blob: 7d81a93afd480c9d9f0335c208f130946b1dc259 [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>
James Bottomley f1970ba2005-06-20 14:06:52 +020028#include <scsi/sg.h> /* for struct sg_iovec */
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
30#define BIO_POOL_SIZE 256
31
32static kmem_cache_t *bio_slab;
33
34#define BIOVEC_NR_POOLS 6
35
36/*
37 * a small number of entries is fine, not going to be performance critical.
38 * basically we just need to survive
39 */
40#define BIO_SPLIT_ENTRIES 8
41mempool_t *bio_split_pool;
42
43struct biovec_slab {
44 int nr_vecs;
45 char *name;
46 kmem_cache_t *slab;
47};
48
49/*
50 * if you change this list, also change bvec_alloc or things will
51 * break badly! cannot be bigger than what you can fit into an
52 * unsigned short
53 */
54
55#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
Christoph Lameter6c036522005-07-07 17:56:59 -070056static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
58};
59#undef BV
60
61/*
62 * bio_set is used to allow other portions of the IO system to
63 * allocate their own private memory pools for bio and iovec structures.
64 * These memory pools in turn all allocate from the bio_slab
65 * and the bvec_slabs[].
66 */
67struct bio_set {
68 mempool_t *bio_pool;
69 mempool_t *bvec_pools[BIOVEC_NR_POOLS];
70};
71
72/*
73 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
74 * IO code that does not need private memory pools.
75 */
76static struct bio_set *fs_bio_set;
77
Al Virodd0fc662005-10-07 07:46:04 +010078static inline struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079{
80 struct bio_vec *bvl;
81 struct biovec_slab *bp;
82
83 /*
84 * see comment near bvec_array define!
85 */
86 switch (nr) {
87 case 1 : *idx = 0; break;
88 case 2 ... 4: *idx = 1; break;
89 case 5 ... 16: *idx = 2; break;
90 case 17 ... 64: *idx = 3; break;
91 case 65 ... 128: *idx = 4; break;
92 case 129 ... BIO_MAX_PAGES: *idx = 5; break;
93 default:
94 return NULL;
95 }
96 /*
97 * idx now points to the pool we want to allocate from
98 */
99
100 bp = bvec_slabs + *idx;
101 bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask);
102 if (bvl)
103 memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec));
104
105 return bvl;
106}
107
Peter Osterlund36763472005-09-06 15:16:42 -0700108void bio_free(struct bio *bio, struct bio_set *bio_set)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109{
110 const int pool_idx = BIO_POOL_IDX(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
112 BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS);
113
Peter Osterlund36763472005-09-06 15:16:42 -0700114 mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]);
115 mempool_free(bio, bio_set->bio_pool);
116}
117
118/*
119 * default destructor for a bio allocated with bio_alloc_bioset()
120 */
121static void bio_fs_destructor(struct bio *bio)
122{
123 bio_free(bio, fs_bio_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124}
125
126inline void bio_init(struct bio *bio)
127{
128 bio->bi_next = NULL;
129 bio->bi_flags = 1 << BIO_UPTODATE;
130 bio->bi_rw = 0;
131 bio->bi_vcnt = 0;
132 bio->bi_idx = 0;
133 bio->bi_phys_segments = 0;
134 bio->bi_hw_segments = 0;
135 bio->bi_hw_front_size = 0;
136 bio->bi_hw_back_size = 0;
137 bio->bi_size = 0;
138 bio->bi_max_vecs = 0;
139 bio->bi_end_io = NULL;
140 atomic_set(&bio->bi_cnt, 1);
141 bio->bi_private = NULL;
142}
143
144/**
145 * bio_alloc_bioset - allocate a bio for I/O
146 * @gfp_mask: the GFP_ mask given to the slab allocator
147 * @nr_iovecs: number of iovecs to pre-allocate
Martin Waitz67be2dd2005-05-01 08:59:26 -0700148 * @bs: the bio_set to allocate from
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 *
150 * Description:
151 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation.
152 * If %__GFP_WAIT is set then we will block on the internal pool waiting
153 * for a &struct bio to become free.
154 *
155 * allocate bio and iovecs from the memory pools specified by the
156 * bio_set structure.
157 **/
Al Virodd0fc662005-10-07 07:46:04 +0100158struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159{
160 struct bio *bio = mempool_alloc(bs->bio_pool, gfp_mask);
161
162 if (likely(bio)) {
163 struct bio_vec *bvl = NULL;
164
165 bio_init(bio);
166 if (likely(nr_iovecs)) {
167 unsigned long idx;
168
169 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
170 if (unlikely(!bvl)) {
171 mempool_free(bio, bs->bio_pool);
172 bio = NULL;
173 goto out;
174 }
175 bio->bi_flags |= idx << BIO_POOL_OFFSET;
176 bio->bi_max_vecs = bvec_slabs[idx].nr_vecs;
177 }
178 bio->bi_io_vec = bvl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 }
180out:
181 return bio;
182}
183
Al Virodd0fc662005-10-07 07:46:04 +0100184struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185{
Peter Osterlund36763472005-09-06 15:16:42 -0700186 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
187
188 if (bio)
189 bio->bi_destructor = bio_fs_destructor;
190
191 return bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192}
193
194void zero_fill_bio(struct bio *bio)
195{
196 unsigned long flags;
197 struct bio_vec *bv;
198 int i;
199
200 bio_for_each_segment(bv, bio, i) {
201 char *data = bvec_kmap_irq(bv, &flags);
202 memset(data, 0, bv->bv_len);
203 flush_dcache_page(bv->bv_page);
204 bvec_kunmap_irq(data, &flags);
205 }
206}
207EXPORT_SYMBOL(zero_fill_bio);
208
209/**
210 * bio_put - release a reference to a bio
211 * @bio: bio to release reference to
212 *
213 * Description:
214 * Put a reference to a &struct bio, either one you have gotten with
215 * bio_alloc or bio_get. The last put of a bio will free it.
216 **/
217void bio_put(struct bio *bio)
218{
219 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
220
221 /*
222 * last put frees it
223 */
224 if (atomic_dec_and_test(&bio->bi_cnt)) {
225 bio->bi_next = NULL;
226 bio->bi_destructor(bio);
227 }
228}
229
230inline int bio_phys_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_phys_segments;
236}
237
238inline int bio_hw_segments(request_queue_t *q, struct bio *bio)
239{
240 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
241 blk_recount_segments(q, bio);
242
243 return bio->bi_hw_segments;
244}
245
246/**
247 * __bio_clone - clone a bio
248 * @bio: destination bio
249 * @bio_src: bio to clone
250 *
251 * Clone a &bio. Caller will own the returned bio, but not
252 * the actual data it points to. Reference count of returned
253 * bio will be one.
254 */
255inline void __bio_clone(struct bio *bio, struct bio *bio_src)
256{
257 request_queue_t *q = bdev_get_queue(bio_src->bi_bdev);
258
Andrew Mortone525e152005-08-07 09:42:12 -0700259 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
260 bio_src->bi_max_vecs * sizeof(struct bio_vec));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
262 bio->bi_sector = bio_src->bi_sector;
263 bio->bi_bdev = bio_src->bi_bdev;
264 bio->bi_flags |= 1 << BIO_CLONED;
265 bio->bi_rw = bio_src->bi_rw;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 bio->bi_vcnt = bio_src->bi_vcnt;
267 bio->bi_size = bio_src->bi_size;
Andrew Mortona5453be2005-07-28 01:07:18 -0700268 bio->bi_idx = bio_src->bi_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 bio_phys_segments(q, bio);
270 bio_hw_segments(q, bio);
271}
272
273/**
274 * bio_clone - clone a bio
275 * @bio: bio to clone
276 * @gfp_mask: allocation priority
277 *
278 * Like __bio_clone, only also allocates the returned bio
279 */
Al Virodd0fc662005-10-07 07:46:04 +0100280struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281{
282 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
283
Peter Osterlund36763472005-09-06 15:16:42 -0700284 if (b) {
285 b->bi_destructor = bio_fs_destructor;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 __bio_clone(b, bio);
Peter Osterlund36763472005-09-06 15:16:42 -0700287 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
289 return b;
290}
291
292/**
293 * bio_get_nr_vecs - return approx number of vecs
294 * @bdev: I/O target
295 *
296 * Return the approximate number of pages we can send to this target.
297 * There's no guarantee that you will be able to fit this number of pages
298 * into a bio, it does not account for dynamic restrictions that vary
299 * on offset.
300 */
301int bio_get_nr_vecs(struct block_device *bdev)
302{
303 request_queue_t *q = bdev_get_queue(bdev);
304 int nr_pages;
305
306 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
307 if (nr_pages > q->max_phys_segments)
308 nr_pages = q->max_phys_segments;
309 if (nr_pages > q->max_hw_segments)
310 nr_pages = q->max_hw_segments;
311
312 return nr_pages;
313}
314
315static int __bio_add_page(request_queue_t *q, struct bio *bio, struct page
316 *page, unsigned int len, unsigned int offset)
317{
318 int retried_segments = 0;
319 struct bio_vec *bvec;
320
321 /*
322 * cloned bio must not modify vec list
323 */
324 if (unlikely(bio_flagged(bio, BIO_CLONED)))
325 return 0;
326
327 if (bio->bi_vcnt >= bio->bi_max_vecs)
328 return 0;
329
330 if (((bio->bi_size + len) >> 9) > q->max_sectors)
331 return 0;
332
333 /*
334 * we might lose a segment or two here, but rather that than
335 * make this too complex.
336 */
337
338 while (bio->bi_phys_segments >= q->max_phys_segments
339 || bio->bi_hw_segments >= q->max_hw_segments
340 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
341
342 if (retried_segments)
343 return 0;
344
345 retried_segments = 1;
346 blk_recount_segments(q, bio);
347 }
348
349 /*
350 * setup the new entry, we might clear it again later if we
351 * cannot add the page
352 */
353 bvec = &bio->bi_io_vec[bio->bi_vcnt];
354 bvec->bv_page = page;
355 bvec->bv_len = len;
356 bvec->bv_offset = offset;
357
358 /*
359 * if queue has other restrictions (eg varying max sector size
360 * depending on offset), it can specify a merge_bvec_fn in the
361 * queue to get further control
362 */
363 if (q->merge_bvec_fn) {
364 /*
365 * merge_bvec_fn() returns number of bytes it can accept
366 * at this offset
367 */
368 if (q->merge_bvec_fn(q, bio, bvec) < len) {
369 bvec->bv_page = NULL;
370 bvec->bv_len = 0;
371 bvec->bv_offset = 0;
372 return 0;
373 }
374 }
375
376 /* If we may be able to merge these biovecs, force a recount */
377 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
378 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
379 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
380
381 bio->bi_vcnt++;
382 bio->bi_phys_segments++;
383 bio->bi_hw_segments++;
384 bio->bi_size += len;
385 return len;
386}
387
388/**
389 * bio_add_page - attempt to add page to bio
390 * @bio: destination bio
391 * @page: page to add
392 * @len: vec entry length
393 * @offset: vec entry offset
394 *
395 * Attempt to add a page to the bio_vec maplist. This can fail for a
396 * number of reasons, such as the bio being full or target block
397 * device limitations. The target block device must allow bio's
398 * smaller than PAGE_SIZE, so it is always possible to add a single
399 * page to an empty bio.
400 */
401int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
402 unsigned int offset)
403{
404 return __bio_add_page(bdev_get_queue(bio->bi_bdev), bio, page,
405 len, offset);
406}
407
408struct bio_map_data {
409 struct bio_vec *iovecs;
410 void __user *userptr;
411};
412
413static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio)
414{
415 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
416 bio->bi_private = bmd;
417}
418
419static void bio_free_map_data(struct bio_map_data *bmd)
420{
421 kfree(bmd->iovecs);
422 kfree(bmd);
423}
424
425static struct bio_map_data *bio_alloc_map_data(int nr_segs)
426{
427 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL);
428
429 if (!bmd)
430 return NULL;
431
432 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL);
433 if (bmd->iovecs)
434 return bmd;
435
436 kfree(bmd);
437 return NULL;
438}
439
440/**
441 * bio_uncopy_user - finish previously mapped bio
442 * @bio: bio being terminated
443 *
444 * Free pages allocated from bio_copy_user() and write back data
445 * to user space in case of a read.
446 */
447int bio_uncopy_user(struct bio *bio)
448{
449 struct bio_map_data *bmd = bio->bi_private;
450 const int read = bio_data_dir(bio) == READ;
451 struct bio_vec *bvec;
452 int i, ret = 0;
453
454 __bio_for_each_segment(bvec, bio, i, 0) {
455 char *addr = page_address(bvec->bv_page);
456 unsigned int len = bmd->iovecs[i].bv_len;
457
458 if (read && !ret && copy_to_user(bmd->userptr, addr, len))
459 ret = -EFAULT;
460
461 __free_page(bvec->bv_page);
462 bmd->userptr += len;
463 }
464 bio_free_map_data(bmd);
465 bio_put(bio);
466 return ret;
467}
468
469/**
470 * bio_copy_user - copy user data to bio
471 * @q: destination block queue
472 * @uaddr: start of user address
473 * @len: length in bytes
474 * @write_to_vm: bool indicating writing to pages or not
475 *
476 * Prepares and returns a bio for indirect user io, bouncing data
477 * to/from kernel pages as necessary. Must be paired with
478 * call bio_uncopy_user() on io completion.
479 */
480struct bio *bio_copy_user(request_queue_t *q, unsigned long uaddr,
481 unsigned int len, int write_to_vm)
482{
483 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
484 unsigned long start = uaddr >> PAGE_SHIFT;
485 struct bio_map_data *bmd;
486 struct bio_vec *bvec;
487 struct page *page;
488 struct bio *bio;
489 int i, ret;
490
491 bmd = bio_alloc_map_data(end - start);
492 if (!bmd)
493 return ERR_PTR(-ENOMEM);
494
495 bmd->userptr = (void __user *) uaddr;
496
497 ret = -ENOMEM;
498 bio = bio_alloc(GFP_KERNEL, end - start);
499 if (!bio)
500 goto out_bmd;
501
502 bio->bi_rw |= (!write_to_vm << BIO_RW);
503
504 ret = 0;
505 while (len) {
506 unsigned int bytes = PAGE_SIZE;
507
508 if (bytes > len)
509 bytes = len;
510
511 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
512 if (!page) {
513 ret = -ENOMEM;
514 break;
515 }
516
517 if (__bio_add_page(q, bio, page, bytes, 0) < bytes) {
518 ret = -EINVAL;
519 break;
520 }
521
522 len -= bytes;
523 }
524
525 if (ret)
526 goto cleanup;
527
528 /*
529 * success
530 */
531 if (!write_to_vm) {
532 char __user *p = (char __user *) uaddr;
533
534 /*
535 * for a write, copy in data to kernel pages
536 */
537 ret = -EFAULT;
538 bio_for_each_segment(bvec, bio, i) {
539 char *addr = page_address(bvec->bv_page);
540
541 if (copy_from_user(addr, p, bvec->bv_len))
542 goto cleanup;
543 p += bvec->bv_len;
544 }
545 }
546
547 bio_set_map_data(bmd, bio);
548 return bio;
549cleanup:
550 bio_for_each_segment(bvec, bio, i)
551 __free_page(bvec->bv_page);
552
553 bio_put(bio);
554out_bmd:
555 bio_free_map_data(bmd);
556 return ERR_PTR(ret);
557}
558
James Bottomley f1970ba2005-06-20 14:06:52 +0200559static struct bio *__bio_map_user_iov(request_queue_t *q,
560 struct block_device *bdev,
561 struct sg_iovec *iov, int iov_count,
562 int write_to_vm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563{
James Bottomley f1970ba2005-06-20 14:06:52 +0200564 int i, j;
565 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 struct page **pages;
567 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200568 int cur_page = 0;
569 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
James Bottomley f1970ba2005-06-20 14:06:52 +0200571 for (i = 0; i < iov_count; i++) {
572 unsigned long uaddr = (unsigned long)iov[i].iov_base;
573 unsigned long len = iov[i].iov_len;
574 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
575 unsigned long start = uaddr >> PAGE_SHIFT;
576
577 nr_pages += end - start;
578 /*
579 * transfer and buffer must be aligned to at least hardsector
580 * size for now, in the future we can relax this restriction
581 */
582 if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q)))
583 return ERR_PTR(-EINVAL);
584 }
585
586 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 return ERR_PTR(-EINVAL);
588
589 bio = bio_alloc(GFP_KERNEL, nr_pages);
590 if (!bio)
591 return ERR_PTR(-ENOMEM);
592
593 ret = -ENOMEM;
594 pages = kmalloc(nr_pages * sizeof(struct page *), GFP_KERNEL);
595 if (!pages)
596 goto out;
597
James Bottomley f1970ba2005-06-20 14:06:52 +0200598 memset(pages, 0, nr_pages * sizeof(struct page *));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
James Bottomley f1970ba2005-06-20 14:06:52 +0200600 for (i = 0; i < iov_count; i++) {
601 unsigned long uaddr = (unsigned long)iov[i].iov_base;
602 unsigned long len = iov[i].iov_len;
603 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
604 unsigned long start = uaddr >> PAGE_SHIFT;
605 const int local_nr_pages = end - start;
606 const int page_limit = cur_page + local_nr_pages;
607
608 down_read(&current->mm->mmap_sem);
609 ret = get_user_pages(current, current->mm, uaddr,
610 local_nr_pages,
611 write_to_vm, 0, &pages[cur_page], NULL);
612 up_read(&current->mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
James Bottomley f1970ba2005-06-20 14:06:52 +0200614 if (ret < local_nr_pages)
615 goto out_unmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
James Bottomley f1970ba2005-06-20 14:06:52 +0200618 offset = uaddr & ~PAGE_MASK;
619 for (j = cur_page; j < page_limit; j++) {
620 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
James Bottomley f1970ba2005-06-20 14:06:52 +0200622 if (len <= 0)
623 break;
624
625 if (bytes > len)
626 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
James Bottomley f1970ba2005-06-20 14:06:52 +0200628 /*
629 * sorry...
630 */
631 if (__bio_add_page(q, bio, pages[j], bytes, offset) < bytes)
632 break;
633
634 len -= bytes;
635 offset = 0;
636 }
637
638 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 /*
James Bottomley f1970ba2005-06-20 14:06:52 +0200640 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 */
James Bottomley f1970ba2005-06-20 14:06:52 +0200642 while (j < page_limit)
643 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 }
645
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 kfree(pages);
647
648 /*
649 * set data direction, and check if mapped pages need bouncing
650 */
651 if (!write_to_vm)
652 bio->bi_rw |= (1 << BIO_RW);
653
James Bottomley f1970ba2005-06-20 14:06:52 +0200654 bio->bi_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 bio->bi_flags |= (1 << BIO_USER_MAPPED);
656 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200657
658 out_unmap:
659 for (i = 0; i < nr_pages; i++) {
660 if(!pages[i])
661 break;
662 page_cache_release(pages[i]);
663 }
664 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 kfree(pages);
666 bio_put(bio);
667 return ERR_PTR(ret);
668}
669
670/**
671 * bio_map_user - map user address into bio
Martin Waitz67be2dd2005-05-01 08:59:26 -0700672 * @q: the request_queue_t for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 * @bdev: destination block device
674 * @uaddr: start of user address
675 * @len: length in bytes
676 * @write_to_vm: bool indicating writing to pages or not
677 *
678 * Map the user space address into a bio suitable for io to a block
679 * device. Returns an error pointer in case of error.
680 */
681struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev,
682 unsigned long uaddr, unsigned int len, int write_to_vm)
683{
James Bottomley f1970ba2005-06-20 14:06:52 +0200684 struct sg_iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
viro@ZenIV.linux.org.uk3f703532005-09-09 16:53:56 +0100686 iov.iov_base = (void __user *)uaddr;
James Bottomley f1970ba2005-06-20 14:06:52 +0200687 iov.iov_len = len;
688
689 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
690}
691
692/**
693 * bio_map_user_iov - map user sg_iovec table into bio
694 * @q: the request_queue_t for the bio
695 * @bdev: destination block device
696 * @iov: the iovec.
697 * @iov_count: number of elements in the iovec
698 * @write_to_vm: bool indicating writing to pages or not
699 *
700 * Map the user space address into a bio suitable for io to a block
701 * device. Returns an error pointer in case of error.
702 */
703struct bio *bio_map_user_iov(request_queue_t *q, struct block_device *bdev,
704 struct sg_iovec *iov, int iov_count,
705 int write_to_vm)
706{
707 struct bio *bio;
708 int len = 0, i;
709
710 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
712 if (IS_ERR(bio))
713 return bio;
714
715 /*
716 * subtle -- if __bio_map_user() ended up bouncing a bio,
717 * it would normally disappear when its bi_end_io is run.
718 * however, we need it for the unmap, so grab an extra
719 * reference to it
720 */
721 bio_get(bio);
722
James Bottomley f1970ba2005-06-20 14:06:52 +0200723 for (i = 0; i < iov_count; i++)
724 len += iov[i].iov_len;
725
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 if (bio->bi_size == len)
727 return bio;
728
729 /*
730 * don't support partial mappings
731 */
732 bio_endio(bio, bio->bi_size, 0);
733 bio_unmap_user(bio);
734 return ERR_PTR(-EINVAL);
735}
736
737static void __bio_unmap_user(struct bio *bio)
738{
739 struct bio_vec *bvec;
740 int i;
741
742 /*
743 * make sure we dirty pages we wrote to
744 */
745 __bio_for_each_segment(bvec, bio, i, 0) {
746 if (bio_data_dir(bio) == READ)
747 set_page_dirty_lock(bvec->bv_page);
748
749 page_cache_release(bvec->bv_page);
750 }
751
752 bio_put(bio);
753}
754
755/**
756 * bio_unmap_user - unmap a bio
757 * @bio: the bio being unmapped
758 *
759 * Unmap a bio previously mapped by bio_map_user(). Must be called with
760 * a process context.
761 *
762 * bio_unmap_user() may sleep.
763 */
764void bio_unmap_user(struct bio *bio)
765{
766 __bio_unmap_user(bio);
767 bio_put(bio);
768}
769
Jens Axboeb8238252005-06-20 14:05:27 +0200770static int bio_map_kern_endio(struct bio *bio, unsigned int bytes_done, int err)
771{
772 if (bio->bi_size)
773 return 1;
774
775 bio_put(bio);
776 return 0;
777}
778
779
Mike Christie df46b9a2005-06-20 14:04:44 +0200780static struct bio *__bio_map_kern(request_queue_t *q, void *data,
781 unsigned int len, unsigned int gfp_mask)
782{
783 unsigned long kaddr = (unsigned long)data;
784 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
785 unsigned long start = kaddr >> PAGE_SHIFT;
786 const int nr_pages = end - start;
787 int offset, i;
788 struct bio *bio;
789
790 bio = bio_alloc(gfp_mask, nr_pages);
791 if (!bio)
792 return ERR_PTR(-ENOMEM);
793
794 offset = offset_in_page(kaddr);
795 for (i = 0; i < nr_pages; i++) {
796 unsigned int bytes = PAGE_SIZE - offset;
797
798 if (len <= 0)
799 break;
800
801 if (bytes > len)
802 bytes = len;
803
804 if (__bio_add_page(q, bio, virt_to_page(data), bytes,
805 offset) < bytes)
806 break;
807
808 data += bytes;
809 len -= bytes;
810 offset = 0;
811 }
812
Jens Axboeb8238252005-06-20 14:05:27 +0200813 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +0200814 return bio;
815}
816
817/**
818 * bio_map_kern - map kernel address into bio
819 * @q: the request_queue_t for the bio
820 * @data: pointer to buffer to map
821 * @len: length in bytes
822 * @gfp_mask: allocation flags for bio allocation
823 *
824 * Map the kernel address into a bio suitable for io to a block
825 * device. Returns an error pointer in case of error.
826 */
827struct bio *bio_map_kern(request_queue_t *q, void *data, unsigned int len,
828 unsigned int gfp_mask)
829{
830 struct bio *bio;
831
832 bio = __bio_map_kern(q, data, len, gfp_mask);
833 if (IS_ERR(bio))
834 return bio;
835
836 if (bio->bi_size == len)
837 return bio;
838
839 /*
840 * Don't support partial mappings.
841 */
842 bio_put(bio);
843 return ERR_PTR(-EINVAL);
844}
845
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846/*
847 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
848 * for performing direct-IO in BIOs.
849 *
850 * The problem is that we cannot run set_page_dirty() from interrupt context
851 * because the required locks are not interrupt-safe. So what we can do is to
852 * mark the pages dirty _before_ performing IO. And in interrupt context,
853 * check that the pages are still dirty. If so, fine. If not, redirty them
854 * in process context.
855 *
856 * We special-case compound pages here: normally this means reads into hugetlb
857 * pages. The logic in here doesn't really work right for compound pages
858 * because the VM does not uniformly chase down the head page in all cases.
859 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
860 * handle them at all. So we skip compound pages here at an early stage.
861 *
862 * Note that this code is very hard to test under normal circumstances because
863 * direct-io pins the pages with get_user_pages(). This makes
864 * is_page_cache_freeable return false, and the VM will not clean the pages.
865 * But other code (eg, pdflush) could clean the pages if they are mapped
866 * pagecache.
867 *
868 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
869 * deferred bio dirtying paths.
870 */
871
872/*
873 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
874 */
875void bio_set_pages_dirty(struct bio *bio)
876{
877 struct bio_vec *bvec = bio->bi_io_vec;
878 int i;
879
880 for (i = 0; i < bio->bi_vcnt; i++) {
881 struct page *page = bvec[i].bv_page;
882
883 if (page && !PageCompound(page))
884 set_page_dirty_lock(page);
885 }
886}
887
888static void bio_release_pages(struct bio *bio)
889{
890 struct bio_vec *bvec = bio->bi_io_vec;
891 int i;
892
893 for (i = 0; i < bio->bi_vcnt; i++) {
894 struct page *page = bvec[i].bv_page;
895
896 if (page)
897 put_page(page);
898 }
899}
900
901/*
902 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
903 * If they are, then fine. If, however, some pages are clean then they must
904 * have been written out during the direct-IO read. So we take another ref on
905 * the BIO and the offending pages and re-dirty the pages in process context.
906 *
907 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
908 * here on. It will run one page_cache_release() against each page and will
909 * run one bio_put() against the BIO.
910 */
911
912static void bio_dirty_fn(void *data);
913
914static DECLARE_WORK(bio_dirty_work, bio_dirty_fn, NULL);
915static DEFINE_SPINLOCK(bio_dirty_lock);
916static struct bio *bio_dirty_list;
917
918/*
919 * This runs in process context
920 */
921static void bio_dirty_fn(void *data)
922{
923 unsigned long flags;
924 struct bio *bio;
925
926 spin_lock_irqsave(&bio_dirty_lock, flags);
927 bio = bio_dirty_list;
928 bio_dirty_list = NULL;
929 spin_unlock_irqrestore(&bio_dirty_lock, flags);
930
931 while (bio) {
932 struct bio *next = bio->bi_private;
933
934 bio_set_pages_dirty(bio);
935 bio_release_pages(bio);
936 bio_put(bio);
937 bio = next;
938 }
939}
940
941void bio_check_pages_dirty(struct bio *bio)
942{
943 struct bio_vec *bvec = bio->bi_io_vec;
944 int nr_clean_pages = 0;
945 int i;
946
947 for (i = 0; i < bio->bi_vcnt; i++) {
948 struct page *page = bvec[i].bv_page;
949
950 if (PageDirty(page) || PageCompound(page)) {
951 page_cache_release(page);
952 bvec[i].bv_page = NULL;
953 } else {
954 nr_clean_pages++;
955 }
956 }
957
958 if (nr_clean_pages) {
959 unsigned long flags;
960
961 spin_lock_irqsave(&bio_dirty_lock, flags);
962 bio->bi_private = bio_dirty_list;
963 bio_dirty_list = bio;
964 spin_unlock_irqrestore(&bio_dirty_lock, flags);
965 schedule_work(&bio_dirty_work);
966 } else {
967 bio_put(bio);
968 }
969}
970
971/**
972 * bio_endio - end I/O on a bio
973 * @bio: bio
974 * @bytes_done: number of bytes completed
975 * @error: error, if any
976 *
977 * Description:
978 * bio_endio() will end I/O on @bytes_done number of bytes. This may be
979 * just a partial part of the bio, or it may be the whole bio. bio_endio()
980 * is the preferred way to end I/O on a bio, it takes care of decrementing
981 * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and
982 * and one of the established -Exxxx (-EIO, for instance) error values in
983 * case something went wrong. Noone should call bi_end_io() directly on
984 * a bio unless they own it and thus know that it has an end_io function.
985 **/
986void bio_endio(struct bio *bio, unsigned int bytes_done, int error)
987{
988 if (error)
989 clear_bit(BIO_UPTODATE, &bio->bi_flags);
990
991 if (unlikely(bytes_done > bio->bi_size)) {
992 printk("%s: want %u bytes done, only %u left\n", __FUNCTION__,
993 bytes_done, bio->bi_size);
994 bytes_done = bio->bi_size;
995 }
996
997 bio->bi_size -= bytes_done;
998 bio->bi_sector += (bytes_done >> 9);
999
1000 if (bio->bi_end_io)
1001 bio->bi_end_io(bio, bytes_done, error);
1002}
1003
1004void bio_pair_release(struct bio_pair *bp)
1005{
1006 if (atomic_dec_and_test(&bp->cnt)) {
1007 struct bio *master = bp->bio1.bi_private;
1008
1009 bio_endio(master, master->bi_size, bp->error);
1010 mempool_free(bp, bp->bio2.bi_private);
1011 }
1012}
1013
1014static int bio_pair_end_1(struct bio * bi, unsigned int done, int err)
1015{
1016 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1017
1018 if (err)
1019 bp->error = err;
1020
1021 if (bi->bi_size)
1022 return 1;
1023
1024 bio_pair_release(bp);
1025 return 0;
1026}
1027
1028static int bio_pair_end_2(struct bio * bi, unsigned int done, int err)
1029{
1030 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1031
1032 if (err)
1033 bp->error = err;
1034
1035 if (bi->bi_size)
1036 return 1;
1037
1038 bio_pair_release(bp);
1039 return 0;
1040}
1041
1042/*
1043 * split a bio - only worry about a bio with a single page
1044 * in it's iovec
1045 */
1046struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
1047{
1048 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
1049
1050 if (!bp)
1051 return bp;
1052
1053 BUG_ON(bi->bi_vcnt != 1);
1054 BUG_ON(bi->bi_idx != 0);
1055 atomic_set(&bp->cnt, 3);
1056 bp->error = 0;
1057 bp->bio1 = *bi;
1058 bp->bio2 = *bi;
1059 bp->bio2.bi_sector += first_sectors;
1060 bp->bio2.bi_size -= first_sectors << 9;
1061 bp->bio1.bi_size = first_sectors << 9;
1062
1063 bp->bv1 = bi->bi_io_vec[0];
1064 bp->bv2 = bi->bi_io_vec[0];
1065 bp->bv2.bv_offset += first_sectors << 9;
1066 bp->bv2.bv_len -= first_sectors << 9;
1067 bp->bv1.bv_len = first_sectors << 9;
1068
1069 bp->bio1.bi_io_vec = &bp->bv1;
1070 bp->bio2.bi_io_vec = &bp->bv2;
1071
1072 bp->bio1.bi_end_io = bio_pair_end_1;
1073 bp->bio2.bi_end_io = bio_pair_end_2;
1074
1075 bp->bio1.bi_private = bi;
1076 bp->bio2.bi_private = pool;
1077
1078 return bp;
1079}
1080
Al Virodd0fc662005-10-07 07:46:04 +01001081static void *bio_pair_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082{
1083 return kmalloc(sizeof(struct bio_pair), gfp_flags);
1084}
1085
1086static void bio_pair_free(void *bp, void *data)
1087{
1088 kfree(bp);
1089}
1090
1091
1092/*
1093 * create memory pools for biovec's in a bio_set.
1094 * use the global biovec slabs created for general use.
1095 */
1096static int biovec_create_pools(struct bio_set *bs, int pool_entries, int scale)
1097{
1098 int i;
1099
1100 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1101 struct biovec_slab *bp = bvec_slabs + i;
1102 mempool_t **bvp = bs->bvec_pools + i;
1103
1104 if (i >= scale)
1105 pool_entries >>= 1;
1106
1107 *bvp = mempool_create(pool_entries, mempool_alloc_slab,
1108 mempool_free_slab, bp->slab);
1109 if (!*bvp)
1110 return -ENOMEM;
1111 }
1112 return 0;
1113}
1114
1115static void biovec_free_pools(struct bio_set *bs)
1116{
1117 int i;
1118
1119 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1120 mempool_t *bvp = bs->bvec_pools[i];
1121
1122 if (bvp)
1123 mempool_destroy(bvp);
1124 }
1125
1126}
1127
1128void bioset_free(struct bio_set *bs)
1129{
1130 if (bs->bio_pool)
1131 mempool_destroy(bs->bio_pool);
1132
1133 biovec_free_pools(bs);
1134
1135 kfree(bs);
1136}
1137
1138struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size, int scale)
1139{
1140 struct bio_set *bs = kmalloc(sizeof(*bs), GFP_KERNEL);
1141
1142 if (!bs)
1143 return NULL;
1144
1145 memset(bs, 0, sizeof(*bs));
1146 bs->bio_pool = mempool_create(bio_pool_size, mempool_alloc_slab,
1147 mempool_free_slab, bio_slab);
1148
1149 if (!bs->bio_pool)
1150 goto bad;
1151
1152 if (!biovec_create_pools(bs, bvec_pool_size, scale))
1153 return bs;
1154
1155bad:
1156 bioset_free(bs);
1157 return NULL;
1158}
1159
1160static void __init biovec_init_slabs(void)
1161{
1162 int i;
1163
1164 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1165 int size;
1166 struct biovec_slab *bvs = bvec_slabs + i;
1167
1168 size = bvs->nr_vecs * sizeof(struct bio_vec);
1169 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1170 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1171 }
1172}
1173
1174static int __init init_bio(void)
1175{
1176 int megabytes, bvec_pool_entries;
1177 int scale = BIOVEC_NR_POOLS;
1178
1179 bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0,
1180 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1181
1182 biovec_init_slabs();
1183
1184 megabytes = nr_free_pages() >> (20 - PAGE_SHIFT);
1185
1186 /*
1187 * find out where to start scaling
1188 */
1189 if (megabytes <= 16)
1190 scale = 0;
1191 else if (megabytes <= 32)
1192 scale = 1;
1193 else if (megabytes <= 64)
1194 scale = 2;
1195 else if (megabytes <= 96)
1196 scale = 3;
1197 else if (megabytes <= 128)
1198 scale = 4;
1199
1200 /*
1201 * scale number of entries
1202 */
1203 bvec_pool_entries = megabytes * 2;
1204 if (bvec_pool_entries > 256)
1205 bvec_pool_entries = 256;
1206
1207 fs_bio_set = bioset_create(BIO_POOL_SIZE, bvec_pool_entries, scale);
1208 if (!fs_bio_set)
1209 panic("bio: can't allocate bios\n");
1210
1211 bio_split_pool = mempool_create(BIO_SPLIT_ENTRIES,
1212 bio_pair_alloc, bio_pair_free, NULL);
1213 if (!bio_split_pool)
1214 panic("bio: can't create split pool\n");
1215
1216 return 0;
1217}
1218
1219subsys_initcall(init_bio);
1220
1221EXPORT_SYMBOL(bio_alloc);
1222EXPORT_SYMBOL(bio_put);
Peter Osterlund36763472005-09-06 15:16:42 -07001223EXPORT_SYMBOL(bio_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224EXPORT_SYMBOL(bio_endio);
1225EXPORT_SYMBOL(bio_init);
1226EXPORT_SYMBOL(__bio_clone);
1227EXPORT_SYMBOL(bio_clone);
1228EXPORT_SYMBOL(bio_phys_segments);
1229EXPORT_SYMBOL(bio_hw_segments);
1230EXPORT_SYMBOL(bio_add_page);
1231EXPORT_SYMBOL(bio_get_nr_vecs);
1232EXPORT_SYMBOL(bio_map_user);
1233EXPORT_SYMBOL(bio_unmap_user);
Mike Christie df46b9a2005-06-20 14:04:44 +02001234EXPORT_SYMBOL(bio_map_kern);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235EXPORT_SYMBOL(bio_pair_release);
1236EXPORT_SYMBOL(bio_split);
1237EXPORT_SYMBOL(bio_split_pool);
1238EXPORT_SYMBOL(bio_copy_user);
1239EXPORT_SYMBOL(bio_uncopy_user);
1240EXPORT_SYMBOL(bioset_create);
1241EXPORT_SYMBOL(bioset_free);
1242EXPORT_SYMBOL(bio_alloc_bioset);