]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blobdiff - fs/bio.c
[PATCH] Add scatter-gather support for the block layer SG_IO
[mirror_ubuntu-bionic-kernel.git] / fs / bio.c
index c0d9140e470c3ecd7373522fb9b012eda000f05b..24e4045788e2aed120056c40ac297ed264fd3272 100644 (file)
--- a/fs/bio.c
+++ b/fs/bio.c
@@ -25,6 +25,7 @@
 #include <linux/module.h>
 #include <linux/mempool.h>
 #include <linux/workqueue.h>
+#include <scsi/sg.h>           /* for struct sg_iovec */
 
 #define BIO_POOL_SIZE 256
 
@@ -549,22 +550,34 @@ out_bmd:
        return ERR_PTR(ret);
 }
 
-static struct bio *__bio_map_user(request_queue_t *q, struct block_device *bdev,
-                                 unsigned long uaddr, unsigned int len,
-                                 int write_to_vm)
+static struct bio *__bio_map_user_iov(request_queue_t *q,
+                                     struct block_device *bdev,
+                                     struct sg_iovec *iov, int iov_count,
+                                     int write_to_vm)
 {
-       unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
-       unsigned long start = uaddr >> PAGE_SHIFT;
-       const int nr_pages = end - start;
-       int ret, offset, i;
+       int i, j;
+       int nr_pages = 0;
        struct page **pages;
        struct bio *bio;
+       int cur_page = 0;
+       int ret, offset;
 
-       /*
-        * transfer and buffer must be aligned to at least hardsector
-        * size for now, in the future we can relax this restriction
-        */
-       if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q)))
+       for (i = 0; i < iov_count; i++) {
+               unsigned long uaddr = (unsigned long)iov[i].iov_base;
+               unsigned long len = iov[i].iov_len;
+               unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
+               unsigned long start = uaddr >> PAGE_SHIFT;
+
+               nr_pages += end - start;
+               /*
+                * transfer and buffer must be aligned to at least hardsector
+                * size for now, in the future we can relax this restriction
+                */
+               if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q)))
+                       return ERR_PTR(-EINVAL);
+       }
+
+       if (!nr_pages)
                return ERR_PTR(-EINVAL);
 
        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,
        if (!pages)
                goto out;
 
-       down_read(&current->mm->mmap_sem);
-       ret = get_user_pages(current, current->mm, uaddr, nr_pages,
-                                               write_to_vm, 0, pages, NULL);
-       up_read(&current->mm->mmap_sem);
-
-       if (ret < nr_pages)
-               goto out;
-
-       bio->bi_bdev = bdev;
-
-       offset = uaddr & ~PAGE_MASK;
-       for (i = 0; i < nr_pages; i++) {
-               unsigned int bytes = PAGE_SIZE - offset;
-
-               if (len <= 0)
-                       break;
-
-               if (bytes > len)
-                       bytes = len;
+       memset(pages, 0, nr_pages * sizeof(struct page *));
+
+       for (i = 0; i < iov_count; i++) {
+               unsigned long uaddr = (unsigned long)iov[i].iov_base;
+               unsigned long len = iov[i].iov_len;
+               unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
+               unsigned long start = uaddr >> PAGE_SHIFT;
+               const int local_nr_pages = end - start;
+               const int page_limit = cur_page + local_nr_pages;
+               
+               down_read(&current->mm->mmap_sem);
+               ret = get_user_pages(current, current->mm, uaddr,
+                                    local_nr_pages,
+                                    write_to_vm, 0, &pages[cur_page], NULL);
+               up_read(&current->mm->mmap_sem);
+
+               if (ret < local_nr_pages)
+                       goto out_unmap;
+
+
+               offset = uaddr & ~PAGE_MASK;
+               for (j = cur_page; j < page_limit; j++) {
+                       unsigned int bytes = PAGE_SIZE - offset;
+
+                       if (len <= 0)
+                               break;
+                       
+                       if (bytes > len)
+                               bytes = len;
+
+                       /*
+                        * sorry...
+                        */
+                       if (__bio_add_page(q, bio, pages[j], bytes, offset) < bytes)
+                               break;
+
+                       len -= bytes;
+                       offset = 0;
+               }
 
+               cur_page = j;
                /*
-                * sorry...
+                * release the pages we didn't map into the bio, if any
                 */
-               if (__bio_add_page(q, bio, pages[i], bytes, offset) < bytes)
-                       break;
-
-               len -= bytes;
-               offset = 0;
+               while (j < page_limit)
+                       page_cache_release(pages[j++]);
        }
 
-       /*
-        * release the pages we didn't map into the bio, if any
-        */
-       while (i < nr_pages)
-               page_cache_release(pages[i++]);
-
        kfree(pages);
 
        /*
@@ -620,9 +645,17 @@ static struct bio *__bio_map_user(request_queue_t *q, struct block_device *bdev,
        if (!write_to_vm)
                bio->bi_rw |= (1 << BIO_RW);
 
+       bio->bi_bdev = bdev;
        bio->bi_flags |= (1 << BIO_USER_MAPPED);
        return bio;
-out:
+
+ out_unmap:
+       for (i = 0; i < nr_pages; i++) {
+               if(!pages[i])
+                       break;
+               page_cache_release(pages[i]);
+       }
+ out:
        kfree(pages);
        bio_put(bio);
        return ERR_PTR(ret);
@@ -641,10 +674,34 @@ out:
  */
 struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev,
                         unsigned long uaddr, unsigned int len, int write_to_vm)
+{
+       struct sg_iovec iov;
+
+       iov.iov_base = (__user void *)uaddr;
+       iov.iov_len = len;
+
+       return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
+}
+
+/**
+ *     bio_map_user_iov - map user sg_iovec table into bio
+ *     @q: the request_queue_t for the bio
+ *     @bdev: destination block device
+ *     @iov:   the iovec.
+ *     @iov_count: number of elements in the iovec
+ *     @write_to_vm: bool indicating writing to pages or not
+ *
+ *     Map the user space address into a bio suitable for io to a block
+ *     device. Returns an error pointer in case of error.
+ */
+struct bio *bio_map_user_iov(request_queue_t *q, struct block_device *bdev,
+                            struct sg_iovec *iov, int iov_count,
+                            int write_to_vm)
 {
        struct bio *bio;
+       int len = 0, i;
 
-       bio = __bio_map_user(q, bdev, uaddr, len, write_to_vm);
+       bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
 
        if (IS_ERR(bio))
                return bio;
@@ -657,6 +714,9 @@ struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev,
         */
        bio_get(bio);
 
+       for (i = 0; i < iov_count; i++)
+               len += iov[i].iov_len;
+
        if (bio->bi_size == len)
                return bio;