summaryrefslogtreecommitdiffhomepage
path: root/remotetcpfwd.c
diff options
context:
space:
mode:
Diffstat (limited to 'remotetcpfwd.c')
-rw-r--r--remotetcpfwd.c301
1 files changed, 301 insertions, 0 deletions
diff --git a/remotetcpfwd.c b/remotetcpfwd.c
new file mode 100644
index 0000000..c58b820
--- /dev/null
+++ b/remotetcpfwd.c
@@ -0,0 +1,301 @@
+#include "includes.h"
+#include "ssh.h"
+#include "remotetcpfwd.h"
+#include "dbutil.h"
+#include "session.h"
+#include "buffer.h"
+#include "packet.h"
+#include "tcpfwd.h"
+
+#ifndef DISABLE_REMOTETCPFWD
+
+struct RemoteTCP {
+
+ unsigned char* addr;
+ unsigned int port;
+
+};
+
+static void send_msg_request_success();
+static void send_msg_request_failure();
+static int cancelremotetcp();
+static int remotetcpreq();
+static int newlistener(unsigned char* bindaddr, unsigned int port);
+static void acceptremote(struct TCPListener *listener);
+
+/* At the moment this is completely used for tcp code (with the name reflecting
+ * that). If new request types are added, this should be replaced with code
+ * similar to the request-switching in chansession.c */
+void recv_msg_global_request_remotetcp() {
+
+ unsigned char* reqname = NULL;
+ unsigned int namelen;
+ unsigned int wantreply = 0;
+ int ret = DROPBEAR_FAILURE;
+
+ TRACE(("enter recv_msg_global_request_remotetcp"));
+
+ if (ses.opts->noremotetcp) {
+ TRACE(("leave recv_msg_global_request_remotetcp: remote tcp forwarding disabled"));
+ goto out;
+ }
+
+ reqname = buf_getstring(ses.payload, &namelen);
+ wantreply = buf_getbyte(ses.payload);
+
+ if (namelen > MAXNAMLEN) {
+ TRACE(("name len is wrong: %d", namelen));
+ goto out;
+ }
+
+ if (strcmp("tcpip-forward", reqname) == 0) {
+ ret = remotetcpreq();
+ } else if (strcmp("cancel-tcpip-forward", reqname) == 0) {
+ ret = cancelremotetcp();
+ } else {
+ TRACE(("reqname isn't tcpip-forward: '%s'", reqname));
+ }
+
+out:
+ if (wantreply) {
+ if (ret == DROPBEAR_SUCCESS) {
+ send_msg_request_success();
+ } else {
+ send_msg_request_failure();
+ }
+ }
+
+ m_free(reqname);
+
+ TRACE(("leave recv_msg_global_request"));
+}
+
+static void acceptremote(struct TCPListener *listener) {
+
+ int fd;
+ struct sockaddr addr;
+ int len;
+ char ipstring[NI_MAXHOST], portstring[NI_MAXSERV];
+ struct RemoteTCP *tcpinfo = (struct RemoteTCP*)(listener->typedata);
+
+ len = sizeof(addr);
+
+ fd = accept(listener->sock, &addr, &len);
+ if (fd < 0) {
+ return;
+ }
+
+ if (getnameinfo(&addr, len, ipstring, sizeof(ipstring), portstring,
+ sizeof(portstring), NI_NUMERICHOST | NI_NUMERICSERV) != 0) {
+ return;
+ }
+
+ if (send_msg_channel_open_init(fd, CHANNEL_ID_TCPFORWARDED,
+ "forwarded-tcpip") == DROPBEAR_SUCCESS) {
+ buf_putstring(ses.writepayload, tcpinfo->addr,
+ strlen(tcpinfo->addr));
+ buf_putint(ses.writepayload, tcpinfo->port);
+ buf_putstring(ses.writepayload, ipstring, strlen(ipstring));
+ buf_putint(ses.writepayload, atol(portstring));
+ encrypt_packet();
+ }
+}
+
+static void cleanupremote(struct TCPListener *listener) {
+
+ struct RemoteTCP *tcpinfo = (struct RemoteTCP*)(listener->typedata);
+
+ m_free(tcpinfo->addr);
+ m_free(tcpinfo);
+}
+
+static void send_msg_request_success() {
+
+ CHECKCLEARTOWRITE();
+ buf_putbyte(ses.writepayload, SSH_MSG_REQUEST_SUCCESS);
+ encrypt_packet();
+
+}
+
+static void send_msg_request_failure() {
+
+ CHECKCLEARTOWRITE();
+ buf_putbyte(ses.writepayload, SSH_MSG_REQUEST_FAILURE);
+ encrypt_packet();
+
+}
+
+static int matchtcp(void* typedata1, void* typedata2) {
+
+ const struct RemoteTCP *info1 = (struct RemoteTCP*)typedata1;
+ const struct RemoteTCP *info2 = (struct RemoteTCP*)typedata2;
+
+ return info1->port == info2->port
+ && (strcmp(info1->addr, info2->addr) == 0);
+}
+
+static int cancelremotetcp() {
+
+ int ret = DROPBEAR_FAILURE;
+ unsigned char * bindaddr = NULL;
+ unsigned int addrlen;
+ unsigned int port;
+ struct TCPListener * listener = NULL;
+ struct RemoteTCP tcpinfo;
+
+ TRACE(("enter cancelremotetcp"));
+
+ bindaddr = buf_getstring(ses.payload, &addrlen);
+ if (addrlen > MAX_IP_LEN) {
+ TRACE(("addr len too long: %d", addrlen));
+ goto out;
+ }
+
+ port = buf_getint(ses.payload);
+
+ tcpinfo.addr = bindaddr;
+ tcpinfo.port = port;
+ listener = get_listener(CHANNEL_ID_TCPFORWARDED, &tcpinfo, matchtcp);
+ if (listener) {
+ remove_listener( listener );
+ ret = DROPBEAR_SUCCESS;
+ }
+
+out:
+ m_free(bindaddr);
+ TRACE(("leave cancelremotetcp"));
+ return ret;
+}
+
+static int remotetcpreq() {
+
+ int ret = DROPBEAR_FAILURE;
+ unsigned char * bindaddr = NULL;
+ unsigned int addrlen;
+ unsigned int port;
+
+ TRACE(("enter remotetcpreq"));
+
+ bindaddr = buf_getstring(ses.payload, &addrlen);
+ if (addrlen > MAX_IP_LEN) {
+ TRACE(("addr len too long: %d", addrlen));
+ goto out;
+ }
+
+ port = buf_getint(ses.payload);
+
+ if (port == 0) {
+ dropbear_log(LOG_INFO, "Server chosen tcpfwd ports are unsupported");
+ goto out;
+ }
+
+ if (port < 1 || port > 65535) {
+ TRACE(("invalid port: %d", port));
+ goto out;
+ }
+
+ /* XXX matt - server change
+ if (ses.authstate.pw->pw_uid != 0
+ && port < IPPORT_RESERVED) {
+ TRACE(("can't assign port < 1024 for non-root"));
+ goto out;
+ }
+ */
+
+ ret = newlistener(bindaddr, port);
+
+out:
+ if (ret == DROPBEAR_FAILURE) {
+ /* we only free it if a listener wasn't created, since the listener
+ * has to remember it if it's to be cancelled */
+ m_free(bindaddr);
+ }
+ TRACE(("leave remotetcpreq"));
+ return ret;
+}
+
+static int newlistener(unsigned char* bindaddr, unsigned int port) {
+
+ struct RemoteTCP * tcpinfo = NULL;
+ char portstring[6]; /* "65535\0" */
+ struct addrinfo *res = NULL, *ai = NULL;
+ struct addrinfo hints;
+ int sock = -1;
+ int ret = DROPBEAR_FAILURE;
+
+ TRACE(("enter newlistener"));
+
+ /* first we try to bind, so don't need to do so much cleanup on failure */
+ snprintf(portstring, sizeof(portstring), "%d", port);
+ memset(&hints, 0x0, sizeof(hints));
+ hints.ai_socktype = SOCK_STREAM;
+ hints.ai_family = PF_INET;
+ hints.ai_flags = AI_PASSIVE | AI_NUMERICHOST;
+
+ if (getaddrinfo(bindaddr, portstring, &hints, &res) < 0) {
+ TRACE(("leave newlistener: getaddrinfo failed: %s",
+ strerror(errno)));
+ goto done;
+ }
+
+ /* find the first one which works */
+ for (ai = res; ai != NULL; ai = ai->ai_next) {
+ if (ai->ai_family != PF_INET && ai->ai_family != PF_INET6) {
+ continue;
+ }
+
+ sock = socket(ai->ai_family, SOCK_STREAM, 0);
+ if (sock < 0) {
+ TRACE(("socket failed: %s", strerror(errno)));
+ goto fail;
+ }
+
+ if (bind(sock, ai->ai_addr, ai->ai_addrlen) < 0) {
+ TRACE(("bind failed: %s", strerror(errno)));
+ goto fail;
+ }
+
+ if (listen(sock, 20) < 0) {
+ TRACE(("listen failed: %s", strerror(errno)));
+ goto fail;
+ }
+
+ if (fcntl(sock, F_SETFL, O_NONBLOCK) < 0) {
+ TRACE(("fcntl nonblocking failed: %s", strerror(errno)));
+ goto fail;
+ }
+
+ /* success */
+ break;
+
+fail:
+ close(sock);
+ }
+
+
+ if (ai == NULL) {
+ TRACE(("no successful sockets"));
+ goto done;
+ }
+
+ tcpinfo = (struct RemoteTCP*)m_malloc(sizeof(struct RemoteTCP));
+ tcpinfo->addr = bindaddr;
+ tcpinfo->port = port;
+
+ ret = new_fwd(sock, CHANNEL_ID_TCPFORWARDED, tcpinfo,
+ acceptremote, cleanupremote);
+
+ if (ret == DROPBEAR_FAILURE) {
+ m_free(tcpinfo);
+ }
+
+done:
+ if (res) {
+ freeaddrinfo(res);
+ }
+
+ TRACE(("leave newlistener"));
+ return ret;
+}
+
+#endif /* DISABLE_REMOTETCPFWD */