aboutsummaryrefslogtreecommitdiffstats
path: root/fs
diff options
context:
space:
mode:
authorJames Bottomley <James.Bottomley@SteelEye.com>2005-06-20 08:06:52 -0400
committerJens Axboe <axboe@suse.de>2005-06-20 08:06:52 -0400
commitf1970baf6d74e03bd32072ab453f2fc01bc1b8d3 (patch)
tree559898cdf83bd0f93b8a72248c6423a6548fb604 /fs
parentdd1cab95f356f1395278633565f198463cf6bd24 (diff)
[PATCH] Add scatter-gather support for the block layer SG_IO
Signed-off-by: Jens Axboe <axboe@suse.de>
Diffstat (limited to 'fs')
-rw-r--r--fs/bio.c150
1 files changed, 105 insertions, 45 deletions
diff --git a/fs/bio.c b/fs/bio.c
index c0d9140e470c..24e4045788e2 100644
--- a/fs/bio.c
+++ b/fs/bio.c
@@ -25,6 +25,7 @@
25#include <linux/module.h> 25#include <linux/module.h>
26#include <linux/mempool.h> 26#include <linux/mempool.h>
27#include <linux/workqueue.h> 27#include <linux/workqueue.h>
28#include <scsi/sg.h> /* for struct sg_iovec */
28 29
29#define BIO_POOL_SIZE 256 30#define BIO_POOL_SIZE 256
30 31
@@ -549,22 +550,34 @@ out_bmd:
549 return ERR_PTR(ret); 550 return ERR_PTR(ret);
550} 551}
551 552
552static struct bio *__bio_map_user(request_queue_t *q, struct block_device *bdev, 553static struct bio *__bio_map_user_iov(request_queue_t *q,
553 unsigned long uaddr, unsigned int len, 554 struct block_device *bdev,
554 int write_to_vm) 555 struct sg_iovec *iov, int iov_count,
556 int write_to_vm)
555{ 557{
556 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 558 int i, j;
557 unsigned long start = uaddr >> PAGE_SHIFT; 559 int nr_pages = 0;
558 const int nr_pages = end - start;
559 int ret, offset, i;
560 struct page **pages; 560 struct page **pages;
561 struct bio *bio; 561 struct bio *bio;
562 int cur_page = 0;
563 int ret, offset;
562 564
563 /* 565 for (i = 0; i < iov_count; i++) {
564 * transfer and buffer must be aligned to at least hardsector 566 unsigned long uaddr = (unsigned long)iov[i].iov_base;
565 * size for now, in the future we can relax this restriction 567 unsigned long len = iov[i].iov_len;
566 */ 568 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
567 if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q))) 569 unsigned long start = uaddr >> PAGE_SHIFT;
570
571 nr_pages += end - start;
572 /*
573 * transfer and buffer must be aligned to at least hardsector
574 * size for now, in the future we can relax this restriction
575 */
576 if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q)))
577 return ERR_PTR(-EINVAL);
578 }
579
580 if (!nr_pages)
568 return ERR_PTR(-EINVAL); 581 return ERR_PTR(-EINVAL);
569 582
570 bio = bio_alloc(GFP_KERNEL, nr_pages); 583 bio = bio_alloc(GFP_KERNEL, nr_pages);
@@ -576,42 +589,54 @@ static struct bio *__bio_map_user(request_queue_t *q, struct block_device *bdev,
576 if (!pages) 589 if (!pages)
577 goto out; 590 goto out;
578 591
579 down_read(&current->mm->mmap_sem); 592 memset(pages, 0, nr_pages * sizeof(struct page *));
580 ret = get_user_pages(current, current->mm, uaddr, nr_pages, 593
581 write_to_vm, 0, pages, NULL); 594 for (i = 0; i < iov_count; i++) {
582 up_read(&current->mm->mmap_sem); 595 unsigned long uaddr = (unsigned long)iov[i].iov_base;
583 596 unsigned long len = iov[i].iov_len;
584 if (ret < nr_pages) 597 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
585 goto out; 598 unsigned long start = uaddr >> PAGE_SHIFT;
586 599 const int local_nr_pages = end - start;
587 bio->bi_bdev = bdev; 600 const int page_limit = cur_page + local_nr_pages;
588 601
589 offset = uaddr & ~PAGE_MASK; 602 down_read(&current->mm->mmap_sem);
590 for (i = 0; i < nr_pages; i++) { 603 ret = get_user_pages(current, current->mm, uaddr,
591 unsigned int bytes = PAGE_SIZE - offset; 604 local_nr_pages,
592 605 write_to_vm, 0, &pages[cur_page], NULL);
593 if (len <= 0) 606 up_read(&current->mm->mmap_sem);
594 break; 607
595 608 if (ret < local_nr_pages)
596 if (bytes > len) 609 goto out_unmap;
597 bytes = len; 610
611
612 offset = uaddr & ~PAGE_MASK;
613 for (j = cur_page; j < page_limit; j++) {
614 unsigned int bytes = PAGE_SIZE - offset;
615
616 if (len <= 0)
617 break;
618
619 if (bytes > len)
620 bytes = len;
621
622 /*
623 * sorry...
624 */
625 if (__bio_add_page(q, bio, pages[j], bytes, offset) < bytes)
626 break;
627
628 len -= bytes;
629 offset = 0;
630 }
598 631
632 cur_page = j;
599 /* 633 /*
600 * sorry... 634 * release the pages we didn't map into the bio, if any
601 */ 635 */
602 if (__bio_add_page(q, bio, pages[i], bytes, offset) < bytes) 636 while (j < page_limit)
603 break; 637 page_cache_release(pages[j++]);
604
605 len -= bytes;
606 offset = 0;
607 } 638 }
608 639
609 /*
610 * release the pages we didn't map into the bio, if any
611 */
612 while (i < nr_pages)
613 page_cache_release(pages[i++]);
614
615 kfree(pages); 640 kfree(pages);
616 641
617 /* 642 /*
@@ -620,9 +645,17 @@ static struct bio *__bio_map_user(request_queue_t *q, struct block_device *bdev,
620 if (!write_to_vm) 645 if (!write_to_vm)
621 bio->bi_rw |= (1 << BIO_RW); 646 bio->bi_rw |= (1 << BIO_RW);
622 647
648 bio->bi_bdev = bdev;
623 bio->bi_flags |= (1 << BIO_USER_MAPPED); 649 bio->bi_flags |= (1 << BIO_USER_MAPPED);
624 return bio; 650 return bio;
625out: 651
652 out_unmap:
653 for (i = 0; i < nr_pages; i++) {
654 if(!pages[i])
655 break;
656 page_cache_release(pages[i]);
657 }
658 out:
626 kfree(pages); 659 kfree(pages);
627 bio_put(bio); 660 bio_put(bio);
628 return ERR_PTR(ret); 661 return ERR_PTR(ret);
@@ -642,9 +675,33 @@ out:
642struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev, 675struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev,
643 unsigned long uaddr, unsigned int len, int write_to_vm) 676 unsigned long uaddr, unsigned int len, int write_to_vm)
644{ 677{
678 struct sg_iovec iov;
679
680 iov.iov_base = (__user void *)uaddr;
681 iov.iov_len = len;
682
683 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
684}
685
686/**
687 * bio_map_user_iov - map user sg_iovec table into bio
688 * @q: the request_queue_t for the bio
689 * @bdev: destination block device
690 * @iov: the iovec.
691 * @iov_count: number of elements in the iovec
692 * @write_to_vm: bool indicating writing to pages or not
693 *
694 * Map the user space address into a bio suitable for io to a block
695 * device. Returns an error pointer in case of error.
696 */
697struct bio *bio_map_user_iov(request_queue_t *q, struct block_device *bdev,
698 struct sg_iovec *iov, int iov_count,
699 int write_to_vm)
700{
645 struct bio *bio; 701 struct bio *bio;
702 int len = 0, i;
646 703
647 bio = __bio_map_user(q, bdev, uaddr, len, write_to_vm); 704 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
648 705
649 if (IS_ERR(bio)) 706 if (IS_ERR(bio))
650 return bio; 707 return bio;
@@ -657,6 +714,9 @@ struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev,
657 */ 714 */
658 bio_get(bio); 715 bio_get(bio);
659 716
717 for (i = 0; i < iov_count; i++)
718 len += iov[i].iov_len;
719
660 if (bio->bi_size == len) 720 if (bio->bi_size == len)
661 return bio; 721 return bio;
662 722