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 | 93 | return sizeof(struct compressed_bio) + |
ed6078f7 | 94 | (DIV_ROUND_UP(disk_size, root->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 | ||
d20f7043 CM |
103 | static int check_compressed_csum(struct inode *inode, |
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 | ||
6cbff00f | 114 | if (BTRFS_I(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); |
d20f7043 | 123 | btrfs_csum_final(csum, (char *)&csum); |
7ac687d9 | 124 | kunmap_atomic(kaddr); |
d20f7043 CM |
125 | |
126 | if (csum != *cb_sum) { | |
efe120a0 FH |
127 | btrfs_info(BTRFS_I(inode)->root->fs_info, |
128 | "csum failed ino %llu extent %llu csum %u wanted %u mirror %d", | |
129 | btrfs_ino(inode), disk_start, csum, *cb_sum, | |
130 | cb->mirror_num); | |
d20f7043 CM |
131 | ret = -EIO; |
132 | goto fail; | |
133 | } | |
134 | cb_sum++; | |
135 | ||
136 | } | |
137 | ret = 0; | |
138 | fail: | |
139 | return ret; | |
140 | } | |
141 | ||
c8b97818 CM |
142 | /* when we finish reading compressed pages from the disk, we |
143 | * decompress them and then run the bio end_io routines on the | |
144 | * decompressed pages (in the inode address space). | |
145 | * | |
146 | * This allows the checksumming and other IO error handling routines | |
147 | * to work normally | |
148 | * | |
149 | * The compressed pages are freed here, and it must be run | |
150 | * in process context | |
151 | */ | |
4246a0b6 | 152 | static void end_compressed_bio_read(struct bio *bio) |
c8b97818 | 153 | { |
c8b97818 CM |
154 | struct compressed_bio *cb = bio->bi_private; |
155 | struct inode *inode; | |
156 | struct page *page; | |
157 | unsigned long index; | |
158 | int ret; | |
159 | ||
4246a0b6 | 160 | if (bio->bi_error) |
c8b97818 CM |
161 | cb->errors = 1; |
162 | ||
163 | /* if there are more bios still pending for this compressed | |
164 | * extent, just exit | |
165 | */ | |
166 | if (!atomic_dec_and_test(&cb->pending_bios)) | |
167 | goto out; | |
168 | ||
d20f7043 | 169 | inode = cb->inode; |
4f024f37 KO |
170 | ret = check_compressed_csum(inode, cb, |
171 | (u64)bio->bi_iter.bi_sector << 9); | |
d20f7043 CM |
172 | if (ret) |
173 | goto csum_failed; | |
174 | ||
c8b97818 CM |
175 | /* ok, we're the last bio for this extent, lets start |
176 | * the decompression. | |
177 | */ | |
261507a0 LZ |
178 | ret = btrfs_decompress_biovec(cb->compress_type, |
179 | cb->compressed_pages, | |
180 | cb->start, | |
181 | cb->orig_bio->bi_io_vec, | |
182 | cb->orig_bio->bi_vcnt, | |
183 | cb->compressed_len); | |
d20f7043 | 184 | csum_failed: |
c8b97818 CM |
185 | if (ret) |
186 | cb->errors = 1; | |
187 | ||
188 | /* release the compressed pages */ | |
189 | index = 0; | |
190 | for (index = 0; index < cb->nr_pages; index++) { | |
191 | page = cb->compressed_pages[index]; | |
192 | page->mapping = NULL; | |
09cbfeaf | 193 | put_page(page); |
c8b97818 CM |
194 | } |
195 | ||
196 | /* do io completion on the original bio */ | |
771ed689 | 197 | if (cb->errors) { |
c8b97818 | 198 | bio_io_error(cb->orig_bio); |
d20f7043 | 199 | } else { |
2c30c71b KO |
200 | int i; |
201 | struct bio_vec *bvec; | |
d20f7043 CM |
202 | |
203 | /* | |
204 | * we have verified the checksum already, set page | |
205 | * checked so the end_io handlers know about it | |
206 | */ | |
2c30c71b | 207 | bio_for_each_segment_all(bvec, cb->orig_bio, i) |
d20f7043 | 208 | SetPageChecked(bvec->bv_page); |
2c30c71b | 209 | |
4246a0b6 | 210 | bio_endio(cb->orig_bio); |
d20f7043 | 211 | } |
c8b97818 CM |
212 | |
213 | /* finally free the cb struct */ | |
214 | kfree(cb->compressed_pages); | |
215 | kfree(cb); | |
216 | out: | |
217 | bio_put(bio); | |
218 | } | |
219 | ||
220 | /* | |
221 | * Clear the writeback bits on all of the file | |
222 | * pages for a compressed write | |
223 | */ | |
7bdcefc1 FM |
224 | static noinline void end_compressed_writeback(struct inode *inode, |
225 | const struct compressed_bio *cb) | |
c8b97818 | 226 | { |
09cbfeaf KS |
227 | unsigned long index = cb->start >> PAGE_SHIFT; |
228 | unsigned long end_index = (cb->start + cb->len - 1) >> PAGE_SHIFT; | |
c8b97818 CM |
229 | struct page *pages[16]; |
230 | unsigned long nr_pages = end_index - index + 1; | |
231 | int i; | |
232 | int ret; | |
233 | ||
7bdcefc1 FM |
234 | if (cb->errors) |
235 | mapping_set_error(inode->i_mapping, -EIO); | |
236 | ||
d397712b | 237 | while (nr_pages > 0) { |
c8b97818 | 238 | ret = find_get_pages_contig(inode->i_mapping, index, |
5b050f04 CM |
239 | min_t(unsigned long, |
240 | nr_pages, ARRAY_SIZE(pages)), pages); | |
c8b97818 CM |
241 | if (ret == 0) { |
242 | nr_pages -= 1; | |
243 | index += 1; | |
244 | continue; | |
245 | } | |
246 | for (i = 0; i < ret; i++) { | |
7bdcefc1 FM |
247 | if (cb->errors) |
248 | SetPageError(pages[i]); | |
c8b97818 | 249 | end_page_writeback(pages[i]); |
09cbfeaf | 250 | put_page(pages[i]); |
c8b97818 CM |
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 | */ | |
4246a0b6 | 266 | static void end_compressed_bio_write(struct bio *bio) |
c8b97818 CM |
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 | ||
4246a0b6 | 274 | if (bio->bi_error) |
c8b97818 CM |
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, | |
7bdcefc1 | 292 | NULL, |
4246a0b6 | 293 | bio->bi_error ? 0 : 1); |
70b99e69 | 294 | cb->compressed_pages[0]->mapping = NULL; |
c8b97818 | 295 | |
7bdcefc1 | 296 | end_compressed_writeback(inode, cb); |
c8b97818 CM |
297 | /* note, our inode could be gone now */ |
298 | ||
299 | /* | |
300 | * release the compressed pages, these came from alloc_page and | |
301 | * are not attached to the inode at all | |
302 | */ | |
303 | index = 0; | |
304 | for (index = 0; index < cb->nr_pages; index++) { | |
305 | page = cb->compressed_pages[index]; | |
306 | page->mapping = NULL; | |
09cbfeaf | 307 | put_page(page); |
c8b97818 CM |
308 | } |
309 | ||
310 | /* finally free the cb struct */ | |
311 | kfree(cb->compressed_pages); | |
312 | kfree(cb); | |
313 | out: | |
314 | bio_put(bio); | |
315 | } | |
316 | ||
317 | /* | |
318 | * worker function to build and submit bios for previously compressed pages. | |
319 | * The corresponding pages in the inode should be marked for writeback | |
320 | * and the compressed pages should have a reference on them for dropping | |
321 | * when the IO is complete. | |
322 | * | |
323 | * This also checksums the file bytes and gets things ready for | |
324 | * the end io hooks. | |
325 | */ | |
326 | int btrfs_submit_compressed_write(struct inode *inode, u64 start, | |
327 | unsigned long len, u64 disk_start, | |
328 | unsigned long compressed_len, | |
329 | struct page **compressed_pages, | |
330 | unsigned long nr_pages) | |
331 | { | |
332 | struct bio *bio = NULL; | |
333 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
334 | struct compressed_bio *cb; | |
335 | unsigned long bytes_left; | |
336 | struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; | |
306e16ce | 337 | int pg_index = 0; |
c8b97818 CM |
338 | struct page *page; |
339 | u64 first_byte = disk_start; | |
340 | struct block_device *bdev; | |
341 | int ret; | |
e55179b3 | 342 | int skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM; |
c8b97818 | 343 | |
09cbfeaf | 344 | WARN_ON(start & ((u64)PAGE_SIZE - 1)); |
d20f7043 | 345 | cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); |
dac97e51 YS |
346 | if (!cb) |
347 | return -ENOMEM; | |
c8b97818 CM |
348 | atomic_set(&cb->pending_bios, 0); |
349 | cb->errors = 0; | |
350 | cb->inode = inode; | |
351 | cb->start = start; | |
352 | cb->len = len; | |
d20f7043 | 353 | cb->mirror_num = 0; |
c8b97818 CM |
354 | cb->compressed_pages = compressed_pages; |
355 | cb->compressed_len = compressed_len; | |
356 | cb->orig_bio = NULL; | |
357 | cb->nr_pages = nr_pages; | |
358 | ||
359 | bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; | |
360 | ||
c8b97818 | 361 | bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); |
67871254 | 362 | if (!bio) { |
dac97e51 YS |
363 | kfree(cb); |
364 | return -ENOMEM; | |
365 | } | |
37226b21 | 366 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
c8b97818 CM |
367 | bio->bi_private = cb; |
368 | bio->bi_end_io = end_compressed_bio_write; | |
369 | atomic_inc(&cb->pending_bios); | |
370 | ||
371 | /* create and submit bios for the compressed pages */ | |
372 | bytes_left = compressed_len; | |
306e16ce DS |
373 | for (pg_index = 0; pg_index < cb->nr_pages; pg_index++) { |
374 | page = compressed_pages[pg_index]; | |
c8b97818 | 375 | page->mapping = inode->i_mapping; |
4f024f37 | 376 | if (bio->bi_iter.bi_size) |
81a75f67 | 377 | ret = io_tree->ops->merge_bio_hook(page, 0, |
09cbfeaf | 378 | PAGE_SIZE, |
c8b97818 CM |
379 | bio, 0); |
380 | else | |
381 | ret = 0; | |
382 | ||
70b99e69 | 383 | page->mapping = NULL; |
09cbfeaf KS |
384 | if (ret || bio_add_page(bio, page, PAGE_SIZE, 0) < |
385 | PAGE_SIZE) { | |
c8b97818 CM |
386 | bio_get(bio); |
387 | ||
af09abfe CM |
388 | /* |
389 | * inc the count before we submit the bio so | |
390 | * we know the end IO handler won't happen before | |
391 | * we inc the count. Otherwise, the cb might get | |
392 | * freed before we're done setting it up | |
393 | */ | |
394 | atomic_inc(&cb->pending_bios); | |
bfebd8b5 DS |
395 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, |
396 | BTRFS_WQ_ENDIO_DATA); | |
79787eaa | 397 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 398 | |
e55179b3 LZ |
399 | if (!skip_sum) { |
400 | ret = btrfs_csum_one_bio(root, inode, bio, | |
401 | start, 1); | |
79787eaa | 402 | BUG_ON(ret); /* -ENOMEM */ |
e55179b3 | 403 | } |
d20f7043 | 404 | |
81a75f67 | 405 | ret = btrfs_map_bio(root, bio, 0, 1); |
79787eaa | 406 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
407 | |
408 | bio_put(bio); | |
409 | ||
410 | bio = compressed_bio_alloc(bdev, first_byte, GFP_NOFS); | |
e627ee7b | 411 | BUG_ON(!bio); |
37226b21 | 412 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
c8b97818 CM |
413 | bio->bi_private = cb; |
414 | bio->bi_end_io = end_compressed_bio_write; | |
09cbfeaf | 415 | bio_add_page(bio, page, PAGE_SIZE, 0); |
c8b97818 | 416 | } |
09cbfeaf | 417 | if (bytes_left < PAGE_SIZE) { |
efe120a0 FH |
418 | btrfs_info(BTRFS_I(inode)->root->fs_info, |
419 | "bytes left %lu compress len %lu nr %lu", | |
cfbc246e CM |
420 | bytes_left, cb->compressed_len, cb->nr_pages); |
421 | } | |
09cbfeaf KS |
422 | bytes_left -= PAGE_SIZE; |
423 | first_byte += PAGE_SIZE; | |
771ed689 | 424 | cond_resched(); |
c8b97818 CM |
425 | } |
426 | bio_get(bio); | |
427 | ||
bfebd8b5 | 428 | ret = btrfs_bio_wq_end_io(root->fs_info, bio, BTRFS_WQ_ENDIO_DATA); |
79787eaa | 429 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 430 | |
e55179b3 LZ |
431 | if (!skip_sum) { |
432 | ret = btrfs_csum_one_bio(root, inode, bio, start, 1); | |
79787eaa | 433 | BUG_ON(ret); /* -ENOMEM */ |
e55179b3 | 434 | } |
d20f7043 | 435 | |
81a75f67 | 436 | ret = btrfs_map_bio(root, bio, 0, 1); |
79787eaa | 437 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 CM |
438 | |
439 | bio_put(bio); | |
440 | return 0; | |
441 | } | |
442 | ||
771ed689 CM |
443 | static noinline int add_ra_bio_pages(struct inode *inode, |
444 | u64 compressed_end, | |
445 | struct compressed_bio *cb) | |
446 | { | |
447 | unsigned long end_index; | |
306e16ce | 448 | unsigned long pg_index; |
771ed689 CM |
449 | u64 last_offset; |
450 | u64 isize = i_size_read(inode); | |
451 | int ret; | |
452 | struct page *page; | |
453 | unsigned long nr_pages = 0; | |
454 | struct extent_map *em; | |
455 | struct address_space *mapping = inode->i_mapping; | |
771ed689 CM |
456 | struct extent_map_tree *em_tree; |
457 | struct extent_io_tree *tree; | |
458 | u64 end; | |
459 | int misses = 0; | |
460 | ||
461 | page = cb->orig_bio->bi_io_vec[cb->orig_bio->bi_vcnt - 1].bv_page; | |
09cbfeaf | 462 | last_offset = (page_offset(page) + PAGE_SIZE); |
771ed689 CM |
463 | em_tree = &BTRFS_I(inode)->extent_tree; |
464 | tree = &BTRFS_I(inode)->io_tree; | |
465 | ||
466 | if (isize == 0) | |
467 | return 0; | |
468 | ||
09cbfeaf | 469 | end_index = (i_size_read(inode) - 1) >> PAGE_SHIFT; |
771ed689 | 470 | |
d397712b | 471 | while (last_offset < compressed_end) { |
09cbfeaf | 472 | pg_index = last_offset >> PAGE_SHIFT; |
771ed689 | 473 | |
306e16ce | 474 | if (pg_index > end_index) |
771ed689 CM |
475 | break; |
476 | ||
477 | rcu_read_lock(); | |
306e16ce | 478 | page = radix_tree_lookup(&mapping->page_tree, pg_index); |
771ed689 | 479 | rcu_read_unlock(); |
0cd6144a | 480 | if (page && !radix_tree_exceptional_entry(page)) { |
771ed689 CM |
481 | misses++; |
482 | if (misses > 4) | |
483 | break; | |
484 | goto next; | |
485 | } | |
486 | ||
c62d2555 MH |
487 | page = __page_cache_alloc(mapping_gfp_constraint(mapping, |
488 | ~__GFP_FS)); | |
771ed689 CM |
489 | if (!page) |
490 | break; | |
491 | ||
c62d2555 | 492 | if (add_to_page_cache_lru(page, mapping, pg_index, GFP_NOFS)) { |
09cbfeaf | 493 | put_page(page); |
771ed689 CM |
494 | goto next; |
495 | } | |
496 | ||
09cbfeaf | 497 | end = last_offset + PAGE_SIZE - 1; |
771ed689 CM |
498 | /* |
499 | * at this point, we have a locked page in the page cache | |
500 | * for these bytes in the file. But, we have to make | |
501 | * sure they map to this compressed extent on disk. | |
502 | */ | |
503 | set_page_extent_mapped(page); | |
d0082371 | 504 | lock_extent(tree, last_offset, end); |
890871be | 505 | read_lock(&em_tree->lock); |
771ed689 | 506 | em = lookup_extent_mapping(em_tree, last_offset, |
09cbfeaf | 507 | PAGE_SIZE); |
890871be | 508 | read_unlock(&em_tree->lock); |
771ed689 CM |
509 | |
510 | if (!em || last_offset < em->start || | |
09cbfeaf | 511 | (last_offset + PAGE_SIZE > extent_map_end(em)) || |
4f024f37 | 512 | (em->block_start >> 9) != cb->orig_bio->bi_iter.bi_sector) { |
771ed689 | 513 | free_extent_map(em); |
d0082371 | 514 | unlock_extent(tree, last_offset, end); |
771ed689 | 515 | unlock_page(page); |
09cbfeaf | 516 | put_page(page); |
771ed689 CM |
517 | break; |
518 | } | |
519 | free_extent_map(em); | |
520 | ||
521 | if (page->index == end_index) { | |
522 | char *userpage; | |
09cbfeaf | 523 | size_t zero_offset = isize & (PAGE_SIZE - 1); |
771ed689 CM |
524 | |
525 | if (zero_offset) { | |
526 | int zeros; | |
09cbfeaf | 527 | zeros = PAGE_SIZE - zero_offset; |
7ac687d9 | 528 | userpage = kmap_atomic(page); |
771ed689 CM |
529 | memset(userpage + zero_offset, 0, zeros); |
530 | flush_dcache_page(page); | |
7ac687d9 | 531 | kunmap_atomic(userpage); |
771ed689 CM |
532 | } |
533 | } | |
534 | ||
535 | ret = bio_add_page(cb->orig_bio, page, | |
09cbfeaf | 536 | PAGE_SIZE, 0); |
771ed689 | 537 | |
09cbfeaf | 538 | if (ret == PAGE_SIZE) { |
771ed689 | 539 | nr_pages++; |
09cbfeaf | 540 | put_page(page); |
771ed689 | 541 | } else { |
d0082371 | 542 | unlock_extent(tree, last_offset, end); |
771ed689 | 543 | unlock_page(page); |
09cbfeaf | 544 | put_page(page); |
771ed689 CM |
545 | break; |
546 | } | |
547 | next: | |
09cbfeaf | 548 | last_offset += PAGE_SIZE; |
771ed689 | 549 | } |
771ed689 CM |
550 | return 0; |
551 | } | |
552 | ||
c8b97818 CM |
553 | /* |
554 | * for a compressed read, the bio we get passed has all the inode pages | |
555 | * in it. We don't actually do IO on those pages but allocate new ones | |
556 | * to hold the compressed pages on disk. | |
557 | * | |
4f024f37 | 558 | * bio->bi_iter.bi_sector points to the compressed extent on disk |
c8b97818 CM |
559 | * bio->bi_io_vec points to all of the inode pages |
560 | * bio->bi_vcnt is a count of pages | |
561 | * | |
562 | * After the compressed pages are read, we copy the bytes into the | |
563 | * bio we were passed and then call the bio end_io calls | |
564 | */ | |
565 | int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio, | |
566 | int mirror_num, unsigned long bio_flags) | |
567 | { | |
568 | struct extent_io_tree *tree; | |
569 | struct extent_map_tree *em_tree; | |
570 | struct compressed_bio *cb; | |
571 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
09cbfeaf | 572 | unsigned long uncompressed_len = bio->bi_vcnt * PAGE_SIZE; |
c8b97818 CM |
573 | unsigned long compressed_len; |
574 | unsigned long nr_pages; | |
306e16ce | 575 | unsigned long pg_index; |
c8b97818 CM |
576 | struct page *page; |
577 | struct block_device *bdev; | |
578 | struct bio *comp_bio; | |
4f024f37 | 579 | u64 cur_disk_byte = (u64)bio->bi_iter.bi_sector << 9; |
e04ca626 CM |
580 | u64 em_len; |
581 | u64 em_start; | |
c8b97818 | 582 | struct extent_map *em; |
6b82ce8d | 583 | int ret = -ENOMEM; |
15e3004a | 584 | int faili = 0; |
d20f7043 | 585 | u32 *sums; |
c8b97818 CM |
586 | |
587 | tree = &BTRFS_I(inode)->io_tree; | |
588 | em_tree = &BTRFS_I(inode)->extent_tree; | |
589 | ||
590 | /* we need the actual starting offset of this extent in the file */ | |
890871be | 591 | read_lock(&em_tree->lock); |
c8b97818 CM |
592 | em = lookup_extent_mapping(em_tree, |
593 | page_offset(bio->bi_io_vec->bv_page), | |
09cbfeaf | 594 | PAGE_SIZE); |
890871be | 595 | read_unlock(&em_tree->lock); |
285190d9 TI |
596 | if (!em) |
597 | return -EIO; | |
c8b97818 | 598 | |
d20f7043 CM |
599 | compressed_len = em->block_len; |
600 | cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS); | |
6b82ce8d | 601 | if (!cb) |
602 | goto out; | |
603 | ||
c8b97818 CM |
604 | atomic_set(&cb->pending_bios, 0); |
605 | cb->errors = 0; | |
606 | cb->inode = inode; | |
d20f7043 CM |
607 | cb->mirror_num = mirror_num; |
608 | sums = &cb->sums; | |
c8b97818 | 609 | |
ff5b7ee3 | 610 | cb->start = em->orig_start; |
e04ca626 CM |
611 | em_len = em->len; |
612 | em_start = em->start; | |
d20f7043 | 613 | |
c8b97818 | 614 | free_extent_map(em); |
e04ca626 | 615 | em = NULL; |
c8b97818 CM |
616 | |
617 | cb->len = uncompressed_len; | |
618 | cb->compressed_len = compressed_len; | |
261507a0 | 619 | cb->compress_type = extent_compress_type(bio_flags); |
c8b97818 CM |
620 | cb->orig_bio = bio; |
621 | ||
09cbfeaf | 622 | nr_pages = DIV_ROUND_UP(compressed_len, PAGE_SIZE); |
31e818fe | 623 | cb->compressed_pages = kcalloc(nr_pages, sizeof(struct page *), |
c8b97818 | 624 | GFP_NOFS); |
6b82ce8d | 625 | if (!cb->compressed_pages) |
626 | goto fail1; | |
627 | ||
c8b97818 CM |
628 | bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; |
629 | ||
306e16ce DS |
630 | for (pg_index = 0; pg_index < nr_pages; pg_index++) { |
631 | cb->compressed_pages[pg_index] = alloc_page(GFP_NOFS | | |
c8b97818 | 632 | __GFP_HIGHMEM); |
15e3004a JB |
633 | if (!cb->compressed_pages[pg_index]) { |
634 | faili = pg_index - 1; | |
635 | ret = -ENOMEM; | |
6b82ce8d | 636 | goto fail2; |
15e3004a | 637 | } |
c8b97818 | 638 | } |
15e3004a | 639 | faili = nr_pages - 1; |
c8b97818 CM |
640 | cb->nr_pages = nr_pages; |
641 | ||
7f042a83 | 642 | add_ra_bio_pages(inode, em_start + em_len, cb); |
771ed689 | 643 | |
771ed689 | 644 | /* include any pages we added in add_ra-bio_pages */ |
09cbfeaf | 645 | uncompressed_len = bio->bi_vcnt * PAGE_SIZE; |
771ed689 CM |
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; | |
37226b21 | 651 | bio_set_op_attrs (comp_bio, REQ_OP_READ, 0); |
c8b97818 CM |
652 | comp_bio->bi_private = cb; |
653 | comp_bio->bi_end_io = end_compressed_bio_read; | |
654 | atomic_inc(&cb->pending_bios); | |
655 | ||
306e16ce DS |
656 | for (pg_index = 0; pg_index < nr_pages; pg_index++) { |
657 | page = cb->compressed_pages[pg_index]; | |
c8b97818 | 658 | page->mapping = inode->i_mapping; |
09cbfeaf | 659 | page->index = em_start >> PAGE_SHIFT; |
d20f7043 | 660 | |
4f024f37 | 661 | if (comp_bio->bi_iter.bi_size) |
81a75f67 | 662 | ret = tree->ops->merge_bio_hook(page, 0, |
09cbfeaf | 663 | PAGE_SIZE, |
c8b97818 CM |
664 | comp_bio, 0); |
665 | else | |
666 | ret = 0; | |
667 | ||
70b99e69 | 668 | page->mapping = NULL; |
09cbfeaf KS |
669 | if (ret || bio_add_page(comp_bio, page, PAGE_SIZE, 0) < |
670 | PAGE_SIZE) { | |
c8b97818 CM |
671 | bio_get(comp_bio); |
672 | ||
bfebd8b5 DS |
673 | ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, |
674 | BTRFS_WQ_ENDIO_DATA); | |
79787eaa | 675 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 676 | |
af09abfe CM |
677 | /* |
678 | * inc the count before we submit the bio so | |
679 | * we know the end IO handler won't happen before | |
680 | * we inc the count. Otherwise, the cb might get | |
681 | * freed before we're done setting it up | |
682 | */ | |
683 | atomic_inc(&cb->pending_bios); | |
684 | ||
6cbff00f | 685 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { |
c2db1073 TI |
686 | ret = btrfs_lookup_bio_sums(root, inode, |
687 | comp_bio, sums); | |
79787eaa | 688 | BUG_ON(ret); /* -ENOMEM */ |
d20f7043 | 689 | } |
ed6078f7 DS |
690 | sums += DIV_ROUND_UP(comp_bio->bi_iter.bi_size, |
691 | root->sectorsize); | |
d20f7043 | 692 | |
81a75f67 | 693 | ret = btrfs_map_bio(root, comp_bio, mirror_num, 0); |
4246a0b6 CH |
694 | if (ret) { |
695 | bio->bi_error = ret; | |
696 | bio_endio(comp_bio); | |
697 | } | |
c8b97818 CM |
698 | |
699 | bio_put(comp_bio); | |
700 | ||
701 | comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, | |
702 | GFP_NOFS); | |
e627ee7b | 703 | BUG_ON(!comp_bio); |
37226b21 | 704 | bio_set_op_attrs(comp_bio, REQ_OP_READ, 0); |
771ed689 CM |
705 | comp_bio->bi_private = cb; |
706 | comp_bio->bi_end_io = end_compressed_bio_read; | |
707 | ||
09cbfeaf | 708 | bio_add_page(comp_bio, page, PAGE_SIZE, 0); |
c8b97818 | 709 | } |
09cbfeaf | 710 | cur_disk_byte += PAGE_SIZE; |
c8b97818 CM |
711 | } |
712 | bio_get(comp_bio); | |
713 | ||
bfebd8b5 DS |
714 | ret = btrfs_bio_wq_end_io(root->fs_info, comp_bio, |
715 | BTRFS_WQ_ENDIO_DATA); | |
79787eaa | 716 | BUG_ON(ret); /* -ENOMEM */ |
c8b97818 | 717 | |
c2db1073 TI |
718 | if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) { |
719 | ret = btrfs_lookup_bio_sums(root, inode, comp_bio, sums); | |
79787eaa | 720 | BUG_ON(ret); /* -ENOMEM */ |
c2db1073 | 721 | } |
d20f7043 | 722 | |
81a75f67 | 723 | ret = btrfs_map_bio(root, comp_bio, mirror_num, 0); |
4246a0b6 CH |
724 | if (ret) { |
725 | bio->bi_error = ret; | |
726 | bio_endio(comp_bio); | |
727 | } | |
c8b97818 CM |
728 | |
729 | bio_put(comp_bio); | |
730 | return 0; | |
6b82ce8d | 731 | |
732 | fail2: | |
15e3004a JB |
733 | while (faili >= 0) { |
734 | __free_page(cb->compressed_pages[faili]); | |
735 | faili--; | |
736 | } | |
6b82ce8d | 737 | |
738 | kfree(cb->compressed_pages); | |
739 | fail1: | |
740 | kfree(cb); | |
741 | out: | |
742 | free_extent_map(em); | |
743 | return ret; | |
c8b97818 | 744 | } |
261507a0 | 745 | |
d9187649 BL |
746 | static struct { |
747 | struct list_head idle_ws; | |
748 | spinlock_t ws_lock; | |
6ac10a6a DS |
749 | /* Number of free workspaces */ |
750 | int free_ws; | |
751 | /* Total number of allocated workspaces */ | |
752 | atomic_t total_ws; | |
753 | /* Waiters for a free workspace */ | |
d9187649 BL |
754 | wait_queue_head_t ws_wait; |
755 | } btrfs_comp_ws[BTRFS_COMPRESS_TYPES]; | |
261507a0 | 756 | |
e8c9f186 | 757 | static const struct btrfs_compress_op * const btrfs_compress_op[] = { |
261507a0 | 758 | &btrfs_zlib_compress, |
a6fa6fae | 759 | &btrfs_lzo_compress, |
261507a0 LZ |
760 | }; |
761 | ||
143bede5 | 762 | void __init btrfs_init_compress(void) |
261507a0 LZ |
763 | { |
764 | int i; | |
765 | ||
766 | for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) { | |
f77dd0d6 DS |
767 | struct list_head *workspace; |
768 | ||
d9187649 BL |
769 | INIT_LIST_HEAD(&btrfs_comp_ws[i].idle_ws); |
770 | spin_lock_init(&btrfs_comp_ws[i].ws_lock); | |
6ac10a6a | 771 | atomic_set(&btrfs_comp_ws[i].total_ws, 0); |
d9187649 | 772 | init_waitqueue_head(&btrfs_comp_ws[i].ws_wait); |
f77dd0d6 DS |
773 | |
774 | /* | |
775 | * Preallocate one workspace for each compression type so | |
776 | * we can guarantee forward progress in the worst case | |
777 | */ | |
778 | workspace = btrfs_compress_op[i]->alloc_workspace(); | |
779 | if (IS_ERR(workspace)) { | |
780 | printk(KERN_WARNING | |
781 | "BTRFS: cannot preallocate compression workspace, will try later"); | |
782 | } else { | |
783 | atomic_set(&btrfs_comp_ws[i].total_ws, 1); | |
784 | btrfs_comp_ws[i].free_ws = 1; | |
785 | list_add(workspace, &btrfs_comp_ws[i].idle_ws); | |
786 | } | |
261507a0 | 787 | } |
261507a0 LZ |
788 | } |
789 | ||
790 | /* | |
e721e49d DS |
791 | * This finds an available workspace or allocates a new one. |
792 | * If it's not possible to allocate a new one, waits until there's one. | |
793 | * Preallocation makes a forward progress guarantees and we do not return | |
794 | * errors. | |
261507a0 LZ |
795 | */ |
796 | static struct list_head *find_workspace(int type) | |
797 | { | |
798 | struct list_head *workspace; | |
799 | int cpus = num_online_cpus(); | |
800 | int idx = type - 1; | |
801 | ||
d9187649 BL |
802 | struct list_head *idle_ws = &btrfs_comp_ws[idx].idle_ws; |
803 | spinlock_t *ws_lock = &btrfs_comp_ws[idx].ws_lock; | |
6ac10a6a | 804 | atomic_t *total_ws = &btrfs_comp_ws[idx].total_ws; |
d9187649 | 805 | wait_queue_head_t *ws_wait = &btrfs_comp_ws[idx].ws_wait; |
6ac10a6a | 806 | int *free_ws = &btrfs_comp_ws[idx].free_ws; |
261507a0 | 807 | again: |
d9187649 BL |
808 | spin_lock(ws_lock); |
809 | if (!list_empty(idle_ws)) { | |
810 | workspace = idle_ws->next; | |
261507a0 | 811 | list_del(workspace); |
6ac10a6a | 812 | (*free_ws)--; |
d9187649 | 813 | spin_unlock(ws_lock); |
261507a0 LZ |
814 | return workspace; |
815 | ||
816 | } | |
6ac10a6a | 817 | if (atomic_read(total_ws) > cpus) { |
261507a0 LZ |
818 | DEFINE_WAIT(wait); |
819 | ||
d9187649 BL |
820 | spin_unlock(ws_lock); |
821 | prepare_to_wait(ws_wait, &wait, TASK_UNINTERRUPTIBLE); | |
6ac10a6a | 822 | if (atomic_read(total_ws) > cpus && !*free_ws) |
261507a0 | 823 | schedule(); |
d9187649 | 824 | finish_wait(ws_wait, &wait); |
261507a0 LZ |
825 | goto again; |
826 | } | |
6ac10a6a | 827 | atomic_inc(total_ws); |
d9187649 | 828 | spin_unlock(ws_lock); |
261507a0 LZ |
829 | |
830 | workspace = btrfs_compress_op[idx]->alloc_workspace(); | |
831 | if (IS_ERR(workspace)) { | |
6ac10a6a | 832 | atomic_dec(total_ws); |
d9187649 | 833 | wake_up(ws_wait); |
e721e49d DS |
834 | |
835 | /* | |
836 | * Do not return the error but go back to waiting. There's a | |
837 | * workspace preallocated for each type and the compression | |
838 | * time is bounded so we get to a workspace eventually. This | |
839 | * makes our caller's life easier. | |
52356716 DS |
840 | * |
841 | * To prevent silent and low-probability deadlocks (when the | |
842 | * initial preallocation fails), check if there are any | |
843 | * workspaces at all. | |
e721e49d | 844 | */ |
52356716 DS |
845 | if (atomic_read(total_ws) == 0) { |
846 | static DEFINE_RATELIMIT_STATE(_rs, | |
847 | /* once per minute */ 60 * HZ, | |
848 | /* no burst */ 1); | |
849 | ||
850 | if (__ratelimit(&_rs)) { | |
851 | printk(KERN_WARNING | |
852 | "no compression workspaces, low memory, retrying"); | |
853 | } | |
854 | } | |
e721e49d | 855 | goto again; |
261507a0 LZ |
856 | } |
857 | return workspace; | |
858 | } | |
859 | ||
860 | /* | |
861 | * put a workspace struct back on the list or free it if we have enough | |
862 | * idle ones sitting around | |
863 | */ | |
864 | static void free_workspace(int type, struct list_head *workspace) | |
865 | { | |
866 | int idx = type - 1; | |
d9187649 BL |
867 | struct list_head *idle_ws = &btrfs_comp_ws[idx].idle_ws; |
868 | spinlock_t *ws_lock = &btrfs_comp_ws[idx].ws_lock; | |
6ac10a6a | 869 | atomic_t *total_ws = &btrfs_comp_ws[idx].total_ws; |
d9187649 | 870 | wait_queue_head_t *ws_wait = &btrfs_comp_ws[idx].ws_wait; |
6ac10a6a | 871 | int *free_ws = &btrfs_comp_ws[idx].free_ws; |
d9187649 BL |
872 | |
873 | spin_lock(ws_lock); | |
6ac10a6a | 874 | if (*free_ws < num_online_cpus()) { |
d9187649 | 875 | list_add(workspace, idle_ws); |
6ac10a6a | 876 | (*free_ws)++; |
d9187649 | 877 | spin_unlock(ws_lock); |
261507a0 LZ |
878 | goto wake; |
879 | } | |
d9187649 | 880 | spin_unlock(ws_lock); |
261507a0 LZ |
881 | |
882 | btrfs_compress_op[idx]->free_workspace(workspace); | |
6ac10a6a | 883 | atomic_dec(total_ws); |
261507a0 | 884 | wake: |
a83342aa DS |
885 | /* |
886 | * Make sure counter is updated before we wake up waiters. | |
887 | */ | |
66657b31 | 888 | smp_mb(); |
d9187649 BL |
889 | if (waitqueue_active(ws_wait)) |
890 | wake_up(ws_wait); | |
261507a0 LZ |
891 | } |
892 | ||
893 | /* | |
894 | * cleanup function for module exit | |
895 | */ | |
896 | static void free_workspaces(void) | |
897 | { | |
898 | struct list_head *workspace; | |
899 | int i; | |
900 | ||
901 | for (i = 0; i < BTRFS_COMPRESS_TYPES; i++) { | |
d9187649 BL |
902 | while (!list_empty(&btrfs_comp_ws[i].idle_ws)) { |
903 | workspace = btrfs_comp_ws[i].idle_ws.next; | |
261507a0 LZ |
904 | list_del(workspace); |
905 | btrfs_compress_op[i]->free_workspace(workspace); | |
6ac10a6a | 906 | atomic_dec(&btrfs_comp_ws[i].total_ws); |
261507a0 LZ |
907 | } |
908 | } | |
909 | } | |
910 | ||
911 | /* | |
912 | * given an address space and start/len, compress the bytes. | |
913 | * | |
914 | * pages are allocated to hold the compressed result and stored | |
915 | * in 'pages' | |
916 | * | |
917 | * out_pages is used to return the number of pages allocated. There | |
918 | * may be pages allocated even if we return an error | |
919 | * | |
920 | * total_in is used to return the number of bytes actually read. It | |
921 | * may be smaller then len if we had to exit early because we | |
922 | * ran out of room in the pages array or because we cross the | |
923 | * max_out threshold. | |
924 | * | |
925 | * total_out is used to return the total number of compressed bytes | |
926 | * | |
927 | * max_out tells us the max number of bytes that we're allowed to | |
928 | * stuff into pages | |
929 | */ | |
930 | int btrfs_compress_pages(int type, struct address_space *mapping, | |
931 | u64 start, unsigned long len, | |
932 | struct page **pages, | |
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, | |
945 | start, len, pages, | |
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 | * | |
958 | * bvec is a bio_vec of pages from the file that we want to decompress into | |
959 | * | |
960 | * vcnt is the count of pages in the biovec | |
961 | * | |
962 | * srclen is the number of bytes in pages_in | |
963 | * | |
964 | * The basic idea is that we have a bio that was created by readpages. | |
965 | * The pages in the bio are for the uncompressed data, and they may not | |
966 | * be contiguous. They all correspond to the range of bytes covered by | |
967 | * the compressed extent. | |
968 | */ | |
48a3b636 ES |
969 | static int btrfs_decompress_biovec(int type, struct page **pages_in, |
970 | u64 disk_start, struct bio_vec *bvec, | |
971 | int vcnt, size_t srclen) | |
261507a0 LZ |
972 | { |
973 | struct list_head *workspace; | |
974 | int ret; | |
975 | ||
976 | workspace = find_workspace(type); | |
261507a0 LZ |
977 | |
978 | ret = btrfs_compress_op[type-1]->decompress_biovec(workspace, pages_in, | |
979 | disk_start, | |
980 | bvec, vcnt, srclen); | |
981 | free_workspace(type, workspace); | |
982 | return ret; | |
983 | } | |
984 | ||
985 | /* | |
986 | * a less complex decompression routine. Our compressed data fits in a | |
987 | * single page, and we want to read a single page out of it. | |
988 | * start_byte tells us the offset into the compressed data we're interested in | |
989 | */ | |
990 | int btrfs_decompress(int type, unsigned char *data_in, struct page *dest_page, | |
991 | unsigned long start_byte, size_t srclen, size_t destlen) | |
992 | { | |
993 | struct list_head *workspace; | |
994 | int ret; | |
995 | ||
996 | workspace = find_workspace(type); | |
261507a0 LZ |
997 | |
998 | ret = btrfs_compress_op[type-1]->decompress(workspace, data_in, | |
999 | dest_page, start_byte, | |
1000 | srclen, destlen); | |
1001 | ||
1002 | free_workspace(type, workspace); | |
1003 | return ret; | |
1004 | } | |
1005 | ||
8e4eef7a | 1006 | void btrfs_exit_compress(void) |
261507a0 LZ |
1007 | { |
1008 | free_workspaces(); | |
1009 | } | |
3a39c18d LZ |
1010 | |
1011 | /* | |
1012 | * Copy uncompressed data from working buffer to pages. | |
1013 | * | |
1014 | * buf_start is the byte offset we're of the start of our workspace buffer. | |
1015 | * | |
1016 | * total_out is the last byte of the buffer | |
1017 | */ | |
1018 | int btrfs_decompress_buf2page(char *buf, unsigned long buf_start, | |
1019 | unsigned long total_out, u64 disk_start, | |
1020 | struct bio_vec *bvec, int vcnt, | |
306e16ce | 1021 | unsigned long *pg_index, |
3a39c18d LZ |
1022 | unsigned long *pg_offset) |
1023 | { | |
1024 | unsigned long buf_offset; | |
1025 | unsigned long current_buf_start; | |
1026 | unsigned long start_byte; | |
1027 | unsigned long working_bytes = total_out - buf_start; | |
1028 | unsigned long bytes; | |
1029 | char *kaddr; | |
306e16ce | 1030 | struct page *page_out = bvec[*pg_index].bv_page; |
3a39c18d LZ |
1031 | |
1032 | /* | |
1033 | * start byte is the first byte of the page we're currently | |
1034 | * copying into relative to the start of the compressed data. | |
1035 | */ | |
1036 | start_byte = page_offset(page_out) - disk_start; | |
1037 | ||
1038 | /* we haven't yet hit data corresponding to this page */ | |
1039 | if (total_out <= start_byte) | |
1040 | return 1; | |
1041 | ||
1042 | /* | |
1043 | * the start of the data we care about is offset into | |
1044 | * the middle of our working buffer | |
1045 | */ | |
1046 | if (total_out > start_byte && buf_start < start_byte) { | |
1047 | buf_offset = start_byte - buf_start; | |
1048 | working_bytes -= buf_offset; | |
1049 | } else { | |
1050 | buf_offset = 0; | |
1051 | } | |
1052 | current_buf_start = buf_start; | |
1053 | ||
1054 | /* copy bytes from the working buffer into the pages */ | |
1055 | while (working_bytes > 0) { | |
09cbfeaf KS |
1056 | bytes = min(PAGE_SIZE - *pg_offset, |
1057 | PAGE_SIZE - buf_offset); | |
3a39c18d | 1058 | bytes = min(bytes, working_bytes); |
7ac687d9 | 1059 | kaddr = kmap_atomic(page_out); |
3a39c18d | 1060 | memcpy(kaddr + *pg_offset, buf + buf_offset, bytes); |
7ac687d9 | 1061 | kunmap_atomic(kaddr); |
3a39c18d LZ |
1062 | flush_dcache_page(page_out); |
1063 | ||
1064 | *pg_offset += bytes; | |
1065 | buf_offset += bytes; | |
1066 | working_bytes -= bytes; | |
1067 | current_buf_start += bytes; | |
1068 | ||
1069 | /* check if we need to pick another page */ | |
09cbfeaf | 1070 | if (*pg_offset == PAGE_SIZE) { |
306e16ce DS |
1071 | (*pg_index)++; |
1072 | if (*pg_index >= vcnt) | |
3a39c18d LZ |
1073 | return 0; |
1074 | ||
306e16ce | 1075 | page_out = bvec[*pg_index].bv_page; |
3a39c18d LZ |
1076 | *pg_offset = 0; |
1077 | start_byte = page_offset(page_out) - disk_start; | |
1078 | ||
1079 | /* | |
1080 | * make sure our new page is covered by this | |
1081 | * working buffer | |
1082 | */ | |
1083 | if (total_out <= start_byte) | |
1084 | return 1; | |
1085 | ||
1086 | /* | |
1087 | * the next page in the biovec might not be adjacent | |
1088 | * to the last page, but it might still be found | |
1089 | * inside this working buffer. bump our offset pointer | |
1090 | */ | |
1091 | if (total_out > start_byte && | |
1092 | current_buf_start < start_byte) { | |
1093 | buf_offset = start_byte - buf_start; | |
1094 | working_bytes = total_out - start_byte; | |
1095 | current_buf_start = buf_start + buf_offset; | |
1096 | } | |
1097 | } | |
1098 | } | |
1099 | ||
1100 | return 1; | |
1101 | } | |
2f19cad9 CM |
1102 | |
1103 | /* | |
1104 | * When uncompressing data, we need to make sure and zero any parts of | |
1105 | * the biovec that were not filled in by the decompression code. pg_index | |
1106 | * and pg_offset indicate the last page and the last offset of that page | |
1107 | * that have been filled in. This will zero everything remaining in the | |
1108 | * biovec. | |
1109 | */ | |
1110 | void btrfs_clear_biovec_end(struct bio_vec *bvec, int vcnt, | |
1111 | unsigned long pg_index, | |
1112 | unsigned long pg_offset) | |
1113 | { | |
1114 | while (pg_index < vcnt) { | |
1115 | struct page *page = bvec[pg_index].bv_page; | |
1116 | unsigned long off = bvec[pg_index].bv_offset; | |
1117 | unsigned long len = bvec[pg_index].bv_len; | |
1118 | ||
1119 | if (pg_offset < off) | |
1120 | pg_offset = off; | |
1121 | if (pg_offset < off + len) { | |
1122 | unsigned long bytes = off + len - pg_offset; | |
1123 | char *kaddr; | |
1124 | ||
1125 | kaddr = kmap_atomic(page); | |
1126 | memset(kaddr + pg_offset, 0, bytes); | |
1127 | kunmap_atomic(kaddr); | |
1128 | } | |
1129 | pg_index++; | |
1130 | pg_offset = 0; | |
1131 | } | |
1132 | } |