blob: ab32567bd566c2a5a44e66f7284753cd814846d6 [file] [log] [blame]
Willy Tarreau59f98392012-07-06 14:13:49 +02001/*
2 * Connection management functions
3 *
4 * Copyright 2000-2012 Willy Tarreau <w@1wt.eu>
5 *
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
Willy Tarreaue1e4a612012-10-05 00:10:55 +020013#include <errno.h>
14
Willy Tarreau59f98392012-07-06 14:13:49 +020015#include <common/compat.h>
16#include <common/config.h>
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +010017#include <common/namespace.h>
Emmanuel Hocdet4399c752018-02-05 15:26:43 +010018#include <common/hash.h>
19#include <common/net_helper.h>
Willy Tarreau59f98392012-07-06 14:13:49 +020020
Willy Tarreauc5788912012-08-24 18:12:41 +020021#include <proto/connection.h>
Willy Tarreaudd2f85e2012-09-02 22:34:23 +020022#include <proto/fd.h>
Willy Tarreau5f1504f2012-10-04 23:55:57 +020023#include <proto/frontend.h>
Willy Tarreau2da156f2012-07-23 15:07:23 +020024#include <proto/proto_tcp.h>
Willy Tarreau2c6be842012-07-06 17:12:34 +020025#include <proto/stream_interface.h>
Emeric Brun4f603012017-01-05 15:11:44 +010026#include <proto/sample.h>
Willy Tarreau59f98392012-07-06 14:13:49 +020027
Emeric Brun46591952012-05-18 15:47:34 +020028#ifdef USE_OPENSSL
29#include <proto/ssl_sock.h>
30#endif
31
Willy Tarreaubafbe012017-11-24 17:34:44 +010032struct pool_head *pool_head_connection;
33struct pool_head *pool_head_connstream;
Willy Tarreau13e14102016-12-22 20:25:26 +010034struct xprt_ops *registered_xprt[XPRT_ENTRIES] = { NULL, };
Willy Tarreauf2943dc2012-10-26 20:10:28 +020035
Christopher Faulet32f61c02018-04-10 14:33:41 +020036/* List head of all known muxes for PROTO */
37struct mux_proto_list mux_proto_list = {
38 .list = LIST_HEAD_INIT(mux_proto_list.list)
Willy Tarreau2386be62017-09-21 19:40:52 +020039};
40
Willy Tarreauf2943dc2012-10-26 20:10:28 +020041/* perform minimal intializations, report 0 in case of error, 1 if OK. */
42int init_connection()
43{
Willy Tarreaubafbe012017-11-24 17:34:44 +010044 pool_head_connection = create_pool("connection", sizeof (struct connection), MEM_F_SHARED);
45 if (!pool_head_connection)
Olivier Houcharde2b40b92017-09-13 18:30:23 +020046 goto fail_conn;
47
Willy Tarreaubafbe012017-11-24 17:34:44 +010048 pool_head_connstream = create_pool("conn_stream", sizeof(struct conn_stream), MEM_F_SHARED);
49 if (!pool_head_connstream)
Olivier Houcharde2b40b92017-09-13 18:30:23 +020050 goto fail_cs;
51
52 return 1;
53 fail_cs:
Willy Tarreaubafbe012017-11-24 17:34:44 +010054 pool_destroy(pool_head_connection);
55 pool_head_connection = NULL;
Olivier Houcharde2b40b92017-09-13 18:30:23 +020056 fail_conn:
57 return 0;
Willy Tarreauf2943dc2012-10-26 20:10:28 +020058}
59
Willy Tarreau59f98392012-07-06 14:13:49 +020060/* I/O callback for fd-based connections. It calls the read/write handlers
Willy Tarreau7a798e52016-04-14 11:13:20 +020061 * provided by the connection's sock_ops, which must be valid.
Willy Tarreau59f98392012-07-06 14:13:49 +020062 */
Willy Tarreau7a798e52016-04-14 11:13:20 +020063void conn_fd_handler(int fd)
Willy Tarreau59f98392012-07-06 14:13:49 +020064{
Willy Tarreau80184712012-07-06 14:54:49 +020065 struct connection *conn = fdtab[fd].owner;
Willy Tarreau9e272bf2012-10-03 21:04:48 +020066 unsigned int flags;
Olivier Houchard910b2bc2018-07-17 18:49:38 +020067 int can_send = 0;
Willy Tarreau59f98392012-07-06 14:13:49 +020068
Willy Tarreaud80cb4e2018-01-20 19:30:13 +010069 if (unlikely(!conn)) {
70 activity[tid].conn_dead++;
Willy Tarreau7a798e52016-04-14 11:13:20 +020071 return;
Willy Tarreaud80cb4e2018-01-20 19:30:13 +010072 }
Willy Tarreau59f98392012-07-06 14:13:49 +020073
Willy Tarreau7d281492012-12-16 19:19:13 +010074 conn_refresh_polling_flags(conn);
Willy Tarreau916e12d2017-10-25 09:22:43 +020075 conn->flags |= CO_FL_WILL_UPDATE;
76
Willy Tarreau7d281492012-12-16 19:19:13 +010077 flags = conn->flags & ~CO_FL_ERROR; /* ensure to call the wake handler upon error */
Willy Tarreaud29a0662012-12-10 16:33:38 +010078
Willy Tarreauc76ae332012-07-12 15:32:13 +020079 process_handshake:
Willy Tarreauf9dabec2012-08-17 17:33:53 +020080 /* The handshake callbacks are called in sequence. If either of them is
81 * missing something, it must enable the required polling at the socket
82 * layer of the connection. Polling state is not guaranteed when entering
83 * these handlers, so any handshake handler which does not complete its
Willy Tarreaud6e999b2013-11-25 08:41:15 +010084 * work must explicitly disable events it's not interested in. Error
85 * handling is also performed here in order to reduce the number of tests
86 * around.
Willy Tarreauf9dabec2012-08-17 17:33:53 +020087 */
Willy Tarreaud6e999b2013-11-25 08:41:15 +010088 while (unlikely(conn->flags & (CO_FL_HANDSHAKE | CO_FL_ERROR))) {
Willy Tarreau310987a2014-01-22 19:46:33 +010089 if (unlikely(conn->flags & CO_FL_ERROR))
Willy Tarreau2c6be842012-07-06 17:12:34 +020090 goto leave;
Willy Tarreau59f98392012-07-06 14:13:49 +020091
Bertrand Jacquin93b227d2016-06-04 15:11:10 +010092 if (conn->flags & CO_FL_ACCEPT_CIP)
93 if (!conn_recv_netscaler_cip(conn, CO_FL_ACCEPT_CIP))
94 goto leave;
95
Willy Tarreau22cda212012-08-31 17:43:29 +020096 if (conn->flags & CO_FL_ACCEPT_PROXY)
97 if (!conn_recv_proxy(conn, CO_FL_ACCEPT_PROXY))
98 goto leave;
99
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200100 if (conn->flags & CO_FL_SEND_PROXY)
101 if (!conn_si_send_proxy(conn, CO_FL_SEND_PROXY))
Willy Tarreau5f1504f2012-10-04 23:55:57 +0200102 goto leave;
Emeric Brun46591952012-05-18 15:47:34 +0200103#ifdef USE_OPENSSL
104 if (conn->flags & CO_FL_SSL_WAIT_HS)
105 if (!ssl_sock_handshake(conn, CO_FL_SSL_WAIT_HS))
106 goto leave;
107#endif
Willy Tarreauc76ae332012-07-12 15:32:13 +0200108 }
109
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200110 /* Once we're purely in the data phase, we disable handshake polling */
111 if (!(conn->flags & CO_FL_POLL_SOCK))
112 __conn_sock_stop_both(conn);
Willy Tarreauc76ae332012-07-12 15:32:13 +0200113
Willy Tarreau8e3c6ce2017-08-28 15:46:01 +0200114 /* The connection owner might want to be notified about an end of
115 * handshake indicating the connection is ready, before we proceed with
116 * any data exchange. The callback may fail and cause the connection to
117 * be destroyed, thus we must not use it anymore and should immediately
118 * leave instead. The caller must immediately unregister itself once
119 * called.
Willy Tarreau2542b532012-08-31 16:01:23 +0200120 */
Willy Tarreau8e3c6ce2017-08-28 15:46:01 +0200121 if (conn->xprt_done_cb && conn->xprt_done_cb(conn) < 0)
Willy Tarreau7a798e52016-04-14 11:13:20 +0200122 return;
Willy Tarreau2542b532012-08-31 16:01:23 +0200123
Willy Tarreau57ec32f2017-04-11 19:59:33 +0200124 if (conn->xprt && fd_send_ready(fd) &&
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200125 ((conn->flags & (CO_FL_XPRT_WR_ENA|CO_FL_ERROR|CO_FL_HANDSHAKE)) == CO_FL_XPRT_WR_ENA)) {
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100126 /* force reporting of activity by clearing the previous flags :
127 * we'll have at least ERROR or CONNECTED at the end of an I/O,
128 * both of which will be detected below.
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200129 */
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100130 flags = 0;
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200131 can_send = LIST_ISEMPTY(&conn->send_wait_list);
Olivier Houchard6ff20392018-07-17 18:46:31 +0200132 while (!LIST_ISEMPTY(&conn->send_wait_list)) {
133 struct wait_list *sw = LIST_ELEM(conn->send_wait_list.n,
134 struct wait_list *, list);
135 LIST_DEL(&sw->list);
136 LIST_INIT(&sw->list);
Olivier Houcharde1c6dbc2018-08-01 17:06:43 +0200137 sw->wait_reason &= ~SUB_CAN_SEND;
Olivier Houchard6ff20392018-07-17 18:46:31 +0200138 tasklet_wakeup(sw->task);
139 }
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200140 }
Willy Tarreau59f98392012-07-06 14:13:49 +0200141
Willy Tarreau57ec32f2017-04-11 19:59:33 +0200142 /* The data transfer starts here and stops on error and handshakes. Note
143 * that we must absolutely test conn->xprt at each step in case it suddenly
144 * changes due to a quick unexpected close().
145 */
146 if (conn->xprt && fd_recv_ready(fd) &&
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200147 ((conn->flags & (CO_FL_XPRT_RD_ENA|CO_FL_WAIT_ROOM|CO_FL_ERROR|CO_FL_HANDSHAKE)) == CO_FL_XPRT_RD_ENA)) {
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100148 /* force reporting of activity by clearing the previous flags :
149 * we'll have at least ERROR or CONNECTED at the end of an I/O,
150 * both of which will be detected below.
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200151 */
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100152 flags = 0;
Willy Tarreau53a47662017-08-28 10:53:00 +0200153 conn->mux->recv(conn);
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200154 }
Willy Tarreau2da156f2012-07-23 15:07:23 +0200155
Willy Tarreauc76ae332012-07-12 15:32:13 +0200156 /* It may happen during the data phase that a handshake is
157 * enabled again (eg: SSL)
158 */
Willy Tarreaud6e999b2013-11-25 08:41:15 +0100159 if (unlikely(conn->flags & (CO_FL_HANDSHAKE | CO_FL_ERROR)))
Willy Tarreauc76ae332012-07-12 15:32:13 +0200160 goto process_handshake;
161
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100162 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreauf8deb0c2012-09-01 17:59:22 +0200163 /* still waiting for a connection to establish and nothing was
164 * attempted yet to probe the connection. Then let's retry the
165 * connect().
Willy Tarreau2da156f2012-07-23 15:07:23 +0200166 */
Willy Tarreau239d7182012-07-23 18:53:03 +0200167 if (!tcp_connect_probe(conn))
Willy Tarreauafad0e02012-08-09 14:45:22 +0200168 goto leave;
Willy Tarreau2da156f2012-07-23 15:07:23 +0200169 }
Willy Tarreau2c6be842012-07-06 17:12:34 +0200170 leave:
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100171 /* Verify if the connection just established. */
Willy Tarreau7bf3fa32017-03-14 20:19:29 +0100172 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED))))
173 conn->flags |= CO_FL_CONNECTED;
174
Willy Tarreau8e3c6ce2017-08-28 15:46:01 +0200175 /* The connection owner might want to be notified about failures to
176 * complete the handshake. The callback may fail and cause the
177 * connection to be destroyed, thus we must not use it anymore and
178 * should immediately leave instead. The caller must immediately
179 * unregister itself once called.
180 */
181 if (((conn->flags ^ flags) & CO_FL_NOTIFY_DONE) &&
182 conn->xprt_done_cb && conn->xprt_done_cb(conn) < 0)
183 return;
184
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100185 /* The wake callback is normally used to notify the data layer about
186 * data layer activity (successful send/recv), connection establishment,
187 * shutdown and fatal errors. We need to consider the following
188 * situations to wake up the data layer :
189 * - change among the CO_FL_NOTIFY_DATA flags :
190 * {DATA,SOCK}_{RD,WR}_SH, ERROR,
191 * - absence of any of {L4,L6}_CONN and CONNECTED, indicating the
192 * end of handshake and transition to CONNECTED
193 * - raise of CONNECTED with HANDSHAKE down
194 * - end of HANDSHAKE with CONNECTED set
195 * - regular data layer activity
196 *
197 * Note that the wake callback is allowed to release the connection and
198 * the fd (and return < 0 in this case).
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200199 */
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200200 if ((can_send || (((conn->flags ^ flags) & CO_FL_NOTIFY_DATA) ||
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100201 ((flags & (CO_FL_CONNECTED|CO_FL_HANDSHAKE)) != CO_FL_CONNECTED &&
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200202 (conn->flags & (CO_FL_CONNECTED|CO_FL_HANDSHAKE)) == CO_FL_CONNECTED))) &&
Willy Tarreau53a47662017-08-28 10:53:00 +0200203 conn->mux->wake(conn) < 0)
Willy Tarreau7a798e52016-04-14 11:13:20 +0200204 return;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200205
Willy Tarreau61ace1b2012-07-23 12:14:26 +0200206 /* remove the events before leaving */
Willy Tarreau26d7cfc2012-12-07 00:09:43 +0100207 fdtab[fd].ev &= FD_POLL_STICKY;
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200208
209 /* commit polling changes */
Willy Tarreau916e12d2017-10-25 09:22:43 +0200210 conn->flags &= ~CO_FL_WILL_UPDATE;
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200211 conn_cond_update_polling(conn);
Willy Tarreau7a798e52016-04-14 11:13:20 +0200212 return;
Willy Tarreau59f98392012-07-06 14:13:49 +0200213}
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200214
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200215/* Update polling on connection <c>'s file descriptor depending on its current
216 * state as reported in the connection's CO_FL_CURR_* flags, reports of EAGAIN
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200217 * in CO_FL_WAIT_*, and the data layer expectations indicated by CO_FL_XPRT_*.
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200218 * The connection flags are updated with the new flags at the end of the
Willy Tarreau0ffde2c2012-10-04 22:21:15 +0200219 * operation. Polling is totally disabled if an error was reported.
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200220 */
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200221void conn_update_xprt_polling(struct connection *c)
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200222{
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200223 unsigned int f = c->flags;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200224
Willy Tarreau3c728722014-01-23 13:50:42 +0100225 if (!conn_ctrl_ready(c))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200226 return;
227
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200228 /* update read status if needed */
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200229 if (unlikely((f & (CO_FL_CURR_RD_ENA|CO_FL_XPRT_RD_ENA)) == CO_FL_XPRT_RD_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200230 fd_want_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100231 f |= CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200232 }
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200233 else if (unlikely((f & (CO_FL_CURR_RD_ENA|CO_FL_XPRT_RD_ENA)) == CO_FL_CURR_RD_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200234 fd_stop_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100235 f &= ~CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200236 }
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200237
238 /* update write status if needed */
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200239 if (unlikely((f & (CO_FL_CURR_WR_ENA|CO_FL_XPRT_WR_ENA)) == CO_FL_XPRT_WR_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200240 fd_want_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100241 f |= CO_FL_CURR_WR_ENA;
242 }
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200243 else if (unlikely((f & (CO_FL_CURR_WR_ENA|CO_FL_XPRT_WR_ENA)) == CO_FL_CURR_WR_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200244 fd_stop_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100245 f &= ~CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200246 }
Willy Tarreau310987a2014-01-22 19:46:33 +0100247 c->flags = f;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200248}
249
250/* Update polling on connection <c>'s file descriptor depending on its current
251 * state as reported in the connection's CO_FL_CURR_* flags, reports of EAGAIN
252 * in CO_FL_WAIT_*, and the sock layer expectations indicated by CO_FL_SOCK_*.
253 * The connection flags are updated with the new flags at the end of the
Willy Tarreau0ffde2c2012-10-04 22:21:15 +0200254 * operation. Polling is totally disabled if an error was reported.
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200255 */
256void conn_update_sock_polling(struct connection *c)
257{
258 unsigned int f = c->flags;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200259
Willy Tarreau3c728722014-01-23 13:50:42 +0100260 if (!conn_ctrl_ready(c))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200261 return;
262
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200263 /* update read status if needed */
Willy Tarreau310987a2014-01-22 19:46:33 +0100264 if (unlikely((f & (CO_FL_CURR_RD_ENA|CO_FL_SOCK_RD_ENA)) == CO_FL_SOCK_RD_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200265 fd_want_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100266 f |= CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200267 }
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100268 else if (unlikely((f & (CO_FL_CURR_RD_ENA|CO_FL_SOCK_RD_ENA)) == CO_FL_CURR_RD_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200269 fd_stop_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100270 f &= ~CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200271 }
272
273 /* update write status if needed */
Willy Tarreau310987a2014-01-22 19:46:33 +0100274 if (unlikely((f & (CO_FL_CURR_WR_ENA|CO_FL_SOCK_WR_ENA)) == CO_FL_SOCK_WR_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200275 fd_want_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100276 f |= CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200277 }
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100278 else if (unlikely((f & (CO_FL_CURR_WR_ENA|CO_FL_SOCK_WR_ENA)) == CO_FL_CURR_WR_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200279 fd_stop_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100280 f &= ~CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200281 }
Willy Tarreau310987a2014-01-22 19:46:33 +0100282 c->flags = f;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200283}
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200284
Willy Tarreauff3e6482015-03-12 23:56:52 +0100285/* Send a message over an established connection. It makes use of send() and
286 * returns the same return code and errno. If the socket layer is not ready yet
287 * then -1 is returned and ENOTSOCK is set into errno. If the fd is not marked
288 * as ready, or if EAGAIN or ENOTCONN is returned, then we return 0. It returns
289 * EMSGSIZE if called with a zero length message. The purpose is to simplify
290 * some rare attempts to directly write on the socket from above the connection
291 * (typically send_proxy). In case of EAGAIN, the fd is marked as "cant_send".
292 * It automatically retries on EINTR. Other errors cause the connection to be
293 * marked as in error state. It takes similar arguments as send() except the
294 * first one which is the connection instead of the file descriptor. Note,
295 * MSG_DONTWAIT and MSG_NOSIGNAL are forced on the flags.
296 */
297int conn_sock_send(struct connection *conn, const void *buf, int len, int flags)
298{
299 int ret;
300
301 ret = -1;
302 errno = ENOTSOCK;
303
304 if (conn->flags & CO_FL_SOCK_WR_SH)
305 goto fail;
306
307 if (!conn_ctrl_ready(conn))
308 goto fail;
309
310 errno = EMSGSIZE;
311 if (!len)
312 goto fail;
313
Willy Tarreau585744b2017-08-24 14:31:19 +0200314 if (!fd_send_ready(conn->handle.fd))
Willy Tarreauff3e6482015-03-12 23:56:52 +0100315 goto wait;
316
317 do {
Willy Tarreau585744b2017-08-24 14:31:19 +0200318 ret = send(conn->handle.fd, buf, len, flags | MSG_DONTWAIT | MSG_NOSIGNAL);
Willy Tarreauff3e6482015-03-12 23:56:52 +0100319 } while (ret < 0 && errno == EINTR);
320
321
322 if (ret > 0)
323 return ret;
324
325 if (ret == 0 || errno == EAGAIN || errno == ENOTCONN) {
326 wait:
Willy Tarreau585744b2017-08-24 14:31:19 +0200327 fd_cant_send(conn->handle.fd);
Willy Tarreauff3e6482015-03-12 23:56:52 +0100328 return 0;
329 }
330 fail:
331 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH | CO_FL_ERROR;
332 return ret;
333}
334
Olivier Houchard6ff20392018-07-17 18:46:31 +0200335int conn_subscribe(struct connection *conn, int event_type, void *param)
336{
337 struct wait_list *sw;
338
339 switch (event_type) {
340 case SUB_CAN_SEND:
341 sw = param;
Olivier Houcharde1c6dbc2018-08-01 17:06:43 +0200342 if (!(sw->wait_reason & SUB_CAN_SEND)) {
343 sw->wait_reason |= SUB_CAN_SEND;
Olivier Houchard6ff20392018-07-17 18:46:31 +0200344 LIST_ADDQ(&conn->send_wait_list, &sw->list);
Olivier Houcharde1c6dbc2018-08-01 17:06:43 +0200345 }
Olivier Houchard6ff20392018-07-17 18:46:31 +0200346 return 0;
347 default:
348 break;
349 }
350 return (-1);
351}
352
Willy Tarreaud85c4852015-03-13 00:40:28 +0100353/* Drains possibly pending incoming data on the file descriptor attached to the
354 * connection and update the connection's flags accordingly. This is used to
355 * know whether we need to disable lingering on close. Returns non-zero if it
356 * is safe to close without disabling lingering, otherwise zero. The SOCK_RD_SH
357 * flag may also be updated if the incoming shutdown was reported by the drain()
358 * function.
359 */
360int conn_sock_drain(struct connection *conn)
361{
Willy Tarreaue215bba2018-08-24 14:31:53 +0200362 int turns = 2;
363 int len;
364
Willy Tarreaud85c4852015-03-13 00:40:28 +0100365 if (!conn_ctrl_ready(conn))
366 return 1;
367
368 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH))
369 return 1;
370
Willy Tarreaue215bba2018-08-24 14:31:53 +0200371 if (fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP))
372 goto shut;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100373
Willy Tarreaue215bba2018-08-24 14:31:53 +0200374 if (!fd_recv_ready(conn->handle.fd))
375 return 0;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100376
Willy Tarreaue215bba2018-08-24 14:31:53 +0200377 if (conn->ctrl->drain) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200378 if (conn->ctrl->drain(conn->handle.fd) <= 0)
Willy Tarreaud85c4852015-03-13 00:40:28 +0100379 return 0;
Willy Tarreaue215bba2018-08-24 14:31:53 +0200380 goto shut;
381 }
382
383 /* no drain function defined, use the generic one */
384
385 while (turns) {
386#ifdef MSG_TRUNC_CLEARS_INPUT
387 len = recv(conn->handle.fd, NULL, INT_MAX, MSG_DONTWAIT | MSG_NOSIGNAL | MSG_TRUNC);
388 if (len == -1 && errno == EFAULT)
389#endif
390 len = recv(conn->handle.fd, trash.area, trash.size,
391 MSG_DONTWAIT | MSG_NOSIGNAL);
392
393 if (len == 0)
394 goto shut;
395
396 if (len < 0) {
397 if (errno == EAGAIN) {
398 /* connection not closed yet */
399 fd_cant_recv(conn->handle.fd);
400 break;
401 }
402 if (errno == EINTR) /* oops, try again */
403 continue;
404 /* other errors indicate a dead connection, fine. */
405 goto shut;
406 }
407 /* OK we read some data, let's try again once */
408 turns--;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100409 }
410
Willy Tarreaue215bba2018-08-24 14:31:53 +0200411 /* some data are still present, give up */
412 return 0;
413
414 shut:
415 /* we're certain the connection was shut down */
416 fdtab[conn->handle.fd].linger_risk = 0;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100417 conn->flags |= CO_FL_SOCK_RD_SH;
418 return 1;
419}
420
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100421/*
422 * Get data length from tlv
423 */
424static int get_tlv_length(const struct tlv *src)
425{
426 return (src->length_hi << 8) | src->length_lo;
427}
428
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200429/* This handshake handler waits a PROXY protocol header at the beginning of the
430 * raw data stream. The header looks like this :
431 *
432 * "PROXY" <SP> PROTO <SP> SRC3 <SP> DST3 <SP> SRC4 <SP> <DST4> "\r\n"
433 *
434 * There must be exactly one space between each field. Fields are :
435 * - PROTO : layer 4 protocol, which must be "TCP4" or "TCP6".
436 * - SRC3 : layer 3 (eg: IP) source address in standard text form
437 * - DST3 : layer 3 (eg: IP) destination address in standard text form
438 * - SRC4 : layer 4 (eg: TCP port) source address in standard text form
439 * - DST4 : layer 4 (eg: TCP port) destination address in standard text form
440 *
441 * This line MUST be at the beginning of the buffer and MUST NOT wrap.
442 *
443 * The header line is small and in all cases smaller than the smallest normal
444 * TCP MSS. So it MUST always be delivered as one segment, which ensures we
445 * can safely use MSG_PEEK and avoid buffering.
446 *
447 * Once the data is fetched, the values are set in the connection's address
448 * fields, and data are removed from the socket's buffer. The function returns
449 * zero if it needs to wait for more data or if it fails, or 1 if it completed
450 * and removed itself.
451 */
452int conn_recv_proxy(struct connection *conn, int flag)
453{
454 char *line, *end;
Willy Tarreau77992672014-06-14 11:06:17 +0200455 struct proxy_hdr_v2 *hdr_v2;
456 const char v2sig[] = PP2_SIGNATURE;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100457 int tlv_length = 0;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200458 int tlv_offset = 0;
Willy Tarreaub406b872018-08-22 05:20:32 +0200459 int ret;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200460
461 /* we might have been called just after an asynchronous shutr */
462 if (conn->flags & CO_FL_SOCK_RD_SH)
463 goto fail;
464
Willy Tarreau3c728722014-01-23 13:50:42 +0100465 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200466 goto fail;
467
Willy Tarreau585744b2017-08-24 14:31:19 +0200468 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100469 return 0;
470
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200471 do {
Willy Tarreaub406b872018-08-22 05:20:32 +0200472 ret = recv(conn->handle.fd, trash.area, trash.size, MSG_PEEK);
473 if (ret < 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200474 if (errno == EINTR)
475 continue;
476 if (errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200477 fd_cant_recv(conn->handle.fd);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200478 return 0;
479 }
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100480 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200481 }
Willy Tarreaub406b872018-08-22 05:20:32 +0200482 trash.data = ret;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200483 } while (0);
484
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200485 if (!trash.data) {
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100486 /* client shutdown */
487 conn->err_code = CO_ER_PRX_EMPTY;
488 goto fail;
489 }
490
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200491 if (trash.data < 6)
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200492 goto missing;
493
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200494 line = trash.area;
495 end = trash.area + trash.data;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200496
497 /* Decode a possible proxy request, fail early if it does not match */
Willy Tarreau77992672014-06-14 11:06:17 +0200498 if (strncmp(line, "PROXY ", 6) != 0)
499 goto not_v1;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200500
501 line += 6;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200502 if (trash.data < 9) /* shortest possible line */
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200503 goto missing;
504
David CARLIER42ff05e2016-03-24 09:22:36 +0000505 if (memcmp(line, "TCP4 ", 5) == 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200506 u32 src3, dst3, sport, dport;
507
508 line += 5;
509
510 src3 = inetaddr_host_lim_ret(line, end, &line);
511 if (line == end)
512 goto missing;
513 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100514 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200515
516 dst3 = inetaddr_host_lim_ret(line, end, &line);
517 if (line == end)
518 goto missing;
519 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100520 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200521
522 sport = read_uint((const char **)&line, end);
523 if (line == end)
524 goto missing;
525 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100526 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200527
528 dport = read_uint((const char **)&line, end);
529 if (line > end - 2)
530 goto missing;
531 if (*line++ != '\r')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100532 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200533 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100534 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200535
536 /* update the session's addresses and mark them set */
537 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
538 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = htonl(src3);
539 ((struct sockaddr_in *)&conn->addr.from)->sin_port = htons(sport);
540
541 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
542 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = htonl(dst3);
543 ((struct sockaddr_in *)&conn->addr.to)->sin_port = htons(dport);
544 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
545 }
David CARLIER42ff05e2016-03-24 09:22:36 +0000546 else if (memcmp(line, "TCP6 ", 5) == 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200547 u32 sport, dport;
548 char *src_s;
549 char *dst_s, *sport_s, *dport_s;
550 struct in6_addr src3, dst3;
551
552 line += 5;
553
554 src_s = line;
555 dst_s = sport_s = dport_s = NULL;
556 while (1) {
557 if (line > end - 2) {
558 goto missing;
559 }
560 else if (*line == '\r') {
561 *line = 0;
562 line++;
563 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100564 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200565 break;
566 }
567
568 if (*line == ' ') {
569 *line = 0;
570 if (!dst_s)
571 dst_s = line + 1;
572 else if (!sport_s)
573 sport_s = line + 1;
574 else if (!dport_s)
575 dport_s = line + 1;
576 }
577 line++;
578 }
579
580 if (!dst_s || !sport_s || !dport_s)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100581 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200582
583 sport = read_uint((const char **)&sport_s,dport_s - 1);
584 if (*sport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100585 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200586
587 dport = read_uint((const char **)&dport_s,line - 2);
588 if (*dport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100589 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200590
591 if (inet_pton(AF_INET6, src_s, (void *)&src3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100592 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200593
594 if (inet_pton(AF_INET6, dst_s, (void *)&dst3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100595 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200596
597 /* update the session's addresses and mark them set */
598 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
599 memcpy(&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr, &src3, sizeof(struct in6_addr));
600 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = htons(sport);
601
602 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
603 memcpy(&((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr, &dst3, sizeof(struct in6_addr));
604 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = htons(dport);
605 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
606 }
Willy Tarreau4c20d292014-06-14 11:41:36 +0200607 else if (memcmp(line, "UNKNOWN\r\n", 9) == 0) {
608 /* This can be a UNIX socket forwarded by an haproxy upstream */
609 line += 9;
610 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200611 else {
Willy Tarreau4c20d292014-06-14 11:41:36 +0200612 /* The protocol does not match something known (TCP4/TCP6/UNKNOWN) */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100613 conn->err_code = CO_ER_PRX_BAD_PROTO;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200614 goto fail;
615 }
616
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200617 trash.data = line - trash.area;
Willy Tarreau77992672014-06-14 11:06:17 +0200618 goto eat_header;
619
620 not_v1:
621 /* try PPv2 */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200622 if (trash.data < PP2_HEADER_LEN)
Willy Tarreau77992672014-06-14 11:06:17 +0200623 goto missing;
624
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200625 hdr_v2 = (struct proxy_hdr_v2 *) trash.area;
Willy Tarreau77992672014-06-14 11:06:17 +0200626
627 if (memcmp(hdr_v2->sig, v2sig, PP2_SIGNATURE_LEN) != 0 ||
628 (hdr_v2->ver_cmd & PP2_VERSION_MASK) != PP2_VERSION) {
629 conn->err_code = CO_ER_PRX_NOT_HDR;
630 goto fail;
631 }
632
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200633 if (trash.data < PP2_HEADER_LEN + ntohs(hdr_v2->len))
Willy Tarreau77992672014-06-14 11:06:17 +0200634 goto missing;
635
636 switch (hdr_v2->ver_cmd & PP2_CMD_MASK) {
637 case 0x01: /* PROXY command */
638 switch (hdr_v2->fam) {
639 case 0x11: /* TCPv4 */
KOVACS Krisztianefd3aa92014-11-19 10:53:20 +0100640 if (ntohs(hdr_v2->len) < PP2_ADDR_LEN_INET)
641 goto bad_header;
642
Willy Tarreau77992672014-06-14 11:06:17 +0200643 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
644 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = hdr_v2->addr.ip4.src_addr;
645 ((struct sockaddr_in *)&conn->addr.from)->sin_port = hdr_v2->addr.ip4.src_port;
646 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
647 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = hdr_v2->addr.ip4.dst_addr;
648 ((struct sockaddr_in *)&conn->addr.to)->sin_port = hdr_v2->addr.ip4.dst_port;
649 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200650 tlv_offset = PP2_HEADER_LEN + PP2_ADDR_LEN_INET;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100651 tlv_length = ntohs(hdr_v2->len) - PP2_ADDR_LEN_INET;
Willy Tarreau77992672014-06-14 11:06:17 +0200652 break;
653 case 0x21: /* TCPv6 */
KOVACS Krisztianefd3aa92014-11-19 10:53:20 +0100654 if (ntohs(hdr_v2->len) < PP2_ADDR_LEN_INET6)
655 goto bad_header;
656
Willy Tarreau77992672014-06-14 11:06:17 +0200657 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
658 memcpy(&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr, hdr_v2->addr.ip6.src_addr, 16);
659 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = hdr_v2->addr.ip6.src_port;
660 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
661 memcpy(&((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr, hdr_v2->addr.ip6.dst_addr, 16);
662 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = hdr_v2->addr.ip6.dst_port;
663 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200664 tlv_offset = PP2_HEADER_LEN + PP2_ADDR_LEN_INET6;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100665 tlv_length = ntohs(hdr_v2->len) - PP2_ADDR_LEN_INET6;
Willy Tarreau77992672014-06-14 11:06:17 +0200666 break;
667 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100668
669 /* TLV parsing */
670 if (tlv_length > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200671 while (tlv_offset + TLV_HEADER_SIZE <= trash.data) {
672 const struct tlv *tlv_packet = (struct tlv *) &trash.area[tlv_offset];
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100673 const int tlv_len = get_tlv_length(tlv_packet);
674 tlv_offset += tlv_len + TLV_HEADER_SIZE;
675
676 switch (tlv_packet->type) {
Emmanuel Hocdet115df3e2018-02-05 16:23:23 +0100677 case PP2_TYPE_CRC32C: {
678 void *tlv_crc32c_p = (void *)tlv_packet->value;
679 uint32_t n_crc32c = ntohl(read_u32(tlv_crc32c_p));
680 write_u32(tlv_crc32c_p, 0);
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200681 if (hash_crc32c(trash.area, PP2_HEADER_LEN + ntohs(hdr_v2->len)) != n_crc32c)
Emmanuel Hocdet115df3e2018-02-05 16:23:23 +0100682 goto bad_header;
683 break;
684 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100685#ifdef CONFIG_HAP_NS
686 case PP2_TYPE_NETNS: {
687 const struct netns_entry *ns;
688 ns = netns_store_lookup((char*)tlv_packet->value, tlv_len);
689 if (ns)
690 conn->proxy_netns = ns;
691 break;
692 }
693#endif
694 default:
695 break;
696 }
697 }
698 }
699
Willy Tarreau77992672014-06-14 11:06:17 +0200700 /* unsupported protocol, keep local connection address */
701 break;
702 case 0x00: /* LOCAL command */
703 /* keep local connection address for LOCAL */
704 break;
705 default:
706 goto bad_header; /* not a supported command */
707 }
708
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200709 trash.data = PP2_HEADER_LEN + ntohs(hdr_v2->len);
Willy Tarreau77992672014-06-14 11:06:17 +0200710 goto eat_header;
711
712 eat_header:
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200713 /* remove the PROXY line from the request. For this we re-read the
714 * exact line at once. If we don't get the exact same result, we
715 * fail.
716 */
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200717 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200718 int len2 = recv(conn->handle.fd, trash.area, trash.data, 0);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200719 if (len2 < 0 && errno == EINTR)
720 continue;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200721 if (len2 != trash.data)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100722 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200723 } while (0);
724
725 conn->flags &= ~flag;
Emeric Brun4f603012017-01-05 15:11:44 +0100726 conn->flags |= CO_FL_RCVD_PROXY;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200727 return 1;
728
729 missing:
730 /* Missing data. Since we're using MSG_PEEK, we can only poll again if
731 * we have not read anything. Otherwise we need to fail because we won't
732 * be able to poll anymore.
733 */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100734 conn->err_code = CO_ER_PRX_TRUNCATED;
735 goto fail;
736
737 bad_header:
738 /* This is not a valid proxy protocol header */
739 conn->err_code = CO_ER_PRX_BAD_HDR;
740 goto fail;
741
742 recv_abort:
743 conn->err_code = CO_ER_PRX_ABORT;
Willy Tarreau26f4a042013-12-04 23:44:10 +0100744 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100745 goto fail;
746
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200747 fail:
Willy Tarreaud486ef52012-12-10 17:03:52 +0100748 __conn_sock_stop_both(conn);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200749 conn->flags |= CO_FL_ERROR;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200750 return 0;
751}
752
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100753/* This handshake handler waits a NetScaler Client IP insertion header
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000754 * at the beginning of the raw data stream. The header format is
755 * described in doc/netscaler-client-ip-insertion-protocol.txt
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100756 *
757 * This line MUST be at the beginning of the buffer and MUST NOT be
758 * fragmented.
759 *
760 * The header line is small and in all cases smaller than the smallest normal
761 * TCP MSS. So it MUST always be delivered as one segment, which ensures we
762 * can safely use MSG_PEEK and avoid buffering.
763 *
764 * Once the data is fetched, the values are set in the connection's address
765 * fields, and data are removed from the socket's buffer. The function returns
766 * zero if it needs to wait for more data or if it fails, or 1 if it completed
767 * and removed itself.
768 */
769int conn_recv_netscaler_cip(struct connection *conn, int flag)
770{
771 char *line;
Bertrand Jacquin7d668f92017-12-13 01:23:39 +0000772 uint32_t hdr_len;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100773 uint8_t ip_v;
Willy Tarreaub406b872018-08-22 05:20:32 +0200774 int ret;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100775
776 /* we might have been called just after an asynchronous shutr */
777 if (conn->flags & CO_FL_SOCK_RD_SH)
778 goto fail;
779
780 if (!conn_ctrl_ready(conn))
781 goto fail;
782
Willy Tarreau585744b2017-08-24 14:31:19 +0200783 if (!fd_recv_ready(conn->handle.fd))
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100784 return 0;
785
786 do {
Willy Tarreaub406b872018-08-22 05:20:32 +0200787 ret = recv(conn->handle.fd, trash.area, trash.size, MSG_PEEK);
788 if (ret < 0) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100789 if (errno == EINTR)
790 continue;
791 if (errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200792 fd_cant_recv(conn->handle.fd);
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100793 return 0;
794 }
795 goto recv_abort;
796 }
Willy Tarreaub406b872018-08-22 05:20:32 +0200797 trash.data = ret;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100798 } while (0);
799
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200800 if (!trash.data) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100801 /* client shutdown */
802 conn->err_code = CO_ER_CIP_EMPTY;
803 goto fail;
804 }
805
806 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000807 * CIP magic, header length or
808 * CIP magic, CIP length, CIP type, header length */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200809 if (trash.data < 12)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100810 goto missing;
811
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200812 line = trash.area;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100813
814 /* Decode a possible NetScaler Client IP request, fail early if
815 * it does not match */
Bertrand Jacquin4b4c2862017-12-13 01:07:12 +0000816 if (ntohl(*(uint32_t *)line) != objt_listener(conn->target)->bind_conf->ns_cip_magic)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100817 goto bad_magic;
818
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000819 /* Legacy CIP protocol */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200820 if ((trash.area[8] & 0xD0) == 0x40) {
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000821 hdr_len = ntohl(*(uint32_t *)(line+4));
822 line += 8;
823 }
824 /* Standard CIP protocol */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200825 else if (trash.area[8] == 0x00) {
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000826 hdr_len = ntohs(*(uint32_t *)(line+10));
827 line += 12;
828 }
829 /* Unknown CIP protocol */
830 else {
831 conn->err_code = CO_ER_CIP_BAD_PROTO;
832 goto fail;
833 }
834
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100835 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000836 * a minimal IP header */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200837 if (trash.data < 20)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100838 goto missing;
839
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100840 /* Get IP version from the first four bits */
841 ip_v = (*line & 0xf0) >> 4;
842
843 if (ip_v == 4) {
844 struct ip *hdr_ip4;
David Carlier3015a2e2016-07-04 22:51:33 +0100845 struct my_tcphdr *hdr_tcp;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100846
847 hdr_ip4 = (struct ip *)line;
848
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200849 if (trash.data < 40 || trash.data < hdr_len) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100850 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin67de5a22017-12-13 01:15:05 +0000851 * IPv4 header, TCP header */
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100852 goto missing;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000853 }
854 else if (hdr_ip4->ip_p != IPPROTO_TCP) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100855 /* The protocol does not include a TCP header */
856 conn->err_code = CO_ER_CIP_BAD_PROTO;
857 goto fail;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000858 }
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100859
David Carlier3015a2e2016-07-04 22:51:33 +0100860 hdr_tcp = (struct my_tcphdr *)(line + (hdr_ip4->ip_hl * 4));
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100861
862 /* update the session's addresses and mark them set */
863 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
864 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = hdr_ip4->ip_src.s_addr;
865 ((struct sockaddr_in *)&conn->addr.from)->sin_port = hdr_tcp->source;
866
867 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
868 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = hdr_ip4->ip_dst.s_addr;
869 ((struct sockaddr_in *)&conn->addr.to)->sin_port = hdr_tcp->dest;
870
871 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
872 }
873 else if (ip_v == 6) {
874 struct ip6_hdr *hdr_ip6;
David Carlier3015a2e2016-07-04 22:51:33 +0100875 struct my_tcphdr *hdr_tcp;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100876
877 hdr_ip6 = (struct ip6_hdr *)line;
878
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200879 if (trash.data < 60 || trash.data < hdr_len) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100880 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin67de5a22017-12-13 01:15:05 +0000881 * IPv6 header, TCP header */
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100882 goto missing;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000883 }
884 else if (hdr_ip6->ip6_nxt != IPPROTO_TCP) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100885 /* The protocol does not include a TCP header */
886 conn->err_code = CO_ER_CIP_BAD_PROTO;
887 goto fail;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000888 }
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100889
David Carlier3015a2e2016-07-04 22:51:33 +0100890 hdr_tcp = (struct my_tcphdr *)(line + sizeof(struct ip6_hdr));
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100891
892 /* update the session's addresses and mark them set */
893 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
894 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr = hdr_ip6->ip6_src;
895 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = hdr_tcp->source;
896
897 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
898 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr = hdr_ip6->ip6_dst;
899 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = hdr_tcp->dest;
900
901 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
902 }
903 else {
904 /* The protocol does not match something known (IPv4/IPv6) */
905 conn->err_code = CO_ER_CIP_BAD_PROTO;
906 goto fail;
907 }
908
Bertrand Jacquin7d668f92017-12-13 01:23:39 +0000909 line += hdr_len;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200910 trash.data = line - trash.area;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100911
912 /* remove the NetScaler Client IP header from the request. For this
913 * we re-read the exact line at once. If we don't get the exact same
914 * result, we fail.
915 */
916 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200917 int len2 = recv(conn->handle.fd, trash.area, trash.data, 0);
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100918 if (len2 < 0 && errno == EINTR)
919 continue;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200920 if (len2 != trash.data)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100921 goto recv_abort;
922 } while (0);
923
924 conn->flags &= ~flag;
925 return 1;
926
927 missing:
928 /* Missing data. Since we're using MSG_PEEK, we can only poll again if
929 * we have not read anything. Otherwise we need to fail because we won't
930 * be able to poll anymore.
931 */
932 conn->err_code = CO_ER_CIP_TRUNCATED;
933 goto fail;
934
935 bad_magic:
936 conn->err_code = CO_ER_CIP_BAD_MAGIC;
937 goto fail;
938
939 recv_abort:
940 conn->err_code = CO_ER_CIP_ABORT;
941 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
942 goto fail;
943
944 fail:
945 __conn_sock_stop_both(conn);
946 conn->flags |= CO_FL_ERROR;
947 return 0;
948}
949
David Safb76832014-05-08 23:42:08 -0400950int make_proxy_line(char *buf, int buf_len, struct server *srv, struct connection *remote)
951{
952 int ret = 0;
953
954 if (srv && (srv->pp_opts & SRV_PP_V2)) {
955 ret = make_proxy_line_v2(buf, buf_len, srv, remote);
956 }
957 else {
958 if (remote)
959 ret = make_proxy_line_v1(buf, buf_len, &remote->addr.from, &remote->addr.to);
960 else
961 ret = make_proxy_line_v1(buf, buf_len, NULL, NULL);
962 }
963
964 return ret;
965}
966
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200967/* Makes a PROXY protocol line from the two addresses. The output is sent to
968 * buffer <buf> for a maximum size of <buf_len> (including the trailing zero).
969 * It returns the number of bytes composing this line (including the trailing
970 * LF), or zero in case of failure (eg: not enough space). It supports TCP4,
Willy Tarreau2e1401a2013-10-01 11:41:55 +0200971 * TCP6 and "UNKNOWN" formats. If any of <src> or <dst> is null, UNKNOWN is
972 * emitted as well.
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200973 */
David Safb76832014-05-08 23:42:08 -0400974int make_proxy_line_v1(char *buf, int buf_len, struct sockaddr_storage *src, struct sockaddr_storage *dst)
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200975{
976 int ret = 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200977 char * protocol;
978 char src_str[MAX(INET_ADDRSTRLEN, INET6_ADDRSTRLEN)];
979 char dst_str[MAX(INET_ADDRSTRLEN, INET6_ADDRSTRLEN)];
980 in_port_t src_port;
981 in_port_t dst_port;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200982
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200983 if ( !src
984 || !dst
985 || (src->ss_family != AF_INET && src->ss_family != AF_INET6)
986 || (dst->ss_family != AF_INET && dst->ss_family != AF_INET6)) {
987 /* unknown family combination */
988 ret = snprintf(buf, buf_len, "PROXY UNKNOWN\r\n");
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200989 if (ret >= buf_len)
990 return 0;
991
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200992 return ret;
993 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200994
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200995 /* IPv4 for both src and dst */
996 if (src->ss_family == AF_INET && dst->ss_family == AF_INET) {
997 protocol = "TCP4";
998 if (!inet_ntop(AF_INET, &((struct sockaddr_in *)src)->sin_addr, src_str, sizeof(src_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200999 return 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001000 src_port = ((struct sockaddr_in *)src)->sin_port;
1001 if (!inet_ntop(AF_INET, &((struct sockaddr_in *)dst)->sin_addr, dst_str, sizeof(dst_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001002 return 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001003 dst_port = ((struct sockaddr_in *)dst)->sin_port;
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001004 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001005 /* IPv6 for at least one of src and dst */
1006 else {
1007 struct in6_addr tmp;
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001008
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001009 protocol = "TCP6";
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001010
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001011 if (src->ss_family == AF_INET) {
1012 /* Convert src to IPv6 */
1013 v4tov6(&tmp, &((struct sockaddr_in *)src)->sin_addr);
1014 src_port = ((struct sockaddr_in *)src)->sin_port;
1015 }
1016 else {
1017 tmp = ((struct sockaddr_in6 *)src)->sin6_addr;
1018 src_port = ((struct sockaddr_in6 *)src)->sin6_port;
1019 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001020
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001021 if (!inet_ntop(AF_INET6, &tmp, src_str, sizeof(src_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001022 return 0;
1023
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001024 if (dst->ss_family == AF_INET) {
1025 /* Convert dst to IPv6 */
1026 v4tov6(&tmp, &((struct sockaddr_in *)dst)->sin_addr);
1027 dst_port = ((struct sockaddr_in *)dst)->sin_port;
1028 }
1029 else {
1030 tmp = ((struct sockaddr_in6 *)dst)->sin6_addr;
1031 dst_port = ((struct sockaddr_in6 *)dst)->sin6_port;
1032 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001033
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001034 if (!inet_ntop(AF_INET6, &tmp, dst_str, sizeof(dst_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001035 return 0;
1036 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001037
1038 ret = snprintf(buf, buf_len, "PROXY %s %s %s %u %u\r\n", protocol, src_str, dst_str, ntohs(src_port), ntohs(dst_port));
1039 if (ret >= buf_len)
1040 return 0;
1041
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001042 return ret;
1043}
David Safb76832014-05-08 23:42:08 -04001044
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001045static int make_tlv(char *dest, int dest_len, char type, uint16_t length, const char *value)
David Safb76832014-05-08 23:42:08 -04001046{
1047 struct tlv *tlv;
1048
1049 if (!dest || (length + sizeof(*tlv) > dest_len))
1050 return 0;
1051
1052 tlv = (struct tlv *)dest;
1053
1054 tlv->type = type;
1055 tlv->length_hi = length >> 8;
1056 tlv->length_lo = length & 0x00ff;
1057 memcpy(tlv->value, value, length);
1058 return length + sizeof(*tlv);
1059}
David Safb76832014-05-08 23:42:08 -04001060
1061int make_proxy_line_v2(char *buf, int buf_len, struct server *srv, struct connection *remote)
1062{
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001063 const char pp2_signature[] = PP2_SIGNATURE;
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001064 void *tlv_crc32c_p = NULL;
David Safb76832014-05-08 23:42:08 -04001065 int ret = 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001066 struct proxy_hdr_v2 *hdr = (struct proxy_hdr_v2 *)buf;
Vincent Bernat6e615892016-05-18 16:17:44 +02001067 struct sockaddr_storage null_addr = { .ss_family = 0 };
David Safb76832014-05-08 23:42:08 -04001068 struct sockaddr_storage *src = &null_addr;
1069 struct sockaddr_storage *dst = &null_addr;
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001070 const char *value;
1071 int value_len;
David Safb76832014-05-08 23:42:08 -04001072
1073 if (buf_len < PP2_HEADER_LEN)
1074 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001075 memcpy(hdr->sig, pp2_signature, PP2_SIGNATURE_LEN);
David Safb76832014-05-08 23:42:08 -04001076
1077 if (remote) {
1078 src = &remote->addr.from;
1079 dst = &remote->addr.to;
1080 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001081
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001082 /* At least one of src or dst is not of AF_INET or AF_INET6 */
1083 if ( !src
1084 || !dst
1085 || (src->ss_family != AF_INET && src->ss_family != AF_INET6)
1086 || (dst->ss_family != AF_INET && dst->ss_family != AF_INET6)) {
David Safb76832014-05-08 23:42:08 -04001087 if (buf_len < PP2_HDR_LEN_UNSPEC)
1088 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001089 hdr->ver_cmd = PP2_VERSION | PP2_CMD_LOCAL;
1090 hdr->fam = PP2_FAM_UNSPEC | PP2_TRANS_UNSPEC;
David Safb76832014-05-08 23:42:08 -04001091 ret = PP2_HDR_LEN_UNSPEC;
1092 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001093 else {
1094 /* IPv4 for both src and dst */
1095 if (src->ss_family == AF_INET && dst->ss_family == AF_INET) {
1096 if (buf_len < PP2_HDR_LEN_INET)
1097 return 0;
1098 hdr->ver_cmd = PP2_VERSION | PP2_CMD_PROXY;
1099 hdr->fam = PP2_FAM_INET | PP2_TRANS_STREAM;
1100 hdr->addr.ip4.src_addr = ((struct sockaddr_in *)src)->sin_addr.s_addr;
1101 hdr->addr.ip4.src_port = ((struct sockaddr_in *)src)->sin_port;
1102 hdr->addr.ip4.dst_addr = ((struct sockaddr_in *)dst)->sin_addr.s_addr;
1103 hdr->addr.ip4.dst_port = ((struct sockaddr_in *)dst)->sin_port;
1104 ret = PP2_HDR_LEN_INET;
1105 }
1106 /* IPv6 for at least one of src and dst */
1107 else {
1108 struct in6_addr tmp;
1109
1110 if (buf_len < PP2_HDR_LEN_INET6)
1111 return 0;
1112 hdr->ver_cmd = PP2_VERSION | PP2_CMD_PROXY;
1113 hdr->fam = PP2_FAM_INET6 | PP2_TRANS_STREAM;
1114 if (src->ss_family == AF_INET) {
1115 v4tov6(&tmp, &((struct sockaddr_in *)src)->sin_addr);
1116 memcpy(hdr->addr.ip6.src_addr, &tmp, 16);
1117 hdr->addr.ip6.src_port = ((struct sockaddr_in *)src)->sin_port;
1118 }
1119 else {
1120 memcpy(hdr->addr.ip6.src_addr, &((struct sockaddr_in6 *)src)->sin6_addr, 16);
1121 hdr->addr.ip6.src_port = ((struct sockaddr_in6 *)src)->sin6_port;
1122 }
1123 if (dst->ss_family == AF_INET) {
1124 v4tov6(&tmp, &((struct sockaddr_in *)dst)->sin_addr);
1125 memcpy(hdr->addr.ip6.dst_addr, &tmp, 16);
1126 hdr->addr.ip6.src_port = ((struct sockaddr_in *)src)->sin_port;
1127 }
1128 else {
1129 memcpy(hdr->addr.ip6.dst_addr, &((struct sockaddr_in6 *)dst)->sin6_addr, 16);
1130 hdr->addr.ip6.dst_port = ((struct sockaddr_in6 *)dst)->sin6_port;
1131 }
1132
1133 ret = PP2_HDR_LEN_INET6;
1134 }
1135 }
David Safb76832014-05-08 23:42:08 -04001136
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001137 if (srv->pp_opts & SRV_PP_V2_CRC32C) {
1138 uint32_t zero_crc32c = 0;
1139 if ((buf_len - ret) < sizeof(struct tlv))
1140 return 0;
1141 tlv_crc32c_p = (void *)((struct tlv *)&buf[ret])->value;
1142 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_CRC32C, sizeof(zero_crc32c), (const char *)&zero_crc32c);
1143 }
1144
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001145 if (conn_get_alpn(remote, &value, &value_len)) {
1146 if ((buf_len - ret) < sizeof(struct tlv))
1147 return 0;
Emmanuel Hocdet571c7ac2017-10-31 18:24:05 +01001148 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_ALPN, value_len, value);
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001149 }
1150
David Safb76832014-05-08 23:42:08 -04001151#ifdef USE_OPENSSL
Emmanuel Hocdet253c3b72018-02-01 18:29:59 +01001152 if (srv->pp_opts & SRV_PP_V2_AUTHORITY) {
1153 value = ssl_sock_get_sni(remote);
1154 if (value) {
1155 if ((buf_len - ret) < sizeof(struct tlv))
1156 return 0;
1157 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_AUTHORITY, strlen(value), value);
1158 }
1159 }
1160
David Safb76832014-05-08 23:42:08 -04001161 if (srv->pp_opts & SRV_PP_V2_SSL) {
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001162 struct tlv_ssl *tlv;
1163 int ssl_tlv_len = 0;
David Safb76832014-05-08 23:42:08 -04001164 if ((buf_len - ret) < sizeof(struct tlv_ssl))
1165 return 0;
1166 tlv = (struct tlv_ssl *)&buf[ret];
1167 memset(tlv, 0, sizeof(struct tlv_ssl));
1168 ssl_tlv_len += sizeof(struct tlv_ssl);
1169 tlv->tlv.type = PP2_TYPE_SSL;
1170 if (ssl_sock_is_ssl(remote)) {
1171 tlv->client |= PP2_CLIENT_SSL;
Emmanuel Hocdet01da5712017-10-13 16:59:49 +02001172 value = ssl_sock_get_proto_version(remote);
David Safb76832014-05-08 23:42:08 -04001173 if (value) {
Emmanuel Hocdet8c0c34b2018-02-28 12:02:14 +01001174 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len-ret-ssl_tlv_len), PP2_SUBTYPE_SSL_VERSION, strlen(value), value);
David Safb76832014-05-08 23:42:08 -04001175 }
Dave McCowan328fb582014-07-30 10:39:13 -04001176 if (ssl_sock_get_cert_used_sess(remote)) {
1177 tlv->client |= PP2_CLIENT_CERT_SESS;
David Safb76832014-05-08 23:42:08 -04001178 tlv->verify = htonl(ssl_sock_get_verify_result(remote));
Dave McCowan328fb582014-07-30 10:39:13 -04001179 if (ssl_sock_get_cert_used_conn(remote))
1180 tlv->client |= PP2_CLIENT_CERT_CONN;
David Safb76832014-05-08 23:42:08 -04001181 }
1182 if (srv->pp_opts & SRV_PP_V2_SSL_CN) {
Willy Tarreau83061a82018-07-13 11:56:34 +02001183 struct buffer *cn_trash = get_trash_chunk();
Willy Tarreau3b9a0c92014-07-19 06:37:33 +02001184 if (ssl_sock_get_remote_common_name(remote, cn_trash) > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001185 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_CN,
1186 cn_trash->data,
1187 cn_trash->area);
David Safb76832014-05-08 23:42:08 -04001188 }
1189 }
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001190 if (srv->pp_opts & SRV_PP_V2_SSL_KEY_ALG) {
Willy Tarreau83061a82018-07-13 11:56:34 +02001191 struct buffer *pkey_trash = get_trash_chunk();
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001192 if (ssl_sock_get_pkey_algo(remote, pkey_trash) > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001193 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_KEY_ALG,
1194 pkey_trash->data,
1195 pkey_trash->area);
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001196 }
1197 }
1198 if (srv->pp_opts & SRV_PP_V2_SSL_SIG_ALG) {
1199 value = ssl_sock_get_cert_sig(remote);
1200 if (value) {
1201 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_SIG_ALG, strlen(value), value);
1202 }
1203 }
1204 if (srv->pp_opts & SRV_PP_V2_SSL_CIPHER) {
1205 value = ssl_sock_get_cipher_name(remote);
1206 if (value) {
1207 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_CIPHER, strlen(value), value);
1208 }
1209 }
David Safb76832014-05-08 23:42:08 -04001210 }
1211 tlv->tlv.length_hi = (uint16_t)(ssl_tlv_len - sizeof(struct tlv)) >> 8;
1212 tlv->tlv.length_lo = (uint16_t)(ssl_tlv_len - sizeof(struct tlv)) & 0x00ff;
1213 ret += ssl_tlv_len;
1214 }
1215#endif
1216
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001217#ifdef CONFIG_HAP_NS
1218 if (remote && (remote->proxy_netns)) {
1219 if ((buf_len - ret) < sizeof(struct tlv))
1220 return 0;
Emmanuel Hocdet571c7ac2017-10-31 18:24:05 +01001221 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_NETNS, remote->proxy_netns->name_len, remote->proxy_netns->node.key);
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001222 }
1223#endif
1224
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001225 hdr->len = htons((uint16_t)(ret - PP2_HEADER_LEN));
David Safb76832014-05-08 23:42:08 -04001226
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001227 if (tlv_crc32c_p) {
1228 write_u32(tlv_crc32c_p, htonl(hash_crc32c(buf, ret)));
1229 }
1230
David Safb76832014-05-08 23:42:08 -04001231 return ret;
1232}
Emeric Brun4f603012017-01-05 15:11:44 +01001233
Willy Tarreau60ca10a2017-08-18 15:26:54 +02001234/* return the major HTTP version as 1 or 2 depending on how the request arrived
1235 * before being processed.
1236 */
1237static int
1238smp_fetch_fc_http_major(const struct arg *args, struct sample *smp, const char *kw, void *private)
1239{
1240 struct connection *conn = objt_conn(smp->sess->origin);
1241
1242 smp->data.type = SMP_T_SINT;
1243 smp->data.u.sint = (conn && strcmp(conn_get_mux_name(conn), "H2") == 0) ? 2 : 1;
1244 return 1;
1245}
1246
Emeric Brun4f603012017-01-05 15:11:44 +01001247/* fetch if the received connection used a PROXY protocol header */
1248int smp_fetch_fc_rcvd_proxy(const struct arg *args, struct sample *smp, const char *kw, void *private)
1249{
1250 struct connection *conn;
1251
1252 conn = objt_conn(smp->sess->origin);
1253 if (!conn)
1254 return 0;
1255
1256 if (!(conn->flags & CO_FL_CONNECTED)) {
1257 smp->flags |= SMP_F_MAY_CHANGE;
1258 return 0;
1259 }
1260
1261 smp->flags = 0;
1262 smp->data.type = SMP_T_BOOL;
1263 smp->data.u.sint = (conn->flags & CO_FL_RCVD_PROXY) ? 1 : 0;
1264
1265 return 1;
1266}
1267
1268/* Note: must not be declared <const> as its list will be overwritten.
1269 * Note: fetches that may return multiple types must be declared as the lowest
1270 * common denominator, the type that can be casted into all other ones. For
1271 * instance v4/v6 must be declared v4.
1272 */
1273static struct sample_fetch_kw_list sample_fetch_keywords = {ILH, {
Willy Tarreau60ca10a2017-08-18 15:26:54 +02001274 { "fc_http_major", smp_fetch_fc_http_major, 0, NULL, SMP_T_SINT, SMP_USE_L4CLI },
Emeric Brun4f603012017-01-05 15:11:44 +01001275 { "fc_rcvd_proxy", smp_fetch_fc_rcvd_proxy, 0, NULL, SMP_T_BOOL, SMP_USE_L4CLI },
1276 { /* END */ },
1277}};
1278
1279
1280__attribute__((constructor))
1281static void __connection_init(void)
1282{
1283 sample_register_fetches(&sample_fetch_keywords);
1284}