[PATCH v1 0/3] virtio-blk: add io_uring passthrough support.

[PATCH v1 0/3] virtio-blk: add io_uring passthrough support.

[Ferry Meng]

This patchset implements io_uring passthrough surppot in virtio-blk
driver, bypass vfs and part of block layer logic, resulting in lower
submit latency and increased flexibility when utilizing virtio-blk.

In this version, currently only supports READ/WRITE vec/no-vec operations,
others like discard or zoned ops not considered in. So the userspace-related
struct is not complicated.

struct virtblk_uring_cmd {
	__u32 type;
	__u32 ioprio;
	__u64 sector;
	/* above is related to out_hdr */
	__u64 data;  // user buffer addr or iovec base addr.
	__u32 data_len; // user buffer length or iovec count.
	__u32 flag;  // only contains whether a vector rw or not.
}; 

To test this patch series, I changed fio's code: 
1. Added virtio-blk support to engines/io_uring.c.
2. Added virtio-blk support to the t/io_uring.c testing tool.
Link: https://github.com/jdmfr/fio


===========
Performance
===========

Using t/io_uring-vblk, the performance of virtio-blk based on uring-cmd
scales better than block device access. (such as below, Virtio-Blk with QEMU,
1-depth fio) 
(passthru) read: IOPS=17.2k, BW=67.4MiB/s (70.6MB/s) 
slat (nsec): min=2907, max=43592, avg=3981.87, stdev=595.10 
clat (usec): min=38, max=285,avg=53.47, stdev= 8.28 
lat (usec): min=44, max=288, avg=57.45, stdev= 8.28
(block) read: IOPS=15.3k, BW=59.8MiB/s (62.7MB/s) 
slat (nsec): min=3408, max=35366, avg=5102.17, stdev=790.79 
clat (usec): min=35, max=343, avg=59.63, stdev=10.26 
lat (usec): min=43, max=349, avg=64.73, stdev=10.21

Testing the virtio-blk device with fio using 'engines=io_uring_cmd'
and 'engines=io_uring' also demonstrates improvements in submit latency.
(passthru) taskset -c 0 t/io_uring-vblk -b4096 -d8 -c4 -s4 -p0 -F1 -B0 -O0 -n1 -u1 /dev/vdcc0 
IOPS=189.80K, BW=741MiB/s, IOS/call=4/3
IOPS=187.68K, BW=733MiB/s, IOS/call=4/3 
(block) taskset -c 0 t/io_uring-vblk -b4096 -d8 -c4 -s4 -p0 -F1 -B0 -O0 -n1 -u0 /dev/vdc 
IOPS=101.51K, BW=396MiB/s, IOS/call=4/3
IOPS=100.01K, BW=390MiB/s, IOS/call=4/4

=======
Changes
=======

Changes in v1:
--------------
* remove virtblk_is_write() helper
* fix rq_flags type definition (blk_opf_t), add REQ_ALLOC_CACHE flag.
https://lore.kernel.org/io-uring/202412042324.uKQ5KdkE-lkp@intel.com/

RFC discussion:
---------------
https://lore.kernel.org/io-uring/20241203121424.19887-1-mengferry@linux.alibaba.com/

Ferry Meng (3):
  virtio-blk: add virtio-blk chardev support.
  virtio-blk: add uring_cmd support for I/O passthru on chardev.
  virtio-blk: add uring_cmd iopoll support.

 drivers/block/virtio_blk.c      | 320 +++++++++++++++++++++++++++++++-
 include/uapi/linux/virtio_blk.h |  16 ++
 2 files changed, 331 insertions(+), 5 deletions(-)

-- 
2.43.5

[PATCH v1 1/3] virtio-blk: add virtio-blk chardev support.

[Ferry Meng]

Introduce character interfaces for block device (per-device), facilitating
access to block devices through io_uring I/O passsthrough.

Besides, vblk initialize only use kmalloc with GFP_KERNEL flag, but for
char device support, we should ensure cdev kobj must be zero before
initialize. So better initial this struct with __GFP_ZERO flag.

Now the character devices only named as

	- /dev/vdXc0

Currently, only one character interface is created for one actual
virtblk device, although it has been partitioned.

Signed-off-by: Ferry Meng <mengferry@linux.alibaba.com>
---
 drivers/block/virtio_blk.c | 84 +++++++++++++++++++++++++++++++++++++-
 1 file changed, 83 insertions(+), 1 deletion(-)

diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
index 194417abc105..3487aaa67514 100644
--- a/drivers/block/virtio_blk.c
+++ b/drivers/block/virtio_blk.c
@@ -17,6 +17,7 @@
 #include <linux/numa.h>
 #include <linux/vmalloc.h>
 #include <uapi/linux/virtio_ring.h>
+#include <linux/cdev.h>
 
 #define PART_BITS 4
 #define VQ_NAME_LEN 16
@@ -25,6 +26,8 @@
 /* The maximum number of sg elements that fit into a virtqueue */
 #define VIRTIO_BLK_MAX_SG_ELEMS 32768
 
+#define VIRTBLK_MINORS		(1U << MINORBITS)
+
 #ifdef CONFIG_ARCH_NO_SG_CHAIN
 #define VIRTIO_BLK_INLINE_SG_CNT	0
 #else
@@ -45,6 +48,10 @@ MODULE_PARM_DESC(poll_queues, "The number of dedicated virtqueues for polling I/
 static int major;
 static DEFINE_IDA(vd_index_ida);
 
+static DEFINE_IDA(vd_chr_minor_ida);
+static dev_t vd_chr_devt;
+static struct class *vd_chr_class;
+
 static struct workqueue_struct *virtblk_wq;
 
 struct virtio_blk_vq {
@@ -84,6 +91,10 @@ struct virtio_blk {
 
 	/* For zoned device */
 	unsigned int zone_sectors;
+
+	/* For passthrough cmd */
+	struct cdev cdev;
+	struct device cdev_device;
 };
 
 struct virtblk_req {
@@ -1239,6 +1250,55 @@ static const struct blk_mq_ops virtio_mq_ops = {
 	.poll		= virtblk_poll,
 };
 
+static void virtblk_cdev_rel(struct device *dev)
+{
+	ida_free(&vd_chr_minor_ida, MINOR(dev->devt));
+}
+
+static void virtblk_cdev_del(struct cdev *cdev, struct device *cdev_device)
+{
+	cdev_device_del(cdev, cdev_device);
+	put_device(cdev_device);
+}
+
+static int virtblk_cdev_add(struct virtio_blk *vblk,
+		const struct file_operations *fops)
+{
+	struct cdev *cdev = &vblk->cdev;
+	struct device *cdev_device = &vblk->cdev_device;
+	int minor, ret;
+
+	minor = ida_alloc(&vd_chr_minor_ida, GFP_KERNEL);
+	if (minor < 0)
+		return minor;
+
+	cdev_device->parent = &vblk->vdev->dev;
+	cdev_device->devt = MKDEV(MAJOR(vd_chr_devt), minor);
+	cdev_device->class = vd_chr_class;
+	cdev_device->release = virtblk_cdev_rel;
+	device_initialize(cdev_device);
+
+	ret = dev_set_name(cdev_device, "%sc0", vblk->disk->disk_name);
+	if (ret)
+		goto err;
+
+	cdev_init(cdev, fops);
+	ret = cdev_device_add(cdev, cdev_device);
+	if (ret) {
+		put_device(cdev_device);
+		goto err;
+	}
+	return ret;
+
+err:
+	ida_free(&vd_chr_minor_ida, minor);
+	return ret;
+}
+
+static const struct file_operations virtblk_chr_fops = {
+	.owner		= THIS_MODULE,
+};
+
 static unsigned int virtblk_queue_depth;
 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
 
@@ -1456,7 +1516,7 @@ static int virtblk_probe(struct virtio_device *vdev)
 		goto out;
 	index = err;
 
-	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
+	vdev->priv = vblk = kzalloc(sizeof(*vblk), GFP_KERNEL);
 	if (!vblk) {
 		err = -ENOMEM;
 		goto out_free_index;
@@ -1544,6 +1604,10 @@ static int virtblk_probe(struct virtio_device *vdev)
 	if (err)
 		goto out_cleanup_disk;
 
+	err = virtblk_cdev_add(vblk, &virtblk_chr_fops);
+	if (err)
+		goto out_cleanup_disk;
+
 	return 0;
 
 out_cleanup_disk:
@@ -1568,6 +1632,8 @@ static void virtblk_remove(struct virtio_device *vdev)
 	/* Make sure no work handler is accessing the device. */
 	flush_work(&vblk->config_work);
 
+	virtblk_cdev_del(&vblk->cdev, &vblk->cdev_device);
+
 	del_gendisk(vblk->disk);
 	blk_mq_free_tag_set(&vblk->tag_set);
 
@@ -1674,13 +1740,27 @@ static int __init virtio_blk_init(void)
 		goto out_destroy_workqueue;
 	}
 
+	error = alloc_chrdev_region(&vd_chr_devt, 0, VIRTBLK_MINORS,
+				"vblk-generic");
+	if (error < 0)
+		goto unregister_chrdev;
+
+	vd_chr_class = class_create("vblk-generic");
+	if (IS_ERR(vd_chr_class)) {
+		error = PTR_ERR(vd_chr_class);
+		goto unregister_chrdev;
+	}
+
 	error = register_virtio_driver(&virtio_blk);
 	if (error)
 		goto out_unregister_blkdev;
+
 	return 0;
 
 out_unregister_blkdev:
 	unregister_blkdev(major, "virtblk");
+unregister_chrdev:
+	unregister_chrdev_region(vd_chr_devt, VIRTBLK_MINORS);
 out_destroy_workqueue:
 	destroy_workqueue(virtblk_wq);
 	return error;
@@ -1690,7 +1770,9 @@ static void __exit virtio_blk_fini(void)
 {
 	unregister_virtio_driver(&virtio_blk);
 	unregister_blkdev(major, "virtblk");
+	unregister_chrdev_region(vd_chr_devt, VIRTBLK_MINORS);
 	destroy_workqueue(virtblk_wq);
+	ida_destroy(&vd_chr_minor_ida);
 }
 module_init(virtio_blk_init);
 module_exit(virtio_blk_fini);
-- 
2.43.5

[PATCH v1 2/3] virtio-blk: add uring_cmd support for I/O passthru on chardev.

[Ferry Meng]

Add ->uring_cmd() support for virtio-blk chardev (/dev/vdXc0).
According to virtio spec, in addition to passing 'hdr' info into kernel,
we also need to pass vaddr & data length of the 'iov' requeired for the
writev/readv op.

Signed-off-by: Ferry Meng <mengferry@linux.alibaba.com>
---
 drivers/block/virtio_blk.c      | 223 +++++++++++++++++++++++++++++++-
 include/uapi/linux/virtio_blk.h |  16 +++
 2 files changed, 235 insertions(+), 4 deletions(-)

diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
index 3487aaa67514..cd88cf939144 100644
--- a/drivers/block/virtio_blk.c
+++ b/drivers/block/virtio_blk.c
@@ -18,6 +18,9 @@
 #include <linux/vmalloc.h>
 #include <uapi/linux/virtio_ring.h>
 #include <linux/cdev.h>
+#include <linux/io_uring/cmd.h>
+#include <linux/types.h>
+#include <linux/uio.h>
 
 #define PART_BITS 4
 #define VQ_NAME_LEN 16
@@ -54,6 +57,20 @@ static struct class *vd_chr_class;
 
 static struct workqueue_struct *virtblk_wq;
 
+struct virtblk_uring_cmd_pdu {
+	struct request *req;
+	struct bio *bio;
+	int status;
+};
+
+struct virtblk_command {
+	struct virtio_blk_outhdr out_hdr;
+
+	__u64	data;
+	__u32	data_len;
+	__u32	flag;
+};
+
 struct virtio_blk_vq {
 	struct virtqueue *vq;
 	spinlock_t lock;
@@ -122,6 +139,11 @@ struct virtblk_req {
 	struct scatterlist sg[];
 };
 
+static void __user *virtblk_to_user_ptr(uintptr_t ptrval)
+{
+	return (void __user *)ptrval;
+}
+
 static inline blk_status_t virtblk_result(u8 status)
 {
 	switch (status) {
@@ -259,9 +281,6 @@ static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
 	if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && op_is_zone_mgmt(req_op(req)))
 		return BLK_STS_NOTSUPP;
 
-	/* Set fields for all request types */
-	vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, req_get_ioprio(req));
-
 	switch (req_op(req)) {
 	case REQ_OP_READ:
 		type = VIRTIO_BLK_T_IN;
@@ -309,9 +328,11 @@ static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
 		type = VIRTIO_BLK_T_ZONE_RESET_ALL;
 		break;
 	case REQ_OP_DRV_IN:
+	case REQ_OP_DRV_OUT:
 		/*
 		 * Out header has already been prepared by the caller (virtblk_get_id()
-		 * or virtblk_submit_zone_report()), nothing to do here.
+		 * virtblk_submit_zone_report() or io_uring passthrough cmd), nothing
+		 * to do here.
 		 */
 		return 0;
 	default:
@@ -323,6 +344,7 @@ static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
 	vbr->in_hdr_len = in_hdr_len;
 	vbr->out_hdr.type = cpu_to_virtio32(vdev, type);
 	vbr->out_hdr.sector = cpu_to_virtio64(vdev, sector);
+	vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, req_get_ioprio(req));
 
 	if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES ||
 	    type == VIRTIO_BLK_T_SECURE_ERASE) {
@@ -832,6 +854,7 @@ static int virtblk_get_id(struct gendisk *disk, char *id_str)
 	vbr = blk_mq_rq_to_pdu(req);
 	vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
 	vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID);
+	vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(req));
 	vbr->out_hdr.sector = 0;
 
 	err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
@@ -1250,6 +1273,197 @@ static const struct blk_mq_ops virtio_mq_ops = {
 	.poll		= virtblk_poll,
 };
 
+static inline struct virtblk_uring_cmd_pdu *virtblk_get_uring_cmd_pdu(
+		struct io_uring_cmd *ioucmd)
+{
+	return (struct virtblk_uring_cmd_pdu *)&ioucmd->pdu;
+}
+
+static void virtblk_uring_task_cb(struct io_uring_cmd *ioucmd,
+		unsigned int issue_flags)
+{
+	struct virtblk_uring_cmd_pdu *pdu = virtblk_get_uring_cmd_pdu(ioucmd);
+	struct virtblk_req *vbr = blk_mq_rq_to_pdu(pdu->req);
+	u64 result = 0;
+
+	if (pdu->bio)
+		blk_rq_unmap_user(pdu->bio);
+
+	/* currently result has no use, it should be zero as cqe->res */
+	io_uring_cmd_done(ioucmd, vbr->in_hdr.status, result, issue_flags);
+}
+
+static enum rq_end_io_ret virtblk_uring_cmd_end_io(struct request *req,
+						   blk_status_t err)
+{
+	struct io_uring_cmd *ioucmd = req->end_io_data;
+	struct virtblk_uring_cmd_pdu *pdu = virtblk_get_uring_cmd_pdu(ioucmd);
+
+	/*
+	 * For iopoll, complete it directly. Note that using the uring_cmd
+	 * helper for this is safe only because we check blk_rq_is_poll().
+	 * As that returns false if we're NOT on a polled queue, then it's
+	 * safe to use the polled completion helper.
+	 *
+	 * Otherwise, move the completion to task work.
+	 */
+	if (blk_rq_is_poll(req)) {
+		if (pdu->bio)
+			blk_rq_unmap_user(pdu->bio);
+		io_uring_cmd_iopoll_done(ioucmd, 0, pdu->status);
+	} else {
+		io_uring_cmd_do_in_task_lazy(ioucmd, virtblk_uring_task_cb);
+	}
+
+	return RQ_END_IO_FREE;
+}
+
+static struct virtblk_req *virtblk_req(struct request *req)
+{
+	return blk_mq_rq_to_pdu(req);
+}
+
+static inline enum req_op virtblk_req_op(const struct virtblk_uring_cmd *cmd)
+{
+	return (cmd->type & VIRTIO_BLK_T_OUT) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
+}
+
+static struct request *virtblk_alloc_user_request(
+		struct request_queue *q, struct virtblk_command *cmd,
+		blk_opf_t rq_flags, blk_mq_req_flags_t blk_flags)
+{
+	struct request *req;
+
+	req = blk_mq_alloc_request(q, rq_flags, blk_flags);
+	if (IS_ERR(req))
+		return req;
+
+	req->rq_flags |= RQF_DONTPREP;
+	memcpy(&virtblk_req(req)->out_hdr, &cmd->out_hdr, sizeof(struct virtio_blk_outhdr));
+	return req;
+}
+
+static int virtblk_map_user_request(struct request *req, u64 ubuffer,
+		unsigned int bufflen, struct io_uring_cmd *ioucmd,
+		bool vec)
+{
+	struct request_queue *q = req->q;
+	struct virtio_blk *vblk = q->queuedata;
+	struct block_device *bdev = vblk ? vblk->disk->part0 : NULL;
+	struct bio *bio = NULL;
+	int ret;
+
+	if (ioucmd && (ioucmd->flags & IORING_URING_CMD_FIXED)) {
+		struct iov_iter iter;
+
+		/* fixedbufs is only for non-vectored io */
+		if (WARN_ON_ONCE(vec))
+			return -EINVAL;
+		ret = io_uring_cmd_import_fixed(ubuffer, bufflen,
+				rq_data_dir(req), &iter, ioucmd);
+		if (ret < 0)
+			goto out;
+		ret = blk_rq_map_user_iov(q, req, NULL,
+			&iter, GFP_KERNEL);
+	} else {
+		ret = blk_rq_map_user_io(req, NULL,
+				virtblk_to_user_ptr(ubuffer),
+				bufflen, GFP_KERNEL, vec, 0,
+				0, rq_data_dir(req));
+	}
+	if (ret)
+		goto out;
+
+	bio = req->bio;
+	if (bdev)
+		bio_set_dev(bio, bdev);
+	return 0;
+
+out:
+	blk_mq_free_request(req);
+	return ret;
+}
+
+static int virtblk_uring_cmd_io(struct virtio_blk *vblk,
+		struct io_uring_cmd *ioucmd, unsigned int issue_flags, bool vec)
+{
+	struct virtblk_uring_cmd_pdu *pdu = virtblk_get_uring_cmd_pdu(ioucmd);
+	const struct virtblk_uring_cmd *cmd = io_uring_sqe_cmd(ioucmd->sqe);
+	struct request_queue *q = vblk->disk->queue;
+	struct virtblk_req *vbr;
+	struct virtblk_command d;
+	struct request *req;
+	blk_opf_t rq_flags = REQ_ALLOC_CACHE | virtblk_req_op(cmd);
+	blk_mq_req_flags_t blk_flags = 0;
+	int ret;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EACCES;
+
+	d.out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, READ_ONCE(cmd->ioprio));
+	d.out_hdr.type = cpu_to_virtio32(vblk->vdev, READ_ONCE(cmd->type));
+	d.out_hdr.sector = cpu_to_virtio64(vblk->vdev, READ_ONCE(cmd->sector));
+	d.data = READ_ONCE(cmd->data);
+	d.data_len = READ_ONCE(cmd->data_len);
+
+	if (issue_flags & IO_URING_F_NONBLOCK) {
+		rq_flags |= REQ_NOWAIT;
+		blk_flags = BLK_MQ_REQ_NOWAIT;
+	}
+	if (issue_flags & IO_URING_F_IOPOLL)
+		rq_flags |= REQ_POLLED;
+
+	req = virtblk_alloc_user_request(q, &d, rq_flags, blk_flags);
+	if (IS_ERR(req))
+		return PTR_ERR(req);
+
+	vbr = virtblk_req(req);
+	vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
+	if (d.data && d.data_len) {
+		ret = virtblk_map_user_request(req, d.data, d.data_len, ioucmd, vec);
+		if (ret)
+			return ret;
+	}
+
+	/* to free bio on completion, as req->bio will be null at that time */
+	pdu->bio = req->bio;
+	pdu->req = req;
+	req->end_io_data = ioucmd;
+	req->end_io = virtblk_uring_cmd_end_io;
+	blk_execute_rq_nowait(req, false);
+	return -EIOCBQUEUED;
+}
+
+
+static int virtblk_uring_cmd(struct virtio_blk *vblk, struct io_uring_cmd *ioucmd,
+			     unsigned int issue_flags)
+{
+	int ret;
+
+	BUILD_BUG_ON(sizeof(struct virtblk_uring_cmd_pdu) > sizeof(ioucmd->pdu));
+
+	switch (ioucmd->cmd_op) {
+	case VIRTBLK_URING_CMD_IO:
+		ret = virtblk_uring_cmd_io(vblk, ioucmd, issue_flags, false);
+		break;
+	case VIRTBLK_URING_CMD_IO_VEC:
+		ret = virtblk_uring_cmd_io(vblk, ioucmd, issue_flags, true);
+		break;
+	default:
+		ret = -ENOTTY;
+	}
+
+	return ret;
+}
+
+static int virtblk_chr_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags)
+{
+	struct virtio_blk *vblk = container_of(file_inode(ioucmd->file)->i_cdev,
+			struct virtio_blk, cdev);
+
+	return virtblk_uring_cmd(vblk, ioucmd, issue_flags);
+}
+
 static void virtblk_cdev_rel(struct device *dev)
 {
 	ida_free(&vd_chr_minor_ida, MINOR(dev->devt));
@@ -1297,6 +1511,7 @@ static int virtblk_cdev_add(struct virtio_blk *vblk,
 
 static const struct file_operations virtblk_chr_fops = {
 	.owner		= THIS_MODULE,
+	.uring_cmd	= virtblk_chr_uring_cmd,
 };
 
 static unsigned int virtblk_queue_depth;
diff --git a/include/uapi/linux/virtio_blk.h b/include/uapi/linux/virtio_blk.h
index 3744e4da1b2a..93b6e1b5b9a4 100644
--- a/include/uapi/linux/virtio_blk.h
+++ b/include/uapi/linux/virtio_blk.h
@@ -313,6 +313,22 @@ struct virtio_scsi_inhdr {
 };
 #endif /* !VIRTIO_BLK_NO_LEGACY */
 
+struct virtblk_uring_cmd {
+	/* VIRTIO_BLK_T* */
+	__u32 type;
+	/* io priority. */
+	__u32 ioprio;
+	/* Sector (ie. 512 byte offset) */
+	__u64 sector;
+
+	__u64 data;
+	__u32 data_len;
+	__u32 flag;
+};
+
+#define VIRTBLK_URING_CMD_IO		1
+#define VIRTBLK_URING_CMD_IO_VEC	2
+
 /* And this is the final byte of the write scatter-gather list. */
 #define VIRTIO_BLK_S_OK		0
 #define VIRTIO_BLK_S_IOERR	1
-- 
2.43.5

[PATCH v1 3/3] virtio-blk: add uring_cmd iopoll support.

[Ferry Meng]

Add polling support for uring_cmd polling support for virtblk, which
will be called during completion-polling.

Signed-off-by: Ferry Meng <mengferry@linux.alibaba.com>
---
 drivers/block/virtio_blk.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
index cd88cf939144..cd4c74e06107 100644
--- a/drivers/block/virtio_blk.c
+++ b/drivers/block/virtio_blk.c
@@ -1464,6 +1464,18 @@ static int virtblk_chr_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue
 	return virtblk_uring_cmd(vblk, ioucmd, issue_flags);
 }
 
+static int virtblk_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd,
+				 struct io_comp_batch *iob,
+				 unsigned int poll_flags)
+{
+	struct virtblk_uring_cmd_pdu *pdu = virtblk_get_uring_cmd_pdu(ioucmd);
+	struct request *req = pdu->req;
+
+	if (req && blk_rq_is_poll(req))
+		return blk_rq_poll(req, iob, poll_flags);
+	return 0;
+}
+
 static void virtblk_cdev_rel(struct device *dev)
 {
 	ida_free(&vd_chr_minor_ida, MINOR(dev->devt));
@@ -1512,6 +1524,7 @@ static int virtblk_cdev_add(struct virtio_blk *vblk,
 static const struct file_operations virtblk_chr_fops = {
 	.owner		= THIS_MODULE,
 	.uring_cmd	= virtblk_chr_uring_cmd,
+	.uring_cmd_iopoll = virtblk_chr_uring_cmd_iopoll,
 };
 
 static unsigned int virtblk_queue_depth;
-- 
2.43.5

Re: [PATCH v1 1/3] virtio-blk: add virtio-blk chardev support.

[Joseph Qi]

On 2024/12/18 17:24, Ferry Meng wrote:
> Introduce character interfaces for block device (per-device), facilitating
> access to block devices through io_uring I/O passsthrough.
> 
> Besides, vblk initialize only use kmalloc with GFP_KERNEL flag, but for
> char device support, we should ensure cdev kobj must be zero before
> initialize. So better initial this struct with __GFP_ZERO flag.
> 
> Now the character devices only named as
> 
> 	- /dev/vdXc0
> 
> Currently, only one character interface is created for one actual
> virtblk device, although it has been partitioned.
> 
> Signed-off-by: Ferry Meng <mengferry@linux.alibaba.com>
> ---
>  drivers/block/virtio_blk.c | 84 +++++++++++++++++++++++++++++++++++++-
>  1 file changed, 83 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> index 194417abc105..3487aaa67514 100644
> --- a/drivers/block/virtio_blk.c
> +++ b/drivers/block/virtio_blk.c
> @@ -17,6 +17,7 @@
>  #include <linux/numa.h>
>  #include <linux/vmalloc.h>
>  #include <uapi/linux/virtio_ring.h>
> +#include <linux/cdev.h>
>  
>  #define PART_BITS 4
>  #define VQ_NAME_LEN 16
> @@ -25,6 +26,8 @@
>  /* The maximum number of sg elements that fit into a virtqueue */
>  #define VIRTIO_BLK_MAX_SG_ELEMS 32768
>  
> +#define VIRTBLK_MINORS		(1U << MINORBITS)
> +
>  #ifdef CONFIG_ARCH_NO_SG_CHAIN
>  #define VIRTIO_BLK_INLINE_SG_CNT	0
>  #else
> @@ -45,6 +48,10 @@ MODULE_PARM_DESC(poll_queues, "The number of dedicated virtqueues for polling I/
>  static int major;
>  static DEFINE_IDA(vd_index_ida);
>  
> +static DEFINE_IDA(vd_chr_minor_ida);
> +static dev_t vd_chr_devt;
> +static struct class *vd_chr_class;
> +
>  static struct workqueue_struct *virtblk_wq;
>  
>  struct virtio_blk_vq {
> @@ -84,6 +91,10 @@ struct virtio_blk {
>  
>  	/* For zoned device */
>  	unsigned int zone_sectors;
> +
> +	/* For passthrough cmd */
> +	struct cdev cdev;
> +	struct device cdev_device;
>  };
>  
>  struct virtblk_req {
> @@ -1239,6 +1250,55 @@ static const struct blk_mq_ops virtio_mq_ops = {
>  	.poll		= virtblk_poll,
>  };
>  
> +static void virtblk_cdev_rel(struct device *dev)
> +{
> +	ida_free(&vd_chr_minor_ida, MINOR(dev->devt));
> +}
> +
> +static void virtblk_cdev_del(struct cdev *cdev, struct device *cdev_device)
> +{
> +	cdev_device_del(cdev, cdev_device);
> +	put_device(cdev_device);
> +}
> +
> +static int virtblk_cdev_add(struct virtio_blk *vblk,
> +		const struct file_operations *fops)
> +{
> +	struct cdev *cdev = &vblk->cdev;
> +	struct device *cdev_device = &vblk->cdev_device;
> +	int minor, ret;
> +
> +	minor = ida_alloc(&vd_chr_minor_ida, GFP_KERNEL);
> +	if (minor < 0)
> +		return minor;
> +
> +	cdev_device->parent = &vblk->vdev->dev;
> +	cdev_device->devt = MKDEV(MAJOR(vd_chr_devt), minor);
> +	cdev_device->class = vd_chr_class;
> +	cdev_device->release = virtblk_cdev_rel;
> +	device_initialize(cdev_device);
> +
> +	ret = dev_set_name(cdev_device, "%sc0", vblk->disk->disk_name);
> +	if (ret)
> +		goto err;
> +
> +	cdev_init(cdev, fops);
> +	ret = cdev_device_add(cdev, cdev_device);
> +	if (ret) {
> +		put_device(cdev_device);
> +		goto err;

put_device() will call cdev_device->release() to free vd_chr_minor_ida.

> +	}
> +	return ret;
> +
> +err:
> +	ida_free(&vd_chr_minor_ida, minor);
> +	return ret;
> +}
> +
> +static const struct file_operations virtblk_chr_fops = {
> +	.owner		= THIS_MODULE,
> +};
> +
>  static unsigned int virtblk_queue_depth;
>  module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
>  
> @@ -1456,7 +1516,7 @@ static int virtblk_probe(struct virtio_device *vdev)
>  		goto out;
>  	index = err;
>  
> -	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
> +	vdev->priv = vblk = kzalloc(sizeof(*vblk), GFP_KERNEL);
>  	if (!vblk) {
>  		err = -ENOMEM;
>  		goto out_free_index;
> @@ -1544,6 +1604,10 @@ static int virtblk_probe(struct virtio_device *vdev)
>  	if (err)
>  		goto out_cleanup_disk;
>  
> +	err = virtblk_cdev_add(vblk, &virtblk_chr_fops);
> +	if (err)
> +		goto out_cleanup_disk;

Missing remove the added disk before.

> +
>  	return 0;
>  
>  out_cleanup_disk:
> @@ -1568,6 +1632,8 @@ static void virtblk_remove(struct virtio_device *vdev)
>  	/* Make sure no work handler is accessing the device. */
>  	flush_work(&vblk->config_work);
>  
> +	virtblk_cdev_del(&vblk->cdev, &vblk->cdev_device);
> +
>  	del_gendisk(vblk->disk);
>  	blk_mq_free_tag_set(&vblk->tag_set);
>  
> @@ -1674,13 +1740,27 @@ static int __init virtio_blk_init(void)
>  		goto out_destroy_workqueue;
>  	}
>  
> +	error = alloc_chrdev_region(&vd_chr_devt, 0, VIRTBLK_MINORS,
> +				"vblk-generic");
> +	if (error < 0)
> +		goto unregister_chrdev;

Should unregister blkdev.

> +
> +	vd_chr_class = class_create("vblk-generic");
> +	if (IS_ERR(vd_chr_class)) {
> +		error = PTR_ERR(vd_chr_class);
> +		goto unregister_chrdev;
> +	}
> +
>  	error = register_virtio_driver(&virtio_blk);
>  	if (error)
>  		goto out_unregister_blkdev;

You've missed destroying vd_chr_class.

> +
>  	return 0;
>  
>  out_unregister_blkdev:
>  	unregister_blkdev(major, "virtblk");
> +unregister_chrdev:
> +	unregister_chrdev_region(vd_chr_devt, VIRTBLK_MINORS);

The out labels should be re-ordered, e.g. move this up.

>  out_destroy_workqueue:
>  	destroy_workqueue(virtblk_wq);
>  	return error;
> @@ -1690,7 +1770,9 @@ static void __exit virtio_blk_fini(void)
>  {
>  	unregister_virtio_driver(&virtio_blk);
>  	unregister_blkdev(major, "virtblk");

Also missed destroying vd_chr_class.

Thanks,
Joseph

> +	unregister_chrdev_region(vd_chr_devt, VIRTBLK_MINORS);
>  	destroy_workqueue(virtblk_wq);
> +	ida_destroy(&vd_chr_minor_ida);
>  }
>  module_init(virtio_blk_init);
>  module_exit(virtio_blk_fini);

Re: [PATCH v1 2/3] virtio-blk: add uring_cmd support for I/O passthru on chardev.

[Joseph Qi]

On 2024/12/18 17:24, Ferry Meng wrote:
> Add ->uring_cmd() support for virtio-blk chardev (/dev/vdXc0).
> According to virtio spec, in addition to passing 'hdr' info into kernel,
> we also need to pass vaddr & data length of the 'iov' requeired for the
> writev/readv op.
> 
> Signed-off-by: Ferry Meng <mengferry@linux.alibaba.com>
> ---
>  drivers/block/virtio_blk.c      | 223 +++++++++++++++++++++++++++++++-
>  include/uapi/linux/virtio_blk.h |  16 +++
>  2 files changed, 235 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/block/virtio_blk.c b/drivers/block/virtio_blk.c
> index 3487aaa67514..cd88cf939144 100644
> --- a/drivers/block/virtio_blk.c
> +++ b/drivers/block/virtio_blk.c
> @@ -18,6 +18,9 @@
>  #include <linux/vmalloc.h>
>  #include <uapi/linux/virtio_ring.h>
>  #include <linux/cdev.h>
> +#include <linux/io_uring/cmd.h>
> +#include <linux/types.h>
> +#include <linux/uio.h>
>  
>  #define PART_BITS 4
>  #define VQ_NAME_LEN 16
> @@ -54,6 +57,20 @@ static struct class *vd_chr_class;
>  
>  static struct workqueue_struct *virtblk_wq;
>  
> +struct virtblk_uring_cmd_pdu {
> +	struct request *req;
> +	struct bio *bio;
> +	int status;
> +};
> +
> +struct virtblk_command {
> +	struct virtio_blk_outhdr out_hdr;
> +
> +	__u64	data;
> +	__u32	data_len;
> +	__u32	flag;
> +};
> +
>  struct virtio_blk_vq {
>  	struct virtqueue *vq;
>  	spinlock_t lock;
> @@ -122,6 +139,11 @@ struct virtblk_req {
>  	struct scatterlist sg[];
>  };
>  
> +static void __user *virtblk_to_user_ptr(uintptr_t ptrval)
> +{

Refer to nvme_to_user_ptr(), the logic for compat syscall is missing.

> +	return (void __user *)ptrval;
> +}
> +
>  static inline blk_status_t virtblk_result(u8 status)
>  {
>  	switch (status) {
> @@ -259,9 +281,6 @@ static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
>  	if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && op_is_zone_mgmt(req_op(req)))
>  		return BLK_STS_NOTSUPP;
>  
> -	/* Set fields for all request types */
> -	vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, req_get_ioprio(req));
> -
>  	switch (req_op(req)) {
>  	case REQ_OP_READ:
>  		type = VIRTIO_BLK_T_IN;
> @@ -309,9 +328,11 @@ static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
>  		type = VIRTIO_BLK_T_ZONE_RESET_ALL;
>  		break;
>  	case REQ_OP_DRV_IN:
> +	case REQ_OP_DRV_OUT:
>  		/*
>  		 * Out header has already been prepared by the caller (virtblk_get_id()
> -		 * or virtblk_submit_zone_report()), nothing to do here.
> +		 * virtblk_submit_zone_report() or io_uring passthrough cmd), nothing
> +		 * to do here.
>  		 */
>  		return 0;
>  	default:
> @@ -323,6 +344,7 @@ static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
>  	vbr->in_hdr_len = in_hdr_len;
>  	vbr->out_hdr.type = cpu_to_virtio32(vdev, type);
>  	vbr->out_hdr.sector = cpu_to_virtio64(vdev, sector);
> +	vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, req_get_ioprio(req));
>  
>  	if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES ||
>  	    type == VIRTIO_BLK_T_SECURE_ERASE) {
> @@ -832,6 +854,7 @@ static int virtblk_get_id(struct gendisk *disk, char *id_str)
>  	vbr = blk_mq_rq_to_pdu(req);
>  	vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
>  	vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID);
> +	vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(req));
>  	vbr->out_hdr.sector = 0;
>  
>  	err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
> @@ -1250,6 +1273,197 @@ static const struct blk_mq_ops virtio_mq_ops = {
>  	.poll		= virtblk_poll,
>  };
>  
> +static inline struct virtblk_uring_cmd_pdu *virtblk_get_uring_cmd_pdu(
> +		struct io_uring_cmd *ioucmd)
> +{
> +	return (struct virtblk_uring_cmd_pdu *)&ioucmd->pdu;
> +}
> +
> +static void virtblk_uring_task_cb(struct io_uring_cmd *ioucmd,
> +		unsigned int issue_flags)
> +{
> +	struct virtblk_uring_cmd_pdu *pdu = virtblk_get_uring_cmd_pdu(ioucmd);
> +	struct virtblk_req *vbr = blk_mq_rq_to_pdu(pdu->req);
> +	u64 result = 0;
> +
> +	if (pdu->bio)
> +		blk_rq_unmap_user(pdu->bio);
> +
> +	/* currently result has no use, it should be zero as cqe->res */
> +	io_uring_cmd_done(ioucmd, vbr->in_hdr.status, result, issue_flags);
> +}
> +
> +static enum rq_end_io_ret virtblk_uring_cmd_end_io(struct request *req,
> +						   blk_status_t err)
> +{
> +	struct io_uring_cmd *ioucmd = req->end_io_data;
> +	struct virtblk_uring_cmd_pdu *pdu = virtblk_get_uring_cmd_pdu(ioucmd);
> +
> +	/*
> +	 * For iopoll, complete it directly. Note that using the uring_cmd
> +	 * helper for this is safe only because we check blk_rq_is_poll().
> +	 * As that returns false if we're NOT on a polled queue, then it's
> +	 * safe to use the polled completion helper.
> +	 *
> +	 * Otherwise, move the completion to task work.
> +	 */
> +	if (blk_rq_is_poll(req)) {
> +		if (pdu->bio)
> +			blk_rq_unmap_user(pdu->bio);
> +		io_uring_cmd_iopoll_done(ioucmd, 0, pdu->status);
> +	} else {
> +		io_uring_cmd_do_in_task_lazy(ioucmd, virtblk_uring_task_cb);
> +	}
> +
> +	return RQ_END_IO_FREE;
> +}
> +
> +static struct virtblk_req *virtblk_req(struct request *req)
> +{
> +	return blk_mq_rq_to_pdu(req);
> +}

Don't think this helper is necessary.
You've already open-coded in other places.

> +
> +static inline enum req_op virtblk_req_op(const struct virtblk_uring_cmd *cmd)
> +{
> +	return (cmd->type & VIRTIO_BLK_T_OUT) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
> +}
> +
> +static struct request *virtblk_alloc_user_request(
> +		struct request_queue *q, struct virtblk_command *cmd,
> +		blk_opf_t rq_flags, blk_mq_req_flags_t blk_flags)
> +{
> +	struct request *req;
> +
> +	req = blk_mq_alloc_request(q, rq_flags, blk_flags);
> +	if (IS_ERR(req))
> +		return req;
> +
> +	req->rq_flags |= RQF_DONTPREP;

Do we have to do some other initialization? e.g. REQ_POLLED.

> +	memcpy(&virtblk_req(req)->out_hdr, &cmd->out_hdr, sizeof(struct virtio_blk_outhdr));
> +	return req;
> +}
> +
> +static int virtblk_map_user_request(struct request *req, u64 ubuffer,
> +		unsigned int bufflen, struct io_uring_cmd *ioucmd,
> +		bool vec)
> +{
> +	struct request_queue *q = req->q;
> +	struct virtio_blk *vblk = q->queuedata;
> +	struct block_device *bdev = vblk ? vblk->disk->part0 : NULL;
> +	struct bio *bio = NULL;
> +	int ret;
> +
> +	if (ioucmd && (ioucmd->flags & IORING_URING_CMD_FIXED)) {
> +		struct iov_iter iter;
> +
> +		/* fixedbufs is only for non-vectored io */
> +		if (WARN_ON_ONCE(vec))
> +			return -EINVAL;
> +		ret = io_uring_cmd_import_fixed(ubuffer, bufflen,
> +				rq_data_dir(req), &iter, ioucmd);
> +		if (ret < 0)
> +			goto out;
> +		ret = blk_rq_map_user_iov(q, req, NULL,
> +			&iter, GFP_KERNEL);
> +	} else {
> +		ret = blk_rq_map_user_io(req, NULL,
> +				virtblk_to_user_ptr(ubuffer),
> +				bufflen, GFP_KERNEL, vec, 0,
> +				0, rq_data_dir(req));
> +	}
> +	if (ret)
> +		goto out;
> +
> +	bio = req->bio;
> +	if (bdev)
> +		bio_set_dev(bio, bdev);
> +	return 0;
> +
> +out:
> +	blk_mq_free_request(req);
> +	return ret;
> +}
> +
> +static int virtblk_uring_cmd_io(struct virtio_blk *vblk,
> +		struct io_uring_cmd *ioucmd, unsigned int issue_flags, bool vec)
> +{
> +	struct virtblk_uring_cmd_pdu *pdu = virtblk_get_uring_cmd_pdu(ioucmd);
> +	const struct virtblk_uring_cmd *cmd = io_uring_sqe_cmd(ioucmd->sqe);
> +	struct request_queue *q = vblk->disk->queue;
> +	struct virtblk_req *vbr;
> +	struct virtblk_command d;

Or use 'c' for command?

> +	struct request *req;
> +	blk_opf_t rq_flags = REQ_ALLOC_CACHE | virtblk_req_op(cmd);
> +	blk_mq_req_flags_t blk_flags = 0;
> +	int ret;
> +
> +	if (!capable(CAP_SYS_ADMIN))
> +		return -EACCES;
> +
> +	d.out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, READ_ONCE(cmd->ioprio));
> +	d.out_hdr.type = cpu_to_virtio32(vblk->vdev, READ_ONCE(cmd->type));
> +	d.out_hdr.sector = cpu_to_virtio64(vblk->vdev, READ_ONCE(cmd->sector));
> +	d.data = READ_ONCE(cmd->data);
> +	d.data_len = READ_ONCE(cmd->data_len);
> +
> +	if (issue_flags & IO_URING_F_NONBLOCK) {
> +		rq_flags |= REQ_NOWAIT;
> +		blk_flags = BLK_MQ_REQ_NOWAIT;
> +	}
> +	if (issue_flags & IO_URING_F_IOPOLL)
> +		rq_flags |= REQ_POLLED;
> +
> +	req = virtblk_alloc_user_request(q, &d, rq_flags, blk_flags);
> +	if (IS_ERR(req))
> +		return PTR_ERR(req);
> +
> +	vbr = virtblk_req(req);
> +	vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
> +	if (d.data && d.data_len) {
> +		ret = virtblk_map_user_request(req, d.data, d.data_len, ioucmd, vec);
> +		if (ret)
> +			return ret;
> +	}
> +
> +	/* to free bio on completion, as req->bio will be null at that time */
> +	pdu->bio = req->bio;
> +	pdu->req = req;
> +	req->end_io_data = ioucmd;
> +	req->end_io = virtblk_uring_cmd_end_io;
> +	blk_execute_rq_nowait(req, false);
> +	return -EIOCBQUEUED;
> +}
> +
> +
> +static int virtblk_uring_cmd(struct virtio_blk *vblk, struct io_uring_cmd *ioucmd,
> +			     unsigned int issue_flags)
> +{
> +	int ret;
> +
> +	BUILD_BUG_ON(sizeof(struct virtblk_uring_cmd_pdu) > sizeof(ioucmd->pdu));

io_uring passthrough requires big sqe/cqe support.
So it is deserved a check here.

Thanks,
Joseph

> +
> +	switch (ioucmd->cmd_op) {
> +	case VIRTBLK_URING_CMD_IO:
> +		ret = virtblk_uring_cmd_io(vblk, ioucmd, issue_flags, false);
> +		break;
> +	case VIRTBLK_URING_CMD_IO_VEC:
> +		ret = virtblk_uring_cmd_io(vblk, ioucmd, issue_flags, true);
> +		break;
> +	default:
> +		ret = -ENOTTY;
> +	}
> +
> +	return ret;
> +}
> +
> +static int virtblk_chr_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags)
> +{
> +	struct virtio_blk *vblk = container_of(file_inode(ioucmd->file)->i_cdev,
> +			struct virtio_blk, cdev);
> +
> +	return virtblk_uring_cmd(vblk, ioucmd, issue_flags);
> +}
> +
>  static void virtblk_cdev_rel(struct device *dev)
>  {
>  	ida_free(&vd_chr_minor_ida, MINOR(dev->devt));
> @@ -1297,6 +1511,7 @@ static int virtblk_cdev_add(struct virtio_blk *vblk,
>  
>  static const struct file_operations virtblk_chr_fops = {
>  	.owner		= THIS_MODULE,
> +	.uring_cmd	= virtblk_chr_uring_cmd,
>  };
>  
>  static unsigned int virtblk_queue_depth;
> diff --git a/include/uapi/linux/virtio_blk.h b/include/uapi/linux/virtio_blk.h
> index 3744e4da1b2a..93b6e1b5b9a4 100644
> --- a/include/uapi/linux/virtio_blk.h
> +++ b/include/uapi/linux/virtio_blk.h
> @@ -313,6 +313,22 @@ struct virtio_scsi_inhdr {
>  };
>  #endif /* !VIRTIO_BLK_NO_LEGACY */
>  
> +struct virtblk_uring_cmd {
> +	/* VIRTIO_BLK_T* */
> +	__u32 type;
> +	/* io priority. */
> +	__u32 ioprio;
> +	/* Sector (ie. 512 byte offset) */
> +	__u64 sector;
> +
> +	__u64 data;
> +	__u32 data_len;
> +	__u32 flag;
> +};
> +
> +#define VIRTBLK_URING_CMD_IO		1
> +#define VIRTBLK_URING_CMD_IO_VEC	2
> +
>  /* And this is the final byte of the write scatter-gather list. */
>  #define VIRTIO_BLK_S_OK		0
>  #define VIRTIO_BLK_S_IOERR	1