blob: 8826706f901982e880f0c2482b8b10981c8111fe [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
Willy Tarreau2386be62017-09-21 19:40:52 +020036/* List head of all known muxes for ALPN */
37struct alpn_mux_list alpn_mux_list = {
38 .list = LIST_HEAD_INIT(alpn_mux_list.list)
39};
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);
137 tasklet_wakeup(sw->task);
138 }
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200139 }
Willy Tarreau59f98392012-07-06 14:13:49 +0200140
Willy Tarreau57ec32f2017-04-11 19:59:33 +0200141 /* The data transfer starts here and stops on error and handshakes. Note
142 * that we must absolutely test conn->xprt at each step in case it suddenly
143 * changes due to a quick unexpected close().
144 */
145 if (conn->xprt && fd_recv_ready(fd) &&
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200146 ((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 +0100147 /* force reporting of activity by clearing the previous flags :
148 * we'll have at least ERROR or CONNECTED at the end of an I/O,
149 * both of which will be detected below.
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200150 */
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100151 flags = 0;
Willy Tarreau53a47662017-08-28 10:53:00 +0200152 conn->mux->recv(conn);
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200153 }
Willy Tarreau2da156f2012-07-23 15:07:23 +0200154
Willy Tarreauc76ae332012-07-12 15:32:13 +0200155 /* It may happen during the data phase that a handshake is
156 * enabled again (eg: SSL)
157 */
Willy Tarreaud6e999b2013-11-25 08:41:15 +0100158 if (unlikely(conn->flags & (CO_FL_HANDSHAKE | CO_FL_ERROR)))
Willy Tarreauc76ae332012-07-12 15:32:13 +0200159 goto process_handshake;
160
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100161 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreauf8deb0c2012-09-01 17:59:22 +0200162 /* still waiting for a connection to establish and nothing was
163 * attempted yet to probe the connection. Then let's retry the
164 * connect().
Willy Tarreau2da156f2012-07-23 15:07:23 +0200165 */
Willy Tarreau239d7182012-07-23 18:53:03 +0200166 if (!tcp_connect_probe(conn))
Willy Tarreauafad0e02012-08-09 14:45:22 +0200167 goto leave;
Willy Tarreau2da156f2012-07-23 15:07:23 +0200168 }
Willy Tarreau2c6be842012-07-06 17:12:34 +0200169 leave:
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100170 /* Verify if the connection just established. */
Willy Tarreau7bf3fa32017-03-14 20:19:29 +0100171 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED))))
172 conn->flags |= CO_FL_CONNECTED;
173
Willy Tarreau8e3c6ce2017-08-28 15:46:01 +0200174 /* The connection owner might want to be notified about failures to
175 * complete the handshake. The callback may fail and cause the
176 * connection to be destroyed, thus we must not use it anymore and
177 * should immediately leave instead. The caller must immediately
178 * unregister itself once called.
179 */
180 if (((conn->flags ^ flags) & CO_FL_NOTIFY_DONE) &&
181 conn->xprt_done_cb && conn->xprt_done_cb(conn) < 0)
182 return;
183
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100184 /* The wake callback is normally used to notify the data layer about
185 * data layer activity (successful send/recv), connection establishment,
186 * shutdown and fatal errors. We need to consider the following
187 * situations to wake up the data layer :
188 * - change among the CO_FL_NOTIFY_DATA flags :
189 * {DATA,SOCK}_{RD,WR}_SH, ERROR,
190 * - absence of any of {L4,L6}_CONN and CONNECTED, indicating the
191 * end of handshake and transition to CONNECTED
192 * - raise of CONNECTED with HANDSHAKE down
193 * - end of HANDSHAKE with CONNECTED set
194 * - regular data layer activity
195 *
196 * Note that the wake callback is allowed to release the connection and
197 * the fd (and return < 0 in this case).
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200198 */
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200199 if ((can_send || (((conn->flags ^ flags) & CO_FL_NOTIFY_DATA) ||
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100200 ((flags & (CO_FL_CONNECTED|CO_FL_HANDSHAKE)) != CO_FL_CONNECTED &&
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200201 (conn->flags & (CO_FL_CONNECTED|CO_FL_HANDSHAKE)) == CO_FL_CONNECTED))) &&
Willy Tarreau53a47662017-08-28 10:53:00 +0200202 conn->mux->wake(conn) < 0)
Willy Tarreau7a798e52016-04-14 11:13:20 +0200203 return;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200204
Willy Tarreau61ace1b2012-07-23 12:14:26 +0200205 /* remove the events before leaving */
Willy Tarreau26d7cfc2012-12-07 00:09:43 +0100206 fdtab[fd].ev &= FD_POLL_STICKY;
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200207
208 /* commit polling changes */
Willy Tarreau916e12d2017-10-25 09:22:43 +0200209 conn->flags &= ~CO_FL_WILL_UPDATE;
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200210 conn_cond_update_polling(conn);
Willy Tarreau7a798e52016-04-14 11:13:20 +0200211 return;
Willy Tarreau59f98392012-07-06 14:13:49 +0200212}
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200213
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200214/* Update polling on connection <c>'s file descriptor depending on its current
215 * state as reported in the connection's CO_FL_CURR_* flags, reports of EAGAIN
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200216 * in CO_FL_WAIT_*, and the data layer expectations indicated by CO_FL_XPRT_*.
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200217 * The connection flags are updated with the new flags at the end of the
Willy Tarreau0ffde2c2012-10-04 22:21:15 +0200218 * operation. Polling is totally disabled if an error was reported.
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200219 */
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200220void conn_update_xprt_polling(struct connection *c)
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200221{
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200222 unsigned int f = c->flags;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200223
Willy Tarreau3c728722014-01-23 13:50:42 +0100224 if (!conn_ctrl_ready(c))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200225 return;
226
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200227 /* update read status if needed */
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200228 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 +0200229 fd_want_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100230 f |= CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200231 }
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200232 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 +0200233 fd_stop_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100234 f &= ~CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200235 }
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200236
237 /* update write status if needed */
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200238 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 +0200239 fd_want_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100240 f |= CO_FL_CURR_WR_ENA;
241 }
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200242 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 +0200243 fd_stop_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100244 f &= ~CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200245 }
Willy Tarreau310987a2014-01-22 19:46:33 +0100246 c->flags = f;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200247}
248
249/* Update polling on connection <c>'s file descriptor depending on its current
250 * state as reported in the connection's CO_FL_CURR_* flags, reports of EAGAIN
251 * in CO_FL_WAIT_*, and the sock layer expectations indicated by CO_FL_SOCK_*.
252 * The connection flags are updated with the new flags at the end of the
Willy Tarreau0ffde2c2012-10-04 22:21:15 +0200253 * operation. Polling is totally disabled if an error was reported.
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200254 */
255void conn_update_sock_polling(struct connection *c)
256{
257 unsigned int f = c->flags;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200258
Willy Tarreau3c728722014-01-23 13:50:42 +0100259 if (!conn_ctrl_ready(c))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200260 return;
261
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200262 /* update read status if needed */
Willy Tarreau310987a2014-01-22 19:46:33 +0100263 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 +0200264 fd_want_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100265 f |= CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200266 }
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100267 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 +0200268 fd_stop_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100269 f &= ~CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200270 }
271
272 /* update write status if needed */
Willy Tarreau310987a2014-01-22 19:46:33 +0100273 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 +0200274 fd_want_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100275 f |= CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200276 }
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100277 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 +0200278 fd_stop_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100279 f &= ~CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200280 }
Willy Tarreau310987a2014-01-22 19:46:33 +0100281 c->flags = f;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200282}
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200283
Willy Tarreauff3e6482015-03-12 23:56:52 +0100284/* Send a message over an established connection. It makes use of send() and
285 * returns the same return code and errno. If the socket layer is not ready yet
286 * then -1 is returned and ENOTSOCK is set into errno. If the fd is not marked
287 * as ready, or if EAGAIN or ENOTCONN is returned, then we return 0. It returns
288 * EMSGSIZE if called with a zero length message. The purpose is to simplify
289 * some rare attempts to directly write on the socket from above the connection
290 * (typically send_proxy). In case of EAGAIN, the fd is marked as "cant_send".
291 * It automatically retries on EINTR. Other errors cause the connection to be
292 * marked as in error state. It takes similar arguments as send() except the
293 * first one which is the connection instead of the file descriptor. Note,
294 * MSG_DONTWAIT and MSG_NOSIGNAL are forced on the flags.
295 */
296int conn_sock_send(struct connection *conn, const void *buf, int len, int flags)
297{
298 int ret;
299
300 ret = -1;
301 errno = ENOTSOCK;
302
303 if (conn->flags & CO_FL_SOCK_WR_SH)
304 goto fail;
305
306 if (!conn_ctrl_ready(conn))
307 goto fail;
308
309 errno = EMSGSIZE;
310 if (!len)
311 goto fail;
312
Willy Tarreau585744b2017-08-24 14:31:19 +0200313 if (!fd_send_ready(conn->handle.fd))
Willy Tarreauff3e6482015-03-12 23:56:52 +0100314 goto wait;
315
316 do {
Willy Tarreau585744b2017-08-24 14:31:19 +0200317 ret = send(conn->handle.fd, buf, len, flags | MSG_DONTWAIT | MSG_NOSIGNAL);
Willy Tarreauff3e6482015-03-12 23:56:52 +0100318 } while (ret < 0 && errno == EINTR);
319
320
321 if (ret > 0)
322 return ret;
323
324 if (ret == 0 || errno == EAGAIN || errno == ENOTCONN) {
325 wait:
Willy Tarreau585744b2017-08-24 14:31:19 +0200326 fd_cant_send(conn->handle.fd);
Willy Tarreauff3e6482015-03-12 23:56:52 +0100327 return 0;
328 }
329 fail:
330 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH | CO_FL_ERROR;
331 return ret;
332}
333
Olivier Houchard6ff20392018-07-17 18:46:31 +0200334int conn_subscribe(struct connection *conn, int event_type, void *param)
335{
336 struct wait_list *sw;
337
338 switch (event_type) {
339 case SUB_CAN_SEND:
340 sw = param;
341 if (LIST_ISEMPTY(&sw->list))
342 LIST_ADDQ(&conn->send_wait_list, &sw->list);
343 return 0;
344 default:
345 break;
346 }
347 return (-1);
348}
349
Willy Tarreaud85c4852015-03-13 00:40:28 +0100350/* Drains possibly pending incoming data on the file descriptor attached to the
351 * connection and update the connection's flags accordingly. This is used to
352 * know whether we need to disable lingering on close. Returns non-zero if it
353 * is safe to close without disabling lingering, otherwise zero. The SOCK_RD_SH
354 * flag may also be updated if the incoming shutdown was reported by the drain()
355 * function.
356 */
357int conn_sock_drain(struct connection *conn)
358{
359 if (!conn_ctrl_ready(conn))
360 return 1;
361
362 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH))
363 return 1;
364
Willy Tarreau585744b2017-08-24 14:31:19 +0200365 if (fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP)) {
366 fdtab[conn->handle.fd].linger_risk = 0;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100367 }
368 else {
Willy Tarreau585744b2017-08-24 14:31:19 +0200369 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaud85c4852015-03-13 00:40:28 +0100370 return 0;
371
372 /* disable draining if we were called and have no drain function */
373 if (!conn->ctrl->drain) {
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200374 __conn_xprt_stop_recv(conn);
Willy Tarreaud85c4852015-03-13 00:40:28 +0100375 return 0;
376 }
377
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;
380 }
381
382 conn->flags |= CO_FL_SOCK_RD_SH;
383 return 1;
384}
385
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100386/*
Willy Tarreau11c9aa42018-03-02 13:55:01 +0100387 * default conn_stream recv() : this one is used when cs->rcv_buf == NULL.
388 * It reads up to <count> bytes from cs->rxbuf, puts them into <buf> and
389 * returns the count. It possibly sets/clears CS_FL_RCV_MORE depending on the
390 * buffer's state, and may set CS_FL_EOS. The number of bytes transferred is
391 * returned. <buf> is not touched if <count> is null, but cs flags will be
392 * updated to indicate any RCV_MORE or EOS.
393 */
394size_t __cs_recv(struct conn_stream *cs, struct buffer *buf, size_t count, int flags)
395{
396 size_t ret = 0;
397
398 /* transfer possibly pending data to the upper layer */
399 ret = b_xfer(buf, &cs->rxbuf, count);
400
401 if (b_data(&cs->rxbuf))
402 cs->flags |= CS_FL_RCV_MORE;
403 else {
404 cs->flags &= ~CS_FL_RCV_MORE;
405 if (cs->flags & CS_FL_REOS)
406 cs->flags |= CS_FL_EOS;
407 cs_drop_rxbuf(cs);
408 }
409
410 return ret;
411}
412
413/*
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100414 * Get data length from tlv
415 */
416static int get_tlv_length(const struct tlv *src)
417{
418 return (src->length_hi << 8) | src->length_lo;
419}
420
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200421/* This handshake handler waits a PROXY protocol header at the beginning of the
422 * raw data stream. The header looks like this :
423 *
424 * "PROXY" <SP> PROTO <SP> SRC3 <SP> DST3 <SP> SRC4 <SP> <DST4> "\r\n"
425 *
426 * There must be exactly one space between each field. Fields are :
427 * - PROTO : layer 4 protocol, which must be "TCP4" or "TCP6".
428 * - SRC3 : layer 3 (eg: IP) source address in standard text form
429 * - DST3 : layer 3 (eg: IP) destination address in standard text form
430 * - SRC4 : layer 4 (eg: TCP port) source address in standard text form
431 * - DST4 : layer 4 (eg: TCP port) destination address in standard text form
432 *
433 * This line MUST be at the beginning of the buffer and MUST NOT wrap.
434 *
435 * The header line is small and in all cases smaller than the smallest normal
436 * TCP MSS. So it MUST always be delivered as one segment, which ensures we
437 * can safely use MSG_PEEK and avoid buffering.
438 *
439 * Once the data is fetched, the values are set in the connection's address
440 * fields, and data are removed from the socket's buffer. The function returns
441 * zero if it needs to wait for more data or if it fails, or 1 if it completed
442 * and removed itself.
443 */
444int conn_recv_proxy(struct connection *conn, int flag)
445{
446 char *line, *end;
Willy Tarreau77992672014-06-14 11:06:17 +0200447 struct proxy_hdr_v2 *hdr_v2;
448 const char v2sig[] = PP2_SIGNATURE;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100449 int tlv_length = 0;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200450 int tlv_offset = 0;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200451
452 /* we might have been called just after an asynchronous shutr */
453 if (conn->flags & CO_FL_SOCK_RD_SH)
454 goto fail;
455
Willy Tarreau3c728722014-01-23 13:50:42 +0100456 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200457 goto fail;
458
Willy Tarreau585744b2017-08-24 14:31:19 +0200459 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100460 return 0;
461
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200462 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200463 trash.data = recv(conn->handle.fd, trash.area, trash.size,
464 MSG_PEEK);
465 if (trash.data < 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200466 if (errno == EINTR)
467 continue;
468 if (errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200469 fd_cant_recv(conn->handle.fd);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200470 return 0;
471 }
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100472 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200473 }
474 } while (0);
475
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200476 if (!trash.data) {
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100477 /* client shutdown */
478 conn->err_code = CO_ER_PRX_EMPTY;
479 goto fail;
480 }
481
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200482 if (trash.data < 6)
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200483 goto missing;
484
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200485 line = trash.area;
486 end = trash.area + trash.data;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200487
488 /* Decode a possible proxy request, fail early if it does not match */
Willy Tarreau77992672014-06-14 11:06:17 +0200489 if (strncmp(line, "PROXY ", 6) != 0)
490 goto not_v1;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200491
492 line += 6;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200493 if (trash.data < 9) /* shortest possible line */
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200494 goto missing;
495
David CARLIER42ff05e2016-03-24 09:22:36 +0000496 if (memcmp(line, "TCP4 ", 5) == 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200497 u32 src3, dst3, sport, dport;
498
499 line += 5;
500
501 src3 = inetaddr_host_lim_ret(line, end, &line);
502 if (line == end)
503 goto missing;
504 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100505 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200506
507 dst3 = inetaddr_host_lim_ret(line, end, &line);
508 if (line == end)
509 goto missing;
510 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100511 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200512
513 sport = read_uint((const char **)&line, end);
514 if (line == end)
515 goto missing;
516 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100517 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200518
519 dport = read_uint((const char **)&line, end);
520 if (line > end - 2)
521 goto missing;
522 if (*line++ != '\r')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100523 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200524 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100525 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200526
527 /* update the session's addresses and mark them set */
528 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
529 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = htonl(src3);
530 ((struct sockaddr_in *)&conn->addr.from)->sin_port = htons(sport);
531
532 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
533 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = htonl(dst3);
534 ((struct sockaddr_in *)&conn->addr.to)->sin_port = htons(dport);
535 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
536 }
David CARLIER42ff05e2016-03-24 09:22:36 +0000537 else if (memcmp(line, "TCP6 ", 5) == 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200538 u32 sport, dport;
539 char *src_s;
540 char *dst_s, *sport_s, *dport_s;
541 struct in6_addr src3, dst3;
542
543 line += 5;
544
545 src_s = line;
546 dst_s = sport_s = dport_s = NULL;
547 while (1) {
548 if (line > end - 2) {
549 goto missing;
550 }
551 else if (*line == '\r') {
552 *line = 0;
553 line++;
554 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100555 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200556 break;
557 }
558
559 if (*line == ' ') {
560 *line = 0;
561 if (!dst_s)
562 dst_s = line + 1;
563 else if (!sport_s)
564 sport_s = line + 1;
565 else if (!dport_s)
566 dport_s = line + 1;
567 }
568 line++;
569 }
570
571 if (!dst_s || !sport_s || !dport_s)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100572 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200573
574 sport = read_uint((const char **)&sport_s,dport_s - 1);
575 if (*sport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100576 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200577
578 dport = read_uint((const char **)&dport_s,line - 2);
579 if (*dport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100580 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200581
582 if (inet_pton(AF_INET6, src_s, (void *)&src3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100583 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200584
585 if (inet_pton(AF_INET6, dst_s, (void *)&dst3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100586 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200587
588 /* update the session's addresses and mark them set */
589 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
590 memcpy(&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr, &src3, sizeof(struct in6_addr));
591 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = htons(sport);
592
593 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
594 memcpy(&((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr, &dst3, sizeof(struct in6_addr));
595 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = htons(dport);
596 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
597 }
Willy Tarreau4c20d292014-06-14 11:41:36 +0200598 else if (memcmp(line, "UNKNOWN\r\n", 9) == 0) {
599 /* This can be a UNIX socket forwarded by an haproxy upstream */
600 line += 9;
601 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200602 else {
Willy Tarreau4c20d292014-06-14 11:41:36 +0200603 /* The protocol does not match something known (TCP4/TCP6/UNKNOWN) */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100604 conn->err_code = CO_ER_PRX_BAD_PROTO;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200605 goto fail;
606 }
607
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200608 trash.data = line - trash.area;
Willy Tarreau77992672014-06-14 11:06:17 +0200609 goto eat_header;
610
611 not_v1:
612 /* try PPv2 */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200613 if (trash.data < PP2_HEADER_LEN)
Willy Tarreau77992672014-06-14 11:06:17 +0200614 goto missing;
615
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200616 hdr_v2 = (struct proxy_hdr_v2 *) trash.area;
Willy Tarreau77992672014-06-14 11:06:17 +0200617
618 if (memcmp(hdr_v2->sig, v2sig, PP2_SIGNATURE_LEN) != 0 ||
619 (hdr_v2->ver_cmd & PP2_VERSION_MASK) != PP2_VERSION) {
620 conn->err_code = CO_ER_PRX_NOT_HDR;
621 goto fail;
622 }
623
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200624 if (trash.data < PP2_HEADER_LEN + ntohs(hdr_v2->len))
Willy Tarreau77992672014-06-14 11:06:17 +0200625 goto missing;
626
627 switch (hdr_v2->ver_cmd & PP2_CMD_MASK) {
628 case 0x01: /* PROXY command */
629 switch (hdr_v2->fam) {
630 case 0x11: /* TCPv4 */
KOVACS Krisztianefd3aa92014-11-19 10:53:20 +0100631 if (ntohs(hdr_v2->len) < PP2_ADDR_LEN_INET)
632 goto bad_header;
633
Willy Tarreau77992672014-06-14 11:06:17 +0200634 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
635 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = hdr_v2->addr.ip4.src_addr;
636 ((struct sockaddr_in *)&conn->addr.from)->sin_port = hdr_v2->addr.ip4.src_port;
637 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
638 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = hdr_v2->addr.ip4.dst_addr;
639 ((struct sockaddr_in *)&conn->addr.to)->sin_port = hdr_v2->addr.ip4.dst_port;
640 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200641 tlv_offset = PP2_HEADER_LEN + PP2_ADDR_LEN_INET;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100642 tlv_length = ntohs(hdr_v2->len) - PP2_ADDR_LEN_INET;
Willy Tarreau77992672014-06-14 11:06:17 +0200643 break;
644 case 0x21: /* TCPv6 */
KOVACS Krisztianefd3aa92014-11-19 10:53:20 +0100645 if (ntohs(hdr_v2->len) < PP2_ADDR_LEN_INET6)
646 goto bad_header;
647
Willy Tarreau77992672014-06-14 11:06:17 +0200648 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
649 memcpy(&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr, hdr_v2->addr.ip6.src_addr, 16);
650 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = hdr_v2->addr.ip6.src_port;
651 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
652 memcpy(&((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr, hdr_v2->addr.ip6.dst_addr, 16);
653 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = hdr_v2->addr.ip6.dst_port;
654 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200655 tlv_offset = PP2_HEADER_LEN + PP2_ADDR_LEN_INET6;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100656 tlv_length = ntohs(hdr_v2->len) - PP2_ADDR_LEN_INET6;
Willy Tarreau77992672014-06-14 11:06:17 +0200657 break;
658 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100659
660 /* TLV parsing */
661 if (tlv_length > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200662 while (tlv_offset + TLV_HEADER_SIZE <= trash.data) {
663 const struct tlv *tlv_packet = (struct tlv *) &trash.area[tlv_offset];
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100664 const int tlv_len = get_tlv_length(tlv_packet);
665 tlv_offset += tlv_len + TLV_HEADER_SIZE;
666
667 switch (tlv_packet->type) {
Emmanuel Hocdet115df3e2018-02-05 16:23:23 +0100668 case PP2_TYPE_CRC32C: {
669 void *tlv_crc32c_p = (void *)tlv_packet->value;
670 uint32_t n_crc32c = ntohl(read_u32(tlv_crc32c_p));
671 write_u32(tlv_crc32c_p, 0);
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200672 if (hash_crc32c(trash.area, PP2_HEADER_LEN + ntohs(hdr_v2->len)) != n_crc32c)
Emmanuel Hocdet115df3e2018-02-05 16:23:23 +0100673 goto bad_header;
674 break;
675 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100676#ifdef CONFIG_HAP_NS
677 case PP2_TYPE_NETNS: {
678 const struct netns_entry *ns;
679 ns = netns_store_lookup((char*)tlv_packet->value, tlv_len);
680 if (ns)
681 conn->proxy_netns = ns;
682 break;
683 }
684#endif
685 default:
686 break;
687 }
688 }
689 }
690
Willy Tarreau77992672014-06-14 11:06:17 +0200691 /* unsupported protocol, keep local connection address */
692 break;
693 case 0x00: /* LOCAL command */
694 /* keep local connection address for LOCAL */
695 break;
696 default:
697 goto bad_header; /* not a supported command */
698 }
699
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200700 trash.data = PP2_HEADER_LEN + ntohs(hdr_v2->len);
Willy Tarreau77992672014-06-14 11:06:17 +0200701 goto eat_header;
702
703 eat_header:
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200704 /* remove the PROXY line from the request. For this we re-read the
705 * exact line at once. If we don't get the exact same result, we
706 * fail.
707 */
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200708 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200709 int len2 = recv(conn->handle.fd, trash.area, trash.data, 0);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200710 if (len2 < 0 && errno == EINTR)
711 continue;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200712 if (len2 != trash.data)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100713 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200714 } while (0);
715
716 conn->flags &= ~flag;
Emeric Brun4f603012017-01-05 15:11:44 +0100717 conn->flags |= CO_FL_RCVD_PROXY;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200718 return 1;
719
720 missing:
721 /* Missing data. Since we're using MSG_PEEK, we can only poll again if
722 * we have not read anything. Otherwise we need to fail because we won't
723 * be able to poll anymore.
724 */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100725 conn->err_code = CO_ER_PRX_TRUNCATED;
726 goto fail;
727
728 bad_header:
729 /* This is not a valid proxy protocol header */
730 conn->err_code = CO_ER_PRX_BAD_HDR;
731 goto fail;
732
733 recv_abort:
734 conn->err_code = CO_ER_PRX_ABORT;
Willy Tarreau26f4a042013-12-04 23:44:10 +0100735 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100736 goto fail;
737
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200738 fail:
Willy Tarreaud486ef52012-12-10 17:03:52 +0100739 __conn_sock_stop_both(conn);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200740 conn->flags |= CO_FL_ERROR;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200741 return 0;
742}
743
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100744/* This handshake handler waits a NetScaler Client IP insertion header
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000745 * at the beginning of the raw data stream. The header format is
746 * described in doc/netscaler-client-ip-insertion-protocol.txt
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100747 *
748 * This line MUST be at the beginning of the buffer and MUST NOT be
749 * fragmented.
750 *
751 * The header line is small and in all cases smaller than the smallest normal
752 * TCP MSS. So it MUST always be delivered as one segment, which ensures we
753 * can safely use MSG_PEEK and avoid buffering.
754 *
755 * Once the data is fetched, the values are set in the connection's address
756 * fields, and data are removed from the socket's buffer. The function returns
757 * zero if it needs to wait for more data or if it fails, or 1 if it completed
758 * and removed itself.
759 */
760int conn_recv_netscaler_cip(struct connection *conn, int flag)
761{
762 char *line;
Bertrand Jacquin7d668f92017-12-13 01:23:39 +0000763 uint32_t hdr_len;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100764 uint8_t ip_v;
765
766 /* we might have been called just after an asynchronous shutr */
767 if (conn->flags & CO_FL_SOCK_RD_SH)
768 goto fail;
769
770 if (!conn_ctrl_ready(conn))
771 goto fail;
772
Willy Tarreau585744b2017-08-24 14:31:19 +0200773 if (!fd_recv_ready(conn->handle.fd))
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100774 return 0;
775
776 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200777 trash.data = recv(conn->handle.fd, trash.area, trash.size,
778 MSG_PEEK);
779 if (trash.data < 0) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100780 if (errno == EINTR)
781 continue;
782 if (errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200783 fd_cant_recv(conn->handle.fd);
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100784 return 0;
785 }
786 goto recv_abort;
787 }
788 } while (0);
789
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200790 if (!trash.data) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100791 /* client shutdown */
792 conn->err_code = CO_ER_CIP_EMPTY;
793 goto fail;
794 }
795
796 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000797 * CIP magic, header length or
798 * CIP magic, CIP length, CIP type, header length */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200799 if (trash.data < 12)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100800 goto missing;
801
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200802 line = trash.area;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100803
804 /* Decode a possible NetScaler Client IP request, fail early if
805 * it does not match */
Bertrand Jacquin4b4c2862017-12-13 01:07:12 +0000806 if (ntohl(*(uint32_t *)line) != objt_listener(conn->target)->bind_conf->ns_cip_magic)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100807 goto bad_magic;
808
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000809 /* Legacy CIP protocol */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200810 if ((trash.area[8] & 0xD0) == 0x40) {
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000811 hdr_len = ntohl(*(uint32_t *)(line+4));
812 line += 8;
813 }
814 /* Standard CIP protocol */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200815 else if (trash.area[8] == 0x00) {
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000816 hdr_len = ntohs(*(uint32_t *)(line+10));
817 line += 12;
818 }
819 /* Unknown CIP protocol */
820 else {
821 conn->err_code = CO_ER_CIP_BAD_PROTO;
822 goto fail;
823 }
824
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100825 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000826 * a minimal IP header */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200827 if (trash.data < 20)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100828 goto missing;
829
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100830 /* Get IP version from the first four bits */
831 ip_v = (*line & 0xf0) >> 4;
832
833 if (ip_v == 4) {
834 struct ip *hdr_ip4;
David Carlier3015a2e2016-07-04 22:51:33 +0100835 struct my_tcphdr *hdr_tcp;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100836
837 hdr_ip4 = (struct ip *)line;
838
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200839 if (trash.data < 40 || trash.data < hdr_len) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100840 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin67de5a22017-12-13 01:15:05 +0000841 * IPv4 header, TCP header */
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100842 goto missing;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000843 }
844 else if (hdr_ip4->ip_p != IPPROTO_TCP) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100845 /* The protocol does not include a TCP header */
846 conn->err_code = CO_ER_CIP_BAD_PROTO;
847 goto fail;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000848 }
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100849
David Carlier3015a2e2016-07-04 22:51:33 +0100850 hdr_tcp = (struct my_tcphdr *)(line + (hdr_ip4->ip_hl * 4));
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100851
852 /* update the session's addresses and mark them set */
853 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
854 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = hdr_ip4->ip_src.s_addr;
855 ((struct sockaddr_in *)&conn->addr.from)->sin_port = hdr_tcp->source;
856
857 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
858 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = hdr_ip4->ip_dst.s_addr;
859 ((struct sockaddr_in *)&conn->addr.to)->sin_port = hdr_tcp->dest;
860
861 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
862 }
863 else if (ip_v == 6) {
864 struct ip6_hdr *hdr_ip6;
David Carlier3015a2e2016-07-04 22:51:33 +0100865 struct my_tcphdr *hdr_tcp;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100866
867 hdr_ip6 = (struct ip6_hdr *)line;
868
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200869 if (trash.data < 60 || trash.data < hdr_len) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100870 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin67de5a22017-12-13 01:15:05 +0000871 * IPv6 header, TCP header */
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100872 goto missing;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000873 }
874 else if (hdr_ip6->ip6_nxt != IPPROTO_TCP) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100875 /* The protocol does not include a TCP header */
876 conn->err_code = CO_ER_CIP_BAD_PROTO;
877 goto fail;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000878 }
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100879
David Carlier3015a2e2016-07-04 22:51:33 +0100880 hdr_tcp = (struct my_tcphdr *)(line + sizeof(struct ip6_hdr));
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100881
882 /* update the session's addresses and mark them set */
883 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
884 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr = hdr_ip6->ip6_src;
885 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = hdr_tcp->source;
886
887 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
888 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr = hdr_ip6->ip6_dst;
889 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = hdr_tcp->dest;
890
891 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
892 }
893 else {
894 /* The protocol does not match something known (IPv4/IPv6) */
895 conn->err_code = CO_ER_CIP_BAD_PROTO;
896 goto fail;
897 }
898
Bertrand Jacquin7d668f92017-12-13 01:23:39 +0000899 line += hdr_len;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200900 trash.data = line - trash.area;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100901
902 /* remove the NetScaler Client IP header from the request. For this
903 * we re-read the exact line at once. If we don't get the exact same
904 * result, we fail.
905 */
906 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200907 int len2 = recv(conn->handle.fd, trash.area, trash.data, 0);
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100908 if (len2 < 0 && errno == EINTR)
909 continue;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200910 if (len2 != trash.data)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100911 goto recv_abort;
912 } while (0);
913
914 conn->flags &= ~flag;
915 return 1;
916
917 missing:
918 /* Missing data. Since we're using MSG_PEEK, we can only poll again if
919 * we have not read anything. Otherwise we need to fail because we won't
920 * be able to poll anymore.
921 */
922 conn->err_code = CO_ER_CIP_TRUNCATED;
923 goto fail;
924
925 bad_magic:
926 conn->err_code = CO_ER_CIP_BAD_MAGIC;
927 goto fail;
928
929 recv_abort:
930 conn->err_code = CO_ER_CIP_ABORT;
931 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
932 goto fail;
933
934 fail:
935 __conn_sock_stop_both(conn);
936 conn->flags |= CO_FL_ERROR;
937 return 0;
938}
939
David Safb76832014-05-08 23:42:08 -0400940int make_proxy_line(char *buf, int buf_len, struct server *srv, struct connection *remote)
941{
942 int ret = 0;
943
944 if (srv && (srv->pp_opts & SRV_PP_V2)) {
945 ret = make_proxy_line_v2(buf, buf_len, srv, remote);
946 }
947 else {
948 if (remote)
949 ret = make_proxy_line_v1(buf, buf_len, &remote->addr.from, &remote->addr.to);
950 else
951 ret = make_proxy_line_v1(buf, buf_len, NULL, NULL);
952 }
953
954 return ret;
955}
956
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200957/* Makes a PROXY protocol line from the two addresses. The output is sent to
958 * buffer <buf> for a maximum size of <buf_len> (including the trailing zero).
959 * It returns the number of bytes composing this line (including the trailing
960 * LF), or zero in case of failure (eg: not enough space). It supports TCP4,
Willy Tarreau2e1401a2013-10-01 11:41:55 +0200961 * TCP6 and "UNKNOWN" formats. If any of <src> or <dst> is null, UNKNOWN is
962 * emitted as well.
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200963 */
David Safb76832014-05-08 23:42:08 -0400964int make_proxy_line_v1(char *buf, int buf_len, struct sockaddr_storage *src, struct sockaddr_storage *dst)
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200965{
966 int ret = 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200967 char * protocol;
968 char src_str[MAX(INET_ADDRSTRLEN, INET6_ADDRSTRLEN)];
969 char dst_str[MAX(INET_ADDRSTRLEN, INET6_ADDRSTRLEN)];
970 in_port_t src_port;
971 in_port_t dst_port;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200972
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200973 if ( !src
974 || !dst
975 || (src->ss_family != AF_INET && src->ss_family != AF_INET6)
976 || (dst->ss_family != AF_INET && dst->ss_family != AF_INET6)) {
977 /* unknown family combination */
978 ret = snprintf(buf, buf_len, "PROXY UNKNOWN\r\n");
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200979 if (ret >= buf_len)
980 return 0;
981
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200982 return ret;
983 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200984
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200985 /* IPv4 for both src and dst */
986 if (src->ss_family == AF_INET && dst->ss_family == AF_INET) {
987 protocol = "TCP4";
988 if (!inet_ntop(AF_INET, &((struct sockaddr_in *)src)->sin_addr, src_str, sizeof(src_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200989 return 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200990 src_port = ((struct sockaddr_in *)src)->sin_port;
991 if (!inet_ntop(AF_INET, &((struct sockaddr_in *)dst)->sin_addr, dst_str, sizeof(dst_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200992 return 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200993 dst_port = ((struct sockaddr_in *)dst)->sin_port;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200994 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200995 /* IPv6 for at least one of src and dst */
996 else {
997 struct in6_addr tmp;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200998
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200999 protocol = "TCP6";
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001000
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001001 if (src->ss_family == AF_INET) {
1002 /* Convert src to IPv6 */
1003 v4tov6(&tmp, &((struct sockaddr_in *)src)->sin_addr);
1004 src_port = ((struct sockaddr_in *)src)->sin_port;
1005 }
1006 else {
1007 tmp = ((struct sockaddr_in6 *)src)->sin6_addr;
1008 src_port = ((struct sockaddr_in6 *)src)->sin6_port;
1009 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001010
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001011 if (!inet_ntop(AF_INET6, &tmp, src_str, sizeof(src_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001012 return 0;
1013
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001014 if (dst->ss_family == AF_INET) {
1015 /* Convert dst to IPv6 */
1016 v4tov6(&tmp, &((struct sockaddr_in *)dst)->sin_addr);
1017 dst_port = ((struct sockaddr_in *)dst)->sin_port;
1018 }
1019 else {
1020 tmp = ((struct sockaddr_in6 *)dst)->sin6_addr;
1021 dst_port = ((struct sockaddr_in6 *)dst)->sin6_port;
1022 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001023
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001024 if (!inet_ntop(AF_INET6, &tmp, dst_str, sizeof(dst_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001025 return 0;
1026 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001027
1028 ret = snprintf(buf, buf_len, "PROXY %s %s %s %u %u\r\n", protocol, src_str, dst_str, ntohs(src_port), ntohs(dst_port));
1029 if (ret >= buf_len)
1030 return 0;
1031
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001032 return ret;
1033}
David Safb76832014-05-08 23:42:08 -04001034
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001035static int make_tlv(char *dest, int dest_len, char type, uint16_t length, const char *value)
David Safb76832014-05-08 23:42:08 -04001036{
1037 struct tlv *tlv;
1038
1039 if (!dest || (length + sizeof(*tlv) > dest_len))
1040 return 0;
1041
1042 tlv = (struct tlv *)dest;
1043
1044 tlv->type = type;
1045 tlv->length_hi = length >> 8;
1046 tlv->length_lo = length & 0x00ff;
1047 memcpy(tlv->value, value, length);
1048 return length + sizeof(*tlv);
1049}
David Safb76832014-05-08 23:42:08 -04001050
1051int make_proxy_line_v2(char *buf, int buf_len, struct server *srv, struct connection *remote)
1052{
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001053 const char pp2_signature[] = PP2_SIGNATURE;
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001054 void *tlv_crc32c_p = NULL;
David Safb76832014-05-08 23:42:08 -04001055 int ret = 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001056 struct proxy_hdr_v2 *hdr = (struct proxy_hdr_v2 *)buf;
Vincent Bernat6e615892016-05-18 16:17:44 +02001057 struct sockaddr_storage null_addr = { .ss_family = 0 };
David Safb76832014-05-08 23:42:08 -04001058 struct sockaddr_storage *src = &null_addr;
1059 struct sockaddr_storage *dst = &null_addr;
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001060 const char *value;
1061 int value_len;
David Safb76832014-05-08 23:42:08 -04001062
1063 if (buf_len < PP2_HEADER_LEN)
1064 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001065 memcpy(hdr->sig, pp2_signature, PP2_SIGNATURE_LEN);
David Safb76832014-05-08 23:42:08 -04001066
1067 if (remote) {
1068 src = &remote->addr.from;
1069 dst = &remote->addr.to;
1070 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001071
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001072 /* At least one of src or dst is not of AF_INET or AF_INET6 */
1073 if ( !src
1074 || !dst
1075 || (src->ss_family != AF_INET && src->ss_family != AF_INET6)
1076 || (dst->ss_family != AF_INET && dst->ss_family != AF_INET6)) {
David Safb76832014-05-08 23:42:08 -04001077 if (buf_len < PP2_HDR_LEN_UNSPEC)
1078 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001079 hdr->ver_cmd = PP2_VERSION | PP2_CMD_LOCAL;
1080 hdr->fam = PP2_FAM_UNSPEC | PP2_TRANS_UNSPEC;
David Safb76832014-05-08 23:42:08 -04001081 ret = PP2_HDR_LEN_UNSPEC;
1082 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001083 else {
1084 /* IPv4 for both src and dst */
1085 if (src->ss_family == AF_INET && dst->ss_family == AF_INET) {
1086 if (buf_len < PP2_HDR_LEN_INET)
1087 return 0;
1088 hdr->ver_cmd = PP2_VERSION | PP2_CMD_PROXY;
1089 hdr->fam = PP2_FAM_INET | PP2_TRANS_STREAM;
1090 hdr->addr.ip4.src_addr = ((struct sockaddr_in *)src)->sin_addr.s_addr;
1091 hdr->addr.ip4.src_port = ((struct sockaddr_in *)src)->sin_port;
1092 hdr->addr.ip4.dst_addr = ((struct sockaddr_in *)dst)->sin_addr.s_addr;
1093 hdr->addr.ip4.dst_port = ((struct sockaddr_in *)dst)->sin_port;
1094 ret = PP2_HDR_LEN_INET;
1095 }
1096 /* IPv6 for at least one of src and dst */
1097 else {
1098 struct in6_addr tmp;
1099
1100 if (buf_len < PP2_HDR_LEN_INET6)
1101 return 0;
1102 hdr->ver_cmd = PP2_VERSION | PP2_CMD_PROXY;
1103 hdr->fam = PP2_FAM_INET6 | PP2_TRANS_STREAM;
1104 if (src->ss_family == AF_INET) {
1105 v4tov6(&tmp, &((struct sockaddr_in *)src)->sin_addr);
1106 memcpy(hdr->addr.ip6.src_addr, &tmp, 16);
1107 hdr->addr.ip6.src_port = ((struct sockaddr_in *)src)->sin_port;
1108 }
1109 else {
1110 memcpy(hdr->addr.ip6.src_addr, &((struct sockaddr_in6 *)src)->sin6_addr, 16);
1111 hdr->addr.ip6.src_port = ((struct sockaddr_in6 *)src)->sin6_port;
1112 }
1113 if (dst->ss_family == AF_INET) {
1114 v4tov6(&tmp, &((struct sockaddr_in *)dst)->sin_addr);
1115 memcpy(hdr->addr.ip6.dst_addr, &tmp, 16);
1116 hdr->addr.ip6.src_port = ((struct sockaddr_in *)src)->sin_port;
1117 }
1118 else {
1119 memcpy(hdr->addr.ip6.dst_addr, &((struct sockaddr_in6 *)dst)->sin6_addr, 16);
1120 hdr->addr.ip6.dst_port = ((struct sockaddr_in6 *)dst)->sin6_port;
1121 }
1122
1123 ret = PP2_HDR_LEN_INET6;
1124 }
1125 }
David Safb76832014-05-08 23:42:08 -04001126
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001127 if (srv->pp_opts & SRV_PP_V2_CRC32C) {
1128 uint32_t zero_crc32c = 0;
1129 if ((buf_len - ret) < sizeof(struct tlv))
1130 return 0;
1131 tlv_crc32c_p = (void *)((struct tlv *)&buf[ret])->value;
1132 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_CRC32C, sizeof(zero_crc32c), (const char *)&zero_crc32c);
1133 }
1134
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001135 if (conn_get_alpn(remote, &value, &value_len)) {
1136 if ((buf_len - ret) < sizeof(struct tlv))
1137 return 0;
Emmanuel Hocdet571c7ac2017-10-31 18:24:05 +01001138 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_ALPN, value_len, value);
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001139 }
1140
David Safb76832014-05-08 23:42:08 -04001141#ifdef USE_OPENSSL
Emmanuel Hocdet253c3b72018-02-01 18:29:59 +01001142 if (srv->pp_opts & SRV_PP_V2_AUTHORITY) {
1143 value = ssl_sock_get_sni(remote);
1144 if (value) {
1145 if ((buf_len - ret) < sizeof(struct tlv))
1146 return 0;
1147 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_AUTHORITY, strlen(value), value);
1148 }
1149 }
1150
David Safb76832014-05-08 23:42:08 -04001151 if (srv->pp_opts & SRV_PP_V2_SSL) {
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001152 struct tlv_ssl *tlv;
1153 int ssl_tlv_len = 0;
David Safb76832014-05-08 23:42:08 -04001154 if ((buf_len - ret) < sizeof(struct tlv_ssl))
1155 return 0;
1156 tlv = (struct tlv_ssl *)&buf[ret];
1157 memset(tlv, 0, sizeof(struct tlv_ssl));
1158 ssl_tlv_len += sizeof(struct tlv_ssl);
1159 tlv->tlv.type = PP2_TYPE_SSL;
1160 if (ssl_sock_is_ssl(remote)) {
1161 tlv->client |= PP2_CLIENT_SSL;
Emmanuel Hocdet01da5712017-10-13 16:59:49 +02001162 value = ssl_sock_get_proto_version(remote);
David Safb76832014-05-08 23:42:08 -04001163 if (value) {
Emmanuel Hocdet8c0c34b2018-02-28 12:02:14 +01001164 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 -04001165 }
Dave McCowan328fb582014-07-30 10:39:13 -04001166 if (ssl_sock_get_cert_used_sess(remote)) {
1167 tlv->client |= PP2_CLIENT_CERT_SESS;
David Safb76832014-05-08 23:42:08 -04001168 tlv->verify = htonl(ssl_sock_get_verify_result(remote));
Dave McCowan328fb582014-07-30 10:39:13 -04001169 if (ssl_sock_get_cert_used_conn(remote))
1170 tlv->client |= PP2_CLIENT_CERT_CONN;
David Safb76832014-05-08 23:42:08 -04001171 }
1172 if (srv->pp_opts & SRV_PP_V2_SSL_CN) {
Willy Tarreau83061a82018-07-13 11:56:34 +02001173 struct buffer *cn_trash = get_trash_chunk();
Willy Tarreau3b9a0c92014-07-19 06:37:33 +02001174 if (ssl_sock_get_remote_common_name(remote, cn_trash) > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001175 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_CN,
1176 cn_trash->data,
1177 cn_trash->area);
David Safb76832014-05-08 23:42:08 -04001178 }
1179 }
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001180 if (srv->pp_opts & SRV_PP_V2_SSL_KEY_ALG) {
Willy Tarreau83061a82018-07-13 11:56:34 +02001181 struct buffer *pkey_trash = get_trash_chunk();
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001182 if (ssl_sock_get_pkey_algo(remote, pkey_trash) > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001183 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_KEY_ALG,
1184 pkey_trash->data,
1185 pkey_trash->area);
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001186 }
1187 }
1188 if (srv->pp_opts & SRV_PP_V2_SSL_SIG_ALG) {
1189 value = ssl_sock_get_cert_sig(remote);
1190 if (value) {
1191 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_SIG_ALG, strlen(value), value);
1192 }
1193 }
1194 if (srv->pp_opts & SRV_PP_V2_SSL_CIPHER) {
1195 value = ssl_sock_get_cipher_name(remote);
1196 if (value) {
1197 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_CIPHER, strlen(value), value);
1198 }
1199 }
David Safb76832014-05-08 23:42:08 -04001200 }
1201 tlv->tlv.length_hi = (uint16_t)(ssl_tlv_len - sizeof(struct tlv)) >> 8;
1202 tlv->tlv.length_lo = (uint16_t)(ssl_tlv_len - sizeof(struct tlv)) & 0x00ff;
1203 ret += ssl_tlv_len;
1204 }
1205#endif
1206
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001207#ifdef CONFIG_HAP_NS
1208 if (remote && (remote->proxy_netns)) {
1209 if ((buf_len - ret) < sizeof(struct tlv))
1210 return 0;
Emmanuel Hocdet571c7ac2017-10-31 18:24:05 +01001211 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 +01001212 }
1213#endif
1214
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001215 hdr->len = htons((uint16_t)(ret - PP2_HEADER_LEN));
David Safb76832014-05-08 23:42:08 -04001216
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001217 if (tlv_crc32c_p) {
1218 write_u32(tlv_crc32c_p, htonl(hash_crc32c(buf, ret)));
1219 }
1220
David Safb76832014-05-08 23:42:08 -04001221 return ret;
1222}
Emeric Brun4f603012017-01-05 15:11:44 +01001223
Willy Tarreau60ca10a2017-08-18 15:26:54 +02001224/* return the major HTTP version as 1 or 2 depending on how the request arrived
1225 * before being processed.
1226 */
1227static int
1228smp_fetch_fc_http_major(const struct arg *args, struct sample *smp, const char *kw, void *private)
1229{
1230 struct connection *conn = objt_conn(smp->sess->origin);
1231
1232 smp->data.type = SMP_T_SINT;
1233 smp->data.u.sint = (conn && strcmp(conn_get_mux_name(conn), "H2") == 0) ? 2 : 1;
1234 return 1;
1235}
1236
Emeric Brun4f603012017-01-05 15:11:44 +01001237/* fetch if the received connection used a PROXY protocol header */
1238int smp_fetch_fc_rcvd_proxy(const struct arg *args, struct sample *smp, const char *kw, void *private)
1239{
1240 struct connection *conn;
1241
1242 conn = objt_conn(smp->sess->origin);
1243 if (!conn)
1244 return 0;
1245
1246 if (!(conn->flags & CO_FL_CONNECTED)) {
1247 smp->flags |= SMP_F_MAY_CHANGE;
1248 return 0;
1249 }
1250
1251 smp->flags = 0;
1252 smp->data.type = SMP_T_BOOL;
1253 smp->data.u.sint = (conn->flags & CO_FL_RCVD_PROXY) ? 1 : 0;
1254
1255 return 1;
1256}
1257
1258/* Note: must not be declared <const> as its list will be overwritten.
1259 * Note: fetches that may return multiple types must be declared as the lowest
1260 * common denominator, the type that can be casted into all other ones. For
1261 * instance v4/v6 must be declared v4.
1262 */
1263static struct sample_fetch_kw_list sample_fetch_keywords = {ILH, {
Willy Tarreau60ca10a2017-08-18 15:26:54 +02001264 { "fc_http_major", smp_fetch_fc_http_major, 0, NULL, SMP_T_SINT, SMP_USE_L4CLI },
Emeric Brun4f603012017-01-05 15:11:44 +01001265 { "fc_rcvd_proxy", smp_fetch_fc_rcvd_proxy, 0, NULL, SMP_T_BOOL, SMP_USE_L4CLI },
1266 { /* END */ },
1267}};
1268
1269
1270__attribute__((constructor))
1271static void __connection_init(void)
1272{
1273 sample_register_fetches(&sample_fetch_keywords);
1274}