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 | ||
d20f7043 CM |
85 | static inline int compressed_bio_size(struct btrfs_root *root, |
86 | unsigned long disk_size) | |
87 | { | |
6c41761f DS |
88 | u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy); |
89 | ||
d20f7043 CM |
90 | return sizeof(struct compressed_bio) + |
91 | ((disk_size + root->sectorsize - 1) / root->sectorsize) * | |
92 | csum_size; | |
93 | } | |
94 | ||
c8b97818 CM |
95 | static struct bio *compressed_bio_alloc(struct block_device *bdev, |
96 | u64 first_byte, gfp_t gfp_flags) | |
97 | { | |
c8b97818 CM |
98 | int nr_vecs; |
99 | ||
100 | nr_vecs = bio_get_nr_vecs(bdev); | |
88f794ed | 101 | return btrfs_bio_alloc(bdev, first_byte >> 9, nr_vecs, gfp_flags); |
c8b97818 CM |
102 | } |
103 | ||
d20f7043 CM |
104 | static 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 | ||
6cbff00f | 116 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
d20f7043 CM |
117 | return 0; |
118 | ||
119 | for (i = 0; i < cb->nr_pages; i++) { | |
120 | page = cb->compressed_pages[i]; | |
121 | csum = ~(u32)0; | |
122 | ||
7ac687d9 | 123 | kaddr = kmap_atomic(page); |
d20f7043 CM |
124 | csum = btrfs_csum_data(root, kaddr, csum, PAGE_CACHE_SIZE); |
125 | btrfs_csum_final(csum, (char *)&csum); | |
7ac687d9 | 126 | kunmap_atomic(kaddr); |
d20f7043 CM |
127 | |
128 | if (csum != *cb_sum) { | |
33345d01 | 129 | printk(KERN_INFO "btrfs csum failed ino %llu " |
d397712b | 130 | "extent %llu csum %u " |
33345d01 LZ |
131 | "wanted %u mirror %d\n", |
132 | (unsigned long long)btrfs_ino(inode), | |
d20f7043 CM |
133 | (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; | |
142 | fail: | |
143 | return ret; | |
144 | } | |
145 | ||
c8b97818 CM |
146 | /* 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 | */ | |
156 | static void end_compressed_bio_read(struct bio *bio, int err) | |
157 | { | |
c8b97818 CM |
158 | 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 | ||
d20f7043 CM |
173 | inode = cb->inode; |
174 | ret = check_compressed_csum(inode, cb, (u64)bio->bi_sector << 9); | |
175 | if (ret) | |
176 | goto csum_failed; | |
177 | ||
c8b97818 CM |
178 | /* ok, we're the last bio for this extent, lets start |
179 | * the decompression. | |
180 | */ | |
261507a0 LZ |
181 | 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); | |
d20f7043 | 187 | csum_failed: |
c8b97818 CM |
188 | 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 */ | |
771ed689 | 200 | if (cb->errors) { |
c8b97818 | 201 | bio_io_error(cb->orig_bio); |
d20f7043 CM |
202 | } 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 | */ | |
d397712b | 210 | while (bio_index < cb->orig_bio->bi_vcnt) { |
d20f7043 CM |
211 | SetPageChecked(bvec->bv_page); |
212 | bvec++; | |
213 | bio_index++; | |
214 | } | |
c8b97818 | 215 | bio_endio(cb->orig_bio, 0); |
d20f7043 | 216 | } |
c8b97818 CM |
217 | |
218 | /* finally free the cb struct */ | |
219 | kfree(cb->compressed_pages); | |
220 | kfree(cb); | |
221 | out: | |
222 | bio_put(bio); | |
223 | } | |
224 | ||
225 | /* | |
226 | * Clear the writeback bits on all of the file | |
227 | * pages for a compressed write | |
228 | */ | |
143bede5 JM |
229 | static noinline void end_compressed_writeback(struct inode *inode, u64 start, |
230 | unsigned long ram_size) | |
c8b97818 CM |
231 | { |
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 | ||
d397712b | 239 | while (nr_pages > 0) { |
c8b97818 | 240 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
241 | min_t(unsigned long, |
242 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
243 | 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 */ | |
c8b97818 CM |
256 | } |
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 | */ | |
266 | static 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; | |
70b99e69 | 288 | cb->compressed_pages[0]->mapping = cb->inode->i_mapping; |
c8b97818 CM |
289 | tree->ops->writepage_end_io_hook(cb->compressed_pages[0], |
290 | cb->start, | |
291 | cb->start + cb->len - 1, | |
292 | NULL, 1); | |
70b99e69 | 293 | cb->compressed_pages[0]->mapping = NULL; |
c8b97818 CM |
294 | |
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); | |
312 | out: | |
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 | */ | |
325 | int 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; | |
306e16ce | 336 | int pg_index = 0; |
c8b97818 CM |
337 | struct page *page; |
338 | u64 first_byte = disk_start; | |
339 | struct block_device *bdev; | |
340 | int ret; | |
e55179b3 | 341 | int skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
c8b97818 CM |
342 | |
343 | WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1)); | |
d20f7043 | 344 | cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); |
dac97e51 YS |
345 | if (!cb) |
346 | return -ENOMEM; | |
c8b97818 CM |
347 | atomic_set(&cb->pending_bios, 0); |
348 | cb->errors = 0; | |
349 | cb->inode = inode; | |
350 | cb->start = start; | |
351 | cb->len = len; | |
d20f7043 | 352 | cb->mirror_num = 0; |
c8b97818 CM |
353 | 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 | ||
c8b97818 | 360 | bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); |
dac97e51 YS |
361 | if(!bio) { |
362 | kfree(cb); | |
363 | return -ENOMEM; | |
364 | } | |
c8b97818 CM |
365 | 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; | |
306e16ce DS |
371 | for (pg_index = 0; pg_index < cb->nr_pages; pg_index++) { |
372 | page = compressed_pages[pg_index]; | |
c8b97818 CM |
373 | 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 | ||
70b99e69 | 381 | page->mapping = NULL; |
c8b97818 CM |
382 | if (ret || bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < |
383 | PAGE_CACHE_SIZE) { | |
384 | bio_get(bio); | |
385 | ||
af09abfe CM |
386 | /* |
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); | |
c8b97818 | 393 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
79787eaa | 394 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 395 | |
e55179b3 LZ |
396 | if (!skip_sum) { |
397 | ret = btrfs_csum_one_bio(root, inode, bio, | |
398 | start, 1); | |
79787eaa | 399 | BUG_ON(ret); /* -ENOMEM */ |
e55179b3 | 400 | } |
d20f7043 | 401 | |
c8b97818 | 402 | ret = btrfs_map_bio(root, WRITE, bio, 0, 1); |
79787eaa | 403 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
404 | |
405 | bio_put(bio); | |
406 | ||
407 | bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); | |
e627ee7b | 408 | BUG_ON(!bio); |
c8b97818 CM |
409 | bio->bi_private = cb; |
410 | bio->bi_end_io = end_compressed_bio_write; | |
411 | bio_add_page(bio, page, PAGE_CACHE_SIZE, 0); | |
412 | } | |
cfbc246e CM |
413 | 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 | } | |
c8b97818 CM |
417 | bytes_left -= PAGE_CACHE_SIZE; |
418 | first_byte += PAGE_CACHE_SIZE; | |
771ed689 | 419 | cond_resched(); |
c8b97818 CM |
420 | } |
421 | bio_get(bio); | |
422 | ||
423 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
79787eaa | 424 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 425 | |
e55179b3 LZ |
426 | if (!skip_sum) { |
427 | ret = btrfs_csum_one_bio(root, inode, bio, start, 1); | |
79787eaa | 428 | BUG_ON(ret); /* -ENOMEM */ |
e55179b3 | 429 | } |
d20f7043 | 430 | |
c8b97818 | 431 | ret = btrfs_map_bio(root, WRITE, bio, 0, 1); |
79787eaa | 432 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
433 | |
434 | bio_put(bio); | |
435 | return 0; | |
436 | } | |
437 | ||
771ed689 CM |
438 | static noinline int add_ra_bio_pages(struct inode *inode, |
439 | u64 compressed_end, | |
440 | struct compressed_bio *cb) | |
441 | { | |
442 | unsigned long end_index; | |
306e16ce | 443 | unsigned long pg_index; |
771ed689 CM |
444 | 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; | |
771ed689 CM |
451 | 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 | ||
d397712b | 466 | while (last_offset < compressed_end) { |
306e16ce | 467 | pg_index = last_offset >> PAGE_CACHE_SHIFT; |
771ed689 | 468 | |
306e16ce | 469 | if (pg_index > end_index) |
771ed689 CM |
470 | break; |
471 | ||
472 | rcu_read_lock(); | |
306e16ce | 473 | page = radix_tree_lookup(&mapping->page_tree, pg_index); |
771ed689 CM |
474 | rcu_read_unlock(); |
475 | if (page) { | |
476 | misses++; | |
477 | if (misses > 4) | |
478 | break; | |
479 | goto next; | |
480 | } | |
481 | ||
28ecb609 NP |
482 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & |
483 | ~__GFP_FS); | |
771ed689 CM |
484 | if (!page) |
485 | break; | |
486 | ||
306e16ce | 487 | if (add_to_page_cache_lru(page, mapping, pg_index, |
28ecb609 | 488 | GFP_NOFS)) { |
771ed689 CM |
489 | page_cache_release(page); |
490 | goto next; | |
491 | } | |
492 | ||
771ed689 CM |
493 | 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); | |
d0082371 | 500 | lock_extent(tree, last_offset, end); |
890871be | 501 | read_lock(&em_tree->lock); |
771ed689 CM |
502 | em = lookup_extent_mapping(em_tree, last_offset, |
503 | PAGE_CACHE_SIZE); | |
890871be | 504 | read_unlock(&em_tree->lock); |
771ed689 CM |
505 | |
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); | |
d0082371 | 510 | unlock_extent(tree, last_offset, end); |
771ed689 CM |
511 | 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; | |
7ac687d9 | 524 | userpage = kmap_atomic(page); |
771ed689 CM |
525 | memset(userpage + zero_offset, 0, zeros); |
526 | flush_dcache_page(page); | |
7ac687d9 | 527 | kunmap_atomic(userpage); |
771ed689 CM |
528 | } |
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 { | |
d0082371 | 538 | unlock_extent(tree, last_offset, end); |
771ed689 CM |
539 | unlock_page(page); |
540 | page_cache_release(page); | |
541 | break; | |
542 | } | |
543 | next: | |
544 | last_offset += PAGE_CACHE_SIZE; | |
545 | } | |
771ed689 CM |
546 | return 0; |
547 | } | |
548 | ||
c8b97818 CM |
549 | /* |
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 | */ | |
561 | int 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; | |
306e16ce | 571 | unsigned long pg_index; |
c8b97818 CM |
572 | struct page *page; |
573 | struct block_device *bdev; | |
574 | struct bio *comp_bio; | |
575 | u64 cur_disk_byte = (u64)bio->bi_sector << 9; | |
e04ca626 CM |
576 | u64 em_len; |
577 | u64 em_start; | |
c8b97818 | 578 | struct extent_map *em; |
6b82ce8d | 579 | int ret = -ENOMEM; |
15e3004a | 580 | int faili = 0; |
d20f7043 | 581 | u32 *sums; |
c8b97818 CM |
582 | |
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 */ | |
890871be | 587 | read_lock(&em_tree->lock); |
c8b97818 CM |
588 | em = lookup_extent_mapping(em_tree, |
589 | page_offset(bio->bi_io_vec->bv_page), | |
590 | PAGE_CACHE_SIZE); | |
890871be | 591 | read_unlock(&em_tree->lock); |
285190d9 TI |
592 | if (!em) |
593 | return -EIO; | |
c8b97818 | 594 | |
d20f7043 CM |
595 | compressed_len = em->block_len; |
596 | cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); | |
6b82ce8d | 597 | if (!cb) |
598 | goto out; | |
599 | ||
c8b97818 CM |
600 | atomic_set(&cb->pending_bios, 0); |
601 | cb->errors = 0; | |
602 | cb->inode = inode; | |
d20f7043 CM |
603 | cb->mirror_num = mirror_num; |
604 | sums = &cb->sums; | |
c8b97818 | 605 | |
ff5b7ee3 | 606 | cb->start = em->orig_start; |
e04ca626 CM |
607 | em_len = em->len; |
608 | em_start = em->start; | |
d20f7043 | 609 | |
c8b97818 | 610 | free_extent_map(em); |
e04ca626 | 611 | em = NULL; |
c8b97818 CM |
612 | |
613 | cb->len = uncompressed_len; | |
614 | cb->compressed_len = compressed_len; | |
261507a0 | 615 | cb->compress_type = extent_compress_type(bio_flags); |
c8b97818 CM |
616 | cb->orig_bio = bio; |
617 | ||
618 | nr_pages = (compressed_len + PAGE_CACHE_SIZE - 1) / | |
619 | PAGE_CACHE_SIZE; | |
6b82ce8d | 620 | cb->compressed_pages = kzalloc(sizeof(struct page *) * nr_pages, |
c8b97818 | 621 | GFP_NOFS); |
6b82ce8d | 622 | if (!cb->compressed_pages) |
623 | goto fail1; | |
624 | ||
c8b97818 CM |
625 | bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; |
626 | ||
306e16ce DS |
627 | for (pg_index = 0; pg_index < nr_pages; pg_index++) { |
628 | cb->compressed_pages[pg_index] = alloc_page(GFP_NOFS | | |
c8b97818 | 629 | __GFP_HIGHMEM); |
15e3004a JB |
630 | if (!cb->compressed_pages[pg_index]) { |
631 | faili = pg_index - 1; | |
632 | ret = -ENOMEM; | |
6b82ce8d | 633 | goto fail2; |
15e3004a | 634 | } |
c8b97818 | 635 | } |
15e3004a | 636 | faili = nr_pages - 1; |
c8b97818 CM |
637 | cb->nr_pages = nr_pages; |
638 | ||
e04ca626 | 639 | add_ra_bio_pages(inode, em_start + em_len, cb); |
771ed689 | 640 | |
771ed689 CM |
641 | /* include any pages we added in add_ra-bio_pages */ |
642 | uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE; | |
643 | cb->len = uncompressed_len; | |
644 | ||
c8b97818 | 645 | comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS); |
6b82ce8d | 646 | if (!comp_bio) |
647 | goto fail2; | |
c8b97818 CM |
648 | comp_bio->bi_private = cb; |
649 | comp_bio->bi_end_io = end_compressed_bio_read; | |
650 | atomic_inc(&cb->pending_bios); | |
651 | ||
306e16ce DS |
652 | for (pg_index = 0; pg_index < nr_pages; pg_index++) { |
653 | page = cb->compressed_pages[pg_index]; | |
c8b97818 | 654 | page->mapping = inode->i_mapping; |
d20f7043 CM |
655 | page->index = em_start >> PAGE_CACHE_SHIFT; |
656 | ||
c8b97818 CM |
657 | if (comp_bio->bi_size) |
658 | ret = tree->ops->merge_bio_hook(page, 0, | |
659 | PAGE_CACHE_SIZE, | |
660 | comp_bio, 0); | |
661 | else | |
662 | ret = 0; | |
663 | ||
70b99e69 | 664 | page->mapping = NULL; |
c8b97818 CM |
665 | if (ret || bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0) < |
666 | PAGE_CACHE_SIZE) { | |
667 | bio_get(comp_bio); | |
668 | ||
669 | ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0); | |
79787eaa | 670 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 671 | |
af09abfe CM |
672 | /* |
673 | * inc the count before we submit the bio so | |
674 | * we know the end IO handler won't happen before | |
675 | * we inc the count. Otherwise, the cb might get | |
676 | * freed before we're done setting it up | |
677 | */ | |
678 | atomic_inc(&cb->pending_bios); | |
679 | ||
6cbff00f | 680 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { |
c2db1073 TI |
681 | ret = btrfs_lookup_bio_sums(root, inode, |
682 | comp_bio, sums); | |
79787eaa | 683 | BUG_ON(ret); /* -ENOMEM */ |
d20f7043 CM |
684 | } |
685 | sums += (comp_bio->bi_size + root->sectorsize - 1) / | |
686 | root->sectorsize; | |
687 | ||
688 | ret = btrfs_map_bio(root, READ, comp_bio, | |
689 | mirror_num, 0); | |
79787eaa | 690 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
691 | |
692 | bio_put(comp_bio); | |
693 | ||
694 | comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, | |
695 | GFP_NOFS); | |
e627ee7b | 696 | BUG_ON(!comp_bio); |
771ed689 CM |
697 | comp_bio->bi_private = cb; |
698 | comp_bio->bi_end_io = end_compressed_bio_read; | |
699 | ||
700 | bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0); | |
c8b97818 CM |
701 | } |
702 | cur_disk_byte += PAGE_CACHE_SIZE; | |
703 | } | |
704 | bio_get(comp_bio); | |
705 | ||
706 | ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, 0); | |
79787eaa | 707 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 708 | |
c2db1073 TI |
709 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { |
710 | ret = btrfs_lookup_bio_sums(root, inode, comp_bio, sums); | |
79787eaa | 711 | BUG_ON(ret); /* -ENOMEM */ |
c2db1073 | 712 | } |
d20f7043 CM |
713 | |
714 | ret = btrfs_map_bio(root, READ, comp_bio, mirror_num, 0); | |
79787eaa | 715 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
716 | |
717 | bio_put(comp_bio); | |
718 | return 0; | |
6b82ce8d | 719 | |
720 | fail2: | |
15e3004a JB |
721 | while (faili >= 0) { |
722 | __free_page(cb->compressed_pages[faili]); | |
723 | faili--; | |
724 | } | |
6b82ce8d | 725 | |
726 | kfree(cb->compressed_pages); | |
727 | fail1: | |
728 | kfree(cb); | |
729 | out: | |
730 | free_extent_map(em); | |
731 | return ret; | |
c8b97818 | 732 | } |
261507a0 LZ |
733 | |
734 | static struct list_head comp_idle_workspace[BTRFS_COMPRESS_TYPES]; | |
735 | static spinlock_t comp_workspace_lock[BTRFS_COMPRESS_TYPES]; | |
736 | static int comp_num_workspace[BTRFS_COMPRESS_TYPES]; | |
737 | static atomic_t comp_alloc_workspace[BTRFS_COMPRESS_TYPES]; | |
738 | static wait_queue_head_t comp_workspace_wait[BTRFS_COMPRESS_TYPES]; | |
739 | ||
740 | struct btrfs_compress_op *btrfs_compress_op[] = { | |
741 | &btrfs_zlib_compress, | |
a6fa6fae | 742 | &btrfs_lzo_compress, |
261507a0 LZ |
743 | }; |
744 | ||
143bede5 | 745 | void __init btrfs_init_compress(void) |
261507a0 LZ |
746 | { |
747 | int i; | |
748 | ||
749 | for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) { | |
750 | INIT_LIST_HEAD(&comp_idle_workspace[i]); | |
751 | spin_lock_init(&comp_workspace_lock[i]); | |
752 | atomic_set(&comp_alloc_workspace[i], 0); | |
753 | init_waitqueue_head(&comp_workspace_wait[i]); | |
754 | } | |
261507a0 LZ |
755 | } |
756 | ||
757 | /* | |
758 | * this finds an available workspace or allocates a new one | |
759 | * ERR_PTR is returned if things go bad. | |
760 | */ | |
761 | static struct list_head *find_workspace(int type) | |
762 | { | |
763 | struct list_head *workspace; | |
764 | int cpus = num_online_cpus(); | |
765 | int idx = type - 1; | |
766 | ||
767 | struct list_head *idle_workspace = &comp_idle_workspace[idx]; | |
768 | spinlock_t *workspace_lock = &comp_workspace_lock[idx]; | |
769 | atomic_t *alloc_workspace = &comp_alloc_workspace[idx]; | |
770 | wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx]; | |
771 | int *num_workspace = &comp_num_workspace[idx]; | |
772 | again: | |
773 | spin_lock(workspace_lock); | |
774 | if (!list_empty(idle_workspace)) { | |
775 | workspace = idle_workspace->next; | |
776 | list_del(workspace); | |
777 | (*num_workspace)--; | |
778 | spin_unlock(workspace_lock); | |
779 | return workspace; | |
780 | ||
781 | } | |
782 | if (atomic_read(alloc_workspace) > cpus) { | |
783 | DEFINE_WAIT(wait); | |
784 | ||
785 | spin_unlock(workspace_lock); | |
786 | prepare_to_wait(workspace_wait, &wait, TASK_UNINTERRUPTIBLE); | |
787 | if (atomic_read(alloc_workspace) > cpus && !*num_workspace) | |
788 | schedule(); | |
789 | finish_wait(workspace_wait, &wait); | |
790 | goto again; | |
791 | } | |
792 | atomic_inc(alloc_workspace); | |
793 | spin_unlock(workspace_lock); | |
794 | ||
795 | workspace = btrfs_compress_op[idx]->alloc_workspace(); | |
796 | if (IS_ERR(workspace)) { | |
797 | atomic_dec(alloc_workspace); | |
798 | wake_up(workspace_wait); | |
799 | } | |
800 | return workspace; | |
801 | } | |
802 | ||
803 | /* | |
804 | * put a workspace struct back on the list or free it if we have enough | |
805 | * idle ones sitting around | |
806 | */ | |
807 | static void free_workspace(int type, struct list_head *workspace) | |
808 | { | |
809 | int idx = type - 1; | |
810 | struct list_head *idle_workspace = &comp_idle_workspace[idx]; | |
811 | spinlock_t *workspace_lock = &comp_workspace_lock[idx]; | |
812 | atomic_t *alloc_workspace = &comp_alloc_workspace[idx]; | |
813 | wait_queue_head_t *workspace_wait = &comp_workspace_wait[idx]; | |
814 | int *num_workspace = &comp_num_workspace[idx]; | |
815 | ||
816 | spin_lock(workspace_lock); | |
817 | if (*num_workspace < num_online_cpus()) { | |
818 | list_add_tail(workspace, idle_workspace); | |
819 | (*num_workspace)++; | |
820 | spin_unlock(workspace_lock); | |
821 | goto wake; | |
822 | } | |
823 | spin_unlock(workspace_lock); | |
824 | ||
825 | btrfs_compress_op[idx]->free_workspace(workspace); | |
826 | atomic_dec(alloc_workspace); | |
827 | wake: | |
66657b31 | 828 | smp_mb(); |
261507a0 LZ |
829 | if (waitqueue_active(workspace_wait)) |
830 | wake_up(workspace_wait); | |
831 | } | |
832 | ||
833 | /* | |
834 | * cleanup function for module exit | |
835 | */ | |
836 | static void free_workspaces(void) | |
837 | { | |
838 | struct list_head *workspace; | |
839 | int i; | |
840 | ||
841 | for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) { | |
842 | while (!list_empty(&comp_idle_workspace[i])) { | |
843 | workspace = comp_idle_workspace[i].next; | |
844 | list_del(workspace); | |
845 | btrfs_compress_op[i]->free_workspace(workspace); | |
846 | atomic_dec(&comp_alloc_workspace[i]); | |
847 | } | |
848 | } | |
849 | } | |
850 | ||
851 | /* | |
852 | * given an address space and start/len, compress the bytes. | |
853 | * | |
854 | * pages are allocated to hold the compressed result and stored | |
855 | * in 'pages' | |
856 | * | |
857 | * out_pages is used to return the number of pages allocated. There | |
858 | * may be pages allocated even if we return an error | |
859 | * | |
860 | * total_in is used to return the number of bytes actually read. It | |
861 | * may be smaller then len if we had to exit early because we | |
862 | * ran out of room in the pages array or because we cross the | |
863 | * max_out threshold. | |
864 | * | |
865 | * total_out is used to return the total number of compressed bytes | |
866 | * | |
867 | * max_out tells us the max number of bytes that we're allowed to | |
868 | * stuff into pages | |
869 | */ | |
870 | int btrfs_compress_pages(int type, struct address_space *mapping, | |
871 | u64 start, unsigned long len, | |
872 | struct page **pages, | |
873 | unsigned long nr_dest_pages, | |
874 | unsigned long *out_pages, | |
875 | unsigned long *total_in, | |
876 | unsigned long *total_out, | |
877 | unsigned long max_out) | |
878 | { | |
879 | struct list_head *workspace; | |
880 | int ret; | |
881 | ||
882 | workspace = find_workspace(type); | |
883 | if (IS_ERR(workspace)) | |
884 | return -1; | |
885 | ||
886 | ret = btrfs_compress_op[type-1]->compress_pages(workspace, mapping, | |
887 | start, len, pages, | |
888 | nr_dest_pages, out_pages, | |
889 | total_in, total_out, | |
890 | max_out); | |
891 | free_workspace(type, workspace); | |
892 | return ret; | |
893 | } | |
894 | ||
895 | /* | |
896 | * pages_in is an array of pages with compressed data. | |
897 | * | |
898 | * disk_start is the starting logical offset of this array in the file | |
899 | * | |
900 | * bvec is a bio_vec of pages from the file that we want to decompress into | |
901 | * | |
902 | * vcnt is the count of pages in the biovec | |
903 | * | |
904 | * srclen is the number of bytes in pages_in | |
905 | * | |
906 | * The basic idea is that we have a bio that was created by readpages. | |
907 | * The pages in the bio are for the uncompressed data, and they may not | |
908 | * be contiguous. They all correspond to the range of bytes covered by | |
909 | * the compressed extent. | |
910 | */ | |
911 | int btrfs_decompress_biovec(int type, struct page **pages_in, u64 disk_start, | |
912 | struct bio_vec *bvec, int vcnt, size_t srclen) | |
913 | { | |
914 | struct list_head *workspace; | |
915 | int ret; | |
916 | ||
917 | workspace = find_workspace(type); | |
918 | if (IS_ERR(workspace)) | |
919 | return -ENOMEM; | |
920 | ||
921 | ret = btrfs_compress_op[type-1]->decompress_biovec(workspace, pages_in, | |
922 | disk_start, | |
923 | bvec, vcnt, srclen); | |
924 | free_workspace(type, workspace); | |
925 | return ret; | |
926 | } | |
927 | ||
928 | /* | |
929 | * a less complex decompression routine. Our compressed data fits in a | |
930 | * single page, and we want to read a single page out of it. | |
931 | * start_byte tells us the offset into the compressed data we're interested in | |
932 | */ | |
933 | int btrfs_decompress(int type, unsigned char *data_in, struct page *dest_page, | |
934 | unsigned long start_byte, size_t srclen, size_t destlen) | |
935 | { | |
936 | struct list_head *workspace; | |
937 | int ret; | |
938 | ||
939 | workspace = find_workspace(type); | |
940 | if (IS_ERR(workspace)) | |
941 | return -ENOMEM; | |
942 | ||
943 | ret = btrfs_compress_op[type-1]->decompress(workspace, data_in, | |
944 | dest_page, start_byte, | |
945 | srclen, destlen); | |
946 | ||
947 | free_workspace(type, workspace); | |
948 | return ret; | |
949 | } | |
950 | ||
8e4eef7a | 951 | void btrfs_exit_compress(void) |
261507a0 LZ |
952 | { |
953 | free_workspaces(); | |
954 | } | |
3a39c18d LZ |
955 | |
956 | /* | |
957 | * Copy uncompressed data from working buffer to pages. | |
958 | * | |
959 | * buf_start is the byte offset we're of the start of our workspace buffer. | |
960 | * | |
961 | * total_out is the last byte of the buffer | |
962 | */ | |
963 | int btrfs_decompress_buf2page(char *buf, unsigned long buf_start, | |
964 | unsigned long total_out, u64 disk_start, | |
965 | struct bio_vec *bvec, int vcnt, | |
306e16ce | 966 | unsigned long *pg_index, |
3a39c18d LZ |
967 | unsigned long *pg_offset) |
968 | { | |
969 | unsigned long buf_offset; | |
970 | unsigned long current_buf_start; | |
971 | unsigned long start_byte; | |
972 | unsigned long working_bytes = total_out - buf_start; | |
973 | unsigned long bytes; | |
974 | char *kaddr; | |
306e16ce | 975 | struct page *page_out = bvec[*pg_index].bv_page; |
3a39c18d LZ |
976 | |
977 | /* | |
978 | * start byte is the first byte of the page we're currently | |
979 | * copying into relative to the start of the compressed data. | |
980 | */ | |
981 | start_byte = page_offset(page_out) - disk_start; | |
982 | ||
983 | /* we haven't yet hit data corresponding to this page */ | |
984 | if (total_out <= start_byte) | |
985 | return 1; | |
986 | ||
987 | /* | |
988 | * the start of the data we care about is offset into | |
989 | * the middle of our working buffer | |
990 | */ | |
991 | if (total_out > start_byte && buf_start < start_byte) { | |
992 | buf_offset = start_byte - buf_start; | |
993 | working_bytes -= buf_offset; | |
994 | } else { | |
995 | buf_offset = 0; | |
996 | } | |
997 | current_buf_start = buf_start; | |
998 | ||
999 | /* copy bytes from the working buffer into the pages */ | |
1000 | while (working_bytes > 0) { | |
1001 | bytes = min(PAGE_CACHE_SIZE - *pg_offset, | |
1002 | PAGE_CACHE_SIZE - buf_offset); | |
1003 | bytes = min(bytes, working_bytes); | |
7ac687d9 | 1004 | kaddr = kmap_atomic(page_out); |
3a39c18d | 1005 | memcpy(kaddr + *pg_offset, buf + buf_offset, bytes); |
7ac687d9 | 1006 | kunmap_atomic(kaddr); |
3a39c18d LZ |
1007 | flush_dcache_page(page_out); |
1008 | ||
1009 | *pg_offset += bytes; | |
1010 | buf_offset += bytes; | |
1011 | working_bytes -= bytes; | |
1012 | current_buf_start += bytes; | |
1013 | ||
1014 | /* check if we need to pick another page */ | |
1015 | if (*pg_offset == PAGE_CACHE_SIZE) { | |
306e16ce DS |
1016 | (*pg_index)++; |
1017 | if (*pg_index >= vcnt) | |
3a39c18d LZ |
1018 | return 0; |
1019 | ||
306e16ce | 1020 | page_out = bvec[*pg_index].bv_page; |
3a39c18d LZ |
1021 | *pg_offset = 0; |
1022 | start_byte = page_offset(page_out) - disk_start; | |
1023 | ||
1024 | /* | |
1025 | * make sure our new page is covered by this | |
1026 | * working buffer | |
1027 | */ | |
1028 | if (total_out <= start_byte) | |
1029 | return 1; | |
1030 | ||
1031 | /* | |
1032 | * the next page in the biovec might not be adjacent | |
1033 | * to the last page, but it might still be found | |
1034 | * inside this working buffer. bump our offset pointer | |
1035 | */ | |
1036 | if (total_out > start_byte && | |
1037 | current_buf_start < start_byte) { | |
1038 | buf_offset = start_byte - buf_start; | |
1039 | working_bytes = total_out - start_byte; | |
1040 | current_buf_start = buf_start + buf_offset; | |
1041 | } | |
1042 | } | |
1043 | } | |
1044 | ||
1045 | return 1; | |
1046 | } |