blob: d7650325dfe13b2fefe57d8cb4e4646e40895571 [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>
28#include <sys/stat.h>
29#include <sys/types.h>
30#include <sys/un.h>
31
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040032#include <netinet/tcp.h>
Willy Tarreau2d392c22015-08-24 01:43:45 +020033#include <netinet/in.h>
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +040034
Willy Tarreaue6b98942007-10-29 01:09:36 +010035#include <common/compat.h>
36#include <common/config.h>
37#include <common/debug.h>
38#include <common/errors.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010039#include <common/mini-clist.h>
40#include <common/standard.h>
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +010041#include <common/namespace.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010042
Willy Tarreau39713102016-11-25 15:49:32 +010043#include <types/action.h>
Willy Tarreau2d392c22015-08-24 01:43:45 +020044#include <types/connection.h>
Willy Tarreau39713102016-11-25 15:49:32 +010045#include <types/global.h>
46#include <types/stream.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010047
Willy Tarreau9fcb9842012-04-20 14:45:49 +020048#include <proto/arg.h>
Willy Tarreauc7e42382012-08-24 19:22:53 +020049#include <proto/channel.h>
Willy Tarreaud2274c62012-07-06 14:29:45 +020050#include <proto/connection.h>
Willy Tarreaudd2f85e2012-09-02 22:34:23 +020051#include <proto/fd.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>
Willy Tarreau2d392c22015-08-24 01:43:45 +020056#include <proto/proto_http.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 Tarreaue6b98942007-10-29 01:09:36 +010066
67/* Note: must not be declared <const> as its list will be overwritten */
68static struct protocol proto_tcpv4 = {
69 .name = "tcpv4",
70 .sock_domain = AF_INET,
71 .sock_type = SOCK_STREAM,
72 .sock_prot = IPPROTO_TCP,
73 .sock_family = AF_INET,
74 .sock_addrlen = sizeof(struct sockaddr_in),
75 .l3_addrlen = 32/8,
Willy Tarreaubbebbbf2012-05-07 21:22:09 +020076 .accept = &listener_accept,
Willy Tarreau26d8c592012-05-07 18:12:14 +020077 .connect = tcp_connect_server,
Emeric Bruncf20bf12010-10-22 16:06:11 +020078 .bind = tcp_bind_listener,
Willy Tarreaue6b98942007-10-29 01:09:36 +010079 .bind_all = tcp_bind_listeners,
80 .unbind_all = unbind_all_listeners,
81 .enable_all = enable_all_listeners,
Willy Tarreau59b94792012-05-11 16:16:40 +020082 .get_src = tcp_get_src,
83 .get_dst = tcp_get_dst,
Willy Tarreau2b57cb82013-06-10 19:56:38 +020084 .drain = tcp_drain,
Willy Tarreau092d8652014-07-07 20:22:12 +020085 .pause = tcp_pause_listener,
Willy Tarreaue6b98942007-10-29 01:09:36 +010086 .listeners = LIST_HEAD_INIT(proto_tcpv4.listeners),
87 .nb_listeners = 0,
88};
89
90/* Note: must not be declared <const> as its list will be overwritten */
91static struct protocol proto_tcpv6 = {
92 .name = "tcpv6",
93 .sock_domain = AF_INET6,
94 .sock_type = SOCK_STREAM,
95 .sock_prot = IPPROTO_TCP,
96 .sock_family = AF_INET6,
97 .sock_addrlen = sizeof(struct sockaddr_in6),
98 .l3_addrlen = 128/8,
Willy Tarreaubbebbbf2012-05-07 21:22:09 +020099 .accept = &listener_accept,
Willy Tarreau26d8c592012-05-07 18:12:14 +0200100 .connect = tcp_connect_server,
Emeric Bruncf20bf12010-10-22 16:06:11 +0200101 .bind = tcp_bind_listener,
Willy Tarreaue6b98942007-10-29 01:09:36 +0100102 .bind_all = tcp_bind_listeners,
103 .unbind_all = unbind_all_listeners,
104 .enable_all = enable_all_listeners,
Willy Tarreau59b94792012-05-11 16:16:40 +0200105 .get_src = tcp_get_src,
106 .get_dst = tcp_get_dst,
Willy Tarreau2b57cb82013-06-10 19:56:38 +0200107 .drain = tcp_drain,
Willy Tarreau092d8652014-07-07 20:22:12 +0200108 .pause = tcp_pause_listener,
Willy Tarreaue6b98942007-10-29 01:09:36 +0100109 .listeners = LIST_HEAD_INIT(proto_tcpv6.listeners),
110 .nb_listeners = 0,
111};
112
David du Colombier6f5ccb12011-03-10 22:26:24 +0100113/* Binds ipv4/ipv6 address <local> to socket <fd>, unless <flags> is set, in which
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100114 * case we try to bind <remote>. <flags> is a 2-bit field consisting of :
115 * - 0 : ignore remote address (may even be a NULL pointer)
116 * - 1 : use provided address
117 * - 2 : use provided port
118 * - 3 : use both
119 *
120 * The function supports multiple foreign binding methods :
121 * - linux_tproxy: we directly bind to the foreign address
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100122 * The second one can be used as a fallback for the first one.
123 * This function returns 0 when everything's OK, 1 if it could not bind, to the
124 * local address, 2 if it could not bind to the foreign address.
125 */
David du Colombier6f5ccb12011-03-10 22:26:24 +0100126int tcp_bind_socket(int fd, int flags, struct sockaddr_storage *local, struct sockaddr_storage *remote)
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100127{
David du Colombier6f5ccb12011-03-10 22:26:24 +0100128 struct sockaddr_storage bind_addr;
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100129 int foreign_ok = 0;
130 int ret;
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100131 static int ip_transp_working = 1;
David du Colombier65c17962012-07-13 14:34:59 +0200132 static int ip6_transp_working = 1;
Pieter Baauwd551fb52013-05-08 22:49:23 +0200133
David du Colombier65c17962012-07-13 14:34:59 +0200134 switch (local->ss_family) {
135 case AF_INET:
136 if (flags && ip_transp_working) {
Pieter Baauwd551fb52013-05-08 22:49:23 +0200137 /* This deserves some explanation. Some platforms will support
138 * multiple combinations of certain methods, so we try the
139 * supported ones until one succeeds.
140 */
141 if (0
142#if defined(IP_TRANSPARENT)
143 || (setsockopt(fd, SOL_IP, IP_TRANSPARENT, &one, sizeof(one)) == 0)
144#endif
145#if defined(IP_FREEBIND)
146 || (setsockopt(fd, SOL_IP, IP_FREEBIND, &one, sizeof(one)) == 0)
147#endif
Pieter Baauwff30b662013-05-08 23:22:39 +0200148#if defined(IP_BINDANY)
149 || (setsockopt(fd, IPPROTO_IP, IP_BINDANY, &one, sizeof(one)) == 0)
150#endif
Pieter Baauw1eb75922013-05-08 23:30:23 +0200151#if defined(SO_BINDANY)
152 || (setsockopt(fd, SOL_SOCKET, SO_BINDANY, &one, sizeof(one)) == 0)
153#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +0200154 )
David du Colombier65c17962012-07-13 14:34:59 +0200155 foreign_ok = 1;
156 else
157 ip_transp_working = 0;
158 }
159 break;
160 case AF_INET6:
161 if (flags && ip6_transp_working) {
Pieter Baauwd551fb52013-05-08 22:49:23 +0200162 if (0
Dinko Korunic7276f3a2016-09-09 09:41:15 +0200163#if defined(IPV6_TRANSPARENT) && defined(SOL_IPV6)
Pieter Baauwd551fb52013-05-08 22:49:23 +0200164 || (setsockopt(fd, SOL_IPV6, IPV6_TRANSPARENT, &one, sizeof(one)) == 0)
165#endif
Lukas Tribus7640e722014-03-03 21:10:51 +0100166#if defined(IP_FREEBIND)
167 || (setsockopt(fd, SOL_IP, IP_FREEBIND, &one, sizeof(one)) == 0)
168#endif
Pieter Baauwff30b662013-05-08 23:22:39 +0200169#if defined(IPV6_BINDANY)
170 || (setsockopt(fd, IPPROTO_IPV6, IPV6_BINDANY, &one, sizeof(one)) == 0)
171#endif
Pieter Baauw1eb75922013-05-08 23:30:23 +0200172#if defined(SO_BINDANY)
173 || (setsockopt(fd, SOL_SOCKET, SO_BINDANY, &one, sizeof(one)) == 0)
174#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +0200175 )
David du Colombier65c17962012-07-13 14:34:59 +0200176 foreign_ok = 1;
177 else
178 ip6_transp_working = 0;
179 }
180 break;
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100181 }
Pieter Baauwd551fb52013-05-08 22:49:23 +0200182
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100183 if (flags) {
184 memset(&bind_addr, 0, sizeof(bind_addr));
Willy Tarreau96dd0792011-04-19 07:20:57 +0200185 bind_addr.ss_family = remote->ss_family;
David du Colombier6f5ccb12011-03-10 22:26:24 +0100186 switch (remote->ss_family) {
187 case AF_INET:
188 if (flags & 1)
189 ((struct sockaddr_in *)&bind_addr)->sin_addr = ((struct sockaddr_in *)remote)->sin_addr;
190 if (flags & 2)
191 ((struct sockaddr_in *)&bind_addr)->sin_port = ((struct sockaddr_in *)remote)->sin_port;
192 break;
193 case AF_INET6:
194 if (flags & 1)
195 ((struct sockaddr_in6 *)&bind_addr)->sin6_addr = ((struct sockaddr_in6 *)remote)->sin6_addr;
196 if (flags & 2)
197 ((struct sockaddr_in6 *)&bind_addr)->sin6_port = ((struct sockaddr_in6 *)remote)->sin6_port;
198 break;
Willy Tarreau5dc1e982011-12-16 21:25:11 +0100199 default:
200 /* we don't want to try to bind to an unknown address family */
201 foreign_ok = 0;
David du Colombier6f5ccb12011-03-10 22:26:24 +0100202 }
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100203 }
204
Simon Hormande072bd2011-06-24 15:11:37 +0900205 setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &one, sizeof(one));
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100206 if (foreign_ok) {
Willy Tarreau9cf8d3f2014-05-09 22:56:10 +0200207 if (is_inet_addr(&bind_addr)) {
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200208 ret = bind(fd, (struct sockaddr *)&bind_addr, get_addr_len(&bind_addr));
209 if (ret < 0)
210 return 2;
211 }
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100212 }
213 else {
Willy Tarreau9cf8d3f2014-05-09 22:56:10 +0200214 if (is_inet_addr(local)) {
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200215 ret = bind(fd, (struct sockaddr *)local, get_addr_len(local));
216 if (ret < 0)
217 return 1;
218 }
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100219 }
220
221 if (!flags)
222 return 0;
223
Willy Tarreaue8c66af2008-01-13 18:40:14 +0100224 if (!foreign_ok)
225 /* we could not bind to a foreign address */
226 return 2;
227
228 return 0;
229}
Willy Tarreaue6b98942007-10-29 01:09:36 +0100230
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100231static int create_server_socket(struct connection *conn)
232{
Willy Tarreau529c1392014-12-24 13:47:55 +0100233 const struct netns_entry *ns = NULL;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100234
Willy Tarreau529c1392014-12-24 13:47:55 +0100235#ifdef CONFIG_HAP_NS
236 if (objt_server(conn->target)) {
237 if (__objt_server(conn->target)->flags & SRV_F_USE_NS_FROM_PP)
238 ns = conn->proxy_netns;
239 else
240 ns = __objt_server(conn->target)->netns;
241 }
242#endif
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100243 return my_socketat(ns, conn->addr.to.ss_family, SOCK_STREAM, IPPROTO_TCP);
244}
Willy Tarreau9650f372009-08-16 14:02:45 +0200245
246/*
Willy Tarreau14f8e862012-08-30 22:23:13 +0200247 * This function initiates a TCP connection establishment to the target assigned
248 * to connection <conn> using (si->{target,addr.to}). A source address may be
249 * pointed to by conn->addr.from in case of transparent proxying. Normal source
250 * bind addresses are still determined locally (due to the possible need of a
251 * source port). conn->target may point either to a valid server or to a backend,
Willy Tarreau3fdb3662012-11-12 00:42:33 +0100252 * depending on conn->target. Only OBJ_TYPE_PROXY and OBJ_TYPE_SERVER are
Willy Tarreauf0837b22012-11-24 10:24:27 +0100253 * supported. The <data> parameter is a boolean indicating whether there are data
254 * waiting for being sent or not, in order to adjust data write polling and on
255 * some platforms, the ability to avoid an empty initial ACK. The <delack> argument
256 * allows the caller to force using a delayed ACK when establishing the connection :
257 * - 0 = no delayed ACK unless data are advertised and backend has tcp-smart-connect
258 * - 1 = delayed ACK if backend has tcp-smart-connect, regardless of data
259 * - 2 = delayed ACK regardless of backend options
Willy Tarreaub1d67742010-03-29 19:36:59 +0200260 *
Willy Tarreau1ec74bf2013-10-24 21:45:00 +0200261 * Note that a pending send_proxy message accounts for data.
262 *
Willy Tarreau9650f372009-08-16 14:02:45 +0200263 * It can return one of :
Willy Tarreaue7dff022015-04-03 01:14:29 +0200264 * - SF_ERR_NONE if everything's OK
265 * - SF_ERR_SRVTO if there are no more servers
266 * - SF_ERR_SRVCL if the connection was refused by the server
267 * - SF_ERR_PRXCOND if the connection has been limited by the proxy (maxconn)
268 * - SF_ERR_RESOURCE if a system resource is lacking (eg: fd limits, ports, ...)
269 * - SF_ERR_INTERNAL for any other purely internal errors
Tim Düsterhus4896c442016-11-29 02:15:19 +0100270 * Additionally, in the case of SF_ERR_RESOURCE, an emergency log will be emitted.
Willy Tarreau6b0a8502012-11-23 08:51:32 +0100271 *
Willy Tarreaue7dff022015-04-03 01:14:29 +0200272 * The connection's fd is inserted only when SF_ERR_NONE is returned, otherwise
Willy Tarreau6b0a8502012-11-23 08:51:32 +0100273 * it's invalid and the caller has nothing to do.
Willy Tarreau9650f372009-08-16 14:02:45 +0200274 */
Willy Tarreauf1536862011-03-03 18:27:32 +0100275
Willy Tarreauf0837b22012-11-24 10:24:27 +0100276int tcp_connect_server(struct connection *conn, int data, int delack)
Willy Tarreau9650f372009-08-16 14:02:45 +0200277{
278 int fd;
Willy Tarreauac825402011-03-04 22:04:29 +0100279 struct server *srv;
280 struct proxy *be;
Willy Tarreaua4380b42012-12-08 22:49:11 +0100281 struct conn_src *src;
Willy Tarreauac825402011-03-04 22:04:29 +0100282
Willy Tarreau9ce70132014-01-24 16:08:19 +0100283 conn->flags = CO_FL_WAIT_L4_CONN; /* connection in progress */
284
Willy Tarreau3fdb3662012-11-12 00:42:33 +0100285 switch (obj_type(conn->target)) {
286 case OBJ_TYPE_PROXY:
287 be = objt_proxy(conn->target);
Willy Tarreauac825402011-03-04 22:04:29 +0100288 srv = NULL;
289 break;
Willy Tarreau3fdb3662012-11-12 00:42:33 +0100290 case OBJ_TYPE_SERVER:
291 srv = objt_server(conn->target);
Willy Tarreauac825402011-03-04 22:04:29 +0100292 be = srv->proxy;
293 break;
294 default:
Willy Tarreau9ce70132014-01-24 16:08:19 +0100295 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200296 return SF_ERR_INTERNAL;
Willy Tarreauac825402011-03-04 22:04:29 +0100297 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200298
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100299 fd = conn->t.sock.fd = create_server_socket(conn);
300
301 if (fd == -1) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200302 qfprintf(stderr, "Cannot get a server socket.\n");
303
Willy Tarreau9ce70132014-01-24 16:08:19 +0100304 if (errno == ENFILE) {
305 conn->err_code = CO_ER_SYS_FDLIM;
Willy Tarreau9650f372009-08-16 14:02:45 +0200306 send_log(be, LOG_EMERG,
307 "Proxy %s reached system FD limit at %d. Please check system tunables.\n",
308 be->id, maxfd);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100309 }
310 else if (errno == EMFILE) {
311 conn->err_code = CO_ER_PROC_FDLIM;
Willy Tarreau9650f372009-08-16 14:02:45 +0200312 send_log(be, LOG_EMERG,
313 "Proxy %s reached process FD limit at %d. Please check 'ulimit-n' and restart.\n",
314 be->id, maxfd);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100315 }
316 else if (errno == ENOBUFS || errno == ENOMEM) {
317 conn->err_code = CO_ER_SYS_MEMLIM;
Willy Tarreau9650f372009-08-16 14:02:45 +0200318 send_log(be, LOG_EMERG,
319 "Proxy %s reached system memory limit at %d sockets. Please check system tunables.\n",
320 be->id, maxfd);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100321 }
322 else if (errno == EAFNOSUPPORT || errno == EPROTONOSUPPORT) {
323 conn->err_code = CO_ER_NOPROTO;
324 }
325 else
326 conn->err_code = CO_ER_SOCK_ERR;
327
Willy Tarreau9650f372009-08-16 14:02:45 +0200328 /* this is a resource error */
Willy Tarreau9ce70132014-01-24 16:08:19 +0100329 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200330 return SF_ERR_RESOURCE;
Willy Tarreau9650f372009-08-16 14:02:45 +0200331 }
332
333 if (fd >= global.maxsock) {
334 /* do not log anything there, it's a normal condition when this option
335 * is used to serialize connections to a server !
336 */
337 Alert("socket(): not enough free sockets. Raise -n argument. Giving up.\n");
338 close(fd);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100339 conn->err_code = CO_ER_CONF_FDLIM;
340 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200341 return SF_ERR_PRXCOND; /* it is a configuration limit */
Willy Tarreau9650f372009-08-16 14:02:45 +0200342 }
343
344 if ((fcntl(fd, F_SETFL, O_NONBLOCK)==-1) ||
Simon Hormande072bd2011-06-24 15:11:37 +0900345 (setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &one, sizeof(one)) == -1)) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200346 qfprintf(stderr,"Cannot set client socket to non blocking mode.\n");
347 close(fd);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100348 conn->err_code = CO_ER_SOCK_ERR;
349 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200350 return SF_ERR_INTERNAL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200351 }
352
353 if (be->options & PR_O_TCP_SRV_KA)
Simon Hormande072bd2011-06-24 15:11:37 +0900354 setsockopt(fd, SOL_SOCKET, SO_KEEPALIVE, &one, sizeof(one));
Willy Tarreau9650f372009-08-16 14:02:45 +0200355
Willy Tarreau9650f372009-08-16 14:02:45 +0200356 /* allow specific binding :
357 * - server-specific at first
358 * - proxy-specific next
359 */
Willy Tarreaua4380b42012-12-08 22:49:11 +0100360 if (srv && srv->conn_src.opts & CO_SRC_BIND)
361 src = &srv->conn_src;
362 else if (be->conn_src.opts & CO_SRC_BIND)
363 src = &be->conn_src;
364 else
365 src = NULL;
366
367 if (src) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200368 int ret, flags = 0;
369
Willy Tarreau9cf8d3f2014-05-09 22:56:10 +0200370 if (is_inet_addr(&conn->addr.from)) {
Willy Tarreaua4380b42012-12-08 22:49:11 +0100371 switch (src->opts & CO_SRC_TPROXY_MASK) {
Willy Tarreauef9a3602012-12-08 22:29:20 +0100372 case CO_SRC_TPROXY_CLI:
Willy Tarreau387ebf82015-08-04 19:24:13 +0200373 conn->flags |= CO_FL_PRIVATE;
374 /* fall through */
375 case CO_SRC_TPROXY_ADDR:
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200376 flags = 3;
377 break;
Willy Tarreauef9a3602012-12-08 22:29:20 +0100378 case CO_SRC_TPROXY_CIP:
379 case CO_SRC_TPROXY_DYN:
Willy Tarreau387ebf82015-08-04 19:24:13 +0200380 conn->flags |= CO_FL_PRIVATE;
Willy Tarreau5f2877a2012-10-26 19:57:58 +0200381 flags = 1;
382 break;
383 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200384 }
Willy Tarreaub1d67742010-03-29 19:36:59 +0200385
Willy Tarreau9650f372009-08-16 14:02:45 +0200386#ifdef SO_BINDTODEVICE
387 /* Note: this might fail if not CAP_NET_RAW */
Willy Tarreaua4380b42012-12-08 22:49:11 +0100388 if (src->iface_name)
389 setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE, src->iface_name, src->iface_len + 1);
Willy Tarreau9650f372009-08-16 14:02:45 +0200390#endif
391
Willy Tarreaua4380b42012-12-08 22:49:11 +0100392 if (src->sport_range) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200393 int attempts = 10; /* should be more than enough to find a spare port */
Willy Tarreaua4380b42012-12-08 22:49:11 +0100394 struct sockaddr_storage sa;
Willy Tarreau9650f372009-08-16 14:02:45 +0200395
396 ret = 1;
Vincent Bernat6e615892016-05-18 16:17:44 +0200397 memcpy(&sa, &src->source_addr, sizeof(sa));
Willy Tarreau9650f372009-08-16 14:02:45 +0200398
399 do {
400 /* note: in case of retry, we may have to release a previously
401 * allocated port, hence this loop's construct.
402 */
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200403 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
404 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200405
406 if (!attempts)
407 break;
408 attempts--;
409
Willy Tarreaua4380b42012-12-08 22:49:11 +0100410 fdinfo[fd].local_port = port_range_alloc_port(src->sport_range);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100411 if (!fdinfo[fd].local_port) {
412 conn->err_code = CO_ER_PORT_RANGE;
Willy Tarreau9650f372009-08-16 14:02:45 +0200413 break;
Willy Tarreau9ce70132014-01-24 16:08:19 +0100414 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200415
Willy Tarreaua4380b42012-12-08 22:49:11 +0100416 fdinfo[fd].port_range = src->sport_range;
417 set_host_port(&sa, fdinfo[fd].local_port);
Willy Tarreau9650f372009-08-16 14:02:45 +0200418
Willy Tarreaua4380b42012-12-08 22:49:11 +0100419 ret = tcp_bind_socket(fd, flags, &sa, &conn->addr.from);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100420 if (ret != 0)
421 conn->err_code = CO_ER_CANT_BIND;
Willy Tarreau9650f372009-08-16 14:02:45 +0200422 } while (ret != 0); /* binding NOK */
423 }
424 else {
Lukas Tribus7d56c6d2016-09-13 09:51:15 +0000425#ifdef IP_BIND_ADDRESS_NO_PORT
426 static int bind_address_no_port = 1;
427 setsockopt(fd, SOL_IP, IP_BIND_ADDRESS_NO_PORT, (const void *) &bind_address_no_port, sizeof(int));
428#endif
Willy Tarreaua4380b42012-12-08 22:49:11 +0100429 ret = tcp_bind_socket(fd, flags, &src->source_addr, &conn->addr.from);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100430 if (ret != 0)
431 conn->err_code = CO_ER_CANT_BIND;
Willy Tarreau9650f372009-08-16 14:02:45 +0200432 }
433
Willy Tarreaua4380b42012-12-08 22:49:11 +0100434 if (unlikely(ret != 0)) {
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200435 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
436 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200437 close(fd);
438
Willy Tarreau9650f372009-08-16 14:02:45 +0200439 if (ret == 1) {
Willy Tarreaua4380b42012-12-08 22:49:11 +0100440 Alert("Cannot bind to source address before connect() for backend %s. Aborting.\n",
Willy Tarreau9650f372009-08-16 14:02:45 +0200441 be->id);
442 send_log(be, LOG_EMERG,
Willy Tarreaua4380b42012-12-08 22:49:11 +0100443 "Cannot bind to source address before connect() for backend %s.\n",
Willy Tarreau9650f372009-08-16 14:02:45 +0200444 be->id);
445 } else {
Willy Tarreaua4380b42012-12-08 22:49:11 +0100446 Alert("Cannot bind to tproxy source address before connect() for backend %s. Aborting.\n",
Willy Tarreau9650f372009-08-16 14:02:45 +0200447 be->id);
448 send_log(be, LOG_EMERG,
Willy Tarreaua4380b42012-12-08 22:49:11 +0100449 "Cannot bind to tproxy source address before connect() for backend %s.\n",
Willy Tarreau9650f372009-08-16 14:02:45 +0200450 be->id);
451 }
Willy Tarreau9ce70132014-01-24 16:08:19 +0100452 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200453 return SF_ERR_RESOURCE;
Willy Tarreau9650f372009-08-16 14:02:45 +0200454 }
455 }
456
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +0400457#if defined(TCP_QUICKACK)
Willy Tarreau9650f372009-08-16 14:02:45 +0200458 /* disabling tcp quick ack now allows the first request to leave the
459 * machine with the first ACK. We only do this if there are pending
Willy Tarreauf0837b22012-11-24 10:24:27 +0100460 * data in the buffer.
Willy Tarreau9650f372009-08-16 14:02:45 +0200461 */
Willy Tarreaufb20e462014-10-24 12:02:24 +0200462 if (delack == 2 || ((delack || data || conn->send_proxy_ofs) && (be->options2 & PR_O2_SMARTCON)))
Simon Hormande072bd2011-06-24 15:11:37 +0900463 setsockopt(fd, IPPROTO_TCP, TCP_QUICKACK, &zero, sizeof(zero));
Willy Tarreau9650f372009-08-16 14:02:45 +0200464#endif
465
Willy Tarreau163d4622015-10-13 16:16:41 +0200466#ifdef TCP_USER_TIMEOUT
467 /* there is not much more we can do here when it fails, it's still minor */
468 if (srv && srv->tcp_ut)
469 setsockopt(fd, IPPROTO_TCP, TCP_USER_TIMEOUT, &srv->tcp_ut, sizeof(srv->tcp_ut));
470#endif
Willy Tarreaue803de22010-01-21 17:43:04 +0100471 if (global.tune.server_sndbuf)
472 setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &global.tune.server_sndbuf, sizeof(global.tune.server_sndbuf));
473
474 if (global.tune.server_rcvbuf)
475 setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &global.tune.server_rcvbuf, sizeof(global.tune.server_rcvbuf));
476
Willy Tarreau986a9d22012-08-30 21:11:38 +0200477 if ((connect(fd, (struct sockaddr *)&conn->addr.to, get_addr_len(&conn->addr.to)) == -1) &&
Willy Tarreau9650f372009-08-16 14:02:45 +0200478 (errno != EINPROGRESS) && (errno != EALREADY) && (errno != EISCONN)) {
479
Willy Tarreaub1719512012-12-08 23:03:28 +0100480 if (errno == EAGAIN || errno == EADDRINUSE || errno == EADDRNOTAVAIL) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200481 char *msg;
Willy Tarreau9ce70132014-01-24 16:08:19 +0100482 if (errno == EAGAIN || errno == EADDRNOTAVAIL) {
Willy Tarreau9650f372009-08-16 14:02:45 +0200483 msg = "no free ports";
Willy Tarreau9ce70132014-01-24 16:08:19 +0100484 conn->err_code = CO_ER_FREE_PORTS;
485 }
486 else {
Willy Tarreau9650f372009-08-16 14:02:45 +0200487 msg = "local address already in use";
Willy Tarreau9ce70132014-01-24 16:08:19 +0100488 conn->err_code = CO_ER_ADDR_INUSE;
489 }
Willy Tarreau9650f372009-08-16 14:02:45 +0200490
Willy Tarreaub1719512012-12-08 23:03:28 +0100491 qfprintf(stderr,"Connect() failed for backend %s: %s.\n", be->id, msg);
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200492 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
493 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200494 close(fd);
Willy Tarreaub1719512012-12-08 23:03:28 +0100495 send_log(be, LOG_ERR, "Connect() failed for backend %s: %s.\n", be->id, msg);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100496 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200497 return SF_ERR_RESOURCE;
Willy Tarreau9650f372009-08-16 14:02:45 +0200498 } else if (errno == ETIMEDOUT) {
499 //qfprintf(stderr,"Connect(): ETIMEDOUT");
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200500 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
501 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200502 close(fd);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100503 conn->err_code = CO_ER_SOCK_ERR;
504 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200505 return SF_ERR_SRVTO;
Willy Tarreau9650f372009-08-16 14:02:45 +0200506 } else {
507 // (errno == ECONNREFUSED || errno == ENETUNREACH || errno == EACCES || errno == EPERM)
508 //qfprintf(stderr,"Connect(): %d", errno);
Willy Tarreau8d5d77e2009-10-18 07:25:52 +0200509 port_range_release_port(fdinfo[fd].port_range, fdinfo[fd].local_port);
510 fdinfo[fd].port_range = NULL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200511 close(fd);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100512 conn->err_code = CO_ER_SOCK_ERR;
513 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200514 return SF_ERR_SRVCL;
Willy Tarreau9650f372009-08-16 14:02:45 +0200515 }
516 }
517
Willy Tarreaufc8f1f02012-12-08 18:53:44 +0100518 conn->flags |= CO_FL_ADDR_TO_SET;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200519
Willy Tarreau1ec74bf2013-10-24 21:45:00 +0200520 /* Prepare to send a few handshakes related to the on-wire protocol. */
521 if (conn->send_proxy_ofs)
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200522 conn->flags |= CO_FL_SEND_PROXY;
Willy Tarreau1ec74bf2013-10-24 21:45:00 +0200523
Willy Tarreauf79c8172013-10-21 16:30:56 +0200524 conn_ctrl_init(conn); /* registers the FD */
Willy Tarreauad38ace2013-12-15 14:19:38 +0100525 fdtab[fd].linger_risk = 1; /* close hard if needed */
Willy Tarreau986a9d22012-08-30 21:11:38 +0200526 conn_sock_want_send(conn); /* for connect status */
Willy Tarreau15678ef2012-08-31 13:54:11 +0200527
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200528 if (conn_xprt_init(conn) < 0) {
Willy Tarreauf79c8172013-10-21 16:30:56 +0200529 conn_force_close(conn);
Willy Tarreau9ce70132014-01-24 16:08:19 +0100530 conn->flags |= CO_FL_ERROR;
Willy Tarreaue7dff022015-04-03 01:14:29 +0200531 return SF_ERR_RESOURCE;
Willy Tarreau184636e2012-09-06 14:04:41 +0200532 }
Willy Tarreau15678ef2012-08-31 13:54:11 +0200533
Willy Tarreau14f8e862012-08-30 22:23:13 +0200534 if (data)
Willy Tarreau986a9d22012-08-30 21:11:38 +0200535 conn_data_want_send(conn); /* prepare to send data if any */
Willy Tarreau9650f372009-08-16 14:02:45 +0200536
Willy Tarreaue7dff022015-04-03 01:14:29 +0200537 return SF_ERR_NONE; /* connection is OK */
Willy Tarreau9650f372009-08-16 14:02:45 +0200538}
539
540
Willy Tarreau59b94792012-05-11 16:16:40 +0200541/*
542 * Retrieves the source address for the socket <fd>, with <dir> indicating
543 * if we're a listener (=0) or an initiator (!=0). It returns 0 in case of
544 * success, -1 in case of error. The socket's source address is stored in
545 * <sa> for <salen> bytes.
546 */
547int tcp_get_src(int fd, struct sockaddr *sa, socklen_t salen, int dir)
548{
549 if (dir)
550 return getsockname(fd, sa, &salen);
551 else
552 return getpeername(fd, sa, &salen);
553}
554
555
556/*
557 * Retrieves the original destination address for the socket <fd>, with <dir>
558 * indicating if we're a listener (=0) or an initiator (!=0). In the case of a
559 * listener, if the original destination address was translated, the original
560 * address is retrieved. It returns 0 in case of success, -1 in case of error.
561 * The socket's source address is stored in <sa> for <salen> bytes.
562 */
563int tcp_get_dst(int fd, struct sockaddr *sa, socklen_t salen, int dir)
564{
565 if (dir)
566 return getpeername(fd, sa, &salen);
Willy Tarreau5e0d0e02014-10-29 21:46:01 +0100567 else {
568 int ret = getsockname(fd, sa, &salen);
569
570 if (ret < 0)
571 return ret;
572
Willy Tarreau59b94792012-05-11 16:16:40 +0200573#if defined(TPROXY) && defined(SO_ORIGINAL_DST)
Willy Tarreau5e0d0e02014-10-29 21:46:01 +0100574 /* For TPROXY and Netfilter's NAT, we can retrieve the original
575 * IPv4 address before DNAT/REDIRECT. We must not do that with
576 * other families because v6-mapped IPv4 addresses are still
577 * reported as v4.
578 */
579 if (((struct sockaddr_storage *)sa)->ss_family == AF_INET
580 && getsockopt(fd, SOL_IP, SO_ORIGINAL_DST, sa, &salen) == 0)
581 return 0;
Willy Tarreau59b94792012-05-11 16:16:40 +0200582#endif
Willy Tarreau5e0d0e02014-10-29 21:46:01 +0100583 return ret;
584 }
Willy Tarreau59b94792012-05-11 16:16:40 +0200585}
586
Willy Tarreau2b57cb82013-06-10 19:56:38 +0200587/* Tries to drain any pending incoming data from the socket to reach the
Willy Tarreau7f4bcc32014-01-20 11:26:12 +0100588 * receive shutdown. Returns positive if the shutdown was found, negative
589 * if EAGAIN was hit, otherwise zero. This is useful to decide whether we
590 * can close a connection cleanly are we must kill it hard.
Willy Tarreau2b57cb82013-06-10 19:56:38 +0200591 */
592int tcp_drain(int fd)
593{
594 int turns = 2;
595 int len;
596
597 while (turns) {
598#ifdef MSG_TRUNC_CLEARS_INPUT
599 len = recv(fd, NULL, INT_MAX, MSG_DONTWAIT | MSG_NOSIGNAL | MSG_TRUNC);
600 if (len == -1 && errno == EFAULT)
601#endif
602 len = recv(fd, trash.str, trash.size, MSG_DONTWAIT | MSG_NOSIGNAL);
603
Willy Tarreau3bd3e572014-01-20 11:56:37 +0100604 if (len == 0) {
605 /* cool, shutdown received */
606 fdtab[fd].linger_risk = 0;
Willy Tarreau2b57cb82013-06-10 19:56:38 +0200607 return 1;
Willy Tarreau3bd3e572014-01-20 11:56:37 +0100608 }
Willy Tarreau2b57cb82013-06-10 19:56:38 +0200609
610 if (len < 0) {
Willy Tarreauf817e9f2014-01-10 16:58:45 +0100611 if (errno == EAGAIN) {
612 /* connection not closed yet */
613 fd_cant_recv(fd);
Willy Tarreau7f4bcc32014-01-20 11:26:12 +0100614 return -1;
Willy Tarreauf817e9f2014-01-10 16:58:45 +0100615 }
Willy Tarreau2b57cb82013-06-10 19:56:38 +0200616 if (errno == EINTR) /* oops, try again */
617 continue;
618 /* other errors indicate a dead connection, fine. */
Willy Tarreau3bd3e572014-01-20 11:56:37 +0100619 fdtab[fd].linger_risk = 0;
Willy Tarreau2b57cb82013-06-10 19:56:38 +0200620 return 1;
621 }
622 /* OK we read some data, let's try again once */
623 turns--;
624 }
625 /* some data are still present, give up */
626 return 0;
627}
628
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200629/* This is the callback which is set when a connection establishment is pending
Willy Tarreauf12a20e2013-12-04 16:11:04 +0100630 * and we have nothing to send. It updates the FD polling status. It returns 0
631 * if it fails in a fatal way or needs to poll to go further, otherwise it
632 * returns non-zero and removes the CO_FL_WAIT_L4_CONN flag from the connection's
633 * flags. In case of error, it sets CO_FL_ERROR and leaves the error code in
634 * errno. The error checking is done in two passes in order to limit the number
635 * of syscalls in the normal case :
636 * - if POLL_ERR was reported by the poller, we check for a pending error on
637 * the socket before proceeding. If found, it's assigned to errno so that
638 * upper layers can see it.
639 * - otherwise connect() is used to check the connection state again, since
640 * the getsockopt return cannot reliably be used to know if the connection
641 * is still pending or ready. This one may often return an error as well,
642 * since we don't always have POLL_ERR (eg: OSX or cached events).
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200643 */
Willy Tarreau239d7182012-07-23 18:53:03 +0200644int tcp_connect_probe(struct connection *conn)
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200645{
Willy Tarreau239d7182012-07-23 18:53:03 +0200646 int fd = conn->t.sock.fd;
Willy Tarreauf12a20e2013-12-04 16:11:04 +0100647 socklen_t lskerr;
648 int skerr;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200649
Willy Tarreau80184712012-07-06 14:54:49 +0200650 if (conn->flags & CO_FL_ERROR)
Willy Tarreauafad0e02012-08-09 14:45:22 +0200651 return 0;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200652
Willy Tarreau3c728722014-01-23 13:50:42 +0100653 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200654 return 0;
655
Willy Tarreau80184712012-07-06 14:54:49 +0200656 if (!(conn->flags & CO_FL_WAIT_L4_CONN))
Willy Tarreau0b0c0972012-07-23 20:05:00 +0200657 return 1; /* strange we were called while ready */
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200658
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100659 if (!fd_send_ready(fd))
660 return 0;
661
Willy Tarreauf12a20e2013-12-04 16:11:04 +0100662 /* we might be the first witness of FD_POLL_ERR. Note that FD_POLL_HUP
663 * without FD_POLL_IN also indicates a hangup without input data meaning
664 * there was no connection.
665 */
666 if (fdtab[fd].ev & FD_POLL_ERR ||
667 (fdtab[fd].ev & (FD_POLL_IN|FD_POLL_HUP)) == FD_POLL_HUP) {
668 skerr = 0;
669 lskerr = sizeof(skerr);
670 getsockopt(fd, SOL_SOCKET, SO_ERROR, &skerr, &lskerr);
671 errno = skerr;
672 if (errno == EAGAIN)
673 errno = 0;
674 if (errno)
675 goto out_error;
676 }
Willy Tarreau2da156f2012-07-23 15:07:23 +0200677
Willy Tarreauf12a20e2013-12-04 16:11:04 +0100678 /* Use connect() to check the state of the socket. This has the
679 * advantage of giving us the following info :
Willy Tarreau2c6be842012-07-06 17:12:34 +0200680 * - error
681 * - connecting (EALREADY, EINPROGRESS)
682 * - connected (EISCONN, 0)
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200683 */
Willy Tarreau986a9d22012-08-30 21:11:38 +0200684 if (connect(fd, (struct sockaddr *)&conn->addr.to, get_addr_len(&conn->addr.to)) < 0) {
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200685 if (errno == EALREADY || errno == EINPROGRESS) {
Willy Tarreaud486ef52012-12-10 17:03:52 +0100686 __conn_sock_stop_recv(conn);
Willy Tarreaue1f50c42014-01-22 20:02:06 +0100687 fd_cant_send(fd);
Willy Tarreau0b0c0972012-07-23 20:05:00 +0200688 return 0;
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200689 }
Willy Tarreaua190d592012-05-20 18:35:19 +0200690
Willy Tarreau2c6be842012-07-06 17:12:34 +0200691 if (errno && errno != EISCONN)
Willy Tarreaua190d592012-05-20 18:35:19 +0200692 goto out_error;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200693
Willy Tarreau2c6be842012-07-06 17:12:34 +0200694 /* otherwise we're connected */
Willy Tarreaua190d592012-05-20 18:35:19 +0200695 }
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200696
Willy Tarreau076be252012-07-06 16:02:29 +0200697 /* The FD is ready now, we'll mark the connection as complete and
Willy Tarreauf7bc57c2012-10-03 00:19:48 +0200698 * forward the event to the transport layer which will notify the
699 * data layer.
Willy Tarreaua190d592012-05-20 18:35:19 +0200700 */
Willy Tarreau80184712012-07-06 14:54:49 +0200701 conn->flags &= ~CO_FL_WAIT_L4_CONN;
Willy Tarreau0b0c0972012-07-23 20:05:00 +0200702 return 1;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200703
704 out_error:
Willy Tarreau0b0c0972012-07-23 20:05:00 +0200705 /* Write error on the file descriptor. Report it to the connection
706 * and disable polling on this FD.
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200707 */
Willy Tarreau3bd3e572014-01-20 11:56:37 +0100708 fdtab[fd].linger_risk = 0;
Willy Tarreau26f4a042013-12-04 23:44:10 +0100709 conn->flags |= CO_FL_ERROR | CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreaud486ef52012-12-10 17:03:52 +0100710 __conn_sock_stop_both(conn);
Willy Tarreauafad0e02012-08-09 14:45:22 +0200711 return 0;
Willy Tarreaueeda90e2012-05-11 19:53:32 +0200712}
713
Willy Tarreau59b94792012-05-11 16:16:40 +0200714
Willy Tarreaue6b98942007-10-29 01:09:36 +0100715/* This function tries to bind a TCPv4/v6 listener. It may return a warning or
Willy Tarreau8ab505b2013-01-24 01:41:38 +0100716 * an error message in <errmsg> if the message is at most <errlen> bytes long
717 * (including '\0'). Note that <errmsg> may be NULL if <errlen> is also zero.
718 * The return value is composed from ERR_ABORT, ERR_WARN,
Willy Tarreaue6b98942007-10-29 01:09:36 +0100719 * ERR_ALERT, ERR_RETRYABLE and ERR_FATAL. ERR_NONE indicates that everything
720 * was alright and that no message was returned. ERR_RETRYABLE means that an
721 * error occurred but that it may vanish after a retry (eg: port in use), and
Aman Guptad94991d2012-04-06 17:39:26 -0700722 * ERR_FATAL indicates a non-fixable error. ERR_WARN and ERR_ALERT do not alter
Willy Tarreaue6b98942007-10-29 01:09:36 +0100723 * the meaning of the error, but just indicate that a message is present which
724 * should be displayed with the respective level. Last, ERR_ABORT indicates
725 * that it's pointless to try to start other listeners. No error message is
726 * returned if errlen is NULL.
727 */
728int tcp_bind_listener(struct listener *listener, char *errmsg, int errlen)
729{
730 __label__ tcp_return, tcp_close_return;
731 int fd, err;
Willy Tarreau40aa0702013-03-10 23:51:38 +0100732 int ext, ready;
733 socklen_t ready_len;
Willy Tarreaue6b98942007-10-29 01:09:36 +0100734 const char *msg = NULL;
735
736 /* ensure we never return garbage */
Willy Tarreau8ab505b2013-01-24 01:41:38 +0100737 if (errlen)
Willy Tarreaue6b98942007-10-29 01:09:36 +0100738 *errmsg = 0;
739
740 if (listener->state != LI_ASSIGNED)
741 return ERR_NONE; /* already bound */
742
743 err = ERR_NONE;
744
Willy Tarreau40aa0702013-03-10 23:51:38 +0100745 /* if the listener already has an fd assigned, then we were offered the
746 * fd by an external process (most likely the parent), and we don't want
747 * to create a new socket. However we still want to set a few flags on
748 * the socket.
749 */
750 fd = listener->fd;
751 ext = (fd >= 0);
752
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100753 if (!ext) {
754 fd = my_socketat(listener->netns, listener->addr.ss_family, SOCK_STREAM, IPPROTO_TCP);
755
756 if (fd == -1) {
757 err |= ERR_RETRYABLE | ERR_ALERT;
758 msg = "cannot create listening socket";
759 goto tcp_return;
760 }
Willy Tarreaue6b98942007-10-29 01:09:36 +0100761 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100762
Willy Tarreaue6b98942007-10-29 01:09:36 +0100763 if (fd >= global.maxsock) {
764 err |= ERR_FATAL | ERR_ABORT | ERR_ALERT;
765 msg = "not enough free sockets (raise '-n' parameter)";
766 goto tcp_close_return;
767 }
768
Willy Tarreaufb14edc2009-06-14 15:24:37 +0200769 if (fcntl(fd, F_SETFL, O_NONBLOCK) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +0100770 err |= ERR_FATAL | ERR_ALERT;
771 msg = "cannot make socket non-blocking";
772 goto tcp_close_return;
773 }
774
Willy Tarreau40aa0702013-03-10 23:51:38 +0100775 if (!ext && setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &one, sizeof(one)) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +0100776 /* not fatal but should be reported */
777 msg = "cannot do so_reuseaddr";
778 err |= ERR_ALERT;
779 }
780
781 if (listener->options & LI_O_NOLINGER)
Simon Hormande072bd2011-06-24 15:11:37 +0900782 setsockopt(fd, SOL_SOCKET, SO_LINGER, &nolinger, sizeof(struct linger));
Willy Tarreauedcf6682008-11-30 23:15:34 +0100783
Willy Tarreaue6b98942007-10-29 01:09:36 +0100784#ifdef SO_REUSEPORT
Lukas Tribusa0bcbdc2016-09-12 21:42:20 +0000785 /* OpenBSD and Linux 3.9 support this. As it's present in old libc versions of
786 * Linux, it might return an error that we will silently ignore.
Willy Tarreaue6b98942007-10-29 01:09:36 +0100787 */
Lukas Tribusa0bcbdc2016-09-12 21:42:20 +0000788 if (!ext && (global.tune.options & GTUNE_USE_REUSEPORT))
Willy Tarreau40aa0702013-03-10 23:51:38 +0100789 setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &one, sizeof(one));
Willy Tarreaue6b98942007-10-29 01:09:36 +0100790#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +0200791
Willy Tarreau40aa0702013-03-10 23:51:38 +0100792 if (!ext && (listener->options & LI_O_FOREIGN)) {
David du Colombier65c17962012-07-13 14:34:59 +0200793 switch (listener->addr.ss_family) {
794 case AF_INET:
Pieter Baauwd551fb52013-05-08 22:49:23 +0200795 if (1
796#if defined(IP_TRANSPARENT)
797 && (setsockopt(fd, SOL_IP, IP_TRANSPARENT, &one, sizeof(one)) == -1)
798#endif
799#if defined(IP_FREEBIND)
800 && (setsockopt(fd, SOL_IP, IP_FREEBIND, &one, sizeof(one)) == -1)
801#endif
Pieter Baauwff30b662013-05-08 23:22:39 +0200802#if defined(IP_BINDANY)
803 && (setsockopt(fd, IPPROTO_IP, IP_BINDANY, &one, sizeof(one)) == -1)
804#endif
Pieter Baauw1eb75922013-05-08 23:30:23 +0200805#if defined(SO_BINDANY)
806 && (setsockopt(fd, SOL_SOCKET, SO_BINDANY, &one, sizeof(one)) == -1)
807#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +0200808 ) {
David du Colombier65c17962012-07-13 14:34:59 +0200809 msg = "cannot make listening socket transparent";
810 err |= ERR_ALERT;
811 }
812 break;
813 case AF_INET6:
Pieter Baauwd551fb52013-05-08 22:49:23 +0200814 if (1
Dinko Korunic7276f3a2016-09-09 09:41:15 +0200815#if defined(IPV6_TRANSPARENT) && defined(SOL_IPV6)
Pieter Baauwd551fb52013-05-08 22:49:23 +0200816 && (setsockopt(fd, SOL_IPV6, IPV6_TRANSPARENT, &one, sizeof(one)) == -1)
817#endif
Lukas Tribus7640e722014-03-03 21:10:51 +0100818#if defined(IP_FREEBIND)
819 && (setsockopt(fd, SOL_IP, IP_FREEBIND, &one, sizeof(one)) == -1)
820#endif
Pieter Baauwff30b662013-05-08 23:22:39 +0200821#if defined(IPV6_BINDANY)
822 && (setsockopt(fd, IPPROTO_IPV6, IPV6_BINDANY, &one, sizeof(one)) == -1)
823#endif
Pieter Baauw1eb75922013-05-08 23:30:23 +0200824#if defined(SO_BINDANY)
825 && (setsockopt(fd, SOL_SOCKET, SO_BINDANY, &one, sizeof(one)) == -1)
826#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +0200827 ) {
David du Colombier65c17962012-07-13 14:34:59 +0200828 msg = "cannot make listening socket transparent";
829 err |= ERR_ALERT;
830 }
831 break;
832 }
Willy Tarreaub1e52e82008-01-13 14:49:51 +0100833 }
Pieter Baauwd551fb52013-05-08 22:49:23 +0200834
Willy Tarreau5e6e2042009-02-04 17:19:29 +0100835#ifdef SO_BINDTODEVICE
836 /* Note: this might fail if not CAP_NET_RAW */
Willy Tarreau40aa0702013-03-10 23:51:38 +0100837 if (!ext && listener->interface) {
Willy Tarreau5e6e2042009-02-04 17:19:29 +0100838 if (setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE,
Willy Tarreau604e8302009-03-06 00:48:23 +0100839 listener->interface, strlen(listener->interface) + 1) == -1) {
Willy Tarreau5e6e2042009-02-04 17:19:29 +0100840 msg = "cannot bind listener to device";
841 err |= ERR_WARN;
842 }
843 }
844#endif
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +0400845#if defined(TCP_MAXSEG)
Willy Tarreau48a7e722010-12-24 15:26:39 +0100846 if (listener->maxseg > 0) {
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +0400847 if (setsockopt(fd, IPPROTO_TCP, TCP_MAXSEG,
Willy Tarreaube1b9182009-06-14 18:48:19 +0200848 &listener->maxseg, sizeof(listener->maxseg)) == -1) {
849 msg = "cannot set MSS";
850 err |= ERR_WARN;
851 }
852 }
853#endif
Willy Tarreau2af207a2015-02-04 00:45:58 +0100854#if defined(TCP_USER_TIMEOUT)
855 if (listener->tcp_ut) {
856 if (setsockopt(fd, IPPROTO_TCP, TCP_USER_TIMEOUT,
857 &listener->tcp_ut, sizeof(listener->tcp_ut)) == -1) {
858 msg = "cannot set TCP User Timeout";
859 err |= ERR_WARN;
860 }
861 }
862#endif
Willy Tarreaucb6cd432009-10-13 07:34:14 +0200863#if defined(TCP_DEFER_ACCEPT)
864 if (listener->options & LI_O_DEF_ACCEPT) {
865 /* defer accept by up to one second */
866 int accept_delay = 1;
867 if (setsockopt(fd, IPPROTO_TCP, TCP_DEFER_ACCEPT, &accept_delay, sizeof(accept_delay)) == -1) {
868 msg = "cannot enable DEFER_ACCEPT";
869 err |= ERR_WARN;
870 }
871 }
872#endif
Willy Tarreau1c862c52012-10-05 16:21:00 +0200873#if defined(TCP_FASTOPEN)
874 if (listener->options & LI_O_TCP_FO) {
875 /* TFO needs a queue length, let's use the configured backlog */
876 int qlen = listener->backlog ? listener->backlog : listener->maxconn;
877 if (setsockopt(fd, IPPROTO_TCP, TCP_FASTOPEN, &qlen, sizeof(qlen)) == -1) {
878 msg = "cannot enable TCP_FASTOPEN";
879 err |= ERR_WARN;
880 }
881 }
882#endif
Willy Tarreau9b6700f2012-11-24 11:55:28 +0100883#if defined(IPV6_V6ONLY)
884 if (listener->options & LI_O_V6ONLY)
885 setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &one, sizeof(one));
Willy Tarreau77e3af92012-11-24 15:07:23 +0100886 else if (listener->options & LI_O_V4V6)
887 setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &zero, sizeof(zero));
Willy Tarreau9b6700f2012-11-24 11:55:28 +0100888#endif
889
Willy Tarreau40aa0702013-03-10 23:51:38 +0100890 if (!ext && bind(fd, (struct sockaddr *)&listener->addr, listener->proto->sock_addrlen) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +0100891 err |= ERR_RETRYABLE | ERR_ALERT;
892 msg = "cannot bind socket";
893 goto tcp_close_return;
894 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100895
Willy Tarreau40aa0702013-03-10 23:51:38 +0100896 ready = 0;
897 ready_len = sizeof(ready);
898 if (getsockopt(fd, SOL_SOCKET, SO_ACCEPTCONN, &ready, &ready_len) == -1)
899 ready = 0;
900
901 if (!(ext && ready) && /* only listen if not already done by external process */
902 listen(fd, listener->backlog ? listener->backlog : listener->maxconn) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +0100903 err |= ERR_RETRYABLE | ERR_ALERT;
904 msg = "cannot listen to socket";
905 goto tcp_close_return;
906 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100907
Dmitry Sivachenkocaf58982009-08-24 15:11:06 +0400908#if defined(TCP_QUICKACK)
Willy Tarreau9ea05a72009-06-14 12:07:01 +0200909 if (listener->options & LI_O_NOQUICKACK)
Simon Hormande072bd2011-06-24 15:11:37 +0900910 setsockopt(fd, IPPROTO_TCP, TCP_QUICKACK, &zero, sizeof(zero));
Willy Tarreau9ea05a72009-06-14 12:07:01 +0200911#endif
912
Willy Tarreaue6b98942007-10-29 01:09:36 +0100913 /* the socket is ready */
914 listener->fd = fd;
915 listener->state = LI_LISTEN;
916
Willy Tarreaueabf3132008-08-29 23:36:51 +0200917 fdtab[fd].owner = listener; /* reference the listener instead of a task */
Willy Tarreauaece46a2012-07-06 12:25:58 +0200918 fdtab[fd].iocb = listener->proto->accept;
Willy Tarreaueb472682010-05-28 18:46:57 +0200919 fd_insert(fd);
920
Willy Tarreaue6b98942007-10-29 01:09:36 +0100921 tcp_return:
Cyril Bonté43ba1b32010-11-01 19:26:01 +0100922 if (msg && errlen) {
923 char pn[INET6_ADDRSTRLEN];
924
Willy Tarreau631f01c2011-09-05 00:36:48 +0200925 addr_to_str(&listener->addr, pn, sizeof(pn));
926 snprintf(errmsg, errlen, "%s [%s:%d]", msg, pn, get_host_port(&listener->addr));
Cyril Bonté43ba1b32010-11-01 19:26:01 +0100927 }
Willy Tarreaue6b98942007-10-29 01:09:36 +0100928 return err;
929
930 tcp_close_return:
931 close(fd);
932 goto tcp_return;
933}
934
935/* This function creates all TCP sockets bound to the protocol entry <proto>.
936 * It is intended to be used as the protocol's bind_all() function.
937 * The sockets will be registered but not added to any fd_set, in order not to
938 * loose them across the fork(). A call to enable_all_listeners() is needed
939 * to complete initialization. The return value is composed from ERR_*.
940 */
Emeric Bruncf20bf12010-10-22 16:06:11 +0200941static int tcp_bind_listeners(struct protocol *proto, char *errmsg, int errlen)
Willy Tarreaue6b98942007-10-29 01:09:36 +0100942{
943 struct listener *listener;
944 int err = ERR_NONE;
945
946 list_for_each_entry(listener, &proto->listeners, proto_list) {
Emeric Bruncf20bf12010-10-22 16:06:11 +0200947 err |= tcp_bind_listener(listener, errmsg, errlen);
948 if (err & ERR_ABORT)
Willy Tarreaue6b98942007-10-29 01:09:36 +0100949 break;
950 }
951
952 return err;
953}
954
955/* Add listener to the list of tcpv4 listeners. The listener's state
956 * is automatically updated from LI_INIT to LI_ASSIGNED. The number of
957 * listeners is updated. This is the function to use to add a new listener.
958 */
959void tcpv4_add_listener(struct listener *listener)
960{
961 if (listener->state != LI_INIT)
962 return;
963 listener->state = LI_ASSIGNED;
964 listener->proto = &proto_tcpv4;
965 LIST_ADDQ(&proto_tcpv4.listeners, &listener->proto_list);
966 proto_tcpv4.nb_listeners++;
967}
968
969/* Add listener to the list of tcpv4 listeners. The listener's state
970 * is automatically updated from LI_INIT to LI_ASSIGNED. The number of
971 * listeners is updated. This is the function to use to add a new listener.
972 */
973void tcpv6_add_listener(struct listener *listener)
974{
975 if (listener->state != LI_INIT)
976 return;
977 listener->state = LI_ASSIGNED;
978 listener->proto = &proto_tcpv6;
979 LIST_ADDQ(&proto_tcpv6.listeners, &listener->proto_list);
980 proto_tcpv6.nb_listeners++;
981}
982
Willy Tarreau092d8652014-07-07 20:22:12 +0200983/* Pause a listener. Returns < 0 in case of failure, 0 if the listener
984 * was totally stopped, or > 0 if correctly paused.
985 */
986int tcp_pause_listener(struct listener *l)
987{
988 if (shutdown(l->fd, SHUT_WR) != 0)
989 return -1; /* Solaris dies here */
990
991 if (listen(l->fd, l->backlog ? l->backlog : l->maxconn) != 0)
992 return -1; /* OpenBSD dies here */
993
994 if (shutdown(l->fd, SHUT_RD) != 0)
995 return -1; /* should always be OK */
996 return 1;
997}
998
William Lallemand2e785f22016-05-25 01:48:42 +0200999/*
Willy Tarreau00005ce2016-10-21 15:07:45 +02001000 * Execute the "set-src" action. May be called from {tcp,http}request.
1001 * It only changes the address and tries to preserve the original port. If the
1002 * previous family was neither AF_INET nor AF_INET6, the port is set to zero.
William Lallemand2e785f22016-05-25 01:48:42 +02001003 */
1004enum act_return tcp_action_req_set_src(struct act_rule *rule, struct proxy *px,
1005 struct session *sess, struct stream *s, int flags)
1006{
1007 struct connection *cli_conn;
1008
1009 if ((cli_conn = objt_conn(sess->origin)) && conn_ctrl_ready(cli_conn)) {
1010 struct sample *smp;
1011
1012 smp = sample_fetch_as_type(px, sess, s, SMP_OPT_DIR_REQ|SMP_OPT_FINAL, rule->arg.expr, SMP_T_ADDR);
1013 if (smp) {
Willy Tarreau00005ce2016-10-21 15:07:45 +02001014 int port = get_net_port(&cli_conn->addr.from);
1015
William Lallemand2e785f22016-05-25 01:48:42 +02001016 if (smp->data.type == SMP_T_IPV4) {
1017 ((struct sockaddr_in *)&cli_conn->addr.from)->sin_family = AF_INET;
1018 ((struct sockaddr_in *)&cli_conn->addr.from)->sin_addr.s_addr = smp->data.u.ipv4.s_addr;
Willy Tarreau00005ce2016-10-21 15:07:45 +02001019 ((struct sockaddr_in *)&cli_conn->addr.from)->sin_port = port;
William Lallemand2e785f22016-05-25 01:48:42 +02001020 } else if (smp->data.type == SMP_T_IPV6) {
1021 ((struct sockaddr_in6 *)&cli_conn->addr.from)->sin6_family = AF_INET6;
1022 memcpy(&((struct sockaddr_in6 *)&cli_conn->addr.from)->sin6_addr, &smp->data.u.ipv6, sizeof(struct in6_addr));
Willy Tarreau00005ce2016-10-21 15:07:45 +02001023 ((struct sockaddr_in6 *)&cli_conn->addr.from)->sin6_port = port;
William Lallemand2e785f22016-05-25 01:48:42 +02001024 }
1025 }
William Lallemand01252ed2016-05-25 02:33:16 +02001026 cli_conn->flags |= CO_FL_ADDR_FROM_SET;
William Lallemand2e785f22016-05-25 01:48:42 +02001027 }
1028 return ACT_RET_CONT;
1029}
1030
William Lallemand44be6402016-05-25 01:51:35 +02001031/*
Willy Tarreau00005ce2016-10-21 15:07:45 +02001032 * Execute the "set-dst" action. May be called from {tcp,http}request.
1033 * It only changes the address and tries to preserve the original port. If the
1034 * previous family was neither AF_INET nor AF_INET6, the port is set to zero.
William Lallemand13e9b0c2016-05-25 02:34:07 +02001035 */
1036enum act_return tcp_action_req_set_dst(struct act_rule *rule, struct proxy *px,
1037 struct session *sess, struct stream *s, int flags)
1038{
1039 struct connection *cli_conn;
1040
1041 if ((cli_conn = objt_conn(sess->origin)) && conn_ctrl_ready(cli_conn)) {
1042 struct sample *smp;
1043
1044 smp = sample_fetch_as_type(px, sess, s, SMP_OPT_DIR_REQ|SMP_OPT_FINAL, rule->arg.expr, SMP_T_ADDR);
1045 if (smp) {
Willy Tarreau00005ce2016-10-21 15:07:45 +02001046 int port = get_net_port(&cli_conn->addr.to);
1047
William Lallemand13e9b0c2016-05-25 02:34:07 +02001048 if (smp->data.type == SMP_T_IPV4) {
1049 ((struct sockaddr_in *)&cli_conn->addr.to)->sin_family = AF_INET;
1050 ((struct sockaddr_in *)&cli_conn->addr.to)->sin_addr.s_addr = smp->data.u.ipv4.s_addr;
1051 } else if (smp->data.type == SMP_T_IPV6) {
1052 ((struct sockaddr_in6 *)&cli_conn->addr.to)->sin6_family = AF_INET6;
1053 memcpy(&((struct sockaddr_in6 *)&cli_conn->addr.to)->sin6_addr, &smp->data.u.ipv6, sizeof(struct in6_addr));
Willy Tarreau00005ce2016-10-21 15:07:45 +02001054 ((struct sockaddr_in6 *)&cli_conn->addr.to)->sin6_port = port;
William Lallemand13e9b0c2016-05-25 02:34:07 +02001055 }
1056 cli_conn->flags |= CO_FL_ADDR_TO_SET;
1057 }
1058 }
1059 return ACT_RET_CONT;
1060}
1061
1062/*
Willy Tarreau00005ce2016-10-21 15:07:45 +02001063 * Execute the "set-src-port" action. May be called from {tcp,http}request.
1064 * We must test the sin_family before setting the port. If the address family
1065 * is neither AF_INET nor AF_INET6, the address is forced to AF_INET "0.0.0.0"
1066 * and the port is assigned.
William Lallemand44be6402016-05-25 01:51:35 +02001067 */
1068enum act_return tcp_action_req_set_src_port(struct act_rule *rule, struct proxy *px,
1069 struct session *sess, struct stream *s, int flags)
1070{
1071 struct connection *cli_conn;
1072
1073 if ((cli_conn = objt_conn(sess->origin)) && conn_ctrl_ready(cli_conn)) {
1074 struct sample *smp;
1075
1076 conn_get_from_addr(cli_conn);
1077
1078 smp = sample_fetch_as_type(px, sess, s, SMP_OPT_DIR_REQ|SMP_OPT_FINAL, rule->arg.expr, SMP_T_SINT);
1079 if (smp) {
Willy Tarreau00005ce2016-10-21 15:07:45 +02001080 if (cli_conn->addr.from.ss_family == AF_INET6) {
William Lallemand44be6402016-05-25 01:51:35 +02001081 ((struct sockaddr_in6 *)&cli_conn->addr.from)->sin6_port = htons(smp->data.u.sint);
Willy Tarreau00005ce2016-10-21 15:07:45 +02001082 } else {
1083 if (cli_conn->addr.from.ss_family != AF_INET) {
1084 cli_conn->addr.from.ss_family = AF_INET;
1085 ((struct sockaddr_in *)&cli_conn->addr.from)->sin_addr.s_addr = 0;
1086 }
1087 ((struct sockaddr_in *)&cli_conn->addr.from)->sin_port = htons(smp->data.u.sint);
William Lallemand44be6402016-05-25 01:51:35 +02001088 }
1089 }
1090 }
1091 return ACT_RET_CONT;
1092}
1093
William Lallemand13e9b0c2016-05-25 02:34:07 +02001094/*
Willy Tarreau00005ce2016-10-21 15:07:45 +02001095 * Execute the "set-dst-port" action. May be called from {tcp,http}request.
1096 * We must test the sin_family before setting the port. If the address family
1097 * is neither AF_INET nor AF_INET6, the address is forced to AF_INET "0.0.0.0"
1098 * and the port is assigned.
William Lallemand13e9b0c2016-05-25 02:34:07 +02001099 */
1100enum act_return tcp_action_req_set_dst_port(struct act_rule *rule, struct proxy *px,
1101 struct session *sess, struct stream *s, int flags)
1102{
1103 struct connection *cli_conn;
1104
1105 if ((cli_conn = objt_conn(sess->origin)) && conn_ctrl_ready(cli_conn)) {
1106 struct sample *smp;
1107
1108 conn_get_to_addr(cli_conn);
1109
1110 smp = sample_fetch_as_type(px, sess, s, SMP_OPT_DIR_REQ|SMP_OPT_FINAL, rule->arg.expr, SMP_T_SINT);
1111 if (smp) {
Willy Tarreau00005ce2016-10-21 15:07:45 +02001112 if (cli_conn->addr.from.ss_family == AF_INET6) {
1113 ((struct sockaddr_in6 *)&cli_conn->addr.from)->sin6_port = htons(smp->data.u.sint);
1114 } else {
1115 if (cli_conn->addr.from.ss_family != AF_INET) {
1116 cli_conn->addr.from.ss_family = AF_INET;
1117 ((struct sockaddr_in *)&cli_conn->addr.from)->sin_addr.s_addr = 0;
1118 }
1119 ((struct sockaddr_in *)&cli_conn->addr.from)->sin_port = htons(smp->data.u.sint);
William Lallemand13e9b0c2016-05-25 02:34:07 +02001120 }
1121 }
1122 }
1123 return ACT_RET_CONT;
1124}
1125
Willy Tarreau2d392c22015-08-24 01:43:45 +02001126/* Executes the "silent-drop" action. May be called from {tcp,http}{request,response} */
1127static enum act_return tcp_exec_action_silent_drop(struct act_rule *rule, struct proxy *px, struct session *sess, struct stream *strm, int flags)
1128{
1129 struct connection *conn = objt_conn(sess->origin);
1130
1131 if (!conn)
1132 goto out;
1133
1134 if (!conn_ctrl_ready(conn))
1135 goto out;
1136
Willy Tarreau2d392c22015-08-24 01:43:45 +02001137#ifdef TCP_QUICKACK
Willy Tarreaufc2a2d92015-09-29 18:15:01 +02001138 /* drain is needed only to send the quick ACK */
1139 conn_sock_drain(conn);
1140
Willy Tarreau2d392c22015-08-24 01:43:45 +02001141 /* re-enable quickack if it was disabled to ack all data and avoid
1142 * retransmits from the client that might trigger a real reset.
1143 */
1144 setsockopt(conn->t.sock.fd, SOL_TCP, TCP_QUICKACK, &one, sizeof(one));
1145#endif
1146 /* lingering must absolutely be disabled so that we don't send a
1147 * shutdown(), this is critical to the TCP_REPAIR trick. When no stream
1148 * is present, returning with ERR will cause lingering to be disabled.
1149 */
1150 if (strm)
1151 strm->si[0].flags |= SI_FL_NOLINGER;
1152
Willy Tarreauf50ec0f2015-09-29 18:11:32 +02001153 /* We're on the client-facing side, we must force to disable lingering to
1154 * ensure we will use an RST exclusively and kill any pending data.
1155 */
1156 fdtab[conn->t.sock.fd].linger_risk = 1;
1157
Willy Tarreau2d392c22015-08-24 01:43:45 +02001158#ifdef TCP_REPAIR
1159 if (setsockopt(conn->t.sock.fd, SOL_TCP, TCP_REPAIR, &one, sizeof(one)) == 0) {
1160 /* socket will be quiet now */
1161 goto out;
1162 }
1163#endif
1164 /* either TCP_REPAIR is not defined or it failed (eg: permissions).
1165 * Let's fall back on the TTL trick, though it only works for routed
1166 * network and has no effect on local net.
1167 */
1168#ifdef IP_TTL
1169 setsockopt(conn->t.sock.fd, SOL_IP, IP_TTL, &one, sizeof(one));
1170#endif
1171 out:
1172 /* kill the stream if any */
1173 if (strm) {
1174 channel_abort(&strm->req);
1175 channel_abort(&strm->res);
1176 strm->req.analysers = 0;
1177 strm->res.analysers = 0;
1178 strm->be->be_counters.denied_req++;
1179 if (!(strm->flags & SF_ERR_MASK))
1180 strm->flags |= SF_ERR_PRXCOND;
1181 if (!(strm->flags & SF_FINST_MASK))
1182 strm->flags |= SF_FINST_R;
1183 }
1184
1185 sess->fe->fe_counters.denied_req++;
1186 if (sess->listener->counters)
1187 sess->listener->counters->denied_req++;
1188
1189 return ACT_RET_STOP;
1190}
1191
William Lallemand13e9b0c2016-05-25 02:34:07 +02001192/* parse "set-{src,dst}[-port]" action */
1193enum 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 +02001194{
1195 int cur_arg;
1196 struct sample_expr *expr;
1197 unsigned int where;
1198
1199 cur_arg = *orig_arg;
1200 expr = sample_parse_expr((char **)args, &cur_arg, px->conf.args.file, px->conf.args.line, err, &px->conf.args);
1201 if (!expr)
1202 return ACT_RET_PRS_ERR;
1203
1204 where = 0;
1205 if (proxy->cap & PR_CAP_FE)
1206 where |= SMP_VAL_FE_HRQ_HDR;
1207 if (proxy->cap & PR_CAP_BE)
1208 where |= SMP_VAL_BE_HRQ_HDR;
1209
1210 if (!(expr->fetch->val & where)) {
1211 memprintf(err,
1212 "fetch method '%s' extracts information from '%s', none of which is available here",
1213 args[cur_arg-1], sample_src_names(expr->fetch->use));
1214 free(expr);
1215 return ACT_RET_PRS_ERR;
1216 }
1217 rule->arg.expr = expr;
1218 rule->action = ACT_CUSTOM;
1219
1220 if (!strcmp(args[*orig_arg-1], "set-src")) {
1221 rule->action_ptr = tcp_action_req_set_src;
William Lallemand44be6402016-05-25 01:51:35 +02001222 } else if (!strcmp(args[*orig_arg-1], "set-src-port")) {
1223 rule->action_ptr = tcp_action_req_set_src_port;
William Lallemand13e9b0c2016-05-25 02:34:07 +02001224 } else if (!strcmp(args[*orig_arg-1], "set-dst")) {
1225 rule->action_ptr = tcp_action_req_set_dst;
1226 } else if (!strcmp(args[*orig_arg-1], "set-dst-port")) {
1227 rule->action_ptr = tcp_action_req_set_dst_port;
William Lallemand2e785f22016-05-25 01:48:42 +02001228 } else {
1229 return ACT_RET_PRS_ERR;
1230 }
1231
1232 (*orig_arg)++;
1233
1234 return ACT_RET_PRS_OK;
1235}
1236
1237
Willy Tarreau2d392c22015-08-24 01:43:45 +02001238/* Parse a "silent-drop" action. It takes no argument. It returns ACT_RET_PRS_OK on
1239 * success, ACT_RET_PRS_ERR on error.
1240 */
1241static enum act_parse_ret tcp_parse_silent_drop(const char **args, int *orig_arg, struct proxy *px,
1242 struct act_rule *rule, char **err)
1243{
1244 rule->action = ACT_CUSTOM;
1245 rule->action_ptr = tcp_exec_action_silent_drop;
1246 return ACT_RET_PRS_OK;
1247}
1248
Willy Tarreau645513a2010-05-24 20:55:15 +02001249
1250/************************************************************************/
Willy Tarreaud4c33c82013-01-07 21:59:07 +01001251/* All supported sample fetch functions must be declared here */
Willy Tarreau32389b72012-04-23 23:13:20 +02001252/************************************************************************/
1253
Willy Tarreau4a129812012-04-25 17:31:42 +02001254/* fetch the connection's source IPv4/IPv6 address */
Thierry FOURNIERa123ad82015-07-24 09:12:15 +02001255int smp_fetch_src(const struct arg *args, struct sample *smp, const char *kw, void *private)
Willy Tarreau645513a2010-05-24 20:55:15 +02001256{
Thierry FOURNIER0a9a2b82015-05-11 15:20:49 +02001257 struct connection *cli_conn = objt_conn(smp->sess->origin);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001258
1259 if (!cli_conn)
1260 return 0;
1261
1262 switch (cli_conn->addr.from.ss_family) {
Willy Tarreauf4362b32011-12-16 17:49:52 +01001263 case AF_INET:
Thierry FOURNIER136f9d32015-08-19 09:07:19 +02001264 smp->data.u.ipv4 = ((struct sockaddr_in *)&cli_conn->addr.from)->sin_addr;
Thierry FOURNIER8c542ca2015-08-19 09:00:18 +02001265 smp->data.type = SMP_T_IPV4;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001266 break;
1267 case AF_INET6:
Thierry FOURNIER136f9d32015-08-19 09:07:19 +02001268 smp->data.u.ipv6 = ((struct sockaddr_in6 *)&cli_conn->addr.from)->sin6_addr;
Thierry FOURNIER8c542ca2015-08-19 09:00:18 +02001269 smp->data.type = SMP_T_IPV6;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001270 break;
1271 default:
Emeric Brunf769f512010-10-22 17:14:01 +02001272 return 0;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001273 }
Emeric Brunf769f512010-10-22 17:14:01 +02001274
Willy Tarreau37406352012-04-23 16:16:37 +02001275 smp->flags = 0;
Willy Tarreau645513a2010-05-24 20:55:15 +02001276 return 1;
1277}
1278
Willy Tarreaua5e37562011-12-16 17:06:15 +01001279/* set temp integer to the connection's source port */
Willy Tarreau645513a2010-05-24 20:55:15 +02001280static int
Thierry FOURNIER0786d052015-05-11 15:42:45 +02001281smp_fetch_sport(const struct arg *args, struct sample *smp, const char *k, void *private)
Willy Tarreau645513a2010-05-24 20:55:15 +02001282{
Thierry FOURNIER0a9a2b82015-05-11 15:20:49 +02001283 struct connection *cli_conn = objt_conn(smp->sess->origin);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001284
1285 if (!cli_conn)
1286 return 0;
1287
Thierry FOURNIER8c542ca2015-08-19 09:00:18 +02001288 smp->data.type = SMP_T_SINT;
Thierry FOURNIER136f9d32015-08-19 09:07:19 +02001289 if (!(smp->data.u.sint = get_host_port(&cli_conn->addr.from)))
Emeric Brunf769f512010-10-22 17:14:01 +02001290 return 0;
1291
Willy Tarreau37406352012-04-23 16:16:37 +02001292 smp->flags = 0;
Willy Tarreau645513a2010-05-24 20:55:15 +02001293 return 1;
1294}
1295
Willy Tarreau4a129812012-04-25 17:31:42 +02001296/* fetch the connection's destination IPv4/IPv6 address */
Willy Tarreau645513a2010-05-24 20:55:15 +02001297static int
Thierry FOURNIER0786d052015-05-11 15:42:45 +02001298smp_fetch_dst(const struct arg *args, struct sample *smp, const char *kw, void *private)
Willy Tarreau645513a2010-05-24 20:55:15 +02001299{
Thierry FOURNIER0a9a2b82015-05-11 15:20:49 +02001300 struct connection *cli_conn = objt_conn(smp->sess->origin);
Willy Tarreau645513a2010-05-24 20:55:15 +02001301
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001302 if (!cli_conn)
1303 return 0;
1304
1305 conn_get_to_addr(cli_conn);
1306
1307 switch (cli_conn->addr.to.ss_family) {
Willy Tarreauf4362b32011-12-16 17:49:52 +01001308 case AF_INET:
Thierry FOURNIER136f9d32015-08-19 09:07:19 +02001309 smp->data.u.ipv4 = ((struct sockaddr_in *)&cli_conn->addr.to)->sin_addr;
Thierry FOURNIER8c542ca2015-08-19 09:00:18 +02001310 smp->data.type = SMP_T_IPV4;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001311 break;
1312 case AF_INET6:
Thierry FOURNIER136f9d32015-08-19 09:07:19 +02001313 smp->data.u.ipv6 = ((struct sockaddr_in6 *)&cli_conn->addr.to)->sin6_addr;
Thierry FOURNIER8c542ca2015-08-19 09:00:18 +02001314 smp->data.type = SMP_T_IPV6;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001315 break;
1316 default:
Emeric Brunf769f512010-10-22 17:14:01 +02001317 return 0;
Willy Tarreauf4362b32011-12-16 17:49:52 +01001318 }
Emeric Brunf769f512010-10-22 17:14:01 +02001319
Willy Tarreau37406352012-04-23 16:16:37 +02001320 smp->flags = 0;
Willy Tarreau645513a2010-05-24 20:55:15 +02001321 return 1;
1322}
1323
Willy Tarreau16e01562016-08-09 16:46:18 +02001324/* check if the destination address of the front connection is local to the
1325 * system or if it was intercepted.
1326 */
1327int smp_fetch_dst_is_local(const struct arg *args, struct sample *smp, const char *kw, void *private)
1328{
1329 struct connection *conn = objt_conn(smp->sess->origin);
1330 struct listener *li = smp->sess->listener;
1331
1332 if (!conn)
1333 return 0;
1334
1335 conn_get_to_addr(conn);
1336 if (!(conn->flags & CO_FL_ADDR_TO_SET))
1337 return 0;
1338
1339 smp->data.type = SMP_T_BOOL;
1340 smp->flags = 0;
1341 smp->data.u.sint = addr_is_local(li->netns, &conn->addr.to);
1342 return smp->data.u.sint >= 0;
1343}
1344
1345/* check if the source address of the front connection is local to the system
1346 * or not.
1347 */
1348int smp_fetch_src_is_local(const struct arg *args, struct sample *smp, const char *kw, void *private)
1349{
1350 struct connection *conn = objt_conn(smp->sess->origin);
1351 struct listener *li = smp->sess->listener;
1352
1353 if (!conn)
1354 return 0;
1355
1356 conn_get_from_addr(conn);
1357 if (!(conn->flags & CO_FL_ADDR_FROM_SET))
1358 return 0;
1359
1360 smp->data.type = SMP_T_BOOL;
1361 smp->flags = 0;
1362 smp->data.u.sint = addr_is_local(li->netns, &conn->addr.from);
1363 return smp->data.u.sint >= 0;
1364}
1365
Willy Tarreaua5e37562011-12-16 17:06:15 +01001366/* set temp integer to the frontend connexion's destination port */
Willy Tarreau645513a2010-05-24 20:55:15 +02001367static int
Thierry FOURNIER0786d052015-05-11 15:42:45 +02001368smp_fetch_dport(const struct arg *args, struct sample *smp, const char *kw, void *private)
Willy Tarreau645513a2010-05-24 20:55:15 +02001369{
Thierry FOURNIER0a9a2b82015-05-11 15:20:49 +02001370 struct connection *cli_conn = objt_conn(smp->sess->origin);
Willy Tarreaub363a1f2013-10-01 10:45:07 +02001371
1372 if (!cli_conn)
1373 return 0;
1374
1375 conn_get_to_addr(cli_conn);
Willy Tarreau645513a2010-05-24 20:55:15 +02001376
Thierry FOURNIER8c542ca2015-08-19 09:00:18 +02001377 smp->data.type = SMP_T_SINT;
Thierry FOURNIER136f9d32015-08-19 09:07:19 +02001378 if (!(smp->data.u.sint = get_host_port(&cli_conn->addr.to)))
Emeric Brunf769f512010-10-22 17:14:01 +02001379 return 0;
1380
Willy Tarreau37406352012-04-23 16:16:37 +02001381 smp->flags = 0;
Willy Tarreau645513a2010-05-24 20:55:15 +02001382 return 1;
Emericf2d7cae2010-11-05 18:13:50 +01001383}
1384
Thierry Fournier / OZON.IO6310bef2016-07-24 20:16:50 +02001385#ifdef TCP_INFO
1386
1387/* Returns some tcp_info data is its avalaible. "dir" must be set to 0 if
1388 * the client connection is require, otherwise it is set to 1. "val" represents
1389 * the required value. Use 0 for rtt and 1 for rttavg. "unit" is the expected unit
1390 * by default, the rtt is in us. Id "unit" is set to 0, the unit is us, if it is
1391 * set to 1, the untis are milliseconds.
1392 * If the function fails it returns 0, otherwise it returns 1 and "result" is filled.
1393 */
1394static inline int get_tcp_info(const struct arg *args, struct sample *smp,
1395 int dir, int val)
1396{
1397 struct connection *conn;
1398 struct tcp_info info;
1399 socklen_t optlen;
1400
1401 /* strm can be null. */
1402 if (!smp->strm)
1403 return 0;
1404
1405 /* get the object associated with the stream interface.The
1406 * object can be other thing than a connection. For example,
1407 * it be a appctx. */
1408 conn = objt_conn(smp->strm->si[dir].end);
1409 if (!conn)
1410 return 0;
1411
1412 /* The fd may not be avalaible for the tcp_info struct, and the
1413 syscal can fail. */
1414 optlen = sizeof(info);
1415 if (getsockopt(conn->t.sock.fd, SOL_TCP, TCP_INFO, &info, &optlen) == -1)
1416 return 0;
1417
1418 /* extract the value. */
1419 smp->data.type = SMP_T_SINT;
1420 switch (val) {
Joe Williams30fcd392016-08-10 07:06:44 -07001421 case 0: smp->data.u.sint = info.tcpi_rtt; break;
1422 case 1: smp->data.u.sint = info.tcpi_rttvar; break;
1423#if defined(__linux__)
1424 /* these ones are common to all Linux versions */
1425 case 2: smp->data.u.sint = info.tcpi_unacked; break;
1426 case 3: smp->data.u.sint = info.tcpi_sacked; break;
1427 case 4: smp->data.u.sint = info.tcpi_lost; break;
1428 case 5: smp->data.u.sint = info.tcpi_retrans; break;
1429 case 6: smp->data.u.sint = info.tcpi_fackets; break;
1430 case 7: smp->data.u.sint = info.tcpi_reordering; break;
1431#elif defined(__FreeBSD__) || defined(__NetBSD__)
1432 /* the ones are found on FreeBSD and NetBSD featuring TCP_INFO */
1433 case 2: smp->data.u.sint = info.__tcpi_unacked; break;
1434 case 3: smp->data.u.sint = info.__tcpi_sacked; break;
1435 case 4: smp->data.u.sint = info.__tcpi_lost; break;
1436 case 5: smp->data.u.sint = info.__tcpi_retrans; break;
1437 case 6: smp->data.u.sint = info.__tcpi_fackets; break;
1438 case 7: smp->data.u.sint = info.__tcpi_reordering; break;
1439#endif
Thierry Fournier / OZON.IO6310bef2016-07-24 20:16:50 +02001440 default: return 0;
1441 }
1442
1443 /* Convert the value as expected. */
1444 if (args) {
1445 if (args[0].type == ARGT_STR) {
1446 if (strcmp(args[0].data.str.str, "us") == 0) {
1447 /* Do nothing. */
1448 } else if (strcmp(args[0].data.str.str, "ms") == 0) {
1449 smp->data.u.sint = (smp->data.u.sint + 500) / 1000;
1450 } else
1451 return 0;
1452 } else if (args[0].type == ARGT_STOP) {
1453 smp->data.u.sint = (smp->data.u.sint + 500) / 1000;
1454 } else
1455 return 0;
1456 }
1457
1458 return 1;
1459}
1460
1461/* get the mean rtt of a client connexion */
1462static int
1463smp_fetch_fc_rtt(const struct arg *args, struct sample *smp, const char *kw, void *private)
1464{
1465 if (!get_tcp_info(args, smp, 0, 0))
1466 return 0;
1467 return 1;
1468}
1469
1470/* get the variance of the mean rtt of a client connexion */
1471static int
1472smp_fetch_fc_rttvar(const struct arg *args, struct sample *smp, const char *kw, void *private)
1473{
1474 if (!get_tcp_info(args, smp, 0, 1))
1475 return 0;
1476 return 1;
1477}
Joe Williams30fcd392016-08-10 07:06:44 -07001478
1479#if defined(__linux__) || defined(__FreeBSD__) || defined(__NetBSD__)
1480
1481/* get the unacked counter on a client connexion */
1482static int
1483smp_fetch_fc_unacked(const struct arg *args, struct sample *smp, const char *kw, void *private)
1484{
1485 if (!get_tcp_info(args, smp, 0, 2))
1486 return 0;
1487 return 1;
1488}
1489
1490/* get the sacked counter on a client connexion */
1491static int
1492smp_fetch_fc_sacked(const struct arg *args, struct sample *smp, const char *kw, void *private)
1493{
1494 if (!get_tcp_info(args, smp, 0, 3))
1495 return 0;
1496 return 1;
1497}
1498
1499/* get the lost counter on a client connexion */
1500static int
1501smp_fetch_fc_lost(const struct arg *args, struct sample *smp, const char *kw, void *private)
1502{
1503 if (!get_tcp_info(args, smp, 0, 4))
1504 return 0;
1505 return 1;
1506}
1507
1508/* get the retrans counter on a client connexion */
1509static int
1510smp_fetch_fc_retrans(const struct arg *args, struct sample *smp, const char *kw, void *private)
1511{
1512 if (!get_tcp_info(args, smp, 0, 5))
1513 return 0;
1514 return 1;
1515}
1516
1517/* get the fackets counter on a client connexion */
1518static int
1519smp_fetch_fc_fackets(const struct arg *args, struct sample *smp, const char *kw, void *private)
1520{
1521 if (!get_tcp_info(args, smp, 0, 6))
1522 return 0;
1523 return 1;
1524}
1525
1526/* get the reordering counter on a client connexion */
1527static int
1528smp_fetch_fc_reordering(const struct arg *args, struct sample *smp, const char *kw, void *private)
1529{
1530 if (!get_tcp_info(args, smp, 0, 7))
1531 return 0;
1532 return 1;
1533}
1534#endif // linux || freebsd || netbsd
1535#endif // TCP_INFO
Thierry Fournier / OZON.IO6310bef2016-07-24 20:16:50 +02001536
Willy Tarreau9b6700f2012-11-24 11:55:28 +01001537#ifdef IPV6_V6ONLY
Willy Tarreau77e3af92012-11-24 15:07:23 +01001538/* parse the "v4v6" bind keyword */
1539static int bind_parse_v4v6(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
1540{
1541 struct listener *l;
1542
1543 list_for_each_entry(l, &conf->listeners, by_bind) {
1544 if (l->addr.ss_family == AF_INET6)
1545 l->options |= LI_O_V4V6;
1546 }
1547
1548 return 0;
1549}
1550
Willy Tarreau9b6700f2012-11-24 11:55:28 +01001551/* parse the "v6only" bind keyword */
1552static int bind_parse_v6only(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
1553{
1554 struct listener *l;
1555
1556 list_for_each_entry(l, &conf->listeners, by_bind) {
1557 if (l->addr.ss_family == AF_INET6)
1558 l->options |= LI_O_V6ONLY;
1559 }
1560
1561 return 0;
1562}
1563#endif
1564
Pieter Baauwd551fb52013-05-08 22:49:23 +02001565#ifdef CONFIG_HAP_TRANSPARENT
Willy Tarreau44791242012-09-12 23:27:21 +02001566/* parse the "transparent" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +02001567static 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 +02001568{
1569 struct listener *l;
1570
Willy Tarreau4348fad2012-09-20 16:48:07 +02001571 list_for_each_entry(l, &conf->listeners, by_bind) {
1572 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1573 l->options |= LI_O_FOREIGN;
Willy Tarreau44791242012-09-12 23:27:21 +02001574 }
1575
Willy Tarreau44791242012-09-12 23:27:21 +02001576 return 0;
1577}
1578#endif
1579
1580#ifdef TCP_DEFER_ACCEPT
1581/* parse the "defer-accept" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +02001582static 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 +02001583{
1584 struct listener *l;
1585
Willy Tarreau4348fad2012-09-20 16:48:07 +02001586 list_for_each_entry(l, &conf->listeners, by_bind) {
1587 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1588 l->options |= LI_O_DEF_ACCEPT;
Willy Tarreau44791242012-09-12 23:27:21 +02001589 }
1590
Willy Tarreau44791242012-09-12 23:27:21 +02001591 return 0;
1592}
1593#endif
1594
Willy Tarreau1c862c52012-10-05 16:21:00 +02001595#ifdef TCP_FASTOPEN
Lukas Tribus0defb902013-02-13 23:35:39 +01001596/* parse the "tfo" bind keyword */
Willy Tarreau1c862c52012-10-05 16:21:00 +02001597static int bind_parse_tfo(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
1598{
1599 struct listener *l;
1600
1601 list_for_each_entry(l, &conf->listeners, by_bind) {
1602 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1603 l->options |= LI_O_TCP_FO;
1604 }
1605
1606 return 0;
1607}
1608#endif
1609
Willy Tarreau44791242012-09-12 23:27:21 +02001610#ifdef TCP_MAXSEG
1611/* parse the "mss" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +02001612static 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 +02001613{
1614 struct listener *l;
1615 int mss;
1616
Willy Tarreau44791242012-09-12 23:27:21 +02001617 if (!*args[cur_arg + 1]) {
Willy Tarreaueb6cead2012-09-20 19:43:14 +02001618 memprintf(err, "'%s' : missing MSS value", args[cur_arg]);
Willy Tarreau44791242012-09-12 23:27:21 +02001619 return ERR_ALERT | ERR_FATAL;
1620 }
1621
1622 mss = atoi(args[cur_arg + 1]);
1623 if (!mss || abs(mss) > 65535) {
Willy Tarreaueb6cead2012-09-20 19:43:14 +02001624 memprintf(err, "'%s' : expects an MSS with and absolute value between 1 and 65535", args[cur_arg]);
Willy Tarreau44791242012-09-12 23:27:21 +02001625 return ERR_ALERT | ERR_FATAL;
1626 }
1627
Willy Tarreau4348fad2012-09-20 16:48:07 +02001628 list_for_each_entry(l, &conf->listeners, by_bind) {
1629 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1630 l->maxseg = mss;
1631 }
Willy Tarreau44791242012-09-12 23:27:21 +02001632
1633 return 0;
1634}
1635#endif
1636
Willy Tarreau2af207a2015-02-04 00:45:58 +01001637#ifdef TCP_USER_TIMEOUT
1638/* parse the "tcp-ut" bind keyword */
1639static int bind_parse_tcp_ut(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
1640{
1641 const char *ptr = NULL;
1642 struct listener *l;
1643 unsigned int timeout;
1644
1645 if (!*args[cur_arg + 1]) {
1646 memprintf(err, "'%s' : missing TCP User Timeout value", args[cur_arg]);
1647 return ERR_ALERT | ERR_FATAL;
1648 }
1649
1650 ptr = parse_time_err(args[cur_arg + 1], &timeout, TIME_UNIT_MS);
1651 if (ptr) {
1652 memprintf(err, "'%s' : expects a positive delay in milliseconds", args[cur_arg]);
1653 return ERR_ALERT | ERR_FATAL;
1654 }
1655
1656 list_for_each_entry(l, &conf->listeners, by_bind) {
1657 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1658 l->tcp_ut = timeout;
1659 }
1660
1661 return 0;
1662}
1663#endif
1664
Willy Tarreau44791242012-09-12 23:27:21 +02001665#ifdef SO_BINDTODEVICE
Willy Tarreau2af207a2015-02-04 00:45:58 +01001666/* parse the "interface" bind keyword */
Willy Tarreau4348fad2012-09-20 16:48:07 +02001667static 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 +02001668{
1669 struct listener *l;
1670
Willy Tarreau44791242012-09-12 23:27:21 +02001671 if (!*args[cur_arg + 1]) {
Willy Tarreaueb6cead2012-09-20 19:43:14 +02001672 memprintf(err, "'%s' : missing interface name", args[cur_arg]);
Willy Tarreau44791242012-09-12 23:27:21 +02001673 return ERR_ALERT | ERR_FATAL;
1674 }
1675
Willy Tarreau4348fad2012-09-20 16:48:07 +02001676 list_for_each_entry(l, &conf->listeners, by_bind) {
1677 if (l->addr.ss_family == AF_INET || l->addr.ss_family == AF_INET6)
1678 l->interface = strdup(args[cur_arg + 1]);
1679 }
Willy Tarreau44791242012-09-12 23:27:21 +02001680
1681 global.last_checks |= LSTCHK_NETADM;
1682 return 0;
1683}
1684#endif
1685
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001686#ifdef CONFIG_HAP_NS
1687/* parse the "namespace" bind keyword */
1688static int bind_parse_namespace(char **args, int cur_arg, struct proxy *px, struct bind_conf *conf, char **err)
1689{
1690 struct listener *l;
1691 char *namespace = NULL;
1692
1693 if (!*args[cur_arg + 1]) {
1694 memprintf(err, "'%s' : missing namespace id", args[cur_arg]);
1695 return ERR_ALERT | ERR_FATAL;
1696 }
1697 namespace = args[cur_arg + 1];
1698
1699 list_for_each_entry(l, &conf->listeners, by_bind) {
1700 l->netns = netns_store_lookup(namespace, strlen(namespace));
1701
1702 if (l->netns == NULL)
1703 l->netns = netns_store_insert(namespace);
1704
1705 if (l->netns == NULL) {
1706 Alert("Cannot open namespace '%s'.\n", args[cur_arg + 1]);
1707 return ERR_ALERT | ERR_FATAL;
1708 }
1709 }
1710 return 0;
1711}
1712#endif
1713
Willy Tarreau163d4622015-10-13 16:16:41 +02001714#ifdef TCP_USER_TIMEOUT
1715/* parse the "tcp-ut" server keyword */
1716static int srv_parse_tcp_ut(char **args, int *cur_arg, struct proxy *px, struct server *newsrv, char **err)
1717{
1718 const char *ptr = NULL;
1719 unsigned int timeout;
1720
1721 if (!*args[*cur_arg + 1]) {
1722 memprintf(err, "'%s' : missing TCP User Timeout value", args[*cur_arg]);
1723 return ERR_ALERT | ERR_FATAL;
1724 }
1725
1726 ptr = parse_time_err(args[*cur_arg + 1], &timeout, TIME_UNIT_MS);
1727 if (ptr) {
1728 memprintf(err, "'%s' : expects a positive delay in milliseconds", args[*cur_arg]);
1729 return ERR_ALERT | ERR_FATAL;
1730 }
1731
1732 if (newsrv->addr.ss_family == AF_INET || newsrv->addr.ss_family == AF_INET6)
1733 newsrv->tcp_ut = timeout;
1734
1735 return 0;
1736}
1737#endif
1738
Willy Tarreaud4c33c82013-01-07 21:59:07 +01001739
Willy Tarreau4a129812012-04-25 17:31:42 +02001740/* Note: must not be declared <const> as its list will be overwritten.
1741 * Note: fetches that may return multiple types must be declared as the lowest
1742 * common denominator, the type that can be casted into all other ones. For
1743 * instance v4/v6 must be declared v4.
1744 */
Willy Tarreaudc13c112013-06-21 23:16:39 +02001745static struct sample_fetch_kw_list sample_fetch_keywords = {ILH, {
Willy Tarreaud4c33c82013-01-07 21:59:07 +01001746 { "dst", smp_fetch_dst, 0, NULL, SMP_T_IPV4, SMP_USE_L4CLI },
Willy Tarreau16e01562016-08-09 16:46:18 +02001747 { "dst_is_local", smp_fetch_dst_is_local, 0, NULL, SMP_T_BOOL, SMP_USE_L4CLI },
Thierry FOURNIER07ee64e2015-07-06 23:43:03 +02001748 { "dst_port", smp_fetch_dport, 0, NULL, SMP_T_SINT, SMP_USE_L4CLI },
Willy Tarreaud4c33c82013-01-07 21:59:07 +01001749 { "src", smp_fetch_src, 0, NULL, SMP_T_IPV4, SMP_USE_L4CLI },
Willy Tarreau16e01562016-08-09 16:46:18 +02001750 { "src_is_local", smp_fetch_src_is_local, 0, NULL, SMP_T_BOOL, SMP_USE_L4CLI },
Thierry FOURNIER07ee64e2015-07-06 23:43:03 +02001751 { "src_port", smp_fetch_sport, 0, NULL, SMP_T_SINT, SMP_USE_L4CLI },
Thierry Fournier / OZON.IO6310bef2016-07-24 20:16:50 +02001752#ifdef TCP_INFO
Joe Williams30fcd392016-08-10 07:06:44 -07001753 { "fc_rtt", smp_fetch_fc_rtt, ARG1(0,STR), NULL, SMP_T_SINT, SMP_USE_L4CLI },
1754 { "fc_rttvar", smp_fetch_fc_rttvar, ARG1(0,STR), NULL, SMP_T_SINT, SMP_USE_L4CLI },
1755#if defined(__linux__) || defined(__FreeBSD__) || defined(__NetBSD__)
1756 { "fc_unacked", smp_fetch_fc_unacked, ARG1(0,STR), NULL, SMP_T_SINT, SMP_USE_L4CLI },
1757 { "fc_sacked", smp_fetch_fc_sacked, ARG1(0,STR), NULL, SMP_T_SINT, SMP_USE_L4CLI },
1758 { "fc_retrans", smp_fetch_fc_retrans, ARG1(0,STR), NULL, SMP_T_SINT, SMP_USE_L4CLI },
1759 { "fc_fackets", smp_fetch_fc_fackets, ARG1(0,STR), NULL, SMP_T_SINT, SMP_USE_L4CLI },
1760 { "fc_lost", smp_fetch_fc_lost, ARG1(0,STR), NULL, SMP_T_SINT, SMP_USE_L4CLI },
1761 { "fc_reordering", smp_fetch_fc_reordering, ARG1(0,STR), NULL, SMP_T_SINT, SMP_USE_L4CLI },
1762#endif // linux || freebsd || netbsd
1763#endif // TCP_INFO
Willy Tarreaud4c33c82013-01-07 21:59:07 +01001764 { /* END */ },
Willy Tarreau645513a2010-05-24 20:55:15 +02001765}};
1766
Willy Tarreau44791242012-09-12 23:27:21 +02001767/************************************************************************/
1768/* All supported bind keywords must be declared here. */
1769/************************************************************************/
1770
1771/* Note: must not be declared <const> as its list will be overwritten.
1772 * Please take care of keeping this list alphabetically sorted, doing so helps
1773 * all code contributors.
1774 * Optional keywords are also declared with a NULL ->parse() function so that
1775 * the config parser can report an appropriate error when a known keyword was
1776 * not enabled.
1777 */
Willy Tarreau51fb7652012-09-18 18:24:39 +02001778static struct bind_kw_list bind_kws = { "TCP", { }, {
Willy Tarreau44791242012-09-12 23:27:21 +02001779#ifdef TCP_DEFER_ACCEPT
1780 { "defer-accept", bind_parse_defer_accept, 0 }, /* wait for some data for 1 second max before doing accept */
1781#endif
1782#ifdef SO_BINDTODEVICE
1783 { "interface", bind_parse_interface, 1 }, /* specifically bind to this interface */
1784#endif
1785#ifdef TCP_MAXSEG
1786 { "mss", bind_parse_mss, 1 }, /* set MSS of listening socket */
1787#endif
Willy Tarreau2af207a2015-02-04 00:45:58 +01001788#ifdef TCP_USER_TIMEOUT
1789 { "tcp-ut", bind_parse_tcp_ut, 1 }, /* set User Timeout on listening socket */
1790#endif
Willy Tarreau1c862c52012-10-05 16:21:00 +02001791#ifdef TCP_FASTOPEN
1792 { "tfo", bind_parse_tfo, 0 }, /* enable TCP_FASTOPEN of listening socket */
1793#endif
Pieter Baauwd551fb52013-05-08 22:49:23 +02001794#ifdef CONFIG_HAP_TRANSPARENT
Willy Tarreau44791242012-09-12 23:27:21 +02001795 { "transparent", bind_parse_transparent, 0 }, /* transparently bind to the specified addresses */
1796#endif
Willy Tarreau9b6700f2012-11-24 11:55:28 +01001797#ifdef IPV6_V6ONLY
Willy Tarreau77e3af92012-11-24 15:07:23 +01001798 { "v4v6", bind_parse_v4v6, 0 }, /* force socket to bind to IPv4+IPv6 */
Willy Tarreau9b6700f2012-11-24 11:55:28 +01001799 { "v6only", bind_parse_v6only, 0 }, /* force socket to bind to IPv6 only */
1800#endif
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001801#ifdef CONFIG_HAP_NS
1802 { "namespace", bind_parse_namespace, 1 },
1803#endif
Willy Tarreau44791242012-09-12 23:27:21 +02001804 /* the versions with the NULL parse function*/
1805 { "defer-accept", NULL, 0 },
1806 { "interface", NULL, 1 },
1807 { "mss", NULL, 1 },
1808 { "transparent", NULL, 0 },
Willy Tarreau77e3af92012-11-24 15:07:23 +01001809 { "v4v6", NULL, 0 },
Willy Tarreau9b6700f2012-11-24 11:55:28 +01001810 { "v6only", NULL, 0 },
Willy Tarreau44791242012-09-12 23:27:21 +02001811 { NULL, NULL, 0 },
1812}};
1813
Willy Tarreau163d4622015-10-13 16:16:41 +02001814static struct srv_kw_list srv_kws = { "TCP", { }, {
1815#ifdef TCP_USER_TIMEOUT
1816 { "tcp-ut", srv_parse_tcp_ut, 1, 0 }, /* set TCP user timeout on server */
1817#endif
1818 { NULL, NULL, 0 },
1819}};
Willy Tarreau2d392c22015-08-24 01:43:45 +02001820
1821static struct action_kw_list tcp_req_conn_actions = {ILH, {
William Lallemand2e785f22016-05-25 01:48:42 +02001822 { "silent-drop", tcp_parse_silent_drop },
William Lallemand13e9b0c2016-05-25 02:34:07 +02001823 { "set-src", tcp_parse_set_src_dst },
1824 { "set-src-port", tcp_parse_set_src_dst },
1825 { "set-dst" , tcp_parse_set_src_dst },
1826 { "set-dst-port", tcp_parse_set_src_dst },
Willy Tarreau2d392c22015-08-24 01:43:45 +02001827 { /* END */ }
1828}};
1829
Willy Tarreau620408f2016-10-21 16:37:51 +02001830static struct action_kw_list tcp_req_sess_actions = {ILH, {
1831 { "silent-drop", tcp_parse_silent_drop },
1832 { "set-src", tcp_parse_set_src_dst },
1833 { "set-src-port", tcp_parse_set_src_dst },
1834 { "set-dst" , tcp_parse_set_src_dst },
1835 { "set-dst-port", tcp_parse_set_src_dst },
1836 { /* END */ }
1837}};
1838
Willy Tarreau2d392c22015-08-24 01:43:45 +02001839static struct action_kw_list tcp_req_cont_actions = {ILH, {
1840 { "silent-drop", tcp_parse_silent_drop },
1841 { /* END */ }
1842}};
1843
1844static struct action_kw_list tcp_res_cont_actions = {ILH, {
1845 { "silent-drop", tcp_parse_silent_drop },
1846 { /* END */ }
1847}};
1848
1849static struct action_kw_list http_req_actions = {ILH, {
William Lallemand2e785f22016-05-25 01:48:42 +02001850 { "silent-drop", tcp_parse_silent_drop },
William Lallemand13e9b0c2016-05-25 02:34:07 +02001851 { "set-src", tcp_parse_set_src_dst },
1852 { "set-src-port", tcp_parse_set_src_dst },
1853 { "set-dst", tcp_parse_set_src_dst },
1854 { "set-dst-port", tcp_parse_set_src_dst },
Willy Tarreau2d392c22015-08-24 01:43:45 +02001855 { /* END */ }
1856}};
1857
1858static struct action_kw_list http_res_actions = {ILH, {
1859 { "silent-drop", tcp_parse_silent_drop },
1860 { /* END */ }
1861}};
1862
1863
Willy Tarreaue6b98942007-10-29 01:09:36 +01001864__attribute__((constructor))
1865static void __tcp_protocol_init(void)
1866{
1867 protocol_register(&proto_tcpv4);
1868 protocol_register(&proto_tcpv6);
Willy Tarreau12785782012-04-27 21:37:17 +02001869 sample_register_fetches(&sample_fetch_keywords);
Willy Tarreau44791242012-09-12 23:27:21 +02001870 bind_register_keywords(&bind_kws);
Willy Tarreau163d4622015-10-13 16:16:41 +02001871 srv_register_keywords(&srv_kws);
Willy Tarreau2d392c22015-08-24 01:43:45 +02001872 tcp_req_conn_keywords_register(&tcp_req_conn_actions);
Willy Tarreau620408f2016-10-21 16:37:51 +02001873 tcp_req_sess_keywords_register(&tcp_req_sess_actions);
Willy Tarreau2d392c22015-08-24 01:43:45 +02001874 tcp_req_cont_keywords_register(&tcp_req_cont_actions);
1875 tcp_res_cont_keywords_register(&tcp_res_cont_actions);
1876 http_req_keywords_register(&http_req_actions);
1877 http_res_keywords_register(&http_res_actions);
Willy Tarreauba962912016-12-21 18:55:02 +01001878
1879
1880 hap_register_build_opts("Built with transparent proxy support using:"
1881#if defined(IP_TRANSPARENT)
1882 " IP_TRANSPARENT"
1883#endif
1884#if defined(IPV6_TRANSPARENT)
1885 " IPV6_TRANSPARENT"
1886#endif
1887#if defined(IP_FREEBIND)
1888 " IP_FREEBIND"
1889#endif
1890#if defined(IP_BINDANY)
1891 " IP_BINDANY"
1892#endif
1893#if defined(IPV6_BINDANY)
1894 " IPV6_BINDANY"
1895#endif
1896#if defined(SO_BINDANY)
1897 " SO_BINDANY"
1898#endif
1899 "", 0);
Willy Tarreaue6b98942007-10-29 01:09:36 +01001900}
1901
1902
1903/*
1904 * Local variables:
1905 * c-indent-level: 8
1906 * c-basic-offset: 8
1907 * End:
1908 */