blob: 26bcadd2256be68c639f551386aee13e3781329d [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>
Willy Tarreau59f98392012-07-06 14:13:49 +020018
Willy Tarreauc5788912012-08-24 18:12:41 +020019#include <proto/connection.h>
Willy Tarreaudd2f85e2012-09-02 22:34:23 +020020#include <proto/fd.h>
Willy Tarreau5f1504f2012-10-04 23:55:57 +020021#include <proto/frontend.h>
Willy Tarreau2da156f2012-07-23 15:07:23 +020022#include <proto/proto_tcp.h>
Willy Tarreau2c6be842012-07-06 17:12:34 +020023#include <proto/stream_interface.h>
Emeric Brun4f603012017-01-05 15:11:44 +010024#include <proto/sample.h>
Willy Tarreau59f98392012-07-06 14:13:49 +020025
Emeric Brun46591952012-05-18 15:47:34 +020026#ifdef USE_OPENSSL
27#include <proto/ssl_sock.h>
28#endif
29
Willy Tarreauf2943dc2012-10-26 20:10:28 +020030struct pool_head *pool2_connection;
Olivier Houcharde2b40b92017-09-13 18:30:23 +020031struct pool_head *pool2_connstream;
Willy Tarreau13e14102016-12-22 20:25:26 +010032struct xprt_ops *registered_xprt[XPRT_ENTRIES] = { NULL, };
Willy Tarreauf2943dc2012-10-26 20:10:28 +020033
Willy Tarreau2386be62017-09-21 19:40:52 +020034/* List head of all known muxes for ALPN */
35struct alpn_mux_list alpn_mux_list = {
36 .list = LIST_HEAD_INIT(alpn_mux_list.list)
37};
38
Willy Tarreauf2943dc2012-10-26 20:10:28 +020039/* perform minimal intializations, report 0 in case of error, 1 if OK. */
40int init_connection()
41{
42 pool2_connection = create_pool("connection", sizeof (struct connection), MEM_F_SHARED);
Olivier Houcharde2b40b92017-09-13 18:30:23 +020043 if (!pool2_connection)
44 goto fail_conn;
45
46 pool2_connstream = create_pool("conn_stream", sizeof(struct conn_stream), MEM_F_SHARED);
47 if (!pool2_connstream)
48 goto fail_cs;
49
50 return 1;
51 fail_cs:
52 pool_destroy2(pool2_connection);
53 pool2_connection = NULL;
54 fail_conn:
55 return 0;
Willy Tarreauf2943dc2012-10-26 20:10:28 +020056}
57
Willy Tarreau59f98392012-07-06 14:13:49 +020058/* I/O callback for fd-based connections. It calls the read/write handlers
Willy Tarreau7a798e52016-04-14 11:13:20 +020059 * provided by the connection's sock_ops, which must be valid.
Willy Tarreau59f98392012-07-06 14:13:49 +020060 */
Willy Tarreau7a798e52016-04-14 11:13:20 +020061void conn_fd_handler(int fd)
Willy Tarreau59f98392012-07-06 14:13:49 +020062{
Willy Tarreau80184712012-07-06 14:54:49 +020063 struct connection *conn = fdtab[fd].owner;
Willy Tarreau9e272bf2012-10-03 21:04:48 +020064 unsigned int flags;
Willy Tarreau59f98392012-07-06 14:13:49 +020065
Willy Tarreauc76ae332012-07-12 15:32:13 +020066 if (unlikely(!conn))
Willy Tarreau7a798e52016-04-14 11:13:20 +020067 return;
Willy Tarreau59f98392012-07-06 14:13:49 +020068
Willy Tarreau7d281492012-12-16 19:19:13 +010069 conn_refresh_polling_flags(conn);
Willy Tarreau916e12d2017-10-25 09:22:43 +020070 conn->flags |= CO_FL_WILL_UPDATE;
71
Willy Tarreau7d281492012-12-16 19:19:13 +010072 flags = conn->flags & ~CO_FL_ERROR; /* ensure to call the wake handler upon error */
Willy Tarreaud29a0662012-12-10 16:33:38 +010073
Willy Tarreauc76ae332012-07-12 15:32:13 +020074 process_handshake:
Willy Tarreauf9dabec2012-08-17 17:33:53 +020075 /* The handshake callbacks are called in sequence. If either of them is
76 * missing something, it must enable the required polling at the socket
77 * layer of the connection. Polling state is not guaranteed when entering
78 * these handlers, so any handshake handler which does not complete its
Willy Tarreaud6e999b2013-11-25 08:41:15 +010079 * work must explicitly disable events it's not interested in. Error
80 * handling is also performed here in order to reduce the number of tests
81 * around.
Willy Tarreauf9dabec2012-08-17 17:33:53 +020082 */
Willy Tarreaud6e999b2013-11-25 08:41:15 +010083 while (unlikely(conn->flags & (CO_FL_HANDSHAKE | CO_FL_ERROR))) {
Willy Tarreau310987a2014-01-22 19:46:33 +010084 if (unlikely(conn->flags & CO_FL_ERROR))
Willy Tarreau2c6be842012-07-06 17:12:34 +020085 goto leave;
Willy Tarreau59f98392012-07-06 14:13:49 +020086
Bertrand Jacquin93b227d2016-06-04 15:11:10 +010087 if (conn->flags & CO_FL_ACCEPT_CIP)
88 if (!conn_recv_netscaler_cip(conn, CO_FL_ACCEPT_CIP))
89 goto leave;
90
Willy Tarreau22cda212012-08-31 17:43:29 +020091 if (conn->flags & CO_FL_ACCEPT_PROXY)
92 if (!conn_recv_proxy(conn, CO_FL_ACCEPT_PROXY))
93 goto leave;
94
Willy Tarreau57cd3e42013-10-24 22:01:26 +020095 if (conn->flags & CO_FL_SEND_PROXY)
96 if (!conn_si_send_proxy(conn, CO_FL_SEND_PROXY))
Willy Tarreau5f1504f2012-10-04 23:55:57 +020097 goto leave;
Emeric Brun46591952012-05-18 15:47:34 +020098#ifdef USE_OPENSSL
99 if (conn->flags & CO_FL_SSL_WAIT_HS)
100 if (!ssl_sock_handshake(conn, CO_FL_SSL_WAIT_HS))
101 goto leave;
102#endif
Willy Tarreauc76ae332012-07-12 15:32:13 +0200103 }
104
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200105 /* Once we're purely in the data phase, we disable handshake polling */
106 if (!(conn->flags & CO_FL_POLL_SOCK))
107 __conn_sock_stop_both(conn);
Willy Tarreauc76ae332012-07-12 15:32:13 +0200108
Willy Tarreau8e3c6ce2017-08-28 15:46:01 +0200109 /* The connection owner might want to be notified about an end of
110 * handshake indicating the connection is ready, before we proceed with
111 * any data exchange. The callback may fail and cause the connection to
112 * be destroyed, thus we must not use it anymore and should immediately
113 * leave instead. The caller must immediately unregister itself once
114 * called.
Willy Tarreau2542b532012-08-31 16:01:23 +0200115 */
Willy Tarreau8e3c6ce2017-08-28 15:46:01 +0200116 if (conn->xprt_done_cb && conn->xprt_done_cb(conn) < 0)
Willy Tarreau7a798e52016-04-14 11:13:20 +0200117 return;
Willy Tarreau2542b532012-08-31 16:01:23 +0200118
Willy Tarreau57ec32f2017-04-11 19:59:33 +0200119 if (conn->xprt && fd_send_ready(fd) &&
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200120 ((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 +0100121 /* force reporting of activity by clearing the previous flags :
122 * we'll have at least ERROR or CONNECTED at the end of an I/O,
123 * both of which will be detected below.
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200124 */
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100125 flags = 0;
Willy Tarreau53a47662017-08-28 10:53:00 +0200126 conn->mux->send(conn);
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200127 }
Willy Tarreau59f98392012-07-06 14:13:49 +0200128
Willy Tarreau57ec32f2017-04-11 19:59:33 +0200129 /* The data transfer starts here and stops on error and handshakes. Note
130 * that we must absolutely test conn->xprt at each step in case it suddenly
131 * changes due to a quick unexpected close().
132 */
133 if (conn->xprt && fd_recv_ready(fd) &&
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200134 ((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 +0100135 /* force reporting of activity by clearing the previous flags :
136 * we'll have at least ERROR or CONNECTED at the end of an I/O,
137 * both of which will be detected below.
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200138 */
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100139 flags = 0;
Willy Tarreau53a47662017-08-28 10:53:00 +0200140 conn->mux->recv(conn);
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200141 }
Willy Tarreau2da156f2012-07-23 15:07:23 +0200142
Willy Tarreauc76ae332012-07-12 15:32:13 +0200143 /* It may happen during the data phase that a handshake is
144 * enabled again (eg: SSL)
145 */
Willy Tarreaud6e999b2013-11-25 08:41:15 +0100146 if (unlikely(conn->flags & (CO_FL_HANDSHAKE | CO_FL_ERROR)))
Willy Tarreauc76ae332012-07-12 15:32:13 +0200147 goto process_handshake;
148
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100149 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreauf8deb0c2012-09-01 17:59:22 +0200150 /* still waiting for a connection to establish and nothing was
151 * attempted yet to probe the connection. Then let's retry the
152 * connect().
Willy Tarreau2da156f2012-07-23 15:07:23 +0200153 */
Willy Tarreau239d7182012-07-23 18:53:03 +0200154 if (!tcp_connect_probe(conn))
Willy Tarreauafad0e02012-08-09 14:45:22 +0200155 goto leave;
Willy Tarreau2da156f2012-07-23 15:07:23 +0200156 }
Willy Tarreau2c6be842012-07-06 17:12:34 +0200157 leave:
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100158 /* Verify if the connection just established. */
Willy Tarreau7bf3fa32017-03-14 20:19:29 +0100159 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED))))
160 conn->flags |= CO_FL_CONNECTED;
161
Willy Tarreau8e3c6ce2017-08-28 15:46:01 +0200162 /* The connection owner might want to be notified about failures to
163 * complete the handshake. The callback may fail and cause the
164 * connection to be destroyed, thus we must not use it anymore and
165 * should immediately leave instead. The caller must immediately
166 * unregister itself once called.
167 */
168 if (((conn->flags ^ flags) & CO_FL_NOTIFY_DONE) &&
169 conn->xprt_done_cb && conn->xprt_done_cb(conn) < 0)
170 return;
171
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100172 /* The wake callback is normally used to notify the data layer about
173 * data layer activity (successful send/recv), connection establishment,
174 * shutdown and fatal errors. We need to consider the following
175 * situations to wake up the data layer :
176 * - change among the CO_FL_NOTIFY_DATA flags :
177 * {DATA,SOCK}_{RD,WR}_SH, ERROR,
178 * - absence of any of {L4,L6}_CONN and CONNECTED, indicating the
179 * end of handshake and transition to CONNECTED
180 * - raise of CONNECTED with HANDSHAKE down
181 * - end of HANDSHAKE with CONNECTED set
182 * - regular data layer activity
183 *
184 * Note that the wake callback is allowed to release the connection and
185 * the fd (and return < 0 in this case).
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200186 */
Willy Tarreau9fa1ee62017-03-18 15:39:57 +0100187 if ((((conn->flags ^ flags) & CO_FL_NOTIFY_DATA) ||
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100188 ((flags & (CO_FL_CONNECTED|CO_FL_HANDSHAKE)) != CO_FL_CONNECTED &&
189 (conn->flags & (CO_FL_CONNECTED|CO_FL_HANDSHAKE)) == CO_FL_CONNECTED)) &&
Willy Tarreau53a47662017-08-28 10:53:00 +0200190 conn->mux->wake(conn) < 0)
Willy Tarreau7a798e52016-04-14 11:13:20 +0200191 return;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200192
Willy Tarreau61ace1b2012-07-23 12:14:26 +0200193 /* remove the events before leaving */
Willy Tarreau26d7cfc2012-12-07 00:09:43 +0100194 fdtab[fd].ev &= FD_POLL_STICKY;
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200195
196 /* commit polling changes */
Willy Tarreau916e12d2017-10-25 09:22:43 +0200197 conn->flags &= ~CO_FL_WILL_UPDATE;
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200198 conn_cond_update_polling(conn);
Willy Tarreau7a798e52016-04-14 11:13:20 +0200199 return;
Willy Tarreau59f98392012-07-06 14:13:49 +0200200}
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200201
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200202/* Update polling on connection <c>'s file descriptor depending on its current
203 * state as reported in the connection's CO_FL_CURR_* flags, reports of EAGAIN
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200204 * in CO_FL_WAIT_*, and the data layer expectations indicated by CO_FL_XPRT_*.
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200205 * The connection flags are updated with the new flags at the end of the
Willy Tarreau0ffde2c2012-10-04 22:21:15 +0200206 * operation. Polling is totally disabled if an error was reported.
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200207 */
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200208void conn_update_xprt_polling(struct connection *c)
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200209{
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200210 unsigned int f = c->flags;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200211
Willy Tarreau3c728722014-01-23 13:50:42 +0100212 if (!conn_ctrl_ready(c))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200213 return;
214
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200215 /* update read status if needed */
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200216 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 +0200217 fd_want_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100218 f |= CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200219 }
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200220 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 +0200221 fd_stop_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100222 f &= ~CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200223 }
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200224
225 /* update write status if needed */
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200226 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 +0200227 fd_want_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100228 f |= CO_FL_CURR_WR_ENA;
229 }
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200230 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 +0200231 fd_stop_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100232 f &= ~CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200233 }
Willy Tarreau310987a2014-01-22 19:46:33 +0100234 c->flags = f;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200235}
236
237/* Update polling on connection <c>'s file descriptor depending on its current
238 * state as reported in the connection's CO_FL_CURR_* flags, reports of EAGAIN
239 * in CO_FL_WAIT_*, and the sock layer expectations indicated by CO_FL_SOCK_*.
240 * The connection flags are updated with the new flags at the end of the
Willy Tarreau0ffde2c2012-10-04 22:21:15 +0200241 * operation. Polling is totally disabled if an error was reported.
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200242 */
243void conn_update_sock_polling(struct connection *c)
244{
245 unsigned int f = c->flags;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200246
Willy Tarreau3c728722014-01-23 13:50:42 +0100247 if (!conn_ctrl_ready(c))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200248 return;
249
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200250 /* update read status if needed */
Willy Tarreau310987a2014-01-22 19:46:33 +0100251 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 +0200252 fd_want_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100253 f |= CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200254 }
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100255 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 +0200256 fd_stop_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100257 f &= ~CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200258 }
259
260 /* update write status if needed */
Willy Tarreau310987a2014-01-22 19:46:33 +0100261 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 +0200262 fd_want_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100263 f |= CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200264 }
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100265 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 +0200266 fd_stop_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100267 f &= ~CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200268 }
Willy Tarreau310987a2014-01-22 19:46:33 +0100269 c->flags = f;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200270}
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200271
Willy Tarreauff3e6482015-03-12 23:56:52 +0100272/* Send a message over an established connection. It makes use of send() and
273 * returns the same return code and errno. If the socket layer is not ready yet
274 * then -1 is returned and ENOTSOCK is set into errno. If the fd is not marked
275 * as ready, or if EAGAIN or ENOTCONN is returned, then we return 0. It returns
276 * EMSGSIZE if called with a zero length message. The purpose is to simplify
277 * some rare attempts to directly write on the socket from above the connection
278 * (typically send_proxy). In case of EAGAIN, the fd is marked as "cant_send".
279 * It automatically retries on EINTR. Other errors cause the connection to be
280 * marked as in error state. It takes similar arguments as send() except the
281 * first one which is the connection instead of the file descriptor. Note,
282 * MSG_DONTWAIT and MSG_NOSIGNAL are forced on the flags.
283 */
284int conn_sock_send(struct connection *conn, const void *buf, int len, int flags)
285{
286 int ret;
287
288 ret = -1;
289 errno = ENOTSOCK;
290
291 if (conn->flags & CO_FL_SOCK_WR_SH)
292 goto fail;
293
294 if (!conn_ctrl_ready(conn))
295 goto fail;
296
297 errno = EMSGSIZE;
298 if (!len)
299 goto fail;
300
Willy Tarreau585744b2017-08-24 14:31:19 +0200301 if (!fd_send_ready(conn->handle.fd))
Willy Tarreauff3e6482015-03-12 23:56:52 +0100302 goto wait;
303
304 do {
Willy Tarreau585744b2017-08-24 14:31:19 +0200305 ret = send(conn->handle.fd, buf, len, flags | MSG_DONTWAIT | MSG_NOSIGNAL);
Willy Tarreauff3e6482015-03-12 23:56:52 +0100306 } while (ret < 0 && errno == EINTR);
307
308
309 if (ret > 0)
310 return ret;
311
312 if (ret == 0 || errno == EAGAIN || errno == ENOTCONN) {
313 wait:
Willy Tarreau585744b2017-08-24 14:31:19 +0200314 fd_cant_send(conn->handle.fd);
Willy Tarreauff3e6482015-03-12 23:56:52 +0100315 return 0;
316 }
317 fail:
318 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH | CO_FL_ERROR;
319 return ret;
320}
321
Willy Tarreaud85c4852015-03-13 00:40:28 +0100322/* Drains possibly pending incoming data on the file descriptor attached to the
323 * connection and update the connection's flags accordingly. This is used to
324 * know whether we need to disable lingering on close. Returns non-zero if it
325 * is safe to close without disabling lingering, otherwise zero. The SOCK_RD_SH
326 * flag may also be updated if the incoming shutdown was reported by the drain()
327 * function.
328 */
329int conn_sock_drain(struct connection *conn)
330{
331 if (!conn_ctrl_ready(conn))
332 return 1;
333
334 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH))
335 return 1;
336
Willy Tarreau585744b2017-08-24 14:31:19 +0200337 if (fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP)) {
338 fdtab[conn->handle.fd].linger_risk = 0;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100339 }
340 else {
Willy Tarreau585744b2017-08-24 14:31:19 +0200341 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaud85c4852015-03-13 00:40:28 +0100342 return 0;
343
344 /* disable draining if we were called and have no drain function */
345 if (!conn->ctrl->drain) {
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200346 __conn_xprt_stop_recv(conn);
Willy Tarreaud85c4852015-03-13 00:40:28 +0100347 return 0;
348 }
349
Willy Tarreau585744b2017-08-24 14:31:19 +0200350 if (conn->ctrl->drain(conn->handle.fd) <= 0)
Willy Tarreaud85c4852015-03-13 00:40:28 +0100351 return 0;
352 }
353
354 conn->flags |= CO_FL_SOCK_RD_SH;
355 return 1;
356}
357
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100358/*
359 * Get data length from tlv
360 */
361static int get_tlv_length(const struct tlv *src)
362{
363 return (src->length_hi << 8) | src->length_lo;
364}
365
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200366/* This handshake handler waits a PROXY protocol header at the beginning of the
367 * raw data stream. The header looks like this :
368 *
369 * "PROXY" <SP> PROTO <SP> SRC3 <SP> DST3 <SP> SRC4 <SP> <DST4> "\r\n"
370 *
371 * There must be exactly one space between each field. Fields are :
372 * - PROTO : layer 4 protocol, which must be "TCP4" or "TCP6".
373 * - SRC3 : layer 3 (eg: IP) source address in standard text form
374 * - DST3 : layer 3 (eg: IP) destination address in standard text form
375 * - SRC4 : layer 4 (eg: TCP port) source address in standard text form
376 * - DST4 : layer 4 (eg: TCP port) destination address in standard text form
377 *
378 * This line MUST be at the beginning of the buffer and MUST NOT wrap.
379 *
380 * The header line is small and in all cases smaller than the smallest normal
381 * TCP MSS. So it MUST always be delivered as one segment, which ensures we
382 * can safely use MSG_PEEK and avoid buffering.
383 *
384 * Once the data is fetched, the values are set in the connection's address
385 * fields, and data are removed from the socket's buffer. The function returns
386 * zero if it needs to wait for more data or if it fails, or 1 if it completed
387 * and removed itself.
388 */
389int conn_recv_proxy(struct connection *conn, int flag)
390{
391 char *line, *end;
Willy Tarreau77992672014-06-14 11:06:17 +0200392 struct proxy_hdr_v2 *hdr_v2;
393 const char v2sig[] = PP2_SIGNATURE;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100394 int tlv_length = 0;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200395 int tlv_offset = 0;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200396
397 /* we might have been called just after an asynchronous shutr */
398 if (conn->flags & CO_FL_SOCK_RD_SH)
399 goto fail;
400
Willy Tarreau3c728722014-01-23 13:50:42 +0100401 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200402 goto fail;
403
Willy Tarreau585744b2017-08-24 14:31:19 +0200404 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100405 return 0;
406
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200407 do {
Willy Tarreau585744b2017-08-24 14:31:19 +0200408 trash.len = recv(conn->handle.fd, trash.str, trash.size, MSG_PEEK);
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100409 if (trash.len < 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200410 if (errno == EINTR)
411 continue;
412 if (errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200413 fd_cant_recv(conn->handle.fd);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200414 return 0;
415 }
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100416 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200417 }
418 } while (0);
419
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100420 if (!trash.len) {
421 /* client shutdown */
422 conn->err_code = CO_ER_PRX_EMPTY;
423 goto fail;
424 }
425
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100426 if (trash.len < 6)
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200427 goto missing;
428
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100429 line = trash.str;
430 end = trash.str + trash.len;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200431
432 /* Decode a possible proxy request, fail early if it does not match */
Willy Tarreau77992672014-06-14 11:06:17 +0200433 if (strncmp(line, "PROXY ", 6) != 0)
434 goto not_v1;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200435
436 line += 6;
Willy Tarreau4c20d292014-06-14 11:41:36 +0200437 if (trash.len < 9) /* shortest possible line */
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200438 goto missing;
439
David CARLIER42ff05e2016-03-24 09:22:36 +0000440 if (memcmp(line, "TCP4 ", 5) == 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200441 u32 src3, dst3, sport, dport;
442
443 line += 5;
444
445 src3 = inetaddr_host_lim_ret(line, end, &line);
446 if (line == end)
447 goto missing;
448 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100449 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200450
451 dst3 = inetaddr_host_lim_ret(line, end, &line);
452 if (line == end)
453 goto missing;
454 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100455 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200456
457 sport = read_uint((const char **)&line, end);
458 if (line == end)
459 goto missing;
460 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100461 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200462
463 dport = read_uint((const char **)&line, end);
464 if (line > end - 2)
465 goto missing;
466 if (*line++ != '\r')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100467 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200468 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100469 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200470
471 /* update the session's addresses and mark them set */
472 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
473 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = htonl(src3);
474 ((struct sockaddr_in *)&conn->addr.from)->sin_port = htons(sport);
475
476 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
477 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = htonl(dst3);
478 ((struct sockaddr_in *)&conn->addr.to)->sin_port = htons(dport);
479 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
480 }
David CARLIER42ff05e2016-03-24 09:22:36 +0000481 else if (memcmp(line, "TCP6 ", 5) == 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200482 u32 sport, dport;
483 char *src_s;
484 char *dst_s, *sport_s, *dport_s;
485 struct in6_addr src3, dst3;
486
487 line += 5;
488
489 src_s = line;
490 dst_s = sport_s = dport_s = NULL;
491 while (1) {
492 if (line > end - 2) {
493 goto missing;
494 }
495 else if (*line == '\r') {
496 *line = 0;
497 line++;
498 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100499 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200500 break;
501 }
502
503 if (*line == ' ') {
504 *line = 0;
505 if (!dst_s)
506 dst_s = line + 1;
507 else if (!sport_s)
508 sport_s = line + 1;
509 else if (!dport_s)
510 dport_s = line + 1;
511 }
512 line++;
513 }
514
515 if (!dst_s || !sport_s || !dport_s)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100516 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200517
518 sport = read_uint((const char **)&sport_s,dport_s - 1);
519 if (*sport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100520 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200521
522 dport = read_uint((const char **)&dport_s,line - 2);
523 if (*dport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100524 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200525
526 if (inet_pton(AF_INET6, src_s, (void *)&src3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100527 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200528
529 if (inet_pton(AF_INET6, dst_s, (void *)&dst3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100530 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200531
532 /* update the session's addresses and mark them set */
533 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
534 memcpy(&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr, &src3, sizeof(struct in6_addr));
535 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = htons(sport);
536
537 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
538 memcpy(&((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr, &dst3, sizeof(struct in6_addr));
539 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = htons(dport);
540 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
541 }
Willy Tarreau4c20d292014-06-14 11:41:36 +0200542 else if (memcmp(line, "UNKNOWN\r\n", 9) == 0) {
543 /* This can be a UNIX socket forwarded by an haproxy upstream */
544 line += 9;
545 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200546 else {
Willy Tarreau4c20d292014-06-14 11:41:36 +0200547 /* The protocol does not match something known (TCP4/TCP6/UNKNOWN) */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100548 conn->err_code = CO_ER_PRX_BAD_PROTO;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200549 goto fail;
550 }
551
Willy Tarreau77992672014-06-14 11:06:17 +0200552 trash.len = line - trash.str;
553 goto eat_header;
554
555 not_v1:
556 /* try PPv2 */
557 if (trash.len < PP2_HEADER_LEN)
558 goto missing;
559
560 hdr_v2 = (struct proxy_hdr_v2 *)trash.str;
561
562 if (memcmp(hdr_v2->sig, v2sig, PP2_SIGNATURE_LEN) != 0 ||
563 (hdr_v2->ver_cmd & PP2_VERSION_MASK) != PP2_VERSION) {
564 conn->err_code = CO_ER_PRX_NOT_HDR;
565 goto fail;
566 }
567
568 if (trash.len < PP2_HEADER_LEN + ntohs(hdr_v2->len))
569 goto missing;
570
571 switch (hdr_v2->ver_cmd & PP2_CMD_MASK) {
572 case 0x01: /* PROXY command */
573 switch (hdr_v2->fam) {
574 case 0x11: /* TCPv4 */
KOVACS Krisztianefd3aa92014-11-19 10:53:20 +0100575 if (ntohs(hdr_v2->len) < PP2_ADDR_LEN_INET)
576 goto bad_header;
577
Willy Tarreau77992672014-06-14 11:06:17 +0200578 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
579 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = hdr_v2->addr.ip4.src_addr;
580 ((struct sockaddr_in *)&conn->addr.from)->sin_port = hdr_v2->addr.ip4.src_port;
581 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
582 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = hdr_v2->addr.ip4.dst_addr;
583 ((struct sockaddr_in *)&conn->addr.to)->sin_port = hdr_v2->addr.ip4.dst_port;
584 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200585 tlv_offset = PP2_HEADER_LEN + PP2_ADDR_LEN_INET;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100586 tlv_length = ntohs(hdr_v2->len) - PP2_ADDR_LEN_INET;
Willy Tarreau77992672014-06-14 11:06:17 +0200587 break;
588 case 0x21: /* TCPv6 */
KOVACS Krisztianefd3aa92014-11-19 10:53:20 +0100589 if (ntohs(hdr_v2->len) < PP2_ADDR_LEN_INET6)
590 goto bad_header;
591
Willy Tarreau77992672014-06-14 11:06:17 +0200592 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
593 memcpy(&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr, hdr_v2->addr.ip6.src_addr, 16);
594 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = hdr_v2->addr.ip6.src_port;
595 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
596 memcpy(&((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr, hdr_v2->addr.ip6.dst_addr, 16);
597 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = hdr_v2->addr.ip6.dst_port;
598 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200599 tlv_offset = PP2_HEADER_LEN + PP2_ADDR_LEN_INET6;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100600 tlv_length = ntohs(hdr_v2->len) - PP2_ADDR_LEN_INET6;
Willy Tarreau77992672014-06-14 11:06:17 +0200601 break;
602 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100603
604 /* TLV parsing */
605 if (tlv_length > 0) {
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100606 while (tlv_offset + TLV_HEADER_SIZE <= trash.len) {
607 const struct tlv *tlv_packet = (struct tlv *) &trash.str[tlv_offset];
608 const int tlv_len = get_tlv_length(tlv_packet);
609 tlv_offset += tlv_len + TLV_HEADER_SIZE;
610
611 switch (tlv_packet->type) {
612#ifdef CONFIG_HAP_NS
613 case PP2_TYPE_NETNS: {
614 const struct netns_entry *ns;
615 ns = netns_store_lookup((char*)tlv_packet->value, tlv_len);
616 if (ns)
617 conn->proxy_netns = ns;
618 break;
619 }
620#endif
621 default:
622 break;
623 }
624 }
625 }
626
Willy Tarreau77992672014-06-14 11:06:17 +0200627 /* unsupported protocol, keep local connection address */
628 break;
629 case 0x00: /* LOCAL command */
630 /* keep local connection address for LOCAL */
631 break;
632 default:
633 goto bad_header; /* not a supported command */
634 }
635
636 trash.len = PP2_HEADER_LEN + ntohs(hdr_v2->len);
637 goto eat_header;
638
639 eat_header:
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200640 /* remove the PROXY line from the request. For this we re-read the
641 * exact line at once. If we don't get the exact same result, we
642 * fail.
643 */
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200644 do {
Willy Tarreau585744b2017-08-24 14:31:19 +0200645 int len2 = recv(conn->handle.fd, trash.str, trash.len, 0);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200646 if (len2 < 0 && errno == EINTR)
647 continue;
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100648 if (len2 != trash.len)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100649 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200650 } while (0);
651
652 conn->flags &= ~flag;
Emeric Brun4f603012017-01-05 15:11:44 +0100653 conn->flags |= CO_FL_RCVD_PROXY;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200654 return 1;
655
656 missing:
657 /* Missing data. Since we're using MSG_PEEK, we can only poll again if
658 * we have not read anything. Otherwise we need to fail because we won't
659 * be able to poll anymore.
660 */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100661 conn->err_code = CO_ER_PRX_TRUNCATED;
662 goto fail;
663
664 bad_header:
665 /* This is not a valid proxy protocol header */
666 conn->err_code = CO_ER_PRX_BAD_HDR;
667 goto fail;
668
669 recv_abort:
670 conn->err_code = CO_ER_PRX_ABORT;
Willy Tarreau26f4a042013-12-04 23:44:10 +0100671 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100672 goto fail;
673
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200674 fail:
Willy Tarreaud486ef52012-12-10 17:03:52 +0100675 __conn_sock_stop_both(conn);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200676 conn->flags |= CO_FL_ERROR;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200677 return 0;
678}
679
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100680/* This handshake handler waits a NetScaler Client IP insertion header
681 * at the beginning of the raw data stream. The header looks like this:
682 *
683 * 4 bytes: CIP magic number
684 * 4 bytes: Header length
685 * 20+ bytes: Header of the last IP packet sent by the client during
686 * TCP handshake.
687 * 20+ bytes: Header of the last TCP packet sent by the client during
688 * TCP handshake.
689 *
690 * This line MUST be at the beginning of the buffer and MUST NOT be
691 * fragmented.
692 *
693 * The header line is small and in all cases smaller than the smallest normal
694 * TCP MSS. So it MUST always be delivered as one segment, which ensures we
695 * can safely use MSG_PEEK and avoid buffering.
696 *
697 * Once the data is fetched, the values are set in the connection's address
698 * fields, and data are removed from the socket's buffer. The function returns
699 * zero if it needs to wait for more data or if it fails, or 1 if it completed
700 * and removed itself.
701 */
702int conn_recv_netscaler_cip(struct connection *conn, int flag)
703{
704 char *line;
705 uint32_t cip_magic;
706 uint32_t cip_len;
707 uint8_t ip_v;
708
709 /* we might have been called just after an asynchronous shutr */
710 if (conn->flags & CO_FL_SOCK_RD_SH)
711 goto fail;
712
713 if (!conn_ctrl_ready(conn))
714 goto fail;
715
Willy Tarreau585744b2017-08-24 14:31:19 +0200716 if (!fd_recv_ready(conn->handle.fd))
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100717 return 0;
718
719 do {
Willy Tarreau585744b2017-08-24 14:31:19 +0200720 trash.len = recv(conn->handle.fd, trash.str, trash.size, MSG_PEEK);
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100721 if (trash.len < 0) {
722 if (errno == EINTR)
723 continue;
724 if (errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200725 fd_cant_recv(conn->handle.fd);
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100726 return 0;
727 }
728 goto recv_abort;
729 }
730 } while (0);
731
732 if (!trash.len) {
733 /* client shutdown */
734 conn->err_code = CO_ER_CIP_EMPTY;
735 goto fail;
736 }
737
738 /* Fail if buffer length is not large enough to contain
739 * CIP magic, CIP length */
740 if (trash.len < 8)
741 goto missing;
742
743 line = trash.str;
744
745 cip_magic = ntohl(*(uint32_t *)line);
746 cip_len = ntohl(*(uint32_t *)(line+4));
747
748 /* Decode a possible NetScaler Client IP request, fail early if
749 * it does not match */
750 if (cip_magic != objt_listener(conn->target)->bind_conf->ns_cip_magic)
751 goto bad_magic;
752
753 /* Fail if buffer length is not large enough to contain
754 * CIP magic, CIP length, minimal IP header */
755 if (trash.len < 28)
756 goto missing;
757
758 line += 8;
759
760 /* Get IP version from the first four bits */
761 ip_v = (*line & 0xf0) >> 4;
762
763 if (ip_v == 4) {
764 struct ip *hdr_ip4;
David Carlier3015a2e2016-07-04 22:51:33 +0100765 struct my_tcphdr *hdr_tcp;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100766
767 hdr_ip4 = (struct ip *)line;
768
769 if (trash.len < (8 + ntohs(hdr_ip4->ip_len))) {
770 /* Fail if buffer length is not large enough to contain
771 * CIP magic, CIP length, IPv4 header */
772 goto missing;
773 } else if (hdr_ip4->ip_p != IPPROTO_TCP) {
774 /* The protocol does not include a TCP header */
775 conn->err_code = CO_ER_CIP_BAD_PROTO;
776 goto fail;
777 } else if (trash.len < (28 + ntohs(hdr_ip4->ip_len))) {
778 /* Fail if buffer length is not large enough to contain
779 * CIP magic, CIP length, IPv4 header, TCP header */
780 goto missing;
781 }
782
David Carlier3015a2e2016-07-04 22:51:33 +0100783 hdr_tcp = (struct my_tcphdr *)(line + (hdr_ip4->ip_hl * 4));
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100784
785 /* update the session's addresses and mark them set */
786 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
787 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = hdr_ip4->ip_src.s_addr;
788 ((struct sockaddr_in *)&conn->addr.from)->sin_port = hdr_tcp->source;
789
790 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
791 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = hdr_ip4->ip_dst.s_addr;
792 ((struct sockaddr_in *)&conn->addr.to)->sin_port = hdr_tcp->dest;
793
794 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
795 }
796 else if (ip_v == 6) {
797 struct ip6_hdr *hdr_ip6;
David Carlier3015a2e2016-07-04 22:51:33 +0100798 struct my_tcphdr *hdr_tcp;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100799
800 hdr_ip6 = (struct ip6_hdr *)line;
801
802 if (trash.len < 28) {
803 /* Fail if buffer length is not large enough to contain
804 * CIP magic, CIP length, IPv6 header */
805 goto missing;
806 } else if (hdr_ip6->ip6_nxt != IPPROTO_TCP) {
807 /* The protocol does not include a TCP header */
808 conn->err_code = CO_ER_CIP_BAD_PROTO;
809 goto fail;
810 } else if (trash.len < 48) {
811 /* Fail if buffer length is not large enough to contain
812 * CIP magic, CIP length, IPv6 header, TCP header */
813 goto missing;
814 }
815
David Carlier3015a2e2016-07-04 22:51:33 +0100816 hdr_tcp = (struct my_tcphdr *)(line + sizeof(struct ip6_hdr));
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100817
818 /* update the session's addresses and mark them set */
819 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
820 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr = hdr_ip6->ip6_src;
821 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = hdr_tcp->source;
822
823 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
824 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr = hdr_ip6->ip6_dst;
825 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = hdr_tcp->dest;
826
827 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
828 }
829 else {
830 /* The protocol does not match something known (IPv4/IPv6) */
831 conn->err_code = CO_ER_CIP_BAD_PROTO;
832 goto fail;
833 }
834
835 line += cip_len;
836 trash.len = line - trash.str;
837
838 /* remove the NetScaler Client IP header from the request. For this
839 * we re-read the exact line at once. If we don't get the exact same
840 * result, we fail.
841 */
842 do {
Willy Tarreau585744b2017-08-24 14:31:19 +0200843 int len2 = recv(conn->handle.fd, trash.str, trash.len, 0);
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100844 if (len2 < 0 && errno == EINTR)
845 continue;
846 if (len2 != trash.len)
847 goto recv_abort;
848 } while (0);
849
850 conn->flags &= ~flag;
851 return 1;
852
853 missing:
854 /* Missing data. Since we're using MSG_PEEK, we can only poll again if
855 * we have not read anything. Otherwise we need to fail because we won't
856 * be able to poll anymore.
857 */
858 conn->err_code = CO_ER_CIP_TRUNCATED;
859 goto fail;
860
861 bad_magic:
862 conn->err_code = CO_ER_CIP_BAD_MAGIC;
863 goto fail;
864
865 recv_abort:
866 conn->err_code = CO_ER_CIP_ABORT;
867 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
868 goto fail;
869
870 fail:
871 __conn_sock_stop_both(conn);
872 conn->flags |= CO_FL_ERROR;
873 return 0;
874}
875
David Safb76832014-05-08 23:42:08 -0400876int make_proxy_line(char *buf, int buf_len, struct server *srv, struct connection *remote)
877{
878 int ret = 0;
879
880 if (srv && (srv->pp_opts & SRV_PP_V2)) {
881 ret = make_proxy_line_v2(buf, buf_len, srv, remote);
882 }
883 else {
884 if (remote)
885 ret = make_proxy_line_v1(buf, buf_len, &remote->addr.from, &remote->addr.to);
886 else
887 ret = make_proxy_line_v1(buf, buf_len, NULL, NULL);
888 }
889
890 return ret;
891}
892
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200893/* Makes a PROXY protocol line from the two addresses. The output is sent to
894 * buffer <buf> for a maximum size of <buf_len> (including the trailing zero).
895 * It returns the number of bytes composing this line (including the trailing
896 * LF), or zero in case of failure (eg: not enough space). It supports TCP4,
Willy Tarreau2e1401a2013-10-01 11:41:55 +0200897 * TCP6 and "UNKNOWN" formats. If any of <src> or <dst> is null, UNKNOWN is
898 * emitted as well.
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200899 */
David Safb76832014-05-08 23:42:08 -0400900int make_proxy_line_v1(char *buf, int buf_len, struct sockaddr_storage *src, struct sockaddr_storage *dst)
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200901{
902 int ret = 0;
903
Willy Tarreau2e1401a2013-10-01 11:41:55 +0200904 if (src && dst && src->ss_family == dst->ss_family && src->ss_family == AF_INET) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200905 ret = snprintf(buf + ret, buf_len - ret, "PROXY TCP4 ");
906 if (ret >= buf_len)
907 return 0;
908
909 /* IPv4 src */
910 if (!inet_ntop(src->ss_family, &((struct sockaddr_in *)src)->sin_addr, buf + ret, buf_len - ret))
911 return 0;
912
913 ret += strlen(buf + ret);
914 if (ret >= buf_len)
915 return 0;
916
917 buf[ret++] = ' ';
918
919 /* IPv4 dst */
920 if (!inet_ntop(dst->ss_family, &((struct sockaddr_in *)dst)->sin_addr, buf + ret, buf_len - ret))
921 return 0;
922
923 ret += strlen(buf + ret);
924 if (ret >= buf_len)
925 return 0;
926
927 /* source and destination ports */
928 ret += snprintf(buf + ret, buf_len - ret, " %u %u\r\n",
929 ntohs(((struct sockaddr_in *)src)->sin_port),
930 ntohs(((struct sockaddr_in *)dst)->sin_port));
931 if (ret >= buf_len)
932 return 0;
933 }
Willy Tarreau2e1401a2013-10-01 11:41:55 +0200934 else if (src && dst && src->ss_family == dst->ss_family && src->ss_family == AF_INET6) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200935 ret = snprintf(buf + ret, buf_len - ret, "PROXY TCP6 ");
936 if (ret >= buf_len)
937 return 0;
938
939 /* IPv6 src */
940 if (!inet_ntop(src->ss_family, &((struct sockaddr_in6 *)src)->sin6_addr, buf + ret, buf_len - ret))
941 return 0;
942
943 ret += strlen(buf + ret);
944 if (ret >= buf_len)
945 return 0;
946
947 buf[ret++] = ' ';
948
949 /* IPv6 dst */
950 if (!inet_ntop(dst->ss_family, &((struct sockaddr_in6 *)dst)->sin6_addr, buf + ret, buf_len - ret))
951 return 0;
952
953 ret += strlen(buf + ret);
954 if (ret >= buf_len)
955 return 0;
956
957 /* source and destination ports */
958 ret += snprintf(buf + ret, buf_len - ret, " %u %u\r\n",
959 ntohs(((struct sockaddr_in6 *)src)->sin6_port),
960 ntohs(((struct sockaddr_in6 *)dst)->sin6_port));
961 if (ret >= buf_len)
962 return 0;
963 }
964 else {
965 /* unknown family combination */
966 ret = snprintf(buf, buf_len, "PROXY UNKNOWN\r\n");
967 if (ret >= buf_len)
968 return 0;
969 }
970 return ret;
971}
David Safb76832014-05-08 23:42:08 -0400972
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100973static int make_tlv(char *dest, int dest_len, char type, uint16_t length, const char *value)
David Safb76832014-05-08 23:42:08 -0400974{
975 struct tlv *tlv;
976
977 if (!dest || (length + sizeof(*tlv) > dest_len))
978 return 0;
979
980 tlv = (struct tlv *)dest;
981
982 tlv->type = type;
983 tlv->length_hi = length >> 8;
984 tlv->length_lo = length & 0x00ff;
985 memcpy(tlv->value, value, length);
986 return length + sizeof(*tlv);
987}
David Safb76832014-05-08 23:42:08 -0400988
989int make_proxy_line_v2(char *buf, int buf_len, struct server *srv, struct connection *remote)
990{
Willy Tarreau8fccfa22014-06-14 08:28:06 +0200991 const char pp2_signature[] = PP2_SIGNATURE;
David Safb76832014-05-08 23:42:08 -0400992 int ret = 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +0200993 struct proxy_hdr_v2 *hdr = (struct proxy_hdr_v2 *)buf;
Vincent Bernat6e615892016-05-18 16:17:44 +0200994 struct sockaddr_storage null_addr = { .ss_family = 0 };
David Safb76832014-05-08 23:42:08 -0400995 struct sockaddr_storage *src = &null_addr;
996 struct sockaddr_storage *dst = &null_addr;
Emmanuel Hocdet404d9782017-10-24 10:55:14 +0200997 const char *value;
998 int value_len;
David Safb76832014-05-08 23:42:08 -0400999
1000 if (buf_len < PP2_HEADER_LEN)
1001 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001002 memcpy(hdr->sig, pp2_signature, PP2_SIGNATURE_LEN);
David Safb76832014-05-08 23:42:08 -04001003
1004 if (remote) {
1005 src = &remote->addr.from;
1006 dst = &remote->addr.to;
1007 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001008
David Safb76832014-05-08 23:42:08 -04001009 if (src && dst && src->ss_family == dst->ss_family && src->ss_family == AF_INET) {
1010 if (buf_len < PP2_HDR_LEN_INET)
1011 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001012 hdr->ver_cmd = PP2_VERSION | PP2_CMD_PROXY;
1013 hdr->fam = PP2_FAM_INET | PP2_TRANS_STREAM;
1014 hdr->addr.ip4.src_addr = ((struct sockaddr_in *)src)->sin_addr.s_addr;
1015 hdr->addr.ip4.dst_addr = ((struct sockaddr_in *)dst)->sin_addr.s_addr;
1016 hdr->addr.ip4.src_port = ((struct sockaddr_in *)src)->sin_port;
1017 hdr->addr.ip4.dst_port = ((struct sockaddr_in *)dst)->sin_port;
David Safb76832014-05-08 23:42:08 -04001018 ret = PP2_HDR_LEN_INET;
1019 }
1020 else if (src && dst && src->ss_family == dst->ss_family && src->ss_family == AF_INET6) {
1021 if (buf_len < PP2_HDR_LEN_INET6)
1022 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001023 hdr->ver_cmd = PP2_VERSION | PP2_CMD_PROXY;
1024 hdr->fam = PP2_FAM_INET6 | PP2_TRANS_STREAM;
1025 memcpy(hdr->addr.ip6.src_addr, &((struct sockaddr_in6 *)src)->sin6_addr, 16);
1026 memcpy(hdr->addr.ip6.dst_addr, &((struct sockaddr_in6 *)dst)->sin6_addr, 16);
1027 hdr->addr.ip6.src_port = ((struct sockaddr_in6 *)src)->sin6_port;
1028 hdr->addr.ip6.dst_port = ((struct sockaddr_in6 *)dst)->sin6_port;
David Safb76832014-05-08 23:42:08 -04001029 ret = PP2_HDR_LEN_INET6;
1030 }
1031 else {
1032 if (buf_len < PP2_HDR_LEN_UNSPEC)
1033 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001034 hdr->ver_cmd = PP2_VERSION | PP2_CMD_LOCAL;
1035 hdr->fam = PP2_FAM_UNSPEC | PP2_TRANS_UNSPEC;
David Safb76832014-05-08 23:42:08 -04001036 ret = PP2_HDR_LEN_UNSPEC;
1037 }
1038
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001039 if (conn_get_alpn(remote, &value, &value_len)) {
1040 if ((buf_len - ret) < sizeof(struct tlv))
1041 return 0;
Emmanuel Hocdet571c7ac2017-10-31 18:24:05 +01001042 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_ALPN, value_len, value);
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001043 }
1044
David Safb76832014-05-08 23:42:08 -04001045#ifdef USE_OPENSSL
1046 if (srv->pp_opts & SRV_PP_V2_SSL) {
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001047 struct tlv_ssl *tlv;
1048 int ssl_tlv_len = 0;
David Safb76832014-05-08 23:42:08 -04001049 if ((buf_len - ret) < sizeof(struct tlv_ssl))
1050 return 0;
1051 tlv = (struct tlv_ssl *)&buf[ret];
1052 memset(tlv, 0, sizeof(struct tlv_ssl));
1053 ssl_tlv_len += sizeof(struct tlv_ssl);
1054 tlv->tlv.type = PP2_TYPE_SSL;
1055 if (ssl_sock_is_ssl(remote)) {
1056 tlv->client |= PP2_CLIENT_SSL;
Emmanuel Hocdet01da5712017-10-13 16:59:49 +02001057 value = ssl_sock_get_proto_version(remote);
David Safb76832014-05-08 23:42:08 -04001058 if (value) {
Emmanuel Hocdet82913e42017-10-31 18:31:36 +01001059 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_VERSION, strlen(value)+1, value);
David Safb76832014-05-08 23:42:08 -04001060 }
Dave McCowan328fb582014-07-30 10:39:13 -04001061 if (ssl_sock_get_cert_used_sess(remote)) {
1062 tlv->client |= PP2_CLIENT_CERT_SESS;
David Safb76832014-05-08 23:42:08 -04001063 tlv->verify = htonl(ssl_sock_get_verify_result(remote));
Dave McCowan328fb582014-07-30 10:39:13 -04001064 if (ssl_sock_get_cert_used_conn(remote))
1065 tlv->client |= PP2_CLIENT_CERT_CONN;
David Safb76832014-05-08 23:42:08 -04001066 }
1067 if (srv->pp_opts & SRV_PP_V2_SSL_CN) {
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001068 struct chunk *cn_trash = get_trash_chunk();
Willy Tarreau3b9a0c92014-07-19 06:37:33 +02001069 if (ssl_sock_get_remote_common_name(remote, cn_trash) > 0) {
Emmanuel Hocdet58118b42017-10-13 12:15:28 +02001070 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_CN, cn_trash->len, cn_trash->str);
David Safb76832014-05-08 23:42:08 -04001071 }
1072 }
1073 }
1074 tlv->tlv.length_hi = (uint16_t)(ssl_tlv_len - sizeof(struct tlv)) >> 8;
1075 tlv->tlv.length_lo = (uint16_t)(ssl_tlv_len - sizeof(struct tlv)) & 0x00ff;
1076 ret += ssl_tlv_len;
1077 }
1078#endif
1079
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001080#ifdef CONFIG_HAP_NS
1081 if (remote && (remote->proxy_netns)) {
1082 if ((buf_len - ret) < sizeof(struct tlv))
1083 return 0;
Emmanuel Hocdet571c7ac2017-10-31 18:24:05 +01001084 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 +01001085 }
1086#endif
1087
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001088 hdr->len = htons((uint16_t)(ret - PP2_HEADER_LEN));
David Safb76832014-05-08 23:42:08 -04001089
1090 return ret;
1091}
Emeric Brun4f603012017-01-05 15:11:44 +01001092
Willy Tarreau60ca10a2017-08-18 15:26:54 +02001093/* return the major HTTP version as 1 or 2 depending on how the request arrived
1094 * before being processed.
1095 */
1096static int
1097smp_fetch_fc_http_major(const struct arg *args, struct sample *smp, const char *kw, void *private)
1098{
1099 struct connection *conn = objt_conn(smp->sess->origin);
1100
1101 smp->data.type = SMP_T_SINT;
1102 smp->data.u.sint = (conn && strcmp(conn_get_mux_name(conn), "H2") == 0) ? 2 : 1;
1103 return 1;
1104}
1105
Emeric Brun4f603012017-01-05 15:11:44 +01001106/* fetch if the received connection used a PROXY protocol header */
1107int smp_fetch_fc_rcvd_proxy(const struct arg *args, struct sample *smp, const char *kw, void *private)
1108{
1109 struct connection *conn;
1110
1111 conn = objt_conn(smp->sess->origin);
1112 if (!conn)
1113 return 0;
1114
1115 if (!(conn->flags & CO_FL_CONNECTED)) {
1116 smp->flags |= SMP_F_MAY_CHANGE;
1117 return 0;
1118 }
1119
1120 smp->flags = 0;
1121 smp->data.type = SMP_T_BOOL;
1122 smp->data.u.sint = (conn->flags & CO_FL_RCVD_PROXY) ? 1 : 0;
1123
1124 return 1;
1125}
1126
1127/* Note: must not be declared <const> as its list will be overwritten.
1128 * Note: fetches that may return multiple types must be declared as the lowest
1129 * common denominator, the type that can be casted into all other ones. For
1130 * instance v4/v6 must be declared v4.
1131 */
1132static struct sample_fetch_kw_list sample_fetch_keywords = {ILH, {
Willy Tarreau60ca10a2017-08-18 15:26:54 +02001133 { "fc_http_major", smp_fetch_fc_http_major, 0, NULL, SMP_T_SINT, SMP_USE_L4CLI },
Emeric Brun4f603012017-01-05 15:11:44 +01001134 { "fc_rcvd_proxy", smp_fetch_fc_rcvd_proxy, 0, NULL, SMP_T_BOOL, SMP_USE_L4CLI },
1135 { /* END */ },
1136}};
1137
1138
1139__attribute__((constructor))
1140static void __connection_init(void)
1141{
1142 sample_register_fetches(&sample_fetch_keywords);
1143}