]>
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 | /* | |
42 | * if you change this list, also change bvec_alloc or things will | |
43 | * break badly! cannot be bigger than what you can fit into an | |
44 | * unsigned short | |
45 | */ | |
46 | #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } | |
47 | static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { | |
48 | BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), | |
49 | }; | |
50 | #undef BV | |
51 | ||
52 | /* | |
53 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
54 | * IO code that does not need private memory pools. | |
55 | */ | |
56 | struct bio_set *fs_bio_set; | |
57 | EXPORT_SYMBOL(fs_bio_set); | |
58 | ||
59 | /* | |
60 | * Our slab pool management | |
61 | */ | |
62 | struct bio_slab { | |
63 | struct kmem_cache *slab; | |
64 | unsigned int slab_ref; | |
65 | unsigned int slab_size; | |
66 | char name[8]; | |
67 | }; | |
68 | static DEFINE_MUTEX(bio_slab_lock); | |
69 | static struct bio_slab *bio_slabs; | |
70 | static unsigned int bio_slab_nr, bio_slab_max; | |
71 | ||
72 | static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) | |
73 | { | |
74 | unsigned int sz = sizeof(struct bio) + extra_size; | |
75 | struct kmem_cache *slab = NULL; | |
76 | struct bio_slab *bslab, *new_bio_slabs; | |
77 | unsigned int new_bio_slab_max; | |
78 | unsigned int i, entry = -1; | |
79 | ||
80 | mutex_lock(&bio_slab_lock); | |
81 | ||
82 | i = 0; | |
83 | while (i < bio_slab_nr) { | |
84 | bslab = &bio_slabs[i]; | |
85 | ||
86 | if (!bslab->slab && entry == -1) | |
87 | entry = i; | |
88 | else if (bslab->slab_size == sz) { | |
89 | slab = bslab->slab; | |
90 | bslab->slab_ref++; | |
91 | break; | |
92 | } | |
93 | i++; | |
94 | } | |
95 | ||
96 | if (slab) | |
97 | goto out_unlock; | |
98 | ||
99 | if (bio_slab_nr == bio_slab_max && entry == -1) { | |
100 | new_bio_slab_max = bio_slab_max << 1; | |
101 | new_bio_slabs = krealloc(bio_slabs, | |
102 | new_bio_slab_max * sizeof(struct bio_slab), | |
103 | GFP_KERNEL); | |
104 | if (!new_bio_slabs) | |
105 | goto out_unlock; | |
106 | bio_slab_max = new_bio_slab_max; | |
107 | bio_slabs = new_bio_slabs; | |
108 | } | |
109 | if (entry == -1) | |
110 | entry = bio_slab_nr++; | |
111 | ||
112 | bslab = &bio_slabs[entry]; | |
113 | ||
114 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); | |
115 | slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL); | |
116 | if (!slab) | |
117 | goto out_unlock; | |
118 | ||
119 | printk(KERN_INFO "bio: create slab <%s> at %d\n", bslab->name, entry); | |
120 | bslab->slab = slab; | |
121 | bslab->slab_ref = 1; | |
122 | bslab->slab_size = sz; | |
123 | out_unlock: | |
124 | mutex_unlock(&bio_slab_lock); | |
125 | return slab; | |
126 | } | |
127 | ||
128 | static void bio_put_slab(struct bio_set *bs) | |
129 | { | |
130 | struct bio_slab *bslab = NULL; | |
131 | unsigned int i; | |
132 | ||
133 | mutex_lock(&bio_slab_lock); | |
134 | ||
135 | for (i = 0; i < bio_slab_nr; i++) { | |
136 | if (bs->bio_slab == bio_slabs[i].slab) { | |
137 | bslab = &bio_slabs[i]; | |
138 | break; | |
139 | } | |
140 | } | |
141 | ||
142 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) | |
143 | goto out; | |
144 | ||
145 | WARN_ON(!bslab->slab_ref); | |
146 | ||
147 | if (--bslab->slab_ref) | |
148 | goto out; | |
149 | ||
150 | kmem_cache_destroy(bslab->slab); | |
151 | bslab->slab = NULL; | |
152 | ||
153 | out: | |
154 | mutex_unlock(&bio_slab_lock); | |
155 | } | |
156 | ||
157 | unsigned int bvec_nr_vecs(unsigned short idx) | |
158 | { | |
159 | return bvec_slabs[idx].nr_vecs; | |
160 | } | |
161 | ||
162 | void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) | |
163 | { | |
164 | BIO_BUG_ON(idx >= BIOVEC_NR_POOLS); | |
165 | ||
166 | if (idx == BIOVEC_MAX_IDX) | |
167 | mempool_free(bv, pool); | |
168 | else { | |
169 | struct biovec_slab *bvs = bvec_slabs + idx; | |
170 | ||
171 | kmem_cache_free(bvs->slab, bv); | |
172 | } | |
173 | } | |
174 | ||
175 | struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, | |
176 | mempool_t *pool) | |
177 | { | |
178 | struct bio_vec *bvl; | |
179 | ||
180 | /* | |
181 | * see comment near bvec_array define! | |
182 | */ | |
183 | switch (nr) { | |
184 | case 1: | |
185 | *idx = 0; | |
186 | break; | |
187 | case 2 ... 4: | |
188 | *idx = 1; | |
189 | break; | |
190 | case 5 ... 16: | |
191 | *idx = 2; | |
192 | break; | |
193 | case 17 ... 64: | |
194 | *idx = 3; | |
195 | break; | |
196 | case 65 ... 128: | |
197 | *idx = 4; | |
198 | break; | |
199 | case 129 ... BIO_MAX_PAGES: | |
200 | *idx = 5; | |
201 | break; | |
202 | default: | |
203 | return NULL; | |
204 | } | |
205 | ||
206 | /* | |
207 | * idx now points to the pool we want to allocate from. only the | |
208 | * 1-vec entry pool is mempool backed. | |
209 | */ | |
210 | if (*idx == BIOVEC_MAX_IDX) { | |
211 | fallback: | |
212 | bvl = mempool_alloc(pool, gfp_mask); | |
213 | } else { | |
214 | struct biovec_slab *bvs = bvec_slabs + *idx; | |
215 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO); | |
216 | ||
217 | /* | |
218 | * Make this allocation restricted and don't dump info on | |
219 | * allocation failures, since we'll fallback to the mempool | |
220 | * in case of failure. | |
221 | */ | |
222 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; | |
223 | ||
224 | /* | |
225 | * Try a slab allocation. If this fails and __GFP_WAIT | |
226 | * is set, retry with the 1-entry mempool | |
227 | */ | |
228 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); | |
229 | if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) { | |
230 | *idx = BIOVEC_MAX_IDX; | |
231 | goto fallback; | |
232 | } | |
233 | } | |
234 | ||
235 | return bvl; | |
236 | } | |
237 | ||
238 | static void __bio_free(struct bio *bio) | |
239 | { | |
240 | bio_disassociate_task(bio); | |
241 | ||
242 | if (bio_integrity(bio)) | |
243 | bio_integrity_free(bio); | |
244 | } | |
245 | ||
246 | static void bio_free(struct bio *bio) | |
247 | { | |
248 | struct bio_set *bs = bio->bi_pool; | |
249 | void *p; | |
250 | ||
251 | __bio_free(bio); | |
252 | ||
253 | if (bs) { | |
254 | if (bio_flagged(bio, BIO_OWNS_VEC)) | |
255 | bvec_free(bs->bvec_pool, bio->bi_io_vec, BIO_POOL_IDX(bio)); | |
256 | ||
257 | /* | |
258 | * If we have front padding, adjust the bio pointer before freeing | |
259 | */ | |
260 | p = bio; | |
261 | p -= bs->front_pad; | |
262 | ||
263 | mempool_free(p, bs->bio_pool); | |
264 | } else { | |
265 | /* Bio was allocated by bio_kmalloc() */ | |
266 | kfree(bio); | |
267 | } | |
268 | } | |
269 | ||
270 | void bio_init(struct bio *bio) | |
271 | { | |
272 | memset(bio, 0, sizeof(*bio)); | |
273 | bio->bi_flags = 1 << BIO_UPTODATE; | |
274 | atomic_set(&bio->bi_remaining, 1); | |
275 | atomic_set(&bio->bi_cnt, 1); | |
276 | } | |
277 | EXPORT_SYMBOL(bio_init); | |
278 | ||
279 | /** | |
280 | * bio_reset - reinitialize a bio | |
281 | * @bio: bio to reset | |
282 | * | |
283 | * Description: | |
284 | * After calling bio_reset(), @bio will be in the same state as a freshly | |
285 | * allocated bio returned bio bio_alloc_bioset() - the only fields that are | |
286 | * preserved are the ones that are initialized by bio_alloc_bioset(). See | |
287 | * comment in struct bio. | |
288 | */ | |
289 | void bio_reset(struct bio *bio) | |
290 | { | |
291 | unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); | |
292 | ||
293 | __bio_free(bio); | |
294 | ||
295 | memset(bio, 0, BIO_RESET_BYTES); | |
296 | bio->bi_flags = flags|(1 << BIO_UPTODATE); | |
297 | atomic_set(&bio->bi_remaining, 1); | |
298 | } | |
299 | EXPORT_SYMBOL(bio_reset); | |
300 | ||
301 | static void bio_chain_endio(struct bio *bio, int error) | |
302 | { | |
303 | bio_endio(bio->bi_private, error); | |
304 | bio_put(bio); | |
305 | } | |
306 | ||
307 | /** | |
308 | * bio_chain - chain bio completions | |
309 | * | |
310 | * The caller won't have a bi_end_io called when @bio completes - instead, | |
311 | * @parent's bi_end_io won't be called until both @parent and @bio have | |
312 | * completed; the chained bio will also be freed when it completes. | |
313 | * | |
314 | * The caller must not set bi_private or bi_end_io in @bio. | |
315 | */ | |
316 | void bio_chain(struct bio *bio, struct bio *parent) | |
317 | { | |
318 | BUG_ON(bio->bi_private || bio->bi_end_io); | |
319 | ||
320 | bio->bi_private = parent; | |
321 | bio->bi_end_io = bio_chain_endio; | |
322 | atomic_inc(&parent->bi_remaining); | |
323 | } | |
324 | EXPORT_SYMBOL(bio_chain); | |
325 | ||
326 | static void bio_alloc_rescue(struct work_struct *work) | |
327 | { | |
328 | struct bio_set *bs = container_of(work, struct bio_set, rescue_work); | |
329 | struct bio *bio; | |
330 | ||
331 | while (1) { | |
332 | spin_lock(&bs->rescue_lock); | |
333 | bio = bio_list_pop(&bs->rescue_list); | |
334 | spin_unlock(&bs->rescue_lock); | |
335 | ||
336 | if (!bio) | |
337 | break; | |
338 | ||
339 | generic_make_request(bio); | |
340 | } | |
341 | } | |
342 | ||
343 | static void punt_bios_to_rescuer(struct bio_set *bs) | |
344 | { | |
345 | struct bio_list punt, nopunt; | |
346 | struct bio *bio; | |
347 | ||
348 | /* | |
349 | * In order to guarantee forward progress we must punt only bios that | |
350 | * were allocated from this bio_set; otherwise, if there was a bio on | |
351 | * there for a stacking driver higher up in the stack, processing it | |
352 | * could require allocating bios from this bio_set, and doing that from | |
353 | * our own rescuer would be bad. | |
354 | * | |
355 | * Since bio lists are singly linked, pop them all instead of trying to | |
356 | * remove from the middle of the list: | |
357 | */ | |
358 | ||
359 | bio_list_init(&punt); | |
360 | bio_list_init(&nopunt); | |
361 | ||
362 | while ((bio = bio_list_pop(current->bio_list))) | |
363 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); | |
364 | ||
365 | *current->bio_list = nopunt; | |
366 | ||
367 | spin_lock(&bs->rescue_lock); | |
368 | bio_list_merge(&bs->rescue_list, &punt); | |
369 | spin_unlock(&bs->rescue_lock); | |
370 | ||
371 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
372 | } | |
373 | ||
374 | /** | |
375 | * bio_alloc_bioset - allocate a bio for I/O | |
376 | * @gfp_mask: the GFP_ mask given to the slab allocator | |
377 | * @nr_iovecs: number of iovecs to pre-allocate | |
378 | * @bs: the bio_set to allocate from. | |
379 | * | |
380 | * Description: | |
381 | * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is | |
382 | * backed by the @bs's mempool. | |
383 | * | |
384 | * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be | |
385 | * able to allocate a bio. This is due to the mempool guarantees. To make this | |
386 | * work, callers must never allocate more than 1 bio at a time from this pool. | |
387 | * Callers that need to allocate more than 1 bio must always submit the | |
388 | * previously allocated bio for IO before attempting to allocate a new one. | |
389 | * Failure to do so can cause deadlocks under memory pressure. | |
390 | * | |
391 | * Note that when running under generic_make_request() (i.e. any block | |
392 | * driver), bios are not submitted until after you return - see the code in | |
393 | * generic_make_request() that converts recursion into iteration, to prevent | |
394 | * stack overflows. | |
395 | * | |
396 | * This would normally mean allocating multiple bios under | |
397 | * generic_make_request() would be susceptible to deadlocks, but we have | |
398 | * deadlock avoidance code that resubmits any blocked bios from a rescuer | |
399 | * thread. | |
400 | * | |
401 | * However, we do not guarantee forward progress for allocations from other | |
402 | * mempools. Doing multiple allocations from the same mempool under | |
403 | * generic_make_request() should be avoided - instead, use bio_set's front_pad | |
404 | * for per bio allocations. | |
405 | * | |
406 | * RETURNS: | |
407 | * Pointer to new bio on success, NULL on failure. | |
408 | */ | |
409 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) | |
410 | { | |
411 | gfp_t saved_gfp = gfp_mask; | |
412 | unsigned front_pad; | |
413 | unsigned inline_vecs; | |
414 | unsigned long idx = BIO_POOL_NONE; | |
415 | struct bio_vec *bvl = NULL; | |
416 | struct bio *bio; | |
417 | void *p; | |
418 | ||
419 | if (!bs) { | |
420 | if (nr_iovecs > UIO_MAXIOV) | |
421 | return NULL; | |
422 | ||
423 | p = kmalloc(sizeof(struct bio) + | |
424 | nr_iovecs * sizeof(struct bio_vec), | |
425 | gfp_mask); | |
426 | front_pad = 0; | |
427 | inline_vecs = nr_iovecs; | |
428 | } else { | |
429 | /* | |
430 | * generic_make_request() converts recursion to iteration; this | |
431 | * means if we're running beneath it, any bios we allocate and | |
432 | * submit will not be submitted (and thus freed) until after we | |
433 | * return. | |
434 | * | |
435 | * This exposes us to a potential deadlock if we allocate | |
436 | * multiple bios from the same bio_set() while running | |
437 | * underneath generic_make_request(). If we were to allocate | |
438 | * multiple bios (say a stacking block driver that was splitting | |
439 | * bios), we would deadlock if we exhausted the mempool's | |
440 | * reserve. | |
441 | * | |
442 | * We solve this, and guarantee forward progress, with a rescuer | |
443 | * workqueue per bio_set. If we go to allocate and there are | |
444 | * bios on current->bio_list, we first try the allocation | |
445 | * without __GFP_WAIT; if that fails, we punt those bios we | |
446 | * would be blocking to the rescuer workqueue before we retry | |
447 | * with the original gfp_flags. | |
448 | */ | |
449 | ||
450 | if (current->bio_list && !bio_list_empty(current->bio_list)) | |
451 | gfp_mask &= ~__GFP_WAIT; | |
452 | ||
453 | p = mempool_alloc(bs->bio_pool, gfp_mask); | |
454 | if (!p && gfp_mask != saved_gfp) { | |
455 | punt_bios_to_rescuer(bs); | |
456 | gfp_mask = saved_gfp; | |
457 | p = mempool_alloc(bs->bio_pool, gfp_mask); | |
458 | } | |
459 | ||
460 | front_pad = bs->front_pad; | |
461 | inline_vecs = BIO_INLINE_VECS; | |
462 | } | |
463 | ||
464 | if (unlikely(!p)) | |
465 | return NULL; | |
466 | ||
467 | bio = p + front_pad; | |
468 | bio_init(bio); | |
469 | ||
470 | if (nr_iovecs > inline_vecs) { | |
471 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); | |
472 | if (!bvl && gfp_mask != saved_gfp) { | |
473 | punt_bios_to_rescuer(bs); | |
474 | gfp_mask = saved_gfp; | |
475 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); | |
476 | } | |
477 | ||
478 | if (unlikely(!bvl)) | |
479 | goto err_free; | |
480 | ||
481 | bio->bi_flags |= 1 << BIO_OWNS_VEC; | |
482 | } else if (nr_iovecs) { | |
483 | bvl = bio->bi_inline_vecs; | |
484 | } | |
485 | ||
486 | bio->bi_pool = bs; | |
487 | bio->bi_flags |= idx << BIO_POOL_OFFSET; | |
488 | bio->bi_max_vecs = nr_iovecs; | |
489 | bio->bi_io_vec = bvl; | |
490 | return bio; | |
491 | ||
492 | err_free: | |
493 | mempool_free(p, bs->bio_pool); | |
494 | return NULL; | |
495 | } | |
496 | EXPORT_SYMBOL(bio_alloc_bioset); | |
497 | ||
498 | void zero_fill_bio(struct bio *bio) | |
499 | { | |
500 | unsigned long flags; | |
501 | struct bio_vec bv; | |
502 | struct bvec_iter iter; | |
503 | ||
504 | bio_for_each_segment(bv, bio, iter) { | |
505 | char *data = bvec_kmap_irq(&bv, &flags); | |
506 | memset(data, 0, bv.bv_len); | |
507 | flush_dcache_page(bv.bv_page); | |
508 | bvec_kunmap_irq(data, &flags); | |
509 | } | |
510 | } | |
511 | EXPORT_SYMBOL(zero_fill_bio); | |
512 | ||
513 | /** | |
514 | * bio_put - release a reference to a bio | |
515 | * @bio: bio to release reference to | |
516 | * | |
517 | * Description: | |
518 | * Put a reference to a &struct bio, either one you have gotten with | |
519 | * bio_alloc, bio_get or bio_clone. The last put of a bio will free it. | |
520 | **/ | |
521 | void bio_put(struct bio *bio) | |
522 | { | |
523 | BIO_BUG_ON(!atomic_read(&bio->bi_cnt)); | |
524 | ||
525 | /* | |
526 | * last put frees it | |
527 | */ | |
528 | if (atomic_dec_and_test(&bio->bi_cnt)) | |
529 | bio_free(bio); | |
530 | } | |
531 | EXPORT_SYMBOL(bio_put); | |
532 | ||
533 | inline int bio_phys_segments(struct request_queue *q, struct bio *bio) | |
534 | { | |
535 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
536 | blk_recount_segments(q, bio); | |
537 | ||
538 | return bio->bi_phys_segments; | |
539 | } | |
540 | EXPORT_SYMBOL(bio_phys_segments); | |
541 | ||
542 | /** | |
543 | * __bio_clone_fast - clone a bio that shares the original bio's biovec | |
544 | * @bio: destination bio | |
545 | * @bio_src: bio to clone | |
546 | * | |
547 | * Clone a &bio. Caller will own the returned bio, but not | |
548 | * the actual data it points to. Reference count of returned | |
549 | * bio will be one. | |
550 | * | |
551 | * Caller must ensure that @bio_src is not freed before @bio. | |
552 | */ | |
553 | void __bio_clone_fast(struct bio *bio, struct bio *bio_src) | |
554 | { | |
555 | BUG_ON(bio->bi_pool && BIO_POOL_IDX(bio) != BIO_POOL_NONE); | |
556 | ||
557 | /* | |
558 | * most users will be overriding ->bi_bdev with a new target, | |
559 | * so we don't set nor calculate new physical/hw segment counts here | |
560 | */ | |
561 | bio->bi_bdev = bio_src->bi_bdev; | |
562 | bio->bi_flags |= 1 << BIO_CLONED; | |
563 | bio->bi_rw = bio_src->bi_rw; | |
564 | bio->bi_iter = bio_src->bi_iter; | |
565 | bio->bi_io_vec = bio_src->bi_io_vec; | |
566 | } | |
567 | EXPORT_SYMBOL(__bio_clone_fast); | |
568 | ||
569 | /** | |
570 | * bio_clone_fast - clone a bio that shares the original bio's biovec | |
571 | * @bio: bio to clone | |
572 | * @gfp_mask: allocation priority | |
573 | * @bs: bio_set to allocate from | |
574 | * | |
575 | * Like __bio_clone_fast, only also allocates the returned bio | |
576 | */ | |
577 | struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) | |
578 | { | |
579 | struct bio *b; | |
580 | ||
581 | b = bio_alloc_bioset(gfp_mask, 0, bs); | |
582 | if (!b) | |
583 | return NULL; | |
584 | ||
585 | __bio_clone_fast(b, bio); | |
586 | ||
587 | if (bio_integrity(bio)) { | |
588 | int ret; | |
589 | ||
590 | ret = bio_integrity_clone(b, bio, gfp_mask); | |
591 | ||
592 | if (ret < 0) { | |
593 | bio_put(b); | |
594 | return NULL; | |
595 | } | |
596 | } | |
597 | ||
598 | return b; | |
599 | } | |
600 | EXPORT_SYMBOL(bio_clone_fast); | |
601 | ||
602 | /** | |
603 | * bio_clone_bioset - clone a bio | |
604 | * @bio_src: bio to clone | |
605 | * @gfp_mask: allocation priority | |
606 | * @bs: bio_set to allocate from | |
607 | * | |
608 | * Clone bio. Caller will own the returned bio, but not the actual data it | |
609 | * points to. Reference count of returned bio will be one. | |
610 | */ | |
611 | struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask, | |
612 | struct bio_set *bs) | |
613 | { | |
614 | unsigned nr_iovecs = 0; | |
615 | struct bvec_iter iter; | |
616 | struct bio_vec bv; | |
617 | struct bio *bio; | |
618 | ||
619 | /* | |
620 | * Pre immutable biovecs, __bio_clone() used to just do a memcpy from | |
621 | * bio_src->bi_io_vec to bio->bi_io_vec. | |
622 | * | |
623 | * We can't do that anymore, because: | |
624 | * | |
625 | * - The point of cloning the biovec is to produce a bio with a biovec | |
626 | * the caller can modify: bi_idx and bi_bvec_done should be 0. | |
627 | * | |
628 | * - The original bio could've had more than BIO_MAX_PAGES biovecs; if | |
629 | * we tried to clone the whole thing bio_alloc_bioset() would fail. | |
630 | * But the clone should succeed as long as the number of biovecs we | |
631 | * actually need to allocate is fewer than BIO_MAX_PAGES. | |
632 | * | |
633 | * - Lastly, bi_vcnt should not be looked at or relied upon by code | |
634 | * that does not own the bio - reason being drivers don't use it for | |
635 | * iterating over the biovec anymore, so expecting it to be kept up | |
636 | * to date (i.e. for clones that share the parent biovec) is just | |
637 | * asking for trouble and would force extra work on | |
638 | * __bio_clone_fast() anyways. | |
639 | */ | |
640 | ||
641 | bio_for_each_segment(bv, bio_src, iter) | |
642 | nr_iovecs++; | |
643 | ||
644 | bio = bio_alloc_bioset(gfp_mask, nr_iovecs, bs); | |
645 | if (!bio) | |
646 | return NULL; | |
647 | ||
648 | bio->bi_bdev = bio_src->bi_bdev; | |
649 | bio->bi_rw = bio_src->bi_rw; | |
650 | bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector; | |
651 | bio->bi_iter.bi_size = bio_src->bi_iter.bi_size; | |
652 | ||
653 | bio_for_each_segment(bv, bio_src, iter) | |
654 | bio->bi_io_vec[bio->bi_vcnt++] = bv; | |
655 | ||
656 | if (bio_integrity(bio_src)) { | |
657 | int ret; | |
658 | ||
659 | ret = bio_integrity_clone(bio, bio_src, gfp_mask); | |
660 | if (ret < 0) { | |
661 | bio_put(bio); | |
662 | return NULL; | |
663 | } | |
664 | } | |
665 | ||
666 | return bio; | |
667 | } | |
668 | EXPORT_SYMBOL(bio_clone_bioset); | |
669 | ||
670 | /** | |
671 | * bio_get_nr_vecs - return approx number of vecs | |
672 | * @bdev: I/O target | |
673 | * | |
674 | * Return the approximate number of pages we can send to this target. | |
675 | * There's no guarantee that you will be able to fit this number of pages | |
676 | * into a bio, it does not account for dynamic restrictions that vary | |
677 | * on offset. | |
678 | */ | |
679 | int bio_get_nr_vecs(struct block_device *bdev) | |
680 | { | |
681 | struct request_queue *q = bdev_get_queue(bdev); | |
682 | int nr_pages; | |
683 | ||
684 | nr_pages = min_t(unsigned, | |
685 | queue_max_segments(q), | |
686 | queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1); | |
687 | ||
688 | return min_t(unsigned, nr_pages, BIO_MAX_PAGES); | |
689 | ||
690 | } | |
691 | EXPORT_SYMBOL(bio_get_nr_vecs); | |
692 | ||
693 | static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page | |
694 | *page, unsigned int len, unsigned int offset, | |
695 | unsigned int max_sectors) | |
696 | { | |
697 | int retried_segments = 0; | |
698 | struct bio_vec *bvec; | |
699 | ||
700 | /* | |
701 | * cloned bio must not modify vec list | |
702 | */ | |
703 | if (unlikely(bio_flagged(bio, BIO_CLONED))) | |
704 | return 0; | |
705 | ||
706 | if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) | |
707 | return 0; | |
708 | ||
709 | /* | |
710 | * For filesystems with a blocksize smaller than the pagesize | |
711 | * we will often be called with the same page as last time and | |
712 | * a consecutive offset. Optimize this special case. | |
713 | */ | |
714 | if (bio->bi_vcnt > 0) { | |
715 | struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
716 | ||
717 | if (page == prev->bv_page && | |
718 | offset == prev->bv_offset + prev->bv_len) { | |
719 | unsigned int prev_bv_len = prev->bv_len; | |
720 | prev->bv_len += len; | |
721 | ||
722 | if (q->merge_bvec_fn) { | |
723 | struct bvec_merge_data bvm = { | |
724 | /* prev_bvec is already charged in | |
725 | bi_size, discharge it in order to | |
726 | simulate merging updated prev_bvec | |
727 | as new bvec. */ | |
728 | .bi_bdev = bio->bi_bdev, | |
729 | .bi_sector = bio->bi_iter.bi_sector, | |
730 | .bi_size = bio->bi_iter.bi_size - | |
731 | prev_bv_len, | |
732 | .bi_rw = bio->bi_rw, | |
733 | }; | |
734 | ||
735 | if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) { | |
736 | prev->bv_len -= len; | |
737 | return 0; | |
738 | } | |
739 | } | |
740 | ||
741 | goto done; | |
742 | } | |
743 | } | |
744 | ||
745 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
746 | return 0; | |
747 | ||
748 | /* | |
749 | * we might lose a segment or two here, but rather that than | |
750 | * make this too complex. | |
751 | */ | |
752 | ||
753 | while (bio->bi_phys_segments >= queue_max_segments(q)) { | |
754 | ||
755 | if (retried_segments) | |
756 | return 0; | |
757 | ||
758 | retried_segments = 1; | |
759 | blk_recount_segments(q, bio); | |
760 | } | |
761 | ||
762 | /* | |
763 | * setup the new entry, we might clear it again later if we | |
764 | * cannot add the page | |
765 | */ | |
766 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; | |
767 | bvec->bv_page = page; | |
768 | bvec->bv_len = len; | |
769 | bvec->bv_offset = offset; | |
770 | ||
771 | /* | |
772 | * if queue has other restrictions (eg varying max sector size | |
773 | * depending on offset), it can specify a merge_bvec_fn in the | |
774 | * queue to get further control | |
775 | */ | |
776 | if (q->merge_bvec_fn) { | |
777 | struct bvec_merge_data bvm = { | |
778 | .bi_bdev = bio->bi_bdev, | |
779 | .bi_sector = bio->bi_iter.bi_sector, | |
780 | .bi_size = bio->bi_iter.bi_size, | |
781 | .bi_rw = bio->bi_rw, | |
782 | }; | |
783 | ||
784 | /* | |
785 | * merge_bvec_fn() returns number of bytes it can accept | |
786 | * at this offset | |
787 | */ | |
788 | if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) { | |
789 | bvec->bv_page = NULL; | |
790 | bvec->bv_len = 0; | |
791 | bvec->bv_offset = 0; | |
792 | return 0; | |
793 | } | |
794 | } | |
795 | ||
796 | /* If we may be able to merge these biovecs, force a recount */ | |
797 | if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) | |
798 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
799 | ||
800 | bio->bi_vcnt++; | |
801 | bio->bi_phys_segments++; | |
802 | done: | |
803 | bio->bi_iter.bi_size += len; | |
804 | return len; | |
805 | } | |
806 | ||
807 | /** | |
808 | * bio_add_pc_page - attempt to add page to bio | |
809 | * @q: the target queue | |
810 | * @bio: destination bio | |
811 | * @page: page to add | |
812 | * @len: vec entry length | |
813 | * @offset: vec entry offset | |
814 | * | |
815 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
816 | * number of reasons, such as the bio being full or target block device | |
817 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
818 | * so it is always possible to add a single page to an empty bio. | |
819 | * | |
820 | * This should only be used by REQ_PC bios. | |
821 | */ | |
822 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page, | |
823 | unsigned int len, unsigned int offset) | |
824 | { | |
825 | return __bio_add_page(q, bio, page, len, offset, | |
826 | queue_max_hw_sectors(q)); | |
827 | } | |
828 | EXPORT_SYMBOL(bio_add_pc_page); | |
829 | ||
830 | /** | |
831 | * bio_add_page - attempt to add page to bio | |
832 | * @bio: destination bio | |
833 | * @page: page to add | |
834 | * @len: vec entry length | |
835 | * @offset: vec entry offset | |
836 | * | |
837 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
838 | * number of reasons, such as the bio being full or target block device | |
839 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
840 | * so it is always possible to add a single page to an empty bio. | |
841 | */ | |
842 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, | |
843 | unsigned int offset) | |
844 | { | |
845 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); | |
846 | return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q)); | |
847 | } | |
848 | EXPORT_SYMBOL(bio_add_page); | |
849 | ||
850 | struct submit_bio_ret { | |
851 | struct completion event; | |
852 | int error; | |
853 | }; | |
854 | ||
855 | static void submit_bio_wait_endio(struct bio *bio, int error) | |
856 | { | |
857 | struct submit_bio_ret *ret = bio->bi_private; | |
858 | ||
859 | ret->error = error; | |
860 | complete(&ret->event); | |
861 | } | |
862 | ||
863 | /** | |
864 | * submit_bio_wait - submit a bio, and wait until it completes | |
865 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | |
866 | * @bio: The &struct bio which describes the I/O | |
867 | * | |
868 | * Simple wrapper around submit_bio(). Returns 0 on success, or the error from | |
869 | * bio_endio() on failure. | |
870 | */ | |
871 | int submit_bio_wait(int rw, struct bio *bio) | |
872 | { | |
873 | struct submit_bio_ret ret; | |
874 | ||
875 | rw |= REQ_SYNC; | |
876 | init_completion(&ret.event); | |
877 | bio->bi_private = &ret; | |
878 | bio->bi_end_io = submit_bio_wait_endio; | |
879 | submit_bio(rw, bio); | |
880 | wait_for_completion(&ret.event); | |
881 | ||
882 | return ret.error; | |
883 | } | |
884 | EXPORT_SYMBOL(submit_bio_wait); | |
885 | ||
886 | /** | |
887 | * bio_advance - increment/complete a bio by some number of bytes | |
888 | * @bio: bio to advance | |
889 | * @bytes: number of bytes to complete | |
890 | * | |
891 | * This updates bi_sector, bi_size and bi_idx; if the number of bytes to | |
892 | * complete doesn't align with a bvec boundary, then bv_len and bv_offset will | |
893 | * be updated on the last bvec as well. | |
894 | * | |
895 | * @bio will then represent the remaining, uncompleted portion of the io. | |
896 | */ | |
897 | void bio_advance(struct bio *bio, unsigned bytes) | |
898 | { | |
899 | if (bio_integrity(bio)) | |
900 | bio_integrity_advance(bio, bytes); | |
901 | ||
902 | bio_advance_iter(bio, &bio->bi_iter, bytes); | |
903 | } | |
904 | EXPORT_SYMBOL(bio_advance); | |
905 | ||
906 | /** | |
907 | * bio_alloc_pages - allocates a single page for each bvec in a bio | |
908 | * @bio: bio to allocate pages for | |
909 | * @gfp_mask: flags for allocation | |
910 | * | |
911 | * Allocates pages up to @bio->bi_vcnt. | |
912 | * | |
913 | * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are | |
914 | * freed. | |
915 | */ | |
916 | int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask) | |
917 | { | |
918 | int i; | |
919 | struct bio_vec *bv; | |
920 | ||
921 | bio_for_each_segment_all(bv, bio, i) { | |
922 | bv->bv_page = alloc_page(gfp_mask); | |
923 | if (!bv->bv_page) { | |
924 | while (--bv >= bio->bi_io_vec) | |
925 | __free_page(bv->bv_page); | |
926 | return -ENOMEM; | |
927 | } | |
928 | } | |
929 | ||
930 | return 0; | |
931 | } | |
932 | EXPORT_SYMBOL(bio_alloc_pages); | |
933 | ||
934 | /** | |
935 | * bio_copy_data - copy contents of data buffers from one chain of bios to | |
936 | * another | |
937 | * @src: source bio list | |
938 | * @dst: destination bio list | |
939 | * | |
940 | * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats | |
941 | * @src and @dst as linked lists of bios. | |
942 | * | |
943 | * Stops when it reaches the end of either @src or @dst - that is, copies | |
944 | * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). | |
945 | */ | |
946 | void bio_copy_data(struct bio *dst, struct bio *src) | |
947 | { | |
948 | struct bvec_iter src_iter, dst_iter; | |
949 | struct bio_vec src_bv, dst_bv; | |
950 | void *src_p, *dst_p; | |
951 | unsigned bytes; | |
952 | ||
953 | src_iter = src->bi_iter; | |
954 | dst_iter = dst->bi_iter; | |
955 | ||
956 | while (1) { | |
957 | if (!src_iter.bi_size) { | |
958 | src = src->bi_next; | |
959 | if (!src) | |
960 | break; | |
961 | ||
962 | src_iter = src->bi_iter; | |
963 | } | |
964 | ||
965 | if (!dst_iter.bi_size) { | |
966 | dst = dst->bi_next; | |
967 | if (!dst) | |
968 | break; | |
969 | ||
970 | dst_iter = dst->bi_iter; | |
971 | } | |
972 | ||
973 | src_bv = bio_iter_iovec(src, src_iter); | |
974 | dst_bv = bio_iter_iovec(dst, dst_iter); | |
975 | ||
976 | bytes = min(src_bv.bv_len, dst_bv.bv_len); | |
977 | ||
978 | src_p = kmap_atomic(src_bv.bv_page); | |
979 | dst_p = kmap_atomic(dst_bv.bv_page); | |
980 | ||
981 | memcpy(dst_p + dst_bv.bv_offset, | |
982 | src_p + src_bv.bv_offset, | |
983 | bytes); | |
984 | ||
985 | kunmap_atomic(dst_p); | |
986 | kunmap_atomic(src_p); | |
987 | ||
988 | bio_advance_iter(src, &src_iter, bytes); | |
989 | bio_advance_iter(dst, &dst_iter, bytes); | |
990 | } | |
991 | } | |
992 | EXPORT_SYMBOL(bio_copy_data); | |
993 | ||
994 | struct bio_map_data { | |
995 | int nr_sgvecs; | |
996 | int is_our_pages; | |
997 | struct sg_iovec sgvecs[]; | |
998 | }; | |
999 | ||
1000 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio, | |
1001 | struct sg_iovec *iov, int iov_count, | |
1002 | int is_our_pages) | |
1003 | { | |
1004 | memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count); | |
1005 | bmd->nr_sgvecs = iov_count; | |
1006 | bmd->is_our_pages = is_our_pages; | |
1007 | bio->bi_private = bmd; | |
1008 | } | |
1009 | ||
1010 | static struct bio_map_data *bio_alloc_map_data(int nr_segs, | |
1011 | unsigned int iov_count, | |
1012 | gfp_t gfp_mask) | |
1013 | { | |
1014 | if (iov_count > UIO_MAXIOV) | |
1015 | return NULL; | |
1016 | ||
1017 | return kmalloc(sizeof(struct bio_map_data) + | |
1018 | sizeof(struct sg_iovec) * iov_count, gfp_mask); | |
1019 | } | |
1020 | ||
1021 | static int __bio_copy_iov(struct bio *bio, struct sg_iovec *iov, int iov_count, | |
1022 | int to_user, int from_user, int do_free_page) | |
1023 | { | |
1024 | int ret = 0, i; | |
1025 | struct bio_vec *bvec; | |
1026 | int iov_idx = 0; | |
1027 | unsigned int iov_off = 0; | |
1028 | ||
1029 | bio_for_each_segment_all(bvec, bio, i) { | |
1030 | char *bv_addr = page_address(bvec->bv_page); | |
1031 | unsigned int bv_len = bvec->bv_len; | |
1032 | ||
1033 | while (bv_len && iov_idx < iov_count) { | |
1034 | unsigned int bytes; | |
1035 | char __user *iov_addr; | |
1036 | ||
1037 | bytes = min_t(unsigned int, | |
1038 | iov[iov_idx].iov_len - iov_off, bv_len); | |
1039 | iov_addr = iov[iov_idx].iov_base + iov_off; | |
1040 | ||
1041 | if (!ret) { | |
1042 | if (to_user) | |
1043 | ret = copy_to_user(iov_addr, bv_addr, | |
1044 | bytes); | |
1045 | ||
1046 | if (from_user) | |
1047 | ret = copy_from_user(bv_addr, iov_addr, | |
1048 | bytes); | |
1049 | ||
1050 | if (ret) | |
1051 | ret = -EFAULT; | |
1052 | } | |
1053 | ||
1054 | bv_len -= bytes; | |
1055 | bv_addr += bytes; | |
1056 | iov_addr += bytes; | |
1057 | iov_off += bytes; | |
1058 | ||
1059 | if (iov[iov_idx].iov_len == iov_off) { | |
1060 | iov_idx++; | |
1061 | iov_off = 0; | |
1062 | } | |
1063 | } | |
1064 | ||
1065 | if (do_free_page) | |
1066 | __free_page(bvec->bv_page); | |
1067 | } | |
1068 | ||
1069 | return ret; | |
1070 | } | |
1071 | ||
1072 | /** | |
1073 | * bio_uncopy_user - finish previously mapped bio | |
1074 | * @bio: bio being terminated | |
1075 | * | |
1076 | * Free pages allocated from bio_copy_user() and write back data | |
1077 | * to user space in case of a read. | |
1078 | */ | |
1079 | int bio_uncopy_user(struct bio *bio) | |
1080 | { | |
1081 | struct bio_map_data *bmd = bio->bi_private; | |
1082 | struct bio_vec *bvec; | |
1083 | int ret = 0, i; | |
1084 | ||
1085 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) { | |
1086 | /* | |
1087 | * if we're in a workqueue, the request is orphaned, so | |
1088 | * don't copy into a random user address space, just free. | |
1089 | */ | |
1090 | if (current->mm) | |
1091 | ret = __bio_copy_iov(bio, bmd->sgvecs, bmd->nr_sgvecs, | |
1092 | bio_data_dir(bio) == READ, | |
1093 | 0, bmd->is_our_pages); | |
1094 | else if (bmd->is_our_pages) | |
1095 | bio_for_each_segment_all(bvec, bio, i) | |
1096 | __free_page(bvec->bv_page); | |
1097 | } | |
1098 | kfree(bmd); | |
1099 | bio_put(bio); | |
1100 | return ret; | |
1101 | } | |
1102 | EXPORT_SYMBOL(bio_uncopy_user); | |
1103 | ||
1104 | /** | |
1105 | * bio_copy_user_iov - copy user data to bio | |
1106 | * @q: destination block queue | |
1107 | * @map_data: pointer to the rq_map_data holding pages (if necessary) | |
1108 | * @iov: the iovec. | |
1109 | * @iov_count: number of elements in the iovec | |
1110 | * @write_to_vm: bool indicating writing to pages or not | |
1111 | * @gfp_mask: memory allocation flags | |
1112 | * | |
1113 | * Prepares and returns a bio for indirect user io, bouncing data | |
1114 | * to/from kernel pages as necessary. Must be paired with | |
1115 | * call bio_uncopy_user() on io completion. | |
1116 | */ | |
1117 | struct bio *bio_copy_user_iov(struct request_queue *q, | |
1118 | struct rq_map_data *map_data, | |
1119 | struct sg_iovec *iov, int iov_count, | |
1120 | int write_to_vm, gfp_t gfp_mask) | |
1121 | { | |
1122 | struct bio_map_data *bmd; | |
1123 | struct bio_vec *bvec; | |
1124 | struct page *page; | |
1125 | struct bio *bio; | |
1126 | int i, ret; | |
1127 | int nr_pages = 0; | |
1128 | unsigned int len = 0; | |
1129 | unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0; | |
1130 | ||
1131 | for (i = 0; i < iov_count; i++) { | |
1132 | unsigned long uaddr; | |
1133 | unsigned long end; | |
1134 | unsigned long start; | |
1135 | ||
1136 | uaddr = (unsigned long)iov[i].iov_base; | |
1137 | end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1138 | start = uaddr >> PAGE_SHIFT; | |
1139 | ||
1140 | /* | |
1141 | * Overflow, abort | |
1142 | */ | |
1143 | if (end < start) | |
1144 | return ERR_PTR(-EINVAL); | |
1145 | ||
1146 | nr_pages += end - start; | |
1147 | len += iov[i].iov_len; | |
1148 | } | |
1149 | ||
1150 | if (offset) | |
1151 | nr_pages++; | |
1152 | ||
1153 | bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask); | |
1154 | if (!bmd) | |
1155 | return ERR_PTR(-ENOMEM); | |
1156 | ||
1157 | ret = -ENOMEM; | |
1158 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1159 | if (!bio) | |
1160 | goto out_bmd; | |
1161 | ||
1162 | if (!write_to_vm) | |
1163 | bio->bi_rw |= REQ_WRITE; | |
1164 | ||
1165 | ret = 0; | |
1166 | ||
1167 | if (map_data) { | |
1168 | nr_pages = 1 << map_data->page_order; | |
1169 | i = map_data->offset / PAGE_SIZE; | |
1170 | } | |
1171 | while (len) { | |
1172 | unsigned int bytes = PAGE_SIZE; | |
1173 | ||
1174 | bytes -= offset; | |
1175 | ||
1176 | if (bytes > len) | |
1177 | bytes = len; | |
1178 | ||
1179 | if (map_data) { | |
1180 | if (i == map_data->nr_entries * nr_pages) { | |
1181 | ret = -ENOMEM; | |
1182 | break; | |
1183 | } | |
1184 | ||
1185 | page = map_data->pages[i / nr_pages]; | |
1186 | page += (i % nr_pages); | |
1187 | ||
1188 | i++; | |
1189 | } else { | |
1190 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
1191 | if (!page) { | |
1192 | ret = -ENOMEM; | |
1193 | break; | |
1194 | } | |
1195 | } | |
1196 | ||
1197 | if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) | |
1198 | break; | |
1199 | ||
1200 | len -= bytes; | |
1201 | offset = 0; | |
1202 | } | |
1203 | ||
1204 | if (ret) | |
1205 | goto cleanup; | |
1206 | ||
1207 | /* | |
1208 | * success | |
1209 | */ | |
1210 | if ((!write_to_vm && (!map_data || !map_data->null_mapped)) || | |
1211 | (map_data && map_data->from_user)) { | |
1212 | ret = __bio_copy_iov(bio, iov, iov_count, 0, 1, 0); | |
1213 | if (ret) | |
1214 | goto cleanup; | |
1215 | } | |
1216 | ||
1217 | bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1); | |
1218 | return bio; | |
1219 | cleanup: | |
1220 | if (!map_data) | |
1221 | bio_for_each_segment_all(bvec, bio, i) | |
1222 | __free_page(bvec->bv_page); | |
1223 | ||
1224 | bio_put(bio); | |
1225 | out_bmd: | |
1226 | kfree(bmd); | |
1227 | return ERR_PTR(ret); | |
1228 | } | |
1229 | ||
1230 | /** | |
1231 | * bio_copy_user - copy user data to bio | |
1232 | * @q: destination block queue | |
1233 | * @map_data: pointer to the rq_map_data holding pages (if necessary) | |
1234 | * @uaddr: start of user address | |
1235 | * @len: length in bytes | |
1236 | * @write_to_vm: bool indicating writing to pages or not | |
1237 | * @gfp_mask: memory allocation flags | |
1238 | * | |
1239 | * Prepares and returns a bio for indirect user io, bouncing data | |
1240 | * to/from kernel pages as necessary. Must be paired with | |
1241 | * call bio_uncopy_user() on io completion. | |
1242 | */ | |
1243 | struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data, | |
1244 | unsigned long uaddr, unsigned int len, | |
1245 | int write_to_vm, gfp_t gfp_mask) | |
1246 | { | |
1247 | struct sg_iovec iov; | |
1248 | ||
1249 | iov.iov_base = (void __user *)uaddr; | |
1250 | iov.iov_len = len; | |
1251 | ||
1252 | return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask); | |
1253 | } | |
1254 | EXPORT_SYMBOL(bio_copy_user); | |
1255 | ||
1256 | static struct bio *__bio_map_user_iov(struct request_queue *q, | |
1257 | struct block_device *bdev, | |
1258 | struct sg_iovec *iov, int iov_count, | |
1259 | int write_to_vm, gfp_t gfp_mask) | |
1260 | { | |
1261 | int i, j; | |
1262 | int nr_pages = 0; | |
1263 | struct page **pages; | |
1264 | struct bio *bio; | |
1265 | int cur_page = 0; | |
1266 | int ret, offset; | |
1267 | ||
1268 | for (i = 0; i < iov_count; i++) { | |
1269 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
1270 | unsigned long len = iov[i].iov_len; | |
1271 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1272 | unsigned long start = uaddr >> PAGE_SHIFT; | |
1273 | ||
1274 | /* | |
1275 | * Overflow, abort | |
1276 | */ | |
1277 | if (end < start) | |
1278 | return ERR_PTR(-EINVAL); | |
1279 | ||
1280 | nr_pages += end - start; | |
1281 | /* | |
1282 | * buffer must be aligned to at least hardsector size for now | |
1283 | */ | |
1284 | if (uaddr & queue_dma_alignment(q)) | |
1285 | return ERR_PTR(-EINVAL); | |
1286 | } | |
1287 | ||
1288 | if (!nr_pages) | |
1289 | return ERR_PTR(-EINVAL); | |
1290 | ||
1291 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1292 | if (!bio) | |
1293 | return ERR_PTR(-ENOMEM); | |
1294 | ||
1295 | ret = -ENOMEM; | |
1296 | pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask); | |
1297 | if (!pages) | |
1298 | goto out; | |
1299 | ||
1300 | for (i = 0; i < iov_count; i++) { | |
1301 | unsigned long uaddr = (unsigned long)iov[i].iov_base; | |
1302 | unsigned long len = iov[i].iov_len; | |
1303 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1304 | unsigned long start = uaddr >> PAGE_SHIFT; | |
1305 | const int local_nr_pages = end - start; | |
1306 | const int page_limit = cur_page + local_nr_pages; | |
1307 | ||
1308 | ret = get_user_pages_fast(uaddr, local_nr_pages, | |
1309 | write_to_vm, &pages[cur_page]); | |
1310 | if (ret < local_nr_pages) { | |
1311 | ret = -EFAULT; | |
1312 | goto out_unmap; | |
1313 | } | |
1314 | ||
1315 | offset = uaddr & ~PAGE_MASK; | |
1316 | for (j = cur_page; j < page_limit; j++) { | |
1317 | unsigned int bytes = PAGE_SIZE - offset; | |
1318 | ||
1319 | if (len <= 0) | |
1320 | break; | |
1321 | ||
1322 | if (bytes > len) | |
1323 | bytes = len; | |
1324 | ||
1325 | /* | |
1326 | * sorry... | |
1327 | */ | |
1328 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < | |
1329 | bytes) | |
1330 | break; | |
1331 | ||
1332 | len -= bytes; | |
1333 | offset = 0; | |
1334 | } | |
1335 | ||
1336 | cur_page = j; | |
1337 | /* | |
1338 | * release the pages we didn't map into the bio, if any | |
1339 | */ | |
1340 | while (j < page_limit) | |
1341 | page_cache_release(pages[j++]); | |
1342 | } | |
1343 | ||
1344 | kfree(pages); | |
1345 | ||
1346 | /* | |
1347 | * set data direction, and check if mapped pages need bouncing | |
1348 | */ | |
1349 | if (!write_to_vm) | |
1350 | bio->bi_rw |= REQ_WRITE; | |
1351 | ||
1352 | bio->bi_bdev = bdev; | |
1353 | bio->bi_flags |= (1 << BIO_USER_MAPPED); | |
1354 | return bio; | |
1355 | ||
1356 | out_unmap: | |
1357 | for (i = 0; i < nr_pages; i++) { | |
1358 | if(!pages[i]) | |
1359 | break; | |
1360 | page_cache_release(pages[i]); | |
1361 | } | |
1362 | out: | |
1363 | kfree(pages); | |
1364 | bio_put(bio); | |
1365 | return ERR_PTR(ret); | |
1366 | } | |
1367 | ||
1368 | /** | |
1369 | * bio_map_user - map user address into bio | |
1370 | * @q: the struct request_queue for the bio | |
1371 | * @bdev: destination block device | |
1372 | * @uaddr: start of user address | |
1373 | * @len: length in bytes | |
1374 | * @write_to_vm: bool indicating writing to pages or not | |
1375 | * @gfp_mask: memory allocation flags | |
1376 | * | |
1377 | * Map the user space address into a bio suitable for io to a block | |
1378 | * device. Returns an error pointer in case of error. | |
1379 | */ | |
1380 | struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev, | |
1381 | unsigned long uaddr, unsigned int len, int write_to_vm, | |
1382 | gfp_t gfp_mask) | |
1383 | { | |
1384 | struct sg_iovec iov; | |
1385 | ||
1386 | iov.iov_base = (void __user *)uaddr; | |
1387 | iov.iov_len = len; | |
1388 | ||
1389 | return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask); | |
1390 | } | |
1391 | EXPORT_SYMBOL(bio_map_user); | |
1392 | ||
1393 | /** | |
1394 | * bio_map_user_iov - map user sg_iovec table into bio | |
1395 | * @q: the struct request_queue for the bio | |
1396 | * @bdev: destination block device | |
1397 | * @iov: the iovec. | |
1398 | * @iov_count: number of elements in the iovec | |
1399 | * @write_to_vm: bool indicating writing to pages or not | |
1400 | * @gfp_mask: memory allocation flags | |
1401 | * | |
1402 | * Map the user space address into a bio suitable for io to a block | |
1403 | * device. Returns an error pointer in case of error. | |
1404 | */ | |
1405 | struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev, | |
1406 | struct sg_iovec *iov, int iov_count, | |
1407 | int write_to_vm, gfp_t gfp_mask) | |
1408 | { | |
1409 | struct bio *bio; | |
1410 | ||
1411 | bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm, | |
1412 | gfp_mask); | |
1413 | if (IS_ERR(bio)) | |
1414 | return bio; | |
1415 | ||
1416 | /* | |
1417 | * subtle -- if __bio_map_user() ended up bouncing a bio, | |
1418 | * it would normally disappear when its bi_end_io is run. | |
1419 | * however, we need it for the unmap, so grab an extra | |
1420 | * reference to it | |
1421 | */ | |
1422 | bio_get(bio); | |
1423 | ||
1424 | return bio; | |
1425 | } | |
1426 | ||
1427 | static void __bio_unmap_user(struct bio *bio) | |
1428 | { | |
1429 | struct bio_vec *bvec; | |
1430 | int i; | |
1431 | ||
1432 | /* | |
1433 | * make sure we dirty pages we wrote to | |
1434 | */ | |
1435 | bio_for_each_segment_all(bvec, bio, i) { | |
1436 | if (bio_data_dir(bio) == READ) | |
1437 | set_page_dirty_lock(bvec->bv_page); | |
1438 | ||
1439 | page_cache_release(bvec->bv_page); | |
1440 | } | |
1441 | ||
1442 | bio_put(bio); | |
1443 | } | |
1444 | ||
1445 | /** | |
1446 | * bio_unmap_user - unmap a bio | |
1447 | * @bio: the bio being unmapped | |
1448 | * | |
1449 | * Unmap a bio previously mapped by bio_map_user(). Must be called with | |
1450 | * a process context. | |
1451 | * | |
1452 | * bio_unmap_user() may sleep. | |
1453 | */ | |
1454 | void bio_unmap_user(struct bio *bio) | |
1455 | { | |
1456 | __bio_unmap_user(bio); | |
1457 | bio_put(bio); | |
1458 | } | |
1459 | EXPORT_SYMBOL(bio_unmap_user); | |
1460 | ||
1461 | static void bio_map_kern_endio(struct bio *bio, int err) | |
1462 | { | |
1463 | bio_put(bio); | |
1464 | } | |
1465 | ||
1466 | static struct bio *__bio_map_kern(struct request_queue *q, void *data, | |
1467 | unsigned int len, gfp_t gfp_mask) | |
1468 | { | |
1469 | unsigned long kaddr = (unsigned long)data; | |
1470 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1471 | unsigned long start = kaddr >> PAGE_SHIFT; | |
1472 | const int nr_pages = end - start; | |
1473 | int offset, i; | |
1474 | struct bio *bio; | |
1475 | ||
1476 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1477 | if (!bio) | |
1478 | return ERR_PTR(-ENOMEM); | |
1479 | ||
1480 | offset = offset_in_page(kaddr); | |
1481 | for (i = 0; i < nr_pages; i++) { | |
1482 | unsigned int bytes = PAGE_SIZE - offset; | |
1483 | ||
1484 | if (len <= 0) | |
1485 | break; | |
1486 | ||
1487 | if (bytes > len) | |
1488 | bytes = len; | |
1489 | ||
1490 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, | |
1491 | offset) < bytes) | |
1492 | break; | |
1493 | ||
1494 | data += bytes; | |
1495 | len -= bytes; | |
1496 | offset = 0; | |
1497 | } | |
1498 | ||
1499 | bio->bi_end_io = bio_map_kern_endio; | |
1500 | return bio; | |
1501 | } | |
1502 | ||
1503 | /** | |
1504 | * bio_map_kern - map kernel address into bio | |
1505 | * @q: the struct request_queue for the bio | |
1506 | * @data: pointer to buffer to map | |
1507 | * @len: length in bytes | |
1508 | * @gfp_mask: allocation flags for bio allocation | |
1509 | * | |
1510 | * Map the kernel address into a bio suitable for io to a block | |
1511 | * device. Returns an error pointer in case of error. | |
1512 | */ | |
1513 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, | |
1514 | gfp_t gfp_mask) | |
1515 | { | |
1516 | struct bio *bio; | |
1517 | ||
1518 | bio = __bio_map_kern(q, data, len, gfp_mask); | |
1519 | if (IS_ERR(bio)) | |
1520 | return bio; | |
1521 | ||
1522 | if (bio->bi_iter.bi_size == len) | |
1523 | return bio; | |
1524 | ||
1525 | /* | |
1526 | * Don't support partial mappings. | |
1527 | */ | |
1528 | bio_put(bio); | |
1529 | return ERR_PTR(-EINVAL); | |
1530 | } | |
1531 | EXPORT_SYMBOL(bio_map_kern); | |
1532 | ||
1533 | static void bio_copy_kern_endio(struct bio *bio, int err) | |
1534 | { | |
1535 | struct bio_vec *bvec; | |
1536 | const int read = bio_data_dir(bio) == READ; | |
1537 | struct bio_map_data *bmd = bio->bi_private; | |
1538 | int i; | |
1539 | char *p = bmd->sgvecs[0].iov_base; | |
1540 | ||
1541 | bio_for_each_segment_all(bvec, bio, i) { | |
1542 | char *addr = page_address(bvec->bv_page); | |
1543 | ||
1544 | if (read) | |
1545 | memcpy(p, addr, bvec->bv_len); | |
1546 | ||
1547 | __free_page(bvec->bv_page); | |
1548 | p += bvec->bv_len; | |
1549 | } | |
1550 | ||
1551 | kfree(bmd); | |
1552 | bio_put(bio); | |
1553 | } | |
1554 | ||
1555 | /** | |
1556 | * bio_copy_kern - copy kernel address into bio | |
1557 | * @q: the struct request_queue for the bio | |
1558 | * @data: pointer to buffer to copy | |
1559 | * @len: length in bytes | |
1560 | * @gfp_mask: allocation flags for bio and page allocation | |
1561 | * @reading: data direction is READ | |
1562 | * | |
1563 | * copy the kernel address into a bio suitable for io to a block | |
1564 | * device. Returns an error pointer in case of error. | |
1565 | */ | |
1566 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
1567 | gfp_t gfp_mask, int reading) | |
1568 | { | |
1569 | struct bio *bio; | |
1570 | struct bio_vec *bvec; | |
1571 | int i; | |
1572 | ||
1573 | bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask); | |
1574 | if (IS_ERR(bio)) | |
1575 | return bio; | |
1576 | ||
1577 | if (!reading) { | |
1578 | void *p = data; | |
1579 | ||
1580 | bio_for_each_segment_all(bvec, bio, i) { | |
1581 | char *addr = page_address(bvec->bv_page); | |
1582 | ||
1583 | memcpy(addr, p, bvec->bv_len); | |
1584 | p += bvec->bv_len; | |
1585 | } | |
1586 | } | |
1587 | ||
1588 | bio->bi_end_io = bio_copy_kern_endio; | |
1589 | ||
1590 | return bio; | |
1591 | } | |
1592 | EXPORT_SYMBOL(bio_copy_kern); | |
1593 | ||
1594 | /* | |
1595 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1596 | * for performing direct-IO in BIOs. | |
1597 | * | |
1598 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1599 | * because the required locks are not interrupt-safe. So what we can do is to | |
1600 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1601 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1602 | * in process context. | |
1603 | * | |
1604 | * We special-case compound pages here: normally this means reads into hugetlb | |
1605 | * pages. The logic in here doesn't really work right for compound pages | |
1606 | * because the VM does not uniformly chase down the head page in all cases. | |
1607 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1608 | * handle them at all. So we skip compound pages here at an early stage. | |
1609 | * | |
1610 | * Note that this code is very hard to test under normal circumstances because | |
1611 | * direct-io pins the pages with get_user_pages(). This makes | |
1612 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
1613 | * But other code (eg, flusher threads) could clean the pages if they are mapped | |
1614 | * pagecache. | |
1615 | * | |
1616 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1617 | * deferred bio dirtying paths. | |
1618 | */ | |
1619 | ||
1620 | /* | |
1621 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1622 | */ | |
1623 | void bio_set_pages_dirty(struct bio *bio) | |
1624 | { | |
1625 | struct bio_vec *bvec; | |
1626 | int i; | |
1627 | ||
1628 | bio_for_each_segment_all(bvec, bio, i) { | |
1629 | struct page *page = bvec->bv_page; | |
1630 | ||
1631 | if (page && !PageCompound(page)) | |
1632 | set_page_dirty_lock(page); | |
1633 | } | |
1634 | } | |
1635 | ||
1636 | static void bio_release_pages(struct bio *bio) | |
1637 | { | |
1638 | struct bio_vec *bvec; | |
1639 | int i; | |
1640 | ||
1641 | bio_for_each_segment_all(bvec, bio, i) { | |
1642 | struct page *page = bvec->bv_page; | |
1643 | ||
1644 | if (page) | |
1645 | put_page(page); | |
1646 | } | |
1647 | } | |
1648 | ||
1649 | /* | |
1650 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1651 | * If they are, then fine. If, however, some pages are clean then they must | |
1652 | * have been written out during the direct-IO read. So we take another ref on | |
1653 | * the BIO and the offending pages and re-dirty the pages in process context. | |
1654 | * | |
1655 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
1656 | * here on. It will run one page_cache_release() against each page and will | |
1657 | * run one bio_put() against the BIO. | |
1658 | */ | |
1659 | ||
1660 | static void bio_dirty_fn(struct work_struct *work); | |
1661 | ||
1662 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); | |
1663 | static DEFINE_SPINLOCK(bio_dirty_lock); | |
1664 | static struct bio *bio_dirty_list; | |
1665 | ||
1666 | /* | |
1667 | * This runs in process context | |
1668 | */ | |
1669 | static void bio_dirty_fn(struct work_struct *work) | |
1670 | { | |
1671 | unsigned long flags; | |
1672 | struct bio *bio; | |
1673 | ||
1674 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1675 | bio = bio_dirty_list; | |
1676 | bio_dirty_list = NULL; | |
1677 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1678 | ||
1679 | while (bio) { | |
1680 | struct bio *next = bio->bi_private; | |
1681 | ||
1682 | bio_set_pages_dirty(bio); | |
1683 | bio_release_pages(bio); | |
1684 | bio_put(bio); | |
1685 | bio = next; | |
1686 | } | |
1687 | } | |
1688 | ||
1689 | void bio_check_pages_dirty(struct bio *bio) | |
1690 | { | |
1691 | struct bio_vec *bvec; | |
1692 | int nr_clean_pages = 0; | |
1693 | int i; | |
1694 | ||
1695 | bio_for_each_segment_all(bvec, bio, i) { | |
1696 | struct page *page = bvec->bv_page; | |
1697 | ||
1698 | if (PageDirty(page) || PageCompound(page)) { | |
1699 | page_cache_release(page); | |
1700 | bvec->bv_page = NULL; | |
1701 | } else { | |
1702 | nr_clean_pages++; | |
1703 | } | |
1704 | } | |
1705 | ||
1706 | if (nr_clean_pages) { | |
1707 | unsigned long flags; | |
1708 | ||
1709 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1710 | bio->bi_private = bio_dirty_list; | |
1711 | bio_dirty_list = bio; | |
1712 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1713 | schedule_work(&bio_dirty_work); | |
1714 | } else { | |
1715 | bio_put(bio); | |
1716 | } | |
1717 | } | |
1718 | ||
1719 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE | |
1720 | void bio_flush_dcache_pages(struct bio *bi) | |
1721 | { | |
1722 | struct bio_vec bvec; | |
1723 | struct bvec_iter iter; | |
1724 | ||
1725 | bio_for_each_segment(bvec, bi, iter) | |
1726 | flush_dcache_page(bvec.bv_page); | |
1727 | } | |
1728 | EXPORT_SYMBOL(bio_flush_dcache_pages); | |
1729 | #endif | |
1730 | ||
1731 | /** | |
1732 | * bio_endio - end I/O on a bio | |
1733 | * @bio: bio | |
1734 | * @error: error, if any | |
1735 | * | |
1736 | * Description: | |
1737 | * bio_endio() will end I/O on the whole bio. bio_endio() is the | |
1738 | * preferred way to end I/O on a bio, it takes care of clearing | |
1739 | * BIO_UPTODATE on error. @error is 0 on success, and and one of the | |
1740 | * established -Exxxx (-EIO, for instance) error values in case | |
1741 | * something went wrong. No one should call bi_end_io() directly on a | |
1742 | * bio unless they own it and thus know that it has an end_io | |
1743 | * function. | |
1744 | **/ | |
1745 | void bio_endio(struct bio *bio, int error) | |
1746 | { | |
1747 | while (bio) { | |
1748 | BUG_ON(atomic_read(&bio->bi_remaining) <= 0); | |
1749 | ||
1750 | if (error) | |
1751 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
1752 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
1753 | error = -EIO; | |
1754 | ||
1755 | if (!atomic_dec_and_test(&bio->bi_remaining)) | |
1756 | return; | |
1757 | ||
1758 | /* | |
1759 | * Need to have a real endio function for chained bios, | |
1760 | * otherwise various corner cases will break (like stacking | |
1761 | * block devices that save/restore bi_end_io) - however, we want | |
1762 | * to avoid unbounded recursion and blowing the stack. Tail call | |
1763 | * optimization would handle this, but compiling with frame | |
1764 | * pointers also disables gcc's sibling call optimization. | |
1765 | */ | |
1766 | if (bio->bi_end_io == bio_chain_endio) { | |
1767 | struct bio *parent = bio->bi_private; | |
1768 | bio_put(bio); | |
1769 | bio = parent; | |
1770 | } else { | |
1771 | if (bio->bi_end_io) | |
1772 | bio->bi_end_io(bio, error); | |
1773 | bio = NULL; | |
1774 | } | |
1775 | } | |
1776 | } | |
1777 | EXPORT_SYMBOL(bio_endio); | |
1778 | ||
1779 | /** | |
1780 | * bio_endio_nodec - end I/O on a bio, without decrementing bi_remaining | |
1781 | * @bio: bio | |
1782 | * @error: error, if any | |
1783 | * | |
1784 | * For code that has saved and restored bi_end_io; thing hard before using this | |
1785 | * function, probably you should've cloned the entire bio. | |
1786 | **/ | |
1787 | void bio_endio_nodec(struct bio *bio, int error) | |
1788 | { | |
1789 | atomic_inc(&bio->bi_remaining); | |
1790 | bio_endio(bio, error); | |
1791 | } | |
1792 | EXPORT_SYMBOL(bio_endio_nodec); | |
1793 | ||
1794 | /** | |
1795 | * bio_split - split a bio | |
1796 | * @bio: bio to split | |
1797 | * @sectors: number of sectors to split from the front of @bio | |
1798 | * @gfp: gfp mask | |
1799 | * @bs: bio set to allocate from | |
1800 | * | |
1801 | * Allocates and returns a new bio which represents @sectors from the start of | |
1802 | * @bio, and updates @bio to represent the remaining sectors. | |
1803 | * | |
1804 | * The newly allocated bio will point to @bio's bi_io_vec; it is the caller's | |
1805 | * responsibility to ensure that @bio is not freed before the split. | |
1806 | */ | |
1807 | struct bio *bio_split(struct bio *bio, int sectors, | |
1808 | gfp_t gfp, struct bio_set *bs) | |
1809 | { | |
1810 | struct bio *split = NULL; | |
1811 | ||
1812 | BUG_ON(sectors <= 0); | |
1813 | BUG_ON(sectors >= bio_sectors(bio)); | |
1814 | ||
1815 | split = bio_clone_fast(bio, gfp, bs); | |
1816 | if (!split) | |
1817 | return NULL; | |
1818 | ||
1819 | split->bi_iter.bi_size = sectors << 9; | |
1820 | ||
1821 | if (bio_integrity(split)) | |
1822 | bio_integrity_trim(split, 0, sectors); | |
1823 | ||
1824 | bio_advance(bio, split->bi_iter.bi_size); | |
1825 | ||
1826 | return split; | |
1827 | } | |
1828 | EXPORT_SYMBOL(bio_split); | |
1829 | ||
1830 | /** | |
1831 | * bio_trim - trim a bio | |
1832 | * @bio: bio to trim | |
1833 | * @offset: number of sectors to trim from the front of @bio | |
1834 | * @size: size we want to trim @bio to, in sectors | |
1835 | */ | |
1836 | void bio_trim(struct bio *bio, int offset, int size) | |
1837 | { | |
1838 | /* 'bio' is a cloned bio which we need to trim to match | |
1839 | * the given offset and size. | |
1840 | */ | |
1841 | ||
1842 | size <<= 9; | |
1843 | if (offset == 0 && size == bio->bi_iter.bi_size) | |
1844 | return; | |
1845 | ||
1846 | clear_bit(BIO_SEG_VALID, &bio->bi_flags); | |
1847 | ||
1848 | bio_advance(bio, offset << 9); | |
1849 | ||
1850 | bio->bi_iter.bi_size = size; | |
1851 | } | |
1852 | EXPORT_SYMBOL_GPL(bio_trim); | |
1853 | ||
1854 | /* | |
1855 | * create memory pools for biovec's in a bio_set. | |
1856 | * use the global biovec slabs created for general use. | |
1857 | */ | |
1858 | mempool_t *biovec_create_pool(struct bio_set *bs, int pool_entries) | |
1859 | { | |
1860 | struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX; | |
1861 | ||
1862 | return mempool_create_slab_pool(pool_entries, bp->slab); | |
1863 | } | |
1864 | ||
1865 | void bioset_free(struct bio_set *bs) | |
1866 | { | |
1867 | if (bs->rescue_workqueue) | |
1868 | destroy_workqueue(bs->rescue_workqueue); | |
1869 | ||
1870 | if (bs->bio_pool) | |
1871 | mempool_destroy(bs->bio_pool); | |
1872 | ||
1873 | if (bs->bvec_pool) | |
1874 | mempool_destroy(bs->bvec_pool); | |
1875 | ||
1876 | bioset_integrity_free(bs); | |
1877 | bio_put_slab(bs); | |
1878 | ||
1879 | kfree(bs); | |
1880 | } | |
1881 | EXPORT_SYMBOL(bioset_free); | |
1882 | ||
1883 | /** | |
1884 | * bioset_create - Create a bio_set | |
1885 | * @pool_size: Number of bio and bio_vecs to cache in the mempool | |
1886 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1887 | * | |
1888 | * Description: | |
1889 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller | |
1890 | * to ask for a number of bytes to be allocated in front of the bio. | |
1891 | * Front pad allocation is useful for embedding the bio inside | |
1892 | * another structure, to avoid allocating extra data to go with the bio. | |
1893 | * Note that the bio must be embedded at the END of that structure always, | |
1894 | * or things will break badly. | |
1895 | */ | |
1896 | struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad) | |
1897 | { | |
1898 | unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); | |
1899 | struct bio_set *bs; | |
1900 | ||
1901 | bs = kzalloc(sizeof(*bs), GFP_KERNEL); | |
1902 | if (!bs) | |
1903 | return NULL; | |
1904 | ||
1905 | bs->front_pad = front_pad; | |
1906 | ||
1907 | spin_lock_init(&bs->rescue_lock); | |
1908 | bio_list_init(&bs->rescue_list); | |
1909 | INIT_WORK(&bs->rescue_work, bio_alloc_rescue); | |
1910 | ||
1911 | bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); | |
1912 | if (!bs->bio_slab) { | |
1913 | kfree(bs); | |
1914 | return NULL; | |
1915 | } | |
1916 | ||
1917 | bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab); | |
1918 | if (!bs->bio_pool) | |
1919 | goto bad; | |
1920 | ||
1921 | bs->bvec_pool = biovec_create_pool(bs, pool_size); | |
1922 | if (!bs->bvec_pool) | |
1923 | goto bad; | |
1924 | ||
1925 | bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); | |
1926 | if (!bs->rescue_workqueue) | |
1927 | goto bad; | |
1928 | ||
1929 | return bs; | |
1930 | bad: | |
1931 | bioset_free(bs); | |
1932 | return NULL; | |
1933 | } | |
1934 | EXPORT_SYMBOL(bioset_create); | |
1935 | ||
1936 | #ifdef CONFIG_BLK_CGROUP | |
1937 | /** | |
1938 | * bio_associate_current - associate a bio with %current | |
1939 | * @bio: target bio | |
1940 | * | |
1941 | * Associate @bio with %current if it hasn't been associated yet. Block | |
1942 | * layer will treat @bio as if it were issued by %current no matter which | |
1943 | * task actually issues it. | |
1944 | * | |
1945 | * This function takes an extra reference of @task's io_context and blkcg | |
1946 | * which will be put when @bio is released. The caller must own @bio, | |
1947 | * ensure %current->io_context exists, and is responsible for synchronizing | |
1948 | * calls to this function. | |
1949 | */ | |
1950 | int bio_associate_current(struct bio *bio) | |
1951 | { | |
1952 | struct io_context *ioc; | |
1953 | struct cgroup_subsys_state *css; | |
1954 | ||
1955 | if (bio->bi_ioc) | |
1956 | return -EBUSY; | |
1957 | ||
1958 | ioc = current->io_context; | |
1959 | if (!ioc) | |
1960 | return -ENOENT; | |
1961 | ||
1962 | /* acquire active ref on @ioc and associate */ | |
1963 | get_io_context_active(ioc); | |
1964 | bio->bi_ioc = ioc; | |
1965 | ||
1966 | /* associate blkcg if exists */ | |
1967 | rcu_read_lock(); | |
1968 | css = task_css(current, blkio_subsys_id); | |
1969 | if (css && css_tryget(css)) | |
1970 | bio->bi_css = css; | |
1971 | rcu_read_unlock(); | |
1972 | ||
1973 | return 0; | |
1974 | } | |
1975 | ||
1976 | /** | |
1977 | * bio_disassociate_task - undo bio_associate_current() | |
1978 | * @bio: target bio | |
1979 | */ | |
1980 | void bio_disassociate_task(struct bio *bio) | |
1981 | { | |
1982 | if (bio->bi_ioc) { | |
1983 | put_io_context(bio->bi_ioc); | |
1984 | bio->bi_ioc = NULL; | |
1985 | } | |
1986 | if (bio->bi_css) { | |
1987 | css_put(bio->bi_css); | |
1988 | bio->bi_css = NULL; | |
1989 | } | |
1990 | } | |
1991 | ||
1992 | #endif /* CONFIG_BLK_CGROUP */ | |
1993 | ||
1994 | static void __init biovec_init_slabs(void) | |
1995 | { | |
1996 | int i; | |
1997 | ||
1998 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1999 | int size; | |
2000 | struct biovec_slab *bvs = bvec_slabs + i; | |
2001 | ||
2002 | if (bvs->nr_vecs <= BIO_INLINE_VECS) { | |
2003 | bvs->slab = NULL; | |
2004 | continue; | |
2005 | } | |
2006 | ||
2007 | size = bvs->nr_vecs * sizeof(struct bio_vec); | |
2008 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
2009 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); | |
2010 | } | |
2011 | } | |
2012 | ||
2013 | static int __init init_bio(void) | |
2014 | { | |
2015 | bio_slab_max = 2; | |
2016 | bio_slab_nr = 0; | |
2017 | bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL); | |
2018 | if (!bio_slabs) | |
2019 | panic("bio: can't allocate bios\n"); | |
2020 | ||
2021 | bio_integrity_init(); | |
2022 | biovec_init_slabs(); | |
2023 | ||
2024 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 0); | |
2025 | if (!fs_bio_set) | |
2026 | panic("bio: can't allocate bios\n"); | |
2027 | ||
2028 | if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE)) | |
2029 | panic("bio: can't create integrity pool\n"); | |
2030 | ||
2031 | return 0; | |
2032 | } | |
2033 | subsys_initcall(init_bio); |