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; | |
328 | if (q->merge_bvec_fn && | |
329 | q->merge_bvec_fn(q, bio, prev) < len) { | |
330 | prev->bv_len -= len; | |
331 | return 0; | |
332 | } | |
333 | ||
334 | goto done; | |
335 | } | |
336 | } | |
337 | ||
338 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
1da177e4 LT |
339 | return 0; |
340 | ||
341 | /* | |
342 | * we might lose a segment or two here, but rather that than | |
343 | * make this too complex. | |
344 | */ | |
345 | ||
346 | while (bio->bi_phys_segments >= q->max_phys_segments | |
347 | || bio->bi_hw_segments >= q->max_hw_segments | |
348 | || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) { | |
349 | ||
350 | if (retried_segments) | |
351 | return 0; | |
352 | ||
353 | retried_segments = 1; | |
354 | blk_recount_segments(q, bio); | |
355 | } | |
356 | ||
357 | /* | |
358 | * setup the new entry, we might clear it again later if we | |
359 | * cannot add the page | |
360 | */ | |
361 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; | |
362 | bvec->bv_page = page; | |
363 | bvec->bv_len = len; | |
364 | bvec->bv_offset = offset; | |
365 | ||
366 | /* | |
367 | * if queue has other restrictions (eg varying max sector size | |
368 | * depending on offset), it can specify a merge_bvec_fn in the | |
369 | * queue to get further control | |
370 | */ | |
371 | if (q->merge_bvec_fn) { | |
372 | /* | |
373 | * merge_bvec_fn() returns number of bytes it can accept | |
374 | * at this offset | |
375 | */ | |
376 | if (q->merge_bvec_fn(q, bio, bvec) < len) { | |
377 | bvec->bv_page = NULL; | |
378 | bvec->bv_len = 0; | |
379 | bvec->bv_offset = 0; | |
380 | return 0; | |
381 | } | |
382 | } | |
383 | ||
384 | /* If we may be able to merge these biovecs, force a recount */ | |
385 | if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) || | |
386 | BIOVEC_VIRT_MERGEABLE(bvec-1, bvec))) | |
387 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
388 | ||
389 | bio->bi_vcnt++; | |
390 | bio->bi_phys_segments++; | |
391 | bio->bi_hw_segments++; | |
80cfd548 | 392 | done: |
1da177e4 LT |
393 | bio->bi_size += len; |
394 | return len; | |
395 | } | |
396 | ||
6e68af66 MC |
397 | /** |
398 | * bio_add_pc_page - attempt to add page to bio | |
fddfdeaf | 399 | * @q: the target queue |
6e68af66 MC |
400 | * @bio: destination bio |
401 | * @page: page to add | |
402 | * @len: vec entry length | |
403 | * @offset: vec entry offset | |
404 | * | |
405 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
406 | * number of reasons, such as the bio being full or target block | |
407 | * device limitations. The target block device must allow bio's | |
408 | * smaller than PAGE_SIZE, so it is always possible to add a single | |
409 | * page to an empty bio. This should only be used by REQ_PC bios. | |
410 | */ | |
165125e1 | 411 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page, |
6e68af66 MC |
412 | unsigned int len, unsigned int offset) |
413 | { | |
defd94b7 | 414 | return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors); |
6e68af66 MC |
415 | } |
416 | ||
1da177e4 LT |
417 | /** |
418 | * bio_add_page - attempt to add page to bio | |
419 | * @bio: destination bio | |
420 | * @page: page to add | |
421 | * @len: vec entry length | |
422 | * @offset: vec entry offset | |
423 | * | |
424 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
425 | * number of reasons, such as the bio being full or target block | |
426 | * device limitations. The target block device must allow bio's | |
427 | * smaller than PAGE_SIZE, so it is always possible to add a single | |
428 | * page to an empty bio. | |
429 | */ | |
430 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, | |
431 | unsigned int offset) | |
432 | { | |
defd94b7 MC |
433 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
434 | return __bio_add_page(q, bio, page, len, offset, q->max_sectors); | |
1da177e4 LT |
435 | } |
436 | ||
437 | struct bio_map_data { | |
438 | struct bio_vec *iovecs; | |
c5dec1c3 FT |
439 | int nr_sgvecs; |
440 | struct sg_iovec *sgvecs; | |
1da177e4 LT |
441 | }; |
442 | ||
c5dec1c3 FT |
443 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio, |
444 | struct sg_iovec *iov, int iov_count) | |
1da177e4 LT |
445 | { |
446 | memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); | |
c5dec1c3 FT |
447 | memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count); |
448 | bmd->nr_sgvecs = iov_count; | |
1da177e4 LT |
449 | bio->bi_private = bmd; |
450 | } | |
451 | ||
452 | static void bio_free_map_data(struct bio_map_data *bmd) | |
453 | { | |
454 | kfree(bmd->iovecs); | |
c5dec1c3 | 455 | kfree(bmd->sgvecs); |
1da177e4 LT |
456 | kfree(bmd); |
457 | } | |
458 | ||
c5dec1c3 | 459 | static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count) |
1da177e4 LT |
460 | { |
461 | struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL); | |
462 | ||
463 | if (!bmd) | |
464 | return NULL; | |
465 | ||
466 | bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL); | |
c5dec1c3 FT |
467 | if (!bmd->iovecs) { |
468 | kfree(bmd); | |
469 | return NULL; | |
470 | } | |
471 | ||
472 | bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, GFP_KERNEL); | |
473 | if (bmd->sgvecs) | |
1da177e4 LT |
474 | return bmd; |
475 | ||
c5dec1c3 | 476 | kfree(bmd->iovecs); |
1da177e4 LT |
477 | kfree(bmd); |
478 | return NULL; | |
479 | } | |
480 | ||
c5dec1c3 FT |
481 | static int __bio_copy_iov(struct bio *bio, struct sg_iovec *iov, int iov_count, |
482 | int uncopy) | |
483 | { | |
484 | int ret = 0, i; | |
485 | struct bio_vec *bvec; | |
486 | int iov_idx = 0; | |
487 | unsigned int iov_off = 0; | |
488 | int read = bio_data_dir(bio) == READ; | |
489 | ||
490 | __bio_for_each_segment(bvec, bio, i, 0) { | |
491 | char *bv_addr = page_address(bvec->bv_page); | |
492 | unsigned int bv_len = bvec->bv_len; | |
493 | ||
494 | while (bv_len && iov_idx < iov_count) { | |
495 | unsigned int bytes; | |
496 | char *iov_addr; | |
497 | ||
498 | bytes = min_t(unsigned int, | |
499 | iov[iov_idx].iov_len - iov_off, bv_len); | |
500 | iov_addr = iov[iov_idx].iov_base + iov_off; | |
501 | ||
502 | if (!ret) { | |
503 | if (!read && !uncopy) | |
504 | ret = copy_from_user(bv_addr, iov_addr, | |
505 | bytes); | |
506 | if (read && uncopy) | |
507 | ret = copy_to_user(iov_addr, bv_addr, | |
508 | bytes); | |
509 | ||
510 | if (ret) | |
511 | ret = -EFAULT; | |
512 | } | |
513 | ||
514 | bv_len -= bytes; | |
515 | bv_addr += bytes; | |
516 | iov_addr += bytes; | |
517 | iov_off += bytes; | |
518 | ||
519 | if (iov[iov_idx].iov_len == iov_off) { | |
520 | iov_idx++; | |
521 | iov_off = 0; | |
522 | } | |
523 | } | |
524 | ||
525 | if (uncopy) | |
526 | __free_page(bvec->bv_page); | |
527 | } | |
528 | ||
529 | return ret; | |
530 | } | |
531 | ||
1da177e4 LT |
532 | /** |
533 | * bio_uncopy_user - finish previously mapped bio | |
534 | * @bio: bio being terminated | |
535 | * | |
536 | * Free pages allocated from bio_copy_user() and write back data | |
537 | * to user space in case of a read. | |
538 | */ | |
539 | int bio_uncopy_user(struct bio *bio) | |
540 | { | |
541 | struct bio_map_data *bmd = bio->bi_private; | |
c5dec1c3 | 542 | int ret; |
1da177e4 | 543 | |
c5dec1c3 | 544 | ret = __bio_copy_iov(bio, bmd->sgvecs, bmd->nr_sgvecs, 1); |
1da177e4 | 545 | |
1da177e4 LT |
546 | bio_free_map_data(bmd); |
547 | bio_put(bio); | |
548 | return ret; | |
549 | } | |
550 | ||
551 | /** | |
c5dec1c3 | 552 | * bio_copy_user_iov - copy user data to bio |
1da177e4 | 553 | * @q: destination block queue |
c5dec1c3 FT |
554 | * @iov: the iovec. |
555 | * @iov_count: number of elements in the iovec | |
1da177e4 LT |
556 | * @write_to_vm: bool indicating writing to pages or not |
557 | * | |
558 | * Prepares and returns a bio for indirect user io, bouncing data | |
559 | * to/from kernel pages as necessary. Must be paired with | |
560 | * call bio_uncopy_user() on io completion. | |
561 | */ | |
c5dec1c3 FT |
562 | struct bio *bio_copy_user_iov(struct request_queue *q, struct sg_iovec *iov, |
563 | int iov_count, int write_to_vm) | |
1da177e4 | 564 | { |
1da177e4 LT |
565 | struct bio_map_data *bmd; |
566 | struct bio_vec *bvec; | |
567 | struct page *page; | |
568 | struct bio *bio; | |
569 | int i, ret; | |
c5dec1c3 FT |
570 | int nr_pages = 0; |
571 | unsigned int len = 0; | |
1da177e4 | 572 | |
c5dec1c3 FT |
573 | for (i = 0; i < iov_count; i++) { |
574 | unsigned long uaddr; | |
575 | unsigned long end; | |
576 | unsigned long start; | |
577 | ||
578 | uaddr = (unsigned long)iov[i].iov_base; | |
579 | end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
580 | start = uaddr >> PAGE_SHIFT; | |
581 | ||
582 | nr_pages += end - start; | |
583 | len += iov[i].iov_len; | |
584 | } | |
585 | ||
586 | bmd = bio_alloc_map_data(nr_pages, iov_count); | |
1da177e4 LT |
587 | if (!bmd) |
588 | return ERR_PTR(-ENOMEM); | |
589 | ||
1da177e4 | 590 | ret = -ENOMEM; |
c5dec1c3 | 591 | bio = bio_alloc(GFP_KERNEL, nr_pages); |
1da177e4 LT |
592 | if (!bio) |
593 | goto out_bmd; | |
594 | ||
595 | bio->bi_rw |= (!write_to_vm << BIO_RW); | |
596 | ||
597 | ret = 0; | |
598 | while (len) { | |
599 | unsigned int bytes = PAGE_SIZE; | |
600 | ||
601 | if (bytes > len) | |
602 | bytes = len; | |
603 | ||
604 | page = alloc_page(q->bounce_gfp | GFP_KERNEL); | |
605 | if (!page) { | |
606 | ret = -ENOMEM; | |
607 | break; | |
608 | } | |
609 | ||
0e75f906 | 610 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) |
1da177e4 | 611 | break; |
1da177e4 LT |
612 | |
613 | len -= bytes; | |
614 | } | |
615 | ||
616 | if (ret) | |
617 | goto cleanup; | |
618 | ||
619 | /* | |
620 | * success | |
621 | */ | |
622 | if (!write_to_vm) { | |
c5dec1c3 FT |
623 | ret = __bio_copy_iov(bio, iov, iov_count, 0); |
624 | if (ret) | |
625 | goto cleanup; | |
1da177e4 LT |
626 | } |
627 | ||
c5dec1c3 | 628 | bio_set_map_data(bmd, bio, iov, iov_count); |
1da177e4 LT |
629 | return bio; |
630 | cleanup: | |
631 | bio_for_each_segment(bvec, bio, i) | |
632 | __free_page(bvec->bv_page); | |
633 | ||
634 | bio_put(bio); | |
635 | out_bmd: | |
636 | bio_free_map_data(bmd); | |
637 | return ERR_PTR(ret); | |
638 | } | |
639 | ||
c5dec1c3 FT |
640 | /** |
641 | * bio_copy_user - copy user data to bio | |
642 | * @q: destination block queue | |
643 | * @uaddr: start of user address | |
644 | * @len: length in bytes | |
645 | * @write_to_vm: bool indicating writing to pages or not | |
646 | * | |
647 | * Prepares and returns a bio for indirect user io, bouncing data | |
648 | * to/from kernel pages as necessary. Must be paired with | |
649 | * call bio_uncopy_user() on io completion. | |
650 | */ | |
651 | struct bio *bio_copy_user(struct request_queue *q, unsigned long uaddr, | |
652 | unsigned int len, int write_to_vm) | |
653 | { | |
654 | struct sg_iovec iov; | |
655 | ||
656 | iov.iov_base = (void __user *)uaddr; | |
657 | iov.iov_len = len; | |
658 | ||
659 | return bio_copy_user_iov(q, &iov, 1, write_to_vm); | |
660 | } | |
661 | ||
165125e1 | 662 | static struct bio *__bio_map_user_iov(struct request_queue *q, |
f1970baf JB |
663 | struct block_device *bdev, |
664 | struct sg_iovec *iov, int iov_count, | |
665 | int write_to_vm) | |
1da177e4 | 666 | { |
f1970baf JB |
667 | int i, j; |
668 | int nr_pages = 0; | |
1da177e4 LT |
669 | struct page **pages; |
670 | struct bio *bio; | |
f1970baf JB |
671 | int cur_page = 0; |
672 | int ret, offset; | |
1da177e4 | 673 | |
f1970baf JB |
674 | for (i = 0; i < iov_count; i++) { |
675 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
676 | unsigned long len = iov[i].iov_len; | |
677 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
678 | unsigned long start = uaddr >> PAGE_SHIFT; | |
679 | ||
680 | nr_pages += end - start; | |
681 | /* | |
ad2d7225 | 682 | * buffer must be aligned to at least hardsector size for now |
f1970baf | 683 | */ |
ad2d7225 | 684 | if (uaddr & queue_dma_alignment(q)) |
f1970baf JB |
685 | return ERR_PTR(-EINVAL); |
686 | } | |
687 | ||
688 | if (!nr_pages) | |
1da177e4 LT |
689 | return ERR_PTR(-EINVAL); |
690 | ||
691 | bio = bio_alloc(GFP_KERNEL, nr_pages); | |
692 | if (!bio) | |
693 | return ERR_PTR(-ENOMEM); | |
694 | ||
695 | ret = -ENOMEM; | |
11b0b5ab | 696 | pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL); |
1da177e4 LT |
697 | if (!pages) |
698 | goto out; | |
699 | ||
f1970baf JB |
700 | for (i = 0; i < iov_count; i++) { |
701 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
702 | unsigned long len = iov[i].iov_len; | |
703 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
704 | unsigned long start = uaddr >> PAGE_SHIFT; | |
705 | const int local_nr_pages = end - start; | |
706 | const int page_limit = cur_page + local_nr_pages; | |
707 | ||
708 | down_read(¤t->mm->mmap_sem); | |
709 | ret = get_user_pages(current, current->mm, uaddr, | |
710 | local_nr_pages, | |
711 | write_to_vm, 0, &pages[cur_page], NULL); | |
712 | up_read(¤t->mm->mmap_sem); | |
713 | ||
99172157 JA |
714 | if (ret < local_nr_pages) { |
715 | ret = -EFAULT; | |
f1970baf | 716 | goto out_unmap; |
99172157 | 717 | } |
f1970baf JB |
718 | |
719 | offset = uaddr & ~PAGE_MASK; | |
720 | for (j = cur_page; j < page_limit; j++) { | |
721 | unsigned int bytes = PAGE_SIZE - offset; | |
722 | ||
723 | if (len <= 0) | |
724 | break; | |
725 | ||
726 | if (bytes > len) | |
727 | bytes = len; | |
728 | ||
729 | /* | |
730 | * sorry... | |
731 | */ | |
defd94b7 MC |
732 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < |
733 | bytes) | |
f1970baf JB |
734 | break; |
735 | ||
736 | len -= bytes; | |
737 | offset = 0; | |
738 | } | |
1da177e4 | 739 | |
f1970baf | 740 | cur_page = j; |
1da177e4 | 741 | /* |
f1970baf | 742 | * release the pages we didn't map into the bio, if any |
1da177e4 | 743 | */ |
f1970baf JB |
744 | while (j < page_limit) |
745 | page_cache_release(pages[j++]); | |
1da177e4 LT |
746 | } |
747 | ||
1da177e4 LT |
748 | kfree(pages); |
749 | ||
750 | /* | |
751 | * set data direction, and check if mapped pages need bouncing | |
752 | */ | |
753 | if (!write_to_vm) | |
754 | bio->bi_rw |= (1 << BIO_RW); | |
755 | ||
f1970baf | 756 | bio->bi_bdev = bdev; |
1da177e4 LT |
757 | bio->bi_flags |= (1 << BIO_USER_MAPPED); |
758 | return bio; | |
f1970baf JB |
759 | |
760 | out_unmap: | |
761 | for (i = 0; i < nr_pages; i++) { | |
762 | if(!pages[i]) | |
763 | break; | |
764 | page_cache_release(pages[i]); | |
765 | } | |
766 | out: | |
1da177e4 LT |
767 | kfree(pages); |
768 | bio_put(bio); | |
769 | return ERR_PTR(ret); | |
770 | } | |
771 | ||
772 | /** | |
773 | * bio_map_user - map user address into bio | |
165125e1 | 774 | * @q: the struct request_queue for the bio |
1da177e4 LT |
775 | * @bdev: destination block device |
776 | * @uaddr: start of user address | |
777 | * @len: length in bytes | |
778 | * @write_to_vm: bool indicating writing to pages or not | |
779 | * | |
780 | * Map the user space address into a bio suitable for io to a block | |
781 | * device. Returns an error pointer in case of error. | |
782 | */ | |
165125e1 | 783 | struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev, |
1da177e4 | 784 | unsigned long uaddr, unsigned int len, int write_to_vm) |
f1970baf JB |
785 | { |
786 | struct sg_iovec iov; | |
787 | ||
3f70353e | 788 | iov.iov_base = (void __user *)uaddr; |
f1970baf JB |
789 | iov.iov_len = len; |
790 | ||
791 | return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm); | |
792 | } | |
793 | ||
794 | /** | |
795 | * bio_map_user_iov - map user sg_iovec table into bio | |
165125e1 | 796 | * @q: the struct request_queue for the bio |
f1970baf JB |
797 | * @bdev: destination block device |
798 | * @iov: the iovec. | |
799 | * @iov_count: number of elements in the iovec | |
800 | * @write_to_vm: bool indicating writing to pages or not | |
801 | * | |
802 | * Map the user space address into a bio suitable for io to a block | |
803 | * device. Returns an error pointer in case of error. | |
804 | */ | |
165125e1 | 805 | struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev, |
f1970baf JB |
806 | struct sg_iovec *iov, int iov_count, |
807 | int write_to_vm) | |
1da177e4 LT |
808 | { |
809 | struct bio *bio; | |
810 | ||
f1970baf | 811 | bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm); |
1da177e4 LT |
812 | |
813 | if (IS_ERR(bio)) | |
814 | return bio; | |
815 | ||
816 | /* | |
817 | * subtle -- if __bio_map_user() ended up bouncing a bio, | |
818 | * it would normally disappear when its bi_end_io is run. | |
819 | * however, we need it for the unmap, so grab an extra | |
820 | * reference to it | |
821 | */ | |
822 | bio_get(bio); | |
823 | ||
0e75f906 | 824 | return bio; |
1da177e4 LT |
825 | } |
826 | ||
827 | static void __bio_unmap_user(struct bio *bio) | |
828 | { | |
829 | struct bio_vec *bvec; | |
830 | int i; | |
831 | ||
832 | /* | |
833 | * make sure we dirty pages we wrote to | |
834 | */ | |
835 | __bio_for_each_segment(bvec, bio, i, 0) { | |
836 | if (bio_data_dir(bio) == READ) | |
837 | set_page_dirty_lock(bvec->bv_page); | |
838 | ||
839 | page_cache_release(bvec->bv_page); | |
840 | } | |
841 | ||
842 | bio_put(bio); | |
843 | } | |
844 | ||
845 | /** | |
846 | * bio_unmap_user - unmap a bio | |
847 | * @bio: the bio being unmapped | |
848 | * | |
849 | * Unmap a bio previously mapped by bio_map_user(). Must be called with | |
850 | * a process context. | |
851 | * | |
852 | * bio_unmap_user() may sleep. | |
853 | */ | |
854 | void bio_unmap_user(struct bio *bio) | |
855 | { | |
856 | __bio_unmap_user(bio); | |
857 | bio_put(bio); | |
858 | } | |
859 | ||
6712ecf8 | 860 | static void bio_map_kern_endio(struct bio *bio, int err) |
b823825e | 861 | { |
b823825e | 862 | bio_put(bio); |
b823825e JA |
863 | } |
864 | ||
865 | ||
165125e1 | 866 | static struct bio *__bio_map_kern(struct request_queue *q, void *data, |
27496a8c | 867 | unsigned int len, gfp_t gfp_mask) |
df46b9a4 MC |
868 | { |
869 | unsigned long kaddr = (unsigned long)data; | |
870 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
871 | unsigned long start = kaddr >> PAGE_SHIFT; | |
872 | const int nr_pages = end - start; | |
873 | int offset, i; | |
874 | struct bio *bio; | |
875 | ||
876 | bio = bio_alloc(gfp_mask, nr_pages); | |
877 | if (!bio) | |
878 | return ERR_PTR(-ENOMEM); | |
879 | ||
880 | offset = offset_in_page(kaddr); | |
881 | for (i = 0; i < nr_pages; i++) { | |
882 | unsigned int bytes = PAGE_SIZE - offset; | |
883 | ||
884 | if (len <= 0) | |
885 | break; | |
886 | ||
887 | if (bytes > len) | |
888 | bytes = len; | |
889 | ||
defd94b7 MC |
890 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, |
891 | offset) < bytes) | |
df46b9a4 MC |
892 | break; |
893 | ||
894 | data += bytes; | |
895 | len -= bytes; | |
896 | offset = 0; | |
897 | } | |
898 | ||
b823825e | 899 | bio->bi_end_io = bio_map_kern_endio; |
df46b9a4 MC |
900 | return bio; |
901 | } | |
902 | ||
903 | /** | |
904 | * bio_map_kern - map kernel address into bio | |
165125e1 | 905 | * @q: the struct request_queue for the bio |
df46b9a4 MC |
906 | * @data: pointer to buffer to map |
907 | * @len: length in bytes | |
908 | * @gfp_mask: allocation flags for bio allocation | |
909 | * | |
910 | * Map the kernel address into a bio suitable for io to a block | |
911 | * device. Returns an error pointer in case of error. | |
912 | */ | |
165125e1 | 913 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, |
27496a8c | 914 | gfp_t gfp_mask) |
df46b9a4 MC |
915 | { |
916 | struct bio *bio; | |
917 | ||
918 | bio = __bio_map_kern(q, data, len, gfp_mask); | |
919 | if (IS_ERR(bio)) | |
920 | return bio; | |
921 | ||
922 | if (bio->bi_size == len) | |
923 | return bio; | |
924 | ||
925 | /* | |
926 | * Don't support partial mappings. | |
927 | */ | |
928 | bio_put(bio); | |
929 | return ERR_PTR(-EINVAL); | |
930 | } | |
931 | ||
68154e90 FT |
932 | static void bio_copy_kern_endio(struct bio *bio, int err) |
933 | { | |
934 | struct bio_vec *bvec; | |
935 | const int read = bio_data_dir(bio) == READ; | |
936 | char *p = bio->bi_private; | |
937 | int i; | |
938 | ||
939 | __bio_for_each_segment(bvec, bio, i, 0) { | |
940 | char *addr = page_address(bvec->bv_page); | |
941 | ||
942 | if (read && !err) | |
943 | memcpy(p, addr, bvec->bv_len); | |
944 | ||
945 | __free_page(bvec->bv_page); | |
946 | p += bvec->bv_len; | |
947 | } | |
948 | ||
949 | bio_put(bio); | |
950 | } | |
951 | ||
952 | /** | |
953 | * bio_copy_kern - copy kernel address into bio | |
954 | * @q: the struct request_queue for the bio | |
955 | * @data: pointer to buffer to copy | |
956 | * @len: length in bytes | |
957 | * @gfp_mask: allocation flags for bio and page allocation | |
ffee0259 | 958 | * @reading: data direction is READ |
68154e90 FT |
959 | * |
960 | * copy the kernel address into a bio suitable for io to a block | |
961 | * device. Returns an error pointer in case of error. | |
962 | */ | |
963 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
964 | gfp_t gfp_mask, int reading) | |
965 | { | |
966 | unsigned long kaddr = (unsigned long)data; | |
967 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
968 | unsigned long start = kaddr >> PAGE_SHIFT; | |
969 | const int nr_pages = end - start; | |
970 | struct bio *bio; | |
971 | struct bio_vec *bvec; | |
972 | int i, ret; | |
973 | ||
974 | bio = bio_alloc(gfp_mask, nr_pages); | |
975 | if (!bio) | |
976 | return ERR_PTR(-ENOMEM); | |
977 | ||
978 | while (len) { | |
979 | struct page *page; | |
980 | unsigned int bytes = PAGE_SIZE; | |
981 | ||
982 | if (bytes > len) | |
983 | bytes = len; | |
984 | ||
985 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
986 | if (!page) { | |
987 | ret = -ENOMEM; | |
988 | goto cleanup; | |
989 | } | |
990 | ||
991 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) { | |
992 | ret = -EINVAL; | |
993 | goto cleanup; | |
994 | } | |
995 | ||
996 | len -= bytes; | |
997 | } | |
998 | ||
999 | if (!reading) { | |
1000 | void *p = data; | |
1001 | ||
1002 | bio_for_each_segment(bvec, bio, i) { | |
1003 | char *addr = page_address(bvec->bv_page); | |
1004 | ||
1005 | memcpy(addr, p, bvec->bv_len); | |
1006 | p += bvec->bv_len; | |
1007 | } | |
1008 | } | |
1009 | ||
1010 | bio->bi_private = data; | |
1011 | bio->bi_end_io = bio_copy_kern_endio; | |
1012 | return bio; | |
1013 | cleanup: | |
1014 | bio_for_each_segment(bvec, bio, i) | |
1015 | __free_page(bvec->bv_page); | |
1016 | ||
1017 | bio_put(bio); | |
1018 | ||
1019 | return ERR_PTR(ret); | |
1020 | } | |
1021 | ||
1da177e4 LT |
1022 | /* |
1023 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1024 | * for performing direct-IO in BIOs. | |
1025 | * | |
1026 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1027 | * because the required locks are not interrupt-safe. So what we can do is to | |
1028 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1029 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1030 | * in process context. | |
1031 | * | |
1032 | * We special-case compound pages here: normally this means reads into hugetlb | |
1033 | * pages. The logic in here doesn't really work right for compound pages | |
1034 | * because the VM does not uniformly chase down the head page in all cases. | |
1035 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1036 | * handle them at all. So we skip compound pages here at an early stage. | |
1037 | * | |
1038 | * Note that this code is very hard to test under normal circumstances because | |
1039 | * direct-io pins the pages with get_user_pages(). This makes | |
1040 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
1041 | * But other code (eg, pdflush) could clean the pages if they are mapped | |
1042 | * pagecache. | |
1043 | * | |
1044 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1045 | * deferred bio dirtying paths. | |
1046 | */ | |
1047 | ||
1048 | /* | |
1049 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1050 | */ | |
1051 | void bio_set_pages_dirty(struct bio *bio) | |
1052 | { | |
1053 | struct bio_vec *bvec = bio->bi_io_vec; | |
1054 | int i; | |
1055 | ||
1056 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1057 | struct page *page = bvec[i].bv_page; | |
1058 | ||
1059 | if (page && !PageCompound(page)) | |
1060 | set_page_dirty_lock(page); | |
1061 | } | |
1062 | } | |
1063 | ||
86b6c7a7 | 1064 | static void bio_release_pages(struct bio *bio) |
1da177e4 LT |
1065 | { |
1066 | struct bio_vec *bvec = bio->bi_io_vec; | |
1067 | int i; | |
1068 | ||
1069 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1070 | struct page *page = bvec[i].bv_page; | |
1071 | ||
1072 | if (page) | |
1073 | put_page(page); | |
1074 | } | |
1075 | } | |
1076 | ||
1077 | /* | |
1078 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1079 | * If they are, then fine. If, however, some pages are clean then they must | |
1080 | * have been written out during the direct-IO read. So we take another ref on | |
1081 | * the BIO and the offending pages and re-dirty the pages in process context. | |
1082 | * | |
1083 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
1084 | * here on. It will run one page_cache_release() against each page and will | |
1085 | * run one bio_put() against the BIO. | |
1086 | */ | |
1087 | ||
65f27f38 | 1088 | static void bio_dirty_fn(struct work_struct *work); |
1da177e4 | 1089 | |
65f27f38 | 1090 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1da177e4 LT |
1091 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1092 | static struct bio *bio_dirty_list; | |
1093 | ||
1094 | /* | |
1095 | * This runs in process context | |
1096 | */ | |
65f27f38 | 1097 | static void bio_dirty_fn(struct work_struct *work) |
1da177e4 LT |
1098 | { |
1099 | unsigned long flags; | |
1100 | struct bio *bio; | |
1101 | ||
1102 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1103 | bio = bio_dirty_list; | |
1104 | bio_dirty_list = NULL; | |
1105 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1106 | ||
1107 | while (bio) { | |
1108 | struct bio *next = bio->bi_private; | |
1109 | ||
1110 | bio_set_pages_dirty(bio); | |
1111 | bio_release_pages(bio); | |
1112 | bio_put(bio); | |
1113 | bio = next; | |
1114 | } | |
1115 | } | |
1116 | ||
1117 | void bio_check_pages_dirty(struct bio *bio) | |
1118 | { | |
1119 | struct bio_vec *bvec = bio->bi_io_vec; | |
1120 | int nr_clean_pages = 0; | |
1121 | int i; | |
1122 | ||
1123 | for (i = 0; i < bio->bi_vcnt; i++) { | |
1124 | struct page *page = bvec[i].bv_page; | |
1125 | ||
1126 | if (PageDirty(page) || PageCompound(page)) { | |
1127 | page_cache_release(page); | |
1128 | bvec[i].bv_page = NULL; | |
1129 | } else { | |
1130 | nr_clean_pages++; | |
1131 | } | |
1132 | } | |
1133 | ||
1134 | if (nr_clean_pages) { | |
1135 | unsigned long flags; | |
1136 | ||
1137 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1138 | bio->bi_private = bio_dirty_list; | |
1139 | bio_dirty_list = bio; | |
1140 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1141 | schedule_work(&bio_dirty_work); | |
1142 | } else { | |
1143 | bio_put(bio); | |
1144 | } | |
1145 | } | |
1146 | ||
1147 | /** | |
1148 | * bio_endio - end I/O on a bio | |
1149 | * @bio: bio | |
1da177e4 LT |
1150 | * @error: error, if any |
1151 | * | |
1152 | * Description: | |
6712ecf8 | 1153 | * bio_endio() will end I/O on the whole bio. bio_endio() is the |
5bb23a68 N |
1154 | * preferred way to end I/O on a bio, it takes care of clearing |
1155 | * BIO_UPTODATE on error. @error is 0 on success, and and one of the | |
1156 | * established -Exxxx (-EIO, for instance) error values in case | |
1157 | * something went wrong. Noone should call bi_end_io() directly on a | |
1158 | * bio unless they own it and thus know that it has an end_io | |
1159 | * function. | |
1da177e4 | 1160 | **/ |
6712ecf8 | 1161 | void bio_endio(struct bio *bio, int error) |
1da177e4 LT |
1162 | { |
1163 | if (error) | |
1164 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
9cc54d40 N |
1165 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) |
1166 | error = -EIO; | |
1da177e4 | 1167 | |
5bb23a68 | 1168 | if (bio->bi_end_io) |
6712ecf8 | 1169 | bio->bi_end_io(bio, error); |
1da177e4 LT |
1170 | } |
1171 | ||
1172 | void bio_pair_release(struct bio_pair *bp) | |
1173 | { | |
1174 | if (atomic_dec_and_test(&bp->cnt)) { | |
1175 | struct bio *master = bp->bio1.bi_private; | |
1176 | ||
6712ecf8 | 1177 | bio_endio(master, bp->error); |
1da177e4 LT |
1178 | mempool_free(bp, bp->bio2.bi_private); |
1179 | } | |
1180 | } | |
1181 | ||
6712ecf8 | 1182 | static void bio_pair_end_1(struct bio *bi, int err) |
1da177e4 LT |
1183 | { |
1184 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); | |
1185 | ||
1186 | if (err) | |
1187 | bp->error = err; | |
1188 | ||
1da177e4 | 1189 | bio_pair_release(bp); |
1da177e4 LT |
1190 | } |
1191 | ||
6712ecf8 | 1192 | static void bio_pair_end_2(struct bio *bi, int err) |
1da177e4 LT |
1193 | { |
1194 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); | |
1195 | ||
1196 | if (err) | |
1197 | bp->error = err; | |
1198 | ||
1da177e4 | 1199 | bio_pair_release(bp); |
1da177e4 LT |
1200 | } |
1201 | ||
1202 | /* | |
1203 | * split a bio - only worry about a bio with a single page | |
1204 | * in it's iovec | |
1205 | */ | |
1206 | struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors) | |
1207 | { | |
1208 | struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO); | |
1209 | ||
1210 | if (!bp) | |
1211 | return bp; | |
1212 | ||
2056a782 JA |
1213 | blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi, |
1214 | bi->bi_sector + first_sectors); | |
1215 | ||
1da177e4 LT |
1216 | BUG_ON(bi->bi_vcnt != 1); |
1217 | BUG_ON(bi->bi_idx != 0); | |
1218 | atomic_set(&bp->cnt, 3); | |
1219 | bp->error = 0; | |
1220 | bp->bio1 = *bi; | |
1221 | bp->bio2 = *bi; | |
1222 | bp->bio2.bi_sector += first_sectors; | |
1223 | bp->bio2.bi_size -= first_sectors << 9; | |
1224 | bp->bio1.bi_size = first_sectors << 9; | |
1225 | ||
1226 | bp->bv1 = bi->bi_io_vec[0]; | |
1227 | bp->bv2 = bi->bi_io_vec[0]; | |
1228 | bp->bv2.bv_offset += first_sectors << 9; | |
1229 | bp->bv2.bv_len -= first_sectors << 9; | |
1230 | bp->bv1.bv_len = first_sectors << 9; | |
1231 | ||
1232 | bp->bio1.bi_io_vec = &bp->bv1; | |
1233 | bp->bio2.bi_io_vec = &bp->bv2; | |
1234 | ||
a2eb0c10 N |
1235 | bp->bio1.bi_max_vecs = 1; |
1236 | bp->bio2.bi_max_vecs = 1; | |
1237 | ||
1da177e4 LT |
1238 | bp->bio1.bi_end_io = bio_pair_end_1; |
1239 | bp->bio2.bi_end_io = bio_pair_end_2; | |
1240 | ||
1241 | bp->bio1.bi_private = bi; | |
1242 | bp->bio2.bi_private = pool; | |
1243 | ||
7ba1ba12 MP |
1244 | if (bio_integrity(bi)) |
1245 | bio_integrity_split(bi, bp, first_sectors); | |
1246 | ||
1da177e4 LT |
1247 | return bp; |
1248 | } | |
1249 | ||
1da177e4 LT |
1250 | |
1251 | /* | |
1252 | * create memory pools for biovec's in a bio_set. | |
1253 | * use the global biovec slabs created for general use. | |
1254 | */ | |
5972511b | 1255 | static int biovec_create_pools(struct bio_set *bs, int pool_entries) |
1da177e4 LT |
1256 | { |
1257 | int i; | |
1258 | ||
1259 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1260 | struct biovec_slab *bp = bvec_slabs + i; | |
1261 | mempool_t **bvp = bs->bvec_pools + i; | |
1262 | ||
93d2341c | 1263 | *bvp = mempool_create_slab_pool(pool_entries, bp->slab); |
1da177e4 LT |
1264 | if (!*bvp) |
1265 | return -ENOMEM; | |
1266 | } | |
1267 | return 0; | |
1268 | } | |
1269 | ||
1270 | static void biovec_free_pools(struct bio_set *bs) | |
1271 | { | |
1272 | int i; | |
1273 | ||
1274 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1275 | mempool_t *bvp = bs->bvec_pools[i]; | |
1276 | ||
1277 | if (bvp) | |
1278 | mempool_destroy(bvp); | |
1279 | } | |
1280 | ||
1281 | } | |
1282 | ||
1283 | void bioset_free(struct bio_set *bs) | |
1284 | { | |
1285 | if (bs->bio_pool) | |
1286 | mempool_destroy(bs->bio_pool); | |
1287 | ||
7ba1ba12 | 1288 | bioset_integrity_free(bs); |
1da177e4 LT |
1289 | biovec_free_pools(bs); |
1290 | ||
1291 | kfree(bs); | |
1292 | } | |
1293 | ||
5972511b | 1294 | struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size) |
1da177e4 | 1295 | { |
11b0b5ab | 1296 | struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL); |
1da177e4 LT |
1297 | |
1298 | if (!bs) | |
1299 | return NULL; | |
1300 | ||
93d2341c | 1301 | bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab); |
1da177e4 LT |
1302 | if (!bs->bio_pool) |
1303 | goto bad; | |
1304 | ||
7ba1ba12 MP |
1305 | if (bioset_integrity_create(bs, bio_pool_size)) |
1306 | goto bad; | |
1307 | ||
5972511b | 1308 | if (!biovec_create_pools(bs, bvec_pool_size)) |
1da177e4 LT |
1309 | return bs; |
1310 | ||
1311 | bad: | |
1312 | bioset_free(bs); | |
1313 | return NULL; | |
1314 | } | |
1315 | ||
1316 | static void __init biovec_init_slabs(void) | |
1317 | { | |
1318 | int i; | |
1319 | ||
1320 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1321 | int size; | |
1322 | struct biovec_slab *bvs = bvec_slabs + i; | |
1323 | ||
1324 | size = bvs->nr_vecs * sizeof(struct bio_vec); | |
1325 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
20c2df83 | 1326 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
1327 | } |
1328 | } | |
1329 | ||
1330 | static int __init init_bio(void) | |
1331 | { | |
0a31bd5f | 1332 | bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC); |
1da177e4 | 1333 | |
7ba1ba12 | 1334 | bio_integrity_init_slab(); |
1da177e4 LT |
1335 | biovec_init_slabs(); |
1336 | ||
5972511b | 1337 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 2); |
1da177e4 LT |
1338 | if (!fs_bio_set) |
1339 | panic("bio: can't allocate bios\n"); | |
1340 | ||
0eaae62a MD |
1341 | bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES, |
1342 | sizeof(struct bio_pair)); | |
1da177e4 LT |
1343 | if (!bio_split_pool) |
1344 | panic("bio: can't create split pool\n"); | |
1345 | ||
1346 | return 0; | |
1347 | } | |
1348 | ||
1349 | subsys_initcall(init_bio); | |
1350 | ||
1351 | EXPORT_SYMBOL(bio_alloc); | |
1352 | EXPORT_SYMBOL(bio_put); | |
3676347a | 1353 | EXPORT_SYMBOL(bio_free); |
1da177e4 LT |
1354 | EXPORT_SYMBOL(bio_endio); |
1355 | EXPORT_SYMBOL(bio_init); | |
1356 | EXPORT_SYMBOL(__bio_clone); | |
1357 | EXPORT_SYMBOL(bio_clone); | |
1358 | EXPORT_SYMBOL(bio_phys_segments); | |
1359 | EXPORT_SYMBOL(bio_hw_segments); | |
1360 | EXPORT_SYMBOL(bio_add_page); | |
6e68af66 | 1361 | EXPORT_SYMBOL(bio_add_pc_page); |
1da177e4 | 1362 | EXPORT_SYMBOL(bio_get_nr_vecs); |
40044ce0 JA |
1363 | EXPORT_SYMBOL(bio_map_user); |
1364 | EXPORT_SYMBOL(bio_unmap_user); | |
df46b9a4 | 1365 | EXPORT_SYMBOL(bio_map_kern); |
68154e90 | 1366 | EXPORT_SYMBOL(bio_copy_kern); |
1da177e4 LT |
1367 | EXPORT_SYMBOL(bio_pair_release); |
1368 | EXPORT_SYMBOL(bio_split); | |
1369 | EXPORT_SYMBOL(bio_split_pool); | |
1370 | EXPORT_SYMBOL(bio_copy_user); | |
1371 | EXPORT_SYMBOL(bio_uncopy_user); | |
1372 | EXPORT_SYMBOL(bioset_create); | |
1373 | EXPORT_SYMBOL(bioset_free); | |
1374 | EXPORT_SYMBOL(bio_alloc_bioset); |