 (unsigned long long)disk_start,
134 csum, *cb_sum, cb->mirror_num);
135 ret = -EIO;
136 goto fail;
137 }
138 cb_sum++;
139
140 }
141 ret = 0;
142fail:
143 return ret;
144}
145
Chris Masonc8b97812008-10-29 14:49:59 -0400146/* when we finish reading compressed pages from the disk, we
147 * decompress them and then run the bio end_io routines on the
148 * decompressed pages (in the inode address space).
149 *
150 * This allows the checksumming and other IO error handling routines
151 * to work normally
152 *
153 * The compressed pages are freed here, and it must be run
154 * in process context
155 */
156static void end_compressed_bio_read(struct bio *bio, int err)
157{
Chris Masonc8b97812008-10-29 14:49:59 -0400158 struct compressed_bio *cb = bio->bi_private;
159 struct inode *inode;
160 struct page *page;
161 unsigned long index;
162 int ret;
163
164 if (err)
165 cb->errors = 1;
166
167 /* if there are more bios still pending for this compressed
168 * extent, just exit
169 */
170 if (!atomic_dec_and_test(&cb->pending_bios))
171 goto out;
172
Chris Masond20f7042008-12-08 16:58:54 -0500173 inode = cb->inode;
174 ret = check_compressed_csum(inode, cb, (u64)bio->bi_sector << 9);
175 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 {
203 int bio_index = 0;
204 struct bio_vec *bvec = cb->orig_bio->bi_io_vec;
205
206 /*
207 * we have verified the checksum already, set page
208 * checked so the end_io handlers know about it
209 */
Chris Masond3977122009-01-05 21:25:51 -0500210 while (bio_index < cb->orig_bio->bi_vcnt) {
Chris Masond20f7042008-12-08 16:58:54 -0500211 SetPageChecked(bvec->bv_page);
212 bvec++;
213 bio_index++;
214 }
Chris Masonc8b97812008-10-29 14:49:59 -0400215 bio_endio(cb->orig_bio, 0);
Chris Masond20f7042008-12-08 16:58:54 -0500216 }
Chris Masonc8b97812008-10-29 14:49:59 -0400217
218 /* finally free the cb struct */
219 kfree(cb->compressed_pages);
220 kfree(cb);
221out:
222 bio_put(bio);
223}
224
225/*
226 * Clear the writeback bits on all of the file
227 * pages for a compressed write
228 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100229static noinline void end_compressed_writeback(struct inode *inode, u64 start,
230 unsigned long ram_size)
Chris Masonc8b97812008-10-29 14:49:59 -0400231{
232 unsigned long index = start >> PAGE_CACHE_SHIFT;
233 unsigned long end_index = (start + ram_size - 1) >> PAGE_CACHE_SHIFT;
234 struct page *pages[16];
235 unsigned long nr_pages = end_index - index + 1;
236 int i;
237 int ret;
238
Chris Masond3977122009-01-05 21:25:51 -0500239 while (nr_pages > 0) {
Chris Masonc8b97812008-10-29 14:49:59 -0400240 ret = find_get_pages_contig(inode->i_mapping, index,
Chris Mason5b050f02008-11-11 09:34:41 -0500241 min_t(unsigned long,
242 nr_pages, ARRAY_SIZE(pages)), pages);
Chris Masonc8b97812008-10-29 14:49:59 -0400243 if (ret == 0) {
244 nr_pages -= 1;
245 index += 1;
246 continue;
247 }
248 for (i = 0; i < ret; i++) {
249 end_page_writeback(pages[i]);
250 page_cache_release(pages[i]);
251 }
252 nr_pages -= ret;
253 index += ret;
254 }
255 /* the inode may be gone now */
Chris Masonc8b97812008-10-29 14:49:59 -0400256}
257
258/*
259 * do the cleanup once all the compressed pages hit the disk.
260 * This will clear writeback on the file pages and free the compressed
261 * pages.
262 *
263 * This also calls the writeback end hooks for the file pages so that
264 * metadata and checksums can be updated in the file.
265 */
266static void end_compressed_bio_write(struct bio *bio, int err)
267{
268 struct extent_io_tree *tree;
269 struct compressed_bio *cb = bio->bi_private;
270 struct inode *inode;
271 struct page *page;
272 unsigned long index;
273
274 if (err)
275 cb->errors = 1;
276
277 /* if there are more bios still pending for this compressed
278 * extent, just exit
279 */
280 if (!atomic_dec_and_test(&cb->pending_bios))
281 goto out;
282
283 /* ok, we're the last bio for this extent, step one is to
284 * call back into the FS and do all the end_io operations
285 */
286 inode = cb->inode;
287 tree = &BTRFS_I(inode)->io_tree;
Chris Mason70b99e62008-10-31 12:46:39 -0400288 cb->compressed_pages[0]->mapping = cb->inode->i_mapping;
Chris Masonc8b97812008-10-29 14:49:59 -0400289 tree->ops->writepage_end_io_hook(cb->compressed_pages[0],
290 cb->start,
291 cb->start + cb->len - 1,
292 NULL, 1);
Chris Mason70b99e62008-10-31 12:46:39 -0400293 cb->compressed_pages[0]->mapping = NULL;
Chris Masonc8b97812008-10-29 14:49:59 -0400294
295 end_compressed_writeback(inode, cb->start, cb->len);
296 /* note, our inode could be gone now */
297
298 /*
299 * release the compressed pages, these came from alloc_page and
300 * are not attached to the inode at all
301 */
302 index = 0;
303 for (index = 0; index < cb->nr_pages; index++) {
304 page = cb->compressed_pages[index];
305 page->mapping = NULL;
306 page_cache_release(page);
307 }
308
309 /* finally free the cb struct */
310 kfree(cb->compressed_pages);
311 kfree(cb);
312out:
313 bio_put(bio);
314}
315
316/*
317 * worker function to build and submit bios for previously compressed pages.
318 * The corresponding pages in the inode should be marked for writeback
319 * and the compressed pages should have a reference on them for dropping
320 * when the IO is complete.
321 *
322 * This also checksums the file bytes and gets things ready for
323 * the end io hooks.
324 */
325int btrfs_submit_compressed_write(struct inode *inode, u64 start,
326 unsigned long len, u64 disk_start,
327 unsigned long compressed_len,
328 struct page **compressed_pages,
329 unsigned long nr_pages)
330{
331 struct bio *bio = NULL;
332 struct btrfs_root *root = BTRFS_I(inode)->root;
333 struct compressed_bio *cb;
334 unsigned long bytes_left;
335 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
David Sterba306e16c2011-04-19 14:29:38 +0200336 int pg_index = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400337 struct page *page;
338 u64 first_byte = disk_start;
339 struct block_device *bdev;
340 int ret;
Li Zefane55179b2011-07-14 03:16:47 +0000341 int skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
Chris Masonc8b97812008-10-29 14:49:59 -0400342
343 WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1));
Chris Masond20f7042008-12-08 16:58:54 -0500344 cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
Yoshinori Sanodac97e512011-02-15 12:01:42 +0000345 if (!cb)
346 return -ENOMEM;
Chris Masonc8b97812008-10-29 14:49:59 -0400347 atomic_set(&cb->pending_bios, 0);
348 cb->errors = 0;
349 cb->inode = inode;
350 cb->start = start;
351 cb->len = len;
Chris Masond20f7042008-12-08 16:58:54 -0500352 cb->mirror_num = 0;
Chris Masonc8b97812008-10-29 14:49:59 -0400353 cb->compressed_pages = compressed_pages;
354 cb->compressed_len = compressed_len;
355 cb->orig_bio = NULL;
356 cb->nr_pages = nr_pages;
357
358 bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
359
Chris Masonc8b97812008-10-29 14:49:59 -0400360 bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
Yoshinori Sanodac97e512011-02-15 12:01:42 +0000361 if(!bio) {
362 kfree(cb);
363 return -ENOMEM;
364 }
Chris Masonc8b97812008-10-29 14:49:59 -0400365 bio->bi_private = cb;
366 bio->bi_end_io = end_compressed_bio_write;
367 atomic_inc(&cb->pending_bios);
368
369 /* create and submit bios for the compressed pages */
370 bytes_left = compressed_len;
David Sterba306e16c2011-04-19 14:29:38 +0200371 for (pg_index = 0; pg_index < cb->nr_pages; pg_index++) {
372 page = compressed_pages[pg_index];
Chris Masonc8b97812008-10-29 14:49:59 -0400373 page->mapping = inode->i_mapping;
374 if (bio->bi_size)
375 ret = io_tree->ops->merge_bio_hook(page, 0,
376 PAGE_CACHE_SIZE,
377 bio, 0);
378 else
379 ret = 0;
380
Chris Mason70b99e62008-10-31 12:46:39 -0400381 page->mapping = NULL;
Chris Masonc8b97812008-10-29 14:49:59 -0400382 if (ret || bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) <
383 PAGE_CACHE_SIZE) {
384 bio_get(bio);
385
Chris Masonaf09abf2008-11-07 12:35:44 -0500386 /*
387 * inc the count before we submit the bio so
388 * we know the end IO handler won't happen before
389 * we inc the count. Otherwise, the cb might get
390 * freed before we're done setting it up
391 */
392 atomic_inc(&cb->pending_bios);
Chris Masonc8b97812008-10-29 14:49:59 -0400393 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100394 BUG_ON(ret); /* -ENOMEM */
Chris Masonc8b97812008-10-29 14:49:59 -0400395
Li Zefane55179b2011-07-14 03:16:47 +0000396 if (!skip_sum) {
397 ret = btrfs_csum_one_bio(root, inode, bio,
398 start, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100399 BUG_ON(ret); /* -ENOMEM */
Li Zefane55179b2011-07-14 03:16:47 +0000400 }
Chris Masond20f7042008-12-08 16:58:54 -0500401
Chris Masonc8b97812008-10-29 14:49:59 -0400402 ret = btrfs_map_bio(root, WRITE, bio, 0, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100403 BUG_ON(ret); /* -ENOMEM */
Chris Masonc8b97812008-10-29 14:49:59 -0400404
405 bio_put(bio);
406
407 bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS);
Tsutomu Itohe627ee72012-04-12 16:03:56 -0400408 BUG_ON(!bio);
Chris Masonc8b97812008-10-29 14:49:59 -0400409 bio->bi_private = cb;
410 bio->bi_end_io = end_compressed_bio_write;
411 bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
412 }
Chris Masoncfbc2462008-10-30 13:22:14 -0400413 if (bytes_left < PAGE_CACHE_SIZE) {
414 printk("bytes left %lu compress len %lu nr %lu\n",
415 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
423 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
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();
475 if (page) {
476 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)) ||
508 (em->block_start >> 9) != cb->orig_bio->bi_sector) {
509 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 *
554 * bio->bi_sector points to the compressed extent on disk
555 * 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;
575 u64 cur_disk_byte = (u64)bio->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;
Chris Masond20f7042008-12-08 16:58:54 -0500580 u32 *sums;
Chris Masonc8b97812008-10-29 14:49:59 -0400581
582 tree = &BTRFS_I(inode)->io_tree;
583 em_tree = &BTRFS_I(inode)->extent_tree;
584
585 /* we need the actual starting offset of this extent in the file */
Chris Mason890871b2009-09-02 16:24:52 -0400586 read_lock(&em_tree->lock);
Chris Masonc8b97812008-10-29 14:49:59 -0400587 em = lookup_extent_mapping(em_tree,
588 page_offset(bio->bi_io_vec->bv_page),
589 PAGE_CACHE_SIZE);
Chris Mason890871b2009-09-02 16:24:52 -0400590 read_unlock(&em_tree->lock);
Tsutomu Itoh285190d2012-02-16 16:23:58 +0900591 if (!em)
592 return -EIO;
Chris Masonc8b97812008-10-29 14:49:59 -0400593
Chris Masond20f7042008-12-08 16:58:54 -0500594 compressed_len = em->block_len;
595 cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
liubo6b82ce82011-01-26 06:21:39 +0000596 if (!cb)
597 goto out;
598
Chris Masonc8b97812008-10-29 14:49:59 -0400599 atomic_set(&cb->pending_bios, 0);
600 cb->errors = 0;
601 cb->inode = inode;
Chris Masond20f7042008-12-08 16:58:54 -0500602 cb->mirror_num = mirror_num;
603 sums = &cb->sums;
Chris Masonc8b97812008-10-29 14:49:59 -0400604
Yan Zhengff5b7ee2008-11-10 07:34:43 -0500605 cb->start = em->orig_start;
Chris Masone04ca622008-11-10 11:44:58 -0500606 em_len = em->len;
607 em_start = em->start;
Chris Masond20f7042008-12-08 16:58:54 -0500608
Chris Masonc8b97812008-10-29 14:49:59 -0400609 free_extent_map(em);
Chris Masone04ca622008-11-10 11:44:58 -0500610 em = NULL;
Chris Masonc8b97812008-10-29 14:49:59 -0400611
612 cb->len = uncompressed_len;
613 cb->compressed_len = compressed_len;
Li Zefan261507a02010-12-17 14:21:50 +0800614 cb->compress_type = extent_compress_type(bio_flags);
Chris Masonc8b97812008-10-29 14:49:59 -0400615 cb->orig_bio = bio;
616
617 nr_pages = (compressed_len + PAGE_CACHE_SIZE - 1) /
618 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);
David Sterba306e16c2011-04-19 14:29:38 +0200629 if (!cb->compressed_pages[pg_index])
liubo6b82ce82011-01-26 06:21:39 +0000630 goto fail2;
Chris Masonc8b97812008-10-29 14:49:59 -0400631 }
632 cb->nr_pages = nr_pages;
633
Chris Masone04ca622008-11-10 11:44:58 -0500634 add_ra_bio_pages(inode, em_start + em_len, cb);
Chris Mason771ed682008-11-06 22:02:51 -0500635
Chris Mason771ed682008-11-06 22:02:51 -0500636 /* include any pages we added in add_ra-bio_pages */
637 uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
638 cb->len = uncompressed_len;
639
Chris Masonc8b97812008-10-29 14:49:59 -0400640 comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS);
liubo6b82ce82011-01-26 06:21:39 +0000641 if (!comp_bio)
642 goto fail2;
Chris Masonc8b97812008-10-29 14:49:59 -0400643 comp_bio->bi_private = cb;
644 comp_bio->bi_end_io = end_compressed_bio_read;
645 atomic_inc(&cb->pending_bios);
646
David Sterba306e16c2011-04-19 14:29:38 +0200647 for (pg_index = 0; pg_index < nr_pages; pg_index++) {
648 page = cb->compressed_pages[pg_index];
Chris Masonc8b97812008-10-29 14:49:59 -0400649 page->mapping = inode->i_mapping;
Chris Masond20f7042008-12-08 16:58:54 -0500650 page->index = em_start >> PAGE_CACHE_SHIFT;
651
Chris Masonc8b97812008-10-29 14:49:59 -0400652 if (comp_bio->bi_size)
653 ret = tree->ops->merge_bio_hook(page, 0,
654 PAGE_CACHE_SIZE,
655 comp_bio, 0);
656 else
657 ret = 0;
658
Chris Mason70b99e62008-10-31 12:46:39 -0400659 page->mapping = NULL;
Chris Masonc8b97812008-10-29 14:49:59 -0400660 if (ret || bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0) <
661 PAGE_CACHE_SIZE) {
662 bio_get(comp_bio);
663
664 ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100665 BUG_ON(ret); /* -ENOMEM */
Chris Masonc8b97812008-10-29 14:49:59 -0400666
Chris Masonaf09abf2008-11-07 12:35:44 -0500667 /*
668 * inc the count before we submit the bio so
669 * we know the end IO handler won't happen before
670 * we inc the count. Otherwise, the cb might get
671 * freed before we're done setting it up
672 */
673 atomic_inc(&cb->pending_bios);
674
Christoph Hellwig6cbff002009-04-17 10:37:41 +0200675 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
Tsutomu Itohc2db1072011-03-01 06:48:31 +0000676 ret = btrfs_lookup_bio_sums(root, inode,
677 comp_bio, sums);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100678 BUG_ON(ret); /* -ENOMEM */
Chris Masond20f7042008-12-08 16:58:54 -0500679 }
680 sums += (comp_bio->bi_size + root->sectorsize - 1) /
681 root->sectorsize;
682
683 ret = btrfs_map_bio(root, READ, comp_bio,
684 mirror_num, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100685 BUG_ON(ret); /* -ENOMEM */
Chris Masonc8b97812008-10-29 14:49:59 -0400686
687 bio_put(comp_bio);
688
689 comp_bio = compressed_bio_alloc(bdev, cur_disk_byte,
690 GFP_NOFS);
Tsutomu Itohe627ee72012-04-12 16:03:56 -0400691 BUG_ON(!comp_bio);
Chris Mason771ed682008-11-06 22:02:51 -0500692 comp_bio->bi_private = cb;
693 comp_bio->bi_end_io = end_compressed_bio_read;
694
695 bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0);
Chris Masonc8b97812008-10-29 14:49:59 -0400696 }
697 cur_disk_byte += PAGE_CACHE_SIZE;
698 }
699 bio_get(comp_bio);
700
701 ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100702 BUG_ON(ret); /* -ENOMEM */
Chris Masonc8b97812008-10-29 14:49:59 -0400703
Tsutomu Itohc2db1072011-03-01 06:48:31 +0000704 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
705 ret = btrfs_lookup_bio_sums(root, inode, comp_bio, sums);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100706 BUG_ON(ret); /* -ENOMEM */
Tsutomu Itohc2db1072011-03-01 06:48:31 +0000707 }
Chris Masond20f7042008-12-08 16:58:54 -0500708
709 ret = btrfs_map_bio(root, READ, comp_bio, mirror_num, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100710 BUG_ON(ret); /* -ENOMEM */
Chris Masonc8b97812008-10-29 14:49:59 -0400711
712 bio_put(comp_bio);
713 return 0;
liubo6b82ce82011-01-26 06:21:39 +0000714
715fail2:
David Sterba306e16c2011-04-19 14:29:38 +0200716 for (pg_index = 0; pg_index < nr_pages; pg_index++)
717 free_page((unsigned long)cb->compressed_pages[pg_index]);
liubo6b82ce82011-01-26 06:21:39 +0000718
719 kfree(cb->compressed_pages);
720fail1:
721 kfree(cb);
722out:
723 free_extent_map(em);
724 return ret;
Chris Masonc8b97812008-10-29 14:49:59 -0400725}
Li Zefan261507a02010-12-17 14:21:50 +0800726
727static struct list_head comp_idle_workspace[BTRFS_COMPRESS_TYPES];
728static spinlock_t comp_workspace_lock[BTRFS_COMPRESS_TYPES];
729static int comp_num_workspace[BTRFS_COMPRESS_TYPES];
730static atomic_t comp_alloc_workspace[BTRFS_COMPRESS_TYPES];
731static wait_queue_head_t comp_workspace_wait[BTRFS_COMPRESS_TYPES];
732
733struct btrfs_compress_op *btrfs_compress_op[] = {
734 &btrfs_zlib_compress,
Li Zefana6fa6fa2010-10-25 15:12:26 +0800735 &btrfs_lzo_compress,
Li Zefan261507a02010-12-17 14:21:50 +0800736};
737
Jeff Mahoney143bede2012-03-01 14:56:26 +0100738void __init btrfs_init_compress(void)
Li Zefan261507a02010-12-17 14:21:50 +0800739{
740 int i;
741
742 for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) {
743 INIT_LIST_HEAD(&comp_idle_workspace[i]);
744 spin_lock_init(&comp_workspace_lock[i]);
745 atomic_set(&comp_alloc_workspace[i], 0);
746 init_waitqueue_head(&comp_workspace_wait[i]);
747 }
Li Zefan261507a02010-12-17 14:21:50 +0800748}
749
750/*
751 * this finds an available workspace or allocates a new one
752 * ERR_PTR is returned if things go bad.
753 */
754static struct list_head *find_workspace(int type)
755{
756 struct list_head *workspace;
757 int cpus = num_online_cpus();
758 int idx = type - 1;
759
760 struct list_head *idle_workspace = &comp_idle_workspace[idx];
761 spinlock_t *workspace_lock = &comp_workspace_lock[idx];
762 atomic_t *alloc_workspace = &comp_alloc_workspace[idx];
763 wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx];
764 int *num_workspace = &comp_num_workspace[idx];
765again:
766 spin_lock(workspace_lock);
767 if (!list_empty(idle_workspace)) {
768 workspace = idle_workspace->next;
769 list_del(workspace);
770 (*num_workspace)--;
771 spin_unlock(workspace_lock);
772 return workspace;
773
774 }
775 if (atomic_read(alloc_workspace) > cpus) {
776 DEFINE_WAIT(wait);
777
778 spin_unlock(workspace_lock);
779 prepare_to_wait(workspace_wait, &wait, TASK_UNINTERRUPTIBLE);
780 if (atomic_read(alloc_workspace) > cpus && !*num_workspace)
781 schedule();
782 finish_wait(workspace_wait, &wait);
783 goto again;
784 }
785 atomic_inc(alloc_workspace);
786 spin_unlock(workspace_lock);
787
788 workspace = btrfs_compress_op[idx]->alloc_workspace();
789 if (IS_ERR(workspace)) {
790 atomic_dec(alloc_workspace);
791 wake_up(workspace_wait);
792 }
793 return workspace;
794}
795
796/*
797 * put a workspace struct back on the list or free it if we have enough
798 * idle ones sitting around
799 */
800static void free_workspace(int type, struct list_head *workspace)
801{
802 int idx = type - 1;
803 struct list_head *idle_workspace = &comp_idle_workspace[idx];
804 spinlock_t *workspace_lock = &comp_workspace_lock[idx];
805 atomic_t *alloc_workspace = &comp_alloc_workspace[idx];
806 wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx];
807 int *num_workspace = &comp_num_workspace[idx];
808
809 spin_lock(workspace_lock);
810 if (*num_workspace < num_online_cpus()) {
811 list_add_tail(workspace, idle_workspace);
812 (*num_workspace)++;
813 spin_unlock(workspace_lock);
814 goto wake;
815 }
816 spin_unlock(workspace_lock);
817
818 btrfs_compress_op[idx]->free_workspace(workspace);
819 atomic_dec(alloc_workspace);
820wake:
821 if (waitqueue_active(workspace_wait))
822 wake_up(workspace_wait);
823}
824
825/*
826 * cleanup function for module exit
827 */
828static void free_workspaces(void)
829{
830 struct list_head *workspace;
831 int i;
832
833 for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) {
834 while (!list_empty(&comp_idle_workspace[i])) {
835 workspace = comp_idle_workspace[i].next;
836 list_del(workspace);
837 btrfs_compress_op[i]->free_workspace(workspace);
838 atomic_dec(&comp_alloc_workspace[i]);
839 }
840 }
841}
842
843/*
844 * given an address space and start/len, compress the bytes.
845 *
846 * pages are allocated to hold the compressed result and stored
847 * in 'pages'
848 *
849 * out_pages is used to return the number of pages allocated. There
850 * may be pages allocated even if we return an error
851 *
852 * total_in is used to return the number of bytes actually read. It
853 * may be smaller then len if we had to exit early because we
854 * ran out of room in the pages array or because we cross the
855 * max_out threshold.
856 *
857 * total_out is used to return the total number of compressed bytes
858 *
859 * max_out tells us the max number of bytes that we're allowed to
860 * stuff into pages
861 */
862int btrfs_compress_pages(int type, struct address_space *mapping,
863 u64 start, unsigned long len,
864 struct page **pages,
865 unsigned long nr_dest_pages,
866 unsigned long *out_pages,
867 unsigned long *total_in,
868 unsigned long *total_out,
869 unsigned long max_out)
870{
871 struct list_head *workspace;
872 int ret;
873
874 workspace = find_workspace(type);
875 if (IS_ERR(workspace))
876 return -1;
877
878 ret = btrfs_compress_op[type-1]->compress_pages(workspace, mapping,
879 start, len, pages,
880 nr_dest_pages, out_pages,
881 total_in, total_out,
882 max_out);
883 free_workspace(type, workspace);
884 return ret;
885}
886
887/*
888 * pages_in is an array of pages with compressed data.
889 *
890 * disk_start is the starting logical offset of this array in the file
891 *
892 * bvec is a bio_vec of pages from the file that we want to decompress into
893 *
894 * vcnt is the count of pages in the biovec
895 *
896 * srclen is the number of bytes in pages_in
897 *
898 * The basic idea is that we have a bio that was created by readpages.
899 * The pages in the bio are for the uncompressed data, and they may not
900 * be contiguous. They all correspond to the range of bytes covered by
901 * the compressed extent.
902 */
903int btrfs_decompress_biovec(int type, struct page **pages_in, u64 disk_start,
904 struct bio_vec *bvec, int vcnt, size_t srclen)
905{
906 struct list_head *workspace;
907 int ret;
908
909 workspace = find_workspace(type);
910 if (IS_ERR(workspace))
911 return -ENOMEM;
912
913 ret = btrfs_compress_op[type-1]->decompress_biovec(workspace, pages_in,
914 disk_start,
915 bvec, vcnt, srclen);
916 free_workspace(type, workspace);
917 return ret;
918}
919
920/*
921 * a less complex decompression routine. Our compressed data fits in a
922 * single page, and we want to read a single page out of it.
923 * start_byte tells us the offset into the compressed data we're interested in
924 */
925int btrfs_decompress(int type, unsigned char *data_in, struct page *dest_page,
926 unsigned long start_byte, size_t srclen, size_t destlen)
927{
928 struct list_head *workspace;
929 int ret;
930
931 workspace = find_workspace(type);
932 if (IS_ERR(workspace))
933 return -ENOMEM;
934
935 ret = btrfs_compress_op[type-1]->decompress(workspace, data_in,
936 dest_page, start_byte,
937 srclen, destlen);
938
939 free_workspace(type, workspace);
940 return ret;
941}
942
Alexey Charkov8e4eef72011-02-02 21:15:35 +0000943void btrfs_exit_compress(void)
Li Zefan261507a02010-12-17 14:21:50 +0800944{
945 free_workspaces();
946}
Li Zefan3a39c182010-11-08 15:22:19 +0800947
948/*
949 * Copy uncompressed data from working buffer to pages.
950 *
951 * buf_start is the byte offset we're of the start of our workspace buffer.
952 *
953 * total_out is the last byte of the buffer
954 */
955int btrfs_decompress_buf2page(char *buf, unsigned long buf_start,
956 unsigned long total_out, u64 disk_start,
957 struct bio_vec *bvec, int vcnt,
David Sterba306e16c2011-04-19 14:29:38 +0200958 unsigned long *pg_index,
Li Zefan3a39c182010-11-08 15:22:19 +0800959 unsigned long *pg_offset)
960{
961 unsigned long buf_offset;
962 unsigned long current_buf_start;
963 unsigned long start_byte;
964 unsigned long working_bytes = total_out - buf_start;
965 unsigned long bytes;
966 char *kaddr;
David Sterba306e16c2011-04-19 14:29:38 +0200967 struct page *page_out = bvec[*pg_index].bv_page;
Li Zefan3a39c182010-11-08 15:22:19 +0800968
969 /*
970 * start byte is the first byte of the page we're currently
971 * copying into relative to the start of the compressed data.
972 */
973 start_byte = page_offset(page_out) - disk_start;
974
975 /* we haven't yet hit data corresponding to this page */
976 if (total_out <= start_byte)
977 return 1;
978
979 /*
980 * the start of the data we care about is offset into
981 * the middle of our working buffer
982 */
983 if (total_out > start_byte && buf_start < start_byte) {
984 buf_offset = start_byte - buf_start;
985 working_bytes -= buf_offset;
986 } else {
987 buf_offset = 0;
988 }
989 current_buf_start = buf_start;
990
991 /* copy bytes from the working buffer into the pages */
992 while (working_bytes > 0) {
993 bytes = min(PAGE_CACHE_SIZE - *pg_offset,
994 PAGE_CACHE_SIZE - buf_offset);
995 bytes = min(bytes, working_bytes);
Cong Wang7ac687d2011-11-25 23:14:28 +0800996 kaddr = kmap_atomic(page_out);
Li Zefan3a39c182010-11-08 15:22:19 +0800997 memcpy(kaddr + *pg_offset, buf + buf_offset, bytes);
Cong Wang7ac687d2011-11-25 23:14:28 +0800998 kunmap_atomic(kaddr);
Li Zefan3a39c182010-11-08 15:22:19 +0800999 flush_dcache_page(page_out);
1000
1001 *pg_offset += bytes;
1002 buf_offset += bytes;
1003 working_bytes -= bytes;
1004 current_buf_start += bytes;
1005
1006 /* check if we need to pick another page */
1007 if (*pg_offset == PAGE_CACHE_SIZE) {
David Sterba306e16c2011-04-19 14:29:38 +02001008 (*pg_index)++;
1009 if (*pg_index >= vcnt)
Li Zefan3a39c182010-11-08 15:22:19 +08001010 return 0;
1011
David Sterba306e16c2011-04-19 14:29:38 +02001012 page_out = bvec[*pg_index].bv_page;
Li Zefan3a39c182010-11-08 15:22:19 +08001013 *pg_offset = 0;
1014 start_byte = page_offset(page_out) - disk_start;
1015
1016 /*
1017 * make sure our new page is covered by this
1018 * working buffer
1019 */
1020 if (total_out <= start_byte)
1021 return 1;
1022
1023 /*
1024 * the next page in the biovec might not be adjacent
1025 * to the last page, but it might still be found
1026 * inside this working buffer. bump our offset pointer
1027 */
1028 if (total_out > start_byte &&
1029 current_buf_start < start_byte) {
1030 buf_offset = start_byte - buf_start;
1031 working_bytes = total_out - start_byte;
1032 current_buf_start = buf_start + buf_offset;
1033 }
1034 }
1035 }
1036
1037 return 1;
1038}