kvm tools: Add scatter-gather support for disk images

Add optional support for scatter-gather to disk_image.
Formats that can't take advantage of scatter-gather fallback to simple IO.

Signed-off-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>

disk-image.c

@@ -72,6 +72,20 @@ static int raw_image__write_sector(struct disk_image *self, uint64_t sector, voi
 	return 0;
 }
 
+static ssize_t raw_image__read_sector_sg(struct disk_image *self, uint64_t sector, const struct iovec *iov, int iovcount)
+{
+	uint64_t offset = sector << SECTOR_SHIFT;
+
+	return preadv_in_full(self->fd, iov, iovcount, offset);
+}
+
+static ssize_t raw_image__write_sector_sg(struct disk_image *self, uint64_t sector, const struct iovec *iov, int iovcount)
+{
+	uint64_t offset = sector << SECTOR_SHIFT;
+
+	return pwritev_in_full(self->fd, iov, iovcount, offset);
+}
+
 static int raw_image__read_sector_ro_mmap(struct disk_image *self, uint64_t sector, void *dst, uint32_t dst_len)
 {
 	uint64_t offset = sector << SECTOR_SHIFT;
@@ -105,6 +119,8 @@ static void raw_image__close_ro_mmap(struct disk_image *self)
 static struct disk_image_operations raw_image_ops = {
 	.read_sector		= raw_image__read_sector,
 	.write_sector		= raw_image__write_sector,
+	.read_sector_sg		= raw_image__read_sector_sg,
+	.write_sector_sg	= raw_image__write_sector_sg
 };
 
 static struct disk_image_operations raw_image_ro_mmap_ops = {

include/kvm/disk-image.h

@@ -3,6 +3,7 @@
 
 #include <stdint.h>
 #include <stdbool.h>
+#include <sys/uio.h>
 
 #define SECTOR_SHIFT		9
 #define SECTOR_SIZE		(1UL << SECTOR_SHIFT)
@@ -12,6 +13,8 @@ struct disk_image;
 struct disk_image_operations {
 	int (*read_sector)(struct disk_image *self, uint64_t sector, void *dst, uint32_t dst_len);
 	int (*write_sector)(struct disk_image *self, uint64_t sector, void *src, uint32_t src_len);
+	ssize_t (*read_sector_sg)(struct disk_image *self, uint64_t sector, const struct iovec *iov, int iovcount);
+	ssize_t (*write_sector_sg)(struct disk_image *self, uint64_t sector, const struct iovec *iov, int iovcount);
 	void (*close)(struct disk_image *self);
 };
 
@@ -37,4 +40,32 @@ static inline int disk_image__write_sector(struct disk_image *self, uint64_t sec
 	return self->ops->write_sector(self, sector, src, src_len);
 }
 
+static inline ssize_t disk_image__read_sector_sg(struct disk_image *self, uint64_t sector, const struct iovec *iov, int iovcount)
+{
+	if (self->ops->read_sector_sg)
+		return self->ops->read_sector_sg(self, sector, iov, iovcount);
+
+	while (iovcount--) {
+		self->ops->read_sector(self, sector, iov->iov_base, iov->iov_len);
+		sector += iov->iov_len >> SECTOR_SHIFT;
+		iov++;
+	}
+
+	return sector << SECTOR_SHIFT;
+}
+
+static inline ssize_t disk_image__write_sector_sg(struct disk_image *self, uint64_t sector, const struct iovec *iov, int iovcount)
+{
+	if (self->ops->write_sector_sg)
+		return self->ops->write_sector_sg(self, sector, iov, iovcount);
+
+	while (iovcount--) {
+		self->ops->write_sector(self, sector, iov->iov_base, iov->iov_len);
+		sector += iov->iov_len >> SECTOR_SHIFT;
+		iov++;
+	}
+
+	return sector << SECTOR_SHIFT;
+}
+
 #endif /* KVM__DISK_IMAGE_H */

virtio-blk.c

@@ -115,50 +115,33 @@ static bool virtio_blk_do_io_request(struct kvm *self, struct virt_queue *queue)
 {
 	struct iovec iov[VIRTIO_BLK_QUEUE_SIZE];
 	struct virtio_blk_outhdr *req;
-	uint32_t block_len, block_cnt;
+	ssize_t block_cnt = -1;
 	uint16_t out, in, head;
 	uint8_t *status;
-	bool io_error;
-	void *block;
-	int err, i;
-
-	io_error	= false;
 
 	head		= virt_queue__get_iov(queue, iov, &out, &in, self);
 
 	/* head */
 	req		= iov[0].iov_base;
 
-	/* block */
-	block_cnt	= 0;
+	switch (req->type) {
+	case VIRTIO_BLK_T_IN:
+		block_cnt = disk_image__read_sector_sg(self->disk_image, req->sector, iov + 1, in + out - 2);
 
-	for (i = 1; i < out + in - 1; i++) {
-		block		= iov[i].iov_base;
-		block_len	= iov[i].iov_len;
+		break;
+	case VIRTIO_BLK_T_OUT:
+		block_cnt = disk_image__write_sector_sg(self->disk_image, req->sector, iov + 1, in + out - 2);
 
-		switch (req->type) {
-		case VIRTIO_BLK_T_IN:
-			err	= disk_image__read_sector(self->disk_image, req->sector, block, block_len);
-			if (err)
-				io_error = true;
-			break;
-		case VIRTIO_BLK_T_OUT:
-			err	= disk_image__write_sector(self->disk_image, req->sector, block, block_len);
-			if (err)
-				io_error = true;
-			break;
-		default:
-			warning("request type %d", req->type);
-			io_error	= true;
-		}
+		break;
 
-		req->sector	+= block_len >> SECTOR_SHIFT;
-		block_cnt	+= block_len;
+	default:
+		warning("request type %d", req->type);
+		block_cnt = -1;
 	}
 
 	/* status */
 	status			= iov[out + in - 1].iov_base;
-	*status			= io_error ? VIRTIO_BLK_S_IOERR : VIRTIO_BLK_S_OK;
+	*status			= (block_cnt < 0) ? VIRTIO_BLK_S_IOERR : VIRTIO_BLK_S_OK;
 
 	virt_queue__set_used_elem(queue, head, block_cnt);