src/proto_quic.c

/*
 * AF_INET/AF_INET6 QUIC protocol layer.
 *
 * Copyright 2020 Frédéric Lécaille <flecaille@haproxy.com>
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * as published by the Free Software Foundation; either version
 * 2 of the License, or (at your option) any later version.
 *
 */

#include <ctype.h>
#include <errno.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>

#include <sys/param.h>
#include <sys/socket.h>
#include <sys/types.h>

#include <netinet/udp.h>
#include <netinet/in.h>

#include <haproxy/api.h>
#include <haproxy/arg.h>
#include <haproxy/cbuf.h>
#include <haproxy/connection.h>
#include <haproxy/errors.h>
#include <haproxy/fd.h>
#include <haproxy/global.h>
#include <haproxy/list.h>
#include <haproxy/listener.h>
#include <haproxy/log.h>
#include <haproxy/namespace.h>
#include <haproxy/port_range.h>
#include <haproxy/protocol.h>
#include <haproxy/proto_quic.h>
#include <haproxy/proto_udp.h>
#include <haproxy/proxy-t.h>
#include <haproxy/sock.h>
#include <haproxy/quic_sock.h>
#include <haproxy/sock_inet.h>
#include <haproxy/tools.h>
#include <haproxy/xprt_quic-t.h>


static void quic_add_listener(struct protocol *proto, struct listener *listener);
static int quic_bind_listener(struct listener *listener, char *errmsg, int errlen);
static int quic_connect_server(struct connection *conn, int flags);
static void quic_enable_listener(struct listener *listener);
static void quic_disable_listener(struct listener *listener);

/* Note: must not be declared <const> as its list will be overwritten */
struct protocol proto_quic4 = {
	.name           = "quic4",

	/* connection layer */
	.ctrl_type      = SOCK_STREAM,
	.listen         = quic_bind_listener,
	.enable         = quic_enable_listener,
	.disable        = quic_disable_listener,
	.add            = quic_add_listener,
	.unbind         = default_unbind_listener,
	.suspend        = default_suspend_listener,
	.resume         = default_resume_listener,
	.accept_conn    = quic_sock_accept_conn,
	.connect        = quic_connect_server,

	/* binding layer */
	.rx_suspend     = udp_suspend_receiver,
	.rx_resume      = udp_resume_receiver,

	/* address family */
	.fam            = &proto_fam_inet4,

	/* socket layer */
	.proto_type     = PROTO_TYPE_DGRAM,
	.sock_type      = SOCK_DGRAM,
	.sock_prot      = IPPROTO_UDP,
	.rx_enable      = sock_enable,
	.rx_disable     = sock_disable,
	.rx_unbind      = sock_unbind,
	.rx_listening   = quic_sock_accepting_conn,
	.default_iocb   = quic_sock_fd_iocb,
	.receivers      = LIST_HEAD_INIT(proto_quic4.receivers),
	.nb_receivers   = 0,
};

INITCALL1(STG_REGISTER, protocol_register, &proto_quic4);

/* Note: must not be declared <const> as its list will be overwritten */
struct protocol proto_quic6 = {
	.name            = "quic6",

	/* connection layer */
	.ctrl_type      = SOCK_STREAM,
	.listen         = quic_bind_listener,
	.enable         = quic_enable_listener,
	.disable        = quic_disable_listener,
	.add            = quic_add_listener,
	.unbind         = default_unbind_listener,
	.suspend        = default_suspend_listener,
	.resume         = default_resume_listener,
	.accept_conn    = quic_sock_accept_conn,
	.connect        = quic_connect_server,

	/* binding layer */
	.rx_suspend     = udp_suspend_receiver,
	.rx_resume      = udp_resume_receiver,

	/* address family */
	.fam            = &proto_fam_inet6,

	/* socket layer */
	.proto_type     = PROTO_TYPE_DGRAM,
	.sock_type      = SOCK_DGRAM,
	.sock_prot      = IPPROTO_UDP,
	.rx_enable      = sock_enable,
	.rx_disable     = sock_disable,
	.rx_unbind      = sock_unbind,
	.rx_listening   = quic_sock_accepting_conn,
	.default_iocb   = quic_sock_fd_iocb,
	.receivers      = LIST_HEAD_INIT(proto_quic6.receivers),
	.nb_receivers   = 0,
};

INITCALL1(STG_REGISTER, protocol_register, &proto_quic6);

/* Binds ipv4/ipv6 address <local> to socket <fd>, unless <flags> is set, in which
 * case we try to bind <remote>. <flags> is a 2-bit field consisting of :
 *  - 0 : ignore remote address (may even be a NULL pointer)
 *  - 1 : use provided address
 *  - 2 : use provided port
 *  - 3 : use both
 *
 * The function supports multiple foreign binding methods :
 *   - linux_tproxy: we directly bind to the foreign address
 * The second one can be used as a fallback for the first one.
 * This function returns 0 when everything's OK, 1 if it could not bind, to the
 * local address, 2 if it could not bind to the foreign address.
 */
int quic_bind_socket(int fd, int flags, struct sockaddr_storage *local, struct sockaddr_storage *remote)
{
	struct sockaddr_storage bind_addr;
	int foreign_ok = 0;
	int ret;
	static THREAD_LOCAL int ip_transp_working = 1;
	static THREAD_LOCAL int ip6_transp_working = 1;

	switch (local->ss_family) {
	case AF_INET:
		if (flags && ip_transp_working) {
			/* This deserves some explanation. Some platforms will support
			 * multiple combinations of certain methods, so we try the
			 * supported ones until one succeeds.
			 */
			if (sock_inet4_make_foreign(fd))
				foreign_ok = 1;
			else
				ip_transp_working = 0;
		}
		break;
	case AF_INET6:
		if (flags && ip6_transp_working) {
			if (sock_inet6_make_foreign(fd))
				foreign_ok = 1;
			else
				ip6_transp_working = 0;
		}
		break;
	}

	if (flags) {
		memset(&bind_addr, 0, sizeof(bind_addr));
		bind_addr.ss_family = remote->ss_family;
		switch (remote->ss_family) {
		case AF_INET:
			if (flags & 1)
				((struct sockaddr_in *)&bind_addr)->sin_addr = ((struct sockaddr_in *)remote)->sin_addr;
			if (flags & 2)
				((struct sockaddr_in *)&bind_addr)->sin_port = ((struct sockaddr_in *)remote)->sin_port;
			break;
		case AF_INET6:
			if (flags & 1)
				((struct sockaddr_in6 *)&bind_addr)->sin6_addr = ((struct sockaddr_in6 *)remote)->sin6_addr;
			if (flags & 2)
				((struct sockaddr_in6 *)&bind_addr)->sin6_port = ((struct sockaddr_in6 *)remote)->sin6_port;
			break;
		default:
			/* we don't want to try to bind to an unknown address family */
			foreign_ok = 0;
		}
	}

	setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &one, sizeof(one));
	if (foreign_ok) {
		if (is_inet_addr(&bind_addr)) {
			ret = bind(fd, (struct sockaddr *)&bind_addr, get_addr_len(&bind_addr));
			if (ret < 0)
				return 2;
		}
	}
	else {
		if (is_inet_addr(local)) {
			ret = bind(fd, (struct sockaddr *)local, get_addr_len(local));
			if (ret < 0)
				return 1;
		}
	}

	if (!flags)
		return 0;

	if (!foreign_ok)
		/* we could not bind to a foreign address */
		return 2;

	return 0;
}

/*
 * This function initiates a QUIC connection establishment to the target assigned
 * to connection <conn> using (si->{target,dst}). A source address may be
 * pointed to by conn->src in case of transparent proxying. Normal source
 * bind addresses are still determined locally (due to the possible need of a
 * source port). conn->target may point either to a valid server or to a backend,
 * depending on conn->target. Only OBJ_TYPE_PROXY and OBJ_TYPE_SERVER are
 * supported. The <data> parameter is a boolean indicating whether there are data
 * waiting for being sent or not, in order to adjust data write polling and on
 * some platforms, the ability to avoid an empty initial ACK. The <flags> argument
 * is not used.
 *
 * Note that a pending send_proxy message accounts for data.
 *
 * It can return one of :
 *  - SF_ERR_NONE if everything's OK
 *  - SF_ERR_SRVTO if there are no more servers
 *  - SF_ERR_SRVCL if the connection was refused by the server
 *  - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
 *  - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
 *  - SF_ERR_INTERNAL for any other purely internal errors
 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
 *
 * The connection's fd is inserted only when SF_ERR_NONE is returned, otherwise
 * it's invalid and the caller has nothing to do.
 */

int quic_connect_server(struct connection *conn, int flags)
{
	int fd;
	struct server *srv;
	struct proxy *be;
	struct conn_src *src;
	struct sockaddr_storage *addr;

	conn->flags |= CO_FL_WAIT_L4_CONN; /* connection in progress */

	switch (obj_type(conn->target)) {
	case OBJ_TYPE_PROXY:
		be = objt_proxy(conn->target);
		srv = NULL;
		break;
	case OBJ_TYPE_SERVER:
		srv = objt_server(conn->target);
		be = srv->proxy;
		break;
	default:
		conn->flags |= CO_FL_ERROR;
		return SF_ERR_INTERNAL;
	}

	if (!conn->dst) {
		conn->flags |= CO_FL_ERROR;
		return SF_ERR_INTERNAL;
	}

	fd = conn->handle.fd = sock_create_server_socket(conn);

	if (fd == -1) {
		qfprintf(stderr, "Cannot get a server socket.\n");

		if (errno == ENFILE) {
			conn->err_code = CO_ER_SYS_FDLIM;
			send_log(be, LOG_EMERG,
				 "Proxy %s reached system FD limit (maxsock=%d). Please check system tunables.\n",
				 be->id, global.maxsock);
		}
		else if (errno == EMFILE) {
			conn->err_code = CO_ER_PROC_FDLIM;
			send_log(be, LOG_EMERG,
				 "Proxy %s reached process FD limit (maxsock=%d). Please check 'ulimit-n' and restart.\n",
				 be->id, global.maxsock);
		}
		else if (errno == ENOBUFS || errno == ENOMEM) {
			conn->err_code = CO_ER_SYS_MEMLIM;
			send_log(be, LOG_EMERG,
				 "Proxy %s reached system memory limit (maxsock=%d). Please check system tunables.\n",
				 be->id, global.maxsock);
		}
		else if (errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
			conn->err_code = CO_ER_NOPROTO;
		}
		else
			conn->err_code = CO_ER_SOCK_ERR;

		/* this is a resource error */
		conn->flags |= CO_FL_ERROR;
		return SF_ERR_RESOURCE;
	}

	if (fd >= global.maxsock) {
		/* do not log anything there, it's a normal condition when this option
		 * is used to serialize connections to a server !
		 */
		ha_alert("socket(): not enough free sockets. Raise -n argument. Giving up.\n");
		close(fd);
		conn->err_code = CO_ER_CONF_FDLIM;
		conn->flags |= CO_FL_ERROR;
		return SF_ERR_PRXCOND; /* it is a configuration limit */
	}

	if ((fcntl(fd, F_SETFL, O_NONBLOCK)==-1)) {
		qfprintf(stderr,"Cannot set client socket to non blocking mode.\n");
		close(fd);
		conn->err_code = CO_ER_SOCK_ERR;
		conn->flags |= CO_FL_ERROR;
		return SF_ERR_INTERNAL;
	}

	if (master == 1 && (fcntl(fd, F_SETFD, FD_CLOEXEC) == -1)) {
		ha_alert("Cannot set CLOEXEC on client socket.\n");
		close(fd);
		conn->err_code = CO_ER_SOCK_ERR;
		conn->flags |= CO_FL_ERROR;
		return SF_ERR_INTERNAL;
	}

	/* allow specific binding :
	 * - server-specific at first
	 * - proxy-specific next
	 */
	if (srv && srv->conn_src.opts & CO_SRC_BIND)
		src = &srv->conn_src;
	else if (be->conn_src.opts & CO_SRC_BIND)
		src = &be->conn_src;
	else
		src = NULL;

	if (src) {
		int ret, flags = 0;

		if (conn->src && is_inet_addr(conn->src)) {
			switch (src->opts & CO_SRC_TPROXY_MASK) {
			case CO_SRC_TPROXY_CLI:
				conn_set_private(conn);
				/* fall through */
			case CO_SRC_TPROXY_ADDR:
				flags = 3;
				break;
			case CO_SRC_TPROXY_CIP:
			case CO_SRC_TPROXY_DYN:
				conn_set_private(conn);
				flags = 1;
				break;
			}
		}

#ifdef SO_BINDTODEVICE
		/* Note: this might fail if not CAP_NET_RAW */
		if (src->iface_name)
			setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, src->iface_name, src->iface_len + 1);
#endif

		if (src->sport_range) {
			int attempts = 10; /* should be more than enough to find a spare port */
			struct sockaddr_storage sa;

			ret = 1;
			memcpy(&sa, &src->source_addr, sizeof(sa));

			do {
				/* note: in case of retry, we may have to release a previously
				 * allocated port, hence this loop's construct.
				 */
				port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
				fdinfo[fd].port_range = NULL;

				if (!attempts)
					break;
				attempts--;

				fdinfo[fd].local_port = port_range_alloc_port(src->sport_range);
				if (!fdinfo[fd].local_port) {
					conn->err_code = CO_ER_PORT_RANGE;
					break;
				}

				fdinfo[fd].port_range = src->sport_range;
				set_host_port(&sa, fdinfo[fd].local_port);

				ret = quic_bind_socket(fd, flags, &sa, conn->src);
				if (ret != 0)
					conn->err_code = CO_ER_CANT_BIND;
			} while (ret != 0); /* binding NOK */
		}
		else {
#ifdef IP_BIND_ADDRESS_NO_PORT
			static THREAD_LOCAL int bind_address_no_port = 1;
			setsockopt(fd, IPPROTO_IP, IP_BIND_ADDRESS_NO_PORT, (const void *) &bind_address_no_port, sizeof(int));
#endif
			ret = quic_bind_socket(fd, flags, &src->source_addr, conn->src);
			if (ret != 0)
				conn->err_code = CO_ER_CANT_BIND;
		}

		if (unlikely(ret != 0)) {
			port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
			fdinfo[fd].port_range = NULL;
			close(fd);

			if (ret == 1) {
				ha_alert("Cannot bind to source address before connect() for backend %s. Aborting.\n",
					 be->id);
				send_log(be, LOG_EMERG,
					 "Cannot bind to source address before connect() for backend %s.\n",
					 be->id);
			} else {
				ha_alert("Cannot bind to tproxy source address before connect() for backend %s. Aborting.\n",
					 be->id);
				send_log(be, LOG_EMERG,
					 "Cannot bind to tproxy source address before connect() for backend %s.\n",
					 be->id);
			}
			conn->flags |= CO_FL_ERROR;
			return SF_ERR_RESOURCE;
		}
	}

	if (global.tune.server_sndbuf)
                setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &global.tune.server_sndbuf, sizeof(global.tune.server_sndbuf));

	if (global.tune.server_rcvbuf)
                setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &global.tune.server_rcvbuf, sizeof(global.tune.server_rcvbuf));

	addr = (conn->flags & CO_FL_SOCKS4) ? &srv->socks4_addr : conn->dst;
	if (connect(fd, (const struct sockaddr *)addr, get_addr_len(addr)) == -1) {
		if (errno == EINPROGRESS || errno == EALREADY) {
			/* common case, let's wait for connect status */
			conn->flags |= CO_FL_WAIT_L4_CONN;
		}
		else if (errno == EISCONN) {
			/* should normally not happen but if so, indicates that it's OK */
			conn->flags &= ~CO_FL_WAIT_L4_CONN;
		}
		else if (errno == EAGAIN || errno == EADDRINUSE || errno == EADDRNOTAVAIL) {
			char *msg;
			if (errno == EAGAIN || errno == EADDRNOTAVAIL) {
				msg = "no free ports";
				conn->err_code = CO_ER_FREE_PORTS;
			}
			else {
				msg = "local address already in use";
				conn->err_code = CO_ER_ADDR_INUSE;
			}

			qfprintf(stderr,"Connect() failed for backend %s: %s.\n", be->id, msg);
			port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
			fdinfo[fd].port_range = NULL;
			close(fd);
			send_log(be, LOG_ERR, "Connect() failed for backend %s: %s.\n", be->id, msg);
			conn->flags |= CO_FL_ERROR;
			return SF_ERR_RESOURCE;
		} else if (errno == ETIMEDOUT) {
			//qfprintf(stderr,"Connect(): ETIMEDOUT");
			port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
			fdinfo[fd].port_range = NULL;
			close(fd);
			conn->err_code = CO_ER_SOCK_ERR;
			conn->flags |= CO_FL_ERROR;
			return SF_ERR_SRVTO;
		} else {
			// (errno == ECONNREFUSED || errno == ENETUNREACH || errno == EACCES || errno == EPERM)
			//qfprintf(stderr,"Connect(): %d", errno);
			port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
			fdinfo[fd].port_range = NULL;
			close(fd);
			conn->err_code = CO_ER_SOCK_ERR;
			conn->flags |= CO_FL_ERROR;
			return SF_ERR_SRVCL;
		}
	}
	else {
		/* connect() == 0, this is great! */
		conn->flags &= ~CO_FL_WAIT_L4_CONN;
	}

	conn->flags |= CO_FL_ADDR_TO_SET;

	conn_ctrl_init(conn);       /* registers the FD */
	HA_ATOMIC_OR(&fdtab[fd].state, FD_LINGER_RISK);  /* close hard if needed */

	if (conn->flags & CO_FL_WAIT_L4_CONN) {
		fd_want_send(fd);
		fd_cant_send(fd);
		fd_cant_recv(fd);
	}

	return SF_ERR_NONE;  /* connection is OK */
}

/* Add listener <listener> to protocol <proto>. Technically speaking we just
 * initialize a few entries which should be doable during quic_bind_listener().
 * The end of the initialization goes on with the default function.
 */
static void quic_add_listener(struct protocol *proto, struct listener *listener)
{
	MT_LIST_INIT(&listener->rx.pkts);
	listener->rx.odcids = EB_ROOT_UNIQUE;
	listener->rx.cids = EB_ROOT_UNIQUE;
	HA_RWLOCK_INIT(&listener->rx.cids_lock);
	default_add_listener(proto, listener);
}

/* Allocate the TX ring buffers for <l> listener.
 * Return 1 if succeeded, 0 if not.
 */
static int quic_alloc_tx_rings_listener(struct listener *l)
{
	struct qring *qr;
	int i;

	l->rx.tx_qrings = calloc(global.nbthread, sizeof *l->rx.tx_qrings);
	if (!l->rx.tx_qrings)
		return 0;

	MT_LIST_INIT(&l->rx.tx_qring_list);
	for (i = 0; i < global.nbthread; i++) {
		unsigned char *buf;
		struct qring *qr = &l->rx.tx_qrings[i];

		buf = pool_alloc(pool_head_quic_tx_ring);
		if (!buf)
			goto err;

		qr->cbuf = cbuf_new(buf, QUIC_TX_RING_BUFSZ);
		if (!qr->cbuf) {
			pool_free(pool_head_quic_tx_ring, buf);
			goto err;
		}

		MT_LIST_APPEND(&l->rx.tx_qring_list, &qr->mt_list);
	}

	return 1;

 err:
	while ((qr = MT_LIST_POP(&l->rx.tx_qring_list, typeof(qr), mt_list))) {
		pool_free(pool_head_quic_tx_ring, qr->cbuf->buf);
		cbuf_free(qr->cbuf);
	}
	free(l->rx.tx_qrings);
	return 0;
}

/* Allocate the RX buffers for <l> listener.
 * Return 1 if succeeded, 0 if not.
 */
static int quic_alloc_rxbufs_listener(struct listener *l)
{
	int i;
	struct rxbuf *rxbuf;

	l->rx.rxbufs = calloc(global.nbthread, sizeof *l->rx.rxbufs);
	if (!l->rx.rxbufs)
		return 0;

	MT_LIST_INIT(&l->rx.rxbuf_list);
	for (i = 0; i < global.nbthread; i++) {
		char *buf;

		rxbuf = &l->rx.rxbufs[i];
		buf = pool_alloc(pool_head_quic_rxbuf);
		if (!buf)
			goto err;

		rxbuf->buf = b_make(buf, QUIC_RX_BUFSZ, 0, 0);
		MT_LIST_APPEND(&l->rx.rxbuf_list, &rxbuf->mt_list);
	}

	return 1;

 err:
	while ((rxbuf = MT_LIST_POP(&l->rx.rxbuf_list, typeof(rxbuf), mt_list)))
		pool_free(pool_head_quic_rxbuf, rxbuf->buf.area);
	free(l->rx.rxbufs);
	return 0;
}

/* This function tries to bind a QUIC4/6 listener. It may return a warning or
 * an error message in <errmsg> if the message is at most <errlen> bytes long
 * (including '\0'). Note that <errmsg> may be NULL if <errlen> is also zero.
 * The return value is composed from ERR_ABORT, ERR_WARN,
 * ERR_ALERT, ERR_RETRYABLE and ERR_FATAL. ERR_NONE indicates that everything
 * was alright and that no message was returned. ERR_RETRYABLE means that an
 * error occurred but that it may vanish after a retry (eg: port in use), and
 * ERR_FATAL indicates a non-fixable error. ERR_WARN and ERR_ALERT do not alter
 * the meaning of the error, but just indicate that a message is present which
 * should be displayed with the respective level. Last, ERR_ABORT indicates
 * that it's pointless to try to start other listeners. No error message is
 * returned if errlen is NULL.
 */
static int quic_bind_listener(struct listener *listener, char *errmsg, int errlen)
{
	int err = ERR_NONE;
	char *msg = NULL;

	/* ensure we never return garbage */
	if (errlen)
		*errmsg = 0;

	if (listener->state != LI_ASSIGNED)
		return ERR_NONE; /* already bound */

	if (!(listener->rx.flags & RX_F_BOUND)) {
		msg = "receiving socket not bound";
		goto udp_return;
	}

	if (!quic_alloc_tx_rings_listener(listener) ||
	    !quic_alloc_rxbufs_listener(listener)) {
		msg = "could not initialize tx/rx rings";
		err |= ERR_WARN;
		goto udp_return;
	}

	listener_set_state(listener, LI_LISTEN);

 udp_return:
	if (msg && errlen) {
		char pn[INET6_ADDRSTRLEN];

		addr_to_str(&listener->rx.addr, pn, sizeof(pn));
		snprintf(errmsg, errlen, "%s for [%s:%d]", msg, pn, get_host_port(&listener->rx.addr));
	}
	return err;
}

/* Enable receipt of incoming connections for listener <l>. The receiver must
 * still be valid. Does nothing in early boot (needs fd_updt).
 */
static void quic_enable_listener(struct listener *l)
{
	/* FIXME: The following statements are incorrect. This
	 * is the responsibility of the QUIC xprt to stop accepting new
	 * connections.
	 */
	if (fd_updt)
		fd_want_recv(l->rx.fd);
}

/* Disable receipt of incoming connections for listener <l>. The receiver must
 * still be valid. Does nothing in early boot (needs fd_updt).
 */
static void quic_disable_listener(struct listener *l)
{
	/* FIXME: The following statements are incorrect. This
	 * is the responsibility of the QUIC xprt to start accepting new
	 * connections again.
	 */
	if (fd_updt)
		fd_stop_recv(l->rx.fd);
}

/*
 * Local variables:
 *  c-indent-level: 8
 *  c-basic-offset: 8
 * End:
 */