Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
0fe23479 | 2 | * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> |
1da177e4 LT |
3 | * |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License version 2 as | |
6 | * 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 | |
11 | * GNU General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public Licens | |
14 | * along with this program; if not, write to the Free Software | |
15 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111- | |
16 | * | |
17 | */ | |
18 | #include <linux/mm.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/bio.h> | |
21 | #include <linux/blkdev.h> | |
22 | #include <linux/slab.h> | |
23 | #include <linux/init.h> | |
24 | #include <linux/kernel.h> | |
25 | #include <linux/module.h> | |
26 | #include <linux/mempool.h> | |
27 | #include <linux/workqueue.h> | |
2056a782 | 28 | #include <linux/blktrace_api.h> |
f1970baf | 29 | #include <scsi/sg.h> /* for struct sg_iovec */ |
1da177e4 | 30 | |
e18b890b | 31 | static struct kmem_cache *bio_slab __read_mostly; |
1da177e4 | 32 | |
fa3536cc | 33 | mempool_t *bio_split_pool __read_mostly; |
1da177e4 | 34 | |
1da177e4 LT |
35 | /* |
36 | * if you change this list, also change bvec_alloc or things will | |
37 | * break badly! cannot be bigger than what you can fit into an | |
38 | * unsigned short | |
39 | */ | |
40 | ||
41 | #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } | |
6c036527 | 42 | static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { |
1da177e4 LT |
43 | BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), |
44 | }; | |
45 | #undef BV | |
46 | ||
1da177e4 LT |
47 | /* |
48 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
49 | * IO code that does not need private memory pools. | |
50 | */ | |
51d654e1 | 51 | struct bio_set *fs_bio_set; |
1da177e4 | 52 | |
7ba1ba12 MP |
53 | unsigned int bvec_nr_vecs(unsigned short idx) |
54 | { | |
55 | return bvec_slabs[idx].nr_vecs; | |
56 | } | |
57 | ||
51d654e1 | 58 | struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs) |
1da177e4 LT |
59 | { |
60 | struct bio_vec *bvl; | |
1da177e4 LT |
61 | |
62 | /* | |
63 | * see comment near bvec_array define! | |
64 | */ | |
65 | switch (nr) { | |
66 | case 1 : *idx = 0; break; | |
67 | case 2 ... 4: *idx = 1; break; | |
68 | case 5 ... 16: *idx = 2; break; | |
69 | case 17 ... 64: *idx = 3; break; | |
70 | case 65 ... 128: *idx = 4; break; | |
71 | case 129 ... BIO_MAX_PAGES: *idx = 5; break; | |
72 | default: | |
73 | return NULL; | |
74 | } | |
75 | /* | |
76 | * idx now points to the pool we want to allocate from | |
77 | */ | |
78 | ||
1da177e4 | 79 | bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask); |
bf02c082 AM |
80 | if (bvl) { |
81 | struct biovec_slab *bp = bvec_slabs + *idx; | |
82 | ||
1da177e4 | 83 | memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec)); |
bf02c082 | 84 | } |
1da177e4 LT |
85 | |
86 | return bvl; | |
87 | } | |
88 | ||
3676347a | 89 | void bio_free(struct bio *bio, struct bio_set *bio_set) |
1da177e4 | 90 | { |
992c5dda JA |
91 | if (bio->bi_io_vec) { |
92 | const int pool_idx = BIO_POOL_IDX(bio); | |
1da177e4 | 93 | |
992c5dda JA |
94 | BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS); |
95 | ||
96 | mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]); | |
97 | } | |
1da177e4 | 98 | |
7ba1ba12 MP |
99 | if (bio_integrity(bio)) |
100 | bio_integrity_free(bio, bio_set); | |
101 | ||
3676347a PO |
102 | mempool_free(bio, bio_set->bio_pool); |
103 | } | |
104 | ||
105 | /* | |
106 | * default destructor for a bio allocated with bio_alloc_bioset() | |
107 | */ | |
108 | static void bio_fs_destructor(struct bio *bio) | |
109 | { | |
110 | bio_free(bio, fs_bio_set); | |
1da177e4 LT |
111 | } |
112 | ||
858119e1 | 113 | void bio_init(struct bio *bio) |
1da177e4 | 114 | { |
2b94de55 | 115 | memset(bio, 0, sizeof(*bio)); |
1da177e4 | 116 | bio->bi_flags = 1 << BIO_UPTODATE; |
1da177e4 | 117 | atomic_set(&bio->bi_cnt, 1); |
1da177e4 LT |
118 | } |
119 | ||
120 | /** | |
121 | * bio_alloc_bioset - allocate a bio for I/O | |
122 | * @gfp_mask: the GFP_ mask given to the slab allocator | |
123 | * @nr_iovecs: number of iovecs to pre-allocate | |
67be2dd1 | 124 | * @bs: the bio_set to allocate from |
1da177e4 LT |
125 | * |
126 | * Description: | |
127 | * bio_alloc_bioset will first try it's on mempool to satisfy the allocation. | |
128 | * If %__GFP_WAIT is set then we will block on the internal pool waiting | |
129 | * for a &struct bio to become free. | |
130 | * | |
131 | * allocate bio and iovecs from the memory pools specified by the | |
132 | * bio_set structure. | |
133 | **/ | |
dd0fc66f | 134 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) |
1da177e4 LT |
135 | { |
136 | struct bio *bio = mempool_alloc(bs->bio_pool, gfp_mask); | |
137 | ||
138 | if (likely(bio)) { | |
139 | struct bio_vec *bvl = NULL; | |
140 | ||
141 | bio_init(bio); | |
142 | if (likely(nr_iovecs)) { | |
eeae1d48 | 143 | unsigned long uninitialized_var(idx); |
1da177e4 LT |
144 | |
145 | bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs); | |
146 | if (unlikely(!bvl)) { | |
147 | mempool_free(bio, bs->bio_pool); | |
148 | bio = NULL; | |
149 | goto out; | |
150 | } | |
151 | bio->bi_flags |= idx << BIO_POOL_OFFSET; | |
152 | bio->bi_max_vecs = bvec_slabs[idx].nr_vecs; | |
153 | } | |
154 | bio->bi_io_vec = bvl; | |
1da177e4 LT |
155 | } |
156 | out: | |
157 | return bio; | |
158 | } | |
159 | ||
dd0fc66f | 160 | struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs) |
1da177e4 | 161 | { |
3676347a PO |
162 | struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set); |
163 | ||
164 | if (bio) | |
165 | bio->bi_destructor = bio_fs_destructor; | |
166 | ||
167 | return bio; | |
1da177e4 LT |
168 | } |
169 | ||
170 | void zero_fill_bio(struct bio *bio) | |
171 | { | |
172 | unsigned long flags; | |
173 | struct bio_vec *bv; | |
174 | int i; | |
175 | ||
176 | bio_for_each_segment(bv, bio, i) { | |
177 | char *data = bvec_kmap_irq(bv, &flags); | |
178 | memset(data, 0, bv->bv_len); | |
179 | flush_dcache_page(bv->bv_page); | |
180 | bvec_kunmap_irq(data, &flags); | |
181 | } | |
182 | } | |
183 | EXPORT_SYMBOL(zero_fill_bio); | |
184 | ||
185 | /** | |
186 | * bio_put - release a reference to a bio | |
187 | * @bio: bio to release reference to | |
188 | * | |
189 | * Description: | |
190 | * Put a reference to a &struct bio, either one you have gotten with | |
191 | * bio_alloc or bio_get. The last put of a bio will free it. | |
192 | **/ | |
193 | void bio_put(struct bio *bio) | |
194 | { | |
195 | BIO_BUG_ON(!atomic_read(&bio->bi_cnt)); | |
196 | ||
197 | /* | |
198 | * last put frees it | |
199 | */ | |
200 | if (atomic_dec_and_test(&bio->bi_cnt)) { | |
201 | bio->bi_next = NULL; | |
202 | bio->bi_destructor(bio); | |
203 | } | |
204 | } | |
205 | ||
165125e1 | 206 | inline int bio_phys_segments(struct request_queue *q, struct bio *bio) |
1da177e4 LT |
207 | { |
208 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
209 | blk_recount_segments(q, bio); | |
210 | ||
211 | return bio->bi_phys_segments; | |
212 | } | |
213 | ||
165125e1 | 214 | inline int bio_hw_segments(struct request_queue *q, struct bio *bio) |
1da177e4 LT |
215 | { |
216 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
217 | blk_recount_segments(q, bio); | |
218 | ||
219 | return bio->bi_hw_segments; | |
220 | } | |
221 | ||
222 | /** | |
223 | * __bio_clone - clone a bio | |
224 | * @bio: destination bio | |
225 | * @bio_src: bio to clone | |
226 | * | |
227 | * Clone a &bio. Caller will own the returned bio, but not | |
228 | * the actual data it points to. Reference count of returned | |
229 | * bio will be one. | |
230 | */ | |
858119e1 | 231 | void __bio_clone(struct bio *bio, struct bio *bio_src) |
1da177e4 | 232 | { |
e525e153 AM |
233 | memcpy(bio->bi_io_vec, bio_src->bi_io_vec, |
234 | bio_src->bi_max_vecs * sizeof(struct bio_vec)); | |
1da177e4 | 235 | |
5d84070e JA |
236 | /* |
237 | * most users will be overriding ->bi_bdev with a new target, | |
238 | * so we don't set nor calculate new physical/hw segment counts here | |
239 | */ | |
1da177e4 LT |
240 | bio->bi_sector = bio_src->bi_sector; |
241 | bio->bi_bdev = bio_src->bi_bdev; | |
242 | bio->bi_flags |= 1 << BIO_CLONED; | |
243 | bio->bi_rw = bio_src->bi_rw; | |
1da177e4 LT |
244 | bio->bi_vcnt = bio_src->bi_vcnt; |
245 | bio->bi_size = bio_src->bi_size; | |
a5453be4 | 246 | bio->bi_idx = bio_src->bi_idx; |
1da177e4 LT |
247 | } |
248 | ||
249 | /** | |
250 | * bio_clone - clone a bio | |
251 | * @bio: bio to clone | |
252 | * @gfp_mask: allocation priority | |
253 | * | |
254 | * Like __bio_clone, only also allocates the returned bio | |
255 | */ | |
dd0fc66f | 256 | struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask) |
1da177e4 LT |
257 | { |
258 | struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set); | |
259 | ||
7ba1ba12 MP |
260 | if (!b) |
261 | return NULL; | |
262 | ||
263 | b->bi_destructor = bio_fs_destructor; | |
264 | __bio_clone(b, bio); | |
265 | ||
266 | if (bio_integrity(bio)) { | |
267 | int ret; | |
268 | ||
269 | ret = bio_integrity_clone(b, bio, fs_bio_set); | |
270 | ||
271 | if (ret < 0) | |
272 | return NULL; | |
3676347a | 273 | } |
1da177e4 LT |
274 | |
275 | return b; | |
276 | } | |
277 | ||
278 | /** | |
279 | * bio_get_nr_vecs - return approx number of vecs | |
280 | * @bdev: I/O target | |
281 | * | |
282 | * Return the approximate number of pages we can send to this target. | |
283 | * There's no guarantee that you will be able to fit this number of pages | |
284 | * into a bio, it does not account for dynamic restrictions that vary | |
285 | * on offset. | |
286 | */ | |
287 | int bio_get_nr_vecs(struct block_device *bdev) | |
288 | { | |
165125e1 | 289 | struct request_queue *q = bdev_get_queue(bdev); |
1da177e4 LT |
290 | int nr_pages; |
291 | ||
292 | nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
293 | if (nr_pages > q->max_phys_segments) | |
294 | nr_pages = q->max_phys_segments; | |
295 | if (nr_pages > q->max_hw_segments) | |
296 | nr_pages = q->max_hw_segments; | |
297 | ||
298 | return nr_pages; | |
299 | } | |
300 | ||
165125e1 | 301 | static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page |
defd94b7 MC |
302 | *page, unsigned int len, unsigned int offset, |
303 | unsigned short max_sectors) | |
1da177e4 LT |
304 | { |
305 | int retried_segments = 0; | |
306 | struct bio_vec *bvec; | |
307 | ||
308 | /* | |
309 | * cloned bio must not modify vec list | |
310 | */ | |
311 | if (unlikely(bio_flagged(bio, BIO_CLONED))) | |
312 | return 0; | |
313 | ||
80cfd548 | 314 | if (((bio->bi_size + len) >> 9) > max_sectors) |
1da177e4 LT |
315 | return 0; |
316 | ||
80cfd548 JA |
317 | /* |
318 | * For filesystems with a blocksize smaller than the pagesize | |
319 | * we will often be called with the same page as last time and | |
320 | * a consecutive offset. Optimize this special case. | |
321 | */ | |
322 | if (bio->bi_vcnt > 0) { | |
323 | struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
324 | ||
325 | if (page == prev->bv_page && | |
326 | offset == prev->bv_offset + prev->bv_len) { | |
327 | prev->bv_len += len; | |
cc371e66 AK |
328 | |
329 | if (q->merge_bvec_fn) { | |
330 | struct bvec_merge_data bvm = { | |
331 | .bi_bdev = bio->bi_bdev, | |
332 | .bi_sector = bio->bi_sector, | |
333 | .bi_size = bio->bi_size, | |
334 | .bi_rw = bio->bi_rw, | |
335 | }; | |
336 | ||
337 | if (q->merge_bvec_fn(q, &bvm, prev) < len) { | |
338 | prev->bv_len -= len; | |
339 | return 0; | |
340 | } | |
80cfd548 JA |
341 | } |
342 | ||
343 | goto done; | |
344 | } | |
345 | } | |
346 | ||
347 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
1da177e4 LT |
348 | return 0; |
349 | ||
350 | /* | |
351 | * we might lose a segment or two here, but rather that than | |
352 | * make this too complex. | |
353 | */ | |
354 | ||
355 | while (bio->bi_phys_segments >= q->max_phys_segments | |
356 | || bio->bi_hw_segments >= q->max_hw_segments | |
357 | || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) { | |
358 | ||
359 | if (retried_segments) | |
360 | return 0; | |
361 | ||
362 | retried_segments = 1; | |
363 | blk_recount_segments(q, bio); | |
364 | } | |
365 | ||
366 | /* | |
367 | * setup the new entry, we might clear it again later if we | |
368 | * cannot add the page | |
369 | */ | |
370 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; | |
371 | bvec->bv_page = page; | |
372 | bvec->bv_len = len; | |
373 | bvec->bv_offset = offset; | |
374 | ||
375 | /* | |
376 | * if queue has other restrictions (eg varying max sector size | |
377 | * depending on offset), it can specify a merge_bvec_fn in the | |
378 | * queue to get further control | |
379 | */ | |
380 | if (q->merge_bvec_fn) { | |
cc371e66 AK |
381 | struct bvec_merge_data bvm = { |
382 | .bi_bdev = bio->bi_bdev, | |
383 | .bi_sector = bio->bi_sector, | |
384 | .bi_size = bio->bi_size, | |
385 | .bi_rw = bio->bi_rw, | |
386 | }; | |
387 | ||
1da177e4 LT |
388 | /* |
389 | * merge_bvec_fn() returns number of bytes it can accept | |
390 | * at this offset | |
391 | */ | |
cc371e66 | 392 | if (q->merge_bvec_fn(q, &bvm, bvec) < len) { |
1da177e4 LT |
393 | bvec->bv_page = NULL; |
394 | bvec->bv_len = 0; | |
395 | bvec->bv_offset = 0; | |
396 | return 0; | |
397 | } | |
398 | } | |
399 | ||
400 | /* If we may be able to merge these biovecs, force a recount */ | |
401 | if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) || | |
402 | BIOVEC_VIRT_MERGEABLE(bvec-1, bvec))) | |
403 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
404 | ||
405 | bio->bi_vcnt++; | |
406 | bio->bi_phys_segments++; | |
407 | bio->bi_hw_segments++; | |
80cfd548 | 408 | done: |
1da177e4 LT |
409 | bio->bi_size += len; |
410 | return len; | |
411 | } | |
412 | ||
6e68af66 MC |
413 | /** |
414 | * bio_add_pc_page - attempt to add page to bio | |
fddfdeaf | 415 | * @q: the target queue |
6e68af66 MC |
416 | * @bio: destination bio |
417 | * @page: page to add | |
418 | * @len: vec entry length | |
419 | * @offset: vec entry offset | |
420 | * | |
421 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
422 | * number of reasons, such as the bio being full or target block | |
423 | * device limitations. The target block device must allow bio's | |
424 | * smaller than PAGE_SIZE, so it is always possible to add a single | |
425 | * page to an empty bio. This should only be used by REQ_PC bios. | |
426 | */ | |
165125e1 | 427 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page, |
6e68af66 MC |
428 | unsigned int len, unsigned int offset) |
429 | { | |
defd94b7 | 430 | return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors); |
6e68af66 MC |
431 | } |
432 | ||
1da177e4 LT |
433 | /** |
434 | * bio_add_page - attempt to add page to bio | |
435 | * @bio: destination bio | |
436 | * @page: page to add | |
437 | * @len: vec entry length | |
438 | * @offset: vec entry offset | |
439 | * | |
440 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
441 | * number of reasons, such as the bio being full or target block | |
442 | * device limitations. The target block device must allow bio's | |
443 | * smaller than PAGE_SIZE, so it is always possible to add a single | |
444 | * page to an empty bio. | |
445 | */ | |
446 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, | |
447 | unsigned int offset) | |
448 | { | |
defd94b7 MC |
449 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
450 | return __bio_add_page(q, bio, page, len, offset, q->max_sectors); | |
1da177e4 LT |
451 | } |
452 | ||
453 | struct bio_map_data { | |
454 | struct bio_vec *iovecs; | |
c5dec1c3 FT |
455 | int nr_sgvecs; |
456 | struct sg_iovec *sgvecs; | |
1da177e4 LT |
457 | }; |
458 | ||
c5dec1c3 FT |
459 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio, |
460 | struct sg_iovec *iov, int iov_count) | |
1da177e4 LT |
461 | { |
462 | memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); | |
c5dec1c3 FT |
463 | memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count); |
464 | bmd->nr_sgvecs = iov_count; | |
1da177e4 LT |
465 | bio->bi_private = bmd; |
466 | } | |
467 | ||
468 | static void bio_free_map_data(struct bio_map_data *bmd) | |
469 | { | |
470 | kfree(bmd->iovecs); | |
c5dec1c3 | 471 | kfree(bmd->sgvecs); |
1da177e4 LT |
472 | kfree(bmd); |
473 | } | |
474 | ||
c5dec1c3 | 475 | static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count) |
1da177e4 LT |
476 | { |
477 | struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL); | |
478 | ||
479 | if (!bmd) | |
480 | return NULL; | |
481 | ||
482 | bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL); | |
c5dec1c3 FT |
483 | if (!bmd->iovecs) { |
484 | kfree(bmd); | |
485 | return NULL; | |
486 | } | |
487 | ||
488 | bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, GFP_KERNEL); | |
489 | if (bmd->sgvecs) | |
1da177e4 LT |
490 | return bmd; |
491 | ||
c5dec1c3 | 492 | kfree(bmd->iovecs); |
1da177e4 LT |
493 | kfree(bmd); |
494 | return NULL; | |
495 | } | |
496 | ||
c5dec1c3 FT |
497 | static int __bio_copy_iov(struct bio *bio, struct sg_iovec *iov, int iov_count, |
498 | int uncopy) | |
499 | { | |
500 | int ret = 0, i; | |
501 | struct bio_vec *bvec; | |
502 | int iov_idx = 0; | |
503 | unsigned int iov_off = 0; | |
504 | int read = bio_data_dir(bio) == READ; | |
505 | ||
506 | __bio_for_each_segment(bvec, bio, i, 0) { | |
507 | char *bv_addr = page_address(bvec->bv_page); | |
508 | unsigned int bv_len = bvec->bv_len; | |
509 | ||
510 | while (bv_len && iov_idx < iov_count) { | |
511 | unsigned int bytes; | |
512 | char *iov_addr; | |
513 | ||
514 | bytes = min_t(unsigned int, | |
515 | iov[iov_idx].iov_len - iov_off, bv_len); | |
516 | iov_addr = iov[iov_idx].iov_base + iov_off; | |
517 | ||
518 | if (!ret) { | |
519 | if (!read && !uncopy) | |
520 | ret = copy_from_user(bv_addr, iov_addr, | |
521 | bytes); | |
522 | if (read && uncopy) | |
523 | ret = copy_to_user(iov_addr, bv_addr, | |
524 | bytes); | |
525 | ||
526 | if (ret) | |
527 | ret = -EFAULT; | |
528 | } | |
529 | ||
530 | bv_len -= bytes; | |
531 | bv_addr += bytes; | |
532 | iov_addr += bytes; | |
533 | iov_off += bytes; | |
534 | ||
535 | if (iov[iov_idx].iov_len == iov_off) { | |
536 | iov_idx++; | |
537 | iov_off = 0; | |
538 | } | |
539 | } | |
540 | ||
541 | if (uncopy) | |
542 | __free_page(bvec->bv_page); | |
543 | } | |
544 | ||
545 | return ret; | |
546 | } | |
547 | ||
1da177e4 LT |
548 | /** |
549 | * bio_uncopy_user - finish previously mapped bio | |
550 | * @bio: bio being terminated | |
551 | * | |
552 | * Free pages allocated from bio_copy_user() and write back data | |
553 | * to user space in case of a read. | |
554 | */ | |
555 | int bio_uncopy_user(struct bio *bio) | |
556 | { | |
557 | struct bio_map_data *bmd = bio->bi_private; | |
c5dec1c3 | 558 | int ret; |
1da177e4 | 559 | |
c5dec1c3 | 560 | ret = __bio_copy_iov(bio, bmd->sgvecs, bmd->nr_sgvecs, 1); |
1da177e4 | 561 | |
1da177e4 LT |
562 | bio_free_map_data(bmd); |
563 | bio_put(bio); | |
564 | return ret; | |
565 | } | |
566 | ||
567 | /** | |
c5dec1c3 | 568 | * bio_copy_user_iov - copy user data to bio |
1da177e4 | 569 | * @q: destination block queue |
c5dec1c3 FT |
570 | * @iov: the iovec. |
571 | * @iov_count: number of elements in the iovec | |
1da177e4 LT |
572 | * @write_to_vm: bool indicating writing to pages or not |
573 | * | |
574 | * Prepares and returns a bio for indirect user io, bouncing data | |
575 | * to/from kernel pages as necessary. Must be paired with | |
576 | * call bio_uncopy_user() on io completion. | |
577 | */ | |
c5dec1c3 FT |
578 | struct bio *bio_copy_user_iov(struct request_queue *q, struct sg_iovec *iov, |
579 | int iov_count, int write_to_vm) | |
1da177e4 | 580 | { |
1da177e4 LT |
581 | struct bio_map_data *bmd; |
582 | struct bio_vec *bvec; | |
583 | struct page *page; | |
584 | struct bio *bio; | |
585 | int i, ret; | |
c5dec1c3 FT |
586 | int nr_pages = 0; |
587 | unsigned int len = 0; | |
1da177e4 | 588 | |
c5dec1c3 FT |
589 | for (i = 0; i < iov_count; i++) { |
590 | unsigned long uaddr; | |
591 | unsigned long end; | |
592 | unsigned long start; | |
593 | ||
594 | uaddr = (unsigned long)iov[i].iov_base; | |
595 | end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
596 | start = uaddr >> PAGE_SHIFT; | |
597 | ||
598 | nr_pages += end - start; | |
599 | len += iov[i].iov_len; | |
600 | } | |
601 | ||
602 | bmd = bio_alloc_map_data(nr_pages, iov_count); | |
1da177e4 LT |
603 | if (!bmd) |
604 | return ERR_PTR(-ENOMEM); | |
605 | ||
1da177e4 | 606 | ret = -ENOMEM; |
c5dec1c3 | 607 | bio = bio_alloc(GFP_KERNEL, nr_pages); |
1da177e4 LT |
608 | if (!bio) |
609 | goto out_bmd; | |
610 | ||
611 | bio->bi_rw |= (!write_to_vm << BIO_RW); | |
612 | ||
613 | ret = 0; | |
614 | while (len) { | |
615 | unsigned int bytes = PAGE_SIZE; | |
616 | ||
617 | if (bytes > len) | |
618 | bytes = len; | |
619 | ||
620 | page = alloc_page(q->bounce_gfp | GFP_KERNEL); | |
621 | if (!page) { | |
622 | ret = -ENOMEM; | |
623 | break; | |
624 | } | |
625 | ||
0e75f906 | 626 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) |
1da177e4 | 627 | break; |
1da177e4 LT |
628 | |
629 | len -= bytes; | |
630 | } | |
631 | ||
632 | if (ret) | |
633 | goto cleanup; | |
634 | ||
635 | /* | |
636 | * success | |
637 | */ | |
638 | if (!write_to_vm) { | |
c5dec1c3 FT |
639 | ret = __bio_copy_iov(bio, iov, iov_count, 0); |
640 | if (ret) | |
641 | goto cleanup; | |
1da177e4 LT |
642 | } |
643 | ||
c5dec1c3 | 644 | bio_set_map_data(bmd, bio, iov, iov_count); |
1da177e4 LT |
645 | return bio; |
646 | cleanup: | |
647 | bio_for_each_segment(bvec, bio, i) | |
648 | __free_page(bvec->bv_page); | |
649 | ||
650 | bio_put(bio); | |
651 | out_bmd: | |
652 | bio_free_map_data(bmd); | |
653 | return ERR_PTR(ret); | |
654 | } | |
655 | ||
c5dec1c3 FT |
656 | /** |
657 | * bio_copy_user - copy user data to bio | |
658 | * @q: destination block queue | |
659 | * @uaddr: start of user address | |
660 | * @len: length in bytes | |
661 | * @write_to_vm: bool indicating writing to pages or not | |
662 | * | |
663 | * Prepares and returns a bio for indirect user io, bouncing data | |
664 | * to/from kernel pages as necessary. Must be paired with | |
665 | * call bio_uncopy_user() on io completion. | |
666 | */ | |
667 | struct bio *bio_copy_user(struct request_queue *q, unsigned long uaddr, | |
668 | unsigned int len, int write_to_vm) | |
669 | { | |
670 | struct sg_iovec iov; | |
671 | ||
672 | iov.iov_base = (void __user *)uaddr; | |
673 | iov.iov_len = len; | |
674 | ||
675 | return bio_copy_user_iov(q, &iov, 1, write_to_vm); | |
676 | } | |
677 | ||
165125e1 | 678 | static struct bio *__bio_map_user_iov(struct request_queue *q, |
f1970baf JB |
679 | struct block_device *bdev, |
680 | struct sg_iovec *iov, int iov_count, | |
681 | int write_to_vm) | |
1da177e4 | 682 | { |
f1970baf JB |
683 | int i, j; |
684 | int nr_pages = 0; | |
1da177e4 LT |
685 | struct page **pages; |
686 | struct bio *bio; | |
f1970baf JB |
687 | int cur_page = 0; |
688 | int ret, offset; | |
1da177e4 | 689 | |
f1970baf JB |
690 | for (i = 0; i < iov_count; i++) { |
691 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
692 | unsigned long len = iov[i].iov_len; | |
693 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
694 | unsigned long start = uaddr >> PAGE_SHIFT; | |
695 | ||
696 | nr_pages += end - start; | |
697 | /* | |
ad2d7225 | 698 | * buffer must be aligned to at least hardsector size for now |
f1970baf | 699 | */ |
ad2d7225 | 700 | if (uaddr & queue_dma_alignment(q)) |
f1970baf JB |
701 | return ERR_PTR(-EINVAL); |
702 | } | |
703 | ||
704 | if (!nr_pages) | |
1da177e4 LT |
705 | return ERR_PTR(-EINVAL); |
706 | ||
707 | bio = bio_alloc(GFP_KERNEL, nr_pages); | |
708 | if (!bio) | |
709 | return ERR_PTR(-ENOMEM); | |
710 | ||
711 | ret = -ENOMEM; | |
11b0b5ab | 712 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL); |
1da177e4 LT |
713 | if (!pages) |
714 | goto out; | |
715 | ||
f1970baf JB |
716 | for (i = 0; i < iov_count; i++) { |
717 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
718 | unsigned long len = iov[i].iov_len; | |
719 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
720 | unsigned long start = uaddr >> PAGE_SHIFT; | |
721 | const int local_nr_pages = end - start; | |
722 | const int page_limit = cur_page + local_nr_pages; | |
723 | ||
724 | down_read(¤t->mm->mmap_sem); | |
725 | ret = get_user_pages(current, current->mm, uaddr, | |
726 | local_nr_pages, | |
727 | write_to_vm, 0, &pages[cur_page], NULL); | |
728 | up_read(¤t->mm->mmap_sem); | |
729 | ||
99172157 JA |
730 | if (ret < local_nr_pages) { |
731 | ret = -EFAULT; | |
f1970baf | 732 | goto out_unmap; |
99172157 | 733 | } |
f1970baf JB |
734 | |
735 | offset = uaddr & ~PAGE_MASK; | |
736 | for (j = cur_page; j < page_limit; j++) { | |
737 | unsigned int bytes = PAGE_SIZE - offset; | |
738 | ||
739 | if (len <= 0) | |
740 | break; | |
741 | ||
742 | if (bytes > len) | |
743 | bytes = len; | |
744 | ||
745 | /* | |
746 | * sorry... | |
747 | */ | |
defd94b7 MC |
748 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < |
749 | bytes) | |
f1970baf JB |
750 | break; |
751 | ||
752 | len -= bytes; | |
753 | offset = 0; | |
754 | } | |
1da177e4 | 755 | |
f1970baf | 756 | cur_page = j; |
1da177e4 | 757 | /* |
f1970baf | 758 | * release the pages we didn't map into the bio, if any |
1da177e4 | 759 | */ |
f1970baf JB |
760 | while (j < page_limit) |
761 | page_cache_release(pages[j++]); | |
1da177e4 LT |
762 | } |
763 | ||
1da177e4 LT |
764 | kfree(pages); |
765 | ||
766 | /* | |
767 | * set data direction, and check if mapped pages need bouncing | |
768 | */ | |
769 | if (!write_to_vm) | |
770 | bio->bi_rw |= (1 << BIO_RW); | |
771 | ||
f1970baf | 772 | bio->bi_bdev = bdev; |
1da177e4 LT |
773 | bio->bi_flags |= (1 << BIO_USER_MAPPED); |
774 | return bio; | |
f1970baf JB |
775 | |
776 | out_unmap: | |
777 | for (i = 0; i < nr_pages; i++) { | |
778 | if(!pages[i]) | |
779 | break; | |
780 | page_cache_release(pages[i]); | |
781 | } | |
782 | out: | |
1da177e4 LT |
783 | kfree(pages); |
784 | bio_put(bio); | |
785 | return ERR_PTR(ret); | |
786 | } | |
787 | ||
788 | /** | |
789 | * bio_map_user - map user address into bio | |
165125e1 | 790 | * @q: the struct request_queue for the bio |
1da177e4 LT |
791 | * @bdev: destination block device |
792 | * @uaddr: start of user address | |
793 | * @len: length in bytes | |
794 | * @write_to_vm: bool indicating writing to pages or not | |
795 | * | |
796 | * Map the user space address into a bio suitable for io to a block | |
797 | * device. Returns an error pointer in case of error. | |
798 | */ | |
165125e1 | 799 | struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev, |
1da177e4 | 800 | unsigned long uaddr, unsigned int len, int write_to_vm) |
f1970baf JB |
801 | { |
802 | struct sg_iovec iov; | |
803 | ||
3f70353e | 804 | iov.iov_base = (void __user *)uaddr; |
f1970baf JB |
805 | iov.iov_len = len; |
806 | ||
807 | return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm); | |
808 | } | |
809 | ||
810 | /** | |
811 | * bio_map_user_iov - map user sg_iovec table into bio | |
165125e1 | 812 | * @q: the struct request_queue for the bio |
f1970baf JB |
813 | * @bdev: destination block device |
814 | * @iov: the iovec. | |
815 | * @iov_count: number of elements in the iovec | |
816 | * @write_to_vm: bool indicating writing to pages or not | |
817 | * | |
818 | * Map the user space address into a bio suitable for io to a block | |
819 | * device. Returns an error pointer in case of error. | |
820 | */ | |
165125e1 | 821 | struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev, |
f1970baf JB |
822 | struct sg_iovec *iov, int iov_count, |
823 | int write_to_vm) | |
1da177e4 LT |
824 | { |
825 | struct bio *bio; | |
826 | ||
f1970baf | 827 | bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm); |
1da177e4 LT |
828 | |
829 | if (IS_ERR(bio)) | |
830 | return bio; | |
831 | ||
832 | /* | |
833 | * subtle -- if __bio_map_user() ended up bouncing a bio, | |
834 | * it would normally disappear when its bi_end_io is run. | |
835 | * however, we need it for the unmap, so grab an extra | |
836 | * reference to it | |
837 | */ | |
838 | bio_get(bio); | |
839 | ||
0e75f906 | 840 | return bio; |
1da177e4 LT |
841 | } |
842 | ||
843 | static void __bio_unmap_user(struct bio *bio) | |
844 | { | |
845 | struct bio_vec *bvec; | |
846 | int i; | |
847 | ||
848 | /* | |
849 | * make sure we dirty pages we wrote to | |
850 | */ | |
851 | __bio_for_each_segment(bvec, bio, i, 0) { | |
852 | if (bio_data_dir(bio) == READ) | |
853 | set_page_dirty_lock(bvec->bv_page); | |
854 | ||
855 | page_cache_release(bvec->bv_page); | |
856 | } | |
857 | ||
858 | bio_put(bio); | |
859 | } | |
860 | ||
861 | /** | |
862 | * bio_unmap_user - unmap a bio | |
863 | * @bio: the bio being unmapped | |
864 | * | |
865 | * Unmap a bio previously mapped by bio_map_user(). Must be called with | |
866 | * a process context. | |
867 | * | |
868 | * bio_unmap_user() may sleep. | |
869 | */ | |
870 | void bio_unmap_user(struct bio *bio) | |
871 | { | |
872 | __bio_unmap_user(bio); | |
873 | bio_put(bio); | |
874 | } | |
875 | ||
6712ecf8 | 876 | static void bio_map_kern_endio(struct bio *bio, int err) |
b823825e | 877 | { |
b823825e | 878 | bio_put(bio); |
b823825e JA |
879 | } |
880 | ||
881 | ||
165125e1 | 882 | static struct bio *__bio_map_kern(struct request_queue *q, void *data, |
27496a8c | 883 | unsigned int len, gfp_t gfp_mask) |
df46b9a4 MC |
884 | { |
885 | unsigned long kaddr = (unsigned long)data; | |
886 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
887 | unsigned long start = kaddr >> PAGE_SHIFT; | |
888 | const int nr_pages = end - start; | |
889 | int offset, i; | |
890 | struct bio *bio; | |
891 | ||
892 | bio = bio_alloc(gfp_mask, nr_pages); | |
893 | if (!bio) | |
894 | return ERR_PTR(-ENOMEM); | |
895 | ||
896 | offset = offset_in_page(kaddr); | |
897 | for (i = 0; i < nr_pages; i++) { | |
898 | unsigned int bytes = PAGE_SIZE - offset; | |
899 | ||
900 | if (len <= 0) | |
901 | break; | |
902 | ||
903 | if (bytes > len) | |
904 | bytes = len; | |
905 | ||
defd94b7 MC |
906 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, |
907 | offset) < bytes) | |
df46b9a4 MC |
908 | break; |
909 | ||
910 | data += bytes; | |
911 | len -= bytes; | |
912 | offset = 0; | |
913 | } | |
914 | ||
b823825e | 915 | bio->bi_end_io = bio_map_kern_endio; |
df46b9a4 MC |
916 | return bio; |
917 | } | |
918 | ||
919 | /** | |
920 | * bio_map_kern - map kernel address into bio | |
165125e1 | 921 | * @q: the struct request_queue for the bio |
df46b9a4 MC |
922 | * @data: pointer to buffer to map |
923 | * @len: length in bytes | |
924 | * @gfp_mask: allocation flags for bio allocation | |
925 | * | |
926 | * Map the kernel address into a bio suitable for io to a block | |
927 | * device. Returns an error pointer in case of error. | |
928 | */ | |
165125e1 | 929 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, |
27496a8c | 930 | gfp_t gfp_mask) |
df46b9a4 MC |
931 | { |
932 | struct bio *bio; | |
933 | ||
934 | bio = __bio_map_kern(q, data, len, gfp_mask); | |
935 | if (IS_ERR(bio)) | |
936 | return bio; | |
937 | ||
938 | if (bio->bi_size == len) | |
939 | return bio; | |
940 | ||
941 | /* | |
942 | * Don't support partial mappings. | |
943 | */ | |
944 | bio_put(bio); | |
945 | return ERR_PTR(-EINVAL); | |
946 | } | |
947 | ||
68154e90 FT |
948 | static void bio_copy_kern_endio(struct bio *bio, int err) |
949 | { | |
950 | struct bio_vec *bvec; | |
951 | const int read = bio_data_dir(bio) == READ; | |
952 | char *p = bio->bi_private; | |
953 | int i; | |
954 | ||
955 | __bio_for_each_segment(bvec, bio, i, 0) { | |
956 | char *addr = page_address(bvec->bv_page); | |
957 | ||
958 | if (read && !err) | |
959 | memcpy(p, addr, bvec->bv_len); | |
960 | ||
961 | __free_page(bvec->bv_page); | |
962 | p += bvec->bv_len; | |
963 | } | |
964 | ||
965 | bio_put(bio); | |
966 | } | |
967 | ||
968 | /** | |
969 | * bio_copy_kern - copy kernel address into bio | |
970 | * @q: the struct request_queue for the bio | |
971 | * @data: pointer to buffer to copy | |
972 | * @len: length in bytes | |
973 | * @gfp_mask: allocation flags for bio and page allocation | |
ffee0259 | 974 | * @reading: data direction is READ |
68154e90 FT |
975 | * |
976 | * copy the kernel address into a bio suitable for io to a block | |
977 | * device. Returns an error pointer in case of error. | |
978 | */ | |
979 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
980 | gfp_t gfp_mask, int reading) | |
981 | { | |
982 | unsigned long kaddr = (unsigned long)data; | |
983 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
984 | unsigned long start = kaddr >> PAGE_SHIFT; | |
985 | const int nr_pages = end - start; | |
986 | struct bio *bio; | |
987 | struct bio_vec *bvec; | |
988 | int i, ret; | |
989 | ||
990 | bio = bio_alloc(gfp_mask, nr_pages); | |
991 | if (!bio) | |
992 | return ERR_PTR(-ENOMEM); | |
993 | ||
994 | while (len) { | |
995 | struct page *page; | |
996 | unsigned int bytes = PAGE_SIZE; | |
997 | ||
998 | if (bytes > len) | |
999 | bytes = len; | |
1000 | ||
1001 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
1002 | if (!page) { | |
1003 | ret = -ENOMEM; | |
1004 | goto cleanup; | |
1005 | } | |
1006 | ||
1007 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) { | |
1008 | ret = -EINVAL; | |
1009 | goto cleanup; | |
1010 | } | |
1011 | ||
1012 | len -= bytes; | |
1013 | } | |
1014 | ||
1015 | if (!reading) { | |
1016 | void *p = data; | |
1017 | ||
1018 | bio_for_each_segment(bvec, bio, i) { | |
1019 | char *addr = page_address(bvec->bv_page); | |
1020 | ||
1021 | memcpy(addr, p, bvec->bv_len); | |
1022 | p += bvec->bv_len; | |
1023 | } | |
1024 | } | |
1025 | ||
1026 | bio->bi_private = data; | |
1027 | bio->bi_end_io = bio_copy_kern_endio; | |
1028 | return bio; | |
1029 | cleanup: | |
1030 | bio_for_each_segment(bvec, bio, i) | |
1031 | __free_page(bvec->bv_page); | |
1032 | ||
1033 | bio_put(bio); | |
1034 | ||
1035 | return ERR_PTR(ret); | |
1036 | } | |
1037 | ||
1da177e4 LT |
1038 | /* |
1039 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1040 | * for performing direct-IO in BIOs. | |
1041 | * | |
1042 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1043 | * because the required locks are not interrupt-safe. So what we can do is to | |
1044 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1045 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1046 | * in process context. | |
1047 | * | |
1048 | * We special-case compound pages here: normally this means reads into hugetlb | |
1049 | * pages. The logic in here doesn't really work right for compound pages | |
1050 | * because the VM does not uniformly chase down the head page in all cases. | |
1051 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1052 | * handle them at all. So we skip compound pages here at an early stage. | |
1053 | * | |
1054 | * Note that this code is very hard to test under normal circumstances because | |
1055 | * direct-io pins the pages with get_user_pages(). This makes | |
1056 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
1057 | * But other code (eg, pdflush) could clean the pages if they are mapped | |
1058 | * pagecache. | |
1059 | * | |
1060 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1061 | * deferred bio dirtying paths. | |
1062 | */ | |
1063 | ||
1064 | /* | |
1065 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1066 | */ | |
1067 | void bio_set_pages_dirty(struct bio *bio) | |
1068 | { | |
1069 | struct bio_vec *bvec = bio->bi_io_vec; | |
1070 | int i; | |
1071 | ||
1072 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1073 | struct page *page = bvec[i].bv_page; | |
1074 | ||
1075 | if (page && !PageCompound(page)) | |
1076 | set_page_dirty_lock(page); | |
1077 | } | |
1078 | } | |
1079 | ||
86b6c7a7 | 1080 | static void bio_release_pages(struct bio *bio) |
1da177e4 LT |
1081 | { |
1082 | struct bio_vec *bvec = bio->bi_io_vec; | |
1083 | int i; | |
1084 | ||
1085 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1086 | struct page *page = bvec[i].bv_page; | |
1087 | ||
1088 | if (page) | |
1089 | put_page(page); | |
1090 | } | |
1091 | } | |
1092 | ||
1093 | /* | |
1094 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1095 | * If they are, then fine. If, however, some pages are clean then they must | |
1096 | * have been written out during the direct-IO read. So we take another ref on | |
1097 | * the BIO and the offending pages and re-dirty the pages in process context. | |
1098 | * | |
1099 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
1100 | * here on. It will run one page_cache_release() against each page and will | |
1101 | * run one bio_put() against the BIO. | |
1102 | */ | |
1103 | ||
65f27f38 | 1104 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1105 | |
65f27f38 | 1106 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1107 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1108 | static struct bio *bio_dirty_list; | |
1109 | ||
1110 | /* | |
1111 | * This runs in process context | |
1112 | */ | |
65f27f38 | 1113 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 LT |
1114 | { |
1115 | unsigned long flags; | |
1116 | struct bio *bio; | |
1117 | ||
1118 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1119 | bio = bio_dirty_list; | |
1120 | bio_dirty_list = NULL; | |
1121 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1122 | ||
1123 | while (bio) { | |
1124 | struct bio *next = bio->bi_private; | |
1125 | ||
1126 | bio_set_pages_dirty(bio); | |
1127 | bio_release_pages(bio); | |
1128 | bio_put(bio); | |
1129 | bio = next; | |
1130 | } | |
1131 | } | |
1132 | ||
1133 | void bio_check_pages_dirty(struct bio *bio) | |
1134 | { | |
1135 | struct bio_vec *bvec = bio->bi_io_vec; | |
1136 | int nr_clean_pages = 0; | |
1137 | int i; | |
1138 | ||
1139 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1140 | struct page *page = bvec[i].bv_page; | |
1141 | ||
1142 | if (PageDirty(page) || PageCompound(page)) { | |
1143 | page_cache_release(page); | |
1144 | bvec[i].bv_page = NULL; | |
1145 | } else { | |
1146 | nr_clean_pages++; | |
1147 | } | |
1148 | } | |
1149 | ||
1150 | if (nr_clean_pages) { | |
1151 | unsigned long flags; | |
1152 | ||
1153 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1154 | bio->bi_private = bio_dirty_list; | |
1155 | bio_dirty_list = bio; | |
1156 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1157 | schedule_work(&bio_dirty_work); | |
1158 | } else { | |
1159 | bio_put(bio); | |
1160 | } | |
1161 | } | |
1162 | ||
1163 | /** | |
1164 | * bio_endio - end I/O on a bio | |
1165 | * @bio: bio | |
1da177e4 LT |
1166 | * @error: error, if any |
1167 | * | |
1168 | * Description: | |
6712ecf8 | 1169 | * bio_endio() will end I/O on the whole bio. bio_endio() is the |
5bb23a68 N |
1170 | * preferred way to end I/O on a bio, it takes care of clearing |
1171 | * BIO_UPTODATE on error. @error is 0 on success, and and one of the | |
1172 | * established -Exxxx (-EIO, for instance) error values in case | |
1173 | * something went wrong. Noone should call bi_end_io() directly on a | |
1174 | * bio unless they own it and thus know that it has an end_io | |
1175 | * function. | |
1da177e4 | 1176 | **/ |
6712ecf8 | 1177 | void bio_endio(struct bio *bio, int error) |
1da177e4 LT |
1178 | { |
1179 | if (error) | |
1180 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
9cc54d40 N |
1181 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) |
1182 | error = -EIO; | |
1da177e4 | 1183 | |
5bb23a68 | 1184 | if (bio->bi_end_io) |
6712ecf8 | 1185 | bio->bi_end_io(bio, error); |
1da177e4 LT |
1186 | } |
1187 | ||
1188 | void bio_pair_release(struct bio_pair *bp) | |
1189 | { | |
1190 | if (atomic_dec_and_test(&bp->cnt)) { | |
1191 | struct bio *master = bp->bio1.bi_private; | |
1192 | ||
6712ecf8 | 1193 | bio_endio(master, bp->error); |
1da177e4 LT |
1194 | mempool_free(bp, bp->bio2.bi_private); |
1195 | } | |
1196 | } | |
1197 | ||
6712ecf8 | 1198 | static void bio_pair_end_1(struct bio *bi, int err) |
1da177e4 LT |
1199 | { |
1200 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); | |
1201 | ||
1202 | if (err) | |
1203 | bp->error = err; | |
1204 | ||
1da177e4 | 1205 | bio_pair_release(bp); |
1da177e4 LT |
1206 | } |
1207 | ||
6712ecf8 | 1208 | static void bio_pair_end_2(struct bio *bi, int err) |
1da177e4 LT |
1209 | { |
1210 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); | |
1211 | ||
1212 | if (err) | |
1213 | bp->error = err; | |
1214 | ||
1da177e4 | 1215 | bio_pair_release(bp); |
1da177e4 LT |
1216 | } |
1217 | ||
1218 | /* | |
1219 | * split a bio - only worry about a bio with a single page | |
1220 | * in it's iovec | |
1221 | */ | |
1222 | struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors) | |
1223 | { | |
1224 | struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO); | |
1225 | ||
1226 | if (!bp) | |
1227 | return bp; | |
1228 | ||
2056a782 JA |
1229 | blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi, |
1230 | bi->bi_sector + first_sectors); | |
1231 | ||
1da177e4 LT |
1232 | BUG_ON(bi->bi_vcnt != 1); |
1233 | BUG_ON(bi->bi_idx != 0); | |
1234 | atomic_set(&bp->cnt, 3); | |
1235 | bp->error = 0; | |
1236 | bp->bio1 = *bi; | |
1237 | bp->bio2 = *bi; | |
1238 | bp->bio2.bi_sector += first_sectors; | |
1239 | bp->bio2.bi_size -= first_sectors << 9; | |
1240 | bp->bio1.bi_size = first_sectors << 9; | |
1241 | ||
1242 | bp->bv1 = bi->bi_io_vec[0]; | |
1243 | bp->bv2 = bi->bi_io_vec[0]; | |
1244 | bp->bv2.bv_offset += first_sectors << 9; | |
1245 | bp->bv2.bv_len -= first_sectors << 9; | |
1246 | bp->bv1.bv_len = first_sectors << 9; | |
1247 | ||
1248 | bp->bio1.bi_io_vec = &bp->bv1; | |
1249 | bp->bio2.bi_io_vec = &bp->bv2; | |
1250 | ||
a2eb0c10 N |
1251 | bp->bio1.bi_max_vecs = 1; |
1252 | bp->bio2.bi_max_vecs = 1; | |
1253 | ||
1da177e4 LT |
1254 | bp->bio1.bi_end_io = bio_pair_end_1; |
1255 | bp->bio2.bi_end_io = bio_pair_end_2; | |
1256 | ||
1257 | bp->bio1.bi_private = bi; | |
1258 | bp->bio2.bi_private = pool; | |
1259 | ||
7ba1ba12 MP |
1260 | if (bio_integrity(bi)) |
1261 | bio_integrity_split(bi, bp, first_sectors); | |
1262 | ||
1da177e4 LT |
1263 | return bp; |
1264 | } | |
1265 | ||
1da177e4 LT |
1266 | |
1267 | /* | |
1268 | * create memory pools for biovec's in a bio_set. | |
1269 | * use the global biovec slabs created for general use. | |
1270 | */ | |
5972511b | 1271 | static int biovec_create_pools(struct bio_set *bs, int pool_entries) |
1da177e4 LT |
1272 | { |
1273 | int i; | |
1274 | ||
1275 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1276 | struct biovec_slab *bp = bvec_slabs + i; | |
1277 | mempool_t **bvp = bs->bvec_pools + i; | |
1278 | ||
93d2341c | 1279 | *bvp = mempool_create_slab_pool(pool_entries, bp->slab); |
1da177e4 LT |
1280 | if (!*bvp) |
1281 | return -ENOMEM; | |
1282 | } | |
1283 | return 0; | |
1284 | } | |
1285 | ||
1286 | static void biovec_free_pools(struct bio_set *bs) | |
1287 | { | |
1288 | int i; | |
1289 | ||
1290 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1291 | mempool_t *bvp = bs->bvec_pools[i]; | |
1292 | ||
1293 | if (bvp) | |
1294 | mempool_destroy(bvp); | |
1295 | } | |
1296 | ||
1297 | } | |
1298 | ||
1299 | void bioset_free(struct bio_set *bs) | |
1300 | { | |
1301 | if (bs->bio_pool) | |
1302 | mempool_destroy(bs->bio_pool); | |
1303 | ||
7ba1ba12 | 1304 | bioset_integrity_free(bs); |
1da177e4 LT |
1305 | biovec_free_pools(bs); |
1306 | ||
1307 | kfree(bs); | |
1308 | } | |
1309 | ||
5972511b | 1310 | struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size) |
1da177e4 | 1311 | { |
11b0b5ab | 1312 | struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL); |
1da177e4 LT |
1313 | |
1314 | if (!bs) | |
1315 | return NULL; | |
1316 | ||
93d2341c | 1317 | bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab); |
1da177e4 LT |
1318 | if (!bs->bio_pool) |
1319 | goto bad; | |
1320 | ||
7ba1ba12 MP |
1321 | if (bioset_integrity_create(bs, bio_pool_size)) |
1322 | goto bad; | |
1323 | ||
5972511b | 1324 | if (!biovec_create_pools(bs, bvec_pool_size)) |
1da177e4 LT |
1325 | return bs; |
1326 | ||
1327 | bad: | |
1328 | bioset_free(bs); | |
1329 | return NULL; | |
1330 | } | |
1331 | ||
1332 | static void __init biovec_init_slabs(void) | |
1333 | { | |
1334 | int i; | |
1335 | ||
1336 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1337 | int size; | |
1338 | struct biovec_slab *bvs = bvec_slabs + i; | |
1339 | ||
1340 | size = bvs->nr_vecs * sizeof(struct bio_vec); | |
1341 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
20c2df83 | 1342 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
1343 | } |
1344 | } | |
1345 | ||
1346 | static int __init init_bio(void) | |
1347 | { | |
0a31bd5f | 1348 | bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
1da177e4 | 1349 | |
7ba1ba12 | 1350 | bio_integrity_init_slab(); |
1da177e4 LT |
1351 | biovec_init_slabs(); |
1352 | ||
5972511b | 1353 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 2); |
1da177e4 LT |
1354 | if (!fs_bio_set) |
1355 | panic("bio: can't allocate bios\n"); | |
1356 | ||
0eaae62a MD |
1357 | bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES, |
1358 | sizeof(struct bio_pair)); | |
1da177e4 LT |
1359 | if (!bio_split_pool) |
1360 | panic("bio: can't create split pool\n"); | |
1361 | ||
1362 | return 0; | |
1363 | } | |
1364 | ||
1365 | subsys_initcall(init_bio); | |
1366 | ||
1367 | EXPORT_SYMBOL(bio_alloc); | |
1368 | EXPORT_SYMBOL(bio_put); | |
3676347a | 1369 | EXPORT_SYMBOL(bio_free); |
1da177e4 LT |
1370 | EXPORT_SYMBOL(bio_endio); |
1371 | EXPORT_SYMBOL(bio_init); | |
1372 | EXPORT_SYMBOL(__bio_clone); | |
1373 | EXPORT_SYMBOL(bio_clone); | |
1374 | EXPORT_SYMBOL(bio_phys_segments); | |
1375 | EXPORT_SYMBOL(bio_hw_segments); | |
1376 | EXPORT_SYMBOL(bio_add_page); | |
6e68af66 | 1377 | EXPORT_SYMBOL(bio_add_pc_page); |
1da177e4 | 1378 | EXPORT_SYMBOL(bio_get_nr_vecs); |
40044ce0 JA |
1379 | EXPORT_SYMBOL(bio_map_user); |
1380 | EXPORT_SYMBOL(bio_unmap_user); | |
df46b9a4 | 1381 | EXPORT_SYMBOL(bio_map_kern); |
68154e90 | 1382 | EXPORT_SYMBOL(bio_copy_kern); |
1da177e4 LT |
1383 | EXPORT_SYMBOL(bio_pair_release); |
1384 | EXPORT_SYMBOL(bio_split); | |
1385 | EXPORT_SYMBOL(bio_split_pool); | |
1386 | EXPORT_SYMBOL(bio_copy_user); | |
1387 | EXPORT_SYMBOL(bio_uncopy_user); | |
1388 | EXPORT_SYMBOL(bioset_create); | |
1389 | EXPORT_SYMBOL(bioset_free); | |
1390 | EXPORT_SYMBOL(bio_alloc_bioset); |