linux-tcpip.c

非常经典的一个分布式系统

        unsigned long  then;
        struct timeval tv;

        LASSERT (nob > 0);
        LASSERT (ticks > 0);

        for (;;) {
                struct iovec  iov = {
                        .iov_base = buffer,
                        .iov_len  = nob
                };
                struct msghdr msg = {
                        .msg_name       = NULL,
                        .msg_namelen    = 0,
                        .msg_iov        = &iov,
                        .msg_iovlen     = 1,
                        .msg_control    = NULL,
                        .msg_controllen = 0,
                        .msg_flags      = 0
                };

                /* Set receive timeout to remaining time */
                tv = (struct timeval) {
                        .tv_sec = ticks / HZ,
                        .tv_usec = ((ticks % HZ) * 1000000) / HZ
                };
                set_fs(KERNEL_DS);
                rc = sock_setsockopt(sock, SOL_SOCKET, SO_RCVTIMEO,
                                     (char *)&tv, sizeof(tv));
                set_fs(oldmm);
                if (rc != 0) {
                        CERROR("Can't set socket recv timeout %ld.%06d: %d\n",
                               (long)tv.tv_sec, (int)tv.tv_usec, rc);
                        return rc;
                }

                set_fs(KERNEL_DS);
                then = jiffies;
                rc = sock_recvmsg(sock, &msg, iov.iov_len, 0);
                ticks -= jiffies - then;
                set_fs(oldmm);

                if (rc < 0)
                        return rc;

                if (rc == 0)
                        return -ECONNRESET;

                buffer = ((char *)buffer) + rc;
                nob -= rc;

                if (nob == 0)
                        return 0;

                if (ticks <= 0)
                        return -ETIMEDOUT;
        }
}

EXPORT_SYMBOL(libcfs_sock_read);

static int
libcfs_sock_create (struct socket **sockp, int *fatal,
                    __u32 local_ip, int local_port)
{
        struct sockaddr_in  locaddr;
        struct socket      *sock;
        int                 rc;
        int                 option;
        mm_segment_t        oldmm = get_fs();

        /* All errors are fatal except bind failure if the port is in use */
        *fatal = 1;

        rc = sock_create (PF_INET, SOCK_STREAM, 0, &sock);
        *sockp = sock;
        if (rc != 0) {
                CERROR ("Can't create socket: %d\n", rc);
                return (rc);
        }

        set_fs (KERNEL_DS);
        option = 1;
        rc = sock_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR,
                             (char *)&option, sizeof (option));
        set_fs (oldmm);
        if (rc != 0) {
                CERROR("Can't set SO_REUSEADDR for socket: %d\n", rc);
                goto failed;
        }

        if (local_ip != 0 || local_port != 0) {
                memset(&locaddr, 0, sizeof(locaddr));
                locaddr.sin_family = AF_INET;
                locaddr.sin_port = htons(local_port);
                locaddr.sin_addr.s_addr = (local_ip == 0) ?
                                          INADDR_ANY : htonl(local_ip);

                rc = sock->ops->bind(sock, (struct sockaddr *)&locaddr,
                                     sizeof(locaddr));
                if (rc == -EADDRINUSE) {
                        CDEBUG(D_NET, "Port %d already in use\n", local_port);
                        *fatal = 0;
                        goto failed;
                }
                if (rc != 0) {
                        CERROR("Error trying to bind to port %d: %d\n",
                               local_port, rc);
                        goto failed;
                }
        }

        return 0;

 failed:
        sock_release(sock);
        return rc;
}

int
libcfs_sock_setbuf (struct socket *sock, int txbufsize, int rxbufsize)
{
        mm_segment_t        oldmm = get_fs();
        int                 option;
        int                 rc;

        if (txbufsize != 0) {
                option = txbufsize;
                set_fs (KERNEL_DS);
                rc = sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
                                     (char *)&option, sizeof (option));
                set_fs (oldmm);
                if (rc != 0) {
                        CERROR ("Can't set send buffer %d: %d\n",
                                option, rc);
                        return (rc);
                }
        }

        if (rxbufsize != 0) {
                option = rxbufsize;
                set_fs (KERNEL_DS);
                rc = sock_setsockopt (sock, SOL_SOCKET, SO_RCVBUF,
                                      (char *)&option, sizeof (option));
                set_fs (oldmm);
                if (rc != 0) {
                        CERROR ("Can't set receive buffer %d: %d\n",
                                option, rc);
                        return (rc);
                }
        }

        return 0;
}

EXPORT_SYMBOL(libcfs_sock_setbuf);

int
libcfs_sock_getaddr (struct socket *sock, int remote, __u32 *ip, int *port)
{
        struct sockaddr_in sin;
        int                len = sizeof (sin);
        int                rc;

        rc = sock->ops->getname (sock, (struct sockaddr *)&sin, &len,
                                 remote ? 2 : 0);
        if (rc != 0) {
                CERROR ("Error %d getting sock %s IP/port\n",
                        rc, remote ? "peer" : "local");
                return rc;
        }

        if (ip != NULL)
                *ip = ntohl (sin.sin_addr.s_addr);

        if (port != NULL)
                *port = ntohs (sin.sin_port);

        return 0;
}

EXPORT_SYMBOL(libcfs_sock_getaddr);

int
libcfs_sock_getbuf (struct socket *sock, int *txbufsize, int *rxbufsize)
{

        if (txbufsize != NULL) {
                *txbufsize = sock->sk->sk_sndbuf;
        }

        if (rxbufsize != NULL) {
                *rxbufsize = sock->sk->sk_rcvbuf;
        }

        return 0;
}

EXPORT_SYMBOL(libcfs_sock_getbuf);

int
libcfs_sock_listen (struct socket **sockp,
                    __u32 local_ip, int local_port, int backlog)
{
        int      fatal;
        int      rc;

        rc = libcfs_sock_create(sockp, &fatal, local_ip, local_port);
        if (rc != 0) {
                if (!fatal)
                        CERROR("Can't create socket: port %d already in use\n",
                               local_port);
                return rc;
        }

        rc = (*sockp)->ops->listen(*sockp, backlog);
        if (rc == 0)
                return 0;

        CERROR("Can't set listen backlog %d: %d\n", backlog, rc);
        sock_release(*sockp);
        return rc;
}

EXPORT_SYMBOL(libcfs_sock_listen);

#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,12)
int sock_create_lite(int family, int type, int protocol, struct socket **res)
{
        struct socket *sock;

        sock = sock_alloc();
        if (sock == NULL) 
                return -ENOMEM;

        sock->type = type;
        *res = sock;

        return 0;
}
#endif

int
libcfs_sock_accept (struct socket **newsockp, struct socket *sock)
{
        wait_queue_t   wait;
        struct socket *newsock;
        int            rc;

        init_waitqueue_entry(&wait, current);

        /* XXX this should add a ref to sock->ops->owner, if
         * TCP could be a module */
        rc = sock_create_lite(PF_PACKET, sock->type, IPPROTO_TCP, &newsock);
        if (rc) {
                CERROR("Can't allocate socket\n");
                return rc;
        }

        newsock->ops = sock->ops;

        set_current_state(TASK_INTERRUPTIBLE);
        add_wait_queue(sock->sk->sk_sleep, &wait);

        rc = sock->ops->accept(sock, newsock, O_NONBLOCK);
        if (rc == -EAGAIN) {
                /* Nothing ready, so wait for activity */
                schedule();
                rc = sock->ops->accept(sock, newsock, O_NONBLOCK);
        }

        remove_wait_queue(sock->sk->sk_sleep, &wait);
        set_current_state(TASK_RUNNING);

        if (rc != 0)
                goto failed;

        *newsockp = newsock;
        return 0;

 failed:
        sock_release(newsock);
        return rc;
}

EXPORT_SYMBOL(libcfs_sock_accept);

void
libcfs_sock_abort_accept (struct socket *sock)
{
        wake_up_all(sock->sk->sk_sleep);
}

EXPORT_SYMBOL(libcfs_sock_abort_accept);

int
libcfs_sock_connect (struct socket **sockp, int *fatal,
                     __u32 local_ip, int local_port,
                     __u32 peer_ip, int peer_port)
{
        struct sockaddr_in  srvaddr;
        int                 rc;

        rc = libcfs_sock_create(sockp, fatal, local_ip, local_port);
        if (rc != 0)
                return rc;

        memset (&srvaddr, 0, sizeof (srvaddr));
        srvaddr.sin_family = AF_INET;
        srvaddr.sin_port = htons(peer_port);
        srvaddr.sin_addr.s_addr = htonl(peer_ip);

        rc = (*sockp)->ops->connect(*sockp,
                                    (struct sockaddr *)&srvaddr, sizeof(srvaddr),
                                    0);
        if (rc == 0)
                return 0;

        /* EADDRNOTAVAIL probably means we're already connected to the same
         * peer/port on the same local port on a differently typed
         * connection.  Let our caller retry with a different local
         * port... */
        *fatal = !(rc == -EADDRNOTAVAIL);

        CDEBUG(*fatal ? D_NETERROR : D_NET,
               "Error %d connecting %u.%u.%u.%u/%d -> %u.%u.%u.%u/%d\n", rc,
               HIPQUAD(local_ip), local_port, HIPQUAD(peer_ip), peer_port);

        sock_release(*sockp);
        return rc;
}

EXPORT_SYMBOL(libcfs_sock_connect);

void
libcfs_sock_release (struct socket *sock)
{
        sock_release(sock);
}

EXPORT_SYMBOL(libcfs_sock_release);