blob: 4b4a31482e5f150d42ade9e90497883246c0b1c7 [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>
Willy Tarreau0108d902018-11-25 19:14:37 +010017#include <common/initcall.h>
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +010018#include <common/namespace.h>
Emmanuel Hocdet4399c752018-02-05 15:26:43 +010019#include <common/hash.h>
20#include <common/net_helper.h>
Willy Tarreau59f98392012-07-06 14:13:49 +020021
Willy Tarreauc5788912012-08-24 18:12:41 +020022#include <proto/connection.h>
Willy Tarreaudd2f85e2012-09-02 22:34:23 +020023#include <proto/fd.h>
Willy Tarreau5f1504f2012-10-04 23:55:57 +020024#include <proto/frontend.h>
Willy Tarreau2da156f2012-07-23 15:07:23 +020025#include <proto/proto_tcp.h>
Willy Tarreau2c6be842012-07-06 17:12:34 +020026#include <proto/stream_interface.h>
Emeric Brun4f603012017-01-05 15:11:44 +010027#include <proto/sample.h>
Emeric Brun46591952012-05-18 15:47:34 +020028#include <proto/ssl_sock.h>
Emeric Brun46591952012-05-18 15:47:34 +020029
Willy Tarreau8ceae722018-11-26 11:58:30 +010030DECLARE_POOL(pool_head_connection, "connection", sizeof(struct connection));
31DECLARE_POOL(pool_head_connstream, "conn_stream", sizeof(struct conn_stream));
32
Willy Tarreau13e14102016-12-22 20:25:26 +010033struct xprt_ops *registered_xprt[XPRT_ENTRIES] = { NULL, };
Willy Tarreauf2943dc2012-10-26 20:10:28 +020034
Christopher Faulet32f61c02018-04-10 14:33:41 +020035/* List head of all known muxes for PROTO */
36struct mux_proto_list mux_proto_list = {
37 .list = LIST_HEAD_INIT(mux_proto_list.list)
Willy Tarreau2386be62017-09-21 19:40:52 +020038};
39
Willy Tarreau59f98392012-07-06 14:13:49 +020040/* I/O callback for fd-based connections. It calls the read/write handlers
Willy Tarreau7a798e52016-04-14 11:13:20 +020041 * provided by the connection's sock_ops, which must be valid.
Willy Tarreau59f98392012-07-06 14:13:49 +020042 */
Willy Tarreau7a798e52016-04-14 11:13:20 +020043void conn_fd_handler(int fd)
Willy Tarreau59f98392012-07-06 14:13:49 +020044{
Willy Tarreau80184712012-07-06 14:54:49 +020045 struct connection *conn = fdtab[fd].owner;
Willy Tarreau9e272bf2012-10-03 21:04:48 +020046 unsigned int flags;
Olivier Houchardaf4021e2018-08-09 13:06:55 +020047 int io_available = 0;
Willy Tarreau59f98392012-07-06 14:13:49 +020048
Willy Tarreaud80cb4e2018-01-20 19:30:13 +010049 if (unlikely(!conn)) {
50 activity[tid].conn_dead++;
Willy Tarreau7a798e52016-04-14 11:13:20 +020051 return;
Willy Tarreaud80cb4e2018-01-20 19:30:13 +010052 }
Willy Tarreau59f98392012-07-06 14:13:49 +020053
Willy Tarreau7d281492012-12-16 19:19:13 +010054 conn_refresh_polling_flags(conn);
Willy Tarreau916e12d2017-10-25 09:22:43 +020055 conn->flags |= CO_FL_WILL_UPDATE;
56
Willy Tarreau7d281492012-12-16 19:19:13 +010057 flags = conn->flags & ~CO_FL_ERROR; /* ensure to call the wake handler upon error */
Willy Tarreaud29a0662012-12-10 16:33:38 +010058
Willy Tarreauc76ae332012-07-12 15:32:13 +020059 process_handshake:
Willy Tarreauf9dabec2012-08-17 17:33:53 +020060 /* The handshake callbacks are called in sequence. If either of them is
61 * missing something, it must enable the required polling at the socket
62 * layer of the connection. Polling state is not guaranteed when entering
63 * these handlers, so any handshake handler which does not complete its
Willy Tarreaud6e999b2013-11-25 08:41:15 +010064 * work must explicitly disable events it's not interested in. Error
65 * handling is also performed here in order to reduce the number of tests
66 * around.
Willy Tarreauf9dabec2012-08-17 17:33:53 +020067 */
Willy Tarreaud6e999b2013-11-25 08:41:15 +010068 while (unlikely(conn->flags & (CO_FL_HANDSHAKE | CO_FL_ERROR))) {
Willy Tarreau310987a2014-01-22 19:46:33 +010069 if (unlikely(conn->flags & CO_FL_ERROR))
Willy Tarreau2c6be842012-07-06 17:12:34 +020070 goto leave;
Willy Tarreau59f98392012-07-06 14:13:49 +020071
Bertrand Jacquin93b227d2016-06-04 15:11:10 +010072 if (conn->flags & CO_FL_ACCEPT_CIP)
73 if (!conn_recv_netscaler_cip(conn, CO_FL_ACCEPT_CIP))
74 goto leave;
75
Willy Tarreau22cda212012-08-31 17:43:29 +020076 if (conn->flags & CO_FL_ACCEPT_PROXY)
77 if (!conn_recv_proxy(conn, CO_FL_ACCEPT_PROXY))
78 goto leave;
79
Willy Tarreau57cd3e42013-10-24 22:01:26 +020080 if (conn->flags & CO_FL_SEND_PROXY)
81 if (!conn_si_send_proxy(conn, CO_FL_SEND_PROXY))
Willy Tarreau5f1504f2012-10-04 23:55:57 +020082 goto leave;
Emeric Brun46591952012-05-18 15:47:34 +020083#ifdef USE_OPENSSL
84 if (conn->flags & CO_FL_SSL_WAIT_HS)
85 if (!ssl_sock_handshake(conn, CO_FL_SSL_WAIT_HS))
86 goto leave;
87#endif
Willy Tarreauc76ae332012-07-12 15:32:13 +020088 }
89
Willy Tarreauf9dabec2012-08-17 17:33:53 +020090 /* Once we're purely in the data phase, we disable handshake polling */
91 if (!(conn->flags & CO_FL_POLL_SOCK))
92 __conn_sock_stop_both(conn);
Willy Tarreauc76ae332012-07-12 15:32:13 +020093
Willy Tarreau8e3c6ce2017-08-28 15:46:01 +020094 /* The connection owner might want to be notified about an end of
95 * handshake indicating the connection is ready, before we proceed with
96 * any data exchange. The callback may fail and cause the connection to
97 * be destroyed, thus we must not use it anymore and should immediately
98 * leave instead. The caller must immediately unregister itself once
99 * called.
Willy Tarreau2542b532012-08-31 16:01:23 +0200100 */
Willy Tarreau8e3c6ce2017-08-28 15:46:01 +0200101 if (conn->xprt_done_cb && conn->xprt_done_cb(conn) < 0)
Willy Tarreau7a798e52016-04-14 11:13:20 +0200102 return;
Willy Tarreau2542b532012-08-31 16:01:23 +0200103
Willy Tarreau57ec32f2017-04-11 19:59:33 +0200104 if (conn->xprt && fd_send_ready(fd) &&
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200105 ((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 +0100106 /* force reporting of activity by clearing the previous flags :
107 * we'll have at least ERROR or CONNECTED at the end of an I/O,
108 * both of which will be detected below.
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200109 */
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100110 flags = 0;
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200111 if (conn->send_wait != NULL) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100112 conn->send_wait->events &= ~SUB_RETRY_SEND;
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200113 tasklet_wakeup(conn->send_wait->task);
114 conn->send_wait = NULL;
115 } else
116 io_available = 1;
Olivier Houchard53216e72018-10-10 15:46:36 +0200117 __conn_xprt_stop_send(conn);
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200118 }
Willy Tarreau59f98392012-07-06 14:13:49 +0200119
Willy Tarreau57ec32f2017-04-11 19:59:33 +0200120 /* The data transfer starts here and stops on error and handshakes. Note
121 * that we must absolutely test conn->xprt at each step in case it suddenly
122 * changes due to a quick unexpected close().
123 */
124 if (conn->xprt && fd_recv_ready(fd) &&
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200125 ((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 +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->recv_wait) {
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100132 conn->recv_wait->events &= ~SUB_RETRY_RECV;
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200133 tasklet_wakeup(conn->recv_wait->task);
134 conn->recv_wait = NULL;
135 } else
136 io_available = 1;
Olivier Houchard53216e72018-10-10 15:46:36 +0200137 __conn_xprt_stop_recv(conn);
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200138 }
Willy Tarreau2da156f2012-07-23 15:07:23 +0200139
Willy Tarreauc76ae332012-07-12 15:32:13 +0200140 /* It may happen during the data phase that a handshake is
141 * enabled again (eg: SSL)
142 */
Willy Tarreaud6e999b2013-11-25 08:41:15 +0100143 if (unlikely(conn->flags & (CO_FL_HANDSHAKE | CO_FL_ERROR)))
Willy Tarreauc76ae332012-07-12 15:32:13 +0200144 goto process_handshake;
145
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100146 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreauf8deb0c2012-09-01 17:59:22 +0200147 /* still waiting for a connection to establish and nothing was
148 * attempted yet to probe the connection. Then let's retry the
149 * connect().
Willy Tarreau2da156f2012-07-23 15:07:23 +0200150 */
Willy Tarreau239d7182012-07-23 18:53:03 +0200151 if (!tcp_connect_probe(conn))
Willy Tarreauafad0e02012-08-09 14:45:22 +0200152 goto leave;
Willy Tarreau2da156f2012-07-23 15:07:23 +0200153 }
Willy Tarreau2c6be842012-07-06 17:12:34 +0200154 leave:
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100155 /* Verify if the connection just established. */
Willy Tarreau7bf3fa32017-03-14 20:19:29 +0100156 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED))))
157 conn->flags |= CO_FL_CONNECTED;
158
Willy Tarreau8e3c6ce2017-08-28 15:46:01 +0200159 /* The connection owner might want to be notified about failures to
160 * complete the handshake. The callback may fail and cause the
161 * connection to be destroyed, thus we must not use it anymore and
162 * should immediately leave instead. The caller must immediately
163 * unregister itself once called.
164 */
165 if (((conn->flags ^ flags) & CO_FL_NOTIFY_DONE) &&
166 conn->xprt_done_cb && conn->xprt_done_cb(conn) < 0)
167 return;
168
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100169 /* The wake callback is normally used to notify the data layer about
170 * data layer activity (successful send/recv), connection establishment,
171 * shutdown and fatal errors. We need to consider the following
172 * situations to wake up the data layer :
173 * - change among the CO_FL_NOTIFY_DATA flags :
174 * {DATA,SOCK}_{RD,WR}_SH, ERROR,
175 * - absence of any of {L4,L6}_CONN and CONNECTED, indicating the
176 * end of handshake and transition to CONNECTED
177 * - raise of CONNECTED with HANDSHAKE down
178 * - end of HANDSHAKE with CONNECTED set
179 * - regular data layer activity
180 *
181 * Note that the wake callback is allowed to release the connection and
182 * the fd (and return < 0 in this case).
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200183 */
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200184 if ((io_available || (((conn->flags ^ flags) & CO_FL_NOTIFY_DATA) ||
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100185 ((flags & (CO_FL_CONNECTED|CO_FL_HANDSHAKE)) != CO_FL_CONNECTED &&
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200186 (conn->flags & (CO_FL_CONNECTED|CO_FL_HANDSHAKE)) == CO_FL_CONNECTED))) &&
Olivier Houchard7505f942018-08-21 18:10:44 +0200187 conn->mux->wake && conn->mux->wake(conn) < 0)
Willy Tarreau7a798e52016-04-14 11:13:20 +0200188 return;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200189
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200190 /* commit polling changes */
Willy Tarreau916e12d2017-10-25 09:22:43 +0200191 conn->flags &= ~CO_FL_WILL_UPDATE;
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200192 conn_cond_update_polling(conn);
Willy Tarreau7a798e52016-04-14 11:13:20 +0200193 return;
Willy Tarreau59f98392012-07-06 14:13:49 +0200194}
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200195
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200196/* Update polling on connection <c>'s file descriptor depending on its current
197 * state as reported in the connection's CO_FL_CURR_* flags, reports of EAGAIN
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200198 * in CO_FL_WAIT_*, and the data layer expectations indicated by CO_FL_XPRT_*.
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200199 * The connection flags are updated with the new flags at the end of the
Willy Tarreau0ffde2c2012-10-04 22:21:15 +0200200 * operation. Polling is totally disabled if an error was reported.
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200201 */
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200202void conn_update_xprt_polling(struct connection *c)
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200203{
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200204 unsigned int f = c->flags;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200205
Willy Tarreau3c728722014-01-23 13:50:42 +0100206 if (!conn_ctrl_ready(c))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200207 return;
208
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200209 /* update read status if needed */
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200210 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 +0200211 fd_want_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100212 f |= CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200213 }
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200214 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 +0200215 fd_stop_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100216 f &= ~CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200217 }
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200218
219 /* update write status if needed */
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200220 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 +0200221 fd_want_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100222 f |= CO_FL_CURR_WR_ENA;
223 }
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200224 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 +0200225 fd_stop_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100226 f &= ~CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200227 }
Willy Tarreau310987a2014-01-22 19:46:33 +0100228 c->flags = f;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200229}
230
231/* Update polling on connection <c>'s file descriptor depending on its current
232 * state as reported in the connection's CO_FL_CURR_* flags, reports of EAGAIN
233 * in CO_FL_WAIT_*, and the sock layer expectations indicated by CO_FL_SOCK_*.
234 * The connection flags are updated with the new flags at the end of the
Willy Tarreau0ffde2c2012-10-04 22:21:15 +0200235 * operation. Polling is totally disabled if an error was reported.
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200236 */
237void conn_update_sock_polling(struct connection *c)
238{
239 unsigned int f = c->flags;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200240
Willy Tarreau3c728722014-01-23 13:50:42 +0100241 if (!conn_ctrl_ready(c))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200242 return;
243
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200244 /* update read status if needed */
Willy Tarreau310987a2014-01-22 19:46:33 +0100245 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 +0200246 fd_want_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100247 f |= CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200248 }
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100249 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 +0200250 fd_stop_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100251 f &= ~CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200252 }
253
254 /* update write status if needed */
Willy Tarreau310987a2014-01-22 19:46:33 +0100255 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 +0200256 fd_want_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100257 f |= CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200258 }
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100259 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 +0200260 fd_stop_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100261 f &= ~CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200262 }
Willy Tarreau310987a2014-01-22 19:46:33 +0100263 c->flags = f;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200264}
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200265
Willy Tarreauff3e6482015-03-12 23:56:52 +0100266/* Send a message over an established connection. It makes use of send() and
267 * returns the same return code and errno. If the socket layer is not ready yet
268 * then -1 is returned and ENOTSOCK is set into errno. If the fd is not marked
269 * as ready, or if EAGAIN or ENOTCONN is returned, then we return 0. It returns
270 * EMSGSIZE if called with a zero length message. The purpose is to simplify
271 * some rare attempts to directly write on the socket from above the connection
272 * (typically send_proxy). In case of EAGAIN, the fd is marked as "cant_send".
273 * It automatically retries on EINTR. Other errors cause the connection to be
274 * marked as in error state. It takes similar arguments as send() except the
275 * first one which is the connection instead of the file descriptor. Note,
276 * MSG_DONTWAIT and MSG_NOSIGNAL are forced on the flags.
277 */
278int conn_sock_send(struct connection *conn, const void *buf, int len, int flags)
279{
280 int ret;
281
282 ret = -1;
283 errno = ENOTSOCK;
284
285 if (conn->flags & CO_FL_SOCK_WR_SH)
286 goto fail;
287
288 if (!conn_ctrl_ready(conn))
289 goto fail;
290
291 errno = EMSGSIZE;
292 if (!len)
293 goto fail;
294
Willy Tarreau585744b2017-08-24 14:31:19 +0200295 if (!fd_send_ready(conn->handle.fd))
Willy Tarreauff3e6482015-03-12 23:56:52 +0100296 goto wait;
297
298 do {
Willy Tarreau585744b2017-08-24 14:31:19 +0200299 ret = send(conn->handle.fd, buf, len, flags | MSG_DONTWAIT | MSG_NOSIGNAL);
Willy Tarreauff3e6482015-03-12 23:56:52 +0100300 } while (ret < 0 && errno == EINTR);
301
302
303 if (ret > 0)
304 return ret;
305
306 if (ret == 0 || errno == EAGAIN || errno == ENOTCONN) {
307 wait:
Willy Tarreau585744b2017-08-24 14:31:19 +0200308 fd_cant_send(conn->handle.fd);
Willy Tarreauff3e6482015-03-12 23:56:52 +0100309 return 0;
310 }
311 fail:
312 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH | CO_FL_ERROR;
313 return ret;
314}
315
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100316int conn_unsubscribe(struct connection *conn, void *xprt_ctx, int event_type, void *param)
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200317{
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200318 struct wait_event *sw;
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200319
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100320 if (event_type & SUB_RETRY_RECV) {
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200321 sw = param;
Olivier Houchard35d11682019-05-14 18:02:47 +0200322 BUG_ON(conn->recv_wait != sw);
323 conn->recv_wait = NULL;
324 sw->events &= ~SUB_RETRY_RECV;
Olivier Houchard53216e72018-10-10 15:46:36 +0200325 __conn_xprt_stop_recv(conn);
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200326 }
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100327 if (event_type & SUB_RETRY_SEND) {
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200328 sw = param;
Olivier Houchard35d11682019-05-14 18:02:47 +0200329 BUG_ON(conn->send_wait != sw);
330 conn->send_wait = NULL;
331 sw->events &= ~SUB_RETRY_SEND;
Olivier Houchard53216e72018-10-10 15:46:36 +0200332 __conn_xprt_stop_send(conn);
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200333 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200334 conn_update_xprt_polling(conn);
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200335 return 0;
336}
337
Olivier Houcharde179d0e2019-03-21 18:27:17 +0100338int conn_subscribe(struct connection *conn, void *xprt_ctx, int event_type, void *param)
Olivier Houchard6ff20392018-07-17 18:46:31 +0200339{
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200340 struct wait_event *sw;
Olivier Houchard6ff20392018-07-17 18:46:31 +0200341
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100342 if (event_type & SUB_RETRY_RECV) {
Olivier Houchard4cf7fb12018-08-02 19:23:05 +0200343 sw = param;
Olivier Houchard35d11682019-05-14 18:02:47 +0200344 BUG_ON(conn->recv_wait != NULL || (sw->events & SUB_RETRY_RECV));
345 sw->events |= SUB_RETRY_RECV;
346 conn->recv_wait = sw;
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100347 event_type &= ~SUB_RETRY_RECV;
Olivier Houchard53216e72018-10-10 15:46:36 +0200348 __conn_xprt_want_recv(conn);
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200349 }
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100350 if (event_type & SUB_RETRY_SEND) {
Olivier Houchard6ff20392018-07-17 18:46:31 +0200351 sw = param;
Olivier Houchard35d11682019-05-14 18:02:47 +0200352 BUG_ON(conn->send_wait != NULL || (sw->events & SUB_RETRY_SEND));
353 sw->events |= SUB_RETRY_SEND;
354 conn->send_wait = sw;
Willy Tarreau4f6516d2018-12-19 13:59:17 +0100355 event_type &= ~SUB_RETRY_SEND;
Olivier Houchard53216e72018-10-10 15:46:36 +0200356 __conn_xprt_want_send(conn);
Olivier Houchard6ff20392018-07-17 18:46:31 +0200357 }
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200358 if (event_type != 0)
359 return (-1);
Olivier Houchard53216e72018-10-10 15:46:36 +0200360 conn_update_xprt_polling(conn);
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200361 return 0;
Olivier Houchard6ff20392018-07-17 18:46:31 +0200362}
363
Willy Tarreaud85c4852015-03-13 00:40:28 +0100364/* Drains possibly pending incoming data on the file descriptor attached to the
365 * connection and update the connection's flags accordingly. This is used to
366 * know whether we need to disable lingering on close. Returns non-zero if it
367 * is safe to close without disabling lingering, otherwise zero. The SOCK_RD_SH
368 * flag may also be updated if the incoming shutdown was reported by the drain()
369 * function.
370 */
371int conn_sock_drain(struct connection *conn)
372{
Willy Tarreaue215bba2018-08-24 14:31:53 +0200373 int turns = 2;
374 int len;
375
Willy Tarreaud85c4852015-03-13 00:40:28 +0100376 if (!conn_ctrl_ready(conn))
377 return 1;
378
379 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH))
380 return 1;
381
Willy Tarreaue215bba2018-08-24 14:31:53 +0200382 if (fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP))
383 goto shut;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100384
Willy Tarreaue215bba2018-08-24 14:31:53 +0200385 if (!fd_recv_ready(conn->handle.fd))
386 return 0;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100387
Willy Tarreaue215bba2018-08-24 14:31:53 +0200388 if (conn->ctrl->drain) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200389 if (conn->ctrl->drain(conn->handle.fd) <= 0)
Willy Tarreaud85c4852015-03-13 00:40:28 +0100390 return 0;
Willy Tarreaue215bba2018-08-24 14:31:53 +0200391 goto shut;
392 }
393
394 /* no drain function defined, use the generic one */
395
396 while (turns) {
397#ifdef MSG_TRUNC_CLEARS_INPUT
398 len = recv(conn->handle.fd, NULL, INT_MAX, MSG_DONTWAIT | MSG_NOSIGNAL | MSG_TRUNC);
399 if (len == -1 && errno == EFAULT)
400#endif
401 len = recv(conn->handle.fd, trash.area, trash.size,
402 MSG_DONTWAIT | MSG_NOSIGNAL);
403
404 if (len == 0)
405 goto shut;
406
407 if (len < 0) {
408 if (errno == EAGAIN) {
409 /* connection not closed yet */
410 fd_cant_recv(conn->handle.fd);
411 break;
412 }
413 if (errno == EINTR) /* oops, try again */
414 continue;
415 /* other errors indicate a dead connection, fine. */
416 goto shut;
417 }
418 /* OK we read some data, let's try again once */
419 turns--;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100420 }
421
Willy Tarreaue215bba2018-08-24 14:31:53 +0200422 /* some data are still present, give up */
423 return 0;
424
425 shut:
426 /* we're certain the connection was shut down */
427 fdtab[conn->handle.fd].linger_risk = 0;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100428 conn->flags |= CO_FL_SOCK_RD_SH;
429 return 1;
430}
431
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100432/*
433 * Get data length from tlv
434 */
435static int get_tlv_length(const struct tlv *src)
436{
437 return (src->length_hi << 8) | src->length_lo;
438}
439
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200440/* This handshake handler waits a PROXY protocol header at the beginning of the
441 * raw data stream. The header looks like this :
442 *
443 * "PROXY" <SP> PROTO <SP> SRC3 <SP> DST3 <SP> SRC4 <SP> <DST4> "\r\n"
444 *
445 * There must be exactly one space between each field. Fields are :
446 * - PROTO : layer 4 protocol, which must be "TCP4" or "TCP6".
447 * - SRC3 : layer 3 (eg: IP) source address in standard text form
448 * - DST3 : layer 3 (eg: IP) destination address in standard text form
449 * - SRC4 : layer 4 (eg: TCP port) source address in standard text form
450 * - DST4 : layer 4 (eg: TCP port) destination address in standard text form
451 *
452 * This line MUST be at the beginning of the buffer and MUST NOT wrap.
453 *
454 * The header line is small and in all cases smaller than the smallest normal
455 * TCP MSS. So it MUST always be delivered as one segment, which ensures we
456 * can safely use MSG_PEEK and avoid buffering.
457 *
458 * Once the data is fetched, the values are set in the connection's address
459 * fields, and data are removed from the socket's buffer. The function returns
460 * zero if it needs to wait for more data or if it fails, or 1 if it completed
461 * and removed itself.
462 */
463int conn_recv_proxy(struct connection *conn, int flag)
464{
465 char *line, *end;
Willy Tarreau77992672014-06-14 11:06:17 +0200466 struct proxy_hdr_v2 *hdr_v2;
467 const char v2sig[] = PP2_SIGNATURE;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100468 int tlv_length = 0;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200469 int tlv_offset = 0;
Willy Tarreaub406b872018-08-22 05:20:32 +0200470 int ret;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200471
472 /* we might have been called just after an asynchronous shutr */
473 if (conn->flags & CO_FL_SOCK_RD_SH)
474 goto fail;
475
Willy Tarreau3c728722014-01-23 13:50:42 +0100476 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200477 goto fail;
478
Willy Tarreau585744b2017-08-24 14:31:19 +0200479 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100480 return 0;
481
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200482 do {
Willy Tarreaub406b872018-08-22 05:20:32 +0200483 ret = recv(conn->handle.fd, trash.area, trash.size, MSG_PEEK);
484 if (ret < 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200485 if (errno == EINTR)
486 continue;
487 if (errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200488 fd_cant_recv(conn->handle.fd);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200489 return 0;
490 }
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100491 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200492 }
Willy Tarreaub406b872018-08-22 05:20:32 +0200493 trash.data = ret;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200494 } while (0);
495
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200496 if (!trash.data) {
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100497 /* client shutdown */
498 conn->err_code = CO_ER_PRX_EMPTY;
499 goto fail;
500 }
501
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200502 if (trash.data < 6)
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200503 goto missing;
504
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200505 line = trash.area;
506 end = trash.area + trash.data;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200507
508 /* Decode a possible proxy request, fail early if it does not match */
Willy Tarreau77992672014-06-14 11:06:17 +0200509 if (strncmp(line, "PROXY ", 6) != 0)
510 goto not_v1;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200511
512 line += 6;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200513 if (trash.data < 9) /* shortest possible line */
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200514 goto missing;
515
David CARLIER42ff05e2016-03-24 09:22:36 +0000516 if (memcmp(line, "TCP4 ", 5) == 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200517 u32 src3, dst3, sport, dport;
518
519 line += 5;
520
521 src3 = inetaddr_host_lim_ret(line, end, &line);
522 if (line == end)
523 goto missing;
524 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100525 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200526
527 dst3 = inetaddr_host_lim_ret(line, end, &line);
528 if (line == end)
529 goto missing;
530 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100531 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200532
533 sport = read_uint((const char **)&line, end);
534 if (line == end)
535 goto missing;
536 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100537 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200538
539 dport = read_uint((const char **)&line, end);
540 if (line > end - 2)
541 goto missing;
542 if (*line++ != '\r')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100543 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200544 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100545 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200546
547 /* update the session's addresses and mark them set */
548 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
549 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = htonl(src3);
550 ((struct sockaddr_in *)&conn->addr.from)->sin_port = htons(sport);
551
552 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
553 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = htonl(dst3);
554 ((struct sockaddr_in *)&conn->addr.to)->sin_port = htons(dport);
555 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
556 }
David CARLIER42ff05e2016-03-24 09:22:36 +0000557 else if (memcmp(line, "TCP6 ", 5) == 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200558 u32 sport, dport;
559 char *src_s;
560 char *dst_s, *sport_s, *dport_s;
561 struct in6_addr src3, dst3;
562
563 line += 5;
564
565 src_s = line;
566 dst_s = sport_s = dport_s = NULL;
567 while (1) {
568 if (line > end - 2) {
569 goto missing;
570 }
571 else if (*line == '\r') {
572 *line = 0;
573 line++;
574 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100575 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200576 break;
577 }
578
579 if (*line == ' ') {
580 *line = 0;
581 if (!dst_s)
582 dst_s = line + 1;
583 else if (!sport_s)
584 sport_s = line + 1;
585 else if (!dport_s)
586 dport_s = line + 1;
587 }
588 line++;
589 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200590 __conn_xprt_stop_recv(conn);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200591
592 if (!dst_s || !sport_s || !dport_s)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100593 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200594
595 sport = read_uint((const char **)&sport_s,dport_s - 1);
596 if (*sport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100597 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200598
599 dport = read_uint((const char **)&dport_s,line - 2);
600 if (*dport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100601 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200602
603 if (inet_pton(AF_INET6, src_s, (void *)&src3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100604 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200605
606 if (inet_pton(AF_INET6, dst_s, (void *)&dst3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100607 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200608
609 /* update the session's addresses and mark them set */
610 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
611 memcpy(&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr, &src3, sizeof(struct in6_addr));
612 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = htons(sport);
613
614 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
615 memcpy(&((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr, &dst3, sizeof(struct in6_addr));
616 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = htons(dport);
617 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
618 }
Willy Tarreau4c20d292014-06-14 11:41:36 +0200619 else if (memcmp(line, "UNKNOWN\r\n", 9) == 0) {
620 /* This can be a UNIX socket forwarded by an haproxy upstream */
621 line += 9;
622 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200623 else {
Willy Tarreau4c20d292014-06-14 11:41:36 +0200624 /* The protocol does not match something known (TCP4/TCP6/UNKNOWN) */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100625 conn->err_code = CO_ER_PRX_BAD_PROTO;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200626 goto fail;
627 }
628
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200629 trash.data = line - trash.area;
Willy Tarreau77992672014-06-14 11:06:17 +0200630 goto eat_header;
631
632 not_v1:
633 /* try PPv2 */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200634 if (trash.data < PP2_HEADER_LEN)
Willy Tarreau77992672014-06-14 11:06:17 +0200635 goto missing;
636
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200637 hdr_v2 = (struct proxy_hdr_v2 *) trash.area;
Willy Tarreau77992672014-06-14 11:06:17 +0200638
639 if (memcmp(hdr_v2->sig, v2sig, PP2_SIGNATURE_LEN) != 0 ||
640 (hdr_v2->ver_cmd & PP2_VERSION_MASK) != PP2_VERSION) {
641 conn->err_code = CO_ER_PRX_NOT_HDR;
642 goto fail;
643 }
644
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200645 if (trash.data < PP2_HEADER_LEN + ntohs(hdr_v2->len))
Willy Tarreau77992672014-06-14 11:06:17 +0200646 goto missing;
647
648 switch (hdr_v2->ver_cmd & PP2_CMD_MASK) {
649 case 0x01: /* PROXY command */
650 switch (hdr_v2->fam) {
651 case 0x11: /* TCPv4 */
KOVACS Krisztianefd3aa92014-11-19 10:53:20 +0100652 if (ntohs(hdr_v2->len) < PP2_ADDR_LEN_INET)
653 goto bad_header;
654
Willy Tarreau77992672014-06-14 11:06:17 +0200655 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
656 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = hdr_v2->addr.ip4.src_addr;
657 ((struct sockaddr_in *)&conn->addr.from)->sin_port = hdr_v2->addr.ip4.src_port;
658 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
659 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = hdr_v2->addr.ip4.dst_addr;
660 ((struct sockaddr_in *)&conn->addr.to)->sin_port = hdr_v2->addr.ip4.dst_port;
661 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200662 tlv_offset = PP2_HEADER_LEN + PP2_ADDR_LEN_INET;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100663 tlv_length = ntohs(hdr_v2->len) - PP2_ADDR_LEN_INET;
Willy Tarreau77992672014-06-14 11:06:17 +0200664 break;
665 case 0x21: /* TCPv6 */
KOVACS Krisztianefd3aa92014-11-19 10:53:20 +0100666 if (ntohs(hdr_v2->len) < PP2_ADDR_LEN_INET6)
667 goto bad_header;
668
Willy Tarreau77992672014-06-14 11:06:17 +0200669 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
670 memcpy(&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr, hdr_v2->addr.ip6.src_addr, 16);
671 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = hdr_v2->addr.ip6.src_port;
672 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
673 memcpy(&((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr, hdr_v2->addr.ip6.dst_addr, 16);
674 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = hdr_v2->addr.ip6.dst_port;
675 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200676 tlv_offset = PP2_HEADER_LEN + PP2_ADDR_LEN_INET6;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100677 tlv_length = ntohs(hdr_v2->len) - PP2_ADDR_LEN_INET6;
Willy Tarreau77992672014-06-14 11:06:17 +0200678 break;
679 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100680
681 /* TLV parsing */
682 if (tlv_length > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200683 while (tlv_offset + TLV_HEADER_SIZE <= trash.data) {
684 const struct tlv *tlv_packet = (struct tlv *) &trash.area[tlv_offset];
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100685 const int tlv_len = get_tlv_length(tlv_packet);
686 tlv_offset += tlv_len + TLV_HEADER_SIZE;
687
688 switch (tlv_packet->type) {
Emmanuel Hocdet115df3e2018-02-05 16:23:23 +0100689 case PP2_TYPE_CRC32C: {
690 void *tlv_crc32c_p = (void *)tlv_packet->value;
691 uint32_t n_crc32c = ntohl(read_u32(tlv_crc32c_p));
692 write_u32(tlv_crc32c_p, 0);
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200693 if (hash_crc32c(trash.area, PP2_HEADER_LEN + ntohs(hdr_v2->len)) != n_crc32c)
Emmanuel Hocdet115df3e2018-02-05 16:23:23 +0100694 goto bad_header;
695 break;
696 }
Willy Tarreaue5733232019-05-22 19:24:06 +0200697#ifdef USE_NS
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100698 case PP2_TYPE_NETNS: {
699 const struct netns_entry *ns;
700 ns = netns_store_lookup((char*)tlv_packet->value, tlv_len);
701 if (ns)
702 conn->proxy_netns = ns;
703 break;
704 }
705#endif
706 default:
707 break;
708 }
709 }
710 }
711
Willy Tarreau77992672014-06-14 11:06:17 +0200712 /* unsupported protocol, keep local connection address */
713 break;
714 case 0x00: /* LOCAL command */
715 /* keep local connection address for LOCAL */
716 break;
717 default:
718 goto bad_header; /* not a supported command */
719 }
720
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200721 trash.data = PP2_HEADER_LEN + ntohs(hdr_v2->len);
Willy Tarreau77992672014-06-14 11:06:17 +0200722 goto eat_header;
723
724 eat_header:
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200725 /* remove the PROXY line from the request. For this we re-read the
726 * exact line at once. If we don't get the exact same result, we
727 * fail.
728 */
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200729 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200730 int len2 = recv(conn->handle.fd, trash.area, trash.data, 0);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200731 if (len2 < 0 && errno == EINTR)
732 continue;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200733 if (len2 != trash.data)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100734 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200735 } while (0);
736
737 conn->flags &= ~flag;
Emeric Brun4f603012017-01-05 15:11:44 +0100738 conn->flags |= CO_FL_RCVD_PROXY;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200739 return 1;
740
741 missing:
742 /* Missing data. Since we're using MSG_PEEK, we can only poll again if
743 * we have not read anything. Otherwise we need to fail because we won't
744 * be able to poll anymore.
745 */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100746 conn->err_code = CO_ER_PRX_TRUNCATED;
747 goto fail;
748
749 bad_header:
750 /* This is not a valid proxy protocol header */
751 conn->err_code = CO_ER_PRX_BAD_HDR;
752 goto fail;
753
754 recv_abort:
755 conn->err_code = CO_ER_PRX_ABORT;
Willy Tarreau26f4a042013-12-04 23:44:10 +0100756 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100757 goto fail;
758
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200759 fail:
Willy Tarreaud486ef52012-12-10 17:03:52 +0100760 __conn_sock_stop_both(conn);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200761 conn->flags |= CO_FL_ERROR;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200762 return 0;
763}
764
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100765/* This handshake handler waits a NetScaler Client IP insertion header
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000766 * at the beginning of the raw data stream. The header format is
767 * described in doc/netscaler-client-ip-insertion-protocol.txt
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100768 *
769 * This line MUST be at the beginning of the buffer and MUST NOT be
770 * fragmented.
771 *
772 * The header line is small and in all cases smaller than the smallest normal
773 * TCP MSS. So it MUST always be delivered as one segment, which ensures we
774 * can safely use MSG_PEEK and avoid buffering.
775 *
776 * Once the data is fetched, the values are set in the connection's address
777 * fields, and data are removed from the socket's buffer. The function returns
778 * zero if it needs to wait for more data or if it fails, or 1 if it completed
779 * and removed itself.
780 */
781int conn_recv_netscaler_cip(struct connection *conn, int flag)
782{
783 char *line;
Bertrand Jacquin7d668f92017-12-13 01:23:39 +0000784 uint32_t hdr_len;
Willy Tarreau0ca24aa2019-03-29 17:35:32 +0100785 uint8_t ip_ver;
Willy Tarreaub406b872018-08-22 05:20:32 +0200786 int ret;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100787
788 /* we might have been called just after an asynchronous shutr */
789 if (conn->flags & CO_FL_SOCK_RD_SH)
790 goto fail;
791
792 if (!conn_ctrl_ready(conn))
793 goto fail;
794
Willy Tarreau585744b2017-08-24 14:31:19 +0200795 if (!fd_recv_ready(conn->handle.fd))
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100796 return 0;
797
798 do {
Willy Tarreaub406b872018-08-22 05:20:32 +0200799 ret = recv(conn->handle.fd, trash.area, trash.size, MSG_PEEK);
800 if (ret < 0) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100801 if (errno == EINTR)
802 continue;
803 if (errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200804 fd_cant_recv(conn->handle.fd);
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100805 return 0;
806 }
807 goto recv_abort;
808 }
Willy Tarreaub406b872018-08-22 05:20:32 +0200809 trash.data = ret;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100810 } while (0);
811
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200812 if (!trash.data) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100813 /* client shutdown */
814 conn->err_code = CO_ER_CIP_EMPTY;
815 goto fail;
816 }
817
818 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000819 * CIP magic, header length or
820 * CIP magic, CIP length, CIP type, header length */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200821 if (trash.data < 12)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100822 goto missing;
823
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200824 line = trash.area;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100825
826 /* Decode a possible NetScaler Client IP request, fail early if
827 * it does not match */
Willy Tarreau55e0da62018-09-20 11:26:52 +0200828 if (ntohl(*(uint32_t *)line) != __objt_listener(conn->target)->bind_conf->ns_cip_magic)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100829 goto bad_magic;
830
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000831 /* Legacy CIP protocol */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200832 if ((trash.area[8] & 0xD0) == 0x40) {
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000833 hdr_len = ntohl(*(uint32_t *)(line+4));
834 line += 8;
835 }
836 /* Standard CIP protocol */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200837 else if (trash.area[8] == 0x00) {
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000838 hdr_len = ntohs(*(uint32_t *)(line+10));
839 line += 12;
840 }
841 /* Unknown CIP protocol */
842 else {
843 conn->err_code = CO_ER_CIP_BAD_PROTO;
844 goto fail;
845 }
846
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100847 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000848 * a minimal IP header */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200849 if (trash.data < 20)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100850 goto missing;
851
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100852 /* Get IP version from the first four bits */
Willy Tarreau0ca24aa2019-03-29 17:35:32 +0100853 ip_ver = (*line & 0xf0) >> 4;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100854
Willy Tarreau0ca24aa2019-03-29 17:35:32 +0100855 if (ip_ver == 4) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100856 struct ip *hdr_ip4;
David Carlier3015a2e2016-07-04 22:51:33 +0100857 struct my_tcphdr *hdr_tcp;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100858
859 hdr_ip4 = (struct ip *)line;
860
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200861 if (trash.data < 40 || trash.data < hdr_len) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100862 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin67de5a22017-12-13 01:15:05 +0000863 * IPv4 header, TCP header */
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100864 goto missing;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000865 }
866 else if (hdr_ip4->ip_p != IPPROTO_TCP) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100867 /* The protocol does not include a TCP header */
868 conn->err_code = CO_ER_CIP_BAD_PROTO;
869 goto fail;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000870 }
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100871
David Carlier3015a2e2016-07-04 22:51:33 +0100872 hdr_tcp = (struct my_tcphdr *)(line + (hdr_ip4->ip_hl * 4));
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100873
874 /* update the session's addresses and mark them set */
875 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
876 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = hdr_ip4->ip_src.s_addr;
877 ((struct sockaddr_in *)&conn->addr.from)->sin_port = hdr_tcp->source;
878
879 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
880 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = hdr_ip4->ip_dst.s_addr;
881 ((struct sockaddr_in *)&conn->addr.to)->sin_port = hdr_tcp->dest;
882
883 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
884 }
Willy Tarreau0ca24aa2019-03-29 17:35:32 +0100885 else if (ip_ver == 6) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100886 struct ip6_hdr *hdr_ip6;
David Carlier3015a2e2016-07-04 22:51:33 +0100887 struct my_tcphdr *hdr_tcp;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100888
889 hdr_ip6 = (struct ip6_hdr *)line;
890
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200891 if (trash.data < 60 || trash.data < hdr_len) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100892 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin67de5a22017-12-13 01:15:05 +0000893 * IPv6 header, TCP header */
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100894 goto missing;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000895 }
896 else if (hdr_ip6->ip6_nxt != IPPROTO_TCP) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100897 /* The protocol does not include a TCP header */
898 conn->err_code = CO_ER_CIP_BAD_PROTO;
899 goto fail;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000900 }
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100901
David Carlier3015a2e2016-07-04 22:51:33 +0100902 hdr_tcp = (struct my_tcphdr *)(line + sizeof(struct ip6_hdr));
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100903
904 /* update the session's addresses and mark them set */
905 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
906 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr = hdr_ip6->ip6_src;
907 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = hdr_tcp->source;
908
909 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
910 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr = hdr_ip6->ip6_dst;
911 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = hdr_tcp->dest;
912
913 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
914 }
915 else {
916 /* The protocol does not match something known (IPv4/IPv6) */
917 conn->err_code = CO_ER_CIP_BAD_PROTO;
918 goto fail;
919 }
920
Bertrand Jacquin7d668f92017-12-13 01:23:39 +0000921 line += hdr_len;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200922 trash.data = line - trash.area;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100923
924 /* remove the NetScaler Client IP header from the request. For this
925 * we re-read the exact line at once. If we don't get the exact same
926 * result, we fail.
927 */
928 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200929 int len2 = recv(conn->handle.fd, trash.area, trash.data, 0);
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100930 if (len2 < 0 && errno == EINTR)
931 continue;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200932 if (len2 != trash.data)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100933 goto recv_abort;
934 } while (0);
935
936 conn->flags &= ~flag;
937 return 1;
938
939 missing:
940 /* Missing data. Since we're using MSG_PEEK, we can only poll again if
941 * we have not read anything. Otherwise we need to fail because we won't
942 * be able to poll anymore.
943 */
944 conn->err_code = CO_ER_CIP_TRUNCATED;
945 goto fail;
946
947 bad_magic:
948 conn->err_code = CO_ER_CIP_BAD_MAGIC;
949 goto fail;
950
951 recv_abort:
952 conn->err_code = CO_ER_CIP_ABORT;
953 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
954 goto fail;
955
956 fail:
957 __conn_sock_stop_both(conn);
958 conn->flags |= CO_FL_ERROR;
959 return 0;
960}
961
Ilya Shipitsinca56fce2018-09-15 00:50:05 +0500962/* Note: <remote> is explicitly allowed to be NULL */
David Safb76832014-05-08 23:42:08 -0400963int make_proxy_line(char *buf, int buf_len, struct server *srv, struct connection *remote)
964{
965 int ret = 0;
966
967 if (srv && (srv->pp_opts & SRV_PP_V2)) {
968 ret = make_proxy_line_v2(buf, buf_len, srv, remote);
969 }
970 else {
971 if (remote)
972 ret = make_proxy_line_v1(buf, buf_len, &remote->addr.from, &remote->addr.to);
973 else
974 ret = make_proxy_line_v1(buf, buf_len, NULL, NULL);
975 }
976
977 return ret;
978}
979
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200980/* Makes a PROXY protocol line from the two addresses. The output is sent to
981 * buffer <buf> for a maximum size of <buf_len> (including the trailing zero).
982 * It returns the number of bytes composing this line (including the trailing
983 * LF), or zero in case of failure (eg: not enough space). It supports TCP4,
Willy Tarreau2e1401a2013-10-01 11:41:55 +0200984 * TCP6 and "UNKNOWN" formats. If any of <src> or <dst> is null, UNKNOWN is
985 * emitted as well.
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200986 */
David Safb76832014-05-08 23:42:08 -0400987int make_proxy_line_v1(char *buf, int buf_len, struct sockaddr_storage *src, struct sockaddr_storage *dst)
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200988{
989 int ret = 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200990 char * protocol;
991 char src_str[MAX(INET_ADDRSTRLEN, INET6_ADDRSTRLEN)];
992 char dst_str[MAX(INET_ADDRSTRLEN, INET6_ADDRSTRLEN)];
993 in_port_t src_port;
994 in_port_t dst_port;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200995
Tim Duesterhus7fec0212018-07-27 18:46:13 +0200996 if ( !src
997 || !dst
998 || (src->ss_family != AF_INET && src->ss_family != AF_INET6)
999 || (dst->ss_family != AF_INET && dst->ss_family != AF_INET6)) {
1000 /* unknown family combination */
1001 ret = snprintf(buf, buf_len, "PROXY UNKNOWN\r\n");
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001002 if (ret >= buf_len)
1003 return 0;
1004
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001005 return ret;
1006 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001007
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001008 /* IPv4 for both src and dst */
1009 if (src->ss_family == AF_INET && dst->ss_family == AF_INET) {
1010 protocol = "TCP4";
1011 if (!inet_ntop(AF_INET, &((struct sockaddr_in *)src)->sin_addr, src_str, sizeof(src_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001012 return 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001013 src_port = ((struct sockaddr_in *)src)->sin_port;
1014 if (!inet_ntop(AF_INET, &((struct sockaddr_in *)dst)->sin_addr, dst_str, sizeof(dst_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001015 return 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001016 dst_port = ((struct sockaddr_in *)dst)->sin_port;
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001017 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001018 /* IPv6 for at least one of src and dst */
1019 else {
1020 struct in6_addr tmp;
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001021
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001022 protocol = "TCP6";
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001023
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001024 if (src->ss_family == AF_INET) {
1025 /* Convert src to IPv6 */
1026 v4tov6(&tmp, &((struct sockaddr_in *)src)->sin_addr);
1027 src_port = ((struct sockaddr_in *)src)->sin_port;
1028 }
1029 else {
1030 tmp = ((struct sockaddr_in6 *)src)->sin6_addr;
1031 src_port = ((struct sockaddr_in6 *)src)->sin6_port;
1032 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001033
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001034 if (!inet_ntop(AF_INET6, &tmp, src_str, sizeof(src_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001035 return 0;
1036
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001037 if (dst->ss_family == AF_INET) {
1038 /* Convert dst to IPv6 */
1039 v4tov6(&tmp, &((struct sockaddr_in *)dst)->sin_addr);
1040 dst_port = ((struct sockaddr_in *)dst)->sin_port;
1041 }
1042 else {
1043 tmp = ((struct sockaddr_in6 *)dst)->sin6_addr;
1044 dst_port = ((struct sockaddr_in6 *)dst)->sin6_port;
1045 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001046
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001047 if (!inet_ntop(AF_INET6, &tmp, dst_str, sizeof(dst_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001048 return 0;
1049 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001050
1051 ret = snprintf(buf, buf_len, "PROXY %s %s %s %u %u\r\n", protocol, src_str, dst_str, ntohs(src_port), ntohs(dst_port));
1052 if (ret >= buf_len)
1053 return 0;
1054
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001055 return ret;
1056}
David Safb76832014-05-08 23:42:08 -04001057
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001058static int make_tlv(char *dest, int dest_len, char type, uint16_t length, const char *value)
David Safb76832014-05-08 23:42:08 -04001059{
1060 struct tlv *tlv;
1061
1062 if (!dest || (length + sizeof(*tlv) > dest_len))
1063 return 0;
1064
1065 tlv = (struct tlv *)dest;
1066
1067 tlv->type = type;
1068 tlv->length_hi = length >> 8;
1069 tlv->length_lo = length & 0x00ff;
1070 memcpy(tlv->value, value, length);
1071 return length + sizeof(*tlv);
1072}
David Safb76832014-05-08 23:42:08 -04001073
Ilya Shipitsinca56fce2018-09-15 00:50:05 +05001074/* Note: <remote> is explicitly allowed to be NULL */
David Safb76832014-05-08 23:42:08 -04001075int make_proxy_line_v2(char *buf, int buf_len, struct server *srv, struct connection *remote)
1076{
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001077 const char pp2_signature[] = PP2_SIGNATURE;
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001078 void *tlv_crc32c_p = NULL;
David Safb76832014-05-08 23:42:08 -04001079 int ret = 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001080 struct proxy_hdr_v2 *hdr = (struct proxy_hdr_v2 *)buf;
Vincent Bernat6e615892016-05-18 16:17:44 +02001081 struct sockaddr_storage null_addr = { .ss_family = 0 };
David Safb76832014-05-08 23:42:08 -04001082 struct sockaddr_storage *src = &null_addr;
1083 struct sockaddr_storage *dst = &null_addr;
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001084 const char *value;
1085 int value_len;
David Safb76832014-05-08 23:42:08 -04001086
1087 if (buf_len < PP2_HEADER_LEN)
1088 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001089 memcpy(hdr->sig, pp2_signature, PP2_SIGNATURE_LEN);
David Safb76832014-05-08 23:42:08 -04001090
1091 if (remote) {
1092 src = &remote->addr.from;
1093 dst = &remote->addr.to;
1094 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001095
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001096 /* At least one of src or dst is not of AF_INET or AF_INET6 */
1097 if ( !src
1098 || !dst
1099 || (src->ss_family != AF_INET && src->ss_family != AF_INET6)
1100 || (dst->ss_family != AF_INET && dst->ss_family != AF_INET6)) {
David Safb76832014-05-08 23:42:08 -04001101 if (buf_len < PP2_HDR_LEN_UNSPEC)
1102 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001103 hdr->ver_cmd = PP2_VERSION | PP2_CMD_LOCAL;
1104 hdr->fam = PP2_FAM_UNSPEC | PP2_TRANS_UNSPEC;
David Safb76832014-05-08 23:42:08 -04001105 ret = PP2_HDR_LEN_UNSPEC;
1106 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001107 else {
1108 /* IPv4 for both src and dst */
1109 if (src->ss_family == AF_INET && dst->ss_family == AF_INET) {
1110 if (buf_len < PP2_HDR_LEN_INET)
1111 return 0;
1112 hdr->ver_cmd = PP2_VERSION | PP2_CMD_PROXY;
1113 hdr->fam = PP2_FAM_INET | PP2_TRANS_STREAM;
1114 hdr->addr.ip4.src_addr = ((struct sockaddr_in *)src)->sin_addr.s_addr;
1115 hdr->addr.ip4.src_port = ((struct sockaddr_in *)src)->sin_port;
1116 hdr->addr.ip4.dst_addr = ((struct sockaddr_in *)dst)->sin_addr.s_addr;
1117 hdr->addr.ip4.dst_port = ((struct sockaddr_in *)dst)->sin_port;
1118 ret = PP2_HDR_LEN_INET;
1119 }
1120 /* IPv6 for at least one of src and dst */
1121 else {
1122 struct in6_addr tmp;
1123
1124 if (buf_len < PP2_HDR_LEN_INET6)
1125 return 0;
1126 hdr->ver_cmd = PP2_VERSION | PP2_CMD_PROXY;
1127 hdr->fam = PP2_FAM_INET6 | PP2_TRANS_STREAM;
1128 if (src->ss_family == AF_INET) {
1129 v4tov6(&tmp, &((struct sockaddr_in *)src)->sin_addr);
1130 memcpy(hdr->addr.ip6.src_addr, &tmp, 16);
1131 hdr->addr.ip6.src_port = ((struct sockaddr_in *)src)->sin_port;
1132 }
1133 else {
1134 memcpy(hdr->addr.ip6.src_addr, &((struct sockaddr_in6 *)src)->sin6_addr, 16);
1135 hdr->addr.ip6.src_port = ((struct sockaddr_in6 *)src)->sin6_port;
1136 }
1137 if (dst->ss_family == AF_INET) {
1138 v4tov6(&tmp, &((struct sockaddr_in *)dst)->sin_addr);
1139 memcpy(hdr->addr.ip6.dst_addr, &tmp, 16);
1140 hdr->addr.ip6.src_port = ((struct sockaddr_in *)src)->sin_port;
1141 }
1142 else {
1143 memcpy(hdr->addr.ip6.dst_addr, &((struct sockaddr_in6 *)dst)->sin6_addr, 16);
1144 hdr->addr.ip6.dst_port = ((struct sockaddr_in6 *)dst)->sin6_port;
1145 }
1146
1147 ret = PP2_HDR_LEN_INET6;
1148 }
1149 }
David Safb76832014-05-08 23:42:08 -04001150
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001151 if (srv->pp_opts & SRV_PP_V2_CRC32C) {
1152 uint32_t zero_crc32c = 0;
1153 if ((buf_len - ret) < sizeof(struct tlv))
1154 return 0;
1155 tlv_crc32c_p = (void *)((struct tlv *)&buf[ret])->value;
1156 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_CRC32C, sizeof(zero_crc32c), (const char *)&zero_crc32c);
1157 }
1158
Ilya Shipitsinca56fce2018-09-15 00:50:05 +05001159 if (remote && conn_get_alpn(remote, &value, &value_len)) {
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001160 if ((buf_len - ret) < sizeof(struct tlv))
1161 return 0;
Emmanuel Hocdet571c7ac2017-10-31 18:24:05 +01001162 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_ALPN, value_len, value);
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001163 }
1164
David Safb76832014-05-08 23:42:08 -04001165#ifdef USE_OPENSSL
Emmanuel Hocdet253c3b72018-02-01 18:29:59 +01001166 if (srv->pp_opts & SRV_PP_V2_AUTHORITY) {
1167 value = ssl_sock_get_sni(remote);
1168 if (value) {
1169 if ((buf_len - ret) < sizeof(struct tlv))
1170 return 0;
1171 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_AUTHORITY, strlen(value), value);
1172 }
1173 }
1174
David Safb76832014-05-08 23:42:08 -04001175 if (srv->pp_opts & SRV_PP_V2_SSL) {
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001176 struct tlv_ssl *tlv;
1177 int ssl_tlv_len = 0;
David Safb76832014-05-08 23:42:08 -04001178 if ((buf_len - ret) < sizeof(struct tlv_ssl))
1179 return 0;
1180 tlv = (struct tlv_ssl *)&buf[ret];
1181 memset(tlv, 0, sizeof(struct tlv_ssl));
1182 ssl_tlv_len += sizeof(struct tlv_ssl);
1183 tlv->tlv.type = PP2_TYPE_SSL;
1184 if (ssl_sock_is_ssl(remote)) {
1185 tlv->client |= PP2_CLIENT_SSL;
Emmanuel Hocdet01da5712017-10-13 16:59:49 +02001186 value = ssl_sock_get_proto_version(remote);
David Safb76832014-05-08 23:42:08 -04001187 if (value) {
Emmanuel Hocdet8c0c34b2018-02-28 12:02:14 +01001188 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 -04001189 }
Dave McCowan328fb582014-07-30 10:39:13 -04001190 if (ssl_sock_get_cert_used_sess(remote)) {
1191 tlv->client |= PP2_CLIENT_CERT_SESS;
David Safb76832014-05-08 23:42:08 -04001192 tlv->verify = htonl(ssl_sock_get_verify_result(remote));
Dave McCowan328fb582014-07-30 10:39:13 -04001193 if (ssl_sock_get_cert_used_conn(remote))
1194 tlv->client |= PP2_CLIENT_CERT_CONN;
David Safb76832014-05-08 23:42:08 -04001195 }
1196 if (srv->pp_opts & SRV_PP_V2_SSL_CN) {
Willy Tarreau83061a82018-07-13 11:56:34 +02001197 struct buffer *cn_trash = get_trash_chunk();
Willy Tarreau3b9a0c92014-07-19 06:37:33 +02001198 if (ssl_sock_get_remote_common_name(remote, cn_trash) > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001199 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_CN,
1200 cn_trash->data,
1201 cn_trash->area);
David Safb76832014-05-08 23:42:08 -04001202 }
1203 }
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001204 if (srv->pp_opts & SRV_PP_V2_SSL_KEY_ALG) {
Willy Tarreau83061a82018-07-13 11:56:34 +02001205 struct buffer *pkey_trash = get_trash_chunk();
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001206 if (ssl_sock_get_pkey_algo(remote, pkey_trash) > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001207 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_KEY_ALG,
1208 pkey_trash->data,
1209 pkey_trash->area);
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001210 }
1211 }
1212 if (srv->pp_opts & SRV_PP_V2_SSL_SIG_ALG) {
1213 value = ssl_sock_get_cert_sig(remote);
1214 if (value) {
1215 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_SIG_ALG, strlen(value), value);
1216 }
1217 }
1218 if (srv->pp_opts & SRV_PP_V2_SSL_CIPHER) {
1219 value = ssl_sock_get_cipher_name(remote);
1220 if (value) {
1221 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_CIPHER, strlen(value), value);
1222 }
1223 }
David Safb76832014-05-08 23:42:08 -04001224 }
1225 tlv->tlv.length_hi = (uint16_t)(ssl_tlv_len - sizeof(struct tlv)) >> 8;
1226 tlv->tlv.length_lo = (uint16_t)(ssl_tlv_len - sizeof(struct tlv)) & 0x00ff;
1227 ret += ssl_tlv_len;
1228 }
1229#endif
1230
Willy Tarreaue5733232019-05-22 19:24:06 +02001231#ifdef USE_NS
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001232 if (remote && (remote->proxy_netns)) {
1233 if ((buf_len - ret) < sizeof(struct tlv))
1234 return 0;
Emmanuel Hocdet571c7ac2017-10-31 18:24:05 +01001235 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 +01001236 }
1237#endif
1238
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001239 hdr->len = htons((uint16_t)(ret - PP2_HEADER_LEN));
David Safb76832014-05-08 23:42:08 -04001240
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001241 if (tlv_crc32c_p) {
1242 write_u32(tlv_crc32c_p, htonl(hash_crc32c(buf, ret)));
1243 }
1244
David Safb76832014-05-08 23:42:08 -04001245 return ret;
1246}
Emeric Brun4f603012017-01-05 15:11:44 +01001247
Willy Tarreau60ca10a2017-08-18 15:26:54 +02001248/* return the major HTTP version as 1 or 2 depending on how the request arrived
1249 * before being processed.
1250 */
1251static int
1252smp_fetch_fc_http_major(const struct arg *args, struct sample *smp, const char *kw, void *private)
1253{
Jérôme Magnin86577422018-12-07 09:03:11 +01001254 struct connection *conn = (kw[0] != 'b') ? objt_conn(smp->sess->origin) :
1255 smp->strm ? cs_conn(objt_cs(smp->strm->si[1].end)) : NULL;
Willy Tarreau60ca10a2017-08-18 15:26:54 +02001256
1257 smp->data.type = SMP_T_SINT;
1258 smp->data.u.sint = (conn && strcmp(conn_get_mux_name(conn), "H2") == 0) ? 2 : 1;
1259 return 1;
1260}
1261
Emeric Brun4f603012017-01-05 15:11:44 +01001262/* fetch if the received connection used a PROXY protocol header */
1263int smp_fetch_fc_rcvd_proxy(const struct arg *args, struct sample *smp, const char *kw, void *private)
1264{
1265 struct connection *conn;
1266
1267 conn = objt_conn(smp->sess->origin);
1268 if (!conn)
1269 return 0;
1270
1271 if (!(conn->flags & CO_FL_CONNECTED)) {
1272 smp->flags |= SMP_F_MAY_CHANGE;
1273 return 0;
1274 }
1275
1276 smp->flags = 0;
1277 smp->data.type = SMP_T_BOOL;
1278 smp->data.u.sint = (conn->flags & CO_FL_RCVD_PROXY) ? 1 : 0;
1279
1280 return 1;
1281}
1282
1283/* Note: must not be declared <const> as its list will be overwritten.
1284 * Note: fetches that may return multiple types must be declared as the lowest
1285 * common denominator, the type that can be casted into all other ones. For
1286 * instance v4/v6 must be declared v4.
1287 */
1288static struct sample_fetch_kw_list sample_fetch_keywords = {ILH, {
Willy Tarreau60ca10a2017-08-18 15:26:54 +02001289 { "fc_http_major", smp_fetch_fc_http_major, 0, NULL, SMP_T_SINT, SMP_USE_L4CLI },
Jérôme Magnin86577422018-12-07 09:03:11 +01001290 { "bc_http_major", smp_fetch_fc_http_major, 0, NULL, SMP_T_SINT, SMP_USE_L4SRV },
Emeric Brun4f603012017-01-05 15:11:44 +01001291 { "fc_rcvd_proxy", smp_fetch_fc_rcvd_proxy, 0, NULL, SMP_T_BOOL, SMP_USE_L4CLI },
1292 { /* END */ },
1293}};
1294
Willy Tarreau0108d902018-11-25 19:14:37 +01001295INITCALL1(STG_REGISTER, sample_register_fetches, &sample_fetch_keywords);