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/slab.h> |
23 | #include <linux/init.h> |
24 | #include <linux/kernel.h> |
25 | #include <linux/module.h> |
26 | #include <linux/mempool.h> |
27 | #include <linux/workqueue.h> |
28 | #include <scsi/sg.h> /* for struct sg_iovec */ |
29 | |
30 | #include <trace/events/block.h> |
31 | |
32 | /* |
33 | * Test patch to inline a certain number of bi_io_vec's inside the bio |
34 | * itself, to shrink a bio data allocation from two mempool calls to one |
35 | */ |
36 | #define BIO_INLINE_VECS 4 |
37 | |
38 | static mempool_t *bio_split_pool __read_mostly; |
39 | |
40 | /* |
41 | * if you change this list, also change bvec_alloc or things will |
42 | * break badly! cannot be bigger than what you can fit into an |
43 | * unsigned short |
44 | */ |
45 | #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } |
46 | struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { |
47 | BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), |
48 | }; |
49 | #undef BV |
50 | |
51 | /* |
52 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by |
53 | * IO code that does not need private memory pools. |
54 | */ |
55 | struct bio_set *fs_bio_set; |
56 | |
57 | /* |
58 | * Our slab pool management |
59 | */ |
60 | struct bio_slab { |
61 | struct kmem_cache *slab; |
62 | unsigned int slab_ref; |
63 | unsigned int slab_size; |
64 | char name[8]; |
65 | }; |
66 | static DEFINE_MUTEX(bio_slab_lock); |
67 | static struct bio_slab *bio_slabs; |
68 | static unsigned int bio_slab_nr, bio_slab_max; |
69 | |
70 | static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) |
71 | { |
72 | unsigned int sz = sizeof(struct bio) + extra_size; |
73 | struct kmem_cache *slab = NULL; |
74 | struct bio_slab *bslab; |
75 | unsigned int i, entry = -1; |
76 | |
77 | mutex_lock(&bio_slab_lock); |
78 | |
79 | i = 0; |
80 | while (i < bio_slab_nr) { |
81 | bslab = &bio_slabs[i]; |
82 | |
83 | if (!bslab->slab && entry == -1) |
84 | entry = i; |
85 | else if (bslab->slab_size == sz) { |
86 | slab = bslab->slab; |
87 | bslab->slab_ref++; |
88 | break; |
89 | } |
90 | i++; |
91 | } |
92 | |
93 | if (slab) |
94 | goto out_unlock; |
95 | |
96 | if (bio_slab_nr == bio_slab_max && entry == -1) { |
97 | bio_slab_max <<= 1; |
98 | bio_slabs = krealloc(bio_slabs, |
99 | bio_slab_max * sizeof(struct bio_slab), |
100 | GFP_KERNEL); |
101 | if (!bio_slabs) |
102 | goto out_unlock; |
103 | } |
104 | if (entry == -1) |
105 | entry = bio_slab_nr++; |
106 | |
107 | bslab = &bio_slabs[entry]; |
108 | |
109 | snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); |
110 | slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL); |
111 | if (!slab) |
112 | goto out_unlock; |
113 | |
114 | printk("bio: create slab <%s> at %d\n", bslab->name, entry); |
115 | bslab->slab = slab; |
116 | bslab->slab_ref = 1; |
117 | bslab->slab_size = sz; |
118 | out_unlock: |
119 | mutex_unlock(&bio_slab_lock); |
120 | return slab; |
121 | } |
122 | |
123 | static void bio_put_slab(struct bio_set *bs) |
124 | { |
125 | struct bio_slab *bslab = NULL; |
126 | unsigned int i; |
127 | |
128 | mutex_lock(&bio_slab_lock); |
129 | |
130 | for (i = 0; i < bio_slab_nr; i++) { |
131 | if (bs->bio_slab == bio_slabs[i].slab) { |
132 | bslab = &bio_slabs[i]; |
133 | break; |
134 | } |
135 | } |
136 | |
137 | if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) |
138 | goto out; |
139 | |
140 | WARN_ON(!bslab->slab_ref); |
141 | |
142 | if (--bslab->slab_ref) |
143 | goto out; |
144 | |
145 | kmem_cache_destroy(bslab->slab); |
146 | bslab->slab = NULL; |
147 | |
148 | out: |
149 | mutex_unlock(&bio_slab_lock); |
150 | } |
151 | |
152 | unsigned int bvec_nr_vecs(unsigned short idx) |
153 | { |
154 | return bvec_slabs[idx].nr_vecs; |
155 | } |
156 | |
157 | void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx) |
158 | { |
159 | BIO_BUG_ON(idx >= BIOVEC_NR_POOLS); |
160 | |
161 | if (idx == BIOVEC_MAX_IDX) |
162 | mempool_free(bv, bs->bvec_pool); |
163 | else { |
164 | struct biovec_slab *bvs = bvec_slabs + idx; |
165 | |
166 | kmem_cache_free(bvs->slab, bv); |
167 | } |
168 | } |
169 | |
170 | struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, |
171 | struct bio_set *bs) |
172 | { |
173 | struct bio_vec *bvl; |
174 | |
175 | /* |
176 | * see comment near bvec_array define! |
177 | */ |
178 | switch (nr) { |
179 | case 1: |
180 | *idx = 0; |
181 | break; |
182 | case 2 ... 4: |
183 | *idx = 1; |
184 | break; |
185 | case 5 ... 16: |
186 | *idx = 2; |
187 | break; |
188 | case 17 ... 64: |
189 | *idx = 3; |
190 | break; |
191 | case 65 ... 128: |
192 | *idx = 4; |
193 | break; |
194 | case 129 ... BIO_MAX_PAGES: |
195 | *idx = 5; |
196 | break; |
197 | default: |
198 | return NULL; |
199 | } |
200 | |
201 | /* |
202 | * idx now points to the pool we want to allocate from. only the |
203 | * 1-vec entry pool is mempool backed. |
204 | */ |
205 | if (*idx == BIOVEC_MAX_IDX) { |
206 | fallback: |
207 | bvl = mempool_alloc(bs->bvec_pool, gfp_mask); |
208 | } else { |
209 | struct biovec_slab *bvs = bvec_slabs + *idx; |
210 | gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO); |
211 | |
212 | /* |
213 | * Make this allocation restricted and don't dump info on |
214 | * allocation failures, since we'll fallback to the mempool |
215 | * in case of failure. |
216 | */ |
217 | __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; |
218 | |
219 | /* |
220 | * Try a slab allocation. If this fails and __GFP_WAIT |
221 | * is set, retry with the 1-entry mempool |
222 | */ |
223 | bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); |
224 | if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) { |
225 | *idx = BIOVEC_MAX_IDX; |
226 | goto fallback; |
227 | } |
228 | } |
229 | |
230 | return bvl; |
231 | } |
232 | |
233 | void bio_free(struct bio *bio, struct bio_set *bs) |
234 | { |
235 | void *p; |
236 | |
237 | if (bio_has_allocated_vec(bio)) |
238 | bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio)); |
239 | |
240 | if (bio_integrity(bio)) |
241 | bio_integrity_free(bio, bs); |
242 | |
243 | /* |
244 | * If we have front padding, adjust the bio pointer before freeing |
245 | */ |
246 | p = bio; |
247 | if (bs->front_pad) |
248 | p -= bs->front_pad; |
249 | |
250 | mempool_free(p, bs->bio_pool); |
251 | } |
252 | EXPORT_SYMBOL(bio_free); |
253 | |
254 | void bio_init(struct bio *bio) |
255 | { |
256 | memset(bio, 0, sizeof(*bio)); |
257 | bio->bi_flags = 1 << BIO_UPTODATE; |
258 | bio->bi_comp_cpu = -1; |
259 | atomic_set(&bio->bi_cnt, 1); |
260 | } |
261 | EXPORT_SYMBOL(bio_init); |
262 | |
263 | /** |
264 | * bio_alloc_bioset - allocate a bio for I/O |
265 | * @gfp_mask: the GFP_ mask given to the slab allocator |
266 | * @nr_iovecs: number of iovecs to pre-allocate |
267 | * @bs: the bio_set to allocate from. |
268 | * |
269 | * Description: |
270 | * bio_alloc_bioset will try its own mempool to satisfy the allocation. |
271 | * If %__GFP_WAIT is set then we will block on the internal pool waiting |
272 | * for a &struct bio to become free. |
273 | * |
274 | * Note that the caller must set ->bi_destructor on successful return |
275 | * of a bio, to do the appropriate freeing of the bio once the reference |
276 | * count drops to zero. |
277 | **/ |
278 | struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) |
279 | { |
280 | unsigned long idx = BIO_POOL_NONE; |
281 | struct bio_vec *bvl = NULL; |
282 | struct bio *bio; |
283 | void *p; |
284 | |
285 | p = mempool_alloc(bs->bio_pool, gfp_mask); |
286 | if (unlikely(!p)) |
287 | return NULL; |
288 | bio = p + bs->front_pad; |
289 | |
290 | bio_init(bio); |
291 | |
292 | if (unlikely(!nr_iovecs)) |
293 | goto out_set; |
294 | |
295 | if (nr_iovecs <= BIO_INLINE_VECS) { |
296 | bvl = bio->bi_inline_vecs; |
297 | nr_iovecs = BIO_INLINE_VECS; |
298 | } else { |
299 | bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs); |
300 | if (unlikely(!bvl)) |
301 | goto err_free; |
302 | |
303 | nr_iovecs = bvec_nr_vecs(idx); |
304 | } |
305 | out_set: |
306 | bio->bi_flags |= idx << BIO_POOL_OFFSET; |
307 | bio->bi_max_vecs = nr_iovecs; |
308 | bio->bi_io_vec = bvl; |
309 | return bio; |
310 | |
311 | err_free: |
312 | mempool_free(p, bs->bio_pool); |
313 | return NULL; |
314 | } |
315 | EXPORT_SYMBOL(bio_alloc_bioset); |
316 | |
317 | static void bio_fs_destructor(struct bio *bio) |
318 | { |
319 | bio_free(bio, fs_bio_set); |
320 | } |
321 | |
322 | /** |
323 | * bio_alloc - allocate a new bio, memory pool backed |
324 | * @gfp_mask: allocation mask to use |
325 | * @nr_iovecs: number of iovecs |
326 | * |
327 | * bio_alloc will allocate a bio and associated bio_vec array that can hold |
328 | * at least @nr_iovecs entries. Allocations will be done from the |
329 | * fs_bio_set. Also see @bio_alloc_bioset and @bio_kmalloc. |
330 | * |
331 | * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate |
332 | * a bio. This is due to the mempool guarantees. To make this work, callers |
333 | * must never allocate more than 1 bio at a time from this pool. Callers |
334 | * that need to allocate more than 1 bio must always submit the previously |
335 | * allocated bio for IO before attempting to allocate a new one. Failure to |
336 | * do so can cause livelocks under memory pressure. |
337 | * |
338 | * RETURNS: |
339 | * Pointer to new bio on success, NULL on failure. |
340 | */ |
341 | struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs) |
342 | { |
343 | struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set); |
344 | |
345 | if (bio) |
346 | bio->bi_destructor = bio_fs_destructor; |
347 | |
348 | return bio; |
349 | } |
350 | EXPORT_SYMBOL(bio_alloc); |
351 | |
352 | static void bio_kmalloc_destructor(struct bio *bio) |
353 | { |
354 | if (bio_integrity(bio)) |
355 | bio_integrity_free(bio, fs_bio_set); |
356 | kfree(bio); |
357 | } |
358 | |
359 | /** |
360 | * bio_kmalloc - allocate a bio for I/O using kmalloc() |
361 | * @gfp_mask: the GFP_ mask given to the slab allocator |
362 | * @nr_iovecs: number of iovecs to pre-allocate |
363 | * |
364 | * Description: |
365 | * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask contains |
366 | * %__GFP_WAIT, the allocation is guaranteed to succeed. |
367 | * |
368 | **/ |
369 | struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs) |
370 | { |
371 | struct bio *bio; |
372 | |
373 | bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec), |
374 | gfp_mask); |
375 | if (unlikely(!bio)) |
376 | return NULL; |
377 | |
378 | bio_init(bio); |
379 | bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET; |
380 | bio->bi_max_vecs = nr_iovecs; |
381 | bio->bi_io_vec = bio->bi_inline_vecs; |
382 | bio->bi_destructor = bio_kmalloc_destructor; |
383 | |
384 | return bio; |
385 | } |
386 | EXPORT_SYMBOL(bio_kmalloc); |
387 | |
388 | void zero_fill_bio(struct bio *bio) |
389 | { |
390 | unsigned long flags; |
391 | struct bio_vec *bv; |
392 | int i; |
393 | |
394 | bio_for_each_segment(bv, bio, i) { |
395 | char *data = bvec_kmap_irq(bv, &flags); |
396 | memset(data, 0, bv->bv_len); |
397 | flush_dcache_page(bv->bv_page); |
398 | bvec_kunmap_irq(data, &flags); |
399 | } |
400 | } |
401 | EXPORT_SYMBOL(zero_fill_bio); |
402 | |
403 | /** |
404 | * bio_put - release a reference to a bio |
405 | * @bio: bio to release reference to |
406 | * |
407 | * Description: |
408 | * Put a reference to a &struct bio, either one you have gotten with |
409 | * bio_alloc, bio_get or bio_clone. The last put of a bio will free it. |
410 | **/ |
411 | void bio_put(struct bio *bio) |
412 | { |
413 | BIO_BUG_ON(!atomic_read(&bio->bi_cnt)); |
414 | |
415 | /* |
416 | * last put frees it |
417 | */ |
418 | if (atomic_dec_and_test(&bio->bi_cnt)) { |
419 | bio->bi_next = NULL; |
420 | bio->bi_destructor(bio); |
421 | } |
422 | } |
423 | EXPORT_SYMBOL(bio_put); |
424 | |
425 | inline int bio_phys_segments(struct request_queue *q, struct bio *bio) |
426 | { |
427 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) |
428 | blk_recount_segments(q, bio); |
429 | |
430 | return bio->bi_phys_segments; |
431 | } |
432 | EXPORT_SYMBOL(bio_phys_segments); |
433 | |
434 | /** |
435 | * __bio_clone - clone a bio |
436 | * @bio: destination bio |
437 | * @bio_src: bio to clone |
438 | * |
439 | * Clone a &bio. Caller will own the returned bio, but not |
440 | * the actual data it points to. Reference count of returned |
441 | * bio will be one. |
442 | */ |
443 | void __bio_clone(struct bio *bio, struct bio *bio_src) |
444 | { |
445 | memcpy(bio->bi_io_vec, bio_src->bi_io_vec, |
446 | bio_src->bi_max_vecs * sizeof(struct bio_vec)); |
447 | |
448 | /* |
449 | * most users will be overriding ->bi_bdev with a new target, |
450 | * so we don't set nor calculate new physical/hw segment counts here |
451 | */ |
452 | bio->bi_sector = bio_src->bi_sector; |
453 | bio->bi_bdev = bio_src->bi_bdev; |
454 | bio->bi_flags |= 1 << BIO_CLONED; |
455 | bio->bi_rw = bio_src->bi_rw; |
456 | bio->bi_vcnt = bio_src->bi_vcnt; |
457 | bio->bi_size = bio_src->bi_size; |
458 | bio->bi_idx = bio_src->bi_idx; |
459 | } |
460 | EXPORT_SYMBOL(__bio_clone); |
461 | |
462 | /** |
463 | * bio_clone - clone a bio |
464 | * @bio: bio to clone |
465 | * @gfp_mask: allocation priority |
466 | * |
467 | * Like __bio_clone, only also allocates the returned bio |
468 | */ |
469 | struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask) |
470 | { |
471 | struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set); |
472 | |
473 | if (!b) |
474 | return NULL; |
475 | |
476 | b->bi_destructor = bio_fs_destructor; |
477 | __bio_clone(b, bio); |
478 | |
479 | if (bio_integrity(bio)) { |
480 | int ret; |
481 | |
482 | ret = bio_integrity_clone(b, bio, gfp_mask, fs_bio_set); |
483 | |
484 | if (ret < 0) { |
485 | bio_put(b); |
486 | return NULL; |
487 | } |
488 | } |
489 | |
490 | return b; |
491 | } |
492 | EXPORT_SYMBOL(bio_clone); |
493 | |
494 | /** |
495 | * bio_get_nr_vecs - return approx number of vecs |
496 | * @bdev: I/O target |
497 | * |
498 | * Return the approximate number of pages we can send to this target. |
499 | * There's no guarantee that you will be able to fit this number of pages |
500 | * into a bio, it does not account for dynamic restrictions that vary |
501 | * on offset. |
502 | */ |
503 | int bio_get_nr_vecs(struct block_device *bdev) |
504 | { |
505 | struct request_queue *q = bdev_get_queue(bdev); |
506 | int nr_pages; |
507 | |
508 | nr_pages = ((queue_max_sectors(q) << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT; |
509 | if (nr_pages > queue_max_segments(q)) |
510 | nr_pages = queue_max_segments(q); |
511 | |
512 | return nr_pages; |
513 | } |
514 | EXPORT_SYMBOL(bio_get_nr_vecs); |
515 | |
516 | static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page |
517 | *page, unsigned int len, unsigned int offset, |
518 | unsigned short max_sectors) |
519 | { |
520 | int retried_segments = 0; |
521 | struct bio_vec *bvec; |
522 | |
523 | /* |
524 | * cloned bio must not modify vec list |
525 | */ |
526 | if (unlikely(bio_flagged(bio, BIO_CLONED))) |
527 | return 0; |
528 | |
529 | if (((bio->bi_size + len) >> 9) > max_sectors) |
530 | return 0; |
531 | |
532 | /* |
533 | * For filesystems with a blocksize smaller than the pagesize |
534 | * we will often be called with the same page as last time and |
535 | * a consecutive offset. Optimize this special case. |
536 | */ |
537 | if (bio->bi_vcnt > 0) { |
538 | struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
539 | |
540 | if (page == prev->bv_page && |
541 | offset == prev->bv_offset + prev->bv_len) { |
542 | unsigned int prev_bv_len = prev->bv_len; |
543 | prev->bv_len += len; |
544 | |
545 | if (q->merge_bvec_fn) { |
546 | struct bvec_merge_data bvm = { |
547 | /* prev_bvec is already charged in |
548 | bi_size, discharge it in order to |
549 | simulate merging updated prev_bvec |
550 | as new bvec. */ |
551 | .bi_bdev = bio->bi_bdev, |
552 | .bi_sector = bio->bi_sector, |
553 | .bi_size = bio->bi_size - prev_bv_len, |
554 | .bi_rw = bio->bi_rw, |
555 | }; |
556 | |
557 | if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) { |
558 | prev->bv_len -= len; |
559 | return 0; |
560 | } |
561 | } |
562 | |
563 | goto done; |
564 | } |
565 | } |
566 | |
567 | if (bio->bi_vcnt >= bio->bi_max_vecs) |
568 | return 0; |
569 | |
570 | /* |
571 | * we might lose a segment or two here, but rather that than |
572 | * make this too complex. |
573 | */ |
574 | |
575 | while (bio->bi_phys_segments >= queue_max_segments(q)) { |
576 | |
577 | if (retried_segments) |
578 | return 0; |
579 | |
580 | retried_segments = 1; |
581 | blk_recount_segments(q, bio); |
582 | } |
583 | |
584 | /* |
585 | * setup the new entry, we might clear it again later if we |
586 | * cannot add the page |
587 | */ |
588 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; |
589 | bvec->bv_page = page; |
590 | bvec->bv_len = len; |
591 | bvec->bv_offset = offset; |
592 | |
593 | /* |
594 | * if queue has other restrictions (eg varying max sector size |
595 | * depending on offset), it can specify a merge_bvec_fn in the |
596 | * queue to get further control |
597 | */ |
598 | if (q->merge_bvec_fn) { |
599 | struct bvec_merge_data bvm = { |
600 | .bi_bdev = bio->bi_bdev, |
601 | .bi_sector = bio->bi_sector, |
602 | .bi_size = bio->bi_size, |
603 | .bi_rw = bio->bi_rw, |
604 | }; |
605 | |
606 | /* |
607 | * merge_bvec_fn() returns number of bytes it can accept |
608 | * at this offset |
609 | */ |
610 | if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) { |
611 | bvec->bv_page = NULL; |
612 | bvec->bv_len = 0; |
613 | bvec->bv_offset = 0; |
614 | return 0; |
615 | } |
616 | } |
617 | |
618 | /* If we may be able to merge these biovecs, force a recount */ |
619 | if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) |
620 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); |
621 | |
622 | bio->bi_vcnt++; |
623 | bio->bi_phys_segments++; |
624 | done: |
625 | bio->bi_size += len; |
626 | return len; |
627 | } |
628 | |
629 | /** |
630 | * bio_add_pc_page - attempt to add page to bio |
631 | * @q: the target queue |
632 | * @bio: destination bio |
633 | * @page: page to add |
634 | * @len: vec entry length |
635 | * @offset: vec entry offset |
636 | * |
637 | * Attempt to add a page to the bio_vec maplist. This can fail for a |
638 | * number of reasons, such as the bio being full or target block |
639 | * device limitations. The target block device must allow bio's |
640 | * smaller than PAGE_SIZE, so it is always possible to add a single |
641 | * page to an empty bio. This should only be used by REQ_PC bios. |
642 | */ |
643 | int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page, |
644 | unsigned int len, unsigned int offset) |
645 | { |
646 | return __bio_add_page(q, bio, page, len, offset, |
647 | queue_max_hw_sectors(q)); |
648 | } |
649 | EXPORT_SYMBOL(bio_add_pc_page); |
650 | |
651 | /** |
652 | * bio_add_page - attempt to add page to bio |
653 | * @bio: destination bio |
654 | * @page: page to add |
655 | * @len: vec entry length |
656 | * @offset: vec entry offset |
657 | * |
658 | * Attempt to add a page to the bio_vec maplist. This can fail for a |
659 | * number of reasons, such as the bio being full or target block |
660 | * device limitations. The target block device must allow bio's |
661 | * smaller than PAGE_SIZE, so it is always possible to add a single |
662 | * page to an empty bio. |
663 | */ |
664 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, |
665 | unsigned int offset) |
666 | { |
667 | struct request_queue *q = bdev_get_queue(bio->bi_bdev); |
668 | return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q)); |
669 | } |
670 | EXPORT_SYMBOL(bio_add_page); |
671 | |
672 | struct bio_map_data { |
673 | struct bio_vec *iovecs; |
674 | struct sg_iovec *sgvecs; |
675 | int nr_sgvecs; |
676 | int is_our_pages; |
677 | }; |
678 | |
679 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio, |
680 | struct sg_iovec *iov, int iov_count, |
681 | int is_our_pages) |
682 | { |
683 | memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); |
684 | memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count); |
685 | bmd->nr_sgvecs = iov_count; |
686 | bmd->is_our_pages = is_our_pages; |
687 | bio->bi_private = bmd; |
688 | } |
689 | |
690 | static void bio_free_map_data(struct bio_map_data *bmd) |
691 | { |
692 | kfree(bmd->iovecs); |
693 | kfree(bmd->sgvecs); |
694 | kfree(bmd); |
695 | } |
696 | |
697 | static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count, |
698 | gfp_t gfp_mask) |
699 | { |
700 | struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask); |
701 | |
702 | if (!bmd) |
703 | return NULL; |
704 | |
705 | bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask); |
706 | if (!bmd->iovecs) { |
707 | kfree(bmd); |
708 | return NULL; |
709 | } |
710 | |
711 | bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask); |
712 | if (bmd->sgvecs) |
713 | return bmd; |
714 | |
715 | kfree(bmd->iovecs); |
716 | kfree(bmd); |
717 | return NULL; |
718 | } |
719 | |
720 | static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs, |
721 | struct sg_iovec *iov, int iov_count, |
722 | int to_user, int from_user, int do_free_page) |
723 | { |
724 | int ret = 0, i; |
725 | struct bio_vec *bvec; |
726 | int iov_idx = 0; |
727 | unsigned int iov_off = 0; |
728 | |
729 | __bio_for_each_segment(bvec, bio, i, 0) { |
730 | char *bv_addr = page_address(bvec->bv_page); |
731 | unsigned int bv_len = iovecs[i].bv_len; |
732 | |
733 | while (bv_len && iov_idx < iov_count) { |
734 | unsigned int bytes; |
735 | char __user *iov_addr; |
736 | |
737 | bytes = min_t(unsigned int, |
738 | iov[iov_idx].iov_len - iov_off, bv_len); |
739 | iov_addr = iov[iov_idx].iov_base + iov_off; |
740 | |
741 | if (!ret) { |
742 | if (to_user) |
743 | ret = copy_to_user(iov_addr, bv_addr, |
744 | bytes); |
745 | |
746 | if (from_user) |
747 | ret = copy_from_user(bv_addr, iov_addr, |
748 | bytes); |
749 | |
750 | if (ret) |
751 | ret = -EFAULT; |
752 | } |
753 | |
754 | bv_len -= bytes; |
755 | bv_addr += bytes; |
756 | iov_addr += bytes; |
757 | iov_off += bytes; |
758 | |
759 | if (iov[iov_idx].iov_len == iov_off) { |
760 | iov_idx++; |
761 | iov_off = 0; |
762 | } |
763 | } |
764 | |
765 | if (do_free_page) |
766 | __free_page(bvec->bv_page); |
767 | } |
768 | |
769 | return ret; |
770 | } |
771 | |
772 | /** |
773 | * bio_uncopy_user - finish previously mapped bio |
774 | * @bio: bio being terminated |
775 | * |
776 | * Free pages allocated from bio_copy_user() and write back data |
777 | * to user space in case of a read. |
778 | */ |
779 | int bio_uncopy_user(struct bio *bio) |
780 | { |
781 | struct bio_map_data *bmd = bio->bi_private; |
782 | int ret = 0; |
783 | |
784 | if (!bio_flagged(bio, BIO_NULL_MAPPED)) |
785 | ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs, |
786 | bmd->nr_sgvecs, bio_data_dir(bio) == READ, |
787 | 0, bmd->is_our_pages); |
788 | bio_free_map_data(bmd); |
789 | bio_put(bio); |
790 | return ret; |
791 | } |
792 | EXPORT_SYMBOL(bio_uncopy_user); |
793 | |
794 | /** |
795 | * bio_copy_user_iov - copy user data to bio |
796 | * @q: destination block queue |
797 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
798 | * @iov: the iovec. |
799 | * @iov_count: number of elements in the iovec |
800 | * @write_to_vm: bool indicating writing to pages or not |
801 | * @gfp_mask: memory allocation flags |
802 | * |
803 | * Prepares and returns a bio for indirect user io, bouncing data |
804 | * to/from kernel pages as necessary. Must be paired with |
805 | * call bio_uncopy_user() on io completion. |
806 | */ |
807 | struct bio *bio_copy_user_iov(struct request_queue *q, |
808 | struct rq_map_data *map_data, |
809 | struct sg_iovec *iov, int iov_count, |
810 | int write_to_vm, gfp_t gfp_mask) |
811 | { |
812 | struct bio_map_data *bmd; |
813 | struct bio_vec *bvec; |
814 | struct page *page; |
815 | struct bio *bio; |
816 | int i, ret; |
817 | int nr_pages = 0; |
818 | unsigned int len = 0; |
819 | unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0; |
820 | |
821 | for (i = 0; i < iov_count; i++) { |
822 | unsigned long uaddr; |
823 | unsigned long end; |
824 | unsigned long start; |
825 | |
826 | uaddr = (unsigned long)iov[i].iov_base; |
827 | end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
828 | start = uaddr >> PAGE_SHIFT; |
829 | |
830 | nr_pages += end - start; |
831 | len += iov[i].iov_len; |
832 | } |
833 | |
834 | if (offset) |
835 | nr_pages++; |
836 | |
837 | bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask); |
838 | if (!bmd) |
839 | return ERR_PTR(-ENOMEM); |
840 | |
841 | ret = -ENOMEM; |
842 | bio = bio_kmalloc(gfp_mask, nr_pages); |
843 | if (!bio) |
844 | goto out_bmd; |
845 | |
846 | bio->bi_rw |= (!write_to_vm << BIO_RW); |
847 | |
848 | ret = 0; |
849 | |
850 | if (map_data) { |
851 | nr_pages = 1 << map_data->page_order; |
852 | i = map_data->offset / PAGE_SIZE; |
853 | } |
854 | while (len) { |
855 | unsigned int bytes = PAGE_SIZE; |
856 | |
857 | bytes -= offset; |
858 | |
859 | if (bytes > len) |
860 | bytes = len; |
861 | |
862 | if (map_data) { |
863 | if (i == map_data->nr_entries * nr_pages) { |
864 | ret = -ENOMEM; |
865 | break; |
866 | } |
867 | |
868 | page = map_data->pages[i / nr_pages]; |
869 | page += (i % nr_pages); |
870 | |
871 | i++; |
872 | } else { |
873 | page = alloc_page(q->bounce_gfp | gfp_mask); |
874 | if (!page) { |
875 | ret = -ENOMEM; |
876 | break; |
877 | } |
878 | } |
879 | |
880 | if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) |
881 | break; |
882 | |
883 | len -= bytes; |
884 | offset = 0; |
885 | } |
886 | |
887 | if (ret) |
888 | goto cleanup; |
889 | |
890 | /* |
891 | * success |
892 | */ |
893 | if ((!write_to_vm && (!map_data || !map_data->null_mapped)) || |
894 | (map_data && map_data->from_user)) { |
895 | ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 1, 0); |
896 | if (ret) |
897 | goto cleanup; |
898 | } |
899 | |
900 | bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1); |
901 | return bio; |
902 | cleanup: |
903 | if (!map_data) |
904 | bio_for_each_segment(bvec, bio, i) |
905 | __free_page(bvec->bv_page); |
906 | |
907 | bio_put(bio); |
908 | out_bmd: |
909 | bio_free_map_data(bmd); |
910 | return ERR_PTR(ret); |
911 | } |
912 | |
913 | /** |
914 | * bio_copy_user - copy user data to bio |
915 | * @q: destination block queue |
916 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
917 | * @uaddr: start of user address |
918 | * @len: length in bytes |
919 | * @write_to_vm: bool indicating writing to pages or not |
920 | * @gfp_mask: memory allocation flags |
921 | * |
922 | * Prepares and returns a bio for indirect user io, bouncing data |
923 | * to/from kernel pages as necessary. Must be paired with |
924 | * call bio_uncopy_user() on io completion. |
925 | */ |
926 | struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data, |
927 | unsigned long uaddr, unsigned int len, |
928 | int write_to_vm, gfp_t gfp_mask) |
929 | { |
930 | struct sg_iovec iov; |
931 | |
932 | iov.iov_base = (void __user *)uaddr; |
933 | iov.iov_len = len; |
934 | |
935 | return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask); |
936 | } |
937 | EXPORT_SYMBOL(bio_copy_user); |
938 | |
939 | static struct bio *__bio_map_user_iov(struct request_queue *q, |
940 | struct block_device *bdev, |
941 | struct sg_iovec *iov, int iov_count, |
942 | int write_to_vm, gfp_t gfp_mask) |
943 | { |
944 | int i, j; |
945 | int nr_pages = 0; |
946 | struct page **pages; |
947 | struct bio *bio; |
948 | int cur_page = 0; |
949 | int ret, offset; |
950 | |
951 | for (i = 0; i < iov_count; i++) { |
952 | unsigned long uaddr = (unsigned long)iov[i].iov_base; |
953 | unsigned long len = iov[i].iov_len; |
954 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
955 | unsigned long start = uaddr >> PAGE_SHIFT; |
956 | |
957 | nr_pages += end - start; |
958 | /* |
959 | * buffer must be aligned to at least hardsector size for now |
960 | */ |
961 | if (uaddr & queue_dma_alignment(q)) |
962 | return ERR_PTR(-EINVAL); |
963 | } |
964 | |
965 | if (!nr_pages) |
966 | return ERR_PTR(-EINVAL); |
967 | |
968 | bio = bio_kmalloc(gfp_mask, nr_pages); |
969 | if (!bio) |
970 | return ERR_PTR(-ENOMEM); |
971 | |
972 | ret = -ENOMEM; |
973 | pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask); |
974 | if (!pages) |
975 | goto out; |
976 | |
977 | for (i = 0; i < iov_count; i++) { |
978 | unsigned long uaddr = (unsigned long)iov[i].iov_base; |
979 | unsigned long len = iov[i].iov_len; |
980 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
981 | unsigned long start = uaddr >> PAGE_SHIFT; |
982 | const int local_nr_pages = end - start; |
983 | const int page_limit = cur_page + local_nr_pages; |
984 | |
985 | ret = get_user_pages_fast(uaddr, local_nr_pages, |
986 | write_to_vm, &pages[cur_page]); |
987 | if (ret < local_nr_pages) { |
988 | ret = -EFAULT; |
989 | goto out_unmap; |
990 | } |
991 | |
992 | offset = uaddr & ~PAGE_MASK; |
993 | for (j = cur_page; j < page_limit; j++) { |
994 | unsigned int bytes = PAGE_SIZE - offset; |
995 | |
996 | if (len <= 0) |
997 | break; |
998 | |
999 | if (bytes > len) |
1000 | bytes = len; |
1001 | |
1002 | /* |
1003 | * sorry... |
1004 | */ |
1005 | if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < |
1006 | bytes) |
1007 | break; |
1008 | |
1009 | len -= bytes; |
1010 | offset = 0; |
1011 | } |
1012 | |
1013 | cur_page = j; |
1014 | /* |
1015 | * release the pages we didn't map into the bio, if any |
1016 | */ |
1017 | while (j < page_limit) |
1018 | page_cache_release(pages[j++]); |
1019 | } |
1020 | |
1021 | kfree(pages); |
1022 | |
1023 | /* |
1024 | * set data direction, and check if mapped pages need bouncing |
1025 | */ |
1026 | if (!write_to_vm) |
1027 | bio->bi_rw |= (1 << BIO_RW); |
1028 | |
1029 | bio->bi_bdev = bdev; |
1030 | bio->bi_flags |= (1 << BIO_USER_MAPPED); |
1031 | return bio; |
1032 | |
1033 | out_unmap: |
1034 | for (i = 0; i < nr_pages; i++) { |
1035 | if(!pages[i]) |
1036 | break; |
1037 | page_cache_release(pages[i]); |
1038 | } |
1039 | out: |
1040 | kfree(pages); |
1041 | bio_put(bio); |
1042 | return ERR_PTR(ret); |
1043 | } |
1044 | |
1045 | /** |
1046 | * bio_map_user - map user address into bio |
1047 | * @q: the struct request_queue for the bio |
1048 | * @bdev: destination block device |
1049 | * @uaddr: start of user address |
1050 | * @len: length in bytes |
1051 | * @write_to_vm: bool indicating writing to pages or not |
1052 | * @gfp_mask: memory allocation flags |
1053 | * |
1054 | * Map the user space address into a bio suitable for io to a block |
1055 | * device. Returns an error pointer in case of error. |
1056 | */ |
1057 | struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev, |
1058 | unsigned long uaddr, unsigned int len, int write_to_vm, |
1059 | gfp_t gfp_mask) |
1060 | { |
1061 | struct sg_iovec iov; |
1062 | |
1063 | iov.iov_base = (void __user *)uaddr; |
1064 | iov.iov_len = len; |
1065 | |
1066 | return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask); |
1067 | } |
1068 | EXPORT_SYMBOL(bio_map_user); |
1069 | |
1070 | /** |
1071 | * bio_map_user_iov - map user sg_iovec table into bio |
1072 | * @q: the struct request_queue for the bio |
1073 | * @bdev: destination block device |
1074 | * @iov: the iovec. |
1075 | * @iov_count: number of elements in the iovec |
1076 | * @write_to_vm: bool indicating writing to pages or not |
1077 | * @gfp_mask: memory allocation flags |
1078 | * |
1079 | * Map the user space address into a bio suitable for io to a block |
1080 | * device. Returns an error pointer in case of error. |
1081 | */ |
1082 | struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev, |
1083 | struct sg_iovec *iov, int iov_count, |
1084 | int write_to_vm, gfp_t gfp_mask) |
1085 | { |
1086 | struct bio *bio; |
1087 | |
1088 | bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm, |
1089 | gfp_mask); |
1090 | if (IS_ERR(bio)) |
1091 | return bio; |
1092 | |
1093 | /* |
1094 | * subtle -- if __bio_map_user() ended up bouncing a bio, |
1095 | * it would normally disappear when its bi_end_io is run. |
1096 | * however, we need it for the unmap, so grab an extra |
1097 | * reference to it |
1098 | */ |
1099 | bio_get(bio); |
1100 | |
1101 | return bio; |
1102 | } |
1103 | |
1104 | static void __bio_unmap_user(struct bio *bio) |
1105 | { |
1106 | struct bio_vec *bvec; |
1107 | int i; |
1108 | |
1109 | /* |
1110 | * make sure we dirty pages we wrote to |
1111 | */ |
1112 | __bio_for_each_segment(bvec, bio, i, 0) { |
1113 | if (bio_data_dir(bio) == READ) |
1114 | set_page_dirty_lock(bvec->bv_page); |
1115 | |
1116 | page_cache_release(bvec->bv_page); |
1117 | } |
1118 | |
1119 | bio_put(bio); |
1120 | } |
1121 | |
1122 | /** |
1123 | * bio_unmap_user - unmap a bio |
1124 | * @bio: the bio being unmapped |
1125 | * |
1126 | * Unmap a bio previously mapped by bio_map_user(). Must be called with |
1127 | * a process context. |
1128 | * |
1129 | * bio_unmap_user() may sleep. |
1130 | */ |
1131 | void bio_unmap_user(struct bio *bio) |
1132 | { |
1133 | __bio_unmap_user(bio); |
1134 | bio_put(bio); |
1135 | } |
1136 | EXPORT_SYMBOL(bio_unmap_user); |
1137 | |
1138 | static void bio_map_kern_endio(struct bio *bio, int err) |
1139 | { |
1140 | bio_put(bio); |
1141 | } |
1142 | |
1143 | static struct bio *__bio_map_kern(struct request_queue *q, void *data, |
1144 | unsigned int len, gfp_t gfp_mask) |
1145 | { |
1146 | unsigned long kaddr = (unsigned long)data; |
1147 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1148 | unsigned long start = kaddr >> PAGE_SHIFT; |
1149 | const int nr_pages = end - start; |
1150 | int offset, i; |
1151 | struct bio *bio; |
1152 | |
1153 | bio = bio_kmalloc(gfp_mask, nr_pages); |
1154 | if (!bio) |
1155 | return ERR_PTR(-ENOMEM); |
1156 | |
1157 | offset = offset_in_page(kaddr); |
1158 | for (i = 0; i < nr_pages; i++) { |
1159 | unsigned int bytes = PAGE_SIZE - offset; |
1160 | |
1161 | if (len <= 0) |
1162 | break; |
1163 | |
1164 | if (bytes > len) |
1165 | bytes = len; |
1166 | |
1167 | if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, |
1168 | offset) < bytes) |
1169 | break; |
1170 | |
1171 | data += bytes; |
1172 | len -= bytes; |
1173 | offset = 0; |
1174 | } |
1175 | |
1176 | bio->bi_end_io = bio_map_kern_endio; |
1177 | return bio; |
1178 | } |
1179 | |
1180 | /** |
1181 | * bio_map_kern - map kernel address into bio |
1182 | * @q: the struct request_queue for the bio |
1183 | * @data: pointer to buffer to map |
1184 | * @len: length in bytes |
1185 | * @gfp_mask: allocation flags for bio allocation |
1186 | * |
1187 | * Map the kernel address into a bio suitable for io to a block |
1188 | * device. Returns an error pointer in case of error. |
1189 | */ |
1190 | struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, |
1191 | gfp_t gfp_mask) |
1192 | { |
1193 | struct bio *bio; |
1194 | |
1195 | bio = __bio_map_kern(q, data, len, gfp_mask); |
1196 | if (IS_ERR(bio)) |
1197 | return bio; |
1198 | |
1199 | if (bio->bi_size == len) |
1200 | return bio; |
1201 | |
1202 | /* |
1203 | * Don't support partial mappings. |
1204 | */ |
1205 | bio_put(bio); |
1206 | return ERR_PTR(-EINVAL); |
1207 | } |
1208 | EXPORT_SYMBOL(bio_map_kern); |
1209 | |
1210 | static void bio_copy_kern_endio(struct bio *bio, int err) |
1211 | { |
1212 | struct bio_vec *bvec; |
1213 | const int read = bio_data_dir(bio) == READ; |
1214 | struct bio_map_data *bmd = bio->bi_private; |
1215 | int i; |
1216 | char *p = bmd->sgvecs[0].iov_base; |
1217 | |
1218 | __bio_for_each_segment(bvec, bio, i, 0) { |
1219 | char *addr = page_address(bvec->bv_page); |
1220 | int len = bmd->iovecs[i].bv_len; |
1221 | |
1222 | if (read) |
1223 | memcpy(p, addr, len); |
1224 | |
1225 | __free_page(bvec->bv_page); |
1226 | p += len; |
1227 | } |
1228 | |
1229 | bio_free_map_data(bmd); |
1230 | bio_put(bio); |
1231 | } |
1232 | |
1233 | /** |
1234 | * bio_copy_kern - copy kernel address into bio |
1235 | * @q: the struct request_queue for the bio |
1236 | * @data: pointer to buffer to copy |
1237 | * @len: length in bytes |
1238 | * @gfp_mask: allocation flags for bio and page allocation |
1239 | * @reading: data direction is READ |
1240 | * |
1241 | * copy the kernel address into a bio suitable for io to a block |
1242 | * device. Returns an error pointer in case of error. |
1243 | */ |
1244 | struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, |
1245 | gfp_t gfp_mask, int reading) |
1246 | { |
1247 | struct bio *bio; |
1248 | struct bio_vec *bvec; |
1249 | int i; |
1250 | |
1251 | bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask); |
1252 | if (IS_ERR(bio)) |
1253 | return bio; |
1254 | |
1255 | if (!reading) { |
1256 | void *p = data; |
1257 | |
1258 | bio_for_each_segment(bvec, bio, i) { |
1259 | char *addr = page_address(bvec->bv_page); |
1260 | |
1261 | memcpy(addr, p, bvec->bv_len); |
1262 | p += bvec->bv_len; |
1263 | } |
1264 | } |
1265 | |
1266 | bio->bi_end_io = bio_copy_kern_endio; |
1267 | |
1268 | return bio; |
1269 | } |
1270 | EXPORT_SYMBOL(bio_copy_kern); |
1271 | |
1272 | /* |
1273 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions |
1274 | * for performing direct-IO in BIOs. |
1275 | * |
1276 | * The problem is that we cannot run set_page_dirty() from interrupt context |
1277 | * because the required locks are not interrupt-safe. So what we can do is to |
1278 | * mark the pages dirty _before_ performing IO. And in interrupt context, |
1279 | * check that the pages are still dirty. If so, fine. If not, redirty them |
1280 | * in process context. |
1281 | * |
1282 | * We special-case compound pages here: normally this means reads into hugetlb |
1283 | * pages. The logic in here doesn't really work right for compound pages |
1284 | * because the VM does not uniformly chase down the head page in all cases. |
1285 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't |
1286 | * handle them at all. So we skip compound pages here at an early stage. |
1287 | * |
1288 | * Note that this code is very hard to test under normal circumstances because |
1289 | * direct-io pins the pages with get_user_pages(). This makes |
1290 | * is_page_cache_freeable return false, and the VM will not clean the pages. |
1291 | * But other code (eg, pdflush) could clean the pages if they are mapped |
1292 | * pagecache. |
1293 | * |
1294 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the |
1295 | * deferred bio dirtying paths. |
1296 | */ |
1297 | |
1298 | /* |
1299 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. |
1300 | */ |
1301 | void bio_set_pages_dirty(struct bio *bio) |
1302 | { |
1303 | struct bio_vec *bvec = bio->bi_io_vec; |
1304 | int i; |
1305 | |
1306 | for (i = 0; i < bio->bi_vcnt; i++) { |
1307 | struct page *page = bvec[i].bv_page; |
1308 | |
1309 | if (page && !PageCompound(page)) |
1310 | set_page_dirty_lock(page); |
1311 | } |
1312 | } |
1313 | |
1314 | static void bio_release_pages(struct bio *bio) |
1315 | { |
1316 | struct bio_vec *bvec = bio->bi_io_vec; |
1317 | int i; |
1318 | |
1319 | for (i = 0; i < bio->bi_vcnt; i++) { |
1320 | struct page *page = bvec[i].bv_page; |
1321 | |
1322 | if (page) |
1323 | put_page(page); |
1324 | } |
1325 | } |
1326 | |
1327 | /* |
1328 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. |
1329 | * If they are, then fine. If, however, some pages are clean then they must |
1330 | * have been written out during the direct-IO read. So we take another ref on |
1331 | * the BIO and the offending pages and re-dirty the pages in process context. |
1332 | * |
1333 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from |
1334 | * here on. It will run one page_cache_release() against each page and will |
1335 | * run one bio_put() against the BIO. |
1336 | */ |
1337 | |
1338 | static void bio_dirty_fn(struct work_struct *work); |
1339 | |
1340 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); |
1341 | static DEFINE_SPINLOCK(bio_dirty_lock); |
1342 | static struct bio *bio_dirty_list; |
1343 | |
1344 | /* |
1345 | * This runs in process context |
1346 | */ |
1347 | static void bio_dirty_fn(struct work_struct *work) |
1348 | { |
1349 | unsigned long flags; |
1350 | struct bio *bio; |
1351 | |
1352 | spin_lock_irqsave(&bio_dirty_lock, flags); |
1353 | bio = bio_dirty_list; |
1354 | bio_dirty_list = NULL; |
1355 | spin_unlock_irqrestore(&bio_dirty_lock, flags); |
1356 | |
1357 | while (bio) { |
1358 | struct bio *next = bio->bi_private; |
1359 | |
1360 | bio_set_pages_dirty(bio); |
1361 | bio_release_pages(bio); |
1362 | bio_put(bio); |
1363 | bio = next; |
1364 | } |
1365 | } |
1366 | |
1367 | void bio_check_pages_dirty(struct bio *bio) |
1368 | { |
1369 | struct bio_vec *bvec = bio->bi_io_vec; |
1370 | int nr_clean_pages = 0; |
1371 | int i; |
1372 | |
1373 | for (i = 0; i < bio->bi_vcnt; i++) { |
1374 | struct page *page = bvec[i].bv_page; |
1375 | |
1376 | if (PageDirty(page) || PageCompound(page)) { |
1377 | page_cache_release(page); |
1378 | bvec[i].bv_page = NULL; |
1379 | } else { |
1380 | nr_clean_pages++; |
1381 | } |
1382 | } |
1383 | |
1384 | if (nr_clean_pages) { |
1385 | unsigned long flags; |
1386 | |
1387 | spin_lock_irqsave(&bio_dirty_lock, flags); |
1388 | bio->bi_private = bio_dirty_list; |
1389 | bio_dirty_list = bio; |
1390 | spin_unlock_irqrestore(&bio_dirty_lock, flags); |
1391 | schedule_work(&bio_dirty_work); |
1392 | } else { |
1393 | bio_put(bio); |
1394 | } |
1395 | } |
1396 | |
1397 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
1398 | void bio_flush_dcache_pages(struct bio *bi) |
1399 | { |
1400 | int i; |
1401 | struct bio_vec *bvec; |
1402 | |
1403 | bio_for_each_segment(bvec, bi, i) |
1404 | flush_dcache_page(bvec->bv_page); |
1405 | } |
1406 | EXPORT_SYMBOL(bio_flush_dcache_pages); |
1407 | #endif |
1408 | |
1409 | /** |
1410 | * bio_endio - end I/O on a bio |
1411 | * @bio: bio |
1412 | * @error: error, if any |
1413 | * |
1414 | * Description: |
1415 | * bio_endio() will end I/O on the whole bio. bio_endio() is the |
1416 | * preferred way to end I/O on a bio, it takes care of clearing |
1417 | * BIO_UPTODATE on error. @error is 0 on success, and and one of the |
1418 | * established -Exxxx (-EIO, for instance) error values in case |
1419 | * something went wrong. Noone should call bi_end_io() directly on a |
1420 | * bio unless they own it and thus know that it has an end_io |
1421 | * function. |
1422 | **/ |
1423 | void bio_endio(struct bio *bio, int error) |
1424 | { |
1425 | if (error) |
1426 | clear_bit(BIO_UPTODATE, &bio->bi_flags); |
1427 | else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) |
1428 | error = -EIO; |
1429 | |
1430 | if (bio->bi_end_io) |
1431 | bio->bi_end_io(bio, error); |
1432 | } |
1433 | EXPORT_SYMBOL(bio_endio); |
1434 | |
1435 | void bio_pair_release(struct bio_pair *bp) |
1436 | { |
1437 | if (atomic_dec_and_test(&bp->cnt)) { |
1438 | struct bio *master = bp->bio1.bi_private; |
1439 | |
1440 | bio_endio(master, bp->error); |
1441 | mempool_free(bp, bp->bio2.bi_private); |
1442 | } |
1443 | } |
1444 | EXPORT_SYMBOL(bio_pair_release); |
1445 | |
1446 | static void bio_pair_end_1(struct bio *bi, int err) |
1447 | { |
1448 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); |
1449 | |
1450 | if (err) |
1451 | bp->error = err; |
1452 | |
1453 | bio_pair_release(bp); |
1454 | } |
1455 | |
1456 | static void bio_pair_end_2(struct bio *bi, int err) |
1457 | { |
1458 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); |
1459 | |
1460 | if (err) |
1461 | bp->error = err; |
1462 | |
1463 | bio_pair_release(bp); |
1464 | } |
1465 | |
1466 | /* |
1467 | * split a bio - only worry about a bio with a single page in its iovec |
1468 | */ |
1469 | struct bio_pair *bio_split(struct bio *bi, int first_sectors) |
1470 | { |
1471 | struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO); |
1472 | |
1473 | if (!bp) |
1474 | return bp; |
1475 | |
1476 | trace_block_split(bdev_get_queue(bi->bi_bdev), bi, |
1477 | bi->bi_sector + first_sectors); |
1478 | |
1479 | BUG_ON(bi->bi_vcnt != 1); |
1480 | BUG_ON(bi->bi_idx != 0); |
1481 | atomic_set(&bp->cnt, 3); |
1482 | bp->error = 0; |
1483 | bp->bio1 = *bi; |
1484 | bp->bio2 = *bi; |
1485 | bp->bio2.bi_sector += first_sectors; |
1486 | bp->bio2.bi_size -= first_sectors << 9; |
1487 | bp->bio1.bi_size = first_sectors << 9; |
1488 | |
1489 | bp->bv1 = bi->bi_io_vec[0]; |
1490 | bp->bv2 = bi->bi_io_vec[0]; |
1491 | bp->bv2.bv_offset += first_sectors << 9; |
1492 | bp->bv2.bv_len -= first_sectors << 9; |
1493 | bp->bv1.bv_len = first_sectors << 9; |
1494 | |
1495 | bp->bio1.bi_io_vec = &bp->bv1; |
1496 | bp->bio2.bi_io_vec = &bp->bv2; |
1497 | |
1498 | bp->bio1.bi_max_vecs = 1; |
1499 | bp->bio2.bi_max_vecs = 1; |
1500 | |
1501 | bp->bio1.bi_end_io = bio_pair_end_1; |
1502 | bp->bio2.bi_end_io = bio_pair_end_2; |
1503 | |
1504 | bp->bio1.bi_private = bi; |
1505 | bp->bio2.bi_private = bio_split_pool; |
1506 | |
1507 | if (bio_integrity(bi)) |
1508 | bio_integrity_split(bi, bp, first_sectors); |
1509 | |
1510 | return bp; |
1511 | } |
1512 | EXPORT_SYMBOL(bio_split); |
1513 | |
1514 | /** |
1515 | * bio_sector_offset - Find hardware sector offset in bio |
1516 | * @bio: bio to inspect |
1517 | * @index: bio_vec index |
1518 | * @offset: offset in bv_page |
1519 | * |
1520 | * Return the number of hardware sectors between beginning of bio |
1521 | * and an end point indicated by a bio_vec index and an offset |
1522 | * within that vector's page. |
1523 | */ |
1524 | sector_t bio_sector_offset(struct bio *bio, unsigned short index, |
1525 | unsigned int offset) |
1526 | { |
1527 | unsigned int sector_sz; |
1528 | struct bio_vec *bv; |
1529 | sector_t sectors; |
1530 | int i; |
1531 | |
1532 | sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue); |
1533 | sectors = 0; |
1534 | |
1535 | if (index >= bio->bi_idx) |
1536 | index = bio->bi_vcnt - 1; |
1537 | |
1538 | __bio_for_each_segment(bv, bio, i, 0) { |
1539 | if (i == index) { |
1540 | if (offset > bv->bv_offset) |
1541 | sectors += (offset - bv->bv_offset) / sector_sz; |
1542 | break; |
1543 | } |
1544 | |
1545 | sectors += bv->bv_len / sector_sz; |
1546 | } |
1547 | |
1548 | return sectors; |
1549 | } |
1550 | EXPORT_SYMBOL(bio_sector_offset); |
1551 | |
1552 | /* |
1553 | * create memory pools for biovec's in a bio_set. |
1554 | * use the global biovec slabs created for general use. |
1555 | */ |
1556 | static int biovec_create_pools(struct bio_set *bs, int pool_entries) |
1557 | { |
1558 | struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX; |
1559 | |
1560 | bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab); |
1561 | if (!bs->bvec_pool) |
1562 | return -ENOMEM; |
1563 | |
1564 | return 0; |
1565 | } |
1566 | |
1567 | static void biovec_free_pools(struct bio_set *bs) |
1568 | { |
1569 | mempool_destroy(bs->bvec_pool); |
1570 | } |
1571 | |
1572 | void bioset_free(struct bio_set *bs) |
1573 | { |
1574 | if (bs->bio_pool) |
1575 | mempool_destroy(bs->bio_pool); |
1576 | |
1577 | bioset_integrity_free(bs); |
1578 | biovec_free_pools(bs); |
1579 | bio_put_slab(bs); |
1580 | |
1581 | kfree(bs); |
1582 | } |
1583 | EXPORT_SYMBOL(bioset_free); |
1584 | |
1585 | /** |
1586 | * bioset_create - Create a bio_set |
1587 | * @pool_size: Number of bio and bio_vecs to cache in the mempool |
1588 | * @front_pad: Number of bytes to allocate in front of the returned bio |
1589 | * |
1590 | * Description: |
1591 | * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller |
1592 | * to ask for a number of bytes to be allocated in front of the bio. |
1593 | * Front pad allocation is useful for embedding the bio inside |
1594 | * another structure, to avoid allocating extra data to go with the bio. |
1595 | * Note that the bio must be embedded at the END of that structure always, |
1596 | * or things will break badly. |
1597 | */ |
1598 | struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad) |
1599 | { |
1600 | unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); |
1601 | struct bio_set *bs; |
1602 | |
1603 | bs = kzalloc(sizeof(*bs), GFP_KERNEL); |
1604 | if (!bs) |
1605 | return NULL; |
1606 | |
1607 | bs->front_pad = front_pad; |
1608 | |
1609 | bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); |
1610 | if (!bs->bio_slab) { |
1611 | kfree(bs); |
1612 | return NULL; |
1613 | } |
1614 | |
1615 | bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab); |
1616 | if (!bs->bio_pool) |
1617 | goto bad; |
1618 | |
1619 | if (bioset_integrity_create(bs, pool_size)) |
1620 | goto bad; |
1621 | |
1622 | if (!biovec_create_pools(bs, pool_size)) |
1623 | return bs; |
1624 | |
1625 | bad: |
1626 | bioset_free(bs); |
1627 | return NULL; |
1628 | } |
1629 | EXPORT_SYMBOL(bioset_create); |
1630 | |
1631 | static void __init biovec_init_slabs(void) |
1632 | { |
1633 | int i; |
1634 | |
1635 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { |
1636 | int size; |
1637 | struct biovec_slab *bvs = bvec_slabs + i; |
1638 | |
1639 | #ifndef CONFIG_BLK_DEV_INTEGRITY |
1640 | if (bvs->nr_vecs <= BIO_INLINE_VECS) { |
1641 | bvs->slab = NULL; |
1642 | continue; |
1643 | } |
1644 | #endif |
1645 | |
1646 | size = bvs->nr_vecs * sizeof(struct bio_vec); |
1647 | bvs->slab = kmem_cache_create(bvs->name, size, 0, |
1648 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1649 | } |
1650 | } |
1651 | |
1652 | static int __init init_bio(void) |
1653 | { |
1654 | bio_slab_max = 2; |
1655 | bio_slab_nr = 0; |
1656 | bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL); |
1657 | if (!bio_slabs) |
1658 | panic("bio: can't allocate bios\n"); |
1659 | |
1660 | bio_integrity_init(); |
1661 | biovec_init_slabs(); |
1662 | |
1663 | fs_bio_set = bioset_create(BIO_POOL_SIZE, 0); |
1664 | if (!fs_bio_set) |
1665 | panic("bio: can't allocate bios\n"); |
1666 | |
1667 | bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES, |
1668 | sizeof(struct bio_pair)); |
1669 | if (!bio_split_pool) |
1670 | panic("bio: can't create split pool\n"); |
1671 | |
1672 | return 0; |
1673 | } |
1674 | subsys_initcall(init_bio); |
1675 |
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