]>
Commit | Line | Data |
---|---|---|
c8b97818 CM |
1 | /* |
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> | |
c8b97818 CM |
29 | #include <linux/backing-dev.h> |
30 | #include <linux/mpage.h> | |
31 | #include <linux/swap.h> | |
32 | #include <linux/writeback.h> | |
33 | #include <linux/bit_spinlock.h> | |
5a0e3ad6 | 34 | #include <linux/slab.h> |
4b4e25f2 | 35 | #include "compat.h" |
c8b97818 CM |
36 | #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" | |
c8b97818 CM |
42 | #include "compression.h" |
43 | #include "extent_io.h" | |
44 | #include "extent_map.h" | |
45 | ||
46 | struct 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 | ||
261507a0 LZ |
65 | /* the compression algorithm for this bio */ |
66 | int compress_type; | |
67 | ||
c8b97818 CM |
68 | /* number of compressed pages in the array */ |
69 | unsigned long nr_pages; | |
70 | ||
71 | /* IO errors */ | |
72 | int errors; | |
d20f7043 | 73 | int mirror_num; |
c8b97818 CM |
74 | |
75 | /* for reads, this is the bio we are copying the data into */ | |
76 | struct bio *orig_bio; | |
d20f7043 CM |
77 | |
78 | /* | |
79 | * the start of a variable length array of checksums only | |
80 | * used by reads | |
81 | */ | |
82 | u32 sums; | |
c8b97818 CM |
83 | }; |
84 | ||
48a3b636 ES |
85 | static int btrfs_decompress_biovec(int type, struct page **pages_in, |
86 | u64 disk_start, struct bio_vec *bvec, | |
87 | int vcnt, size_t srclen); | |
88 | ||
d20f7043 CM |
89 | static inline int compressed_bio_size(struct btrfs_root *root, |
90 | unsigned long disk_size) | |
91 | { | |
6c41761f DS |
92 | u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy); |
93 | ||
d20f7043 CM |
94 | return sizeof(struct compressed_bio) + |
95 | ((disk_size + root->sectorsize - 1) / root->sectorsize) * | |
96 | csum_size; | |
97 | } | |
98 | ||
c8b97818 CM |
99 | static struct bio *compressed_bio_alloc(struct block_device *bdev, |
100 | u64 first_byte, gfp_t gfp_flags) | |
101 | { | |
c8b97818 CM |
102 | int nr_vecs; |
103 | ||
104 | nr_vecs = bio_get_nr_vecs(bdev); | |
88f794ed | 105 | return btrfs_bio_alloc(bdev, first_byte >> 9, nr_vecs, gfp_flags); |
c8b97818 CM |
106 | } |
107 | ||
d20f7043 CM |
108 | static int check_compressed_csum(struct inode *inode, |
109 | struct compressed_bio *cb, | |
110 | u64 disk_start) | |
111 | { | |
112 | int ret; | |
d20f7043 CM |
113 | struct page *page; |
114 | unsigned long i; | |
115 | char *kaddr; | |
116 | u32 csum; | |
117 | u32 *cb_sum = &cb->sums; | |
118 | ||
6cbff00f | 119 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
d20f7043 CM |
120 | return 0; |
121 | ||
122 | for (i = 0; i < cb->nr_pages; i++) { | |
123 | page = cb->compressed_pages[i]; | |
124 | csum = ~(u32)0; | |
125 | ||
7ac687d9 | 126 | kaddr = kmap_atomic(page); |
b0496686 | 127 | csum = btrfs_csum_data(kaddr, csum, PAGE_CACHE_SIZE); |
d20f7043 | 128 | btrfs_csum_final(csum, (char *)&csum); |
7ac687d9 | 129 | kunmap_atomic(kaddr); |
d20f7043 CM |
130 | |
131 | if (csum != *cb_sum) { | |
33345d01 | 132 | printk(KERN_INFO "btrfs csum failed ino %llu " |
d397712b | 133 | "extent %llu csum %u " |
33345d01 LZ |
134 | "wanted %u mirror %d\n", |
135 | (unsigned long long)btrfs_ino(inode), | |
d20f7043 CM |
136 | (unsigned long long)disk_start, |
137 | csum, *cb_sum, cb->mirror_num); | |
138 | ret = -EIO; | |
139 | goto fail; | |
140 | } | |
141 | cb_sum++; | |
142 | ||
143 | } | |
144 | ret = 0; | |
145 | fail: | |
146 | return ret; | |
147 | } | |
148 | ||
c8b97818 CM |
149 | /* when we finish reading compressed pages from the disk, we |
150 | * decompress them and then run the bio end_io routines on the | |
151 | * decompressed pages (in the inode address space). | |
152 | * | |
153 | * This allows the checksumming and other IO error handling routines | |
154 | * to work normally | |
155 | * | |
156 | * The compressed pages are freed here, and it must be run | |
157 | * in process context | |
158 | */ | |
159 | static void end_compressed_bio_read(struct bio *bio, int err) | |
160 | { | |
c8b97818 CM |
161 | struct compressed_bio *cb = bio->bi_private; |
162 | struct inode *inode; | |
163 | struct page *page; | |
164 | unsigned long index; | |
165 | int ret; | |
166 | ||
167 | if (err) | |
168 | cb->errors = 1; | |
169 | ||
170 | /* if there are more bios still pending for this compressed | |
171 | * extent, just exit | |
172 | */ | |
173 | if (!atomic_dec_and_test(&cb->pending_bios)) | |
174 | goto out; | |
175 | ||
d20f7043 CM |
176 | inode = cb->inode; |
177 | ret = check_compressed_csum(inode, cb, (u64)bio->bi_sector << 9); | |
178 | if (ret) | |
179 | goto csum_failed; | |
180 | ||
c8b97818 CM |
181 | /* ok, we're the last bio for this extent, lets start |
182 | * the decompression. | |
183 | */ | |
261507a0 LZ |
184 | ret = btrfs_decompress_biovec(cb->compress_type, |
185 | cb->compressed_pages, | |
186 | cb->start, | |
187 | cb->orig_bio->bi_io_vec, | |
188 | cb->orig_bio->bi_vcnt, | |
189 | cb->compressed_len); | |
d20f7043 | 190 | csum_failed: |
c8b97818 CM |
191 | if (ret) |
192 | cb->errors = 1; | |
193 | ||
194 | /* release the compressed pages */ | |
195 | index = 0; | |
196 | for (index = 0; index < cb->nr_pages; index++) { | |
197 | page = cb->compressed_pages[index]; | |
198 | page->mapping = NULL; | |
199 | page_cache_release(page); | |
200 | } | |
201 | ||
202 | /* do io completion on the original bio */ | |
771ed689 | 203 | if (cb->errors) { |
c8b97818 | 204 | bio_io_error(cb->orig_bio); |
d20f7043 CM |
205 | } else { |
206 | int bio_index = 0; | |
207 | struct bio_vec *bvec = cb->orig_bio->bi_io_vec; | |
208 | ||
209 | /* | |
210 | * we have verified the checksum already, set page | |
211 | * checked so the end_io handlers know about it | |
212 | */ | |
d397712b | 213 | while (bio_index < cb->orig_bio->bi_vcnt) { |
d20f7043 CM |
214 | SetPageChecked(bvec->bv_page); |
215 | bvec++; | |
216 | bio_index++; | |
217 | } | |
c8b97818 | 218 | bio_endio(cb->orig_bio, 0); |
d20f7043 | 219 | } |
c8b97818 CM |
220 | |
221 | /* finally free the cb struct */ | |
222 | kfree(cb->compressed_pages); | |
223 | kfree(cb); | |
224 | out: | |
225 | bio_put(bio); | |
226 | } | |
227 | ||
228 | /* | |
229 | * Clear the writeback bits on all of the file | |
230 | * pages for a compressed write | |
231 | */ | |
143bede5 JM |
232 | static noinline void end_compressed_writeback(struct inode *inode, u64 start, |
233 | unsigned long ram_size) | |
c8b97818 CM |
234 | { |
235 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
236 | unsigned long end_index = (start + ram_size - 1) >> PAGE_CACHE_SHIFT; | |
237 | struct page *pages[16]; | |
238 | unsigned long nr_pages = end_index - index + 1; | |
239 | int i; | |
240 | int ret; | |
241 | ||
d397712b | 242 | while (nr_pages > 0) { |
c8b97818 | 243 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
244 | min_t(unsigned long, |
245 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
246 | if (ret == 0) { |
247 | nr_pages -= 1; | |
248 | index += 1; | |
249 | continue; | |
250 | } | |
251 | for (i = 0; i < ret; i++) { | |
252 | end_page_writeback(pages[i]); | |
253 | page_cache_release(pages[i]); | |
254 | } | |
255 | nr_pages -= ret; | |
256 | index += ret; | |
257 | } | |
258 | /* the inode may be gone now */ | |
c8b97818 CM |
259 | } |
260 | ||
261 | /* | |
262 | * do the cleanup once all the compressed pages hit the disk. | |
263 | * This will clear writeback on the file pages and free the compressed | |
264 | * pages. | |
265 | * | |
266 | * This also calls the writeback end hooks for the file pages so that | |
267 | * metadata and checksums can be updated in the file. | |
268 | */ | |
269 | static void end_compressed_bio_write(struct bio *bio, int err) | |
270 | { | |
271 | struct extent_io_tree *tree; | |
272 | struct compressed_bio *cb = bio->bi_private; | |
273 | struct inode *inode; | |
274 | struct page *page; | |
275 | unsigned long index; | |
276 | ||
277 | if (err) | |
278 | cb->errors = 1; | |
279 | ||
280 | /* if there are more bios still pending for this compressed | |
281 | * extent, just exit | |
282 | */ | |
283 | if (!atomic_dec_and_test(&cb->pending_bios)) | |
284 | goto out; | |
285 | ||
286 | /* ok, we're the last bio for this extent, step one is to | |
287 | * call back into the FS and do all the end_io operations | |
288 | */ | |
289 | inode = cb->inode; | |
290 | tree = &BTRFS_I(inode)->io_tree; | |
70b99e69 | 291 | cb->compressed_pages[0]->mapping = cb->inode->i_mapping; |
c8b97818 CM |
292 | tree->ops->writepage_end_io_hook(cb->compressed_pages[0], |
293 | cb->start, | |
294 | cb->start + cb->len - 1, | |
295 | NULL, 1); | |
70b99e69 | 296 | cb->compressed_pages[0]->mapping = NULL; |
c8b97818 CM |
297 | |
298 | end_compressed_writeback(inode, cb->start, cb->len); | |
299 | /* note, our inode could be gone now */ | |
300 | ||
301 | /* | |
302 | * release the compressed pages, these came from alloc_page and | |
303 | * are not attached to the inode at all | |
304 | */ | |
305 | index = 0; | |
306 | for (index = 0; index < cb->nr_pages; index++) { | |
307 | page = cb->compressed_pages[index]; | |
308 | page->mapping = NULL; | |
309 | page_cache_release(page); | |
310 | } | |
311 | ||
312 | /* finally free the cb struct */ | |
313 | kfree(cb->compressed_pages); | |
314 | kfree(cb); | |
315 | out: | |
316 | bio_put(bio); | |
317 | } | |
318 | ||
319 | /* | |
320 | * worker function to build and submit bios for previously compressed pages. | |
321 | * The corresponding pages in the inode should be marked for writeback | |
322 | * and the compressed pages should have a reference on them for dropping | |
323 | * when the IO is complete. | |
324 | * | |
325 | * This also checksums the file bytes and gets things ready for | |
326 | * the end io hooks. | |
327 | */ | |
328 | int btrfs_submit_compressed_write(struct inode *inode, u64 start, | |
329 | unsigned long len, u64 disk_start, | |
330 | unsigned long compressed_len, | |
331 | struct page **compressed_pages, | |
332 | unsigned long nr_pages) | |
333 | { | |
334 | struct bio *bio = NULL; | |
335 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
336 | struct compressed_bio *cb; | |
337 | unsigned long bytes_left; | |
338 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
306e16ce | 339 | int pg_index = 0; |
c8b97818 CM |
340 | struct page *page; |
341 | u64 first_byte = disk_start; | |
342 | struct block_device *bdev; | |
343 | int ret; | |
e55179b3 | 344 | int skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
c8b97818 CM |
345 | |
346 | WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1)); | |
d20f7043 | 347 | cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); |
dac97e51 YS |
348 | if (!cb) |
349 | return -ENOMEM; | |
c8b97818 CM |
350 | atomic_set(&cb->pending_bios, 0); |
351 | cb->errors = 0; | |
352 | cb->inode = inode; | |
353 | cb->start = start; | |
354 | cb->len = len; | |
d20f7043 | 355 | cb->mirror_num = 0; |
c8b97818 CM |
356 | cb->compressed_pages = compressed_pages; |
357 | cb->compressed_len = compressed_len; | |
358 | cb->orig_bio = NULL; | |
359 | cb->nr_pages = nr_pages; | |
360 | ||
361 | bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; | |
362 | ||
c8b97818 | 363 | bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); |
dac97e51 YS |
364 | if(!bio) { |
365 | kfree(cb); | |
366 | return -ENOMEM; | |
367 | } | |
c8b97818 CM |
368 | bio->bi_private = cb; |
369 | bio->bi_end_io = end_compressed_bio_write; | |
370 | atomic_inc(&cb->pending_bios); | |
371 | ||
372 | /* create and submit bios for the compressed pages */ | |
373 | bytes_left = compressed_len; | |
306e16ce DS |
374 | for (pg_index = 0; pg_index < cb->nr_pages; pg_index++) { |
375 | page = compressed_pages[pg_index]; | |
c8b97818 CM |
376 | page->mapping = inode->i_mapping; |
377 | if (bio->bi_size) | |
64a16701 | 378 | ret = io_tree->ops->merge_bio_hook(WRITE, page, 0, |
c8b97818 CM |
379 | PAGE_CACHE_SIZE, |
380 | bio, 0); | |
381 | else | |
382 | ret = 0; | |
383 | ||
70b99e69 | 384 | page->mapping = NULL; |
c8b97818 CM |
385 | if (ret || bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < |
386 | PAGE_CACHE_SIZE) { | |
387 | bio_get(bio); | |
388 | ||
af09abfe CM |
389 | /* |
390 | * inc the count before we submit the bio so | |
391 | * we know the end IO handler won't happen before | |
392 | * we inc the count. Otherwise, the cb might get | |
393 | * freed before we're done setting it up | |
394 | */ | |
395 | atomic_inc(&cb->pending_bios); | |
c8b97818 | 396 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
79787eaa | 397 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 398 | |
e55179b3 LZ |
399 | if (!skip_sum) { |
400 | ret = btrfs_csum_one_bio(root, inode, bio, | |
401 | start, 1); | |
79787eaa | 402 | BUG_ON(ret); /* -ENOMEM */ |
e55179b3 | 403 | } |
d20f7043 | 404 | |
c8b97818 | 405 | ret = btrfs_map_bio(root, WRITE, bio, 0, 1); |
79787eaa | 406 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
407 | |
408 | bio_put(bio); | |
409 | ||
410 | bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); | |
e627ee7b | 411 | BUG_ON(!bio); |
c8b97818 CM |
412 | bio->bi_private = cb; |
413 | bio->bi_end_io = end_compressed_bio_write; | |
414 | bio_add_page(bio, page, PAGE_CACHE_SIZE, 0); | |
415 | } | |
cfbc246e CM |
416 | if (bytes_left < PAGE_CACHE_SIZE) { |
417 | printk("bytes left %lu compress len %lu nr %lu\n", | |
418 | bytes_left, cb->compressed_len, cb->nr_pages); | |
419 | } | |
c8b97818 CM |
420 | bytes_left -= PAGE_CACHE_SIZE; |
421 | first_byte += PAGE_CACHE_SIZE; | |
771ed689 | 422 | cond_resched(); |
c8b97818 CM |
423 | } |
424 | bio_get(bio); | |
425 | ||
426 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
79787eaa | 427 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 428 | |
e55179b3 LZ |
429 | if (!skip_sum) { |
430 | ret = btrfs_csum_one_bio(root, inode, bio, start, 1); | |
79787eaa | 431 | BUG_ON(ret); /* -ENOMEM */ |
e55179b3 | 432 | } |
d20f7043 | 433 | |
c8b97818 | 434 | ret = btrfs_map_bio(root, WRITE, bio, 0, 1); |
79787eaa | 435 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
436 | |
437 | bio_put(bio); | |
438 | return 0; | |
439 | } | |
440 | ||
771ed689 CM |
441 | static noinline int add_ra_bio_pages(struct inode *inode, |
442 | u64 compressed_end, | |
443 | struct compressed_bio *cb) | |
444 | { | |
445 | unsigned long end_index; | |
306e16ce | 446 | unsigned long pg_index; |
771ed689 CM |
447 | u64 last_offset; |
448 | u64 isize = i_size_read(inode); | |
449 | int ret; | |
450 | struct page *page; | |
451 | unsigned long nr_pages = 0; | |
452 | struct extent_map *em; | |
453 | struct address_space *mapping = inode->i_mapping; | |
771ed689 CM |
454 | struct extent_map_tree *em_tree; |
455 | struct extent_io_tree *tree; | |
456 | u64 end; | |
457 | int misses = 0; | |
458 | ||
459 | page = cb->orig_bio->bi_io_vec[cb->orig_bio->bi_vcnt - 1].bv_page; | |
460 | last_offset = (page_offset(page) + PAGE_CACHE_SIZE); | |
461 | em_tree = &BTRFS_I(inode)->extent_tree; | |
462 | tree = &BTRFS_I(inode)->io_tree; | |
463 | ||
464 | if (isize == 0) | |
465 | return 0; | |
466 | ||
467 | end_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; | |
468 | ||
d397712b | 469 | while (last_offset < compressed_end) { |
306e16ce | 470 | pg_index = last_offset >> PAGE_CACHE_SHIFT; |
771ed689 | 471 | |
306e16ce | 472 | if (pg_index > end_index) |
771ed689 CM |
473 | break; |
474 | ||
475 | rcu_read_lock(); | |
306e16ce | 476 | page = radix_tree_lookup(&mapping->page_tree, pg_index); |
771ed689 CM |
477 | rcu_read_unlock(); |
478 | if (page) { | |
479 | misses++; | |
480 | if (misses > 4) | |
481 | break; | |
482 | goto next; | |
483 | } | |
484 | ||
28ecb609 NP |
485 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & |
486 | ~__GFP_FS); | |
771ed689 CM |
487 | if (!page) |
488 | break; | |
489 | ||
306e16ce | 490 | if (add_to_page_cache_lru(page, mapping, pg_index, |
28ecb609 | 491 | GFP_NOFS)) { |
771ed689 CM |
492 | page_cache_release(page); |
493 | goto next; | |
494 | } | |
495 | ||
771ed689 CM |
496 | end = last_offset + PAGE_CACHE_SIZE - 1; |
497 | /* | |
498 | * at this point, we have a locked page in the page cache | |
499 | * for these bytes in the file. But, we have to make | |
500 | * sure they map to this compressed extent on disk. | |
501 | */ | |
502 | set_page_extent_mapped(page); | |
d0082371 | 503 | lock_extent(tree, last_offset, end); |
890871be | 504 | read_lock(&em_tree->lock); |
771ed689 CM |
505 | em = lookup_extent_mapping(em_tree, last_offset, |
506 | PAGE_CACHE_SIZE); | |
890871be | 507 | read_unlock(&em_tree->lock); |
771ed689 CM |
508 | |
509 | if (!em || last_offset < em->start || | |
510 | (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) || | |
511 | (em->block_start >> 9) != cb->orig_bio->bi_sector) { | |
512 | free_extent_map(em); | |
d0082371 | 513 | unlock_extent(tree, last_offset, end); |
771ed689 CM |
514 | unlock_page(page); |
515 | page_cache_release(page); | |
516 | break; | |
517 | } | |
518 | free_extent_map(em); | |
519 | ||
520 | if (page->index == end_index) { | |
521 | char *userpage; | |
522 | size_t zero_offset = isize & (PAGE_CACHE_SIZE - 1); | |
523 | ||
524 | if (zero_offset) { | |
525 | int zeros; | |
526 | zeros = PAGE_CACHE_SIZE - zero_offset; | |
7ac687d9 | 527 | userpage = kmap_atomic(page); |
771ed689 CM |
528 | memset(userpage + zero_offset, 0, zeros); |
529 | flush_dcache_page(page); | |
7ac687d9 | 530 | kunmap_atomic(userpage); |
771ed689 CM |
531 | } |
532 | } | |
533 | ||
534 | ret = bio_add_page(cb->orig_bio, page, | |
535 | PAGE_CACHE_SIZE, 0); | |
536 | ||
537 | if (ret == PAGE_CACHE_SIZE) { | |
538 | nr_pages++; | |
539 | page_cache_release(page); | |
540 | } else { | |
d0082371 | 541 | unlock_extent(tree, last_offset, end); |
771ed689 CM |
542 | unlock_page(page); |
543 | page_cache_release(page); | |
544 | break; | |
545 | } | |
546 | next: | |
547 | last_offset += PAGE_CACHE_SIZE; | |
548 | } | |
771ed689 CM |
549 | return 0; |
550 | } | |
551 | ||
c8b97818 CM |
552 | /* |
553 | * for a compressed read, the bio we get passed has all the inode pages | |
554 | * in it. We don't actually do IO on those pages but allocate new ones | |
555 | * to hold the compressed pages on disk. | |
556 | * | |
557 | * bio->bi_sector points to the compressed extent on disk | |
558 | * bio->bi_io_vec points to all of the inode pages | |
559 | * bio->bi_vcnt is a count of pages | |
560 | * | |
561 | * After the compressed pages are read, we copy the bytes into the | |
562 | * bio we were passed and then call the bio end_io calls | |
563 | */ | |
564 | int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio, | |
565 | int mirror_num, unsigned long bio_flags) | |
566 | { | |
567 | struct extent_io_tree *tree; | |
568 | struct extent_map_tree *em_tree; | |
569 | struct compressed_bio *cb; | |
570 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
571 | unsigned long uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE; | |
572 | unsigned long compressed_len; | |
573 | unsigned long nr_pages; | |
306e16ce | 574 | unsigned long pg_index; |
c8b97818 CM |
575 | struct page *page; |
576 | struct block_device *bdev; | |
577 | struct bio *comp_bio; | |
578 | u64 cur_disk_byte = (u64)bio->bi_sector << 9; | |
e04ca626 CM |
579 | u64 em_len; |
580 | u64 em_start; | |
c8b97818 | 581 | struct extent_map *em; |
6b82ce8d | 582 | int ret = -ENOMEM; |
15e3004a | 583 | int faili = 0; |
d20f7043 | 584 | u32 *sums; |
c8b97818 CM |
585 | |
586 | tree = &BTRFS_I(inode)->io_tree; | |
587 | em_tree = &BTRFS_I(inode)->extent_tree; | |
588 | ||
589 | /* we need the actual starting offset of this extent in the file */ | |
890871be | 590 | read_lock(&em_tree->lock); |
c8b97818 CM |
591 | em = lookup_extent_mapping(em_tree, |
592 | page_offset(bio->bi_io_vec->bv_page), | |
593 | PAGE_CACHE_SIZE); | |
890871be | 594 | read_unlock(&em_tree->lock); |
285190d9 TI |
595 | if (!em) |
596 | return -EIO; | |
c8b97818 | 597 | |
d20f7043 CM |
598 | compressed_len = em->block_len; |
599 | cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); | |
6b82ce8d | 600 | if (!cb) |
601 | goto out; | |
602 | ||
c8b97818 CM |
603 | atomic_set(&cb->pending_bios, 0); |
604 | cb->errors = 0; | |
605 | cb->inode = inode; | |
d20f7043 CM |
606 | cb->mirror_num = mirror_num; |
607 | sums = &cb->sums; | |
c8b97818 | 608 | |
ff5b7ee3 | 609 | cb->start = em->orig_start; |
e04ca626 CM |
610 | em_len = em->len; |
611 | em_start = em->start; | |
d20f7043 | 612 | |
c8b97818 | 613 | free_extent_map(em); |
e04ca626 | 614 | em = NULL; |
c8b97818 CM |
615 | |
616 | cb->len = uncompressed_len; | |
617 | cb->compressed_len = compressed_len; | |
261507a0 | 618 | cb->compress_type = extent_compress_type(bio_flags); |
c8b97818 CM |
619 | cb->orig_bio = bio; |
620 | ||
621 | nr_pages = (compressed_len + PAGE_CACHE_SIZE - 1) / | |
622 | PAGE_CACHE_SIZE; | |
6b82ce8d | 623 | cb->compressed_pages = kzalloc(sizeof(struct page *) * nr_pages, |
c8b97818 | 624 | GFP_NOFS); |
6b82ce8d | 625 | if (!cb->compressed_pages) |
626 | goto fail1; | |
627 | ||
c8b97818 CM |
628 | bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; |
629 | ||
306e16ce DS |
630 | for (pg_index = 0; pg_index < nr_pages; pg_index++) { |
631 | cb->compressed_pages[pg_index] = alloc_page(GFP_NOFS | | |
c8b97818 | 632 | __GFP_HIGHMEM); |
15e3004a JB |
633 | if (!cb->compressed_pages[pg_index]) { |
634 | faili = pg_index - 1; | |
635 | ret = -ENOMEM; | |
6b82ce8d | 636 | goto fail2; |
15e3004a | 637 | } |
c8b97818 | 638 | } |
15e3004a | 639 | faili = nr_pages - 1; |
c8b97818 CM |
640 | cb->nr_pages = nr_pages; |
641 | ||
e04ca626 | 642 | add_ra_bio_pages(inode, em_start + em_len, cb); |
771ed689 | 643 | |
771ed689 CM |
644 | /* include any pages we added in add_ra-bio_pages */ |
645 | uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE; | |
646 | cb->len = uncompressed_len; | |
647 | ||
c8b97818 | 648 | comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS); |
6b82ce8d | 649 | if (!comp_bio) |
650 | goto fail2; | |
c8b97818 CM |
651 | comp_bio->bi_private = cb; |
652 | comp_bio->bi_end_io = end_compressed_bio_read; | |
653 | atomic_inc(&cb->pending_bios); | |
654 | ||
306e16ce DS |
655 | for (pg_index = 0; pg_index < nr_pages; pg_index++) { |
656 | page = cb->compressed_pages[pg_index]; | |
c8b97818 | 657 | page->mapping = inode->i_mapping; |
d20f7043 CM |
658 | page->index = em_start >> PAGE_CACHE_SHIFT; |
659 | ||
c8b97818 | 660 | if (comp_bio->bi_size) |
64a16701 | 661 | ret = tree->ops->merge_bio_hook(READ, page, 0, |
c8b97818 CM |
662 | PAGE_CACHE_SIZE, |
663 | comp_bio, 0); | |
664 | else | |
665 | ret = 0; | |
666 | ||
70b99e69 | 667 | page->mapping = NULL; |
c8b97818 CM |
668 | if (ret || bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0) < |
669 | PAGE_CACHE_SIZE) { | |
670 | bio_get(comp_bio); | |
671 | ||
672 | ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0); | |
79787eaa | 673 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 674 | |
af09abfe CM |
675 | /* |
676 | * inc the count before we submit the bio so | |
677 | * we know the end IO handler won't happen before | |
678 | * we inc the count. Otherwise, the cb might get | |
679 | * freed before we're done setting it up | |
680 | */ | |
681 | atomic_inc(&cb->pending_bios); | |
682 | ||
6cbff00f | 683 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { |
c2db1073 TI |
684 | ret = btrfs_lookup_bio_sums(root, inode, |
685 | comp_bio, sums); | |
79787eaa | 686 | BUG_ON(ret); /* -ENOMEM */ |
d20f7043 CM |
687 | } |
688 | sums += (comp_bio->bi_size + root->sectorsize - 1) / | |
689 | root->sectorsize; | |
690 | ||
691 | ret = btrfs_map_bio(root, READ, comp_bio, | |
692 | mirror_num, 0); | |
61891923 SB |
693 | if (ret) |
694 | bio_endio(comp_bio, ret); | |
c8b97818 CM |
695 | |
696 | bio_put(comp_bio); | |
697 | ||
698 | comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, | |
699 | GFP_NOFS); | |
e627ee7b | 700 | BUG_ON(!comp_bio); |
771ed689 CM |
701 | comp_bio->bi_private = cb; |
702 | comp_bio->bi_end_io = end_compressed_bio_read; | |
703 | ||
704 | bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0); | |
c8b97818 CM |
705 | } |
706 | cur_disk_byte += PAGE_CACHE_SIZE; | |
707 | } | |
708 | bio_get(comp_bio); | |
709 | ||
710 | ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0); | |
79787eaa | 711 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 712 | |
c2db1073 TI |
713 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { |
714 | ret = btrfs_lookup_bio_sums(root, inode, comp_bio, sums); | |
79787eaa | 715 | BUG_ON(ret); /* -ENOMEM */ |
c2db1073 | 716 | } |
d20f7043 CM |
717 | |
718 | ret = btrfs_map_bio(root, READ, comp_bio, mirror_num, 0); | |
61891923 SB |
719 | if (ret) |
720 | bio_endio(comp_bio, ret); | |
c8b97818 CM |
721 | |
722 | bio_put(comp_bio); | |
723 | return 0; | |
6b82ce8d | 724 | |
725 | fail2: | |
15e3004a JB |
726 | while (faili >= 0) { |
727 | __free_page(cb->compressed_pages[faili]); | |
728 | faili--; | |
729 | } | |
6b82ce8d | 730 | |
731 | kfree(cb->compressed_pages); | |
732 | fail1: | |
733 | kfree(cb); | |
734 | out: | |
735 | free_extent_map(em); | |
736 | return ret; | |
c8b97818 | 737 | } |
261507a0 LZ |
738 | |
739 | static struct list_head comp_idle_workspace[BTRFS_COMPRESS_TYPES]; | |
740 | static spinlock_t comp_workspace_lock[BTRFS_COMPRESS_TYPES]; | |
741 | static int comp_num_workspace[BTRFS_COMPRESS_TYPES]; | |
742 | static atomic_t comp_alloc_workspace[BTRFS_COMPRESS_TYPES]; | |
743 | static wait_queue_head_t comp_workspace_wait[BTRFS_COMPRESS_TYPES]; | |
744 | ||
48a3b636 | 745 | static struct btrfs_compress_op *btrfs_compress_op[] = { |
261507a0 | 746 | &btrfs_zlib_compress, |
a6fa6fae | 747 | &btrfs_lzo_compress, |
261507a0 LZ |
748 | }; |
749 | ||
143bede5 | 750 | void __init btrfs_init_compress(void) |
261507a0 LZ |
751 | { |
752 | int i; | |
753 | ||
754 | for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) { | |
755 | INIT_LIST_HEAD(&comp_idle_workspace[i]); | |
756 | spin_lock_init(&comp_workspace_lock[i]); | |
757 | atomic_set(&comp_alloc_workspace[i], 0); | |
758 | init_waitqueue_head(&comp_workspace_wait[i]); | |
759 | } | |
261507a0 LZ |
760 | } |
761 | ||
762 | /* | |
763 | * this finds an available workspace or allocates a new one | |
764 | * ERR_PTR is returned if things go bad. | |
765 | */ | |
766 | static struct list_head *find_workspace(int type) | |
767 | { | |
768 | struct list_head *workspace; | |
769 | int cpus = num_online_cpus(); | |
770 | int idx = type - 1; | |
771 | ||
772 | struct list_head *idle_workspace = &comp_idle_workspace[idx]; | |
773 | spinlock_t *workspace_lock = &comp_workspace_lock[idx]; | |
774 | atomic_t *alloc_workspace = &comp_alloc_workspace[idx]; | |
775 | wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx]; | |
776 | int *num_workspace = &comp_num_workspace[idx]; | |
777 | again: | |
778 | spin_lock(workspace_lock); | |
779 | if (!list_empty(idle_workspace)) { | |
780 | workspace = idle_workspace->next; | |
781 | list_del(workspace); | |
782 | (*num_workspace)--; | |
783 | spin_unlock(workspace_lock); | |
784 | return workspace; | |
785 | ||
786 | } | |
787 | if (atomic_read(alloc_workspace) > cpus) { | |
788 | DEFINE_WAIT(wait); | |
789 | ||
790 | spin_unlock(workspace_lock); | |
791 | prepare_to_wait(workspace_wait, &wait, TASK_UNINTERRUPTIBLE); | |
792 | if (atomic_read(alloc_workspace) > cpus && !*num_workspace) | |
793 | schedule(); | |
794 | finish_wait(workspace_wait, &wait); | |
795 | goto again; | |
796 | } | |
797 | atomic_inc(alloc_workspace); | |
798 | spin_unlock(workspace_lock); | |
799 | ||
800 | workspace = btrfs_compress_op[idx]->alloc_workspace(); | |
801 | if (IS_ERR(workspace)) { | |
802 | atomic_dec(alloc_workspace); | |
803 | wake_up(workspace_wait); | |
804 | } | |
805 | return workspace; | |
806 | } | |
807 | ||
808 | /* | |
809 | * put a workspace struct back on the list or free it if we have enough | |
810 | * idle ones sitting around | |
811 | */ | |
812 | static void free_workspace(int type, struct list_head *workspace) | |
813 | { | |
814 | int idx = type - 1; | |
815 | struct list_head *idle_workspace = &comp_idle_workspace[idx]; | |
816 | spinlock_t *workspace_lock = &comp_workspace_lock[idx]; | |
817 | atomic_t *alloc_workspace = &comp_alloc_workspace[idx]; | |
818 | wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx]; | |
819 | int *num_workspace = &comp_num_workspace[idx]; | |
820 | ||
821 | spin_lock(workspace_lock); | |
822 | if (*num_workspace < num_online_cpus()) { | |
823 | list_add_tail(workspace, idle_workspace); | |
824 | (*num_workspace)++; | |
825 | spin_unlock(workspace_lock); | |
826 | goto wake; | |
827 | } | |
828 | spin_unlock(workspace_lock); | |
829 | ||
830 | btrfs_compress_op[idx]->free_workspace(workspace); | |
831 | atomic_dec(alloc_workspace); | |
832 | wake: | |
66657b31 | 833 | smp_mb(); |
261507a0 LZ |
834 | if (waitqueue_active(workspace_wait)) |
835 | wake_up(workspace_wait); | |
836 | } | |
837 | ||
838 | /* | |
839 | * cleanup function for module exit | |
840 | */ | |
841 | static void free_workspaces(void) | |
842 | { | |
843 | struct list_head *workspace; | |
844 | int i; | |
845 | ||
846 | for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) { | |
847 | while (!list_empty(&comp_idle_workspace[i])) { | |
848 | workspace = comp_idle_workspace[i].next; | |
849 | list_del(workspace); | |
850 | btrfs_compress_op[i]->free_workspace(workspace); | |
851 | atomic_dec(&comp_alloc_workspace[i]); | |
852 | } | |
853 | } | |
854 | } | |
855 | ||
856 | /* | |
857 | * given an address space and start/len, compress the bytes. | |
858 | * | |
859 | * pages are allocated to hold the compressed result and stored | |
860 | * in 'pages' | |
861 | * | |
862 | * out_pages is used to return the number of pages allocated. There | |
863 | * may be pages allocated even if we return an error | |
864 | * | |
865 | * total_in is used to return the number of bytes actually read. It | |
866 | * may be smaller then len if we had to exit early because we | |
867 | * ran out of room in the pages array or because we cross the | |
868 | * max_out threshold. | |
869 | * | |
870 | * total_out is used to return the total number of compressed bytes | |
871 | * | |
872 | * max_out tells us the max number of bytes that we're allowed to | |
873 | * stuff into pages | |
874 | */ | |
875 | int btrfs_compress_pages(int type, struct address_space *mapping, | |
876 | u64 start, unsigned long len, | |
877 | struct page **pages, | |
878 | unsigned long nr_dest_pages, | |
879 | unsigned long *out_pages, | |
880 | unsigned long *total_in, | |
881 | unsigned long *total_out, | |
882 | unsigned long max_out) | |
883 | { | |
884 | struct list_head *workspace; | |
885 | int ret; | |
886 | ||
887 | workspace = find_workspace(type); | |
888 | if (IS_ERR(workspace)) | |
889 | return -1; | |
890 | ||
891 | ret = btrfs_compress_op[type-1]->compress_pages(workspace, mapping, | |
892 | start, len, pages, | |
893 | nr_dest_pages, out_pages, | |
894 | total_in, total_out, | |
895 | max_out); | |
896 | free_workspace(type, workspace); | |
897 | return ret; | |
898 | } | |
899 | ||
900 | /* | |
901 | * pages_in is an array of pages with compressed data. | |
902 | * | |
903 | * disk_start is the starting logical offset of this array in the file | |
904 | * | |
905 | * bvec is a bio_vec of pages from the file that we want to decompress into | |
906 | * | |
907 | * vcnt is the count of pages in the biovec | |
908 | * | |
909 | * srclen is the number of bytes in pages_in | |
910 | * | |
911 | * The basic idea is that we have a bio that was created by readpages. | |
912 | * The pages in the bio are for the uncompressed data, and they may not | |
913 | * be contiguous. They all correspond to the range of bytes covered by | |
914 | * the compressed extent. | |
915 | */ | |
48a3b636 ES |
916 | static int btrfs_decompress_biovec(int type, struct page **pages_in, |
917 | u64 disk_start, struct bio_vec *bvec, | |
918 | int vcnt, size_t srclen) | |
261507a0 LZ |
919 | { |
920 | struct list_head *workspace; | |
921 | int ret; | |
922 | ||
923 | workspace = find_workspace(type); | |
924 | if (IS_ERR(workspace)) | |
925 | return -ENOMEM; | |
926 | ||
927 | ret = btrfs_compress_op[type-1]->decompress_biovec(workspace, pages_in, | |
928 | disk_start, | |
929 | bvec, vcnt, srclen); | |
930 | free_workspace(type, workspace); | |
931 | return ret; | |
932 | } | |
933 | ||
934 | /* | |
935 | * a less complex decompression routine. Our compressed data fits in a | |
936 | * single page, and we want to read a single page out of it. | |
937 | * start_byte tells us the offset into the compressed data we're interested in | |
938 | */ | |
939 | int btrfs_decompress(int type, unsigned char *data_in, struct page *dest_page, | |
940 | unsigned long start_byte, size_t srclen, size_t destlen) | |
941 | { | |
942 | struct list_head *workspace; | |
943 | int ret; | |
944 | ||
945 | workspace = find_workspace(type); | |
946 | if (IS_ERR(workspace)) | |
947 | return -ENOMEM; | |
948 | ||
949 | ret = btrfs_compress_op[type-1]->decompress(workspace, data_in, | |
950 | dest_page, start_byte, | |
951 | srclen, destlen); | |
952 | ||
953 | free_workspace(type, workspace); | |
954 | return ret; | |
955 | } | |
956 | ||
8e4eef7a | 957 | void btrfs_exit_compress(void) |
261507a0 LZ |
958 | { |
959 | free_workspaces(); | |
960 | } | |
3a39c18d LZ |
961 | |
962 | /* | |
963 | * Copy uncompressed data from working buffer to pages. | |
964 | * | |
965 | * buf_start is the byte offset we're of the start of our workspace buffer. | |
966 | * | |
967 | * total_out is the last byte of the buffer | |
968 | */ | |
969 | int btrfs_decompress_buf2page(char *buf, unsigned long buf_start, | |
970 | unsigned long total_out, u64 disk_start, | |
971 | struct bio_vec *bvec, int vcnt, | |
306e16ce | 972 | unsigned long *pg_index, |
3a39c18d LZ |
973 | unsigned long *pg_offset) |
974 | { | |
975 | unsigned long buf_offset; | |
976 | unsigned long current_buf_start; | |
977 | unsigned long start_byte; | |
978 | unsigned long working_bytes = total_out - buf_start; | |
979 | unsigned long bytes; | |
980 | char *kaddr; | |
306e16ce | 981 | struct page *page_out = bvec[*pg_index].bv_page; |
3a39c18d LZ |
982 | |
983 | /* | |
984 | * start byte is the first byte of the page we're currently | |
985 | * copying into relative to the start of the compressed data. | |
986 | */ | |
987 | start_byte = page_offset(page_out) - disk_start; | |
988 | ||
989 | /* we haven't yet hit data corresponding to this page */ | |
990 | if (total_out <= start_byte) | |
991 | return 1; | |
992 | ||
993 | /* | |
994 | * the start of the data we care about is offset into | |
995 | * the middle of our working buffer | |
996 | */ | |
997 | if (total_out > start_byte && buf_start < start_byte) { | |
998 | buf_offset = start_byte - buf_start; | |
999 | working_bytes -= buf_offset; | |
1000 | } else { | |
1001 | buf_offset = 0; | |
1002 | } | |
1003 | current_buf_start = buf_start; | |
1004 | ||
1005 | /* copy bytes from the working buffer into the pages */ | |
1006 | while (working_bytes > 0) { | |
1007 | bytes = min(PAGE_CACHE_SIZE - *pg_offset, | |
1008 | PAGE_CACHE_SIZE - buf_offset); | |
1009 | bytes = min(bytes, working_bytes); | |
7ac687d9 | 1010 | kaddr = kmap_atomic(page_out); |
3a39c18d | 1011 | memcpy(kaddr + *pg_offset, buf + buf_offset, bytes); |
7ac687d9 | 1012 | kunmap_atomic(kaddr); |
3a39c18d LZ |
1013 | flush_dcache_page(page_out); |
1014 | ||
1015 | *pg_offset += bytes; | |
1016 | buf_offset += bytes; | |
1017 | working_bytes -= bytes; | |
1018 | current_buf_start += bytes; | |
1019 | ||
1020 | /* check if we need to pick another page */ | |
1021 | if (*pg_offset == PAGE_CACHE_SIZE) { | |
306e16ce DS |
1022 | (*pg_index)++; |
1023 | if (*pg_index >= vcnt) | |
3a39c18d LZ |
1024 | return 0; |
1025 | ||
306e16ce | 1026 | page_out = bvec[*pg_index].bv_page; |
3a39c18d LZ |
1027 | *pg_offset = 0; |
1028 | start_byte = page_offset(page_out) - disk_start; | |
1029 | ||
1030 | /* | |
1031 | * make sure our new page is covered by this | |
1032 | * working buffer | |
1033 | */ | |
1034 | if (total_out <= start_byte) | |
1035 | return 1; | |
1036 | ||
1037 | /* | |
1038 | * the next page in the biovec might not be adjacent | |
1039 | * to the last page, but it might still be found | |
1040 | * inside this working buffer. bump our offset pointer | |
1041 | */ | |
1042 | if (total_out > start_byte && | |
1043 | current_buf_start < start_byte) { | |
1044 | buf_offset = start_byte - buf_start; | |
1045 | working_bytes = total_out - start_byte; | |
1046 | current_buf_start = buf_start + buf_offset; | |
1047 | } | |
1048 | } | |
1049 | } | |
1050 | ||
1051 | return 1; | |
1052 | } |