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 | ||
974b1adc CH |
84 | static int btrfs_decompress_bio(int type, struct page **pages_in, |
85 | u64 disk_start, struct bio *orig_bio, | |
86 | size_t srclen); | |
48a3b636 | 87 | |
2ff7e61e | 88 | static inline int compressed_bio_size(struct btrfs_fs_info *fs_info, |
d20f7043 CM |
89 | unsigned long disk_size) |
90 | { | |
0b246afa | 91 | u16 csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6c41761f | 92 | |
d20f7043 | 93 | return sizeof(struct compressed_bio) + |
0b246afa | 94 | (DIV_ROUND_UP(disk_size, fs_info->sectorsize)) * csum_size; |
d20f7043 CM |
95 | } |
96 | ||
c8b97818 CM |
97 | static struct bio *compressed_bio_alloc(struct block_device *bdev, |
98 | u64 first_byte, gfp_t gfp_flags) | |
99 | { | |
b54ffb73 | 100 | return btrfs_bio_alloc(bdev, first_byte >> 9, BIO_MAX_PAGES, gfp_flags); |
c8b97818 CM |
101 | } |
102 | ||
f898ac6a | 103 | static int check_compressed_csum(struct btrfs_inode *inode, |
d20f7043 CM |
104 | struct compressed_bio *cb, |
105 | u64 disk_start) | |
106 | { | |
107 | int ret; | |
d20f7043 CM |
108 | struct page *page; |
109 | unsigned long i; | |
110 | char *kaddr; | |
111 | u32 csum; | |
112 | u32 *cb_sum = &cb->sums; | |
113 | ||
f898ac6a | 114 | if (inode->flags & BTRFS_INODE_NODATASUM) |
d20f7043 CM |
115 | return 0; |
116 | ||
117 | for (i = 0; i < cb->nr_pages; i++) { | |
118 | page = cb->compressed_pages[i]; | |
119 | csum = ~(u32)0; | |
120 | ||
7ac687d9 | 121 | kaddr = kmap_atomic(page); |
09cbfeaf | 122 | csum = btrfs_csum_data(kaddr, csum, PAGE_SIZE); |
0b5e3daf | 123 | btrfs_csum_final(csum, (u8 *)&csum); |
7ac687d9 | 124 | kunmap_atomic(kaddr); |
d20f7043 CM |
125 | |
126 | if (csum != *cb_sum) { | |
f898ac6a | 127 | btrfs_print_data_csum_error(inode, disk_start, csum, |
0970a22e | 128 | *cb_sum, cb->mirror_num); |
d20f7043 CM |
129 | ret = -EIO; |
130 | goto fail; | |
131 | } | |
132 | cb_sum++; | |
133 | ||
134 | } | |
135 | ret = 0; | |
136 | fail: | |
137 | return ret; | |
138 | } | |
139 | ||
c8b97818 CM |
140 | /* when we finish reading compressed pages from the disk, we |
141 | * decompress them and then run the bio end_io routines on the | |
142 | * decompressed pages (in the inode address space). | |
143 | * | |
144 | * This allows the checksumming and other IO error handling routines | |
145 | * to work normally | |
146 | * | |
147 | * The compressed pages are freed here, and it must be run | |
148 | * in process context | |
149 | */ | |
4246a0b6 | 150 | static void end_compressed_bio_read(struct bio *bio) |
c8b97818 | 151 | { |
c8b97818 CM |
152 | struct compressed_bio *cb = bio->bi_private; |
153 | struct inode *inode; | |
154 | struct page *page; | |
155 | unsigned long index; | |
156 | int ret; | |
157 | ||
4246a0b6 | 158 | if (bio->bi_error) |
c8b97818 CM |
159 | cb->errors = 1; |
160 | ||
161 | /* if there are more bios still pending for this compressed | |
162 | * extent, just exit | |
163 | */ | |
164 | if (!atomic_dec_and_test(&cb->pending_bios)) | |
165 | goto out; | |
166 | ||
d20f7043 | 167 | inode = cb->inode; |
f898ac6a | 168 | ret = check_compressed_csum(BTRFS_I(inode), cb, |
4f024f37 | 169 | (u64)bio->bi_iter.bi_sector << 9); |
d20f7043 CM |
170 | if (ret) |
171 | goto csum_failed; | |
172 | ||
c8b97818 CM |
173 | /* ok, we're the last bio for this extent, lets start |
174 | * the decompression. | |
175 | */ | |
974b1adc | 176 | ret = btrfs_decompress_bio(cb->compress_type, |
261507a0 LZ |
177 | cb->compressed_pages, |
178 | cb->start, | |
974b1adc | 179 | cb->orig_bio, |
261507a0 | 180 | cb->compressed_len); |
d20f7043 | 181 | csum_failed: |
c8b97818 CM |
182 | if (ret) |
183 | cb->errors = 1; | |
184 | ||
185 | /* release the compressed pages */ | |
186 | index = 0; | |
187 | for (index = 0; index < cb->nr_pages; index++) { | |
188 | page = cb->compressed_pages[index]; | |
189 | page->mapping = NULL; | |
09cbfeaf | 190 | put_page(page); |
c8b97818 CM |
191 | } |
192 | ||
193 | /* do io completion on the original bio */ | |
771ed689 | 194 | if (cb->errors) { |
c8b97818 | 195 | bio_io_error(cb->orig_bio); |
d20f7043 | 196 | } else { |
2c30c71b KO |
197 | int i; |
198 | struct bio_vec *bvec; | |
d20f7043 CM |
199 | |
200 | /* | |
201 | * we have verified the checksum already, set page | |
202 | * checked so the end_io handlers know about it | |
203 | */ | |
2c30c71b | 204 | bio_for_each_segment_all(bvec, cb->orig_bio, i) |
d20f7043 | 205 | SetPageChecked(bvec->bv_page); |
2c30c71b | 206 | |
4246a0b6 | 207 | bio_endio(cb->orig_bio); |
d20f7043 | 208 | } |
c8b97818 CM |
209 | |
210 | /* finally free the cb struct */ | |
211 | kfree(cb->compressed_pages); | |
212 | kfree(cb); | |
213 | out: | |
214 | bio_put(bio); | |
215 | } | |
216 | ||
217 | /* | |
218 | * Clear the writeback bits on all of the file | |
219 | * pages for a compressed write | |
220 | */ | |
7bdcefc1 FM |
221 | static noinline void end_compressed_writeback(struct inode *inode, |
222 | const struct compressed_bio *cb) | |
c8b97818 | 223 | { |
09cbfeaf KS |
224 | unsigned long index = cb->start >> PAGE_SHIFT; |
225 | unsigned long end_index = (cb->start + cb->len - 1) >> PAGE_SHIFT; | |
c8b97818 CM |
226 | struct page *pages[16]; |
227 | unsigned long nr_pages = end_index - index + 1; | |
228 | int i; | |
229 | int ret; | |
230 | ||
7bdcefc1 FM |
231 | if (cb->errors) |
232 | mapping_set_error(inode->i_mapping, -EIO); | |
233 | ||
d397712b | 234 | while (nr_pages > 0) { |
c8b97818 | 235 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
236 | min_t(unsigned long, |
237 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
238 | if (ret == 0) { |
239 | nr_pages -= 1; | |
240 | index += 1; | |
241 | continue; | |
242 | } | |
243 | for (i = 0; i < ret; i++) { | |
7bdcefc1 FM |
244 | if (cb->errors) |
245 | SetPageError(pages[i]); | |
c8b97818 | 246 | end_page_writeback(pages[i]); |
09cbfeaf | 247 | put_page(pages[i]); |
c8b97818 CM |
248 | } |
249 | nr_pages -= ret; | |
250 | index += ret; | |
251 | } | |
252 | /* the inode may be gone now */ | |
c8b97818 CM |
253 | } |
254 | ||
255 | /* | |
256 | * do the cleanup once all the compressed pages hit the disk. | |
257 | * This will clear writeback on the file pages and free the compressed | |
258 | * pages. | |
259 | * | |
260 | * This also calls the writeback end hooks for the file pages so that | |
261 | * metadata and checksums can be updated in the file. | |
262 | */ | |
4246a0b6 | 263 | static void end_compressed_bio_write(struct bio *bio) |
c8b97818 CM |
264 | { |
265 | struct extent_io_tree *tree; | |
266 | struct compressed_bio *cb = bio->bi_private; | |
267 | struct inode *inode; | |
268 | struct page *page; | |
269 | unsigned long index; | |
270 | ||
4246a0b6 | 271 | if (bio->bi_error) |
c8b97818 CM |
272 | cb->errors = 1; |
273 | ||
274 | /* if there are more bios still pending for this compressed | |
275 | * extent, just exit | |
276 | */ | |
277 | if (!atomic_dec_and_test(&cb->pending_bios)) | |
278 | goto out; | |
279 | ||
280 | /* ok, we're the last bio for this extent, step one is to | |
281 | * call back into the FS and do all the end_io operations | |
282 | */ | |
283 | inode = cb->inode; | |
284 | tree = &BTRFS_I(inode)->io_tree; | |
70b99e69 | 285 | cb->compressed_pages[0]->mapping = cb->inode->i_mapping; |
c8b97818 CM |
286 | tree->ops->writepage_end_io_hook(cb->compressed_pages[0], |
287 | cb->start, | |
288 | cb->start + cb->len - 1, | |
7bdcefc1 | 289 | NULL, |
4246a0b6 | 290 | bio->bi_error ? 0 : 1); |
70b99e69 | 291 | cb->compressed_pages[0]->mapping = NULL; |
c8b97818 | 292 | |
7bdcefc1 | 293 | end_compressed_writeback(inode, cb); |
c8b97818 CM |
294 | /* note, our inode could be gone now */ |
295 | ||
296 | /* | |
297 | * release the compressed pages, these came from alloc_page and | |
298 | * are not attached to the inode at all | |
299 | */ | |
300 | index = 0; | |
301 | for (index = 0; index < cb->nr_pages; index++) { | |
302 | page = cb->compressed_pages[index]; | |
303 | page->mapping = NULL; | |
09cbfeaf | 304 | put_page(page); |
c8b97818 CM |
305 | } |
306 | ||
307 | /* finally free the cb struct */ | |
308 | kfree(cb->compressed_pages); | |
309 | kfree(cb); | |
310 | out: | |
311 | bio_put(bio); | |
312 | } | |
313 | ||
314 | /* | |
315 | * worker function to build and submit bios for previously compressed pages. | |
316 | * The corresponding pages in the inode should be marked for writeback | |
317 | * and the compressed pages should have a reference on them for dropping | |
318 | * when the IO is complete. | |
319 | * | |
320 | * This also checksums the file bytes and gets things ready for | |
321 | * the end io hooks. | |
322 | */ | |
323 | int btrfs_submit_compressed_write(struct inode *inode, u64 start, | |
324 | unsigned long len, u64 disk_start, | |
325 | unsigned long compressed_len, | |
326 | struct page **compressed_pages, | |
327 | unsigned long nr_pages) | |
328 | { | |
0b246afa | 329 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
c8b97818 | 330 | struct bio *bio = NULL; |
c8b97818 CM |
331 | struct compressed_bio *cb; |
332 | unsigned long bytes_left; | |
333 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
306e16ce | 334 | int pg_index = 0; |
c8b97818 CM |
335 | struct page *page; |
336 | u64 first_byte = disk_start; | |
337 | struct block_device *bdev; | |
338 | int ret; | |
e55179b3 | 339 | int skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
c8b97818 | 340 | |
09cbfeaf | 341 | WARN_ON(start & ((u64)PAGE_SIZE - 1)); |
2ff7e61e | 342 | cb = kmalloc(compressed_bio_size(fs_info, compressed_len), GFP_NOFS); |
dac97e51 YS |
343 | if (!cb) |
344 | return -ENOMEM; | |
c8b97818 CM |
345 | atomic_set(&cb->pending_bios, 0); |
346 | cb->errors = 0; | |
347 | cb->inode = inode; | |
348 | cb->start = start; | |
349 | cb->len = len; | |
d20f7043 | 350 | cb->mirror_num = 0; |
c8b97818 CM |
351 | cb->compressed_pages = compressed_pages; |
352 | cb->compressed_len = compressed_len; | |
353 | cb->orig_bio = NULL; | |
354 | cb->nr_pages = nr_pages; | |
355 | ||
0b246afa | 356 | bdev = fs_info->fs_devices->latest_bdev; |
c8b97818 | 357 | |
c8b97818 | 358 | bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); |
67871254 | 359 | if (!bio) { |
dac97e51 YS |
360 | kfree(cb); |
361 | return -ENOMEM; | |
362 | } | |
37226b21 | 363 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
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 | 372 | page->mapping = inode->i_mapping; |
4f024f37 | 373 | if (bio->bi_iter.bi_size) |
81a75f67 | 374 | ret = io_tree->ops->merge_bio_hook(page, 0, |
09cbfeaf | 375 | PAGE_SIZE, |
c8b97818 CM |
376 | bio, 0); |
377 | else | |
378 | ret = 0; | |
379 | ||
70b99e69 | 380 | page->mapping = NULL; |
09cbfeaf KS |
381 | if (ret || bio_add_page(bio, page, PAGE_SIZE, 0) < |
382 | PAGE_SIZE) { | |
c8b97818 CM |
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); | |
0b246afa JM |
392 | ret = btrfs_bio_wq_end_io(fs_info, bio, |
393 | BTRFS_WQ_ENDIO_DATA); | |
79787eaa | 394 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 395 | |
e55179b3 | 396 | if (!skip_sum) { |
2ff7e61e | 397 | ret = btrfs_csum_one_bio(inode, bio, start, 1); |
79787eaa | 398 | BUG_ON(ret); /* -ENOMEM */ |
e55179b3 | 399 | } |
d20f7043 | 400 | |
2ff7e61e | 401 | ret = btrfs_map_bio(fs_info, bio, 0, 1); |
f5daf2c7 LB |
402 | if (ret) { |
403 | bio->bi_error = ret; | |
404 | bio_endio(bio); | |
405 | } | |
c8b97818 CM |
406 | |
407 | bio_put(bio); | |
408 | ||
409 | bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); | |
e627ee7b | 410 | BUG_ON(!bio); |
37226b21 | 411 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
c8b97818 CM |
412 | bio->bi_private = cb; |
413 | bio->bi_end_io = end_compressed_bio_write; | |
09cbfeaf | 414 | bio_add_page(bio, page, PAGE_SIZE, 0); |
c8b97818 | 415 | } |
09cbfeaf | 416 | if (bytes_left < PAGE_SIZE) { |
0b246afa | 417 | btrfs_info(fs_info, |
efe120a0 | 418 | "bytes left %lu compress len %lu nr %lu", |
cfbc246e CM |
419 | bytes_left, cb->compressed_len, cb->nr_pages); |
420 | } | |
09cbfeaf KS |
421 | bytes_left -= PAGE_SIZE; |
422 | first_byte += PAGE_SIZE; | |
771ed689 | 423 | cond_resched(); |
c8b97818 CM |
424 | } |
425 | bio_get(bio); | |
426 | ||
0b246afa | 427 | ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
79787eaa | 428 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 429 | |
e55179b3 | 430 | if (!skip_sum) { |
2ff7e61e | 431 | ret = btrfs_csum_one_bio(inode, bio, start, 1); |
79787eaa | 432 | BUG_ON(ret); /* -ENOMEM */ |
e55179b3 | 433 | } |
d20f7043 | 434 | |
2ff7e61e | 435 | ret = btrfs_map_bio(fs_info, bio, 0, 1); |
f5daf2c7 LB |
436 | if (ret) { |
437 | bio->bi_error = ret; | |
438 | bio_endio(bio); | |
439 | } | |
c8b97818 CM |
440 | |
441 | bio_put(bio); | |
442 | return 0; | |
443 | } | |
444 | ||
2a4d0c90 CH |
445 | static u64 bio_end_offset(struct bio *bio) |
446 | { | |
447 | struct bio_vec *last = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
448 | ||
449 | return page_offset(last->bv_page) + last->bv_len + last->bv_offset; | |
450 | } | |
451 | ||
771ed689 CM |
452 | static noinline int add_ra_bio_pages(struct inode *inode, |
453 | u64 compressed_end, | |
454 | struct compressed_bio *cb) | |
455 | { | |
456 | unsigned long end_index; | |
306e16ce | 457 | unsigned long pg_index; |
771ed689 CM |
458 | u64 last_offset; |
459 | u64 isize = i_size_read(inode); | |
460 | int ret; | |
461 | struct page *page; | |
462 | unsigned long nr_pages = 0; | |
463 | struct extent_map *em; | |
464 | struct address_space *mapping = inode->i_mapping; | |
771ed689 CM |
465 | struct extent_map_tree *em_tree; |
466 | struct extent_io_tree *tree; | |
467 | u64 end; | |
468 | int misses = 0; | |
469 | ||
2a4d0c90 | 470 | last_offset = bio_end_offset(cb->orig_bio); |
771ed689 CM |
471 | em_tree = &BTRFS_I(inode)->extent_tree; |
472 | tree = &BTRFS_I(inode)->io_tree; | |
473 | ||
474 | if (isize == 0) | |
475 | return 0; | |
476 | ||
09cbfeaf | 477 | end_index = (i_size_read(inode) - 1) >> PAGE_SHIFT; |
771ed689 | 478 | |
d397712b | 479 | while (last_offset < compressed_end) { |
09cbfeaf | 480 | pg_index = last_offset >> PAGE_SHIFT; |
771ed689 | 481 | |
306e16ce | 482 | if (pg_index > end_index) |
771ed689 CM |
483 | break; |
484 | ||
485 | rcu_read_lock(); | |
306e16ce | 486 | page = radix_tree_lookup(&mapping->page_tree, pg_index); |
771ed689 | 487 | rcu_read_unlock(); |
0cd6144a | 488 | if (page && !radix_tree_exceptional_entry(page)) { |
771ed689 CM |
489 | misses++; |
490 | if (misses > 4) | |
491 | break; | |
492 | goto next; | |
493 | } | |
494 | ||
c62d2555 MH |
495 | page = __page_cache_alloc(mapping_gfp_constraint(mapping, |
496 | ~__GFP_FS)); | |
771ed689 CM |
497 | if (!page) |
498 | break; | |
499 | ||
c62d2555 | 500 | if (add_to_page_cache_lru(page, mapping, pg_index, GFP_NOFS)) { |
09cbfeaf | 501 | put_page(page); |
771ed689 CM |
502 | goto next; |
503 | } | |
504 | ||
09cbfeaf | 505 | end = last_offset + PAGE_SIZE - 1; |
771ed689 CM |
506 | /* |
507 | * at this point, we have a locked page in the page cache | |
508 | * for these bytes in the file. But, we have to make | |
509 | * sure they map to this compressed extent on disk. | |
510 | */ | |
511 | set_page_extent_mapped(page); | |
d0082371 | 512 | lock_extent(tree, last_offset, end); |
890871be | 513 | read_lock(&em_tree->lock); |
771ed689 | 514 | em = lookup_extent_mapping(em_tree, last_offset, |
09cbfeaf | 515 | PAGE_SIZE); |
890871be | 516 | read_unlock(&em_tree->lock); |
771ed689 CM |
517 | |
518 | if (!em || last_offset < em->start || | |
09cbfeaf | 519 | (last_offset + PAGE_SIZE > extent_map_end(em)) || |
4f024f37 | 520 | (em->block_start >> 9) != cb->orig_bio->bi_iter.bi_sector) { |
771ed689 | 521 | free_extent_map(em); |
d0082371 | 522 | unlock_extent(tree, last_offset, end); |
771ed689 | 523 | unlock_page(page); |
09cbfeaf | 524 | put_page(page); |
771ed689 CM |
525 | break; |
526 | } | |
527 | free_extent_map(em); | |
528 | ||
529 | if (page->index == end_index) { | |
530 | char *userpage; | |
09cbfeaf | 531 | size_t zero_offset = isize & (PAGE_SIZE - 1); |
771ed689 CM |
532 | |
533 | if (zero_offset) { | |
534 | int zeros; | |
09cbfeaf | 535 | zeros = PAGE_SIZE - zero_offset; |
7ac687d9 | 536 | userpage = kmap_atomic(page); |
771ed689 CM |
537 | memset(userpage + zero_offset, 0, zeros); |
538 | flush_dcache_page(page); | |
7ac687d9 | 539 | kunmap_atomic(userpage); |
771ed689 CM |
540 | } |
541 | } | |
542 | ||
543 | ret = bio_add_page(cb->orig_bio, page, | |
09cbfeaf | 544 | PAGE_SIZE, 0); |
771ed689 | 545 | |
09cbfeaf | 546 | if (ret == PAGE_SIZE) { |
771ed689 | 547 | nr_pages++; |
09cbfeaf | 548 | put_page(page); |
771ed689 | 549 | } else { |
d0082371 | 550 | unlock_extent(tree, last_offset, end); |
771ed689 | 551 | unlock_page(page); |
09cbfeaf | 552 | put_page(page); |
771ed689 CM |
553 | break; |
554 | } | |
555 | next: | |
09cbfeaf | 556 | last_offset += PAGE_SIZE; |
771ed689 | 557 | } |
771ed689 CM |
558 | return 0; |
559 | } | |
560 | ||
c8b97818 CM |
561 | /* |
562 | * for a compressed read, the bio we get passed has all the inode pages | |
563 | * in it. We don't actually do IO on those pages but allocate new ones | |
564 | * to hold the compressed pages on disk. | |
565 | * | |
4f024f37 | 566 | * bio->bi_iter.bi_sector points to the compressed extent on disk |
c8b97818 | 567 | * bio->bi_io_vec points to all of the inode pages |
c8b97818 CM |
568 | * |
569 | * After the compressed pages are read, we copy the bytes into the | |
570 | * bio we were passed and then call the bio end_io calls | |
571 | */ | |
572 | int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio, | |
573 | int mirror_num, unsigned long bio_flags) | |
574 | { | |
0b246afa | 575 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
c8b97818 CM |
576 | struct extent_io_tree *tree; |
577 | struct extent_map_tree *em_tree; | |
578 | struct compressed_bio *cb; | |
c8b97818 CM |
579 | unsigned long compressed_len; |
580 | unsigned long nr_pages; | |
306e16ce | 581 | unsigned long pg_index; |
c8b97818 CM |
582 | struct page *page; |
583 | struct block_device *bdev; | |
584 | struct bio *comp_bio; | |
4f024f37 | 585 | u64 cur_disk_byte = (u64)bio->bi_iter.bi_sector << 9; |
e04ca626 CM |
586 | u64 em_len; |
587 | u64 em_start; | |
c8b97818 | 588 | struct extent_map *em; |
6b82ce8d | 589 | int ret = -ENOMEM; |
15e3004a | 590 | int faili = 0; |
d20f7043 | 591 | u32 *sums; |
c8b97818 CM |
592 | |
593 | tree = &BTRFS_I(inode)->io_tree; | |
594 | em_tree = &BTRFS_I(inode)->extent_tree; | |
595 | ||
596 | /* we need the actual starting offset of this extent in the file */ | |
890871be | 597 | read_lock(&em_tree->lock); |
c8b97818 CM |
598 | em = lookup_extent_mapping(em_tree, |
599 | page_offset(bio->bi_io_vec->bv_page), | |
09cbfeaf | 600 | PAGE_SIZE); |
890871be | 601 | read_unlock(&em_tree->lock); |
285190d9 TI |
602 | if (!em) |
603 | return -EIO; | |
c8b97818 | 604 | |
d20f7043 | 605 | compressed_len = em->block_len; |
2ff7e61e | 606 | cb = kmalloc(compressed_bio_size(fs_info, compressed_len), GFP_NOFS); |
6b82ce8d | 607 | if (!cb) |
608 | goto out; | |
609 | ||
c8b97818 CM |
610 | atomic_set(&cb->pending_bios, 0); |
611 | cb->errors = 0; | |
612 | cb->inode = inode; | |
d20f7043 CM |
613 | cb->mirror_num = mirror_num; |
614 | sums = &cb->sums; | |
c8b97818 | 615 | |
ff5b7ee3 | 616 | cb->start = em->orig_start; |
e04ca626 CM |
617 | em_len = em->len; |
618 | em_start = em->start; | |
d20f7043 | 619 | |
c8b97818 | 620 | free_extent_map(em); |
e04ca626 | 621 | em = NULL; |
c8b97818 | 622 | |
81381053 | 623 | cb->len = bio->bi_iter.bi_size; |
c8b97818 | 624 | cb->compressed_len = compressed_len; |
261507a0 | 625 | cb->compress_type = extent_compress_type(bio_flags); |
c8b97818 CM |
626 | cb->orig_bio = bio; |
627 | ||
09cbfeaf | 628 | nr_pages = DIV_ROUND_UP(compressed_len, PAGE_SIZE); |
31e818fe | 629 | cb->compressed_pages = kcalloc(nr_pages, sizeof(struct page *), |
c8b97818 | 630 | GFP_NOFS); |
6b82ce8d | 631 | if (!cb->compressed_pages) |
632 | goto fail1; | |
633 | ||
0b246afa | 634 | bdev = fs_info->fs_devices->latest_bdev; |
c8b97818 | 635 | |
306e16ce DS |
636 | for (pg_index = 0; pg_index < nr_pages; pg_index++) { |
637 | cb->compressed_pages[pg_index] = alloc_page(GFP_NOFS | | |
c8b97818 | 638 | __GFP_HIGHMEM); |
15e3004a JB |
639 | if (!cb->compressed_pages[pg_index]) { |
640 | faili = pg_index - 1; | |
641 | ret = -ENOMEM; | |
6b82ce8d | 642 | goto fail2; |
15e3004a | 643 | } |
c8b97818 | 644 | } |
15e3004a | 645 | faili = nr_pages - 1; |
c8b97818 CM |
646 | cb->nr_pages = nr_pages; |
647 | ||
7f042a83 | 648 | add_ra_bio_pages(inode, em_start + em_len, cb); |
771ed689 | 649 | |
771ed689 | 650 | /* include any pages we added in add_ra-bio_pages */ |
81381053 | 651 | cb->len = bio->bi_iter.bi_size; |
771ed689 | 652 | |
c8b97818 | 653 | comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS); |
6b82ce8d | 654 | if (!comp_bio) |
655 | goto fail2; | |
37226b21 | 656 | bio_set_op_attrs (comp_bio, REQ_OP_READ, 0); |
c8b97818 CM |
657 | comp_bio->bi_private = cb; |
658 | comp_bio->bi_end_io = end_compressed_bio_read; | |
659 | atomic_inc(&cb->pending_bios); | |
660 | ||
306e16ce DS |
661 | for (pg_index = 0; pg_index < nr_pages; pg_index++) { |
662 | page = cb->compressed_pages[pg_index]; | |
c8b97818 | 663 | page->mapping = inode->i_mapping; |
09cbfeaf | 664 | page->index = em_start >> PAGE_SHIFT; |
d20f7043 | 665 | |
4f024f37 | 666 | if (comp_bio->bi_iter.bi_size) |
81a75f67 | 667 | ret = tree->ops->merge_bio_hook(page, 0, |
09cbfeaf | 668 | PAGE_SIZE, |
c8b97818 CM |
669 | comp_bio, 0); |
670 | else | |
671 | ret = 0; | |
672 | ||
70b99e69 | 673 | page->mapping = NULL; |
09cbfeaf KS |
674 | if (ret || bio_add_page(comp_bio, page, PAGE_SIZE, 0) < |
675 | PAGE_SIZE) { | |
c8b97818 CM |
676 | bio_get(comp_bio); |
677 | ||
0b246afa JM |
678 | ret = btrfs_bio_wq_end_io(fs_info, comp_bio, |
679 | BTRFS_WQ_ENDIO_DATA); | |
79787eaa | 680 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 681 | |
af09abfe CM |
682 | /* |
683 | * inc the count before we submit the bio so | |
684 | * we know the end IO handler won't happen before | |
685 | * we inc the count. Otherwise, the cb might get | |
686 | * freed before we're done setting it up | |
687 | */ | |
688 | atomic_inc(&cb->pending_bios); | |
689 | ||
6cbff00f | 690 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { |
2ff7e61e JM |
691 | ret = btrfs_lookup_bio_sums(inode, comp_bio, |
692 | sums); | |
79787eaa | 693 | BUG_ON(ret); /* -ENOMEM */ |
d20f7043 | 694 | } |
ed6078f7 | 695 | sums += DIV_ROUND_UP(comp_bio->bi_iter.bi_size, |
0b246afa | 696 | fs_info->sectorsize); |
d20f7043 | 697 | |
2ff7e61e | 698 | ret = btrfs_map_bio(fs_info, comp_bio, mirror_num, 0); |
4246a0b6 | 699 | if (ret) { |
14155caf | 700 | comp_bio->bi_error = ret; |
4246a0b6 CH |
701 | bio_endio(comp_bio); |
702 | } | |
c8b97818 CM |
703 | |
704 | bio_put(comp_bio); | |
705 | ||
706 | comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, | |
707 | GFP_NOFS); | |
e627ee7b | 708 | BUG_ON(!comp_bio); |
37226b21 | 709 | bio_set_op_attrs(comp_bio, REQ_OP_READ, 0); |
771ed689 CM |
710 | comp_bio->bi_private = cb; |
711 | comp_bio->bi_end_io = end_compressed_bio_read; | |
712 | ||
09cbfeaf | 713 | bio_add_page(comp_bio, page, PAGE_SIZE, 0); |
c8b97818 | 714 | } |
09cbfeaf | 715 | cur_disk_byte += PAGE_SIZE; |
c8b97818 CM |
716 | } |
717 | bio_get(comp_bio); | |
718 | ||
0b246afa | 719 | ret = btrfs_bio_wq_end_io(fs_info, comp_bio, BTRFS_WQ_ENDIO_DATA); |
79787eaa | 720 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 721 | |
c2db1073 | 722 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { |
2ff7e61e | 723 | ret = btrfs_lookup_bio_sums(inode, comp_bio, sums); |
79787eaa | 724 | BUG_ON(ret); /* -ENOMEM */ |
c2db1073 | 725 | } |
d20f7043 | 726 | |
2ff7e61e | 727 | ret = btrfs_map_bio(fs_info, comp_bio, mirror_num, 0); |
4246a0b6 | 728 | if (ret) { |
14155caf | 729 | comp_bio->bi_error = ret; |
4246a0b6 CH |
730 | bio_endio(comp_bio); |
731 | } | |
c8b97818 CM |
732 | |
733 | bio_put(comp_bio); | |
734 | return 0; | |
6b82ce8d | 735 | |
736 | fail2: | |
15e3004a JB |
737 | while (faili >= 0) { |
738 | __free_page(cb->compressed_pages[faili]); | |
739 | faili--; | |
740 | } | |
6b82ce8d | 741 | |
742 | kfree(cb->compressed_pages); | |
743 | fail1: | |
744 | kfree(cb); | |
745 | out: | |
746 | free_extent_map(em); | |
747 | return ret; | |
c8b97818 | 748 | } |
261507a0 | 749 | |
d9187649 BL |
750 | static struct { |
751 | struct list_head idle_ws; | |
752 | spinlock_t ws_lock; | |
6ac10a6a DS |
753 | /* Number of free workspaces */ |
754 | int free_ws; | |
755 | /* Total number of allocated workspaces */ | |
756 | atomic_t total_ws; | |
757 | /* Waiters for a free workspace */ | |
d9187649 BL |
758 | wait_queue_head_t ws_wait; |
759 | } btrfs_comp_ws[BTRFS_COMPRESS_TYPES]; | |
261507a0 | 760 | |
e8c9f186 | 761 | static const struct btrfs_compress_op * const btrfs_compress_op[] = { |
261507a0 | 762 | &btrfs_zlib_compress, |
a6fa6fae | 763 | &btrfs_lzo_compress, |
261507a0 LZ |
764 | }; |
765 | ||
143bede5 | 766 | void __init btrfs_init_compress(void) |
261507a0 LZ |
767 | { |
768 | int i; | |
769 | ||
770 | for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) { | |
f77dd0d6 DS |
771 | struct list_head *workspace; |
772 | ||
d9187649 BL |
773 | INIT_LIST_HEAD(&btrfs_comp_ws[i].idle_ws); |
774 | spin_lock_init(&btrfs_comp_ws[i].ws_lock); | |
6ac10a6a | 775 | atomic_set(&btrfs_comp_ws[i].total_ws, 0); |
d9187649 | 776 | init_waitqueue_head(&btrfs_comp_ws[i].ws_wait); |
f77dd0d6 DS |
777 | |
778 | /* | |
779 | * Preallocate one workspace for each compression type so | |
780 | * we can guarantee forward progress in the worst case | |
781 | */ | |
782 | workspace = btrfs_compress_op[i]->alloc_workspace(); | |
783 | if (IS_ERR(workspace)) { | |
62e85577 | 784 | pr_warn("BTRFS: cannot preallocate compression workspace, will try later\n"); |
f77dd0d6 DS |
785 | } else { |
786 | atomic_set(&btrfs_comp_ws[i].total_ws, 1); | |
787 | btrfs_comp_ws[i].free_ws = 1; | |
788 | list_add(workspace, &btrfs_comp_ws[i].idle_ws); | |
789 | } | |
261507a0 | 790 | } |
261507a0 LZ |
791 | } |
792 | ||
793 | /* | |
e721e49d DS |
794 | * This finds an available workspace or allocates a new one. |
795 | * If it's not possible to allocate a new one, waits until there's one. | |
796 | * Preallocation makes a forward progress guarantees and we do not return | |
797 | * errors. | |
261507a0 LZ |
798 | */ |
799 | static struct list_head *find_workspace(int type) | |
800 | { | |
801 | struct list_head *workspace; | |
802 | int cpus = num_online_cpus(); | |
803 | int idx = type - 1; | |
804 | ||
d9187649 BL |
805 | struct list_head *idle_ws = &btrfs_comp_ws[idx].idle_ws; |
806 | spinlock_t *ws_lock = &btrfs_comp_ws[idx].ws_lock; | |
6ac10a6a | 807 | atomic_t *total_ws = &btrfs_comp_ws[idx].total_ws; |
d9187649 | 808 | wait_queue_head_t *ws_wait = &btrfs_comp_ws[idx].ws_wait; |
6ac10a6a | 809 | int *free_ws = &btrfs_comp_ws[idx].free_ws; |
261507a0 | 810 | again: |
d9187649 BL |
811 | spin_lock(ws_lock); |
812 | if (!list_empty(idle_ws)) { | |
813 | workspace = idle_ws->next; | |
261507a0 | 814 | list_del(workspace); |
6ac10a6a | 815 | (*free_ws)--; |
d9187649 | 816 | spin_unlock(ws_lock); |
261507a0 LZ |
817 | return workspace; |
818 | ||
819 | } | |
6ac10a6a | 820 | if (atomic_read(total_ws) > cpus) { |
261507a0 LZ |
821 | DEFINE_WAIT(wait); |
822 | ||
d9187649 BL |
823 | spin_unlock(ws_lock); |
824 | prepare_to_wait(ws_wait, &wait, TASK_UNINTERRUPTIBLE); | |
6ac10a6a | 825 | if (atomic_read(total_ws) > cpus && !*free_ws) |
261507a0 | 826 | schedule(); |
d9187649 | 827 | finish_wait(ws_wait, &wait); |
261507a0 LZ |
828 | goto again; |
829 | } | |
6ac10a6a | 830 | atomic_inc(total_ws); |
d9187649 | 831 | spin_unlock(ws_lock); |
261507a0 LZ |
832 | |
833 | workspace = btrfs_compress_op[idx]->alloc_workspace(); | |
834 | if (IS_ERR(workspace)) { | |
6ac10a6a | 835 | atomic_dec(total_ws); |
d9187649 | 836 | wake_up(ws_wait); |
e721e49d DS |
837 | |
838 | /* | |
839 | * Do not return the error but go back to waiting. There's a | |
840 | * workspace preallocated for each type and the compression | |
841 | * time is bounded so we get to a workspace eventually. This | |
842 | * makes our caller's life easier. | |
52356716 DS |
843 | * |
844 | * To prevent silent and low-probability deadlocks (when the | |
845 | * initial preallocation fails), check if there are any | |
846 | * workspaces at all. | |
e721e49d | 847 | */ |
52356716 DS |
848 | if (atomic_read(total_ws) == 0) { |
849 | static DEFINE_RATELIMIT_STATE(_rs, | |
850 | /* once per minute */ 60 * HZ, | |
851 | /* no burst */ 1); | |
852 | ||
853 | if (__ratelimit(&_rs)) { | |
ab8d0fc4 | 854 | pr_warn("BTRFS: no compression workspaces, low memory, retrying\n"); |
52356716 DS |
855 | } |
856 | } | |
e721e49d | 857 | goto again; |
261507a0 LZ |
858 | } |
859 | return workspace; | |
860 | } | |
861 | ||
862 | /* | |
863 | * put a workspace struct back on the list or free it if we have enough | |
864 | * idle ones sitting around | |
865 | */ | |
866 | static void free_workspace(int type, struct list_head *workspace) | |
867 | { | |
868 | int idx = type - 1; | |
d9187649 BL |
869 | struct list_head *idle_ws = &btrfs_comp_ws[idx].idle_ws; |
870 | spinlock_t *ws_lock = &btrfs_comp_ws[idx].ws_lock; | |
6ac10a6a | 871 | atomic_t *total_ws = &btrfs_comp_ws[idx].total_ws; |
d9187649 | 872 | wait_queue_head_t *ws_wait = &btrfs_comp_ws[idx].ws_wait; |
6ac10a6a | 873 | int *free_ws = &btrfs_comp_ws[idx].free_ws; |
d9187649 BL |
874 | |
875 | spin_lock(ws_lock); | |
6ac10a6a | 876 | if (*free_ws < num_online_cpus()) { |
d9187649 | 877 | list_add(workspace, idle_ws); |
6ac10a6a | 878 | (*free_ws)++; |
d9187649 | 879 | spin_unlock(ws_lock); |
261507a0 LZ |
880 | goto wake; |
881 | } | |
d9187649 | 882 | spin_unlock(ws_lock); |
261507a0 LZ |
883 | |
884 | btrfs_compress_op[idx]->free_workspace(workspace); | |
6ac10a6a | 885 | atomic_dec(total_ws); |
261507a0 | 886 | wake: |
a83342aa DS |
887 | /* |
888 | * Make sure counter is updated before we wake up waiters. | |
889 | */ | |
66657b31 | 890 | smp_mb(); |
d9187649 BL |
891 | if (waitqueue_active(ws_wait)) |
892 | wake_up(ws_wait); | |
261507a0 LZ |
893 | } |
894 | ||
895 | /* | |
896 | * cleanup function for module exit | |
897 | */ | |
898 | static void free_workspaces(void) | |
899 | { | |
900 | struct list_head *workspace; | |
901 | int i; | |
902 | ||
903 | for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) { | |
d9187649 BL |
904 | while (!list_empty(&btrfs_comp_ws[i].idle_ws)) { |
905 | workspace = btrfs_comp_ws[i].idle_ws.next; | |
261507a0 LZ |
906 | list_del(workspace); |
907 | btrfs_compress_op[i]->free_workspace(workspace); | |
6ac10a6a | 908 | atomic_dec(&btrfs_comp_ws[i].total_ws); |
261507a0 LZ |
909 | } |
910 | } | |
911 | } | |
912 | ||
913 | /* | |
38c31464 DS |
914 | * Given an address space and start and length, compress the bytes into @pages |
915 | * that are allocated on demand. | |
261507a0 | 916 | * |
38c31464 DS |
917 | * @out_pages is used to return the number of pages allocated. There |
918 | * may be pages allocated even if we return an error. | |
261507a0 | 919 | * |
38c31464 DS |
920 | * @total_in is used to return the number of bytes actually read. It |
921 | * may be smaller than the input length if we had to exit early because we | |
261507a0 LZ |
922 | * ran out of room in the pages array or because we cross the |
923 | * max_out threshold. | |
924 | * | |
38c31464 DS |
925 | * @total_out is an in/out parameter, must be set to the input length and will |
926 | * be also used to return the total number of compressed bytes | |
261507a0 | 927 | * |
38c31464 | 928 | * @max_out tells us the max number of bytes that we're allowed to |
261507a0 LZ |
929 | * stuff into pages |
930 | */ | |
931 | int btrfs_compress_pages(int type, struct address_space *mapping, | |
38c31464 | 932 | u64 start, struct page **pages, |
261507a0 LZ |
933 | unsigned long nr_dest_pages, |
934 | unsigned long *out_pages, | |
935 | unsigned long *total_in, | |
936 | unsigned long *total_out, | |
937 | unsigned long max_out) | |
938 | { | |
939 | struct list_head *workspace; | |
940 | int ret; | |
941 | ||
942 | workspace = find_workspace(type); | |
261507a0 LZ |
943 | |
944 | ret = btrfs_compress_op[type-1]->compress_pages(workspace, mapping, | |
38c31464 | 945 | start, pages, |
261507a0 LZ |
946 | nr_dest_pages, out_pages, |
947 | total_in, total_out, | |
948 | max_out); | |
949 | free_workspace(type, workspace); | |
950 | return ret; | |
951 | } | |
952 | ||
953 | /* | |
954 | * pages_in is an array of pages with compressed data. | |
955 | * | |
956 | * disk_start is the starting logical offset of this array in the file | |
957 | * | |
974b1adc | 958 | * orig_bio contains the pages from the file that we want to decompress into |
261507a0 LZ |
959 | * |
960 | * srclen is the number of bytes in pages_in | |
961 | * | |
962 | * The basic idea is that we have a bio that was created by readpages. | |
963 | * The pages in the bio are for the uncompressed data, and they may not | |
964 | * be contiguous. They all correspond to the range of bytes covered by | |
965 | * the compressed extent. | |
966 | */ | |
974b1adc CH |
967 | static int btrfs_decompress_bio(int type, struct page **pages_in, |
968 | u64 disk_start, struct bio *orig_bio, | |
969 | size_t srclen) | |
261507a0 LZ |
970 | { |
971 | struct list_head *workspace; | |
972 | int ret; | |
973 | ||
974 | workspace = find_workspace(type); | |
261507a0 | 975 | |
974b1adc CH |
976 | ret = btrfs_compress_op[type-1]->decompress_bio(workspace, pages_in, |
977 | disk_start, orig_bio, | |
978 | srclen); | |
261507a0 LZ |
979 | free_workspace(type, workspace); |
980 | return ret; | |
981 | } | |
982 | ||
983 | /* | |
984 | * a less complex decompression routine. Our compressed data fits in a | |
985 | * single page, and we want to read a single page out of it. | |
986 | * start_byte tells us the offset into the compressed data we're interested in | |
987 | */ | |
988 | int btrfs_decompress(int type, unsigned char *data_in, struct page *dest_page, | |
989 | unsigned long start_byte, size_t srclen, size_t destlen) | |
990 | { | |
991 | struct list_head *workspace; | |
992 | int ret; | |
993 | ||
994 | workspace = find_workspace(type); | |
261507a0 LZ |
995 | |
996 | ret = btrfs_compress_op[type-1]->decompress(workspace, data_in, | |
997 | dest_page, start_byte, | |
998 | srclen, destlen); | |
999 | ||
1000 | free_workspace(type, workspace); | |
1001 | return ret; | |
1002 | } | |
1003 | ||
8e4eef7a | 1004 | void btrfs_exit_compress(void) |
261507a0 LZ |
1005 | { |
1006 | free_workspaces(); | |
1007 | } | |
3a39c18d LZ |
1008 | |
1009 | /* | |
1010 | * Copy uncompressed data from working buffer to pages. | |
1011 | * | |
1012 | * buf_start is the byte offset we're of the start of our workspace buffer. | |
1013 | * | |
1014 | * total_out is the last byte of the buffer | |
1015 | */ | |
14a3357b | 1016 | int btrfs_decompress_buf2page(const char *buf, unsigned long buf_start, |
3a39c18d | 1017 | unsigned long total_out, u64 disk_start, |
974b1adc | 1018 | struct bio *bio) |
3a39c18d LZ |
1019 | { |
1020 | unsigned long buf_offset; | |
1021 | unsigned long current_buf_start; | |
1022 | unsigned long start_byte; | |
6e78b3f7 | 1023 | unsigned long prev_start_byte; |
3a39c18d LZ |
1024 | unsigned long working_bytes = total_out - buf_start; |
1025 | unsigned long bytes; | |
1026 | char *kaddr; | |
974b1adc | 1027 | struct bio_vec bvec = bio_iter_iovec(bio, bio->bi_iter); |
3a39c18d LZ |
1028 | |
1029 | /* | |
1030 | * start byte is the first byte of the page we're currently | |
1031 | * copying into relative to the start of the compressed data. | |
1032 | */ | |
974b1adc | 1033 | start_byte = page_offset(bvec.bv_page) - disk_start; |
3a39c18d LZ |
1034 | |
1035 | /* we haven't yet hit data corresponding to this page */ | |
1036 | if (total_out <= start_byte) | |
1037 | return 1; | |
1038 | ||
1039 | /* | |
1040 | * the start of the data we care about is offset into | |
1041 | * the middle of our working buffer | |
1042 | */ | |
1043 | if (total_out > start_byte && buf_start < start_byte) { | |
1044 | buf_offset = start_byte - buf_start; | |
1045 | working_bytes -= buf_offset; | |
1046 | } else { | |
1047 | buf_offset = 0; | |
1048 | } | |
1049 | current_buf_start = buf_start; | |
1050 | ||
1051 | /* copy bytes from the working buffer into the pages */ | |
1052 | while (working_bytes > 0) { | |
974b1adc CH |
1053 | bytes = min_t(unsigned long, bvec.bv_len, |
1054 | PAGE_SIZE - buf_offset); | |
3a39c18d | 1055 | bytes = min(bytes, working_bytes); |
974b1adc CH |
1056 | |
1057 | kaddr = kmap_atomic(bvec.bv_page); | |
1058 | memcpy(kaddr + bvec.bv_offset, buf + buf_offset, bytes); | |
7ac687d9 | 1059 | kunmap_atomic(kaddr); |
974b1adc | 1060 | flush_dcache_page(bvec.bv_page); |
3a39c18d | 1061 | |
3a39c18d LZ |
1062 | buf_offset += bytes; |
1063 | working_bytes -= bytes; | |
1064 | current_buf_start += bytes; | |
1065 | ||
1066 | /* check if we need to pick another page */ | |
974b1adc CH |
1067 | bio_advance(bio, bytes); |
1068 | if (!bio->bi_iter.bi_size) | |
1069 | return 0; | |
1070 | bvec = bio_iter_iovec(bio, bio->bi_iter); | |
6e78b3f7 | 1071 | prev_start_byte = start_byte; |
974b1adc | 1072 | start_byte = page_offset(bvec.bv_page) - disk_start; |
3a39c18d | 1073 | |
974b1adc | 1074 | /* |
6e78b3f7 OS |
1075 | * We need to make sure we're only adjusting |
1076 | * our offset into compression working buffer when | |
1077 | * we're switching pages. Otherwise we can incorrectly | |
1078 | * keep copying when we were actually done. | |
974b1adc | 1079 | */ |
6e78b3f7 OS |
1080 | if (start_byte != prev_start_byte) { |
1081 | /* | |
1082 | * make sure our new page is covered by this | |
1083 | * working buffer | |
1084 | */ | |
1085 | if (total_out <= start_byte) | |
1086 | return 1; | |
3a39c18d | 1087 | |
6e78b3f7 OS |
1088 | /* |
1089 | * the next page in the biovec might not be adjacent | |
1090 | * to the last page, but it might still be found | |
1091 | * inside this working buffer. bump our offset pointer | |
1092 | */ | |
1093 | if (total_out > start_byte && | |
1094 | current_buf_start < start_byte) { | |
1095 | buf_offset = start_byte - buf_start; | |
1096 | working_bytes = total_out - start_byte; | |
1097 | current_buf_start = buf_start + buf_offset; | |
1098 | } | |
3a39c18d LZ |
1099 | } |
1100 | } | |
1101 | ||
1102 | return 1; | |
1103 | } |