blob: b62cccecdac9f543c32ea551f5b983238c38c3c3 [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 Houchardaf4021e2018-08-09 13:06:55 +020067 int io_available = 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 Houchardfa8aa862018-10-10 18:25:41 +0200131 if (conn->send_wait != NULL) {
132 conn->send_wait->wait_reason &= ~SUB_CAN_SEND;
133 tasklet_wakeup(conn->send_wait->task);
134 conn->send_wait = NULL;
135 } else
136 io_available = 1;
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200137 }
Willy Tarreau59f98392012-07-06 14:13:49 +0200138
Willy Tarreau57ec32f2017-04-11 19:59:33 +0200139 /* The data transfer starts here and stops on error and handshakes. Note
140 * that we must absolutely test conn->xprt at each step in case it suddenly
141 * changes due to a quick unexpected close().
142 */
143 if (conn->xprt && fd_recv_ready(fd) &&
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200144 ((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 +0100145 /* force reporting of activity by clearing the previous flags :
146 * we'll have at least ERROR or CONNECTED at the end of an I/O,
147 * both of which will be detected below.
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200148 */
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100149 flags = 0;
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200150 if (conn->recv_wait) {
151 conn->recv_wait->wait_reason &= ~SUB_CAN_RECV;
152 tasklet_wakeup(conn->recv_wait->task);
153 conn->recv_wait = NULL;
154 } else
155 io_available = 1;
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200156 }
Willy Tarreau2da156f2012-07-23 15:07:23 +0200157
Willy Tarreauc76ae332012-07-12 15:32:13 +0200158 /* It may happen during the data phase that a handshake is
159 * enabled again (eg: SSL)
160 */
Willy Tarreaud6e999b2013-11-25 08:41:15 +0100161 if (unlikely(conn->flags & (CO_FL_HANDSHAKE | CO_FL_ERROR)))
Willy Tarreauc76ae332012-07-12 15:32:13 +0200162 goto process_handshake;
163
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100164 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreauf8deb0c2012-09-01 17:59:22 +0200165 /* still waiting for a connection to establish and nothing was
166 * attempted yet to probe the connection. Then let's retry the
167 * connect().
Willy Tarreau2da156f2012-07-23 15:07:23 +0200168 */
Willy Tarreau239d7182012-07-23 18:53:03 +0200169 if (!tcp_connect_probe(conn))
Willy Tarreauafad0e02012-08-09 14:45:22 +0200170 goto leave;
Willy Tarreau2da156f2012-07-23 15:07:23 +0200171 }
Willy Tarreau2c6be842012-07-06 17:12:34 +0200172 leave:
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100173 /* Verify if the connection just established. */
Willy Tarreau7bf3fa32017-03-14 20:19:29 +0100174 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED))))
175 conn->flags |= CO_FL_CONNECTED;
176
Willy Tarreau8e3c6ce2017-08-28 15:46:01 +0200177 /* The connection owner might want to be notified about failures to
178 * complete the handshake. The callback may fail and cause the
179 * connection to be destroyed, thus we must not use it anymore and
180 * should immediately leave instead. The caller must immediately
181 * unregister itself once called.
182 */
183 if (((conn->flags ^ flags) & CO_FL_NOTIFY_DONE) &&
184 conn->xprt_done_cb && conn->xprt_done_cb(conn) < 0)
185 return;
186
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100187 /* The wake callback is normally used to notify the data layer about
188 * data layer activity (successful send/recv), connection establishment,
189 * shutdown and fatal errors. We need to consider the following
190 * situations to wake up the data layer :
191 * - change among the CO_FL_NOTIFY_DATA flags :
192 * {DATA,SOCK}_{RD,WR}_SH, ERROR,
193 * - absence of any of {L4,L6}_CONN and CONNECTED, indicating the
194 * end of handshake and transition to CONNECTED
195 * - raise of CONNECTED with HANDSHAKE down
196 * - end of HANDSHAKE with CONNECTED set
197 * - regular data layer activity
198 *
199 * Note that the wake callback is allowed to release the connection and
200 * the fd (and return < 0 in this case).
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200201 */
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200202 if ((io_available || (((conn->flags ^ flags) & CO_FL_NOTIFY_DATA) ||
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100203 ((flags & (CO_FL_CONNECTED|CO_FL_HANDSHAKE)) != CO_FL_CONNECTED &&
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200204 (conn->flags & (CO_FL_CONNECTED|CO_FL_HANDSHAKE)) == CO_FL_CONNECTED))) &&
Olivier Houchard7505f942018-08-21 18:10:44 +0200205 conn->mux->wake && conn->mux->wake(conn) < 0)
Willy Tarreau7a798e52016-04-14 11:13:20 +0200206 return;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200207
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200208 /* 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 Houchard83a0cd82018-09-28 17:57:58 +0200334int conn_unsubscribe(struct connection *conn, int event_type, void *param)
335{
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200336 struct wait_event *sw;
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200337
338 if (event_type & SUB_CAN_RECV) {
339 sw = param;
340 if (sw->wait_reason & SUB_CAN_RECV) {
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200341 conn->recv_wait = NULL;
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200342 sw->wait_reason &= ~SUB_CAN_RECV;
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200343 }
344 }
345 if (event_type & SUB_CAN_SEND) {
346 sw = param;
347 if (sw->wait_reason & SUB_CAN_SEND) {
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200348 conn->send_wait = NULL;
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200349 sw->wait_reason &= ~SUB_CAN_SEND;
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200350 }
351 }
352 return 0;
353}
354
Olivier Houchard6ff20392018-07-17 18:46:31 +0200355int conn_subscribe(struct connection *conn, int event_type, void *param)
356{
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200357 struct wait_event *sw;
Olivier Houchard6ff20392018-07-17 18:46:31 +0200358
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200359 if (event_type & SUB_CAN_RECV) {
Olivier Houchard4cf7fb12018-08-02 19:23:05 +0200360 sw = param;
361 if (!(sw->wait_reason & SUB_CAN_RECV)) {
362 sw->wait_reason |= SUB_CAN_RECV;
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200363 conn->recv_wait = sw;
Olivier Houchard4cf7fb12018-08-02 19:23:05 +0200364 }
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200365 event_type &= ~SUB_CAN_RECV;
366 }
367 if (event_type & SUB_CAN_SEND) {
Olivier Houchard6ff20392018-07-17 18:46:31 +0200368 sw = param;
Olivier Houcharde1c6dbc2018-08-01 17:06:43 +0200369 if (!(sw->wait_reason & SUB_CAN_SEND)) {
370 sw->wait_reason |= SUB_CAN_SEND;
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200371 conn->send_wait = sw;
Olivier Houcharde1c6dbc2018-08-01 17:06:43 +0200372 }
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200373 event_type &= ~SUB_CAN_SEND;
Olivier Houchard6ff20392018-07-17 18:46:31 +0200374 }
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200375 if (event_type != 0)
376 return (-1);
377 return 0;
Olivier Houchard6ff20392018-07-17 18:46:31 +0200378}
379
Willy Tarreaud85c4852015-03-13 00:40:28 +0100380/* Drains possibly pending incoming data on the file descriptor attached to the
381 * connection and update the connection's flags accordingly. This is used to
382 * know whether we need to disable lingering on close. Returns non-zero if it
383 * is safe to close without disabling lingering, otherwise zero. The SOCK_RD_SH
384 * flag may also be updated if the incoming shutdown was reported by the drain()
385 * function.
386 */
387int conn_sock_drain(struct connection *conn)
388{
Willy Tarreaue215bba2018-08-24 14:31:53 +0200389 int turns = 2;
390 int len;
391
Willy Tarreaud85c4852015-03-13 00:40:28 +0100392 if (!conn_ctrl_ready(conn))
393 return 1;
394
395 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH))
396 return 1;
397
Willy Tarreaue215bba2018-08-24 14:31:53 +0200398 if (fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP))
399 goto shut;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100400
Willy Tarreaue215bba2018-08-24 14:31:53 +0200401 if (!fd_recv_ready(conn->handle.fd))
402 return 0;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100403
Willy Tarreaue215bba2018-08-24 14:31:53 +0200404 if (conn->ctrl->drain) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200405 if (conn->ctrl->drain(conn->handle.fd) <= 0)
Willy Tarreaud85c4852015-03-13 00:40:28 +0100406 return 0;
Willy Tarreaue215bba2018-08-24 14:31:53 +0200407 goto shut;
408 }
409
410 /* no drain function defined, use the generic one */
411
412 while (turns) {
413#ifdef MSG_TRUNC_CLEARS_INPUT
414 len = recv(conn->handle.fd, NULL, INT_MAX, MSG_DONTWAIT | MSG_NOSIGNAL | MSG_TRUNC);
415 if (len == -1 && errno == EFAULT)
416#endif
417 len = recv(conn->handle.fd, trash.area, trash.size,
418 MSG_DONTWAIT | MSG_NOSIGNAL);
419
420 if (len == 0)
421 goto shut;
422
423 if (len < 0) {
424 if (errno == EAGAIN) {
425 /* connection not closed yet */
426 fd_cant_recv(conn->handle.fd);
427 break;
428 }
429 if (errno == EINTR) /* oops, try again */
430 continue;
431 /* other errors indicate a dead connection, fine. */
432 goto shut;
433 }
434 /* OK we read some data, let's try again once */
435 turns--;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100436 }
437
Willy Tarreaue215bba2018-08-24 14:31:53 +0200438 /* some data are still present, give up */
439 return 0;
440
441 shut:
442 /* we're certain the connection was shut down */
443 fdtab[conn->handle.fd].linger_risk = 0;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100444 conn->flags |= CO_FL_SOCK_RD_SH;
445 return 1;
446}
447
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100448/*
449 * Get data length from tlv
450 */
451static int get_tlv_length(const struct tlv *src)
452{
453 return (src->length_hi << 8) | src->length_lo;
454}
455
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200456/* This handshake handler waits a PROXY protocol header at the beginning of the
457 * raw data stream. The header looks like this :
458 *
459 * "PROXY" <SP> PROTO <SP> SRC3 <SP> DST3 <SP> SRC4 <SP> <DST4> "\r\n"
460 *
461 * There must be exactly one space between each field. Fields are :
462 * - PROTO : layer 4 protocol, which must be "TCP4" or "TCP6".
463 * - SRC3 : layer 3 (eg: IP) source address in standard text form
464 * - DST3 : layer 3 (eg: IP) destination address in standard text form
465 * - SRC4 : layer 4 (eg: TCP port) source address in standard text form
466 * - DST4 : layer 4 (eg: TCP port) destination address in standard text form
467 *
468 * This line MUST be at the beginning of the buffer and MUST NOT wrap.
469 *
470 * The header line is small and in all cases smaller than the smallest normal
471 * TCP MSS. So it MUST always be delivered as one segment, which ensures we
472 * can safely use MSG_PEEK and avoid buffering.
473 *
474 * Once the data is fetched, the values are set in the connection's address
475 * fields, and data are removed from the socket's buffer. The function returns
476 * zero if it needs to wait for more data or if it fails, or 1 if it completed
477 * and removed itself.
478 */
479int conn_recv_proxy(struct connection *conn, int flag)
480{
481 char *line, *end;
Willy Tarreau77992672014-06-14 11:06:17 +0200482 struct proxy_hdr_v2 *hdr_v2;
483 const char v2sig[] = PP2_SIGNATURE;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100484 int tlv_length = 0;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200485 int tlv_offset = 0;
Willy Tarreaub406b872018-08-22 05:20:32 +0200486 int ret;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200487
488 /* we might have been called just after an asynchronous shutr */
489 if (conn->flags & CO_FL_SOCK_RD_SH)
490 goto fail;
491
Willy Tarreau3c728722014-01-23 13:50:42 +0100492 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200493 goto fail;
494
Willy Tarreau585744b2017-08-24 14:31:19 +0200495 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100496 return 0;
497
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200498 do {
Willy Tarreaub406b872018-08-22 05:20:32 +0200499 ret = recv(conn->handle.fd, trash.area, trash.size, MSG_PEEK);
500 if (ret < 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200501 if (errno == EINTR)
502 continue;
503 if (errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200504 fd_cant_recv(conn->handle.fd);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200505 return 0;
506 }
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100507 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200508 }
Willy Tarreaub406b872018-08-22 05:20:32 +0200509 trash.data = ret;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200510 } while (0);
511
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200512 if (!trash.data) {
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100513 /* client shutdown */
514 conn->err_code = CO_ER_PRX_EMPTY;
515 goto fail;
516 }
517
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200518 if (trash.data < 6)
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200519 goto missing;
520
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200521 line = trash.area;
522 end = trash.area + trash.data;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200523
524 /* Decode a possible proxy request, fail early if it does not match */
Willy Tarreau77992672014-06-14 11:06:17 +0200525 if (strncmp(line, "PROXY ", 6) != 0)
526 goto not_v1;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200527
528 line += 6;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200529 if (trash.data < 9) /* shortest possible line */
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200530 goto missing;
531
David CARLIER42ff05e2016-03-24 09:22:36 +0000532 if (memcmp(line, "TCP4 ", 5) == 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200533 u32 src3, dst3, sport, dport;
534
535 line += 5;
536
537 src3 = inetaddr_host_lim_ret(line, end, &line);
538 if (line == end)
539 goto missing;
540 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100541 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200542
543 dst3 = inetaddr_host_lim_ret(line, end, &line);
544 if (line == end)
545 goto missing;
546 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100547 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200548
549 sport = read_uint((const char **)&line, end);
550 if (line == end)
551 goto missing;
552 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100553 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200554
555 dport = read_uint((const char **)&line, end);
556 if (line > end - 2)
557 goto missing;
558 if (*line++ != '\r')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100559 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200560 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100561 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200562
563 /* update the session's addresses and mark them set */
564 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
565 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = htonl(src3);
566 ((struct sockaddr_in *)&conn->addr.from)->sin_port = htons(sport);
567
568 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
569 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = htonl(dst3);
570 ((struct sockaddr_in *)&conn->addr.to)->sin_port = htons(dport);
571 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
572 }
David CARLIER42ff05e2016-03-24 09:22:36 +0000573 else if (memcmp(line, "TCP6 ", 5) == 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200574 u32 sport, dport;
575 char *src_s;
576 char *dst_s, *sport_s, *dport_s;
577 struct in6_addr src3, dst3;
578
579 line += 5;
580
581 src_s = line;
582 dst_s = sport_s = dport_s = NULL;
583 while (1) {
584 if (line > end - 2) {
585 goto missing;
586 }
587 else if (*line == '\r') {
588 *line = 0;
589 line++;
590 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100591 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200592 break;
593 }
594
595 if (*line == ' ') {
596 *line = 0;
597 if (!dst_s)
598 dst_s = line + 1;
599 else if (!sport_s)
600 sport_s = line + 1;
601 else if (!dport_s)
602 dport_s = line + 1;
603 }
604 line++;
605 }
606
607 if (!dst_s || !sport_s || !dport_s)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100608 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200609
610 sport = read_uint((const char **)&sport_s,dport_s - 1);
611 if (*sport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100612 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200613
614 dport = read_uint((const char **)&dport_s,line - 2);
615 if (*dport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100616 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200617
618 if (inet_pton(AF_INET6, src_s, (void *)&src3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100619 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200620
621 if (inet_pton(AF_INET6, dst_s, (void *)&dst3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100622 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200623
624 /* update the session's addresses and mark them set */
625 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
626 memcpy(&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr, &src3, sizeof(struct in6_addr));
627 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = htons(sport);
628
629 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
630 memcpy(&((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr, &dst3, sizeof(struct in6_addr));
631 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = htons(dport);
632 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
633 }
Willy Tarreau4c20d292014-06-14 11:41:36 +0200634 else if (memcmp(line, "UNKNOWN\r\n", 9) == 0) {
635 /* This can be a UNIX socket forwarded by an haproxy upstream */
636 line += 9;
637 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200638 else {
Willy Tarreau4c20d292014-06-14 11:41:36 +0200639 /* The protocol does not match something known (TCP4/TCP6/UNKNOWN) */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100640 conn->err_code = CO_ER_PRX_BAD_PROTO;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200641 goto fail;
642 }
643
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200644 trash.data = line - trash.area;
Willy Tarreau77992672014-06-14 11:06:17 +0200645 goto eat_header;
646
647 not_v1:
648 /* try PPv2 */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200649 if (trash.data < PP2_HEADER_LEN)
Willy Tarreau77992672014-06-14 11:06:17 +0200650 goto missing;
651
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200652 hdr_v2 = (struct proxy_hdr_v2 *) trash.area;
Willy Tarreau77992672014-06-14 11:06:17 +0200653
654 if (memcmp(hdr_v2->sig, v2sig, PP2_SIGNATURE_LEN) != 0 ||
655 (hdr_v2->ver_cmd & PP2_VERSION_MASK) != PP2_VERSION) {
656 conn->err_code = CO_ER_PRX_NOT_HDR;
657 goto fail;
658 }
659
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200660 if (trash.data < PP2_HEADER_LEN + ntohs(hdr_v2->len))
Willy Tarreau77992672014-06-14 11:06:17 +0200661 goto missing;
662
663 switch (hdr_v2->ver_cmd & PP2_CMD_MASK) {
664 case 0x01: /* PROXY command */
665 switch (hdr_v2->fam) {
666 case 0x11: /* TCPv4 */
KOVACS Krisztianefd3aa92014-11-19 10:53:20 +0100667 if (ntohs(hdr_v2->len) < PP2_ADDR_LEN_INET)
668 goto bad_header;
669
Willy Tarreau77992672014-06-14 11:06:17 +0200670 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
671 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = hdr_v2->addr.ip4.src_addr;
672 ((struct sockaddr_in *)&conn->addr.from)->sin_port = hdr_v2->addr.ip4.src_port;
673 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
674 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = hdr_v2->addr.ip4.dst_addr;
675 ((struct sockaddr_in *)&conn->addr.to)->sin_port = hdr_v2->addr.ip4.dst_port;
676 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200677 tlv_offset = PP2_HEADER_LEN + PP2_ADDR_LEN_INET;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100678 tlv_length = ntohs(hdr_v2->len) - PP2_ADDR_LEN_INET;
Willy Tarreau77992672014-06-14 11:06:17 +0200679 break;
680 case 0x21: /* TCPv6 */
KOVACS Krisztianefd3aa92014-11-19 10:53:20 +0100681 if (ntohs(hdr_v2->len) < PP2_ADDR_LEN_INET6)
682 goto bad_header;
683
Willy Tarreau77992672014-06-14 11:06:17 +0200684 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
685 memcpy(&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr, hdr_v2->addr.ip6.src_addr, 16);
686 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = hdr_v2->addr.ip6.src_port;
687 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
688 memcpy(&((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr, hdr_v2->addr.ip6.dst_addr, 16);
689 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = hdr_v2->addr.ip6.dst_port;
690 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200691 tlv_offset = PP2_HEADER_LEN + PP2_ADDR_LEN_INET6;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100692 tlv_length = ntohs(hdr_v2->len) - PP2_ADDR_LEN_INET6;
Willy Tarreau77992672014-06-14 11:06:17 +0200693 break;
694 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100695
696 /* TLV parsing */
697 if (tlv_length > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200698 while (tlv_offset + TLV_HEADER_SIZE <= trash.data) {
699 const struct tlv *tlv_packet = (struct tlv *) &trash.area[tlv_offset];
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100700 const int tlv_len = get_tlv_length(tlv_packet);
701 tlv_offset += tlv_len + TLV_HEADER_SIZE;
702
703 switch (tlv_packet->type) {
Emmanuel Hocdet115df3e2018-02-05 16:23:23 +0100704 case PP2_TYPE_CRC32C: {
705 void *tlv_crc32c_p = (void *)tlv_packet->value;
706 uint32_t n_crc32c = ntohl(read_u32(tlv_crc32c_p));
707 write_u32(tlv_crc32c_p, 0);
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200708 if (hash_crc32c(trash.area, PP2_HEADER_LEN + ntohs(hdr_v2->len)) != n_crc32c)
Emmanuel Hocdet115df3e2018-02-05 16:23:23 +0100709 goto bad_header;
710 break;
711 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100712#ifdef CONFIG_HAP_NS
713 case PP2_TYPE_NETNS: {
714 const struct netns_entry *ns;
715 ns = netns_store_lookup((char*)tlv_packet->value, tlv_len);
716 if (ns)
717 conn->proxy_netns = ns;
718 break;
719 }
720#endif
721 default:
722 break;
723 }
724 }
725 }
726
Willy Tarreau77992672014-06-14 11:06:17 +0200727 /* unsupported protocol, keep local connection address */
728 break;
729 case 0x00: /* LOCAL command */
730 /* keep local connection address for LOCAL */
731 break;
732 default:
733 goto bad_header; /* not a supported command */
734 }
735
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200736 trash.data = PP2_HEADER_LEN + ntohs(hdr_v2->len);
Willy Tarreau77992672014-06-14 11:06:17 +0200737 goto eat_header;
738
739 eat_header:
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200740 /* remove the PROXY line from the request. For this we re-read the
741 * exact line at once. If we don't get the exact same result, we
742 * fail.
743 */
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200744 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200745 int len2 = recv(conn->handle.fd, trash.area, trash.data, 0);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200746 if (len2 < 0 && errno == EINTR)
747 continue;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200748 if (len2 != trash.data)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100749 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200750 } while (0);
751
752 conn->flags &= ~flag;
Emeric Brun4f603012017-01-05 15:11:44 +0100753 conn->flags |= CO_FL_RCVD_PROXY;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200754 return 1;
755
756 missing:
757 /* Missing data. Since we're using MSG_PEEK, we can only poll again if
758 * we have not read anything. Otherwise we need to fail because we won't
759 * be able to poll anymore.
760 */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100761 conn->err_code = CO_ER_PRX_TRUNCATED;
762 goto fail;
763
764 bad_header:
765 /* This is not a valid proxy protocol header */
766 conn->err_code = CO_ER_PRX_BAD_HDR;
767 goto fail;
768
769 recv_abort:
770 conn->err_code = CO_ER_PRX_ABORT;
Willy Tarreau26f4a042013-12-04 23:44:10 +0100771 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100772 goto fail;
773
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200774 fail:
Willy Tarreaud486ef52012-12-10 17:03:52 +0100775 __conn_sock_stop_both(conn);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200776 conn->flags |= CO_FL_ERROR;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200777 return 0;
778}
779
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100780/* This handshake handler waits a NetScaler Client IP insertion header
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000781 * at the beginning of the raw data stream. The header format is
782 * described in doc/netscaler-client-ip-insertion-protocol.txt
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100783 *
784 * This line MUST be at the beginning of the buffer and MUST NOT be
785 * fragmented.
786 *
787 * The header line is small and in all cases smaller than the smallest normal
788 * TCP MSS. So it MUST always be delivered as one segment, which ensures we
789 * can safely use MSG_PEEK and avoid buffering.
790 *
791 * Once the data is fetched, the values are set in the connection's address
792 * fields, and data are removed from the socket's buffer. The function returns
793 * zero if it needs to wait for more data or if it fails, or 1 if it completed
794 * and removed itself.
795 */
796int conn_recv_netscaler_cip(struct connection *conn, int flag)
797{
798 char *line;
Bertrand Jacquin7d668f92017-12-13 01:23:39 +0000799 uint32_t hdr_len;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100800 uint8_t ip_v;
Willy Tarreaub406b872018-08-22 05:20:32 +0200801 int ret;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100802
803 /* we might have been called just after an asynchronous shutr */
804 if (conn->flags & CO_FL_SOCK_RD_SH)
805 goto fail;
806
807 if (!conn_ctrl_ready(conn))
808 goto fail;
809
Willy Tarreau585744b2017-08-24 14:31:19 +0200810 if (!fd_recv_ready(conn->handle.fd))
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100811 return 0;
812
813 do {
Willy Tarreaub406b872018-08-22 05:20:32 +0200814 ret = recv(conn->handle.fd, trash.area, trash.size, MSG_PEEK);
815 if (ret < 0) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100816 if (errno == EINTR)
817 continue;
818 if (errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200819 fd_cant_recv(conn->handle.fd);
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100820 return 0;
821 }
822 goto recv_abort;
823 }
Willy Tarreaub406b872018-08-22 05:20:32 +0200824 trash.data = ret;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100825 } while (0);
826
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200827 if (!trash.data) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100828 /* client shutdown */
829 conn->err_code = CO_ER_CIP_EMPTY;
830 goto fail;
831 }
832
833 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000834 * CIP magic, header length or
835 * CIP magic, CIP length, CIP type, header length */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200836 if (trash.data < 12)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100837 goto missing;
838
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200839 line = trash.area;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100840
841 /* Decode a possible NetScaler Client IP request, fail early if
842 * it does not match */
Willy Tarreau55e0da62018-09-20 11:26:52 +0200843 if (ntohl(*(uint32_t *)line) != __objt_listener(conn->target)->bind_conf->ns_cip_magic)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100844 goto bad_magic;
845
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000846 /* Legacy CIP protocol */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200847 if ((trash.area[8] & 0xD0) == 0x40) {
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000848 hdr_len = ntohl(*(uint32_t *)(line+4));
849 line += 8;
850 }
851 /* Standard CIP protocol */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200852 else if (trash.area[8] == 0x00) {
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000853 hdr_len = ntohs(*(uint32_t *)(line+10));
854 line += 12;
855 }
856 /* Unknown CIP protocol */
857 else {
858 conn->err_code = CO_ER_CIP_BAD_PROTO;
859 goto fail;
860 }
861
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100862 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000863 * a minimal IP header */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200864 if (trash.data < 20)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100865 goto missing;
866
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100867 /* Get IP version from the first four bits */
868 ip_v = (*line & 0xf0) >> 4;
869
870 if (ip_v == 4) {
871 struct ip *hdr_ip4;
David Carlier3015a2e2016-07-04 22:51:33 +0100872 struct my_tcphdr *hdr_tcp;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100873
874 hdr_ip4 = (struct ip *)line;
875
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200876 if (trash.data < 40 || trash.data < hdr_len) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100877 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin67de5a22017-12-13 01:15:05 +0000878 * IPv4 header, TCP header */
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100879 goto missing;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000880 }
881 else if (hdr_ip4->ip_p != IPPROTO_TCP) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100882 /* The protocol does not include a TCP header */
883 conn->err_code = CO_ER_CIP_BAD_PROTO;
884 goto fail;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000885 }
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100886
David Carlier3015a2e2016-07-04 22:51:33 +0100887 hdr_tcp = (struct my_tcphdr *)(line + (hdr_ip4->ip_hl * 4));
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100888
889 /* update the session's addresses and mark them set */
890 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
891 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = hdr_ip4->ip_src.s_addr;
892 ((struct sockaddr_in *)&conn->addr.from)->sin_port = hdr_tcp->source;
893
894 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
895 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = hdr_ip4->ip_dst.s_addr;
896 ((struct sockaddr_in *)&conn->addr.to)->sin_port = hdr_tcp->dest;
897
898 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
899 }
900 else if (ip_v == 6) {
901 struct ip6_hdr *hdr_ip6;
David Carlier3015a2e2016-07-04 22:51:33 +0100902 struct my_tcphdr *hdr_tcp;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100903
904 hdr_ip6 = (struct ip6_hdr *)line;
905
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200906 if (trash.data < 60 || trash.data < hdr_len) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100907 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin67de5a22017-12-13 01:15:05 +0000908 * IPv6 header, TCP header */
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100909 goto missing;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000910 }
911 else if (hdr_ip6->ip6_nxt != IPPROTO_TCP) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100912 /* The protocol does not include a TCP header */
913 conn->err_code = CO_ER_CIP_BAD_PROTO;
914 goto fail;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000915 }
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100916
David Carlier3015a2e2016-07-04 22:51:33 +0100917 hdr_tcp = (struct my_tcphdr *)(line + sizeof(struct ip6_hdr));
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100918
919 /* update the session's addresses and mark them set */
920 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
921 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr = hdr_ip6->ip6_src;
922 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = hdr_tcp->source;
923
924 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
925 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr = hdr_ip6->ip6_dst;
926 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = hdr_tcp->dest;
927
928 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
929 }
930 else {
931 /* The protocol does not match something known (IPv4/IPv6) */
932 conn->err_code = CO_ER_CIP_BAD_PROTO;
933 goto fail;
934 }
935
Bertrand Jacquin7d668f92017-12-13 01:23:39 +0000936 line += hdr_len;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200937 trash.data = line - trash.area;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100938
939 /* remove the NetScaler Client IP header from the request. For this
940 * we re-read the exact line at once. If we don't get the exact same
941 * result, we fail.
942 */
943 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200944 int len2 = recv(conn->handle.fd, trash.area, trash.data, 0);
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100945 if (len2 < 0 && errno == EINTR)
946 continue;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200947 if (len2 != trash.data)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100948 goto recv_abort;
949 } while (0);
950
951 conn->flags &= ~flag;
952 return 1;
953
954 missing:
955 /* Missing data. Since we're using MSG_PEEK, we can only poll again if
956 * we have not read anything. Otherwise we need to fail because we won't
957 * be able to poll anymore.
958 */
959 conn->err_code = CO_ER_CIP_TRUNCATED;
960 goto fail;
961
962 bad_magic:
963 conn->err_code = CO_ER_CIP_BAD_MAGIC;
964 goto fail;
965
966 recv_abort:
967 conn->err_code = CO_ER_CIP_ABORT;
968 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
969 goto fail;
970
971 fail:
972 __conn_sock_stop_both(conn);
973 conn->flags |= CO_FL_ERROR;
974 return 0;
975}
976
Ilya Shipitsinca56fce2018-09-15 00:50:05 +0500977/* Note: <remote> is explicitly allowed to be NULL */
David Safb76832014-05-08 23:42:08 -0400978int make_proxy_line(char *buf, int buf_len, struct server *srv, struct connection *remote)
979{
980 int ret = 0;
981
982 if (srv && (srv->pp_opts & SRV_PP_V2)) {
983 ret = make_proxy_line_v2(buf, buf_len, srv, remote);
984 }
985 else {
986 if (remote)
987 ret = make_proxy_line_v1(buf, buf_len, &remote->addr.from, &remote->addr.to);
988 else
989 ret = make_proxy_line_v1(buf, buf_len, NULL, NULL);
990 }
991
992 return ret;
993}
994
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200995/* Makes a PROXY protocol line from the two addresses. The output is sent to
996 * buffer <buf> for a maximum size of <buf_len> (including the trailing zero).
997 * It returns the number of bytes composing this line (including the trailing
998 * LF), or zero in case of failure (eg: not enough space). It supports TCP4,
Willy Tarreau2e1401a2013-10-01 11:41:55 +0200999 * TCP6 and "UNKNOWN" formats. If any of <src> or <dst> is null, UNKNOWN is
1000 * emitted as well.
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001001 */
David Safb76832014-05-08 23:42:08 -04001002int make_proxy_line_v1(char *buf, int buf_len, struct sockaddr_storage *src, struct sockaddr_storage *dst)
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001003{
1004 int ret = 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001005 char * protocol;
1006 char src_str[MAX(INET_ADDRSTRLEN, INET6_ADDRSTRLEN)];
1007 char dst_str[MAX(INET_ADDRSTRLEN, INET6_ADDRSTRLEN)];
1008 in_port_t src_port;
1009 in_port_t dst_port;
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001010
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001011 if ( !src
1012 || !dst
1013 || (src->ss_family != AF_INET && src->ss_family != AF_INET6)
1014 || (dst->ss_family != AF_INET && dst->ss_family != AF_INET6)) {
1015 /* unknown family combination */
1016 ret = snprintf(buf, buf_len, "PROXY UNKNOWN\r\n");
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001017 if (ret >= buf_len)
1018 return 0;
1019
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001020 return ret;
1021 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001022
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001023 /* IPv4 for both src and dst */
1024 if (src->ss_family == AF_INET && dst->ss_family == AF_INET) {
1025 protocol = "TCP4";
1026 if (!inet_ntop(AF_INET, &((struct sockaddr_in *)src)->sin_addr, src_str, sizeof(src_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001027 return 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001028 src_port = ((struct sockaddr_in *)src)->sin_port;
1029 if (!inet_ntop(AF_INET, &((struct sockaddr_in *)dst)->sin_addr, dst_str, sizeof(dst_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001030 return 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001031 dst_port = ((struct sockaddr_in *)dst)->sin_port;
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001032 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001033 /* IPv6 for at least one of src and dst */
1034 else {
1035 struct in6_addr tmp;
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001036
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001037 protocol = "TCP6";
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001038
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001039 if (src->ss_family == AF_INET) {
1040 /* Convert src to IPv6 */
1041 v4tov6(&tmp, &((struct sockaddr_in *)src)->sin_addr);
1042 src_port = ((struct sockaddr_in *)src)->sin_port;
1043 }
1044 else {
1045 tmp = ((struct sockaddr_in6 *)src)->sin6_addr;
1046 src_port = ((struct sockaddr_in6 *)src)->sin6_port;
1047 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001048
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001049 if (!inet_ntop(AF_INET6, &tmp, src_str, sizeof(src_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001050 return 0;
1051
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001052 if (dst->ss_family == AF_INET) {
1053 /* Convert dst to IPv6 */
1054 v4tov6(&tmp, &((struct sockaddr_in *)dst)->sin_addr);
1055 dst_port = ((struct sockaddr_in *)dst)->sin_port;
1056 }
1057 else {
1058 tmp = ((struct sockaddr_in6 *)dst)->sin6_addr;
1059 dst_port = ((struct sockaddr_in6 *)dst)->sin6_port;
1060 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001061
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001062 if (!inet_ntop(AF_INET6, &tmp, dst_str, sizeof(dst_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001063 return 0;
1064 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001065
1066 ret = snprintf(buf, buf_len, "PROXY %s %s %s %u %u\r\n", protocol, src_str, dst_str, ntohs(src_port), ntohs(dst_port));
1067 if (ret >= buf_len)
1068 return 0;
1069
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001070 return ret;
1071}
David Safb76832014-05-08 23:42:08 -04001072
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001073static int make_tlv(char *dest, int dest_len, char type, uint16_t length, const char *value)
David Safb76832014-05-08 23:42:08 -04001074{
1075 struct tlv *tlv;
1076
1077 if (!dest || (length + sizeof(*tlv) > dest_len))
1078 return 0;
1079
1080 tlv = (struct tlv *)dest;
1081
1082 tlv->type = type;
1083 tlv->length_hi = length >> 8;
1084 tlv->length_lo = length & 0x00ff;
1085 memcpy(tlv->value, value, length);
1086 return length + sizeof(*tlv);
1087}
David Safb76832014-05-08 23:42:08 -04001088
Ilya Shipitsinca56fce2018-09-15 00:50:05 +05001089/* Note: <remote> is explicitly allowed to be NULL */
David Safb76832014-05-08 23:42:08 -04001090int make_proxy_line_v2(char *buf, int buf_len, struct server *srv, struct connection *remote)
1091{
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001092 const char pp2_signature[] = PP2_SIGNATURE;
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001093 void *tlv_crc32c_p = NULL;
David Safb76832014-05-08 23:42:08 -04001094 int ret = 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001095 struct proxy_hdr_v2 *hdr = (struct proxy_hdr_v2 *)buf;
Vincent Bernat6e615892016-05-18 16:17:44 +02001096 struct sockaddr_storage null_addr = { .ss_family = 0 };
David Safb76832014-05-08 23:42:08 -04001097 struct sockaddr_storage *src = &null_addr;
1098 struct sockaddr_storage *dst = &null_addr;
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001099 const char *value;
1100 int value_len;
David Safb76832014-05-08 23:42:08 -04001101
1102 if (buf_len < PP2_HEADER_LEN)
1103 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001104 memcpy(hdr->sig, pp2_signature, PP2_SIGNATURE_LEN);
David Safb76832014-05-08 23:42:08 -04001105
1106 if (remote) {
1107 src = &remote->addr.from;
1108 dst = &remote->addr.to;
1109 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001110
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001111 /* At least one of src or dst is not of AF_INET or AF_INET6 */
1112 if ( !src
1113 || !dst
1114 || (src->ss_family != AF_INET && src->ss_family != AF_INET6)
1115 || (dst->ss_family != AF_INET && dst->ss_family != AF_INET6)) {
David Safb76832014-05-08 23:42:08 -04001116 if (buf_len < PP2_HDR_LEN_UNSPEC)
1117 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001118 hdr->ver_cmd = PP2_VERSION | PP2_CMD_LOCAL;
1119 hdr->fam = PP2_FAM_UNSPEC | PP2_TRANS_UNSPEC;
David Safb76832014-05-08 23:42:08 -04001120 ret = PP2_HDR_LEN_UNSPEC;
1121 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001122 else {
1123 /* IPv4 for both src and dst */
1124 if (src->ss_family == AF_INET && dst->ss_family == AF_INET) {
1125 if (buf_len < PP2_HDR_LEN_INET)
1126 return 0;
1127 hdr->ver_cmd = PP2_VERSION | PP2_CMD_PROXY;
1128 hdr->fam = PP2_FAM_INET | PP2_TRANS_STREAM;
1129 hdr->addr.ip4.src_addr = ((struct sockaddr_in *)src)->sin_addr.s_addr;
1130 hdr->addr.ip4.src_port = ((struct sockaddr_in *)src)->sin_port;
1131 hdr->addr.ip4.dst_addr = ((struct sockaddr_in *)dst)->sin_addr.s_addr;
1132 hdr->addr.ip4.dst_port = ((struct sockaddr_in *)dst)->sin_port;
1133 ret = PP2_HDR_LEN_INET;
1134 }
1135 /* IPv6 for at least one of src and dst */
1136 else {
1137 struct in6_addr tmp;
1138
1139 if (buf_len < PP2_HDR_LEN_INET6)
1140 return 0;
1141 hdr->ver_cmd = PP2_VERSION | PP2_CMD_PROXY;
1142 hdr->fam = PP2_FAM_INET6 | PP2_TRANS_STREAM;
1143 if (src->ss_family == AF_INET) {
1144 v4tov6(&tmp, &((struct sockaddr_in *)src)->sin_addr);
1145 memcpy(hdr->addr.ip6.src_addr, &tmp, 16);
1146 hdr->addr.ip6.src_port = ((struct sockaddr_in *)src)->sin_port;
1147 }
1148 else {
1149 memcpy(hdr->addr.ip6.src_addr, &((struct sockaddr_in6 *)src)->sin6_addr, 16);
1150 hdr->addr.ip6.src_port = ((struct sockaddr_in6 *)src)->sin6_port;
1151 }
1152 if (dst->ss_family == AF_INET) {
1153 v4tov6(&tmp, &((struct sockaddr_in *)dst)->sin_addr);
1154 memcpy(hdr->addr.ip6.dst_addr, &tmp, 16);
1155 hdr->addr.ip6.src_port = ((struct sockaddr_in *)src)->sin_port;
1156 }
1157 else {
1158 memcpy(hdr->addr.ip6.dst_addr, &((struct sockaddr_in6 *)dst)->sin6_addr, 16);
1159 hdr->addr.ip6.dst_port = ((struct sockaddr_in6 *)dst)->sin6_port;
1160 }
1161
1162 ret = PP2_HDR_LEN_INET6;
1163 }
1164 }
David Safb76832014-05-08 23:42:08 -04001165
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001166 if (srv->pp_opts & SRV_PP_V2_CRC32C) {
1167 uint32_t zero_crc32c = 0;
1168 if ((buf_len - ret) < sizeof(struct tlv))
1169 return 0;
1170 tlv_crc32c_p = (void *)((struct tlv *)&buf[ret])->value;
1171 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_CRC32C, sizeof(zero_crc32c), (const char *)&zero_crc32c);
1172 }
1173
Ilya Shipitsinca56fce2018-09-15 00:50:05 +05001174 if (remote && conn_get_alpn(remote, &value, &value_len)) {
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001175 if ((buf_len - ret) < sizeof(struct tlv))
1176 return 0;
Emmanuel Hocdet571c7ac2017-10-31 18:24:05 +01001177 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_ALPN, value_len, value);
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001178 }
1179
David Safb76832014-05-08 23:42:08 -04001180#ifdef USE_OPENSSL
Emmanuel Hocdet253c3b72018-02-01 18:29:59 +01001181 if (srv->pp_opts & SRV_PP_V2_AUTHORITY) {
1182 value = ssl_sock_get_sni(remote);
1183 if (value) {
1184 if ((buf_len - ret) < sizeof(struct tlv))
1185 return 0;
1186 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_AUTHORITY, strlen(value), value);
1187 }
1188 }
1189
David Safb76832014-05-08 23:42:08 -04001190 if (srv->pp_opts & SRV_PP_V2_SSL) {
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001191 struct tlv_ssl *tlv;
1192 int ssl_tlv_len = 0;
David Safb76832014-05-08 23:42:08 -04001193 if ((buf_len - ret) < sizeof(struct tlv_ssl))
1194 return 0;
1195 tlv = (struct tlv_ssl *)&buf[ret];
1196 memset(tlv, 0, sizeof(struct tlv_ssl));
1197 ssl_tlv_len += sizeof(struct tlv_ssl);
1198 tlv->tlv.type = PP2_TYPE_SSL;
1199 if (ssl_sock_is_ssl(remote)) {
1200 tlv->client |= PP2_CLIENT_SSL;
Emmanuel Hocdet01da5712017-10-13 16:59:49 +02001201 value = ssl_sock_get_proto_version(remote);
David Safb76832014-05-08 23:42:08 -04001202 if (value) {
Emmanuel Hocdet8c0c34b2018-02-28 12:02:14 +01001203 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 -04001204 }
Dave McCowan328fb582014-07-30 10:39:13 -04001205 if (ssl_sock_get_cert_used_sess(remote)) {
1206 tlv->client |= PP2_CLIENT_CERT_SESS;
David Safb76832014-05-08 23:42:08 -04001207 tlv->verify = htonl(ssl_sock_get_verify_result(remote));
Dave McCowan328fb582014-07-30 10:39:13 -04001208 if (ssl_sock_get_cert_used_conn(remote))
1209 tlv->client |= PP2_CLIENT_CERT_CONN;
David Safb76832014-05-08 23:42:08 -04001210 }
1211 if (srv->pp_opts & SRV_PP_V2_SSL_CN) {
Willy Tarreau83061a82018-07-13 11:56:34 +02001212 struct buffer *cn_trash = get_trash_chunk();
Willy Tarreau3b9a0c92014-07-19 06:37:33 +02001213 if (ssl_sock_get_remote_common_name(remote, cn_trash) > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001214 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_CN,
1215 cn_trash->data,
1216 cn_trash->area);
David Safb76832014-05-08 23:42:08 -04001217 }
1218 }
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001219 if (srv->pp_opts & SRV_PP_V2_SSL_KEY_ALG) {
Willy Tarreau83061a82018-07-13 11:56:34 +02001220 struct buffer *pkey_trash = get_trash_chunk();
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001221 if (ssl_sock_get_pkey_algo(remote, pkey_trash) > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001222 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_KEY_ALG,
1223 pkey_trash->data,
1224 pkey_trash->area);
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001225 }
1226 }
1227 if (srv->pp_opts & SRV_PP_V2_SSL_SIG_ALG) {
1228 value = ssl_sock_get_cert_sig(remote);
1229 if (value) {
1230 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_SIG_ALG, strlen(value), value);
1231 }
1232 }
1233 if (srv->pp_opts & SRV_PP_V2_SSL_CIPHER) {
1234 value = ssl_sock_get_cipher_name(remote);
1235 if (value) {
1236 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_CIPHER, strlen(value), value);
1237 }
1238 }
David Safb76832014-05-08 23:42:08 -04001239 }
1240 tlv->tlv.length_hi = (uint16_t)(ssl_tlv_len - sizeof(struct tlv)) >> 8;
1241 tlv->tlv.length_lo = (uint16_t)(ssl_tlv_len - sizeof(struct tlv)) & 0x00ff;
1242 ret += ssl_tlv_len;
1243 }
1244#endif
1245
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001246#ifdef CONFIG_HAP_NS
1247 if (remote && (remote->proxy_netns)) {
1248 if ((buf_len - ret) < sizeof(struct tlv))
1249 return 0;
Emmanuel Hocdet571c7ac2017-10-31 18:24:05 +01001250 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 +01001251 }
1252#endif
1253
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001254 hdr->len = htons((uint16_t)(ret - PP2_HEADER_LEN));
David Safb76832014-05-08 23:42:08 -04001255
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001256 if (tlv_crc32c_p) {
1257 write_u32(tlv_crc32c_p, htonl(hash_crc32c(buf, ret)));
1258 }
1259
David Safb76832014-05-08 23:42:08 -04001260 return ret;
1261}
Emeric Brun4f603012017-01-05 15:11:44 +01001262
Willy Tarreau60ca10a2017-08-18 15:26:54 +02001263/* return the major HTTP version as 1 or 2 depending on how the request arrived
1264 * before being processed.
1265 */
1266static int
1267smp_fetch_fc_http_major(const struct arg *args, struct sample *smp, const char *kw, void *private)
1268{
1269 struct connection *conn = objt_conn(smp->sess->origin);
1270
1271 smp->data.type = SMP_T_SINT;
1272 smp->data.u.sint = (conn && strcmp(conn_get_mux_name(conn), "H2") == 0) ? 2 : 1;
1273 return 1;
1274}
1275
Emeric Brun4f603012017-01-05 15:11:44 +01001276/* fetch if the received connection used a PROXY protocol header */
1277int smp_fetch_fc_rcvd_proxy(const struct arg *args, struct sample *smp, const char *kw, void *private)
1278{
1279 struct connection *conn;
1280
1281 conn = objt_conn(smp->sess->origin);
1282 if (!conn)
1283 return 0;
1284
1285 if (!(conn->flags & CO_FL_CONNECTED)) {
1286 smp->flags |= SMP_F_MAY_CHANGE;
1287 return 0;
1288 }
1289
1290 smp->flags = 0;
1291 smp->data.type = SMP_T_BOOL;
1292 smp->data.u.sint = (conn->flags & CO_FL_RCVD_PROXY) ? 1 : 0;
1293
1294 return 1;
1295}
1296
1297/* Note: must not be declared <const> as its list will be overwritten.
1298 * Note: fetches that may return multiple types must be declared as the lowest
1299 * common denominator, the type that can be casted into all other ones. For
1300 * instance v4/v6 must be declared v4.
1301 */
1302static struct sample_fetch_kw_list sample_fetch_keywords = {ILH, {
Willy Tarreau60ca10a2017-08-18 15:26:54 +02001303 { "fc_http_major", smp_fetch_fc_http_major, 0, NULL, SMP_T_SINT, SMP_USE_L4CLI },
Emeric Brun4f603012017-01-05 15:11:44 +01001304 { "fc_rcvd_proxy", smp_fetch_fc_rcvd_proxy, 0, NULL, SMP_T_BOOL, SMP_USE_L4CLI },
1305 { /* END */ },
1306}};
1307
1308
1309__attribute__((constructor))
1310static void __connection_init(void)
1311{
1312 sample_register_fetches(&sample_fetch_keywords);
1313}