]>
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 | ||
32 | #include <trace/events/block.h> | |
33 | ||
34 | /* | |
35 | * Test patch to inline a certain number of bi_io_vec's inside the bio | |
36 | * itself, to shrink a bio data allocation from two mempool calls to one | |
37 | */ | |
38 | #define BIO_INLINE_VECS 4 | |
39 | ||
40 | /* | |
41 | * if you change this list, also change bvec_alloc or things will | |
42 | * break badly! cannot be bigger than what you can fit into an | |
43 | * unsigned short | |
44 | */ | |
45 | #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } | |
46 | static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { | |
47 | BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), | |
48 | }; | |
49 | #undef BV | |
50 | ||
51 | /* | |
52 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
53 | * IO code that does not need private memory pools. | |
54 | */ | |
55 | struct bio_set *fs_bio_set; | |
56 | EXPORT_SYMBOL(fs_bio_set); | |
57 | ||
58 | /* | |
59 | * Our slab pool management | |
60 | */ | |
61 | struct bio_slab { | |
62 | struct kmem_cache *slab; | |
63 | unsigned int slab_ref; | |
64 | unsigned int slab_size; | |
65 | char name[8]; | |
66 | }; | |
67 | static DEFINE_MUTEX(bio_slab_lock); | |
68 | static struct bio_slab *bio_slabs; | |
69 | static unsigned int bio_slab_nr, bio_slab_max; | |
70 | ||
71 | static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) | |
72 | { | |
73 | unsigned int sz = sizeof(struct bio) + extra_size; | |
74 | struct kmem_cache *slab = NULL; | |
75 | struct bio_slab *bslab, *new_bio_slabs; | |
76 | unsigned int new_bio_slab_max; | |
77 | unsigned int i, entry = -1; | |
78 | ||
79 | mutex_lock(&bio_slab_lock); | |
80 | ||
81 | i = 0; | |
82 | while (i < bio_slab_nr) { | |
83 | bslab = &bio_slabs[i]; | |
84 | ||
85 | if (!bslab->slab && entry == -1) | |
86 | entry = i; | |
87 | else if (bslab->slab_size == sz) { | |
88 | slab = bslab->slab; | |
89 | bslab->slab_ref++; | |
90 | break; | |
91 | } | |
92 | i++; | |
93 | } | |
94 | ||
95 | if (slab) | |
96 | goto out_unlock; | |
97 | ||
98 | if (bio_slab_nr == bio_slab_max && entry == -1) { | |
99 | new_bio_slab_max = bio_slab_max << 1; | |
100 | new_bio_slabs = krealloc(bio_slabs, | |
101 | new_bio_slab_max * sizeof(struct bio_slab), | |
102 | GFP_KERNEL); | |
103 | if (!new_bio_slabs) | |
104 | goto out_unlock; | |
105 | bio_slab_max = new_bio_slab_max; | |
106 | bio_slabs = new_bio_slabs; | |
107 | } | |
108 | if (entry == -1) | |
109 | entry = bio_slab_nr++; | |
110 | ||
111 | bslab = &bio_slabs[entry]; | |
112 | ||
113 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); | |
114 | slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, | |
115 | SLAB_HWCACHE_ALIGN, NULL); | |
116 | if (!slab) | |
117 | goto out_unlock; | |
118 | ||
119 | bslab->slab = slab; | |
120 | bslab->slab_ref = 1; | |
121 | bslab->slab_size = sz; | |
122 | out_unlock: | |
123 | mutex_unlock(&bio_slab_lock); | |
124 | return slab; | |
125 | } | |
126 | ||
127 | static void bio_put_slab(struct bio_set *bs) | |
128 | { | |
129 | struct bio_slab *bslab = NULL; | |
130 | unsigned int i; | |
131 | ||
132 | mutex_lock(&bio_slab_lock); | |
133 | ||
134 | for (i = 0; i < bio_slab_nr; i++) { | |
135 | if (bs->bio_slab == bio_slabs[i].slab) { | |
136 | bslab = &bio_slabs[i]; | |
137 | break; | |
138 | } | |
139 | } | |
140 | ||
141 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) | |
142 | goto out; | |
143 | ||
144 | WARN_ON(!bslab->slab_ref); | |
145 | ||
146 | if (--bslab->slab_ref) | |
147 | goto out; | |
148 | ||
149 | kmem_cache_destroy(bslab->slab); | |
150 | bslab->slab = NULL; | |
151 | ||
152 | out: | |
153 | mutex_unlock(&bio_slab_lock); | |
154 | } | |
155 | ||
156 | unsigned int bvec_nr_vecs(unsigned short idx) | |
157 | { | |
158 | return bvec_slabs[idx].nr_vecs; | |
159 | } | |
160 | ||
161 | void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) | |
162 | { | |
163 | if (!idx) | |
164 | return; | |
165 | idx--; | |
166 | ||
167 | BIO_BUG_ON(idx >= BVEC_POOL_NR); | |
168 | ||
169 | if (idx == BVEC_POOL_MAX) { | |
170 | mempool_free(bv, pool); | |
171 | } else { | |
172 | struct biovec_slab *bvs = bvec_slabs + idx; | |
173 | ||
174 | kmem_cache_free(bvs->slab, bv); | |
175 | } | |
176 | } | |
177 | ||
178 | struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, | |
179 | mempool_t *pool) | |
180 | { | |
181 | struct bio_vec *bvl; | |
182 | ||
183 | /* | |
184 | * see comment near bvec_array define! | |
185 | */ | |
186 | switch (nr) { | |
187 | case 1: | |
188 | *idx = 0; | |
189 | break; | |
190 | case 2 ... 4: | |
191 | *idx = 1; | |
192 | break; | |
193 | case 5 ... 16: | |
194 | *idx = 2; | |
195 | break; | |
196 | case 17 ... 64: | |
197 | *idx = 3; | |
198 | break; | |
199 | case 65 ... 128: | |
200 | *idx = 4; | |
201 | break; | |
202 | case 129 ... BIO_MAX_PAGES: | |
203 | *idx = 5; | |
204 | break; | |
205 | default: | |
206 | return NULL; | |
207 | } | |
208 | ||
209 | /* | |
210 | * idx now points to the pool we want to allocate from. only the | |
211 | * 1-vec entry pool is mempool backed. | |
212 | */ | |
213 | if (*idx == BVEC_POOL_MAX) { | |
214 | fallback: | |
215 | bvl = mempool_alloc(pool, gfp_mask); | |
216 | } else { | |
217 | struct biovec_slab *bvs = bvec_slabs + *idx; | |
218 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); | |
219 | ||
220 | /* | |
221 | * Make this allocation restricted and don't dump info on | |
222 | * allocation failures, since we'll fallback to the mempool | |
223 | * in case of failure. | |
224 | */ | |
225 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; | |
226 | ||
227 | /* | |
228 | * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM | |
229 | * is set, retry with the 1-entry mempool | |
230 | */ | |
231 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); | |
232 | if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { | |
233 | *idx = BVEC_POOL_MAX; | |
234 | goto fallback; | |
235 | } | |
236 | } | |
237 | ||
238 | (*idx)++; | |
239 | return bvl; | |
240 | } | |
241 | ||
242 | static void __bio_free(struct bio *bio) | |
243 | { | |
244 | bio_disassociate_task(bio); | |
245 | ||
246 | if (bio_integrity(bio)) | |
247 | bio_integrity_free(bio); | |
248 | } | |
249 | ||
250 | static void bio_free(struct bio *bio) | |
251 | { | |
252 | struct bio_set *bs = bio->bi_pool; | |
253 | void *p; | |
254 | ||
255 | __bio_free(bio); | |
256 | ||
257 | if (bs) { | |
258 | bvec_free(bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); | |
259 | ||
260 | /* | |
261 | * If we have front padding, adjust the bio pointer before freeing | |
262 | */ | |
263 | p = bio; | |
264 | p -= bs->front_pad; | |
265 | ||
266 | mempool_free(p, bs->bio_pool); | |
267 | } else { | |
268 | /* Bio was allocated by bio_kmalloc() */ | |
269 | kfree(bio); | |
270 | } | |
271 | } | |
272 | ||
273 | void bio_init(struct bio *bio, struct bio_vec *table, | |
274 | unsigned short max_vecs) | |
275 | { | |
276 | memset(bio, 0, sizeof(*bio)); | |
277 | atomic_set(&bio->__bi_remaining, 1); | |
278 | atomic_set(&bio->__bi_cnt, 1); | |
279 | ||
280 | bio->bi_io_vec = table; | |
281 | bio->bi_max_vecs = max_vecs; | |
282 | } | |
283 | EXPORT_SYMBOL(bio_init); | |
284 | ||
285 | /** | |
286 | * bio_reset - reinitialize a bio | |
287 | * @bio: bio to reset | |
288 | * | |
289 | * Description: | |
290 | * After calling bio_reset(), @bio will be in the same state as a freshly | |
291 | * allocated bio returned bio bio_alloc_bioset() - the only fields that are | |
292 | * preserved are the ones that are initialized by bio_alloc_bioset(). See | |
293 | * comment in struct bio. | |
294 | */ | |
295 | void bio_reset(struct bio *bio) | |
296 | { | |
297 | unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); | |
298 | ||
299 | __bio_free(bio); | |
300 | ||
301 | memset(bio, 0, BIO_RESET_BYTES); | |
302 | bio->bi_flags = flags; | |
303 | atomic_set(&bio->__bi_remaining, 1); | |
304 | } | |
305 | EXPORT_SYMBOL(bio_reset); | |
306 | ||
307 | static struct bio *__bio_chain_endio(struct bio *bio) | |
308 | { | |
309 | struct bio *parent = bio->bi_private; | |
310 | ||
311 | if (!parent->bi_error) | |
312 | parent->bi_error = bio->bi_error; | |
313 | bio_put(bio); | |
314 | return parent; | |
315 | } | |
316 | ||
317 | static void bio_chain_endio(struct bio *bio) | |
318 | { | |
319 | bio_endio(__bio_chain_endio(bio)); | |
320 | } | |
321 | ||
322 | /** | |
323 | * bio_chain - chain bio completions | |
324 | * @bio: the target bio | |
325 | * @parent: the @bio's parent bio | |
326 | * | |
327 | * The caller won't have a bi_end_io called when @bio completes - instead, | |
328 | * @parent's bi_end_io won't be called until both @parent and @bio have | |
329 | * completed; the chained bio will also be freed when it completes. | |
330 | * | |
331 | * The caller must not set bi_private or bi_end_io in @bio. | |
332 | */ | |
333 | void bio_chain(struct bio *bio, struct bio *parent) | |
334 | { | |
335 | BUG_ON(bio->bi_private || bio->bi_end_io); | |
336 | ||
337 | bio->bi_private = parent; | |
338 | bio->bi_end_io = bio_chain_endio; | |
339 | bio_inc_remaining(parent); | |
340 | } | |
341 | EXPORT_SYMBOL(bio_chain); | |
342 | ||
343 | static void bio_alloc_rescue(struct work_struct *work) | |
344 | { | |
345 | struct bio_set *bs = container_of(work, struct bio_set, rescue_work); | |
346 | struct bio *bio; | |
347 | ||
348 | while (1) { | |
349 | spin_lock(&bs->rescue_lock); | |
350 | bio = bio_list_pop(&bs->rescue_list); | |
351 | spin_unlock(&bs->rescue_lock); | |
352 | ||
353 | if (!bio) | |
354 | break; | |
355 | ||
356 | generic_make_request(bio); | |
357 | } | |
358 | } | |
359 | ||
360 | static void punt_bios_to_rescuer(struct bio_set *bs) | |
361 | { | |
362 | struct bio_list punt, nopunt; | |
363 | struct bio *bio; | |
364 | ||
365 | /* | |
366 | * In order to guarantee forward progress we must punt only bios that | |
367 | * were allocated from this bio_set; otherwise, if there was a bio on | |
368 | * there for a stacking driver higher up in the stack, processing it | |
369 | * could require allocating bios from this bio_set, and doing that from | |
370 | * our own rescuer would be bad. | |
371 | * | |
372 | * Since bio lists are singly linked, pop them all instead of trying to | |
373 | * remove from the middle of the list: | |
374 | */ | |
375 | ||
376 | bio_list_init(&punt); | |
377 | bio_list_init(&nopunt); | |
378 | ||
379 | while ((bio = bio_list_pop(¤t->bio_list[0]))) | |
380 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); | |
381 | current->bio_list[0] = nopunt; | |
382 | ||
383 | bio_list_init(&nopunt); | |
384 | while ((bio = bio_list_pop(¤t->bio_list[1]))) | |
385 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); | |
386 | current->bio_list[1] = nopunt; | |
387 | ||
388 | spin_lock(&bs->rescue_lock); | |
389 | bio_list_merge(&bs->rescue_list, &punt); | |
390 | spin_unlock(&bs->rescue_lock); | |
391 | ||
392 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
393 | } | |
394 | ||
395 | /** | |
396 | * bio_alloc_bioset - allocate a bio for I/O | |
397 | * @gfp_mask: the GFP_ mask given to the slab allocator | |
398 | * @nr_iovecs: number of iovecs to pre-allocate | |
399 | * @bs: the bio_set to allocate from. | |
400 | * | |
401 | * Description: | |
402 | * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is | |
403 | * backed by the @bs's mempool. | |
404 | * | |
405 | * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will | |
406 | * always be able to allocate a bio. This is due to the mempool guarantees. | |
407 | * To make this work, callers must never allocate more than 1 bio at a time | |
408 | * from this pool. Callers that need to allocate more than 1 bio must always | |
409 | * submit the previously allocated bio for IO before attempting to allocate | |
410 | * a new one. Failure to do so can cause deadlocks under memory pressure. | |
411 | * | |
412 | * Note that when running under generic_make_request() (i.e. any block | |
413 | * driver), bios are not submitted until after you return - see the code in | |
414 | * generic_make_request() that converts recursion into iteration, to prevent | |
415 | * stack overflows. | |
416 | * | |
417 | * This would normally mean allocating multiple bios under | |
418 | * generic_make_request() would be susceptible to deadlocks, but we have | |
419 | * deadlock avoidance code that resubmits any blocked bios from a rescuer | |
420 | * thread. | |
421 | * | |
422 | * However, we do not guarantee forward progress for allocations from other | |
423 | * mempools. Doing multiple allocations from the same mempool under | |
424 | * generic_make_request() should be avoided - instead, use bio_set's front_pad | |
425 | * for per bio allocations. | |
426 | * | |
427 | * RETURNS: | |
428 | * Pointer to new bio on success, NULL on failure. | |
429 | */ | |
430 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) | |
431 | { | |
432 | gfp_t saved_gfp = gfp_mask; | |
433 | unsigned front_pad; | |
434 | unsigned inline_vecs; | |
435 | struct bio_vec *bvl = NULL; | |
436 | struct bio *bio; | |
437 | void *p; | |
438 | ||
439 | if (!bs) { | |
440 | if (nr_iovecs > UIO_MAXIOV) | |
441 | return NULL; | |
442 | ||
443 | p = kmalloc(sizeof(struct bio) + | |
444 | nr_iovecs * sizeof(struct bio_vec), | |
445 | gfp_mask); | |
446 | front_pad = 0; | |
447 | inline_vecs = nr_iovecs; | |
448 | } else { | |
449 | /* should not use nobvec bioset for nr_iovecs > 0 */ | |
450 | if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0)) | |
451 | return NULL; | |
452 | /* | |
453 | * generic_make_request() converts recursion to iteration; this | |
454 | * means if we're running beneath it, any bios we allocate and | |
455 | * submit will not be submitted (and thus freed) until after we | |
456 | * return. | |
457 | * | |
458 | * This exposes us to a potential deadlock if we allocate | |
459 | * multiple bios from the same bio_set() while running | |
460 | * underneath generic_make_request(). If we were to allocate | |
461 | * multiple bios (say a stacking block driver that was splitting | |
462 | * bios), we would deadlock if we exhausted the mempool's | |
463 | * reserve. | |
464 | * | |
465 | * We solve this, and guarantee forward progress, with a rescuer | |
466 | * workqueue per bio_set. If we go to allocate and there are | |
467 | * bios on current->bio_list, we first try the allocation | |
468 | * without __GFP_DIRECT_RECLAIM; if that fails, we punt those | |
469 | * bios we would be blocking to the rescuer workqueue before | |
470 | * we retry with the original gfp_flags. | |
471 | */ | |
472 | ||
473 | if (current->bio_list && | |
474 | (!bio_list_empty(¤t->bio_list[0]) || | |
475 | !bio_list_empty(¤t->bio_list[1]))) | |
476 | gfp_mask &= ~__GFP_DIRECT_RECLAIM; | |
477 | ||
478 | p = mempool_alloc(bs->bio_pool, gfp_mask); | |
479 | if (!p && gfp_mask != saved_gfp) { | |
480 | punt_bios_to_rescuer(bs); | |
481 | gfp_mask = saved_gfp; | |
482 | p = mempool_alloc(bs->bio_pool, gfp_mask); | |
483 | } | |
484 | ||
485 | front_pad = bs->front_pad; | |
486 | inline_vecs = BIO_INLINE_VECS; | |
487 | } | |
488 | ||
489 | if (unlikely(!p)) | |
490 | return NULL; | |
491 | ||
492 | bio = p + front_pad; | |
493 | bio_init(bio, NULL, 0); | |
494 | ||
495 | if (nr_iovecs > inline_vecs) { | |
496 | unsigned long idx = 0; | |
497 | ||
498 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); | |
499 | if (!bvl && gfp_mask != saved_gfp) { | |
500 | punt_bios_to_rescuer(bs); | |
501 | gfp_mask = saved_gfp; | |
502 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); | |
503 | } | |
504 | ||
505 | if (unlikely(!bvl)) | |
506 | goto err_free; | |
507 | ||
508 | bio->bi_flags |= idx << BVEC_POOL_OFFSET; | |
509 | } else if (nr_iovecs) { | |
510 | bvl = bio->bi_inline_vecs; | |
511 | } | |
512 | ||
513 | bio->bi_pool = bs; | |
514 | bio->bi_max_vecs = nr_iovecs; | |
515 | bio->bi_io_vec = bvl; | |
516 | return bio; | |
517 | ||
518 | err_free: | |
519 | mempool_free(p, bs->bio_pool); | |
520 | return NULL; | |
521 | } | |
522 | EXPORT_SYMBOL(bio_alloc_bioset); | |
523 | ||
524 | void zero_fill_bio(struct bio *bio) | |
525 | { | |
526 | unsigned long flags; | |
527 | struct bio_vec bv; | |
528 | struct bvec_iter iter; | |
529 | ||
530 | bio_for_each_segment(bv, bio, iter) { | |
531 | char *data = bvec_kmap_irq(&bv, &flags); | |
532 | memset(data, 0, bv.bv_len); | |
533 | flush_dcache_page(bv.bv_page); | |
534 | bvec_kunmap_irq(data, &flags); | |
535 | } | |
536 | } | |
537 | EXPORT_SYMBOL(zero_fill_bio); | |
538 | ||
539 | /** | |
540 | * bio_put - release a reference to a bio | |
541 | * @bio: bio to release reference to | |
542 | * | |
543 | * Description: | |
544 | * Put a reference to a &struct bio, either one you have gotten with | |
545 | * bio_alloc, bio_get or bio_clone. The last put of a bio will free it. | |
546 | **/ | |
547 | void bio_put(struct bio *bio) | |
548 | { | |
549 | if (!bio_flagged(bio, BIO_REFFED)) | |
550 | bio_free(bio); | |
551 | else { | |
552 | BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); | |
553 | ||
554 | /* | |
555 | * last put frees it | |
556 | */ | |
557 | if (atomic_dec_and_test(&bio->__bi_cnt)) | |
558 | bio_free(bio); | |
559 | } | |
560 | } | |
561 | EXPORT_SYMBOL(bio_put); | |
562 | ||
563 | inline int bio_phys_segments(struct request_queue *q, struct bio *bio) | |
564 | { | |
565 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
566 | blk_recount_segments(q, bio); | |
567 | ||
568 | return bio->bi_phys_segments; | |
569 | } | |
570 | EXPORT_SYMBOL(bio_phys_segments); | |
571 | ||
572 | /** | |
573 | * __bio_clone_fast - clone a bio that shares the original bio's biovec | |
574 | * @bio: destination bio | |
575 | * @bio_src: bio to clone | |
576 | * | |
577 | * Clone a &bio. Caller will own the returned bio, but not | |
578 | * the actual data it points to. Reference count of returned | |
579 | * bio will be one. | |
580 | * | |
581 | * Caller must ensure that @bio_src is not freed before @bio. | |
582 | */ | |
583 | void __bio_clone_fast(struct bio *bio, struct bio *bio_src) | |
584 | { | |
585 | BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); | |
586 | ||
587 | /* | |
588 | * most users will be overriding ->bi_bdev with a new target, | |
589 | * so we don't set nor calculate new physical/hw segment counts here | |
590 | */ | |
591 | bio->bi_bdev = bio_src->bi_bdev; | |
592 | bio_set_flag(bio, BIO_CLONED); | |
593 | bio->bi_opf = bio_src->bi_opf; | |
594 | bio->bi_iter = bio_src->bi_iter; | |
595 | bio->bi_io_vec = bio_src->bi_io_vec; | |
596 | ||
597 | bio_clone_blkcg_association(bio, bio_src); | |
598 | } | |
599 | EXPORT_SYMBOL(__bio_clone_fast); | |
600 | ||
601 | /** | |
602 | * bio_clone_fast - clone a bio that shares the original bio's biovec | |
603 | * @bio: bio to clone | |
604 | * @gfp_mask: allocation priority | |
605 | * @bs: bio_set to allocate from | |
606 | * | |
607 | * Like __bio_clone_fast, only also allocates the returned bio | |
608 | */ | |
609 | struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) | |
610 | { | |
611 | struct bio *b; | |
612 | ||
613 | b = bio_alloc_bioset(gfp_mask, 0, bs); | |
614 | if (!b) | |
615 | return NULL; | |
616 | ||
617 | __bio_clone_fast(b, bio); | |
618 | ||
619 | if (bio_integrity(bio)) { | |
620 | int ret; | |
621 | ||
622 | ret = bio_integrity_clone(b, bio, gfp_mask); | |
623 | ||
624 | if (ret < 0) { | |
625 | bio_put(b); | |
626 | return NULL; | |
627 | } | |
628 | } | |
629 | ||
630 | return b; | |
631 | } | |
632 | EXPORT_SYMBOL(bio_clone_fast); | |
633 | ||
634 | /** | |
635 | * bio_clone_bioset - clone a bio | |
636 | * @bio_src: bio to clone | |
637 | * @gfp_mask: allocation priority | |
638 | * @bs: bio_set to allocate from | |
639 | * | |
640 | * Clone bio. Caller will own the returned bio, but not the actual data it | |
641 | * points to. Reference count of returned bio will be one. | |
642 | */ | |
643 | struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask, | |
644 | struct bio_set *bs) | |
645 | { | |
646 | struct bvec_iter iter; | |
647 | struct bio_vec bv; | |
648 | struct bio *bio; | |
649 | ||
650 | /* | |
651 | * Pre immutable biovecs, __bio_clone() used to just do a memcpy from | |
652 | * bio_src->bi_io_vec to bio->bi_io_vec. | |
653 | * | |
654 | * We can't do that anymore, because: | |
655 | * | |
656 | * - The point of cloning the biovec is to produce a bio with a biovec | |
657 | * the caller can modify: bi_idx and bi_bvec_done should be 0. | |
658 | * | |
659 | * - The original bio could've had more than BIO_MAX_PAGES biovecs; if | |
660 | * we tried to clone the whole thing bio_alloc_bioset() would fail. | |
661 | * But the clone should succeed as long as the number of biovecs we | |
662 | * actually need to allocate is fewer than BIO_MAX_PAGES. | |
663 | * | |
664 | * - Lastly, bi_vcnt should not be looked at or relied upon by code | |
665 | * that does not own the bio - reason being drivers don't use it for | |
666 | * iterating over the biovec anymore, so expecting it to be kept up | |
667 | * to date (i.e. for clones that share the parent biovec) is just | |
668 | * asking for trouble and would force extra work on | |
669 | * __bio_clone_fast() anyways. | |
670 | */ | |
671 | ||
672 | bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs); | |
673 | if (!bio) | |
674 | return NULL; | |
675 | bio->bi_bdev = bio_src->bi_bdev; | |
676 | bio->bi_opf = bio_src->bi_opf; | |
677 | bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector; | |
678 | bio->bi_iter.bi_size = bio_src->bi_iter.bi_size; | |
679 | ||
680 | switch (bio_op(bio)) { | |
681 | case REQ_OP_DISCARD: | |
682 | case REQ_OP_SECURE_ERASE: | |
683 | case REQ_OP_WRITE_ZEROES: | |
684 | break; | |
685 | case REQ_OP_WRITE_SAME: | |
686 | bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0]; | |
687 | break; | |
688 | default: | |
689 | bio_for_each_segment(bv, bio_src, iter) | |
690 | bio->bi_io_vec[bio->bi_vcnt++] = bv; | |
691 | break; | |
692 | } | |
693 | ||
694 | if (bio_integrity(bio_src)) { | |
695 | int ret; | |
696 | ||
697 | ret = bio_integrity_clone(bio, bio_src, gfp_mask); | |
698 | if (ret < 0) { | |
699 | bio_put(bio); | |
700 | return NULL; | |
701 | } | |
702 | } | |
703 | ||
704 | bio_clone_blkcg_association(bio, bio_src); | |
705 | ||
706 | return bio; | |
707 | } | |
708 | EXPORT_SYMBOL(bio_clone_bioset); | |
709 | ||
710 | /** | |
711 | * bio_add_pc_page - attempt to add page to bio | |
712 | * @q: the target queue | |
713 | * @bio: destination bio | |
714 | * @page: page to add | |
715 | * @len: vec entry length | |
716 | * @offset: vec entry offset | |
717 | * | |
718 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
719 | * number of reasons, such as the bio being full or target block device | |
720 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
721 | * so it is always possible to add a single page to an empty bio. | |
722 | * | |
723 | * This should only be used by REQ_PC bios. | |
724 | */ | |
725 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page | |
726 | *page, unsigned int len, unsigned int offset) | |
727 | { | |
728 | int retried_segments = 0; | |
729 | struct bio_vec *bvec; | |
730 | ||
731 | /* | |
732 | * cloned bio must not modify vec list | |
733 | */ | |
734 | if (unlikely(bio_flagged(bio, BIO_CLONED))) | |
735 | return 0; | |
736 | ||
737 | if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) | |
738 | return 0; | |
739 | ||
740 | /* | |
741 | * For filesystems with a blocksize smaller than the pagesize | |
742 | * we will often be called with the same page as last time and | |
743 | * a consecutive offset. Optimize this special case. | |
744 | */ | |
745 | if (bio->bi_vcnt > 0) { | |
746 | struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
747 | ||
748 | if (page == prev->bv_page && | |
749 | offset == prev->bv_offset + prev->bv_len) { | |
750 | prev->bv_len += len; | |
751 | bio->bi_iter.bi_size += len; | |
752 | goto done; | |
753 | } | |
754 | ||
755 | /* | |
756 | * If the queue doesn't support SG gaps and adding this | |
757 | * offset would create a gap, disallow it. | |
758 | */ | |
759 | if (bvec_gap_to_prev(q, prev, offset)) | |
760 | return 0; | |
761 | } | |
762 | ||
763 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
764 | return 0; | |
765 | ||
766 | /* | |
767 | * setup the new entry, we might clear it again later if we | |
768 | * cannot add the page | |
769 | */ | |
770 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; | |
771 | bvec->bv_page = page; | |
772 | bvec->bv_len = len; | |
773 | bvec->bv_offset = offset; | |
774 | bio->bi_vcnt++; | |
775 | bio->bi_phys_segments++; | |
776 | bio->bi_iter.bi_size += len; | |
777 | ||
778 | /* | |
779 | * Perform a recount if the number of segments is greater | |
780 | * than queue_max_segments(q). | |
781 | */ | |
782 | ||
783 | while (bio->bi_phys_segments > queue_max_segments(q)) { | |
784 | ||
785 | if (retried_segments) | |
786 | goto failed; | |
787 | ||
788 | retried_segments = 1; | |
789 | blk_recount_segments(q, bio); | |
790 | } | |
791 | ||
792 | /* If we may be able to merge these biovecs, force a recount */ | |
793 | if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) | |
794 | bio_clear_flag(bio, BIO_SEG_VALID); | |
795 | ||
796 | done: | |
797 | return len; | |
798 | ||
799 | failed: | |
800 | bvec->bv_page = NULL; | |
801 | bvec->bv_len = 0; | |
802 | bvec->bv_offset = 0; | |
803 | bio->bi_vcnt--; | |
804 | bio->bi_iter.bi_size -= len; | |
805 | blk_recount_segments(q, bio); | |
806 | return 0; | |
807 | } | |
808 | EXPORT_SYMBOL(bio_add_pc_page); | |
809 | ||
810 | /** | |
811 | * bio_add_page - attempt to add page to bio | |
812 | * @bio: destination bio | |
813 | * @page: page to add | |
814 | * @len: vec entry length | |
815 | * @offset: vec entry offset | |
816 | * | |
817 | * Attempt to add a page to the bio_vec maplist. This will only fail | |
818 | * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. | |
819 | */ | |
820 | int bio_add_page(struct bio *bio, struct page *page, | |
821 | unsigned int len, unsigned int offset) | |
822 | { | |
823 | struct bio_vec *bv; | |
824 | ||
825 | /* | |
826 | * cloned bio must not modify vec list | |
827 | */ | |
828 | if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) | |
829 | return 0; | |
830 | ||
831 | /* | |
832 | * For filesystems with a blocksize smaller than the pagesize | |
833 | * we will often be called with the same page as last time and | |
834 | * a consecutive offset. Optimize this special case. | |
835 | */ | |
836 | if (bio->bi_vcnt > 0) { | |
837 | bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
838 | ||
839 | if (page == bv->bv_page && | |
840 | offset == bv->bv_offset + bv->bv_len) { | |
841 | bv->bv_len += len; | |
842 | goto done; | |
843 | } | |
844 | } | |
845 | ||
846 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
847 | return 0; | |
848 | ||
849 | bv = &bio->bi_io_vec[bio->bi_vcnt]; | |
850 | bv->bv_page = page; | |
851 | bv->bv_len = len; | |
852 | bv->bv_offset = offset; | |
853 | ||
854 | bio->bi_vcnt++; | |
855 | done: | |
856 | bio->bi_iter.bi_size += len; | |
857 | return len; | |
858 | } | |
859 | EXPORT_SYMBOL(bio_add_page); | |
860 | ||
861 | /** | |
862 | * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio | |
863 | * @bio: bio to add pages to | |
864 | * @iter: iov iterator describing the region to be mapped | |
865 | * | |
866 | * Pins as many pages from *iter and appends them to @bio's bvec array. The | |
867 | * pages will have to be released using put_page() when done. | |
868 | */ | |
869 | int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) | |
870 | { | |
871 | unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; | |
872 | struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; | |
873 | struct page **pages = (struct page **)bv; | |
874 | size_t offset, diff; | |
875 | ssize_t size; | |
876 | ||
877 | size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); | |
878 | if (unlikely(size <= 0)) | |
879 | return size ? size : -EFAULT; | |
880 | nr_pages = (size + offset + PAGE_SIZE - 1) / PAGE_SIZE; | |
881 | ||
882 | /* | |
883 | * Deep magic below: We need to walk the pinned pages backwards | |
884 | * because we are abusing the space allocated for the bio_vecs | |
885 | * for the page array. Because the bio_vecs are larger than the | |
886 | * page pointers by definition this will always work. But it also | |
887 | * means we can't use bio_add_page, so any changes to it's semantics | |
888 | * need to be reflected here as well. | |
889 | */ | |
890 | bio->bi_iter.bi_size += size; | |
891 | bio->bi_vcnt += nr_pages; | |
892 | ||
893 | diff = (nr_pages * PAGE_SIZE - offset) - size; | |
894 | while (nr_pages--) { | |
895 | bv[nr_pages].bv_page = pages[nr_pages]; | |
896 | bv[nr_pages].bv_len = PAGE_SIZE; | |
897 | bv[nr_pages].bv_offset = 0; | |
898 | } | |
899 | ||
900 | bv[0].bv_offset += offset; | |
901 | bv[0].bv_len -= offset; | |
902 | if (diff) | |
903 | bv[bio->bi_vcnt - 1].bv_len -= diff; | |
904 | ||
905 | iov_iter_advance(iter, size); | |
906 | return 0; | |
907 | } | |
908 | EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); | |
909 | ||
910 | struct submit_bio_ret { | |
911 | struct completion event; | |
912 | int error; | |
913 | }; | |
914 | ||
915 | static void submit_bio_wait_endio(struct bio *bio) | |
916 | { | |
917 | struct submit_bio_ret *ret = bio->bi_private; | |
918 | ||
919 | ret->error = bio->bi_error; | |
920 | complete(&ret->event); | |
921 | } | |
922 | ||
923 | /** | |
924 | * submit_bio_wait - submit a bio, and wait until it completes | |
925 | * @bio: The &struct bio which describes the I/O | |
926 | * | |
927 | * Simple wrapper around submit_bio(). Returns 0 on success, or the error from | |
928 | * bio_endio() on failure. | |
929 | */ | |
930 | int submit_bio_wait(struct bio *bio) | |
931 | { | |
932 | struct submit_bio_ret ret; | |
933 | ||
934 | init_completion(&ret.event); | |
935 | bio->bi_private = &ret; | |
936 | bio->bi_end_io = submit_bio_wait_endio; | |
937 | bio->bi_opf |= REQ_SYNC; | |
938 | submit_bio(bio); | |
939 | wait_for_completion_io(&ret.event); | |
940 | ||
941 | return ret.error; | |
942 | } | |
943 | EXPORT_SYMBOL(submit_bio_wait); | |
944 | ||
945 | /** | |
946 | * bio_advance - increment/complete a bio by some number of bytes | |
947 | * @bio: bio to advance | |
948 | * @bytes: number of bytes to complete | |
949 | * | |
950 | * This updates bi_sector, bi_size and bi_idx; if the number of bytes to | |
951 | * complete doesn't align with a bvec boundary, then bv_len and bv_offset will | |
952 | * be updated on the last bvec as well. | |
953 | * | |
954 | * @bio will then represent the remaining, uncompleted portion of the io. | |
955 | */ | |
956 | void bio_advance(struct bio *bio, unsigned bytes) | |
957 | { | |
958 | if (bio_integrity(bio)) | |
959 | bio_integrity_advance(bio, bytes); | |
960 | ||
961 | bio_advance_iter(bio, &bio->bi_iter, bytes); | |
962 | } | |
963 | EXPORT_SYMBOL(bio_advance); | |
964 | ||
965 | /** | |
966 | * bio_alloc_pages - allocates a single page for each bvec in a bio | |
967 | * @bio: bio to allocate pages for | |
968 | * @gfp_mask: flags for allocation | |
969 | * | |
970 | * Allocates pages up to @bio->bi_vcnt. | |
971 | * | |
972 | * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are | |
973 | * freed. | |
974 | */ | |
975 | int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask) | |
976 | { | |
977 | int i; | |
978 | struct bio_vec *bv; | |
979 | ||
980 | bio_for_each_segment_all(bv, bio, i) { | |
981 | bv->bv_page = alloc_page(gfp_mask); | |
982 | if (!bv->bv_page) { | |
983 | while (--bv >= bio->bi_io_vec) | |
984 | __free_page(bv->bv_page); | |
985 | return -ENOMEM; | |
986 | } | |
987 | } | |
988 | ||
989 | return 0; | |
990 | } | |
991 | EXPORT_SYMBOL(bio_alloc_pages); | |
992 | ||
993 | /** | |
994 | * bio_copy_data - copy contents of data buffers from one chain of bios to | |
995 | * another | |
996 | * @src: source bio list | |
997 | * @dst: destination bio list | |
998 | * | |
999 | * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats | |
1000 | * @src and @dst as linked lists of bios. | |
1001 | * | |
1002 | * Stops when it reaches the end of either @src or @dst - that is, copies | |
1003 | * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). | |
1004 | */ | |
1005 | void bio_copy_data(struct bio *dst, struct bio *src) | |
1006 | { | |
1007 | struct bvec_iter src_iter, dst_iter; | |
1008 | struct bio_vec src_bv, dst_bv; | |
1009 | void *src_p, *dst_p; | |
1010 | unsigned bytes; | |
1011 | ||
1012 | src_iter = src->bi_iter; | |
1013 | dst_iter = dst->bi_iter; | |
1014 | ||
1015 | while (1) { | |
1016 | if (!src_iter.bi_size) { | |
1017 | src = src->bi_next; | |
1018 | if (!src) | |
1019 | break; | |
1020 | ||
1021 | src_iter = src->bi_iter; | |
1022 | } | |
1023 | ||
1024 | if (!dst_iter.bi_size) { | |
1025 | dst = dst->bi_next; | |
1026 | if (!dst) | |
1027 | break; | |
1028 | ||
1029 | dst_iter = dst->bi_iter; | |
1030 | } | |
1031 | ||
1032 | src_bv = bio_iter_iovec(src, src_iter); | |
1033 | dst_bv = bio_iter_iovec(dst, dst_iter); | |
1034 | ||
1035 | bytes = min(src_bv.bv_len, dst_bv.bv_len); | |
1036 | ||
1037 | src_p = kmap_atomic(src_bv.bv_page); | |
1038 | dst_p = kmap_atomic(dst_bv.bv_page); | |
1039 | ||
1040 | memcpy(dst_p + dst_bv.bv_offset, | |
1041 | src_p + src_bv.bv_offset, | |
1042 | bytes); | |
1043 | ||
1044 | kunmap_atomic(dst_p); | |
1045 | kunmap_atomic(src_p); | |
1046 | ||
1047 | bio_advance_iter(src, &src_iter, bytes); | |
1048 | bio_advance_iter(dst, &dst_iter, bytes); | |
1049 | } | |
1050 | } | |
1051 | EXPORT_SYMBOL(bio_copy_data); | |
1052 | ||
1053 | struct bio_map_data { | |
1054 | int is_our_pages; | |
1055 | struct iov_iter iter; | |
1056 | struct iovec iov[]; | |
1057 | }; | |
1058 | ||
1059 | static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count, | |
1060 | gfp_t gfp_mask) | |
1061 | { | |
1062 | if (iov_count > UIO_MAXIOV) | |
1063 | return NULL; | |
1064 | ||
1065 | return kmalloc(sizeof(struct bio_map_data) + | |
1066 | sizeof(struct iovec) * iov_count, gfp_mask); | |
1067 | } | |
1068 | ||
1069 | /** | |
1070 | * bio_copy_from_iter - copy all pages from iov_iter to bio | |
1071 | * @bio: The &struct bio which describes the I/O as destination | |
1072 | * @iter: iov_iter as source | |
1073 | * | |
1074 | * Copy all pages from iov_iter to bio. | |
1075 | * Returns 0 on success, or error on failure. | |
1076 | */ | |
1077 | static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter) | |
1078 | { | |
1079 | int i; | |
1080 | struct bio_vec *bvec; | |
1081 | ||
1082 | bio_for_each_segment_all(bvec, bio, i) { | |
1083 | ssize_t ret; | |
1084 | ||
1085 | ret = copy_page_from_iter(bvec->bv_page, | |
1086 | bvec->bv_offset, | |
1087 | bvec->bv_len, | |
1088 | &iter); | |
1089 | ||
1090 | if (!iov_iter_count(&iter)) | |
1091 | break; | |
1092 | ||
1093 | if (ret < bvec->bv_len) | |
1094 | return -EFAULT; | |
1095 | } | |
1096 | ||
1097 | return 0; | |
1098 | } | |
1099 | ||
1100 | /** | |
1101 | * bio_copy_to_iter - copy all pages from bio to iov_iter | |
1102 | * @bio: The &struct bio which describes the I/O as source | |
1103 | * @iter: iov_iter as destination | |
1104 | * | |
1105 | * Copy all pages from bio to iov_iter. | |
1106 | * Returns 0 on success, or error on failure. | |
1107 | */ | |
1108 | static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) | |
1109 | { | |
1110 | int i; | |
1111 | struct bio_vec *bvec; | |
1112 | ||
1113 | bio_for_each_segment_all(bvec, bio, i) { | |
1114 | ssize_t ret; | |
1115 | ||
1116 | ret = copy_page_to_iter(bvec->bv_page, | |
1117 | bvec->bv_offset, | |
1118 | bvec->bv_len, | |
1119 | &iter); | |
1120 | ||
1121 | if (!iov_iter_count(&iter)) | |
1122 | break; | |
1123 | ||
1124 | if (ret < bvec->bv_len) | |
1125 | return -EFAULT; | |
1126 | } | |
1127 | ||
1128 | return 0; | |
1129 | } | |
1130 | ||
1131 | void bio_free_pages(struct bio *bio) | |
1132 | { | |
1133 | struct bio_vec *bvec; | |
1134 | int i; | |
1135 | ||
1136 | bio_for_each_segment_all(bvec, bio, i) | |
1137 | __free_page(bvec->bv_page); | |
1138 | } | |
1139 | EXPORT_SYMBOL(bio_free_pages); | |
1140 | ||
1141 | /** | |
1142 | * bio_uncopy_user - finish previously mapped bio | |
1143 | * @bio: bio being terminated | |
1144 | * | |
1145 | * Free pages allocated from bio_copy_user_iov() and write back data | |
1146 | * to user space in case of a read. | |
1147 | */ | |
1148 | int bio_uncopy_user(struct bio *bio) | |
1149 | { | |
1150 | struct bio_map_data *bmd = bio->bi_private; | |
1151 | int ret = 0; | |
1152 | ||
1153 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) { | |
1154 | /* | |
1155 | * if we're in a workqueue, the request is orphaned, so | |
1156 | * don't copy into a random user address space, just free | |
1157 | * and return -EINTR so user space doesn't expect any data. | |
1158 | */ | |
1159 | if (!current->mm) | |
1160 | ret = -EINTR; | |
1161 | else if (bio_data_dir(bio) == READ) | |
1162 | ret = bio_copy_to_iter(bio, bmd->iter); | |
1163 | if (bmd->is_our_pages) | |
1164 | bio_free_pages(bio); | |
1165 | } | |
1166 | kfree(bmd); | |
1167 | bio_put(bio); | |
1168 | return ret; | |
1169 | } | |
1170 | ||
1171 | /** | |
1172 | * bio_copy_user_iov - copy user data to bio | |
1173 | * @q: destination block queue | |
1174 | * @map_data: pointer to the rq_map_data holding pages (if necessary) | |
1175 | * @iter: iovec iterator | |
1176 | * @gfp_mask: memory allocation flags | |
1177 | * | |
1178 | * Prepares and returns a bio for indirect user io, bouncing data | |
1179 | * to/from kernel pages as necessary. Must be paired with | |
1180 | * call bio_uncopy_user() on io completion. | |
1181 | */ | |
1182 | struct bio *bio_copy_user_iov(struct request_queue *q, | |
1183 | struct rq_map_data *map_data, | |
1184 | const struct iov_iter *iter, | |
1185 | gfp_t gfp_mask) | |
1186 | { | |
1187 | struct bio_map_data *bmd; | |
1188 | struct page *page; | |
1189 | struct bio *bio; | |
1190 | int i, ret; | |
1191 | int nr_pages = 0; | |
1192 | unsigned int len = iter->count; | |
1193 | unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; | |
1194 | ||
1195 | for (i = 0; i < iter->nr_segs; i++) { | |
1196 | unsigned long uaddr; | |
1197 | unsigned long end; | |
1198 | unsigned long start; | |
1199 | ||
1200 | uaddr = (unsigned long) iter->iov[i].iov_base; | |
1201 | end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1) | |
1202 | >> PAGE_SHIFT; | |
1203 | start = uaddr >> PAGE_SHIFT; | |
1204 | ||
1205 | /* | |
1206 | * Overflow, abort | |
1207 | */ | |
1208 | if (end < start) | |
1209 | return ERR_PTR(-EINVAL); | |
1210 | ||
1211 | nr_pages += end - start; | |
1212 | } | |
1213 | ||
1214 | if (offset) | |
1215 | nr_pages++; | |
1216 | ||
1217 | bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask); | |
1218 | if (!bmd) | |
1219 | return ERR_PTR(-ENOMEM); | |
1220 | ||
1221 | /* | |
1222 | * We need to do a deep copy of the iov_iter including the iovecs. | |
1223 | * The caller provided iov might point to an on-stack or otherwise | |
1224 | * shortlived one. | |
1225 | */ | |
1226 | bmd->is_our_pages = map_data ? 0 : 1; | |
1227 | memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs); | |
1228 | iov_iter_init(&bmd->iter, iter->type, bmd->iov, | |
1229 | iter->nr_segs, iter->count); | |
1230 | ||
1231 | ret = -ENOMEM; | |
1232 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1233 | if (!bio) | |
1234 | goto out_bmd; | |
1235 | ||
1236 | if (iter->type & WRITE) | |
1237 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); | |
1238 | ||
1239 | ret = 0; | |
1240 | ||
1241 | if (map_data) { | |
1242 | nr_pages = 1 << map_data->page_order; | |
1243 | i = map_data->offset / PAGE_SIZE; | |
1244 | } | |
1245 | while (len) { | |
1246 | unsigned int bytes = PAGE_SIZE; | |
1247 | ||
1248 | bytes -= offset; | |
1249 | ||
1250 | if (bytes > len) | |
1251 | bytes = len; | |
1252 | ||
1253 | if (map_data) { | |
1254 | if (i == map_data->nr_entries * nr_pages) { | |
1255 | ret = -ENOMEM; | |
1256 | break; | |
1257 | } | |
1258 | ||
1259 | page = map_data->pages[i / nr_pages]; | |
1260 | page += (i % nr_pages); | |
1261 | ||
1262 | i++; | |
1263 | } else { | |
1264 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
1265 | if (!page) { | |
1266 | ret = -ENOMEM; | |
1267 | break; | |
1268 | } | |
1269 | } | |
1270 | ||
1271 | if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) | |
1272 | break; | |
1273 | ||
1274 | len -= bytes; | |
1275 | offset = 0; | |
1276 | } | |
1277 | ||
1278 | if (ret) | |
1279 | goto cleanup; | |
1280 | ||
1281 | /* | |
1282 | * success | |
1283 | */ | |
1284 | if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) || | |
1285 | (map_data && map_data->from_user)) { | |
1286 | ret = bio_copy_from_iter(bio, *iter); | |
1287 | if (ret) | |
1288 | goto cleanup; | |
1289 | } | |
1290 | ||
1291 | bio->bi_private = bmd; | |
1292 | return bio; | |
1293 | cleanup: | |
1294 | if (!map_data) | |
1295 | bio_free_pages(bio); | |
1296 | bio_put(bio); | |
1297 | out_bmd: | |
1298 | kfree(bmd); | |
1299 | return ERR_PTR(ret); | |
1300 | } | |
1301 | ||
1302 | /** | |
1303 | * bio_map_user_iov - map user iovec into bio | |
1304 | * @q: the struct request_queue for the bio | |
1305 | * @iter: iovec iterator | |
1306 | * @gfp_mask: memory allocation flags | |
1307 | * | |
1308 | * Map the user space address into a bio suitable for io to a block | |
1309 | * device. Returns an error pointer in case of error. | |
1310 | */ | |
1311 | struct bio *bio_map_user_iov(struct request_queue *q, | |
1312 | const struct iov_iter *iter, | |
1313 | gfp_t gfp_mask) | |
1314 | { | |
1315 | int j; | |
1316 | int nr_pages = 0; | |
1317 | struct page **pages; | |
1318 | struct bio *bio; | |
1319 | int cur_page = 0; | |
1320 | int ret, offset; | |
1321 | struct iov_iter i; | |
1322 | struct iovec iov; | |
1323 | ||
1324 | iov_for_each(iov, i, *iter) { | |
1325 | unsigned long uaddr = (unsigned long) iov.iov_base; | |
1326 | unsigned long len = iov.iov_len; | |
1327 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1328 | unsigned long start = uaddr >> PAGE_SHIFT; | |
1329 | ||
1330 | /* | |
1331 | * Overflow, abort | |
1332 | */ | |
1333 | if (end < start) | |
1334 | return ERR_PTR(-EINVAL); | |
1335 | ||
1336 | nr_pages += end - start; | |
1337 | /* | |
1338 | * buffer must be aligned to at least logical block size for now | |
1339 | */ | |
1340 | if (uaddr & queue_dma_alignment(q)) | |
1341 | return ERR_PTR(-EINVAL); | |
1342 | } | |
1343 | ||
1344 | if (!nr_pages) | |
1345 | return ERR_PTR(-EINVAL); | |
1346 | ||
1347 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1348 | if (!bio) | |
1349 | return ERR_PTR(-ENOMEM); | |
1350 | ||
1351 | ret = -ENOMEM; | |
1352 | pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask); | |
1353 | if (!pages) | |
1354 | goto out; | |
1355 | ||
1356 | iov_for_each(iov, i, *iter) { | |
1357 | unsigned long uaddr = (unsigned long) iov.iov_base; | |
1358 | unsigned long len = iov.iov_len; | |
1359 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1360 | unsigned long start = uaddr >> PAGE_SHIFT; | |
1361 | const int local_nr_pages = end - start; | |
1362 | const int page_limit = cur_page + local_nr_pages; | |
1363 | ||
1364 | ret = get_user_pages_fast(uaddr, local_nr_pages, | |
1365 | (iter->type & WRITE) != WRITE, | |
1366 | &pages[cur_page]); | |
1367 | if (ret < local_nr_pages) { | |
1368 | ret = -EFAULT; | |
1369 | goto out_unmap; | |
1370 | } | |
1371 | ||
1372 | offset = offset_in_page(uaddr); | |
1373 | for (j = cur_page; j < page_limit; j++) { | |
1374 | unsigned int bytes = PAGE_SIZE - offset; | |
1375 | ||
1376 | if (len <= 0) | |
1377 | break; | |
1378 | ||
1379 | if (bytes > len) | |
1380 | bytes = len; | |
1381 | ||
1382 | /* | |
1383 | * sorry... | |
1384 | */ | |
1385 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < | |
1386 | bytes) | |
1387 | break; | |
1388 | ||
1389 | len -= bytes; | |
1390 | offset = 0; | |
1391 | } | |
1392 | ||
1393 | cur_page = j; | |
1394 | /* | |
1395 | * release the pages we didn't map into the bio, if any | |
1396 | */ | |
1397 | while (j < page_limit) | |
1398 | put_page(pages[j++]); | |
1399 | } | |
1400 | ||
1401 | kfree(pages); | |
1402 | ||
1403 | /* | |
1404 | * set data direction, and check if mapped pages need bouncing | |
1405 | */ | |
1406 | if (iter->type & WRITE) | |
1407 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); | |
1408 | ||
1409 | bio_set_flag(bio, BIO_USER_MAPPED); | |
1410 | ||
1411 | /* | |
1412 | * subtle -- if __bio_map_user() ended up bouncing a bio, | |
1413 | * it would normally disappear when its bi_end_io is run. | |
1414 | * however, we need it for the unmap, so grab an extra | |
1415 | * reference to it | |
1416 | */ | |
1417 | bio_get(bio); | |
1418 | return bio; | |
1419 | ||
1420 | out_unmap: | |
1421 | for (j = 0; j < nr_pages; j++) { | |
1422 | if (!pages[j]) | |
1423 | break; | |
1424 | put_page(pages[j]); | |
1425 | } | |
1426 | out: | |
1427 | kfree(pages); | |
1428 | bio_put(bio); | |
1429 | return ERR_PTR(ret); | |
1430 | } | |
1431 | ||
1432 | static void __bio_unmap_user(struct bio *bio) | |
1433 | { | |
1434 | struct bio_vec *bvec; | |
1435 | int i; | |
1436 | ||
1437 | /* | |
1438 | * make sure we dirty pages we wrote to | |
1439 | */ | |
1440 | bio_for_each_segment_all(bvec, bio, i) { | |
1441 | if (bio_data_dir(bio) == READ) | |
1442 | set_page_dirty_lock(bvec->bv_page); | |
1443 | ||
1444 | put_page(bvec->bv_page); | |
1445 | } | |
1446 | ||
1447 | bio_put(bio); | |
1448 | } | |
1449 | ||
1450 | /** | |
1451 | * bio_unmap_user - unmap a bio | |
1452 | * @bio: the bio being unmapped | |
1453 | * | |
1454 | * Unmap a bio previously mapped by bio_map_user(). Must be called with | |
1455 | * a process context. | |
1456 | * | |
1457 | * bio_unmap_user() may sleep. | |
1458 | */ | |
1459 | void bio_unmap_user(struct bio *bio) | |
1460 | { | |
1461 | __bio_unmap_user(bio); | |
1462 | bio_put(bio); | |
1463 | } | |
1464 | ||
1465 | static void bio_map_kern_endio(struct bio *bio) | |
1466 | { | |
1467 | bio_put(bio); | |
1468 | } | |
1469 | ||
1470 | /** | |
1471 | * bio_map_kern - map kernel address into bio | |
1472 | * @q: the struct request_queue for the bio | |
1473 | * @data: pointer to buffer to map | |
1474 | * @len: length in bytes | |
1475 | * @gfp_mask: allocation flags for bio allocation | |
1476 | * | |
1477 | * Map the kernel address into a bio suitable for io to a block | |
1478 | * device. Returns an error pointer in case of error. | |
1479 | */ | |
1480 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, | |
1481 | gfp_t gfp_mask) | |
1482 | { | |
1483 | unsigned long kaddr = (unsigned long)data; | |
1484 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1485 | unsigned long start = kaddr >> PAGE_SHIFT; | |
1486 | const int nr_pages = end - start; | |
1487 | int offset, i; | |
1488 | struct bio *bio; | |
1489 | ||
1490 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1491 | if (!bio) | |
1492 | return ERR_PTR(-ENOMEM); | |
1493 | ||
1494 | offset = offset_in_page(kaddr); | |
1495 | for (i = 0; i < nr_pages; i++) { | |
1496 | unsigned int bytes = PAGE_SIZE - offset; | |
1497 | ||
1498 | if (len <= 0) | |
1499 | break; | |
1500 | ||
1501 | if (bytes > len) | |
1502 | bytes = len; | |
1503 | ||
1504 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, | |
1505 | offset) < bytes) { | |
1506 | /* we don't support partial mappings */ | |
1507 | bio_put(bio); | |
1508 | return ERR_PTR(-EINVAL); | |
1509 | } | |
1510 | ||
1511 | data += bytes; | |
1512 | len -= bytes; | |
1513 | offset = 0; | |
1514 | } | |
1515 | ||
1516 | bio->bi_end_io = bio_map_kern_endio; | |
1517 | return bio; | |
1518 | } | |
1519 | EXPORT_SYMBOL(bio_map_kern); | |
1520 | ||
1521 | static void bio_copy_kern_endio(struct bio *bio) | |
1522 | { | |
1523 | bio_free_pages(bio); | |
1524 | bio_put(bio); | |
1525 | } | |
1526 | ||
1527 | static void bio_copy_kern_endio_read(struct bio *bio) | |
1528 | { | |
1529 | char *p = bio->bi_private; | |
1530 | struct bio_vec *bvec; | |
1531 | int i; | |
1532 | ||
1533 | bio_for_each_segment_all(bvec, bio, i) { | |
1534 | memcpy(p, page_address(bvec->bv_page), bvec->bv_len); | |
1535 | p += bvec->bv_len; | |
1536 | } | |
1537 | ||
1538 | bio_copy_kern_endio(bio); | |
1539 | } | |
1540 | ||
1541 | /** | |
1542 | * bio_copy_kern - copy kernel address into bio | |
1543 | * @q: the struct request_queue for the bio | |
1544 | * @data: pointer to buffer to copy | |
1545 | * @len: length in bytes | |
1546 | * @gfp_mask: allocation flags for bio and page allocation | |
1547 | * @reading: data direction is READ | |
1548 | * | |
1549 | * copy the kernel address into a bio suitable for io to a block | |
1550 | * device. Returns an error pointer in case of error. | |
1551 | */ | |
1552 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
1553 | gfp_t gfp_mask, int reading) | |
1554 | { | |
1555 | unsigned long kaddr = (unsigned long)data; | |
1556 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1557 | unsigned long start = kaddr >> PAGE_SHIFT; | |
1558 | struct bio *bio; | |
1559 | void *p = data; | |
1560 | int nr_pages = 0; | |
1561 | ||
1562 | /* | |
1563 | * Overflow, abort | |
1564 | */ | |
1565 | if (end < start) | |
1566 | return ERR_PTR(-EINVAL); | |
1567 | ||
1568 | nr_pages = end - start; | |
1569 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1570 | if (!bio) | |
1571 | return ERR_PTR(-ENOMEM); | |
1572 | ||
1573 | while (len) { | |
1574 | struct page *page; | |
1575 | unsigned int bytes = PAGE_SIZE; | |
1576 | ||
1577 | if (bytes > len) | |
1578 | bytes = len; | |
1579 | ||
1580 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
1581 | if (!page) | |
1582 | goto cleanup; | |
1583 | ||
1584 | if (!reading) | |
1585 | memcpy(page_address(page), p, bytes); | |
1586 | ||
1587 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) | |
1588 | break; | |
1589 | ||
1590 | len -= bytes; | |
1591 | p += bytes; | |
1592 | } | |
1593 | ||
1594 | if (reading) { | |
1595 | bio->bi_end_io = bio_copy_kern_endio_read; | |
1596 | bio->bi_private = data; | |
1597 | } else { | |
1598 | bio->bi_end_io = bio_copy_kern_endio; | |
1599 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); | |
1600 | } | |
1601 | ||
1602 | return bio; | |
1603 | ||
1604 | cleanup: | |
1605 | bio_free_pages(bio); | |
1606 | bio_put(bio); | |
1607 | return ERR_PTR(-ENOMEM); | |
1608 | } | |
1609 | ||
1610 | /* | |
1611 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1612 | * for performing direct-IO in BIOs. | |
1613 | * | |
1614 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1615 | * because the required locks are not interrupt-safe. So what we can do is to | |
1616 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1617 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1618 | * in process context. | |
1619 | * | |
1620 | * We special-case compound pages here: normally this means reads into hugetlb | |
1621 | * pages. The logic in here doesn't really work right for compound pages | |
1622 | * because the VM does not uniformly chase down the head page in all cases. | |
1623 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1624 | * handle them at all. So we skip compound pages here at an early stage. | |
1625 | * | |
1626 | * Note that this code is very hard to test under normal circumstances because | |
1627 | * direct-io pins the pages with get_user_pages(). This makes | |
1628 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
1629 | * But other code (eg, flusher threads) could clean the pages if they are mapped | |
1630 | * pagecache. | |
1631 | * | |
1632 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1633 | * deferred bio dirtying paths. | |
1634 | */ | |
1635 | ||
1636 | /* | |
1637 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1638 | */ | |
1639 | void bio_set_pages_dirty(struct bio *bio) | |
1640 | { | |
1641 | struct bio_vec *bvec; | |
1642 | int i; | |
1643 | ||
1644 | bio_for_each_segment_all(bvec, bio, i) { | |
1645 | struct page *page = bvec->bv_page; | |
1646 | ||
1647 | if (page && !PageCompound(page)) | |
1648 | set_page_dirty_lock(page); | |
1649 | } | |
1650 | } | |
1651 | ||
1652 | static void bio_release_pages(struct bio *bio) | |
1653 | { | |
1654 | struct bio_vec *bvec; | |
1655 | int i; | |
1656 | ||
1657 | bio_for_each_segment_all(bvec, bio, i) { | |
1658 | struct page *page = bvec->bv_page; | |
1659 | ||
1660 | if (page) | |
1661 | put_page(page); | |
1662 | } | |
1663 | } | |
1664 | ||
1665 | /* | |
1666 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1667 | * If they are, then fine. If, however, some pages are clean then they must | |
1668 | * have been written out during the direct-IO read. So we take another ref on | |
1669 | * the BIO and the offending pages and re-dirty the pages in process context. | |
1670 | * | |
1671 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
1672 | * here on. It will run one put_page() against each page and will run one | |
1673 | * bio_put() against the BIO. | |
1674 | */ | |
1675 | ||
1676 | static void bio_dirty_fn(struct work_struct *work); | |
1677 | ||
1678 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); | |
1679 | static DEFINE_SPINLOCK(bio_dirty_lock); | |
1680 | static struct bio *bio_dirty_list; | |
1681 | ||
1682 | /* | |
1683 | * This runs in process context | |
1684 | */ | |
1685 | static void bio_dirty_fn(struct work_struct *work) | |
1686 | { | |
1687 | unsigned long flags; | |
1688 | struct bio *bio; | |
1689 | ||
1690 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1691 | bio = bio_dirty_list; | |
1692 | bio_dirty_list = NULL; | |
1693 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1694 | ||
1695 | while (bio) { | |
1696 | struct bio *next = bio->bi_private; | |
1697 | ||
1698 | bio_set_pages_dirty(bio); | |
1699 | bio_release_pages(bio); | |
1700 | bio_put(bio); | |
1701 | bio = next; | |
1702 | } | |
1703 | } | |
1704 | ||
1705 | void bio_check_pages_dirty(struct bio *bio) | |
1706 | { | |
1707 | struct bio_vec *bvec; | |
1708 | int nr_clean_pages = 0; | |
1709 | int i; | |
1710 | ||
1711 | bio_for_each_segment_all(bvec, bio, i) { | |
1712 | struct page *page = bvec->bv_page; | |
1713 | ||
1714 | if (PageDirty(page) || PageCompound(page)) { | |
1715 | put_page(page); | |
1716 | bvec->bv_page = NULL; | |
1717 | } else { | |
1718 | nr_clean_pages++; | |
1719 | } | |
1720 | } | |
1721 | ||
1722 | if (nr_clean_pages) { | |
1723 | unsigned long flags; | |
1724 | ||
1725 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1726 | bio->bi_private = bio_dirty_list; | |
1727 | bio_dirty_list = bio; | |
1728 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1729 | schedule_work(&bio_dirty_work); | |
1730 | } else { | |
1731 | bio_put(bio); | |
1732 | } | |
1733 | } | |
1734 | ||
1735 | void generic_start_io_acct(int rw, unsigned long sectors, | |
1736 | struct hd_struct *part) | |
1737 | { | |
1738 | int cpu = part_stat_lock(); | |
1739 | ||
1740 | part_round_stats(cpu, part); | |
1741 | part_stat_inc(cpu, part, ios[rw]); | |
1742 | part_stat_add(cpu, part, sectors[rw], sectors); | |
1743 | part_inc_in_flight(part, rw); | |
1744 | ||
1745 | part_stat_unlock(); | |
1746 | } | |
1747 | EXPORT_SYMBOL(generic_start_io_acct); | |
1748 | ||
1749 | void generic_end_io_acct(int rw, struct hd_struct *part, | |
1750 | unsigned long start_time) | |
1751 | { | |
1752 | unsigned long duration = jiffies - start_time; | |
1753 | int cpu = part_stat_lock(); | |
1754 | ||
1755 | part_stat_add(cpu, part, ticks[rw], duration); | |
1756 | part_round_stats(cpu, part); | |
1757 | part_dec_in_flight(part, rw); | |
1758 | ||
1759 | part_stat_unlock(); | |
1760 | } | |
1761 | EXPORT_SYMBOL(generic_end_io_acct); | |
1762 | ||
1763 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE | |
1764 | void bio_flush_dcache_pages(struct bio *bi) | |
1765 | { | |
1766 | struct bio_vec bvec; | |
1767 | struct bvec_iter iter; | |
1768 | ||
1769 | bio_for_each_segment(bvec, bi, iter) | |
1770 | flush_dcache_page(bvec.bv_page); | |
1771 | } | |
1772 | EXPORT_SYMBOL(bio_flush_dcache_pages); | |
1773 | #endif | |
1774 | ||
1775 | static inline bool bio_remaining_done(struct bio *bio) | |
1776 | { | |
1777 | /* | |
1778 | * If we're not chaining, then ->__bi_remaining is always 1 and | |
1779 | * we always end io on the first invocation. | |
1780 | */ | |
1781 | if (!bio_flagged(bio, BIO_CHAIN)) | |
1782 | return true; | |
1783 | ||
1784 | BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); | |
1785 | ||
1786 | if (atomic_dec_and_test(&bio->__bi_remaining)) { | |
1787 | bio_clear_flag(bio, BIO_CHAIN); | |
1788 | return true; | |
1789 | } | |
1790 | ||
1791 | return false; | |
1792 | } | |
1793 | ||
1794 | /** | |
1795 | * bio_endio - end I/O on a bio | |
1796 | * @bio: bio | |
1797 | * | |
1798 | * Description: | |
1799 | * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred | |
1800 | * way to end I/O on a bio. No one should call bi_end_io() directly on a | |
1801 | * bio unless they own it and thus know that it has an end_io function. | |
1802 | **/ | |
1803 | void bio_endio(struct bio *bio) | |
1804 | { | |
1805 | again: | |
1806 | if (!bio_remaining_done(bio)) | |
1807 | return; | |
1808 | ||
1809 | /* | |
1810 | * Need to have a real endio function for chained bios, otherwise | |
1811 | * various corner cases will break (like stacking block devices that | |
1812 | * save/restore bi_end_io) - however, we want to avoid unbounded | |
1813 | * recursion and blowing the stack. Tail call optimization would | |
1814 | * handle this, but compiling with frame pointers also disables | |
1815 | * gcc's sibling call optimization. | |
1816 | */ | |
1817 | if (bio->bi_end_io == bio_chain_endio) { | |
1818 | bio = __bio_chain_endio(bio); | |
1819 | goto again; | |
1820 | } | |
1821 | ||
1822 | if (bio->bi_end_io) | |
1823 | bio->bi_end_io(bio); | |
1824 | } | |
1825 | EXPORT_SYMBOL(bio_endio); | |
1826 | ||
1827 | /** | |
1828 | * bio_split - split a bio | |
1829 | * @bio: bio to split | |
1830 | * @sectors: number of sectors to split from the front of @bio | |
1831 | * @gfp: gfp mask | |
1832 | * @bs: bio set to allocate from | |
1833 | * | |
1834 | * Allocates and returns a new bio which represents @sectors from the start of | |
1835 | * @bio, and updates @bio to represent the remaining sectors. | |
1836 | * | |
1837 | * Unless this is a discard request the newly allocated bio will point | |
1838 | * to @bio's bi_io_vec; it is the caller's responsibility to ensure that | |
1839 | * @bio is not freed before the split. | |
1840 | */ | |
1841 | struct bio *bio_split(struct bio *bio, int sectors, | |
1842 | gfp_t gfp, struct bio_set *bs) | |
1843 | { | |
1844 | struct bio *split = NULL; | |
1845 | ||
1846 | BUG_ON(sectors <= 0); | |
1847 | BUG_ON(sectors >= bio_sectors(bio)); | |
1848 | ||
1849 | split = bio_clone_fast(bio, gfp, bs); | |
1850 | if (!split) | |
1851 | return NULL; | |
1852 | ||
1853 | split->bi_iter.bi_size = sectors << 9; | |
1854 | ||
1855 | if (bio_integrity(split)) | |
1856 | bio_integrity_trim(split, 0, sectors); | |
1857 | ||
1858 | bio_advance(bio, split->bi_iter.bi_size); | |
1859 | ||
1860 | return split; | |
1861 | } | |
1862 | EXPORT_SYMBOL(bio_split); | |
1863 | ||
1864 | /** | |
1865 | * bio_trim - trim a bio | |
1866 | * @bio: bio to trim | |
1867 | * @offset: number of sectors to trim from the front of @bio | |
1868 | * @size: size we want to trim @bio to, in sectors | |
1869 | */ | |
1870 | void bio_trim(struct bio *bio, int offset, int size) | |
1871 | { | |
1872 | /* 'bio' is a cloned bio which we need to trim to match | |
1873 | * the given offset and size. | |
1874 | */ | |
1875 | ||
1876 | size <<= 9; | |
1877 | if (offset == 0 && size == bio->bi_iter.bi_size) | |
1878 | return; | |
1879 | ||
1880 | bio_clear_flag(bio, BIO_SEG_VALID); | |
1881 | ||
1882 | bio_advance(bio, offset << 9); | |
1883 | ||
1884 | bio->bi_iter.bi_size = size; | |
1885 | } | |
1886 | EXPORT_SYMBOL_GPL(bio_trim); | |
1887 | ||
1888 | /* | |
1889 | * create memory pools for biovec's in a bio_set. | |
1890 | * use the global biovec slabs created for general use. | |
1891 | */ | |
1892 | mempool_t *biovec_create_pool(int pool_entries) | |
1893 | { | |
1894 | struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; | |
1895 | ||
1896 | return mempool_create_slab_pool(pool_entries, bp->slab); | |
1897 | } | |
1898 | ||
1899 | void bioset_free(struct bio_set *bs) | |
1900 | { | |
1901 | if (bs->rescue_workqueue) | |
1902 | destroy_workqueue(bs->rescue_workqueue); | |
1903 | ||
1904 | if (bs->bio_pool) | |
1905 | mempool_destroy(bs->bio_pool); | |
1906 | ||
1907 | if (bs->bvec_pool) | |
1908 | mempool_destroy(bs->bvec_pool); | |
1909 | ||
1910 | bioset_integrity_free(bs); | |
1911 | bio_put_slab(bs); | |
1912 | ||
1913 | kfree(bs); | |
1914 | } | |
1915 | EXPORT_SYMBOL(bioset_free); | |
1916 | ||
1917 | static struct bio_set *__bioset_create(unsigned int pool_size, | |
1918 | unsigned int front_pad, | |
1919 | bool create_bvec_pool) | |
1920 | { | |
1921 | unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); | |
1922 | struct bio_set *bs; | |
1923 | ||
1924 | bs = kzalloc(sizeof(*bs), GFP_KERNEL); | |
1925 | if (!bs) | |
1926 | return NULL; | |
1927 | ||
1928 | bs->front_pad = front_pad; | |
1929 | ||
1930 | spin_lock_init(&bs->rescue_lock); | |
1931 | bio_list_init(&bs->rescue_list); | |
1932 | INIT_WORK(&bs->rescue_work, bio_alloc_rescue); | |
1933 | ||
1934 | bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); | |
1935 | if (!bs->bio_slab) { | |
1936 | kfree(bs); | |
1937 | return NULL; | |
1938 | } | |
1939 | ||
1940 | bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab); | |
1941 | if (!bs->bio_pool) | |
1942 | goto bad; | |
1943 | ||
1944 | if (create_bvec_pool) { | |
1945 | bs->bvec_pool = biovec_create_pool(pool_size); | |
1946 | if (!bs->bvec_pool) | |
1947 | goto bad; | |
1948 | } | |
1949 | ||
1950 | bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); | |
1951 | if (!bs->rescue_workqueue) | |
1952 | goto bad; | |
1953 | ||
1954 | return bs; | |
1955 | bad: | |
1956 | bioset_free(bs); | |
1957 | return NULL; | |
1958 | } | |
1959 | ||
1960 | /** | |
1961 | * bioset_create - Create a bio_set | |
1962 | * @pool_size: Number of bio and bio_vecs to cache in the mempool | |
1963 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1964 | * | |
1965 | * Description: | |
1966 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller | |
1967 | * to ask for a number of bytes to be allocated in front of the bio. | |
1968 | * Front pad allocation is useful for embedding the bio inside | |
1969 | * another structure, to avoid allocating extra data to go with the bio. | |
1970 | * Note that the bio must be embedded at the END of that structure always, | |
1971 | * or things will break badly. | |
1972 | */ | |
1973 | struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad) | |
1974 | { | |
1975 | return __bioset_create(pool_size, front_pad, true); | |
1976 | } | |
1977 | EXPORT_SYMBOL(bioset_create); | |
1978 | ||
1979 | /** | |
1980 | * bioset_create_nobvec - Create a bio_set without bio_vec mempool | |
1981 | * @pool_size: Number of bio to cache in the mempool | |
1982 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1983 | * | |
1984 | * Description: | |
1985 | * Same functionality as bioset_create() except that mempool is not | |
1986 | * created for bio_vecs. Saving some memory for bio_clone_fast() users. | |
1987 | */ | |
1988 | struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad) | |
1989 | { | |
1990 | return __bioset_create(pool_size, front_pad, false); | |
1991 | } | |
1992 | EXPORT_SYMBOL(bioset_create_nobvec); | |
1993 | ||
1994 | #ifdef CONFIG_BLK_CGROUP | |
1995 | ||
1996 | /** | |
1997 | * bio_associate_blkcg - associate a bio with the specified blkcg | |
1998 | * @bio: target bio | |
1999 | * @blkcg_css: css of the blkcg to associate | |
2000 | * | |
2001 | * Associate @bio with the blkcg specified by @blkcg_css. Block layer will | |
2002 | * treat @bio as if it were issued by a task which belongs to the blkcg. | |
2003 | * | |
2004 | * This function takes an extra reference of @blkcg_css which will be put | |
2005 | * when @bio is released. The caller must own @bio and is responsible for | |
2006 | * synchronizing calls to this function. | |
2007 | */ | |
2008 | int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css) | |
2009 | { | |
2010 | if (unlikely(bio->bi_css)) | |
2011 | return -EBUSY; | |
2012 | css_get(blkcg_css); | |
2013 | bio->bi_css = blkcg_css; | |
2014 | return 0; | |
2015 | } | |
2016 | EXPORT_SYMBOL_GPL(bio_associate_blkcg); | |
2017 | ||
2018 | /** | |
2019 | * bio_associate_current - associate a bio with %current | |
2020 | * @bio: target bio | |
2021 | * | |
2022 | * Associate @bio with %current if it hasn't been associated yet. Block | |
2023 | * layer will treat @bio as if it were issued by %current no matter which | |
2024 | * task actually issues it. | |
2025 | * | |
2026 | * This function takes an extra reference of @task's io_context and blkcg | |
2027 | * which will be put when @bio is released. The caller must own @bio, | |
2028 | * ensure %current->io_context exists, and is responsible for synchronizing | |
2029 | * calls to this function. | |
2030 | */ | |
2031 | int bio_associate_current(struct bio *bio) | |
2032 | { | |
2033 | struct io_context *ioc; | |
2034 | ||
2035 | if (bio->bi_css) | |
2036 | return -EBUSY; | |
2037 | ||
2038 | ioc = current->io_context; | |
2039 | if (!ioc) | |
2040 | return -ENOENT; | |
2041 | ||
2042 | get_io_context_active(ioc); | |
2043 | bio->bi_ioc = ioc; | |
2044 | bio->bi_css = task_get_css(current, io_cgrp_id); | |
2045 | return 0; | |
2046 | } | |
2047 | EXPORT_SYMBOL_GPL(bio_associate_current); | |
2048 | ||
2049 | /** | |
2050 | * bio_disassociate_task - undo bio_associate_current() | |
2051 | * @bio: target bio | |
2052 | */ | |
2053 | void bio_disassociate_task(struct bio *bio) | |
2054 | { | |
2055 | if (bio->bi_ioc) { | |
2056 | put_io_context(bio->bi_ioc); | |
2057 | bio->bi_ioc = NULL; | |
2058 | } | |
2059 | if (bio->bi_css) { | |
2060 | css_put(bio->bi_css); | |
2061 | bio->bi_css = NULL; | |
2062 | } | |
2063 | } | |
2064 | ||
2065 | /** | |
2066 | * bio_clone_blkcg_association - clone blkcg association from src to dst bio | |
2067 | * @dst: destination bio | |
2068 | * @src: source bio | |
2069 | */ | |
2070 | void bio_clone_blkcg_association(struct bio *dst, struct bio *src) | |
2071 | { | |
2072 | if (src->bi_css) | |
2073 | WARN_ON(bio_associate_blkcg(dst, src->bi_css)); | |
2074 | } | |
2075 | ||
2076 | #endif /* CONFIG_BLK_CGROUP */ | |
2077 | ||
2078 | static void __init biovec_init_slabs(void) | |
2079 | { | |
2080 | int i; | |
2081 | ||
2082 | for (i = 0; i < BVEC_POOL_NR; i++) { | |
2083 | int size; | |
2084 | struct biovec_slab *bvs = bvec_slabs + i; | |
2085 | ||
2086 | if (bvs->nr_vecs <= BIO_INLINE_VECS) { | |
2087 | bvs->slab = NULL; | |
2088 | continue; | |
2089 | } | |
2090 | ||
2091 | size = bvs->nr_vecs * sizeof(struct bio_vec); | |
2092 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
2093 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); | |
2094 | } | |
2095 | } | |
2096 | ||
2097 | static int __init init_bio(void) | |
2098 | { | |
2099 | bio_slab_max = 2; | |
2100 | bio_slab_nr = 0; | |
2101 | bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL); | |
2102 | if (!bio_slabs) | |
2103 | panic("bio: can't allocate bios\n"); | |
2104 | ||
2105 | bio_integrity_init(); | |
2106 | biovec_init_slabs(); | |
2107 | ||
2108 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 0); | |
2109 | if (!fs_bio_set) | |
2110 | panic("bio: can't allocate bios\n"); | |
2111 | ||
2112 | if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE)) | |
2113 | panic("bio: can't create integrity pool\n"); | |
2114 | ||
2115 | return 0; | |
2116 | } | |
2117 | subsys_initcall(init_bio); |