Contents of /alx-src/tags/kernel26-2.6.12-alx-r9/fs/bio.c
Parent Directory | Revision Log
Revision 630 -
(show annotations)
(download)
Wed Mar 4 11:03:09 2009 UTC (15 years, 6 months ago) by niro
File MIME type: text/plain
File size: 26069 byte(s)
Wed Mar 4 11:03:09 2009 UTC (15 years, 6 months ago) by niro
File MIME type: text/plain
File size: 26069 byte(s)
Tag kernel26-2.6.12-alx-r9
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] = { |
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, bio_src->bi_max_vecs * sizeof(struct bio_vec)); |
252 | |
253 | bio->bi_sector = bio_src->bi_sector; |
254 | bio->bi_bdev = bio_src->bi_bdev; |
255 | bio->bi_flags |= 1 << BIO_CLONED; |
256 | bio->bi_rw = bio_src->bi_rw; |
257 | |
258 | /* |
259 | * notes -- maybe just leave bi_idx alone. assume identical mapping |
260 | * for the clone |
261 | */ |
262 | bio->bi_vcnt = bio_src->bi_vcnt; |
263 | bio->bi_size = bio_src->bi_size; |
264 | bio->bi_idx = bio_src->bi_idx; |
265 | bio_phys_segments(q, bio); |
266 | bio_hw_segments(q, bio); |
267 | } |
268 | |
269 | /** |
270 | * bio_clone - clone a bio |
271 | * @bio: bio to clone |
272 | * @gfp_mask: allocation priority |
273 | * |
274 | * Like __bio_clone, only also allocates the returned bio |
275 | */ |
276 | struct bio *bio_clone(struct bio *bio, unsigned int __nocast gfp_mask) |
277 | { |
278 | struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set); |
279 | |
280 | if (b) |
281 | __bio_clone(b, bio); |
282 | |
283 | return b; |
284 | } |
285 | |
286 | /** |
287 | * bio_get_nr_vecs - return approx number of vecs |
288 | * @bdev: I/O target |
289 | * |
290 | * Return the approximate number of pages we can send to this target. |
291 | * There's no guarantee that you will be able to fit this number of pages |
292 | * into a bio, it does not account for dynamic restrictions that vary |
293 | * on offset. |
294 | */ |
295 | int bio_get_nr_vecs(struct block_device *bdev) |
296 | { |
297 | request_queue_t *q = bdev_get_queue(bdev); |
298 | int nr_pages; |
299 | |
300 | nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT; |
301 | if (nr_pages > q->max_phys_segments) |
302 | nr_pages = q->max_phys_segments; |
303 | if (nr_pages > q->max_hw_segments) |
304 | nr_pages = q->max_hw_segments; |
305 | |
306 | return nr_pages; |
307 | } |
308 | |
309 | static int __bio_add_page(request_queue_t *q, struct bio *bio, struct page |
310 | *page, unsigned int len, unsigned int offset) |
311 | { |
312 | int retried_segments = 0; |
313 | struct bio_vec *bvec; |
314 | |
315 | /* |
316 | * cloned bio must not modify vec list |
317 | */ |
318 | if (unlikely(bio_flagged(bio, BIO_CLONED))) |
319 | return 0; |
320 | |
321 | if (bio->bi_vcnt >= bio->bi_max_vecs) |
322 | return 0; |
323 | |
324 | if (((bio->bi_size + len) >> 9) > q->max_sectors) |
325 | return 0; |
326 | |
327 | /* |
328 | * we might lose a segment or two here, but rather that than |
329 | * make this too complex. |
330 | */ |
331 | |
332 | while (bio->bi_phys_segments >= q->max_phys_segments |
333 | || bio->bi_hw_segments >= q->max_hw_segments |
334 | || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) { |
335 | |
336 | if (retried_segments) |
337 | return 0; |
338 | |
339 | retried_segments = 1; |
340 | blk_recount_segments(q, bio); |
341 | } |
342 | |
343 | /* |
344 | * setup the new entry, we might clear it again later if we |
345 | * cannot add the page |
346 | */ |
347 | bvec = &bio->bi_io_vec[bio->bi_vcnt]; |
348 | bvec->bv_page = page; |
349 | bvec->bv_len = len; |
350 | bvec->bv_offset = offset; |
351 | |
352 | /* |
353 | * if queue has other restrictions (eg varying max sector size |
354 | * depending on offset), it can specify a merge_bvec_fn in the |
355 | * queue to get further control |
356 | */ |
357 | if (q->merge_bvec_fn) { |
358 | /* |
359 | * merge_bvec_fn() returns number of bytes it can accept |
360 | * at this offset |
361 | */ |
362 | if (q->merge_bvec_fn(q, bio, bvec) < len) { |
363 | bvec->bv_page = NULL; |
364 | bvec->bv_len = 0; |
365 | bvec->bv_offset = 0; |
366 | return 0; |
367 | } |
368 | } |
369 | |
370 | /* If we may be able to merge these biovecs, force a recount */ |
371 | if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) || |
372 | BIOVEC_VIRT_MERGEABLE(bvec-1, bvec))) |
373 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); |
374 | |
375 | bio->bi_vcnt++; |
376 | bio->bi_phys_segments++; |
377 | bio->bi_hw_segments++; |
378 | bio->bi_size += len; |
379 | return len; |
380 | } |
381 | |
382 | /** |
383 | * bio_add_page - attempt to add page to bio |
384 | * @bio: destination bio |
385 | * @page: page to add |
386 | * @len: vec entry length |
387 | * @offset: vec entry offset |
388 | * |
389 | * Attempt to add a page to the bio_vec maplist. This can fail for a |
390 | * number of reasons, such as the bio being full or target block |
391 | * device limitations. The target block device must allow bio's |
392 | * smaller than PAGE_SIZE, so it is always possible to add a single |
393 | * page to an empty bio. |
394 | */ |
395 | int bio_add_page(struct bio *bio, struct page *page, unsigned int len, |
396 | unsigned int offset) |
397 | { |
398 | return __bio_add_page(bdev_get_queue(bio->bi_bdev), bio, page, |
399 | len, offset); |
400 | } |
401 | |
402 | struct bio_map_data { |
403 | struct bio_vec *iovecs; |
404 | void __user *userptr; |
405 | }; |
406 | |
407 | static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio) |
408 | { |
409 | memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt); |
410 | bio->bi_private = bmd; |
411 | } |
412 | |
413 | static void bio_free_map_data(struct bio_map_data *bmd) |
414 | { |
415 | kfree(bmd->iovecs); |
416 | kfree(bmd); |
417 | } |
418 | |
419 | static struct bio_map_data *bio_alloc_map_data(int nr_segs) |
420 | { |
421 | struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL); |
422 | |
423 | if (!bmd) |
424 | return NULL; |
425 | |
426 | bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL); |
427 | if (bmd->iovecs) |
428 | return bmd; |
429 | |
430 | kfree(bmd); |
431 | return NULL; |
432 | } |
433 | |
434 | /** |
435 | * bio_uncopy_user - finish previously mapped bio |
436 | * @bio: bio being terminated |
437 | * |
438 | * Free pages allocated from bio_copy_user() and write back data |
439 | * to user space in case of a read. |
440 | */ |
441 | int bio_uncopy_user(struct bio *bio) |
442 | { |
443 | struct bio_map_data *bmd = bio->bi_private; |
444 | const int read = bio_data_dir(bio) == READ; |
445 | struct bio_vec *bvec; |
446 | int i, ret = 0; |
447 | |
448 | __bio_for_each_segment(bvec, bio, i, 0) { |
449 | char *addr = page_address(bvec->bv_page); |
450 | unsigned int len = bmd->iovecs[i].bv_len; |
451 | |
452 | if (read && !ret && copy_to_user(bmd->userptr, addr, len)) |
453 | ret = -EFAULT; |
454 | |
455 | __free_page(bvec->bv_page); |
456 | bmd->userptr += len; |
457 | } |
458 | bio_free_map_data(bmd); |
459 | bio_put(bio); |
460 | return ret; |
461 | } |
462 | |
463 | /** |
464 | * bio_copy_user - copy user data to bio |
465 | * @q: destination block queue |
466 | * @uaddr: start of user address |
467 | * @len: length in bytes |
468 | * @write_to_vm: bool indicating writing to pages or not |
469 | * |
470 | * Prepares and returns a bio for indirect user io, bouncing data |
471 | * to/from kernel pages as necessary. Must be paired with |
472 | * call bio_uncopy_user() on io completion. |
473 | */ |
474 | struct bio *bio_copy_user(request_queue_t *q, unsigned long uaddr, |
475 | unsigned int len, int write_to_vm) |
476 | { |
477 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
478 | unsigned long start = uaddr >> PAGE_SHIFT; |
479 | struct bio_map_data *bmd; |
480 | struct bio_vec *bvec; |
481 | struct page *page; |
482 | struct bio *bio; |
483 | int i, ret; |
484 | |
485 | bmd = bio_alloc_map_data(end - start); |
486 | if (!bmd) |
487 | return ERR_PTR(-ENOMEM); |
488 | |
489 | bmd->userptr = (void __user *) uaddr; |
490 | |
491 | ret = -ENOMEM; |
492 | bio = bio_alloc(GFP_KERNEL, end - start); |
493 | if (!bio) |
494 | goto out_bmd; |
495 | |
496 | bio->bi_rw |= (!write_to_vm << BIO_RW); |
497 | |
498 | ret = 0; |
499 | while (len) { |
500 | unsigned int bytes = PAGE_SIZE; |
501 | |
502 | if (bytes > len) |
503 | bytes = len; |
504 | |
505 | page = alloc_page(q->bounce_gfp | GFP_KERNEL); |
506 | if (!page) { |
507 | ret = -ENOMEM; |
508 | break; |
509 | } |
510 | |
511 | if (__bio_add_page(q, bio, page, bytes, 0) < bytes) { |
512 | ret = -EINVAL; |
513 | break; |
514 | } |
515 | |
516 | len -= bytes; |
517 | } |
518 | |
519 | if (ret) |
520 | goto cleanup; |
521 | |
522 | /* |
523 | * success |
524 | */ |
525 | if (!write_to_vm) { |
526 | char __user *p = (char __user *) uaddr; |
527 | |
528 | /* |
529 | * for a write, copy in data to kernel pages |
530 | */ |
531 | ret = -EFAULT; |
532 | bio_for_each_segment(bvec, bio, i) { |
533 | char *addr = page_address(bvec->bv_page); |
534 | |
535 | if (copy_from_user(addr, p, bvec->bv_len)) |
536 | goto cleanup; |
537 | p += bvec->bv_len; |
538 | } |
539 | } |
540 | |
541 | bio_set_map_data(bmd, bio); |
542 | return bio; |
543 | cleanup: |
544 | bio_for_each_segment(bvec, bio, i) |
545 | __free_page(bvec->bv_page); |
546 | |
547 | bio_put(bio); |
548 | out_bmd: |
549 | bio_free_map_data(bmd); |
550 | return ERR_PTR(ret); |
551 | } |
552 | |
553 | static struct bio *__bio_map_user(request_queue_t *q, struct block_device *bdev, |
554 | unsigned long uaddr, unsigned int len, |
555 | int write_to_vm) |
556 | { |
557 | unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
558 | unsigned long start = uaddr >> PAGE_SHIFT; |
559 | const int nr_pages = end - start; |
560 | int ret, offset, i; |
561 | struct page **pages; |
562 | struct bio *bio; |
563 | |
564 | /* |
565 | * transfer and buffer must be aligned to at least hardsector |
566 | * size for now, in the future we can relax this restriction |
567 | */ |
568 | if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q))) |
569 | return ERR_PTR(-EINVAL); |
570 | |
571 | bio = bio_alloc(GFP_KERNEL, nr_pages); |
572 | if (!bio) |
573 | return ERR_PTR(-ENOMEM); |
574 | |
575 | ret = -ENOMEM; |
576 | pages = kmalloc(nr_pages * sizeof(struct page *), GFP_KERNEL); |
577 | if (!pages) |
578 | goto out; |
579 | |
580 | down_read(¤t->mm->mmap_sem); |
581 | ret = get_user_pages(current, current->mm, uaddr, nr_pages, |
582 | write_to_vm, 0, pages, NULL); |
583 | up_read(¤t->mm->mmap_sem); |
584 | |
585 | if (ret < nr_pages) |
586 | goto out; |
587 | |
588 | bio->bi_bdev = bdev; |
589 | |
590 | offset = uaddr & ~PAGE_MASK; |
591 | for (i = 0; i < nr_pages; i++) { |
592 | unsigned int bytes = PAGE_SIZE - offset; |
593 | |
594 | if (len <= 0) |
595 | break; |
596 | |
597 | if (bytes > len) |
598 | bytes = len; |
599 | |
600 | /* |
601 | * sorry... |
602 | */ |
603 | if (__bio_add_page(q, bio, pages[i], bytes, offset) < bytes) |
604 | break; |
605 | |
606 | len -= bytes; |
607 | offset = 0; |
608 | } |
609 | |
610 | /* |
611 | * release the pages we didn't map into the bio, if any |
612 | */ |
613 | while (i < nr_pages) |
614 | page_cache_release(pages[i++]); |
615 | |
616 | kfree(pages); |
617 | |
618 | /* |
619 | * set data direction, and check if mapped pages need bouncing |
620 | */ |
621 | if (!write_to_vm) |
622 | bio->bi_rw |= (1 << BIO_RW); |
623 | |
624 | bio->bi_flags |= (1 << BIO_USER_MAPPED); |
625 | return bio; |
626 | out: |
627 | kfree(pages); |
628 | bio_put(bio); |
629 | return ERR_PTR(ret); |
630 | } |
631 | |
632 | /** |
633 | * bio_map_user - map user address into bio |
634 | * @q: the request_queue_t for the bio |
635 | * @bdev: destination block device |
636 | * @uaddr: start of user address |
637 | * @len: length in bytes |
638 | * @write_to_vm: bool indicating writing to pages or not |
639 | * |
640 | * Map the user space address into a bio suitable for io to a block |
641 | * device. Returns an error pointer in case of error. |
642 | */ |
643 | struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev, |
644 | unsigned long uaddr, unsigned int len, int write_to_vm) |
645 | { |
646 | struct bio *bio; |
647 | |
648 | bio = __bio_map_user(q, bdev, uaddr, len, write_to_vm); |
649 | |
650 | if (IS_ERR(bio)) |
651 | return bio; |
652 | |
653 | /* |
654 | * subtle -- if __bio_map_user() ended up bouncing a bio, |
655 | * it would normally disappear when its bi_end_io is run. |
656 | * however, we need it for the unmap, so grab an extra |
657 | * reference to it |
658 | */ |
659 | bio_get(bio); |
660 | |
661 | if (bio->bi_size == len) |
662 | return bio; |
663 | |
664 | /* |
665 | * don't support partial mappings |
666 | */ |
667 | bio_endio(bio, bio->bi_size, 0); |
668 | bio_unmap_user(bio); |
669 | return ERR_PTR(-EINVAL); |
670 | } |
671 | |
672 | static void __bio_unmap_user(struct bio *bio) |
673 | { |
674 | struct bio_vec *bvec; |
675 | int i; |
676 | |
677 | /* |
678 | * make sure we dirty pages we wrote to |
679 | */ |
680 | __bio_for_each_segment(bvec, bio, i, 0) { |
681 | if (bio_data_dir(bio) == READ) |
682 | set_page_dirty_lock(bvec->bv_page); |
683 | |
684 | page_cache_release(bvec->bv_page); |
685 | } |
686 | |
687 | bio_put(bio); |
688 | } |
689 | |
690 | /** |
691 | * bio_unmap_user - unmap a bio |
692 | * @bio: the bio being unmapped |
693 | * |
694 | * Unmap a bio previously mapped by bio_map_user(). Must be called with |
695 | * a process context. |
696 | * |
697 | * bio_unmap_user() may sleep. |
698 | */ |
699 | void bio_unmap_user(struct bio *bio) |
700 | { |
701 | __bio_unmap_user(bio); |
702 | bio_put(bio); |
703 | } |
704 | |
705 | /* |
706 | * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions |
707 | * for performing direct-IO in BIOs. |
708 | * |
709 | * The problem is that we cannot run set_page_dirty() from interrupt context |
710 | * because the required locks are not interrupt-safe. So what we can do is to |
711 | * mark the pages dirty _before_ performing IO. And in interrupt context, |
712 | * check that the pages are still dirty. If so, fine. If not, redirty them |
713 | * in process context. |
714 | * |
715 | * We special-case compound pages here: normally this means reads into hugetlb |
716 | * pages. The logic in here doesn't really work right for compound pages |
717 | * because the VM does not uniformly chase down the head page in all cases. |
718 | * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't |
719 | * handle them at all. So we skip compound pages here at an early stage. |
720 | * |
721 | * Note that this code is very hard to test under normal circumstances because |
722 | * direct-io pins the pages with get_user_pages(). This makes |
723 | * is_page_cache_freeable return false, and the VM will not clean the pages. |
724 | * But other code (eg, pdflush) could clean the pages if they are mapped |
725 | * pagecache. |
726 | * |
727 | * Simply disabling the call to bio_set_pages_dirty() is a good way to test the |
728 | * deferred bio dirtying paths. |
729 | */ |
730 | |
731 | /* |
732 | * bio_set_pages_dirty() will mark all the bio's pages as dirty. |
733 | */ |
734 | void bio_set_pages_dirty(struct bio *bio) |
735 | { |
736 | struct bio_vec *bvec = bio->bi_io_vec; |
737 | int i; |
738 | |
739 | for (i = 0; i < bio->bi_vcnt; i++) { |
740 | struct page *page = bvec[i].bv_page; |
741 | |
742 | if (page && !PageCompound(page)) |
743 | set_page_dirty_lock(page); |
744 | } |
745 | } |
746 | |
747 | static void bio_release_pages(struct bio *bio) |
748 | { |
749 | struct bio_vec *bvec = bio->bi_io_vec; |
750 | int i; |
751 | |
752 | for (i = 0; i < bio->bi_vcnt; i++) { |
753 | struct page *page = bvec[i].bv_page; |
754 | |
755 | if (page) |
756 | put_page(page); |
757 | } |
758 | } |
759 | |
760 | /* |
761 | * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. |
762 | * If they are, then fine. If, however, some pages are clean then they must |
763 | * have been written out during the direct-IO read. So we take another ref on |
764 | * the BIO and the offending pages and re-dirty the pages in process context. |
765 | * |
766 | * It is expected that bio_check_pages_dirty() will wholly own the BIO from |
767 | * here on. It will run one page_cache_release() against each page and will |
768 | * run one bio_put() against the BIO. |
769 | */ |
770 | |
771 | static void bio_dirty_fn(void *data); |
772 | |
773 | static DECLARE_WORK(bio_dirty_work, bio_dirty_fn, NULL); |
774 | static DEFINE_SPINLOCK(bio_dirty_lock); |
775 | static struct bio *bio_dirty_list; |
776 | |
777 | /* |
778 | * This runs in process context |
779 | */ |
780 | static void bio_dirty_fn(void *data) |
781 | { |
782 | unsigned long flags; |
783 | struct bio *bio; |
784 | |
785 | spin_lock_irqsave(&bio_dirty_lock, flags); |
786 | bio = bio_dirty_list; |
787 | bio_dirty_list = NULL; |
788 | spin_unlock_irqrestore(&bio_dirty_lock, flags); |
789 | |
790 | while (bio) { |
791 | struct bio *next = bio->bi_private; |
792 | |
793 | bio_set_pages_dirty(bio); |
794 | bio_release_pages(bio); |
795 | bio_put(bio); |
796 | bio = next; |
797 | } |
798 | } |
799 | |
800 | void bio_check_pages_dirty(struct bio *bio) |
801 | { |
802 | struct bio_vec *bvec = bio->bi_io_vec; |
803 | int nr_clean_pages = 0; |
804 | int i; |
805 | |
806 | for (i = 0; i < bio->bi_vcnt; i++) { |
807 | struct page *page = bvec[i].bv_page; |
808 | |
809 | if (PageDirty(page) || PageCompound(page)) { |
810 | page_cache_release(page); |
811 | bvec[i].bv_page = NULL; |
812 | } else { |
813 | nr_clean_pages++; |
814 | } |
815 | } |
816 | |
817 | if (nr_clean_pages) { |
818 | unsigned long flags; |
819 | |
820 | spin_lock_irqsave(&bio_dirty_lock, flags); |
821 | bio->bi_private = bio_dirty_list; |
822 | bio_dirty_list = bio; |
823 | spin_unlock_irqrestore(&bio_dirty_lock, flags); |
824 | schedule_work(&bio_dirty_work); |
825 | } else { |
826 | bio_put(bio); |
827 | } |
828 | } |
829 | |
830 | /** |
831 | * bio_endio - end I/O on a bio |
832 | * @bio: bio |
833 | * @bytes_done: number of bytes completed |
834 | * @error: error, if any |
835 | * |
836 | * Description: |
837 | * bio_endio() will end I/O on @bytes_done number of bytes. This may be |
838 | * just a partial part of the bio, or it may be the whole bio. bio_endio() |
839 | * is the preferred way to end I/O on a bio, it takes care of decrementing |
840 | * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and |
841 | * and one of the established -Exxxx (-EIO, for instance) error values in |
842 | * case something went wrong. Noone should call bi_end_io() directly on |
843 | * a bio unless they own it and thus know that it has an end_io function. |
844 | **/ |
845 | void bio_endio(struct bio *bio, unsigned int bytes_done, int error) |
846 | { |
847 | if (error) |
848 | clear_bit(BIO_UPTODATE, &bio->bi_flags); |
849 | |
850 | if (unlikely(bytes_done > bio->bi_size)) { |
851 | printk("%s: want %u bytes done, only %u left\n", __FUNCTION__, |
852 | bytes_done, bio->bi_size); |
853 | bytes_done = bio->bi_size; |
854 | } |
855 | |
856 | bio->bi_size -= bytes_done; |
857 | bio->bi_sector += (bytes_done >> 9); |
858 | |
859 | if (bio->bi_end_io) |
860 | bio->bi_end_io(bio, bytes_done, error); |
861 | } |
862 | |
863 | void bio_pair_release(struct bio_pair *bp) |
864 | { |
865 | if (atomic_dec_and_test(&bp->cnt)) { |
866 | struct bio *master = bp->bio1.bi_private; |
867 | |
868 | bio_endio(master, master->bi_size, bp->error); |
869 | mempool_free(bp, bp->bio2.bi_private); |
870 | } |
871 | } |
872 | |
873 | static int bio_pair_end_1(struct bio * bi, unsigned int done, int err) |
874 | { |
875 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio1); |
876 | |
877 | if (err) |
878 | bp->error = err; |
879 | |
880 | if (bi->bi_size) |
881 | return 1; |
882 | |
883 | bio_pair_release(bp); |
884 | return 0; |
885 | } |
886 | |
887 | static int bio_pair_end_2(struct bio * bi, unsigned int done, int err) |
888 | { |
889 | struct bio_pair *bp = container_of(bi, struct bio_pair, bio2); |
890 | |
891 | if (err) |
892 | bp->error = err; |
893 | |
894 | if (bi->bi_size) |
895 | return 1; |
896 | |
897 | bio_pair_release(bp); |
898 | return 0; |
899 | } |
900 | |
901 | /* |
902 | * split a bio - only worry about a bio with a single page |
903 | * in it's iovec |
904 | */ |
905 | struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors) |
906 | { |
907 | struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO); |
908 | |
909 | if (!bp) |
910 | return bp; |
911 | |
912 | BUG_ON(bi->bi_vcnt != 1); |
913 | BUG_ON(bi->bi_idx != 0); |
914 | atomic_set(&bp->cnt, 3); |
915 | bp->error = 0; |
916 | bp->bio1 = *bi; |
917 | bp->bio2 = *bi; |
918 | bp->bio2.bi_sector += first_sectors; |
919 | bp->bio2.bi_size -= first_sectors << 9; |
920 | bp->bio1.bi_size = first_sectors << 9; |
921 | |
922 | bp->bv1 = bi->bi_io_vec[0]; |
923 | bp->bv2 = bi->bi_io_vec[0]; |
924 | bp->bv2.bv_offset += first_sectors << 9; |
925 | bp->bv2.bv_len -= first_sectors << 9; |
926 | bp->bv1.bv_len = first_sectors << 9; |
927 | |
928 | bp->bio1.bi_io_vec = &bp->bv1; |
929 | bp->bio2.bi_io_vec = &bp->bv2; |
930 | |
931 | bp->bio1.bi_end_io = bio_pair_end_1; |
932 | bp->bio2.bi_end_io = bio_pair_end_2; |
933 | |
934 | bp->bio1.bi_private = bi; |
935 | bp->bio2.bi_private = pool; |
936 | |
937 | return bp; |
938 | } |
939 | |
940 | static void *bio_pair_alloc(unsigned int __nocast gfp_flags, void *data) |
941 | { |
942 | return kmalloc(sizeof(struct bio_pair), gfp_flags); |
943 | } |
944 | |
945 | static void bio_pair_free(void *bp, void *data) |
946 | { |
947 | kfree(bp); |
948 | } |
949 | |
950 | |
951 | /* |
952 | * create memory pools for biovec's in a bio_set. |
953 | * use the global biovec slabs created for general use. |
954 | */ |
955 | static int biovec_create_pools(struct bio_set *bs, int pool_entries, int scale) |
956 | { |
957 | int i; |
958 | |
959 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { |
960 | struct biovec_slab *bp = bvec_slabs + i; |
961 | mempool_t **bvp = bs->bvec_pools + i; |
962 | |
963 | if (i >= scale) |
964 | pool_entries >>= 1; |
965 | |
966 | *bvp = mempool_create(pool_entries, mempool_alloc_slab, |
967 | mempool_free_slab, bp->slab); |
968 | if (!*bvp) |
969 | return -ENOMEM; |
970 | } |
971 | return 0; |
972 | } |
973 | |
974 | static void biovec_free_pools(struct bio_set *bs) |
975 | { |
976 | int i; |
977 | |
978 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { |
979 | mempool_t *bvp = bs->bvec_pools[i]; |
980 | |
981 | if (bvp) |
982 | mempool_destroy(bvp); |
983 | } |
984 | |
985 | } |
986 | |
987 | void bioset_free(struct bio_set *bs) |
988 | { |
989 | if (bs->bio_pool) |
990 | mempool_destroy(bs->bio_pool); |
991 | |
992 | biovec_free_pools(bs); |
993 | |
994 | kfree(bs); |
995 | } |
996 | |
997 | struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size, int scale) |
998 | { |
999 | struct bio_set *bs = kmalloc(sizeof(*bs), GFP_KERNEL); |
1000 | |
1001 | if (!bs) |
1002 | return NULL; |
1003 | |
1004 | memset(bs, 0, sizeof(*bs)); |
1005 | bs->bio_pool = mempool_create(bio_pool_size, mempool_alloc_slab, |
1006 | mempool_free_slab, bio_slab); |
1007 | |
1008 | if (!bs->bio_pool) |
1009 | goto bad; |
1010 | |
1011 | if (!biovec_create_pools(bs, bvec_pool_size, scale)) |
1012 | return bs; |
1013 | |
1014 | bad: |
1015 | bioset_free(bs); |
1016 | return NULL; |
1017 | } |
1018 | |
1019 | static void __init biovec_init_slabs(void) |
1020 | { |
1021 | int i; |
1022 | |
1023 | for (i = 0; i < BIOVEC_NR_POOLS; i++) { |
1024 | int size; |
1025 | struct biovec_slab *bvs = bvec_slabs + i; |
1026 | |
1027 | size = bvs->nr_vecs * sizeof(struct bio_vec); |
1028 | bvs->slab = kmem_cache_create(bvs->name, size, 0, |
1029 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); |
1030 | } |
1031 | } |
1032 | |
1033 | static int __init init_bio(void) |
1034 | { |
1035 | int megabytes, bvec_pool_entries; |
1036 | int scale = BIOVEC_NR_POOLS; |
1037 | |
1038 | bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0, |
1039 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); |
1040 | |
1041 | biovec_init_slabs(); |
1042 | |
1043 | megabytes = nr_free_pages() >> (20 - PAGE_SHIFT); |
1044 | |
1045 | /* |
1046 | * find out where to start scaling |
1047 | */ |
1048 | if (megabytes <= 16) |
1049 | scale = 0; |
1050 | else if (megabytes <= 32) |
1051 | scale = 1; |
1052 | else if (megabytes <= 64) |
1053 | scale = 2; |
1054 | else if (megabytes <= 96) |
1055 | scale = 3; |
1056 | else if (megabytes <= 128) |
1057 | scale = 4; |
1058 | |
1059 | /* |
1060 | * scale number of entries |
1061 | */ |
1062 | bvec_pool_entries = megabytes * 2; |
1063 | if (bvec_pool_entries > 256) |
1064 | bvec_pool_entries = 256; |
1065 | |
1066 | fs_bio_set = bioset_create(BIO_POOL_SIZE, bvec_pool_entries, scale); |
1067 | if (!fs_bio_set) |
1068 | panic("bio: can't allocate bios\n"); |
1069 | |
1070 | bio_split_pool = mempool_create(BIO_SPLIT_ENTRIES, |
1071 | bio_pair_alloc, bio_pair_free, NULL); |
1072 | if (!bio_split_pool) |
1073 | panic("bio: can't create split pool\n"); |
1074 | |
1075 | return 0; |
1076 | } |
1077 | |
1078 | subsys_initcall(init_bio); |
1079 | |
1080 | EXPORT_SYMBOL(bio_alloc); |
1081 | EXPORT_SYMBOL(bio_put); |
1082 | EXPORT_SYMBOL(bio_endio); |
1083 | EXPORT_SYMBOL(bio_init); |
1084 | EXPORT_SYMBOL(__bio_clone); |
1085 | EXPORT_SYMBOL(bio_clone); |
1086 | EXPORT_SYMBOL(bio_phys_segments); |
1087 | EXPORT_SYMBOL(bio_hw_segments); |
1088 | EXPORT_SYMBOL(bio_add_page); |
1089 | EXPORT_SYMBOL(bio_get_nr_vecs); |
1090 | EXPORT_SYMBOL(bio_map_user); |
1091 | EXPORT_SYMBOL(bio_unmap_user); |
1092 | EXPORT_SYMBOL(bio_pair_release); |
1093 | EXPORT_SYMBOL(bio_split); |
1094 | EXPORT_SYMBOL(bio_split_pool); |
1095 | EXPORT_SYMBOL(bio_copy_user); |
1096 | EXPORT_SYMBOL(bio_uncopy_user); |
1097 | EXPORT_SYMBOL(bioset_create); |
1098 | EXPORT_SYMBOL(bioset_free); |
1099 | EXPORT_SYMBOL(bio_alloc_bioset); |