[内核版本]
linux-2.6.31
[尚存缺憾]
1、getsockopt和setsockopt的某些特定参数的调用(net/unix/af_unix.c中定义的*sockop函数均保留接口,返回EOPNOTSUPP错误);
2、ss命令查看到的本地socket状态需要进一步确认;
[注意事项]
1、使用本地socket进行通信时,其通信过程并不通过报文交互进行状态机切换:
a)server端在执行listen函数之后,该socket即处于监听状态;
b)client端在执行connect函数时,正常情况下,内核将sock状态设置为SS_CONNECTED,将sk状态设置为TCP_ESTABLISHED,然后通知server端有client请求(向server的socket发送SIGIO信号);
c)一旦出现某一情况导致server端的最大链接请求(应该是由sk_max_ack_backlog成员定义,在listen函数中指定)使用完毕,如果不设置连接超时时间,client端应该在2147483647秒的时间超时
// include/linux/kernel.h
#define LONG_MAX((long)(~0UL>>1))
// include/linux/sched.h
#defineMAX_SCHEDULE_TIMEOUTLONG_MAX
// net/core/sock.c
void sock_init_data(struct socket *sock, struct sock *sk)
{
...
sk->sk_sndtimeo=MAX_SCHEDULE_TIMEOUT;
...
}
2、本地socket和网络socket的通信机制有些不同(比如,tcp状态机不通过报文交互实现,getsockopt不能取到tcp的连接状态,内核在处理本地socket时,可能不会有一些协议栈的过程,即可能忽略了报文的完整性检查),很多网络socket上使用的函数在迁移至本地socket时需要酌情考虑。
[server端代码]
int ct_ipc_sock_create(struct ct_fd * sockfd, int server_flag) { log_debug("server_flag: %d", server_flag); int ret = CT_RET_SUCCESS; int result = 0; struct sockaddr_un srv_addr; memset(&srv_addr, 0, sizeof(struct sockaddr_un)); if (SERVER_SOCKET == server_flag) { sockfd->fd = socket(PF_UNIX, SOCK_DGRAM, 0); if (0 > sockfd->fd) { log_err("create Unix Socket error"); ret = CT_RET_SYS_SOCK_CREATE_ERR; goto ct_ipc_sock_create_err; } srv_addr.sun_family=AF_UNIX; strncpy(srv_addr.sun_path, UNIX_DOMAIN, sizeof(srv_addr.sun_path)-1); unlink(UNIX_DOMAIN); ret=bind(sockfd->fd,(struct sockaddr*)&srv_addr,sizeof(srv_addr)); if(ret==-1) { log_err("cannot bind server socket (%d)", sockfd->fd); ret = CT_RET_SYS_SOCK_BIND_ERR; goto ct_ipc_sock_bind_listen_err; } } else if (CLIENT_SOCKET == server_flag) { sockfd->fd = socket(PF_UNIX, SOCK_STREAM, 0); if (0 > sockfd->fd) { log_err("create Unix Socket error"); ret = CT_RET_SYS_SOCK_CREATE_ERR; goto ct_ipc_sock_create_err; } log_debug("sockfd: %d", sockfd->fd); setsockopt(sockfd->fd, SOL_SOCKET, TCP_NODELAY, &result, sizeof(int)); srv_addr.sun_family=AF_UNIX; strncpy(srv_addr.sun_path, UNIX_DOMAIN_CLOUD, sizeof(srv_addr.sun_path)-1); unlink(UNIX_DOMAIN_CLOUD); ret=bind(sockfd->fd,(struct sockaddr*)&srv_addr,sizeof(srv_addr)); if(ret==-1) { log_err("cannot bind server socket (%d)", sockfd->fd); ret = CT_RET_SYS_SOCK_BIND_ERR; goto ct_ipc_sock_bind_listen_err; } log_debug("%s", strerror(errno)); ret=listen(sockfd->fd, MAX_LISTEN_BACKLOG); if(ret==-1) { log_err("cannot listen the client connect request (%d)", sockfd); ret = CT_RET_SYS_SOCK_LISTEN_ERR; goto ct_ipc_sock_bind_listen_err; } } else { } log_debug("%s", strerror(errno)); return ret; ct_ipc_sock_bind_listen_err: close(sockfd->fd); ct_ipc_sock_create_err: sockfd->fd = -1; return ret; } int ct_select_run() { int ret = CT_RET_SUCCESS; fd_set rset, allset; int nfds = 0; int maxfd = 0; int cloudfd = 0; int read_num = 0; int write_num = 0; socklen_t addrLen; struct sockaddr_in addr; char rsp[] = "ACK"; char data_buf[MAX_BUF] = {0}; log_debug("sockfd: %d, nlfd.fd: %d, cloud_sockfd.fd: %d", ipc_sockfd.fd, nlfd.fd, cloud_sockfd.fd); group_filter_init(); maxfd = ipc_sockfd.fd ; FD_ZERO(&allset); FD_SET(cloud_sockfd.fd, &allset); log_debug("sockfd: %d, maxfd: %d, cloud_sockfd.fd: %d", ipc_sockfd.fd, maxfd, cloud_sockfd.fd); while (1/*nfds == 0*/) { rset = allset; log_debug("sockfd: %d", ipc_sockfd.fd); nfds = select(maxfd + 1, &rset, NULL, NULL, NULL); if (0 > nfds) { log_err("select error: %s", strerror(errno)); ret = CT_RET_SYS_SOCK_SELECT_ERR; return ret; } if (FD_ISSET(cloud_sockfd.fd, &rset)) { log_debug("cloud_sockfd: %d", cloud_sockfd.fd); cloudfd = accept(cloud_sockfd.fd, (struct sockaddr *)&addr, &addrLen); if (cloudfd < 0) { log_err("accept (%d) client error: %d", cloud_sockfd.fd, ret); continue; } read_num = read(cloudfd, data_buf, MAX_BUF); log_debug("client socket is: %d, read_num: %d", cloudfd, read_num); if (read_num > 0) { write_num = write(cloudfd, sta_info, (sizeof(sta_list_attr)*num + sizeof(int))); log_debug("client socket is: %d, write_num: %d", cloudfd, write_num);; } } close(cloudfd); } } return ret; }
[client端代码]
int do_filter_notify(cloud_cmd_attr * attr, char * buf, int buf_len) { int ret = 0; int client_fd = -1; char recv_buf[128] = {0}; struct sockaddr_un srv_addr; int maxfd = 0; int nfds = 0; fd_set rset, allset; struct timeval tv; struct timeval t; /* Wait up to five seconds. */ tv.tv_sec = CLI_TIME_OUT; tv.tv_usec = 0; memset(&srv_addr, 0, sizeof(struct sockaddr_un)); client_fd = socket(AF_UNIX, SOCK_STREAM, 0); log_debug("client_fd: %d\n", client_fd); printf("client_fd: %d\n", client_fd); if (-1 != client_fd) { t.tv_sec = UNIX_SOCK_CONNECT_TIMEO; t.tv_usec = 0; ret = setsockopt(client_fd, SOL_SOCKET, SO_SNDTIMEO, &t, sizeof(t)); srv_addr.sun_family = AF_UNIX; memcpy(srv_addr.sun_path, UNIX_DOMAIN_CLOUD, strlen(UNIX_DOMAIN_CLOUD)); log_debug("srv_addr.sun_path: %s\n", srv_addr.sun_path); printf("srv_addr.sun_path: %s\n", srv_addr.sun_path); ret = connect(client_fd, (struct sockaddr *)&srv_addr, sizeof(srv_addr)); if (-1 == ret) { log_err("error: %d, %s\n", errno, strerror(errno)); ret = CT_RET_SYS_SOCK_CONNECT_ERR; goto close_fd; } else { int write_num = 0; write_num = write(client_fd, attr, sizeof(cloud_cmd_attr)); if (sizeof(cloud_cmd_attr) == write_num) { FD_ZERO(&allset); FD_SET(client_fd, &allset); maxfd = client_fd; rset = allset; nfds = select(maxfd + 1, &rset, NULL, NULL, &tv); log_debug("nfds: %d\n", nfds); printf("nfds: %d\n", nfds); if (FD_ISSET(client_fd, &rset)) { int num = 0; if (NULL == buf) { num = read(client_fd, recv_buf, sizeof(recv_buf)); } else { num = read(client_fd, buf, buf_len); } if (num <= 0) { log_err("read %d error", client_fd); ret = CT_RET_FD_READ_ERR; } else { log_debug("client_fd: %d, read length: %d", client_fd, num); printf("client_fd: %d, read length: %d", client_fd, num); if (NULL == buf) { log_debug("recv_buf: %s\n", recv_buf); if (!strcasecmp(recv_buf, "ack")) { ret = CT_RET_SUCCESS; } } else { // do nothing } } goto close_fd; } else { log_err("no response from: %d", client_fd); printf("no response from: %d", client_fd); ret = CT_RET_SOCK_SELECT_TIMEOUT; goto close_fd; } } else { log_err("(%d) write to peer %s error (num: %d)", client_fd, UNIX_DOMAIN_CLOUD, ret); printf("(%d) write to peer %s error (num: %d)", client_fd, UNIX_DOMAIN_CLOUD, ret); ret = CT_RET_FD_WRITE_ERR; goto close_fd; } } } else { log_err("create Unix socket error"); printf("create Unix socket error"); ret = CT_RET_SYS_SOCK_CREATE_ERR; } return ret; close_fd: close(client_fd); client_fd = -1; return ret; }
原文地址:http://scottgh.blog.51cto.com/406525/1577623