net/mlx5: handle Rx CQE compression
[dpdk.git] / lib / librte_vhost / vhost_user / vhost-net-user.c
index f485a3b..94f1b92 100644 (file)
@@ -33,6 +33,7 @@
 
 #include <stdint.h>
 #include <stdio.h>
+#include <stdbool.h>
 #include <limits.h>
 #include <stdlib.h>
 #include <unistd.h>
@@ -40,6 +41,7 @@
 #include <sys/types.h>
 #include <sys/socket.h>
 #include <sys/un.h>
+#include <sys/queue.h>
 #include <errno.h>
 #include <pthread.h>
 
@@ -58,6 +60,8 @@
 struct vhost_user_socket {
        char *path;
        int listenfd;
+       bool is_server;
+       bool reconnect;
 };
 
 struct vhost_user_connection {
@@ -75,8 +79,9 @@ struct vhost_user {
 
 #define MAX_VIRTIO_BACKLOG 128
 
-static void vhost_user_new_connection(int fd, void *data, int *remove);
+static void vhost_user_server_new_connection(int fd, void *data, int *remove);
 static void vhost_user_msg_handler(int fd, void *dat, int *remove);
+static int vhost_user_create_client(struct vhost_user_socket *vsocket);
 
 static struct vhost_user vhost_user = {
        .fdset = {
@@ -111,48 +116,6 @@ static const char *vhost_message_str[VHOST_USER_MAX] = {
        [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
 };
 
-/**
- * Create a unix domain socket, bind to path and listen for connection.
- * @return
- *  socket fd or -1 on failure
- */
-static int
-uds_socket(const char *path)
-{
-       struct sockaddr_un un;
-       int sockfd;
-       int ret;
-
-       if (path == NULL)
-               return -1;
-
-       sockfd = socket(AF_UNIX, SOCK_STREAM, 0);
-       if (sockfd < 0)
-               return -1;
-       RTE_LOG(INFO, VHOST_CONFIG, "socket created, fd:%d\n", sockfd);
-
-       memset(&un, 0, sizeof(un));
-       un.sun_family = AF_UNIX;
-       snprintf(un.sun_path, sizeof(un.sun_path), "%s", path);
-       ret = bind(sockfd, (struct sockaddr *)&un, sizeof(un));
-       if (ret == -1) {
-               RTE_LOG(ERR, VHOST_CONFIG, "fail to bind fd:%d, remove file:%s and try again.\n",
-                       sockfd, path);
-               goto err;
-       }
-       RTE_LOG(INFO, VHOST_CONFIG, "bind to %s\n", path);
-
-       ret = listen(sockfd, MAX_VIRTIO_BACKLOG);
-       if (ret == -1)
-               goto err;
-
-       return sockfd;
-
-err:
-       close(sockfd);
-       return -1;
-}
-
 /* return bytes# of read on success or negative val on failure. */
 static int
 read_fd_message(int sockfd, char *buf, int buflen, int *fds, int fd_num)
@@ -287,32 +250,24 @@ send_vhost_message(int sockfd, struct VhostUserMsg *msg)
        return ret;
 }
 
-/* call back when there is new vhost-user connection.  */
+
 static void
-vhost_user_new_connection(int fd, void *dat, int *remove __rte_unused)
+vhost_user_add_connection(int fd, struct vhost_user_socket *vsocket)
 {
-       struct vhost_user_socket *vsocket = dat;
-       int conn_fd;
-       struct vhost_user_connection *conn;
        int vid;
-       unsigned int size;
-
-       conn_fd = accept(fd, NULL, NULL);
-       RTE_LOG(INFO, VHOST_CONFIG,
-               "new virtio connection is %d\n", conn_fd);
-       if (conn_fd < 0)
-               return;
+       size_t size;
+       struct vhost_user_connection *conn;
 
-       conn = calloc(1, sizeof(*conn));
+       conn = malloc(sizeof(*conn));
        if (conn == NULL) {
-               close(conn_fd);
+               close(fd);
                return;
        }
 
        vid = vhost_new_device();
        if (vid == -1) {
+               close(fd);
                free(conn);
-               close(conn_fd);
                return;
        }
 
@@ -323,8 +278,21 @@ vhost_user_new_connection(int fd, void *dat, int *remove __rte_unused)
 
        conn->vsocket = vsocket;
        conn->vid = vid;
-       fdset_add(&vhost_user.fdset,
-               conn_fd, vhost_user_msg_handler, NULL, conn);
+       fdset_add(&vhost_user.fdset, fd, vhost_user_msg_handler, NULL, conn);
+}
+
+/* call back when there is new vhost-user connection from client  */
+static void
+vhost_user_server_new_connection(int fd, void *dat, int *remove __rte_unused)
+{
+       struct vhost_user_socket *vsocket = dat;
+
+       fd = accept(fd, NULL, NULL);
+       if (fd < 0)
+               return;
+
+       RTE_LOG(INFO, VHOST_CONFIG, "new vhost user connection is %d\n", fd);
+       vhost_user_add_connection(fd, vsocket);
 }
 
 /* callback when there is message on the connfd */
@@ -340,6 +308,8 @@ vhost_user_msg_handler(int connfd, void *dat, int *remove)
        vid = conn->vid;
        ret = read_vhost_message(connfd, &msg);
        if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
+               struct vhost_user_socket *vsocket = conn->vsocket;
+
                if (ret < 0)
                        RTE_LOG(ERR, VHOST_CONFIG,
                                "vhost read message failed\n");
@@ -355,6 +325,9 @@ vhost_user_msg_handler(int connfd, void *dat, int *remove)
                free(conn);
                vhost_destroy_device(vid);
 
+               if (vsocket->reconnect)
+                       vhost_user_create_client(vsocket);
+
                return;
        }
 
@@ -452,50 +425,222 @@ vhost_user_msg_handler(int connfd, void *dat, int *remove)
        }
 }
 
-/**
- * Creates and initialise the vhost server.
- */
-int
-rte_vhost_driver_register(const char *path)
+static int
+create_unix_socket(const char *path, struct sockaddr_un *un, bool is_server)
 {
-       struct vhost_user_socket *vsocket;
+       int fd;
 
-       pthread_mutex_lock(&vhost_user.mutex);
+       fd = socket(AF_UNIX, SOCK_STREAM, 0);
+       if (fd < 0)
+               return -1;
+       RTE_LOG(INFO, VHOST_CONFIG, "vhost-user %s: socket created, fd: %d\n",
+               is_server ? "server" : "client", fd);
 
-       if (vhost_user.vsocket_cnt == MAX_VHOST_SOCKET) {
+       memset(un, 0, sizeof(*un));
+       un->sun_family = AF_UNIX;
+       strncpy(un->sun_path, path, sizeof(un->sun_path));
+
+       return fd;
+}
+
+static int
+vhost_user_create_server(struct vhost_user_socket *vsocket)
+{
+       int fd;
+       int ret;
+       struct sockaddr_un un;
+       const char *path = vsocket->path;
+
+       fd = create_unix_socket(path, &un, vsocket->is_server);
+       if (fd < 0)
+               return -1;
+
+       ret = bind(fd, (struct sockaddr *)&un, sizeof(un));
+       if (ret < 0) {
                RTE_LOG(ERR, VHOST_CONFIG,
-                       "error: the number of servers reaches maximum\n");
-               pthread_mutex_unlock(&vhost_user.mutex);
+                       "failed to bind to %s: %s; remove it and try again\n",
+                       path, strerror(errno));
+               goto err;
+       }
+       RTE_LOG(INFO, VHOST_CONFIG, "bind to %s\n", path);
+
+       ret = listen(fd, MAX_VIRTIO_BACKLOG);
+       if (ret < 0)
+               goto err;
+
+       vsocket->listenfd = fd;
+       fdset_add(&vhost_user.fdset, fd, vhost_user_server_new_connection,
+                 NULL, vsocket);
+
+       return 0;
+
+err:
+       close(fd);
+       return -1;
+}
+
+struct vhost_user_reconnect {
+       struct sockaddr_un un;
+       int fd;
+       struct vhost_user_socket *vsocket;
+
+       TAILQ_ENTRY(vhost_user_reconnect) next;
+};
+
+TAILQ_HEAD(vhost_user_reconnect_tailq_list, vhost_user_reconnect);
+struct vhost_user_reconnect_list {
+       struct vhost_user_reconnect_tailq_list head;
+       pthread_mutex_t mutex;
+};
+
+static struct vhost_user_reconnect_list reconn_list;
+static pthread_t reconn_tid;
+
+static void *
+vhost_user_client_reconnect(void *arg __rte_unused)
+{
+       struct vhost_user_reconnect *reconn, *next;
+
+       while (1) {
+               pthread_mutex_lock(&reconn_list.mutex);
+
+               /*
+                * An equal implementation of TAILQ_FOREACH_SAFE,
+                * which does not exist on all platforms.
+                */
+               for (reconn = TAILQ_FIRST(&reconn_list.head);
+                    reconn != NULL; reconn = next) {
+                       next = TAILQ_NEXT(reconn, next);
+
+                       if (connect(reconn->fd, (struct sockaddr *)&reconn->un,
+                                   sizeof(reconn->un)) < 0)
+                               continue;
+
+                       RTE_LOG(INFO, VHOST_CONFIG,
+                               "%s: connected\n", reconn->vsocket->path);
+                       vhost_user_add_connection(reconn->fd, reconn->vsocket);
+                       TAILQ_REMOVE(&reconn_list.head, reconn, next);
+                       free(reconn);
+               }
+
+               pthread_mutex_unlock(&reconn_list.mutex);
+               sleep(1);
+       }
+
+       return NULL;
+}
+
+static int
+vhost_user_reconnect_init(void)
+{
+       int ret;
+
+       pthread_mutex_init(&reconn_list.mutex, NULL);
+       TAILQ_INIT(&reconn_list.head);
+
+       ret = pthread_create(&reconn_tid, NULL,
+                            vhost_user_client_reconnect, NULL);
+       if (ret < 0)
+               RTE_LOG(ERR, VHOST_CONFIG, "failed to create reconnect thread");
+
+       return ret;
+}
+
+static int
+vhost_user_create_client(struct vhost_user_socket *vsocket)
+{
+       int fd;
+       int ret;
+       struct sockaddr_un un;
+       const char *path = vsocket->path;
+       struct vhost_user_reconnect *reconn;
+
+       fd = create_unix_socket(path, &un, vsocket->is_server);
+       if (fd < 0)
                return -1;
+
+       ret = connect(fd, (struct sockaddr *)&un, sizeof(un));
+       if (ret == 0) {
+               vhost_user_add_connection(fd, vsocket);
+               return 0;
        }
 
-       vsocket = calloc(sizeof(struct vhost_user_socket), 1);
-       if (vsocket == NULL) {
-               pthread_mutex_unlock(&vhost_user.mutex);
+       RTE_LOG(ERR, VHOST_CONFIG,
+               "failed to connect to %s: %s\n",
+               path, strerror(errno));
+
+       if (!vsocket->reconnect) {
+               close(fd);
                return -1;
        }
 
-       vsocket->listenfd = uds_socket(path);
-       if (vsocket->listenfd < 0) {
-               free(vsocket);
-               pthread_mutex_unlock(&vhost_user.mutex);
+       RTE_LOG(ERR, VHOST_CONFIG, "%s: reconnecting...\n", path);
+       reconn = malloc(sizeof(*reconn));
+       reconn->un = un;
+       reconn->fd = fd;
+       reconn->vsocket = vsocket;
+       pthread_mutex_lock(&reconn_list.mutex);
+       TAILQ_INSERT_TAIL(&reconn_list.head, reconn, next);
+       pthread_mutex_unlock(&reconn_list.mutex);
+
+       return 0;
+}
+
+/*
+ * Register a new vhost-user socket; here we could act as server
+ * (the default case), or client (when RTE_VHOST_USER_CLIENT) flag
+ * is set.
+ */
+int
+rte_vhost_driver_register(const char *path, uint64_t flags)
+{
+       int ret = -1;
+       struct vhost_user_socket *vsocket;
+
+       if (!path)
                return -1;
+
+       pthread_mutex_lock(&vhost_user.mutex);
+
+       if (vhost_user.vsocket_cnt == MAX_VHOST_SOCKET) {
+               RTE_LOG(ERR, VHOST_CONFIG,
+                       "error: the number of vhost sockets reaches maximum\n");
+               goto out;
        }
 
+       vsocket = malloc(sizeof(struct vhost_user_socket));
+       if (!vsocket)
+               goto out;
+       memset(vsocket, 0, sizeof(struct vhost_user_socket));
        vsocket->path = strdup(path);
 
-       fdset_add(&vhost_user.fdset, vsocket->listenfd,
-               vhost_user_new_connection, NULL, vsocket);
+       if ((flags & RTE_VHOST_USER_CLIENT) != 0) {
+               vsocket->reconnect = !(flags & RTE_VHOST_USER_NO_RECONNECT);
+               if (vsocket->reconnect && reconn_tid == 0) {
+                       if (vhost_user_reconnect_init() < 0)
+                               goto out;
+               }
+               ret = vhost_user_create_client(vsocket);
+       } else {
+               vsocket->is_server = true;
+               ret = vhost_user_create_server(vsocket);
+       }
+       if (ret < 0) {
+               free(vsocket->path);
+               free(vsocket);
+               goto out;
+       }
 
        vhost_user.vsockets[vhost_user.vsocket_cnt++] = vsocket;
+
+out:
        pthread_mutex_unlock(&vhost_user.mutex);
 
-       return 0;
+       return ret;
 }
 
-
 /**
- * Unregister the specified vhost server
+ * Unregister the specified vhost socket
  */
 int
 rte_vhost_driver_unregister(const char *path)
@@ -507,15 +652,16 @@ rte_vhost_driver_unregister(const char *path)
 
        for (i = 0; i < vhost_user.vsocket_cnt; i++) {
                if (!strcmp(vhost_user.vsockets[i]->path, path)) {
-                       fdset_del(&vhost_user.fdset,
-                               vhost_user.vsockets[i]->listenfd);
+                       if (vhost_user.vsockets[i]->is_server) {
+                               fdset_del(&vhost_user.fdset,
+                                       vhost_user.vsockets[i]->listenfd);
+                               close(vhost_user.vsockets[i]->listenfd);
+                               unlink(path);
+                       }
 
-                       close(vhost_user.vsockets[i]->listenfd);
                        free(vhost_user.vsockets[i]->path);
                        free(vhost_user.vsockets[i]);
 
-                       unlink(path);
-
                        count = --vhost_user.vsocket_cnt;
                        vhost_user.vsockets[i] = vhost_user.vsockets[count];
                        vhost_user.vsockets[count] = NULL;