]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> | |
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/uio.h> | |
23 | #include <linux/iocontext.h> | |
24 | #include <linux/slab.h> | |
25 | #include <linux/init.h> | |
26 | #include <linux/kernel.h> | |
27 | #include <linux/export.h> | |
28 | #include <linux/mempool.h> | |
29 | #include <linux/workqueue.h> | |
30 | #include <linux/cgroup.h> | |
31 | #include <scsi/sg.h> /* for struct sg_iovec */ | |
32 | ||
33 | #include <trace/events/block.h> | |
34 | ||
35 | /* | |
36 | * Test patch to inline a certain number of bi_io_vec's inside the bio | |
37 | * itself, to shrink a bio data allocation from two mempool calls to one | |
38 | */ | |
39 | #define BIO_INLINE_VECS 4 | |
40 | ||
41 | static mempool_t *bio_split_pool __read_mostly; | |
42 | ||
43 | /* | |
44 | * if you change this list, also change bvec_alloc or things will | |
45 | * break badly! cannot be bigger than what you can fit into an | |
46 | * unsigned short | |
47 | */ | |
48 | #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } | |
49 | static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { | |
50 | BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), | |
51 | }; | |
52 | #undef BV | |
53 | ||
54 | /* | |
55 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
56 | * IO code that does not need private memory pools. | |
57 | */ | |
58 | struct bio_set *fs_bio_set; | |
59 | EXPORT_SYMBOL(fs_bio_set); | |
60 | ||
61 | /* | |
62 | * Our slab pool management | |
63 | */ | |
64 | struct bio_slab { | |
65 | struct kmem_cache *slab; | |
66 | unsigned int slab_ref; | |
67 | unsigned int slab_size; | |
68 | char name[8]; | |
69 | }; | |
70 | static DEFINE_MUTEX(bio_slab_lock); | |
71 | static struct bio_slab *bio_slabs; | |
72 | static unsigned int bio_slab_nr, bio_slab_max; | |
73 | ||
74 | static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) | |
75 | { | |
76 | unsigned int sz = sizeof(struct bio) + extra_size; | |
77 | struct kmem_cache *slab = NULL; | |
78 | struct bio_slab *bslab, *new_bio_slabs; | |
79 | unsigned int new_bio_slab_max; | |
80 | unsigned int i, entry = -1; | |
81 | ||
82 | mutex_lock(&bio_slab_lock); | |
83 | ||
84 | i = 0; | |
85 | while (i < bio_slab_nr) { | |
86 | bslab = &bio_slabs[i]; | |
87 | ||
88 | if (!bslab->slab && entry == -1) | |
89 | entry = i; | |
90 | else if (bslab->slab_size == sz) { | |
91 | slab = bslab->slab; | |
92 | bslab->slab_ref++; | |
93 | break; | |
94 | } | |
95 | i++; | |
96 | } | |
97 | ||
98 | if (slab) | |
99 | goto out_unlock; | |
100 | ||
101 | if (bio_slab_nr == bio_slab_max && entry == -1) { | |
102 | new_bio_slab_max = bio_slab_max << 1; | |
103 | new_bio_slabs = krealloc(bio_slabs, | |
104 | new_bio_slab_max * sizeof(struct bio_slab), | |
105 | GFP_KERNEL); | |
106 | if (!new_bio_slabs) | |
107 | goto out_unlock; | |
108 | bio_slab_max = new_bio_slab_max; | |
109 | bio_slabs = new_bio_slabs; | |
110 | } | |
111 | if (entry == -1) | |
112 | entry = bio_slab_nr++; | |
113 | ||
114 | bslab = &bio_slabs[entry]; | |
115 | ||
116 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); | |
117 | slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL); | |
118 | if (!slab) | |
119 | goto out_unlock; | |
120 | ||
121 | printk(KERN_INFO "bio: create slab <%s> at %d\n", bslab->name, entry); | |
122 | bslab->slab = slab; | |
123 | bslab->slab_ref = 1; | |
124 | bslab->slab_size = sz; | |
125 | out_unlock: | |
126 | mutex_unlock(&bio_slab_lock); | |
127 | return slab; | |
128 | } | |
129 | ||
130 | static void bio_put_slab(struct bio_set *bs) | |
131 | { | |
132 | struct bio_slab *bslab = NULL; | |
133 | unsigned int i; | |
134 | ||
135 | mutex_lock(&bio_slab_lock); | |
136 | ||
137 | for (i = 0; i < bio_slab_nr; i++) { | |
138 | if (bs->bio_slab == bio_slabs[i].slab) { | |
139 | bslab = &bio_slabs[i]; | |
140 | break; | |
141 | } | |
142 | } | |
143 | ||
144 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) | |
145 | goto out; | |
146 | ||
147 | WARN_ON(!bslab->slab_ref); | |
148 | ||
149 | if (--bslab->slab_ref) | |
150 | goto out; | |
151 | ||
152 | kmem_cache_destroy(bslab->slab); | |
153 | bslab->slab = NULL; | |
154 | ||
155 | out: | |
156 | mutex_unlock(&bio_slab_lock); | |
157 | } | |
158 | ||
159 | unsigned int bvec_nr_vecs(unsigned short idx) | |
160 | { | |
161 | return bvec_slabs[idx].nr_vecs; | |
162 | } | |
163 | ||
164 | void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) | |
165 | { | |
166 | BIO_BUG_ON(idx >= BIOVEC_NR_POOLS); | |
167 | ||
168 | if (idx == BIOVEC_MAX_IDX) | |
169 | mempool_free(bv, pool); | |
170 | else { | |
171 | struct biovec_slab *bvs = bvec_slabs + idx; | |
172 | ||
173 | kmem_cache_free(bvs->slab, bv); | |
174 | } | |
175 | } | |
176 | ||
177 | struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, | |
178 | mempool_t *pool) | |
179 | { | |
180 | struct bio_vec *bvl; | |
181 | ||
182 | /* | |
183 | * see comment near bvec_array define! | |
184 | */ | |
185 | switch (nr) { | |
186 | case 1: | |
187 | *idx = 0; | |
188 | break; | |
189 | case 2 ... 4: | |
190 | *idx = 1; | |
191 | break; | |
192 | case 5 ... 16: | |
193 | *idx = 2; | |
194 | break; | |
195 | case 17 ... 64: | |
196 | *idx = 3; | |
197 | break; | |
198 | case 65 ... 128: | |
199 | *idx = 4; | |
200 | break; | |
201 | case 129 ... BIO_MAX_PAGES: | |
202 | *idx = 5; | |
203 | break; | |
204 | default: | |
205 | return NULL; | |
206 | } | |
207 | ||
208 | /* | |
209 | * idx now points to the pool we want to allocate from. only the | |
210 | * 1-vec entry pool is mempool backed. | |
211 | */ | |
212 | if (*idx == BIOVEC_MAX_IDX) { | |
213 | fallback: | |
214 | bvl = mempool_alloc(pool, gfp_mask); | |
215 | } else { | |
216 | struct biovec_slab *bvs = bvec_slabs + *idx; | |
217 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO); | |
218 | ||
219 | /* | |
220 | * Make this allocation restricted and don't dump info on | |
221 | * allocation failures, since we'll fallback to the mempool | |
222 | * in case of failure. | |
223 | */ | |
224 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; | |
225 | ||
226 | /* | |
227 | * Try a slab allocation. If this fails and __GFP_WAIT | |
228 | * is set, retry with the 1-entry mempool | |
229 | */ | |
230 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); | |
231 | if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) { | |
232 | *idx = BIOVEC_MAX_IDX; | |
233 | goto fallback; | |
234 | } | |
235 | } | |
236 | ||
237 | return bvl; | |
238 | } | |
239 | ||
240 | static void __bio_free(struct bio *bio) | |
241 | { | |
242 | bio_disassociate_task(bio); | |
243 | ||
244 | if (bio_integrity(bio)) | |
245 | bio_integrity_free(bio); | |
246 | } | |
247 | ||
248 | static void bio_free(struct bio *bio) | |
249 | { | |
250 | struct bio_set *bs = bio->bi_pool; | |
251 | void *p; | |
252 | ||
253 | __bio_free(bio); | |
254 | ||
255 | if (bs) { | |
256 | if (bio_flagged(bio, BIO_OWNS_VEC)) | |
257 | bvec_free(bs->bvec_pool, bio->bi_io_vec, BIO_POOL_IDX(bio)); | |
258 | ||
259 | /* | |
260 | * If we have front padding, adjust the bio pointer before freeing | |
261 | */ | |
262 | p = bio; | |
263 | p -= bs->front_pad; | |
264 | ||
265 | mempool_free(p, bs->bio_pool); | |
266 | } else { | |
267 | /* Bio was allocated by bio_kmalloc() */ | |
268 | kfree(bio); | |
269 | } | |
270 | } | |
271 | ||
272 | void bio_init(struct bio *bio) | |
273 | { | |
274 | memset(bio, 0, sizeof(*bio)); | |
275 | bio->bi_flags = 1 << BIO_UPTODATE; | |
276 | atomic_set(&bio->bi_cnt, 1); | |
277 | } | |
278 | EXPORT_SYMBOL(bio_init); | |
279 | ||
280 | /** | |
281 | * bio_reset - reinitialize a bio | |
282 | * @bio: bio to reset | |
283 | * | |
284 | * Description: | |
285 | * After calling bio_reset(), @bio will be in the same state as a freshly | |
286 | * allocated bio returned bio bio_alloc_bioset() - the only fields that are | |
287 | * preserved are the ones that are initialized by bio_alloc_bioset(). See | |
288 | * comment in struct bio. | |
289 | */ | |
290 | void bio_reset(struct bio *bio) | |
291 | { | |
292 | unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); | |
293 | ||
294 | __bio_free(bio); | |
295 | ||
296 | memset(bio, 0, BIO_RESET_BYTES); | |
297 | bio->bi_flags = flags|(1 << BIO_UPTODATE); | |
298 | } | |
299 | EXPORT_SYMBOL(bio_reset); | |
300 | ||
301 | static void bio_alloc_rescue(struct work_struct *work) | |
302 | { | |
303 | struct bio_set *bs = container_of(work, struct bio_set, rescue_work); | |
304 | struct bio *bio; | |
305 | ||
306 | while (1) { | |
307 | spin_lock(&bs->rescue_lock); | |
308 | bio = bio_list_pop(&bs->rescue_list); | |
309 | spin_unlock(&bs->rescue_lock); | |
310 | ||
311 | if (!bio) | |
312 | break; | |
313 | ||
314 | generic_make_request(bio); | |
315 | } | |
316 | } | |
317 | ||
318 | static void punt_bios_to_rescuer(struct bio_set *bs) | |
319 | { | |
320 | struct bio_list punt, nopunt; | |
321 | struct bio *bio; | |
322 | ||
323 | /* | |
324 | * In order to guarantee forward progress we must punt only bios that | |
325 | * were allocated from this bio_set; otherwise, if there was a bio on | |
326 | * there for a stacking driver higher up in the stack, processing it | |
327 | * could require allocating bios from this bio_set, and doing that from | |
328 | * our own rescuer would be bad. | |
329 | * | |
330 | * Since bio lists are singly linked, pop them all instead of trying to | |
331 | * remove from the middle of the list: | |
332 | */ | |
333 | ||
334 | bio_list_init(&punt); | |
335 | bio_list_init(&nopunt); | |
336 | ||
337 | while ((bio = bio_list_pop(current->bio_list))) | |
338 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); | |
339 | ||
340 | *current->bio_list = nopunt; | |
341 | ||
342 | spin_lock(&bs->rescue_lock); | |
343 | bio_list_merge(&bs->rescue_list, &punt); | |
344 | spin_unlock(&bs->rescue_lock); | |
345 | ||
346 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
347 | } | |
348 | ||
349 | /** | |
350 | * bio_alloc_bioset - allocate a bio for I/O | |
351 | * @gfp_mask: the GFP_ mask given to the slab allocator | |
352 | * @nr_iovecs: number of iovecs to pre-allocate | |
353 | * @bs: the bio_set to allocate from. | |
354 | * | |
355 | * Description: | |
356 | * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is | |
357 | * backed by the @bs's mempool. | |
358 | * | |
359 | * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be | |
360 | * able to allocate a bio. This is due to the mempool guarantees. To make this | |
361 | * work, callers must never allocate more than 1 bio at a time from this pool. | |
362 | * Callers that need to allocate more than 1 bio must always submit the | |
363 | * previously allocated bio for IO before attempting to allocate a new one. | |
364 | * Failure to do so can cause deadlocks under memory pressure. | |
365 | * | |
366 | * Note that when running under generic_make_request() (i.e. any block | |
367 | * driver), bios are not submitted until after you return - see the code in | |
368 | * generic_make_request() that converts recursion into iteration, to prevent | |
369 | * stack overflows. | |
370 | * | |
371 | * This would normally mean allocating multiple bios under | |
372 | * generic_make_request() would be susceptible to deadlocks, but we have | |
373 | * deadlock avoidance code that resubmits any blocked bios from a rescuer | |
374 | * thread. | |
375 | * | |
376 | * However, we do not guarantee forward progress for allocations from other | |
377 | * mempools. Doing multiple allocations from the same mempool under | |
378 | * generic_make_request() should be avoided - instead, use bio_set's front_pad | |
379 | * for per bio allocations. | |
380 | * | |
381 | * RETURNS: | |
382 | * Pointer to new bio on success, NULL on failure. | |
383 | */ | |
384 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) | |
385 | { | |
386 | gfp_t saved_gfp = gfp_mask; | |
387 | unsigned front_pad; | |
388 | unsigned inline_vecs; | |
389 | unsigned long idx = BIO_POOL_NONE; | |
390 | struct bio_vec *bvl = NULL; | |
391 | struct bio *bio; | |
392 | void *p; | |
393 | ||
394 | if (!bs) { | |
395 | if (nr_iovecs > UIO_MAXIOV) | |
396 | return NULL; | |
397 | ||
398 | p = kmalloc(sizeof(struct bio) + | |
399 | nr_iovecs * sizeof(struct bio_vec), | |
400 | gfp_mask); | |
401 | front_pad = 0; | |
402 | inline_vecs = nr_iovecs; | |
403 | } else { | |
404 | /* | |
405 | * generic_make_request() converts recursion to iteration; this | |
406 | * means if we're running beneath it, any bios we allocate and | |
407 | * submit will not be submitted (and thus freed) until after we | |
408 | * return. | |
409 | * | |
410 | * This exposes us to a potential deadlock if we allocate | |
411 | * multiple bios from the same bio_set() while running | |
412 | * underneath generic_make_request(). If we were to allocate | |
413 | * multiple bios (say a stacking block driver that was splitting | |
414 | * bios), we would deadlock if we exhausted the mempool's | |
415 | * reserve. | |
416 | * | |
417 | * We solve this, and guarantee forward progress, with a rescuer | |
418 | * workqueue per bio_set. If we go to allocate and there are | |
419 | * bios on current->bio_list, we first try the allocation | |
420 | * without __GFP_WAIT; if that fails, we punt those bios we | |
421 | * would be blocking to the rescuer workqueue before we retry | |
422 | * with the original gfp_flags. | |
423 | */ | |
424 | ||
425 | if (current->bio_list && !bio_list_empty(current->bio_list)) | |
426 | gfp_mask &= ~__GFP_WAIT; | |
427 | ||
428 | p = mempool_alloc(bs->bio_pool, gfp_mask); | |
429 | if (!p && gfp_mask != saved_gfp) { | |
430 | punt_bios_to_rescuer(bs); | |
431 | gfp_mask = saved_gfp; | |
432 | p = mempool_alloc(bs->bio_pool, gfp_mask); | |
433 | } | |
434 | ||
435 | front_pad = bs->front_pad; | |
436 | inline_vecs = BIO_INLINE_VECS; | |
437 | } | |
438 | ||
439 | if (unlikely(!p)) | |
440 | return NULL; | |
441 | ||
442 | bio = p + front_pad; | |
443 | bio_init(bio); | |
444 | ||
445 | if (nr_iovecs > inline_vecs) { | |
446 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); | |
447 | if (!bvl && gfp_mask != saved_gfp) { | |
448 | punt_bios_to_rescuer(bs); | |
449 | gfp_mask = saved_gfp; | |
450 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); | |
451 | } | |
452 | ||
453 | if (unlikely(!bvl)) | |
454 | goto err_free; | |
455 | ||
456 | bio->bi_flags |= 1 << BIO_OWNS_VEC; | |
457 | } else if (nr_iovecs) { | |
458 | bvl = bio->bi_inline_vecs; | |
459 | } | |
460 | ||
461 | bio->bi_pool = bs; | |
462 | bio->bi_flags |= idx << BIO_POOL_OFFSET; | |
463 | bio->bi_max_vecs = nr_iovecs; | |
464 | bio->bi_io_vec = bvl; | |
465 | return bio; | |
466 | ||
467 | err_free: | |
468 | mempool_free(p, bs->bio_pool); | |
469 | return NULL; | |
470 | } | |
471 | EXPORT_SYMBOL(bio_alloc_bioset); | |
472 | ||
473 | void zero_fill_bio(struct bio *bio) | |
474 | { | |
475 | unsigned long flags; | |
476 | struct bio_vec *bv; | |
477 | int i; | |
478 | ||
479 | bio_for_each_segment(bv, bio, i) { | |
480 | char *data = bvec_kmap_irq(bv, &flags); | |
481 | memset(data, 0, bv->bv_len); | |
482 | flush_dcache_page(bv->bv_page); | |
483 | bvec_kunmap_irq(data, &flags); | |
484 | } | |
485 | } | |
486 | EXPORT_SYMBOL(zero_fill_bio); | |
487 | ||
488 | /** | |
489 | * bio_put - release a reference to a bio | |
490 | * @bio: bio to release reference to | |
491 | * | |
492 | * Description: | |
493 | * Put a reference to a &struct bio, either one you have gotten with | |
494 | * bio_alloc, bio_get or bio_clone. The last put of a bio will free it. | |
495 | **/ | |
496 | void bio_put(struct bio *bio) | |
497 | { | |
498 | BIO_BUG_ON(!atomic_read(&bio->bi_cnt)); | |
499 | ||
500 | /* | |
501 | * last put frees it | |
502 | */ | |
503 | if (atomic_dec_and_test(&bio->bi_cnt)) | |
504 | bio_free(bio); | |
505 | } | |
506 | EXPORT_SYMBOL(bio_put); | |
507 | ||
508 | inline int bio_phys_segments(struct request_queue *q, struct bio *bio) | |
509 | { | |
510 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
511 | blk_recount_segments(q, bio); | |
512 | ||
513 | return bio->bi_phys_segments; | |
514 | } | |
515 | EXPORT_SYMBOL(bio_phys_segments); | |
516 | ||
517 | /** | |
518 | * __bio_clone - clone a bio | |
519 | * @bio: destination bio | |
520 | * @bio_src: bio to clone | |
521 | * | |
522 | * Clone a &bio. Caller will own the returned bio, but not | |
523 | * the actual data it points to. Reference count of returned | |
524 | * bio will be one. | |
525 | */ | |
526 | void __bio_clone(struct bio *bio, struct bio *bio_src) | |
527 | { | |
528 | memcpy(bio->bi_io_vec, bio_src->bi_io_vec, | |
529 | bio_src->bi_max_vecs * sizeof(struct bio_vec)); | |
530 | ||
531 | /* | |
532 | * most users will be overriding ->bi_bdev with a new target, | |
533 | * so we don't set nor calculate new physical/hw segment counts here | |
534 | */ | |
535 | bio->bi_sector = bio_src->bi_sector; | |
536 | bio->bi_bdev = bio_src->bi_bdev; | |
537 | bio->bi_flags |= 1 << BIO_CLONED; | |
538 | bio->bi_rw = bio_src->bi_rw; | |
539 | bio->bi_vcnt = bio_src->bi_vcnt; | |
540 | bio->bi_size = bio_src->bi_size; | |
541 | bio->bi_idx = bio_src->bi_idx; | |
542 | } | |
543 | EXPORT_SYMBOL(__bio_clone); | |
544 | ||
545 | /** | |
546 | * bio_clone_bioset - clone a bio | |
547 | * @bio: bio to clone | |
548 | * @gfp_mask: allocation priority | |
549 | * @bs: bio_set to allocate from | |
550 | * | |
551 | * Like __bio_clone, only also allocates the returned bio | |
552 | */ | |
553 | struct bio *bio_clone_bioset(struct bio *bio, gfp_t gfp_mask, | |
554 | struct bio_set *bs) | |
555 | { | |
556 | struct bio *b; | |
557 | ||
558 | b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, bs); | |
559 | if (!b) | |
560 | return NULL; | |
561 | ||
562 | __bio_clone(b, bio); | |
563 | ||
564 | if (bio_integrity(bio)) { | |
565 | int ret; | |
566 | ||
567 | ret = bio_integrity_clone(b, bio, gfp_mask); | |
568 | ||
569 | if (ret < 0) { | |
570 | bio_put(b); | |
571 | return NULL; | |
572 | } | |
573 | } | |
574 | ||
575 | return b; | |
576 | } | |
577 | EXPORT_SYMBOL(bio_clone_bioset); | |
578 | ||
579 | /** | |
580 | * bio_get_nr_vecs - return approx number of vecs | |
581 | * @bdev: I/O target | |
582 | * | |
583 | * Return the approximate number of pages we can send to this target. | |
584 | * There's no guarantee that you will be able to fit this number of pages | |
585 | * into a bio, it does not account for dynamic restrictions that vary | |
586 | * on offset. | |
587 | */ | |
588 | int bio_get_nr_vecs(struct block_device *bdev) | |
589 | { | |
590 | struct request_queue *q = bdev_get_queue(bdev); | |
591 | int nr_pages; | |
592 | ||
593 | nr_pages = min_t(unsigned, | |
594 | queue_max_segments(q), | |
595 | queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1); | |
596 | ||
597 | return min_t(unsigned, nr_pages, BIO_MAX_PAGES); | |
598 | ||
599 | } | |
600 | EXPORT_SYMBOL(bio_get_nr_vecs); | |
601 | ||
602 | static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page | |
603 | *page, unsigned int len, unsigned int offset, | |
604 | unsigned short max_sectors) | |
605 | { | |
606 | int retried_segments = 0; | |
607 | struct bio_vec *bvec; | |
608 | ||
609 | /* | |
610 | * cloned bio must not modify vec list | |
611 | */ | |
612 | if (unlikely(bio_flagged(bio, BIO_CLONED))) | |
613 | return 0; | |
614 | ||
615 | if (((bio->bi_size + len) >> 9) > max_sectors) | |
616 | return 0; | |
617 | ||
618 | /* | |
619 | * For filesystems with a blocksize smaller than the pagesize | |
620 | * we will often be called with the same page as last time and | |
621 | * a consecutive offset. Optimize this special case. | |
622 | */ | |
623 | if (bio->bi_vcnt > 0) { | |
624 | struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
625 | ||
626 | if (page == prev->bv_page && | |
627 | offset == prev->bv_offset + prev->bv_len) { | |
628 | unsigned int prev_bv_len = prev->bv_len; | |
629 | prev->bv_len += len; | |
630 | ||
631 | if (q->merge_bvec_fn) { | |
632 | struct bvec_merge_data bvm = { | |
633 | /* prev_bvec is already charged in | |
634 | bi_size, discharge it in order to | |
635 | simulate merging updated prev_bvec | |
636 | as new bvec. */ | |
637 | .bi_bdev = bio->bi_bdev, | |
638 | .bi_sector = bio->bi_sector, | |
639 | .bi_size = bio->bi_size - prev_bv_len, | |
640 | .bi_rw = bio->bi_rw, | |
641 | }; | |
642 | ||
643 | if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) { | |
644 | prev->bv_len -= len; | |
645 | return 0; | |
646 | } | |
647 | } | |
648 | ||
649 | goto done; | |
650 | } | |
651 | } | |
652 | ||
653 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
654 | return 0; | |
655 | ||
656 | /* | |
657 | * we might lose a segment or two here, but rather that than | |
658 | * make this too complex. | |
659 | */ | |
660 | ||
661 | while (bio->bi_phys_segments >= queue_max_segments(q)) { | |
662 | ||
663 | if (retried_segments) | |
664 | return 0; | |
665 | ||
666 | retried_segments = 1; | |
667 | blk_recount_segments(q, bio); | |
668 | } | |
669 | ||
670 | /* | |
671 | * setup the new entry, we might clear it again later if we | |
672 | * cannot add the page | |
673 | */ | |
674 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; | |
675 | bvec->bv_page = page; | |
676 | bvec->bv_len = len; | |
677 | bvec->bv_offset = offset; | |
678 | ||
679 | /* | |
680 | * if queue has other restrictions (eg varying max sector size | |
681 | * depending on offset), it can specify a merge_bvec_fn in the | |
682 | * queue to get further control | |
683 | */ | |
684 | if (q->merge_bvec_fn) { | |
685 | struct bvec_merge_data bvm = { | |
686 | .bi_bdev = bio->bi_bdev, | |
687 | .bi_sector = bio->bi_sector, | |
688 | .bi_size = bio->bi_size, | |
689 | .bi_rw = bio->bi_rw, | |
690 | }; | |
691 | ||
692 | /* | |
693 | * merge_bvec_fn() returns number of bytes it can accept | |
694 | * at this offset | |
695 | */ | |
696 | if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) { | |
697 | bvec->bv_page = NULL; | |
698 | bvec->bv_len = 0; | |
699 | bvec->bv_offset = 0; | |
700 | return 0; | |
701 | } | |
702 | } | |
703 | ||
704 | /* If we may be able to merge these biovecs, force a recount */ | |
705 | if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) | |
706 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
707 | ||
708 | bio->bi_vcnt++; | |
709 | bio->bi_phys_segments++; | |
710 | done: | |
711 | bio->bi_size += len; | |
712 | return len; | |
713 | } | |
714 | ||
715 | /** | |
716 | * bio_add_pc_page - attempt to add page to bio | |
717 | * @q: the target queue | |
718 | * @bio: destination bio | |
719 | * @page: page to add | |
720 | * @len: vec entry length | |
721 | * @offset: vec entry offset | |
722 | * | |
723 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
724 | * number of reasons, such as the bio being full or target block device | |
725 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
726 | * so it is always possible to add a single page to an empty bio. | |
727 | * | |
728 | * This should only be used by REQ_PC bios. | |
729 | */ | |
730 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page, | |
731 | unsigned int len, unsigned int offset) | |
732 | { | |
733 | return __bio_add_page(q, bio, page, len, offset, | |
734 | queue_max_hw_sectors(q)); | |
735 | } | |
736 | EXPORT_SYMBOL(bio_add_pc_page); | |
737 | ||
738 | /** | |
739 | * bio_add_page - attempt to add page to bio | |
740 | * @bio: destination bio | |
741 | * @page: page to add | |
742 | * @len: vec entry length | |
743 | * @offset: vec entry offset | |
744 | * | |
745 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
746 | * number of reasons, such as the bio being full or target block device | |
747 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
748 | * so it is always possible to add a single page to an empty bio. | |
749 | */ | |
750 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, | |
751 | unsigned int offset) | |
752 | { | |
753 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); | |
754 | return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q)); | |
755 | } | |
756 | EXPORT_SYMBOL(bio_add_page); | |
757 | ||
758 | struct submit_bio_ret { | |
759 | struct completion event; | |
760 | int error; | |
761 | }; | |
762 | ||
763 | static void submit_bio_wait_endio(struct bio *bio, int error) | |
764 | { | |
765 | struct submit_bio_ret *ret = bio->bi_private; | |
766 | ||
767 | ret->error = error; | |
768 | complete(&ret->event); | |
769 | } | |
770 | ||
771 | /** | |
772 | * submit_bio_wait - submit a bio, and wait until it completes | |
773 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | |
774 | * @bio: The &struct bio which describes the I/O | |
775 | * | |
776 | * Simple wrapper around submit_bio(). Returns 0 on success, or the error from | |
777 | * bio_endio() on failure. | |
778 | */ | |
779 | int submit_bio_wait(int rw, struct bio *bio) | |
780 | { | |
781 | struct submit_bio_ret ret; | |
782 | ||
783 | rw |= REQ_SYNC; | |
784 | init_completion(&ret.event); | |
785 | bio->bi_private = &ret; | |
786 | bio->bi_end_io = submit_bio_wait_endio; | |
787 | submit_bio(rw, bio); | |
788 | wait_for_completion(&ret.event); | |
789 | ||
790 | return ret.error; | |
791 | } | |
792 | EXPORT_SYMBOL(submit_bio_wait); | |
793 | ||
794 | /** | |
795 | * bio_advance - increment/complete a bio by some number of bytes | |
796 | * @bio: bio to advance | |
797 | * @bytes: number of bytes to complete | |
798 | * | |
799 | * This updates bi_sector, bi_size and bi_idx; if the number of bytes to | |
800 | * complete doesn't align with a bvec boundary, then bv_len and bv_offset will | |
801 | * be updated on the last bvec as well. | |
802 | * | |
803 | * @bio will then represent the remaining, uncompleted portion of the io. | |
804 | */ | |
805 | void bio_advance(struct bio *bio, unsigned bytes) | |
806 | { | |
807 | if (bio_integrity(bio)) | |
808 | bio_integrity_advance(bio, bytes); | |
809 | ||
810 | bio->bi_sector += bytes >> 9; | |
811 | bio->bi_size -= bytes; | |
812 | ||
813 | if (bio->bi_rw & BIO_NO_ADVANCE_ITER_MASK) | |
814 | return; | |
815 | ||
816 | while (bytes) { | |
817 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | |
818 | WARN_ONCE(1, "bio idx %d >= vcnt %d\n", | |
819 | bio->bi_idx, bio->bi_vcnt); | |
820 | break; | |
821 | } | |
822 | ||
823 | if (bytes >= bio_iovec(bio)->bv_len) { | |
824 | bytes -= bio_iovec(bio)->bv_len; | |
825 | bio->bi_idx++; | |
826 | } else { | |
827 | bio_iovec(bio)->bv_len -= bytes; | |
828 | bio_iovec(bio)->bv_offset += bytes; | |
829 | bytes = 0; | |
830 | } | |
831 | } | |
832 | } | |
833 | EXPORT_SYMBOL(bio_advance); | |
834 | ||
835 | /** | |
836 | * bio_alloc_pages - allocates a single page for each bvec in a bio | |
837 | * @bio: bio to allocate pages for | |
838 | * @gfp_mask: flags for allocation | |
839 | * | |
840 | * Allocates pages up to @bio->bi_vcnt. | |
841 | * | |
842 | * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are | |
843 | * freed. | |
844 | */ | |
845 | int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask) | |
846 | { | |
847 | int i; | |
848 | struct bio_vec *bv; | |
849 | ||
850 | bio_for_each_segment_all(bv, bio, i) { | |
851 | bv->bv_page = alloc_page(gfp_mask); | |
852 | if (!bv->bv_page) { | |
853 | while (--bv >= bio->bi_io_vec) | |
854 | __free_page(bv->bv_page); | |
855 | return -ENOMEM; | |
856 | } | |
857 | } | |
858 | ||
859 | return 0; | |
860 | } | |
861 | EXPORT_SYMBOL(bio_alloc_pages); | |
862 | ||
863 | /** | |
864 | * bio_copy_data - copy contents of data buffers from one chain of bios to | |
865 | * another | |
866 | * @src: source bio list | |
867 | * @dst: destination bio list | |
868 | * | |
869 | * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats | |
870 | * @src and @dst as linked lists of bios. | |
871 | * | |
872 | * Stops when it reaches the end of either @src or @dst - that is, copies | |
873 | * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). | |
874 | */ | |
875 | void bio_copy_data(struct bio *dst, struct bio *src) | |
876 | { | |
877 | struct bio_vec *src_bv, *dst_bv; | |
878 | unsigned src_offset, dst_offset, bytes; | |
879 | void *src_p, *dst_p; | |
880 | ||
881 | src_bv = bio_iovec(src); | |
882 | dst_bv = bio_iovec(dst); | |
883 | ||
884 | src_offset = src_bv->bv_offset; | |
885 | dst_offset = dst_bv->bv_offset; | |
886 | ||
887 | while (1) { | |
888 | if (src_offset == src_bv->bv_offset + src_bv->bv_len) { | |
889 | src_bv++; | |
890 | if (src_bv == bio_iovec_idx(src, src->bi_vcnt)) { | |
891 | src = src->bi_next; | |
892 | if (!src) | |
893 | break; | |
894 | ||
895 | src_bv = bio_iovec(src); | |
896 | } | |
897 | ||
898 | src_offset = src_bv->bv_offset; | |
899 | } | |
900 | ||
901 | if (dst_offset == dst_bv->bv_offset + dst_bv->bv_len) { | |
902 | dst_bv++; | |
903 | if (dst_bv == bio_iovec_idx(dst, dst->bi_vcnt)) { | |
904 | dst = dst->bi_next; | |
905 | if (!dst) | |
906 | break; | |
907 | ||
908 | dst_bv = bio_iovec(dst); | |
909 | } | |
910 | ||
911 | dst_offset = dst_bv->bv_offset; | |
912 | } | |
913 | ||
914 | bytes = min(dst_bv->bv_offset + dst_bv->bv_len - dst_offset, | |
915 | src_bv->bv_offset + src_bv->bv_len - src_offset); | |
916 | ||
917 | src_p = kmap_atomic(src_bv->bv_page); | |
918 | dst_p = kmap_atomic(dst_bv->bv_page); | |
919 | ||
920 | memcpy(dst_p + dst_offset, | |
921 | src_p + src_offset, | |
922 | bytes); | |
923 | ||
924 | kunmap_atomic(dst_p); | |
925 | kunmap_atomic(src_p); | |
926 | ||
927 | src_offset += bytes; | |
928 | dst_offset += bytes; | |
929 | } | |
930 | } | |
931 | EXPORT_SYMBOL(bio_copy_data); | |
932 | ||
933 | struct bio_map_data { | |
934 | struct bio_vec *iovecs; | |
935 | struct sg_iovec *sgvecs; | |
936 | int nr_sgvecs; | |
937 | int is_our_pages; | |
938 | }; | |
939 | ||
940 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio, | |
941 | struct sg_iovec *iov, int iov_count, | |
942 | int is_our_pages) | |
943 | { | |
944 | memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); | |
945 | memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count); | |
946 | bmd->nr_sgvecs = iov_count; | |
947 | bmd->is_our_pages = is_our_pages; | |
948 | bio->bi_private = bmd; | |
949 | } | |
950 | ||
951 | static void bio_free_map_data(struct bio_map_data *bmd) | |
952 | { | |
953 | kfree(bmd->iovecs); | |
954 | kfree(bmd->sgvecs); | |
955 | kfree(bmd); | |
956 | } | |
957 | ||
958 | static struct bio_map_data *bio_alloc_map_data(int nr_segs, | |
959 | unsigned int iov_count, | |
960 | gfp_t gfp_mask) | |
961 | { | |
962 | struct bio_map_data *bmd; | |
963 | ||
964 | if (iov_count > UIO_MAXIOV) | |
965 | return NULL; | |
966 | ||
967 | bmd = kmalloc(sizeof(*bmd), gfp_mask); | |
968 | if (!bmd) | |
969 | return NULL; | |
970 | ||
971 | bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask); | |
972 | if (!bmd->iovecs) { | |
973 | kfree(bmd); | |
974 | return NULL; | |
975 | } | |
976 | ||
977 | bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask); | |
978 | if (bmd->sgvecs) | |
979 | return bmd; | |
980 | ||
981 | kfree(bmd->iovecs); | |
982 | kfree(bmd); | |
983 | return NULL; | |
984 | } | |
985 | ||
986 | static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs, | |
987 | struct sg_iovec *iov, int iov_count, | |
988 | int to_user, int from_user, int do_free_page) | |
989 | { | |
990 | int ret = 0, i; | |
991 | struct bio_vec *bvec; | |
992 | int iov_idx = 0; | |
993 | unsigned int iov_off = 0; | |
994 | ||
995 | bio_for_each_segment_all(bvec, bio, i) { | |
996 | char *bv_addr = page_address(bvec->bv_page); | |
997 | unsigned int bv_len = iovecs[i].bv_len; | |
998 | ||
999 | while (bv_len && iov_idx < iov_count) { | |
1000 | unsigned int bytes; | |
1001 | char __user *iov_addr; | |
1002 | ||
1003 | bytes = min_t(unsigned int, | |
1004 | iov[iov_idx].iov_len - iov_off, bv_len); | |
1005 | iov_addr = iov[iov_idx].iov_base + iov_off; | |
1006 | ||
1007 | if (!ret) { | |
1008 | if (to_user) | |
1009 | ret = copy_to_user(iov_addr, bv_addr, | |
1010 | bytes); | |
1011 | ||
1012 | if (from_user) | |
1013 | ret = copy_from_user(bv_addr, iov_addr, | |
1014 | bytes); | |
1015 | ||
1016 | if (ret) | |
1017 | ret = -EFAULT; | |
1018 | } | |
1019 | ||
1020 | bv_len -= bytes; | |
1021 | bv_addr += bytes; | |
1022 | iov_addr += bytes; | |
1023 | iov_off += bytes; | |
1024 | ||
1025 | if (iov[iov_idx].iov_len == iov_off) { | |
1026 | iov_idx++; | |
1027 | iov_off = 0; | |
1028 | } | |
1029 | } | |
1030 | ||
1031 | if (do_free_page) | |
1032 | __free_page(bvec->bv_page); | |
1033 | } | |
1034 | ||
1035 | return ret; | |
1036 | } | |
1037 | ||
1038 | /** | |
1039 | * bio_uncopy_user - finish previously mapped bio | |
1040 | * @bio: bio being terminated | |
1041 | * | |
1042 | * Free pages allocated from bio_copy_user() and write back data | |
1043 | * to user space in case of a read. | |
1044 | */ | |
1045 | int bio_uncopy_user(struct bio *bio) | |
1046 | { | |
1047 | struct bio_map_data *bmd = bio->bi_private; | |
1048 | struct bio_vec *bvec; | |
1049 | int ret = 0, i; | |
1050 | ||
1051 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) { | |
1052 | /* | |
1053 | * if we're in a workqueue, the request is orphaned, so | |
1054 | * don't copy into a random user address space, just free. | |
1055 | */ | |
1056 | if (current->mm) | |
1057 | ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs, | |
1058 | bmd->nr_sgvecs, bio_data_dir(bio) == READ, | |
1059 | 0, bmd->is_our_pages); | |
1060 | else if (bmd->is_our_pages) | |
1061 | bio_for_each_segment_all(bvec, bio, i) | |
1062 | __free_page(bvec->bv_page); | |
1063 | } | |
1064 | bio_free_map_data(bmd); | |
1065 | bio_put(bio); | |
1066 | return ret; | |
1067 | } | |
1068 | EXPORT_SYMBOL(bio_uncopy_user); | |
1069 | ||
1070 | /** | |
1071 | * bio_copy_user_iov - copy user data to bio | |
1072 | * @q: destination block queue | |
1073 | * @map_data: pointer to the rq_map_data holding pages (if necessary) | |
1074 | * @iov: the iovec. | |
1075 | * @iov_count: number of elements in the iovec | |
1076 | * @write_to_vm: bool indicating writing to pages or not | |
1077 | * @gfp_mask: memory allocation flags | |
1078 | * | |
1079 | * Prepares and returns a bio for indirect user io, bouncing data | |
1080 | * to/from kernel pages as necessary. Must be paired with | |
1081 | * call bio_uncopy_user() on io completion. | |
1082 | */ | |
1083 | struct bio *bio_copy_user_iov(struct request_queue *q, | |
1084 | struct rq_map_data *map_data, | |
1085 | struct sg_iovec *iov, int iov_count, | |
1086 | int write_to_vm, gfp_t gfp_mask) | |
1087 | { | |
1088 | struct bio_map_data *bmd; | |
1089 | struct bio_vec *bvec; | |
1090 | struct page *page; | |
1091 | struct bio *bio; | |
1092 | int i, ret; | |
1093 | int nr_pages = 0; | |
1094 | unsigned int len = 0; | |
1095 | unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0; | |
1096 | ||
1097 | for (i = 0; i < iov_count; i++) { | |
1098 | unsigned long uaddr; | |
1099 | unsigned long end; | |
1100 | unsigned long start; | |
1101 | ||
1102 | uaddr = (unsigned long)iov[i].iov_base; | |
1103 | end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1104 | start = uaddr >> PAGE_SHIFT; | |
1105 | ||
1106 | /* | |
1107 | * Overflow, abort | |
1108 | */ | |
1109 | if (end < start) | |
1110 | return ERR_PTR(-EINVAL); | |
1111 | ||
1112 | nr_pages += end - start; | |
1113 | len += iov[i].iov_len; | |
1114 | } | |
1115 | ||
1116 | if (offset) | |
1117 | nr_pages++; | |
1118 | ||
1119 | bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask); | |
1120 | if (!bmd) | |
1121 | return ERR_PTR(-ENOMEM); | |
1122 | ||
1123 | ret = -ENOMEM; | |
1124 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1125 | if (!bio) | |
1126 | goto out_bmd; | |
1127 | ||
1128 | if (!write_to_vm) | |
1129 | bio->bi_rw |= REQ_WRITE; | |
1130 | ||
1131 | ret = 0; | |
1132 | ||
1133 | if (map_data) { | |
1134 | nr_pages = 1 << map_data->page_order; | |
1135 | i = map_data->offset / PAGE_SIZE; | |
1136 | } | |
1137 | while (len) { | |
1138 | unsigned int bytes = PAGE_SIZE; | |
1139 | ||
1140 | bytes -= offset; | |
1141 | ||
1142 | if (bytes > len) | |
1143 | bytes = len; | |
1144 | ||
1145 | if (map_data) { | |
1146 | if (i == map_data->nr_entries * nr_pages) { | |
1147 | ret = -ENOMEM; | |
1148 | break; | |
1149 | } | |
1150 | ||
1151 | page = map_data->pages[i / nr_pages]; | |
1152 | page += (i % nr_pages); | |
1153 | ||
1154 | i++; | |
1155 | } else { | |
1156 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
1157 | if (!page) { | |
1158 | ret = -ENOMEM; | |
1159 | break; | |
1160 | } | |
1161 | } | |
1162 | ||
1163 | if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) | |
1164 | break; | |
1165 | ||
1166 | len -= bytes; | |
1167 | offset = 0; | |
1168 | } | |
1169 | ||
1170 | if (ret) | |
1171 | goto cleanup; | |
1172 | ||
1173 | /* | |
1174 | * success | |
1175 | */ | |
1176 | if ((!write_to_vm && (!map_data || !map_data->null_mapped)) || | |
1177 | (map_data && map_data->from_user)) { | |
1178 | ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0); | |
1179 | if (ret) | |
1180 | goto cleanup; | |
1181 | } | |
1182 | ||
1183 | bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1); | |
1184 | return bio; | |
1185 | cleanup: | |
1186 | if (!map_data) | |
1187 | bio_for_each_segment_all(bvec, bio, i) | |
1188 | __free_page(bvec->bv_page); | |
1189 | ||
1190 | bio_put(bio); | |
1191 | out_bmd: | |
1192 | bio_free_map_data(bmd); | |
1193 | return ERR_PTR(ret); | |
1194 | } | |
1195 | ||
1196 | /** | |
1197 | * bio_copy_user - copy user data to bio | |
1198 | * @q: destination block queue | |
1199 | * @map_data: pointer to the rq_map_data holding pages (if necessary) | |
1200 | * @uaddr: start of user address | |
1201 | * @len: length in bytes | |
1202 | * @write_to_vm: bool indicating writing to pages or not | |
1203 | * @gfp_mask: memory allocation flags | |
1204 | * | |
1205 | * Prepares and returns a bio for indirect user io, bouncing data | |
1206 | * to/from kernel pages as necessary. Must be paired with | |
1207 | * call bio_uncopy_user() on io completion. | |
1208 | */ | |
1209 | struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data, | |
1210 | unsigned long uaddr, unsigned int len, | |
1211 | int write_to_vm, gfp_t gfp_mask) | |
1212 | { | |
1213 | struct sg_iovec iov; | |
1214 | ||
1215 | iov.iov_base = (void __user *)uaddr; | |
1216 | iov.iov_len = len; | |
1217 | ||
1218 | return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask); | |
1219 | } | |
1220 | EXPORT_SYMBOL(bio_copy_user); | |
1221 | ||
1222 | static struct bio *__bio_map_user_iov(struct request_queue *q, | |
1223 | struct block_device *bdev, | |
1224 | struct sg_iovec *iov, int iov_count, | |
1225 | int write_to_vm, gfp_t gfp_mask) | |
1226 | { | |
1227 | int i, j; | |
1228 | int nr_pages = 0; | |
1229 | struct page **pages; | |
1230 | struct bio *bio; | |
1231 | int cur_page = 0; | |
1232 | int ret, offset; | |
1233 | ||
1234 | for (i = 0; i < iov_count; i++) { | |
1235 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
1236 | unsigned long len = iov[i].iov_len; | |
1237 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1238 | unsigned long start = uaddr >> PAGE_SHIFT; | |
1239 | ||
1240 | /* | |
1241 | * Overflow, abort | |
1242 | */ | |
1243 | if (end < start) | |
1244 | return ERR_PTR(-EINVAL); | |
1245 | ||
1246 | nr_pages += end - start; | |
1247 | /* | |
1248 | * buffer must be aligned to at least hardsector size for now | |
1249 | */ | |
1250 | if (uaddr & queue_dma_alignment(q)) | |
1251 | return ERR_PTR(-EINVAL); | |
1252 | } | |
1253 | ||
1254 | if (!nr_pages) | |
1255 | return ERR_PTR(-EINVAL); | |
1256 | ||
1257 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1258 | if (!bio) | |
1259 | return ERR_PTR(-ENOMEM); | |
1260 | ||
1261 | ret = -ENOMEM; | |
1262 | pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask); | |
1263 | if (!pages) | |
1264 | goto out; | |
1265 | ||
1266 | for (i = 0; i < iov_count; i++) { | |
1267 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
1268 | unsigned long len = iov[i].iov_len; | |
1269 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1270 | unsigned long start = uaddr >> PAGE_SHIFT; | |
1271 | const int local_nr_pages = end - start; | |
1272 | const int page_limit = cur_page + local_nr_pages; | |
1273 | ||
1274 | ret = get_user_pages_fast(uaddr, local_nr_pages, | |
1275 | write_to_vm, &pages[cur_page]); | |
1276 | if (ret < local_nr_pages) { | |
1277 | ret = -EFAULT; | |
1278 | goto out_unmap; | |
1279 | } | |
1280 | ||
1281 | offset = uaddr & ~PAGE_MASK; | |
1282 | for (j = cur_page; j < page_limit; j++) { | |
1283 | unsigned int bytes = PAGE_SIZE - offset; | |
1284 | ||
1285 | if (len <= 0) | |
1286 | break; | |
1287 | ||
1288 | if (bytes > len) | |
1289 | bytes = len; | |
1290 | ||
1291 | /* | |
1292 | * sorry... | |
1293 | */ | |
1294 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < | |
1295 | bytes) | |
1296 | break; | |
1297 | ||
1298 | len -= bytes; | |
1299 | offset = 0; | |
1300 | } | |
1301 | ||
1302 | cur_page = j; | |
1303 | /* | |
1304 | * release the pages we didn't map into the bio, if any | |
1305 | */ | |
1306 | while (j < page_limit) | |
1307 | page_cache_release(pages[j++]); | |
1308 | } | |
1309 | ||
1310 | kfree(pages); | |
1311 | ||
1312 | /* | |
1313 | * set data direction, and check if mapped pages need bouncing | |
1314 | */ | |
1315 | if (!write_to_vm) | |
1316 | bio->bi_rw |= REQ_WRITE; | |
1317 | ||
1318 | bio->bi_bdev = bdev; | |
1319 | bio->bi_flags |= (1 << BIO_USER_MAPPED); | |
1320 | return bio; | |
1321 | ||
1322 | out_unmap: | |
1323 | for (i = 0; i < nr_pages; i++) { | |
1324 | if(!pages[i]) | |
1325 | break; | |
1326 | page_cache_release(pages[i]); | |
1327 | } | |
1328 | out: | |
1329 | kfree(pages); | |
1330 | bio_put(bio); | |
1331 | return ERR_PTR(ret); | |
1332 | } | |
1333 | ||
1334 | /** | |
1335 | * bio_map_user - map user address into bio | |
1336 | * @q: the struct request_queue for the bio | |
1337 | * @bdev: destination block device | |
1338 | * @uaddr: start of user address | |
1339 | * @len: length in bytes | |
1340 | * @write_to_vm: bool indicating writing to pages or not | |
1341 | * @gfp_mask: memory allocation flags | |
1342 | * | |
1343 | * Map the user space address into a bio suitable for io to a block | |
1344 | * device. Returns an error pointer in case of error. | |
1345 | */ | |
1346 | struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev, | |
1347 | unsigned long uaddr, unsigned int len, int write_to_vm, | |
1348 | gfp_t gfp_mask) | |
1349 | { | |
1350 | struct sg_iovec iov; | |
1351 | ||
1352 | iov.iov_base = (void __user *)uaddr; | |
1353 | iov.iov_len = len; | |
1354 | ||
1355 | return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask); | |
1356 | } | |
1357 | EXPORT_SYMBOL(bio_map_user); | |
1358 | ||
1359 | /** | |
1360 | * bio_map_user_iov - map user sg_iovec table into bio | |
1361 | * @q: the struct request_queue for the bio | |
1362 | * @bdev: destination block device | |
1363 | * @iov: the iovec. | |
1364 | * @iov_count: number of elements in the iovec | |
1365 | * @write_to_vm: bool indicating writing to pages or not | |
1366 | * @gfp_mask: memory allocation flags | |
1367 | * | |
1368 | * Map the user space address into a bio suitable for io to a block | |
1369 | * device. Returns an error pointer in case of error. | |
1370 | */ | |
1371 | struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev, | |
1372 | struct sg_iovec *iov, int iov_count, | |
1373 | int write_to_vm, gfp_t gfp_mask) | |
1374 | { | |
1375 | struct bio *bio; | |
1376 | ||
1377 | bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm, | |
1378 | gfp_mask); | |
1379 | if (IS_ERR(bio)) | |
1380 | return bio; | |
1381 | ||
1382 | /* | |
1383 | * subtle -- if __bio_map_user() ended up bouncing a bio, | |
1384 | * it would normally disappear when its bi_end_io is run. | |
1385 | * however, we need it for the unmap, so grab an extra | |
1386 | * reference to it | |
1387 | */ | |
1388 | bio_get(bio); | |
1389 | ||
1390 | return bio; | |
1391 | } | |
1392 | ||
1393 | static void __bio_unmap_user(struct bio *bio) | |
1394 | { | |
1395 | struct bio_vec *bvec; | |
1396 | int i; | |
1397 | ||
1398 | /* | |
1399 | * make sure we dirty pages we wrote to | |
1400 | */ | |
1401 | bio_for_each_segment_all(bvec, bio, i) { | |
1402 | if (bio_data_dir(bio) == READ) | |
1403 | set_page_dirty_lock(bvec->bv_page); | |
1404 | ||
1405 | page_cache_release(bvec->bv_page); | |
1406 | } | |
1407 | ||
1408 | bio_put(bio); | |
1409 | } | |
1410 | ||
1411 | /** | |
1412 | * bio_unmap_user - unmap a bio | |
1413 | * @bio: the bio being unmapped | |
1414 | * | |
1415 | * Unmap a bio previously mapped by bio_map_user(). Must be called with | |
1416 | * a process context. | |
1417 | * | |
1418 | * bio_unmap_user() may sleep. | |
1419 | */ | |
1420 | void bio_unmap_user(struct bio *bio) | |
1421 | { | |
1422 | __bio_unmap_user(bio); | |
1423 | bio_put(bio); | |
1424 | } | |
1425 | EXPORT_SYMBOL(bio_unmap_user); | |
1426 | ||
1427 | static void bio_map_kern_endio(struct bio *bio, int err) | |
1428 | { | |
1429 | bio_put(bio); | |
1430 | } | |
1431 | ||
1432 | static struct bio *__bio_map_kern(struct request_queue *q, void *data, | |
1433 | unsigned int len, gfp_t gfp_mask) | |
1434 | { | |
1435 | unsigned long kaddr = (unsigned long)data; | |
1436 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1437 | unsigned long start = kaddr >> PAGE_SHIFT; | |
1438 | const int nr_pages = end - start; | |
1439 | int offset, i; | |
1440 | struct bio *bio; | |
1441 | ||
1442 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1443 | if (!bio) | |
1444 | return ERR_PTR(-ENOMEM); | |
1445 | ||
1446 | offset = offset_in_page(kaddr); | |
1447 | for (i = 0; i < nr_pages; i++) { | |
1448 | unsigned int bytes = PAGE_SIZE - offset; | |
1449 | ||
1450 | if (len <= 0) | |
1451 | break; | |
1452 | ||
1453 | if (bytes > len) | |
1454 | bytes = len; | |
1455 | ||
1456 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, | |
1457 | offset) < bytes) | |
1458 | break; | |
1459 | ||
1460 | data += bytes; | |
1461 | len -= bytes; | |
1462 | offset = 0; | |
1463 | } | |
1464 | ||
1465 | bio->bi_end_io = bio_map_kern_endio; | |
1466 | return bio; | |
1467 | } | |
1468 | ||
1469 | /** | |
1470 | * bio_map_kern - map kernel address into bio | |
1471 | * @q: the struct request_queue for the bio | |
1472 | * @data: pointer to buffer to map | |
1473 | * @len: length in bytes | |
1474 | * @gfp_mask: allocation flags for bio allocation | |
1475 | * | |
1476 | * Map the kernel address into a bio suitable for io to a block | |
1477 | * device. Returns an error pointer in case of error. | |
1478 | */ | |
1479 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, | |
1480 | gfp_t gfp_mask) | |
1481 | { | |
1482 | struct bio *bio; | |
1483 | ||
1484 | bio = __bio_map_kern(q, data, len, gfp_mask); | |
1485 | if (IS_ERR(bio)) | |
1486 | return bio; | |
1487 | ||
1488 | if (bio->bi_size == len) | |
1489 | return bio; | |
1490 | ||
1491 | /* | |
1492 | * Don't support partial mappings. | |
1493 | */ | |
1494 | bio_put(bio); | |
1495 | return ERR_PTR(-EINVAL); | |
1496 | } | |
1497 | EXPORT_SYMBOL(bio_map_kern); | |
1498 | ||
1499 | static void bio_copy_kern_endio(struct bio *bio, int err) | |
1500 | { | |
1501 | struct bio_vec *bvec; | |
1502 | const int read = bio_data_dir(bio) == READ; | |
1503 | struct bio_map_data *bmd = bio->bi_private; | |
1504 | int i; | |
1505 | char *p = bmd->sgvecs[0].iov_base; | |
1506 | ||
1507 | bio_for_each_segment_all(bvec, bio, i) { | |
1508 | char *addr = page_address(bvec->bv_page); | |
1509 | int len = bmd->iovecs[i].bv_len; | |
1510 | ||
1511 | if (read) | |
1512 | memcpy(p, addr, len); | |
1513 | ||
1514 | __free_page(bvec->bv_page); | |
1515 | p += len; | |
1516 | } | |
1517 | ||
1518 | bio_free_map_data(bmd); | |
1519 | bio_put(bio); | |
1520 | } | |
1521 | ||
1522 | /** | |
1523 | * bio_copy_kern - copy kernel address into bio | |
1524 | * @q: the struct request_queue for the bio | |
1525 | * @data: pointer to buffer to copy | |
1526 | * @len: length in bytes | |
1527 | * @gfp_mask: allocation flags for bio and page allocation | |
1528 | * @reading: data direction is READ | |
1529 | * | |
1530 | * copy the kernel address into a bio suitable for io to a block | |
1531 | * device. Returns an error pointer in case of error. | |
1532 | */ | |
1533 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
1534 | gfp_t gfp_mask, int reading) | |
1535 | { | |
1536 | struct bio *bio; | |
1537 | struct bio_vec *bvec; | |
1538 | int i; | |
1539 | ||
1540 | bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask); | |
1541 | if (IS_ERR(bio)) | |
1542 | return bio; | |
1543 | ||
1544 | if (!reading) { | |
1545 | void *p = data; | |
1546 | ||
1547 | bio_for_each_segment_all(bvec, bio, i) { | |
1548 | char *addr = page_address(bvec->bv_page); | |
1549 | ||
1550 | memcpy(addr, p, bvec->bv_len); | |
1551 | p += bvec->bv_len; | |
1552 | } | |
1553 | } | |
1554 | ||
1555 | bio->bi_end_io = bio_copy_kern_endio; | |
1556 | ||
1557 | return bio; | |
1558 | } | |
1559 | EXPORT_SYMBOL(bio_copy_kern); | |
1560 | ||
1561 | /* | |
1562 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1563 | * for performing direct-IO in BIOs. | |
1564 | * | |
1565 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1566 | * because the required locks are not interrupt-safe. So what we can do is to | |
1567 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1568 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1569 | * in process context. | |
1570 | * | |
1571 | * We special-case compound pages here: normally this means reads into hugetlb | |
1572 | * pages. The logic in here doesn't really work right for compound pages | |
1573 | * because the VM does not uniformly chase down the head page in all cases. | |
1574 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1575 | * handle them at all. So we skip compound pages here at an early stage. | |
1576 | * | |
1577 | * Note that this code is very hard to test under normal circumstances because | |
1578 | * direct-io pins the pages with get_user_pages(). This makes | |
1579 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
1580 | * But other code (eg, flusher threads) could clean the pages if they are mapped | |
1581 | * pagecache. | |
1582 | * | |
1583 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1584 | * deferred bio dirtying paths. | |
1585 | */ | |
1586 | ||
1587 | /* | |
1588 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1589 | */ | |
1590 | void bio_set_pages_dirty(struct bio *bio) | |
1591 | { | |
1592 | struct bio_vec *bvec; | |
1593 | int i; | |
1594 | ||
1595 | bio_for_each_segment_all(bvec, bio, i) { | |
1596 | struct page *page = bvec->bv_page; | |
1597 | ||
1598 | if (page && !PageCompound(page)) | |
1599 | set_page_dirty_lock(page); | |
1600 | } | |
1601 | } | |
1602 | ||
1603 | static void bio_release_pages(struct bio *bio) | |
1604 | { | |
1605 | struct bio_vec *bvec; | |
1606 | int i; | |
1607 | ||
1608 | bio_for_each_segment_all(bvec, bio, i) { | |
1609 | struct page *page = bvec->bv_page; | |
1610 | ||
1611 | if (page) | |
1612 | put_page(page); | |
1613 | } | |
1614 | } | |
1615 | ||
1616 | /* | |
1617 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1618 | * If they are, then fine. If, however, some pages are clean then they must | |
1619 | * have been written out during the direct-IO read. So we take another ref on | |
1620 | * the BIO and the offending pages and re-dirty the pages in process context. | |
1621 | * | |
1622 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
1623 | * here on. It will run one page_cache_release() against each page and will | |
1624 | * run one bio_put() against the BIO. | |
1625 | */ | |
1626 | ||
1627 | static void bio_dirty_fn(struct work_struct *work); | |
1628 | ||
1629 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); | |
1630 | static DEFINE_SPINLOCK(bio_dirty_lock); | |
1631 | static struct bio *bio_dirty_list; | |
1632 | ||
1633 | /* | |
1634 | * This runs in process context | |
1635 | */ | |
1636 | static void bio_dirty_fn(struct work_struct *work) | |
1637 | { | |
1638 | unsigned long flags; | |
1639 | struct bio *bio; | |
1640 | ||
1641 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1642 | bio = bio_dirty_list; | |
1643 | bio_dirty_list = NULL; | |
1644 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1645 | ||
1646 | while (bio) { | |
1647 | struct bio *next = bio->bi_private; | |
1648 | ||
1649 | bio_set_pages_dirty(bio); | |
1650 | bio_release_pages(bio); | |
1651 | bio_put(bio); | |
1652 | bio = next; | |
1653 | } | |
1654 | } | |
1655 | ||
1656 | void bio_check_pages_dirty(struct bio *bio) | |
1657 | { | |
1658 | struct bio_vec *bvec; | |
1659 | int nr_clean_pages = 0; | |
1660 | int i; | |
1661 | ||
1662 | bio_for_each_segment_all(bvec, bio, i) { | |
1663 | struct page *page = bvec->bv_page; | |
1664 | ||
1665 | if (PageDirty(page) || PageCompound(page)) { | |
1666 | page_cache_release(page); | |
1667 | bvec->bv_page = NULL; | |
1668 | } else { | |
1669 | nr_clean_pages++; | |
1670 | } | |
1671 | } | |
1672 | ||
1673 | if (nr_clean_pages) { | |
1674 | unsigned long flags; | |
1675 | ||
1676 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1677 | bio->bi_private = bio_dirty_list; | |
1678 | bio_dirty_list = bio; | |
1679 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1680 | schedule_work(&bio_dirty_work); | |
1681 | } else { | |
1682 | bio_put(bio); | |
1683 | } | |
1684 | } | |
1685 | ||
1686 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE | |
1687 | void bio_flush_dcache_pages(struct bio *bi) | |
1688 | { | |
1689 | int i; | |
1690 | struct bio_vec *bvec; | |
1691 | ||
1692 | bio_for_each_segment(bvec, bi, i) | |
1693 | flush_dcache_page(bvec->bv_page); | |
1694 | } | |
1695 | EXPORT_SYMBOL(bio_flush_dcache_pages); | |
1696 | #endif | |
1697 | ||
1698 | /** | |
1699 | * bio_endio - end I/O on a bio | |
1700 | * @bio: bio | |
1701 | * @error: error, if any | |
1702 | * | |
1703 | * Description: | |
1704 | * bio_endio() will end I/O on the whole bio. bio_endio() is the | |
1705 | * preferred way to end I/O on a bio, it takes care of clearing | |
1706 | * BIO_UPTODATE on error. @error is 0 on success, and and one of the | |
1707 | * established -Exxxx (-EIO, for instance) error values in case | |
1708 | * something went wrong. No one should call bi_end_io() directly on a | |
1709 | * bio unless they own it and thus know that it has an end_io | |
1710 | * function. | |
1711 | **/ | |
1712 | void bio_endio(struct bio *bio, int error) | |
1713 | { | |
1714 | if (error) | |
1715 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
1716 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
1717 | error = -EIO; | |
1718 | ||
1719 | if (bio->bi_end_io) | |
1720 | bio->bi_end_io(bio, error); | |
1721 | } | |
1722 | EXPORT_SYMBOL(bio_endio); | |
1723 | ||
1724 | void bio_pair_release(struct bio_pair *bp) | |
1725 | { | |
1726 | if (atomic_dec_and_test(&bp->cnt)) { | |
1727 | struct bio *master = bp->bio1.bi_private; | |
1728 | ||
1729 | bio_endio(master, bp->error); | |
1730 | mempool_free(bp, bp->bio2.bi_private); | |
1731 | } | |
1732 | } | |
1733 | EXPORT_SYMBOL(bio_pair_release); | |
1734 | ||
1735 | static void bio_pair_end_1(struct bio *bi, int err) | |
1736 | { | |
1737 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); | |
1738 | ||
1739 | if (err) | |
1740 | bp->error = err; | |
1741 | ||
1742 | bio_pair_release(bp); | |
1743 | } | |
1744 | ||
1745 | static void bio_pair_end_2(struct bio *bi, int err) | |
1746 | { | |
1747 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); | |
1748 | ||
1749 | if (err) | |
1750 | bp->error = err; | |
1751 | ||
1752 | bio_pair_release(bp); | |
1753 | } | |
1754 | ||
1755 | /* | |
1756 | * split a bio - only worry about a bio with a single page in its iovec | |
1757 | */ | |
1758 | struct bio_pair *bio_split(struct bio *bi, int first_sectors) | |
1759 | { | |
1760 | struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO); | |
1761 | ||
1762 | if (!bp) | |
1763 | return bp; | |
1764 | ||
1765 | trace_block_split(bdev_get_queue(bi->bi_bdev), bi, | |
1766 | bi->bi_sector + first_sectors); | |
1767 | ||
1768 | BUG_ON(bio_segments(bi) > 1); | |
1769 | atomic_set(&bp->cnt, 3); | |
1770 | bp->error = 0; | |
1771 | bp->bio1 = *bi; | |
1772 | bp->bio2 = *bi; | |
1773 | bp->bio2.bi_sector += first_sectors; | |
1774 | bp->bio2.bi_size -= first_sectors << 9; | |
1775 | bp->bio1.bi_size = first_sectors << 9; | |
1776 | ||
1777 | if (bi->bi_vcnt != 0) { | |
1778 | bp->bv1 = *bio_iovec(bi); | |
1779 | bp->bv2 = *bio_iovec(bi); | |
1780 | ||
1781 | if (bio_is_rw(bi)) { | |
1782 | bp->bv2.bv_offset += first_sectors << 9; | |
1783 | bp->bv2.bv_len -= first_sectors << 9; | |
1784 | bp->bv1.bv_len = first_sectors << 9; | |
1785 | } | |
1786 | ||
1787 | bp->bio1.bi_io_vec = &bp->bv1; | |
1788 | bp->bio2.bi_io_vec = &bp->bv2; | |
1789 | ||
1790 | bp->bio1.bi_max_vecs = 1; | |
1791 | bp->bio2.bi_max_vecs = 1; | |
1792 | } | |
1793 | ||
1794 | bp->bio1.bi_end_io = bio_pair_end_1; | |
1795 | bp->bio2.bi_end_io = bio_pair_end_2; | |
1796 | ||
1797 | bp->bio1.bi_private = bi; | |
1798 | bp->bio2.bi_private = bio_split_pool; | |
1799 | ||
1800 | if (bio_integrity(bi)) | |
1801 | bio_integrity_split(bi, bp, first_sectors); | |
1802 | ||
1803 | return bp; | |
1804 | } | |
1805 | EXPORT_SYMBOL(bio_split); | |
1806 | ||
1807 | /** | |
1808 | * bio_sector_offset - Find hardware sector offset in bio | |
1809 | * @bio: bio to inspect | |
1810 | * @index: bio_vec index | |
1811 | * @offset: offset in bv_page | |
1812 | * | |
1813 | * Return the number of hardware sectors between beginning of bio | |
1814 | * and an end point indicated by a bio_vec index and an offset | |
1815 | * within that vector's page. | |
1816 | */ | |
1817 | sector_t bio_sector_offset(struct bio *bio, unsigned short index, | |
1818 | unsigned int offset) | |
1819 | { | |
1820 | unsigned int sector_sz; | |
1821 | struct bio_vec *bv; | |
1822 | sector_t sectors; | |
1823 | int i; | |
1824 | ||
1825 | sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue); | |
1826 | sectors = 0; | |
1827 | ||
1828 | if (index >= bio->bi_idx) | |
1829 | index = bio->bi_vcnt - 1; | |
1830 | ||
1831 | bio_for_each_segment_all(bv, bio, i) { | |
1832 | if (i == index) { | |
1833 | if (offset > bv->bv_offset) | |
1834 | sectors += (offset - bv->bv_offset) / sector_sz; | |
1835 | break; | |
1836 | } | |
1837 | ||
1838 | sectors += bv->bv_len / sector_sz; | |
1839 | } | |
1840 | ||
1841 | return sectors; | |
1842 | } | |
1843 | EXPORT_SYMBOL(bio_sector_offset); | |
1844 | ||
1845 | /* | |
1846 | * create memory pools for biovec's in a bio_set. | |
1847 | * use the global biovec slabs created for general use. | |
1848 | */ | |
1849 | mempool_t *biovec_create_pool(struct bio_set *bs, int pool_entries) | |
1850 | { | |
1851 | struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX; | |
1852 | ||
1853 | return mempool_create_slab_pool(pool_entries, bp->slab); | |
1854 | } | |
1855 | ||
1856 | void bioset_free(struct bio_set *bs) | |
1857 | { | |
1858 | if (bs->rescue_workqueue) | |
1859 | destroy_workqueue(bs->rescue_workqueue); | |
1860 | ||
1861 | if (bs->bio_pool) | |
1862 | mempool_destroy(bs->bio_pool); | |
1863 | ||
1864 | if (bs->bvec_pool) | |
1865 | mempool_destroy(bs->bvec_pool); | |
1866 | ||
1867 | bioset_integrity_free(bs); | |
1868 | bio_put_slab(bs); | |
1869 | ||
1870 | kfree(bs); | |
1871 | } | |
1872 | EXPORT_SYMBOL(bioset_free); | |
1873 | ||
1874 | /** | |
1875 | * bioset_create - Create a bio_set | |
1876 | * @pool_size: Number of bio and bio_vecs to cache in the mempool | |
1877 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1878 | * | |
1879 | * Description: | |
1880 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller | |
1881 | * to ask for a number of bytes to be allocated in front of the bio. | |
1882 | * Front pad allocation is useful for embedding the bio inside | |
1883 | * another structure, to avoid allocating extra data to go with the bio. | |
1884 | * Note that the bio must be embedded at the END of that structure always, | |
1885 | * or things will break badly. | |
1886 | */ | |
1887 | struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad) | |
1888 | { | |
1889 | unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); | |
1890 | struct bio_set *bs; | |
1891 | ||
1892 | bs = kzalloc(sizeof(*bs), GFP_KERNEL); | |
1893 | if (!bs) | |
1894 | return NULL; | |
1895 | ||
1896 | bs->front_pad = front_pad; | |
1897 | ||
1898 | spin_lock_init(&bs->rescue_lock); | |
1899 | bio_list_init(&bs->rescue_list); | |
1900 | INIT_WORK(&bs->rescue_work, bio_alloc_rescue); | |
1901 | ||
1902 | bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); | |
1903 | if (!bs->bio_slab) { | |
1904 | kfree(bs); | |
1905 | return NULL; | |
1906 | } | |
1907 | ||
1908 | bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab); | |
1909 | if (!bs->bio_pool) | |
1910 | goto bad; | |
1911 | ||
1912 | bs->bvec_pool = biovec_create_pool(bs, pool_size); | |
1913 | if (!bs->bvec_pool) | |
1914 | goto bad; | |
1915 | ||
1916 | bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); | |
1917 | if (!bs->rescue_workqueue) | |
1918 | goto bad; | |
1919 | ||
1920 | return bs; | |
1921 | bad: | |
1922 | bioset_free(bs); | |
1923 | return NULL; | |
1924 | } | |
1925 | EXPORT_SYMBOL(bioset_create); | |
1926 | ||
1927 | #ifdef CONFIG_BLK_CGROUP | |
1928 | /** | |
1929 | * bio_associate_current - associate a bio with %current | |
1930 | * @bio: target bio | |
1931 | * | |
1932 | * Associate @bio with %current if it hasn't been associated yet. Block | |
1933 | * layer will treat @bio as if it were issued by %current no matter which | |
1934 | * task actually issues it. | |
1935 | * | |
1936 | * This function takes an extra reference of @task's io_context and blkcg | |
1937 | * which will be put when @bio is released. The caller must own @bio, | |
1938 | * ensure %current->io_context exists, and is responsible for synchronizing | |
1939 | * calls to this function. | |
1940 | */ | |
1941 | int bio_associate_current(struct bio *bio) | |
1942 | { | |
1943 | struct io_context *ioc; | |
1944 | struct cgroup_subsys_state *css; | |
1945 | ||
1946 | if (bio->bi_ioc) | |
1947 | return -EBUSY; | |
1948 | ||
1949 | ioc = current->io_context; | |
1950 | if (!ioc) | |
1951 | return -ENOENT; | |
1952 | ||
1953 | /* acquire active ref on @ioc and associate */ | |
1954 | get_io_context_active(ioc); | |
1955 | bio->bi_ioc = ioc; | |
1956 | ||
1957 | /* associate blkcg if exists */ | |
1958 | rcu_read_lock(); | |
1959 | css = task_css(current, blkio_subsys_id); | |
1960 | if (css && css_tryget(css)) | |
1961 | bio->bi_css = css; | |
1962 | rcu_read_unlock(); | |
1963 | ||
1964 | return 0; | |
1965 | } | |
1966 | ||
1967 | /** | |
1968 | * bio_disassociate_task - undo bio_associate_current() | |
1969 | * @bio: target bio | |
1970 | */ | |
1971 | void bio_disassociate_task(struct bio *bio) | |
1972 | { | |
1973 | if (bio->bi_ioc) { | |
1974 | put_io_context(bio->bi_ioc); | |
1975 | bio->bi_ioc = NULL; | |
1976 | } | |
1977 | if (bio->bi_css) { | |
1978 | css_put(bio->bi_css); | |
1979 | bio->bi_css = NULL; | |
1980 | } | |
1981 | } | |
1982 | ||
1983 | #endif /* CONFIG_BLK_CGROUP */ | |
1984 | ||
1985 | static void __init biovec_init_slabs(void) | |
1986 | { | |
1987 | int i; | |
1988 | ||
1989 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1990 | int size; | |
1991 | struct biovec_slab *bvs = bvec_slabs + i; | |
1992 | ||
1993 | if (bvs->nr_vecs <= BIO_INLINE_VECS) { | |
1994 | bvs->slab = NULL; | |
1995 | continue; | |
1996 | } | |
1997 | ||
1998 | size = bvs->nr_vecs * sizeof(struct bio_vec); | |
1999 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
2000 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); | |
2001 | } | |
2002 | } | |
2003 | ||
2004 | static int __init init_bio(void) | |
2005 | { | |
2006 | bio_slab_max = 2; | |
2007 | bio_slab_nr = 0; | |
2008 | bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL); | |
2009 | if (!bio_slabs) | |
2010 | panic("bio: can't allocate bios\n"); | |
2011 | ||
2012 | bio_integrity_init(); | |
2013 | biovec_init_slabs(); | |
2014 | ||
2015 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 0); | |
2016 | if (!fs_bio_set) | |
2017 | panic("bio: can't allocate bios\n"); | |
2018 | ||
2019 | if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE)) | |
2020 | panic("bio: can't create integrity pool\n"); | |
2021 | ||
2022 | bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES, | |
2023 | sizeof(struct bio_pair)); | |
2024 | if (!bio_split_pool) | |
2025 | panic("bio: can't create split pool\n"); | |
2026 | ||
2027 | return 0; | |
2028 | } | |
2029 | subsys_initcall(init_bio); |