blob: 21b17715c9a0c6bd7dfa160a3c303e28ddf355b3 [file] [log] [blame]
Willy Tarreaue6b98942007-10-29 01:09:36 +01001/*
2 * AF_INET/AF_INET6 SOCK_STREAM protocol layer (tcp)
3 *
Willy Tarreaud4c33c82013-01-07 21:59:07 +01004 * Copyright 2000-2013 Willy Tarreau <w@1wt.eu>
Willy Tarreaue6b98942007-10-29 01:09:36 +01005 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 */
12
13#include <ctype.h>
14#include <errno.h>
15#include <fcntl.h>
16#include <stdio.h>
17#include <stdlib.h>
18#include <string.h>
19#include <time.h>
20
21#include <sys/param.h>
22#include <sys/socket.h>
23#include <sys/stat.h>
24#include <sys/types.h>
25#include <sys/un.h>
26
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040027#include <netinet/tcp.h>
28
Willy Tarreaub6866442008-07-14 23:54:42 +020029#include <common/cfgparse.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010030#include <common/compat.h>
31#include <common/config.h>
32#include <common/debug.h>
33#include <common/errors.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010034#include <common/mini-clist.h>
35#include <common/standard.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010036
Willy Tarreaue6b98942007-10-29 01:09:36 +010037#include <types/global.h>
Willy Tarreau9650f372009-08-16 14:02:45 +020038#include <types/server.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010039
40#include <proto/acl.h>
Willy Tarreau9fcb9842012-04-20 14:45:49 +020041#include <proto/arg.h>
Willy Tarreauc7e42382012-08-24 19:22:53 +020042#include <proto/channel.h>
Willy Tarreaud2274c62012-07-06 14:29:45 +020043#include <proto/connection.h>
Willy Tarreaudd2f85e2012-09-02 22:34:23 +020044#include <proto/fd.h>
Willy Tarreaud1d54542012-09-12 22:58:11 +020045#include <proto/listener.h>
Willy Tarreau9650f372009-08-16 14:02:45 +020046#include <proto/log.h>
47#include <proto/port_range.h>
Willy Tarreaud1d54542012-09-12 22:58:11 +020048#include <proto/protocol.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010049#include <proto/proto_tcp.h>
Willy Tarreaub6866442008-07-14 23:54:42 +020050#include <proto/proxy.h>
Willy Tarreaucd3b0942012-04-27 21:52:18 +020051#include <proto/sample.h>
Willy Tarreau9ba2dcc2010-06-14 21:04:55 +020052#include <proto/session.h>
Willy Tarreaua975b8f2010-06-05 19:13:27 +020053#include <proto/stick_table.h>
Willy Tarreaucc1e04b2013-09-11 23:20:29 +020054#include <proto/stream_interface.h>
Willy Tarreaua975b8f2010-06-05 19:13:27 +020055#include <proto/task.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010056
Willy Tarreaue8c66af2008-01-13 18:40:14 +010057#ifdef CONFIG_HAP_CTTPROXY
58#include <import/ip_tproxy.h>
59#endif
60
Emeric Bruncf20bf12010-10-22 16:06:11 +020061static int tcp_bind_listeners(struct protocol *proto, char *errmsg, int errlen);
62static int tcp_bind_listener(struct listener *listener, char *errmsg, int errlen);
Willy Tarreaue6b98942007-10-29 01:09:36 +010063
64/* Note: must not be declared <const> as its list will be overwritten */
65static struct protocol proto_tcpv4 = {
66 .name = "tcpv4",
67 .sock_domain = AF_INET,
68 .sock_type = SOCK_STREAM,
69 .sock_prot = IPPROTO_TCP,
70 .sock_family = AF_INET,
71 .sock_addrlen = sizeof(struct sockaddr_in),
72 .l3_addrlen = 32/8,
Willy Tarreaubbebbbf2012-05-07 21:22:09 +020073 .accept = &listener_accept,
Willy Tarreau26d8c592012-05-07 18:12:14 +020074 .connect = tcp_connect_server,
Emeric Bruncf20bf12010-10-22 16:06:11 +020075 .bind = tcp_bind_listener,
Willy Tarreaue6b98942007-10-29 01:09:36 +010076 .bind_all = tcp_bind_listeners,
77 .unbind_all = unbind_all_listeners,
78 .enable_all = enable_all_listeners,
Willy Tarreau59b94792012-05-11 16:16:40 +020079 .get_src = tcp_get_src,
80 .get_dst = tcp_get_dst,
Willy Tarreau2b57cb82013-06-10 19:56:38 +020081 .drain = tcp_drain,
Willy Tarreaue6b98942007-10-29 01:09:36 +010082 .listeners = LIST_HEAD_INIT(proto_tcpv4.listeners),
83 .nb_listeners = 0,
84};
85
86/* Note: must not be declared <const> as its list will be overwritten */
87static struct protocol proto_tcpv6 = {
88 .name = "tcpv6",
89 .sock_domain = AF_INET6,
90 .sock_type = SOCK_STREAM,
91 .sock_prot = IPPROTO_TCP,
92 .sock_family = AF_INET6,
93 .sock_addrlen = sizeof(struct sockaddr_in6),
94 .l3_addrlen = 128/8,
Willy Tarreaubbebbbf2012-05-07 21:22:09 +020095 .accept = &listener_accept,
Willy Tarreau26d8c592012-05-07 18:12:14 +020096 .connect = tcp_connect_server,
Emeric Bruncf20bf12010-10-22 16:06:11 +020097 .bind = tcp_bind_listener,
Willy Tarreaue6b98942007-10-29 01:09:36 +010098 .bind_all = tcp_bind_listeners,
99 .unbind_all = unbind_all_listeners,
100 .enable_all = enable_all_listeners,
Willy Tarreau59b94792012-05-11 16:16:40 +0200101 .get_src = tcp_get_src,
102 .get_dst = tcp_get_dst,
Willy Tarreau2b57cb82013-06-10 19:56:38 +0200103 .drain = tcp_drain,
Willy Tarreaue6b98942007-10-29 01:09:36 +0100104 .listeners = LIST_HEAD_INIT(proto_tcpv6.listeners),
105 .nb_listeners = 0,
106};
107
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100108
David du Colombier6f5ccb12011-03-10 22:26:24 +0100109/* Binds ipv4/ipv6 address <local> to socket <fd>, unless <flags> is set, in which
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100110 * case we try to bind <remote>. <flags> is a 2-bit field consisting of :
111 * - 0 : ignore remote address (may even be a NULL pointer)
112 * - 1 : use provided address
113 * - 2 : use provided port
114 * - 3 : use both
115 *
116 * The function supports multiple foreign binding methods :
117 * - linux_tproxy: we directly bind to the foreign address
118 * - cttproxy: we bind to a local address then nat.
119 * The second one can be used as a fallback for the first one.
120 * This function returns 0 when everything's OK, 1 if it could not bind, to the
121 * local address, 2 if it could not bind to the foreign address.
122 */
David du Colombier6f5ccb12011-03-10 22:26:24 +0100123int tcp_bind_socket(int fd, int flags, struct sockaddr_storage *local, struct sockaddr_storage *remote)
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100124{
David du Colombier6f5ccb12011-03-10 22:26:24 +0100125 struct sockaddr_storage bind_addr;
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100126 int foreign_ok = 0;
127 int ret;
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100128 static int ip_transp_working = 1;
David du Colombier65c17962012-07-13 14:34:59 +0200129 static int ip6_transp_working = 1;
Pieter Baauwd551fb52013-05-08 22:49:23 +0200130
David du Colombier65c17962012-07-13 14:34:59 +0200131 switch (local->ss_family) {
132 case AF_INET:
133 if (flags && ip_transp_working) {
Pieter Baauwd551fb52013-05-08 22:49:23 +0200134 /* This deserves some explanation. Some platforms will support
135 * multiple combinations of certain methods, so we try the
136 * supported ones until one succeeds.
137 */
138 if (0
139#if defined(IP_TRANSPARENT)
140 || (setsockopt(fd, SOL_IP, IP_TRANSPARENT, &one, sizeof(one)) == 0)
141#endif
142#if defined(IP_FREEBIND)
143 || (setsockopt(fd, SOL_IP, IP_FREEBIND, &one, sizeof(one)) == 0)
144#endif
Pieter Baauwff30b662013-05-08 23:22:39 +0200145#if defined(IP_BINDANY)
146 || (setsockopt(fd, IPPROTO_IP, IP_BINDANY, &one, sizeof(one)) == 0)
147#endif
Pieter Baauw1eb75922013-05-08 23:30:23 +0200148#if defined(SO_BINDANY)
149 || (setsockopt(fd, SOL_SOCKET, SO_BINDANY, &one, sizeof(one)) == 0)
150#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +0200151 )
David du Colombier65c17962012-07-13 14:34:59 +0200152 foreign_ok = 1;
153 else
154 ip_transp_working = 0;
155 }
156 break;
157 case AF_INET6:
158 if (flags && ip6_transp_working) {
Pieter Baauwd551fb52013-05-08 22:49:23 +0200159 if (0
160#if defined(IPV6_TRANSPARENT)
161 || (setsockopt(fd, SOL_IPV6, IPV6_TRANSPARENT, &one, sizeof(one)) == 0)
162#endif
Pieter Baauwff30b662013-05-08 23:22:39 +0200163#if defined(IPV6_BINDANY)
164 || (setsockopt(fd, IPPROTO_IPV6, IPV6_BINDANY, &one, sizeof(one)) == 0)
165#endif
Pieter Baauw1eb75922013-05-08 23:30:23 +0200166#if defined(SO_BINDANY)
167 || (setsockopt(fd, SOL_SOCKET, SO_BINDANY, &one, sizeof(one)) == 0)
168#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +0200169 )
David du Colombier65c17962012-07-13 14:34:59 +0200170 foreign_ok = 1;
171 else
172 ip6_transp_working = 0;
173 }
174 break;
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100175 }
Pieter Baauwd551fb52013-05-08 22:49:23 +0200176
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100177 if (flags) {
178 memset(&bind_addr, 0, sizeof(bind_addr));
Willy Tarreau96dd0792011-04-19 07:20:57 +0200179 bind_addr.ss_family = remote->ss_family;
David du Colombier6f5ccb12011-03-10 22:26:24 +0100180 switch (remote->ss_family) {
181 case AF_INET:
182 if (flags & 1)
183 ((struct sockaddr_in *)&bind_addr)->sin_addr = ((struct sockaddr_in *)remote)->sin_addr;
184 if (flags & 2)
185 ((struct sockaddr_in *)&bind_addr)->sin_port = ((struct sockaddr_in *)remote)->sin_port;
186 break;
187 case AF_INET6:
188 if (flags & 1)
189 ((struct sockaddr_in6 *)&bind_addr)->sin6_addr = ((struct sockaddr_in6 *)remote)->sin6_addr;
190 if (flags & 2)
191 ((struct sockaddr_in6 *)&bind_addr)->sin6_port = ((struct sockaddr_in6 *)remote)->sin6_port;
192 break;
Willy Tarreau5dc1e982011-12-16 21:25:11 +0100193 default:
194 /* we don't want to try to bind to an unknown address family */
195 foreign_ok = 0;
David du Colombier6f5ccb12011-03-10 22:26:24 +0100196 }
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100197 }
198
Simon Hormande072bd2011-06-24 15:11:37 +0900199 setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &one, sizeof(one));
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100200 if (foreign_ok) {
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200201 if (is_addr(&bind_addr)) {
202 ret = bind(fd, (struct sockaddr *)&bind_addr, get_addr_len(&bind_addr));
203 if (ret < 0)
204 return 2;
205 }
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100206 }
207 else {
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200208 if (is_addr(local)) {
209 ret = bind(fd, (struct sockaddr *)local, get_addr_len(local));
210 if (ret < 0)
211 return 1;
212 }
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100213 }
214
215 if (!flags)
216 return 0;
217
218#ifdef CONFIG_HAP_CTTPROXY
Willy Tarreau6f831b42011-03-20 14:03:54 +0100219 if (!foreign_ok && remote->ss_family == AF_INET) {
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100220 struct in_tproxy itp1, itp2;
221 memset(&itp1, 0, sizeof(itp1));
222
223 itp1.op = TPROXY_ASSIGN;
Willy Tarreau6f831b42011-03-20 14:03:54 +0100224 itp1.v.addr.faddr = ((struct sockaddr_in *)&bind_addr)->sin_addr;
225 itp1.v.addr.fport = ((struct sockaddr_in *)&bind_addr)->sin_port;
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100226
227 /* set connect flag on socket */
228 itp2.op = TPROXY_FLAGS;
229 itp2.v.flags = ITP_CONNECT | ITP_ONCE;
230
231 if (setsockopt(fd, SOL_IP, IP_TPROXY, &itp1, sizeof(itp1)) != -1 &&
232 setsockopt(fd, SOL_IP, IP_TPROXY, &itp2, sizeof(itp2)) != -1) {
233 foreign_ok = 1;
234 }
235 }
236#endif
237 if (!foreign_ok)
238 /* we could not bind to a foreign address */
239 return 2;
240
241 return 0;
242}
Willy Tarreaue6b98942007-10-29 01:09:36 +0100243
Willy Tarreau9650f372009-08-16 14:02:45 +0200244
245/*
Willy Tarreau14f8e862012-08-30 22:23:13 +0200246 * This function initiates a TCP connection establishment to the target assigned
247 * to connection <conn> using (si->{target,addr.to}). A source address may be
248 * pointed to by conn->addr.from in case of transparent proxying. Normal source
249 * bind addresses are still determined locally (due to the possible need of a
250 * source port). conn->target may point either to a valid server or to a backend,
Willy Tarreau3fdb3662012-11-12 00:42:33 +0100251 * depending on conn->target. Only OBJ_TYPE_PROXY and OBJ_TYPE_SERVER are
Willy Tarreauf0837b22012-11-24 10:24:27 +0100252 * supported. The <data> parameter is a boolean indicating whether there are data
253 * waiting for being sent or not, in order to adjust data write polling and on
254 * some platforms, the ability to avoid an empty initial ACK. The <delack> argument
255 * allows the caller to force using a delayed ACK when establishing the connection :
256 * - 0 = no delayed ACK unless data are advertised and backend has tcp-smart-connect
257 * - 1 = delayed ACK if backend has tcp-smart-connect, regardless of data
258 * - 2 = delayed ACK regardless of backend options
Willy Tarreaub1d67742010-03-29 19:36:59 +0200259 *
Willy Tarreau9650f372009-08-16 14:02:45 +0200260 * It can return one of :
261 * - SN_ERR_NONE if everything's OK
262 * - SN_ERR_SRVTO if there are no more servers
263 * - SN_ERR_SRVCL if the connection was refused by the server
264 * - SN_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
265 * - SN_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
266 * - SN_ERR_INTERNAL for any other purely internal errors
267 * Additionnally, in the case of SN_ERR_RESOURCE, an emergency log will be emitted.
Willy Tarreau6b0a8502012-11-23 08:51:32 +0100268 *
269 * The connection's fd is inserted only when SN_ERR_NONE is returned, otherwise
270 * it's invalid and the caller has nothing to do.
Willy Tarreau9650f372009-08-16 14:02:45 +0200271 */
Willy Tarreauf1536862011-03-03 18:27:32 +0100272
Willy Tarreauf0837b22012-11-24 10:24:27 +0100273int tcp_connect_server(struct connection *conn, int data, int delack)
Willy Tarreau9650f372009-08-16 14:02:45 +0200274{
275 int fd;
Willy Tarreauac825402011-03-04 22:04:29 +0100276 struct server *srv;
277 struct proxy *be;
Willy Tarreaua4380b42012-12-08 22:49:11 +0100278 struct conn_src *src;
Willy Tarreauac825402011-03-04 22:04:29 +0100279
Willy Tarreau3fdb3662012-11-12 00:42:33 +0100280 switch (obj_type(conn->target)) {
281 case OBJ_TYPE_PROXY:
282 be = objt_proxy(conn->target);
Willy Tarreauac825402011-03-04 22:04:29 +0100283 srv = NULL;
284 break;
Willy Tarreau3fdb3662012-11-12 00:42:33 +0100285 case OBJ_TYPE_SERVER:
286 srv = objt_server(conn->target);
Willy Tarreauac825402011-03-04 22:04:29 +0100287 be = srv->proxy;
288 break;
289 default:
290 return SN_ERR_INTERNAL;
291 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200292
Willy Tarreau986a9d22012-08-30 21:11:38 +0200293 if ((fd = conn->t.sock.fd = socket(conn->addr.to.ss_family, SOCK_STREAM, IPPROTO_TCP)) == -1) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200294 qfprintf(stderr, "Cannot get a server socket.\n");
295
296 if (errno == ENFILE)
297 send_log(be, LOG_EMERG,
298 "Proxy %s reached system FD limit at %d. Please check system tunables.\n",
299 be->id, maxfd);
300 else if (errno == EMFILE)
301 send_log(be, LOG_EMERG,
302 "Proxy %s reached process FD limit at %d. Please check 'ulimit-n' and restart.\n",
303 be->id, maxfd);
304 else if (errno == ENOBUFS || errno == ENOMEM)
305 send_log(be, LOG_EMERG,
306 "Proxy %s reached system memory limit at %d sockets. Please check system tunables.\n",
307 be->id, maxfd);
308 /* this is a resource error */
309 return SN_ERR_RESOURCE;
310 }
311
312 if (fd >= global.maxsock) {
313 /* do not log anything there, it's a normal condition when this option
314 * is used to serialize connections to a server !
315 */
316 Alert("socket(): not enough free sockets. Raise -n argument. Giving up.\n");
317 close(fd);
318 return SN_ERR_PRXCOND; /* it is a configuration limit */
319 }
320
321 if ((fcntl(fd, F_SETFL, O_NONBLOCK)==-1) ||
Simon Hormande072bd2011-06-24 15:11:37 +0900322 (setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &one, sizeof(one)) == -1)) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200323 qfprintf(stderr,"Cannot set client socket to non blocking mode.\n");
324 close(fd);
325 return SN_ERR_INTERNAL;
326 }
327
328 if (be->options & PR_O_TCP_SRV_KA)
Simon Hormande072bd2011-06-24 15:11:37 +0900329 setsockopt(fd, SOL_SOCKET, SO_KEEPALIVE, &one, sizeof(one));
Willy Tarreau9650f372009-08-16 14:02:45 +0200330
Willy Tarreau9650f372009-08-16 14:02:45 +0200331 /* allow specific binding :
332 * - server-specific at first
333 * - proxy-specific next
334 */
Willy Tarreaua4380b42012-12-08 22:49:11 +0100335 if (srv && srv->conn_src.opts & CO_SRC_BIND)
336 src = &srv->conn_src;
337 else if (be->conn_src.opts & CO_SRC_BIND)
338 src = &be->conn_src;
339 else
340 src = NULL;
341
342 if (src) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200343 int ret, flags = 0;
344
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200345 if (is_addr(&conn->addr.from)) {
Willy Tarreaua4380b42012-12-08 22:49:11 +0100346 switch (src->opts & CO_SRC_TPROXY_MASK) {
Willy Tarreauef9a3602012-12-08 22:29:20 +0100347 case CO_SRC_TPROXY_ADDR:
348 case CO_SRC_TPROXY_CLI:
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200349 flags = 3;
350 break;
Willy Tarreauef9a3602012-12-08 22:29:20 +0100351 case CO_SRC_TPROXY_CIP:
352 case CO_SRC_TPROXY_DYN:
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200353 flags = 1;
354 break;
355 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200356 }
Willy Tarreaub1d67742010-03-29 19:36:59 +0200357
Willy Tarreau9650f372009-08-16 14:02:45 +0200358#ifdef SO_BINDTODEVICE
359 /* Note: this might fail if not CAP_NET_RAW */
Willy Tarreaua4380b42012-12-08 22:49:11 +0100360 if (src->iface_name)
361 setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, src->iface_name, src->iface_len + 1);
Willy Tarreau9650f372009-08-16 14:02:45 +0200362#endif
363
Willy Tarreaua4380b42012-12-08 22:49:11 +0100364 if (src->sport_range) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200365 int attempts = 10; /* should be more than enough to find a spare port */
Willy Tarreaua4380b42012-12-08 22:49:11 +0100366 struct sockaddr_storage sa;
Willy Tarreau9650f372009-08-16 14:02:45 +0200367
368 ret = 1;
Willy Tarreaua4380b42012-12-08 22:49:11 +0100369 sa = src->source_addr;
Willy Tarreau9650f372009-08-16 14:02:45 +0200370
371 do {
372 /* note: in case of retry, we may have to release a previously
373 * allocated port, hence this loop's construct.
374 */
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200375 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
376 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200377
378 if (!attempts)
379 break;
380 attempts--;
381
Willy Tarreaua4380b42012-12-08 22:49:11 +0100382 fdinfo[fd].local_port = port_range_alloc_port(src->sport_range);
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200383 if (!fdinfo[fd].local_port)
Willy Tarreau9650f372009-08-16 14:02:45 +0200384 break;
385
Willy Tarreaua4380b42012-12-08 22:49:11 +0100386 fdinfo[fd].port_range = src->sport_range;
387 set_host_port(&sa, fdinfo[fd].local_port);
Willy Tarreau9650f372009-08-16 14:02:45 +0200388
Willy Tarreaua4380b42012-12-08 22:49:11 +0100389 ret = tcp_bind_socket(fd, flags, &sa, &conn->addr.from);
Willy Tarreau9650f372009-08-16 14:02:45 +0200390 } while (ret != 0); /* binding NOK */
391 }
392 else {
Willy Tarreaua4380b42012-12-08 22:49:11 +0100393 ret = tcp_bind_socket(fd, flags, &src->source_addr, &conn->addr.from);
Willy Tarreau9650f372009-08-16 14:02:45 +0200394 }
395
Willy Tarreaua4380b42012-12-08 22:49:11 +0100396 if (unlikely(ret != 0)) {
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200397 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
398 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200399 close(fd);
400
Willy Tarreau9650f372009-08-16 14:02:45 +0200401 if (ret == 1) {
Willy Tarreaua4380b42012-12-08 22:49:11 +0100402 Alert("Cannot bind to source address before connect() for backend %s. Aborting.\n",
Willy Tarreau9650f372009-08-16 14:02:45 +0200403 be->id);
404 send_log(be, LOG_EMERG,
Willy Tarreaua4380b42012-12-08 22:49:11 +0100405 "Cannot bind to source address before connect() for backend %s.\n",
Willy Tarreau9650f372009-08-16 14:02:45 +0200406 be->id);
407 } else {
Willy Tarreaua4380b42012-12-08 22:49:11 +0100408 Alert("Cannot bind to tproxy source address before connect() for backend %s. Aborting.\n",
Willy Tarreau9650f372009-08-16 14:02:45 +0200409 be->id);
410 send_log(be, LOG_EMERG,
Willy Tarreaua4380b42012-12-08 22:49:11 +0100411 "Cannot bind to tproxy source address before connect() for backend %s.\n",
Willy Tarreau9650f372009-08-16 14:02:45 +0200412 be->id);
413 }
414 return SN_ERR_RESOURCE;
415 }
416 }
417
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +0400418#if defined(TCP_QUICKACK)
Willy Tarreau9650f372009-08-16 14:02:45 +0200419 /* disabling tcp quick ack now allows the first request to leave the
420 * machine with the first ACK. We only do this if there are pending
Willy Tarreauf0837b22012-11-24 10:24:27 +0100421 * data in the buffer.
Willy Tarreau9650f372009-08-16 14:02:45 +0200422 */
Willy Tarreauf0837b22012-11-24 10:24:27 +0100423 if (delack == 2 || ((delack || data) && (be->options2 & PR_O2_SMARTCON)))
Simon Hormande072bd2011-06-24 15:11:37 +0900424 setsockopt(fd, IPPROTO_TCP, TCP_QUICKACK, &zero, sizeof(zero));
Willy Tarreau9650f372009-08-16 14:02:45 +0200425#endif
426
Willy Tarreaue803de22010-01-21 17:43:04 +0100427 if (global.tune.server_sndbuf)
428 setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &global.tune.server_sndbuf, sizeof(global.tune.server_sndbuf));
429
430 if (global.tune.server_rcvbuf)
431 setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &global.tune.server_rcvbuf, sizeof(global.tune.server_rcvbuf));
432
Willy Tarreau986a9d22012-08-30 21:11:38 +0200433 if ((connect(fd, (struct sockaddr *)&conn->addr.to, get_addr_len(&conn->addr.to)) == -1) &&
Willy Tarreau9650f372009-08-16 14:02:45 +0200434 (errno != EINPROGRESS) && (errno != EALREADY) && (errno != EISCONN)) {
435
Willy Tarreaub1719512012-12-08 23:03:28 +0100436 if (errno == EAGAIN || errno == EADDRINUSE || errno == EADDRNOTAVAIL) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200437 char *msg;
Willy Tarreaub1719512012-12-08 23:03:28 +0100438 if (errno == EAGAIN || errno == EADDRNOTAVAIL)
Willy Tarreau9650f372009-08-16 14:02:45 +0200439 msg = "no free ports";
440 else
441 msg = "local address already in use";
442
Willy Tarreaub1719512012-12-08 23:03:28 +0100443 qfprintf(stderr,"Connect() failed for backend %s: %s.\n", be->id, msg);
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200444 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
445 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200446 close(fd);
Willy Tarreaub1719512012-12-08 23:03:28 +0100447 send_log(be, LOG_ERR, "Connect() failed for backend %s: %s.\n", be->id, msg);
Willy Tarreau9650f372009-08-16 14:02:45 +0200448 return SN_ERR_RESOURCE;
449 } else if (errno == ETIMEDOUT) {
450 //qfprintf(stderr,"Connect(): ETIMEDOUT");
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200451 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
452 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200453 close(fd);
454 return SN_ERR_SRVTO;
455 } else {
456 // (errno == ECONNREFUSED || errno == ENETUNREACH || errno == EACCES || errno == EPERM)
457 //qfprintf(stderr,"Connect(): %d", errno);
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200458 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
459 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200460 close(fd);
461 return SN_ERR_SRVCL;
462 }
463 }
464
Willy Tarreau986a9d22012-08-30 21:11:38 +0200465 fdtab[fd].owner = conn;
Willy Tarreau986a9d22012-08-30 21:11:38 +0200466 conn->flags = CO_FL_WAIT_L4_CONN; /* connection in progress */
Willy Tarreaufc8f1f02012-12-08 18:53:44 +0100467 conn->flags |= CO_FL_ADDR_TO_SET;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200468
Willy Tarreaud2274c62012-07-06 14:29:45 +0200469 fdtab[fd].iocb = conn_fd_handler;
Willy Tarreau9650f372009-08-16 14:02:45 +0200470 fd_insert(fd);
Willy Tarreau986a9d22012-08-30 21:11:38 +0200471 conn_sock_want_send(conn); /* for connect status */
Willy Tarreau15678ef2012-08-31 13:54:11 +0200472
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200473 if (conn_xprt_init(conn) < 0) {
Willy Tarreau184636e2012-09-06 14:04:41 +0200474 fd_delete(fd);
Willy Tarreau15678ef2012-08-31 13:54:11 +0200475 return SN_ERR_RESOURCE;
Willy Tarreau184636e2012-09-06 14:04:41 +0200476 }
Willy Tarreau15678ef2012-08-31 13:54:11 +0200477
Willy Tarreau14f8e862012-08-30 22:23:13 +0200478 if (data)
Willy Tarreau986a9d22012-08-30 21:11:38 +0200479 conn_data_want_send(conn); /* prepare to send data if any */
Willy Tarreau9650f372009-08-16 14:02:45 +0200480
Willy Tarreau9650f372009-08-16 14:02:45 +0200481 return SN_ERR_NONE; /* connection is OK */
482}
483
484
Willy Tarreau59b94792012-05-11 16:16:40 +0200485/*
486 * Retrieves the source address for the socket <fd>, with <dir> indicating
487 * if we're a listener (=0) or an initiator (!=0). It returns 0 in case of
488 * success, -1 in case of error. The socket's source address is stored in
489 * <sa> for <salen> bytes.
490 */
491int tcp_get_src(int fd, struct sockaddr *sa, socklen_t salen, int dir)
492{
493 if (dir)
494 return getsockname(fd, sa, &salen);
495 else
496 return getpeername(fd, sa, &salen);
497}
498
499
500/*
501 * Retrieves the original destination address for the socket <fd>, with <dir>
502 * indicating if we're a listener (=0) or an initiator (!=0). In the case of a
503 * listener, if the original destination address was translated, the original
504 * address is retrieved. It returns 0 in case of success, -1 in case of error.
505 * The socket's source address is stored in <sa> for <salen> bytes.
506 */
507int tcp_get_dst(int fd, struct sockaddr *sa, socklen_t salen, int dir)
508{
509 if (dir)
510 return getpeername(fd, sa, &salen);
511#if defined(TPROXY) && defined(SO_ORIGINAL_DST)
512 else if (getsockopt(fd, SOL_IP, SO_ORIGINAL_DST, sa, &salen) == 0)
513 return 0;
514#endif
515 else
516 return getsockname(fd, sa, &salen);
517}
518
Willy Tarreau2b57cb82013-06-10 19:56:38 +0200519/* Tries to drain any pending incoming data from the socket to reach the
520 * receive shutdown. Returns non-zero if the shutdown was found, otherwise
521 * zero. This is useful to decide whether we can close a connection cleanly
522 * are we must kill it hard.
523 */
524int tcp_drain(int fd)
525{
526 int turns = 2;
527 int len;
528
529 while (turns) {
530#ifdef MSG_TRUNC_CLEARS_INPUT
531 len = recv(fd, NULL, INT_MAX, MSG_DONTWAIT | MSG_NOSIGNAL | MSG_TRUNC);
532 if (len == -1 && errno == EFAULT)
533#endif
534 len = recv(fd, trash.str, trash.size, MSG_DONTWAIT | MSG_NOSIGNAL);
535
536 if (len == 0) /* cool, shutdown received */
537 return 1;
538
539 if (len < 0) {
540 if (errno == EAGAIN) /* connection not closed yet */
541 return 0;
542 if (errno == EINTR) /* oops, try again */
543 continue;
544 /* other errors indicate a dead connection, fine. */
545 return 1;
546 }
547 /* OK we read some data, let's try again once */
548 turns--;
549 }
550 /* some data are still present, give up */
551 return 0;
552}
553
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200554/* This is the callback which is set when a connection establishment is pending
Willy Tarreauf12a20e2013-12-04 16:11:04 +0100555 * and we have nothing to send. It updates the FD polling status. It returns 0
556 * if it fails in a fatal way or needs to poll to go further, otherwise it
557 * returns non-zero and removes the CO_FL_WAIT_L4_CONN flag from the connection's
558 * flags. In case of error, it sets CO_FL_ERROR and leaves the error code in
559 * errno. The error checking is done in two passes in order to limit the number
560 * of syscalls in the normal case :
561 * - if POLL_ERR was reported by the poller, we check for a pending error on
562 * the socket before proceeding. If found, it's assigned to errno so that
563 * upper layers can see it.
564 * - otherwise connect() is used to check the connection state again, since
565 * the getsockopt return cannot reliably be used to know if the connection
566 * is still pending or ready. This one may often return an error as well,
567 * since we don't always have POLL_ERR (eg: OSX or cached events).
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200568 */
Willy Tarreau239d7182012-07-23 18:53:03 +0200569int tcp_connect_probe(struct connection *conn)
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200570{
Willy Tarreau239d7182012-07-23 18:53:03 +0200571 int fd = conn->t.sock.fd;
Willy Tarreauf12a20e2013-12-04 16:11:04 +0100572 socklen_t lskerr;
573 int skerr;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200574
Willy Tarreau80184712012-07-06 14:54:49 +0200575 if (conn->flags & CO_FL_ERROR)
Willy Tarreauafad0e02012-08-09 14:45:22 +0200576 return 0;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200577
Willy Tarreau80184712012-07-06 14:54:49 +0200578 if (!(conn->flags & CO_FL_WAIT_L4_CONN))
Willy Tarreau0b0c0972012-07-23 20:05:00 +0200579 return 1; /* strange we were called while ready */
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200580
Willy Tarreauf12a20e2013-12-04 16:11:04 +0100581 /* we might be the first witness of FD_POLL_ERR. Note that FD_POLL_HUP
582 * without FD_POLL_IN also indicates a hangup without input data meaning
583 * there was no connection.
584 */
585 if (fdtab[fd].ev & FD_POLL_ERR ||
586 (fdtab[fd].ev & (FD_POLL_IN|FD_POLL_HUP)) == FD_POLL_HUP) {
587 skerr = 0;
588 lskerr = sizeof(skerr);
589 getsockopt(fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr);
590 errno = skerr;
591 if (errno == EAGAIN)
592 errno = 0;
593 if (errno)
594 goto out_error;
595 }
Willy Tarreau2da156f2012-07-23 15:07:23 +0200596
Willy Tarreauf12a20e2013-12-04 16:11:04 +0100597 /* Use connect() to check the state of the socket. This has the
598 * advantage of giving us the following info :
Willy Tarreau2c6be842012-07-06 17:12:34 +0200599 * - error
600 * - connecting (EALREADY, EINPROGRESS)
601 * - connected (EISCONN, 0)
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200602 */
Willy Tarreau986a9d22012-08-30 21:11:38 +0200603 if (connect(fd, (struct sockaddr *)&conn->addr.to, get_addr_len(&conn->addr.to)) < 0) {
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200604 if (errno == EALREADY || errno == EINPROGRESS) {
Willy Tarreaud486ef52012-12-10 17:03:52 +0100605 __conn_sock_stop_recv(conn);
606 __conn_sock_poll_send(conn);
Willy Tarreau0b0c0972012-07-23 20:05:00 +0200607 return 0;
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200608 }
Willy Tarreaua190d592012-05-20 18:35:19 +0200609
Willy Tarreau2c6be842012-07-06 17:12:34 +0200610 if (errno && errno != EISCONN)
Willy Tarreaua190d592012-05-20 18:35:19 +0200611 goto out_error;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200612
Willy Tarreau2c6be842012-07-06 17:12:34 +0200613 /* otherwise we're connected */
Willy Tarreaua190d592012-05-20 18:35:19 +0200614 }
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200615
Willy Tarreau076be252012-07-06 16:02:29 +0200616 /* The FD is ready now, we'll mark the connection as complete and
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200617 * forward the event to the transport layer which will notify the
618 * data layer.
Willy Tarreaua190d592012-05-20 18:35:19 +0200619 */
Willy Tarreau80184712012-07-06 14:54:49 +0200620 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau0b0c0972012-07-23 20:05:00 +0200621 return 1;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200622
623 out_error:
Willy Tarreau0b0c0972012-07-23 20:05:00 +0200624 /* Write error on the file descriptor. Report it to the connection
625 * and disable polling on this FD.
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200626 */
627
Willy Tarreau80184712012-07-06 14:54:49 +0200628 conn->flags |= CO_FL_ERROR;
Willy Tarreaud486ef52012-12-10 17:03:52 +0100629 __conn_sock_stop_both(conn);
Willy Tarreauafad0e02012-08-09 14:45:22 +0200630 return 0;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200631}
632
Willy Tarreau59b94792012-05-11 16:16:40 +0200633
Willy Tarreaue6b98942007-10-29 01:09:36 +0100634/* This function tries to bind a TCPv4/v6 listener. It may return a warning or
Willy Tarreau8ab505b2013-01-24 01:41:38 +0100635 * an error message in <errmsg> if the message is at most <errlen> bytes long
636 * (including '\0'). Note that <errmsg> may be NULL if <errlen> is also zero.
637 * The return value is composed from ERR_ABORT, ERR_WARN,
Willy Tarreaue6b98942007-10-29 01:09:36 +0100638 * ERR_ALERT, ERR_RETRYABLE and ERR_FATAL. ERR_NONE indicates that everything
639 * was alright and that no message was returned. ERR_RETRYABLE means that an
640 * error occurred but that it may vanish after a retry (eg: port in use), and
Aman Guptad94991d2012-04-06 17:39:26 -0700641 * ERR_FATAL indicates a non-fixable error. ERR_WARN and ERR_ALERT do not alter
Willy Tarreaue6b98942007-10-29 01:09:36 +0100642 * the meaning of the error, but just indicate that a message is present which
643 * should be displayed with the respective level. Last, ERR_ABORT indicates
644 * that it's pointless to try to start other listeners. No error message is
645 * returned if errlen is NULL.
646 */
647int tcp_bind_listener(struct listener *listener, char *errmsg, int errlen)
648{
649 __label__ tcp_return, tcp_close_return;
650 int fd, err;
Willy Tarreau40aa0702013-03-10 23:51:38 +0100651 int ext, ready;
652 socklen_t ready_len;
Willy Tarreaue6b98942007-10-29 01:09:36 +0100653 const char *msg = NULL;
654
655 /* ensure we never return garbage */
Willy Tarreau8ab505b2013-01-24 01:41:38 +0100656 if (errlen)
Willy Tarreaue6b98942007-10-29 01:09:36 +0100657 *errmsg = 0;
658
659 if (listener->state != LI_ASSIGNED)
660 return ERR_NONE; /* already bound */
661
662 err = ERR_NONE;
663
Willy Tarreau40aa0702013-03-10 23:51:38 +0100664 /* if the listener already has an fd assigned, then we were offered the
665 * fd by an external process (most likely the parent), and we don't want
666 * to create a new socket. However we still want to set a few flags on
667 * the socket.
668 */
669 fd = listener->fd;
670 ext = (fd >= 0);
671
672 if (!ext && (fd = socket(listener->addr.ss_family, SOCK_STREAM, IPPROTO_TCP)) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +0100673 err |= ERR_RETRYABLE | ERR_ALERT;
674 msg = "cannot create listening socket";
675 goto tcp_return;
676 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100677
Willy Tarreaue6b98942007-10-29 01:09:36 +0100678 if (fd >= global.maxsock) {
679 err |= ERR_FATAL | ERR_ABORT | ERR_ALERT;
680 msg = "not enough free sockets (raise '-n' parameter)";
681 goto tcp_close_return;
682 }
683
Willy Tarreaufb14edc2009-06-14 15:24:37 +0200684 if (fcntl(fd, F_SETFL, O_NONBLOCK) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +0100685 err |= ERR_FATAL | ERR_ALERT;
686 msg = "cannot make socket non-blocking";
687 goto tcp_close_return;
688 }
689
Willy Tarreau40aa0702013-03-10 23:51:38 +0100690 if (!ext && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &one, sizeof(one)) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +0100691 /* not fatal but should be reported */
692 msg = "cannot do so_reuseaddr";
693 err |= ERR_ALERT;
694 }
695
696 if (listener->options & LI_O_NOLINGER)
Simon Hormande072bd2011-06-24 15:11:37 +0900697 setsockopt(fd, SOL_SOCKET, SO_LINGER, &nolinger, sizeof(struct linger));
Willy Tarreauedcf6682008-11-30 23:15:34 +0100698
Willy Tarreaue6b98942007-10-29 01:09:36 +0100699#ifdef SO_REUSEPORT
700 /* OpenBSD supports this. As it's present in old libc versions of Linux,
701 * it might return an error that we will silently ignore.
702 */
Willy Tarreau40aa0702013-03-10 23:51:38 +0100703 if (!ext)
704 setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &one, sizeof(one));
Willy Tarreaue6b98942007-10-29 01:09:36 +0100705#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +0200706
Willy Tarreau40aa0702013-03-10 23:51:38 +0100707 if (!ext && (listener->options & LI_O_FOREIGN)) {
David du Colombier65c17962012-07-13 14:34:59 +0200708 switch (listener->addr.ss_family) {
709 case AF_INET:
Pieter Baauwd551fb52013-05-08 22:49:23 +0200710 if (1
711#if defined(IP_TRANSPARENT)
712 && (setsockopt(fd, SOL_IP, IP_TRANSPARENT, &one, sizeof(one)) == -1)
713#endif
714#if defined(IP_FREEBIND)
715 && (setsockopt(fd, SOL_IP, IP_FREEBIND, &one, sizeof(one)) == -1)
716#endif
Pieter Baauwff30b662013-05-08 23:22:39 +0200717#if defined(IP_BINDANY)
718 && (setsockopt(fd, IPPROTO_IP, IP_BINDANY, &one, sizeof(one)) == -1)
719#endif
Pieter Baauw1eb75922013-05-08 23:30:23 +0200720#if defined(SO_BINDANY)
721 && (setsockopt(fd, SOL_SOCKET, SO_BINDANY, &one, sizeof(one)) == -1)
722#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +0200723 ) {
David du Colombier65c17962012-07-13 14:34:59 +0200724 msg = "cannot make listening socket transparent";
725 err |= ERR_ALERT;
726 }
727 break;
728 case AF_INET6:
Pieter Baauwd551fb52013-05-08 22:49:23 +0200729 if (1
730#if defined(IPV6_TRANSPARENT)
731 && (setsockopt(fd, SOL_IPV6, IPV6_TRANSPARENT, &one, sizeof(one)) == -1)
732#endif
Pieter Baauwff30b662013-05-08 23:22:39 +0200733#if defined(IPV6_BINDANY)
734 && (setsockopt(fd, IPPROTO_IPV6, IPV6_BINDANY, &one, sizeof(one)) == -1)
735#endif
Pieter Baauw1eb75922013-05-08 23:30:23 +0200736#if defined(SO_BINDANY)
737 && (setsockopt(fd, SOL_SOCKET, SO_BINDANY, &one, sizeof(one)) == -1)
738#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +0200739 ) {
David du Colombier65c17962012-07-13 14:34:59 +0200740 msg = "cannot make listening socket transparent";
741 err |= ERR_ALERT;
742 }
743 break;
744 }
Willy Tarreaub1e52e82008-01-13 14:49:51 +0100745 }
Pieter Baauwd551fb52013-05-08 22:49:23 +0200746
Willy Tarreau5e6e2042009-02-04 17:19:29 +0100747#ifdef SO_BINDTODEVICE
748 /* Note: this might fail if not CAP_NET_RAW */
Willy Tarreau40aa0702013-03-10 23:51:38 +0100749 if (!ext && listener->interface) {
Willy Tarreau5e6e2042009-02-04 17:19:29 +0100750 if (setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE,
Willy Tarreau604e8302009-03-06 00:48:23 +0100751 listener->interface, strlen(listener->interface) + 1) == -1) {
Willy Tarreau5e6e2042009-02-04 17:19:29 +0100752 msg = "cannot bind listener to device";
753 err |= ERR_WARN;
754 }
755 }
756#endif
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +0400757#if defined(TCP_MAXSEG)
Willy Tarreau48a7e722010-12-24 15:26:39 +0100758 if (listener->maxseg > 0) {
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +0400759 if (setsockopt(fd, IPPROTO_TCP, TCP_MAXSEG,
Willy Tarreaube1b9182009-06-14 18:48:19 +0200760 &listener->maxseg, sizeof(listener->maxseg)) == -1) {
761 msg = "cannot set MSS";
762 err |= ERR_WARN;
763 }
764 }
765#endif
Willy Tarreaucb6cd432009-10-13 07:34:14 +0200766#if defined(TCP_DEFER_ACCEPT)
767 if (listener->options & LI_O_DEF_ACCEPT) {
768 /* defer accept by up to one second */
769 int accept_delay = 1;
770 if (setsockopt(fd, IPPROTO_TCP, TCP_DEFER_ACCEPT, &accept_delay, sizeof(accept_delay)) == -1) {
771 msg = "cannot enable DEFER_ACCEPT";
772 err |= ERR_WARN;
773 }
774 }
775#endif
Willy Tarreau1c862c52012-10-05 16:21:00 +0200776#if defined(TCP_FASTOPEN)
777 if (listener->options & LI_O_TCP_FO) {
778 /* TFO needs a queue length, let's use the configured backlog */
779 int qlen = listener->backlog ? listener->backlog : listener->maxconn;
780 if (setsockopt(fd, IPPROTO_TCP, TCP_FASTOPEN, &qlen, sizeof(qlen)) == -1) {
781 msg = "cannot enable TCP_FASTOPEN";
782 err |= ERR_WARN;
783 }
784 }
785#endif
Willy Tarreau9b6700f2012-11-24 11:55:28 +0100786#if defined(IPV6_V6ONLY)
787 if (listener->options & LI_O_V6ONLY)
788 setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &one, sizeof(one));
Willy Tarreau77e3af92012-11-24 15:07:23 +0100789 else if (listener->options & LI_O_V4V6)
790 setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &zero, sizeof(zero));
Willy Tarreau9b6700f2012-11-24 11:55:28 +0100791#endif
792
Willy Tarreau40aa0702013-03-10 23:51:38 +0100793 if (!ext && bind(fd, (struct sockaddr *)&listener->addr, listener->proto->sock_addrlen) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +0100794 err |= ERR_RETRYABLE | ERR_ALERT;
795 msg = "cannot bind socket";
796 goto tcp_close_return;
797 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100798
Willy Tarreau40aa0702013-03-10 23:51:38 +0100799 ready = 0;
800 ready_len = sizeof(ready);
801 if (getsockopt(fd, SOL_SOCKET, SO_ACCEPTCONN, &ready, &ready_len) == -1)
802 ready = 0;
803
804 if (!(ext && ready) && /* only listen if not already done by external process */
805 listen(fd, listener->backlog ? listener->backlog : listener->maxconn) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +0100806 err |= ERR_RETRYABLE | ERR_ALERT;
807 msg = "cannot listen to socket";
808 goto tcp_close_return;
809 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100810
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +0400811#if defined(TCP_QUICKACK)
Willy Tarreau9ea05a72009-06-14 12:07:01 +0200812 if (listener->options & LI_O_NOQUICKACK)
Simon Hormande072bd2011-06-24 15:11:37 +0900813 setsockopt(fd, IPPROTO_TCP, TCP_QUICKACK, &zero, sizeof(zero));
Willy Tarreau9ea05a72009-06-14 12:07:01 +0200814#endif
815
Willy Tarreaue6b98942007-10-29 01:09:36 +0100816 /* the socket is ready */
817 listener->fd = fd;
818 listener->state = LI_LISTEN;
819
Willy Tarreaueabf3132008-08-29 23:36:51 +0200820 fdtab[fd].owner = listener; /* reference the listener instead of a task */
Willy Tarreauaece46a2012-07-06 12:25:58 +0200821 fdtab[fd].iocb = listener->proto->accept;
Willy Tarreaueb472682010-05-28 18:46:57 +0200822 fd_insert(fd);
823
Willy Tarreaue6b98942007-10-29 01:09:36 +0100824 tcp_return:
Cyril Bonté43ba1b32010-11-01 19:26:01 +0100825 if (msg && errlen) {
826 char pn[INET6_ADDRSTRLEN];
827
Willy Tarreau631f01c2011-09-05 00:36:48 +0200828 addr_to_str(&listener->addr, pn, sizeof(pn));
829 snprintf(errmsg, errlen, "%s [%s:%d]", msg, pn, get_host_port(&listener->addr));
Cyril Bonté43ba1b32010-11-01 19:26:01 +0100830 }
Willy Tarreaue6b98942007-10-29 01:09:36 +0100831 return err;
832
833 tcp_close_return:
834 close(fd);
835 goto tcp_return;
836}
837
838/* This function creates all TCP sockets bound to the protocol entry <proto>.
839 * It is intended to be used as the protocol's bind_all() function.
840 * The sockets will be registered but not added to any fd_set, in order not to
841 * loose them across the fork(). A call to enable_all_listeners() is needed
842 * to complete initialization. The return value is composed from ERR_*.
843 */
Emeric Bruncf20bf12010-10-22 16:06:11 +0200844static int tcp_bind_listeners(struct protocol *proto, char *errmsg, int errlen)
Willy Tarreaue6b98942007-10-29 01:09:36 +0100845{
846 struct listener *listener;
847 int err = ERR_NONE;
848
849 list_for_each_entry(listener, &proto->listeners, proto_list) {
Emeric Bruncf20bf12010-10-22 16:06:11 +0200850 err |= tcp_bind_listener(listener, errmsg, errlen);
851 if (err & ERR_ABORT)
Willy Tarreaue6b98942007-10-29 01:09:36 +0100852 break;
853 }
854
855 return err;
856}
857
858/* Add listener to the list of tcpv4 listeners. The listener's state
859 * is automatically updated from LI_INIT to LI_ASSIGNED. The number of
860 * listeners is updated. This is the function to use to add a new listener.
861 */
862void tcpv4_add_listener(struct listener *listener)
863{
864 if (listener->state != LI_INIT)
865 return;
866 listener->state = LI_ASSIGNED;
867 listener->proto = &proto_tcpv4;
868 LIST_ADDQ(&proto_tcpv4.listeners, &listener->proto_list);
869 proto_tcpv4.nb_listeners++;
870}
871
872/* Add listener to the list of tcpv4 listeners. The listener's state
873 * is automatically updated from LI_INIT to LI_ASSIGNED. The number of
874 * listeners is updated. This is the function to use to add a new listener.
875 */
876void tcpv6_add_listener(struct listener *listener)
877{
878 if (listener->state != LI_INIT)
879 return;
880 listener->state = LI_ASSIGNED;
881 listener->proto = &proto_tcpv6;
882 LIST_ADDQ(&proto_tcpv6.listeners, &listener->proto_list);
883 proto_tcpv6.nb_listeners++;
884}
885
Willy Tarreauedcf6682008-11-30 23:15:34 +0100886/* This function performs the TCP request analysis on the current request. It
887 * returns 1 if the processing can continue on next analysers, or zero if it
888 * needs more data, encounters an error, or wants to immediately abort the
Willy Tarreaufb356202010-08-03 14:02:05 +0200889 * request. It relies on buffers flags, and updates s->req->analysers. The
890 * function may be called for frontend rules and backend rules. It only relies
891 * on the backend pointer so this works for both cases.
Willy Tarreauedcf6682008-11-30 23:15:34 +0100892 */
Willy Tarreau7421efb2012-07-02 15:11:27 +0200893int tcp_inspect_request(struct session *s, struct channel *req, int an_bit)
Willy Tarreauedcf6682008-11-30 23:15:34 +0100894{
895 struct tcp_rule *rule;
Willy Tarreaud1f96522010-08-03 19:34:32 +0200896 struct stksess *ts;
897 struct stktable *t;
Willy Tarreauedcf6682008-11-30 23:15:34 +0100898 int partial;
899
Willy Tarreau02d6cfc2012-03-01 18:19:58 +0100900 DPRINTF(stderr,"[%u] %s: session=%p b=%p, exp(r,w)=%u,%u bf=%08x bh=%d analysers=%02x\n",
Willy Tarreauedcf6682008-11-30 23:15:34 +0100901 now_ms, __FUNCTION__,
902 s,
903 req,
904 req->rex, req->wex,
905 req->flags,
Willy Tarreau9b28e032012-10-12 23:49:43 +0200906 req->buf->i,
Willy Tarreauedcf6682008-11-30 23:15:34 +0100907 req->analysers);
908
Willy Tarreauedcf6682008-11-30 23:15:34 +0100909 /* We don't know whether we have enough data, so must proceed
910 * this way :
911 * - iterate through all rules in their declaration order
912 * - if one rule returns MISS, it means the inspect delay is
913 * not over yet, then return immediately, otherwise consider
914 * it as a non-match.
915 * - if one rule returns OK, then return OK
916 * - if one rule returns KO, then return KO
917 */
918
Willy Tarreau9b28e032012-10-12 23:49:43 +0200919 if ((req->flags & CF_SHUTR) || buffer_full(req->buf, global.tune.maxrewrite) ||
Willy Tarreau3bf1b2b2012-08-27 20:46:07 +0200920 !s->be->tcp_req.inspect_delay || tick_is_expired(req->analyse_exp, now_ms))
Willy Tarreau32a6f2e2012-04-25 10:13:36 +0200921 partial = SMP_OPT_FINAL;
Willy Tarreauedcf6682008-11-30 23:15:34 +0100922 else
Willy Tarreau32a6f2e2012-04-25 10:13:36 +0200923 partial = 0;
Willy Tarreauedcf6682008-11-30 23:15:34 +0100924
Willy Tarreaufb356202010-08-03 14:02:05 +0200925 list_for_each_entry(rule, &s->be->tcp_req.inspect_rules, list) {
Willy Tarreau0cba6072013-11-28 22:21:02 +0100926 enum acl_test_res ret = ACL_TEST_PASS;
Willy Tarreauedcf6682008-11-30 23:15:34 +0100927
928 if (rule->cond) {
Willy Tarreau32a6f2e2012-04-25 10:13:36 +0200929 ret = acl_exec_cond(rule->cond, s->be, s, &s->txn, SMP_OPT_DIR_REQ | partial);
Thierry FOURNIERa65b3432013-11-28 18:22:00 +0100930 if (ret == ACL_TEST_MISS) {
Willy Tarreau8263d2b2012-08-28 00:06:31 +0200931 channel_dont_connect(req);
Willy Tarreauedcf6682008-11-30 23:15:34 +0100932 /* just set the request timeout once at the beginning of the request */
Willy Tarreaufb356202010-08-03 14:02:05 +0200933 if (!tick_isset(req->analyse_exp) && s->be->tcp_req.inspect_delay)
Willy Tarreau0bb166b2013-11-04 15:56:53 +0100934 req->analyse_exp = tick_add(now_ms, s->be->tcp_req.inspect_delay);
Willy Tarreauedcf6682008-11-30 23:15:34 +0100935 return 0;
936 }
937
938 ret = acl_pass(ret);
939 if (rule->cond->pol == ACL_COND_UNLESS)
940 ret = !ret;
941 }
942
943 if (ret) {
944 /* we have a matching rule. */
945 if (rule->action == TCP_ACT_REJECT) {
Willy Tarreau8263d2b2012-08-28 00:06:31 +0200946 channel_abort(req);
947 channel_abort(s->rep);
Willy Tarreauedcf6682008-11-30 23:15:34 +0100948 req->analysers = 0;
Krzysztof Piotr Oledzkiaeebf9b2009-10-04 15:43:17 +0200949
Willy Tarreau7d0aaf32011-03-10 23:25:56 +0100950 s->be->be_counters.denied_req++;
951 s->fe->fe_counters.denied_req++;
Krzysztof Piotr Oledzkiaeebf9b2009-10-04 15:43:17 +0200952 if (s->listener->counters)
Willy Tarreau23968d82010-05-23 23:50:44 +0200953 s->listener->counters->denied_req++;
Krzysztof Piotr Oledzkiaeebf9b2009-10-04 15:43:17 +0200954
Willy Tarreauedcf6682008-11-30 23:15:34 +0100955 if (!(s->flags & SN_ERR_MASK))
956 s->flags |= SN_ERR_PRXCOND;
957 if (!(s->flags & SN_FINST_MASK))
958 s->flags |= SN_FINST_R;
959 return 0;
960 }
Willy Tarreau44778ad2013-10-30 19:24:00 +0100961 else if (rule->action >= TCP_ACT_TRK_SC0 && rule->action <= TCP_ACT_TRK_SCMAX) {
Willy Tarreau5d5b5d82012-12-09 12:00:04 +0100962 /* Note: only the first valid tracking parameter of each
963 * applies.
964 */
965 struct stktable_key *key;
966
Willy Tarreau44778ad2013-10-30 19:24:00 +0100967 if (s->stkctr[tcp_trk_idx(rule->action)].entry)
968 continue;
969
Willy Tarreau5d5b5d82012-12-09 12:00:04 +0100970 t = rule->act_prm.trk_ctr.table.t;
971 key = stktable_fetch_key(t, s->be, s, &s->txn, SMP_OPT_DIR_REQ|SMP_OPT_FINAL, rule->act_prm.trk_ctr.expr);
972
973 if (key && (ts = stktable_get_entry(t, key))) {
Willy Tarreaud5ca9ab2013-05-28 17:40:25 +0200974 session_track_stkctr(&s->stkctr[tcp_trk_idx(rule->action)], t, ts);
975 if (s->fe != s->be)
Willy Tarreaube4a3ef2013-06-17 15:04:07 +0200976 s->flags |= SN_BE_TRACK_SC0 << tcp_trk_idx(rule->action);
Willy Tarreaud1f96522010-08-03 19:34:32 +0200977 }
978 }
979 else {
Willy Tarreauedcf6682008-11-30 23:15:34 +0100980 /* otherwise accept */
Willy Tarreaud1f96522010-08-03 19:34:32 +0200981 break;
982 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100983 }
984 }
985
986 /* if we get there, it means we have no rule which matches, or
987 * we have an explicit accept, so we apply the default accept.
988 */
Willy Tarreau3a816292009-07-07 10:55:49 +0200989 req->analysers &= ~an_bit;
Willy Tarreauedcf6682008-11-30 23:15:34 +0100990 req->analyse_exp = TICK_ETERNITY;
991 return 1;
992}
993
Emeric Brun97679e72010-09-23 17:56:44 +0200994/* This function performs the TCP response analysis on the current response. It
995 * returns 1 if the processing can continue on next analysers, or zero if it
996 * needs more data, encounters an error, or wants to immediately abort the
997 * response. It relies on buffers flags, and updates s->rep->analysers. The
998 * function may be called for backend rules.
999 */
Willy Tarreau7421efb2012-07-02 15:11:27 +02001000int tcp_inspect_response(struct session *s, struct channel *rep, int an_bit)
Emeric Brun97679e72010-09-23 17:56:44 +02001001{
1002 struct tcp_rule *rule;
1003 int partial;
1004
Willy Tarreau02d6cfc2012-03-01 18:19:58 +01001005 DPRINTF(stderr,"[%u] %s: session=%p b=%p, exp(r,w)=%u,%u bf=%08x bh=%d analysers=%02x\n",
Emeric Brun97679e72010-09-23 17:56:44 +02001006 now_ms, __FUNCTION__,
1007 s,
1008 rep,
1009 rep->rex, rep->wex,
1010 rep->flags,
Willy Tarreau9b28e032012-10-12 23:49:43 +02001011 rep->buf->i,
Emeric Brun97679e72010-09-23 17:56:44 +02001012 rep->analysers);
1013
1014 /* We don't know whether we have enough data, so must proceed
1015 * this way :
1016 * - iterate through all rules in their declaration order
1017 * - if one rule returns MISS, it means the inspect delay is
1018 * not over yet, then return immediately, otherwise consider
1019 * it as a non-match.
1020 * - if one rule returns OK, then return OK
1021 * - if one rule returns KO, then return KO
1022 */
1023
Willy Tarreau03cdb7c2012-08-27 23:14:58 +02001024 if (rep->flags & CF_SHUTR || tick_is_expired(rep->analyse_exp, now_ms))
Willy Tarreau32a6f2e2012-04-25 10:13:36 +02001025 partial = SMP_OPT_FINAL;
Emeric Brun97679e72010-09-23 17:56:44 +02001026 else
Willy Tarreau32a6f2e2012-04-25 10:13:36 +02001027 partial = 0;
Emeric Brun97679e72010-09-23 17:56:44 +02001028
1029 list_for_each_entry(rule, &s->be->tcp_rep.inspect_rules, list) {
Willy Tarreau0cba6072013-11-28 22:21:02 +01001030 enum acl_test_res ret = ACL_TEST_PASS;
Emeric Brun97679e72010-09-23 17:56:44 +02001031
1032 if (rule->cond) {
Willy Tarreau32a6f2e2012-04-25 10:13:36 +02001033 ret = acl_exec_cond(rule->cond, s->be, s, &s->txn, SMP_OPT_DIR_RES | partial);
Thierry FOURNIERa65b3432013-11-28 18:22:00 +01001034 if (ret == ACL_TEST_MISS) {
Emeric Brun97679e72010-09-23 17:56:44 +02001035 /* just set the analyser timeout once at the beginning of the response */
1036 if (!tick_isset(rep->analyse_exp) && s->be->tcp_rep.inspect_delay)
Willy Tarreau0bb166b2013-11-04 15:56:53 +01001037 rep->analyse_exp = tick_add(now_ms, s->be->tcp_rep.inspect_delay);
Emeric Brun97679e72010-09-23 17:56:44 +02001038 return 0;
1039 }
1040
1041 ret = acl_pass(ret);
1042 if (rule->cond->pol == ACL_COND_UNLESS)
1043 ret = !ret;
1044 }
1045
1046 if (ret) {
1047 /* we have a matching rule. */
1048 if (rule->action == TCP_ACT_REJECT) {
Willy Tarreau8263d2b2012-08-28 00:06:31 +02001049 channel_abort(rep);
1050 channel_abort(s->req);
Emeric Brun97679e72010-09-23 17:56:44 +02001051 rep->analysers = 0;
1052
Willy Tarreau7d0aaf32011-03-10 23:25:56 +01001053 s->be->be_counters.denied_resp++;
1054 s->fe->fe_counters.denied_resp++;
Emeric Brun97679e72010-09-23 17:56:44 +02001055 if (s->listener->counters)
1056 s->listener->counters->denied_resp++;
1057
1058 if (!(s->flags & SN_ERR_MASK))
1059 s->flags |= SN_ERR_PRXCOND;
1060 if (!(s->flags & SN_FINST_MASK))
1061 s->flags |= SN_FINST_D;
1062 return 0;
1063 }
Willy Tarreaucc1e04b2013-09-11 23:20:29 +02001064 else if (rule->action == TCP_ACT_CLOSE) {
1065 rep->prod->flags |= SI_FL_NOLINGER | SI_FL_NOHALF;
1066 si_shutr(rep->prod);
1067 si_shutw(rep->prod);
1068 break;
1069 }
Emeric Brun97679e72010-09-23 17:56:44 +02001070 else {
1071 /* otherwise accept */
1072 break;
1073 }
1074 }
1075 }
1076
1077 /* if we get there, it means we have no rule which matches, or
1078 * we have an explicit accept, so we apply the default accept.
1079 */
1080 rep->analysers &= ~an_bit;
1081 rep->analyse_exp = TICK_ETERNITY;
1082 return 1;
1083}
1084
1085
Willy Tarreaua5c0ab22010-05-31 10:30:33 +02001086/* This function performs the TCP layer4 analysis on the current request. It
1087 * returns 0 if a reject rule matches, otherwise 1 if either an accept rule
1088 * matches or if no more rule matches. It can only use rules which don't need
1089 * any data.
1090 */
1091int tcp_exec_req_rules(struct session *s)
1092{
1093 struct tcp_rule *rule;
Willy Tarreauf059a0f2010-08-03 16:29:52 +02001094 struct stksess *ts;
Willy Tarreau9ba2dcc2010-06-14 21:04:55 +02001095 struct stktable *t = NULL;
1096 int result = 1;
Willy Tarreau0cba6072013-11-28 22:21:02 +01001097 enum acl_test_res ret;
Willy Tarreaua5c0ab22010-05-31 10:30:33 +02001098
1099 list_for_each_entry(rule, &s->fe->tcp_req.l4_rules, list) {
Thierry FOURNIERa65b3432013-11-28 18:22:00 +01001100 ret = ACL_TEST_PASS;
Willy Tarreaua5c0ab22010-05-31 10:30:33 +02001101
1102 if (rule->cond) {
Willy Tarreau32a6f2e2012-04-25 10:13:36 +02001103 ret = acl_exec_cond(rule->cond, s->fe, s, NULL, SMP_OPT_DIR_REQ|SMP_OPT_FINAL);
Willy Tarreaua5c0ab22010-05-31 10:30:33 +02001104 ret = acl_pass(ret);
1105 if (rule->cond->pol == ACL_COND_UNLESS)
1106 ret = !ret;
1107 }
1108
1109 if (ret) {
1110 /* we have a matching rule. */
1111 if (rule->action == TCP_ACT_REJECT) {
Willy Tarreau7d0aaf32011-03-10 23:25:56 +01001112 s->fe->fe_counters.denied_conn++;
Willy Tarreaua5c0ab22010-05-31 10:30:33 +02001113 if (s->listener->counters)
Willy Tarreau2799e982010-06-05 15:43:21 +02001114 s->listener->counters->denied_conn++;
Willy Tarreaua5c0ab22010-05-31 10:30:33 +02001115
1116 if (!(s->flags & SN_ERR_MASK))
1117 s->flags |= SN_ERR_PRXCOND;
1118 if (!(s->flags & SN_FINST_MASK))
1119 s->flags |= SN_FINST_R;
Willy Tarreau9ba2dcc2010-06-14 21:04:55 +02001120 result = 0;
1121 break;
Willy Tarreaua5c0ab22010-05-31 10:30:33 +02001122 }
Willy Tarreau44778ad2013-10-30 19:24:00 +01001123 else if (rule->action >= TCP_ACT_TRK_SC0 && rule->action <= TCP_ACT_TRK_SCMAX) {
Willy Tarreau5d5b5d82012-12-09 12:00:04 +01001124 /* Note: only the first valid tracking parameter of each
1125 * applies.
1126 */
1127 struct stktable_key *key;
1128
Willy Tarreau44778ad2013-10-30 19:24:00 +01001129 if (s->stkctr[tcp_trk_idx(rule->action)].entry)
1130 continue;
1131
Willy Tarreau5d5b5d82012-12-09 12:00:04 +01001132 t = rule->act_prm.trk_ctr.table.t;
1133 key = stktable_fetch_key(t, s->be, s, &s->txn, SMP_OPT_DIR_REQ|SMP_OPT_FINAL, rule->act_prm.trk_ctr.expr);
1134
Willy Tarreaud5ca9ab2013-05-28 17:40:25 +02001135 if (key && (ts = stktable_get_entry(t, key)))
1136 session_track_stkctr(&s->stkctr[tcp_trk_idx(rule->action)], t, ts);
Willy Tarreauf059a0f2010-08-03 16:29:52 +02001137 }
Willy Tarreau4f0d9192013-06-11 20:40:55 +02001138 else if (rule->action == TCP_ACT_EXPECT_PX) {
1139 s->si[0].conn->flags |= CO_FL_ACCEPT_PROXY;
1140 conn_sock_want_recv(s->si[0].conn);
1141 }
Willy Tarreau9ba2dcc2010-06-14 21:04:55 +02001142 else {
1143 /* otherwise it's an accept */
1144 break;
1145 }
Willy Tarreaua5c0ab22010-05-31 10:30:33 +02001146 }
1147 }
Willy Tarreau9ba2dcc2010-06-14 21:04:55 +02001148 return result;
Willy Tarreaua5c0ab22010-05-31 10:30:33 +02001149}
1150
Emeric Brun97679e72010-09-23 17:56:44 +02001151/* Parse a tcp-response rule. Return a negative value in case of failure */
1152static int tcp_parse_response_rule(char **args, int arg, int section_type,
1153 struct proxy *curpx, struct proxy *defpx,
Willy Tarreau80aca902013-01-07 15:42:20 +01001154 struct tcp_rule *rule, char **err,
1155 unsigned int where)
Emeric Brun97679e72010-09-23 17:56:44 +02001156{
1157 if (curpx == defpx || !(curpx->cap & PR_CAP_BE)) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001158 memprintf(err, "%s %s is only allowed in 'backend' sections",
1159 args[0], args[1]);
Emeric Brun97679e72010-09-23 17:56:44 +02001160 return -1;
1161 }
1162
1163 if (strcmp(args[arg], "accept") == 0) {
1164 arg++;
1165 rule->action = TCP_ACT_ACCEPT;
1166 }
1167 else if (strcmp(args[arg], "reject") == 0) {
1168 arg++;
1169 rule->action = TCP_ACT_REJECT;
1170 }
Willy Tarreaucc1e04b2013-09-11 23:20:29 +02001171 else if (strcmp(args[arg], "close") == 0) {
1172 arg++;
1173 rule->action = TCP_ACT_CLOSE;
1174 }
Emeric Brun97679e72010-09-23 17:56:44 +02001175 else {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001176 memprintf(err,
Willy Tarreaucc1e04b2013-09-11 23:20:29 +02001177 "'%s %s' expects 'accept', 'close' or 'reject' in %s '%s' (got '%s')",
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001178 args[0], args[1], proxy_type_str(curpx), curpx->id, args[arg]);
Emeric Brun97679e72010-09-23 17:56:44 +02001179 return -1;
1180 }
1181
1182 if (strcmp(args[arg], "if") == 0 || strcmp(args[arg], "unless") == 0) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001183 if ((rule->cond = build_acl_cond(NULL, 0, curpx, (const char **)args+arg, err)) == NULL) {
1184 memprintf(err,
1185 "'%s %s %s' : error detected in %s '%s' while parsing '%s' condition : %s",
1186 args[0], args[1], args[2], proxy_type_str(curpx), curpx->id, args[arg], *err);
Emeric Brun97679e72010-09-23 17:56:44 +02001187 return -1;
1188 }
1189 }
1190 else if (*args[arg]) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001191 memprintf(err,
1192 "'%s %s %s' only accepts 'if' or 'unless', in %s '%s' (got '%s')",
Emeric Brun97679e72010-09-23 17:56:44 +02001193 args[0], args[1], args[2], proxy_type_str(curpx), curpx->id, args[arg]);
1194 return -1;
1195 }
1196 return 0;
1197}
1198
1199
1200
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001201/* Parse a tcp-request rule. Return a negative value in case of failure */
1202static int tcp_parse_request_rule(char **args, int arg, int section_type,
Willy Tarreau80aca902013-01-07 15:42:20 +01001203 struct proxy *curpx, struct proxy *defpx,
1204 struct tcp_rule *rule, char **err,
1205 unsigned int where)
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001206{
1207 if (curpx == defpx) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001208 memprintf(err, "%s %s is not allowed in 'defaults' sections",
1209 args[0], args[1]);
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001210 return -1;
1211 }
1212
1213 if (!strcmp(args[arg], "accept")) {
1214 arg++;
1215 rule->action = TCP_ACT_ACCEPT;
1216 }
1217 else if (!strcmp(args[arg], "reject")) {
1218 arg++;
1219 rule->action = TCP_ACT_REJECT;
1220 }
Willy Tarreaub4c84932013-07-23 19:15:30 +02001221 else if (strncmp(args[arg], "track-sc", 8) == 0 &&
1222 args[arg][9] == '\0' && args[arg][8] >= '0' &&
1223 args[arg][8] <= '0' + MAX_SESS_STKCTR) { /* track-sc 0..9 */
Willy Tarreau5d5b5d82012-12-09 12:00:04 +01001224 struct sample_expr *expr;
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001225 int kw = arg;
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001226
1227 arg++;
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001228
Willy Tarreaua4312fa2013-04-02 16:34:32 +02001229 curpx->conf.args.ctx = ARGC_TRK;
1230 expr = sample_parse_expr(args, &arg, trash.str, trash.size, &curpx->conf.args);
Willy Tarreau5d5b5d82012-12-09 12:00:04 +01001231 if (!expr) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001232 memprintf(err,
Willy Tarreau5d5b5d82012-12-09 12:00:04 +01001233 "'%s %s %s' : %s",
1234 args[0], args[1], args[kw], trash.str);
1235 return -1;
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001236 }
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001237
Willy Tarreau80aca902013-01-07 15:42:20 +01001238 if (!(expr->fetch->val & where)) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001239 memprintf(err,
Willy Tarreau80aca902013-01-07 15:42:20 +01001240 "'%s %s %s' : fetch method '%s' extracts information from '%s', none of which is available here",
Willy Tarreau33c60de2013-04-12 08:26:32 +02001241 args[0], args[1], args[kw], args[arg-1], sample_src_names(expr->fetch->use));
Willy Tarreau5d5b5d82012-12-09 12:00:04 +01001242 free(expr);
1243 return -1;
1244 }
1245
1246 /* check if we need to allocate an hdr_idx struct for HTTP parsing */
Willy Tarreau25320b22013-03-24 07:22:08 +01001247 curpx->http_needed |= !!(expr->fetch->use & SMP_USE_HTTP_ANY);
Willy Tarreau5d5b5d82012-12-09 12:00:04 +01001248
1249 if (strcmp(args[arg], "table") == 0) {
Willy Tarreau598718a2012-12-09 16:57:27 +01001250 arg++;
1251 if (!args[arg]) {
Willy Tarreau5d5b5d82012-12-09 12:00:04 +01001252 memprintf(err,
1253 "'%s %s %s' : missing table name",
1254 args[0], args[1], args[kw]);
1255 free(expr);
1256 return -1;
1257 }
1258 /* we copy the table name for now, it will be resolved later */
Willy Tarreau598718a2012-12-09 16:57:27 +01001259 rule->act_prm.trk_ctr.table.n = strdup(args[arg]);
Willy Tarreau5d5b5d82012-12-09 12:00:04 +01001260 arg++;
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001261 }
Willy Tarreau5d5b5d82012-12-09 12:00:04 +01001262 rule->act_prm.trk_ctr.expr = expr;
Willy Tarreaube4a3ef2013-06-17 15:04:07 +02001263 rule->action = TCP_ACT_TRK_SC0 + args[kw][8] - '0';
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001264 }
Willy Tarreau4f0d9192013-06-11 20:40:55 +02001265 else if (strcmp(args[arg], "expect-proxy") == 0) {
1266 if (strcmp(args[arg+1], "layer4") != 0) {
1267 memprintf(err,
1268 "'%s %s %s' only supports 'layer4' in %s '%s' (got '%s')",
1269 args[0], args[1], args[arg], proxy_type_str(curpx), curpx->id, args[arg+1]);
1270 return -1;
1271 }
1272
1273 if (!(where & SMP_VAL_FE_CON_ACC)) {
1274 memprintf(err,
1275 "'%s %s' is not allowed in '%s %s' rules in %s '%s'",
1276 args[arg], args[arg+1], args[0], args[1], proxy_type_str(curpx), curpx->id);
1277 return -1;
1278 }
1279
1280 arg += 2;
1281 rule->action = TCP_ACT_EXPECT_PX;
1282 }
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001283 else {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001284 memprintf(err,
Willy Tarreaub4c84932013-07-23 19:15:30 +02001285 "'%s %s' expects 'accept', 'reject', 'track-sc0' ... 'track-sc%d' "
1286 " in %s '%s' (got '%s')",
1287 args[0], args[1], MAX_SESS_STKCTR, proxy_type_str(curpx), curpx->id, args[arg]);
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001288 return -1;
1289 }
1290
1291 if (strcmp(args[arg], "if") == 0 || strcmp(args[arg], "unless") == 0) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001292 if ((rule->cond = build_acl_cond(NULL, 0, curpx, (const char **)args+arg, err)) == NULL) {
1293 memprintf(err,
1294 "'%s %s %s' : error detected in %s '%s' while parsing '%s' condition : %s",
1295 args[0], args[1], args[2], proxy_type_str(curpx), curpx->id, args[arg], *err);
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001296 return -1;
1297 }
1298 }
1299 else if (*args[arg]) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001300 memprintf(err,
1301 "'%s %s %s' only accepts 'if' or 'unless', in %s '%s' (got '%s')",
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001302 args[0], args[1], args[2], proxy_type_str(curpx), curpx->id, args[arg]);
1303 return -1;
1304 }
1305 return 0;
1306}
1307
Emeric Brun97679e72010-09-23 17:56:44 +02001308/* This function should be called to parse a line starting with the "tcp-response"
1309 * keyword.
1310 */
1311static int tcp_parse_tcp_rep(char **args, int section_type, struct proxy *curpx,
Willy Tarreau28a47d62012-09-18 20:02:48 +02001312 struct proxy *defpx, const char *file, int line,
1313 char **err)
Emeric Brun97679e72010-09-23 17:56:44 +02001314{
1315 const char *ptr = NULL;
1316 unsigned int val;
Emeric Brun97679e72010-09-23 17:56:44 +02001317 int warn = 0;
1318 int arg;
1319 struct tcp_rule *rule;
Willy Tarreau80aca902013-01-07 15:42:20 +01001320 unsigned int where;
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001321 const struct acl *acl;
Willy Tarreau93fddf12013-03-31 22:59:32 +02001322 const char *kw;
Emeric Brun97679e72010-09-23 17:56:44 +02001323
1324 if (!*args[1]) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001325 memprintf(err, "missing argument for '%s' in %s '%s'",
1326 args[0], proxy_type_str(curpx), curpx->id);
Emeric Brun97679e72010-09-23 17:56:44 +02001327 return -1;
1328 }
1329
1330 if (strcmp(args[1], "inspect-delay") == 0) {
1331 if (curpx == defpx || !(curpx->cap & PR_CAP_BE)) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001332 memprintf(err, "%s %s is only allowed in 'backend' sections",
1333 args[0], args[1]);
Emeric Brun97679e72010-09-23 17:56:44 +02001334 return -1;
1335 }
1336
1337 if (!*args[2] || (ptr = parse_time_err(args[2], &val, TIME_UNIT_MS))) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001338 memprintf(err,
1339 "'%s %s' expects a positive delay in milliseconds, in %s '%s'",
1340 args[0], args[1], proxy_type_str(curpx), curpx->id);
1341 if (ptr)
1342 memprintf(err, "%s (unexpected character '%c')", *err, *ptr);
Emeric Brun97679e72010-09-23 17:56:44 +02001343 return -1;
1344 }
1345
1346 if (curpx->tcp_rep.inspect_delay) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001347 memprintf(err, "ignoring %s %s (was already defined) in %s '%s'",
1348 args[0], args[1], proxy_type_str(curpx), curpx->id);
Emeric Brun97679e72010-09-23 17:56:44 +02001349 return 1;
1350 }
1351 curpx->tcp_rep.inspect_delay = val;
1352 return 0;
1353 }
1354
Simon Hormande072bd2011-06-24 15:11:37 +09001355 rule = calloc(1, sizeof(*rule));
Emeric Brun97679e72010-09-23 17:56:44 +02001356 LIST_INIT(&rule->list);
1357 arg = 1;
Willy Tarreau80aca902013-01-07 15:42:20 +01001358 where = 0;
Emeric Brun97679e72010-09-23 17:56:44 +02001359
1360 if (strcmp(args[1], "content") == 0) {
1361 arg++;
Willy Tarreau80aca902013-01-07 15:42:20 +01001362
1363 if (curpx->cap & PR_CAP_FE)
1364 where |= SMP_VAL_FE_RES_CNT;
1365 if (curpx->cap & PR_CAP_BE)
1366 where |= SMP_VAL_BE_RES_CNT;
1367
1368 if (tcp_parse_response_rule(args, arg, section_type, curpx, defpx, rule, err, where) < 0)
Emeric Brun97679e72010-09-23 17:56:44 +02001369 goto error;
1370
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001371 acl = rule->cond ? acl_cond_conflicts(rule->cond, where) : NULL;
1372 if (acl) {
1373 if (acl->name && *acl->name)
1374 memprintf(err,
1375 "acl '%s' will never match in '%s %s' because it only involves keywords that are incompatible with '%s'",
1376 acl->name, args[0], args[1], sample_ckp_names(where));
1377 else
1378 memprintf(err,
1379 "anonymous acl will never match in '%s %s' because it uses keyword '%s' which is incompatible with '%s'",
1380 args[0], args[1],
Willy Tarreau93fddf12013-03-31 22:59:32 +02001381 LIST_ELEM(acl->expr.n, struct acl_expr *, list)->kw,
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001382 sample_ckp_names(where));
Emeric Brun97679e72010-09-23 17:56:44 +02001383
Emeric Brun97679e72010-09-23 17:56:44 +02001384 warn++;
1385 }
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001386 else if (rule->cond && acl_cond_kw_conflicts(rule->cond, where, &acl, &kw)) {
1387 if (acl->name && *acl->name)
1388 memprintf(err,
1389 "acl '%s' involves keyword '%s' which is incompatible with '%s'",
Willy Tarreau93fddf12013-03-31 22:59:32 +02001390 acl->name, kw, sample_ckp_names(where));
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001391 else
1392 memprintf(err,
1393 "anonymous acl involves keyword '%s' which is incompatible with '%s'",
Willy Tarreau93fddf12013-03-31 22:59:32 +02001394 kw, sample_ckp_names(where));
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001395 warn++;
1396 }
Emeric Brun97679e72010-09-23 17:56:44 +02001397
1398 LIST_ADDQ(&curpx->tcp_rep.inspect_rules, &rule->list);
1399 }
1400 else {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001401 memprintf(err,
1402 "'%s' expects 'inspect-delay' or 'content' in %s '%s' (got '%s')",
1403 args[0], proxy_type_str(curpx), curpx->id, args[1]);
Emeric Brun97679e72010-09-23 17:56:44 +02001404 goto error;
1405 }
1406
1407 return warn;
1408 error:
1409 free(rule);
1410 return -1;
1411}
1412
1413
Willy Tarreaub6866442008-07-14 23:54:42 +02001414/* This function should be called to parse a line starting with the "tcp-request"
1415 * keyword.
1416 */
1417static int tcp_parse_tcp_req(char **args, int section_type, struct proxy *curpx,
Willy Tarreau28a47d62012-09-18 20:02:48 +02001418 struct proxy *defpx, const char *file, int line,
1419 char **err)
Willy Tarreaub6866442008-07-14 23:54:42 +02001420{
1421 const char *ptr = NULL;
Willy Tarreauc7e961e2008-08-17 17:13:47 +02001422 unsigned int val;
Willy Tarreau1a687942010-05-23 22:40:30 +02001423 int warn = 0;
Willy Tarreau6a984fa2010-06-14 16:44:27 +02001424 int arg;
Willy Tarreau1a687942010-05-23 22:40:30 +02001425 struct tcp_rule *rule;
Willy Tarreau80aca902013-01-07 15:42:20 +01001426 unsigned int where;
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001427 const struct acl *acl;
Willy Tarreau93fddf12013-03-31 22:59:32 +02001428 const char *kw;
Willy Tarreaub6866442008-07-14 23:54:42 +02001429
1430 if (!*args[1]) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001431 if (curpx == defpx)
1432 memprintf(err, "missing argument for '%s' in defaults section", args[0]);
1433 else
1434 memprintf(err, "missing argument for '%s' in %s '%s'",
1435 args[0], proxy_type_str(curpx), curpx->id);
Willy Tarreaub6866442008-07-14 23:54:42 +02001436 return -1;
1437 }
1438
1439 if (!strcmp(args[1], "inspect-delay")) {
1440 if (curpx == defpx) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001441 memprintf(err, "%s %s is not allowed in 'defaults' sections",
1442 args[0], args[1]);
Willy Tarreaub6866442008-07-14 23:54:42 +02001443 return -1;
1444 }
1445
Willy Tarreaub6866442008-07-14 23:54:42 +02001446 if (!*args[2] || (ptr = parse_time_err(args[2], &val, TIME_UNIT_MS))) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001447 memprintf(err,
1448 "'%s %s' expects a positive delay in milliseconds, in %s '%s'",
1449 args[0], args[1], proxy_type_str(curpx), curpx->id);
1450 if (ptr)
1451 memprintf(err, "%s (unexpected character '%c')", *err, *ptr);
Willy Tarreaub6866442008-07-14 23:54:42 +02001452 return -1;
1453 }
1454
1455 if (curpx->tcp_req.inspect_delay) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001456 memprintf(err, "ignoring %s %s (was already defined) in %s '%s'",
1457 args[0], args[1], proxy_type_str(curpx), curpx->id);
Willy Tarreaub6866442008-07-14 23:54:42 +02001458 return 1;
1459 }
1460 curpx->tcp_req.inspect_delay = val;
1461 return 0;
1462 }
1463
Simon Hormande072bd2011-06-24 15:11:37 +09001464 rule = calloc(1, sizeof(*rule));
Willy Tarreaufb024dc2010-08-20 13:35:41 +02001465 LIST_INIT(&rule->list);
Willy Tarreau6a984fa2010-06-14 16:44:27 +02001466 arg = 1;
Willy Tarreau80aca902013-01-07 15:42:20 +01001467 where = 0;
Willy Tarreau6a984fa2010-06-14 16:44:27 +02001468
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001469 if (strcmp(args[1], "content") == 0) {
Willy Tarreaud1f96522010-08-03 19:34:32 +02001470 arg++;
Willy Tarreau80aca902013-01-07 15:42:20 +01001471
1472 if (curpx->cap & PR_CAP_FE)
1473 where |= SMP_VAL_FE_REQ_CNT;
1474 if (curpx->cap & PR_CAP_BE)
1475 where |= SMP_VAL_BE_REQ_CNT;
1476
1477 if (tcp_parse_request_rule(args, arg, section_type, curpx, defpx, rule, err, where) < 0)
Willy Tarreau6a984fa2010-06-14 16:44:27 +02001478 goto error;
Willy Tarreaub6866442008-07-14 23:54:42 +02001479
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001480 acl = rule->cond ? acl_cond_conflicts(rule->cond, where) : NULL;
1481 if (acl) {
1482 if (acl->name && *acl->name)
1483 memprintf(err,
1484 "acl '%s' will never match in '%s %s' because it only involves keywords that are incompatible with '%s'",
1485 acl->name, args[0], args[1], sample_ckp_names(where));
1486 else
1487 memprintf(err,
1488 "anonymous acl will never match in '%s %s' because it uses keyword '%s' which is incompatible with '%s'",
1489 args[0], args[1],
Willy Tarreau93fddf12013-03-31 22:59:32 +02001490 LIST_ELEM(acl->expr.n, struct acl_expr *, list)->kw,
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001491 sample_ckp_names(where));
Willy Tarreaudd64f8d2008-07-27 22:02:32 +02001492
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001493 warn++;
1494 }
1495 else if (rule->cond && acl_cond_kw_conflicts(rule->cond, where, &acl, &kw)) {
1496 if (acl->name && *acl->name)
1497 memprintf(err,
1498 "acl '%s' involves keyword '%s' which is incompatible with '%s'",
Willy Tarreau93fddf12013-03-31 22:59:32 +02001499 acl->name, kw, sample_ckp_names(where));
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001500 else
1501 memprintf(err,
1502 "anonymous acl involves keyword '%s' which is incompatible with '%s'",
Willy Tarreau93fddf12013-03-31 22:59:32 +02001503 kw, sample_ckp_names(where));
Willy Tarreaudd64f8d2008-07-27 22:02:32 +02001504 warn++;
1505 }
Willy Tarreau5d5b5d82012-12-09 12:00:04 +01001506
Willy Tarreaufb024dc2010-08-20 13:35:41 +02001507 LIST_ADDQ(&curpx->tcp_req.inspect_rules, &rule->list);
Willy Tarreau9ba2dcc2010-06-14 21:04:55 +02001508 }
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001509 else if (strcmp(args[1], "connection") == 0) {
Willy Tarreau9ba2dcc2010-06-14 21:04:55 +02001510 arg++;
Willy Tarreau9ba2dcc2010-06-14 21:04:55 +02001511
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001512 if (!(curpx->cap & PR_CAP_FE)) {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001513 memprintf(err, "%s %s is not allowed because %s %s is not a frontend",
1514 args[0], args[1], proxy_type_str(curpx), curpx->id);
Simon Horman6c54d8b2011-07-15 13:14:06 +09001515 goto error;
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001516 }
Willy Tarreau9ba2dcc2010-06-14 21:04:55 +02001517
Willy Tarreau80aca902013-01-07 15:42:20 +01001518 where |= SMP_VAL_FE_CON_ACC;
1519
1520 if (tcp_parse_request_rule(args, arg, section_type, curpx, defpx, rule, err, where) < 0)
Willy Tarreau9ba2dcc2010-06-14 21:04:55 +02001521 goto error;
1522
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001523 acl = rule->cond ? acl_cond_conflicts(rule->cond, where) : NULL;
1524 if (acl) {
1525 if (acl->name && *acl->name)
1526 memprintf(err,
1527 "acl '%s' will never match in '%s %s' because it only involves keywords that are incompatible with '%s'",
1528 acl->name, args[0], args[1], sample_ckp_names(where));
1529 else
1530 memprintf(err,
1531 "anonymous acl will never match in '%s %s' because it uses keyword '%s' which is incompatible with '%s'",
1532 args[0], args[1],
Willy Tarreau93fddf12013-03-31 22:59:32 +02001533 LIST_ELEM(acl->expr.n, struct acl_expr *, list)->kw,
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001534 sample_ckp_names(where));
Willy Tarreauf059a0f2010-08-03 16:29:52 +02001535
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001536 warn++;
1537 }
1538 else if (rule->cond && acl_cond_kw_conflicts(rule->cond, where, &acl, &kw)) {
1539 if (acl->name && *acl->name)
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001540 memprintf(err,
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001541 "acl '%s' involves keyword '%s' which is incompatible with '%s'",
Willy Tarreau93fddf12013-03-31 22:59:32 +02001542 acl->name, kw, sample_ckp_names(where));
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001543 else
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001544 memprintf(err,
Willy Tarreaua91d0a52013-03-25 08:12:18 +01001545 "anonymous acl involves keyword '%s' which is incompatible with '%s'",
Willy Tarreau93fddf12013-03-31 22:59:32 +02001546 kw, sample_ckp_names(where));
Willy Tarreau68c03ab2010-08-06 15:08:45 +02001547 warn++;
1548 }
Willy Tarreau5d5b5d82012-12-09 12:00:04 +01001549
Willy Tarreaufb024dc2010-08-20 13:35:41 +02001550 LIST_ADDQ(&curpx->tcp_req.l4_rules, &rule->list);
Willy Tarreau9ba2dcc2010-06-14 21:04:55 +02001551 }
Willy Tarreau1a687942010-05-23 22:40:30 +02001552 else {
Willy Tarreau0a3dd742012-05-08 19:47:01 +02001553 if (curpx == defpx)
1554 memprintf(err,
1555 "'%s' expects 'inspect-delay', 'connection', or 'content' in defaults section (got '%s')",
1556 args[0], args[1]);
1557 else
1558 memprintf(err,
1559 "'%s' expects 'inspect-delay', 'connection', or 'content' in %s '%s' (got '%s')",
Willy Tarreaudeaec2f2013-04-10 16:31:11 +02001560 args[0], proxy_type_str(curpx), curpx->id, args[1]);
Willy Tarreau6a984fa2010-06-14 16:44:27 +02001561 goto error;
Willy Tarreau1a687942010-05-23 22:40:30 +02001562 }
1563
Willy Tarreau1a687942010-05-23 22:40:30 +02001564 return warn;
Willy Tarreau6a984fa2010-06-14 16:44:27 +02001565 error:
1566 free(rule);
1567 return -1;
Willy Tarreaub6866442008-07-14 23:54:42 +02001568}
1569
Willy Tarreau645513a2010-05-24 20:55:15 +02001570
1571/************************************************************************/
Willy Tarreaud4c33c82013-01-07 21:59:07 +01001572/* All supported sample fetch functions must be declared here */
Willy Tarreau32389b72012-04-23 23:13:20 +02001573/************************************************************************/
1574
Willy Tarreau4a129812012-04-25 17:31:42 +02001575/* fetch the connection's source IPv4/IPv6 address */
Willy Tarreau645513a2010-05-24 20:55:15 +02001576static int
Willy Tarreau4a129812012-04-25 17:31:42 +02001577smp_fetch_src(struct proxy *px, struct session *l4, void *l7, unsigned int opt,
Willy Tarreauef38c392013-07-22 16:29:32 +02001578 const struct arg *args, struct sample *smp, const char *kw)
Willy Tarreau645513a2010-05-24 20:55:15 +02001579{
Willy Tarreauf2943dc2012-10-26 20:10:28 +02001580 switch (l4->si[0].conn->addr.from.ss_family) {
Willy Tarreauf4362b32011-12-16 17:49:52 +01001581 case AF_INET:
Willy Tarreauf2943dc2012-10-26 20:10:28 +02001582 smp->data.ipv4 = ((struct sockaddr_in *)&l4->si[0].conn->addr.from)->sin_addr;
Willy Tarreauf853c462012-04-23 18:53:56 +02001583 smp->type = SMP_T_IPV4;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001584 break;
1585 case AF_INET6:
Willy Tarreauf2943dc2012-10-26 20:10:28 +02001586 smp->data.ipv6 = ((struct sockaddr_in6 *)(&l4->si[0].conn->addr.from))->sin6_addr;
Willy Tarreauf853c462012-04-23 18:53:56 +02001587 smp->type = SMP_T_IPV6;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001588 break;
1589 default:
Emeric Brunf769f512010-10-22 17:14:01 +02001590 return 0;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001591 }
Emeric Brunf769f512010-10-22 17:14:01 +02001592
Willy Tarreau37406352012-04-23 16:16:37 +02001593 smp->flags = 0;
Willy Tarreau645513a2010-05-24 20:55:15 +02001594 return 1;
1595}
1596
Willy Tarreaua5e37562011-12-16 17:06:15 +01001597/* set temp integer to the connection's source port */
Willy Tarreau645513a2010-05-24 20:55:15 +02001598static int
Willy Tarreau25c1ebc2012-04-25 16:21:44 +02001599smp_fetch_sport(struct proxy *px, struct session *l4, void *l7, unsigned int opt,
Willy Tarreauef38c392013-07-22 16:29:32 +02001600 const struct arg *args, struct sample *smp, const char *kw)
Willy Tarreau645513a2010-05-24 20:55:15 +02001601{
Willy Tarreauf853c462012-04-23 18:53:56 +02001602 smp->type = SMP_T_UINT;
Willy Tarreauf2943dc2012-10-26 20:10:28 +02001603 if (!(smp->data.uint = get_host_port(&l4->si[0].conn->addr.from)))
Emeric Brunf769f512010-10-22 17:14:01 +02001604 return 0;
1605
Willy Tarreau37406352012-04-23 16:16:37 +02001606 smp->flags = 0;
Willy Tarreau645513a2010-05-24 20:55:15 +02001607 return 1;
1608}
1609
Willy Tarreau4a129812012-04-25 17:31:42 +02001610/* fetch the connection's destination IPv4/IPv6 address */
Willy Tarreau645513a2010-05-24 20:55:15 +02001611static int
Willy Tarreau4a129812012-04-25 17:31:42 +02001612smp_fetch_dst(struct proxy *px, struct session *l4, void *l7, unsigned int opt,
Willy Tarreauef38c392013-07-22 16:29:32 +02001613 const struct arg *args, struct sample *smp, const char *kw)
Willy Tarreau645513a2010-05-24 20:55:15 +02001614{
Willy Tarreauf2943dc2012-10-26 20:10:28 +02001615 conn_get_to_addr(l4->si[0].conn);
Willy Tarreau645513a2010-05-24 20:55:15 +02001616
Willy Tarreauf2943dc2012-10-26 20:10:28 +02001617 switch (l4->si[0].conn->addr.to.ss_family) {
Willy Tarreauf4362b32011-12-16 17:49:52 +01001618 case AF_INET:
Willy Tarreauf2943dc2012-10-26 20:10:28 +02001619 smp->data.ipv4 = ((struct sockaddr_in *)&l4->si[0].conn->addr.to)->sin_addr;
Willy Tarreauf853c462012-04-23 18:53:56 +02001620 smp->type = SMP_T_IPV4;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001621 break;
1622 case AF_INET6:
Willy Tarreauf2943dc2012-10-26 20:10:28 +02001623 smp->data.ipv6 = ((struct sockaddr_in6 *)(&l4->si[0].conn->addr.to))->sin6_addr;
Willy Tarreauf853c462012-04-23 18:53:56 +02001624 smp->type = SMP_T_IPV6;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001625 break;
1626 default:
Emeric Brunf769f512010-10-22 17:14:01 +02001627 return 0;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001628 }
Emeric Brunf769f512010-10-22 17:14:01 +02001629
Willy Tarreau37406352012-04-23 16:16:37 +02001630 smp->flags = 0;
Willy Tarreau645513a2010-05-24 20:55:15 +02001631 return 1;
1632}
1633
Willy Tarreaua5e37562011-12-16 17:06:15 +01001634/* set temp integer to the frontend connexion's destination port */
Willy Tarreau645513a2010-05-24 20:55:15 +02001635static int
Willy Tarreau25c1ebc2012-04-25 16:21:44 +02001636smp_fetch_dport(struct proxy *px, struct session *l4, void *l7, unsigned int opt,
Willy Tarreauef38c392013-07-22 16:29:32 +02001637 const struct arg *args, struct sample *smp, const char *kw)
Willy Tarreau645513a2010-05-24 20:55:15 +02001638{
Willy Tarreauf2943dc2012-10-26 20:10:28 +02001639 conn_get_to_addr(l4->si[0].conn);
Willy Tarreau645513a2010-05-24 20:55:15 +02001640
Willy Tarreauf853c462012-04-23 18:53:56 +02001641 smp->type = SMP_T_UINT;
Willy Tarreauf2943dc2012-10-26 20:10:28 +02001642 if (!(smp->data.uint = get_host_port(&l4->si[0].conn->addr.to)))
Emeric Brunf769f512010-10-22 17:14:01 +02001643 return 0;
1644
Willy Tarreau37406352012-04-23 16:16:37 +02001645 smp->flags = 0;
Willy Tarreau645513a2010-05-24 20:55:15 +02001646 return 1;
Emericf2d7cae2010-11-05 18:13:50 +01001647}
1648
Willy Tarreau9b6700f2012-11-24 11:55:28 +01001649#ifdef IPV6_V6ONLY
Willy Tarreau77e3af92012-11-24 15:07:23 +01001650/* parse the "v4v6" bind keyword */
1651static int bind_parse_v4v6(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
1652{
1653 struct listener *l;
1654
1655 list_for_each_entry(l, &conf->listeners, by_bind) {
1656 if (l->addr.ss_family == AF_INET6)
1657 l->options |= LI_O_V4V6;
1658 }
1659
1660 return 0;
1661}
1662
Willy Tarreau9b6700f2012-11-24 11:55:28 +01001663/* parse the "v6only" bind keyword */
1664static int bind_parse_v6only(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
1665{
1666 struct listener *l;
1667
1668 list_for_each_entry(l, &conf->listeners, by_bind) {
1669 if (l->addr.ss_family == AF_INET6)
1670 l->options |= LI_O_V6ONLY;
1671 }
1672
1673 return 0;
1674}
1675#endif
1676
Pieter Baauwd551fb52013-05-08 22:49:23 +02001677#ifdef CONFIG_HAP_TRANSPARENT
Willy Tarreau44791242012-09-12 23:27:21 +02001678/* parse the "transparent" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +02001679static int bind_parse_transparent(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
Willy Tarreau44791242012-09-12 23:27:21 +02001680{
1681 struct listener *l;
1682
Willy Tarreau4348fad2012-09-20 16:48:07 +02001683 list_for_each_entry(l, &conf->listeners, by_bind) {
1684 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1685 l->options |= LI_O_FOREIGN;
Willy Tarreau44791242012-09-12 23:27:21 +02001686 }
1687
Willy Tarreau44791242012-09-12 23:27:21 +02001688 return 0;
1689}
1690#endif
1691
1692#ifdef TCP_DEFER_ACCEPT
1693/* parse the "defer-accept" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +02001694static int bind_parse_defer_accept(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
Willy Tarreau44791242012-09-12 23:27:21 +02001695{
1696 struct listener *l;
1697
Willy Tarreau4348fad2012-09-20 16:48:07 +02001698 list_for_each_entry(l, &conf->listeners, by_bind) {
1699 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1700 l->options |= LI_O_DEF_ACCEPT;
Willy Tarreau44791242012-09-12 23:27:21 +02001701 }
1702
Willy Tarreau44791242012-09-12 23:27:21 +02001703 return 0;
1704}
1705#endif
1706
Willy Tarreau1c862c52012-10-05 16:21:00 +02001707#ifdef TCP_FASTOPEN
Lukas Tribus0defb902013-02-13 23:35:39 +01001708/* parse the "tfo" bind keyword */
Willy Tarreau1c862c52012-10-05 16:21:00 +02001709static int bind_parse_tfo(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
1710{
1711 struct listener *l;
1712
1713 list_for_each_entry(l, &conf->listeners, by_bind) {
1714 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1715 l->options |= LI_O_TCP_FO;
1716 }
1717
1718 return 0;
1719}
1720#endif
1721
Willy Tarreau44791242012-09-12 23:27:21 +02001722#ifdef TCP_MAXSEG
1723/* parse the "mss" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +02001724static int bind_parse_mss(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
Willy Tarreau44791242012-09-12 23:27:21 +02001725{
1726 struct listener *l;
1727 int mss;
1728
Willy Tarreau44791242012-09-12 23:27:21 +02001729 if (!*args[cur_arg + 1]) {
Willy Tarreaueb6cead2012-09-20 19:43:14 +02001730 memprintf(err, "'%s' : missing MSS value", args[cur_arg]);
Willy Tarreau44791242012-09-12 23:27:21 +02001731 return ERR_ALERT | ERR_FATAL;
1732 }
1733
1734 mss = atoi(args[cur_arg + 1]);
1735 if (!mss || abs(mss) > 65535) {
Willy Tarreaueb6cead2012-09-20 19:43:14 +02001736 memprintf(err, "'%s' : expects an MSS with and absolute value between 1 and 65535", args[cur_arg]);
Willy Tarreau44791242012-09-12 23:27:21 +02001737 return ERR_ALERT | ERR_FATAL;
1738 }
1739
Willy Tarreau4348fad2012-09-20 16:48:07 +02001740 list_for_each_entry(l, &conf->listeners, by_bind) {
1741 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1742 l->maxseg = mss;
1743 }
Willy Tarreau44791242012-09-12 23:27:21 +02001744
1745 return 0;
1746}
1747#endif
1748
1749#ifdef SO_BINDTODEVICE
1750/* parse the "mss" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +02001751static int bind_parse_interface(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
Willy Tarreau44791242012-09-12 23:27:21 +02001752{
1753 struct listener *l;
1754
Willy Tarreau44791242012-09-12 23:27:21 +02001755 if (!*args[cur_arg + 1]) {
Willy Tarreaueb6cead2012-09-20 19:43:14 +02001756 memprintf(err, "'%s' : missing interface name", args[cur_arg]);
Willy Tarreau44791242012-09-12 23:27:21 +02001757 return ERR_ALERT | ERR_FATAL;
1758 }
1759
Willy Tarreau4348fad2012-09-20 16:48:07 +02001760 list_for_each_entry(l, &conf->listeners, by_bind) {
1761 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1762 l->interface = strdup(args[cur_arg + 1]);
1763 }
Willy Tarreau44791242012-09-12 23:27:21 +02001764
1765 global.last_checks |= LSTCHK_NETADM;
1766 return 0;
1767}
1768#endif
1769
Willy Tarreaudc13c112013-06-21 23:16:39 +02001770static struct cfg_kw_list cfg_kws = {ILH, {
Willy Tarreaud4c33c82013-01-07 21:59:07 +01001771 { CFG_LISTEN, "tcp-request", tcp_parse_tcp_req },
Emeric Brun97679e72010-09-23 17:56:44 +02001772 { CFG_LISTEN, "tcp-response", tcp_parse_tcp_rep },
Willy Tarreaub6866442008-07-14 23:54:42 +02001773 { 0, NULL, NULL },
1774}};
1775
Willy Tarreaud4c33c82013-01-07 21:59:07 +01001776
Willy Tarreau61612d42012-04-19 18:42:05 +02001777/* Note: must not be declared <const> as its list will be overwritten.
1778 * Please take care of keeping this list alphabetically sorted.
1779 */
Willy Tarreaudc13c112013-06-21 23:16:39 +02001780static struct acl_kw_list acl_kws = {ILH, {
Willy Tarreaud4c33c82013-01-07 21:59:07 +01001781 { /* END */ },
Willy Tarreaub6866442008-07-14 23:54:42 +02001782}};
1783
Willy Tarreaud4c33c82013-01-07 21:59:07 +01001784
Willy Tarreau4a129812012-04-25 17:31:42 +02001785/* Note: must not be declared <const> as its list will be overwritten.
1786 * Note: fetches that may return multiple types must be declared as the lowest
1787 * common denominator, the type that can be casted into all other ones. For
1788 * instance v4/v6 must be declared v4.
1789 */
Willy Tarreaudc13c112013-06-21 23:16:39 +02001790static struct sample_fetch_kw_list sample_fetch_keywords = {ILH, {
Willy Tarreaud4c33c82013-01-07 21:59:07 +01001791 { "dst", smp_fetch_dst, 0, NULL, SMP_T_IPV4, SMP_USE_L4CLI },
1792 { "dst_port", smp_fetch_dport, 0, NULL, SMP_T_UINT, SMP_USE_L4CLI },
1793 { "src", smp_fetch_src, 0, NULL, SMP_T_IPV4, SMP_USE_L4CLI },
1794 { "src_port", smp_fetch_sport, 0, NULL, SMP_T_UINT, SMP_USE_L4CLI },
1795 { /* END */ },
Willy Tarreau645513a2010-05-24 20:55:15 +02001796}};
1797
Willy Tarreau44791242012-09-12 23:27:21 +02001798/************************************************************************/
1799/* All supported bind keywords must be declared here. */
1800/************************************************************************/
1801
1802/* Note: must not be declared <const> as its list will be overwritten.
1803 * Please take care of keeping this list alphabetically sorted, doing so helps
1804 * all code contributors.
1805 * Optional keywords are also declared with a NULL ->parse() function so that
1806 * the config parser can report an appropriate error when a known keyword was
1807 * not enabled.
1808 */
Willy Tarreau51fb7652012-09-18 18:24:39 +02001809static struct bind_kw_list bind_kws = { "TCP", { }, {
Willy Tarreau44791242012-09-12 23:27:21 +02001810#ifdef TCP_DEFER_ACCEPT
1811 { "defer-accept", bind_parse_defer_accept, 0 }, /* wait for some data for 1 second max before doing accept */
1812#endif
1813#ifdef SO_BINDTODEVICE
1814 { "interface", bind_parse_interface, 1 }, /* specifically bind to this interface */
1815#endif
1816#ifdef TCP_MAXSEG
1817 { "mss", bind_parse_mss, 1 }, /* set MSS of listening socket */
1818#endif
Willy Tarreau1c862c52012-10-05 16:21:00 +02001819#ifdef TCP_FASTOPEN
1820 { "tfo", bind_parse_tfo, 0 }, /* enable TCP_FASTOPEN of listening socket */
1821#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +02001822#ifdef CONFIG_HAP_TRANSPARENT
Willy Tarreau44791242012-09-12 23:27:21 +02001823 { "transparent", bind_parse_transparent, 0 }, /* transparently bind to the specified addresses */
1824#endif
Willy Tarreau9b6700f2012-11-24 11:55:28 +01001825#ifdef IPV6_V6ONLY
Willy Tarreau77e3af92012-11-24 15:07:23 +01001826 { "v4v6", bind_parse_v4v6, 0 }, /* force socket to bind to IPv4+IPv6 */
Willy Tarreau9b6700f2012-11-24 11:55:28 +01001827 { "v6only", bind_parse_v6only, 0 }, /* force socket to bind to IPv6 only */
1828#endif
Willy Tarreau44791242012-09-12 23:27:21 +02001829 /* the versions with the NULL parse function*/
1830 { "defer-accept", NULL, 0 },
1831 { "interface", NULL, 1 },
1832 { "mss", NULL, 1 },
1833 { "transparent", NULL, 0 },
Willy Tarreau77e3af92012-11-24 15:07:23 +01001834 { "v4v6", NULL, 0 },
Willy Tarreau9b6700f2012-11-24 11:55:28 +01001835 { "v6only", NULL, 0 },
Willy Tarreau44791242012-09-12 23:27:21 +02001836 { NULL, NULL, 0 },
1837}};
1838
Willy Tarreaue6b98942007-10-29 01:09:36 +01001839__attribute__((constructor))
1840static void __tcp_protocol_init(void)
1841{
1842 protocol_register(&proto_tcpv4);
1843 protocol_register(&proto_tcpv6);
Willy Tarreau12785782012-04-27 21:37:17 +02001844 sample_register_fetches(&sample_fetch_keywords);
Willy Tarreaub6866442008-07-14 23:54:42 +02001845 cfg_register_keywords(&cfg_kws);
1846 acl_register_keywords(&acl_kws);
Willy Tarreau44791242012-09-12 23:27:21 +02001847 bind_register_keywords(&bind_kws);
Willy Tarreaue6b98942007-10-29 01:09:36 +01001848}
1849
1850
1851/*
1852 * Local variables:
1853 * c-indent-level: 8
1854 * c-basic-offset: 8
1855 * End:
1856 */