blob: dfe242a21eb4cb49d00d4ee54def5c024be1ea0b [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
Mike Christiedefd94b2005-12-05 02:37:06 -0600316 *page, unsigned int len, unsigned int offset,
317 unsigned short max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
319 int retried_segments = 0;
320 struct bio_vec *bvec;
321
322 /*
323 * cloned bio must not modify vec list
324 */
325 if (unlikely(bio_flagged(bio, BIO_CLONED)))
326 return 0;
327
Jens Axboe80cfd542006-01-06 09:43:28 +0100328 if (((bio->bi_size + len) >> 9) > max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 return 0;
330
Jens Axboe80cfd542006-01-06 09:43:28 +0100331 /*
332 * For filesystems with a blocksize smaller than the pagesize
333 * we will often be called with the same page as last time and
334 * a consecutive offset. Optimize this special case.
335 */
336 if (bio->bi_vcnt > 0) {
337 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
338
339 if (page == prev->bv_page &&
340 offset == prev->bv_offset + prev->bv_len) {
341 prev->bv_len += len;
342 if (q->merge_bvec_fn &&
343 q->merge_bvec_fn(q, bio, prev) < len) {
344 prev->bv_len -= len;
345 return 0;
346 }
347
348 goto done;
349 }
350 }
351
352 if (bio->bi_vcnt >= bio->bi_max_vecs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 return 0;
354
355 /*
356 * we might lose a segment or two here, but rather that than
357 * make this too complex.
358 */
359
360 while (bio->bi_phys_segments >= q->max_phys_segments
361 || bio->bi_hw_segments >= q->max_hw_segments
362 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
363
364 if (retried_segments)
365 return 0;
366
367 retried_segments = 1;
368 blk_recount_segments(q, bio);
369 }
370
371 /*
372 * setup the new entry, we might clear it again later if we
373 * cannot add the page
374 */
375 bvec = &bio->bi_io_vec[bio->bi_vcnt];
376 bvec->bv_page = page;
377 bvec->bv_len = len;
378 bvec->bv_offset = offset;
379
380 /*
381 * if queue has other restrictions (eg varying max sector size
382 * depending on offset), it can specify a merge_bvec_fn in the
383 * queue to get further control
384 */
385 if (q->merge_bvec_fn) {
386 /*
387 * merge_bvec_fn() returns number of bytes it can accept
388 * at this offset
389 */
390 if (q->merge_bvec_fn(q, bio, bvec) < len) {
391 bvec->bv_page = NULL;
392 bvec->bv_len = 0;
393 bvec->bv_offset = 0;
394 return 0;
395 }
396 }
397
398 /* If we may be able to merge these biovecs, force a recount */
399 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
400 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
401 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
402
403 bio->bi_vcnt++;
404 bio->bi_phys_segments++;
405 bio->bi_hw_segments++;
Jens Axboe80cfd542006-01-06 09:43:28 +0100406 done:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 bio->bi_size += len;
408 return len;
409}
410
411/**
Mike Christie6e68af62005-11-11 05:30:27 -0600412 * bio_add_pc_page - attempt to add page to bio
413 * @bio: destination bio
414 * @page: page to add
415 * @len: vec entry length
416 * @offset: vec entry offset
417 *
418 * Attempt to add a page to the bio_vec maplist. This can fail for a
419 * number of reasons, such as the bio being full or target block
420 * device limitations. The target block device must allow bio's
421 * smaller than PAGE_SIZE, so it is always possible to add a single
422 * page to an empty bio. This should only be used by REQ_PC bios.
423 */
424int bio_add_pc_page(request_queue_t *q, struct bio *bio, struct page *page,
425 unsigned int len, unsigned int offset)
426{
Mike Christiedefd94b2005-12-05 02:37:06 -0600427 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
Mike Christie6e68af62005-11-11 05:30:27 -0600428}
429
430/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 * bio_add_page - attempt to add page to bio
432 * @bio: destination bio
433 * @page: page to add
434 * @len: vec entry length
435 * @offset: vec entry offset
436 *
437 * Attempt to add a page to the bio_vec maplist. This can fail for a
438 * number of reasons, such as the bio being full or target block
439 * device limitations. The target block device must allow bio's
440 * smaller than PAGE_SIZE, so it is always possible to add a single
441 * page to an empty bio.
442 */
443int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
444 unsigned int offset)
445{
Mike Christiedefd94b2005-12-05 02:37:06 -0600446 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
447 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448}
449
450struct bio_map_data {
451 struct bio_vec *iovecs;
452 void __user *userptr;
453};
454
455static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio)
456{
457 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
458 bio->bi_private = bmd;
459}
460
461static void bio_free_map_data(struct bio_map_data *bmd)
462{
463 kfree(bmd->iovecs);
464 kfree(bmd);
465}
466
467static struct bio_map_data *bio_alloc_map_data(int nr_segs)
468{
469 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL);
470
471 if (!bmd)
472 return NULL;
473
474 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL);
475 if (bmd->iovecs)
476 return bmd;
477
478 kfree(bmd);
479 return NULL;
480}
481
482/**
483 * bio_uncopy_user - finish previously mapped bio
484 * @bio: bio being terminated
485 *
486 * Free pages allocated from bio_copy_user() and write back data
487 * to user space in case of a read.
488 */
489int bio_uncopy_user(struct bio *bio)
490{
491 struct bio_map_data *bmd = bio->bi_private;
492 const int read = bio_data_dir(bio) == READ;
493 struct bio_vec *bvec;
494 int i, ret = 0;
495
496 __bio_for_each_segment(bvec, bio, i, 0) {
497 char *addr = page_address(bvec->bv_page);
498 unsigned int len = bmd->iovecs[i].bv_len;
499
500 if (read && !ret && copy_to_user(bmd->userptr, addr, len))
501 ret = -EFAULT;
502
503 __free_page(bvec->bv_page);
504 bmd->userptr += len;
505 }
506 bio_free_map_data(bmd);
507 bio_put(bio);
508 return ret;
509}
510
511/**
512 * bio_copy_user - copy user data to bio
513 * @q: destination block queue
514 * @uaddr: start of user address
515 * @len: length in bytes
516 * @write_to_vm: bool indicating writing to pages or not
517 *
518 * Prepares and returns a bio for indirect user io, bouncing data
519 * to/from kernel pages as necessary. Must be paired with
520 * call bio_uncopy_user() on io completion.
521 */
522struct bio *bio_copy_user(request_queue_t *q, unsigned long uaddr,
523 unsigned int len, int write_to_vm)
524{
525 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
526 unsigned long start = uaddr >> PAGE_SHIFT;
527 struct bio_map_data *bmd;
528 struct bio_vec *bvec;
529 struct page *page;
530 struct bio *bio;
531 int i, ret;
532
533 bmd = bio_alloc_map_data(end - start);
534 if (!bmd)
535 return ERR_PTR(-ENOMEM);
536
537 bmd->userptr = (void __user *) uaddr;
538
539 ret = -ENOMEM;
540 bio = bio_alloc(GFP_KERNEL, end - start);
541 if (!bio)
542 goto out_bmd;
543
544 bio->bi_rw |= (!write_to_vm << BIO_RW);
545
546 ret = 0;
547 while (len) {
548 unsigned int bytes = PAGE_SIZE;
549
550 if (bytes > len)
551 bytes = len;
552
553 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
554 if (!page) {
555 ret = -ENOMEM;
556 break;
557 }
558
Mike Christiedefd94b2005-12-05 02:37:06 -0600559 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 ret = -EINVAL;
561 break;
562 }
563
564 len -= bytes;
565 }
566
567 if (ret)
568 goto cleanup;
569
570 /*
571 * success
572 */
573 if (!write_to_vm) {
574 char __user *p = (char __user *) uaddr;
575
576 /*
577 * for a write, copy in data to kernel pages
578 */
579 ret = -EFAULT;
580 bio_for_each_segment(bvec, bio, i) {
581 char *addr = page_address(bvec->bv_page);
582
583 if (copy_from_user(addr, p, bvec->bv_len))
584 goto cleanup;
585 p += bvec->bv_len;
586 }
587 }
588
589 bio_set_map_data(bmd, bio);
590 return bio;
591cleanup:
592 bio_for_each_segment(bvec, bio, i)
593 __free_page(bvec->bv_page);
594
595 bio_put(bio);
596out_bmd:
597 bio_free_map_data(bmd);
598 return ERR_PTR(ret);
599}
600
James Bottomley f1970ba2005-06-20 14:06:52 +0200601static struct bio *__bio_map_user_iov(request_queue_t *q,
602 struct block_device *bdev,
603 struct sg_iovec *iov, int iov_count,
604 int write_to_vm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605{
James Bottomley f1970ba2005-06-20 14:06:52 +0200606 int i, j;
607 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 struct page **pages;
609 struct bio *bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200610 int cur_page = 0;
611 int ret, offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
James Bottomley f1970ba2005-06-20 14:06:52 +0200613 for (i = 0; i < iov_count; i++) {
614 unsigned long uaddr = (unsigned long)iov[i].iov_base;
615 unsigned long len = iov[i].iov_len;
616 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
617 unsigned long start = uaddr >> PAGE_SHIFT;
618
619 nr_pages += end - start;
620 /*
621 * transfer and buffer must be aligned to at least hardsector
622 * size for now, in the future we can relax this restriction
623 */
624 if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q)))
625 return ERR_PTR(-EINVAL);
626 }
627
628 if (!nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 return ERR_PTR(-EINVAL);
630
631 bio = bio_alloc(GFP_KERNEL, nr_pages);
632 if (!bio)
633 return ERR_PTR(-ENOMEM);
634
635 ret = -ENOMEM;
636 pages = kmalloc(nr_pages * sizeof(struct page *), GFP_KERNEL);
637 if (!pages)
638 goto out;
639
James Bottomley f1970ba2005-06-20 14:06:52 +0200640 memset(pages, 0, nr_pages * sizeof(struct page *));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
James Bottomley f1970ba2005-06-20 14:06:52 +0200642 for (i = 0; i < iov_count; i++) {
643 unsigned long uaddr = (unsigned long)iov[i].iov_base;
644 unsigned long len = iov[i].iov_len;
645 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
646 unsigned long start = uaddr >> PAGE_SHIFT;
647 const int local_nr_pages = end - start;
648 const int page_limit = cur_page + local_nr_pages;
649
650 down_read(&current->mm->mmap_sem);
651 ret = get_user_pages(current, current->mm, uaddr,
652 local_nr_pages,
653 write_to_vm, 0, &pages[cur_page], NULL);
654 up_read(&current->mm->mmap_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
James Bottomley f1970ba2005-06-20 14:06:52 +0200656 if (ret < local_nr_pages)
657 goto out_unmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
James Bottomley f1970ba2005-06-20 14:06:52 +0200660 offset = uaddr & ~PAGE_MASK;
661 for (j = cur_page; j < page_limit; j++) {
662 unsigned int bytes = PAGE_SIZE - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
James Bottomley f1970ba2005-06-20 14:06:52 +0200664 if (len <= 0)
665 break;
666
667 if (bytes > len)
668 bytes = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669
James Bottomley f1970ba2005-06-20 14:06:52 +0200670 /*
671 * sorry...
672 */
Mike Christiedefd94b2005-12-05 02:37:06 -0600673 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
674 bytes)
James Bottomley f1970ba2005-06-20 14:06:52 +0200675 break;
676
677 len -= bytes;
678 offset = 0;
679 }
680
681 cur_page = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 /*
James Bottomley f1970ba2005-06-20 14:06:52 +0200683 * release the pages we didn't map into the bio, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 */
James Bottomley f1970ba2005-06-20 14:06:52 +0200685 while (j < page_limit)
686 page_cache_release(pages[j++]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 }
688
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 kfree(pages);
690
691 /*
692 * set data direction, and check if mapped pages need bouncing
693 */
694 if (!write_to_vm)
695 bio->bi_rw |= (1 << BIO_RW);
696
James Bottomley f1970ba2005-06-20 14:06:52 +0200697 bio->bi_bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 bio->bi_flags |= (1 << BIO_USER_MAPPED);
699 return bio;
James Bottomley f1970ba2005-06-20 14:06:52 +0200700
701 out_unmap:
702 for (i = 0; i < nr_pages; i++) {
703 if(!pages[i])
704 break;
705 page_cache_release(pages[i]);
706 }
707 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 kfree(pages);
709 bio_put(bio);
710 return ERR_PTR(ret);
711}
712
713/**
714 * bio_map_user - map user address into bio
Martin Waitz67be2dd2005-05-01 08:59:26 -0700715 * @q: the request_queue_t for the bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 * @bdev: destination block device
717 * @uaddr: start of user address
718 * @len: length in bytes
719 * @write_to_vm: bool indicating writing to pages or not
720 *
721 * Map the user space address into a bio suitable for io to a block
722 * device. Returns an error pointer in case of error.
723 */
724struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev,
725 unsigned long uaddr, unsigned int len, int write_to_vm)
726{
James Bottomley f1970ba2005-06-20 14:06:52 +0200727 struct sg_iovec iov;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
viro@ZenIV.linux.org.uk3f703532005-09-09 16:53:56 +0100729 iov.iov_base = (void __user *)uaddr;
James Bottomley f1970ba2005-06-20 14:06:52 +0200730 iov.iov_len = len;
731
732 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
733}
734
735/**
736 * bio_map_user_iov - map user sg_iovec table into bio
737 * @q: the request_queue_t for the bio
738 * @bdev: destination block device
739 * @iov: the iovec.
740 * @iov_count: number of elements in the iovec
741 * @write_to_vm: bool indicating writing to pages or not
742 *
743 * Map the user space address into a bio suitable for io to a block
744 * device. Returns an error pointer in case of error.
745 */
746struct bio *bio_map_user_iov(request_queue_t *q, struct block_device *bdev,
747 struct sg_iovec *iov, int iov_count,
748 int write_to_vm)
749{
750 struct bio *bio;
751 int len = 0, i;
752
753 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
755 if (IS_ERR(bio))
756 return bio;
757
758 /*
759 * subtle -- if __bio_map_user() ended up bouncing a bio,
760 * it would normally disappear when its bi_end_io is run.
761 * however, we need it for the unmap, so grab an extra
762 * reference to it
763 */
764 bio_get(bio);
765
James Bottomley f1970ba2005-06-20 14:06:52 +0200766 for (i = 0; i < iov_count; i++)
767 len += iov[i].iov_len;
768
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 if (bio->bi_size == len)
770 return bio;
771
772 /*
773 * don't support partial mappings
774 */
775 bio_endio(bio, bio->bi_size, 0);
776 bio_unmap_user(bio);
777 return ERR_PTR(-EINVAL);
778}
779
780static void __bio_unmap_user(struct bio *bio)
781{
782 struct bio_vec *bvec;
783 int i;
784
785 /*
786 * make sure we dirty pages we wrote to
787 */
788 __bio_for_each_segment(bvec, bio, i, 0) {
789 if (bio_data_dir(bio) == READ)
790 set_page_dirty_lock(bvec->bv_page);
791
792 page_cache_release(bvec->bv_page);
793 }
794
795 bio_put(bio);
796}
797
798/**
799 * bio_unmap_user - unmap a bio
800 * @bio: the bio being unmapped
801 *
802 * Unmap a bio previously mapped by bio_map_user(). Must be called with
803 * a process context.
804 *
805 * bio_unmap_user() may sleep.
806 */
807void bio_unmap_user(struct bio *bio)
808{
809 __bio_unmap_user(bio);
810 bio_put(bio);
811}
812
Jens Axboeb8238252005-06-20 14:05:27 +0200813static int bio_map_kern_endio(struct bio *bio, unsigned int bytes_done, int err)
814{
815 if (bio->bi_size)
816 return 1;
817
818 bio_put(bio);
819 return 0;
820}
821
822
Mike Christie df46b9a2005-06-20 14:04:44 +0200823static struct bio *__bio_map_kern(request_queue_t *q, void *data,
Al Viro27496a82005-10-21 03:20:48 -0400824 unsigned int len, gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +0200825{
826 unsigned long kaddr = (unsigned long)data;
827 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
828 unsigned long start = kaddr >> PAGE_SHIFT;
829 const int nr_pages = end - start;
830 int offset, i;
831 struct bio *bio;
832
833 bio = bio_alloc(gfp_mask, nr_pages);
834 if (!bio)
835 return ERR_PTR(-ENOMEM);
836
837 offset = offset_in_page(kaddr);
838 for (i = 0; i < nr_pages; i++) {
839 unsigned int bytes = PAGE_SIZE - offset;
840
841 if (len <= 0)
842 break;
843
844 if (bytes > len)
845 bytes = len;
846
Mike Christiedefd94b2005-12-05 02:37:06 -0600847 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
848 offset) < bytes)
Mike Christie df46b9a2005-06-20 14:04:44 +0200849 break;
850
851 data += bytes;
852 len -= bytes;
853 offset = 0;
854 }
855
Jens Axboeb8238252005-06-20 14:05:27 +0200856 bio->bi_end_io = bio_map_kern_endio;
Mike Christie df46b9a2005-06-20 14:04:44 +0200857 return bio;
858}
859
860/**
861 * bio_map_kern - map kernel address into bio
862 * @q: the request_queue_t for the bio
863 * @data: pointer to buffer to map
864 * @len: length in bytes
865 * @gfp_mask: allocation flags for bio allocation
866 *
867 * Map the kernel address into a bio suitable for io to a block
868 * device. Returns an error pointer in case of error.
869 */
870struct bio *bio_map_kern(request_queue_t *q, void *data, unsigned int len,
Al Viro27496a82005-10-21 03:20:48 -0400871 gfp_t gfp_mask)
Mike Christie df46b9a2005-06-20 14:04:44 +0200872{
873 struct bio *bio;
874
875 bio = __bio_map_kern(q, data, len, gfp_mask);
876 if (IS_ERR(bio))
877 return bio;
878
879 if (bio->bi_size == len)
880 return bio;
881
882 /*
883 * Don't support partial mappings.
884 */
885 bio_put(bio);
886 return ERR_PTR(-EINVAL);
887}
888
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889/*
890 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
891 * for performing direct-IO in BIOs.
892 *
893 * The problem is that we cannot run set_page_dirty() from interrupt context
894 * because the required locks are not interrupt-safe. So what we can do is to
895 * mark the pages dirty _before_ performing IO. And in interrupt context,
896 * check that the pages are still dirty. If so, fine. If not, redirty them
897 * in process context.
898 *
899 * We special-case compound pages here: normally this means reads into hugetlb
900 * pages. The logic in here doesn't really work right for compound pages
901 * because the VM does not uniformly chase down the head page in all cases.
902 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
903 * handle them at all. So we skip compound pages here at an early stage.
904 *
905 * Note that this code is very hard to test under normal circumstances because
906 * direct-io pins the pages with get_user_pages(). This makes
907 * is_page_cache_freeable return false, and the VM will not clean the pages.
908 * But other code (eg, pdflush) could clean the pages if they are mapped
909 * pagecache.
910 *
911 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
912 * deferred bio dirtying paths.
913 */
914
915/*
916 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
917 */
918void bio_set_pages_dirty(struct bio *bio)
919{
920 struct bio_vec *bvec = bio->bi_io_vec;
921 int i;
922
923 for (i = 0; i < bio->bi_vcnt; i++) {
924 struct page *page = bvec[i].bv_page;
925
926 if (page && !PageCompound(page))
927 set_page_dirty_lock(page);
928 }
929}
930
931static void bio_release_pages(struct bio *bio)
932{
933 struct bio_vec *bvec = bio->bi_io_vec;
934 int i;
935
936 for (i = 0; i < bio->bi_vcnt; i++) {
937 struct page *page = bvec[i].bv_page;
938
939 if (page)
940 put_page(page);
941 }
942}
943
944/*
945 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
946 * If they are, then fine. If, however, some pages are clean then they must
947 * have been written out during the direct-IO read. So we take another ref on
948 * the BIO and the offending pages and re-dirty the pages in process context.
949 *
950 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
951 * here on. It will run one page_cache_release() against each page and will
952 * run one bio_put() against the BIO.
953 */
954
955static void bio_dirty_fn(void *data);
956
957static DECLARE_WORK(bio_dirty_work, bio_dirty_fn, NULL);
958static DEFINE_SPINLOCK(bio_dirty_lock);
959static struct bio *bio_dirty_list;
960
961/*
962 * This runs in process context
963 */
964static void bio_dirty_fn(void *data)
965{
966 unsigned long flags;
967 struct bio *bio;
968
969 spin_lock_irqsave(&bio_dirty_lock, flags);
970 bio = bio_dirty_list;
971 bio_dirty_list = NULL;
972 spin_unlock_irqrestore(&bio_dirty_lock, flags);
973
974 while (bio) {
975 struct bio *next = bio->bi_private;
976
977 bio_set_pages_dirty(bio);
978 bio_release_pages(bio);
979 bio_put(bio);
980 bio = next;
981 }
982}
983
984void bio_check_pages_dirty(struct bio *bio)
985{
986 struct bio_vec *bvec = bio->bi_io_vec;
987 int nr_clean_pages = 0;
988 int i;
989
990 for (i = 0; i < bio->bi_vcnt; i++) {
991 struct page *page = bvec[i].bv_page;
992
993 if (PageDirty(page) || PageCompound(page)) {
994 page_cache_release(page);
995 bvec[i].bv_page = NULL;
996 } else {
997 nr_clean_pages++;
998 }
999 }
1000
1001 if (nr_clean_pages) {
1002 unsigned long flags;
1003
1004 spin_lock_irqsave(&bio_dirty_lock, flags);
1005 bio->bi_private = bio_dirty_list;
1006 bio_dirty_list = bio;
1007 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1008 schedule_work(&bio_dirty_work);
1009 } else {
1010 bio_put(bio);
1011 }
1012}
1013
1014/**
1015 * bio_endio - end I/O on a bio
1016 * @bio: bio
1017 * @bytes_done: number of bytes completed
1018 * @error: error, if any
1019 *
1020 * Description:
1021 * bio_endio() will end I/O on @bytes_done number of bytes. This may be
1022 * just a partial part of the bio, or it may be the whole bio. bio_endio()
1023 * is the preferred way to end I/O on a bio, it takes care of decrementing
1024 * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and
1025 * and one of the established -Exxxx (-EIO, for instance) error values in
1026 * case something went wrong. Noone should call bi_end_io() directly on
1027 * a bio unless they own it and thus know that it has an end_io function.
1028 **/
1029void bio_endio(struct bio *bio, unsigned int bytes_done, int error)
1030{
1031 if (error)
1032 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1033
1034 if (unlikely(bytes_done > bio->bi_size)) {
1035 printk("%s: want %u bytes done, only %u left\n", __FUNCTION__,
1036 bytes_done, bio->bi_size);
1037 bytes_done = bio->bi_size;
1038 }
1039
1040 bio->bi_size -= bytes_done;
1041 bio->bi_sector += (bytes_done >> 9);
1042
1043 if (bio->bi_end_io)
1044 bio->bi_end_io(bio, bytes_done, error);
1045}
1046
1047void bio_pair_release(struct bio_pair *bp)
1048{
1049 if (atomic_dec_and_test(&bp->cnt)) {
1050 struct bio *master = bp->bio1.bi_private;
1051
1052 bio_endio(master, master->bi_size, bp->error);
1053 mempool_free(bp, bp->bio2.bi_private);
1054 }
1055}
1056
1057static int bio_pair_end_1(struct bio * bi, unsigned int done, int err)
1058{
1059 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1060
1061 if (err)
1062 bp->error = err;
1063
1064 if (bi->bi_size)
1065 return 1;
1066
1067 bio_pair_release(bp);
1068 return 0;
1069}
1070
1071static int bio_pair_end_2(struct bio * bi, unsigned int done, int err)
1072{
1073 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1074
1075 if (err)
1076 bp->error = err;
1077
1078 if (bi->bi_size)
1079 return 1;
1080
1081 bio_pair_release(bp);
1082 return 0;
1083}
1084
1085/*
1086 * split a bio - only worry about a bio with a single page
1087 * in it's iovec
1088 */
1089struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
1090{
1091 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
1092
1093 if (!bp)
1094 return bp;
1095
1096 BUG_ON(bi->bi_vcnt != 1);
1097 BUG_ON(bi->bi_idx != 0);
1098 atomic_set(&bp->cnt, 3);
1099 bp->error = 0;
1100 bp->bio1 = *bi;
1101 bp->bio2 = *bi;
1102 bp->bio2.bi_sector += first_sectors;
1103 bp->bio2.bi_size -= first_sectors << 9;
1104 bp->bio1.bi_size = first_sectors << 9;
1105
1106 bp->bv1 = bi->bi_io_vec[0];
1107 bp->bv2 = bi->bi_io_vec[0];
1108 bp->bv2.bv_offset += first_sectors << 9;
1109 bp->bv2.bv_len -= first_sectors << 9;
1110 bp->bv1.bv_len = first_sectors << 9;
1111
1112 bp->bio1.bi_io_vec = &bp->bv1;
1113 bp->bio2.bi_io_vec = &bp->bv2;
1114
1115 bp->bio1.bi_end_io = bio_pair_end_1;
1116 bp->bio2.bi_end_io = bio_pair_end_2;
1117
1118 bp->bio1.bi_private = bi;
1119 bp->bio2.bi_private = pool;
1120
1121 return bp;
1122}
1123
Al Virodd0fc662005-10-07 07:46:04 +01001124static void *bio_pair_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125{
1126 return kmalloc(sizeof(struct bio_pair), gfp_flags);
1127}
1128
1129static void bio_pair_free(void *bp, void *data)
1130{
1131 kfree(bp);
1132}
1133
1134
1135/*
1136 * create memory pools for biovec's in a bio_set.
1137 * use the global biovec slabs created for general use.
1138 */
1139static int biovec_create_pools(struct bio_set *bs, int pool_entries, int scale)
1140{
1141 int i;
1142
1143 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1144 struct biovec_slab *bp = bvec_slabs + i;
1145 mempool_t **bvp = bs->bvec_pools + i;
1146
1147 if (i >= scale)
1148 pool_entries >>= 1;
1149
1150 *bvp = mempool_create(pool_entries, mempool_alloc_slab,
1151 mempool_free_slab, bp->slab);
1152 if (!*bvp)
1153 return -ENOMEM;
1154 }
1155 return 0;
1156}
1157
1158static void biovec_free_pools(struct bio_set *bs)
1159{
1160 int i;
1161
1162 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1163 mempool_t *bvp = bs->bvec_pools[i];
1164
1165 if (bvp)
1166 mempool_destroy(bvp);
1167 }
1168
1169}
1170
1171void bioset_free(struct bio_set *bs)
1172{
1173 if (bs->bio_pool)
1174 mempool_destroy(bs->bio_pool);
1175
1176 biovec_free_pools(bs);
1177
1178 kfree(bs);
1179}
1180
1181struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size, int scale)
1182{
1183 struct bio_set *bs = kmalloc(sizeof(*bs), GFP_KERNEL);
1184
1185 if (!bs)
1186 return NULL;
1187
1188 memset(bs, 0, sizeof(*bs));
1189 bs->bio_pool = mempool_create(bio_pool_size, mempool_alloc_slab,
1190 mempool_free_slab, bio_slab);
1191
1192 if (!bs->bio_pool)
1193 goto bad;
1194
1195 if (!biovec_create_pools(bs, bvec_pool_size, scale))
1196 return bs;
1197
1198bad:
1199 bioset_free(bs);
1200 return NULL;
1201}
1202
1203static void __init biovec_init_slabs(void)
1204{
1205 int i;
1206
1207 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1208 int size;
1209 struct biovec_slab *bvs = bvec_slabs + i;
1210
1211 size = bvs->nr_vecs * sizeof(struct bio_vec);
1212 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1213 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1214 }
1215}
1216
1217static int __init init_bio(void)
1218{
1219 int megabytes, bvec_pool_entries;
1220 int scale = BIOVEC_NR_POOLS;
1221
1222 bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0,
1223 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1224
1225 biovec_init_slabs();
1226
1227 megabytes = nr_free_pages() >> (20 - PAGE_SHIFT);
1228
1229 /*
1230 * find out where to start scaling
1231 */
1232 if (megabytes <= 16)
1233 scale = 0;
1234 else if (megabytes <= 32)
1235 scale = 1;
1236 else if (megabytes <= 64)
1237 scale = 2;
1238 else if (megabytes <= 96)
1239 scale = 3;
1240 else if (megabytes <= 128)
1241 scale = 4;
1242
1243 /*
1244 * scale number of entries
1245 */
1246 bvec_pool_entries = megabytes * 2;
1247 if (bvec_pool_entries > 256)
1248 bvec_pool_entries = 256;
1249
1250 fs_bio_set = bioset_create(BIO_POOL_SIZE, bvec_pool_entries, scale);
1251 if (!fs_bio_set)
1252 panic("bio: can't allocate bios\n");
1253
1254 bio_split_pool = mempool_create(BIO_SPLIT_ENTRIES,
1255 bio_pair_alloc, bio_pair_free, NULL);
1256 if (!bio_split_pool)
1257 panic("bio: can't create split pool\n");
1258
1259 return 0;
1260}
1261
1262subsys_initcall(init_bio);
1263
1264EXPORT_SYMBOL(bio_alloc);
1265EXPORT_SYMBOL(bio_put);
Peter Osterlund36763472005-09-06 15:16:42 -07001266EXPORT_SYMBOL(bio_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267EXPORT_SYMBOL(bio_endio);
1268EXPORT_SYMBOL(bio_init);
1269EXPORT_SYMBOL(__bio_clone);
1270EXPORT_SYMBOL(bio_clone);
1271EXPORT_SYMBOL(bio_phys_segments);
1272EXPORT_SYMBOL(bio_hw_segments);
1273EXPORT_SYMBOL(bio_add_page);
Mike Christie6e68af62005-11-11 05:30:27 -06001274EXPORT_SYMBOL(bio_add_pc_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275EXPORT_SYMBOL(bio_get_nr_vecs);
1276EXPORT_SYMBOL(bio_map_user);
1277EXPORT_SYMBOL(bio_unmap_user);
Mike Christie df46b9a2005-06-20 14:04:44 +02001278EXPORT_SYMBOL(bio_map_kern);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279EXPORT_SYMBOL(bio_pair_release);
1280EXPORT_SYMBOL(bio_split);
1281EXPORT_SYMBOL(bio_split_pool);
1282EXPORT_SYMBOL(bio_copy_user);
1283EXPORT_SYMBOL(bio_uncopy_user);
1284EXPORT_SYMBOL(bioset_create);
1285EXPORT_SYMBOL(bioset_free);
1286EXPORT_SYMBOL(bio_alloc_bioset);