blob: f750e3c99ba37a35e897b4c03013c8a2be03bff3 [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);
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/*
Christopher Faulet063f7862018-05-23 15:03:43 +0200414 * default cs send() : this one is used when mux->snd_buf == NULL. It puts up to
415 * <count> bytes from <buf> into cs->txbuf. The number of bytes transferred is
416 * returned. Here we don't care if cs->txbuf is allocated or not. If not, it
417 * will be swapped with <buf>.
418 */
419size_t __cs_send(struct conn_stream *cs, struct buffer *buf, size_t count, int flags)
420{
421 return b_xfer(&cs->txbuf, buf, count);
422}
423
424/*
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100425 * Get data length from tlv
426 */
427static int get_tlv_length(const struct tlv *src)
428{
429 return (src->length_hi << 8) | src->length_lo;
430}
431
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200432/* This handshake handler waits a PROXY protocol header at the beginning of the
433 * raw data stream. The header looks like this :
434 *
435 * "PROXY" <SP> PROTO <SP> SRC3 <SP> DST3 <SP> SRC4 <SP> <DST4> "\r\n"
436 *
437 * There must be exactly one space between each field. Fields are :
438 * - PROTO : layer 4 protocol, which must be "TCP4" or "TCP6".
439 * - SRC3 : layer 3 (eg: IP) source address in standard text form
440 * - DST3 : layer 3 (eg: IP) destination address in standard text form
441 * - SRC4 : layer 4 (eg: TCP port) source address in standard text form
442 * - DST4 : layer 4 (eg: TCP port) destination address in standard text form
443 *
444 * This line MUST be at the beginning of the buffer and MUST NOT wrap.
445 *
446 * The header line is small and in all cases smaller than the smallest normal
447 * TCP MSS. So it MUST always be delivered as one segment, which ensures we
448 * can safely use MSG_PEEK and avoid buffering.
449 *
450 * Once the data is fetched, the values are set in the connection's address
451 * fields, and data are removed from the socket's buffer. The function returns
452 * zero if it needs to wait for more data or if it fails, or 1 if it completed
453 * and removed itself.
454 */
455int conn_recv_proxy(struct connection *conn, int flag)
456{
457 char *line, *end;
Willy Tarreau77992672014-06-14 11:06:17 +0200458 struct proxy_hdr_v2 *hdr_v2;
459 const char v2sig[] = PP2_SIGNATURE;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100460 int tlv_length = 0;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200461 int tlv_offset = 0;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200462
463 /* we might have been called just after an asynchronous shutr */
464 if (conn->flags & CO_FL_SOCK_RD_SH)
465 goto fail;
466
Willy Tarreau3c728722014-01-23 13:50:42 +0100467 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200468 goto fail;
469
Willy Tarreau585744b2017-08-24 14:31:19 +0200470 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100471 return 0;
472
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200473 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200474 trash.data = recv(conn->handle.fd, trash.area, trash.size,
475 MSG_PEEK);
476 if (trash.data < 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200477 if (errno == EINTR)
478 continue;
479 if (errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200480 fd_cant_recv(conn->handle.fd);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200481 return 0;
482 }
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100483 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200484 }
485 } while (0);
486
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200487 if (!trash.data) {
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100488 /* client shutdown */
489 conn->err_code = CO_ER_PRX_EMPTY;
490 goto fail;
491 }
492
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200493 if (trash.data < 6)
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200494 goto missing;
495
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200496 line = trash.area;
497 end = trash.area + trash.data;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200498
499 /* Decode a possible proxy request, fail early if it does not match */
Willy Tarreau77992672014-06-14 11:06:17 +0200500 if (strncmp(line, "PROXY ", 6) != 0)
501 goto not_v1;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200502
503 line += 6;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200504 if (trash.data < 9) /* shortest possible line */
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200505 goto missing;
506
David CARLIER42ff05e2016-03-24 09:22:36 +0000507 if (memcmp(line, "TCP4 ", 5) == 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200508 u32 src3, dst3, sport, dport;
509
510 line += 5;
511
512 src3 = inetaddr_host_lim_ret(line, end, &line);
513 if (line == end)
514 goto missing;
515 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100516 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200517
518 dst3 = inetaddr_host_lim_ret(line, end, &line);
519 if (line == end)
520 goto missing;
521 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100522 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200523
524 sport = read_uint((const char **)&line, end);
525 if (line == end)
526 goto missing;
527 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100528 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200529
530 dport = read_uint((const char **)&line, end);
531 if (line > end - 2)
532 goto missing;
533 if (*line++ != '\r')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100534 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200535 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100536 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200537
538 /* update the session's addresses and mark them set */
539 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
540 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = htonl(src3);
541 ((struct sockaddr_in *)&conn->addr.from)->sin_port = htons(sport);
542
543 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
544 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = htonl(dst3);
545 ((struct sockaddr_in *)&conn->addr.to)->sin_port = htons(dport);
546 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
547 }
David CARLIER42ff05e2016-03-24 09:22:36 +0000548 else if (memcmp(line, "TCP6 ", 5) == 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200549 u32 sport, dport;
550 char *src_s;
551 char *dst_s, *sport_s, *dport_s;
552 struct in6_addr src3, dst3;
553
554 line += 5;
555
556 src_s = line;
557 dst_s = sport_s = dport_s = NULL;
558 while (1) {
559 if (line > end - 2) {
560 goto missing;
561 }
562 else if (*line == '\r') {
563 *line = 0;
564 line++;
565 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100566 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200567 break;
568 }
569
570 if (*line == ' ') {
571 *line = 0;
572 if (!dst_s)
573 dst_s = line + 1;
574 else if (!sport_s)
575 sport_s = line + 1;
576 else if (!dport_s)
577 dport_s = line + 1;
578 }
579 line++;
580 }
581
582 if (!dst_s || !sport_s || !dport_s)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100583 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200584
585 sport = read_uint((const char **)&sport_s,dport_s - 1);
586 if (*sport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100587 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200588
589 dport = read_uint((const char **)&dport_s,line - 2);
590 if (*dport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100591 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200592
593 if (inet_pton(AF_INET6, src_s, (void *)&src3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100594 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200595
596 if (inet_pton(AF_INET6, dst_s, (void *)&dst3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100597 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200598
599 /* update the session's addresses and mark them set */
600 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
601 memcpy(&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr, &src3, sizeof(struct in6_addr));
602 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = htons(sport);
603
604 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
605 memcpy(&((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr, &dst3, sizeof(struct in6_addr));
606 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = htons(dport);
607 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
608 }
Willy Tarreau4c20d292014-06-14 11:41:36 +0200609 else if (memcmp(line, "UNKNOWN\r\n", 9) == 0) {
610 /* This can be a UNIX socket forwarded by an haproxy upstream */
611 line += 9;
612 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200613 else {
Willy Tarreau4c20d292014-06-14 11:41:36 +0200614 /* The protocol does not match something known (TCP4/TCP6/UNKNOWN) */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100615 conn->err_code = CO_ER_PRX_BAD_PROTO;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200616 goto fail;
617 }
618
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200619 trash.data = line - trash.area;
Willy Tarreau77992672014-06-14 11:06:17 +0200620 goto eat_header;
621
622 not_v1:
623 /* try PPv2 */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200624 if (trash.data < PP2_HEADER_LEN)
Willy Tarreau77992672014-06-14 11:06:17 +0200625 goto missing;
626
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200627 hdr_v2 = (struct proxy_hdr_v2 *) trash.area;
Willy Tarreau77992672014-06-14 11:06:17 +0200628
629 if (memcmp(hdr_v2->sig, v2sig, PP2_SIGNATURE_LEN) != 0 ||
630 (hdr_v2->ver_cmd & PP2_VERSION_MASK) != PP2_VERSION) {
631 conn->err_code = CO_ER_PRX_NOT_HDR;
632 goto fail;
633 }
634
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200635 if (trash.data < PP2_HEADER_LEN + ntohs(hdr_v2->len))
Willy Tarreau77992672014-06-14 11:06:17 +0200636 goto missing;
637
638 switch (hdr_v2->ver_cmd & PP2_CMD_MASK) {
639 case 0x01: /* PROXY command */
640 switch (hdr_v2->fam) {
641 case 0x11: /* TCPv4 */
KOVACS Krisztianefd3aa92014-11-19 10:53:20 +0100642 if (ntohs(hdr_v2->len) < PP2_ADDR_LEN_INET)
643 goto bad_header;
644
Willy Tarreau77992672014-06-14 11:06:17 +0200645 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
646 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = hdr_v2->addr.ip4.src_addr;
647 ((struct sockaddr_in *)&conn->addr.from)->sin_port = hdr_v2->addr.ip4.src_port;
648 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
649 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = hdr_v2->addr.ip4.dst_addr;
650 ((struct sockaddr_in *)&conn->addr.to)->sin_port = hdr_v2->addr.ip4.dst_port;
651 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200652 tlv_offset = PP2_HEADER_LEN + PP2_ADDR_LEN_INET;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100653 tlv_length = ntohs(hdr_v2->len) - PP2_ADDR_LEN_INET;
Willy Tarreau77992672014-06-14 11:06:17 +0200654 break;
655 case 0x21: /* TCPv6 */
KOVACS Krisztianefd3aa92014-11-19 10:53:20 +0100656 if (ntohs(hdr_v2->len) < PP2_ADDR_LEN_INET6)
657 goto bad_header;
658
Willy Tarreau77992672014-06-14 11:06:17 +0200659 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
660 memcpy(&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr, hdr_v2->addr.ip6.src_addr, 16);
661 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = hdr_v2->addr.ip6.src_port;
662 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
663 memcpy(&((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr, hdr_v2->addr.ip6.dst_addr, 16);
664 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = hdr_v2->addr.ip6.dst_port;
665 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200666 tlv_offset = PP2_HEADER_LEN + PP2_ADDR_LEN_INET6;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100667 tlv_length = ntohs(hdr_v2->len) - PP2_ADDR_LEN_INET6;
Willy Tarreau77992672014-06-14 11:06:17 +0200668 break;
669 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100670
671 /* TLV parsing */
672 if (tlv_length > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200673 while (tlv_offset + TLV_HEADER_SIZE <= trash.data) {
674 const struct tlv *tlv_packet = (struct tlv *) &trash.area[tlv_offset];
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100675 const int tlv_len = get_tlv_length(tlv_packet);
676 tlv_offset += tlv_len + TLV_HEADER_SIZE;
677
678 switch (tlv_packet->type) {
Emmanuel Hocdet115df3e2018-02-05 16:23:23 +0100679 case PP2_TYPE_CRC32C: {
680 void *tlv_crc32c_p = (void *)tlv_packet->value;
681 uint32_t n_crc32c = ntohl(read_u32(tlv_crc32c_p));
682 write_u32(tlv_crc32c_p, 0);
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200683 if (hash_crc32c(trash.area, PP2_HEADER_LEN + ntohs(hdr_v2->len)) != n_crc32c)
Emmanuel Hocdet115df3e2018-02-05 16:23:23 +0100684 goto bad_header;
685 break;
686 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100687#ifdef CONFIG_HAP_NS
688 case PP2_TYPE_NETNS: {
689 const struct netns_entry *ns;
690 ns = netns_store_lookup((char*)tlv_packet->value, tlv_len);
691 if (ns)
692 conn->proxy_netns = ns;
693 break;
694 }
695#endif
696 default:
697 break;
698 }
699 }
700 }
701
Willy Tarreau77992672014-06-14 11:06:17 +0200702 /* unsupported protocol, keep local connection address */
703 break;
704 case 0x00: /* LOCAL command */
705 /* keep local connection address for LOCAL */
706 break;
707 default:
708 goto bad_header; /* not a supported command */
709 }
710
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200711 trash.data = PP2_HEADER_LEN + ntohs(hdr_v2->len);
Willy Tarreau77992672014-06-14 11:06:17 +0200712 goto eat_header;
713
714 eat_header:
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200715 /* remove the PROXY line from the request. For this we re-read the
716 * exact line at once. If we don't get the exact same result, we
717 * fail.
718 */
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200719 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200720 int len2 = recv(conn->handle.fd, trash.area, trash.data, 0);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200721 if (len2 < 0 && errno == EINTR)
722 continue;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200723 if (len2 != trash.data)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100724 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200725 } while (0);
726
727 conn->flags &= ~flag;
Emeric Brun4f603012017-01-05 15:11:44 +0100728 conn->flags |= CO_FL_RCVD_PROXY;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200729 return 1;
730
731 missing:
732 /* Missing data. Since we're using MSG_PEEK, we can only poll again if
733 * we have not read anything. Otherwise we need to fail because we won't
734 * be able to poll anymore.
735 */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100736 conn->err_code = CO_ER_PRX_TRUNCATED;
737 goto fail;
738
739 bad_header:
740 /* This is not a valid proxy protocol header */
741 conn->err_code = CO_ER_PRX_BAD_HDR;
742 goto fail;
743
744 recv_abort:
745 conn->err_code = CO_ER_PRX_ABORT;
Willy Tarreau26f4a042013-12-04 23:44:10 +0100746 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100747 goto fail;
748
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200749 fail:
Willy Tarreaud486ef52012-12-10 17:03:52 +0100750 __conn_sock_stop_both(conn);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200751 conn->flags |= CO_FL_ERROR;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200752 return 0;
753}
754
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100755/* This handshake handler waits a NetScaler Client IP insertion header
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000756 * at the beginning of the raw data stream. The header format is
757 * described in doc/netscaler-client-ip-insertion-protocol.txt
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100758 *
759 * This line MUST be at the beginning of the buffer and MUST NOT be
760 * fragmented.
761 *
762 * The header line is small and in all cases smaller than the smallest normal
763 * TCP MSS. So it MUST always be delivered as one segment, which ensures we
764 * can safely use MSG_PEEK and avoid buffering.
765 *
766 * Once the data is fetched, the values are set in the connection's address
767 * fields, and data are removed from the socket's buffer. The function returns
768 * zero if it needs to wait for more data or if it fails, or 1 if it completed
769 * and removed itself.
770 */
771int conn_recv_netscaler_cip(struct connection *conn, int flag)
772{
773 char *line;
Bertrand Jacquin7d668f92017-12-13 01:23:39 +0000774 uint32_t hdr_len;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100775 uint8_t ip_v;
776
777 /* we might have been called just after an asynchronous shutr */
778 if (conn->flags & CO_FL_SOCK_RD_SH)
779 goto fail;
780
781 if (!conn_ctrl_ready(conn))
782 goto fail;
783
Willy Tarreau585744b2017-08-24 14:31:19 +0200784 if (!fd_recv_ready(conn->handle.fd))
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100785 return 0;
786
787 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200788 trash.data = recv(conn->handle.fd, trash.area, trash.size,
789 MSG_PEEK);
790 if (trash.data < 0) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100791 if (errno == EINTR)
792 continue;
793 if (errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200794 fd_cant_recv(conn->handle.fd);
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100795 return 0;
796 }
797 goto recv_abort;
798 }
799 } while (0);
800
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200801 if (!trash.data) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100802 /* client shutdown */
803 conn->err_code = CO_ER_CIP_EMPTY;
804 goto fail;
805 }
806
807 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000808 * CIP magic, header length or
809 * CIP magic, CIP length, CIP type, header length */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200810 if (trash.data < 12)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100811 goto missing;
812
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200813 line = trash.area;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100814
815 /* Decode a possible NetScaler Client IP request, fail early if
816 * it does not match */
Bertrand Jacquin4b4c2862017-12-13 01:07:12 +0000817 if (ntohl(*(uint32_t *)line) != objt_listener(conn->target)->bind_conf->ns_cip_magic)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100818 goto bad_magic;
819
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000820 /* Legacy CIP protocol */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200821 if ((trash.area[8] & 0xD0) == 0x40) {
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000822 hdr_len = ntohl(*(uint32_t *)(line+4));
823 line += 8;
824 }
825 /* Standard CIP protocol */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200826 else if (trash.area[8] == 0x00) {
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000827 hdr_len = ntohs(*(uint32_t *)(line+10));
828 line += 12;
829 }
830 /* Unknown CIP protocol */
831 else {
832 conn->err_code = CO_ER_CIP_BAD_PROTO;
833 goto fail;
834 }
835
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100836 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000837 * a minimal IP header */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200838 if (trash.data < 20)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100839 goto missing;
840
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100841 /* Get IP version from the first four bits */
842 ip_v = (*line & 0xf0) >> 4;
843
844 if (ip_v == 4) {
845 struct ip *hdr_ip4;
David Carlier3015a2e2016-07-04 22:51:33 +0100846 struct my_tcphdr *hdr_tcp;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100847
848 hdr_ip4 = (struct ip *)line;
849
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200850 if (trash.data < 40 || trash.data < hdr_len) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100851 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin67de5a22017-12-13 01:15:05 +0000852 * IPv4 header, TCP header */
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100853 goto missing;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000854 }
855 else if (hdr_ip4->ip_p != IPPROTO_TCP) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100856 /* The protocol does not include a TCP header */
857 conn->err_code = CO_ER_CIP_BAD_PROTO;
858 goto fail;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000859 }
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100860
David Carlier3015a2e2016-07-04 22:51:33 +0100861 hdr_tcp = (struct my_tcphdr *)(line + (hdr_ip4->ip_hl * 4));
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100862
863 /* update the session's addresses and mark them set */
864 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
865 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = hdr_ip4->ip_src.s_addr;
866 ((struct sockaddr_in *)&conn->addr.from)->sin_port = hdr_tcp->source;
867
868 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
869 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = hdr_ip4->ip_dst.s_addr;
870 ((struct sockaddr_in *)&conn->addr.to)->sin_port = hdr_tcp->dest;
871
872 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
873 }
874 else if (ip_v == 6) {
875 struct ip6_hdr *hdr_ip6;
David Carlier3015a2e2016-07-04 22:51:33 +0100876 struct my_tcphdr *hdr_tcp;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100877
878 hdr_ip6 = (struct ip6_hdr *)line;
879
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200880 if (trash.data < 60 || trash.data < hdr_len) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100881 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin67de5a22017-12-13 01:15:05 +0000882 * IPv6 header, TCP header */
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100883 goto missing;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000884 }
885 else if (hdr_ip6->ip6_nxt != IPPROTO_TCP) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100886 /* The protocol does not include a TCP header */
887 conn->err_code = CO_ER_CIP_BAD_PROTO;
888 goto fail;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000889 }
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100890
David Carlier3015a2e2016-07-04 22:51:33 +0100891 hdr_tcp = (struct my_tcphdr *)(line + sizeof(struct ip6_hdr));
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100892
893 /* update the session's addresses and mark them set */
894 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
895 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr = hdr_ip6->ip6_src;
896 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = hdr_tcp->source;
897
898 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
899 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr = hdr_ip6->ip6_dst;
900 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = hdr_tcp->dest;
901
902 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
903 }
904 else {
905 /* The protocol does not match something known (IPv4/IPv6) */
906 conn->err_code = CO_ER_CIP_BAD_PROTO;
907 goto fail;
908 }
909
Bertrand Jacquin7d668f92017-12-13 01:23:39 +0000910 line += hdr_len;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200911 trash.data = line - trash.area;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100912
913 /* remove the NetScaler Client IP header from the request. For this
914 * we re-read the exact line at once. If we don't get the exact same
915 * result, we fail.
916 */
917 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200918 int len2 = recv(conn->handle.fd, trash.area, trash.data, 0);
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100919 if (len2 < 0 && errno == EINTR)
920 continue;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200921 if (len2 != trash.data)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100922 goto recv_abort;
923 } while (0);
924
925 conn->flags &= ~flag;
926 return 1;
927
928 missing:
929 /* Missing data. Since we're using MSG_PEEK, we can only poll again if
930 * we have not read anything. Otherwise we need to fail because we won't
931 * be able to poll anymore.
932 */
933 conn->err_code = CO_ER_CIP_TRUNCATED;
934 goto fail;
935
936 bad_magic:
937 conn->err_code = CO_ER_CIP_BAD_MAGIC;
938 goto fail;
939
940 recv_abort:
941 conn->err_code = CO_ER_CIP_ABORT;
942 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
943 goto fail;
944
945 fail:
946 __conn_sock_stop_both(conn);
947 conn->flags |= CO_FL_ERROR;
948 return 0;
949}
950
David Safb76832014-05-08 23:42:08 -0400951int make_proxy_line(char *buf, int buf_len, struct server *srv, struct connection *remote)
952{
953 int ret = 0;
954
955 if (srv && (srv->pp_opts & SRV_PP_V2)) {
956 ret = make_proxy_line_v2(buf, buf_len, srv, remote);
957 }
958 else {
959 if (remote)
960 ret = make_proxy_line_v1(buf, buf_len, &remote->addr.from, &remote->addr.to);
961 else
962 ret = make_proxy_line_v1(buf, buf_len, NULL, NULL);
963 }
964
965 return ret;
966}
967
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200968/* Makes a PROXY protocol line from the two addresses. The output is sent to
969 * buffer <buf> for a maximum size of <buf_len> (including the trailing zero).
970 * It returns the number of bytes composing this line (including the trailing
971 * LF), or zero in case of failure (eg: not enough space). It supports TCP4,
Willy Tarreau2e1401a2013-10-01 11:41:55 +0200972 * TCP6 and "UNKNOWN" formats. If any of <src> or <dst> is null, UNKNOWN is
973 * emitted as well.
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200974 */
David Safb76832014-05-08 23:42:08 -0400975int make_proxy_line_v1(char *buf, int buf_len, struct sockaddr_storage *src, struct sockaddr_storage *dst)
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200976{
977 int ret = 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200978 char * protocol;
979 char src_str[MAX(INET_ADDRSTRLEN, INET6_ADDRSTRLEN)];
980 char dst_str[MAX(INET_ADDRSTRLEN, INET6_ADDRSTRLEN)];
981 in_port_t src_port;
982 in_port_t dst_port;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200983
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200984 if ( !src
985 || !dst
986 || (src->ss_family != AF_INET && src->ss_family != AF_INET6)
987 || (dst->ss_family != AF_INET && dst->ss_family != AF_INET6)) {
988 /* unknown family combination */
989 ret = snprintf(buf, buf_len, "PROXY UNKNOWN\r\n");
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200990 if (ret >= buf_len)
991 return 0;
992
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200993 return ret;
994 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200995
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200996 /* IPv4 for both src and dst */
997 if (src->ss_family == AF_INET && dst->ss_family == AF_INET) {
998 protocol = "TCP4";
999 if (!inet_ntop(AF_INET, &((struct sockaddr_in *)src)->sin_addr, src_str, sizeof(src_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001000 return 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001001 src_port = ((struct sockaddr_in *)src)->sin_port;
1002 if (!inet_ntop(AF_INET, &((struct sockaddr_in *)dst)->sin_addr, dst_str, sizeof(dst_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001003 return 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001004 dst_port = ((struct sockaddr_in *)dst)->sin_port;
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001005 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001006 /* IPv6 for at least one of src and dst */
1007 else {
1008 struct in6_addr tmp;
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001009
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001010 protocol = "TCP6";
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001011
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001012 if (src->ss_family == AF_INET) {
1013 /* Convert src to IPv6 */
1014 v4tov6(&tmp, &((struct sockaddr_in *)src)->sin_addr);
1015 src_port = ((struct sockaddr_in *)src)->sin_port;
1016 }
1017 else {
1018 tmp = ((struct sockaddr_in6 *)src)->sin6_addr;
1019 src_port = ((struct sockaddr_in6 *)src)->sin6_port;
1020 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001021
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001022 if (!inet_ntop(AF_INET6, &tmp, src_str, sizeof(src_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001023 return 0;
1024
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001025 if (dst->ss_family == AF_INET) {
1026 /* Convert dst to IPv6 */
1027 v4tov6(&tmp, &((struct sockaddr_in *)dst)->sin_addr);
1028 dst_port = ((struct sockaddr_in *)dst)->sin_port;
1029 }
1030 else {
1031 tmp = ((struct sockaddr_in6 *)dst)->sin6_addr;
1032 dst_port = ((struct sockaddr_in6 *)dst)->sin6_port;
1033 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001034
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001035 if (!inet_ntop(AF_INET6, &tmp, dst_str, sizeof(dst_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001036 return 0;
1037 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001038
1039 ret = snprintf(buf, buf_len, "PROXY %s %s %s %u %u\r\n", protocol, src_str, dst_str, ntohs(src_port), ntohs(dst_port));
1040 if (ret >= buf_len)
1041 return 0;
1042
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001043 return ret;
1044}
David Safb76832014-05-08 23:42:08 -04001045
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001046static int make_tlv(char *dest, int dest_len, char type, uint16_t length, const char *value)
David Safb76832014-05-08 23:42:08 -04001047{
1048 struct tlv *tlv;
1049
1050 if (!dest || (length + sizeof(*tlv) > dest_len))
1051 return 0;
1052
1053 tlv = (struct tlv *)dest;
1054
1055 tlv->type = type;
1056 tlv->length_hi = length >> 8;
1057 tlv->length_lo = length & 0x00ff;
1058 memcpy(tlv->value, value, length);
1059 return length + sizeof(*tlv);
1060}
David Safb76832014-05-08 23:42:08 -04001061
1062int make_proxy_line_v2(char *buf, int buf_len, struct server *srv, struct connection *remote)
1063{
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001064 const char pp2_signature[] = PP2_SIGNATURE;
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001065 void *tlv_crc32c_p = NULL;
David Safb76832014-05-08 23:42:08 -04001066 int ret = 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001067 struct proxy_hdr_v2 *hdr = (struct proxy_hdr_v2 *)buf;
Vincent Bernat6e615892016-05-18 16:17:44 +02001068 struct sockaddr_storage null_addr = { .ss_family = 0 };
David Safb76832014-05-08 23:42:08 -04001069 struct sockaddr_storage *src = &null_addr;
1070 struct sockaddr_storage *dst = &null_addr;
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001071 const char *value;
1072 int value_len;
David Safb76832014-05-08 23:42:08 -04001073
1074 if (buf_len < PP2_HEADER_LEN)
1075 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001076 memcpy(hdr->sig, pp2_signature, PP2_SIGNATURE_LEN);
David Safb76832014-05-08 23:42:08 -04001077
1078 if (remote) {
1079 src = &remote->addr.from;
1080 dst = &remote->addr.to;
1081 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001082
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001083 /* At least one of src or dst is not of AF_INET or AF_INET6 */
1084 if ( !src
1085 || !dst
1086 || (src->ss_family != AF_INET && src->ss_family != AF_INET6)
1087 || (dst->ss_family != AF_INET && dst->ss_family != AF_INET6)) {
David Safb76832014-05-08 23:42:08 -04001088 if (buf_len < PP2_HDR_LEN_UNSPEC)
1089 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001090 hdr->ver_cmd = PP2_VERSION | PP2_CMD_LOCAL;
1091 hdr->fam = PP2_FAM_UNSPEC | PP2_TRANS_UNSPEC;
David Safb76832014-05-08 23:42:08 -04001092 ret = PP2_HDR_LEN_UNSPEC;
1093 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001094 else {
1095 /* IPv4 for both src and dst */
1096 if (src->ss_family == AF_INET && dst->ss_family == AF_INET) {
1097 if (buf_len < PP2_HDR_LEN_INET)
1098 return 0;
1099 hdr->ver_cmd = PP2_VERSION | PP2_CMD_PROXY;
1100 hdr->fam = PP2_FAM_INET | PP2_TRANS_STREAM;
1101 hdr->addr.ip4.src_addr = ((struct sockaddr_in *)src)->sin_addr.s_addr;
1102 hdr->addr.ip4.src_port = ((struct sockaddr_in *)src)->sin_port;
1103 hdr->addr.ip4.dst_addr = ((struct sockaddr_in *)dst)->sin_addr.s_addr;
1104 hdr->addr.ip4.dst_port = ((struct sockaddr_in *)dst)->sin_port;
1105 ret = PP2_HDR_LEN_INET;
1106 }
1107 /* IPv6 for at least one of src and dst */
1108 else {
1109 struct in6_addr tmp;
1110
1111 if (buf_len < PP2_HDR_LEN_INET6)
1112 return 0;
1113 hdr->ver_cmd = PP2_VERSION | PP2_CMD_PROXY;
1114 hdr->fam = PP2_FAM_INET6 | PP2_TRANS_STREAM;
1115 if (src->ss_family == AF_INET) {
1116 v4tov6(&tmp, &((struct sockaddr_in *)src)->sin_addr);
1117 memcpy(hdr->addr.ip6.src_addr, &tmp, 16);
1118 hdr->addr.ip6.src_port = ((struct sockaddr_in *)src)->sin_port;
1119 }
1120 else {
1121 memcpy(hdr->addr.ip6.src_addr, &((struct sockaddr_in6 *)src)->sin6_addr, 16);
1122 hdr->addr.ip6.src_port = ((struct sockaddr_in6 *)src)->sin6_port;
1123 }
1124 if (dst->ss_family == AF_INET) {
1125 v4tov6(&tmp, &((struct sockaddr_in *)dst)->sin_addr);
1126 memcpy(hdr->addr.ip6.dst_addr, &tmp, 16);
1127 hdr->addr.ip6.src_port = ((struct sockaddr_in *)src)->sin_port;
1128 }
1129 else {
1130 memcpy(hdr->addr.ip6.dst_addr, &((struct sockaddr_in6 *)dst)->sin6_addr, 16);
1131 hdr->addr.ip6.dst_port = ((struct sockaddr_in6 *)dst)->sin6_port;
1132 }
1133
1134 ret = PP2_HDR_LEN_INET6;
1135 }
1136 }
David Safb76832014-05-08 23:42:08 -04001137
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001138 if (srv->pp_opts & SRV_PP_V2_CRC32C) {
1139 uint32_t zero_crc32c = 0;
1140 if ((buf_len - ret) < sizeof(struct tlv))
1141 return 0;
1142 tlv_crc32c_p = (void *)((struct tlv *)&buf[ret])->value;
1143 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_CRC32C, sizeof(zero_crc32c), (const char *)&zero_crc32c);
1144 }
1145
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001146 if (conn_get_alpn(remote, &value, &value_len)) {
1147 if ((buf_len - ret) < sizeof(struct tlv))
1148 return 0;
Emmanuel Hocdet571c7ac2017-10-31 18:24:05 +01001149 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_ALPN, value_len, value);
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001150 }
1151
David Safb76832014-05-08 23:42:08 -04001152#ifdef USE_OPENSSL
Emmanuel Hocdet253c3b72018-02-01 18:29:59 +01001153 if (srv->pp_opts & SRV_PP_V2_AUTHORITY) {
1154 value = ssl_sock_get_sni(remote);
1155 if (value) {
1156 if ((buf_len - ret) < sizeof(struct tlv))
1157 return 0;
1158 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_AUTHORITY, strlen(value), value);
1159 }
1160 }
1161
David Safb76832014-05-08 23:42:08 -04001162 if (srv->pp_opts & SRV_PP_V2_SSL) {
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001163 struct tlv_ssl *tlv;
1164 int ssl_tlv_len = 0;
David Safb76832014-05-08 23:42:08 -04001165 if ((buf_len - ret) < sizeof(struct tlv_ssl))
1166 return 0;
1167 tlv = (struct tlv_ssl *)&buf[ret];
1168 memset(tlv, 0, sizeof(struct tlv_ssl));
1169 ssl_tlv_len += sizeof(struct tlv_ssl);
1170 tlv->tlv.type = PP2_TYPE_SSL;
1171 if (ssl_sock_is_ssl(remote)) {
1172 tlv->client |= PP2_CLIENT_SSL;
Emmanuel Hocdet01da5712017-10-13 16:59:49 +02001173 value = ssl_sock_get_proto_version(remote);
David Safb76832014-05-08 23:42:08 -04001174 if (value) {
Emmanuel Hocdet8c0c34b2018-02-28 12:02:14 +01001175 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 -04001176 }
Dave McCowan328fb582014-07-30 10:39:13 -04001177 if (ssl_sock_get_cert_used_sess(remote)) {
1178 tlv->client |= PP2_CLIENT_CERT_SESS;
David Safb76832014-05-08 23:42:08 -04001179 tlv->verify = htonl(ssl_sock_get_verify_result(remote));
Dave McCowan328fb582014-07-30 10:39:13 -04001180 if (ssl_sock_get_cert_used_conn(remote))
1181 tlv->client |= PP2_CLIENT_CERT_CONN;
David Safb76832014-05-08 23:42:08 -04001182 }
1183 if (srv->pp_opts & SRV_PP_V2_SSL_CN) {
Willy Tarreau83061a82018-07-13 11:56:34 +02001184 struct buffer *cn_trash = get_trash_chunk();
Willy Tarreau3b9a0c92014-07-19 06:37:33 +02001185 if (ssl_sock_get_remote_common_name(remote, cn_trash) > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001186 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_CN,
1187 cn_trash->data,
1188 cn_trash->area);
David Safb76832014-05-08 23:42:08 -04001189 }
1190 }
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001191 if (srv->pp_opts & SRV_PP_V2_SSL_KEY_ALG) {
Willy Tarreau83061a82018-07-13 11:56:34 +02001192 struct buffer *pkey_trash = get_trash_chunk();
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001193 if (ssl_sock_get_pkey_algo(remote, pkey_trash) > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001194 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_KEY_ALG,
1195 pkey_trash->data,
1196 pkey_trash->area);
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001197 }
1198 }
1199 if (srv->pp_opts & SRV_PP_V2_SSL_SIG_ALG) {
1200 value = ssl_sock_get_cert_sig(remote);
1201 if (value) {
1202 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_SIG_ALG, strlen(value), value);
1203 }
1204 }
1205 if (srv->pp_opts & SRV_PP_V2_SSL_CIPHER) {
1206 value = ssl_sock_get_cipher_name(remote);
1207 if (value) {
1208 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_CIPHER, strlen(value), value);
1209 }
1210 }
David Safb76832014-05-08 23:42:08 -04001211 }
1212 tlv->tlv.length_hi = (uint16_t)(ssl_tlv_len - sizeof(struct tlv)) >> 8;
1213 tlv->tlv.length_lo = (uint16_t)(ssl_tlv_len - sizeof(struct tlv)) & 0x00ff;
1214 ret += ssl_tlv_len;
1215 }
1216#endif
1217
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001218#ifdef CONFIG_HAP_NS
1219 if (remote && (remote->proxy_netns)) {
1220 if ((buf_len - ret) < sizeof(struct tlv))
1221 return 0;
Emmanuel Hocdet571c7ac2017-10-31 18:24:05 +01001222 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 +01001223 }
1224#endif
1225
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001226 hdr->len = htons((uint16_t)(ret - PP2_HEADER_LEN));
David Safb76832014-05-08 23:42:08 -04001227
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001228 if (tlv_crc32c_p) {
1229 write_u32(tlv_crc32c_p, htonl(hash_crc32c(buf, ret)));
1230 }
1231
David Safb76832014-05-08 23:42:08 -04001232 return ret;
1233}
Emeric Brun4f603012017-01-05 15:11:44 +01001234
Willy Tarreau60ca10a2017-08-18 15:26:54 +02001235/* return the major HTTP version as 1 or 2 depending on how the request arrived
1236 * before being processed.
1237 */
1238static int
1239smp_fetch_fc_http_major(const struct arg *args, struct sample *smp, const char *kw, void *private)
1240{
1241 struct connection *conn = objt_conn(smp->sess->origin);
1242
1243 smp->data.type = SMP_T_SINT;
1244 smp->data.u.sint = (conn && strcmp(conn_get_mux_name(conn), "H2") == 0) ? 2 : 1;
1245 return 1;
1246}
1247
Emeric Brun4f603012017-01-05 15:11:44 +01001248/* fetch if the received connection used a PROXY protocol header */
1249int smp_fetch_fc_rcvd_proxy(const struct arg *args, struct sample *smp, const char *kw, void *private)
1250{
1251 struct connection *conn;
1252
1253 conn = objt_conn(smp->sess->origin);
1254 if (!conn)
1255 return 0;
1256
1257 if (!(conn->flags & CO_FL_CONNECTED)) {
1258 smp->flags |= SMP_F_MAY_CHANGE;
1259 return 0;
1260 }
1261
1262 smp->flags = 0;
1263 smp->data.type = SMP_T_BOOL;
1264 smp->data.u.sint = (conn->flags & CO_FL_RCVD_PROXY) ? 1 : 0;
1265
1266 return 1;
1267}
1268
1269/* Note: must not be declared <const> as its list will be overwritten.
1270 * Note: fetches that may return multiple types must be declared as the lowest
1271 * common denominator, the type that can be casted into all other ones. For
1272 * instance v4/v6 must be declared v4.
1273 */
1274static struct sample_fetch_kw_list sample_fetch_keywords = {ILH, {
Willy Tarreau60ca10a2017-08-18 15:26:54 +02001275 { "fc_http_major", smp_fetch_fc_http_major, 0, NULL, SMP_T_SINT, SMP_USE_L4CLI },
Emeric Brun4f603012017-01-05 15:11:44 +01001276 { "fc_rcvd_proxy", smp_fetch_fc_rcvd_proxy, 0, NULL, SMP_T_BOOL, SMP_USE_L4CLI },
1277 { /* END */ },
1278}};
1279
1280
1281__attribute__((constructor))
1282static void __connection_init(void)
1283{
1284 sample_register_fetches(&sample_fetch_keywords);
1285}