blob: 3dfc6a62f922b74a335a8b4e922eccd03896c818 [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
Baptiste Assmann39a5f222016-08-08 14:12:08 +020013/* this is to have tcp_info defined on systems using musl
14 * library, such as Alpine Linux
15 */
16#define _GNU_SOURCE
17
Willy Tarreaue6b98942007-10-29 01:09:36 +010018#include <ctype.h>
19#include <errno.h>
20#include <fcntl.h>
21#include <stdio.h>
22#include <stdlib.h>
23#include <string.h>
24#include <time.h>
25
26#include <sys/param.h>
27#include <sys/socket.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010028#include <sys/types.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010029
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040030#include <netinet/tcp.h>
Willy Tarreau2d392c22015-08-24 01:43:45 +020031#include <netinet/in.h>
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040032
Willy Tarreaue6b98942007-10-29 01:09:36 +010033#include <common/compat.h>
34#include <common/config.h>
35#include <common/debug.h>
36#include <common/errors.h>
Willy Tarreau0108d902018-11-25 19:14:37 +010037#include <common/initcall.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010038#include <common/mini-clist.h>
39#include <common/standard.h>
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +010040#include <common/namespace.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010041
Willy Tarreau39713102016-11-25 15:49:32 +010042#include <types/action.h>
Willy Tarreau2d392c22015-08-24 01:43:45 +020043#include <types/connection.h>
Willy Tarreau39713102016-11-25 15:49:32 +010044#include <types/global.h>
45#include <types/stream.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010046
Willy Tarreau9fcb9842012-04-20 14:45:49 +020047#include <proto/arg.h>
Willy Tarreauc7e42382012-08-24 19:22:53 +020048#include <proto/channel.h>
Willy Tarreaud2274c62012-07-06 14:29:45 +020049#include <proto/connection.h>
Willy Tarreaudd2f85e2012-09-02 22:34:23 +020050#include <proto/fd.h>
Willy Tarreau61c112a2018-10-02 16:43:32 +020051#include <proto/http_rules.h>
Willy Tarreaud1d54542012-09-12 22:58:11 +020052#include <proto/listener.h>
Willy Tarreau9650f372009-08-16 14:02:45 +020053#include <proto/log.h>
54#include <proto/port_range.h>
Willy Tarreaud1d54542012-09-12 22:58:11 +020055#include <proto/protocol.h>
Christopher Fauletfc9cfe42019-07-16 14:54:53 +020056#include <proto/http_ana.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010057#include <proto/proto_tcp.h>
Willy Tarreaub6866442008-07-14 23:54:42 +020058#include <proto/proxy.h>
Willy Tarreaucd3b0942012-04-27 21:52:18 +020059#include <proto/sample.h>
Willy Tarreau163d4622015-10-13 16:16:41 +020060#include <proto/server.h>
Willy Tarreaua975b8f2010-06-05 19:13:27 +020061#include <proto/task.h>
Willy Tarreau39713102016-11-25 15:49:32 +010062#include <proto/tcp_rules.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010063
Emeric Bruncf20bf12010-10-22 16:06:11 +020064static int tcp_bind_listeners(struct protocol *proto, char *errmsg, int errlen);
65static int tcp_bind_listener(struct listener *listener, char *errmsg, int errlen);
Willy Tarreau9d5be5c2017-09-15 07:55:51 +020066static void tcpv4_add_listener(struct listener *listener, int port);
67static void tcpv6_add_listener(struct listener *listener, int port);
Willy Tarreaue6b98942007-10-29 01:09:36 +010068
69/* Note: must not be declared <const> as its list will be overwritten */
70static struct protocol proto_tcpv4 = {
71 .name = "tcpv4",
72 .sock_domain = AF_INET,
73 .sock_type = SOCK_STREAM,
74 .sock_prot = IPPROTO_TCP,
75 .sock_family = AF_INET,
76 .sock_addrlen = sizeof(struct sockaddr_in),
77 .l3_addrlen = 32/8,
Willy Tarreaubbebbbf2012-05-07 21:22:09 +020078 .accept = &listener_accept,
Willy Tarreau26d8c592012-05-07 18:12:14 +020079 .connect = tcp_connect_server,
Emeric Bruncf20bf12010-10-22 16:06:11 +020080 .bind = tcp_bind_listener,
Willy Tarreaue6b98942007-10-29 01:09:36 +010081 .bind_all = tcp_bind_listeners,
82 .unbind_all = unbind_all_listeners,
83 .enable_all = enable_all_listeners,
Willy Tarreau59b94792012-05-11 16:16:40 +020084 .get_src = tcp_get_src,
85 .get_dst = tcp_get_dst,
Willy Tarreau092d8652014-07-07 20:22:12 +020086 .pause = tcp_pause_listener,
Willy Tarreau9d5be5c2017-09-15 07:55:51 +020087 .add = tcpv4_add_listener,
Willy Tarreaue6b98942007-10-29 01:09:36 +010088 .listeners = LIST_HEAD_INIT(proto_tcpv4.listeners),
89 .nb_listeners = 0,
90};
91
Willy Tarreau0108d902018-11-25 19:14:37 +010092INITCALL1(STG_REGISTER, protocol_register, &proto_tcpv4);
93
Willy Tarreaue6b98942007-10-29 01:09:36 +010094/* Note: must not be declared <const> as its list will be overwritten */
95static struct protocol proto_tcpv6 = {
96 .name = "tcpv6",
97 .sock_domain = AF_INET6,
98 .sock_type = SOCK_STREAM,
99 .sock_prot = IPPROTO_TCP,
100 .sock_family = AF_INET6,
101 .sock_addrlen = sizeof(struct sockaddr_in6),
102 .l3_addrlen = 128/8,
Willy Tarreaubbebbbf2012-05-07 21:22:09 +0200103 .accept = &listener_accept,
Willy Tarreau26d8c592012-05-07 18:12:14 +0200104 .connect = tcp_connect_server,
Emeric Bruncf20bf12010-10-22 16:06:11 +0200105 .bind = tcp_bind_listener,
Willy Tarreaue6b98942007-10-29 01:09:36 +0100106 .bind_all = tcp_bind_listeners,
107 .unbind_all = unbind_all_listeners,
108 .enable_all = enable_all_listeners,
Willy Tarreau59b94792012-05-11 16:16:40 +0200109 .get_src = tcp_get_src,
110 .get_dst = tcp_get_dst,
Willy Tarreau092d8652014-07-07 20:22:12 +0200111 .pause = tcp_pause_listener,
Willy Tarreau9d5be5c2017-09-15 07:55:51 +0200112 .add = tcpv6_add_listener,
Willy Tarreaue6b98942007-10-29 01:09:36 +0100113 .listeners = LIST_HEAD_INIT(proto_tcpv6.listeners),
114 .nb_listeners = 0,
115};
116
Willy Tarreau0108d902018-11-25 19:14:37 +0100117INITCALL1(STG_REGISTER, protocol_register, &proto_tcpv6);
118
Olivier Houchard153659f2017-04-05 22:39:56 +0200119/* Default TCP parameters, got by opening a temporary TCP socket. */
120#ifdef TCP_MAXSEG
Christopher Faulet1bc04c72017-10-29 20:14:08 +0100121static THREAD_LOCAL int default_tcp_maxseg = -1;
122static THREAD_LOCAL int default_tcp6_maxseg = -1;
Olivier Houchard153659f2017-04-05 22:39:56 +0200123#endif
124
David du Colombier6f5ccb12011-03-10 22:26:24 +0100125/* Binds ipv4/ipv6 address <local> to socket <fd>, unless <flags> is set, in which
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100126 * case we try to bind <remote>. <flags> is a 2-bit field consisting of :
127 * - 0 : ignore remote address (may even be a NULL pointer)
128 * - 1 : use provided address
129 * - 2 : use provided port
130 * - 3 : use both
131 *
132 * The function supports multiple foreign binding methods :
133 * - linux_tproxy: we directly bind to the foreign address
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100134 * The second one can be used as a fallback for the first one.
135 * This function returns 0 when everything's OK, 1 if it could not bind, to the
136 * local address, 2 if it could not bind to the foreign address.
137 */
David du Colombier6f5ccb12011-03-10 22:26:24 +0100138int tcp_bind_socket(int fd, int flags, struct sockaddr_storage *local, struct sockaddr_storage *remote)
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100139{
David du Colombier6f5ccb12011-03-10 22:26:24 +0100140 struct sockaddr_storage bind_addr;
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100141 int foreign_ok = 0;
142 int ret;
Christopher Faulet1bc04c72017-10-29 20:14:08 +0100143 static THREAD_LOCAL int ip_transp_working = 1;
144 static THREAD_LOCAL int ip6_transp_working = 1;
Pieter Baauwd551fb52013-05-08 22:49:23 +0200145
David du Colombier65c17962012-07-13 14:34:59 +0200146 switch (local->ss_family) {
147 case AF_INET:
148 if (flags && ip_transp_working) {
Pieter Baauwd551fb52013-05-08 22:49:23 +0200149 /* This deserves some explanation. Some platforms will support
150 * multiple combinations of certain methods, so we try the
151 * supported ones until one succeeds.
152 */
153 if (0
154#if defined(IP_TRANSPARENT)
155 || (setsockopt(fd, SOL_IP, IP_TRANSPARENT, &one, sizeof(one)) == 0)
156#endif
157#if defined(IP_FREEBIND)
158 || (setsockopt(fd, SOL_IP, IP_FREEBIND, &one, sizeof(one)) == 0)
159#endif
Pieter Baauwff30b662013-05-08 23:22:39 +0200160#if defined(IP_BINDANY)
161 || (setsockopt(fd, IPPROTO_IP, IP_BINDANY, &one, sizeof(one)) == 0)
162#endif
Pieter Baauw1eb75922013-05-08 23:30:23 +0200163#if defined(SO_BINDANY)
164 || (setsockopt(fd, SOL_SOCKET, SO_BINDANY, &one, sizeof(one)) == 0)
165#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +0200166 )
David du Colombier65c17962012-07-13 14:34:59 +0200167 foreign_ok = 1;
168 else
169 ip_transp_working = 0;
170 }
171 break;
172 case AF_INET6:
173 if (flags && ip6_transp_working) {
Pieter Baauwd551fb52013-05-08 22:49:23 +0200174 if (0
Dinko Korunic7276f3a2016-09-09 09:41:15 +0200175#if defined(IPV6_TRANSPARENT) && defined(SOL_IPV6)
Pieter Baauwd551fb52013-05-08 22:49:23 +0200176 || (setsockopt(fd, SOL_IPV6, IPV6_TRANSPARENT, &one, sizeof(one)) == 0)
177#endif
Lukas Tribus7640e722014-03-03 21:10:51 +0100178#if defined(IP_FREEBIND)
179 || (setsockopt(fd, SOL_IP, IP_FREEBIND, &one, sizeof(one)) == 0)
180#endif
Pieter Baauwff30b662013-05-08 23:22:39 +0200181#if defined(IPV6_BINDANY)
182 || (setsockopt(fd, IPPROTO_IPV6, IPV6_BINDANY, &one, sizeof(one)) == 0)
183#endif
Pieter Baauw1eb75922013-05-08 23:30:23 +0200184#if defined(SO_BINDANY)
185 || (setsockopt(fd, SOL_SOCKET, SO_BINDANY, &one, sizeof(one)) == 0)
186#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +0200187 )
David du Colombier65c17962012-07-13 14:34:59 +0200188 foreign_ok = 1;
189 else
190 ip6_transp_working = 0;
191 }
192 break;
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100193 }
Pieter Baauwd551fb52013-05-08 22:49:23 +0200194
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100195 if (flags) {
196 memset(&bind_addr, 0, sizeof(bind_addr));
Willy Tarreau96dd0792011-04-19 07:20:57 +0200197 bind_addr.ss_family = remote->ss_family;
David du Colombier6f5ccb12011-03-10 22:26:24 +0100198 switch (remote->ss_family) {
199 case AF_INET:
200 if (flags & 1)
201 ((struct sockaddr_in *)&bind_addr)->sin_addr = ((struct sockaddr_in *)remote)->sin_addr;
202 if (flags & 2)
203 ((struct sockaddr_in *)&bind_addr)->sin_port = ((struct sockaddr_in *)remote)->sin_port;
204 break;
205 case AF_INET6:
206 if (flags & 1)
207 ((struct sockaddr_in6 *)&bind_addr)->sin6_addr = ((struct sockaddr_in6 *)remote)->sin6_addr;
208 if (flags & 2)
209 ((struct sockaddr_in6 *)&bind_addr)->sin6_port = ((struct sockaddr_in6 *)remote)->sin6_port;
210 break;
Willy Tarreau5dc1e982011-12-16 21:25:11 +0100211 default:
212 /* we don't want to try to bind to an unknown address family */
213 foreign_ok = 0;
David du Colombier6f5ccb12011-03-10 22:26:24 +0100214 }
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100215 }
216
Simon Hormande072bd2011-06-24 15:11:37 +0900217 setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &one, sizeof(one));
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100218 if (foreign_ok) {
Willy Tarreau9cf8d3f2014-05-09 22:56:10 +0200219 if (is_inet_addr(&bind_addr)) {
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200220 ret = bind(fd, (struct sockaddr *)&bind_addr, get_addr_len(&bind_addr));
221 if (ret < 0)
222 return 2;
223 }
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100224 }
225 else {
Willy Tarreau9cf8d3f2014-05-09 22:56:10 +0200226 if (is_inet_addr(local)) {
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200227 ret = bind(fd, (struct sockaddr *)local, get_addr_len(local));
228 if (ret < 0)
229 return 1;
230 }
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100231 }
232
233 if (!flags)
234 return 0;
235
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100236 if (!foreign_ok)
237 /* we could not bind to a foreign address */
238 return 2;
239
240 return 0;
241}
Willy Tarreaue6b98942007-10-29 01:09:36 +0100242
Willy Tarreau7bbc4a52019-07-17 15:41:35 +0200243/* conn->dst MUST be valid */
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100244static int create_server_socket(struct connection *conn)
245{
Willy Tarreau529c1392014-12-24 13:47:55 +0100246 const struct netns_entry *ns = NULL;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100247
Willy Tarreaue5733232019-05-22 19:24:06 +0200248#ifdef USE_NS
Willy Tarreau529c1392014-12-24 13:47:55 +0100249 if (objt_server(conn->target)) {
250 if (__objt_server(conn->target)->flags & SRV_F_USE_NS_FROM_PP)
251 ns = conn->proxy_netns;
252 else
253 ns = __objt_server(conn->target)->netns;
254 }
255#endif
Willy Tarreau7bbc4a52019-07-17 15:41:35 +0200256 return my_socketat(ns, conn->dst->ss_family, SOCK_STREAM, IPPROTO_TCP);
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100257}
Willy Tarreau9650f372009-08-16 14:02:45 +0200258
259/*
Willy Tarreau14f8e862012-08-30 22:23:13 +0200260 * This function initiates a TCP connection establishment to the target assigned
Willy Tarreau7bbc4a52019-07-17 15:41:35 +0200261 * to connection <conn> using (si->{target,dst}). A source address may be
262 * pointed to by conn->src in case of transparent proxying. Normal source
Willy Tarreau14f8e862012-08-30 22:23:13 +0200263 * bind addresses are still determined locally (due to the possible need of a
264 * source port). conn->target may point either to a valid server or to a backend,
Willy Tarreau3fdb3662012-11-12 00:42:33 +0100265 * depending on conn->target. Only OBJ_TYPE_PROXY and OBJ_TYPE_SERVER are
Willy Tarreauf0837b22012-11-24 10:24:27 +0100266 * supported. The <data> parameter is a boolean indicating whether there are data
267 * waiting for being sent or not, in order to adjust data write polling and on
Olivier Houchardfdcb0072019-05-06 18:32:29 +0200268 * some platforms, the ability to avoid an empty initial ACK. The <flags> argument
269 * allows the caller to force using a delayed ACK when establishing the connection
Willy Tarreauf0837b22012-11-24 10:24:27 +0100270 * - 0 = no delayed ACK unless data are advertised and backend has tcp-smart-connect
Olivier Houchardfdcb0072019-05-06 18:32:29 +0200271 * - CONNECT_DELACK_SMART_CONNECT = delayed ACK if backend has tcp-smart-connect, regardless of data
272 * - CONNECT_DELACK_ALWAYS = delayed ACK regardless of backend options
Willy Tarreaub1d67742010-03-29 19:36:59 +0200273 *
Willy Tarreau1ec74bf2013-10-24 21:45:00 +0200274 * Note that a pending send_proxy message accounts for data.
275 *
Willy Tarreau9650f372009-08-16 14:02:45 +0200276 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +0200277 * - SF_ERR_NONE if everything's OK
278 * - SF_ERR_SRVTO if there are no more servers
279 * - SF_ERR_SRVCL if the connection was refused by the server
280 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
281 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
282 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +0100283 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Willy Tarreau6b0a8502012-11-23 08:51:32 +0100284 *
Willy Tarreaue7dff022015-04-03 01:14:29 +0200285 * The connection's fd is inserted only when SF_ERR_NONE is returned, otherwise
Willy Tarreau6b0a8502012-11-23 08:51:32 +0100286 * it's invalid and the caller has nothing to do.
Willy Tarreau9650f372009-08-16 14:02:45 +0200287 */
Willy Tarreauf1536862011-03-03 18:27:32 +0100288
Olivier Houchardfdcb0072019-05-06 18:32:29 +0200289int tcp_connect_server(struct connection *conn, int flags)
Willy Tarreau9650f372009-08-16 14:02:45 +0200290{
291 int fd;
Willy Tarreauac825402011-03-04 22:04:29 +0100292 struct server *srv;
293 struct proxy *be;
Willy Tarreaua4380b42012-12-08 22:49:11 +0100294 struct conn_src *src;
Willy Tarreau034c88c2017-01-23 23:36:45 +0100295 int use_fastopen = 0;
Alexander Liu2a54bb72019-05-22 19:44:48 +0800296 struct sockaddr_storage *addr;
Willy Tarreauac825402011-03-04 22:04:29 +0100297
Olivier Houchard637b6952018-11-23 14:23:07 +0100298 conn->flags |= CO_FL_WAIT_L4_CONN; /* connection in progress */
Willy Tarreau9ce70132014-01-24 16:08:19 +0100299
Willy Tarreau3fdb3662012-11-12 00:42:33 +0100300 switch (obj_type(conn->target)) {
301 case OBJ_TYPE_PROXY:
302 be = objt_proxy(conn->target);
Willy Tarreauac825402011-03-04 22:04:29 +0100303 srv = NULL;
304 break;
Willy Tarreau3fdb3662012-11-12 00:42:33 +0100305 case OBJ_TYPE_SERVER:
306 srv = objt_server(conn->target);
Willy Tarreauac825402011-03-04 22:04:29 +0100307 be = srv->proxy;
Willy Tarreau034c88c2017-01-23 23:36:45 +0100308 /* Make sure we check that we have data before activating
309 * TFO, or we could trigger a kernel issue whereby after
310 * a successful connect() == 0, any subsequent connect()
311 * will return EINPROGRESS instead of EISCONN.
312 */
313 use_fastopen = (srv->flags & SRV_F_FASTOPEN) &&
314 ((flags & (CONNECT_CAN_USE_TFO | CONNECT_HAS_DATA)) ==
315 (CONNECT_CAN_USE_TFO | CONNECT_HAS_DATA));
Willy Tarreauac825402011-03-04 22:04:29 +0100316 break;
317 default:
Willy Tarreau9ce70132014-01-24 16:08:19 +0100318 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200319 return SF_ERR_INTERNAL;
Willy Tarreauac825402011-03-04 22:04:29 +0100320 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200321
Willy Tarreau7bbc4a52019-07-17 15:41:35 +0200322 if (!conn->dst) {
323 conn->flags |= CO_FL_ERROR;
324 return SF_ERR_INTERNAL;
325 }
326
Willy Tarreau585744b2017-08-24 14:31:19 +0200327 fd = conn->handle.fd = create_server_socket(conn);
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100328
329 if (fd == -1) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200330 qfprintf(stderr, "Cannot get a server socket.\n");
331
Willy Tarreau9ce70132014-01-24 16:08:19 +0100332 if (errno == ENFILE) {
333 conn->err_code = CO_ER_SYS_FDLIM;
Willy Tarreau9650f372009-08-16 14:02:45 +0200334 send_log(be, LOG_EMERG,
Willy Tarreauc5532ac2018-01-29 15:06:04 +0100335 "Proxy %s reached system FD limit (maxsock=%d). Please check system tunables.\n",
336 be->id, global.maxsock);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100337 }
338 else if (errno == EMFILE) {
339 conn->err_code = CO_ER_PROC_FDLIM;
Willy Tarreau9650f372009-08-16 14:02:45 +0200340 send_log(be, LOG_EMERG,
Willy Tarreauc5532ac2018-01-29 15:06:04 +0100341 "Proxy %s reached process FD limit (maxsock=%d). Please check 'ulimit-n' and restart.\n",
342 be->id, global.maxsock);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100343 }
344 else if (errno == ENOBUFS || errno == ENOMEM) {
345 conn->err_code = CO_ER_SYS_MEMLIM;
Willy Tarreau9650f372009-08-16 14:02:45 +0200346 send_log(be, LOG_EMERG,
Willy Tarreauc5532ac2018-01-29 15:06:04 +0100347 "Proxy %s reached system memory limit (maxsock=%d). Please check system tunables.\n",
348 be->id, global.maxsock);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100349 }
350 else if (errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
351 conn->err_code = CO_ER_NOPROTO;
352 }
353 else
354 conn->err_code = CO_ER_SOCK_ERR;
355
Willy Tarreau9650f372009-08-16 14:02:45 +0200356 /* this is a resource error */
Willy Tarreau9ce70132014-01-24 16:08:19 +0100357 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200358 return SF_ERR_RESOURCE;
Willy Tarreau9650f372009-08-16 14:02:45 +0200359 }
360
361 if (fd >= global.maxsock) {
362 /* do not log anything there, it's a normal condition when this option
363 * is used to serialize connections to a server !
364 */
Christopher Faulet767a84b2017-11-24 16:50:31 +0100365 ha_alert("socket(): not enough free sockets. Raise -n argument. Giving up.\n");
Willy Tarreau9650f372009-08-16 14:02:45 +0200366 close(fd);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100367 conn->err_code = CO_ER_CONF_FDLIM;
368 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200369 return SF_ERR_PRXCOND; /* it is a configuration limit */
Willy Tarreau9650f372009-08-16 14:02:45 +0200370 }
371
372 if ((fcntl(fd, F_SETFL, O_NONBLOCK)==-1) ||
Simon Hormande072bd2011-06-24 15:11:37 +0900373 (setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &one, sizeof(one)) == -1)) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200374 qfprintf(stderr,"Cannot set client socket to non blocking mode.\n");
375 close(fd);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100376 conn->err_code = CO_ER_SOCK_ERR;
377 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200378 return SF_ERR_INTERNAL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200379 }
380
William Lallemandc03eb012018-11-27 12:02:37 +0100381 if (master == 1 && (fcntl(fd, F_SETFD, FD_CLOEXEC) == -1)) {
382 ha_alert("Cannot set CLOEXEC on client socket.\n");
383 close(fd);
384 conn->err_code = CO_ER_SOCK_ERR;
385 conn->flags |= CO_FL_ERROR;
386 return SF_ERR_INTERNAL;
387 }
388
Willy Tarreau9650f372009-08-16 14:02:45 +0200389 if (be->options & PR_O_TCP_SRV_KA)
Simon Hormande072bd2011-06-24 15:11:37 +0900390 setsockopt(fd, SOL_SOCKET, SO_KEEPALIVE, &one, sizeof(one));
Willy Tarreau9650f372009-08-16 14:02:45 +0200391
Willy Tarreau9650f372009-08-16 14:02:45 +0200392 /* allow specific binding :
393 * - server-specific at first
394 * - proxy-specific next
395 */
Willy Tarreaua4380b42012-12-08 22:49:11 +0100396 if (srv && srv->conn_src.opts & CO_SRC_BIND)
397 src = &srv->conn_src;
398 else if (be->conn_src.opts & CO_SRC_BIND)
399 src = &be->conn_src;
400 else
401 src = NULL;
402
403 if (src) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200404 int ret, flags = 0;
405
Willy Tarreau7bbc4a52019-07-17 15:41:35 +0200406 if (conn->src && is_inet_addr(conn->src)) {
Willy Tarreaua4380b42012-12-08 22:49:11 +0100407 switch (src->opts & CO_SRC_TPROXY_MASK) {
Willy Tarreauef9a3602012-12-08 22:29:20 +0100408 case CO_SRC_TPROXY_CLI:
Willy Tarreau387ebf82015-08-04 19:24:13 +0200409 conn->flags |= CO_FL_PRIVATE;
410 /* fall through */
411 case CO_SRC_TPROXY_ADDR:
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200412 flags = 3;
413 break;
Willy Tarreauef9a3602012-12-08 22:29:20 +0100414 case CO_SRC_TPROXY_CIP:
415 case CO_SRC_TPROXY_DYN:
Willy Tarreau387ebf82015-08-04 19:24:13 +0200416 conn->flags |= CO_FL_PRIVATE;
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200417 flags = 1;
418 break;
419 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200420 }
Willy Tarreaub1d67742010-03-29 19:36:59 +0200421
Willy Tarreau9650f372009-08-16 14:02:45 +0200422#ifdef SO_BINDTODEVICE
423 /* Note: this might fail if not CAP_NET_RAW */
Willy Tarreaua4380b42012-12-08 22:49:11 +0100424 if (src->iface_name)
425 setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, src->iface_name, src->iface_len + 1);
Willy Tarreau9650f372009-08-16 14:02:45 +0200426#endif
427
Willy Tarreaua4380b42012-12-08 22:49:11 +0100428 if (src->sport_range) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200429 int attempts = 10; /* should be more than enough to find a spare port */
Willy Tarreaua4380b42012-12-08 22:49:11 +0100430 struct sockaddr_storage sa;
Willy Tarreau9650f372009-08-16 14:02:45 +0200431
432 ret = 1;
Vincent Bernat6e615892016-05-18 16:17:44 +0200433 memcpy(&sa, &src->source_addr, sizeof(sa));
Willy Tarreau9650f372009-08-16 14:02:45 +0200434
435 do {
436 /* note: in case of retry, we may have to release a previously
437 * allocated port, hence this loop's construct.
438 */
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200439 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
440 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200441
442 if (!attempts)
443 break;
444 attempts--;
445
Willy Tarreaua4380b42012-12-08 22:49:11 +0100446 fdinfo[fd].local_port = port_range_alloc_port(src->sport_range);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100447 if (!fdinfo[fd].local_port) {
448 conn->err_code = CO_ER_PORT_RANGE;
Willy Tarreau9650f372009-08-16 14:02:45 +0200449 break;
Willy Tarreau9ce70132014-01-24 16:08:19 +0100450 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200451
Willy Tarreaua4380b42012-12-08 22:49:11 +0100452 fdinfo[fd].port_range = src->sport_range;
453 set_host_port(&sa, fdinfo[fd].local_port);
Willy Tarreau9650f372009-08-16 14:02:45 +0200454
Willy Tarreau7bbc4a52019-07-17 15:41:35 +0200455 ret = tcp_bind_socket(fd, flags, &sa, conn->src);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100456 if (ret != 0)
457 conn->err_code = CO_ER_CANT_BIND;
Willy Tarreau9650f372009-08-16 14:02:45 +0200458 } while (ret != 0); /* binding NOK */
459 }
460 else {
Lukas Tribus7d56c6d2016-09-13 09:51:15 +0000461#ifdef IP_BIND_ADDRESS_NO_PORT
Christopher Faulet1bc04c72017-10-29 20:14:08 +0100462 static THREAD_LOCAL int bind_address_no_port = 1;
Lukas Tribus7d56c6d2016-09-13 09:51:15 +0000463 setsockopt(fd, SOL_IP, IP_BIND_ADDRESS_NO_PORT, (const void *) &bind_address_no_port, sizeof(int));
464#endif
Willy Tarreau7bbc4a52019-07-17 15:41:35 +0200465 ret = tcp_bind_socket(fd, flags, &src->source_addr, conn->src);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100466 if (ret != 0)
467 conn->err_code = CO_ER_CANT_BIND;
Willy Tarreau9650f372009-08-16 14:02:45 +0200468 }
469
Willy Tarreaua4380b42012-12-08 22:49:11 +0100470 if (unlikely(ret != 0)) {
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200471 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
472 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200473 close(fd);
474
Willy Tarreau9650f372009-08-16 14:02:45 +0200475 if (ret == 1) {
Christopher Faulet767a84b2017-11-24 16:50:31 +0100476 ha_alert("Cannot bind to source address before connect() for backend %s. Aborting.\n",
477 be->id);
Willy Tarreau9650f372009-08-16 14:02:45 +0200478 send_log(be, LOG_EMERG,
Willy Tarreaua4380b42012-12-08 22:49:11 +0100479 "Cannot bind to source address before connect() for backend %s.\n",
Willy Tarreau9650f372009-08-16 14:02:45 +0200480 be->id);
481 } else {
Christopher Faulet767a84b2017-11-24 16:50:31 +0100482 ha_alert("Cannot bind to tproxy source address before connect() for backend %s. Aborting.\n",
483 be->id);
Willy Tarreau9650f372009-08-16 14:02:45 +0200484 send_log(be, LOG_EMERG,
Willy Tarreaua4380b42012-12-08 22:49:11 +0100485 "Cannot bind to tproxy source address before connect() for backend %s.\n",
Willy Tarreau9650f372009-08-16 14:02:45 +0200486 be->id);
487 }
Willy Tarreau9ce70132014-01-24 16:08:19 +0100488 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200489 return SF_ERR_RESOURCE;
Willy Tarreau9650f372009-08-16 14:02:45 +0200490 }
491 }
492
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +0400493#if defined(TCP_QUICKACK)
Willy Tarreau9650f372009-08-16 14:02:45 +0200494 /* disabling tcp quick ack now allows the first request to leave the
495 * machine with the first ACK. We only do this if there are pending
Willy Tarreauf0837b22012-11-24 10:24:27 +0100496 * data in the buffer.
Willy Tarreau9650f372009-08-16 14:02:45 +0200497 */
Olivier Houchardfdcb0072019-05-06 18:32:29 +0200498 if (flags & (CONNECT_DELACK_ALWAYS) ||
499 ((flags & CONNECT_DELACK_SMART_CONNECT ||
500 (flags & CONNECT_HAS_DATA) || conn->send_proxy_ofs) &&
501 (be->options2 & PR_O2_SMARTCON)))
Simon Hormande072bd2011-06-24 15:11:37 +0900502 setsockopt(fd, IPPROTO_TCP, TCP_QUICKACK, &zero, sizeof(zero));
Willy Tarreau9650f372009-08-16 14:02:45 +0200503#endif
504
Willy Tarreau163d4622015-10-13 16:16:41 +0200505#ifdef TCP_USER_TIMEOUT
506 /* there is not much more we can do here when it fails, it's still minor */
507 if (srv && srv->tcp_ut)
508 setsockopt(fd, IPPROTO_TCP, TCP_USER_TIMEOUT, &srv->tcp_ut, sizeof(srv->tcp_ut));
509#endif
Willy Tarreau034c88c2017-01-23 23:36:45 +0100510
511 if (use_fastopen) {
512#if defined(TCP_FASTOPEN_CONNECT)
513 setsockopt(fd, IPPROTO_TCP, TCP_FASTOPEN_CONNECT, &one, sizeof(one));
514#endif
515 }
Willy Tarreaue803de22010-01-21 17:43:04 +0100516 if (global.tune.server_sndbuf)
517 setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &global.tune.server_sndbuf, sizeof(global.tune.server_sndbuf));
518
519 if (global.tune.server_rcvbuf)
520 setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &global.tune.server_rcvbuf, sizeof(global.tune.server_rcvbuf));
521
Willy Tarreau7bbc4a52019-07-17 15:41:35 +0200522 addr = (conn->flags & CO_FL_SOCKS4) ? &srv->socks4_addr : conn->dst;
Alexander Liu2a54bb72019-05-22 19:44:48 +0800523 if (connect(fd, (const struct sockaddr *)addr, get_addr_len(addr)) == -1) {
Willy Tarreau819efbf2017-01-25 14:12:22 +0100524 if (errno == EINPROGRESS || errno == EALREADY) {
525 /* common case, let's wait for connect status */
526 conn->flags |= CO_FL_WAIT_L4_CONN;
527 }
528 else if (errno == EISCONN) {
529 /* should normally not happen but if so, indicates that it's OK */
530 conn->flags &= ~CO_FL_WAIT_L4_CONN;
531 }
532 else if (errno == EAGAIN || errno == EADDRINUSE || errno == EADDRNOTAVAIL) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200533 char *msg;
Willy Tarreau9ce70132014-01-24 16:08:19 +0100534 if (errno == EAGAIN || errno == EADDRNOTAVAIL) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200535 msg = "no free ports";
Willy Tarreau9ce70132014-01-24 16:08:19 +0100536 conn->err_code = CO_ER_FREE_PORTS;
537 }
538 else {
Willy Tarreau9650f372009-08-16 14:02:45 +0200539 msg = "local address already in use";
Willy Tarreau9ce70132014-01-24 16:08:19 +0100540 conn->err_code = CO_ER_ADDR_INUSE;
541 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200542
Willy Tarreaub1719512012-12-08 23:03:28 +0100543 qfprintf(stderr,"Connect() failed for backend %s: %s.\n", be->id, msg);
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200544 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
545 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200546 close(fd);
Willy Tarreaub1719512012-12-08 23:03:28 +0100547 send_log(be, LOG_ERR, "Connect() failed for backend %s: %s.\n", be->id, msg);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100548 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200549 return SF_ERR_RESOURCE;
Willy Tarreau9650f372009-08-16 14:02:45 +0200550 } else if (errno == ETIMEDOUT) {
551 //qfprintf(stderr,"Connect(): ETIMEDOUT");
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200552 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
553 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200554 close(fd);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100555 conn->err_code = CO_ER_SOCK_ERR;
556 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200557 return SF_ERR_SRVTO;
Willy Tarreau9650f372009-08-16 14:02:45 +0200558 } else {
559 // (errno == ECONNREFUSED || errno == ENETUNREACH || errno == EACCES || errno == EPERM)
560 //qfprintf(stderr,"Connect(): %d", errno);
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200561 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
562 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200563 close(fd);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100564 conn->err_code = CO_ER_SOCK_ERR;
565 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200566 return SF_ERR_SRVCL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200567 }
568 }
Willy Tarreau819efbf2017-01-25 14:12:22 +0100569 else {
570 /* connect() == 0, this is great! */
571 conn->flags &= ~CO_FL_WAIT_L4_CONN;
572 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200573
Willy Tarreaufc8f1f02012-12-08 18:53:44 +0100574 conn->flags |= CO_FL_ADDR_TO_SET;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200575
Willy Tarreauf79c8172013-10-21 16:30:56 +0200576 conn_ctrl_init(conn); /* registers the FD */
Willy Tarreauad38ace2013-12-15 14:19:38 +0100577 fdtab[fd].linger_risk = 1; /* close hard if needed */
Willy Tarreau15678ef2012-08-31 13:54:11 +0200578
Willy Tarreauccf3f6d2019-09-05 17:05:05 +0200579 if (conn->flags & CO_FL_WAIT_L4_CONN)
580 fd_cant_recv(fd); // we'll change this once the connection is validated
581
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200582 if (conn_xprt_init(conn) < 0) {
Willy Tarreau3f2770b2017-10-05 18:01:29 +0200583 conn_full_close(conn);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100584 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200585 return SF_ERR_RESOURCE;
Willy Tarreau184636e2012-09-06 14:04:41 +0200586 }
Willy Tarreau15678ef2012-08-31 13:54:11 +0200587
Olivier Houchard03abf2d2019-05-28 10:12:02 +0200588 conn_xprt_want_send(conn); /* for connect status, proxy protocol or SSL */
Willy Tarreaue7dff022015-04-03 01:14:29 +0200589 return SF_ERR_NONE; /* connection is OK */
Willy Tarreau9650f372009-08-16 14:02:45 +0200590}
591
592
Willy Tarreau59b94792012-05-11 16:16:40 +0200593/*
594 * Retrieves the source address for the socket <fd>, with <dir> indicating
595 * if we're a listener (=0) or an initiator (!=0). It returns 0 in case of
596 * success, -1 in case of error. The socket's source address is stored in
597 * <sa> for <salen> bytes.
598 */
599int tcp_get_src(int fd, struct sockaddr *sa, socklen_t salen, int dir)
600{
601 if (dir)
602 return getsockname(fd, sa, &salen);
603 else
604 return getpeername(fd, sa, &salen);
605}
606
607
608/*
609 * Retrieves the original destination address for the socket <fd>, with <dir>
610 * indicating if we're a listener (=0) or an initiator (!=0). In the case of a
611 * listener, if the original destination address was translated, the original
612 * address is retrieved. It returns 0 in case of success, -1 in case of error.
613 * The socket's source address is stored in <sa> for <salen> bytes.
614 */
615int tcp_get_dst(int fd, struct sockaddr *sa, socklen_t salen, int dir)
616{
617 if (dir)
618 return getpeername(fd, sa, &salen);
Willy Tarreau5e0d0e02014-10-29 21:46:01 +0100619 else {
620 int ret = getsockname(fd, sa, &salen);
621
622 if (ret < 0)
623 return ret;
624
Willy Tarreaue5733232019-05-22 19:24:06 +0200625#if defined(USE_TPROXY) && defined(SO_ORIGINAL_DST)
Willy Tarreau5e0d0e02014-10-29 21:46:01 +0100626 /* For TPROXY and Netfilter's NAT, we can retrieve the original
627 * IPv4 address before DNAT/REDIRECT. We must not do that with
628 * other families because v6-mapped IPv4 addresses are still
629 * reported as v4.
630 */
631 if (((struct sockaddr_storage *)sa)->ss_family == AF_INET
632 && getsockopt(fd, SOL_IP, SO_ORIGINAL_DST, sa, &salen) == 0)
633 return 0;
Willy Tarreau59b94792012-05-11 16:16:40 +0200634#endif
Willy Tarreau5e0d0e02014-10-29 21:46:01 +0100635 return ret;
636 }
Willy Tarreau59b94792012-05-11 16:16:40 +0200637}
638
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200639/* This is the callback which is set when a connection establishment is pending
Willy Tarreauf12a20e2013-12-04 16:11:04 +0100640 * and we have nothing to send. It updates the FD polling status. It returns 0
641 * if it fails in a fatal way or needs to poll to go further, otherwise it
642 * returns non-zero and removes the CO_FL_WAIT_L4_CONN flag from the connection's
643 * flags. In case of error, it sets CO_FL_ERROR and leaves the error code in
644 * errno. The error checking is done in two passes in order to limit the number
645 * of syscalls in the normal case :
646 * - if POLL_ERR was reported by the poller, we check for a pending error on
647 * the socket before proceeding. If found, it's assigned to errno so that
648 * upper layers can see it.
649 * - otherwise connect() is used to check the connection state again, since
650 * the getsockopt return cannot reliably be used to know if the connection
651 * is still pending or ready. This one may often return an error as well,
652 * since we don't always have POLL_ERR (eg: OSX or cached events).
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200653 */
Willy Tarreau239d7182012-07-23 18:53:03 +0200654int tcp_connect_probe(struct connection *conn)
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200655{
Alexander Liu2a54bb72019-05-22 19:44:48 +0800656 struct sockaddr_storage *addr;
Willy Tarreau585744b2017-08-24 14:31:19 +0200657 int fd = conn->handle.fd;
Willy Tarreauf12a20e2013-12-04 16:11:04 +0100658 socklen_t lskerr;
659 int skerr;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200660
Willy Tarreau80184712012-07-06 14:54:49 +0200661 if (conn->flags & CO_FL_ERROR)
Willy Tarreauafad0e02012-08-09 14:45:22 +0200662 return 0;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200663
Willy Tarreau3c728722014-01-23 13:50:42 +0100664 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200665 return 0;
666
Willy Tarreau80184712012-07-06 14:54:49 +0200667 if (!(conn->flags & CO_FL_WAIT_L4_CONN))
Willy Tarreau0b0c0972012-07-23 20:05:00 +0200668 return 1; /* strange we were called while ready */
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200669
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100670 if (!fd_send_ready(fd))
671 return 0;
672
Willy Tarreauf12a20e2013-12-04 16:11:04 +0100673 /* we might be the first witness of FD_POLL_ERR. Note that FD_POLL_HUP
674 * without FD_POLL_IN also indicates a hangup without input data meaning
675 * there was no connection.
676 */
677 if (fdtab[fd].ev & FD_POLL_ERR ||
678 (fdtab[fd].ev & (FD_POLL_IN|FD_POLL_HUP)) == FD_POLL_HUP) {
679 skerr = 0;
680 lskerr = sizeof(skerr);
681 getsockopt(fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr);
682 errno = skerr;
683 if (errno == EAGAIN)
684 errno = 0;
685 if (errno)
686 goto out_error;
687 }
Willy Tarreau2da156f2012-07-23 15:07:23 +0200688
Willy Tarreauf12a20e2013-12-04 16:11:04 +0100689 /* Use connect() to check the state of the socket. This has the
690 * advantage of giving us the following info :
Willy Tarreau2c6be842012-07-06 17:12:34 +0200691 * - error
692 * - connecting (EALREADY, EINPROGRESS)
693 * - connected (EISCONN, 0)
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200694 */
Willy Tarreau7bbc4a52019-07-17 15:41:35 +0200695 addr = conn->dst;
Alexander Liu2a54bb72019-05-22 19:44:48 +0800696 if ((conn->flags & CO_FL_SOCKS4) && obj_type(conn->target) == OBJ_TYPE_SERVER)
697 addr = &objt_server(conn->target)->socks4_addr;
698
699 if (connect(fd, (const struct sockaddr *)addr, get_addr_len(addr)) == -1) {
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200700 if (errno == EALREADY || errno == EINPROGRESS) {
Olivier Houchard7b3a79f2019-06-06 18:15:01 +0200701 __conn_xprt_want_send(conn);
Willy Tarreaue1f50c42014-01-22 20:02:06 +0100702 fd_cant_send(fd);
Willy Tarreau0b0c0972012-07-23 20:05:00 +0200703 return 0;
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200704 }
Willy Tarreaua190d592012-05-20 18:35:19 +0200705
Willy Tarreau2c6be842012-07-06 17:12:34 +0200706 if (errno && errno != EISCONN)
Willy Tarreaua190d592012-05-20 18:35:19 +0200707 goto out_error;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200708
Willy Tarreau2c6be842012-07-06 17:12:34 +0200709 /* otherwise we're connected */
Willy Tarreaua190d592012-05-20 18:35:19 +0200710 }
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200711
Willy Tarreau076be252012-07-06 16:02:29 +0200712 /* The FD is ready now, we'll mark the connection as complete and
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200713 * forward the event to the transport layer which will notify the
714 * data layer.
Willy Tarreaua190d592012-05-20 18:35:19 +0200715 */
Willy Tarreau80184712012-07-06 14:54:49 +0200716 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreauccf3f6d2019-09-05 17:05:05 +0200717 fd_may_send(fd);
718 fd_cond_recv(fd);
Willy Tarreau0b0c0972012-07-23 20:05:00 +0200719 return 1;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200720
721 out_error:
Willy Tarreau0b0c0972012-07-23 20:05:00 +0200722 /* Write error on the file descriptor. Report it to the connection
723 * and disable polling on this FD.
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200724 */
Willy Tarreau3bd3e572014-01-20 11:56:37 +0100725 fdtab[fd].linger_risk = 0;
Willy Tarreau26f4a042013-12-04 23:44:10 +0100726 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Olivier Houchard03abf2d2019-05-28 10:12:02 +0200727 __conn_xprt_stop_both(conn);
Willy Tarreauafad0e02012-08-09 14:45:22 +0200728 return 0;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200729}
730
Olivier Houchard153659f2017-04-05 22:39:56 +0200731/* XXX: Should probably be elsewhere */
732static int compare_sockaddr(struct sockaddr_storage *a, struct sockaddr_storage *b)
733{
734 if (a->ss_family != b->ss_family) {
735 return (-1);
736 }
737 switch (a->ss_family) {
738 case AF_INET:
739 {
740 struct sockaddr_in *a4 = (void *)a, *b4 = (void *)b;
741 if (a4->sin_port != b4->sin_port)
742 return (-1);
743 return (memcmp(&a4->sin_addr, &b4->sin_addr,
744 sizeof(a4->sin_addr)));
745 }
746 case AF_INET6:
747 {
748 struct sockaddr_in6 *a6 = (void *)a, *b6 = (void *)b;
749 if (a6->sin6_port != b6->sin6_port)
750 return (-1);
751 return (memcmp(&a6->sin6_addr, &b6->sin6_addr,
752 sizeof(a6->sin6_addr)));
753 }
754 default:
755 return (-1);
756 }
757
758}
759
760#define LI_MANDATORY_FLAGS (LI_O_FOREIGN | LI_O_V6ONLY | LI_O_V4V6)
761/* When binding the listeners, check if a socket has been sent to us by the
762 * previous process that we could reuse, instead of creating a new one.
763 */
764static int tcp_find_compatible_fd(struct listener *l)
765{
766 struct xfer_sock_list *xfer_sock = xfer_sock_list;
767 int ret = -1;
768
769 while (xfer_sock) {
770 if (!compare_sockaddr(&xfer_sock->addr, &l->addr)) {
771 if ((l->interface == NULL && xfer_sock->iface == NULL) ||
772 (l->interface != NULL && xfer_sock->iface != NULL &&
773 !strcmp(l->interface, xfer_sock->iface))) {
774 if ((l->options & LI_MANDATORY_FLAGS) ==
775 (xfer_sock->options & LI_MANDATORY_FLAGS)) {
776 if ((xfer_sock->namespace == NULL &&
777 l->netns == NULL)
Willy Tarreaue5733232019-05-22 19:24:06 +0200778#ifdef USE_NS
Olivier Houchard153659f2017-04-05 22:39:56 +0200779 || (xfer_sock->namespace != NULL &&
780 l->netns != NULL &&
781 !strcmp(xfer_sock->namespace,
782 l->netns->node.key))
783#endif
784 ) {
785 break;
786 }
787
788 }
789 }
790 }
791 xfer_sock = xfer_sock->next;
792 }
793 if (xfer_sock != NULL) {
794 ret = xfer_sock->fd;
795 if (xfer_sock == xfer_sock_list)
796 xfer_sock_list = xfer_sock->next;
797 if (xfer_sock->prev)
798 xfer_sock->prev->next = xfer_sock->next;
799 if (xfer_sock->next)
800 xfer_sock->next->prev = xfer_sock->prev;
801 free(xfer_sock->iface);
802 free(xfer_sock->namespace);
803 free(xfer_sock);
804 }
805 return ret;
806}
807#undef L1_MANDATORY_FLAGS
Willy Tarreau59b94792012-05-11 16:16:40 +0200808
Willy Tarreaue6b98942007-10-29 01:09:36 +0100809/* This function tries to bind a TCPv4/v6 listener. It may return a warning or
Willy Tarreau8ab505b2013-01-24 01:41:38 +0100810 * an error message in <errmsg> if the message is at most <errlen> bytes long
811 * (including '\0'). Note that <errmsg> may be NULL if <errlen> is also zero.
812 * The return value is composed from ERR_ABORT, ERR_WARN,
Willy Tarreaue6b98942007-10-29 01:09:36 +0100813 * ERR_ALERT, ERR_RETRYABLE and ERR_FATAL. ERR_NONE indicates that everything
814 * was alright and that no message was returned. ERR_RETRYABLE means that an
815 * error occurred but that it may vanish after a retry (eg: port in use), and
Aman Guptad94991d2012-04-06 17:39:26 -0700816 * ERR_FATAL indicates a non-fixable error. ERR_WARN and ERR_ALERT do not alter
Willy Tarreaue6b98942007-10-29 01:09:36 +0100817 * the meaning of the error, but just indicate that a message is present which
818 * should be displayed with the respective level. Last, ERR_ABORT indicates
819 * that it's pointless to try to start other listeners. No error message is
820 * returned if errlen is NULL.
821 */
822int tcp_bind_listener(struct listener *listener, char *errmsg, int errlen)
823{
824 __label__ tcp_return, tcp_close_return;
825 int fd, err;
Willy Tarreau40aa0702013-03-10 23:51:38 +0100826 int ext, ready;
827 socklen_t ready_len;
Willy Tarreaue6b98942007-10-29 01:09:36 +0100828 const char *msg = NULL;
Olivier Houchard153659f2017-04-05 22:39:56 +0200829#ifdef TCP_MAXSEG
830
831 /* Create a temporary TCP socket to get default parameters we can't
832 * guess.
833 * */
834 ready_len = sizeof(default_tcp_maxseg);
835 if (default_tcp_maxseg == -1) {
836 default_tcp_maxseg = -2;
837 fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP);
838 if (fd < 0)
Christopher Faulet767a84b2017-11-24 16:50:31 +0100839 ha_warning("Failed to create a temporary socket!\n");
Olivier Houchard153659f2017-04-05 22:39:56 +0200840 else {
841 if (getsockopt(fd, IPPROTO_TCP, TCP_MAXSEG, &default_tcp_maxseg,
842 &ready_len) == -1)
Christopher Faulet767a84b2017-11-24 16:50:31 +0100843 ha_warning("Failed to get the default value of TCP_MAXSEG\n");
Olivier Houchard153659f2017-04-05 22:39:56 +0200844 }
845 close(fd);
846 }
847 if (default_tcp6_maxseg == -1) {
848 default_tcp6_maxseg = -2;
849 fd = socket(AF_INET6, SOCK_STREAM, IPPROTO_TCP);
850 if (fd >= 0) {
851 if (getsockopt(fd, IPPROTO_TCP, TCP_MAXSEG, &default_tcp6_maxseg,
852 &ready_len) == -1)
Christopher Faulet767a84b2017-11-24 16:50:31 +0100853 ha_warning("Failed ot get the default value of TCP_MAXSEG for IPv6\n");
Olivier Houchard153659f2017-04-05 22:39:56 +0200854 close(fd);
855 }
856 }
857#endif
858
Willy Tarreaue6b98942007-10-29 01:09:36 +0100859
860 /* ensure we never return garbage */
Willy Tarreau8ab505b2013-01-24 01:41:38 +0100861 if (errlen)
Willy Tarreaue6b98942007-10-29 01:09:36 +0100862 *errmsg = 0;
863
864 if (listener->state != LI_ASSIGNED)
865 return ERR_NONE; /* already bound */
866
867 err = ERR_NONE;
868
Olivier Houchard153659f2017-04-05 22:39:56 +0200869 if (listener->fd == -1)
870 listener->fd = tcp_find_compatible_fd(listener);
871
Willy Tarreau40aa0702013-03-10 23:51:38 +0100872 /* if the listener already has an fd assigned, then we were offered the
873 * fd by an external process (most likely the parent), and we don't want
874 * to create a new socket. However we still want to set a few flags on
875 * the socket.
876 */
877 fd = listener->fd;
878 ext = (fd >= 0);
879
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100880 if (!ext) {
881 fd = my_socketat(listener->netns, listener->addr.ss_family, SOCK_STREAM, IPPROTO_TCP);
882
883 if (fd == -1) {
884 err |= ERR_RETRYABLE | ERR_ALERT;
885 msg = "cannot create listening socket";
886 goto tcp_return;
887 }
Willy Tarreaue6b98942007-10-29 01:09:36 +0100888 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100889
Willy Tarreaue6b98942007-10-29 01:09:36 +0100890 if (fd >= global.maxsock) {
891 err |= ERR_FATAL | ERR_ABORT | ERR_ALERT;
892 msg = "not enough free sockets (raise '-n' parameter)";
893 goto tcp_close_return;
894 }
895
Willy Tarreaufb14edc2009-06-14 15:24:37 +0200896 if (fcntl(fd, F_SETFL, O_NONBLOCK) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +0100897 err |= ERR_FATAL | ERR_ALERT;
898 msg = "cannot make socket non-blocking";
899 goto tcp_close_return;
900 }
901
Willy Tarreau40aa0702013-03-10 23:51:38 +0100902 if (!ext && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &one, sizeof(one)) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +0100903 /* not fatal but should be reported */
904 msg = "cannot do so_reuseaddr";
905 err |= ERR_ALERT;
906 }
907
908 if (listener->options & LI_O_NOLINGER)
Simon Hormande072bd2011-06-24 15:11:37 +0900909 setsockopt(fd, SOL_SOCKET, SO_LINGER, &nolinger, sizeof(struct linger));
Olivier Houchard153659f2017-04-05 22:39:56 +0200910 else {
911 struct linger tmplinger;
912 socklen_t len = sizeof(tmplinger);
913 if (getsockopt(fd, SOL_SOCKET, SO_LINGER, &tmplinger, &len) == 0 &&
914 (tmplinger.l_onoff == 1 || tmplinger.l_linger == 0)) {
915 tmplinger.l_onoff = 0;
916 tmplinger.l_linger = 0;
917 setsockopt(fd, SOL_SOCKET, SO_LINGER, &tmplinger,
918 sizeof(tmplinger));
919 }
920 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100921
Willy Tarreaue6b98942007-10-29 01:09:36 +0100922#ifdef SO_REUSEPORT
Lukas Tribusa0bcbdc2016-09-12 21:42:20 +0000923 /* OpenBSD and Linux 3.9 support this. As it's present in old libc versions of
924 * Linux, it might return an error that we will silently ignore.
Willy Tarreaue6b98942007-10-29 01:09:36 +0100925 */
Lukas Tribusa0bcbdc2016-09-12 21:42:20 +0000926 if (!ext && (global.tune.options & GTUNE_USE_REUSEPORT))
Willy Tarreau40aa0702013-03-10 23:51:38 +0100927 setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &one, sizeof(one));
Willy Tarreaue6b98942007-10-29 01:09:36 +0100928#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +0200929
Willy Tarreau40aa0702013-03-10 23:51:38 +0100930 if (!ext && (listener->options & LI_O_FOREIGN)) {
David du Colombier65c17962012-07-13 14:34:59 +0200931 switch (listener->addr.ss_family) {
932 case AF_INET:
Pieter Baauwd551fb52013-05-08 22:49:23 +0200933 if (1
934#if defined(IP_TRANSPARENT)
935 && (setsockopt(fd, SOL_IP, IP_TRANSPARENT, &one, sizeof(one)) == -1)
936#endif
937#if defined(IP_FREEBIND)
938 && (setsockopt(fd, SOL_IP, IP_FREEBIND, &one, sizeof(one)) == -1)
939#endif
Pieter Baauwff30b662013-05-08 23:22:39 +0200940#if defined(IP_BINDANY)
941 && (setsockopt(fd, IPPROTO_IP, IP_BINDANY, &one, sizeof(one)) == -1)
942#endif
Pieter Baauw1eb75922013-05-08 23:30:23 +0200943#if defined(SO_BINDANY)
944 && (setsockopt(fd, SOL_SOCKET, SO_BINDANY, &one, sizeof(one)) == -1)
945#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +0200946 ) {
David du Colombier65c17962012-07-13 14:34:59 +0200947 msg = "cannot make listening socket transparent";
948 err |= ERR_ALERT;
949 }
950 break;
951 case AF_INET6:
Pieter Baauwd551fb52013-05-08 22:49:23 +0200952 if (1
Dinko Korunic7276f3a2016-09-09 09:41:15 +0200953#if defined(IPV6_TRANSPARENT) && defined(SOL_IPV6)
Pieter Baauwd551fb52013-05-08 22:49:23 +0200954 && (setsockopt(fd, SOL_IPV6, IPV6_TRANSPARENT, &one, sizeof(one)) == -1)
955#endif
Lukas Tribus7640e722014-03-03 21:10:51 +0100956#if defined(IP_FREEBIND)
957 && (setsockopt(fd, SOL_IP, IP_FREEBIND, &one, sizeof(one)) == -1)
958#endif
Pieter Baauwff30b662013-05-08 23:22:39 +0200959#if defined(IPV6_BINDANY)
960 && (setsockopt(fd, IPPROTO_IPV6, IPV6_BINDANY, &one, sizeof(one)) == -1)
961#endif
Pieter Baauw1eb75922013-05-08 23:30:23 +0200962#if defined(SO_BINDANY)
963 && (setsockopt(fd, SOL_SOCKET, SO_BINDANY, &one, sizeof(one)) == -1)
964#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +0200965 ) {
David du Colombier65c17962012-07-13 14:34:59 +0200966 msg = "cannot make listening socket transparent";
967 err |= ERR_ALERT;
968 }
969 break;
970 }
Willy Tarreaub1e52e82008-01-13 14:49:51 +0100971 }
Pieter Baauwd551fb52013-05-08 22:49:23 +0200972
Willy Tarreau5e6e2042009-02-04 17:19:29 +0100973#ifdef SO_BINDTODEVICE
974 /* Note: this might fail if not CAP_NET_RAW */
Willy Tarreau40aa0702013-03-10 23:51:38 +0100975 if (!ext && listener->interface) {
Willy Tarreau5e6e2042009-02-04 17:19:29 +0100976 if (setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE,
Willy Tarreau604e8302009-03-06 00:48:23 +0100977 listener->interface, strlen(listener->interface) + 1) == -1) {
Willy Tarreau5e6e2042009-02-04 17:19:29 +0100978 msg = "cannot bind listener to device";
979 err |= ERR_WARN;
980 }
981 }
982#endif
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +0400983#if defined(TCP_MAXSEG)
Willy Tarreau48a7e722010-12-24 15:26:39 +0100984 if (listener->maxseg > 0) {
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +0400985 if (setsockopt(fd, IPPROTO_TCP, TCP_MAXSEG,
Willy Tarreaube1b9182009-06-14 18:48:19 +0200986 &listener->maxseg, sizeof(listener->maxseg)) == -1) {
987 msg = "cannot set MSS";
988 err |= ERR_WARN;
989 }
Olivier Houchard153659f2017-04-05 22:39:56 +0200990 } else if (ext) {
991 int tmpmaxseg = -1;
992 int defaultmss;
993 socklen_t len = sizeof(tmpmaxseg);
994
995 if (listener->addr.ss_family == AF_INET)
996 defaultmss = default_tcp_maxseg;
997 else
998 defaultmss = default_tcp6_maxseg;
999
1000 getsockopt(fd, IPPROTO_TCP, TCP_MAXSEG, &tmpmaxseg, &len);
1001 if (tmpmaxseg != defaultmss && setsockopt(fd, IPPROTO_TCP,
1002 TCP_MAXSEG, &defaultmss,
1003 sizeof(defaultmss)) == -1) {
1004 msg = "cannot set MSS";
1005 err |= ERR_WARN;
1006 }
Willy Tarreaube1b9182009-06-14 18:48:19 +02001007 }
1008#endif
Willy Tarreau2af207a2015-02-04 00:45:58 +01001009#if defined(TCP_USER_TIMEOUT)
1010 if (listener->tcp_ut) {
1011 if (setsockopt(fd, IPPROTO_TCP, TCP_USER_TIMEOUT,
1012 &listener->tcp_ut, sizeof(listener->tcp_ut)) == -1) {
1013 msg = "cannot set TCP User Timeout";
1014 err |= ERR_WARN;
1015 }
Olivier Houchard153659f2017-04-05 22:39:56 +02001016 } else
1017 setsockopt(fd, IPPROTO_TCP, TCP_USER_TIMEOUT, &zero,
1018 sizeof(zero));
Willy Tarreau2af207a2015-02-04 00:45:58 +01001019#endif
Willy Tarreaucb6cd432009-10-13 07:34:14 +02001020#if defined(TCP_DEFER_ACCEPT)
1021 if (listener->options & LI_O_DEF_ACCEPT) {
1022 /* defer accept by up to one second */
1023 int accept_delay = 1;
1024 if (setsockopt(fd, IPPROTO_TCP, TCP_DEFER_ACCEPT, &accept_delay, sizeof(accept_delay)) == -1) {
1025 msg = "cannot enable DEFER_ACCEPT";
1026 err |= ERR_WARN;
1027 }
Olivier Houchard153659f2017-04-05 22:39:56 +02001028 } else
1029 setsockopt(fd, IPPROTO_TCP, TCP_DEFER_ACCEPT, &zero,
1030 sizeof(zero));
Willy Tarreaucb6cd432009-10-13 07:34:14 +02001031#endif
Willy Tarreau1c862c52012-10-05 16:21:00 +02001032#if defined(TCP_FASTOPEN)
1033 if (listener->options & LI_O_TCP_FO) {
1034 /* TFO needs a queue length, let's use the configured backlog */
Willy Tarreaue2711c72019-02-27 15:39:41 +01001035 int qlen = listener_backlog(listener);
Willy Tarreau1c862c52012-10-05 16:21:00 +02001036 if (setsockopt(fd, IPPROTO_TCP, TCP_FASTOPEN, &qlen, sizeof(qlen)) == -1) {
1037 msg = "cannot enable TCP_FASTOPEN";
1038 err |= ERR_WARN;
1039 }
Olivier Houchard153659f2017-04-05 22:39:56 +02001040 } else {
1041 socklen_t len;
1042 int qlen;
1043 len = sizeof(qlen);
1044 /* Only disable fast open if it was enabled, we don't want
1045 * the kernel to create a fast open queue if there's none.
1046 */
1047 if (getsockopt(fd, IPPROTO_TCP, TCP_FASTOPEN, &qlen, &len) == 0 &&
1048 qlen != 0) {
1049 if (setsockopt(fd, IPPROTO_TCP, TCP_FASTOPEN, &zero,
1050 sizeof(zero)) == -1) {
1051 msg = "cannot disable TCP_FASTOPEN";
1052 err |= ERR_WARN;
1053 }
1054 }
Willy Tarreau1c862c52012-10-05 16:21:00 +02001055 }
1056#endif
Willy Tarreau9b6700f2012-11-24 11:55:28 +01001057#if defined(IPV6_V6ONLY)
1058 if (listener->options & LI_O_V6ONLY)
1059 setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &one, sizeof(one));
Willy Tarreau77e3af92012-11-24 15:07:23 +01001060 else if (listener->options & LI_O_V4V6)
1061 setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &zero, sizeof(zero));
Willy Tarreau9b6700f2012-11-24 11:55:28 +01001062#endif
1063
Willy Tarreau40aa0702013-03-10 23:51:38 +01001064 if (!ext && bind(fd, (struct sockaddr *)&listener->addr, listener->proto->sock_addrlen) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +01001065 err |= ERR_RETRYABLE | ERR_ALERT;
1066 msg = "cannot bind socket";
1067 goto tcp_close_return;
1068 }
Willy Tarreauedcf6682008-11-30 23:15:34 +01001069
Willy Tarreau40aa0702013-03-10 23:51:38 +01001070 ready = 0;
1071 ready_len = sizeof(ready);
1072 if (getsockopt(fd, SOL_SOCKET, SO_ACCEPTCONN, &ready, &ready_len) == -1)
1073 ready = 0;
1074
1075 if (!(ext && ready) && /* only listen if not already done by external process */
Willy Tarreaue2711c72019-02-27 15:39:41 +01001076 listen(fd, listener_backlog(listener)) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +01001077 err |= ERR_RETRYABLE | ERR_ALERT;
1078 msg = "cannot listen to socket";
1079 goto tcp_close_return;
1080 }
Willy Tarreauedcf6682008-11-30 23:15:34 +01001081
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +04001082#if defined(TCP_QUICKACK)
Willy Tarreau9ea05a72009-06-14 12:07:01 +02001083 if (listener->options & LI_O_NOQUICKACK)
Simon Hormande072bd2011-06-24 15:11:37 +09001084 setsockopt(fd, IPPROTO_TCP, TCP_QUICKACK, &zero, sizeof(zero));
Olivier Houchard153659f2017-04-05 22:39:56 +02001085 else
1086 setsockopt(fd, IPPROTO_TCP, TCP_QUICKACK, &one, sizeof(one));
Willy Tarreau9ea05a72009-06-14 12:07:01 +02001087#endif
1088
Willy Tarreaue6b98942007-10-29 01:09:36 +01001089 /* the socket is ready */
1090 listener->fd = fd;
1091 listener->state = LI_LISTEN;
1092
Willy Tarreaua9786b62018-01-25 07:22:13 +01001093 fd_insert(fd, listener, listener->proto->accept,
Willy Tarreaua36b3242019-02-02 13:14:34 +01001094 thread_mask(listener->bind_conf->bind_thread));
Willy Tarreaueb472682010-05-28 18:46:57 +02001095
Willy Tarreaue6b98942007-10-29 01:09:36 +01001096 tcp_return:
Cyril Bonté43ba1b32010-11-01 19:26:01 +01001097 if (msg && errlen) {
1098 char pn[INET6_ADDRSTRLEN];
1099
Willy Tarreau631f01c2011-09-05 00:36:48 +02001100 addr_to_str(&listener->addr, pn, sizeof(pn));
1101 snprintf(errmsg, errlen, "%s [%s:%d]", msg, pn, get_host_port(&listener->addr));
Cyril Bonté43ba1b32010-11-01 19:26:01 +01001102 }
Willy Tarreaue6b98942007-10-29 01:09:36 +01001103 return err;
1104
1105 tcp_close_return:
1106 close(fd);
1107 goto tcp_return;
1108}
1109
1110/* This function creates all TCP sockets bound to the protocol entry <proto>.
1111 * It is intended to be used as the protocol's bind_all() function.
1112 * The sockets will be registered but not added to any fd_set, in order not to
1113 * loose them across the fork(). A call to enable_all_listeners() is needed
1114 * to complete initialization. The return value is composed from ERR_*.
Willy Tarreaudaacf362019-07-24 16:45:02 +02001115 *
1116 * Must be called with proto_lock held.
1117 *
Willy Tarreaue6b98942007-10-29 01:09:36 +01001118 */
Emeric Bruncf20bf12010-10-22 16:06:11 +02001119static int tcp_bind_listeners(struct protocol *proto, char *errmsg, int errlen)
Willy Tarreaue6b98942007-10-29 01:09:36 +01001120{
1121 struct listener *listener;
1122 int err = ERR_NONE;
1123
1124 list_for_each_entry(listener, &proto->listeners, proto_list) {
Emeric Bruncf20bf12010-10-22 16:06:11 +02001125 err |= tcp_bind_listener(listener, errmsg, errlen);
1126 if (err & ERR_ABORT)
Willy Tarreaue6b98942007-10-29 01:09:36 +01001127 break;
1128 }
1129
1130 return err;
1131}
1132
Willy Tarreau32282382017-09-15 07:44:44 +02001133/* Add <listener> to the list of tcpv4 listeners, on port <port>. The
1134 * listener's state is automatically updated from LI_INIT to LI_ASSIGNED.
1135 * The number of listeners for the protocol is updated.
Willy Tarreaudaacf362019-07-24 16:45:02 +02001136 *
1137 * Must be called with proto_lock held.
1138 *
Willy Tarreaue6b98942007-10-29 01:09:36 +01001139 */
Willy Tarreau9d5be5c2017-09-15 07:55:51 +02001140static void tcpv4_add_listener(struct listener *listener, int port)
Willy Tarreaue6b98942007-10-29 01:09:36 +01001141{
1142 if (listener->state != LI_INIT)
1143 return;
1144 listener->state = LI_ASSIGNED;
1145 listener->proto = &proto_tcpv4;
Willy Tarreau32282382017-09-15 07:44:44 +02001146 ((struct sockaddr_in *)(&listener->addr))->sin_port = htons(port);
Willy Tarreaue6b98942007-10-29 01:09:36 +01001147 LIST_ADDQ(&proto_tcpv4.listeners, &listener->proto_list);
1148 proto_tcpv4.nb_listeners++;
1149}
1150
Willy Tarreau32282382017-09-15 07:44:44 +02001151/* Add <listener> to the list of tcpv6 listeners, on port <port>. The
1152 * listener's state is automatically updated from LI_INIT to LI_ASSIGNED.
1153 * The number of listeners for the protocol is updated.
Willy Tarreaudaacf362019-07-24 16:45:02 +02001154 *
1155 * Must be called with proto_lock held.
1156 *
Willy Tarreaue6b98942007-10-29 01:09:36 +01001157 */
Willy Tarreau9d5be5c2017-09-15 07:55:51 +02001158static void tcpv6_add_listener(struct listener *listener, int port)
Willy Tarreaue6b98942007-10-29 01:09:36 +01001159{
1160 if (listener->state != LI_INIT)
1161 return;
1162 listener->state = LI_ASSIGNED;
1163 listener->proto = &proto_tcpv6;
Willy Tarreau32282382017-09-15 07:44:44 +02001164 ((struct sockaddr_in *)(&listener->addr))->sin_port = htons(port);
Willy Tarreaue6b98942007-10-29 01:09:36 +01001165 LIST_ADDQ(&proto_tcpv6.listeners, &listener->proto_list);
1166 proto_tcpv6.nb_listeners++;
1167}
1168
Willy Tarreau092d8652014-07-07 20:22:12 +02001169/* Pause a listener. Returns < 0 in case of failure, 0 if the listener
1170 * was totally stopped, or > 0 if correctly paused.
1171 */
1172int tcp_pause_listener(struct listener *l)
1173{
1174 if (shutdown(l->fd, SHUT_WR) != 0)
1175 return -1; /* Solaris dies here */
1176
Willy Tarreaue2711c72019-02-27 15:39:41 +01001177 if (listen(l->fd, listener_backlog(l)) != 0)
Willy Tarreau092d8652014-07-07 20:22:12 +02001178 return -1; /* OpenBSD dies here */
1179
1180 if (shutdown(l->fd, SHUT_RD) != 0)
1181 return -1; /* should always be OK */
1182 return 1;
1183}
1184
William Lallemand2e785f22016-05-25 01:48:42 +02001185/*
Willy Tarreau00005ce2016-10-21 15:07:45 +02001186 * Execute the "set-src" action. May be called from {tcp,http}request.
1187 * It only changes the address and tries to preserve the original port. If the
1188 * previous family was neither AF_INET nor AF_INET6, the port is set to zero.
William Lallemand2e785f22016-05-25 01:48:42 +02001189 */
1190enum act_return tcp_action_req_set_src(struct act_rule *rule, struct proxy *px,
1191 struct session *sess, struct stream *s, int flags)
1192{
1193 struct connection *cli_conn;
1194
Willy Tarreaudddd2b42019-07-17 11:37:29 +02001195 if ((cli_conn = objt_conn(sess->origin)) && conn_get_src(cli_conn)) {
William Lallemand2e785f22016-05-25 01:48:42 +02001196 struct sample *smp;
1197
1198 smp = sample_fetch_as_type(px, sess, s, SMP_OPT_DIR_REQ|SMP_OPT_FINAL, rule->arg.expr, SMP_T_ADDR);
1199 if (smp) {
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001200 int port = get_net_port(cli_conn->src);
Willy Tarreau00005ce2016-10-21 15:07:45 +02001201
William Lallemand2e785f22016-05-25 01:48:42 +02001202 if (smp->data.type == SMP_T_IPV4) {
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001203 ((struct sockaddr_in *)cli_conn->src)->sin_family = AF_INET;
1204 ((struct sockaddr_in *)cli_conn->src)->sin_addr.s_addr = smp->data.u.ipv4.s_addr;
1205 ((struct sockaddr_in *)cli_conn->src)->sin_port = port;
William Lallemand2e785f22016-05-25 01:48:42 +02001206 } else if (smp->data.type == SMP_T_IPV6) {
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001207 ((struct sockaddr_in6 *)cli_conn->src)->sin6_family = AF_INET6;
1208 memcpy(&((struct sockaddr_in6 *)cli_conn->src)->sin6_addr, &smp->data.u.ipv6, sizeof(struct in6_addr));
1209 ((struct sockaddr_in6 *)cli_conn->src)->sin6_port = port;
William Lallemand2e785f22016-05-25 01:48:42 +02001210 }
1211 }
William Lallemand01252ed2016-05-25 02:33:16 +02001212 cli_conn->flags |= CO_FL_ADDR_FROM_SET;
William Lallemand2e785f22016-05-25 01:48:42 +02001213 }
1214 return ACT_RET_CONT;
1215}
1216
William Lallemand44be6402016-05-25 01:51:35 +02001217/*
Willy Tarreau00005ce2016-10-21 15:07:45 +02001218 * Execute the "set-dst" action. May be called from {tcp,http}request.
1219 * It only changes the address and tries to preserve the original port. If the
1220 * previous family was neither AF_INET nor AF_INET6, the port is set to zero.
William Lallemand13e9b0c2016-05-25 02:34:07 +02001221 */
1222enum act_return tcp_action_req_set_dst(struct act_rule *rule, struct proxy *px,
1223 struct session *sess, struct stream *s, int flags)
1224{
1225 struct connection *cli_conn;
1226
Willy Tarreaudddd2b42019-07-17 11:37:29 +02001227 if ((cli_conn = objt_conn(sess->origin)) && conn_get_dst(cli_conn)) {
William Lallemand13e9b0c2016-05-25 02:34:07 +02001228 struct sample *smp;
1229
1230 smp = sample_fetch_as_type(px, sess, s, SMP_OPT_DIR_REQ|SMP_OPT_FINAL, rule->arg.expr, SMP_T_ADDR);
1231 if (smp) {
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001232 int port = get_net_port(cli_conn->dst);
Willy Tarreau00005ce2016-10-21 15:07:45 +02001233
William Lallemand13e9b0c2016-05-25 02:34:07 +02001234 if (smp->data.type == SMP_T_IPV4) {
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001235 ((struct sockaddr_in *)cli_conn->dst)->sin_family = AF_INET;
1236 ((struct sockaddr_in *)cli_conn->dst)->sin_addr.s_addr = smp->data.u.ipv4.s_addr;
William Lallemand13e9b0c2016-05-25 02:34:07 +02001237 } else if (smp->data.type == SMP_T_IPV6) {
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001238 ((struct sockaddr_in6 *)cli_conn->dst)->sin6_family = AF_INET6;
1239 memcpy(&((struct sockaddr_in6 *)cli_conn->dst)->sin6_addr, &smp->data.u.ipv6, sizeof(struct in6_addr));
1240 ((struct sockaddr_in6 *)cli_conn->dst)->sin6_port = port;
William Lallemand13e9b0c2016-05-25 02:34:07 +02001241 }
1242 cli_conn->flags |= CO_FL_ADDR_TO_SET;
1243 }
1244 }
1245 return ACT_RET_CONT;
1246}
1247
1248/*
Willy Tarreau00005ce2016-10-21 15:07:45 +02001249 * Execute the "set-src-port" action. May be called from {tcp,http}request.
1250 * We must test the sin_family before setting the port. If the address family
1251 * is neither AF_INET nor AF_INET6, the address is forced to AF_INET "0.0.0.0"
1252 * and the port is assigned.
William Lallemand44be6402016-05-25 01:51:35 +02001253 */
1254enum act_return tcp_action_req_set_src_port(struct act_rule *rule, struct proxy *px,
1255 struct session *sess, struct stream *s, int flags)
1256{
1257 struct connection *cli_conn;
1258
Willy Tarreaudddd2b42019-07-17 11:37:29 +02001259 if ((cli_conn = objt_conn(sess->origin)) && conn_get_src(cli_conn)) {
William Lallemand44be6402016-05-25 01:51:35 +02001260 struct sample *smp;
1261
William Lallemand44be6402016-05-25 01:51:35 +02001262 smp = sample_fetch_as_type(px, sess, s, SMP_OPT_DIR_REQ|SMP_OPT_FINAL, rule->arg.expr, SMP_T_SINT);
1263 if (smp) {
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001264 if (cli_conn->src->ss_family == AF_INET6) {
1265 ((struct sockaddr_in6 *)cli_conn->src)->sin6_port = htons(smp->data.u.sint);
Willy Tarreau00005ce2016-10-21 15:07:45 +02001266 } else {
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001267 if (cli_conn->src->ss_family != AF_INET) {
1268 cli_conn->src->ss_family = AF_INET;
1269 ((struct sockaddr_in *)cli_conn->src)->sin_addr.s_addr = 0;
Willy Tarreau00005ce2016-10-21 15:07:45 +02001270 }
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001271 ((struct sockaddr_in *)cli_conn->src)->sin_port = htons(smp->data.u.sint);
William Lallemand44be6402016-05-25 01:51:35 +02001272 }
1273 }
1274 }
1275 return ACT_RET_CONT;
1276}
1277
William Lallemand13e9b0c2016-05-25 02:34:07 +02001278/*
Willy Tarreau00005ce2016-10-21 15:07:45 +02001279 * Execute the "set-dst-port" action. May be called from {tcp,http}request.
1280 * We must test the sin_family before setting the port. If the address family
1281 * is neither AF_INET nor AF_INET6, the address is forced to AF_INET "0.0.0.0"
1282 * and the port is assigned.
William Lallemand13e9b0c2016-05-25 02:34:07 +02001283 */
1284enum act_return tcp_action_req_set_dst_port(struct act_rule *rule, struct proxy *px,
1285 struct session *sess, struct stream *s, int flags)
1286{
1287 struct connection *cli_conn;
1288
Willy Tarreaudddd2b42019-07-17 11:37:29 +02001289 if ((cli_conn = objt_conn(sess->origin)) && conn_get_dst(cli_conn)) {
William Lallemand13e9b0c2016-05-25 02:34:07 +02001290 struct sample *smp;
1291
William Lallemand13e9b0c2016-05-25 02:34:07 +02001292 smp = sample_fetch_as_type(px, sess, s, SMP_OPT_DIR_REQ|SMP_OPT_FINAL, rule->arg.expr, SMP_T_SINT);
1293 if (smp) {
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001294 if (cli_conn->dst->ss_family == AF_INET6) {
1295 ((struct sockaddr_in6 *)cli_conn->dst)->sin6_port = htons(smp->data.u.sint);
Willy Tarreau00005ce2016-10-21 15:07:45 +02001296 } else {
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001297 if (cli_conn->dst->ss_family != AF_INET) {
1298 cli_conn->dst->ss_family = AF_INET;
1299 ((struct sockaddr_in *)cli_conn->dst)->sin_addr.s_addr = 0;
Willy Tarreau00005ce2016-10-21 15:07:45 +02001300 }
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001301 ((struct sockaddr_in *)cli_conn->dst)->sin_port = htons(smp->data.u.sint);
William Lallemand13e9b0c2016-05-25 02:34:07 +02001302 }
1303 }
1304 }
1305 return ACT_RET_CONT;
1306}
1307
Willy Tarreau2d392c22015-08-24 01:43:45 +02001308/* Executes the "silent-drop" action. May be called from {tcp,http}{request,response} */
1309static enum act_return tcp_exec_action_silent_drop(struct act_rule *rule, struct proxy *px, struct session *sess, struct stream *strm, int flags)
1310{
1311 struct connection *conn = objt_conn(sess->origin);
1312
1313 if (!conn)
1314 goto out;
1315
1316 if (!conn_ctrl_ready(conn))
1317 goto out;
1318
Willy Tarreau2d392c22015-08-24 01:43:45 +02001319#ifdef TCP_QUICKACK
Willy Tarreaufc2a2d92015-09-29 18:15:01 +02001320 /* drain is needed only to send the quick ACK */
1321 conn_sock_drain(conn);
1322
Willy Tarreau2d392c22015-08-24 01:43:45 +02001323 /* re-enable quickack if it was disabled to ack all data and avoid
1324 * retransmits from the client that might trigger a real reset.
1325 */
Willy Tarreau585744b2017-08-24 14:31:19 +02001326 setsockopt(conn->handle.fd, SOL_TCP, TCP_QUICKACK, &one, sizeof(one));
Willy Tarreau2d392c22015-08-24 01:43:45 +02001327#endif
1328 /* lingering must absolutely be disabled so that we don't send a
1329 * shutdown(), this is critical to the TCP_REPAIR trick. When no stream
1330 * is present, returning with ERR will cause lingering to be disabled.
1331 */
1332 if (strm)
1333 strm->si[0].flags |= SI_FL_NOLINGER;
1334
Willy Tarreauf50ec0f2015-09-29 18:11:32 +02001335 /* We're on the client-facing side, we must force to disable lingering to
1336 * ensure we will use an RST exclusively and kill any pending data.
1337 */
Willy Tarreau585744b2017-08-24 14:31:19 +02001338 fdtab[conn->handle.fd].linger_risk = 1;
Willy Tarreauf50ec0f2015-09-29 18:11:32 +02001339
Willy Tarreau2d392c22015-08-24 01:43:45 +02001340#ifdef TCP_REPAIR
Willy Tarreau585744b2017-08-24 14:31:19 +02001341 if (setsockopt(conn->handle.fd, SOL_TCP, TCP_REPAIR, &one, sizeof(one)) == 0) {
Willy Tarreau2d392c22015-08-24 01:43:45 +02001342 /* socket will be quiet now */
1343 goto out;
1344 }
1345#endif
1346 /* either TCP_REPAIR is not defined or it failed (eg: permissions).
1347 * Let's fall back on the TTL trick, though it only works for routed
1348 * network and has no effect on local net.
1349 */
1350#ifdef IP_TTL
Willy Tarreau585744b2017-08-24 14:31:19 +02001351 setsockopt(conn->handle.fd, SOL_IP, IP_TTL, &one, sizeof(one));
Willy Tarreau2d392c22015-08-24 01:43:45 +02001352#endif
1353 out:
1354 /* kill the stream if any */
1355 if (strm) {
1356 channel_abort(&strm->req);
1357 channel_abort(&strm->res);
1358 strm->req.analysers = 0;
1359 strm->res.analysers = 0;
Olivier Houchard40514102019-03-08 18:54:04 +01001360 _HA_ATOMIC_ADD(&strm->be->be_counters.denied_req, 1);
Willy Tarreau2d392c22015-08-24 01:43:45 +02001361 if (!(strm->flags & SF_ERR_MASK))
1362 strm->flags |= SF_ERR_PRXCOND;
1363 if (!(strm->flags & SF_FINST_MASK))
1364 strm->flags |= SF_FINST_R;
1365 }
1366
Olivier Houchard40514102019-03-08 18:54:04 +01001367 _HA_ATOMIC_ADD(&sess->fe->fe_counters.denied_req, 1);
Willy Tarreau2d392c22015-08-24 01:43:45 +02001368 if (sess->listener->counters)
Olivier Houchard40514102019-03-08 18:54:04 +01001369 _HA_ATOMIC_ADD(&sess->listener->counters->denied_req, 1);
Willy Tarreau2d392c22015-08-24 01:43:45 +02001370
1371 return ACT_RET_STOP;
1372}
1373
William Lallemand13e9b0c2016-05-25 02:34:07 +02001374/* parse "set-{src,dst}[-port]" action */
1375enum act_parse_ret tcp_parse_set_src_dst(const char **args, int *orig_arg, struct proxy *px, struct act_rule *rule, char **err)
William Lallemand2e785f22016-05-25 01:48:42 +02001376{
1377 int cur_arg;
1378 struct sample_expr *expr;
1379 unsigned int where;
1380
1381 cur_arg = *orig_arg;
1382 expr = sample_parse_expr((char **)args, &cur_arg, px->conf.args.file, px->conf.args.line, err, &px->conf.args);
1383 if (!expr)
1384 return ACT_RET_PRS_ERR;
1385
1386 where = 0;
Olivier Houchardfbc74e82017-11-24 16:54:05 +01001387 if (px->cap & PR_CAP_FE)
William Lallemand2e785f22016-05-25 01:48:42 +02001388 where |= SMP_VAL_FE_HRQ_HDR;
Olivier Houchardfbc74e82017-11-24 16:54:05 +01001389 if (px->cap & PR_CAP_BE)
William Lallemand2e785f22016-05-25 01:48:42 +02001390 where |= SMP_VAL_BE_HRQ_HDR;
1391
1392 if (!(expr->fetch->val & where)) {
1393 memprintf(err,
1394 "fetch method '%s' extracts information from '%s', none of which is available here",
1395 args[cur_arg-1], sample_src_names(expr->fetch->use));
1396 free(expr);
1397 return ACT_RET_PRS_ERR;
1398 }
1399 rule->arg.expr = expr;
1400 rule->action = ACT_CUSTOM;
1401
1402 if (!strcmp(args[*orig_arg-1], "set-src")) {
1403 rule->action_ptr = tcp_action_req_set_src;
William Lallemand44be6402016-05-25 01:51:35 +02001404 } else if (!strcmp(args[*orig_arg-1], "set-src-port")) {
1405 rule->action_ptr = tcp_action_req_set_src_port;
William Lallemand13e9b0c2016-05-25 02:34:07 +02001406 } else if (!strcmp(args[*orig_arg-1], "set-dst")) {
1407 rule->action_ptr = tcp_action_req_set_dst;
1408 } else if (!strcmp(args[*orig_arg-1], "set-dst-port")) {
1409 rule->action_ptr = tcp_action_req_set_dst_port;
William Lallemand2e785f22016-05-25 01:48:42 +02001410 } else {
1411 return ACT_RET_PRS_ERR;
1412 }
1413
1414 (*orig_arg)++;
1415
1416 return ACT_RET_PRS_OK;
1417}
1418
1419
Willy Tarreau2d392c22015-08-24 01:43:45 +02001420/* Parse a "silent-drop" action. It takes no argument. It returns ACT_RET_PRS_OK on
1421 * success, ACT_RET_PRS_ERR on error.
1422 */
1423static enum act_parse_ret tcp_parse_silent_drop(const char **args, int *orig_arg, struct proxy *px,
1424 struct act_rule *rule, char **err)
1425{
1426 rule->action = ACT_CUSTOM;
1427 rule->action_ptr = tcp_exec_action_silent_drop;
1428 return ACT_RET_PRS_OK;
1429}
1430
Willy Tarreau645513a2010-05-24 20:55:15 +02001431
1432/************************************************************************/
Willy Tarreaud4c33c82013-01-07 21:59:07 +01001433/* All supported sample fetch functions must be declared here */
Willy Tarreau32389b72012-04-23 23:13:20 +02001434/************************************************************************/
1435
Willy Tarreau4a129812012-04-25 17:31:42 +02001436/* fetch the connection's source IPv4/IPv6 address */
Thierry FOURNIERa123ad82015-07-24 09:12:15 +02001437int smp_fetch_src(const struct arg *args, struct sample *smp, const char *kw, void *private)
Willy Tarreau645513a2010-05-24 20:55:15 +02001438{
Thierry FOURNIER0a9a2b82015-05-11 15:20:49 +02001439 struct connection *cli_conn = objt_conn(smp->sess->origin);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001440
1441 if (!cli_conn)
1442 return 0;
1443
Willy Tarreaudddd2b42019-07-17 11:37:29 +02001444 if (!conn_get_src(cli_conn))
1445 return 0;
1446
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001447 switch (cli_conn->src->ss_family) {
Willy Tarreauf4362b32011-12-16 17:49:52 +01001448 case AF_INET:
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001449 smp->data.u.ipv4 = ((struct sockaddr_in *)cli_conn->src)->sin_addr;
Thierry FOURNIER8c542ca2015-08-19 09:00:18 +02001450 smp->data.type = SMP_T_IPV4;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001451 break;
1452 case AF_INET6:
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001453 smp->data.u.ipv6 = ((struct sockaddr_in6 *)cli_conn->src)->sin6_addr;
Thierry FOURNIER8c542ca2015-08-19 09:00:18 +02001454 smp->data.type = SMP_T_IPV6;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001455 break;
1456 default:
Emeric Brunf769f512010-10-22 17:14:01 +02001457 return 0;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001458 }
Emeric Brunf769f512010-10-22 17:14:01 +02001459
Willy Tarreau37406352012-04-23 16:16:37 +02001460 smp->flags = 0;
Willy Tarreau645513a2010-05-24 20:55:15 +02001461 return 1;
1462}
1463
Willy Tarreaua5e37562011-12-16 17:06:15 +01001464/* set temp integer to the connection's source port */
Willy Tarreau645513a2010-05-24 20:55:15 +02001465static int
Thierry FOURNIER0786d052015-05-11 15:42:45 +02001466smp_fetch_sport(const struct arg *args, struct sample *smp, const char *k, void *private)
Willy Tarreau645513a2010-05-24 20:55:15 +02001467{
Thierry FOURNIER0a9a2b82015-05-11 15:20:49 +02001468 struct connection *cli_conn = objt_conn(smp->sess->origin);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001469
1470 if (!cli_conn)
1471 return 0;
1472
Willy Tarreaudddd2b42019-07-17 11:37:29 +02001473 if (!conn_get_src(cli_conn))
1474 return 0;
1475
Thierry FOURNIER8c542ca2015-08-19 09:00:18 +02001476 smp->data.type = SMP_T_SINT;
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001477 if (!(smp->data.u.sint = get_host_port(cli_conn->src)))
Emeric Brunf769f512010-10-22 17:14:01 +02001478 return 0;
1479
Willy Tarreau37406352012-04-23 16:16:37 +02001480 smp->flags = 0;
Willy Tarreau645513a2010-05-24 20:55:15 +02001481 return 1;
1482}
1483
Willy Tarreau4a129812012-04-25 17:31:42 +02001484/* fetch the connection's destination IPv4/IPv6 address */
Willy Tarreau645513a2010-05-24 20:55:15 +02001485static int
Thierry FOURNIER0786d052015-05-11 15:42:45 +02001486smp_fetch_dst(const struct arg *args, struct sample *smp, const char *kw, void *private)
Willy Tarreau645513a2010-05-24 20:55:15 +02001487{
Thierry FOURNIER0a9a2b82015-05-11 15:20:49 +02001488 struct connection *cli_conn = objt_conn(smp->sess->origin);
Willy Tarreau645513a2010-05-24 20:55:15 +02001489
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001490 if (!cli_conn)
1491 return 0;
1492
Willy Tarreaudddd2b42019-07-17 11:37:29 +02001493 if (!conn_get_dst(cli_conn))
1494 return 0;
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001495
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001496 switch (cli_conn->dst->ss_family) {
Willy Tarreauf4362b32011-12-16 17:49:52 +01001497 case AF_INET:
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001498 smp->data.u.ipv4 = ((struct sockaddr_in *)cli_conn->dst)->sin_addr;
Thierry FOURNIER8c542ca2015-08-19 09:00:18 +02001499 smp->data.type = SMP_T_IPV4;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001500 break;
1501 case AF_INET6:
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001502 smp->data.u.ipv6 = ((struct sockaddr_in6 *)cli_conn->dst)->sin6_addr;
Thierry FOURNIER8c542ca2015-08-19 09:00:18 +02001503 smp->data.type = SMP_T_IPV6;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001504 break;
1505 default:
Emeric Brunf769f512010-10-22 17:14:01 +02001506 return 0;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001507 }
Emeric Brunf769f512010-10-22 17:14:01 +02001508
Willy Tarreau37406352012-04-23 16:16:37 +02001509 smp->flags = 0;
Willy Tarreau645513a2010-05-24 20:55:15 +02001510 return 1;
1511}
1512
Willy Tarreau16e01562016-08-09 16:46:18 +02001513/* check if the destination address of the front connection is local to the
1514 * system or if it was intercepted.
1515 */
1516int smp_fetch_dst_is_local(const struct arg *args, struct sample *smp, const char *kw, void *private)
1517{
1518 struct connection *conn = objt_conn(smp->sess->origin);
1519 struct listener *li = smp->sess->listener;
1520
1521 if (!conn)
1522 return 0;
1523
Willy Tarreaudddd2b42019-07-17 11:37:29 +02001524 if (!conn_get_dst(conn))
Willy Tarreau16e01562016-08-09 16:46:18 +02001525 return 0;
1526
1527 smp->data.type = SMP_T_BOOL;
1528 smp->flags = 0;
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001529 smp->data.u.sint = addr_is_local(li->netns, conn->dst);
Willy Tarreau16e01562016-08-09 16:46:18 +02001530 return smp->data.u.sint >= 0;
1531}
1532
1533/* check if the source address of the front connection is local to the system
1534 * or not.
1535 */
1536int smp_fetch_src_is_local(const struct arg *args, struct sample *smp, const char *kw, void *private)
1537{
1538 struct connection *conn = objt_conn(smp->sess->origin);
1539 struct listener *li = smp->sess->listener;
1540
1541 if (!conn)
1542 return 0;
1543
Willy Tarreaudddd2b42019-07-17 11:37:29 +02001544 if (!conn_get_src(conn))
Willy Tarreau16e01562016-08-09 16:46:18 +02001545 return 0;
1546
1547 smp->data.type = SMP_T_BOOL;
1548 smp->flags = 0;
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001549 smp->data.u.sint = addr_is_local(li->netns, conn->src);
Willy Tarreau16e01562016-08-09 16:46:18 +02001550 return smp->data.u.sint >= 0;
1551}
1552
Willy Tarreaua5e37562011-12-16 17:06:15 +01001553/* set temp integer to the frontend connexion's destination port */
Willy Tarreau645513a2010-05-24 20:55:15 +02001554static int
Thierry FOURNIER0786d052015-05-11 15:42:45 +02001555smp_fetch_dport(const struct arg *args, struct sample *smp, const char *kw, void *private)
Willy Tarreau645513a2010-05-24 20:55:15 +02001556{
Thierry FOURNIER0a9a2b82015-05-11 15:20:49 +02001557 struct connection *cli_conn = objt_conn(smp->sess->origin);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001558
1559 if (!cli_conn)
1560 return 0;
1561
Willy Tarreaudddd2b42019-07-17 11:37:29 +02001562 if (!conn_get_dst(cli_conn))
1563 return 0;
Willy Tarreau645513a2010-05-24 20:55:15 +02001564
Thierry FOURNIER8c542ca2015-08-19 09:00:18 +02001565 smp->data.type = SMP_T_SINT;
Willy Tarreau7bbc4a52019-07-17 15:41:35 +02001566 if (!(smp->data.u.sint = get_host_port(cli_conn->dst)))
Emeric Brunf769f512010-10-22 17:14:01 +02001567 return 0;
1568
Willy Tarreau37406352012-04-23 16:16:37 +02001569 smp->flags = 0;
Willy Tarreau645513a2010-05-24 20:55:15 +02001570 return 1;
Emericf2d7cae2010-11-05 18:13:50 +01001571}
1572
Thierry Fournier / OZON.IO6310bef2016-07-24 20:16:50 +02001573#ifdef TCP_INFO
1574
Christopher Fauletba0c53e2019-10-17 14:40:48 +02001575
1576/* Validates the arguments passed to "fc_*" fetch keywords returning a time
1577 * value. These keywords support an optional string representing the unit of the
1578 * result: "us" for microseconds and "ms" for milliseconds". Returns 0 on error
1579 * and non-zero if OK.
1580 */
1581static int val_fc_time_value(struct arg *args, char **err)
1582{
1583 if (args[0].type == ARGT_STOP) {
1584 args[0].type = ARGT_SINT;
1585 args[0].data.sint = TIME_UNIT_MS;
1586 }
1587 if (args[0].type == ARGT_STR) {
1588 if (strcmp(args[0].data.str.area, "us") == 0) {
1589 free(args[0].data.str.area);
1590 args[0].type = ARGT_SINT;
1591 args[0].data.sint = TIME_UNIT_US;
1592 }
1593 else if (strcmp(args[0].data.str.area, "ms") == 0) {
1594 free(args[0].data.str.area);
1595 args[0].type = ARGT_SINT;
1596 args[0].data.sint = TIME_UNIT_MS;
1597 }
1598 else {
1599 memprintf(err, "expects 'us' or 'ms', got '%s'",
1600 args[0].data.str.area);
1601 return 0;
1602 }
1603 }
1604 else {
1605 memprintf(err, "Unexpected arg type");
1606 return 0;
1607 }
1608
1609 return 1;
1610}
1611
1612/* Validates the arguments passed to "fc_*" fetch keywords returning a
1613 * counter. These keywords should be used without any keyword, but because of a
1614 * bug in previous versions, an optional string argument may be passed. In such
1615 * case, the argument is ignored and a warning is emitted. Returns 0 on error
1616 * and non-zero if OK.
1617 */
1618static int var_fc_counter(struct arg *args, char **err)
1619{
1620 if (args[0].type != ARGT_STOP) {
1621 ha_warning("no argument supported for 'fc_*' sample expressions returning counters.\n");
1622 if (args[0].type == ARGT_STR)
1623 free(args[0].data.str.area);
1624 args[0].type = ARGT_STOP;
1625 }
1626
1627 return 1;
1628}
1629
Joseph Herlanta6331472018-11-25 12:59:12 -08001630/* Returns some tcp_info data if it's available. "dir" must be set to 0 if
1631 * the client connection is required, otherwise it is set to 1. "val" represents
Christopher Fauletba0c53e2019-10-17 14:40:48 +02001632 * the required value.
Thierry Fournier / OZON.IO6310bef2016-07-24 20:16:50 +02001633 * If the function fails it returns 0, otherwise it returns 1 and "result" is filled.
1634 */
1635static inline int get_tcp_info(const struct arg *args, struct sample *smp,
1636 int dir, int val)
1637{
1638 struct connection *conn;
1639 struct tcp_info info;
1640 socklen_t optlen;
1641
1642 /* strm can be null. */
1643 if (!smp->strm)
1644 return 0;
1645
1646 /* get the object associated with the stream interface.The
1647 * object can be other thing than a connection. For example,
1648 * it be a appctx. */
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001649 conn = cs_conn(objt_cs(smp->strm->si[dir].end));
Thierry Fournier / OZON.IO6310bef2016-07-24 20:16:50 +02001650 if (!conn)
1651 return 0;
1652
Olivier Houchard9aaf7782017-09-13 18:30:23 +02001653 /* The fd may not be available for the tcp_info struct, and the
Thierry Fournier / OZON.IO6310bef2016-07-24 20:16:50 +02001654 syscal can fail. */
1655 optlen = sizeof(info);
Willy Tarreau585744b2017-08-24 14:31:19 +02001656 if (getsockopt(conn->handle.fd, SOL_TCP, TCP_INFO, &info, &optlen) == -1)
Thierry Fournier / OZON.IO6310bef2016-07-24 20:16:50 +02001657 return 0;
1658
1659 /* extract the value. */
1660 smp->data.type = SMP_T_SINT;
1661 switch (val) {
Joe Williams30fcd392016-08-10 07:06:44 -07001662 case 0: smp->data.u.sint = info.tcpi_rtt; break;
1663 case 1: smp->data.u.sint = info.tcpi_rttvar; break;
1664#if defined(__linux__)
1665 /* these ones are common to all Linux versions */
1666 case 2: smp->data.u.sint = info.tcpi_unacked; break;
1667 case 3: smp->data.u.sint = info.tcpi_sacked; break;
1668 case 4: smp->data.u.sint = info.tcpi_lost; break;
1669 case 5: smp->data.u.sint = info.tcpi_retrans; break;
1670 case 6: smp->data.u.sint = info.tcpi_fackets; break;
1671 case 7: smp->data.u.sint = info.tcpi_reordering; break;
1672#elif defined(__FreeBSD__) || defined(__NetBSD__)
1673 /* the ones are found on FreeBSD and NetBSD featuring TCP_INFO */
1674 case 2: smp->data.u.sint = info.__tcpi_unacked; break;
1675 case 3: smp->data.u.sint = info.__tcpi_sacked; break;
1676 case 4: smp->data.u.sint = info.__tcpi_lost; break;
1677 case 5: smp->data.u.sint = info.__tcpi_retrans; break;
1678 case 6: smp->data.u.sint = info.__tcpi_fackets; break;
1679 case 7: smp->data.u.sint = info.__tcpi_reordering; break;
1680#endif
Thierry Fournier / OZON.IO6310bef2016-07-24 20:16:50 +02001681 default: return 0;
1682 }
1683
Thierry Fournier / OZON.IO6310bef2016-07-24 20:16:50 +02001684 return 1;
1685}
1686
1687/* get the mean rtt of a client connexion */
1688static int
1689smp_fetch_fc_rtt(const struct arg *args, struct sample *smp, const char *kw, void *private)
1690{
1691 if (!get_tcp_info(args, smp, 0, 0))
1692 return 0;
Christopher Fauletba0c53e2019-10-17 14:40:48 +02001693
1694 /* By default or if explicitly specified, convert rtt to ms */
1695 if (!args || args[0].type == ARGT_STOP || args[0].data.sint == TIME_UNIT_MS)
1696 smp->data.u.sint = (smp->data.u.sint + 500) / 1000;
1697
Thierry Fournier / OZON.IO6310bef2016-07-24 20:16:50 +02001698 return 1;
1699}
1700
1701/* get the variance of the mean rtt of a client connexion */
1702static int
1703smp_fetch_fc_rttvar(const struct arg *args, struct sample *smp, const char *kw, void *private)
1704{
1705 if (!get_tcp_info(args, smp, 0, 1))
1706 return 0;
Christopher Fauletba0c53e2019-10-17 14:40:48 +02001707
1708 /* By default or if explicitly specified, convert rttvar to ms */
1709 if (!args || args[0].type == ARGT_STOP || args[0].data.sint == TIME_UNIT_MS)
1710 smp->data.u.sint = (smp->data.u.sint + 500) / 1000;
1711
Thierry Fournier / OZON.IO6310bef2016-07-24 20:16:50 +02001712 return 1;
1713}
Joe Williams30fcd392016-08-10 07:06:44 -07001714
1715#if defined(__linux__) || defined(__FreeBSD__) || defined(__NetBSD__)
1716
1717/* get the unacked counter on a client connexion */
1718static int
1719smp_fetch_fc_unacked(const struct arg *args, struct sample *smp, const char *kw, void *private)
1720{
1721 if (!get_tcp_info(args, smp, 0, 2))
1722 return 0;
1723 return 1;
1724}
1725
1726/* get the sacked counter on a client connexion */
1727static int
1728smp_fetch_fc_sacked(const struct arg *args, struct sample *smp, const char *kw, void *private)
1729{
1730 if (!get_tcp_info(args, smp, 0, 3))
1731 return 0;
1732 return 1;
1733}
1734
1735/* get the lost counter on a client connexion */
1736static int
1737smp_fetch_fc_lost(const struct arg *args, struct sample *smp, const char *kw, void *private)
1738{
1739 if (!get_tcp_info(args, smp, 0, 4))
1740 return 0;
1741 return 1;
1742}
1743
1744/* get the retrans counter on a client connexion */
1745static int
1746smp_fetch_fc_retrans(const struct arg *args, struct sample *smp, const char *kw, void *private)
1747{
1748 if (!get_tcp_info(args, smp, 0, 5))
1749 return 0;
1750 return 1;
1751}
1752
1753/* get the fackets counter on a client connexion */
1754static int
1755smp_fetch_fc_fackets(const struct arg *args, struct sample *smp, const char *kw, void *private)
1756{
1757 if (!get_tcp_info(args, smp, 0, 6))
1758 return 0;
1759 return 1;
1760}
1761
1762/* get the reordering counter on a client connexion */
1763static int
1764smp_fetch_fc_reordering(const struct arg *args, struct sample *smp, const char *kw, void *private)
1765{
1766 if (!get_tcp_info(args, smp, 0, 7))
1767 return 0;
1768 return 1;
1769}
1770#endif // linux || freebsd || netbsd
1771#endif // TCP_INFO
Thierry Fournier / OZON.IO6310bef2016-07-24 20:16:50 +02001772
Willy Tarreau9b6700f2012-11-24 11:55:28 +01001773#ifdef IPV6_V6ONLY
Willy Tarreau77e3af92012-11-24 15:07:23 +01001774/* parse the "v4v6" bind keyword */
1775static int bind_parse_v4v6(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
1776{
1777 struct listener *l;
1778
1779 list_for_each_entry(l, &conf->listeners, by_bind) {
1780 if (l->addr.ss_family == AF_INET6)
1781 l->options |= LI_O_V4V6;
1782 }
1783
1784 return 0;
1785}
1786
Willy Tarreau9b6700f2012-11-24 11:55:28 +01001787/* parse the "v6only" bind keyword */
1788static int bind_parse_v6only(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
1789{
1790 struct listener *l;
1791
1792 list_for_each_entry(l, &conf->listeners, by_bind) {
1793 if (l->addr.ss_family == AF_INET6)
1794 l->options |= LI_O_V6ONLY;
1795 }
1796
1797 return 0;
1798}
1799#endif
1800
Pieter Baauwd551fb52013-05-08 22:49:23 +02001801#ifdef CONFIG_HAP_TRANSPARENT
Willy Tarreau44791242012-09-12 23:27:21 +02001802/* parse the "transparent" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +02001803static 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 +02001804{
1805 struct listener *l;
1806
Willy Tarreau4348fad2012-09-20 16:48:07 +02001807 list_for_each_entry(l, &conf->listeners, by_bind) {
1808 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1809 l->options |= LI_O_FOREIGN;
Willy Tarreau44791242012-09-12 23:27:21 +02001810 }
1811
Willy Tarreau44791242012-09-12 23:27:21 +02001812 return 0;
1813}
1814#endif
1815
1816#ifdef TCP_DEFER_ACCEPT
1817/* parse the "defer-accept" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +02001818static 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 +02001819{
1820 struct listener *l;
1821
Willy Tarreau4348fad2012-09-20 16:48:07 +02001822 list_for_each_entry(l, &conf->listeners, by_bind) {
1823 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1824 l->options |= LI_O_DEF_ACCEPT;
Willy Tarreau44791242012-09-12 23:27:21 +02001825 }
1826
Willy Tarreau44791242012-09-12 23:27:21 +02001827 return 0;
1828}
1829#endif
1830
Willy Tarreau1c862c52012-10-05 16:21:00 +02001831#ifdef TCP_FASTOPEN
Lukas Tribus0defb902013-02-13 23:35:39 +01001832/* parse the "tfo" bind keyword */
Willy Tarreau1c862c52012-10-05 16:21:00 +02001833static int bind_parse_tfo(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
1834{
1835 struct listener *l;
1836
1837 list_for_each_entry(l, &conf->listeners, by_bind) {
1838 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1839 l->options |= LI_O_TCP_FO;
1840 }
1841
1842 return 0;
1843}
1844#endif
1845
Willy Tarreau44791242012-09-12 23:27:21 +02001846#ifdef TCP_MAXSEG
1847/* parse the "mss" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +02001848static 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 +02001849{
1850 struct listener *l;
1851 int mss;
1852
Willy Tarreau44791242012-09-12 23:27:21 +02001853 if (!*args[cur_arg + 1]) {
Willy Tarreaueb6cead2012-09-20 19:43:14 +02001854 memprintf(err, "'%s' : missing MSS value", args[cur_arg]);
Willy Tarreau44791242012-09-12 23:27:21 +02001855 return ERR_ALERT | ERR_FATAL;
1856 }
1857
1858 mss = atoi(args[cur_arg + 1]);
1859 if (!mss || abs(mss) > 65535) {
Willy Tarreaueb6cead2012-09-20 19:43:14 +02001860 memprintf(err, "'%s' : expects an MSS with and absolute value between 1 and 65535", args[cur_arg]);
Willy Tarreau44791242012-09-12 23:27:21 +02001861 return ERR_ALERT | ERR_FATAL;
1862 }
1863
Willy Tarreau4348fad2012-09-20 16:48:07 +02001864 list_for_each_entry(l, &conf->listeners, by_bind) {
1865 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1866 l->maxseg = mss;
1867 }
Willy Tarreau44791242012-09-12 23:27:21 +02001868
1869 return 0;
1870}
1871#endif
1872
Willy Tarreau2af207a2015-02-04 00:45:58 +01001873#ifdef TCP_USER_TIMEOUT
1874/* parse the "tcp-ut" bind keyword */
1875static int bind_parse_tcp_ut(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
1876{
1877 const char *ptr = NULL;
1878 struct listener *l;
1879 unsigned int timeout;
1880
1881 if (!*args[cur_arg + 1]) {
1882 memprintf(err, "'%s' : missing TCP User Timeout value", args[cur_arg]);
1883 return ERR_ALERT | ERR_FATAL;
1884 }
1885
1886 ptr = parse_time_err(args[cur_arg + 1], &timeout, TIME_UNIT_MS);
Willy Tarreau9faebe32019-06-07 19:00:37 +02001887 if (ptr == PARSE_TIME_OVER) {
1888 memprintf(err, "timer overflow in argument '%s' to '%s' (maximum value is 2147483647 ms or ~24.8 days)",
1889 args[cur_arg+1], args[cur_arg]);
1890 return ERR_ALERT | ERR_FATAL;
1891 }
1892 else if (ptr == PARSE_TIME_UNDER) {
1893 memprintf(err, "timer underflow in argument '%s' to '%s' (minimum non-null value is 1 ms)",
1894 args[cur_arg+1], args[cur_arg]);
1895 return ERR_ALERT | ERR_FATAL;
1896 }
1897 else if (ptr) {
Willy Tarreau2af207a2015-02-04 00:45:58 +01001898 memprintf(err, "'%s' : expects a positive delay in milliseconds", args[cur_arg]);
1899 return ERR_ALERT | ERR_FATAL;
1900 }
1901
1902 list_for_each_entry(l, &conf->listeners, by_bind) {
1903 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1904 l->tcp_ut = timeout;
1905 }
1906
1907 return 0;
1908}
1909#endif
1910
Willy Tarreau44791242012-09-12 23:27:21 +02001911#ifdef SO_BINDTODEVICE
Willy Tarreau2af207a2015-02-04 00:45:58 +01001912/* parse the "interface" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +02001913static 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 +02001914{
1915 struct listener *l;
1916
Willy Tarreau44791242012-09-12 23:27:21 +02001917 if (!*args[cur_arg + 1]) {
Willy Tarreaueb6cead2012-09-20 19:43:14 +02001918 memprintf(err, "'%s' : missing interface name", args[cur_arg]);
Willy Tarreau44791242012-09-12 23:27:21 +02001919 return ERR_ALERT | ERR_FATAL;
1920 }
1921
Willy Tarreau4348fad2012-09-20 16:48:07 +02001922 list_for_each_entry(l, &conf->listeners, by_bind) {
1923 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1924 l->interface = strdup(args[cur_arg + 1]);
1925 }
Willy Tarreau44791242012-09-12 23:27:21 +02001926
Willy Tarreau44791242012-09-12 23:27:21 +02001927 return 0;
1928}
1929#endif
1930
Willy Tarreaue5733232019-05-22 19:24:06 +02001931#ifdef USE_NS
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001932/* parse the "namespace" bind keyword */
1933static int bind_parse_namespace(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
1934{
1935 struct listener *l;
1936 char *namespace = NULL;
1937
1938 if (!*args[cur_arg + 1]) {
1939 memprintf(err, "'%s' : missing namespace id", args[cur_arg]);
1940 return ERR_ALERT | ERR_FATAL;
1941 }
1942 namespace = args[cur_arg + 1];
1943
1944 list_for_each_entry(l, &conf->listeners, by_bind) {
1945 l->netns = netns_store_lookup(namespace, strlen(namespace));
1946
1947 if (l->netns == NULL)
1948 l->netns = netns_store_insert(namespace);
1949
1950 if (l->netns == NULL) {
Christopher Faulet767a84b2017-11-24 16:50:31 +01001951 ha_alert("Cannot open namespace '%s'.\n", args[cur_arg + 1]);
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001952 return ERR_ALERT | ERR_FATAL;
1953 }
1954 }
1955 return 0;
1956}
1957#endif
1958
Willy Tarreau163d4622015-10-13 16:16:41 +02001959#ifdef TCP_USER_TIMEOUT
1960/* parse the "tcp-ut" server keyword */
1961static int srv_parse_tcp_ut(char **args, int *cur_arg, struct proxy *px, struct server *newsrv, char **err)
1962{
1963 const char *ptr = NULL;
1964 unsigned int timeout;
1965
1966 if (!*args[*cur_arg + 1]) {
1967 memprintf(err, "'%s' : missing TCP User Timeout value", args[*cur_arg]);
1968 return ERR_ALERT | ERR_FATAL;
1969 }
1970
1971 ptr = parse_time_err(args[*cur_arg + 1], &timeout, TIME_UNIT_MS);
Willy Tarreau9faebe32019-06-07 19:00:37 +02001972 if (ptr == PARSE_TIME_OVER) {
1973 memprintf(err, "timer overflow in argument '%s' to '%s' (maximum value is 2147483647 ms or ~24.8 days)",
1974 args[*cur_arg+1], args[*cur_arg]);
1975 return ERR_ALERT | ERR_FATAL;
1976 }
1977 else if (ptr == PARSE_TIME_UNDER) {
1978 memprintf(err, "timer underflow in argument '%s' to '%s' (minimum non-null value is 1 ms)",
1979 args[*cur_arg+1], args[*cur_arg]);
1980 return ERR_ALERT | ERR_FATAL;
1981 }
1982 else if (ptr) {
Willy Tarreau163d4622015-10-13 16:16:41 +02001983 memprintf(err, "'%s' : expects a positive delay in milliseconds", args[*cur_arg]);
1984 return ERR_ALERT | ERR_FATAL;
1985 }
1986
1987 if (newsrv->addr.ss_family == AF_INET || newsrv->addr.ss_family == AF_INET6)
1988 newsrv->tcp_ut = timeout;
1989
1990 return 0;
1991}
1992#endif
1993
Willy Tarreaud4c33c82013-01-07 21:59:07 +01001994
Willy Tarreau4a129812012-04-25 17:31:42 +02001995/* Note: must not be declared <const> as its list will be overwritten.
1996 * Note: fetches that may return multiple types must be declared as the lowest
1997 * common denominator, the type that can be casted into all other ones. For
1998 * instance v4/v6 must be declared v4.
1999 */
Willy Tarreaudc13c112013-06-21 23:16:39 +02002000static struct sample_fetch_kw_list sample_fetch_keywords = {ILH, {
Willy Tarreaud4c33c82013-01-07 21:59:07 +01002001 { "dst", smp_fetch_dst, 0, NULL, SMP_T_IPV4, SMP_USE_L4CLI },
Willy Tarreau16e01562016-08-09 16:46:18 +02002002 { "dst_is_local", smp_fetch_dst_is_local, 0, NULL, SMP_T_BOOL, SMP_USE_L4CLI },
Thierry FOURNIER07ee64e2015-07-06 23:43:03 +02002003 { "dst_port", smp_fetch_dport, 0, NULL, SMP_T_SINT, SMP_USE_L4CLI },
Willy Tarreaud4c33c82013-01-07 21:59:07 +01002004 { "src", smp_fetch_src, 0, NULL, SMP_T_IPV4, SMP_USE_L4CLI },
Willy Tarreau16e01562016-08-09 16:46:18 +02002005 { "src_is_local", smp_fetch_src_is_local, 0, NULL, SMP_T_BOOL, SMP_USE_L4CLI },
Thierry FOURNIER07ee64e2015-07-06 23:43:03 +02002006 { "src_port", smp_fetch_sport, 0, NULL, SMP_T_SINT, SMP_USE_L4CLI },
Thierry Fournier / OZON.IO6310bef2016-07-24 20:16:50 +02002007#ifdef TCP_INFO
Christopher Fauletba0c53e2019-10-17 14:40:48 +02002008 { "fc_rtt", smp_fetch_fc_rtt, ARG1(0,STR), val_fc_time_value, SMP_T_SINT, SMP_USE_L4CLI },
2009 { "fc_rttvar", smp_fetch_fc_rttvar, ARG1(0,STR), val_fc_time_value, SMP_T_SINT, SMP_USE_L4CLI },
Joe Williams30fcd392016-08-10 07:06:44 -07002010#if defined(__linux__) || defined(__FreeBSD__) || defined(__NetBSD__)
Christopher Fauletba0c53e2019-10-17 14:40:48 +02002011 { "fc_unacked", smp_fetch_fc_unacked, ARG1(0,STR), var_fc_counter, SMP_T_SINT, SMP_USE_L4CLI },
2012 { "fc_sacked", smp_fetch_fc_sacked, ARG1(0,STR), var_fc_counter, SMP_T_SINT, SMP_USE_L4CLI },
2013 { "fc_retrans", smp_fetch_fc_retrans, ARG1(0,STR), var_fc_counter, SMP_T_SINT, SMP_USE_L4CLI },
2014 { "fc_fackets", smp_fetch_fc_fackets, ARG1(0,STR), var_fc_counter, SMP_T_SINT, SMP_USE_L4CLI },
2015 { "fc_lost", smp_fetch_fc_lost, ARG1(0,STR), var_fc_counter, SMP_T_SINT, SMP_USE_L4CLI },
2016 { "fc_reordering", smp_fetch_fc_reordering, ARG1(0,STR), var_fc_counter, SMP_T_SINT, SMP_USE_L4CLI },
Joe Williams30fcd392016-08-10 07:06:44 -07002017#endif // linux || freebsd || netbsd
2018#endif // TCP_INFO
Willy Tarreaud4c33c82013-01-07 21:59:07 +01002019 { /* END */ },
Willy Tarreau645513a2010-05-24 20:55:15 +02002020}};
2021
Willy Tarreau0108d902018-11-25 19:14:37 +01002022INITCALL1(STG_REGISTER, sample_register_fetches, &sample_fetch_keywords);
2023
Willy Tarreau44791242012-09-12 23:27:21 +02002024/************************************************************************/
2025/* All supported bind keywords must be declared here. */
2026/************************************************************************/
2027
2028/* Note: must not be declared <const> as its list will be overwritten.
2029 * Please take care of keeping this list alphabetically sorted, doing so helps
2030 * all code contributors.
2031 * Optional keywords are also declared with a NULL ->parse() function so that
2032 * the config parser can report an appropriate error when a known keyword was
2033 * not enabled.
2034 */
Willy Tarreau51fb7652012-09-18 18:24:39 +02002035static struct bind_kw_list bind_kws = { "TCP", { }, {
Willy Tarreau44791242012-09-12 23:27:21 +02002036#ifdef TCP_DEFER_ACCEPT
2037 { "defer-accept", bind_parse_defer_accept, 0 }, /* wait for some data for 1 second max before doing accept */
2038#endif
2039#ifdef SO_BINDTODEVICE
2040 { "interface", bind_parse_interface, 1 }, /* specifically bind to this interface */
2041#endif
2042#ifdef TCP_MAXSEG
2043 { "mss", bind_parse_mss, 1 }, /* set MSS of listening socket */
2044#endif
Willy Tarreau2af207a2015-02-04 00:45:58 +01002045#ifdef TCP_USER_TIMEOUT
2046 { "tcp-ut", bind_parse_tcp_ut, 1 }, /* set User Timeout on listening socket */
2047#endif
Willy Tarreau1c862c52012-10-05 16:21:00 +02002048#ifdef TCP_FASTOPEN
2049 { "tfo", bind_parse_tfo, 0 }, /* enable TCP_FASTOPEN of listening socket */
2050#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +02002051#ifdef CONFIG_HAP_TRANSPARENT
Willy Tarreau44791242012-09-12 23:27:21 +02002052 { "transparent", bind_parse_transparent, 0 }, /* transparently bind to the specified addresses */
2053#endif
Willy Tarreau9b6700f2012-11-24 11:55:28 +01002054#ifdef IPV6_V6ONLY
Willy Tarreau77e3af92012-11-24 15:07:23 +01002055 { "v4v6", bind_parse_v4v6, 0 }, /* force socket to bind to IPv4+IPv6 */
Willy Tarreau9b6700f2012-11-24 11:55:28 +01002056 { "v6only", bind_parse_v6only, 0 }, /* force socket to bind to IPv6 only */
2057#endif
Willy Tarreaue5733232019-05-22 19:24:06 +02002058#ifdef USE_NS
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01002059 { "namespace", bind_parse_namespace, 1 },
2060#endif
Willy Tarreau44791242012-09-12 23:27:21 +02002061 /* the versions with the NULL parse function*/
2062 { "defer-accept", NULL, 0 },
2063 { "interface", NULL, 1 },
2064 { "mss", NULL, 1 },
2065 { "transparent", NULL, 0 },
Willy Tarreau77e3af92012-11-24 15:07:23 +01002066 { "v4v6", NULL, 0 },
Willy Tarreau9b6700f2012-11-24 11:55:28 +01002067 { "v6only", NULL, 0 },
Willy Tarreau44791242012-09-12 23:27:21 +02002068 { NULL, NULL, 0 },
2069}};
2070
Willy Tarreau0108d902018-11-25 19:14:37 +01002071INITCALL1(STG_REGISTER, bind_register_keywords, &bind_kws);
2072
Willy Tarreau163d4622015-10-13 16:16:41 +02002073static struct srv_kw_list srv_kws = { "TCP", { }, {
2074#ifdef TCP_USER_TIMEOUT
Frédéric Lécaille5c3cd972017-03-15 16:36:09 +01002075 { "tcp-ut", srv_parse_tcp_ut, 1, 1 }, /* set TCP user timeout on server */
Willy Tarreau163d4622015-10-13 16:16:41 +02002076#endif
2077 { NULL, NULL, 0 },
2078}};
Willy Tarreau2d392c22015-08-24 01:43:45 +02002079
Willy Tarreau0108d902018-11-25 19:14:37 +01002080INITCALL1(STG_REGISTER, srv_register_keywords, &srv_kws);
2081
Willy Tarreau2d392c22015-08-24 01:43:45 +02002082static struct action_kw_list tcp_req_conn_actions = {ILH, {
William Lallemand13e9b0c2016-05-25 02:34:07 +02002083 { "set-src", tcp_parse_set_src_dst },
2084 { "set-src-port", tcp_parse_set_src_dst },
2085 { "set-dst" , tcp_parse_set_src_dst },
2086 { "set-dst-port", tcp_parse_set_src_dst },
Baptiste Assmanne1afd4f2019-04-18 16:21:13 +02002087 { "silent-drop", tcp_parse_silent_drop },
Willy Tarreau2d392c22015-08-24 01:43:45 +02002088 { /* END */ }
2089}};
2090
Willy Tarreau0108d902018-11-25 19:14:37 +01002091INITCALL1(STG_REGISTER, tcp_req_conn_keywords_register, &tcp_req_conn_actions);
2092
Willy Tarreau620408f2016-10-21 16:37:51 +02002093static struct action_kw_list tcp_req_sess_actions = {ILH, {
Willy Tarreau620408f2016-10-21 16:37:51 +02002094 { "set-src", tcp_parse_set_src_dst },
2095 { "set-src-port", tcp_parse_set_src_dst },
2096 { "set-dst" , tcp_parse_set_src_dst },
2097 { "set-dst-port", tcp_parse_set_src_dst },
Baptiste Assmanne1afd4f2019-04-18 16:21:13 +02002098 { "silent-drop", tcp_parse_silent_drop },
Willy Tarreau620408f2016-10-21 16:37:51 +02002099 { /* END */ }
2100}};
2101
Willy Tarreau0108d902018-11-25 19:14:37 +01002102INITCALL1(STG_REGISTER, tcp_req_sess_keywords_register, &tcp_req_sess_actions);
2103
Willy Tarreau2d392c22015-08-24 01:43:45 +02002104static struct action_kw_list tcp_req_cont_actions = {ILH, {
Baptiste Assmanne1afd4f2019-04-18 16:21:13 +02002105 { "set-dst" , tcp_parse_set_src_dst },
2106 { "set-dst-port", tcp_parse_set_src_dst },
2107 { "silent-drop", tcp_parse_silent_drop },
Willy Tarreau2d392c22015-08-24 01:43:45 +02002108 { /* END */ }
2109}};
2110
Willy Tarreau0108d902018-11-25 19:14:37 +01002111INITCALL1(STG_REGISTER, tcp_req_cont_keywords_register, &tcp_req_cont_actions);
2112
Willy Tarreau2d392c22015-08-24 01:43:45 +02002113static struct action_kw_list tcp_res_cont_actions = {ILH, {
2114 { "silent-drop", tcp_parse_silent_drop },
2115 { /* END */ }
2116}};
2117
Willy Tarreau0108d902018-11-25 19:14:37 +01002118INITCALL1(STG_REGISTER, tcp_res_cont_keywords_register, &tcp_res_cont_actions);
2119
Willy Tarreau2d392c22015-08-24 01:43:45 +02002120static struct action_kw_list http_req_actions = {ILH, {
William Lallemand2e785f22016-05-25 01:48:42 +02002121 { "silent-drop", tcp_parse_silent_drop },
William Lallemand13e9b0c2016-05-25 02:34:07 +02002122 { "set-src", tcp_parse_set_src_dst },
2123 { "set-src-port", tcp_parse_set_src_dst },
2124 { "set-dst", tcp_parse_set_src_dst },
2125 { "set-dst-port", tcp_parse_set_src_dst },
Willy Tarreau2d392c22015-08-24 01:43:45 +02002126 { /* END */ }
2127}};
2128
Willy Tarreau0108d902018-11-25 19:14:37 +01002129INITCALL1(STG_REGISTER, http_req_keywords_register, &http_req_actions);
2130
Willy Tarreau2d392c22015-08-24 01:43:45 +02002131static struct action_kw_list http_res_actions = {ILH, {
2132 { "silent-drop", tcp_parse_silent_drop },
2133 { /* END */ }
2134}};
2135
Willy Tarreau0108d902018-11-25 19:14:37 +01002136INITCALL1(STG_REGISTER, http_res_keywords_register, &http_res_actions);
Willy Tarreau2d392c22015-08-24 01:43:45 +02002137
Willy Tarreau80713382018-11-26 10:19:54 +01002138REGISTER_BUILD_OPTS("Built with transparent proxy support using:"
Willy Tarreauba962912016-12-21 18:55:02 +01002139#if defined(IP_TRANSPARENT)
Willy Tarreau80713382018-11-26 10:19:54 +01002140 " IP_TRANSPARENT"
Willy Tarreauba962912016-12-21 18:55:02 +01002141#endif
2142#if defined(IPV6_TRANSPARENT)
Willy Tarreau80713382018-11-26 10:19:54 +01002143 " IPV6_TRANSPARENT"
Willy Tarreauba962912016-12-21 18:55:02 +01002144#endif
2145#if defined(IP_FREEBIND)
Willy Tarreau80713382018-11-26 10:19:54 +01002146 " IP_FREEBIND"
Willy Tarreauba962912016-12-21 18:55:02 +01002147#endif
2148#if defined(IP_BINDANY)
Willy Tarreau80713382018-11-26 10:19:54 +01002149 " IP_BINDANY"
Willy Tarreauba962912016-12-21 18:55:02 +01002150#endif
2151#if defined(IPV6_BINDANY)
Willy Tarreau80713382018-11-26 10:19:54 +01002152 " IPV6_BINDANY"
Willy Tarreauba962912016-12-21 18:55:02 +01002153#endif
2154#if defined(SO_BINDANY)
Willy Tarreau80713382018-11-26 10:19:54 +01002155 " SO_BINDANY"
Willy Tarreauba962912016-12-21 18:55:02 +01002156#endif
Willy Tarreau80713382018-11-26 10:19:54 +01002157 "");
Willy Tarreaue6b98942007-10-29 01:09:36 +01002158
2159
2160/*
2161 * Local variables:
2162 * c-indent-level: 8
2163 * c-basic-offset: 8
2164 * End:
2165 */