]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (C) 2001 Jens Axboe <axboe@suse.de> | |
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 | ||
29 | #define BIO_POOL_SIZE 256 | |
30 | ||
31 | static kmem_cache_t *bio_slab; | |
32 | ||
33 | #define BIOVEC_NR_POOLS 6 | |
34 | ||
35 | /* | |
36 | * a small number of entries is fine, not going to be performance critical. | |
37 | * basically we just need to survive | |
38 | */ | |
39 | #define BIO_SPLIT_ENTRIES 8 | |
40 | mempool_t *bio_split_pool; | |
41 | ||
42 | struct biovec_slab { | |
43 | int nr_vecs; | |
44 | char *name; | |
45 | kmem_cache_t *slab; | |
46 | }; | |
47 | ||
48 | /* | |
49 | * if you change this list, also change bvec_alloc or things will | |
50 | * break badly! cannot be bigger than what you can fit into an | |
51 | * unsigned short | |
52 | */ | |
53 | ||
54 | #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } | |
55 | static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { | |
56 | BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), | |
57 | }; | |
58 | #undef BV | |
59 | ||
60 | /* | |
61 | * bio_set is used to allow other portions of the IO system to | |
62 | * allocate their own private memory pools for bio and iovec structures. | |
63 | * These memory pools in turn all allocate from the bio_slab | |
64 | * and the bvec_slabs[]. | |
65 | */ | |
66 | struct bio_set { | |
67 | mempool_t *bio_pool; | |
68 | mempool_t *bvec_pools[BIOVEC_NR_POOLS]; | |
69 | }; | |
70 | ||
71 | /* | |
72 | * fs_bio_set is the bio_set containing bio and iovec memory pools used by | |
73 | * IO code that does not need private memory pools. | |
74 | */ | |
75 | static struct bio_set *fs_bio_set; | |
76 | ||
77 | static inline struct bio_vec *bvec_alloc_bs(unsigned int __nocast gfp_mask, int nr, unsigned long *idx, struct bio_set *bs) | |
78 | { | |
79 | struct bio_vec *bvl; | |
80 | struct biovec_slab *bp; | |
81 | ||
82 | /* | |
83 | * see comment near bvec_array define! | |
84 | */ | |
85 | switch (nr) { | |
86 | case 1 : *idx = 0; break; | |
87 | case 2 ... 4: *idx = 1; break; | |
88 | case 5 ... 16: *idx = 2; break; | |
89 | case 17 ... 64: *idx = 3; break; | |
90 | case 65 ... 128: *idx = 4; break; | |
91 | case 129 ... BIO_MAX_PAGES: *idx = 5; break; | |
92 | default: | |
93 | return NULL; | |
94 | } | |
95 | /* | |
96 | * idx now points to the pool we want to allocate from | |
97 | */ | |
98 | ||
99 | bp = bvec_slabs + *idx; | |
100 | bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask); | |
101 | if (bvl) | |
102 | memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec)); | |
103 | ||
104 | return bvl; | |
105 | } | |
106 | ||
107 | /* | |
108 | * default destructor for a bio allocated with bio_alloc_bioset() | |
109 | */ | |
110 | static void bio_destructor(struct bio *bio) | |
111 | { | |
112 | const int pool_idx = BIO_POOL_IDX(bio); | |
113 | struct bio_set *bs = bio->bi_set; | |
114 | ||
115 | BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS); | |
116 | ||
117 | mempool_free(bio->bi_io_vec, bs->bvec_pools[pool_idx]); | |
118 | mempool_free(bio, bs->bio_pool); | |
119 | } | |
120 | ||
121 | inline void bio_init(struct bio *bio) | |
122 | { | |
123 | bio->bi_next = NULL; | |
124 | bio->bi_flags = 1 << BIO_UPTODATE; | |
125 | bio->bi_rw = 0; | |
126 | bio->bi_vcnt = 0; | |
127 | bio->bi_idx = 0; | |
128 | bio->bi_phys_segments = 0; | |
129 | bio->bi_hw_segments = 0; | |
130 | bio->bi_hw_front_size = 0; | |
131 | bio->bi_hw_back_size = 0; | |
132 | bio->bi_size = 0; | |
133 | bio->bi_max_vecs = 0; | |
134 | bio->bi_end_io = NULL; | |
135 | atomic_set(&bio->bi_cnt, 1); | |
136 | bio->bi_private = NULL; | |
137 | } | |
138 | ||
139 | /** | |
140 | * bio_alloc_bioset - allocate a bio for I/O | |
141 | * @gfp_mask: the GFP_ mask given to the slab allocator | |
142 | * @nr_iovecs: number of iovecs to pre-allocate | |
143 | * @bs: the bio_set to allocate from | |
144 | * | |
145 | * Description: | |
146 | * bio_alloc_bioset will first try it's on mempool to satisfy the allocation. | |
147 | * If %__GFP_WAIT is set then we will block on the internal pool waiting | |
148 | * for a &struct bio to become free. | |
149 | * | |
150 | * allocate bio and iovecs from the memory pools specified by the | |
151 | * bio_set structure. | |
152 | **/ | |
153 | struct bio *bio_alloc_bioset(unsigned int __nocast gfp_mask, int nr_iovecs, struct bio_set *bs) | |
154 | { | |
155 | struct bio *bio = mempool_alloc(bs->bio_pool, gfp_mask); | |
156 | ||
157 | if (likely(bio)) { | |
158 | struct bio_vec *bvl = NULL; | |
159 | ||
160 | bio_init(bio); | |
161 | if (likely(nr_iovecs)) { | |
162 | unsigned long idx; | |
163 | ||
164 | bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs); | |
165 | if (unlikely(!bvl)) { | |
166 | mempool_free(bio, bs->bio_pool); | |
167 | bio = NULL; | |
168 | goto out; | |
169 | } | |
170 | bio->bi_flags |= idx << BIO_POOL_OFFSET; | |
171 | bio->bi_max_vecs = bvec_slabs[idx].nr_vecs; | |
172 | } | |
173 | bio->bi_io_vec = bvl; | |
174 | bio->bi_destructor = bio_destructor; | |
175 | bio->bi_set = bs; | |
176 | } | |
177 | out: | |
178 | return bio; | |
179 | } | |
180 | ||
181 | struct bio *bio_alloc(unsigned int __nocast gfp_mask, int nr_iovecs) | |
182 | { | |
183 | return bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set); | |
184 | } | |
185 | ||
186 | void zero_fill_bio(struct bio *bio) | |
187 | { | |
188 | unsigned long flags; | |
189 | struct bio_vec *bv; | |
190 | int i; | |
191 | ||
192 | bio_for_each_segment(bv, bio, i) { | |
193 | char *data = bvec_kmap_irq(bv, &flags); | |
194 | memset(data, 0, bv->bv_len); | |
195 | flush_dcache_page(bv->bv_page); | |
196 | bvec_kunmap_irq(data, &flags); | |
197 | } | |
198 | } | |
199 | EXPORT_SYMBOL(zero_fill_bio); | |
200 | ||
201 | /** | |
202 | * bio_put - release a reference to a bio | |
203 | * @bio: bio to release reference to | |
204 | * | |
205 | * Description: | |
206 | * Put a reference to a &struct bio, either one you have gotten with | |
207 | * bio_alloc or bio_get. The last put of a bio will free it. | |
208 | **/ | |
209 | void bio_put(struct bio *bio) | |
210 | { | |
211 | BIO_BUG_ON(!atomic_read(&bio->bi_cnt)); | |
212 | ||
213 | /* | |
214 | * last put frees it | |
215 | */ | |
216 | if (atomic_dec_and_test(&bio->bi_cnt)) { | |
217 | bio->bi_next = NULL; | |
218 | bio->bi_destructor(bio); | |
219 | } | |
220 | } | |
221 | ||
222 | inline int bio_phys_segments(request_queue_t *q, struct bio *bio) | |
223 | { | |
224 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
225 | blk_recount_segments(q, bio); | |
226 | ||
227 | return bio->bi_phys_segments; | |
228 | } | |
229 | ||
230 | inline int bio_hw_segments(request_queue_t *q, struct bio *bio) | |
231 | { | |
232 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | |
233 | blk_recount_segments(q, bio); | |
234 | ||
235 | return bio->bi_hw_segments; | |
236 | } | |
237 | ||
238 | /** | |
239 | * __bio_clone - clone a bio | |
240 | * @bio: destination bio | |
241 | * @bio_src: bio to clone | |
242 | * | |
243 | * Clone a &bio. Caller will own the returned bio, but not | |
244 | * the actual data it points to. Reference count of returned | |
245 | * bio will be one. | |
246 | */ | |
247 | inline void __bio_clone(struct bio *bio, struct bio *bio_src) | |
248 | { | |
249 | request_queue_t *q = bdev_get_queue(bio_src->bi_bdev); | |
250 | ||
251 | memcpy(bio->bi_io_vec, bio_src->bi_io_vec, | |
252 | bio_src->bi_max_vecs * sizeof(struct bio_vec)); | |
253 | ||
254 | bio->bi_sector = bio_src->bi_sector; | |
255 | bio->bi_bdev = bio_src->bi_bdev; | |
256 | bio->bi_flags |= 1 << BIO_CLONED; | |
257 | bio->bi_rw = bio_src->bi_rw; | |
258 | bio->bi_vcnt = bio_src->bi_vcnt; | |
259 | bio->bi_size = bio_src->bi_size; | |
260 | bio->bi_idx = bio_src->bi_idx; | |
261 | bio_phys_segments(q, bio); | |
262 | bio_hw_segments(q, bio); | |
263 | } | |
264 | ||
265 | /** | |
266 | * bio_clone - clone a bio | |
267 | * @bio: bio to clone | |
268 | * @gfp_mask: allocation priority | |
269 | * | |
270 | * Like __bio_clone, only also allocates the returned bio | |
271 | */ | |
272 | struct bio *bio_clone(struct bio *bio, unsigned int __nocast gfp_mask) | |
273 | { | |
274 | struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set); | |
275 | ||
276 | if (b) | |
277 | __bio_clone(b, bio); | |
278 | ||
279 | return b; | |
280 | } | |
281 | ||
282 | /** | |
283 | * bio_get_nr_vecs - return approx number of vecs | |
284 | * @bdev: I/O target | |
285 | * | |
286 | * Return the approximate number of pages we can send to this target. | |
287 | * There's no guarantee that you will be able to fit this number of pages | |
288 | * into a bio, it does not account for dynamic restrictions that vary | |
289 | * on offset. | |
290 | */ | |
291 | int bio_get_nr_vecs(struct block_device *bdev) | |
292 | { | |
293 | request_queue_t *q = bdev_get_queue(bdev); | |
294 | int nr_pages; | |
295 | ||
296 | nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
297 | if (nr_pages > q->max_phys_segments) | |
298 | nr_pages = q->max_phys_segments; | |
299 | if (nr_pages > q->max_hw_segments) | |
300 | nr_pages = q->max_hw_segments; | |
301 | ||
302 | return nr_pages; | |
303 | } | |
304 | ||
305 | static int __bio_add_page(request_queue_t *q, struct bio *bio, struct page | |
306 | *page, unsigned int len, unsigned int offset) | |
307 | { | |
308 | int retried_segments = 0; | |
309 | struct bio_vec *bvec; | |
310 | ||
311 | /* | |
312 | * cloned bio must not modify vec list | |
313 | */ | |
314 | if (unlikely(bio_flagged(bio, BIO_CLONED))) | |
315 | return 0; | |
316 | ||
317 | if (bio->bi_vcnt >= bio->bi_max_vecs) | |
318 | return 0; | |
319 | ||
320 | if (((bio->bi_size + len) >> 9) > q->max_sectors) | |
321 | return 0; | |
322 | ||
323 | /* | |
324 | * we might lose a segment or two here, but rather that than | |
325 | * make this too complex. | |
326 | */ | |
327 | ||
328 | while (bio->bi_phys_segments >= q->max_phys_segments | |
329 | || bio->bi_hw_segments >= q->max_hw_segments | |
330 | || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) { | |
331 | ||
332 | if (retried_segments) | |
333 | return 0; | |
334 | ||
335 | retried_segments = 1; | |
336 | blk_recount_segments(q, bio); | |
337 | } | |
338 | ||
339 | /* | |
340 | * setup the new entry, we might clear it again later if we | |
341 | * cannot add the page | |
342 | */ | |
343 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; | |
344 | bvec->bv_page = page; | |
345 | bvec->bv_len = len; | |
346 | bvec->bv_offset = offset; | |
347 | ||
348 | /* | |
349 | * if queue has other restrictions (eg varying max sector size | |
350 | * depending on offset), it can specify a merge_bvec_fn in the | |
351 | * queue to get further control | |
352 | */ | |
353 | if (q->merge_bvec_fn) { | |
354 | /* | |
355 | * merge_bvec_fn() returns number of bytes it can accept | |
356 | * at this offset | |
357 | */ | |
358 | if (q->merge_bvec_fn(q, bio, bvec) < len) { | |
359 | bvec->bv_page = NULL; | |
360 | bvec->bv_len = 0; | |
361 | bvec->bv_offset = 0; | |
362 | return 0; | |
363 | } | |
364 | } | |
365 | ||
366 | /* If we may be able to merge these biovecs, force a recount */ | |
367 | if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) || | |
368 | BIOVEC_VIRT_MERGEABLE(bvec-1, bvec))) | |
369 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
370 | ||
371 | bio->bi_vcnt++; | |
372 | bio->bi_phys_segments++; | |
373 | bio->bi_hw_segments++; | |
374 | bio->bi_size += len; | |
375 | return len; | |
376 | } | |
377 | ||
378 | /** | |
379 | * bio_add_page - attempt to add page to bio | |
380 | * @bio: destination bio | |
381 | * @page: page to add | |
382 | * @len: vec entry length | |
383 | * @offset: vec entry offset | |
384 | * | |
385 | * Attempt to add a page to the bio_vec maplist. This can fail for a | |
386 | * number of reasons, such as the bio being full or target block | |
387 | * device limitations. The target block device must allow bio's | |
388 | * smaller than PAGE_SIZE, so it is always possible to add a single | |
389 | * page to an empty bio. | |
390 | */ | |
391 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, | |
392 | unsigned int offset) | |
393 | { | |
394 | return __bio_add_page(bdev_get_queue(bio->bi_bdev), bio, page, | |
395 | len, offset); | |
396 | } | |
397 | ||
398 | struct bio_map_data { | |
399 | struct bio_vec *iovecs; | |
400 | void __user *userptr; | |
401 | }; | |
402 | ||
403 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio) | |
404 | { | |
405 | memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); | |
406 | bio->bi_private = bmd; | |
407 | } | |
408 | ||
409 | static void bio_free_map_data(struct bio_map_data *bmd) | |
410 | { | |
411 | kfree(bmd->iovecs); | |
412 | kfree(bmd); | |
413 | } | |
414 | ||
415 | static struct bio_map_data *bio_alloc_map_data(int nr_segs) | |
416 | { | |
417 | struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL); | |
418 | ||
419 | if (!bmd) | |
420 | return NULL; | |
421 | ||
422 | bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL); | |
423 | if (bmd->iovecs) | |
424 | return bmd; | |
425 | ||
426 | kfree(bmd); | |
427 | return NULL; | |
428 | } | |
429 | ||
430 | /** | |
431 | * bio_uncopy_user - finish previously mapped bio | |
432 | * @bio: bio being terminated | |
433 | * | |
434 | * Free pages allocated from bio_copy_user() and write back data | |
435 | * to user space in case of a read. | |
436 | */ | |
437 | int bio_uncopy_user(struct bio *bio) | |
438 | { | |
439 | struct bio_map_data *bmd = bio->bi_private; | |
440 | const int read = bio_data_dir(bio) == READ; | |
441 | struct bio_vec *bvec; | |
442 | int i, ret = 0; | |
443 | ||
444 | __bio_for_each_segment(bvec, bio, i, 0) { | |
445 | char *addr = page_address(bvec->bv_page); | |
446 | unsigned int len = bmd->iovecs[i].bv_len; | |
447 | ||
448 | if (read && !ret && copy_to_user(bmd->userptr, addr, len)) | |
449 | ret = -EFAULT; | |
450 | ||
451 | __free_page(bvec->bv_page); | |
452 | bmd->userptr += len; | |
453 | } | |
454 | bio_free_map_data(bmd); | |
455 | bio_put(bio); | |
456 | return ret; | |
457 | } | |
458 | ||
459 | /** | |
460 | * bio_copy_user - copy user data to bio | |
461 | * @q: destination block queue | |
462 | * @uaddr: start of user address | |
463 | * @len: length in bytes | |
464 | * @write_to_vm: bool indicating writing to pages or not | |
465 | * | |
466 | * Prepares and returns a bio for indirect user io, bouncing data | |
467 | * to/from kernel pages as necessary. Must be paired with | |
468 | * call bio_uncopy_user() on io completion. | |
469 | */ | |
470 | struct bio *bio_copy_user(request_queue_t *q, unsigned long uaddr, | |
471 | unsigned int len, int write_to_vm) | |
472 | { | |
473 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
474 | unsigned long start = uaddr >> PAGE_SHIFT; | |
475 | struct bio_map_data *bmd; | |
476 | struct bio_vec *bvec; | |
477 | struct page *page; | |
478 | struct bio *bio; | |
479 | int i, ret; | |
480 | ||
481 | bmd = bio_alloc_map_data(end - start); | |
482 | if (!bmd) | |
483 | return ERR_PTR(-ENOMEM); | |
484 | ||
485 | bmd->userptr = (void __user *) uaddr; | |
486 | ||
487 | ret = -ENOMEM; | |
488 | bio = bio_alloc(GFP_KERNEL, end - start); | |
489 | if (!bio) | |
490 | goto out_bmd; | |
491 | ||
492 | bio->bi_rw |= (!write_to_vm << BIO_RW); | |
493 | ||
494 | ret = 0; | |
495 | while (len) { | |
496 | unsigned int bytes = PAGE_SIZE; | |
497 | ||
498 | if (bytes > len) | |
499 | bytes = len; | |
500 | ||
501 | page = alloc_page(q->bounce_gfp | GFP_KERNEL); | |
502 | if (!page) { | |
503 | ret = -ENOMEM; | |
504 | break; | |
505 | } | |
506 | ||
507 | if (__bio_add_page(q, bio, page, bytes, 0) < bytes) { | |
508 | ret = -EINVAL; | |
509 | break; | |
510 | } | |
511 | ||
512 | len -= bytes; | |
513 | } | |
514 | ||
515 | if (ret) | |
516 | goto cleanup; | |
517 | ||
518 | /* | |
519 | * success | |
520 | */ | |
521 | if (!write_to_vm) { | |
522 | char __user *p = (char __user *) uaddr; | |
523 | ||
524 | /* | |
525 | * for a write, copy in data to kernel pages | |
526 | */ | |
527 | ret = -EFAULT; | |
528 | bio_for_each_segment(bvec, bio, i) { | |
529 | char *addr = page_address(bvec->bv_page); | |
530 | ||
531 | if (copy_from_user(addr, p, bvec->bv_len)) | |
532 | goto cleanup; | |
533 | p += bvec->bv_len; | |
534 | } | |
535 | } | |
536 | ||
537 | bio_set_map_data(bmd, bio); | |
538 | return bio; | |
539 | cleanup: | |
540 | bio_for_each_segment(bvec, bio, i) | |
541 | __free_page(bvec->bv_page); | |
542 | ||
543 | bio_put(bio); | |
544 | out_bmd: | |
545 | bio_free_map_data(bmd); | |
546 | return ERR_PTR(ret); | |
547 | } | |
548 | ||
549 | static struct bio *__bio_map_user(request_queue_t *q, struct block_device *bdev, | |
550 | unsigned long uaddr, unsigned int len, | |
551 | int write_to_vm) | |
552 | { | |
553 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
554 | unsigned long start = uaddr >> PAGE_SHIFT; | |
555 | const int nr_pages = end - start; | |
556 | int ret, offset, i; | |
557 | struct page **pages; | |
558 | struct bio *bio; | |
559 | ||
560 | /* | |
561 | * transfer and buffer must be aligned to at least hardsector | |
562 | * size for now, in the future we can relax this restriction | |
563 | */ | |
564 | if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q))) | |
565 | return ERR_PTR(-EINVAL); | |
566 | ||
567 | bio = bio_alloc(GFP_KERNEL, nr_pages); | |
568 | if (!bio) | |
569 | return ERR_PTR(-ENOMEM); | |
570 | ||
571 | ret = -ENOMEM; | |
572 | pages = kmalloc(nr_pages * sizeof(struct page *), GFP_KERNEL); | |
573 | if (!pages) | |
574 | goto out; | |
575 | ||
576 | down_read(¤t->mm->mmap_sem); | |
577 | ret = get_user_pages(current, current->mm, uaddr, nr_pages, | |
578 | write_to_vm, 0, pages, NULL); | |
579 | up_read(¤t->mm->mmap_sem); | |
580 | ||
581 | if (ret < nr_pages) | |
582 | goto out; | |
583 | ||
584 | bio->bi_bdev = bdev; | |
585 | ||
586 | offset = uaddr & ~PAGE_MASK; | |
587 | for (i = 0; i < nr_pages; i++) { | |
588 | unsigned int bytes = PAGE_SIZE - offset; | |
589 | ||
590 | if (len <= 0) | |
591 | break; | |
592 | ||
593 | if (bytes > len) | |
594 | bytes = len; | |
595 | ||
596 | /* | |
597 | * sorry... | |
598 | */ | |
599 | if (__bio_add_page(q, bio, pages[i], bytes, offset) < bytes) | |
600 | break; | |
601 | ||
602 | len -= bytes; | |
603 | offset = 0; | |
604 | } | |
605 | ||
606 | /* | |
607 | * release the pages we didn't map into the bio, if any | |
608 | */ | |
609 | while (i < nr_pages) | |
610 | page_cache_release(pages[i++]); | |
611 | ||
612 | kfree(pages); | |
613 | ||
614 | /* | |
615 | * set data direction, and check if mapped pages need bouncing | |
616 | */ | |
617 | if (!write_to_vm) | |
618 | bio->bi_rw |= (1 << BIO_RW); | |
619 | ||
620 | bio->bi_flags |= (1 << BIO_USER_MAPPED); | |
621 | return bio; | |
622 | out: | |
623 | kfree(pages); | |
624 | bio_put(bio); | |
625 | return ERR_PTR(ret); | |
626 | } | |
627 | ||
628 | /** | |
629 | * bio_map_user - map user address into bio | |
630 | * @q: the request_queue_t for the bio | |
631 | * @bdev: destination block device | |
632 | * @uaddr: start of user address | |
633 | * @len: length in bytes | |
634 | * @write_to_vm: bool indicating writing to pages or not | |
635 | * | |
636 | * Map the user space address into a bio suitable for io to a block | |
637 | * device. Returns an error pointer in case of error. | |
638 | */ | |
639 | struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev, | |
640 | unsigned long uaddr, unsigned int len, int write_to_vm) | |
641 | { | |
642 | struct bio *bio; | |
643 | ||
644 | bio = __bio_map_user(q, bdev, uaddr, len, write_to_vm); | |
645 | ||
646 | if (IS_ERR(bio)) | |
647 | return bio; | |
648 | ||
649 | /* | |
650 | * subtle -- if __bio_map_user() ended up bouncing a bio, | |
651 | * it would normally disappear when its bi_end_io is run. | |
652 | * however, we need it for the unmap, so grab an extra | |
653 | * reference to it | |
654 | */ | |
655 | bio_get(bio); | |
656 | ||
657 | if (bio->bi_size == len) | |
658 | return bio; | |
659 | ||
660 | /* | |
661 | * don't support partial mappings | |
662 | */ | |
663 | bio_endio(bio, bio->bi_size, 0); | |
664 | bio_unmap_user(bio); | |
665 | return ERR_PTR(-EINVAL); | |
666 | } | |
667 | ||
668 | static void __bio_unmap_user(struct bio *bio) | |
669 | { | |
670 | struct bio_vec *bvec; | |
671 | int i; | |
672 | ||
673 | /* | |
674 | * make sure we dirty pages we wrote to | |
675 | */ | |
676 | __bio_for_each_segment(bvec, bio, i, 0) { | |
677 | if (bio_data_dir(bio) == READ) | |
678 | set_page_dirty_lock(bvec->bv_page); | |
679 | ||
680 | page_cache_release(bvec->bv_page); | |
681 | } | |
682 | ||
683 | bio_put(bio); | |
684 | } | |
685 | ||
686 | /** | |
687 | * bio_unmap_user - unmap a bio | |
688 | * @bio: the bio being unmapped | |
689 | * | |
690 | * Unmap a bio previously mapped by bio_map_user(). Must be called with | |
691 | * a process context. | |
692 | * | |
693 | * bio_unmap_user() may sleep. | |
694 | */ | |
695 | void bio_unmap_user(struct bio *bio) | |
696 | { | |
697 | __bio_unmap_user(bio); | |
698 | bio_put(bio); | |
699 | } | |
700 | ||
701 | /* | |
702 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions | |
703 | * for performing direct-IO in BIOs. | |
704 | * | |
705 | * The problem is that we cannot run set_page_dirty() from interrupt context | |
706 | * because the required locks are not interrupt-safe. So what we can do is to | |
707 | * mark the pages dirty _before_ performing IO. And in interrupt context, | |
708 | * check that the pages are still dirty. If so, fine. If not, redirty them | |
709 | * in process context. | |
710 | * | |
711 | * We special-case compound pages here: normally this means reads into hugetlb | |
712 | * pages. The logic in here doesn't really work right for compound pages | |
713 | * because the VM does not uniformly chase down the head page in all cases. | |
714 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't | |
715 | * handle them at all. So we skip compound pages here at an early stage. | |
716 | * | |
717 | * Note that this code is very hard to test under normal circumstances because | |
718 | * direct-io pins the pages with get_user_pages(). This makes | |
719 | * is_page_cache_freeable return false, and the VM will not clean the pages. | |
720 | * But other code (eg, pdflush) could clean the pages if they are mapped | |
721 | * pagecache. | |
722 | * | |
723 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the | |
724 | * deferred bio dirtying paths. | |
725 | */ | |
726 | ||
727 | /* | |
728 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. | |
729 | */ | |
730 | void bio_set_pages_dirty(struct bio *bio) | |
731 | { | |
732 | struct bio_vec *bvec = bio->bi_io_vec; | |
733 | int i; | |
734 | ||
735 | for (i = 0; i < bio->bi_vcnt; i++) { | |
736 | struct page *page = bvec[i].bv_page; | |
737 | ||
738 | if (page && !PageCompound(page)) | |
739 | set_page_dirty_lock(page); | |
740 | } | |
741 | } | |
742 | ||
743 | static void bio_release_pages(struct bio *bio) | |
744 | { | |
745 | struct bio_vec *bvec = bio->bi_io_vec; | |
746 | int i; | |
747 | ||
748 | for (i = 0; i < bio->bi_vcnt; i++) { | |
749 | struct page *page = bvec[i].bv_page; | |
750 | ||
751 | if (page) | |
752 | put_page(page); | |
753 | } | |
754 | } | |
755 | ||
756 | /* | |
757 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. | |
758 | * If they are, then fine. If, however, some pages are clean then they must | |
759 | * have been written out during the direct-IO read. So we take another ref on | |
760 | * the BIO and the offending pages and re-dirty the pages in process context. | |
761 | * | |
762 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from | |
763 | * here on. It will run one page_cache_release() against each page and will | |
764 | * run one bio_put() against the BIO. | |
765 | */ | |
766 | ||
767 | static void bio_dirty_fn(void *data); | |
768 | ||
769 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn, NULL); | |
770 | static DEFINE_SPINLOCK(bio_dirty_lock); | |
771 | static struct bio *bio_dirty_list; | |
772 | ||
773 | /* | |
774 | * This runs in process context | |
775 | */ | |
776 | static void bio_dirty_fn(void *data) | |
777 | { | |
778 | unsigned long flags; | |
779 | struct bio *bio; | |
780 | ||
781 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
782 | bio = bio_dirty_list; | |
783 | bio_dirty_list = NULL; | |
784 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
785 | ||
786 | while (bio) { | |
787 | struct bio *next = bio->bi_private; | |
788 | ||
789 | bio_set_pages_dirty(bio); | |
790 | bio_release_pages(bio); | |
791 | bio_put(bio); | |
792 | bio = next; | |
793 | } | |
794 | } | |
795 | ||
796 | void bio_check_pages_dirty(struct bio *bio) | |
797 | { | |
798 | struct bio_vec *bvec = bio->bi_io_vec; | |
799 | int nr_clean_pages = 0; | |
800 | int i; | |
801 | ||
802 | for (i = 0; i < bio->bi_vcnt; i++) { | |
803 | struct page *page = bvec[i].bv_page; | |
804 | ||
805 | if (PageDirty(page) || PageCompound(page)) { | |
806 | page_cache_release(page); | |
807 | bvec[i].bv_page = NULL; | |
808 | } else { | |
809 | nr_clean_pages++; | |
810 | } | |
811 | } | |
812 | ||
813 | if (nr_clean_pages) { | |
814 | unsigned long flags; | |
815 | ||
816 | spin_lock_irqsave(&bio_dirty_lock, flags); | |
817 | bio->bi_private = bio_dirty_list; | |
818 | bio_dirty_list = bio; | |
819 | spin_unlock_irqrestore(&bio_dirty_lock, flags); | |
820 | schedule_work(&bio_dirty_work); | |
821 | } else { | |
822 | bio_put(bio); | |
823 | } | |
824 | } | |
825 | ||
826 | /** | |
827 | * bio_endio - end I/O on a bio | |
828 | * @bio: bio | |
829 | * @bytes_done: number of bytes completed | |
830 | * @error: error, if any | |
831 | * | |
832 | * Description: | |
833 | * bio_endio() will end I/O on @bytes_done number of bytes. This may be | |
834 | * just a partial part of the bio, or it may be the whole bio. bio_endio() | |
835 | * is the preferred way to end I/O on a bio, it takes care of decrementing | |
836 | * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and | |
837 | * and one of the established -Exxxx (-EIO, for instance) error values in | |
838 | * case something went wrong. Noone should call bi_end_io() directly on | |
839 | * a bio unless they own it and thus know that it has an end_io function. | |
840 | **/ | |
841 | void bio_endio(struct bio *bio, unsigned int bytes_done, int error) | |
842 | { | |
843 | if (error) | |
844 | clear_bit(BIO_UPTODATE, &bio->bi_flags); | |
845 | ||
846 | if (unlikely(bytes_done > bio->bi_size)) { | |
847 | printk("%s: want %u bytes done, only %u left\n", __FUNCTION__, | |
848 | bytes_done, bio->bi_size); | |
849 | bytes_done = bio->bi_size; | |
850 | } | |
851 | ||
852 | bio->bi_size -= bytes_done; | |
853 | bio->bi_sector += (bytes_done >> 9); | |
854 | ||
855 | if (bio->bi_end_io) | |
856 | bio->bi_end_io(bio, bytes_done, error); | |
857 | } | |
858 | ||
859 | void bio_pair_release(struct bio_pair *bp) | |
860 | { | |
861 | if (atomic_dec_and_test(&bp->cnt)) { | |
862 | struct bio *master = bp->bio1.bi_private; | |
863 | ||
864 | bio_endio(master, master->bi_size, bp->error); | |
865 | mempool_free(bp, bp->bio2.bi_private); | |
866 | } | |
867 | } | |
868 | ||
869 | static int bio_pair_end_1(struct bio * bi, unsigned int done, int err) | |
870 | { | |
871 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); | |
872 | ||
873 | if (err) | |
874 | bp->error = err; | |
875 | ||
876 | if (bi->bi_size) | |
877 | return 1; | |
878 | ||
879 | bio_pair_release(bp); | |
880 | return 0; | |
881 | } | |
882 | ||
883 | static int bio_pair_end_2(struct bio * bi, unsigned int done, int err) | |
884 | { | |
885 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); | |
886 | ||
887 | if (err) | |
888 | bp->error = err; | |
889 | ||
890 | if (bi->bi_size) | |
891 | return 1; | |
892 | ||
893 | bio_pair_release(bp); | |
894 | return 0; | |
895 | } | |
896 | ||
897 | /* | |
898 | * split a bio - only worry about a bio with a single page | |
899 | * in it's iovec | |
900 | */ | |
901 | struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors) | |
902 | { | |
903 | struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO); | |
904 | ||
905 | if (!bp) | |
906 | return bp; | |
907 | ||
908 | BUG_ON(bi->bi_vcnt != 1); | |
909 | BUG_ON(bi->bi_idx != 0); | |
910 | atomic_set(&bp->cnt, 3); | |
911 | bp->error = 0; | |
912 | bp->bio1 = *bi; | |
913 | bp->bio2 = *bi; | |
914 | bp->bio2.bi_sector += first_sectors; | |
915 | bp->bio2.bi_size -= first_sectors << 9; | |
916 | bp->bio1.bi_size = first_sectors << 9; | |
917 | ||
918 | bp->bv1 = bi->bi_io_vec[0]; | |
919 | bp->bv2 = bi->bi_io_vec[0]; | |
920 | bp->bv2.bv_offset += first_sectors << 9; | |
921 | bp->bv2.bv_len -= first_sectors << 9; | |
922 | bp->bv1.bv_len = first_sectors << 9; | |
923 | ||
924 | bp->bio1.bi_io_vec = &bp->bv1; | |
925 | bp->bio2.bi_io_vec = &bp->bv2; | |
926 | ||
927 | bp->bio1.bi_end_io = bio_pair_end_1; | |
928 | bp->bio2.bi_end_io = bio_pair_end_2; | |
929 | ||
930 | bp->bio1.bi_private = bi; | |
931 | bp->bio2.bi_private = pool; | |
932 | ||
933 | return bp; | |
934 | } | |
935 | ||
936 | static void *bio_pair_alloc(unsigned int __nocast gfp_flags, void *data) | |
937 | { | |
938 | return kmalloc(sizeof(struct bio_pair), gfp_flags); | |
939 | } | |
940 | ||
941 | static void bio_pair_free(void *bp, void *data) | |
942 | { | |
943 | kfree(bp); | |
944 | } | |
945 | ||
946 | ||
947 | /* | |
948 | * create memory pools for biovec's in a bio_set. | |
949 | * use the global biovec slabs created for general use. | |
950 | */ | |
951 | static int biovec_create_pools(struct bio_set *bs, int pool_entries, int scale) | |
952 | { | |
953 | int i; | |
954 | ||
955 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
956 | struct biovec_slab *bp = bvec_slabs + i; | |
957 | mempool_t **bvp = bs->bvec_pools + i; | |
958 | ||
959 | if (i >= scale) | |
960 | pool_entries >>= 1; | |
961 | ||
962 | *bvp = mempool_create(pool_entries, mempool_alloc_slab, | |
963 | mempool_free_slab, bp->slab); | |
964 | if (!*bvp) | |
965 | return -ENOMEM; | |
966 | } | |
967 | return 0; | |
968 | } | |
969 | ||
970 | static void biovec_free_pools(struct bio_set *bs) | |
971 | { | |
972 | int i; | |
973 | ||
974 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
975 | mempool_t *bvp = bs->bvec_pools[i]; | |
976 | ||
977 | if (bvp) | |
978 | mempool_destroy(bvp); | |
979 | } | |
980 | ||
981 | } | |
982 | ||
983 | void bioset_free(struct bio_set *bs) | |
984 | { | |
985 | if (bs->bio_pool) | |
986 | mempool_destroy(bs->bio_pool); | |
987 | ||
988 | biovec_free_pools(bs); | |
989 | ||
990 | kfree(bs); | |
991 | } | |
992 | ||
993 | struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size, int scale) | |
994 | { | |
995 | struct bio_set *bs = kmalloc(sizeof(*bs), GFP_KERNEL); | |
996 | ||
997 | if (!bs) | |
998 | return NULL; | |
999 | ||
1000 | memset(bs, 0, sizeof(*bs)); | |
1001 | bs->bio_pool = mempool_create(bio_pool_size, mempool_alloc_slab, | |
1002 | mempool_free_slab, bio_slab); | |
1003 | ||
1004 | if (!bs->bio_pool) | |
1005 | goto bad; | |
1006 | ||
1007 | if (!biovec_create_pools(bs, bvec_pool_size, scale)) | |
1008 | return bs; | |
1009 | ||
1010 | bad: | |
1011 | bioset_free(bs); | |
1012 | return NULL; | |
1013 | } | |
1014 | ||
1015 | static void __init biovec_init_slabs(void) | |
1016 | { | |
1017 | int i; | |
1018 | ||
1019 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { | |
1020 | int size; | |
1021 | struct biovec_slab *bvs = bvec_slabs + i; | |
1022 | ||
1023 | size = bvs->nr_vecs * sizeof(struct bio_vec); | |
1024 | bvs->slab = kmem_cache_create(bvs->name, size, 0, | |
1025 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); | |
1026 | } | |
1027 | } | |
1028 | ||
1029 | static int __init init_bio(void) | |
1030 | { | |
1031 | int megabytes, bvec_pool_entries; | |
1032 | int scale = BIOVEC_NR_POOLS; | |
1033 | ||
1034 | bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0, | |
1035 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); | |
1036 | ||
1037 | biovec_init_slabs(); | |
1038 | ||
1039 | megabytes = nr_free_pages() >> (20 - PAGE_SHIFT); | |
1040 | ||
1041 | /* | |
1042 | * find out where to start scaling | |
1043 | */ | |
1044 | if (megabytes <= 16) | |
1045 | scale = 0; | |
1046 | else if (megabytes <= 32) | |
1047 | scale = 1; | |
1048 | else if (megabytes <= 64) | |
1049 | scale = 2; | |
1050 | else if (megabytes <= 96) | |
1051 | scale = 3; | |
1052 | else if (megabytes <= 128) | |
1053 | scale = 4; | |
1054 | ||
1055 | /* | |
1056 | * scale number of entries | |
1057 | */ | |
1058 | bvec_pool_entries = megabytes * 2; | |
1059 | if (bvec_pool_entries > 256) | |
1060 | bvec_pool_entries = 256; | |
1061 | ||
1062 | fs_bio_set = bioset_create(BIO_POOL_SIZE, bvec_pool_entries, scale); | |
1063 | if (!fs_bio_set) | |
1064 | panic("bio: can't allocate bios\n"); | |
1065 | ||
1066 | bio_split_pool = mempool_create(BIO_SPLIT_ENTRIES, | |
1067 | bio_pair_alloc, bio_pair_free, NULL); | |
1068 | if (!bio_split_pool) | |
1069 | panic("bio: can't create split pool\n"); | |
1070 | ||
1071 | return 0; | |
1072 | } | |
1073 | ||
1074 | subsys_initcall(init_bio); | |
1075 | ||
1076 | EXPORT_SYMBOL(bio_alloc); | |
1077 | EXPORT_SYMBOL(bio_put); | |
1078 | EXPORT_SYMBOL(bio_endio); | |
1079 | EXPORT_SYMBOL(bio_init); | |
1080 | EXPORT_SYMBOL(__bio_clone); | |
1081 | EXPORT_SYMBOL(bio_clone); | |
1082 | EXPORT_SYMBOL(bio_phys_segments); | |
1083 | EXPORT_SYMBOL(bio_hw_segments); | |
1084 | EXPORT_SYMBOL(bio_add_page); | |
1085 | EXPORT_SYMBOL(bio_get_nr_vecs); | |
1086 | EXPORT_SYMBOL(bio_map_user); | |
1087 | EXPORT_SYMBOL(bio_unmap_user); | |
1088 | EXPORT_SYMBOL(bio_pair_release); | |
1089 | EXPORT_SYMBOL(bio_split); | |
1090 | EXPORT_SYMBOL(bio_split_pool); | |
1091 | EXPORT_SYMBOL(bio_copy_user); | |
1092 | EXPORT_SYMBOL(bio_uncopy_user); | |
1093 | EXPORT_SYMBOL(bioset_create); | |
1094 | EXPORT_SYMBOL(bioset_free); | |
1095 | EXPORT_SYMBOL(bio_alloc_bioset); |