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> |
c8b97818 CM |
35 | #include "ctree.h" |
36 | #include "disk-io.h" | |
37 | #include "transaction.h" | |
38 | #include "btrfs_inode.h" | |
39 | #include "volumes.h" | |
40 | #include "ordered-data.h" | |
c8b97818 CM |
41 | #include "compression.h" |
42 | #include "extent_io.h" | |
43 | #include "extent_map.h" | |
44 | ||
45 | struct compressed_bio { | |
46 | /* number of bios pending for this compressed extent */ | |
47 | atomic_t pending_bios; | |
48 | ||
49 | /* the pages with the compressed data on them */ | |
50 | struct page **compressed_pages; | |
51 | ||
52 | /* inode that owns this data */ | |
53 | struct inode *inode; | |
54 | ||
55 | /* starting offset in the inode for our pages */ | |
56 | u64 start; | |
57 | ||
58 | /* number of bytes in the inode we're working on */ | |
59 | unsigned long len; | |
60 | ||
61 | /* number of bytes on disk */ | |
62 | unsigned long compressed_len; | |
63 | ||
261507a0 LZ |
64 | /* the compression algorithm for this bio */ |
65 | int compress_type; | |
66 | ||
c8b97818 CM |
67 | /* number of compressed pages in the array */ |
68 | unsigned long nr_pages; | |
69 | ||
70 | /* IO errors */ | |
71 | int errors; | |
d20f7043 | 72 | int mirror_num; |
c8b97818 CM |
73 | |
74 | /* for reads, this is the bio we are copying the data into */ | |
75 | struct bio *orig_bio; | |
d20f7043 CM |
76 | |
77 | /* | |
78 | * the start of a variable length array of checksums only | |
79 | * used by reads | |
80 | */ | |
81 | u32 sums; | |
c8b97818 CM |
82 | }; |
83 | ||
48a3b636 ES |
84 | static int btrfs_decompress_biovec(int type, struct page **pages_in, |
85 | u64 disk_start, struct bio_vec *bvec, | |
86 | int vcnt, size_t srclen); | |
87 | ||
d20f7043 CM |
88 | static inline int compressed_bio_size(struct btrfs_root *root, |
89 | unsigned long disk_size) | |
90 | { | |
6c41761f DS |
91 | u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy); |
92 | ||
d20f7043 CM |
93 | return sizeof(struct compressed_bio) + |
94 | ((disk_size + root->sectorsize - 1) / root->sectorsize) * | |
95 | csum_size; | |
96 | } | |
97 | ||
c8b97818 CM |
98 | static struct bio *compressed_bio_alloc(struct block_device *bdev, |
99 | u64 first_byte, gfp_t gfp_flags) | |
100 | { | |
c8b97818 CM |
101 | int nr_vecs; |
102 | ||
103 | nr_vecs = bio_get_nr_vecs(bdev); | |
88f794ed | 104 | return btrfs_bio_alloc(bdev, first_byte >> 9, nr_vecs, gfp_flags); |
c8b97818 CM |
105 | } |
106 | ||
d20f7043 CM |
107 | static int check_compressed_csum(struct inode *inode, |
108 | struct compressed_bio *cb, | |
109 | u64 disk_start) | |
110 | { | |
111 | int ret; | |
d20f7043 CM |
112 | struct page *page; |
113 | unsigned long i; | |
114 | char *kaddr; | |
115 | u32 csum; | |
116 | u32 *cb_sum = &cb->sums; | |
117 | ||
6cbff00f | 118 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) |
d20f7043 CM |
119 | return 0; |
120 | ||
121 | for (i = 0; i < cb->nr_pages; i++) { | |
122 | page = cb->compressed_pages[i]; | |
123 | csum = ~(u32)0; | |
124 | ||
7ac687d9 | 125 | kaddr = kmap_atomic(page); |
b0496686 | 126 | csum = btrfs_csum_data(kaddr, csum, PAGE_CACHE_SIZE); |
d20f7043 | 127 | btrfs_csum_final(csum, (char *)&csum); |
7ac687d9 | 128 | kunmap_atomic(kaddr); |
d20f7043 CM |
129 | |
130 | if (csum != *cb_sum) { | |
33345d01 | 131 | printk(KERN_INFO "btrfs csum failed ino %llu " |
d397712b | 132 | "extent %llu csum %u " |
33345d01 | 133 | "wanted %u mirror %d\n", |
c1c9ff7c GU |
134 | btrfs_ino(inode), disk_start, csum, *cb_sum, |
135 | cb->mirror_num); | |
d20f7043 CM |
136 | ret = -EIO; |
137 | goto fail; | |
138 | } | |
139 | cb_sum++; | |
140 | ||
141 | } | |
142 | ret = 0; | |
143 | fail: | |
144 | return ret; | |
145 | } | |
146 | ||
c8b97818 CM |
147 | /* when we finish reading compressed pages from the disk, we |
148 | * decompress them and then run the bio end_io routines on the | |
149 | * decompressed pages (in the inode address space). | |
150 | * | |
151 | * This allows the checksumming and other IO error handling routines | |
152 | * to work normally | |
153 | * | |
154 | * The compressed pages are freed here, and it must be run | |
155 | * in process context | |
156 | */ | |
157 | static void end_compressed_bio_read(struct bio *bio, int err) | |
158 | { | |
c8b97818 CM |
159 | struct compressed_bio *cb = bio->bi_private; |
160 | struct inode *inode; | |
161 | struct page *page; | |
162 | unsigned long index; | |
163 | int ret; | |
164 | ||
165 | if (err) | |
166 | cb->errors = 1; | |
167 | ||
168 | /* if there are more bios still pending for this compressed | |
169 | * extent, just exit | |
170 | */ | |
171 | if (!atomic_dec_and_test(&cb->pending_bios)) | |
172 | goto out; | |
173 | ||
d20f7043 CM |
174 | inode = cb->inode; |
175 | ret = check_compressed_csum(inode, cb, (u64)bio->bi_sector << 9); | |
176 | if (ret) | |
177 | goto csum_failed; | |
178 | ||
c8b97818 CM |
179 | /* ok, we're the last bio for this extent, lets start |
180 | * the decompression. | |
181 | */ | |
261507a0 LZ |
182 | ret = btrfs_decompress_biovec(cb->compress_type, |
183 | cb->compressed_pages, | |
184 | cb->start, | |
185 | cb->orig_bio->bi_io_vec, | |
186 | cb->orig_bio->bi_vcnt, | |
187 | cb->compressed_len); | |
d20f7043 | 188 | csum_failed: |
c8b97818 CM |
189 | if (ret) |
190 | cb->errors = 1; | |
191 | ||
192 | /* release the compressed pages */ | |
193 | index = 0; | |
194 | for (index = 0; index < cb->nr_pages; index++) { | |
195 | page = cb->compressed_pages[index]; | |
196 | page->mapping = NULL; | |
197 | page_cache_release(page); | |
198 | } | |
199 | ||
200 | /* do io completion on the original bio */ | |
771ed689 | 201 | if (cb->errors) { |
c8b97818 | 202 | bio_io_error(cb->orig_bio); |
d20f7043 | 203 | } else { |
2c30c71b KO |
204 | int i; |
205 | struct bio_vec *bvec; | |
d20f7043 CM |
206 | |
207 | /* | |
208 | * we have verified the checksum already, set page | |
209 | * checked so the end_io handlers know about it | |
210 | */ | |
2c30c71b | 211 | bio_for_each_segment_all(bvec, cb->orig_bio, i) |
d20f7043 | 212 | SetPageChecked(bvec->bv_page); |
2c30c71b | 213 | |
c8b97818 | 214 | bio_endio(cb->orig_bio, 0); |
d20f7043 | 215 | } |
c8b97818 CM |
216 | |
217 | /* finally free the cb struct */ | |
218 | kfree(cb->compressed_pages); | |
219 | kfree(cb); | |
220 | out: | |
221 | bio_put(bio); | |
222 | } | |
223 | ||
224 | /* | |
225 | * Clear the writeback bits on all of the file | |
226 | * pages for a compressed write | |
227 | */ | |
143bede5 JM |
228 | static noinline void end_compressed_writeback(struct inode *inode, u64 start, |
229 | unsigned long ram_size) | |
c8b97818 CM |
230 | { |
231 | unsigned long index = start >> PAGE_CACHE_SHIFT; | |
232 | unsigned long end_index = (start + ram_size - 1) >> PAGE_CACHE_SHIFT; | |
233 | struct page *pages[16]; | |
234 | unsigned long nr_pages = end_index - index + 1; | |
235 | int i; | |
236 | int ret; | |
237 | ||
d397712b | 238 | while (nr_pages > 0) { |
c8b97818 | 239 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
240 | min_t(unsigned long, |
241 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
242 | if (ret == 0) { |
243 | nr_pages -= 1; | |
244 | index += 1; | |
245 | continue; | |
246 | } | |
247 | for (i = 0; i < ret; i++) { | |
248 | end_page_writeback(pages[i]); | |
249 | page_cache_release(pages[i]); | |
250 | } | |
251 | nr_pages -= ret; | |
252 | index += ret; | |
253 | } | |
254 | /* the inode may be gone now */ | |
c8b97818 CM |
255 | } |
256 | ||
257 | /* | |
258 | * do the cleanup once all the compressed pages hit the disk. | |
259 | * This will clear writeback on the file pages and free the compressed | |
260 | * pages. | |
261 | * | |
262 | * This also calls the writeback end hooks for the file pages so that | |
263 | * metadata and checksums can be updated in the file. | |
264 | */ | |
265 | static void end_compressed_bio_write(struct bio *bio, int err) | |
266 | { | |
267 | struct extent_io_tree *tree; | |
268 | struct compressed_bio *cb = bio->bi_private; | |
269 | struct inode *inode; | |
270 | struct page *page; | |
271 | unsigned long index; | |
272 | ||
273 | if (err) | |
274 | cb->errors = 1; | |
275 | ||
276 | /* if there are more bios still pending for this compressed | |
277 | * extent, just exit | |
278 | */ | |
279 | if (!atomic_dec_and_test(&cb->pending_bios)) | |
280 | goto out; | |
281 | ||
282 | /* ok, we're the last bio for this extent, step one is to | |
283 | * call back into the FS and do all the end_io operations | |
284 | */ | |
285 | inode = cb->inode; | |
286 | tree = &BTRFS_I(inode)->io_tree; | |
70b99e69 | 287 | cb->compressed_pages[0]->mapping = cb->inode->i_mapping; |
c8b97818 CM |
288 | tree->ops->writepage_end_io_hook(cb->compressed_pages[0], |
289 | cb->start, | |
290 | cb->start + cb->len - 1, | |
291 | NULL, 1); | |
70b99e69 | 292 | cb->compressed_pages[0]->mapping = NULL; |
c8b97818 CM |
293 | |
294 | end_compressed_writeback(inode, cb->start, cb->len); | |
295 | /* note, our inode could be gone now */ | |
296 | ||
297 | /* | |
298 | * release the compressed pages, these came from alloc_page and | |
299 | * are not attached to the inode at all | |
300 | */ | |
301 | index = 0; | |
302 | for (index = 0; index < cb->nr_pages; index++) { | |
303 | page = cb->compressed_pages[index]; | |
304 | page->mapping = NULL; | |
305 | page_cache_release(page); | |
306 | } | |
307 | ||
308 | /* finally free the cb struct */ | |
309 | kfree(cb->compressed_pages); | |
310 | kfree(cb); | |
311 | out: | |
312 | bio_put(bio); | |
313 | } | |
314 | ||
315 | /* | |
316 | * worker function to build and submit bios for previously compressed pages. | |
317 | * The corresponding pages in the inode should be marked for writeback | |
318 | * and the compressed pages should have a reference on them for dropping | |
319 | * when the IO is complete. | |
320 | * | |
321 | * This also checksums the file bytes and gets things ready for | |
322 | * the end io hooks. | |
323 | */ | |
324 | int btrfs_submit_compressed_write(struct inode *inode, u64 start, | |
325 | unsigned long len, u64 disk_start, | |
326 | unsigned long compressed_len, | |
327 | struct page **compressed_pages, | |
328 | unsigned long nr_pages) | |
329 | { | |
330 | struct bio *bio = NULL; | |
331 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
332 | struct compressed_bio *cb; | |
333 | unsigned long bytes_left; | |
334 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
306e16ce | 335 | int pg_index = 0; |
c8b97818 CM |
336 | struct page *page; |
337 | u64 first_byte = disk_start; | |
338 | struct block_device *bdev; | |
339 | int ret; | |
e55179b3 | 340 | int skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
c8b97818 CM |
341 | |
342 | WARN_ON(start & ((u64)PAGE_CACHE_SIZE - 1)); | |
d20f7043 | 343 | cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); |
dac97e51 YS |
344 | if (!cb) |
345 | return -ENOMEM; | |
c8b97818 CM |
346 | atomic_set(&cb->pending_bios, 0); |
347 | cb->errors = 0; | |
348 | cb->inode = inode; | |
349 | cb->start = start; | |
350 | cb->len = len; | |
d20f7043 | 351 | cb->mirror_num = 0; |
c8b97818 CM |
352 | cb->compressed_pages = compressed_pages; |
353 | cb->compressed_len = compressed_len; | |
354 | cb->orig_bio = NULL; | |
355 | cb->nr_pages = nr_pages; | |
356 | ||
357 | bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; | |
358 | ||
c8b97818 | 359 | bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); |
67871254 | 360 | if (!bio) { |
dac97e51 YS |
361 | kfree(cb); |
362 | return -ENOMEM; | |
363 | } | |
c8b97818 CM |
364 | bio->bi_private = cb; |
365 | bio->bi_end_io = end_compressed_bio_write; | |
366 | atomic_inc(&cb->pending_bios); | |
367 | ||
368 | /* create and submit bios for the compressed pages */ | |
369 | bytes_left = compressed_len; | |
306e16ce DS |
370 | for (pg_index = 0; pg_index < cb->nr_pages; pg_index++) { |
371 | page = compressed_pages[pg_index]; | |
c8b97818 CM |
372 | page->mapping = inode->i_mapping; |
373 | if (bio->bi_size) | |
64a16701 | 374 | ret = io_tree->ops->merge_bio_hook(WRITE, page, 0, |
c8b97818 CM |
375 | PAGE_CACHE_SIZE, |
376 | bio, 0); | |
377 | else | |
378 | ret = 0; | |
379 | ||
70b99e69 | 380 | page->mapping = NULL; |
c8b97818 CM |
381 | if (ret || bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < |
382 | PAGE_CACHE_SIZE) { | |
383 | bio_get(bio); | |
384 | ||
af09abfe CM |
385 | /* |
386 | * inc the count before we submit the bio so | |
387 | * we know the end IO handler won't happen before | |
388 | * we inc the count. Otherwise, the cb might get | |
389 | * freed before we're done setting it up | |
390 | */ | |
391 | atomic_inc(&cb->pending_bios); | |
c8b97818 | 392 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); |
79787eaa | 393 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 394 | |
e55179b3 LZ |
395 | if (!skip_sum) { |
396 | ret = btrfs_csum_one_bio(root, inode, bio, | |
397 | start, 1); | |
79787eaa | 398 | BUG_ON(ret); /* -ENOMEM */ |
e55179b3 | 399 | } |
d20f7043 | 400 | |
c8b97818 | 401 | ret = btrfs_map_bio(root, WRITE, bio, 0, 1); |
79787eaa | 402 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
403 | |
404 | bio_put(bio); | |
405 | ||
406 | bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); | |
e627ee7b | 407 | BUG_ON(!bio); |
c8b97818 CM |
408 | bio->bi_private = cb; |
409 | bio->bi_end_io = end_compressed_bio_write; | |
410 | bio_add_page(bio, page, PAGE_CACHE_SIZE, 0); | |
411 | } | |
cfbc246e CM |
412 | if (bytes_left < PAGE_CACHE_SIZE) { |
413 | printk("bytes left %lu compress len %lu nr %lu\n", | |
414 | bytes_left, cb->compressed_len, cb->nr_pages); | |
415 | } | |
c8b97818 CM |
416 | bytes_left -= PAGE_CACHE_SIZE; |
417 | first_byte += PAGE_CACHE_SIZE; | |
771ed689 | 418 | cond_resched(); |
c8b97818 CM |
419 | } |
420 | bio_get(bio); | |
421 | ||
422 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0); | |
79787eaa | 423 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 424 | |
e55179b3 LZ |
425 | if (!skip_sum) { |
426 | ret = btrfs_csum_one_bio(root, inode, bio, start, 1); | |
79787eaa | 427 | BUG_ON(ret); /* -ENOMEM */ |
e55179b3 | 428 | } |
d20f7043 | 429 | |
c8b97818 | 430 | ret = btrfs_map_bio(root, WRITE, bio, 0, 1); |
79787eaa | 431 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
432 | |
433 | bio_put(bio); | |
434 | return 0; | |
435 | } | |
436 | ||
771ed689 CM |
437 | static noinline int add_ra_bio_pages(struct inode *inode, |
438 | u64 compressed_end, | |
439 | struct compressed_bio *cb) | |
440 | { | |
441 | unsigned long end_index; | |
306e16ce | 442 | unsigned long pg_index; |
771ed689 CM |
443 | u64 last_offset; |
444 | u64 isize = i_size_read(inode); | |
445 | int ret; | |
446 | struct page *page; | |
447 | unsigned long nr_pages = 0; | |
448 | struct extent_map *em; | |
449 | struct address_space *mapping = inode->i_mapping; | |
771ed689 CM |
450 | struct extent_map_tree *em_tree; |
451 | struct extent_io_tree *tree; | |
452 | u64 end; | |
453 | int misses = 0; | |
454 | ||
455 | page = cb->orig_bio->bi_io_vec[cb->orig_bio->bi_vcnt - 1].bv_page; | |
456 | last_offset = (page_offset(page) + PAGE_CACHE_SIZE); | |
457 | em_tree = &BTRFS_I(inode)->extent_tree; | |
458 | tree = &BTRFS_I(inode)->io_tree; | |
459 | ||
460 | if (isize == 0) | |
461 | return 0; | |
462 | ||
463 | end_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT; | |
464 | ||
d397712b | 465 | while (last_offset < compressed_end) { |
306e16ce | 466 | pg_index = last_offset >> PAGE_CACHE_SHIFT; |
771ed689 | 467 | |
306e16ce | 468 | if (pg_index > end_index) |
771ed689 CM |
469 | break; |
470 | ||
471 | rcu_read_lock(); | |
306e16ce | 472 | page = radix_tree_lookup(&mapping->page_tree, pg_index); |
771ed689 CM |
473 | rcu_read_unlock(); |
474 | if (page) { | |
475 | misses++; | |
476 | if (misses > 4) | |
477 | break; | |
478 | goto next; | |
479 | } | |
480 | ||
28ecb609 NP |
481 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & |
482 | ~__GFP_FS); | |
771ed689 CM |
483 | if (!page) |
484 | break; | |
485 | ||
306e16ce | 486 | if (add_to_page_cache_lru(page, mapping, pg_index, |
28ecb609 | 487 | GFP_NOFS)) { |
771ed689 CM |
488 | page_cache_release(page); |
489 | goto next; | |
490 | } | |
491 | ||
771ed689 CM |
492 | end = last_offset + PAGE_CACHE_SIZE - 1; |
493 | /* | |
494 | * at this point, we have a locked page in the page cache | |
495 | * for these bytes in the file. But, we have to make | |
496 | * sure they map to this compressed extent on disk. | |
497 | */ | |
498 | set_page_extent_mapped(page); | |
d0082371 | 499 | lock_extent(tree, last_offset, end); |
890871be | 500 | read_lock(&em_tree->lock); |
771ed689 CM |
501 | em = lookup_extent_mapping(em_tree, last_offset, |
502 | PAGE_CACHE_SIZE); | |
890871be | 503 | read_unlock(&em_tree->lock); |
771ed689 CM |
504 | |
505 | if (!em || last_offset < em->start || | |
506 | (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) || | |
507 | (em->block_start >> 9) != cb->orig_bio->bi_sector) { | |
508 | free_extent_map(em); | |
d0082371 | 509 | unlock_extent(tree, last_offset, end); |
771ed689 CM |
510 | unlock_page(page); |
511 | page_cache_release(page); | |
512 | break; | |
513 | } | |
514 | free_extent_map(em); | |
515 | ||
516 | if (page->index == end_index) { | |
517 | char *userpage; | |
518 | size_t zero_offset = isize & (PAGE_CACHE_SIZE - 1); | |
519 | ||
520 | if (zero_offset) { | |
521 | int zeros; | |
522 | zeros = PAGE_CACHE_SIZE - zero_offset; | |
7ac687d9 | 523 | userpage = kmap_atomic(page); |
771ed689 CM |
524 | memset(userpage + zero_offset, 0, zeros); |
525 | flush_dcache_page(page); | |
7ac687d9 | 526 | kunmap_atomic(userpage); |
771ed689 CM |
527 | } |
528 | } | |
529 | ||
530 | ret = bio_add_page(cb->orig_bio, page, | |
531 | PAGE_CACHE_SIZE, 0); | |
532 | ||
533 | if (ret == PAGE_CACHE_SIZE) { | |
534 | nr_pages++; | |
535 | page_cache_release(page); | |
536 | } else { | |
d0082371 | 537 | unlock_extent(tree, last_offset, end); |
771ed689 CM |
538 | unlock_page(page); |
539 | page_cache_release(page); | |
540 | break; | |
541 | } | |
542 | next: | |
543 | last_offset += PAGE_CACHE_SIZE; | |
544 | } | |
771ed689 CM |
545 | return 0; |
546 | } | |
547 | ||
c8b97818 CM |
548 | /* |
549 | * for a compressed read, the bio we get passed has all the inode pages | |
550 | * in it. We don't actually do IO on those pages but allocate new ones | |
551 | * to hold the compressed pages on disk. | |
552 | * | |
553 | * bio->bi_sector points to the compressed extent on disk | |
554 | * bio->bi_io_vec points to all of the inode pages | |
555 | * bio->bi_vcnt is a count of pages | |
556 | * | |
557 | * After the compressed pages are read, we copy the bytes into the | |
558 | * bio we were passed and then call the bio end_io calls | |
559 | */ | |
560 | int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio, | |
561 | int mirror_num, unsigned long bio_flags) | |
562 | { | |
563 | struct extent_io_tree *tree; | |
564 | struct extent_map_tree *em_tree; | |
565 | struct compressed_bio *cb; | |
566 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
567 | unsigned long uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE; | |
568 | unsigned long compressed_len; | |
569 | unsigned long nr_pages; | |
306e16ce | 570 | unsigned long pg_index; |
c8b97818 CM |
571 | struct page *page; |
572 | struct block_device *bdev; | |
573 | struct bio *comp_bio; | |
574 | u64 cur_disk_byte = (u64)bio->bi_sector << 9; | |
e04ca626 CM |
575 | u64 em_len; |
576 | u64 em_start; | |
c8b97818 | 577 | struct extent_map *em; |
6b82ce8d | 578 | int ret = -ENOMEM; |
15e3004a | 579 | int faili = 0; |
d20f7043 | 580 | u32 *sums; |
c8b97818 CM |
581 | |
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 */ | |
890871be | 586 | read_lock(&em_tree->lock); |
c8b97818 CM |
587 | em = lookup_extent_mapping(em_tree, |
588 | page_offset(bio->bi_io_vec->bv_page), | |
589 | PAGE_CACHE_SIZE); | |
890871be | 590 | read_unlock(&em_tree->lock); |
285190d9 TI |
591 | if (!em) |
592 | return -EIO; | |
c8b97818 | 593 | |
d20f7043 CM |
594 | compressed_len = em->block_len; |
595 | cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); | |
6b82ce8d | 596 | if (!cb) |
597 | goto out; | |
598 | ||
c8b97818 CM |
599 | atomic_set(&cb->pending_bios, 0); |
600 | cb->errors = 0; | |
601 | cb->inode = inode; | |
d20f7043 CM |
602 | cb->mirror_num = mirror_num; |
603 | sums = &cb->sums; | |
c8b97818 | 604 | |
ff5b7ee3 | 605 | cb->start = em->orig_start; |
e04ca626 CM |
606 | em_len = em->len; |
607 | em_start = em->start; | |
d20f7043 | 608 | |
c8b97818 | 609 | free_extent_map(em); |
e04ca626 | 610 | em = NULL; |
c8b97818 CM |
611 | |
612 | cb->len = uncompressed_len; | |
613 | cb->compressed_len = compressed_len; | |
261507a0 | 614 | cb->compress_type = extent_compress_type(bio_flags); |
c8b97818 CM |
615 | cb->orig_bio = bio; |
616 | ||
617 | nr_pages = (compressed_len + PAGE_CACHE_SIZE - 1) / | |
618 | PAGE_CACHE_SIZE; | |
6b82ce8d | 619 | cb->compressed_pages = kzalloc(sizeof(struct page *) * nr_pages, |
c8b97818 | 620 | GFP_NOFS); |
6b82ce8d | 621 | if (!cb->compressed_pages) |
622 | goto fail1; | |
623 | ||
c8b97818 CM |
624 | bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; |
625 | ||
306e16ce DS |
626 | for (pg_index = 0; pg_index < nr_pages; pg_index++) { |
627 | cb->compressed_pages[pg_index] = alloc_page(GFP_NOFS | | |
c8b97818 | 628 | __GFP_HIGHMEM); |
15e3004a JB |
629 | if (!cb->compressed_pages[pg_index]) { |
630 | faili = pg_index - 1; | |
631 | ret = -ENOMEM; | |
6b82ce8d | 632 | goto fail2; |
15e3004a | 633 | } |
c8b97818 | 634 | } |
15e3004a | 635 | faili = nr_pages - 1; |
c8b97818 CM |
636 | cb->nr_pages = nr_pages; |
637 | ||
4b384318 MF |
638 | /* In the parent-locked case, we only locked the range we are |
639 | * interested in. In all other cases, we can opportunistically | |
640 | * cache decompressed data that goes beyond the requested range. */ | |
641 | if (!(bio_flags & EXTENT_BIO_PARENT_LOCKED)) | |
642 | add_ra_bio_pages(inode, em_start + em_len, cb); | |
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 | } |