blob: 53569a99897cad78aa6d4a4712ae094709ad4658 [file] [log] [blame]
Willy Tarreaue6b98942007-10-29 01:09:36 +01001/*
2 * AF_INET/AF_INET6 SOCK_STREAM protocol layer (tcp)
3 *
Willy Tarreaud4c33c82013-01-07 21:59:07 +01004 * Copyright 2000-2013 Willy Tarreau <w@1wt.eu>
Willy Tarreaue6b98942007-10-29 01:09:36 +01005 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 */
12
13#include <ctype.h>
14#include <errno.h>
15#include <fcntl.h>
16#include <stdio.h>
17#include <stdlib.h>
18#include <string.h>
19#include <time.h>
20
21#include <sys/param.h>
22#include <sys/socket.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010023#include <sys/types.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010024
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040025#include <netinet/tcp.h>
Willy Tarreau2d392c22015-08-24 01:43:45 +020026#include <netinet/in.h>
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040027
Willy Tarreau4c7e4b72020-05-27 12:58:42 +020028#include <haproxy/api.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020029#include <haproxy/arg.h>
Willy Tarreau7ea393d2020-06-04 18:02:10 +020030#include <haproxy/connection.h>
Willy Tarreau8d366972020-05-27 16:10:29 +020031#include <haproxy/errors.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020032#include <haproxy/fd.h>
Willy Tarreauf268ee82020-06-04 17:05:57 +020033#include <haproxy/global.h>
Willy Tarreau853b2972020-05-27 18:01:47 +020034#include <haproxy/list.h>
Willy Tarreau213e9902020-06-04 14:58:24 +020035#include <haproxy/listener.h>
Willy Tarreauaeed4a82020-06-04 22:01:04 +020036#include <haproxy/log.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020037#include <haproxy/namespace.h>
38#include <haproxy/port_range.h>
39#include <haproxy/proto_tcp.h>
40#include <haproxy/protocol.h>
Willy Tarreaua264d962020-06-04 22:29:18 +020041#include <haproxy/proxy-t.h>
Willy Tarreau18b7df72020-08-28 12:07:22 +020042#include <haproxy/sock.h>
Willy Tarreauf1725582020-08-28 15:30:11 +020043#include <haproxy/sock_inet.h>
Willy Tarreau48fbcae2020-06-03 18:09:46 +020044#include <haproxy/tools.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010045
Willy Tarreaue6b98942007-10-29 01:09:36 +010046
Emeric Bruncf20bf12010-10-22 16:06:11 +020047static int tcp_bind_listener(struct listener *listener, char *errmsg, int errlen);
Willy Tarreaucb66ea62020-09-25 17:12:32 +020048static int tcp_suspend_receiver(struct receiver *rx);
Willy Tarreau58e6b712020-09-25 19:40:31 +020049static int tcp_resume_receiver(struct receiver *rx);
Willy Tarreau5ddf1ce2020-09-25 19:27:39 +020050static void tcp_enable_listener(struct listener *listener);
51static void tcp_disable_listener(struct listener *listener);
Willy Tarreaue6b98942007-10-29 01:09:36 +010052
53/* Note: must not be declared <const> as its list will be overwritten */
Willy Tarreaub9b2fd72020-12-08 14:13:11 +010054struct protocol proto_tcpv4 = {
Willy Tarreaub366c9a2020-12-08 14:54:20 +010055 .name = "tcpv4",
56
57 /* connection layer */
58 .ctrl_type = SOCK_STREAM,
59 .listen = tcp_bind_listener,
60 .enable = tcp_enable_listener,
61 .disable = tcp_disable_listener,
62 .add = default_add_listener,
63 .unbind = default_unbind_listener,
64 .suspend = default_suspend_listener,
65 .resume = default_resume_listener,
66 .accept_conn = sock_accept_conn,
Willy Tarreaude471c42020-12-08 15:50:56 +010067 .ctrl_init = sock_conn_ctrl_init,
68 .ctrl_close = sock_conn_ctrl_close,
Willy Tarreaub366c9a2020-12-08 14:54:20 +010069 .connect = tcp_connect_server,
Willy Tarreau427c8462020-12-11 16:19:12 +010070 .drain = sock_drain,
Willy Tarreau472125b2020-12-11 17:02:50 +010071 .check_events = sock_check_events,
72 .ignore_events = sock_ignore_events,
Willy Tarreaub366c9a2020-12-08 14:54:20 +010073
74 /* binding layer */
75 .rx_suspend = tcp_suspend_receiver,
76 .rx_resume = tcp_resume_receiver,
77
78 /* address family */
79 .fam = &proto_fam_inet4,
80
81 /* socket layer */
82 .sock_type = SOCK_STREAM,
83 .sock_prot = IPPROTO_TCP,
84 .rx_enable = sock_enable,
85 .rx_disable = sock_disable,
86 .rx_unbind = sock_unbind,
87 .rx_listening = sock_accepting_conn,
88 .default_iocb = sock_accept_iocb,
89 .receivers = LIST_HEAD_INIT(proto_tcpv4.receivers),
90 .nb_receivers = 0,
Willy Tarreaue6b98942007-10-29 01:09:36 +010091};
92
Willy Tarreau0108d902018-11-25 19:14:37 +010093INITCALL1(STG_REGISTER, protocol_register, &proto_tcpv4);
94
Willy Tarreaue6b98942007-10-29 01:09:36 +010095/* Note: must not be declared <const> as its list will be overwritten */
Willy Tarreaub9b2fd72020-12-08 14:13:11 +010096struct protocol proto_tcpv6 = {
Willy Tarreaub366c9a2020-12-08 14:54:20 +010097 .name = "tcpv6",
98
99 /* connection layer */
100 .ctrl_type = SOCK_STREAM,
101 .listen = tcp_bind_listener,
102 .enable = tcp_enable_listener,
103 .disable = tcp_disable_listener,
104 .add = default_add_listener,
105 .unbind = default_unbind_listener,
106 .suspend = default_suspend_listener,
107 .resume = default_resume_listener,
108 .accept_conn = sock_accept_conn,
Willy Tarreaude471c42020-12-08 15:50:56 +0100109 .ctrl_init = sock_conn_ctrl_init,
110 .ctrl_close = sock_conn_ctrl_close,
Willy Tarreaub366c9a2020-12-08 14:54:20 +0100111 .connect = tcp_connect_server,
Willy Tarreau427c8462020-12-11 16:19:12 +0100112 .drain = sock_drain,
Willy Tarreau472125b2020-12-11 17:02:50 +0100113 .check_events = sock_check_events,
114 .ignore_events = sock_ignore_events,
Willy Tarreaub366c9a2020-12-08 14:54:20 +0100115
116 /* binding layer */
117 .rx_suspend = tcp_suspend_receiver,
118 .rx_resume = tcp_resume_receiver,
119
120 /* address family */
121 .fam = &proto_fam_inet6,
122
123 /* socket layer */
124 .sock_type = SOCK_STREAM,
125 .sock_prot = IPPROTO_TCP,
126 .rx_enable = sock_enable,
127 .rx_disable = sock_disable,
128 .rx_unbind = sock_unbind,
129 .rx_listening = sock_accepting_conn,
130 .default_iocb = sock_accept_iocb,
131 .receivers = LIST_HEAD_INIT(proto_tcpv6.receivers),
132 .nb_receivers = 0,
Willy Tarreaue6b98942007-10-29 01:09:36 +0100133};
134
Willy Tarreau0108d902018-11-25 19:14:37 +0100135INITCALL1(STG_REGISTER, protocol_register, &proto_tcpv6);
136
David du Colombier6f5ccb12011-03-10 22:26:24 +0100137/* Binds ipv4/ipv6 address <local> to socket <fd>, unless <flags> is set, in which
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100138 * case we try to bind <remote>. <flags> is a 2-bit field consisting of :
139 * - 0 : ignore remote address (may even be a NULL pointer)
140 * - 1 : use provided address
141 * - 2 : use provided port
142 * - 3 : use both
143 *
144 * The function supports multiple foreign binding methods :
145 * - linux_tproxy: we directly bind to the foreign address
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100146 * The second one can be used as a fallback for the first one.
147 * This function returns 0 when everything's OK, 1 if it could not bind, to the
148 * local address, 2 if it could not bind to the foreign address.
149 */
David du Colombier6f5ccb12011-03-10 22:26:24 +0100150int tcp_bind_socket(int fd, int flags, struct sockaddr_storage *local, struct sockaddr_storage *remote)
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100151{
David du Colombier6f5ccb12011-03-10 22:26:24 +0100152 struct sockaddr_storage bind_addr;
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100153 int foreign_ok = 0;
154 int ret;
Christopher Faulet1bc04c72017-10-29 20:14:08 +0100155 static THREAD_LOCAL int ip_transp_working = 1;
156 static THREAD_LOCAL int ip6_transp_working = 1;
Pieter Baauwd551fb52013-05-08 22:49:23 +0200157
David du Colombier65c17962012-07-13 14:34:59 +0200158 switch (local->ss_family) {
159 case AF_INET:
160 if (flags && ip_transp_working) {
Pieter Baauwd551fb52013-05-08 22:49:23 +0200161 /* This deserves some explanation. Some platforms will support
162 * multiple combinations of certain methods, so we try the
163 * supported ones until one succeeds.
164 */
Willy Tarreau37bafdc2020-08-28 17:23:40 +0200165 if (sock_inet4_make_foreign(fd))
David du Colombier65c17962012-07-13 14:34:59 +0200166 foreign_ok = 1;
167 else
168 ip_transp_working = 0;
169 }
170 break;
171 case AF_INET6:
172 if (flags && ip6_transp_working) {
Willy Tarreau37bafdc2020-08-28 17:23:40 +0200173 if (sock_inet6_make_foreign(fd))
David du Colombier65c17962012-07-13 14:34:59 +0200174 foreign_ok = 1;
175 else
176 ip6_transp_working = 0;
177 }
178 break;
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100179 }
Pieter Baauwd551fb52013-05-08 22:49:23 +0200180
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100181 if (flags) {
182 memset(&bind_addr, 0, sizeof(bind_addr));
Willy Tarreau96dd0792011-04-19 07:20:57 +0200183 bind_addr.ss_family = remote->ss_family;
David du Colombier6f5ccb12011-03-10 22:26:24 +0100184 switch (remote->ss_family) {
185 case AF_INET:
186 if (flags & 1)
187 ((struct sockaddr_in *)&bind_addr)->sin_addr = ((struct sockaddr_in *)remote)->sin_addr;
188 if (flags & 2)
189 ((struct sockaddr_in *)&bind_addr)->sin_port = ((struct sockaddr_in *)remote)->sin_port;
190 break;
191 case AF_INET6:
192 if (flags & 1)
193 ((struct sockaddr_in6 *)&bind_addr)->sin6_addr = ((struct sockaddr_in6 *)remote)->sin6_addr;
194 if (flags & 2)
195 ((struct sockaddr_in6 *)&bind_addr)->sin6_port = ((struct sockaddr_in6 *)remote)->sin6_port;
196 break;
Willy Tarreau5dc1e982011-12-16 21:25:11 +0100197 default:
198 /* we don't want to try to bind to an unknown address family */
199 foreign_ok = 0;
David du Colombier6f5ccb12011-03-10 22:26:24 +0100200 }
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100201 }
202
Simon Hormande072bd2011-06-24 15:11:37 +0900203 setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &one, sizeof(one));
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100204 if (foreign_ok) {
Willy Tarreau9cf8d3f2014-05-09 22:56:10 +0200205 if (is_inet_addr(&bind_addr)) {
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200206 ret = bind(fd, (struct sockaddr *)&bind_addr, get_addr_len(&bind_addr));
207 if (ret < 0)
208 return 2;
209 }
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100210 }
211 else {
Willy Tarreau9cf8d3f2014-05-09 22:56:10 +0200212 if (is_inet_addr(local)) {
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200213 ret = bind(fd, (struct sockaddr *)local, get_addr_len(local));
214 if (ret < 0)
215 return 1;
216 }
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100217 }
218
219 if (!flags)
220 return 0;
221
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100222 if (!foreign_ok)
223 /* we could not bind to a foreign address */
224 return 2;
225
226 return 0;
227}
Willy Tarreaue6b98942007-10-29 01:09:36 +0100228
Willy Tarreau9650f372009-08-16 14:02:45 +0200229/*
Willy Tarreau14f8e862012-08-30 22:23:13 +0200230 * This function initiates a TCP connection establishment to the target assigned
Willy Tarreau7bbc4a52019-07-17 15:41:35 +0200231 * to connection <conn> using (si->{target,dst}). A source address may be
232 * pointed to by conn->src in case of transparent proxying. Normal source
Willy Tarreau14f8e862012-08-30 22:23:13 +0200233 * bind addresses are still determined locally (due to the possible need of a
234 * source port). conn->target may point either to a valid server or to a backend,
Willy Tarreau3fdb3662012-11-12 00:42:33 +0100235 * depending on conn->target. Only OBJ_TYPE_PROXY and OBJ_TYPE_SERVER are
Willy Tarreauf0837b22012-11-24 10:24:27 +0100236 * supported. The <data> parameter is a boolean indicating whether there are data
237 * waiting for being sent or not, in order to adjust data write polling and on
Olivier Houchardfdcb0072019-05-06 18:32:29 +0200238 * some platforms, the ability to avoid an empty initial ACK. The <flags> argument
239 * allows the caller to force using a delayed ACK when establishing the connection
Willy Tarreauf0837b22012-11-24 10:24:27 +0100240 * - 0 = no delayed ACK unless data are advertised and backend has tcp-smart-connect
Olivier Houchardfdcb0072019-05-06 18:32:29 +0200241 * - CONNECT_DELACK_SMART_CONNECT = delayed ACK if backend has tcp-smart-connect, regardless of data
242 * - CONNECT_DELACK_ALWAYS = delayed ACK regardless of backend options
Willy Tarreaub1d67742010-03-29 19:36:59 +0200243 *
Willy Tarreau1ec74bf2013-10-24 21:45:00 +0200244 * Note that a pending send_proxy message accounts for data.
245 *
Willy Tarreau9650f372009-08-16 14:02:45 +0200246 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +0200247 * - SF_ERR_NONE if everything's OK
248 * - SF_ERR_SRVTO if there are no more servers
249 * - SF_ERR_SRVCL if the connection was refused by the server
250 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
251 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
252 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +0100253 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Willy Tarreau6b0a8502012-11-23 08:51:32 +0100254 *
Willy Tarreaue7dff022015-04-03 01:14:29 +0200255 * The connection's fd is inserted only when SF_ERR_NONE is returned, otherwise
Willy Tarreau6b0a8502012-11-23 08:51:32 +0100256 * it's invalid and the caller has nothing to do.
Willy Tarreau9650f372009-08-16 14:02:45 +0200257 */
Willy Tarreauf1536862011-03-03 18:27:32 +0100258
Olivier Houchardfdcb0072019-05-06 18:32:29 +0200259int tcp_connect_server(struct connection *conn, int flags)
Willy Tarreau9650f372009-08-16 14:02:45 +0200260{
261 int fd;
Willy Tarreauac825402011-03-04 22:04:29 +0100262 struct server *srv;
263 struct proxy *be;
Willy Tarreaua4380b42012-12-08 22:49:11 +0100264 struct conn_src *src;
Willy Tarreau034c88c2017-01-23 23:36:45 +0100265 int use_fastopen = 0;
Alexander Liu2a54bb72019-05-22 19:44:48 +0800266 struct sockaddr_storage *addr;
Willy Tarreauac825402011-03-04 22:04:29 +0100267
Olivier Houchard637b6952018-11-23 14:23:07 +0100268 conn->flags |= CO_FL_WAIT_L4_CONN; /* connection in progress */
Willy Tarreau9ce70132014-01-24 16:08:19 +0100269
Willy Tarreau3fdb3662012-11-12 00:42:33 +0100270 switch (obj_type(conn->target)) {
271 case OBJ_TYPE_PROXY:
272 be = objt_proxy(conn->target);
Willy Tarreauac825402011-03-04 22:04:29 +0100273 srv = NULL;
274 break;
Willy Tarreau3fdb3662012-11-12 00:42:33 +0100275 case OBJ_TYPE_SERVER:
276 srv = objt_server(conn->target);
Willy Tarreauac825402011-03-04 22:04:29 +0100277 be = srv->proxy;
Willy Tarreau034c88c2017-01-23 23:36:45 +0100278 /* Make sure we check that we have data before activating
279 * TFO, or we could trigger a kernel issue whereby after
280 * a successful connect() == 0, any subsequent connect()
281 * will return EINPROGRESS instead of EISCONN.
282 */
283 use_fastopen = (srv->flags & SRV_F_FASTOPEN) &&
284 ((flags & (CONNECT_CAN_USE_TFO | CONNECT_HAS_DATA)) ==
285 (CONNECT_CAN_USE_TFO | CONNECT_HAS_DATA));
Willy Tarreauac825402011-03-04 22:04:29 +0100286 break;
287 default:
Willy Tarreau9ce70132014-01-24 16:08:19 +0100288 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200289 return SF_ERR_INTERNAL;
Willy Tarreauac825402011-03-04 22:04:29 +0100290 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200291
Willy Tarreau7bbc4a52019-07-17 15:41:35 +0200292 if (!conn->dst) {
293 conn->flags |= CO_FL_ERROR;
294 return SF_ERR_INTERNAL;
295 }
296
Willy Tarreau18b7df72020-08-28 12:07:22 +0200297 fd = conn->handle.fd = sock_create_server_socket(conn);
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100298
299 if (fd == -1) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200300 qfprintf(stderr, "Cannot get a server socket.\n");
301
Willy Tarreau9ce70132014-01-24 16:08:19 +0100302 if (errno == ENFILE) {
303 conn->err_code = CO_ER_SYS_FDLIM;
Willy Tarreau9650f372009-08-16 14:02:45 +0200304 send_log(be, LOG_EMERG,
Willy Tarreauc5532ac2018-01-29 15:06:04 +0100305 "Proxy %s reached system FD limit (maxsock=%d). Please check system tunables.\n",
306 be->id, global.maxsock);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100307 }
308 else if (errno == EMFILE) {
309 conn->err_code = CO_ER_PROC_FDLIM;
Willy Tarreau9650f372009-08-16 14:02:45 +0200310 send_log(be, LOG_EMERG,
Willy Tarreauc5532ac2018-01-29 15:06:04 +0100311 "Proxy %s reached process FD limit (maxsock=%d). Please check 'ulimit-n' and restart.\n",
312 be->id, global.maxsock);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100313 }
314 else if (errno == ENOBUFS || errno == ENOMEM) {
315 conn->err_code = CO_ER_SYS_MEMLIM;
Willy Tarreau9650f372009-08-16 14:02:45 +0200316 send_log(be, LOG_EMERG,
Willy Tarreauc5532ac2018-01-29 15:06:04 +0100317 "Proxy %s reached system memory limit (maxsock=%d). Please check system tunables.\n",
318 be->id, global.maxsock);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100319 }
320 else if (errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
321 conn->err_code = CO_ER_NOPROTO;
322 }
323 else
324 conn->err_code = CO_ER_SOCK_ERR;
325
Willy Tarreau9650f372009-08-16 14:02:45 +0200326 /* this is a resource error */
Willy Tarreau9ce70132014-01-24 16:08:19 +0100327 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200328 return SF_ERR_RESOURCE;
Willy Tarreau9650f372009-08-16 14:02:45 +0200329 }
330
331 if (fd >= global.maxsock) {
332 /* do not log anything there, it's a normal condition when this option
333 * is used to serialize connections to a server !
334 */
Christopher Faulet767a84b2017-11-24 16:50:31 +0100335 ha_alert("socket(): not enough free sockets. Raise -n argument. Giving up.\n");
Willy Tarreau9650f372009-08-16 14:02:45 +0200336 close(fd);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100337 conn->err_code = CO_ER_CONF_FDLIM;
338 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200339 return SF_ERR_PRXCOND; /* it is a configuration limit */
Willy Tarreau9650f372009-08-16 14:02:45 +0200340 }
341
342 if ((fcntl(fd, F_SETFL, O_NONBLOCK)==-1) ||
Simon Hormande072bd2011-06-24 15:11:37 +0900343 (setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &one, sizeof(one)) == -1)) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200344 qfprintf(stderr,"Cannot set client socket to non blocking mode.\n");
345 close(fd);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100346 conn->err_code = CO_ER_SOCK_ERR;
347 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200348 return SF_ERR_INTERNAL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200349 }
350
William Lallemandc03eb012018-11-27 12:02:37 +0100351 if (master == 1 && (fcntl(fd, F_SETFD, FD_CLOEXEC) == -1)) {
352 ha_alert("Cannot set CLOEXEC on client socket.\n");
353 close(fd);
354 conn->err_code = CO_ER_SOCK_ERR;
355 conn->flags |= CO_FL_ERROR;
356 return SF_ERR_INTERNAL;
357 }
358
MIZUTA Takeshib24bc0d2020-07-09 11:13:20 +0900359 if (be->options & PR_O_TCP_SRV_KA) {
Simon Hormande072bd2011-06-24 15:11:37 +0900360 setsockopt(fd, SOL_SOCKET, SO_KEEPALIVE, &one, sizeof(one));
Willy Tarreau9650f372009-08-16 14:02:45 +0200361
Willy Tarreau52543212020-07-09 05:58:51 +0200362#ifdef TCP_KEEPCNT
MIZUTA Takeshib24bc0d2020-07-09 11:13:20 +0900363 if (be->srvtcpka_cnt)
364 setsockopt(fd, IPPROTO_TCP, TCP_KEEPCNT, &be->srvtcpka_cnt, sizeof(be->srvtcpka_cnt));
Willy Tarreau52543212020-07-09 05:58:51 +0200365#endif
MIZUTA Takeshib24bc0d2020-07-09 11:13:20 +0900366
Willy Tarreau52543212020-07-09 05:58:51 +0200367#ifdef TCP_KEEPIDLE
MIZUTA Takeshib24bc0d2020-07-09 11:13:20 +0900368 if (be->srvtcpka_idle)
369 setsockopt(fd, IPPROTO_TCP, TCP_KEEPIDLE, &be->srvtcpka_idle, sizeof(be->srvtcpka_idle));
Willy Tarreau52543212020-07-09 05:58:51 +0200370#endif
MIZUTA Takeshib24bc0d2020-07-09 11:13:20 +0900371
Willy Tarreau52543212020-07-09 05:58:51 +0200372#ifdef TCP_KEEPINTVL
MIZUTA Takeshib24bc0d2020-07-09 11:13:20 +0900373 if (be->srvtcpka_intvl)
374 setsockopt(fd, IPPROTO_TCP, TCP_KEEPINTVL, &be->srvtcpka_intvl, sizeof(be->srvtcpka_intvl));
Willy Tarreau52543212020-07-09 05:58:51 +0200375#endif
MIZUTA Takeshib24bc0d2020-07-09 11:13:20 +0900376 }
377
Willy Tarreau9650f372009-08-16 14:02:45 +0200378 /* allow specific binding :
379 * - server-specific at first
380 * - proxy-specific next
381 */
Willy Tarreaua4380b42012-12-08 22:49:11 +0100382 if (srv && srv->conn_src.opts & CO_SRC_BIND)
383 src = &srv->conn_src;
384 else if (be->conn_src.opts & CO_SRC_BIND)
385 src = &be->conn_src;
386 else
387 src = NULL;
388
389 if (src) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200390 int ret, flags = 0;
391
Willy Tarreau7bbc4a52019-07-17 15:41:35 +0200392 if (conn->src && is_inet_addr(conn->src)) {
Willy Tarreaua4380b42012-12-08 22:49:11 +0100393 switch (src->opts & CO_SRC_TPROXY_MASK) {
Willy Tarreauef9a3602012-12-08 22:29:20 +0100394 case CO_SRC_TPROXY_CLI:
Willy Tarreau387ebf82015-08-04 19:24:13 +0200395 case CO_SRC_TPROXY_ADDR:
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200396 flags = 3;
397 break;
Willy Tarreauef9a3602012-12-08 22:29:20 +0100398 case CO_SRC_TPROXY_CIP:
399 case CO_SRC_TPROXY_DYN:
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200400 flags = 1;
401 break;
402 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200403 }
Willy Tarreaub1d67742010-03-29 19:36:59 +0200404
Willy Tarreau9650f372009-08-16 14:02:45 +0200405#ifdef SO_BINDTODEVICE
406 /* Note: this might fail if not CAP_NET_RAW */
Willy Tarreaua4380b42012-12-08 22:49:11 +0100407 if (src->iface_name)
408 setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, src->iface_name, src->iface_len + 1);
Willy Tarreau9650f372009-08-16 14:02:45 +0200409#endif
410
Willy Tarreaua4380b42012-12-08 22:49:11 +0100411 if (src->sport_range) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200412 int attempts = 10; /* should be more than enough to find a spare port */
Willy Tarreaua4380b42012-12-08 22:49:11 +0100413 struct sockaddr_storage sa;
Willy Tarreau9650f372009-08-16 14:02:45 +0200414
415 ret = 1;
Vincent Bernat6e615892016-05-18 16:17:44 +0200416 memcpy(&sa, &src->source_addr, sizeof(sa));
Willy Tarreau9650f372009-08-16 14:02:45 +0200417
418 do {
419 /* note: in case of retry, we may have to release a previously
420 * allocated port, hence this loop's construct.
421 */
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200422 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
423 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200424
425 if (!attempts)
426 break;
427 attempts--;
428
Willy Tarreaua4380b42012-12-08 22:49:11 +0100429 fdinfo[fd].local_port = port_range_alloc_port(src->sport_range);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100430 if (!fdinfo[fd].local_port) {
431 conn->err_code = CO_ER_PORT_RANGE;
Willy Tarreau9650f372009-08-16 14:02:45 +0200432 break;
Willy Tarreau9ce70132014-01-24 16:08:19 +0100433 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200434
Willy Tarreaua4380b42012-12-08 22:49:11 +0100435 fdinfo[fd].port_range = src->sport_range;
436 set_host_port(&sa, fdinfo[fd].local_port);
Willy Tarreau9650f372009-08-16 14:02:45 +0200437
Willy Tarreau7bbc4a52019-07-17 15:41:35 +0200438 ret = tcp_bind_socket(fd, flags, &sa, conn->src);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100439 if (ret != 0)
440 conn->err_code = CO_ER_CANT_BIND;
Willy Tarreau9650f372009-08-16 14:02:45 +0200441 } while (ret != 0); /* binding NOK */
442 }
443 else {
Lukas Tribus7d56c6d2016-09-13 09:51:15 +0000444#ifdef IP_BIND_ADDRESS_NO_PORT
Christopher Faulet1bc04c72017-10-29 20:14:08 +0100445 static THREAD_LOCAL int bind_address_no_port = 1;
Lukas Tribus7d56c6d2016-09-13 09:51:15 +0000446 setsockopt(fd, SOL_IP, IP_BIND_ADDRESS_NO_PORT, (const void *) &bind_address_no_port, sizeof(int));
447#endif
Willy Tarreau7bbc4a52019-07-17 15:41:35 +0200448 ret = tcp_bind_socket(fd, flags, &src->source_addr, conn->src);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100449 if (ret != 0)
450 conn->err_code = CO_ER_CANT_BIND;
Willy Tarreau9650f372009-08-16 14:02:45 +0200451 }
452
Willy Tarreaua4380b42012-12-08 22:49:11 +0100453 if (unlikely(ret != 0)) {
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200454 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
455 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200456 close(fd);
457
Willy Tarreau9650f372009-08-16 14:02:45 +0200458 if (ret == 1) {
Christopher Faulet767a84b2017-11-24 16:50:31 +0100459 ha_alert("Cannot bind to source address before connect() for backend %s. Aborting.\n",
460 be->id);
Willy Tarreau9650f372009-08-16 14:02:45 +0200461 send_log(be, LOG_EMERG,
Willy Tarreaua4380b42012-12-08 22:49:11 +0100462 "Cannot bind to source address before connect() for backend %s.\n",
Willy Tarreau9650f372009-08-16 14:02:45 +0200463 be->id);
464 } else {
Christopher Faulet767a84b2017-11-24 16:50:31 +0100465 ha_alert("Cannot bind to tproxy source address before connect() for backend %s. Aborting.\n",
466 be->id);
Willy Tarreau9650f372009-08-16 14:02:45 +0200467 send_log(be, LOG_EMERG,
Willy Tarreaua4380b42012-12-08 22:49:11 +0100468 "Cannot bind to tproxy source address before connect() for backend %s.\n",
Willy Tarreau9650f372009-08-16 14:02:45 +0200469 be->id);
470 }
Willy Tarreau9ce70132014-01-24 16:08:19 +0100471 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200472 return SF_ERR_RESOURCE;
Willy Tarreau9650f372009-08-16 14:02:45 +0200473 }
474 }
475
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +0400476#if defined(TCP_QUICKACK)
Willy Tarreau9650f372009-08-16 14:02:45 +0200477 /* disabling tcp quick ack now allows the first request to leave the
478 * machine with the first ACK. We only do this if there are pending
Willy Tarreauf0837b22012-11-24 10:24:27 +0100479 * data in the buffer.
Willy Tarreau9650f372009-08-16 14:02:45 +0200480 */
Olivier Houchardfdcb0072019-05-06 18:32:29 +0200481 if (flags & (CONNECT_DELACK_ALWAYS) ||
482 ((flags & CONNECT_DELACK_SMART_CONNECT ||
483 (flags & CONNECT_HAS_DATA) || conn->send_proxy_ofs) &&
484 (be->options2 & PR_O2_SMARTCON)))
Simon Hormande072bd2011-06-24 15:11:37 +0900485 setsockopt(fd, IPPROTO_TCP, TCP_QUICKACK, &zero, sizeof(zero));
Willy Tarreau9650f372009-08-16 14:02:45 +0200486#endif
487
Willy Tarreau163d4622015-10-13 16:16:41 +0200488#ifdef TCP_USER_TIMEOUT
489 /* there is not much more we can do here when it fails, it's still minor */
490 if (srv && srv->tcp_ut)
491 setsockopt(fd, IPPROTO_TCP, TCP_USER_TIMEOUT, &srv->tcp_ut, sizeof(srv->tcp_ut));
492#endif
Willy Tarreau034c88c2017-01-23 23:36:45 +0100493
494 if (use_fastopen) {
495#if defined(TCP_FASTOPEN_CONNECT)
496 setsockopt(fd, IPPROTO_TCP, TCP_FASTOPEN_CONNECT, &one, sizeof(one));
497#endif
498 }
Willy Tarreaue803de22010-01-21 17:43:04 +0100499 if (global.tune.server_sndbuf)
500 setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &global.tune.server_sndbuf, sizeof(global.tune.server_sndbuf));
501
502 if (global.tune.server_rcvbuf)
503 setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &global.tune.server_rcvbuf, sizeof(global.tune.server_rcvbuf));
504
Willy Tarreau7bbc4a52019-07-17 15:41:35 +0200505 addr = (conn->flags & CO_FL_SOCKS4) ? &srv->socks4_addr : conn->dst;
Alexander Liu2a54bb72019-05-22 19:44:48 +0800506 if (connect(fd, (const struct sockaddr *)addr, get_addr_len(addr)) == -1) {
Willy Tarreau819efbf2017-01-25 14:12:22 +0100507 if (errno == EINPROGRESS || errno == EALREADY) {
508 /* common case, let's wait for connect status */
509 conn->flags |= CO_FL_WAIT_L4_CONN;
510 }
511 else if (errno == EISCONN) {
512 /* should normally not happen but if so, indicates that it's OK */
513 conn->flags &= ~CO_FL_WAIT_L4_CONN;
514 }
515 else if (errno == EAGAIN || errno == EADDRINUSE || errno == EADDRNOTAVAIL) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200516 char *msg;
Willy Tarreau9ce70132014-01-24 16:08:19 +0100517 if (errno == EAGAIN || errno == EADDRNOTAVAIL) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200518 msg = "no free ports";
Willy Tarreau9ce70132014-01-24 16:08:19 +0100519 conn->err_code = CO_ER_FREE_PORTS;
520 }
521 else {
Willy Tarreau9650f372009-08-16 14:02:45 +0200522 msg = "local address already in use";
Willy Tarreau9ce70132014-01-24 16:08:19 +0100523 conn->err_code = CO_ER_ADDR_INUSE;
524 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200525
Willy Tarreaub1719512012-12-08 23:03:28 +0100526 qfprintf(stderr,"Connect() failed for backend %s: %s.\n", be->id, msg);
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200527 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
528 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200529 close(fd);
Willy Tarreaub1719512012-12-08 23:03:28 +0100530 send_log(be, LOG_ERR, "Connect() failed for backend %s: %s.\n", be->id, msg);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100531 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200532 return SF_ERR_RESOURCE;
Willy Tarreau9650f372009-08-16 14:02:45 +0200533 } else if (errno == ETIMEDOUT) {
534 //qfprintf(stderr,"Connect(): ETIMEDOUT");
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200535 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
536 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200537 close(fd);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100538 conn->err_code = CO_ER_SOCK_ERR;
539 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200540 return SF_ERR_SRVTO;
Willy Tarreau9650f372009-08-16 14:02:45 +0200541 } else {
542 // (errno == ECONNREFUSED || errno == ENETUNREACH || errno == EACCES || errno == EPERM)
543 //qfprintf(stderr,"Connect(): %d", errno);
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 Tarreau9ce70132014-01-24 16:08:19 +0100547 conn->err_code = CO_ER_SOCK_ERR;
548 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200549 return SF_ERR_SRVCL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200550 }
551 }
Willy Tarreau819efbf2017-01-25 14:12:22 +0100552 else {
553 /* connect() == 0, this is great! */
554 conn->flags &= ~CO_FL_WAIT_L4_CONN;
555 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200556
Willy Tarreaufc8f1f02012-12-08 18:53:44 +0100557 conn->flags |= CO_FL_ADDR_TO_SET;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200558
Willy Tarreauf79c8172013-10-21 16:30:56 +0200559 conn_ctrl_init(conn); /* registers the FD */
Willy Tarreauad38ace2013-12-15 14:19:38 +0100560 fdtab[fd].linger_risk = 1; /* close hard if needed */
Willy Tarreau15678ef2012-08-31 13:54:11 +0200561
Willy Tarreau4c69cff2020-03-04 16:38:00 +0100562 if (conn->flags & CO_FL_WAIT_L4_CONN) {
563 fd_want_send(fd);
564 fd_cant_send(fd);
Willy Tarreau8dbd1a22020-07-31 08:59:09 +0200565 fd_cant_recv(fd);
Willy Tarreau4c69cff2020-03-04 16:38:00 +0100566 }
Willy Tarreauccf3f6d2019-09-05 17:05:05 +0200567
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200568 if (conn_xprt_init(conn) < 0) {
Willy Tarreau3f2770b2017-10-05 18:01:29 +0200569 conn_full_close(conn);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100570 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200571 return SF_ERR_RESOURCE;
Willy Tarreau184636e2012-09-06 14:04:41 +0200572 }
Willy Tarreau15678ef2012-08-31 13:54:11 +0200573
Willy Tarreaue7dff022015-04-03 01:14:29 +0200574 return SF_ERR_NONE; /* connection is OK */
Willy Tarreau9650f372009-08-16 14:02:45 +0200575}
576
Willy Tarreaue6b98942007-10-29 01:09:36 +0100577/* This function tries to bind a TCPv4/v6 listener. It may return a warning or
Willy Tarreau8ab505b2013-01-24 01:41:38 +0100578 * an error message in <errmsg> if the message is at most <errlen> bytes long
579 * (including '\0'). Note that <errmsg> may be NULL if <errlen> is also zero.
580 * The return value is composed from ERR_ABORT, ERR_WARN,
Willy Tarreaue6b98942007-10-29 01:09:36 +0100581 * ERR_ALERT, ERR_RETRYABLE and ERR_FATAL. ERR_NONE indicates that everything
582 * was alright and that no message was returned. ERR_RETRYABLE means that an
583 * error occurred but that it may vanish after a retry (eg: port in use), and
Aman Guptad94991d2012-04-06 17:39:26 -0700584 * ERR_FATAL indicates a non-fixable error. ERR_WARN and ERR_ALERT do not alter
Willy Tarreaue6b98942007-10-29 01:09:36 +0100585 * the meaning of the error, but just indicate that a message is present which
586 * should be displayed with the respective level. Last, ERR_ABORT indicates
587 * that it's pointless to try to start other listeners. No error message is
588 * returned if errlen is NULL.
589 */
590int tcp_bind_listener(struct listener *listener, char *errmsg, int errlen)
591{
Willy Tarreaue6b98942007-10-29 01:09:36 +0100592 int fd, err;
Willy Tarreauaf9a7f52020-09-01 16:12:50 +0200593 int ready;
Willy Tarreauaf9a7f52020-09-01 16:12:50 +0200594 char *msg = NULL;
Willy Tarreaue6b98942007-10-29 01:09:36 +0100595
Willy Tarreauad33acf2020-09-02 18:40:02 +0200596 err = ERR_NONE;
597
Willy Tarreaue6b98942007-10-29 01:09:36 +0100598 /* ensure we never return garbage */
Willy Tarreau8ab505b2013-01-24 01:41:38 +0100599 if (errlen)
Willy Tarreaue6b98942007-10-29 01:09:36 +0100600 *errmsg = 0;
601
602 if (listener->state != LI_ASSIGNED)
603 return ERR_NONE; /* already bound */
604
Willy Tarreauad33acf2020-09-02 18:40:02 +0200605 if (!(listener->rx.flags & RX_F_BOUND)) {
606 msg = "receiving socket not bound";
607 goto tcp_return;
Willy Tarreaue6b98942007-10-29 01:09:36 +0100608 }
609
Willy Tarreauaf9a7f52020-09-01 16:12:50 +0200610 fd = listener->rx.fd;
Willy Tarreaue6b98942007-10-29 01:09:36 +0100611
612 if (listener->options & LI_O_NOLINGER)
Simon Hormande072bd2011-06-24 15:11:37 +0900613 setsockopt(fd, SOL_SOCKET, SO_LINGER, &nolinger, sizeof(struct linger));
Olivier Houchard153659f2017-04-05 22:39:56 +0200614 else {
615 struct linger tmplinger;
616 socklen_t len = sizeof(tmplinger);
617 if (getsockopt(fd, SOL_SOCKET, SO_LINGER, &tmplinger, &len) == 0 &&
618 (tmplinger.l_onoff == 1 || tmplinger.l_linger == 0)) {
619 tmplinger.l_onoff = 0;
620 tmplinger.l_linger = 0;
621 setsockopt(fd, SOL_SOCKET, SO_LINGER, &tmplinger,
622 sizeof(tmplinger));
623 }
624 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100625
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +0400626#if defined(TCP_MAXSEG)
Willy Tarreau48a7e722010-12-24 15:26:39 +0100627 if (listener->maxseg > 0) {
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +0400628 if (setsockopt(fd, IPPROTO_TCP, TCP_MAXSEG,
Willy Tarreaube1b9182009-06-14 18:48:19 +0200629 &listener->maxseg, sizeof(listener->maxseg)) == -1) {
630 msg = "cannot set MSS";
631 err |= ERR_WARN;
632 }
Willy Tarreauaf9a7f52020-09-01 16:12:50 +0200633 } else {
634 /* we may want to try to restore the default MSS if the socket was inherited */
Olivier Houchard153659f2017-04-05 22:39:56 +0200635 int tmpmaxseg = -1;
636 int defaultmss;
637 socklen_t len = sizeof(tmpmaxseg);
638
Willy Tarreau37159062020-08-27 07:48:42 +0200639 if (listener->rx.addr.ss_family == AF_INET)
Willy Tarreaue5bdc512020-08-28 18:03:10 +0200640 defaultmss = sock_inet_tcp_maxseg_default;
Olivier Houchard153659f2017-04-05 22:39:56 +0200641 else
Willy Tarreaue5bdc512020-08-28 18:03:10 +0200642 defaultmss = sock_inet6_tcp_maxseg_default;
Olivier Houchard153659f2017-04-05 22:39:56 +0200643
644 getsockopt(fd, IPPROTO_TCP, TCP_MAXSEG, &tmpmaxseg, &len);
William Dauchy97a7bda2020-02-12 15:53:04 +0100645 if (defaultmss > 0 &&
646 tmpmaxseg != defaultmss &&
647 setsockopt(fd, IPPROTO_TCP, TCP_MAXSEG, &defaultmss, sizeof(defaultmss)) == -1) {
Olivier Houchard153659f2017-04-05 22:39:56 +0200648 msg = "cannot set MSS";
649 err |= ERR_WARN;
650 }
Willy Tarreaube1b9182009-06-14 18:48:19 +0200651 }
652#endif
Willy Tarreau2af207a2015-02-04 00:45:58 +0100653#if defined(TCP_USER_TIMEOUT)
654 if (listener->tcp_ut) {
655 if (setsockopt(fd, IPPROTO_TCP, TCP_USER_TIMEOUT,
656 &listener->tcp_ut, sizeof(listener->tcp_ut)) == -1) {
657 msg = "cannot set TCP User Timeout";
658 err |= ERR_WARN;
659 }
Olivier Houchard153659f2017-04-05 22:39:56 +0200660 } else
661 setsockopt(fd, IPPROTO_TCP, TCP_USER_TIMEOUT, &zero,
662 sizeof(zero));
Willy Tarreau2af207a2015-02-04 00:45:58 +0100663#endif
Willy Tarreaucb6cd432009-10-13 07:34:14 +0200664#if defined(TCP_DEFER_ACCEPT)
665 if (listener->options & LI_O_DEF_ACCEPT) {
666 /* defer accept by up to one second */
667 int accept_delay = 1;
668 if (setsockopt(fd, IPPROTO_TCP, TCP_DEFER_ACCEPT, &accept_delay, sizeof(accept_delay)) == -1) {
669 msg = "cannot enable DEFER_ACCEPT";
670 err |= ERR_WARN;
671 }
Olivier Houchard153659f2017-04-05 22:39:56 +0200672 } else
673 setsockopt(fd, IPPROTO_TCP, TCP_DEFER_ACCEPT, &zero,
674 sizeof(zero));
Willy Tarreaucb6cd432009-10-13 07:34:14 +0200675#endif
Willy Tarreau1c862c52012-10-05 16:21:00 +0200676#if defined(TCP_FASTOPEN)
677 if (listener->options & LI_O_TCP_FO) {
678 /* TFO needs a queue length, let's use the configured backlog */
Willy Tarreaue2711c72019-02-27 15:39:41 +0100679 int qlen = listener_backlog(listener);
Willy Tarreau1c862c52012-10-05 16:21:00 +0200680 if (setsockopt(fd, IPPROTO_TCP, TCP_FASTOPEN, &qlen, sizeof(qlen)) == -1) {
681 msg = "cannot enable TCP_FASTOPEN";
682 err |= ERR_WARN;
683 }
Olivier Houchard153659f2017-04-05 22:39:56 +0200684 } else {
685 socklen_t len;
686 int qlen;
687 len = sizeof(qlen);
688 /* Only disable fast open if it was enabled, we don't want
689 * the kernel to create a fast open queue if there's none.
690 */
691 if (getsockopt(fd, IPPROTO_TCP, TCP_FASTOPEN, &qlen, &len) == 0 &&
692 qlen != 0) {
693 if (setsockopt(fd, IPPROTO_TCP, TCP_FASTOPEN, &zero,
694 sizeof(zero)) == -1) {
695 msg = "cannot disable TCP_FASTOPEN";
696 err |= ERR_WARN;
697 }
698 }
Willy Tarreau1c862c52012-10-05 16:21:00 +0200699 }
700#endif
Willy Tarreau3e12de22020-10-13 17:42:21 +0200701
Willy Tarreau7d053e42020-10-15 09:19:43 +0200702 ready = sock_accepting_conn(&listener->rx) > 0;
Willy Tarreau40aa0702013-03-10 23:51:38 +0100703
Willy Tarreauaf9a7f52020-09-01 16:12:50 +0200704 if (!ready && /* only listen if not already done by external process */
Willy Tarreaue2711c72019-02-27 15:39:41 +0100705 listen(fd, listener_backlog(listener)) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +0100706 err |= ERR_RETRYABLE | ERR_ALERT;
707 msg = "cannot listen to socket";
708 goto tcp_close_return;
709 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100710
David Carlier1eb595b2021-02-06 12:11:11 +0000711#if !defined(TCP_DEFER_ACCEPT) && defined(SO_ACCEPTFILTER)
712 /* the socket needs to listen first */
713 if (listener->options & LI_O_DEF_ACCEPT) {
714 struct accept_filter_arg accept;
715 memset(&accept, 0, sizeof(accept));
716 strcpy(accept.af_name, "dataready");
717 if (setsockopt(fd, SOL_SOCKET, SO_ACCEPTFILTER, &accept, sizeof(accept)) == -1) {
718 msg = "cannot enable ACCEPT_FILTER";
719 err |= ERR_WARN;
720 }
721 }
722#endif
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +0400723#if defined(TCP_QUICKACK)
Willy Tarreau9ea05a72009-06-14 12:07:01 +0200724 if (listener->options & LI_O_NOQUICKACK)
Simon Hormande072bd2011-06-24 15:11:37 +0900725 setsockopt(fd, IPPROTO_TCP, TCP_QUICKACK, &zero, sizeof(zero));
Olivier Houchard153659f2017-04-05 22:39:56 +0200726 else
727 setsockopt(fd, IPPROTO_TCP, TCP_QUICKACK, &one, sizeof(one));
Willy Tarreau9ea05a72009-06-14 12:07:01 +0200728#endif
729
Willy Tarreaue6b98942007-10-29 01:09:36 +0100730 /* the socket is ready */
Willy Tarreaua37b2442020-09-24 07:23:45 +0200731 listener_set_state(listener, LI_LISTEN);
Christopher Fauleta1000032020-10-07 11:14:47 +0200732 goto tcp_return;
Willy Tarreaubb1caff2020-08-19 10:00:57 +0200733
Willy Tarreauaf9a7f52020-09-01 16:12:50 +0200734 tcp_close_return:
735 close(fd);
Willy Tarreauad33acf2020-09-02 18:40:02 +0200736 tcp_return:
Cyril Bonté43ba1b32010-11-01 19:26:01 +0100737 if (msg && errlen) {
738 char pn[INET6_ADDRSTRLEN];
739
Willy Tarreau37159062020-08-27 07:48:42 +0200740 addr_to_str(&listener->rx.addr, pn, sizeof(pn));
741 snprintf(errmsg, errlen, "%s [%s:%d]", msg, pn, get_host_port(&listener->rx.addr));
Cyril Bonté43ba1b32010-11-01 19:26:01 +0100742 }
Willy Tarreaue6b98942007-10-29 01:09:36 +0100743 return err;
Willy Tarreaue6b98942007-10-29 01:09:36 +0100744}
745
Willy Tarreau5ddf1ce2020-09-25 19:27:39 +0200746/* Enable receipt of incoming connections for listener <l>. The receiver must
Willy Tarreaua4380b22020-11-04 13:59:04 +0100747 * still be valid.
Willy Tarreau5ddf1ce2020-09-25 19:27:39 +0200748 */
749static void tcp_enable_listener(struct listener *l)
750{
Willy Tarreaua4380b22020-11-04 13:59:04 +0100751 fd_want_recv_safe(l->rx.fd);
Willy Tarreau5ddf1ce2020-09-25 19:27:39 +0200752}
753
754/* Disable receipt of incoming connections for listener <l>. The receiver must
Willy Tarreaua4380b22020-11-04 13:59:04 +0100755 * still be valid.
Willy Tarreau5ddf1ce2020-09-25 19:27:39 +0200756 */
757static void tcp_disable_listener(struct listener *l)
758{
Willy Tarreaua4380b22020-11-04 13:59:04 +0100759 fd_stop_recv(l->rx.fd);
Willy Tarreau5ddf1ce2020-09-25 19:27:39 +0200760}
761
Willy Tarreaucb66ea62020-09-25 17:12:32 +0200762/* Suspend a receiver. Returns < 0 in case of failure, 0 if the receiver
763 * was totally stopped, or > 0 if correctly suspended.
Willy Tarreau092d8652014-07-07 20:22:12 +0200764 */
Willy Tarreaucb66ea62020-09-25 17:12:32 +0200765static int tcp_suspend_receiver(struct receiver *rx)
Willy Tarreau092d8652014-07-07 20:22:12 +0200766{
Willy Tarreau2f6f3622020-10-14 10:50:41 +0200767 const struct sockaddr sa = { .sa_family = AF_UNSPEC };
Willy Tarreau3e12de22020-10-13 17:42:21 +0200768 int ret;
Willy Tarreau092d8652014-07-07 20:22:12 +0200769
Willy Tarreau59b5da42020-11-04 14:14:55 +0100770 /* we never do that with a shared FD otherwise we'd break it in the
771 * parent process and any possible subsequent worker inheriting it.
772 */
773 if (rx->flags & RX_F_INHERITED)
774 return -1;
775
Willy Tarreau8b6fc3d2020-10-13 16:34:19 +0200776 if (connect(rx->fd, &sa, sizeof(sa)) < 0)
777 goto check_already_done;
Willy Tarreau91c614d2020-10-08 16:51:09 +0200778
Willy Tarreaucb66ea62020-09-25 17:12:32 +0200779 fd_stop_recv(rx->fd);
Willy Tarreau092d8652014-07-07 20:22:12 +0200780 return 1;
Willy Tarreau91c614d2020-10-08 16:51:09 +0200781
782 check_already_done:
783 /* in case one of the shutdown() above fails, it might be because we're
784 * dealing with a socket that is shared with other processes doing the
785 * same. Let's check if it's still accepting connections.
786 */
Willy Tarreau7d053e42020-10-15 09:19:43 +0200787 ret = sock_accepting_conn(rx);
Willy Tarreau3e12de22020-10-13 17:42:21 +0200788 if (ret <= 0) {
789 /* unrecoverable or paused by another process */
Willy Tarreaucb66ea62020-09-25 17:12:32 +0200790 fd_stop_recv(rx->fd);
Willy Tarreau3e12de22020-10-13 17:42:21 +0200791 return ret == 0;
Willy Tarreaue53608b2020-09-24 18:20:37 +0200792 }
Willy Tarreau91c614d2020-10-08 16:51:09 +0200793
Willy Tarreau3e12de22020-10-13 17:42:21 +0200794 /* still listening, that's not good */
Willy Tarreau91c614d2020-10-08 16:51:09 +0200795 return -1;
Willy Tarreau092d8652014-07-07 20:22:12 +0200796}
797
Willy Tarreau58e6b712020-09-25 19:40:31 +0200798/* Resume a receiver. Returns < 0 in case of failure, 0 if the receiver
799 * was totally stopped, or > 0 if correctly suspended.
800 */
801static int tcp_resume_receiver(struct receiver *rx)
802{
803 struct listener *l = LIST_ELEM(rx, struct listener *, rx);
804
805 if (rx->fd < 0)
806 return 0;
807
808 if (listen(rx->fd, listener_backlog(l)) == 0) {
809 fd_want_recv(l->rx.fd);
810 return 1;
811 }
812 return -1;
813}
814
Willy Tarreaue6b98942007-10-29 01:09:36 +0100815
816/*
817 * Local variables:
818 * c-indent-level: 8
819 * c-basic-offset: 8
820 * End:
821 */