blob: c0d9140e470c3ecd7373522fb9b012eda000f05b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
31static 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
40mempool_t *bio_split_pool;
41
42struct 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) }
55static 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 */
66struct 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 */
75static struct bio_set *fs_bio_set;
76
77static 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 */
110static 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
121inline 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
Martin Waitz67be2dd2005-05-01 08:59:26 -0700143 * @bs: the bio_set to allocate from
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 *
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 **/
153struct 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 }
177out:
178 return bio;
179}
180
181struct 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
186void 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}
199EXPORT_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 **/
209void 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
222inline 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
230inline 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 */
247inline 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_phys_segments(q, bio);
265 bio_hw_segments(q, bio);
266}
267
268/**
269 * bio_clone - clone a bio
270 * @bio: bio to clone
271 * @gfp_mask: allocation priority
272 *
273 * Like __bio_clone, only also allocates the returned bio
274 */
275struct bio *bio_clone(struct bio *bio, unsigned int __nocast gfp_mask)
276{
277 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
278
279 if (b)
280 __bio_clone(b, bio);
281
282 return b;
283}
284
285/**
286 * bio_get_nr_vecs - return approx number of vecs
287 * @bdev: I/O target
288 *
289 * Return the approximate number of pages we can send to this target.
290 * There's no guarantee that you will be able to fit this number of pages
291 * into a bio, it does not account for dynamic restrictions that vary
292 * on offset.
293 */
294int bio_get_nr_vecs(struct block_device *bdev)
295{
296 request_queue_t *q = bdev_get_queue(bdev);
297 int nr_pages;
298
299 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
300 if (nr_pages > q->max_phys_segments)
301 nr_pages = q->max_phys_segments;
302 if (nr_pages > q->max_hw_segments)
303 nr_pages = q->max_hw_segments;
304
305 return nr_pages;
306}
307
308static int __bio_add_page(request_queue_t *q, struct bio *bio, struct page
309 *page, unsigned int len, unsigned int offset)
310{
311 int retried_segments = 0;
312 struct bio_vec *bvec;
313
314 /*
315 * cloned bio must not modify vec list
316 */
317 if (unlikely(bio_flagged(bio, BIO_CLONED)))
318 return 0;
319
320 if (bio->bi_vcnt >= bio->bi_max_vecs)
321 return 0;
322
323 if (((bio->bi_size + len) >> 9) > q->max_sectors)
324 return 0;
325
326 /*
327 * we might lose a segment or two here, but rather that than
328 * make this too complex.
329 */
330
331 while (bio->bi_phys_segments >= q->max_phys_segments
332 || bio->bi_hw_segments >= q->max_hw_segments
333 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
334
335 if (retried_segments)
336 return 0;
337
338 retried_segments = 1;
339 blk_recount_segments(q, bio);
340 }
341
342 /*
343 * setup the new entry, we might clear it again later if we
344 * cannot add the page
345 */
346 bvec = &bio->bi_io_vec[bio->bi_vcnt];
347 bvec->bv_page = page;
348 bvec->bv_len = len;
349 bvec->bv_offset = offset;
350
351 /*
352 * if queue has other restrictions (eg varying max sector size
353 * depending on offset), it can specify a merge_bvec_fn in the
354 * queue to get further control
355 */
356 if (q->merge_bvec_fn) {
357 /*
358 * merge_bvec_fn() returns number of bytes it can accept
359 * at this offset
360 */
361 if (q->merge_bvec_fn(q, bio, bvec) < len) {
362 bvec->bv_page = NULL;
363 bvec->bv_len = 0;
364 bvec->bv_offset = 0;
365 return 0;
366 }
367 }
368
369 /* If we may be able to merge these biovecs, force a recount */
370 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
371 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
372 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
373
374 bio->bi_vcnt++;
375 bio->bi_phys_segments++;
376 bio->bi_hw_segments++;
377 bio->bi_size += len;
378 return len;
379}
380
381/**
382 * bio_add_page - attempt to add page to bio
383 * @bio: destination bio
384 * @page: page to add
385 * @len: vec entry length
386 * @offset: vec entry offset
387 *
388 * Attempt to add a page to the bio_vec maplist. This can fail for a
389 * number of reasons, such as the bio being full or target block
390 * device limitations. The target block device must allow bio's
391 * smaller than PAGE_SIZE, so it is always possible to add a single
392 * page to an empty bio.
393 */
394int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
395 unsigned int offset)
396{
397 return __bio_add_page(bdev_get_queue(bio->bi_bdev), bio, page,
398 len, offset);
399}
400
401struct bio_map_data {
402 struct bio_vec *iovecs;
403 void __user *userptr;
404};
405
406static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio)
407{
408 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
409 bio->bi_private = bmd;
410}
411
412static void bio_free_map_data(struct bio_map_data *bmd)
413{
414 kfree(bmd->iovecs);
415 kfree(bmd);
416}
417
418static struct bio_map_data *bio_alloc_map_data(int nr_segs)
419{
420 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL);
421
422 if (!bmd)
423 return NULL;
424
425 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL);
426 if (bmd->iovecs)
427 return bmd;
428
429 kfree(bmd);
430 return NULL;
431}
432
433/**
434 * bio_uncopy_user - finish previously mapped bio
435 * @bio: bio being terminated
436 *
437 * Free pages allocated from bio_copy_user() and write back data
438 * to user space in case of a read.
439 */
440int bio_uncopy_user(struct bio *bio)
441{
442 struct bio_map_data *bmd = bio->bi_private;
443 const int read = bio_data_dir(bio) == READ;
444 struct bio_vec *bvec;
445 int i, ret = 0;
446
447 __bio_for_each_segment(bvec, bio, i, 0) {
448 char *addr = page_address(bvec->bv_page);
449 unsigned int len = bmd->iovecs[i].bv_len;
450
451 if (read && !ret && copy_to_user(bmd->userptr, addr, len))
452 ret = -EFAULT;
453
454 __free_page(bvec->bv_page);
455 bmd->userptr += len;
456 }
457 bio_free_map_data(bmd);
458 bio_put(bio);
459 return ret;
460}
461
462/**
463 * bio_copy_user - copy user data to bio
464 * @q: destination block queue
465 * @uaddr: start of user address
466 * @len: length in bytes
467 * @write_to_vm: bool indicating writing to pages or not
468 *
469 * Prepares and returns a bio for indirect user io, bouncing data
470 * to/from kernel pages as necessary. Must be paired with
471 * call bio_uncopy_user() on io completion.
472 */
473struct bio *bio_copy_user(request_queue_t *q, unsigned long uaddr,
474 unsigned int len, int write_to_vm)
475{
476 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
477 unsigned long start = uaddr >> PAGE_SHIFT;
478 struct bio_map_data *bmd;
479 struct bio_vec *bvec;
480 struct page *page;
481 struct bio *bio;
482 int i, ret;
483
484 bmd = bio_alloc_map_data(end - start);
485 if (!bmd)
486 return ERR_PTR(-ENOMEM);
487
488 bmd->userptr = (void __user *) uaddr;
489
490 ret = -ENOMEM;
491 bio = bio_alloc(GFP_KERNEL, end - start);
492 if (!bio)
493 goto out_bmd;
494
495 bio->bi_rw |= (!write_to_vm << BIO_RW);
496
497 ret = 0;
498 while (len) {
499 unsigned int bytes = PAGE_SIZE;
500
501 if (bytes > len)
502 bytes = len;
503
504 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
505 if (!page) {
506 ret = -ENOMEM;
507 break;
508 }
509
510 if (__bio_add_page(q, bio, page, bytes, 0) < bytes) {
511 ret = -EINVAL;
512 break;
513 }
514
515 len -= bytes;
516 }
517
518 if (ret)
519 goto cleanup;
520
521 /*
522 * success
523 */
524 if (!write_to_vm) {
525 char __user *p = (char __user *) uaddr;
526
527 /*
528 * for a write, copy in data to kernel pages
529 */
530 ret = -EFAULT;
531 bio_for_each_segment(bvec, bio, i) {
532 char *addr = page_address(bvec->bv_page);
533
534 if (copy_from_user(addr, p, bvec->bv_len))
535 goto cleanup;
536 p += bvec->bv_len;
537 }
538 }
539
540 bio_set_map_data(bmd, bio);
541 return bio;
542cleanup:
543 bio_for_each_segment(bvec, bio, i)
544 __free_page(bvec->bv_page);
545
546 bio_put(bio);
547out_bmd:
548 bio_free_map_data(bmd);
549 return ERR_PTR(ret);
550}
551
552static struct bio *__bio_map_user(request_queue_t *q, struct block_device *bdev,
553 unsigned long uaddr, unsigned int len,
554 int write_to_vm)
555{
556 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
557 unsigned long start = uaddr >> PAGE_SHIFT;
558 const int nr_pages = end - start;
559 int ret, offset, i;
560 struct page **pages;
561 struct bio *bio;
562
563 /*
564 * transfer and buffer must be aligned to at least hardsector
565 * size for now, in the future we can relax this restriction
566 */
567 if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q)))
568 return ERR_PTR(-EINVAL);
569
570 bio = bio_alloc(GFP_KERNEL, nr_pages);
571 if (!bio)
572 return ERR_PTR(-ENOMEM);
573
574 ret = -ENOMEM;
575 pages = kmalloc(nr_pages * sizeof(struct page *), GFP_KERNEL);
576 if (!pages)
577 goto out;
578
579 down_read(&current->mm->mmap_sem);
580 ret = get_user_pages(current, current->mm, uaddr, nr_pages,
581 write_to_vm, 0, pages, NULL);
582 up_read(&current->mm->mmap_sem);
583
584 if (ret < nr_pages)
585 goto out;
586
587 bio->bi_bdev = bdev;
588
589 offset = uaddr & ~PAGE_MASK;
590 for (i = 0; i < nr_pages; i++) {
591 unsigned int bytes = PAGE_SIZE - offset;
592
593 if (len <= 0)
594 break;
595
596 if (bytes > len)
597 bytes = len;
598
599 /*
600 * sorry...
601 */
602 if (__bio_add_page(q, bio, pages[i], bytes, offset) < bytes)
603 break;
604
605 len -= bytes;
606 offset = 0;
607 }
608
609 /*
610 * release the pages we didn't map into the bio, if any
611 */
612 while (i < nr_pages)
613 page_cache_release(pages[i++]);
614
615 kfree(pages);
616
617 /*
618 * set data direction, and check if mapped pages need bouncing
619 */
620 if (!write_to_vm)
621 bio->bi_rw |= (1 << BIO_RW);
622
623 bio->bi_flags |= (1 << BIO_USER_MAPPED);
624 return bio;
625out:
626 kfree(pages);
627 bio_put(bio);
628 return ERR_PTR(ret);
629}
630
631/**
632 * bio_map_user - map user address into bio
Martin Waitz67be2dd2005-05-01 08:59:26 -0700633 * @q: the request_queue_t for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 * @bdev: destination block device
635 * @uaddr: start of user address
636 * @len: length in bytes
637 * @write_to_vm: bool indicating writing to pages or not
638 *
639 * Map the user space address into a bio suitable for io to a block
640 * device. Returns an error pointer in case of error.
641 */
642struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev,
643 unsigned long uaddr, unsigned int len, int write_to_vm)
644{
645 struct bio *bio;
646
647 bio = __bio_map_user(q, bdev, uaddr, len, write_to_vm);
648
649 if (IS_ERR(bio))
650 return bio;
651
652 /*
653 * subtle -- if __bio_map_user() ended up bouncing a bio,
654 * it would normally disappear when its bi_end_io is run.
655 * however, we need it for the unmap, so grab an extra
656 * reference to it
657 */
658 bio_get(bio);
659
660 if (bio->bi_size == len)
661 return bio;
662
663 /*
664 * don't support partial mappings
665 */
666 bio_endio(bio, bio->bi_size, 0);
667 bio_unmap_user(bio);
668 return ERR_PTR(-EINVAL);
669}
670
671static void __bio_unmap_user(struct bio *bio)
672{
673 struct bio_vec *bvec;
674 int i;
675
676 /*
677 * make sure we dirty pages we wrote to
678 */
679 __bio_for_each_segment(bvec, bio, i, 0) {
680 if (bio_data_dir(bio) == READ)
681 set_page_dirty_lock(bvec->bv_page);
682
683 page_cache_release(bvec->bv_page);
684 }
685
686 bio_put(bio);
687}
688
689/**
690 * bio_unmap_user - unmap a bio
691 * @bio: the bio being unmapped
692 *
693 * Unmap a bio previously mapped by bio_map_user(). Must be called with
694 * a process context.
695 *
696 * bio_unmap_user() may sleep.
697 */
698void bio_unmap_user(struct bio *bio)
699{
700 __bio_unmap_user(bio);
701 bio_put(bio);
702}
703
Jens Axboeb8238252005-06-20 14:05:27 +0200704static int bio_map_kern_endio(struct bio *bio, unsigned int bytes_done, int err)
705{
706 if (bio->bi_size)
707 return 1;
708
709 bio_put(bio);
710 return 0;
711}
712
713
Mike Christie df46b9a2005-06-20 14:04:44 +0200714static struct bio *__bio_map_kern(request_queue_t *q, void *data,
715 unsigned int len, unsigned int gfp_mask)
716{
717 unsigned long kaddr = (unsigned long)data;
718 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
719 unsigned long start = kaddr >> PAGE_SHIFT;
720 const int nr_pages = end - start;
721 int offset, i;
722 struct bio *bio;
723
724 bio = bio_alloc(gfp_mask, nr_pages);
725 if (!bio)
726 return ERR_PTR(-ENOMEM);
727
728 offset = offset_in_page(kaddr);
729 for (i = 0; i < nr_pages; i++) {
730 unsigned int bytes = PAGE_SIZE - offset;
731
732 if (len <= 0)
733 break;
734
735 if (bytes > len)
736 bytes = len;
737
738 if (__bio_add_page(q, bio, virt_to_page(data), bytes,
739 offset) < bytes)
740 break;
741
742 data += bytes;
743 len -= bytes;
744 offset = 0;
745 }
746
Jens Axboeb8238252005-06-20 14:05:27 +0200747 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +0200748 return bio;
749}
750
751/**
752 * bio_map_kern - map kernel address into bio
753 * @q: the request_queue_t for the bio
754 * @data: pointer to buffer to map
755 * @len: length in bytes
756 * @gfp_mask: allocation flags for bio allocation
757 *
758 * Map the kernel address into a bio suitable for io to a block
759 * device. Returns an error pointer in case of error.
760 */
761struct bio *bio_map_kern(request_queue_t *q, void *data, unsigned int len,
762 unsigned int gfp_mask)
763{
764 struct bio *bio;
765
766 bio = __bio_map_kern(q, data, len, gfp_mask);
767 if (IS_ERR(bio))
768 return bio;
769
770 if (bio->bi_size == len)
771 return bio;
772
773 /*
774 * Don't support partial mappings.
775 */
776 bio_put(bio);
777 return ERR_PTR(-EINVAL);
778}
779
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780/*
781 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
782 * for performing direct-IO in BIOs.
783 *
784 * The problem is that we cannot run set_page_dirty() from interrupt context
785 * because the required locks are not interrupt-safe. So what we can do is to
786 * mark the pages dirty _before_ performing IO. And in interrupt context,
787 * check that the pages are still dirty. If so, fine. If not, redirty them
788 * in process context.
789 *
790 * We special-case compound pages here: normally this means reads into hugetlb
791 * pages. The logic in here doesn't really work right for compound pages
792 * because the VM does not uniformly chase down the head page in all cases.
793 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
794 * handle them at all. So we skip compound pages here at an early stage.
795 *
796 * Note that this code is very hard to test under normal circumstances because
797 * direct-io pins the pages with get_user_pages(). This makes
798 * is_page_cache_freeable return false, and the VM will not clean the pages.
799 * But other code (eg, pdflush) could clean the pages if they are mapped
800 * pagecache.
801 *
802 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
803 * deferred bio dirtying paths.
804 */
805
806/*
807 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
808 */
809void bio_set_pages_dirty(struct bio *bio)
810{
811 struct bio_vec *bvec = bio->bi_io_vec;
812 int i;
813
814 for (i = 0; i < bio->bi_vcnt; i++) {
815 struct page *page = bvec[i].bv_page;
816
817 if (page && !PageCompound(page))
818 set_page_dirty_lock(page);
819 }
820}
821
822static void bio_release_pages(struct bio *bio)
823{
824 struct bio_vec *bvec = bio->bi_io_vec;
825 int i;
826
827 for (i = 0; i < bio->bi_vcnt; i++) {
828 struct page *page = bvec[i].bv_page;
829
830 if (page)
831 put_page(page);
832 }
833}
834
835/*
836 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
837 * If they are, then fine. If, however, some pages are clean then they must
838 * have been written out during the direct-IO read. So we take another ref on
839 * the BIO and the offending pages and re-dirty the pages in process context.
840 *
841 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
842 * here on. It will run one page_cache_release() against each page and will
843 * run one bio_put() against the BIO.
844 */
845
846static void bio_dirty_fn(void *data);
847
848static DECLARE_WORK(bio_dirty_work, bio_dirty_fn, NULL);
849static DEFINE_SPINLOCK(bio_dirty_lock);
850static struct bio *bio_dirty_list;
851
852/*
853 * This runs in process context
854 */
855static void bio_dirty_fn(void *data)
856{
857 unsigned long flags;
858 struct bio *bio;
859
860 spin_lock_irqsave(&bio_dirty_lock, flags);
861 bio = bio_dirty_list;
862 bio_dirty_list = NULL;
863 spin_unlock_irqrestore(&bio_dirty_lock, flags);
864
865 while (bio) {
866 struct bio *next = bio->bi_private;
867
868 bio_set_pages_dirty(bio);
869 bio_release_pages(bio);
870 bio_put(bio);
871 bio = next;
872 }
873}
874
875void bio_check_pages_dirty(struct bio *bio)
876{
877 struct bio_vec *bvec = bio->bi_io_vec;
878 int nr_clean_pages = 0;
879 int i;
880
881 for (i = 0; i < bio->bi_vcnt; i++) {
882 struct page *page = bvec[i].bv_page;
883
884 if (PageDirty(page) || PageCompound(page)) {
885 page_cache_release(page);
886 bvec[i].bv_page = NULL;
887 } else {
888 nr_clean_pages++;
889 }
890 }
891
892 if (nr_clean_pages) {
893 unsigned long flags;
894
895 spin_lock_irqsave(&bio_dirty_lock, flags);
896 bio->bi_private = bio_dirty_list;
897 bio_dirty_list = bio;
898 spin_unlock_irqrestore(&bio_dirty_lock, flags);
899 schedule_work(&bio_dirty_work);
900 } else {
901 bio_put(bio);
902 }
903}
904
905/**
906 * bio_endio - end I/O on a bio
907 * @bio: bio
908 * @bytes_done: number of bytes completed
909 * @error: error, if any
910 *
911 * Description:
912 * bio_endio() will end I/O on @bytes_done number of bytes. This may be
913 * just a partial part of the bio, or it may be the whole bio. bio_endio()
914 * is the preferred way to end I/O on a bio, it takes care of decrementing
915 * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and
916 * and one of the established -Exxxx (-EIO, for instance) error values in
917 * case something went wrong. Noone should call bi_end_io() directly on
918 * a bio unless they own it and thus know that it has an end_io function.
919 **/
920void bio_endio(struct bio *bio, unsigned int bytes_done, int error)
921{
922 if (error)
923 clear_bit(BIO_UPTODATE, &bio->bi_flags);
924
925 if (unlikely(bytes_done > bio->bi_size)) {
926 printk("%s: want %u bytes done, only %u left\n", __FUNCTION__,
927 bytes_done, bio->bi_size);
928 bytes_done = bio->bi_size;
929 }
930
931 bio->bi_size -= bytes_done;
932 bio->bi_sector += (bytes_done >> 9);
933
934 if (bio->bi_end_io)
935 bio->bi_end_io(bio, bytes_done, error);
936}
937
938void bio_pair_release(struct bio_pair *bp)
939{
940 if (atomic_dec_and_test(&bp->cnt)) {
941 struct bio *master = bp->bio1.bi_private;
942
943 bio_endio(master, master->bi_size, bp->error);
944 mempool_free(bp, bp->bio2.bi_private);
945 }
946}
947
948static int bio_pair_end_1(struct bio * bi, unsigned int done, int err)
949{
950 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
951
952 if (err)
953 bp->error = err;
954
955 if (bi->bi_size)
956 return 1;
957
958 bio_pair_release(bp);
959 return 0;
960}
961
962static int bio_pair_end_2(struct bio * bi, unsigned int done, int err)
963{
964 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
965
966 if (err)
967 bp->error = err;
968
969 if (bi->bi_size)
970 return 1;
971
972 bio_pair_release(bp);
973 return 0;
974}
975
976/*
977 * split a bio - only worry about a bio with a single page
978 * in it's iovec
979 */
980struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
981{
982 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
983
984 if (!bp)
985 return bp;
986
987 BUG_ON(bi->bi_vcnt != 1);
988 BUG_ON(bi->bi_idx != 0);
989 atomic_set(&bp->cnt, 3);
990 bp->error = 0;
991 bp->bio1 = *bi;
992 bp->bio2 = *bi;
993 bp->bio2.bi_sector += first_sectors;
994 bp->bio2.bi_size -= first_sectors << 9;
995 bp->bio1.bi_size = first_sectors << 9;
996
997 bp->bv1 = bi->bi_io_vec[0];
998 bp->bv2 = bi->bi_io_vec[0];
999 bp->bv2.bv_offset += first_sectors << 9;
1000 bp->bv2.bv_len -= first_sectors << 9;
1001 bp->bv1.bv_len = first_sectors << 9;
1002
1003 bp->bio1.bi_io_vec = &bp->bv1;
1004 bp->bio2.bi_io_vec = &bp->bv2;
1005
1006 bp->bio1.bi_end_io = bio_pair_end_1;
1007 bp->bio2.bi_end_io = bio_pair_end_2;
1008
1009 bp->bio1.bi_private = bi;
1010 bp->bio2.bi_private = pool;
1011
1012 return bp;
1013}
1014
1015static void *bio_pair_alloc(unsigned int __nocast gfp_flags, void *data)
1016{
1017 return kmalloc(sizeof(struct bio_pair), gfp_flags);
1018}
1019
1020static void bio_pair_free(void *bp, void *data)
1021{
1022 kfree(bp);
1023}
1024
1025
1026/*
1027 * create memory pools for biovec's in a bio_set.
1028 * use the global biovec slabs created for general use.
1029 */
1030static int biovec_create_pools(struct bio_set *bs, int pool_entries, int scale)
1031{
1032 int i;
1033
1034 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1035 struct biovec_slab *bp = bvec_slabs + i;
1036 mempool_t **bvp = bs->bvec_pools + i;
1037
1038 if (i >= scale)
1039 pool_entries >>= 1;
1040
1041 *bvp = mempool_create(pool_entries, mempool_alloc_slab,
1042 mempool_free_slab, bp->slab);
1043 if (!*bvp)
1044 return -ENOMEM;
1045 }
1046 return 0;
1047}
1048
1049static void biovec_free_pools(struct bio_set *bs)
1050{
1051 int i;
1052
1053 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1054 mempool_t *bvp = bs->bvec_pools[i];
1055
1056 if (bvp)
1057 mempool_destroy(bvp);
1058 }
1059
1060}
1061
1062void bioset_free(struct bio_set *bs)
1063{
1064 if (bs->bio_pool)
1065 mempool_destroy(bs->bio_pool);
1066
1067 biovec_free_pools(bs);
1068
1069 kfree(bs);
1070}
1071
1072struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size, int scale)
1073{
1074 struct bio_set *bs = kmalloc(sizeof(*bs), GFP_KERNEL);
1075
1076 if (!bs)
1077 return NULL;
1078
1079 memset(bs, 0, sizeof(*bs));
1080 bs->bio_pool = mempool_create(bio_pool_size, mempool_alloc_slab,
1081 mempool_free_slab, bio_slab);
1082
1083 if (!bs->bio_pool)
1084 goto bad;
1085
1086 if (!biovec_create_pools(bs, bvec_pool_size, scale))
1087 return bs;
1088
1089bad:
1090 bioset_free(bs);
1091 return NULL;
1092}
1093
1094static void __init biovec_init_slabs(void)
1095{
1096 int i;
1097
1098 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1099 int size;
1100 struct biovec_slab *bvs = bvec_slabs + i;
1101
1102 size = bvs->nr_vecs * sizeof(struct bio_vec);
1103 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1104 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1105 }
1106}
1107
1108static int __init init_bio(void)
1109{
1110 int megabytes, bvec_pool_entries;
1111 int scale = BIOVEC_NR_POOLS;
1112
1113 bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0,
1114 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1115
1116 biovec_init_slabs();
1117
1118 megabytes = nr_free_pages() >> (20 - PAGE_SHIFT);
1119
1120 /*
1121 * find out where to start scaling
1122 */
1123 if (megabytes <= 16)
1124 scale = 0;
1125 else if (megabytes <= 32)
1126 scale = 1;
1127 else if (megabytes <= 64)
1128 scale = 2;
1129 else if (megabytes <= 96)
1130 scale = 3;
1131 else if (megabytes <= 128)
1132 scale = 4;
1133
1134 /*
1135 * scale number of entries
1136 */
1137 bvec_pool_entries = megabytes * 2;
1138 if (bvec_pool_entries > 256)
1139 bvec_pool_entries = 256;
1140
1141 fs_bio_set = bioset_create(BIO_POOL_SIZE, bvec_pool_entries, scale);
1142 if (!fs_bio_set)
1143 panic("bio: can't allocate bios\n");
1144
1145 bio_split_pool = mempool_create(BIO_SPLIT_ENTRIES,
1146 bio_pair_alloc, bio_pair_free, NULL);
1147 if (!bio_split_pool)
1148 panic("bio: can't create split pool\n");
1149
1150 return 0;
1151}
1152
1153subsys_initcall(init_bio);
1154
1155EXPORT_SYMBOL(bio_alloc);
1156EXPORT_SYMBOL(bio_put);
1157EXPORT_SYMBOL(bio_endio);
1158EXPORT_SYMBOL(bio_init);
1159EXPORT_SYMBOL(__bio_clone);
1160EXPORT_SYMBOL(bio_clone);
1161EXPORT_SYMBOL(bio_phys_segments);
1162EXPORT_SYMBOL(bio_hw_segments);
1163EXPORT_SYMBOL(bio_add_page);
1164EXPORT_SYMBOL(bio_get_nr_vecs);
1165EXPORT_SYMBOL(bio_map_user);
1166EXPORT_SYMBOL(bio_unmap_user);
Mike Christie df46b9a2005-06-20 14:04:44 +02001167EXPORT_SYMBOL(bio_map_kern);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168EXPORT_SYMBOL(bio_pair_release);
1169EXPORT_SYMBOL(bio_split);
1170EXPORT_SYMBOL(bio_split_pool);
1171EXPORT_SYMBOL(bio_copy_user);
1172EXPORT_SYMBOL(bio_uncopy_user);
1173EXPORT_SYMBOL(bioset_create);
1174EXPORT_SYMBOL(bioset_free);
1175EXPORT_SYMBOL(bio_alloc_bioset);