Root/
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/iocontext.h> |
23 | #include <linux/slab.h> |
24 | #include <linux/init.h> |
25 | #include <linux/kernel.h> |
26 | #include <linux/export.h> |
27 | #include <linux/mempool.h> |
28 | #include <linux/workqueue.h> |
29 | #include <linux/cgroup.h> |
30 | #include <scsi/sg.h> /* for struct sg_iovec */ |
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 | static mempool_t *bio_split_pool __read_mostly; |
41 | |
42 | /* |
43 | * if you change this list, also change bvec_alloc or things will |
44 | * break badly! cannot be bigger than what you can fit into an |
45 | * unsigned short |
46 | */ |
47 | #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } |
48 | static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { |
49 | BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), |
50 | }; |
51 | #undef BV |
52 | |
53 | /* |
54 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by |
55 | * IO code that does not need private memory pools. |
56 | */ |
57 | struct bio_set *fs_bio_set; |
58 | EXPORT_SYMBOL(fs_bio_set); |
59 | |
60 | /* |
61 | * Our slab pool management |
62 | */ |
63 | struct bio_slab { |
64 | struct kmem_cache *slab; |
65 | unsigned int slab_ref; |
66 | unsigned int slab_size; |
67 | char name[8]; |
68 | }; |
69 | static DEFINE_MUTEX(bio_slab_lock); |
70 | static struct bio_slab *bio_slabs; |
71 | static unsigned int bio_slab_nr, bio_slab_max; |
72 | |
73 | static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) |
74 | { |
75 | unsigned int sz = sizeof(struct bio) + extra_size; |
76 | struct kmem_cache *slab = NULL; |
77 | struct bio_slab *bslab, *new_bio_slabs; |
78 | unsigned int new_bio_slab_max; |
79 | unsigned int i, entry = -1; |
80 | |
81 | mutex_lock(&bio_slab_lock); |
82 | |
83 | i = 0; |
84 | while (i < bio_slab_nr) { |
85 | bslab = &bio_slabs[i]; |
86 | |
87 | if (!bslab->slab && entry == -1) |
88 | entry = i; |
89 | else if (bslab->slab_size == sz) { |
90 | slab = bslab->slab; |
91 | bslab->slab_ref++; |
92 | break; |
93 | } |
94 | i++; |
95 | } |
96 | |
97 | if (slab) |
98 | goto out_unlock; |
99 | |
100 | if (bio_slab_nr == bio_slab_max && entry == -1) { |
101 | new_bio_slab_max = bio_slab_max << 1; |
102 | new_bio_slabs = krealloc(bio_slabs, |
103 | new_bio_slab_max * sizeof(struct bio_slab), |
104 | GFP_KERNEL); |
105 | if (!new_bio_slabs) |
106 | goto out_unlock; |
107 | bio_slab_max = new_bio_slab_max; |
108 | bio_slabs = new_bio_slabs; |
109 | } |
110 | if (entry == -1) |
111 | entry = bio_slab_nr++; |
112 | |
113 | bslab = &bio_slabs[entry]; |
114 | |
115 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); |
116 | slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL); |
117 | if (!slab) |
118 | goto out_unlock; |
119 | |
120 | printk(KERN_INFO "bio: create slab <%s> at %d\n", bslab->name, entry); |
121 | bslab->slab = slab; |
122 | bslab->slab_ref = 1; |
123 | bslab->slab_size = sz; |
124 | out_unlock: |
125 | mutex_unlock(&bio_slab_lock); |
126 | return slab; |
127 | } |
128 | |
129 | static void bio_put_slab(struct bio_set *bs) |
130 | { |
131 | struct bio_slab *bslab = NULL; |
132 | unsigned int i; |
133 | |
134 | mutex_lock(&bio_slab_lock); |
135 | |
136 | for (i = 0; i < bio_slab_nr; i++) { |
137 | if (bs->bio_slab == bio_slabs[i].slab) { |
138 | bslab = &bio_slabs[i]; |
139 | break; |
140 | } |
141 | } |
142 | |
143 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) |
144 | goto out; |
145 | |
146 | WARN_ON(!bslab->slab_ref); |
147 | |
148 | if (--bslab->slab_ref) |
149 | goto out; |
150 | |
151 | kmem_cache_destroy(bslab->slab); |
152 | bslab->slab = NULL; |
153 | |
154 | out: |
155 | mutex_unlock(&bio_slab_lock); |
156 | } |
157 | |
158 | unsigned int bvec_nr_vecs(unsigned short idx) |
159 | { |
160 | return bvec_slabs[idx].nr_vecs; |
161 | } |
162 | |
163 | void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx) |
164 | { |
165 | BIO_BUG_ON(idx >= BIOVEC_NR_POOLS); |
166 | |
167 | if (idx == BIOVEC_MAX_IDX) |
168 | mempool_free(bv, bs->bvec_pool); |
169 | else { |
170 | struct biovec_slab *bvs = bvec_slabs + idx; |
171 | |
172 | kmem_cache_free(bvs->slab, bv); |
173 | } |
174 | } |
175 | |
176 | struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, |
177 | struct bio_set *bs) |
178 | { |
179 | struct bio_vec *bvl; |
180 | |
181 | /* |
182 | * see comment near bvec_array define! |
183 | */ |
184 | switch (nr) { |
185 | case 1: |
186 | *idx = 0; |
187 | break; |
188 | case 2 ... 4: |
189 | *idx = 1; |
190 | break; |
191 | case 5 ... 16: |
192 | *idx = 2; |
193 | break; |
194 | case 17 ... 64: |
195 | *idx = 3; |
196 | break; |
197 | case 65 ... 128: |
198 | *idx = 4; |
199 | break; |
200 | case 129 ... BIO_MAX_PAGES: |
201 | *idx = 5; |
202 | break; |
203 | default: |
204 | return NULL; |
205 | } |
206 | |
207 | /* |
208 | * idx now points to the pool we want to allocate from. only the |
209 | * 1-vec entry pool is mempool backed. |
210 | */ |
211 | if (*idx == BIOVEC_MAX_IDX) { |
212 | fallback: |
213 | bvl = mempool_alloc(bs->bvec_pool, gfp_mask); |
214 | } else { |
215 | struct biovec_slab *bvs = bvec_slabs + *idx; |
216 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO); |
217 | |
218 | /* |
219 | * Make this allocation restricted and don't dump info on |
220 | * allocation failures, since we'll fallback to the mempool |
221 | * in case of failure. |
222 | */ |
223 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; |
224 | |
225 | /* |
226 | * Try a slab allocation. If this fails and __GFP_WAIT |
227 | * is set, retry with the 1-entry mempool |
228 | */ |
229 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); |
230 | if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) { |
231 | *idx = BIOVEC_MAX_IDX; |
232 | goto fallback; |
233 | } |
234 | } |
235 | |
236 | return bvl; |
237 | } |
238 | |
239 | static void __bio_free(struct bio *bio) |
240 | { |
241 | bio_disassociate_task(bio); |
242 | |
243 | if (bio_integrity(bio)) |
244 | bio_integrity_free(bio); |
245 | } |
246 | |
247 | static void bio_free(struct bio *bio) |
248 | { |
249 | struct bio_set *bs = bio->bi_pool; |
250 | void *p; |
251 | |
252 | __bio_free(bio); |
253 | |
254 | if (bs) { |
255 | if (bio_has_allocated_vec(bio)) |
256 | bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio)); |
257 | |
258 | /* |
259 | * If we have front padding, adjust the bio pointer before freeing |
260 | */ |
261 | p = bio; |
262 | p -= bs->front_pad; |
263 | |
264 | mempool_free(p, bs->bio_pool); |
265 | } else { |
266 | /* Bio was allocated by bio_kmalloc() */ |
267 | kfree(bio); |
268 | } |
269 | } |
270 | |
271 | void bio_init(struct bio *bio) |
272 | { |
273 | memset(bio, 0, sizeof(*bio)); |
274 | bio->bi_flags = 1 << BIO_UPTODATE; |
275 | atomic_set(&bio->bi_cnt, 1); |
276 | } |
277 | EXPORT_SYMBOL(bio_init); |
278 | |
279 | /** |
280 | * bio_reset - reinitialize a bio |
281 | * @bio: bio to reset |
282 | * |
283 | * Description: |
284 | * After calling bio_reset(), @bio will be in the same state as a freshly |
285 | * allocated bio returned bio bio_alloc_bioset() - the only fields that are |
286 | * preserved are the ones that are initialized by bio_alloc_bioset(). See |
287 | * comment in struct bio. |
288 | */ |
289 | void bio_reset(struct bio *bio) |
290 | { |
291 | unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); |
292 | |
293 | __bio_free(bio); |
294 | |
295 | memset(bio, 0, BIO_RESET_BYTES); |
296 | bio->bi_flags = flags|(1 << BIO_UPTODATE); |
297 | } |
298 | EXPORT_SYMBOL(bio_reset); |
299 | |
300 | /** |
301 | * bio_alloc_bioset - allocate a bio for I/O |
302 | * @gfp_mask: the GFP_ mask given to the slab allocator |
303 | * @nr_iovecs: number of iovecs to pre-allocate |
304 | * @bs: the bio_set to allocate from. |
305 | * |
306 | * Description: |
307 | * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is |
308 | * backed by the @bs's mempool. |
309 | * |
310 | * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be |
311 | * able to allocate a bio. This is due to the mempool guarantees. To make this |
312 | * work, callers must never allocate more than 1 bio at a time from this pool. |
313 | * Callers that need to allocate more than 1 bio must always submit the |
314 | * previously allocated bio for IO before attempting to allocate a new one. |
315 | * Failure to do so can cause deadlocks under memory pressure. |
316 | * |
317 | * RETURNS: |
318 | * Pointer to new bio on success, NULL on failure. |
319 | */ |
320 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) |
321 | { |
322 | unsigned front_pad; |
323 | unsigned inline_vecs; |
324 | unsigned long idx = BIO_POOL_NONE; |
325 | struct bio_vec *bvl = NULL; |
326 | struct bio *bio; |
327 | void *p; |
328 | |
329 | if (!bs) { |
330 | if (nr_iovecs > UIO_MAXIOV) |
331 | return NULL; |
332 | |
333 | p = kmalloc(sizeof(struct bio) + |
334 | nr_iovecs * sizeof(struct bio_vec), |
335 | gfp_mask); |
336 | front_pad = 0; |
337 | inline_vecs = nr_iovecs; |
338 | } else { |
339 | p = mempool_alloc(bs->bio_pool, gfp_mask); |
340 | front_pad = bs->front_pad; |
341 | inline_vecs = BIO_INLINE_VECS; |
342 | } |
343 | |
344 | if (unlikely(!p)) |
345 | return NULL; |
346 | |
347 | bio = p + front_pad; |
348 | bio_init(bio); |
349 | |
350 | if (nr_iovecs > inline_vecs) { |
351 | bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs); |
352 | if (unlikely(!bvl)) |
353 | goto err_free; |
354 | } else if (nr_iovecs) { |
355 | bvl = bio->bi_inline_vecs; |
356 | } |
357 | |
358 | bio->bi_pool = bs; |
359 | bio->bi_flags |= idx << BIO_POOL_OFFSET; |
360 | bio->bi_max_vecs = nr_iovecs; |
361 | bio->bi_io_vec = bvl; |
362 | return bio; |
363 | |
364 | err_free: |
365 | mempool_free(p, bs->bio_pool); |
366 | return NULL; |
367 | } |
368 | EXPORT_SYMBOL(bio_alloc_bioset); |
369 | |
370 | void zero_fill_bio(struct bio *bio) |
371 | { |
372 | unsigned long flags; |
373 | struct bio_vec *bv; |
374 | int i; |
375 | |
376 | bio_for_each_segment(bv, bio, i) { |
377 | char *data = bvec_kmap_irq(bv, &flags); |
378 | memset(data, 0, bv->bv_len); |
379 | flush_dcache_page(bv->bv_page); |
380 | bvec_kunmap_irq(data, &flags); |
381 | } |
382 | } |
383 | EXPORT_SYMBOL(zero_fill_bio); |
384 | |
385 | /** |
386 | * bio_put - release a reference to a bio |
387 | * @bio: bio to release reference to |
388 | * |
389 | * Description: |
390 | * Put a reference to a &struct bio, either one you have gotten with |
391 | * bio_alloc, bio_get or bio_clone. The last put of a bio will free it. |
392 | **/ |
393 | void bio_put(struct bio *bio) |
394 | { |
395 | BIO_BUG_ON(!atomic_read(&bio->bi_cnt)); |
396 | |
397 | /* |
398 | * last put frees it |
399 | */ |
400 | if (atomic_dec_and_test(&bio->bi_cnt)) |
401 | bio_free(bio); |
402 | } |
403 | EXPORT_SYMBOL(bio_put); |
404 | |
405 | inline int bio_phys_segments(struct request_queue *q, struct bio *bio) |
406 | { |
407 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) |
408 | blk_recount_segments(q, bio); |
409 | |
410 | return bio->bi_phys_segments; |
411 | } |
412 | EXPORT_SYMBOL(bio_phys_segments); |
413 | |
414 | /** |
415 | * __bio_clone - clone a bio |
416 | * @bio: destination bio |
417 | * @bio_src: bio to clone |
418 | * |
419 | * Clone a &bio. Caller will own the returned bio, but not |
420 | * the actual data it points to. Reference count of returned |
421 | * bio will be one. |
422 | */ |
423 | void __bio_clone(struct bio *bio, struct bio *bio_src) |
424 | { |
425 | memcpy(bio->bi_io_vec, bio_src->bi_io_vec, |
426 | bio_src->bi_max_vecs * sizeof(struct bio_vec)); |
427 | |
428 | /* |
429 | * most users will be overriding ->bi_bdev with a new target, |
430 | * so we don't set nor calculate new physical/hw segment counts here |
431 | */ |
432 | bio->bi_sector = bio_src->bi_sector; |
433 | bio->bi_bdev = bio_src->bi_bdev; |
434 | bio->bi_flags |= 1 << BIO_CLONED; |
435 | bio->bi_rw = bio_src->bi_rw; |
436 | bio->bi_vcnt = bio_src->bi_vcnt; |
437 | bio->bi_size = bio_src->bi_size; |
438 | bio->bi_idx = bio_src->bi_idx; |
439 | } |
440 | EXPORT_SYMBOL(__bio_clone); |
441 | |
442 | /** |
443 | * bio_clone_bioset - clone a bio |
444 | * @bio: bio to clone |
445 | * @gfp_mask: allocation priority |
446 | * @bs: bio_set to allocate from |
447 | * |
448 | * Like __bio_clone, only also allocates the returned bio |
449 | */ |
450 | struct bio *bio_clone_bioset(struct bio *bio, gfp_t gfp_mask, |
451 | struct bio_set *bs) |
452 | { |
453 | struct bio *b; |
454 | |
455 | b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, bs); |
456 | if (!b) |
457 | return NULL; |
458 | |
459 | __bio_clone(b, bio); |
460 | |
461 | if (bio_integrity(bio)) { |
462 | int ret; |
463 | |
464 | ret = bio_integrity_clone(b, bio, gfp_mask); |
465 | |
466 | if (ret < 0) { |
467 | bio_put(b); |
468 | return NULL; |
469 | } |
470 | } |
471 | |
472 | return b; |
473 | } |
474 | EXPORT_SYMBOL(bio_clone_bioset); |
475 | |
476 | /** |
477 | * bio_get_nr_vecs - return approx number of vecs |
478 | * @bdev: I/O target |
479 | * |
480 | * Return the approximate number of pages we can send to this target. |
481 | * There's no guarantee that you will be able to fit this number of pages |
482 | * into a bio, it does not account for dynamic restrictions that vary |
483 | * on offset. |
484 | */ |
485 | int bio_get_nr_vecs(struct block_device *bdev) |
486 | { |
487 | struct request_queue *q = bdev_get_queue(bdev); |
488 | int nr_pages; |
489 | |
490 | nr_pages = min_t(unsigned, |
491 | queue_max_segments(q), |
492 | queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1); |
493 | |
494 | return min_t(unsigned, nr_pages, BIO_MAX_PAGES); |
495 | |
496 | } |
497 | EXPORT_SYMBOL(bio_get_nr_vecs); |
498 | |
499 | static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page |
500 | *page, unsigned int len, unsigned int offset, |
501 | unsigned short max_sectors) |
502 | { |
503 | int retried_segments = 0; |
504 | struct bio_vec *bvec; |
505 | |
506 | /* |
507 | * cloned bio must not modify vec list |
508 | */ |
509 | if (unlikely(bio_flagged(bio, BIO_CLONED))) |
510 | return 0; |
511 | |
512 | if (((bio->bi_size + len) >> 9) > max_sectors) |
513 | return 0; |
514 | |
515 | /* |
516 | * For filesystems with a blocksize smaller than the pagesize |
517 | * we will often be called with the same page as last time and |
518 | * a consecutive offset. Optimize this special case. |
519 | */ |
520 | if (bio->bi_vcnt > 0) { |
521 | struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
522 | |
523 | if (page == prev->bv_page && |
524 | offset == prev->bv_offset + prev->bv_len) { |
525 | unsigned int prev_bv_len = prev->bv_len; |
526 | prev->bv_len += len; |
527 | |
528 | if (q->merge_bvec_fn) { |
529 | struct bvec_merge_data bvm = { |
530 | /* prev_bvec is already charged in |
531 | bi_size, discharge it in order to |
532 | simulate merging updated prev_bvec |
533 | as new bvec. */ |
534 | .bi_bdev = bio->bi_bdev, |
535 | .bi_sector = bio->bi_sector, |
536 | .bi_size = bio->bi_size - prev_bv_len, |
537 | .bi_rw = bio->bi_rw, |
538 | }; |
539 | |
540 | if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) { |
541 | prev->bv_len -= len; |
542 | return 0; |
543 | } |
544 | } |
545 | |
546 | goto done; |
547 | } |
548 | } |
549 | |
550 | if (bio->bi_vcnt >= bio->bi_max_vecs) |
551 | return 0; |
552 | |
553 | /* |
554 | * we might lose a segment or two here, but rather that than |
555 | * make this too complex. |
556 | */ |
557 | |
558 | while (bio->bi_phys_segments >= queue_max_segments(q)) { |
559 | |
560 | if (retried_segments) |
561 | return 0; |
562 | |
563 | retried_segments = 1; |
564 | blk_recount_segments(q, bio); |
565 | } |
566 | |
567 | /* |
568 | * setup the new entry, we might clear it again later if we |
569 | * cannot add the page |
570 | */ |
571 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; |
572 | bvec->bv_page = page; |
573 | bvec->bv_len = len; |
574 | bvec->bv_offset = offset; |
575 | |
576 | /* |
577 | * if queue has other restrictions (eg varying max sector size |
578 | * depending on offset), it can specify a merge_bvec_fn in the |
579 | * queue to get further control |
580 | */ |
581 | if (q->merge_bvec_fn) { |
582 | struct bvec_merge_data bvm = { |
583 | .bi_bdev = bio->bi_bdev, |
584 | .bi_sector = bio->bi_sector, |
585 | .bi_size = bio->bi_size, |
586 | .bi_rw = bio->bi_rw, |
587 | }; |
588 | |
589 | /* |
590 | * merge_bvec_fn() returns number of bytes it can accept |
591 | * at this offset |
592 | */ |
593 | if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) { |
594 | bvec->bv_page = NULL; |
595 | bvec->bv_len = 0; |
596 | bvec->bv_offset = 0; |
597 | return 0; |
598 | } |
599 | } |
600 | |
601 | /* If we may be able to merge these biovecs, force a recount */ |
602 | if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) |
603 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); |
604 | |
605 | bio->bi_vcnt++; |
606 | bio->bi_phys_segments++; |
607 | done: |
608 | bio->bi_size += len; |
609 | return len; |
610 | } |
611 | |
612 | /** |
613 | * bio_add_pc_page - attempt to add page to bio |
614 | * @q: the target queue |
615 | * @bio: destination bio |
616 | * @page: page to add |
617 | * @len: vec entry length |
618 | * @offset: vec entry offset |
619 | * |
620 | * Attempt to add a page to the bio_vec maplist. This can fail for a |
621 | * number of reasons, such as the bio being full or target block device |
622 | * limitations. The target block device must allow bio's up to PAGE_SIZE, |
623 | * so it is always possible to add a single page to an empty bio. |
624 | * |
625 | * This should only be used by REQ_PC bios. |
626 | */ |
627 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page, |
628 | unsigned int len, unsigned int offset) |
629 | { |
630 | return __bio_add_page(q, bio, page, len, offset, |
631 | queue_max_hw_sectors(q)); |
632 | } |
633 | EXPORT_SYMBOL(bio_add_pc_page); |
634 | |
635 | /** |
636 | * bio_add_page - attempt to add page to bio |
637 | * @bio: destination bio |
638 | * @page: page to add |
639 | * @len: vec entry length |
640 | * @offset: vec entry offset |
641 | * |
642 | * Attempt to add a page to the bio_vec maplist. This can fail for a |
643 | * number of reasons, such as the bio being full or target block device |
644 | * limitations. The target block device must allow bio's up to PAGE_SIZE, |
645 | * so it is always possible to add a single page to an empty bio. |
646 | */ |
647 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, |
648 | unsigned int offset) |
649 | { |
650 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
651 | return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q)); |
652 | } |
653 | EXPORT_SYMBOL(bio_add_page); |
654 | |
655 | struct bio_map_data { |
656 | struct bio_vec *iovecs; |
657 | struct sg_iovec *sgvecs; |
658 | int nr_sgvecs; |
659 | int is_our_pages; |
660 | }; |
661 | |
662 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio, |
663 | struct sg_iovec *iov, int iov_count, |
664 | int is_our_pages) |
665 | { |
666 | memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); |
667 | memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count); |
668 | bmd->nr_sgvecs = iov_count; |
669 | bmd->is_our_pages = is_our_pages; |
670 | bio->bi_private = bmd; |
671 | } |
672 | |
673 | static void bio_free_map_data(struct bio_map_data *bmd) |
674 | { |
675 | kfree(bmd->iovecs); |
676 | kfree(bmd->sgvecs); |
677 | kfree(bmd); |
678 | } |
679 | |
680 | static struct bio_map_data *bio_alloc_map_data(int nr_segs, |
681 | unsigned int iov_count, |
682 | gfp_t gfp_mask) |
683 | { |
684 | struct bio_map_data *bmd; |
685 | |
686 | if (iov_count > UIO_MAXIOV) |
687 | return NULL; |
688 | |
689 | bmd = kmalloc(sizeof(*bmd), gfp_mask); |
690 | if (!bmd) |
691 | return NULL; |
692 | |
693 | bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask); |
694 | if (!bmd->iovecs) { |
695 | kfree(bmd); |
696 | return NULL; |
697 | } |
698 | |
699 | bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask); |
700 | if (bmd->sgvecs) |
701 | return bmd; |
702 | |
703 | kfree(bmd->iovecs); |
704 | kfree(bmd); |
705 | return NULL; |
706 | } |
707 | |
708 | static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs, |
709 | struct sg_iovec *iov, int iov_count, |
710 | int to_user, int from_user, int do_free_page) |
711 | { |
712 | int ret = 0, i; |
713 | struct bio_vec *bvec; |
714 | int iov_idx = 0; |
715 | unsigned int iov_off = 0; |
716 | |
717 | __bio_for_each_segment(bvec, bio, i, 0) { |
718 | char *bv_addr = page_address(bvec->bv_page); |
719 | unsigned int bv_len = iovecs[i].bv_len; |
720 | |
721 | while (bv_len && iov_idx < iov_count) { |
722 | unsigned int bytes; |
723 | char __user *iov_addr; |
724 | |
725 | bytes = min_t(unsigned int, |
726 | iov[iov_idx].iov_len - iov_off, bv_len); |
727 | iov_addr = iov[iov_idx].iov_base + iov_off; |
728 | |
729 | if (!ret) { |
730 | if (to_user) |
731 | ret = copy_to_user(iov_addr, bv_addr, |
732 | bytes); |
733 | |
734 | if (from_user) |
735 | ret = copy_from_user(bv_addr, iov_addr, |
736 | bytes); |
737 | |
738 | if (ret) |
739 | ret = -EFAULT; |
740 | } |
741 | |
742 | bv_len -= bytes; |
743 | bv_addr += bytes; |
744 | iov_addr += bytes; |
745 | iov_off += bytes; |
746 | |
747 | if (iov[iov_idx].iov_len == iov_off) { |
748 | iov_idx++; |
749 | iov_off = 0; |
750 | } |
751 | } |
752 | |
753 | if (do_free_page) |
754 | __free_page(bvec->bv_page); |
755 | } |
756 | |
757 | return ret; |
758 | } |
759 | |
760 | /** |
761 | * bio_uncopy_user - finish previously mapped bio |
762 | * @bio: bio being terminated |
763 | * |
764 | * Free pages allocated from bio_copy_user() and write back data |
765 | * to user space in case of a read. |
766 | */ |
767 | int bio_uncopy_user(struct bio *bio) |
768 | { |
769 | struct bio_map_data *bmd = bio->bi_private; |
770 | int ret = 0; |
771 | |
772 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) |
773 | ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs, |
774 | bmd->nr_sgvecs, bio_data_dir(bio) == READ, |
775 | 0, bmd->is_our_pages); |
776 | bio_free_map_data(bmd); |
777 | bio_put(bio); |
778 | return ret; |
779 | } |
780 | EXPORT_SYMBOL(bio_uncopy_user); |
781 | |
782 | /** |
783 | * bio_copy_user_iov - copy user data to bio |
784 | * @q: destination block queue |
785 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
786 | * @iov: the iovec. |
787 | * @iov_count: number of elements in the iovec |
788 | * @write_to_vm: bool indicating writing to pages or not |
789 | * @gfp_mask: memory allocation flags |
790 | * |
791 | * Prepares and returns a bio for indirect user io, bouncing data |
792 | * to/from kernel pages as necessary. Must be paired with |
793 | * call bio_uncopy_user() on io completion. |
794 | */ |
795 | struct bio *bio_copy_user_iov(struct request_queue *q, |
796 | struct rq_map_data *map_data, |
797 | struct sg_iovec *iov, int iov_count, |
798 | int write_to_vm, gfp_t gfp_mask) |
799 | { |
800 | struct bio_map_data *bmd; |
801 | struct bio_vec *bvec; |
802 | struct page *page; |
803 | struct bio *bio; |
804 | int i, ret; |
805 | int nr_pages = 0; |
806 | unsigned int len = 0; |
807 | unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0; |
808 | |
809 | for (i = 0; i < iov_count; i++) { |
810 | unsigned long uaddr; |
811 | unsigned long end; |
812 | unsigned long start; |
813 | |
814 | uaddr = (unsigned long)iov[i].iov_base; |
815 | end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
816 | start = uaddr >> PAGE_SHIFT; |
817 | |
818 | /* |
819 | * Overflow, abort |
820 | */ |
821 | if (end < start) |
822 | return ERR_PTR(-EINVAL); |
823 | |
824 | nr_pages += end - start; |
825 | len += iov[i].iov_len; |
826 | } |
827 | |
828 | if (offset) |
829 | nr_pages++; |
830 | |
831 | bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask); |
832 | if (!bmd) |
833 | return ERR_PTR(-ENOMEM); |
834 | |
835 | ret = -ENOMEM; |
836 | bio = bio_kmalloc(gfp_mask, nr_pages); |
837 | if (!bio) |
838 | goto out_bmd; |
839 | |
840 | if (!write_to_vm) |
841 | bio->bi_rw |= REQ_WRITE; |
842 | |
843 | ret = 0; |
844 | |
845 | if (map_data) { |
846 | nr_pages = 1 << map_data->page_order; |
847 | i = map_data->offset / PAGE_SIZE; |
848 | } |
849 | while (len) { |
850 | unsigned int bytes = PAGE_SIZE; |
851 | |
852 | bytes -= offset; |
853 | |
854 | if (bytes > len) |
855 | bytes = len; |
856 | |
857 | if (map_data) { |
858 | if (i == map_data->nr_entries * nr_pages) { |
859 | ret = -ENOMEM; |
860 | break; |
861 | } |
862 | |
863 | page = map_data->pages[i / nr_pages]; |
864 | page += (i % nr_pages); |
865 | |
866 | i++; |
867 | } else { |
868 | page = alloc_page(q->bounce_gfp | gfp_mask); |
869 | if (!page) { |
870 | ret = -ENOMEM; |
871 | break; |
872 | } |
873 | } |
874 | |
875 | if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) |
876 | break; |
877 | |
878 | len -= bytes; |
879 | offset = 0; |
880 | } |
881 | |
882 | if (ret) |
883 | goto cleanup; |
884 | |
885 | /* |
886 | * success |
887 | */ |
888 | if ((!write_to_vm && (!map_data || !map_data->null_mapped)) || |
889 | (map_data && map_data->from_user)) { |
890 | ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0); |
891 | if (ret) |
892 | goto cleanup; |
893 | } |
894 | |
895 | bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1); |
896 | return bio; |
897 | cleanup: |
898 | if (!map_data) |
899 | bio_for_each_segment(bvec, bio, i) |
900 | __free_page(bvec->bv_page); |
901 | |
902 | bio_put(bio); |
903 | out_bmd: |
904 | bio_free_map_data(bmd); |
905 | return ERR_PTR(ret); |
906 | } |
907 | |
908 | /** |
909 | * bio_copy_user - copy user data to bio |
910 | * @q: destination block queue |
911 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
912 | * @uaddr: start of user address |
913 | * @len: length in bytes |
914 | * @write_to_vm: bool indicating writing to pages or not |
915 | * @gfp_mask: memory allocation flags |
916 | * |
917 | * Prepares and returns a bio for indirect user io, bouncing data |
918 | * to/from kernel pages as necessary. Must be paired with |
919 | * call bio_uncopy_user() on io completion. |
920 | */ |
921 | struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data, |
922 | unsigned long uaddr, unsigned int len, |
923 | int write_to_vm, gfp_t gfp_mask) |
924 | { |
925 | struct sg_iovec iov; |
926 | |
927 | iov.iov_base = (void __user *)uaddr; |
928 | iov.iov_len = len; |
929 | |
930 | return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask); |
931 | } |
932 | EXPORT_SYMBOL(bio_copy_user); |
933 | |
934 | static struct bio *__bio_map_user_iov(struct request_queue *q, |
935 | struct block_device *bdev, |
936 | struct sg_iovec *iov, int iov_count, |
937 | int write_to_vm, gfp_t gfp_mask) |
938 | { |
939 | int i, j; |
940 | int nr_pages = 0; |
941 | struct page **pages; |
942 | struct bio *bio; |
943 | int cur_page = 0; |
944 | int ret, offset; |
945 | |
946 | for (i = 0; i < iov_count; i++) { |
947 | unsigned long uaddr = (unsigned long)iov[i].iov_base; |
948 | unsigned long len = iov[i].iov_len; |
949 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
950 | unsigned long start = uaddr >> PAGE_SHIFT; |
951 | |
952 | /* |
953 | * Overflow, abort |
954 | */ |
955 | if (end < start) |
956 | return ERR_PTR(-EINVAL); |
957 | |
958 | nr_pages += end - start; |
959 | /* |
960 | * buffer must be aligned to at least hardsector size for now |
961 | */ |
962 | if (uaddr & queue_dma_alignment(q)) |
963 | return ERR_PTR(-EINVAL); |
964 | } |
965 | |
966 | if (!nr_pages) |
967 | return ERR_PTR(-EINVAL); |
968 | |
969 | bio = bio_kmalloc(gfp_mask, nr_pages); |
970 | if (!bio) |
971 | return ERR_PTR(-ENOMEM); |
972 | |
973 | ret = -ENOMEM; |
974 | pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask); |
975 | if (!pages) |
976 | goto out; |
977 | |
978 | for (i = 0; i < iov_count; i++) { |
979 | unsigned long uaddr = (unsigned long)iov[i].iov_base; |
980 | unsigned long len = iov[i].iov_len; |
981 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
982 | unsigned long start = uaddr >> PAGE_SHIFT; |
983 | const int local_nr_pages = end - start; |
984 | const int page_limit = cur_page + local_nr_pages; |
985 | |
986 | ret = get_user_pages_fast(uaddr, local_nr_pages, |
987 | write_to_vm, &pages[cur_page]); |
988 | if (ret < local_nr_pages) { |
989 | ret = -EFAULT; |
990 | goto out_unmap; |
991 | } |
992 | |
993 | offset = uaddr & ~PAGE_MASK; |
994 | for (j = cur_page; j < page_limit; j++) { |
995 | unsigned int bytes = PAGE_SIZE - offset; |
996 | |
997 | if (len <= 0) |
998 | break; |
999 | |
1000 | if (bytes > len) |
1001 | bytes = len; |
1002 | |
1003 | /* |
1004 | * sorry... |
1005 | */ |
1006 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < |
1007 | bytes) |
1008 | break; |
1009 | |
1010 | len -= bytes; |
1011 | offset = 0; |
1012 | } |
1013 | |
1014 | cur_page = j; |
1015 | /* |
1016 | * release the pages we didn't map into the bio, if any |
1017 | */ |
1018 | while (j < page_limit) |
1019 | page_cache_release(pages[j++]); |
1020 | } |
1021 | |
1022 | kfree(pages); |
1023 | |
1024 | /* |
1025 | * set data direction, and check if mapped pages need bouncing |
1026 | */ |
1027 | if (!write_to_vm) |
1028 | bio->bi_rw |= REQ_WRITE; |
1029 | |
1030 | bio->bi_bdev = bdev; |
1031 | bio->bi_flags |= (1 << BIO_USER_MAPPED); |
1032 | return bio; |
1033 | |
1034 | out_unmap: |
1035 | for (i = 0; i < nr_pages; i++) { |
1036 | if(!pages[i]) |
1037 | break; |
1038 | page_cache_release(pages[i]); |
1039 | } |
1040 | out: |
1041 | kfree(pages); |
1042 | bio_put(bio); |
1043 | return ERR_PTR(ret); |
1044 | } |
1045 | |
1046 | /** |
1047 | * bio_map_user - map user address into bio |
1048 | * @q: the struct request_queue for the bio |
1049 | * @bdev: destination block device |
1050 | * @uaddr: start of user address |
1051 | * @len: length in bytes |
1052 | * @write_to_vm: bool indicating writing to pages or not |
1053 | * @gfp_mask: memory allocation flags |
1054 | * |
1055 | * Map the user space address into a bio suitable for io to a block |
1056 | * device. Returns an error pointer in case of error. |
1057 | */ |
1058 | struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev, |
1059 | unsigned long uaddr, unsigned int len, int write_to_vm, |
1060 | gfp_t gfp_mask) |
1061 | { |
1062 | struct sg_iovec iov; |
1063 | |
1064 | iov.iov_base = (void __user *)uaddr; |
1065 | iov.iov_len = len; |
1066 | |
1067 | return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask); |
1068 | } |
1069 | EXPORT_SYMBOL(bio_map_user); |
1070 | |
1071 | /** |
1072 | * bio_map_user_iov - map user sg_iovec table into bio |
1073 | * @q: the struct request_queue for the bio |
1074 | * @bdev: destination block device |
1075 | * @iov: the iovec. |
1076 | * @iov_count: number of elements in the iovec |
1077 | * @write_to_vm: bool indicating writing to pages or not |
1078 | * @gfp_mask: memory allocation flags |
1079 | * |
1080 | * Map the user space address into a bio suitable for io to a block |
1081 | * device. Returns an error pointer in case of error. |
1082 | */ |
1083 | struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev, |
1084 | struct sg_iovec *iov, int iov_count, |
1085 | int write_to_vm, gfp_t gfp_mask) |
1086 | { |
1087 | struct bio *bio; |
1088 | |
1089 | bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm, |
1090 | gfp_mask); |
1091 | if (IS_ERR(bio)) |
1092 | return bio; |
1093 | |
1094 | /* |
1095 | * subtle -- if __bio_map_user() ended up bouncing a bio, |
1096 | * it would normally disappear when its bi_end_io is run. |
1097 | * however, we need it for the unmap, so grab an extra |
1098 | * reference to it |
1099 | */ |
1100 | bio_get(bio); |
1101 | |
1102 | return bio; |
1103 | } |
1104 | |
1105 | static void __bio_unmap_user(struct bio *bio) |
1106 | { |
1107 | struct bio_vec *bvec; |
1108 | int i; |
1109 | |
1110 | /* |
1111 | * make sure we dirty pages we wrote to |
1112 | */ |
1113 | __bio_for_each_segment(bvec, bio, i, 0) { |
1114 | if (bio_data_dir(bio) == READ) |
1115 | set_page_dirty_lock(bvec->bv_page); |
1116 | |
1117 | page_cache_release(bvec->bv_page); |
1118 | } |
1119 | |
1120 | bio_put(bio); |
1121 | } |
1122 | |
1123 | /** |
1124 | * bio_unmap_user - unmap a bio |
1125 | * @bio: the bio being unmapped |
1126 | * |
1127 | * Unmap a bio previously mapped by bio_map_user(). Must be called with |
1128 | * a process context. |
1129 | * |
1130 | * bio_unmap_user() may sleep. |
1131 | */ |
1132 | void bio_unmap_user(struct bio *bio) |
1133 | { |
1134 | __bio_unmap_user(bio); |
1135 | bio_put(bio); |
1136 | } |
1137 | EXPORT_SYMBOL(bio_unmap_user); |
1138 | |
1139 | static void bio_map_kern_endio(struct bio *bio, int err) |
1140 | { |
1141 | bio_put(bio); |
1142 | } |
1143 | |
1144 | static struct bio *__bio_map_kern(struct request_queue *q, void *data, |
1145 | unsigned int len, gfp_t gfp_mask) |
1146 | { |
1147 | unsigned long kaddr = (unsigned long)data; |
1148 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1149 | unsigned long start = kaddr >> PAGE_SHIFT; |
1150 | const int nr_pages = end - start; |
1151 | int offset, i; |
1152 | struct bio *bio; |
1153 | |
1154 | bio = bio_kmalloc(gfp_mask, nr_pages); |
1155 | if (!bio) |
1156 | return ERR_PTR(-ENOMEM); |
1157 | |
1158 | offset = offset_in_page(kaddr); |
1159 | for (i = 0; i < nr_pages; i++) { |
1160 | unsigned int bytes = PAGE_SIZE - offset; |
1161 | |
1162 | if (len <= 0) |
1163 | break; |
1164 | |
1165 | if (bytes > len) |
1166 | bytes = len; |
1167 | |
1168 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, |
1169 | offset) < bytes) |
1170 | break; |
1171 | |
1172 | data += bytes; |
1173 | len -= bytes; |
1174 | offset = 0; |
1175 | } |
1176 | |
1177 | bio->bi_end_io = bio_map_kern_endio; |
1178 | return bio; |
1179 | } |
1180 | |
1181 | /** |
1182 | * bio_map_kern - map kernel address into bio |
1183 | * @q: the struct request_queue for the bio |
1184 | * @data: pointer to buffer to map |
1185 | * @len: length in bytes |
1186 | * @gfp_mask: allocation flags for bio allocation |
1187 | * |
1188 | * Map the kernel address into a bio suitable for io to a block |
1189 | * device. Returns an error pointer in case of error. |
1190 | */ |
1191 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, |
1192 | gfp_t gfp_mask) |
1193 | { |
1194 | struct bio *bio; |
1195 | |
1196 | bio = __bio_map_kern(q, data, len, gfp_mask); |
1197 | if (IS_ERR(bio)) |
1198 | return bio; |
1199 | |
1200 | if (bio->bi_size == len) |
1201 | return bio; |
1202 | |
1203 | /* |
1204 | * Don't support partial mappings. |
1205 | */ |
1206 | bio_put(bio); |
1207 | return ERR_PTR(-EINVAL); |
1208 | } |
1209 | EXPORT_SYMBOL(bio_map_kern); |
1210 | |
1211 | static void bio_copy_kern_endio(struct bio *bio, int err) |
1212 | { |
1213 | struct bio_vec *bvec; |
1214 | const int read = bio_data_dir(bio) == READ; |
1215 | struct bio_map_data *bmd = bio->bi_private; |
1216 | int i; |
1217 | char *p = bmd->sgvecs[0].iov_base; |
1218 | |
1219 | __bio_for_each_segment(bvec, bio, i, 0) { |
1220 | char *addr = page_address(bvec->bv_page); |
1221 | int len = bmd->iovecs[i].bv_len; |
1222 | |
1223 | if (read) |
1224 | memcpy(p, addr, len); |
1225 | |
1226 | __free_page(bvec->bv_page); |
1227 | p += len; |
1228 | } |
1229 | |
1230 | bio_free_map_data(bmd); |
1231 | bio_put(bio); |
1232 | } |
1233 | |
1234 | /** |
1235 | * bio_copy_kern - copy kernel address into bio |
1236 | * @q: the struct request_queue for the bio |
1237 | * @data: pointer to buffer to copy |
1238 | * @len: length in bytes |
1239 | * @gfp_mask: allocation flags for bio and page allocation |
1240 | * @reading: data direction is READ |
1241 | * |
1242 | * copy the kernel address into a bio suitable for io to a block |
1243 | * device. Returns an error pointer in case of error. |
1244 | */ |
1245 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, |
1246 | gfp_t gfp_mask, int reading) |
1247 | { |
1248 | struct bio *bio; |
1249 | struct bio_vec *bvec; |
1250 | int i; |
1251 | |
1252 | bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask); |
1253 | if (IS_ERR(bio)) |
1254 | return bio; |
1255 | |
1256 | if (!reading) { |
1257 | void *p = data; |
1258 | |
1259 | bio_for_each_segment(bvec, bio, i) { |
1260 | char *addr = page_address(bvec->bv_page); |
1261 | |
1262 | memcpy(addr, p, bvec->bv_len); |
1263 | p += bvec->bv_len; |
1264 | } |
1265 | } |
1266 | |
1267 | bio->bi_end_io = bio_copy_kern_endio; |
1268 | |
1269 | return bio; |
1270 | } |
1271 | EXPORT_SYMBOL(bio_copy_kern); |
1272 | |
1273 | /* |
1274 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions |
1275 | * for performing direct-IO in BIOs. |
1276 | * |
1277 | * The problem is that we cannot run set_page_dirty() from interrupt context |
1278 | * because the required locks are not interrupt-safe. So what we can do is to |
1279 | * mark the pages dirty _before_ performing IO. And in interrupt context, |
1280 | * check that the pages are still dirty. If so, fine. If not, redirty them |
1281 | * in process context. |
1282 | * |
1283 | * We special-case compound pages here: normally this means reads into hugetlb |
1284 | * pages. The logic in here doesn't really work right for compound pages |
1285 | * because the VM does not uniformly chase down the head page in all cases. |
1286 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't |
1287 | * handle them at all. So we skip compound pages here at an early stage. |
1288 | * |
1289 | * Note that this code is very hard to test under normal circumstances because |
1290 | * direct-io pins the pages with get_user_pages(). This makes |
1291 | * is_page_cache_freeable return false, and the VM will not clean the pages. |
1292 | * But other code (eg, flusher threads) could clean the pages if they are mapped |
1293 | * pagecache. |
1294 | * |
1295 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the |
1296 | * deferred bio dirtying paths. |
1297 | */ |
1298 | |
1299 | /* |
1300 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. |
1301 | */ |
1302 | void bio_set_pages_dirty(struct bio *bio) |
1303 | { |
1304 | struct bio_vec *bvec = bio->bi_io_vec; |
1305 | int i; |
1306 | |
1307 | for (i = 0; i < bio->bi_vcnt; i++) { |
1308 | struct page *page = bvec[i].bv_page; |
1309 | |
1310 | if (page && !PageCompound(page)) |
1311 | set_page_dirty_lock(page); |
1312 | } |
1313 | } |
1314 | |
1315 | static void bio_release_pages(struct bio *bio) |
1316 | { |
1317 | struct bio_vec *bvec = bio->bi_io_vec; |
1318 | int i; |
1319 | |
1320 | for (i = 0; i < bio->bi_vcnt; i++) { |
1321 | struct page *page = bvec[i].bv_page; |
1322 | |
1323 | if (page) |
1324 | put_page(page); |
1325 | } |
1326 | } |
1327 | |
1328 | /* |
1329 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. |
1330 | * If they are, then fine. If, however, some pages are clean then they must |
1331 | * have been written out during the direct-IO read. So we take another ref on |
1332 | * the BIO and the offending pages and re-dirty the pages in process context. |
1333 | * |
1334 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from |
1335 | * here on. It will run one page_cache_release() against each page and will |
1336 | * run one bio_put() against the BIO. |
1337 | */ |
1338 | |
1339 | static void bio_dirty_fn(struct work_struct *work); |
1340 | |
1341 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1342 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1343 | static struct bio *bio_dirty_list; |
1344 | |
1345 | /* |
1346 | * This runs in process context |
1347 | */ |
1348 | static void bio_dirty_fn(struct work_struct *work) |
1349 | { |
1350 | unsigned long flags; |
1351 | struct bio *bio; |
1352 | |
1353 | spin_lock_irqsave(&bio_dirty_lock, flags); |
1354 | bio = bio_dirty_list; |
1355 | bio_dirty_list = NULL; |
1356 | spin_unlock_irqrestore(&bio_dirty_lock, flags); |
1357 | |
1358 | while (bio) { |
1359 | struct bio *next = bio->bi_private; |
1360 | |
1361 | bio_set_pages_dirty(bio); |
1362 | bio_release_pages(bio); |
1363 | bio_put(bio); |
1364 | bio = next; |
1365 | } |
1366 | } |
1367 | |
1368 | void bio_check_pages_dirty(struct bio *bio) |
1369 | { |
1370 | struct bio_vec *bvec = bio->bi_io_vec; |
1371 | int nr_clean_pages = 0; |
1372 | int i; |
1373 | |
1374 | for (i = 0; i < bio->bi_vcnt; i++) { |
1375 | struct page *page = bvec[i].bv_page; |
1376 | |
1377 | if (PageDirty(page) || PageCompound(page)) { |
1378 | page_cache_release(page); |
1379 | bvec[i].bv_page = NULL; |
1380 | } else { |
1381 | nr_clean_pages++; |
1382 | } |
1383 | } |
1384 | |
1385 | if (nr_clean_pages) { |
1386 | unsigned long flags; |
1387 | |
1388 | spin_lock_irqsave(&bio_dirty_lock, flags); |
1389 | bio->bi_private = bio_dirty_list; |
1390 | bio_dirty_list = bio; |
1391 | spin_unlock_irqrestore(&bio_dirty_lock, flags); |
1392 | schedule_work(&bio_dirty_work); |
1393 | } else { |
1394 | bio_put(bio); |
1395 | } |
1396 | } |
1397 | |
1398 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
1399 | void bio_flush_dcache_pages(struct bio *bi) |
1400 | { |
1401 | int i; |
1402 | struct bio_vec *bvec; |
1403 | |
1404 | bio_for_each_segment(bvec, bi, i) |
1405 | flush_dcache_page(bvec->bv_page); |
1406 | } |
1407 | EXPORT_SYMBOL(bio_flush_dcache_pages); |
1408 | #endif |
1409 | |
1410 | /** |
1411 | * bio_endio - end I/O on a bio |
1412 | * @bio: bio |
1413 | * @error: error, if any |
1414 | * |
1415 | * Description: |
1416 | * bio_endio() will end I/O on the whole bio. bio_endio() is the |
1417 | * preferred way to end I/O on a bio, it takes care of clearing |
1418 | * BIO_UPTODATE on error. @error is 0 on success, and and one of the |
1419 | * established -Exxxx (-EIO, for instance) error values in case |
1420 | * something went wrong. No one should call bi_end_io() directly on a |
1421 | * bio unless they own it and thus know that it has an end_io |
1422 | * function. |
1423 | **/ |
1424 | void bio_endio(struct bio *bio, int error) |
1425 | { |
1426 | if (error) |
1427 | clear_bit(BIO_UPTODATE, &bio->bi_flags); |
1428 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) |
1429 | error = -EIO; |
1430 | |
1431 | if (bio->bi_end_io) |
1432 | bio->bi_end_io(bio, error); |
1433 | } |
1434 | EXPORT_SYMBOL(bio_endio); |
1435 | |
1436 | void bio_pair_release(struct bio_pair *bp) |
1437 | { |
1438 | if (atomic_dec_and_test(&bp->cnt)) { |
1439 | struct bio *master = bp->bio1.bi_private; |
1440 | |
1441 | bio_endio(master, bp->error); |
1442 | mempool_free(bp, bp->bio2.bi_private); |
1443 | } |
1444 | } |
1445 | EXPORT_SYMBOL(bio_pair_release); |
1446 | |
1447 | static void bio_pair_end_1(struct bio *bi, int err) |
1448 | { |
1449 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); |
1450 | |
1451 | if (err) |
1452 | bp->error = err; |
1453 | |
1454 | bio_pair_release(bp); |
1455 | } |
1456 | |
1457 | static void bio_pair_end_2(struct bio *bi, int err) |
1458 | { |
1459 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); |
1460 | |
1461 | if (err) |
1462 | bp->error = err; |
1463 | |
1464 | bio_pair_release(bp); |
1465 | } |
1466 | |
1467 | /* |
1468 | * split a bio - only worry about a bio with a single page in its iovec |
1469 | */ |
1470 | struct bio_pair *bio_split(struct bio *bi, int first_sectors) |
1471 | { |
1472 | struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO); |
1473 | |
1474 | if (!bp) |
1475 | return bp; |
1476 | |
1477 | trace_block_split(bdev_get_queue(bi->bi_bdev), bi, |
1478 | bi->bi_sector + first_sectors); |
1479 | |
1480 | BUG_ON(bi->bi_vcnt != 1 && bi->bi_vcnt != 0); |
1481 | BUG_ON(bi->bi_idx != 0); |
1482 | atomic_set(&bp->cnt, 3); |
1483 | bp->error = 0; |
1484 | bp->bio1 = *bi; |
1485 | bp->bio2 = *bi; |
1486 | bp->bio2.bi_sector += first_sectors; |
1487 | bp->bio2.bi_size -= first_sectors << 9; |
1488 | bp->bio1.bi_size = first_sectors << 9; |
1489 | |
1490 | if (bi->bi_vcnt != 0) { |
1491 | bp->bv1 = bi->bi_io_vec[0]; |
1492 | bp->bv2 = bi->bi_io_vec[0]; |
1493 | |
1494 | if (bio_is_rw(bi)) { |
1495 | bp->bv2.bv_offset += first_sectors << 9; |
1496 | bp->bv2.bv_len -= first_sectors << 9; |
1497 | bp->bv1.bv_len = first_sectors << 9; |
1498 | } |
1499 | |
1500 | bp->bio1.bi_io_vec = &bp->bv1; |
1501 | bp->bio2.bi_io_vec = &bp->bv2; |
1502 | |
1503 | bp->bio1.bi_max_vecs = 1; |
1504 | bp->bio2.bi_max_vecs = 1; |
1505 | } |
1506 | |
1507 | bp->bio1.bi_end_io = bio_pair_end_1; |
1508 | bp->bio2.bi_end_io = bio_pair_end_2; |
1509 | |
1510 | bp->bio1.bi_private = bi; |
1511 | bp->bio2.bi_private = bio_split_pool; |
1512 | |
1513 | if (bio_integrity(bi)) |
1514 | bio_integrity_split(bi, bp, first_sectors); |
1515 | |
1516 | return bp; |
1517 | } |
1518 | EXPORT_SYMBOL(bio_split); |
1519 | |
1520 | /** |
1521 | * bio_sector_offset - Find hardware sector offset in bio |
1522 | * @bio: bio to inspect |
1523 | * @index: bio_vec index |
1524 | * @offset: offset in bv_page |
1525 | * |
1526 | * Return the number of hardware sectors between beginning of bio |
1527 | * and an end point indicated by a bio_vec index and an offset |
1528 | * within that vector's page. |
1529 | */ |
1530 | sector_t bio_sector_offset(struct bio *bio, unsigned short index, |
1531 | unsigned int offset) |
1532 | { |
1533 | unsigned int sector_sz; |
1534 | struct bio_vec *bv; |
1535 | sector_t sectors; |
1536 | int i; |
1537 | |
1538 | sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue); |
1539 | sectors = 0; |
1540 | |
1541 | if (index >= bio->bi_idx) |
1542 | index = bio->bi_vcnt - 1; |
1543 | |
1544 | __bio_for_each_segment(bv, bio, i, 0) { |
1545 | if (i == index) { |
1546 | if (offset > bv->bv_offset) |
1547 | sectors += (offset - bv->bv_offset) / sector_sz; |
1548 | break; |
1549 | } |
1550 | |
1551 | sectors += bv->bv_len / sector_sz; |
1552 | } |
1553 | |
1554 | return sectors; |
1555 | } |
1556 | EXPORT_SYMBOL(bio_sector_offset); |
1557 | |
1558 | /* |
1559 | * create memory pools for biovec's in a bio_set. |
1560 | * use the global biovec slabs created for general use. |
1561 | */ |
1562 | static int biovec_create_pools(struct bio_set *bs, int pool_entries) |
1563 | { |
1564 | struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX; |
1565 | |
1566 | bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab); |
1567 | if (!bs->bvec_pool) |
1568 | return -ENOMEM; |
1569 | |
1570 | return 0; |
1571 | } |
1572 | |
1573 | static void biovec_free_pools(struct bio_set *bs) |
1574 | { |
1575 | mempool_destroy(bs->bvec_pool); |
1576 | } |
1577 | |
1578 | void bioset_free(struct bio_set *bs) |
1579 | { |
1580 | if (bs->bio_pool) |
1581 | mempool_destroy(bs->bio_pool); |
1582 | |
1583 | bioset_integrity_free(bs); |
1584 | biovec_free_pools(bs); |
1585 | bio_put_slab(bs); |
1586 | |
1587 | kfree(bs); |
1588 | } |
1589 | EXPORT_SYMBOL(bioset_free); |
1590 | |
1591 | /** |
1592 | * bioset_create - Create a bio_set |
1593 | * @pool_size: Number of bio and bio_vecs to cache in the mempool |
1594 | * @front_pad: Number of bytes to allocate in front of the returned bio |
1595 | * |
1596 | * Description: |
1597 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller |
1598 | * to ask for a number of bytes to be allocated in front of the bio. |
1599 | * Front pad allocation is useful for embedding the bio inside |
1600 | * another structure, to avoid allocating extra data to go with the bio. |
1601 | * Note that the bio must be embedded at the END of that structure always, |
1602 | * or things will break badly. |
1603 | */ |
1604 | struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad) |
1605 | { |
1606 | unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); |
1607 | struct bio_set *bs; |
1608 | |
1609 | bs = kzalloc(sizeof(*bs), GFP_KERNEL); |
1610 | if (!bs) |
1611 | return NULL; |
1612 | |
1613 | bs->front_pad = front_pad; |
1614 | |
1615 | bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); |
1616 | if (!bs->bio_slab) { |
1617 | kfree(bs); |
1618 | return NULL; |
1619 | } |
1620 | |
1621 | bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab); |
1622 | if (!bs->bio_pool) |
1623 | goto bad; |
1624 | |
1625 | if (!biovec_create_pools(bs, pool_size)) |
1626 | return bs; |
1627 | |
1628 | bad: |
1629 | bioset_free(bs); |
1630 | return NULL; |
1631 | } |
1632 | EXPORT_SYMBOL(bioset_create); |
1633 | |
1634 | #ifdef CONFIG_BLK_CGROUP |
1635 | /** |
1636 | * bio_associate_current - associate a bio with %current |
1637 | * @bio: target bio |
1638 | * |
1639 | * Associate @bio with %current if it hasn't been associated yet. Block |
1640 | * layer will treat @bio as if it were issued by %current no matter which |
1641 | * task actually issues it. |
1642 | * |
1643 | * This function takes an extra reference of @task's io_context and blkcg |
1644 | * which will be put when @bio is released. The caller must own @bio, |
1645 | * ensure %current->io_context exists, and is responsible for synchronizing |
1646 | * calls to this function. |
1647 | */ |
1648 | int bio_associate_current(struct bio *bio) |
1649 | { |
1650 | struct io_context *ioc; |
1651 | struct cgroup_subsys_state *css; |
1652 | |
1653 | if (bio->bi_ioc) |
1654 | return -EBUSY; |
1655 | |
1656 | ioc = current->io_context; |
1657 | if (!ioc) |
1658 | return -ENOENT; |
1659 | |
1660 | /* acquire active ref on @ioc and associate */ |
1661 | get_io_context_active(ioc); |
1662 | bio->bi_ioc = ioc; |
1663 | |
1664 | /* associate blkcg if exists */ |
1665 | rcu_read_lock(); |
1666 | css = task_subsys_state(current, blkio_subsys_id); |
1667 | if (css && css_tryget(css)) |
1668 | bio->bi_css = css; |
1669 | rcu_read_unlock(); |
1670 | |
1671 | return 0; |
1672 | } |
1673 | |
1674 | /** |
1675 | * bio_disassociate_task - undo bio_associate_current() |
1676 | * @bio: target bio |
1677 | */ |
1678 | void bio_disassociate_task(struct bio *bio) |
1679 | { |
1680 | if (bio->bi_ioc) { |
1681 | put_io_context(bio->bi_ioc); |
1682 | bio->bi_ioc = NULL; |
1683 | } |
1684 | if (bio->bi_css) { |
1685 | css_put(bio->bi_css); |
1686 | bio->bi_css = NULL; |
1687 | } |
1688 | } |
1689 | |
1690 | #endif /* CONFIG_BLK_CGROUP */ |
1691 | |
1692 | static void __init biovec_init_slabs(void) |
1693 | { |
1694 | int i; |
1695 | |
1696 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { |
1697 | int size; |
1698 | struct biovec_slab *bvs = bvec_slabs + i; |
1699 | |
1700 | if (bvs->nr_vecs <= BIO_INLINE_VECS) { |
1701 | bvs->slab = NULL; |
1702 | continue; |
1703 | } |
1704 | |
1705 | size = bvs->nr_vecs * sizeof(struct bio_vec); |
1706 | bvs->slab = kmem_cache_create(bvs->name, size, 0, |
1707 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1708 | } |
1709 | } |
1710 | |
1711 | static int __init init_bio(void) |
1712 | { |
1713 | bio_slab_max = 2; |
1714 | bio_slab_nr = 0; |
1715 | bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL); |
1716 | if (!bio_slabs) |
1717 | panic("bio: can't allocate bios\n"); |
1718 | |
1719 | bio_integrity_init(); |
1720 | biovec_init_slabs(); |
1721 | |
1722 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 0); |
1723 | if (!fs_bio_set) |
1724 | panic("bio: can't allocate bios\n"); |
1725 | |
1726 | if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE)) |
1727 | panic("bio: can't create integrity pool\n"); |
1728 | |
1729 | bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES, |
1730 | sizeof(struct bio_pair)); |
1731 | if (!bio_split_pool) |
1732 | panic("bio: can't create split pool\n"); |
1733 | |
1734 | return 0; |
1735 | } |
1736 | subsys_initcall(init_bio); |
1737 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9