]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/bio.c
[PATCH] knfsd: fix auto-sizing of nfsd request/reply buffers
[mirror_ubuntu-bionic-kernel.git] / fs / bio.c
CommitLineData
1da177e4 1/*
0fe23479 2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
1da177e4
LT
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>
2056a782 28#include <linux/blktrace_api.h>
f1970baf 29#include <scsi/sg.h> /* for struct sg_iovec */
1da177e4
LT
30
31#define BIO_POOL_SIZE 256
32
fa3536cc 33static kmem_cache_t *bio_slab __read_mostly;
1da177e4
LT
34
35#define BIOVEC_NR_POOLS 6
36
37/*
38 * a small number of entries is fine, not going to be performance critical.
39 * basically we just need to survive
40 */
41#define BIO_SPLIT_ENTRIES 8
fa3536cc 42mempool_t *bio_split_pool __read_mostly;
1da177e4
LT
43
44struct biovec_slab {
45 int nr_vecs;
46 char *name;
47 kmem_cache_t *slab;
48};
49
50/*
51 * if you change this list, also change bvec_alloc or things will
52 * break badly! cannot be bigger than what you can fit into an
53 * unsigned short
54 */
55
56#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
6c036527 57static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
1da177e4
LT
58 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
59};
60#undef BV
61
62/*
63 * bio_set is used to allow other portions of the IO system to
64 * allocate their own private memory pools for bio and iovec structures.
65 * These memory pools in turn all allocate from the bio_slab
66 * and the bvec_slabs[].
67 */
68struct bio_set {
69 mempool_t *bio_pool;
70 mempool_t *bvec_pools[BIOVEC_NR_POOLS];
71};
72
73/*
74 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
75 * IO code that does not need private memory pools.
76 */
77static struct bio_set *fs_bio_set;
78
dd0fc66f 79static inline struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs)
1da177e4
LT
80{
81 struct bio_vec *bvl;
82 struct biovec_slab *bp;
83
84 /*
85 * see comment near bvec_array define!
86 */
87 switch (nr) {
88 case 1 : *idx = 0; break;
89 case 2 ... 4: *idx = 1; break;
90 case 5 ... 16: *idx = 2; break;
91 case 17 ... 64: *idx = 3; break;
92 case 65 ... 128: *idx = 4; break;
93 case 129 ... BIO_MAX_PAGES: *idx = 5; break;
94 default:
95 return NULL;
96 }
97 /*
98 * idx now points to the pool we want to allocate from
99 */
100
101 bp = bvec_slabs + *idx;
102 bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask);
103 if (bvl)
104 memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec));
105
106 return bvl;
107}
108
3676347a 109void bio_free(struct bio *bio, struct bio_set *bio_set)
1da177e4
LT
110{
111 const int pool_idx = BIO_POOL_IDX(bio);
1da177e4
LT
112
113 BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS);
114
3676347a
PO
115 mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]);
116 mempool_free(bio, bio_set->bio_pool);
117}
118
119/*
120 * default destructor for a bio allocated with bio_alloc_bioset()
121 */
122static void bio_fs_destructor(struct bio *bio)
123{
124 bio_free(bio, fs_bio_set);
1da177e4
LT
125}
126
858119e1 127void bio_init(struct bio *bio)
1da177e4
LT
128{
129 bio->bi_next = NULL;
0ea60b5a 130 bio->bi_bdev = NULL;
1da177e4
LT
131 bio->bi_flags = 1 << BIO_UPTODATE;
132 bio->bi_rw = 0;
133 bio->bi_vcnt = 0;
134 bio->bi_idx = 0;
135 bio->bi_phys_segments = 0;
136 bio->bi_hw_segments = 0;
137 bio->bi_hw_front_size = 0;
138 bio->bi_hw_back_size = 0;
139 bio->bi_size = 0;
140 bio->bi_max_vecs = 0;
141 bio->bi_end_io = NULL;
142 atomic_set(&bio->bi_cnt, 1);
143 bio->bi_private = NULL;
144}
145
146/**
147 * bio_alloc_bioset - allocate a bio for I/O
148 * @gfp_mask: the GFP_ mask given to the slab allocator
149 * @nr_iovecs: number of iovecs to pre-allocate
67be2dd1 150 * @bs: the bio_set to allocate from
1da177e4
LT
151 *
152 * Description:
153 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation.
154 * If %__GFP_WAIT is set then we will block on the internal pool waiting
155 * for a &struct bio to become free.
156 *
157 * allocate bio and iovecs from the memory pools specified by the
158 * bio_set structure.
159 **/
dd0fc66f 160struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
1da177e4
LT
161{
162 struct bio *bio = mempool_alloc(bs->bio_pool, gfp_mask);
163
164 if (likely(bio)) {
165 struct bio_vec *bvl = NULL;
166
167 bio_init(bio);
168 if (likely(nr_iovecs)) {
169 unsigned long idx;
170
171 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
172 if (unlikely(!bvl)) {
173 mempool_free(bio, bs->bio_pool);
174 bio = NULL;
175 goto out;
176 }
177 bio->bi_flags |= idx << BIO_POOL_OFFSET;
178 bio->bi_max_vecs = bvec_slabs[idx].nr_vecs;
179 }
180 bio->bi_io_vec = bvl;
1da177e4
LT
181 }
182out:
183 return bio;
184}
185
dd0fc66f 186struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
1da177e4 187{
3676347a
PO
188 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
189
190 if (bio)
191 bio->bi_destructor = bio_fs_destructor;
192
193 return bio;
1da177e4
LT
194}
195
196void zero_fill_bio(struct bio *bio)
197{
198 unsigned long flags;
199 struct bio_vec *bv;
200 int i;
201
202 bio_for_each_segment(bv, bio, i) {
203 char *data = bvec_kmap_irq(bv, &flags);
204 memset(data, 0, bv->bv_len);
205 flush_dcache_page(bv->bv_page);
206 bvec_kunmap_irq(data, &flags);
207 }
208}
209EXPORT_SYMBOL(zero_fill_bio);
210
211/**
212 * bio_put - release a reference to a bio
213 * @bio: bio to release reference to
214 *
215 * Description:
216 * Put a reference to a &struct bio, either one you have gotten with
217 * bio_alloc or bio_get. The last put of a bio will free it.
218 **/
219void bio_put(struct bio *bio)
220{
221 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
222
223 /*
224 * last put frees it
225 */
226 if (atomic_dec_and_test(&bio->bi_cnt)) {
227 bio->bi_next = NULL;
228 bio->bi_destructor(bio);
229 }
230}
231
232inline int bio_phys_segments(request_queue_t *q, struct bio *bio)
233{
234 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
235 blk_recount_segments(q, bio);
236
237 return bio->bi_phys_segments;
238}
239
240inline int bio_hw_segments(request_queue_t *q, struct bio *bio)
241{
242 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
243 blk_recount_segments(q, bio);
244
245 return bio->bi_hw_segments;
246}
247
248/**
249 * __bio_clone - clone a bio
250 * @bio: destination bio
251 * @bio_src: bio to clone
252 *
253 * Clone a &bio. Caller will own the returned bio, but not
254 * the actual data it points to. Reference count of returned
255 * bio will be one.
256 */
858119e1 257void __bio_clone(struct bio *bio, struct bio *bio_src)
1da177e4
LT
258{
259 request_queue_t *q = bdev_get_queue(bio_src->bi_bdev);
260
e525e153
AM
261 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
262 bio_src->bi_max_vecs * sizeof(struct bio_vec));
1da177e4
LT
263
264 bio->bi_sector = bio_src->bi_sector;
265 bio->bi_bdev = bio_src->bi_bdev;
266 bio->bi_flags |= 1 << BIO_CLONED;
267 bio->bi_rw = bio_src->bi_rw;
1da177e4
LT
268 bio->bi_vcnt = bio_src->bi_vcnt;
269 bio->bi_size = bio_src->bi_size;
a5453be4 270 bio->bi_idx = bio_src->bi_idx;
1da177e4
LT
271 bio_phys_segments(q, bio);
272 bio_hw_segments(q, bio);
273}
274
275/**
276 * bio_clone - clone a bio
277 * @bio: bio to clone
278 * @gfp_mask: allocation priority
279 *
280 * Like __bio_clone, only also allocates the returned bio
281 */
dd0fc66f 282struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
1da177e4
LT
283{
284 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
285
3676347a
PO
286 if (b) {
287 b->bi_destructor = bio_fs_destructor;
1da177e4 288 __bio_clone(b, bio);
3676347a 289 }
1da177e4
LT
290
291 return b;
292}
293
294/**
295 * bio_get_nr_vecs - return approx number of vecs
296 * @bdev: I/O target
297 *
298 * Return the approximate number of pages we can send to this target.
299 * There's no guarantee that you will be able to fit this number of pages
300 * into a bio, it does not account for dynamic restrictions that vary
301 * on offset.
302 */
303int bio_get_nr_vecs(struct block_device *bdev)
304{
305 request_queue_t *q = bdev_get_queue(bdev);
306 int nr_pages;
307
308 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
309 if (nr_pages > q->max_phys_segments)
310 nr_pages = q->max_phys_segments;
311 if (nr_pages > q->max_hw_segments)
312 nr_pages = q->max_hw_segments;
313
314 return nr_pages;
315}
316
317static int __bio_add_page(request_queue_t *q, struct bio *bio, struct page
defd94b7
MC
318 *page, unsigned int len, unsigned int offset,
319 unsigned short max_sectors)
1da177e4
LT
320{
321 int retried_segments = 0;
322 struct bio_vec *bvec;
323
324 /*
325 * cloned bio must not modify vec list
326 */
327 if (unlikely(bio_flagged(bio, BIO_CLONED)))
328 return 0;
329
80cfd548 330 if (((bio->bi_size + len) >> 9) > max_sectors)
1da177e4
LT
331 return 0;
332
80cfd548
JA
333 /*
334 * For filesystems with a blocksize smaller than the pagesize
335 * we will often be called with the same page as last time and
336 * a consecutive offset. Optimize this special case.
337 */
338 if (bio->bi_vcnt > 0) {
339 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
340
341 if (page == prev->bv_page &&
342 offset == prev->bv_offset + prev->bv_len) {
343 prev->bv_len += len;
344 if (q->merge_bvec_fn &&
345 q->merge_bvec_fn(q, bio, prev) < len) {
346 prev->bv_len -= len;
347 return 0;
348 }
349
350 goto done;
351 }
352 }
353
354 if (bio->bi_vcnt >= bio->bi_max_vecs)
1da177e4
LT
355 return 0;
356
357 /*
358 * we might lose a segment or two here, but rather that than
359 * make this too complex.
360 */
361
362 while (bio->bi_phys_segments >= q->max_phys_segments
363 || bio->bi_hw_segments >= q->max_hw_segments
364 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
365
366 if (retried_segments)
367 return 0;
368
369 retried_segments = 1;
370 blk_recount_segments(q, bio);
371 }
372
373 /*
374 * setup the new entry, we might clear it again later if we
375 * cannot add the page
376 */
377 bvec = &bio->bi_io_vec[bio->bi_vcnt];
378 bvec->bv_page = page;
379 bvec->bv_len = len;
380 bvec->bv_offset = offset;
381
382 /*
383 * if queue has other restrictions (eg varying max sector size
384 * depending on offset), it can specify a merge_bvec_fn in the
385 * queue to get further control
386 */
387 if (q->merge_bvec_fn) {
388 /*
389 * merge_bvec_fn() returns number of bytes it can accept
390 * at this offset
391 */
392 if (q->merge_bvec_fn(q, bio, bvec) < len) {
393 bvec->bv_page = NULL;
394 bvec->bv_len = 0;
395 bvec->bv_offset = 0;
396 return 0;
397 }
398 }
399
400 /* If we may be able to merge these biovecs, force a recount */
401 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
402 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
403 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
404
405 bio->bi_vcnt++;
406 bio->bi_phys_segments++;
407 bio->bi_hw_segments++;
80cfd548 408 done:
1da177e4
LT
409 bio->bi_size += len;
410 return len;
411}
412
6e68af66
MC
413/**
414 * bio_add_pc_page - attempt to add page to bio
fddfdeaf 415 * @q: the target queue
6e68af66
MC
416 * @bio: destination bio
417 * @page: page to add
418 * @len: vec entry length
419 * @offset: vec entry offset
420 *
421 * Attempt to add a page to the bio_vec maplist. This can fail for a
422 * number of reasons, such as the bio being full or target block
423 * device limitations. The target block device must allow bio's
424 * smaller than PAGE_SIZE, so it is always possible to add a single
425 * page to an empty bio. This should only be used by REQ_PC bios.
426 */
427int bio_add_pc_page(request_queue_t *q, struct bio *bio, struct page *page,
428 unsigned int len, unsigned int offset)
429{
defd94b7 430 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
6e68af66
MC
431}
432
1da177e4
LT
433/**
434 * bio_add_page - attempt to add page to bio
435 * @bio: destination bio
436 * @page: page to add
437 * @len: vec entry length
438 * @offset: vec entry offset
439 *
440 * Attempt to add a page to the bio_vec maplist. This can fail for a
441 * number of reasons, such as the bio being full or target block
442 * device limitations. The target block device must allow bio's
443 * smaller than PAGE_SIZE, so it is always possible to add a single
444 * page to an empty bio.
445 */
446int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
447 unsigned int offset)
448{
defd94b7
MC
449 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
450 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
1da177e4
LT
451}
452
453struct bio_map_data {
454 struct bio_vec *iovecs;
455 void __user *userptr;
456};
457
458static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio)
459{
460 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
461 bio->bi_private = bmd;
462}
463
464static void bio_free_map_data(struct bio_map_data *bmd)
465{
466 kfree(bmd->iovecs);
467 kfree(bmd);
468}
469
470static struct bio_map_data *bio_alloc_map_data(int nr_segs)
471{
472 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL);
473
474 if (!bmd)
475 return NULL;
476
477 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL);
478 if (bmd->iovecs)
479 return bmd;
480
481 kfree(bmd);
482 return NULL;
483}
484
485/**
486 * bio_uncopy_user - finish previously mapped bio
487 * @bio: bio being terminated
488 *
489 * Free pages allocated from bio_copy_user() and write back data
490 * to user space in case of a read.
491 */
492int bio_uncopy_user(struct bio *bio)
493{
494 struct bio_map_data *bmd = bio->bi_private;
495 const int read = bio_data_dir(bio) == READ;
496 struct bio_vec *bvec;
497 int i, ret = 0;
498
499 __bio_for_each_segment(bvec, bio, i, 0) {
500 char *addr = page_address(bvec->bv_page);
501 unsigned int len = bmd->iovecs[i].bv_len;
502
503 if (read && !ret && copy_to_user(bmd->userptr, addr, len))
504 ret = -EFAULT;
505
506 __free_page(bvec->bv_page);
507 bmd->userptr += len;
508 }
509 bio_free_map_data(bmd);
510 bio_put(bio);
511 return ret;
512}
513
514/**
515 * bio_copy_user - copy user data to bio
516 * @q: destination block queue
517 * @uaddr: start of user address
518 * @len: length in bytes
519 * @write_to_vm: bool indicating writing to pages or not
520 *
521 * Prepares and returns a bio for indirect user io, bouncing data
522 * to/from kernel pages as necessary. Must be paired with
523 * call bio_uncopy_user() on io completion.
524 */
525struct bio *bio_copy_user(request_queue_t *q, unsigned long uaddr,
526 unsigned int len, int write_to_vm)
527{
528 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
529 unsigned long start = uaddr >> PAGE_SHIFT;
530 struct bio_map_data *bmd;
531 struct bio_vec *bvec;
532 struct page *page;
533 struct bio *bio;
534 int i, ret;
535
536 bmd = bio_alloc_map_data(end - start);
537 if (!bmd)
538 return ERR_PTR(-ENOMEM);
539
540 bmd->userptr = (void __user *) uaddr;
541
542 ret = -ENOMEM;
543 bio = bio_alloc(GFP_KERNEL, end - start);
544 if (!bio)
545 goto out_bmd;
546
547 bio->bi_rw |= (!write_to_vm << BIO_RW);
548
549 ret = 0;
550 while (len) {
551 unsigned int bytes = PAGE_SIZE;
552
553 if (bytes > len)
554 bytes = len;
555
556 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
557 if (!page) {
558 ret = -ENOMEM;
559 break;
560 }
561
defd94b7 562 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) {
1da177e4
LT
563 ret = -EINVAL;
564 break;
565 }
566
567 len -= bytes;
568 }
569
570 if (ret)
571 goto cleanup;
572
573 /*
574 * success
575 */
576 if (!write_to_vm) {
577 char __user *p = (char __user *) uaddr;
578
579 /*
580 * for a write, copy in data to kernel pages
581 */
582 ret = -EFAULT;
583 bio_for_each_segment(bvec, bio, i) {
584 char *addr = page_address(bvec->bv_page);
585
586 if (copy_from_user(addr, p, bvec->bv_len))
587 goto cleanup;
588 p += bvec->bv_len;
589 }
590 }
591
592 bio_set_map_data(bmd, bio);
593 return bio;
594cleanup:
595 bio_for_each_segment(bvec, bio, i)
596 __free_page(bvec->bv_page);
597
598 bio_put(bio);
599out_bmd:
600 bio_free_map_data(bmd);
601 return ERR_PTR(ret);
602}
603
f1970baf
JB
604static struct bio *__bio_map_user_iov(request_queue_t *q,
605 struct block_device *bdev,
606 struct sg_iovec *iov, int iov_count,
607 int write_to_vm)
1da177e4 608{
f1970baf
JB
609 int i, j;
610 int nr_pages = 0;
1da177e4
LT
611 struct page **pages;
612 struct bio *bio;
f1970baf
JB
613 int cur_page = 0;
614 int ret, offset;
1da177e4 615
f1970baf
JB
616 for (i = 0; i < iov_count; i++) {
617 unsigned long uaddr = (unsigned long)iov[i].iov_base;
618 unsigned long len = iov[i].iov_len;
619 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
620 unsigned long start = uaddr >> PAGE_SHIFT;
621
622 nr_pages += end - start;
623 /*
624 * transfer and buffer must be aligned to at least hardsector
625 * size for now, in the future we can relax this restriction
626 */
627 if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q)))
628 return ERR_PTR(-EINVAL);
629 }
630
631 if (!nr_pages)
1da177e4
LT
632 return ERR_PTR(-EINVAL);
633
634 bio = bio_alloc(GFP_KERNEL, nr_pages);
635 if (!bio)
636 return ERR_PTR(-ENOMEM);
637
638 ret = -ENOMEM;
11b0b5ab 639 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
1da177e4
LT
640 if (!pages)
641 goto out;
642
f1970baf
JB
643 for (i = 0; i < iov_count; i++) {
644 unsigned long uaddr = (unsigned long)iov[i].iov_base;
645 unsigned long len = iov[i].iov_len;
646 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
647 unsigned long start = uaddr >> PAGE_SHIFT;
648 const int local_nr_pages = end - start;
649 const int page_limit = cur_page + local_nr_pages;
650
651 down_read(&current->mm->mmap_sem);
652 ret = get_user_pages(current, current->mm, uaddr,
653 local_nr_pages,
654 write_to_vm, 0, &pages[cur_page], NULL);
655 up_read(&current->mm->mmap_sem);
656
99172157
JA
657 if (ret < local_nr_pages) {
658 ret = -EFAULT;
f1970baf 659 goto out_unmap;
99172157 660 }
f1970baf
JB
661
662 offset = uaddr & ~PAGE_MASK;
663 for (j = cur_page; j < page_limit; j++) {
664 unsigned int bytes = PAGE_SIZE - offset;
665
666 if (len <= 0)
667 break;
668
669 if (bytes > len)
670 bytes = len;
671
672 /*
673 * sorry...
674 */
defd94b7
MC
675 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
676 bytes)
f1970baf
JB
677 break;
678
679 len -= bytes;
680 offset = 0;
681 }
1da177e4 682
f1970baf 683 cur_page = j;
1da177e4 684 /*
f1970baf 685 * release the pages we didn't map into the bio, if any
1da177e4 686 */
f1970baf
JB
687 while (j < page_limit)
688 page_cache_release(pages[j++]);
1da177e4
LT
689 }
690
1da177e4
LT
691 kfree(pages);
692
693 /*
694 * set data direction, and check if mapped pages need bouncing
695 */
696 if (!write_to_vm)
697 bio->bi_rw |= (1 << BIO_RW);
698
f1970baf 699 bio->bi_bdev = bdev;
1da177e4
LT
700 bio->bi_flags |= (1 << BIO_USER_MAPPED);
701 return bio;
f1970baf
JB
702
703 out_unmap:
704 for (i = 0; i < nr_pages; i++) {
705 if(!pages[i])
706 break;
707 page_cache_release(pages[i]);
708 }
709 out:
1da177e4
LT
710 kfree(pages);
711 bio_put(bio);
712 return ERR_PTR(ret);
713}
714
715/**
716 * bio_map_user - map user address into bio
67be2dd1 717 * @q: the request_queue_t for the bio
1da177e4
LT
718 * @bdev: destination block device
719 * @uaddr: start of user address
720 * @len: length in bytes
721 * @write_to_vm: bool indicating writing to pages or not
722 *
723 * Map the user space address into a bio suitable for io to a block
724 * device. Returns an error pointer in case of error.
725 */
726struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev,
727 unsigned long uaddr, unsigned int len, int write_to_vm)
f1970baf
JB
728{
729 struct sg_iovec iov;
730
3f70353e 731 iov.iov_base = (void __user *)uaddr;
f1970baf
JB
732 iov.iov_len = len;
733
734 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
735}
736
737/**
738 * bio_map_user_iov - map user sg_iovec table into bio
739 * @q: the request_queue_t for the bio
740 * @bdev: destination block device
741 * @iov: the iovec.
742 * @iov_count: number of elements in the iovec
743 * @write_to_vm: bool indicating writing to pages or not
744 *
745 * Map the user space address into a bio suitable for io to a block
746 * device. Returns an error pointer in case of error.
747 */
748struct bio *bio_map_user_iov(request_queue_t *q, struct block_device *bdev,
749 struct sg_iovec *iov, int iov_count,
750 int write_to_vm)
1da177e4
LT
751{
752 struct bio *bio;
f1970baf 753 int len = 0, i;
1da177e4 754
f1970baf 755 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
1da177e4
LT
756
757 if (IS_ERR(bio))
758 return bio;
759
760 /*
761 * subtle -- if __bio_map_user() ended up bouncing a bio,
762 * it would normally disappear when its bi_end_io is run.
763 * however, we need it for the unmap, so grab an extra
764 * reference to it
765 */
766 bio_get(bio);
767
f1970baf
JB
768 for (i = 0; i < iov_count; i++)
769 len += iov[i].iov_len;
770
1da177e4
LT
771 if (bio->bi_size == len)
772 return bio;
773
774 /*
775 * don't support partial mappings
776 */
777 bio_endio(bio, bio->bi_size, 0);
778 bio_unmap_user(bio);
779 return ERR_PTR(-EINVAL);
780}
781
782static void __bio_unmap_user(struct bio *bio)
783{
784 struct bio_vec *bvec;
785 int i;
786
787 /*
788 * make sure we dirty pages we wrote to
789 */
790 __bio_for_each_segment(bvec, bio, i, 0) {
791 if (bio_data_dir(bio) == READ)
792 set_page_dirty_lock(bvec->bv_page);
793
794 page_cache_release(bvec->bv_page);
795 }
796
797 bio_put(bio);
798}
799
800/**
801 * bio_unmap_user - unmap a bio
802 * @bio: the bio being unmapped
803 *
804 * Unmap a bio previously mapped by bio_map_user(). Must be called with
805 * a process context.
806 *
807 * bio_unmap_user() may sleep.
808 */
809void bio_unmap_user(struct bio *bio)
810{
811 __bio_unmap_user(bio);
812 bio_put(bio);
813}
814
b823825e
JA
815static int bio_map_kern_endio(struct bio *bio, unsigned int bytes_done, int err)
816{
817 if (bio->bi_size)
818 return 1;
819
820 bio_put(bio);
821 return 0;
822}
823
824
df46b9a4 825static struct bio *__bio_map_kern(request_queue_t *q, void *data,
27496a8c 826 unsigned int len, gfp_t gfp_mask)
df46b9a4
MC
827{
828 unsigned long kaddr = (unsigned long)data;
829 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
830 unsigned long start = kaddr >> PAGE_SHIFT;
831 const int nr_pages = end - start;
832 int offset, i;
833 struct bio *bio;
834
835 bio = bio_alloc(gfp_mask, nr_pages);
836 if (!bio)
837 return ERR_PTR(-ENOMEM);
838
839 offset = offset_in_page(kaddr);
840 for (i = 0; i < nr_pages; i++) {
841 unsigned int bytes = PAGE_SIZE - offset;
842
843 if (len <= 0)
844 break;
845
846 if (bytes > len)
847 bytes = len;
848
defd94b7
MC
849 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
850 offset) < bytes)
df46b9a4
MC
851 break;
852
853 data += bytes;
854 len -= bytes;
855 offset = 0;
856 }
857
b823825e 858 bio->bi_end_io = bio_map_kern_endio;
df46b9a4
MC
859 return bio;
860}
861
862/**
863 * bio_map_kern - map kernel address into bio
864 * @q: the request_queue_t for the bio
865 * @data: pointer to buffer to map
866 * @len: length in bytes
867 * @gfp_mask: allocation flags for bio allocation
868 *
869 * Map the kernel address into a bio suitable for io to a block
870 * device. Returns an error pointer in case of error.
871 */
872struct bio *bio_map_kern(request_queue_t *q, void *data, unsigned int len,
27496a8c 873 gfp_t gfp_mask)
df46b9a4
MC
874{
875 struct bio *bio;
876
877 bio = __bio_map_kern(q, data, len, gfp_mask);
878 if (IS_ERR(bio))
879 return bio;
880
881 if (bio->bi_size == len)
882 return bio;
883
884 /*
885 * Don't support partial mappings.
886 */
887 bio_put(bio);
888 return ERR_PTR(-EINVAL);
889}
890
1da177e4
LT
891/*
892 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
893 * for performing direct-IO in BIOs.
894 *
895 * The problem is that we cannot run set_page_dirty() from interrupt context
896 * because the required locks are not interrupt-safe. So what we can do is to
897 * mark the pages dirty _before_ performing IO. And in interrupt context,
898 * check that the pages are still dirty. If so, fine. If not, redirty them
899 * in process context.
900 *
901 * We special-case compound pages here: normally this means reads into hugetlb
902 * pages. The logic in here doesn't really work right for compound pages
903 * because the VM does not uniformly chase down the head page in all cases.
904 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
905 * handle them at all. So we skip compound pages here at an early stage.
906 *
907 * Note that this code is very hard to test under normal circumstances because
908 * direct-io pins the pages with get_user_pages(). This makes
909 * is_page_cache_freeable return false, and the VM will not clean the pages.
910 * But other code (eg, pdflush) could clean the pages if they are mapped
911 * pagecache.
912 *
913 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
914 * deferred bio dirtying paths.
915 */
916
917/*
918 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
919 */
920void bio_set_pages_dirty(struct bio *bio)
921{
922 struct bio_vec *bvec = bio->bi_io_vec;
923 int i;
924
925 for (i = 0; i < bio->bi_vcnt; i++) {
926 struct page *page = bvec[i].bv_page;
927
928 if (page && !PageCompound(page))
929 set_page_dirty_lock(page);
930 }
931}
932
933static void bio_release_pages(struct bio *bio)
934{
935 struct bio_vec *bvec = bio->bi_io_vec;
936 int i;
937
938 for (i = 0; i < bio->bi_vcnt; i++) {
939 struct page *page = bvec[i].bv_page;
940
941 if (page)
942 put_page(page);
943 }
944}
945
946/*
947 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
948 * If they are, then fine. If, however, some pages are clean then they must
949 * have been written out during the direct-IO read. So we take another ref on
950 * the BIO and the offending pages and re-dirty the pages in process context.
951 *
952 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
953 * here on. It will run one page_cache_release() against each page and will
954 * run one bio_put() against the BIO.
955 */
956
957static void bio_dirty_fn(void *data);
958
959static DECLARE_WORK(bio_dirty_work, bio_dirty_fn, NULL);
960static DEFINE_SPINLOCK(bio_dirty_lock);
961static struct bio *bio_dirty_list;
962
963/*
964 * This runs in process context
965 */
966static void bio_dirty_fn(void *data)
967{
968 unsigned long flags;
969 struct bio *bio;
970
971 spin_lock_irqsave(&bio_dirty_lock, flags);
972 bio = bio_dirty_list;
973 bio_dirty_list = NULL;
974 spin_unlock_irqrestore(&bio_dirty_lock, flags);
975
976 while (bio) {
977 struct bio *next = bio->bi_private;
978
979 bio_set_pages_dirty(bio);
980 bio_release_pages(bio);
981 bio_put(bio);
982 bio = next;
983 }
984}
985
986void bio_check_pages_dirty(struct bio *bio)
987{
988 struct bio_vec *bvec = bio->bi_io_vec;
989 int nr_clean_pages = 0;
990 int i;
991
992 for (i = 0; i < bio->bi_vcnt; i++) {
993 struct page *page = bvec[i].bv_page;
994
995 if (PageDirty(page) || PageCompound(page)) {
996 page_cache_release(page);
997 bvec[i].bv_page = NULL;
998 } else {
999 nr_clean_pages++;
1000 }
1001 }
1002
1003 if (nr_clean_pages) {
1004 unsigned long flags;
1005
1006 spin_lock_irqsave(&bio_dirty_lock, flags);
1007 bio->bi_private = bio_dirty_list;
1008 bio_dirty_list = bio;
1009 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1010 schedule_work(&bio_dirty_work);
1011 } else {
1012 bio_put(bio);
1013 }
1014}
1015
1016/**
1017 * bio_endio - end I/O on a bio
1018 * @bio: bio
1019 * @bytes_done: number of bytes completed
1020 * @error: error, if any
1021 *
1022 * Description:
1023 * bio_endio() will end I/O on @bytes_done number of bytes. This may be
1024 * just a partial part of the bio, or it may be the whole bio. bio_endio()
1025 * is the preferred way to end I/O on a bio, it takes care of decrementing
1026 * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and
1027 * and one of the established -Exxxx (-EIO, for instance) error values in
1028 * case something went wrong. Noone should call bi_end_io() directly on
1029 * a bio unless they own it and thus know that it has an end_io function.
1030 **/
1031void bio_endio(struct bio *bio, unsigned int bytes_done, int error)
1032{
1033 if (error)
1034 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1035
1036 if (unlikely(bytes_done > bio->bi_size)) {
1037 printk("%s: want %u bytes done, only %u left\n", __FUNCTION__,
1038 bytes_done, bio->bi_size);
1039 bytes_done = bio->bi_size;
1040 }
1041
1042 bio->bi_size -= bytes_done;
1043 bio->bi_sector += (bytes_done >> 9);
1044
1045 if (bio->bi_end_io)
1046 bio->bi_end_io(bio, bytes_done, error);
1047}
1048
1049void bio_pair_release(struct bio_pair *bp)
1050{
1051 if (atomic_dec_and_test(&bp->cnt)) {
1052 struct bio *master = bp->bio1.bi_private;
1053
1054 bio_endio(master, master->bi_size, bp->error);
1055 mempool_free(bp, bp->bio2.bi_private);
1056 }
1057}
1058
1059static int bio_pair_end_1(struct bio * bi, unsigned int done, int err)
1060{
1061 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1062
1063 if (err)
1064 bp->error = err;
1065
1066 if (bi->bi_size)
1067 return 1;
1068
1069 bio_pair_release(bp);
1070 return 0;
1071}
1072
1073static int bio_pair_end_2(struct bio * bi, unsigned int done, int err)
1074{
1075 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1076
1077 if (err)
1078 bp->error = err;
1079
1080 if (bi->bi_size)
1081 return 1;
1082
1083 bio_pair_release(bp);
1084 return 0;
1085}
1086
1087/*
1088 * split a bio - only worry about a bio with a single page
1089 * in it's iovec
1090 */
1091struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
1092{
1093 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
1094
1095 if (!bp)
1096 return bp;
1097
2056a782
JA
1098 blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi,
1099 bi->bi_sector + first_sectors);
1100
1da177e4
LT
1101 BUG_ON(bi->bi_vcnt != 1);
1102 BUG_ON(bi->bi_idx != 0);
1103 atomic_set(&bp->cnt, 3);
1104 bp->error = 0;
1105 bp->bio1 = *bi;
1106 bp->bio2 = *bi;
1107 bp->bio2.bi_sector += first_sectors;
1108 bp->bio2.bi_size -= first_sectors << 9;
1109 bp->bio1.bi_size = first_sectors << 9;
1110
1111 bp->bv1 = bi->bi_io_vec[0];
1112 bp->bv2 = bi->bi_io_vec[0];
1113 bp->bv2.bv_offset += first_sectors << 9;
1114 bp->bv2.bv_len -= first_sectors << 9;
1115 bp->bv1.bv_len = first_sectors << 9;
1116
1117 bp->bio1.bi_io_vec = &bp->bv1;
1118 bp->bio2.bi_io_vec = &bp->bv2;
1119
a2eb0c10
N
1120 bp->bio1.bi_max_vecs = 1;
1121 bp->bio2.bi_max_vecs = 1;
1122
1da177e4
LT
1123 bp->bio1.bi_end_io = bio_pair_end_1;
1124 bp->bio2.bi_end_io = bio_pair_end_2;
1125
1126 bp->bio1.bi_private = bi;
1127 bp->bio2.bi_private = pool;
1128
1129 return bp;
1130}
1131
1da177e4
LT
1132
1133/*
1134 * create memory pools for biovec's in a bio_set.
1135 * use the global biovec slabs created for general use.
1136 */
1137static int biovec_create_pools(struct bio_set *bs, int pool_entries, int scale)
1138{
1139 int i;
1140
1141 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1142 struct biovec_slab *bp = bvec_slabs + i;
1143 mempool_t **bvp = bs->bvec_pools + i;
1144
50be3455 1145 if (pool_entries > 1 && i >= scale)
1da177e4
LT
1146 pool_entries >>= 1;
1147
93d2341c 1148 *bvp = mempool_create_slab_pool(pool_entries, bp->slab);
1da177e4
LT
1149 if (!*bvp)
1150 return -ENOMEM;
1151 }
1152 return 0;
1153}
1154
1155static void biovec_free_pools(struct bio_set *bs)
1156{
1157 int i;
1158
1159 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1160 mempool_t *bvp = bs->bvec_pools[i];
1161
1162 if (bvp)
1163 mempool_destroy(bvp);
1164 }
1165
1166}
1167
1168void bioset_free(struct bio_set *bs)
1169{
1170 if (bs->bio_pool)
1171 mempool_destroy(bs->bio_pool);
1172
1173 biovec_free_pools(bs);
1174
1175 kfree(bs);
1176}
1177
1178struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size, int scale)
1179{
11b0b5ab 1180 struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1da177e4
LT
1181
1182 if (!bs)
1183 return NULL;
1184
93d2341c 1185 bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab);
1da177e4
LT
1186 if (!bs->bio_pool)
1187 goto bad;
1188
1189 if (!biovec_create_pools(bs, bvec_pool_size, scale))
1190 return bs;
1191
1192bad:
1193 bioset_free(bs);
1194 return NULL;
1195}
1196
1197static void __init biovec_init_slabs(void)
1198{
1199 int i;
1200
1201 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1202 int size;
1203 struct biovec_slab *bvs = bvec_slabs + i;
1204
1205 size = bvs->nr_vecs * sizeof(struct bio_vec);
1206 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1207 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1208 }
1209}
1210
1211static int __init init_bio(void)
1212{
1213 int megabytes, bvec_pool_entries;
1214 int scale = BIOVEC_NR_POOLS;
1215
1216 bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0,
1217 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1218
1219 biovec_init_slabs();
1220
1221 megabytes = nr_free_pages() >> (20 - PAGE_SHIFT);
1222
1223 /*
1224 * find out where to start scaling
1225 */
1226 if (megabytes <= 16)
1227 scale = 0;
1228 else if (megabytes <= 32)
1229 scale = 1;
1230 else if (megabytes <= 64)
1231 scale = 2;
1232 else if (megabytes <= 96)
1233 scale = 3;
1234 else if (megabytes <= 128)
1235 scale = 4;
1236
1237 /*
b0e6e962
BL
1238 * Limit number of entries reserved -- mempools are only used when
1239 * the system is completely unable to allocate memory, so we only
1240 * need enough to make progress.
1da177e4 1241 */
b0e6e962 1242 bvec_pool_entries = 1 + scale;
1da177e4
LT
1243
1244 fs_bio_set = bioset_create(BIO_POOL_SIZE, bvec_pool_entries, scale);
1245 if (!fs_bio_set)
1246 panic("bio: can't allocate bios\n");
1247
0eaae62a
MD
1248 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1249 sizeof(struct bio_pair));
1da177e4
LT
1250 if (!bio_split_pool)
1251 panic("bio: can't create split pool\n");
1252
1253 return 0;
1254}
1255
1256subsys_initcall(init_bio);
1257
1258EXPORT_SYMBOL(bio_alloc);
1259EXPORT_SYMBOL(bio_put);
3676347a 1260EXPORT_SYMBOL(bio_free);
1da177e4
LT
1261EXPORT_SYMBOL(bio_endio);
1262EXPORT_SYMBOL(bio_init);
1263EXPORT_SYMBOL(__bio_clone);
1264EXPORT_SYMBOL(bio_clone);
1265EXPORT_SYMBOL(bio_phys_segments);
1266EXPORT_SYMBOL(bio_hw_segments);
1267EXPORT_SYMBOL(bio_add_page);
6e68af66 1268EXPORT_SYMBOL(bio_add_pc_page);
1da177e4
LT
1269EXPORT_SYMBOL(bio_get_nr_vecs);
1270EXPORT_SYMBOL(bio_map_user);
1271EXPORT_SYMBOL(bio_unmap_user);
df46b9a4 1272EXPORT_SYMBOL(bio_map_kern);
1da177e4
LT
1273EXPORT_SYMBOL(bio_pair_release);
1274EXPORT_SYMBOL(bio_split);
1275EXPORT_SYMBOL(bio_split_pool);
1276EXPORT_SYMBOL(bio_copy_user);
1277EXPORT_SYMBOL(bio_uncopy_user);
1278EXPORT_SYMBOL(bioset_create);
1279EXPORT_SYMBOL(bioset_free);
1280EXPORT_SYMBOL(bio_alloc_bioset);