]>
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 | #include "blk.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[BVEC_POOL_NR] __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, ARCH_KMALLOC_MINALIGN, | |
116 | SLAB_HWCACHE_ALIGN, NULL); | |
117 | if (!slab) | |
118 | goto out_unlock; | |
119 | ||
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 | if (!idx) | |
165 | return; | |
166 | idx--; | |
167 | ||
168 | BIO_BUG_ON(idx >= BVEC_POOL_NR); | |
169 | ||
170 | if (idx == BVEC_POOL_MAX) { | |
171 | mempool_free(bv, pool); | |
172 | } else { | |
173 | struct biovec_slab *bvs = bvec_slabs + idx; | |
174 | ||
175 | kmem_cache_free(bvs->slab, bv); | |
176 | } | |
177 | } | |
178 | ||
179 | struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, | |
180 | mempool_t *pool) | |
181 | { | |
182 | struct bio_vec *bvl; | |
183 | ||
184 | /* | |
185 | * see comment near bvec_array define! | |
186 | */ | |
187 | switch (nr) { | |
188 | case 1: | |
189 | *idx = 0; | |
190 | break; | |
191 | case 2 ... 4: | |
192 | *idx = 1; | |
193 | break; | |
194 | case 5 ... 16: | |
195 | *idx = 2; | |
196 | break; | |
197 | case 17 ... 64: | |
198 | *idx = 3; | |
199 | break; | |
200 | case 65 ... 128: | |
201 | *idx = 4; | |
202 | break; | |
203 | case 129 ... BIO_MAX_PAGES: | |
204 | *idx = 5; | |
205 | break; | |
206 | default: | |
207 | return NULL; | |
208 | } | |
209 | ||
210 | /* | |
211 | * idx now points to the pool we want to allocate from. only the | |
212 | * 1-vec entry pool is mempool backed. | |
213 | */ | |
214 | if (*idx == BVEC_POOL_MAX) { | |
215 | fallback: | |
216 | bvl = mempool_alloc(pool, gfp_mask); | |
217 | } else { | |
218 | struct biovec_slab *bvs = bvec_slabs + *idx; | |
219 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); | |
220 | ||
221 | /* | |
222 | * Make this allocation restricted and don't dump info on | |
223 | * allocation failures, since we'll fallback to the mempool | |
224 | * in case of failure. | |
225 | */ | |
226 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; | |
227 | ||
228 | /* | |
229 | * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM | |
230 | * is set, retry with the 1-entry mempool | |
231 | */ | |
232 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); | |
233 | if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { | |
234 | *idx = BVEC_POOL_MAX; | |
235 | goto fallback; | |
236 | } | |
237 | } | |
238 | ||
239 | (*idx)++; | |
240 | return bvl; | |
241 | } | |
242 | ||
243 | void bio_uninit(struct bio *bio) | |
244 | { | |
245 | bio_disassociate_task(bio); | |
246 | } | |
247 | EXPORT_SYMBOL(bio_uninit); | |
248 | ||
249 | static void bio_free(struct bio *bio) | |
250 | { | |
251 | struct bio_set *bs = bio->bi_pool; | |
252 | void *p; | |
253 | ||
254 | bio_uninit(bio); | |
255 | ||
256 | if (bs) { | |
257 | bvec_free(bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); | |
258 | ||
259 | /* | |
260 | * If we have front padding, adjust the bio pointer before freeing | |
261 | */ | |
262 | p = bio; | |
263 | p -= bs->front_pad; | |
264 | ||
265 | mempool_free(p, bs->bio_pool); | |
266 | } else { | |
267 | /* Bio was allocated by bio_kmalloc() */ | |
268 | kfree(bio); | |
269 | } | |
270 | } | |
271 | ||
272 | /* | |
273 | * Users of this function have their own bio allocation. Subsequently, | |
274 | * they must remember to pair any call to bio_init() with bio_uninit() | |
275 | * when IO has completed, or when the bio is released. | |
276 | */ | |
277 | void bio_init(struct bio *bio, struct bio_vec *table, | |
278 | unsigned short max_vecs) | |
279 | { | |
280 | memset(bio, 0, sizeof(*bio)); | |
281 | atomic_set(&bio->__bi_remaining, 1); | |
282 | atomic_set(&bio->__bi_cnt, 1); | |
283 | ||
284 | bio->bi_io_vec = table; | |
285 | bio->bi_max_vecs = max_vecs; | |
286 | } | |
287 | EXPORT_SYMBOL(bio_init); | |
288 | ||
289 | /** | |
290 | * bio_reset - reinitialize a bio | |
291 | * @bio: bio to reset | |
292 | * | |
293 | * Description: | |
294 | * After calling bio_reset(), @bio will be in the same state as a freshly | |
295 | * allocated bio returned bio bio_alloc_bioset() - the only fields that are | |
296 | * preserved are the ones that are initialized by bio_alloc_bioset(). See | |
297 | * comment in struct bio. | |
298 | */ | |
299 | void bio_reset(struct bio *bio) | |
300 | { | |
301 | unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); | |
302 | ||
303 | bio_uninit(bio); | |
304 | ||
305 | memset(bio, 0, BIO_RESET_BYTES); | |
306 | bio->bi_flags = flags; | |
307 | atomic_set(&bio->__bi_remaining, 1); | |
308 | } | |
309 | EXPORT_SYMBOL(bio_reset); | |
310 | ||
311 | static struct bio *__bio_chain_endio(struct bio *bio) | |
312 | { | |
313 | struct bio *parent = bio->bi_private; | |
314 | ||
315 | if (!parent->bi_status) | |
316 | parent->bi_status = bio->bi_status; | |
317 | bio_put(bio); | |
318 | return parent; | |
319 | } | |
320 | ||
321 | static void bio_chain_endio(struct bio *bio) | |
322 | { | |
323 | bio_endio(__bio_chain_endio(bio)); | |
324 | } | |
325 | ||
326 | /** | |
327 | * bio_chain - chain bio completions | |
328 | * @bio: the target bio | |
329 | * @parent: the @bio's parent bio | |
330 | * | |
331 | * The caller won't have a bi_end_io called when @bio completes - instead, | |
332 | * @parent's bi_end_io won't be called until both @parent and @bio have | |
333 | * completed; the chained bio will also be freed when it completes. | |
334 | * | |
335 | * The caller must not set bi_private or bi_end_io in @bio. | |
336 | */ | |
337 | void bio_chain(struct bio *bio, struct bio *parent) | |
338 | { | |
339 | BUG_ON(bio->bi_private || bio->bi_end_io); | |
340 | ||
341 | bio->bi_private = parent; | |
342 | bio->bi_end_io = bio_chain_endio; | |
343 | bio_inc_remaining(parent); | |
344 | } | |
345 | EXPORT_SYMBOL(bio_chain); | |
346 | ||
347 | static void bio_alloc_rescue(struct work_struct *work) | |
348 | { | |
349 | struct bio_set *bs = container_of(work, struct bio_set, rescue_work); | |
350 | struct bio *bio; | |
351 | ||
352 | while (1) { | |
353 | spin_lock(&bs->rescue_lock); | |
354 | bio = bio_list_pop(&bs->rescue_list); | |
355 | spin_unlock(&bs->rescue_lock); | |
356 | ||
357 | if (!bio) | |
358 | break; | |
359 | ||
360 | generic_make_request(bio); | |
361 | } | |
362 | } | |
363 | ||
364 | static void punt_bios_to_rescuer(struct bio_set *bs) | |
365 | { | |
366 | struct bio_list punt, nopunt; | |
367 | struct bio *bio; | |
368 | ||
369 | if (WARN_ON_ONCE(!bs->rescue_workqueue)) | |
370 | return; | |
371 | /* | |
372 | * In order to guarantee forward progress we must punt only bios that | |
373 | * were allocated from this bio_set; otherwise, if there was a bio on | |
374 | * there for a stacking driver higher up in the stack, processing it | |
375 | * could require allocating bios from this bio_set, and doing that from | |
376 | * our own rescuer would be bad. | |
377 | * | |
378 | * Since bio lists are singly linked, pop them all instead of trying to | |
379 | * remove from the middle of the list: | |
380 | */ | |
381 | ||
382 | bio_list_init(&punt); | |
383 | bio_list_init(&nopunt); | |
384 | ||
385 | while ((bio = bio_list_pop(¤t->bio_list[0]))) | |
386 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); | |
387 | current->bio_list[0] = nopunt; | |
388 | ||
389 | bio_list_init(&nopunt); | |
390 | while ((bio = bio_list_pop(¤t->bio_list[1]))) | |
391 | bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); | |
392 | current->bio_list[1] = nopunt; | |
393 | ||
394 | spin_lock(&bs->rescue_lock); | |
395 | bio_list_merge(&bs->rescue_list, &punt); | |
396 | spin_unlock(&bs->rescue_lock); | |
397 | ||
398 | queue_work(bs->rescue_workqueue, &bs->rescue_work); | |
399 | } | |
400 | ||
401 | /** | |
402 | * bio_alloc_bioset - allocate a bio for I/O | |
403 | * @gfp_mask: the GFP_ mask given to the slab allocator | |
404 | * @nr_iovecs: number of iovecs to pre-allocate | |
405 | * @bs: the bio_set to allocate from. | |
406 | * | |
407 | * Description: | |
408 | * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is | |
409 | * backed by the @bs's mempool. | |
410 | * | |
411 | * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will | |
412 | * always be able to allocate a bio. This is due to the mempool guarantees. | |
413 | * To make this work, callers must never allocate more than 1 bio at a time | |
414 | * from this pool. Callers that need to allocate more than 1 bio must always | |
415 | * submit the previously allocated bio for IO before attempting to allocate | |
416 | * a new one. Failure to do so can cause deadlocks under memory pressure. | |
417 | * | |
418 | * Note that when running under generic_make_request() (i.e. any block | |
419 | * driver), bios are not submitted until after you return - see the code in | |
420 | * generic_make_request() that converts recursion into iteration, to prevent | |
421 | * stack overflows. | |
422 | * | |
423 | * This would normally mean allocating multiple bios under | |
424 | * generic_make_request() would be susceptible to deadlocks, but we have | |
425 | * deadlock avoidance code that resubmits any blocked bios from a rescuer | |
426 | * thread. | |
427 | * | |
428 | * However, we do not guarantee forward progress for allocations from other | |
429 | * mempools. Doing multiple allocations from the same mempool under | |
430 | * generic_make_request() should be avoided - instead, use bio_set's front_pad | |
431 | * for per bio allocations. | |
432 | * | |
433 | * RETURNS: | |
434 | * Pointer to new bio on success, NULL on failure. | |
435 | */ | |
436 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, | |
437 | struct bio_set *bs) | |
438 | { | |
439 | gfp_t saved_gfp = gfp_mask; | |
440 | unsigned front_pad; | |
441 | unsigned inline_vecs; | |
442 | struct bio_vec *bvl = NULL; | |
443 | struct bio *bio; | |
444 | void *p; | |
445 | ||
446 | if (!bs) { | |
447 | if (nr_iovecs > UIO_MAXIOV) | |
448 | return NULL; | |
449 | ||
450 | p = kmalloc(sizeof(struct bio) + | |
451 | nr_iovecs * sizeof(struct bio_vec), | |
452 | gfp_mask); | |
453 | front_pad = 0; | |
454 | inline_vecs = nr_iovecs; | |
455 | } else { | |
456 | /* should not use nobvec bioset for nr_iovecs > 0 */ | |
457 | if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0)) | |
458 | return NULL; | |
459 | /* | |
460 | * generic_make_request() converts recursion to iteration; this | |
461 | * means if we're running beneath it, any bios we allocate and | |
462 | * submit will not be submitted (and thus freed) until after we | |
463 | * return. | |
464 | * | |
465 | * This exposes us to a potential deadlock if we allocate | |
466 | * multiple bios from the same bio_set() while running | |
467 | * underneath generic_make_request(). If we were to allocate | |
468 | * multiple bios (say a stacking block driver that was splitting | |
469 | * bios), we would deadlock if we exhausted the mempool's | |
470 | * reserve. | |
471 | * | |
472 | * We solve this, and guarantee forward progress, with a rescuer | |
473 | * workqueue per bio_set. If we go to allocate and there are | |
474 | * bios on current->bio_list, we first try the allocation | |
475 | * without __GFP_DIRECT_RECLAIM; if that fails, we punt those | |
476 | * bios we would be blocking to the rescuer workqueue before | |
477 | * we retry with the original gfp_flags. | |
478 | */ | |
479 | ||
480 | if (current->bio_list && | |
481 | (!bio_list_empty(¤t->bio_list[0]) || | |
482 | !bio_list_empty(¤t->bio_list[1])) && | |
483 | bs->rescue_workqueue) | |
484 | gfp_mask &= ~__GFP_DIRECT_RECLAIM; | |
485 | ||
486 | p = mempool_alloc(bs->bio_pool, gfp_mask); | |
487 | if (!p && gfp_mask != saved_gfp) { | |
488 | punt_bios_to_rescuer(bs); | |
489 | gfp_mask = saved_gfp; | |
490 | p = mempool_alloc(bs->bio_pool, gfp_mask); | |
491 | } | |
492 | ||
493 | front_pad = bs->front_pad; | |
494 | inline_vecs = BIO_INLINE_VECS; | |
495 | } | |
496 | ||
497 | if (unlikely(!p)) | |
498 | return NULL; | |
499 | ||
500 | bio = p + front_pad; | |
501 | bio_init(bio, NULL, 0); | |
502 | ||
503 | if (nr_iovecs > inline_vecs) { | |
504 | unsigned long idx = 0; | |
505 | ||
506 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); | |
507 | if (!bvl && gfp_mask != saved_gfp) { | |
508 | punt_bios_to_rescuer(bs); | |
509 | gfp_mask = saved_gfp; | |
510 | bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); | |
511 | } | |
512 | ||
513 | if (unlikely(!bvl)) | |
514 | goto err_free; | |
515 | ||
516 | bio->bi_flags |= idx << BVEC_POOL_OFFSET; | |
517 | } else if (nr_iovecs) { | |
518 | bvl = bio->bi_inline_vecs; | |
519 | } | |
520 | ||
521 | bio->bi_pool = bs; | |
522 | bio->bi_max_vecs = nr_iovecs; | |
523 | bio->bi_io_vec = bvl; | |
524 | return bio; | |
525 | ||
526 | err_free: | |
527 | mempool_free(p, bs->bio_pool); | |
528 | return NULL; | |
529 | } | |
530 | EXPORT_SYMBOL(bio_alloc_bioset); | |
531 | ||
532 | void zero_fill_bio(struct bio *bio) | |
533 | { | |
534 | unsigned long flags; | |
535 | struct bio_vec bv; | |
536 | struct bvec_iter iter; | |
537 | ||
538 | bio_for_each_segment(bv, bio, iter) { | |
539 | char *data = bvec_kmap_irq(&bv, &flags); | |
540 | memset(data, 0, bv.bv_len); | |
541 | flush_dcache_page(bv.bv_page); | |
542 | bvec_kunmap_irq(data, &flags); | |
543 | } | |
544 | } | |
545 | EXPORT_SYMBOL(zero_fill_bio); | |
546 | ||
547 | /** | |
548 | * bio_put - release a reference to a bio | |
549 | * @bio: bio to release reference to | |
550 | * | |
551 | * Description: | |
552 | * Put a reference to a &struct bio, either one you have gotten with | |
553 | * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. | |
554 | **/ | |
555 | void bio_put(struct bio *bio) | |
556 | { | |
557 | if (!bio_flagged(bio, BIO_REFFED)) | |
558 | bio_free(bio); | |
559 | else { | |
560 | BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); | |
561 | ||
562 | /* | |
563 | * last put frees it | |
564 | */ | |
565 | if (atomic_dec_and_test(&bio->__bi_cnt)) | |
566 | bio_free(bio); | |
567 | } | |
568 | } | |
569 | EXPORT_SYMBOL(bio_put); | |
570 | ||
571 | inline int bio_phys_segments(struct request_queue *q, struct bio *bio) | |
572 | { | |
573 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
574 | blk_recount_segments(q, bio); | |
575 | ||
576 | return bio->bi_phys_segments; | |
577 | } | |
578 | EXPORT_SYMBOL(bio_phys_segments); | |
579 | ||
580 | /** | |
581 | * __bio_clone_fast - clone a bio that shares the original bio's biovec | |
582 | * @bio: destination bio | |
583 | * @bio_src: bio to clone | |
584 | * | |
585 | * Clone a &bio. Caller will own the returned bio, but not | |
586 | * the actual data it points to. Reference count of returned | |
587 | * bio will be one. | |
588 | * | |
589 | * Caller must ensure that @bio_src is not freed before @bio. | |
590 | */ | |
591 | void __bio_clone_fast(struct bio *bio, struct bio *bio_src) | |
592 | { | |
593 | BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); | |
594 | ||
595 | /* | |
596 | * most users will be overriding ->bi_bdev with a new target, | |
597 | * so we don't set nor calculate new physical/hw segment counts here | |
598 | */ | |
599 | bio->bi_bdev = bio_src->bi_bdev; | |
600 | bio_set_flag(bio, BIO_CLONED); | |
601 | bio->bi_opf = bio_src->bi_opf; | |
602 | bio->bi_write_hint = bio_src->bi_write_hint; | |
603 | bio->bi_iter = bio_src->bi_iter; | |
604 | bio->bi_io_vec = bio_src->bi_io_vec; | |
605 | ||
606 | bio_clone_blkcg_association(bio, bio_src); | |
607 | } | |
608 | EXPORT_SYMBOL(__bio_clone_fast); | |
609 | ||
610 | /** | |
611 | * bio_clone_fast - clone a bio that shares the original bio's biovec | |
612 | * @bio: bio to clone | |
613 | * @gfp_mask: allocation priority | |
614 | * @bs: bio_set to allocate from | |
615 | * | |
616 | * Like __bio_clone_fast, only also allocates the returned bio | |
617 | */ | |
618 | struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) | |
619 | { | |
620 | struct bio *b; | |
621 | ||
622 | b = bio_alloc_bioset(gfp_mask, 0, bs); | |
623 | if (!b) | |
624 | return NULL; | |
625 | ||
626 | __bio_clone_fast(b, bio); | |
627 | ||
628 | if (bio_integrity(bio)) { | |
629 | int ret; | |
630 | ||
631 | ret = bio_integrity_clone(b, bio, gfp_mask); | |
632 | ||
633 | if (ret < 0) { | |
634 | bio_put(b); | |
635 | return NULL; | |
636 | } | |
637 | } | |
638 | ||
639 | return b; | |
640 | } | |
641 | EXPORT_SYMBOL(bio_clone_fast); | |
642 | ||
643 | /** | |
644 | * bio_clone_bioset - clone a bio | |
645 | * @bio_src: bio to clone | |
646 | * @gfp_mask: allocation priority | |
647 | * @bs: bio_set to allocate from | |
648 | * | |
649 | * Clone bio. Caller will own the returned bio, but not the actual data it | |
650 | * points to. Reference count of returned bio will be one. | |
651 | */ | |
652 | struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask, | |
653 | struct bio_set *bs) | |
654 | { | |
655 | struct bvec_iter iter; | |
656 | struct bio_vec bv; | |
657 | struct bio *bio; | |
658 | ||
659 | /* | |
660 | * Pre immutable biovecs, __bio_clone() used to just do a memcpy from | |
661 | * bio_src->bi_io_vec to bio->bi_io_vec. | |
662 | * | |
663 | * We can't do that anymore, because: | |
664 | * | |
665 | * - The point of cloning the biovec is to produce a bio with a biovec | |
666 | * the caller can modify: bi_idx and bi_bvec_done should be 0. | |
667 | * | |
668 | * - The original bio could've had more than BIO_MAX_PAGES biovecs; if | |
669 | * we tried to clone the whole thing bio_alloc_bioset() would fail. | |
670 | * But the clone should succeed as long as the number of biovecs we | |
671 | * actually need to allocate is fewer than BIO_MAX_PAGES. | |
672 | * | |
673 | * - Lastly, bi_vcnt should not be looked at or relied upon by code | |
674 | * that does not own the bio - reason being drivers don't use it for | |
675 | * iterating over the biovec anymore, so expecting it to be kept up | |
676 | * to date (i.e. for clones that share the parent biovec) is just | |
677 | * asking for trouble and would force extra work on | |
678 | * __bio_clone_fast() anyways. | |
679 | */ | |
680 | ||
681 | bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs); | |
682 | if (!bio) | |
683 | return NULL; | |
684 | bio->bi_bdev = bio_src->bi_bdev; | |
685 | bio->bi_opf = bio_src->bi_opf; | |
686 | bio->bi_write_hint = bio_src->bi_write_hint; | |
687 | bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector; | |
688 | bio->bi_iter.bi_size = bio_src->bi_iter.bi_size; | |
689 | ||
690 | switch (bio_op(bio)) { | |
691 | case REQ_OP_DISCARD: | |
692 | case REQ_OP_SECURE_ERASE: | |
693 | case REQ_OP_WRITE_ZEROES: | |
694 | break; | |
695 | case REQ_OP_WRITE_SAME: | |
696 | bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0]; | |
697 | break; | |
698 | default: | |
699 | bio_for_each_segment(bv, bio_src, iter) | |
700 | bio->bi_io_vec[bio->bi_vcnt++] = bv; | |
701 | break; | |
702 | } | |
703 | ||
704 | if (bio_integrity(bio_src)) { | |
705 | int ret; | |
706 | ||
707 | ret = bio_integrity_clone(bio, bio_src, gfp_mask); | |
708 | if (ret < 0) { | |
709 | bio_put(bio); | |
710 | return NULL; | |
711 | } | |
712 | } | |
713 | ||
714 | bio_clone_blkcg_association(bio, bio_src); | |
715 | ||
716 | return bio; | |
717 | } | |
718 | EXPORT_SYMBOL(bio_clone_bioset); | |
719 | ||
720 | /** | |
721 | * bio_add_pc_page - attempt to add page to bio | |
722 | * @q: the target queue | |
723 | * @bio: destination bio | |
724 | * @page: page to add | |
725 | * @len: vec entry length | |
726 | * @offset: vec entry offset | |
727 | * | |
728 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
729 | * number of reasons, such as the bio being full or target block device | |
730 | * limitations. The target block device must allow bio's up to PAGE_SIZE, | |
731 | * so it is always possible to add a single page to an empty bio. | |
732 | * | |
733 | * This should only be used by REQ_PC bios. | |
734 | */ | |
735 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page | |
736 | *page, unsigned int len, unsigned int offset) | |
737 | { | |
738 | int retried_segments = 0; | |
739 | struct bio_vec *bvec; | |
740 | ||
741 | /* | |
742 | * cloned bio must not modify vec list | |
743 | */ | |
744 | if (unlikely(bio_flagged(bio, BIO_CLONED))) | |
745 | return 0; | |
746 | ||
747 | if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) | |
748 | return 0; | |
749 | ||
750 | /* | |
751 | * For filesystems with a blocksize smaller than the pagesize | |
752 | * we will often be called with the same page as last time and | |
753 | * a consecutive offset. Optimize this special case. | |
754 | */ | |
755 | if (bio->bi_vcnt > 0) { | |
756 | struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
757 | ||
758 | if (page == prev->bv_page && | |
759 | offset == prev->bv_offset + prev->bv_len) { | |
760 | prev->bv_len += len; | |
761 | bio->bi_iter.bi_size += len; | |
762 | goto done; | |
763 | } | |
764 | ||
765 | /* | |
766 | * If the queue doesn't support SG gaps and adding this | |
767 | * offset would create a gap, disallow it. | |
768 | */ | |
769 | if (bvec_gap_to_prev(q, prev, offset)) | |
770 | return 0; | |
771 | } | |
772 | ||
773 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
774 | return 0; | |
775 | ||
776 | /* | |
777 | * setup the new entry, we might clear it again later if we | |
778 | * cannot add the page | |
779 | */ | |
780 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; | |
781 | bvec->bv_page = page; | |
782 | bvec->bv_len = len; | |
783 | bvec->bv_offset = offset; | |
784 | bio->bi_vcnt++; | |
785 | bio->bi_phys_segments++; | |
786 | bio->bi_iter.bi_size += len; | |
787 | ||
788 | /* | |
789 | * Perform a recount if the number of segments is greater | |
790 | * than queue_max_segments(q). | |
791 | */ | |
792 | ||
793 | while (bio->bi_phys_segments > queue_max_segments(q)) { | |
794 | ||
795 | if (retried_segments) | |
796 | goto failed; | |
797 | ||
798 | retried_segments = 1; | |
799 | blk_recount_segments(q, bio); | |
800 | } | |
801 | ||
802 | /* If we may be able to merge these biovecs, force a recount */ | |
803 | if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) | |
804 | bio_clear_flag(bio, BIO_SEG_VALID); | |
805 | ||
806 | done: | |
807 | return len; | |
808 | ||
809 | failed: | |
810 | bvec->bv_page = NULL; | |
811 | bvec->bv_len = 0; | |
812 | bvec->bv_offset = 0; | |
813 | bio->bi_vcnt--; | |
814 | bio->bi_iter.bi_size -= len; | |
815 | blk_recount_segments(q, bio); | |
816 | return 0; | |
817 | } | |
818 | EXPORT_SYMBOL(bio_add_pc_page); | |
819 | ||
820 | /** | |
821 | * bio_add_page - attempt to add page to bio | |
822 | * @bio: destination bio | |
823 | * @page: page to add | |
824 | * @len: vec entry length | |
825 | * @offset: vec entry offset | |
826 | * | |
827 | * Attempt to add a page to the bio_vec maplist. This will only fail | |
828 | * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. | |
829 | */ | |
830 | int bio_add_page(struct bio *bio, struct page *page, | |
831 | unsigned int len, unsigned int offset) | |
832 | { | |
833 | struct bio_vec *bv; | |
834 | ||
835 | /* | |
836 | * cloned bio must not modify vec list | |
837 | */ | |
838 | if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) | |
839 | return 0; | |
840 | ||
841 | /* | |
842 | * For filesystems with a blocksize smaller than the pagesize | |
843 | * we will often be called with the same page as last time and | |
844 | * a consecutive offset. Optimize this special case. | |
845 | */ | |
846 | if (bio->bi_vcnt > 0) { | |
847 | bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
848 | ||
849 | if (page == bv->bv_page && | |
850 | offset == bv->bv_offset + bv->bv_len) { | |
851 | bv->bv_len += len; | |
852 | goto done; | |
853 | } | |
854 | } | |
855 | ||
856 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
857 | return 0; | |
858 | ||
859 | bv = &bio->bi_io_vec[bio->bi_vcnt]; | |
860 | bv->bv_page = page; | |
861 | bv->bv_len = len; | |
862 | bv->bv_offset = offset; | |
863 | ||
864 | bio->bi_vcnt++; | |
865 | done: | |
866 | bio->bi_iter.bi_size += len; | |
867 | return len; | |
868 | } | |
869 | EXPORT_SYMBOL(bio_add_page); | |
870 | ||
871 | /** | |
872 | * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio | |
873 | * @bio: bio to add pages to | |
874 | * @iter: iov iterator describing the region to be mapped | |
875 | * | |
876 | * Pins as many pages from *iter and appends them to @bio's bvec array. The | |
877 | * pages will have to be released using put_page() when done. | |
878 | */ | |
879 | int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) | |
880 | { | |
881 | unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; | |
882 | struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; | |
883 | struct page **pages = (struct page **)bv; | |
884 | size_t offset, diff; | |
885 | ssize_t size; | |
886 | ||
887 | size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); | |
888 | if (unlikely(size <= 0)) | |
889 | return size ? size : -EFAULT; | |
890 | nr_pages = (size + offset + PAGE_SIZE - 1) / PAGE_SIZE; | |
891 | ||
892 | /* | |
893 | * Deep magic below: We need to walk the pinned pages backwards | |
894 | * because we are abusing the space allocated for the bio_vecs | |
895 | * for the page array. Because the bio_vecs are larger than the | |
896 | * page pointers by definition this will always work. But it also | |
897 | * means we can't use bio_add_page, so any changes to it's semantics | |
898 | * need to be reflected here as well. | |
899 | */ | |
900 | bio->bi_iter.bi_size += size; | |
901 | bio->bi_vcnt += nr_pages; | |
902 | ||
903 | diff = (nr_pages * PAGE_SIZE - offset) - size; | |
904 | while (nr_pages--) { | |
905 | bv[nr_pages].bv_page = pages[nr_pages]; | |
906 | bv[nr_pages].bv_len = PAGE_SIZE; | |
907 | bv[nr_pages].bv_offset = 0; | |
908 | } | |
909 | ||
910 | bv[0].bv_offset += offset; | |
911 | bv[0].bv_len -= offset; | |
912 | if (diff) | |
913 | bv[bio->bi_vcnt - 1].bv_len -= diff; | |
914 | ||
915 | iov_iter_advance(iter, size); | |
916 | return 0; | |
917 | } | |
918 | EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); | |
919 | ||
920 | struct submit_bio_ret { | |
921 | struct completion event; | |
922 | int error; | |
923 | }; | |
924 | ||
925 | static void submit_bio_wait_endio(struct bio *bio) | |
926 | { | |
927 | struct submit_bio_ret *ret = bio->bi_private; | |
928 | ||
929 | ret->error = blk_status_to_errno(bio->bi_status); | |
930 | complete(&ret->event); | |
931 | } | |
932 | ||
933 | /** | |
934 | * submit_bio_wait - submit a bio, and wait until it completes | |
935 | * @bio: The &struct bio which describes the I/O | |
936 | * | |
937 | * Simple wrapper around submit_bio(). Returns 0 on success, or the error from | |
938 | * bio_endio() on failure. | |
939 | */ | |
940 | int submit_bio_wait(struct bio *bio) | |
941 | { | |
942 | struct submit_bio_ret ret; | |
943 | ||
944 | init_completion(&ret.event); | |
945 | bio->bi_private = &ret; | |
946 | bio->bi_end_io = submit_bio_wait_endio; | |
947 | bio->bi_opf |= REQ_SYNC; | |
948 | submit_bio(bio); | |
949 | wait_for_completion_io(&ret.event); | |
950 | ||
951 | return ret.error; | |
952 | } | |
953 | EXPORT_SYMBOL(submit_bio_wait); | |
954 | ||
955 | /** | |
956 | * bio_advance - increment/complete a bio by some number of bytes | |
957 | * @bio: bio to advance | |
958 | * @bytes: number of bytes to complete | |
959 | * | |
960 | * This updates bi_sector, bi_size and bi_idx; if the number of bytes to | |
961 | * complete doesn't align with a bvec boundary, then bv_len and bv_offset will | |
962 | * be updated on the last bvec as well. | |
963 | * | |
964 | * @bio will then represent the remaining, uncompleted portion of the io. | |
965 | */ | |
966 | void bio_advance(struct bio *bio, unsigned bytes) | |
967 | { | |
968 | if (bio_integrity(bio)) | |
969 | bio_integrity_advance(bio, bytes); | |
970 | ||
971 | bio_advance_iter(bio, &bio->bi_iter, bytes); | |
972 | } | |
973 | EXPORT_SYMBOL(bio_advance); | |
974 | ||
975 | /** | |
976 | * bio_alloc_pages - allocates a single page for each bvec in a bio | |
977 | * @bio: bio to allocate pages for | |
978 | * @gfp_mask: flags for allocation | |
979 | * | |
980 | * Allocates pages up to @bio->bi_vcnt. | |
981 | * | |
982 | * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are | |
983 | * freed. | |
984 | */ | |
985 | int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask) | |
986 | { | |
987 | int i; | |
988 | struct bio_vec *bv; | |
989 | ||
990 | bio_for_each_segment_all(bv, bio, i) { | |
991 | bv->bv_page = alloc_page(gfp_mask); | |
992 | if (!bv->bv_page) { | |
993 | while (--bv >= bio->bi_io_vec) | |
994 | __free_page(bv->bv_page); | |
995 | return -ENOMEM; | |
996 | } | |
997 | } | |
998 | ||
999 | return 0; | |
1000 | } | |
1001 | EXPORT_SYMBOL(bio_alloc_pages); | |
1002 | ||
1003 | /** | |
1004 | * bio_copy_data - copy contents of data buffers from one chain of bios to | |
1005 | * another | |
1006 | * @src: source bio list | |
1007 | * @dst: destination bio list | |
1008 | * | |
1009 | * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats | |
1010 | * @src and @dst as linked lists of bios. | |
1011 | * | |
1012 | * Stops when it reaches the end of either @src or @dst - that is, copies | |
1013 | * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). | |
1014 | */ | |
1015 | void bio_copy_data(struct bio *dst, struct bio *src) | |
1016 | { | |
1017 | struct bvec_iter src_iter, dst_iter; | |
1018 | struct bio_vec src_bv, dst_bv; | |
1019 | void *src_p, *dst_p; | |
1020 | unsigned bytes; | |
1021 | ||
1022 | src_iter = src->bi_iter; | |
1023 | dst_iter = dst->bi_iter; | |
1024 | ||
1025 | while (1) { | |
1026 | if (!src_iter.bi_size) { | |
1027 | src = src->bi_next; | |
1028 | if (!src) | |
1029 | break; | |
1030 | ||
1031 | src_iter = src->bi_iter; | |
1032 | } | |
1033 | ||
1034 | if (!dst_iter.bi_size) { | |
1035 | dst = dst->bi_next; | |
1036 | if (!dst) | |
1037 | break; | |
1038 | ||
1039 | dst_iter = dst->bi_iter; | |
1040 | } | |
1041 | ||
1042 | src_bv = bio_iter_iovec(src, src_iter); | |
1043 | dst_bv = bio_iter_iovec(dst, dst_iter); | |
1044 | ||
1045 | bytes = min(src_bv.bv_len, dst_bv.bv_len); | |
1046 | ||
1047 | src_p = kmap_atomic(src_bv.bv_page); | |
1048 | dst_p = kmap_atomic(dst_bv.bv_page); | |
1049 | ||
1050 | memcpy(dst_p + dst_bv.bv_offset, | |
1051 | src_p + src_bv.bv_offset, | |
1052 | bytes); | |
1053 | ||
1054 | kunmap_atomic(dst_p); | |
1055 | kunmap_atomic(src_p); | |
1056 | ||
1057 | bio_advance_iter(src, &src_iter, bytes); | |
1058 | bio_advance_iter(dst, &dst_iter, bytes); | |
1059 | } | |
1060 | } | |
1061 | EXPORT_SYMBOL(bio_copy_data); | |
1062 | ||
1063 | struct bio_map_data { | |
1064 | int is_our_pages; | |
1065 | struct iov_iter iter; | |
1066 | struct iovec iov[]; | |
1067 | }; | |
1068 | ||
1069 | static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count, | |
1070 | gfp_t gfp_mask) | |
1071 | { | |
1072 | if (iov_count > UIO_MAXIOV) | |
1073 | return NULL; | |
1074 | ||
1075 | return kmalloc(sizeof(struct bio_map_data) + | |
1076 | sizeof(struct iovec) * iov_count, gfp_mask); | |
1077 | } | |
1078 | ||
1079 | /** | |
1080 | * bio_copy_from_iter - copy all pages from iov_iter to bio | |
1081 | * @bio: The &struct bio which describes the I/O as destination | |
1082 | * @iter: iov_iter as source | |
1083 | * | |
1084 | * Copy all pages from iov_iter to bio. | |
1085 | * Returns 0 on success, or error on failure. | |
1086 | */ | |
1087 | static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter) | |
1088 | { | |
1089 | int i; | |
1090 | struct bio_vec *bvec; | |
1091 | ||
1092 | bio_for_each_segment_all(bvec, bio, i) { | |
1093 | ssize_t ret; | |
1094 | ||
1095 | ret = copy_page_from_iter(bvec->bv_page, | |
1096 | bvec->bv_offset, | |
1097 | bvec->bv_len, | |
1098 | &iter); | |
1099 | ||
1100 | if (!iov_iter_count(&iter)) | |
1101 | break; | |
1102 | ||
1103 | if (ret < bvec->bv_len) | |
1104 | return -EFAULT; | |
1105 | } | |
1106 | ||
1107 | return 0; | |
1108 | } | |
1109 | ||
1110 | /** | |
1111 | * bio_copy_to_iter - copy all pages from bio to iov_iter | |
1112 | * @bio: The &struct bio which describes the I/O as source | |
1113 | * @iter: iov_iter as destination | |
1114 | * | |
1115 | * Copy all pages from bio to iov_iter. | |
1116 | * Returns 0 on success, or error on failure. | |
1117 | */ | |
1118 | static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) | |
1119 | { | |
1120 | int i; | |
1121 | struct bio_vec *bvec; | |
1122 | ||
1123 | bio_for_each_segment_all(bvec, bio, i) { | |
1124 | ssize_t ret; | |
1125 | ||
1126 | ret = copy_page_to_iter(bvec->bv_page, | |
1127 | bvec->bv_offset, | |
1128 | bvec->bv_len, | |
1129 | &iter); | |
1130 | ||
1131 | if (!iov_iter_count(&iter)) | |
1132 | break; | |
1133 | ||
1134 | if (ret < bvec->bv_len) | |
1135 | return -EFAULT; | |
1136 | } | |
1137 | ||
1138 | return 0; | |
1139 | } | |
1140 | ||
1141 | void bio_free_pages(struct bio *bio) | |
1142 | { | |
1143 | struct bio_vec *bvec; | |
1144 | int i; | |
1145 | ||
1146 | bio_for_each_segment_all(bvec, bio, i) | |
1147 | __free_page(bvec->bv_page); | |
1148 | } | |
1149 | EXPORT_SYMBOL(bio_free_pages); | |
1150 | ||
1151 | /** | |
1152 | * bio_uncopy_user - finish previously mapped bio | |
1153 | * @bio: bio being terminated | |
1154 | * | |
1155 | * Free pages allocated from bio_copy_user_iov() and write back data | |
1156 | * to user space in case of a read. | |
1157 | */ | |
1158 | int bio_uncopy_user(struct bio *bio) | |
1159 | { | |
1160 | struct bio_map_data *bmd = bio->bi_private; | |
1161 | int ret = 0; | |
1162 | ||
1163 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) { | |
1164 | /* | |
1165 | * if we're in a workqueue, the request is orphaned, so | |
1166 | * don't copy into a random user address space, just free | |
1167 | * and return -EINTR so user space doesn't expect any data. | |
1168 | */ | |
1169 | if (!current->mm) | |
1170 | ret = -EINTR; | |
1171 | else if (bio_data_dir(bio) == READ) | |
1172 | ret = bio_copy_to_iter(bio, bmd->iter); | |
1173 | if (bmd->is_our_pages) | |
1174 | bio_free_pages(bio); | |
1175 | } | |
1176 | kfree(bmd); | |
1177 | bio_put(bio); | |
1178 | return ret; | |
1179 | } | |
1180 | ||
1181 | /** | |
1182 | * bio_copy_user_iov - copy user data to bio | |
1183 | * @q: destination block queue | |
1184 | * @map_data: pointer to the rq_map_data holding pages (if necessary) | |
1185 | * @iter: iovec iterator | |
1186 | * @gfp_mask: memory allocation flags | |
1187 | * | |
1188 | * Prepares and returns a bio for indirect user io, bouncing data | |
1189 | * to/from kernel pages as necessary. Must be paired with | |
1190 | * call bio_uncopy_user() on io completion. | |
1191 | */ | |
1192 | struct bio *bio_copy_user_iov(struct request_queue *q, | |
1193 | struct rq_map_data *map_data, | |
1194 | const struct iov_iter *iter, | |
1195 | gfp_t gfp_mask) | |
1196 | { | |
1197 | struct bio_map_data *bmd; | |
1198 | struct page *page; | |
1199 | struct bio *bio; | |
1200 | int i, ret; | |
1201 | int nr_pages = 0; | |
1202 | unsigned int len = iter->count; | |
1203 | unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; | |
1204 | ||
1205 | for (i = 0; i < iter->nr_segs; i++) { | |
1206 | unsigned long uaddr; | |
1207 | unsigned long end; | |
1208 | unsigned long start; | |
1209 | ||
1210 | uaddr = (unsigned long) iter->iov[i].iov_base; | |
1211 | end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1) | |
1212 | >> PAGE_SHIFT; | |
1213 | start = uaddr >> PAGE_SHIFT; | |
1214 | ||
1215 | /* | |
1216 | * Overflow, abort | |
1217 | */ | |
1218 | if (end < start) | |
1219 | return ERR_PTR(-EINVAL); | |
1220 | ||
1221 | nr_pages += end - start; | |
1222 | } | |
1223 | ||
1224 | if (offset) | |
1225 | nr_pages++; | |
1226 | ||
1227 | bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask); | |
1228 | if (!bmd) | |
1229 | return ERR_PTR(-ENOMEM); | |
1230 | ||
1231 | /* | |
1232 | * We need to do a deep copy of the iov_iter including the iovecs. | |
1233 | * The caller provided iov might point to an on-stack or otherwise | |
1234 | * shortlived one. | |
1235 | */ | |
1236 | bmd->is_our_pages = map_data ? 0 : 1; | |
1237 | memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs); | |
1238 | bmd->iter = *iter; | |
1239 | bmd->iter.iov = bmd->iov; | |
1240 | ||
1241 | ret = -ENOMEM; | |
1242 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1243 | if (!bio) | |
1244 | goto out_bmd; | |
1245 | ||
1246 | ret = 0; | |
1247 | ||
1248 | if (map_data) { | |
1249 | nr_pages = 1 << map_data->page_order; | |
1250 | i = map_data->offset / PAGE_SIZE; | |
1251 | } | |
1252 | while (len) { | |
1253 | unsigned int bytes = PAGE_SIZE; | |
1254 | ||
1255 | bytes -= offset; | |
1256 | ||
1257 | if (bytes > len) | |
1258 | bytes = len; | |
1259 | ||
1260 | if (map_data) { | |
1261 | if (i == map_data->nr_entries * nr_pages) { | |
1262 | ret = -ENOMEM; | |
1263 | break; | |
1264 | } | |
1265 | ||
1266 | page = map_data->pages[i / nr_pages]; | |
1267 | page += (i % nr_pages); | |
1268 | ||
1269 | i++; | |
1270 | } else { | |
1271 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
1272 | if (!page) { | |
1273 | ret = -ENOMEM; | |
1274 | break; | |
1275 | } | |
1276 | } | |
1277 | ||
1278 | if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) | |
1279 | break; | |
1280 | ||
1281 | len -= bytes; | |
1282 | offset = 0; | |
1283 | } | |
1284 | ||
1285 | if (ret) | |
1286 | goto cleanup; | |
1287 | ||
1288 | /* | |
1289 | * success | |
1290 | */ | |
1291 | if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) || | |
1292 | (map_data && map_data->from_user)) { | |
1293 | ret = bio_copy_from_iter(bio, *iter); | |
1294 | if (ret) | |
1295 | goto cleanup; | |
1296 | } | |
1297 | ||
1298 | bio->bi_private = bmd; | |
1299 | return bio; | |
1300 | cleanup: | |
1301 | if (!map_data) | |
1302 | bio_free_pages(bio); | |
1303 | bio_put(bio); | |
1304 | out_bmd: | |
1305 | kfree(bmd); | |
1306 | return ERR_PTR(ret); | |
1307 | } | |
1308 | ||
1309 | /** | |
1310 | * bio_map_user_iov - map user iovec into bio | |
1311 | * @q: the struct request_queue for the bio | |
1312 | * @iter: iovec iterator | |
1313 | * @gfp_mask: memory allocation flags | |
1314 | * | |
1315 | * Map the user space address into a bio suitable for io to a block | |
1316 | * device. Returns an error pointer in case of error. | |
1317 | */ | |
1318 | struct bio *bio_map_user_iov(struct request_queue *q, | |
1319 | const struct iov_iter *iter, | |
1320 | gfp_t gfp_mask) | |
1321 | { | |
1322 | int j; | |
1323 | int nr_pages = 0; | |
1324 | struct page **pages; | |
1325 | struct bio *bio; | |
1326 | int cur_page = 0; | |
1327 | int ret, offset; | |
1328 | struct iov_iter i; | |
1329 | struct iovec iov; | |
1330 | struct bio_vec *bvec; | |
1331 | ||
1332 | iov_for_each(iov, i, *iter) { | |
1333 | unsigned long uaddr = (unsigned long) iov.iov_base; | |
1334 | unsigned long len = iov.iov_len; | |
1335 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1336 | unsigned long start = uaddr >> PAGE_SHIFT; | |
1337 | ||
1338 | /* | |
1339 | * Overflow, abort | |
1340 | */ | |
1341 | if (end < start) | |
1342 | return ERR_PTR(-EINVAL); | |
1343 | ||
1344 | nr_pages += end - start; | |
1345 | /* | |
1346 | * buffer must be aligned to at least logical block size for now | |
1347 | */ | |
1348 | if (uaddr & queue_dma_alignment(q)) | |
1349 | return ERR_PTR(-EINVAL); | |
1350 | } | |
1351 | ||
1352 | if (!nr_pages) | |
1353 | return ERR_PTR(-EINVAL); | |
1354 | ||
1355 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1356 | if (!bio) | |
1357 | return ERR_PTR(-ENOMEM); | |
1358 | ||
1359 | ret = -ENOMEM; | |
1360 | pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask); | |
1361 | if (!pages) | |
1362 | goto out; | |
1363 | ||
1364 | iov_for_each(iov, i, *iter) { | |
1365 | unsigned long uaddr = (unsigned long) iov.iov_base; | |
1366 | unsigned long len = iov.iov_len; | |
1367 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1368 | unsigned long start = uaddr >> PAGE_SHIFT; | |
1369 | const int local_nr_pages = end - start; | |
1370 | const int page_limit = cur_page + local_nr_pages; | |
1371 | ||
1372 | ret = get_user_pages_fast(uaddr, local_nr_pages, | |
1373 | (iter->type & WRITE) != WRITE, | |
1374 | &pages[cur_page]); | |
1375 | if (unlikely(ret < local_nr_pages)) { | |
1376 | for (j = cur_page; j < page_limit; j++) { | |
1377 | if (!pages[j]) | |
1378 | break; | |
1379 | put_page(pages[j]); | |
1380 | } | |
1381 | ret = -EFAULT; | |
1382 | goto out_unmap; | |
1383 | } | |
1384 | ||
1385 | offset = offset_in_page(uaddr); | |
1386 | for (j = cur_page; j < page_limit; j++) { | |
1387 | unsigned int bytes = PAGE_SIZE - offset; | |
1388 | unsigned short prev_bi_vcnt = bio->bi_vcnt; | |
1389 | ||
1390 | if (len <= 0) | |
1391 | break; | |
1392 | ||
1393 | if (bytes > len) | |
1394 | bytes = len; | |
1395 | ||
1396 | /* | |
1397 | * sorry... | |
1398 | */ | |
1399 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < | |
1400 | bytes) | |
1401 | break; | |
1402 | ||
1403 | /* | |
1404 | * check if vector was merged with previous | |
1405 | * drop page reference if needed | |
1406 | */ | |
1407 | if (bio->bi_vcnt == prev_bi_vcnt) | |
1408 | put_page(pages[j]); | |
1409 | ||
1410 | len -= bytes; | |
1411 | offset = 0; | |
1412 | } | |
1413 | ||
1414 | cur_page = j; | |
1415 | /* | |
1416 | * release the pages we didn't map into the bio, if any | |
1417 | */ | |
1418 | while (j < page_limit) | |
1419 | put_page(pages[j++]); | |
1420 | } | |
1421 | ||
1422 | kfree(pages); | |
1423 | ||
1424 | bio_set_flag(bio, BIO_USER_MAPPED); | |
1425 | ||
1426 | /* | |
1427 | * subtle -- if bio_map_user_iov() ended up bouncing a bio, | |
1428 | * it would normally disappear when its bi_end_io is run. | |
1429 | * however, we need it for the unmap, so grab an extra | |
1430 | * reference to it | |
1431 | */ | |
1432 | bio_get(bio); | |
1433 | return bio; | |
1434 | ||
1435 | out_unmap: | |
1436 | bio_for_each_segment_all(bvec, bio, j) { | |
1437 | put_page(bvec->bv_page); | |
1438 | } | |
1439 | out: | |
1440 | kfree(pages); | |
1441 | bio_put(bio); | |
1442 | return ERR_PTR(ret); | |
1443 | } | |
1444 | ||
1445 | static void __bio_unmap_user(struct bio *bio) | |
1446 | { | |
1447 | struct bio_vec *bvec; | |
1448 | int i; | |
1449 | ||
1450 | /* | |
1451 | * make sure we dirty pages we wrote to | |
1452 | */ | |
1453 | bio_for_each_segment_all(bvec, bio, i) { | |
1454 | if (bio_data_dir(bio) == READ) | |
1455 | set_page_dirty_lock(bvec->bv_page); | |
1456 | ||
1457 | put_page(bvec->bv_page); | |
1458 | } | |
1459 | ||
1460 | bio_put(bio); | |
1461 | } | |
1462 | ||
1463 | /** | |
1464 | * bio_unmap_user - unmap a bio | |
1465 | * @bio: the bio being unmapped | |
1466 | * | |
1467 | * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from | |
1468 | * process context. | |
1469 | * | |
1470 | * bio_unmap_user() may sleep. | |
1471 | */ | |
1472 | void bio_unmap_user(struct bio *bio) | |
1473 | { | |
1474 | __bio_unmap_user(bio); | |
1475 | bio_put(bio); | |
1476 | } | |
1477 | ||
1478 | static void bio_map_kern_endio(struct bio *bio) | |
1479 | { | |
1480 | bio_put(bio); | |
1481 | } | |
1482 | ||
1483 | /** | |
1484 | * bio_map_kern - map kernel address into bio | |
1485 | * @q: the struct request_queue for the bio | |
1486 | * @data: pointer to buffer to map | |
1487 | * @len: length in bytes | |
1488 | * @gfp_mask: allocation flags for bio allocation | |
1489 | * | |
1490 | * Map the kernel address into a bio suitable for io to a block | |
1491 | * device. Returns an error pointer in case of error. | |
1492 | */ | |
1493 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, | |
1494 | gfp_t gfp_mask) | |
1495 | { | |
1496 | unsigned long kaddr = (unsigned long)data; | |
1497 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1498 | unsigned long start = kaddr >> PAGE_SHIFT; | |
1499 | const int nr_pages = end - start; | |
1500 | int offset, i; | |
1501 | struct bio *bio; | |
1502 | ||
1503 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1504 | if (!bio) | |
1505 | return ERR_PTR(-ENOMEM); | |
1506 | ||
1507 | offset = offset_in_page(kaddr); | |
1508 | for (i = 0; i < nr_pages; i++) { | |
1509 | unsigned int bytes = PAGE_SIZE - offset; | |
1510 | ||
1511 | if (len <= 0) | |
1512 | break; | |
1513 | ||
1514 | if (bytes > len) | |
1515 | bytes = len; | |
1516 | ||
1517 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, | |
1518 | offset) < bytes) { | |
1519 | /* we don't support partial mappings */ | |
1520 | bio_put(bio); | |
1521 | return ERR_PTR(-EINVAL); | |
1522 | } | |
1523 | ||
1524 | data += bytes; | |
1525 | len -= bytes; | |
1526 | offset = 0; | |
1527 | } | |
1528 | ||
1529 | bio->bi_end_io = bio_map_kern_endio; | |
1530 | return bio; | |
1531 | } | |
1532 | EXPORT_SYMBOL(bio_map_kern); | |
1533 | ||
1534 | static void bio_copy_kern_endio(struct bio *bio) | |
1535 | { | |
1536 | bio_free_pages(bio); | |
1537 | bio_put(bio); | |
1538 | } | |
1539 | ||
1540 | static void bio_copy_kern_endio_read(struct bio *bio) | |
1541 | { | |
1542 | char *p = bio->bi_private; | |
1543 | struct bio_vec *bvec; | |
1544 | int i; | |
1545 | ||
1546 | bio_for_each_segment_all(bvec, bio, i) { | |
1547 | memcpy(p, page_address(bvec->bv_page), bvec->bv_len); | |
1548 | p += bvec->bv_len; | |
1549 | } | |
1550 | ||
1551 | bio_copy_kern_endio(bio); | |
1552 | } | |
1553 | ||
1554 | /** | |
1555 | * bio_copy_kern - copy kernel address into bio | |
1556 | * @q: the struct request_queue for the bio | |
1557 | * @data: pointer to buffer to copy | |
1558 | * @len: length in bytes | |
1559 | * @gfp_mask: allocation flags for bio and page allocation | |
1560 | * @reading: data direction is READ | |
1561 | * | |
1562 | * copy the kernel address into a bio suitable for io to a block | |
1563 | * device. Returns an error pointer in case of error. | |
1564 | */ | |
1565 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, | |
1566 | gfp_t gfp_mask, int reading) | |
1567 | { | |
1568 | unsigned long kaddr = (unsigned long)data; | |
1569 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1570 | unsigned long start = kaddr >> PAGE_SHIFT; | |
1571 | struct bio *bio; | |
1572 | void *p = data; | |
1573 | int nr_pages = 0; | |
1574 | ||
1575 | /* | |
1576 | * Overflow, abort | |
1577 | */ | |
1578 | if (end < start) | |
1579 | return ERR_PTR(-EINVAL); | |
1580 | ||
1581 | nr_pages = end - start; | |
1582 | bio = bio_kmalloc(gfp_mask, nr_pages); | |
1583 | if (!bio) | |
1584 | return ERR_PTR(-ENOMEM); | |
1585 | ||
1586 | while (len) { | |
1587 | struct page *page; | |
1588 | unsigned int bytes = PAGE_SIZE; | |
1589 | ||
1590 | if (bytes > len) | |
1591 | bytes = len; | |
1592 | ||
1593 | page = alloc_page(q->bounce_gfp | gfp_mask); | |
1594 | if (!page) | |
1595 | goto cleanup; | |
1596 | ||
1597 | if (!reading) | |
1598 | memcpy(page_address(page), p, bytes); | |
1599 | ||
1600 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) | |
1601 | break; | |
1602 | ||
1603 | len -= bytes; | |
1604 | p += bytes; | |
1605 | } | |
1606 | ||
1607 | if (reading) { | |
1608 | bio->bi_end_io = bio_copy_kern_endio_read; | |
1609 | bio->bi_private = data; | |
1610 | } else { | |
1611 | bio->bi_end_io = bio_copy_kern_endio; | |
1612 | } | |
1613 | ||
1614 | return bio; | |
1615 | ||
1616 | cleanup: | |
1617 | bio_free_pages(bio); | |
1618 | bio_put(bio); | |
1619 | return ERR_PTR(-ENOMEM); | |
1620 | } | |
1621 | ||
1622 | /* | |
1623 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
1624 | * for performing direct-IO in BIOs. | |
1625 | * | |
1626 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
1627 | * because the required locks are not interrupt-safe. So what we can do is to | |
1628 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
1629 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
1630 | * in process context. | |
1631 | * | |
1632 | * We special-case compound pages here: normally this means reads into hugetlb | |
1633 | * pages. The logic in here doesn't really work right for compound pages | |
1634 | * because the VM does not uniformly chase down the head page in all cases. | |
1635 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
1636 | * handle them at all. So we skip compound pages here at an early stage. | |
1637 | * | |
1638 | * Note that this code is very hard to test under normal circumstances because | |
1639 | * direct-io pins the pages with get_user_pages(). This makes | |
1640 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
1641 | * But other code (eg, flusher threads) could clean the pages if they are mapped | |
1642 | * pagecache. | |
1643 | * | |
1644 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
1645 | * deferred bio dirtying paths. | |
1646 | */ | |
1647 | ||
1648 | /* | |
1649 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
1650 | */ | |
1651 | void bio_set_pages_dirty(struct bio *bio) | |
1652 | { | |
1653 | struct bio_vec *bvec; | |
1654 | int i; | |
1655 | ||
1656 | bio_for_each_segment_all(bvec, bio, i) { | |
1657 | struct page *page = bvec->bv_page; | |
1658 | ||
1659 | if (page && !PageCompound(page)) | |
1660 | set_page_dirty_lock(page); | |
1661 | } | |
1662 | } | |
1663 | ||
1664 | static void bio_release_pages(struct bio *bio) | |
1665 | { | |
1666 | struct bio_vec *bvec; | |
1667 | int i; | |
1668 | ||
1669 | bio_for_each_segment_all(bvec, bio, i) { | |
1670 | struct page *page = bvec->bv_page; | |
1671 | ||
1672 | if (page) | |
1673 | put_page(page); | |
1674 | } | |
1675 | } | |
1676 | ||
1677 | /* | |
1678 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
1679 | * If they are, then fine. If, however, some pages are clean then they must | |
1680 | * have been written out during the direct-IO read. So we take another ref on | |
1681 | * the BIO and the offending pages and re-dirty the pages in process context. | |
1682 | * | |
1683 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
1684 | * here on. It will run one put_page() against each page and will run one | |
1685 | * bio_put() against the BIO. | |
1686 | */ | |
1687 | ||
1688 | static void bio_dirty_fn(struct work_struct *work); | |
1689 | ||
1690 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); | |
1691 | static DEFINE_SPINLOCK(bio_dirty_lock); | |
1692 | static struct bio *bio_dirty_list; | |
1693 | ||
1694 | /* | |
1695 | * This runs in process context | |
1696 | */ | |
1697 | static void bio_dirty_fn(struct work_struct *work) | |
1698 | { | |
1699 | unsigned long flags; | |
1700 | struct bio *bio; | |
1701 | ||
1702 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1703 | bio = bio_dirty_list; | |
1704 | bio_dirty_list = NULL; | |
1705 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1706 | ||
1707 | while (bio) { | |
1708 | struct bio *next = bio->bi_private; | |
1709 | ||
1710 | bio_set_pages_dirty(bio); | |
1711 | bio_release_pages(bio); | |
1712 | bio_put(bio); | |
1713 | bio = next; | |
1714 | } | |
1715 | } | |
1716 | ||
1717 | void bio_check_pages_dirty(struct bio *bio) | |
1718 | { | |
1719 | struct bio_vec *bvec; | |
1720 | int nr_clean_pages = 0; | |
1721 | int i; | |
1722 | ||
1723 | bio_for_each_segment_all(bvec, bio, i) { | |
1724 | struct page *page = bvec->bv_page; | |
1725 | ||
1726 | if (PageDirty(page) || PageCompound(page)) { | |
1727 | put_page(page); | |
1728 | bvec->bv_page = NULL; | |
1729 | } else { | |
1730 | nr_clean_pages++; | |
1731 | } | |
1732 | } | |
1733 | ||
1734 | if (nr_clean_pages) { | |
1735 | unsigned long flags; | |
1736 | ||
1737 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
1738 | bio->bi_private = bio_dirty_list; | |
1739 | bio_dirty_list = bio; | |
1740 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
1741 | schedule_work(&bio_dirty_work); | |
1742 | } else { | |
1743 | bio_put(bio); | |
1744 | } | |
1745 | } | |
1746 | ||
1747 | void generic_start_io_acct(int rw, unsigned long sectors, | |
1748 | struct hd_struct *part) | |
1749 | { | |
1750 | int cpu = part_stat_lock(); | |
1751 | ||
1752 | part_round_stats(cpu, part); | |
1753 | part_stat_inc(cpu, part, ios[rw]); | |
1754 | part_stat_add(cpu, part, sectors[rw], sectors); | |
1755 | part_inc_in_flight(part, rw); | |
1756 | ||
1757 | part_stat_unlock(); | |
1758 | } | |
1759 | EXPORT_SYMBOL(generic_start_io_acct); | |
1760 | ||
1761 | void generic_end_io_acct(int rw, struct hd_struct *part, | |
1762 | unsigned long start_time) | |
1763 | { | |
1764 | unsigned long duration = jiffies - start_time; | |
1765 | int cpu = part_stat_lock(); | |
1766 | ||
1767 | part_stat_add(cpu, part, ticks[rw], duration); | |
1768 | part_round_stats(cpu, part); | |
1769 | part_dec_in_flight(part, rw); | |
1770 | ||
1771 | part_stat_unlock(); | |
1772 | } | |
1773 | EXPORT_SYMBOL(generic_end_io_acct); | |
1774 | ||
1775 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE | |
1776 | void bio_flush_dcache_pages(struct bio *bi) | |
1777 | { | |
1778 | struct bio_vec bvec; | |
1779 | struct bvec_iter iter; | |
1780 | ||
1781 | bio_for_each_segment(bvec, bi, iter) | |
1782 | flush_dcache_page(bvec.bv_page); | |
1783 | } | |
1784 | EXPORT_SYMBOL(bio_flush_dcache_pages); | |
1785 | #endif | |
1786 | ||
1787 | static inline bool bio_remaining_done(struct bio *bio) | |
1788 | { | |
1789 | /* | |
1790 | * If we're not chaining, then ->__bi_remaining is always 1 and | |
1791 | * we always end io on the first invocation. | |
1792 | */ | |
1793 | if (!bio_flagged(bio, BIO_CHAIN)) | |
1794 | return true; | |
1795 | ||
1796 | BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); | |
1797 | ||
1798 | if (atomic_dec_and_test(&bio->__bi_remaining)) { | |
1799 | bio_clear_flag(bio, BIO_CHAIN); | |
1800 | return true; | |
1801 | } | |
1802 | ||
1803 | return false; | |
1804 | } | |
1805 | ||
1806 | /** | |
1807 | * bio_endio - end I/O on a bio | |
1808 | * @bio: bio | |
1809 | * | |
1810 | * Description: | |
1811 | * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred | |
1812 | * way to end I/O on a bio. No one should call bi_end_io() directly on a | |
1813 | * bio unless they own it and thus know that it has an end_io function. | |
1814 | * | |
1815 | * bio_endio() can be called several times on a bio that has been chained | |
1816 | * using bio_chain(). The ->bi_end_io() function will only be called the | |
1817 | * last time. At this point the BLK_TA_COMPLETE tracing event will be | |
1818 | * generated if BIO_TRACE_COMPLETION is set. | |
1819 | **/ | |
1820 | void bio_endio(struct bio *bio) | |
1821 | { | |
1822 | again: | |
1823 | if (!bio_remaining_done(bio)) | |
1824 | return; | |
1825 | if (!bio_integrity_endio(bio)) | |
1826 | return; | |
1827 | ||
1828 | /* | |
1829 | * Need to have a real endio function for chained bios, otherwise | |
1830 | * various corner cases will break (like stacking block devices that | |
1831 | * save/restore bi_end_io) - however, we want to avoid unbounded | |
1832 | * recursion and blowing the stack. Tail call optimization would | |
1833 | * handle this, but compiling with frame pointers also disables | |
1834 | * gcc's sibling call optimization. | |
1835 | */ | |
1836 | if (bio->bi_end_io == bio_chain_endio) { | |
1837 | bio = __bio_chain_endio(bio); | |
1838 | goto again; | |
1839 | } | |
1840 | ||
1841 | if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { | |
1842 | trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio, | |
1843 | blk_status_to_errno(bio->bi_status)); | |
1844 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); | |
1845 | } | |
1846 | ||
1847 | blk_throtl_bio_endio(bio); | |
1848 | /* release cgroup info */ | |
1849 | bio_uninit(bio); | |
1850 | if (bio->bi_end_io) | |
1851 | bio->bi_end_io(bio); | |
1852 | } | |
1853 | EXPORT_SYMBOL(bio_endio); | |
1854 | ||
1855 | /** | |
1856 | * bio_split - split a bio | |
1857 | * @bio: bio to split | |
1858 | * @sectors: number of sectors to split from the front of @bio | |
1859 | * @gfp: gfp mask | |
1860 | * @bs: bio set to allocate from | |
1861 | * | |
1862 | * Allocates and returns a new bio which represents @sectors from the start of | |
1863 | * @bio, and updates @bio to represent the remaining sectors. | |
1864 | * | |
1865 | * Unless this is a discard request the newly allocated bio will point | |
1866 | * to @bio's bi_io_vec; it is the caller's responsibility to ensure that | |
1867 | * @bio is not freed before the split. | |
1868 | */ | |
1869 | struct bio *bio_split(struct bio *bio, int sectors, | |
1870 | gfp_t gfp, struct bio_set *bs) | |
1871 | { | |
1872 | struct bio *split = NULL; | |
1873 | ||
1874 | BUG_ON(sectors <= 0); | |
1875 | BUG_ON(sectors >= bio_sectors(bio)); | |
1876 | ||
1877 | split = bio_clone_fast(bio, gfp, bs); | |
1878 | if (!split) | |
1879 | return NULL; | |
1880 | ||
1881 | split->bi_iter.bi_size = sectors << 9; | |
1882 | ||
1883 | if (bio_integrity(split)) | |
1884 | bio_integrity_trim(split); | |
1885 | ||
1886 | bio_advance(bio, split->bi_iter.bi_size); | |
1887 | ||
1888 | if (bio_flagged(bio, BIO_TRACE_COMPLETION)) | |
1889 | bio_set_flag(bio, BIO_TRACE_COMPLETION); | |
1890 | ||
1891 | return split; | |
1892 | } | |
1893 | EXPORT_SYMBOL(bio_split); | |
1894 | ||
1895 | /** | |
1896 | * bio_trim - trim a bio | |
1897 | * @bio: bio to trim | |
1898 | * @offset: number of sectors to trim from the front of @bio | |
1899 | * @size: size we want to trim @bio to, in sectors | |
1900 | */ | |
1901 | void bio_trim(struct bio *bio, int offset, int size) | |
1902 | { | |
1903 | /* 'bio' is a cloned bio which we need to trim to match | |
1904 | * the given offset and size. | |
1905 | */ | |
1906 | ||
1907 | size <<= 9; | |
1908 | if (offset == 0 && size == bio->bi_iter.bi_size) | |
1909 | return; | |
1910 | ||
1911 | bio_clear_flag(bio, BIO_SEG_VALID); | |
1912 | ||
1913 | bio_advance(bio, offset << 9); | |
1914 | ||
1915 | bio->bi_iter.bi_size = size; | |
1916 | ||
1917 | if (bio_integrity(bio)) | |
1918 | bio_integrity_trim(bio); | |
1919 | ||
1920 | } | |
1921 | EXPORT_SYMBOL_GPL(bio_trim); | |
1922 | ||
1923 | /* | |
1924 | * create memory pools for biovec's in a bio_set. | |
1925 | * use the global biovec slabs created for general use. | |
1926 | */ | |
1927 | mempool_t *biovec_create_pool(int pool_entries) | |
1928 | { | |
1929 | struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; | |
1930 | ||
1931 | return mempool_create_slab_pool(pool_entries, bp->slab); | |
1932 | } | |
1933 | ||
1934 | void bioset_free(struct bio_set *bs) | |
1935 | { | |
1936 | if (bs->rescue_workqueue) | |
1937 | destroy_workqueue(bs->rescue_workqueue); | |
1938 | ||
1939 | if (bs->bio_pool) | |
1940 | mempool_destroy(bs->bio_pool); | |
1941 | ||
1942 | if (bs->bvec_pool) | |
1943 | mempool_destroy(bs->bvec_pool); | |
1944 | ||
1945 | bioset_integrity_free(bs); | |
1946 | bio_put_slab(bs); | |
1947 | ||
1948 | kfree(bs); | |
1949 | } | |
1950 | EXPORT_SYMBOL(bioset_free); | |
1951 | ||
1952 | /** | |
1953 | * bioset_create - Create a bio_set | |
1954 | * @pool_size: Number of bio and bio_vecs to cache in the mempool | |
1955 | * @front_pad: Number of bytes to allocate in front of the returned bio | |
1956 | * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS | |
1957 | * and %BIOSET_NEED_RESCUER | |
1958 | * | |
1959 | * Description: | |
1960 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller | |
1961 | * to ask for a number of bytes to be allocated in front of the bio. | |
1962 | * Front pad allocation is useful for embedding the bio inside | |
1963 | * another structure, to avoid allocating extra data to go with the bio. | |
1964 | * Note that the bio must be embedded at the END of that structure always, | |
1965 | * or things will break badly. | |
1966 | * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated | |
1967 | * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). | |
1968 | * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to | |
1969 | * dispatch queued requests when the mempool runs out of space. | |
1970 | * | |
1971 | */ | |
1972 | struct bio_set *bioset_create(unsigned int pool_size, | |
1973 | unsigned int front_pad, | |
1974 | int flags) | |
1975 | { | |
1976 | unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); | |
1977 | struct bio_set *bs; | |
1978 | ||
1979 | bs = kzalloc(sizeof(*bs), GFP_KERNEL); | |
1980 | if (!bs) | |
1981 | return NULL; | |
1982 | ||
1983 | bs->front_pad = front_pad; | |
1984 | ||
1985 | spin_lock_init(&bs->rescue_lock); | |
1986 | bio_list_init(&bs->rescue_list); | |
1987 | INIT_WORK(&bs->rescue_work, bio_alloc_rescue); | |
1988 | ||
1989 | bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); | |
1990 | if (!bs->bio_slab) { | |
1991 | kfree(bs); | |
1992 | return NULL; | |
1993 | } | |
1994 | ||
1995 | bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab); | |
1996 | if (!bs->bio_pool) | |
1997 | goto bad; | |
1998 | ||
1999 | if (flags & BIOSET_NEED_BVECS) { | |
2000 | bs->bvec_pool = biovec_create_pool(pool_size); | |
2001 | if (!bs->bvec_pool) | |
2002 | goto bad; | |
2003 | } | |
2004 | ||
2005 | if (!(flags & BIOSET_NEED_RESCUER)) | |
2006 | return bs; | |
2007 | ||
2008 | bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); | |
2009 | if (!bs->rescue_workqueue) | |
2010 | goto bad; | |
2011 | ||
2012 | return bs; | |
2013 | bad: | |
2014 | bioset_free(bs); | |
2015 | return NULL; | |
2016 | } | |
2017 | EXPORT_SYMBOL(bioset_create); | |
2018 | ||
2019 | #ifdef CONFIG_BLK_CGROUP | |
2020 | ||
2021 | /** | |
2022 | * bio_associate_blkcg - associate a bio with the specified blkcg | |
2023 | * @bio: target bio | |
2024 | * @blkcg_css: css of the blkcg to associate | |
2025 | * | |
2026 | * Associate @bio with the blkcg specified by @blkcg_css. Block layer will | |
2027 | * treat @bio as if it were issued by a task which belongs to the blkcg. | |
2028 | * | |
2029 | * This function takes an extra reference of @blkcg_css which will be put | |
2030 | * when @bio is released. The caller must own @bio and is responsible for | |
2031 | * synchronizing calls to this function. | |
2032 | */ | |
2033 | int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css) | |
2034 | { | |
2035 | if (unlikely(bio->bi_css)) | |
2036 | return -EBUSY; | |
2037 | css_get(blkcg_css); | |
2038 | bio->bi_css = blkcg_css; | |
2039 | return 0; | |
2040 | } | |
2041 | EXPORT_SYMBOL_GPL(bio_associate_blkcg); | |
2042 | ||
2043 | /** | |
2044 | * bio_associate_current - associate a bio with %current | |
2045 | * @bio: target bio | |
2046 | * | |
2047 | * Associate @bio with %current if it hasn't been associated yet. Block | |
2048 | * layer will treat @bio as if it were issued by %current no matter which | |
2049 | * task actually issues it. | |
2050 | * | |
2051 | * This function takes an extra reference of @task's io_context and blkcg | |
2052 | * which will be put when @bio is released. The caller must own @bio, | |
2053 | * ensure %current->io_context exists, and is responsible for synchronizing | |
2054 | * calls to this function. | |
2055 | */ | |
2056 | int bio_associate_current(struct bio *bio) | |
2057 | { | |
2058 | struct io_context *ioc; | |
2059 | ||
2060 | if (bio->bi_css) | |
2061 | return -EBUSY; | |
2062 | ||
2063 | ioc = current->io_context; | |
2064 | if (!ioc) | |
2065 | return -ENOENT; | |
2066 | ||
2067 | get_io_context_active(ioc); | |
2068 | bio->bi_ioc = ioc; | |
2069 | bio->bi_css = task_get_css(current, io_cgrp_id); | |
2070 | return 0; | |
2071 | } | |
2072 | EXPORT_SYMBOL_GPL(bio_associate_current); | |
2073 | ||
2074 | /** | |
2075 | * bio_disassociate_task - undo bio_associate_current() | |
2076 | * @bio: target bio | |
2077 | */ | |
2078 | void bio_disassociate_task(struct bio *bio) | |
2079 | { | |
2080 | if (bio->bi_ioc) { | |
2081 | put_io_context(bio->bi_ioc); | |
2082 | bio->bi_ioc = NULL; | |
2083 | } | |
2084 | if (bio->bi_css) { | |
2085 | css_put(bio->bi_css); | |
2086 | bio->bi_css = NULL; | |
2087 | } | |
2088 | } | |
2089 | ||
2090 | /** | |
2091 | * bio_clone_blkcg_association - clone blkcg association from src to dst bio | |
2092 | * @dst: destination bio | |
2093 | * @src: source bio | |
2094 | */ | |
2095 | void bio_clone_blkcg_association(struct bio *dst, struct bio *src) | |
2096 | { | |
2097 | if (src->bi_css) | |
2098 | WARN_ON(bio_associate_blkcg(dst, src->bi_css)); | |
2099 | } | |
2100 | ||
2101 | #endif /* CONFIG_BLK_CGROUP */ | |
2102 | ||
2103 | static void __init biovec_init_slabs(void) | |
2104 | { | |
2105 | int i; | |
2106 | ||
2107 | for (i = 0; i < BVEC_POOL_NR; i++) { | |
2108 | int size; | |
2109 | struct biovec_slab *bvs = bvec_slabs + i; | |
2110 | ||
2111 | if (bvs->nr_vecs <= BIO_INLINE_VECS) { | |
2112 | bvs->slab = NULL; | |
2113 | continue; | |
2114 | } | |
2115 | ||
2116 | size = bvs->nr_vecs * sizeof(struct bio_vec); | |
2117 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
2118 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); | |
2119 | } | |
2120 | } | |
2121 | ||
2122 | static int __init init_bio(void) | |
2123 | { | |
2124 | bio_slab_max = 2; | |
2125 | bio_slab_nr = 0; | |
2126 | bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL); | |
2127 | if (!bio_slabs) | |
2128 | panic("bio: can't allocate bios\n"); | |
2129 | ||
2130 | bio_integrity_init(); | |
2131 | biovec_init_slabs(); | |
2132 | ||
2133 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); | |
2134 | if (!fs_bio_set) | |
2135 | panic("bio: can't allocate bios\n"); | |
2136 | ||
2137 | if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE)) | |
2138 | panic("bio: can't create integrity pool\n"); | |
2139 | ||
2140 | return 0; | |
2141 | } | |
2142 | subsys_initcall(init_bio); |