[dpdk-dev] [RFC PATCH v2 05/14] implement eventfd copying(from fd in qemu process to fd in vhost process) in vhost-net-cdev.c

[Huawei Xie <huawei.xie@intel.com>]

Signed-off-by: Huawei Xie <huawei.xie@intel.com>
---
 lib/librte_vhost/vhost_cuse/vhost-net-cdev.c | 85 ++++++++++++++++++++++++----
 lib/librte_vhost/virtio-net.c                | 57 +------------------
 2 files changed, 77 insertions(+), 65 deletions(-)

diff --git a/lib/librte_vhost/vhost_cuse/vhost-net-cdev.c b/lib/librte_vhost/vhost_cuse/vhost-net-cdev.c
index 2bb07af..802c7dc 100644
--- a/lib/librte_vhost/vhost_cuse/vhost-net-cdev.c
+++ b/lib/librte_vhost/vhost_cuse/vhost-net-cdev.c
@@ -38,6 +38,8 @@
 #include <stdint.h>
 #include <string.h>
 #include <unistd.h>
+#include <sys/eventfd.h>
+#include <sys/ioctl.h>
 
 #include <rte_ethdev.h>
 #include <rte_log.h>
@@ -45,6 +47,7 @@
 #include <rte_virtio_net.h>
 
 #include "vhost-net.h"
+#include "eventfd_link/eventfd_link.h"
 
 #define FUSE_OPT_DUMMY "\0\0"
 #define FUSE_OPT_FORE  "-f\0\0"
@@ -54,6 +57,7 @@ static const uint32_t default_major = 231;
 static const uint32_t default_minor = 1;
 static const char cuse_device_name[] = "/dev/cuse";
 static const char default_cdev[] = "vhost-net";
+static const char eventfd_cdev[] = "/dev/eventfd-link";
 
 static struct fuse_session *session;
 static struct vhost_net_device_ops const *ops;
@@ -173,6 +177,47 @@ vhost_net_release(fuse_req_t req, struct fuse_file_info *fi)
 } while (0)
 
 /*
+ * This function uses the eventfd_link kernel module to copy an eventfd file
+ * descriptor provided by QEMU in to our process space.
+ */
+static int
+eventfd_copy(int target_fd, int target_pid)
+{
+	int eventfd_link, ret;
+	struct eventfd_copy eventfd_copy;
+	int fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
+
+	if (fd == -1)
+		return -1;
+
+	/* Open the character device to the kernel module. */
+	/* TODO: check this earlier rather than fail until VM boots! */
+	eventfd_link = open(eventfd_cdev, O_RDWR);
+	if (eventfd_link < 0) {
+		RTE_LOG(ERR, VHOST_CONFIG,
+			"eventfd_link module is not loaded\n");
+		close(fd);
+		return -1;
+	}
+
+	eventfd_copy.source_fd = fd;
+	eventfd_copy.target_fd = target_fd;
+	eventfd_copy.target_pid = target_pid;
+	/* Call the IOCTL to copy the eventfd. */
+	ret = ioctl(eventfd_link, EVENTFD_COPY, &eventfd_copy);
+	close(eventfd_link);
+
+	if (ret < 0) {
+		RTE_LOG(ERR, VHOST_CONFIG,
+			"EVENTFD_COPY ioctl failed\n");
+		close(fd);
+		return -1;
+	}
+
+	return fd;
+}
+
+/*
  * The IOCTLs are handled using CUSE/FUSE in userspace. Depending on the type
  * of IOCTL a buffer is requested to read or to write. This request is handled
  * by FUSE and the buffer is then given to CUSE.
@@ -284,17 +329,37 @@ vhost_net_ioctl(fuse_req_t req, int cmd, void *arg,
 		break;
 
 	case VHOST_SET_VRING_KICK:
-		LOG_DEBUG(VHOST_CONFIG,
-			"(%"PRIu64") IOCTL: VHOST_SET_VRING_KICK\n", ctx.fh);
-		VHOST_IOCTL_R(struct vhost_vring_file, file,
-			ops->set_vring_kick);
-		break;
-
 	case VHOST_SET_VRING_CALL:
-		LOG_DEBUG(VHOST_CONFIG,
-			"(%"PRIu64") IOCTL: VHOST_SET_VRING_CALL\n", ctx.fh);
-		VHOST_IOCTL_R(struct vhost_vring_file, file,
-			ops->set_vring_call);
+		if (cmd == VHOST_SET_VRING_KICK)
+			LOG_DEBUG(VHOST_CONFIG,
+				"(%"PRIu64") IOCTL: VHOST_SET_VRING_KICK\n",
+			ctx.fh);
+		else
+			LOG_DEBUG(VHOST_CONFIG,
+				"(%"PRIu64") IOCTL: VHOST_SET_VRING_CALL\n",
+			ctx.fh);
+		if (!in_buf)
+			VHOST_IOCTL_RETRY(sizeof(struct vhost_vring_file), 0);
+		else {
+			int fd;
+			file = *(const struct vhost_vring_file *)in_buf;
+			LOG_DEBUG(VHOST_CONFIG,
+				"idx:%d fd:%d\n", file.index, file.fd);
+			fd = eventfd_copy(file.fd, ctx.pid);
+			if (fd < 0) {
+				fuse_reply_ioctl(req, -1, NULL, 0);
+				result = -1;
+				break;
+			}
+			file.fd = fd;
+			if (cmd == VHOST_SET_VRING_KICK) {
+				result = ops->set_vring_kick(ctx, &file);
+				fuse_reply_ioctl(req, result, NULL, 0);
+			} else {
+				result = ops->set_vring_call(ctx, &file);
+				fuse_reply_ioctl(req, result, NULL, 0);
+			}
+		}
 		break;
 
 	default:
diff --git a/lib/librte_vhost/virtio-net.c b/lib/librte_vhost/virtio-net.c
index 6bc9d51..da9e3a6 100644
--- a/lib/librte_vhost/virtio-net.c
+++ b/lib/librte_vhost/virtio-net.c
@@ -38,8 +38,6 @@
 #include <stddef.h>
 #include <stdint.h>
 #include <stdlib.h>
-#include <sys/eventfd.h>
-#include <sys/ioctl.h>
 #include <sys/mman.h>
 #include <unistd.h>
 
@@ -54,7 +52,6 @@
 #include <rte_virtio_net.h>
 
 #include "vhost-net.h"
-#include "eventfd_link/eventfd_link.h"
 
 /*
  * Device linked list structure for configuration.
@@ -64,8 +61,6 @@ struct virtio_net_config_ll {
 	struct virtio_net_config_ll *next;	/* Next dev on linked list.*/
 };
 
-const char eventfd_cdev[] = "/dev/eventfd-link";
-
 /* device ops to add/remove device to/from data core. */
 static struct virtio_net_device_ops const *notify_ops;
 /* root address of the linked list of managed virtio devices */
@@ -904,37 +899,6 @@ get_vring_base(struct vhost_device_ctx ctx, uint32_t index,
 	return 0;
 }
 
-/*
- * This function uses the eventfd_link kernel module to copy an eventfd file
- * descriptor provided by QEMU in to our process space.
- */
-static int
-eventfd_copy(struct virtio_net *dev, struct eventfd_copy *eventfd_copy)
-{
-	int eventfd_link, ret;
-
-	/* Open the character device to the kernel module. */
-	eventfd_link = open(eventfd_cdev, O_RDWR);
-	if (eventfd_link < 0) {
-		RTE_LOG(ERR, VHOST_CONFIG,
-			"(%"PRIu64") eventfd_link module is not loaded\n",
-			dev->device_fh);
-		return -1;
-	}
-
-	/* Call the IOCTL to copy the eventfd. */
-	ret = ioctl(eventfd_link, EVENTFD_COPY, eventfd_copy);
-	close(eventfd_link);
-
-	if (ret < 0) {
-		RTE_LOG(ERR, VHOST_CONFIG,
-			"(%"PRIu64") EVENTFD_COPY ioctl failed\n",
-			dev->device_fh);
-		return -1;
-	}
-
-	return 0;
-}
 
 /*
  * Called from CUSE IOCTL: VHOST_SET_VRING_CALL
@@ -945,7 +909,6 @@ static int
 set_vring_call(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
 {
 	struct virtio_net *dev;
-	struct eventfd_copy	eventfd_kick;
 	struct vhost_virtqueue *vq;
 
 	dev = get_device(ctx);
@@ -958,14 +921,7 @@ set_vring_call(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
 	if (vq->kickfd)
 		close((int)vq->kickfd);
 
-	/* Populate the eventfd_copy structure and call eventfd_copy. */
-	vq->kickfd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
-	eventfd_kick.source_fd = vq->kickfd;
-	eventfd_kick.target_fd = file->fd;
-	eventfd_kick.target_pid = ctx.pid;
-
-	if (eventfd_copy(dev, &eventfd_kick))
-		return -1;
+	vq->kickfd = file->fd;
 
 	return 0;
 }
@@ -979,7 +935,6 @@ static int
 set_vring_kick(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
 {
 	struct virtio_net *dev;
-	struct eventfd_copy eventfd_call;
 	struct vhost_virtqueue *vq;
 
 	dev = get_device(ctx);
@@ -991,15 +946,7 @@ set_vring_kick(struct vhost_device_ctx ctx, struct vhost_vring_file *file)
 
 	if (vq->callfd)
 		close((int)vq->callfd);
-
-	/* Populate the eventfd_copy structure and call eventfd_copy. */
-	vq->callfd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
-	eventfd_call.source_fd = vq->callfd;
-	eventfd_call.target_fd = file->fd;
-	eventfd_call.target_pid = ctx.pid;
-
-	if (eventfd_copy(dev, &eventfd_call))
-		return -1;
+	vq->callfd = file->fd;
 
 	return 0;
 }
-- 
1.8.1.4