view netio.c @ 1861:2b3a8026a6ce

Add re-exec for server This allows ASLR to re-randomize the address space for every connection, preventing some vulnerabilities from being exploitable by repeated probing. Overhead (memory and time) is yet to be confirmed. At present this is only enabled on Linux. Other BSD platforms with fexecve() would probably also work though have not been tested.
author Matt Johnston <matt@ucc.asn.au>
date Sun, 30 Jan 2022 10:14:56 +0800
parents 5001e9c5641f
children
line wrap: on
line source

#include "netio.h"
#include "list.h"
#include "dbutil.h"
#include "session.h"
#include "debug.h"

struct dropbear_progress_connection {
	struct addrinfo *res;
	struct addrinfo *res_iter;

	char *remotehost, *remoteport; /* For error reporting */

	connect_callback cb;
	void *cb_data;

	struct Queue *writequeue; /* A queue of encrypted packets to send with TCP fastopen,
								or NULL. */

	int sock;

	char* errstring;
	char *bind_address, *bind_port;
	enum dropbear_prio prio;
};

/* Deallocate a progress connection. Removes from the pending list if iter!=NULL.
Does not close sockets */
static void remove_connect(struct dropbear_progress_connection *c, m_list_elem *iter) {
	if (c->res) {
		freeaddrinfo(c->res);
	}
	m_free(c->remotehost);
	m_free(c->remoteport);
	m_free(c->errstring);
	m_free(c->bind_address);
	m_free(c->bind_port);
	m_free(c);

	if (iter) {
		list_remove(iter);
	}
}

static void cancel_callback(int result, int sock, void* UNUSED(data), const char* UNUSED(errstring)) {
	if (result == DROPBEAR_SUCCESS)
	{
		m_close(sock);
	}
}

void cancel_connect(struct dropbear_progress_connection *c) {
	c->cb = cancel_callback;
	c->cb_data = NULL;
}

static void connect_try_next(struct dropbear_progress_connection *c) {
	struct addrinfo *r;
	int err;
	int res = 0;
	int fastopen = 0;
#if DROPBEAR_CLIENT_TCP_FAST_OPEN
	struct msghdr message;
#endif

	for (r = c->res_iter; r; r = r->ai_next)
	{
		dropbear_assert(c->sock == -1);

		c->sock = socket(r->ai_family, r->ai_socktype, r->ai_protocol);
		if (c->sock < 0) {
			continue;
		}

		if (c->bind_address || c->bind_port) {
			/* bind to a source port/address */
			struct addrinfo hints;
			struct addrinfo *bindaddr = NULL;
			memset(&hints, 0, sizeof(hints));
			hints.ai_socktype = SOCK_STREAM;
			hints.ai_family = r->ai_family;
			hints.ai_flags = AI_PASSIVE;

			err = getaddrinfo(c->bind_address, c->bind_port, &hints, &bindaddr);
			if (err) {
				int len = 100 + strlen(gai_strerror(err));
				m_free(c->errstring);
				c->errstring = (char*)m_malloc(len);
				snprintf(c->errstring, len, "Error resolving bind address '%s' (port %s). %s", 
						c->bind_address, c->bind_port, gai_strerror(err));
				TRACE(("Error resolving bind: %s", gai_strerror(err)))
				close(c->sock);
				c->sock = -1;
				continue;
			}
			res = bind(c->sock, bindaddr->ai_addr, bindaddr->ai_addrlen);
			freeaddrinfo(bindaddr);
			bindaddr = NULL;
			if (res < 0) {
				/* failure */
				int keep_errno = errno;
				int len = 300;
				m_free(c->errstring);
				c->errstring = m_malloc(len);
				snprintf(c->errstring, len, "Error binding local address '%s' (port %s). %s", 
						c->bind_address, c->bind_port, strerror(keep_errno));
				close(c->sock);
				c->sock = -1;
				continue;
			}
		}

		ses.maxfd = MAX(ses.maxfd, c->sock);
		set_sock_nodelay(c->sock);
		set_sock_priority(c->sock, c->prio);
		setnonblocking(c->sock);

#if DROPBEAR_CLIENT_TCP_FAST_OPEN
		fastopen = (c->writequeue != NULL);

		if (fastopen) {
			memset(&message, 0x0, sizeof(message));
			message.msg_name = r->ai_addr;
			message.msg_namelen = r->ai_addrlen;
			/* 6 is arbitrary, enough to hold initial packets */
			unsigned int iovlen = 6; /* Linux msg_iovlen is a size_t */
			struct iovec iov[6];
			packet_queue_to_iovec(c->writequeue, iov, &iovlen);
			message.msg_iov = iov;
			message.msg_iovlen = iovlen;
			res = sendmsg(c->sock, &message, MSG_FASTOPEN);
			/* Returns EINPROGRESS if FASTOPEN wasn't available */
			if (res < 0) {
				if (errno != EINPROGRESS) {
					m_free(c->errstring);
					c->errstring = m_strdup(strerror(errno));
					/* Not entirely sure which kind of errors are normal - 2.6.32 seems to 
					return EPIPE for any (nonblocking?) sendmsg(). just fall back */
					TRACE(("sendmsg tcp_fastopen failed, falling back. %s", strerror(errno)));
					/* No kernel MSG_FASTOPEN support. Fall back below */
					fastopen = 0;
					/* Set to NULL to avoid trying again */
					c->writequeue = NULL;
				}
			} else {
				packet_queue_consume(c->writequeue, res);
			}
		}
#endif

		/* Normal connect(), used as fallback for TCP fastopen too */
		if (!fastopen) {
			res = connect(c->sock, r->ai_addr, r->ai_addrlen);
		}

		if (res < 0 && errno != EINPROGRESS) {
			/* failure */
			m_free(c->errstring);
			c->errstring = m_strdup(strerror(errno));
			close(c->sock);
			c->sock = -1;
			continue;
		} else {
			/* new connection was successful, wait for it to complete */
			break;
		}
	}

	if (r) {
		c->res_iter = r->ai_next;
	} else {
		c->res_iter = NULL;
	}
}

/* Connect via TCP to a host. */
struct dropbear_progress_connection *connect_remote(const char* remotehost, const char* remoteport,
	connect_callback cb, void* cb_data,
	const char* bind_address, const char* bind_port, enum dropbear_prio prio)
{
	struct dropbear_progress_connection *c = NULL;
	int err;
	struct addrinfo hints;

	c = m_malloc(sizeof(*c));
	c->remotehost = m_strdup(remotehost);
	c->remoteport = m_strdup(remoteport);
	c->sock = -1;
	c->cb = cb;
	c->cb_data = cb_data;
	c->prio = prio;

	list_append(&ses.conn_pending, c);

#if DROPBEAR_FUZZ
	if (fuzz.fuzzing) {
		c->errstring = m_strdup("fuzzing connect_remote always fails");
		return c;
	}
#endif

	memset(&hints, 0, sizeof(hints));
	hints.ai_socktype = SOCK_STREAM;
	hints.ai_family = AF_UNSPEC;

	err = getaddrinfo(remotehost, remoteport, &hints, &c->res);
	if (err) {
		int len;
		len = 100 + strlen(gai_strerror(err));
		c->errstring = (char*)m_malloc(len);
		snprintf(c->errstring, len, "Error resolving '%s' port '%s'. %s", 
				remotehost, remoteport, gai_strerror(err));
		TRACE(("Error resolving: %s", gai_strerror(err)))
	} else {
		c->res_iter = c->res;
	}
	
	if (bind_address) {
		c->bind_address = m_strdup(bind_address);
	}
	if (bind_port) {
		c->bind_port = m_strdup(bind_port);
	}

	return c;
}

void remove_connect_pending() {
	while (ses.conn_pending.first) {
		struct dropbear_progress_connection *c = ses.conn_pending.first->item;
		remove_connect(c, ses.conn_pending.first);
	}
}


void set_connect_fds(fd_set *writefd) {
	m_list_elem *iter;
	iter = ses.conn_pending.first;
	while (iter) {
		m_list_elem *next_iter = iter->next;
		struct dropbear_progress_connection *c = iter->item;
		/* Set one going */
		while (c->res_iter && c->sock < 0) {
			connect_try_next(c);
		}
		if (c->sock >= 0) {
			FD_SET(c->sock, writefd);
		} else {
			/* Final failure */
			if (!c->errstring) {
				c->errstring = m_strdup("unexpected failure");
			}
			c->cb(DROPBEAR_FAILURE, -1, c->cb_data, c->errstring);
			remove_connect(c, iter);
		}
		iter = next_iter;
	}
}

void handle_connect_fds(const fd_set *writefd) {
	m_list_elem *iter;
	for (iter = ses.conn_pending.first; iter; iter = iter->next) {
		int val;
		socklen_t vallen = sizeof(val);
		struct dropbear_progress_connection *c = iter->item;

		if (c->sock < 0 || !FD_ISSET(c->sock, writefd)) {
			continue;
		}

		TRACE(("handling %s port %s socket %d", c->remotehost, c->remoteport, c->sock));

		if (getsockopt(c->sock, SOL_SOCKET, SO_ERROR, &val, &vallen) != 0) {
			TRACE(("handle_connect_fds getsockopt(%d) SO_ERROR failed: %s", c->sock, strerror(errno)))
			/* This isn't expected to happen - Unix has surprises though, continue gracefully. */
			m_close(c->sock);
			c->sock = -1;
		} else if (val != 0) {
			/* Connect failed */
			TRACE(("connect to %s port %s failed.", c->remotehost, c->remoteport))
			m_close(c->sock);
			c->sock = -1;

			m_free(c->errstring);
			c->errstring = m_strdup(strerror(val));
		} else {
			/* New connection has been established */
			c->cb(DROPBEAR_SUCCESS, c->sock, c->cb_data, NULL);
			remove_connect(c, iter);
			TRACE(("leave handle_connect_fds - success"))
			/* Must return here - remove_connect() invalidates iter */
			return; 
		}
	}
}

void connect_set_writequeue(struct dropbear_progress_connection *c, struct Queue *writequeue) {
	c->writequeue = writequeue;
}

void packet_queue_to_iovec(const struct Queue *queue, struct iovec *iov, unsigned int *iov_count) {
	struct Link *l;
	unsigned int i;
	int len;
	buffer *writebuf;

#ifndef IOV_MAX
	#if defined(__CYGWIN__) && !defined(UIO_MAXIOV)
		#define IOV_MAX 1024
	#elif defined(__sgi)
		#define IOV_MAX 512 
	#else 
		#define IOV_MAX UIO_MAXIOV
	#endif
#endif

	*iov_count = MIN(MIN(queue->count, IOV_MAX), *iov_count);

	for (l = queue->head, i = 0; i < *iov_count; l = l->link, i++)
	{
		writebuf = (buffer*)l->item;
		len = writebuf->len - writebuf->pos;
		dropbear_assert(len > 0);
		TRACE2(("write_packet writev #%d len %d/%d", i,
				len, writebuf->len))
		iov[i].iov_base = buf_getptr(writebuf, len);
		iov[i].iov_len = len;
	}
}

void packet_queue_consume(struct Queue *queue, ssize_t written) {
	buffer *writebuf;
	int len;
	while (written > 0) {
		writebuf = (buffer*)examine(queue);
		len = writebuf->len - writebuf->pos;
		if (len > written) {
			/* partial buffer write */
			buf_incrpos(writebuf, written);
			written = 0;
		} else {
			written -= len;
			dequeue(queue);
			buf_free(writebuf);
		}
	}
}

void set_sock_nodelay(int sock) {
	int val;

	/* disable nagle */
	val = 1;
	setsockopt(sock, IPPROTO_TCP, TCP_NODELAY, (void*)&val, sizeof(val));
}

#if DROPBEAR_SERVER_TCP_FAST_OPEN
void set_listen_fast_open(int sock) {
	int qlen = MAX(MAX_UNAUTH_PER_IP, 5);
	if (setsockopt(sock, SOL_TCP, TCP_FASTOPEN, &qlen, sizeof(qlen)) != 0) {
		TRACE(("set_listen_fast_open failed for socket %d: %s", sock, strerror(errno)))
	}
}

#endif

void set_sock_priority(int sock, enum dropbear_prio prio) {

	int rc;
	int val;

#if DROPBEAR_FUZZ
	if (fuzz.fuzzing) {
		TRACE(("fuzzing skips set_sock_prio"))
		return;
	}
#endif
	/* Don't log ENOTSOCK errors so that this can harmlessly be called
	 * on a client '-J' proxy pipe */

#ifdef IP_TOS
	/* Set the DSCP field for outbound IP packet priority.
	rfc4594 has some guidance to meanings.

	We set AF21 as "Low-Latency" class for interactive (tty session,
	also handshake/setup packets). Other traffic is left at the default.

	OpenSSH at present uses AF21/CS1, rationale
	https://cvsweb.openbsd.org/src/usr.bin/ssh/readconf.c#rev1.284

	Old Dropbear/OpenSSH and Debian/Ubuntu OpenSSH (at Jan 2022) use
	IPTOS_LOWDELAY/IPTOS_THROUGHPUT

	DSCP constants are from Linux headers, applicable to other platforms
	such as macos.
	*/
	if (prio == DROPBEAR_PRIO_LOWDELAY) {
		val = 0x48; /* IPTOS_DSCP_AF21 */
	} else {
		val = 0; /* default */
	}
#if defined(IPPROTO_IPV6) && defined(IPV6_TCLASS)
	rc = setsockopt(sock, IPPROTO_IPV6, IPV6_TCLASS, (void*)&val, sizeof(val));
	if (rc < 0 && errno != ENOTSOCK) {
		TRACE(("Couldn't set IPV6_TCLASS (%s)", strerror(errno)));
	}
#endif
	rc = setsockopt(sock, IPPROTO_IP, IP_TOS, (void*)&val, sizeof(val));
	if (rc < 0 && errno != ENOTSOCK) {
		TRACE(("Couldn't set IP_TOS (%s)", strerror(errno)));
	}
#endif /* IP_TOS */

#ifdef HAVE_LINUX_PKT_SCHED_H
	/* Set scheduling priority within the local Linux network stack */
	if (prio == DROPBEAR_PRIO_LOWDELAY) {
		val = TC_PRIO_INTERACTIVE;
	} else {
		val = 0;
	}
	/* linux specific, sets QoS class. see tc-prio(8) */
	rc = setsockopt(sock, SOL_SOCKET, SO_PRIORITY, (void*) &val, sizeof(val));
	if (rc < 0 && errno != ENOTSOCK) {
		TRACE(("Couldn't set SO_PRIORITY (%s)", strerror(errno)))
    }
#endif

}

/* from openssh/canohost.c avoid premature-optimization */
int get_sock_port(int sock) {
	struct sockaddr_storage from;
	socklen_t fromlen;
	char strport[NI_MAXSERV];
	int r;

	/* Get IP address of client. */
	fromlen = sizeof(from);
	memset(&from, 0, sizeof(from));
	if (getsockname(sock, (struct sockaddr *)&from, &fromlen) < 0) {
		TRACE(("getsockname failed: %d", errno))
		return 0;
	}

	/* Work around Linux IPv6 weirdness */
	if (from.ss_family == AF_INET6)
		fromlen = sizeof(struct sockaddr_in6);

	/* Non-inet sockets don't have a port number. */
	if (from.ss_family != AF_INET && from.ss_family != AF_INET6)
		return 0;

	/* Return port number. */
	if ((r = getnameinfo((struct sockaddr *)&from, fromlen, NULL, 0,
	    strport, sizeof(strport), NI_NUMERICSERV)) != 0) {
		TRACE(("netio.c/get_sock_port/getnameinfo NI_NUMERICSERV failed: %d", r))
	}
	return atoi(strport);
}

/* Listen on address:port. 
 * Special cases are address of "" listening on everything,
 * and address of NULL listening on localhost only.
 * Returns the number of sockets bound on success, or -1 on failure. On
 * failure, if errstring wasn't NULL, it'll be a newly malloced error
 * string.*/
int dropbear_listen(const char* address, const char* port,
		int *socks, unsigned int sockcount, char **errstring, int *maxfd) {

	struct addrinfo hints, *res = NULL, *res0 = NULL;
	int err;
	unsigned int nsock;
	struct linger linger;
	int val;
	int sock;
	uint16_t *allocated_lport_p = NULL;
	int allocated_lport = 0;
	
	TRACE(("enter dropbear_listen"))

#if DROPBEAR_FUZZ
	if (fuzz.fuzzing) {
		return fuzz_dropbear_listen(address, port, socks, sockcount, errstring, maxfd);
	}
#endif
	
	memset(&hints, 0, sizeof(hints));
	hints.ai_family = AF_UNSPEC; /* TODO: let them flag v4 only etc */
	hints.ai_socktype = SOCK_STREAM;

	/* for calling getaddrinfo:
	 address == NULL and !AI_PASSIVE: local loopback
	 address == NULL and AI_PASSIVE: all interfaces
	 address != NULL: whatever the address says */
	if (!address) {
		TRACE(("dropbear_listen: local loopback"))
	} else {
		if (address[0] == '\0') {
			TRACE(("dropbear_listen: all interfaces"))
			address = NULL;
		}
		hints.ai_flags = AI_PASSIVE;
	}
	err = getaddrinfo(address, port, &hints, &res0);

	if (err) {
		if (errstring != NULL && *errstring == NULL) {
			int len;
			len = 20 + strlen(gai_strerror(err));
			*errstring = (char*)m_malloc(len);
			snprintf(*errstring, len, "Error resolving: %s", gai_strerror(err));
		}
		if (res0) {
			freeaddrinfo(res0);
			res0 = NULL;
		}
		TRACE(("leave dropbear_listen: failed resolving"))
		return -1;
	}

	/* When listening on server-assigned-port 0
	 * the assigned ports may differ for address families (v4/v6)
	 * causing problems for tcpip-forward.
	 * Caller can do a get_socket_address to discover assigned-port
	 * hence, use same port for all address families */
	allocated_lport = 0;
	nsock = 0;
	for (res = res0; res != NULL && nsock < sockcount;
			res = res->ai_next) {
		if (allocated_lport > 0) {
			if (AF_INET == res->ai_family) {
				allocated_lport_p = &((struct sockaddr_in *)res->ai_addr)->sin_port;
			} else if (AF_INET6 == res->ai_family) {
				allocated_lport_p = &((struct sockaddr_in6 *)res->ai_addr)->sin6_port;
			}
			*allocated_lport_p = htons(allocated_lport);
		}

		/* Get a socket */
		socks[nsock] = socket(res->ai_family, res->ai_socktype, res->ai_protocol);
		sock = socks[nsock]; /* For clarity */
		if (sock < 0) {
			err = errno;
			TRACE(("socket() failed"))
			continue;
		}

		/* Various useful socket options */
		val = 1;
		/* set to reuse, quick timeout */
		setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, (void*) &val, sizeof(val));
		linger.l_onoff = 1;
		linger.l_linger = 5;
		setsockopt(sock, SOL_SOCKET, SO_LINGER, (void*)&linger, sizeof(linger));

#if defined(IPPROTO_IPV6) && defined(IPV6_V6ONLY)
		if (res->ai_family == AF_INET6) {
			int on = 1;
			if (setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, 
						&on, sizeof(on)) == -1) {
				dropbear_log(LOG_WARNING, "Couldn't set IPV6_V6ONLY");
			}
		}
#endif
		set_sock_nodelay(sock);

		if (bind(sock, res->ai_addr, res->ai_addrlen) < 0) {
			err = errno;
			close(sock);
			TRACE(("bind(%s) failed", port))
			continue;
		}

		if (listen(sock, DROPBEAR_LISTEN_BACKLOG) < 0) {
			err = errno;
			close(sock);
			TRACE(("listen() failed"))
			continue;
		}

		if (0 == allocated_lport) {
			allocated_lport = get_sock_port(sock);
		}

		*maxfd = MAX(*maxfd, sock);
		nsock++;
	}

	if (res0) {
		freeaddrinfo(res0);
		res0 = NULL;
	}

	if (nsock == 0) {
		if (errstring != NULL && *errstring == NULL) {
			int len;
			len = 20 + strlen(strerror(err));
			*errstring = (char*)m_malloc(len);
			snprintf(*errstring, len, "Error listening: %s", strerror(err));
		}
		TRACE(("leave dropbear_listen: failure, %s", strerror(err)))
		return -1;
	}

	TRACE(("leave dropbear_listen: success, %d socks bound", nsock))
	return nsock;
}

void get_socket_address(int fd, char **local_host, char **local_port,
						char **remote_host, char **remote_port, int host_lookup)
{
	struct sockaddr_storage addr;
	socklen_t addrlen;

#if DROPBEAR_FUZZ
	if (fuzz.fuzzing) {
		fuzz_get_socket_address(fd, local_host, local_port, remote_host, remote_port, host_lookup);
		return;
	}
#endif
	
	if (local_host || local_port) {
		addrlen = sizeof(addr);
		if (getsockname(fd, (struct sockaddr*)&addr, &addrlen) < 0) {
			dropbear_exit("Failed socket address: %s", strerror(errno));
		}
		getaddrstring(&addr, local_host, local_port, host_lookup);		
	}
	if (remote_host || remote_port) {
		addrlen = sizeof(addr);
		if (getpeername(fd, (struct sockaddr*)&addr, &addrlen) < 0) {
			dropbear_exit("Failed socket address: %s", strerror(errno));
		}
		getaddrstring(&addr, remote_host, remote_port, host_lookup);		
	}
}

/* Return a string representation of the socket address passed. The return
 * value is allocated with malloc() */
void getaddrstring(struct sockaddr_storage* addr, 
			char **ret_host, char **ret_port,
			int host_lookup) {

	char host[NI_MAXHOST+1], serv[NI_MAXSERV+1];
	unsigned int len;
	int ret;
	
	int flags = NI_NUMERICSERV | NI_NUMERICHOST;

#if !DO_HOST_LOOKUP
	host_lookup = 0;
#endif
	
	if (host_lookup) {
		flags = NI_NUMERICSERV;
	}

	len = sizeof(struct sockaddr_storage);
	/* Some platforms such as Solaris 8 require that len is the length
	 * of the specific structure. Some older linux systems (glibc 2.1.3
	 * such as debian potato) have sockaddr_storage.__ss_family instead
	 * but we'll ignore them */
#ifdef HAVE_STRUCT_SOCKADDR_STORAGE_SS_FAMILY
	if (addr->ss_family == AF_INET) {
		len = sizeof(struct sockaddr_in);
	}
#ifdef AF_INET6
	if (addr->ss_family == AF_INET6) {
		len = sizeof(struct sockaddr_in6);
	}
#endif
#endif

	ret = getnameinfo((struct sockaddr*)addr, len, host, sizeof(host)-1, 
			serv, sizeof(serv)-1, flags);

	if (ret != 0) {
		if (host_lookup) {
			/* On some systems (Darwin does it) we get EINTR from getnameinfo
			 * somehow. Eew. So we'll just return the IP, since that doesn't seem
			 * to exhibit that behaviour. */
			getaddrstring(addr, ret_host, ret_port, 0);
			return;
		} else {
			/* if we can't do a numeric lookup, something's gone terribly wrong */
			dropbear_exit("Failed lookup: %s", gai_strerror(ret));
		}
	}

	if (ret_host) {
		*ret_host = m_strdup(host);
	}
	if (ret_port) {
		*ret_port = m_strdup(serv);
	}
}