blob: b477b463106bf6e628adb61abcb06b94c395afb9 [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>
17
Willy Tarreauc5788912012-08-24 18:12:41 +020018#include <proto/connection.h>
Willy Tarreaudd2f85e2012-09-02 22:34:23 +020019#include <proto/fd.h>
Willy Tarreau5f1504f2012-10-04 23:55:57 +020020#include <proto/frontend.h>
Willy Tarreau2da156f2012-07-23 15:07:23 +020021#include <proto/proto_tcp.h>
Willy Tarreau2542b532012-08-31 16:01:23 +020022#include <proto/session.h>
Willy Tarreau2c6be842012-07-06 17:12:34 +020023#include <proto/stream_interface.h>
Willy Tarreau59f98392012-07-06 14:13:49 +020024
Emeric Brun46591952012-05-18 15:47:34 +020025#ifdef USE_OPENSSL
26#include <proto/ssl_sock.h>
27#endif
28
Willy Tarreauf2943dc2012-10-26 20:10:28 +020029struct pool_head *pool2_connection;
30
31/* perform minimal intializations, report 0 in case of error, 1 if OK. */
32int init_connection()
33{
34 pool2_connection = create_pool("connection", sizeof (struct connection), MEM_F_SHARED);
35 return pool2_connection != NULL;
36}
37
Willy Tarreau59f98392012-07-06 14:13:49 +020038/* I/O callback for fd-based connections. It calls the read/write handlers
Willy Tarreauafad0e02012-08-09 14:45:22 +020039 * provided by the connection's sock_ops, which must be valid. It returns 0.
Willy Tarreau59f98392012-07-06 14:13:49 +020040 */
41int conn_fd_handler(int fd)
42{
Willy Tarreau80184712012-07-06 14:54:49 +020043 struct connection *conn = fdtab[fd].owner;
Willy Tarreau9e272bf2012-10-03 21:04:48 +020044 unsigned int flags;
Willy Tarreau59f98392012-07-06 14:13:49 +020045
Willy Tarreauc76ae332012-07-12 15:32:13 +020046 if (unlikely(!conn))
Willy Tarreau2542b532012-08-31 16:01:23 +020047 return 0;
Willy Tarreau59f98392012-07-06 14:13:49 +020048
Willy Tarreau7d281492012-12-16 19:19:13 +010049 conn_refresh_polling_flags(conn);
50 flags = conn->flags & ~CO_FL_ERROR; /* ensure to call the wake handler upon error */
Willy Tarreaud29a0662012-12-10 16:33:38 +010051
Willy Tarreauc76ae332012-07-12 15:32:13 +020052 process_handshake:
Willy Tarreauf9dabec2012-08-17 17:33:53 +020053 /* The handshake callbacks are called in sequence. If either of them is
54 * missing something, it must enable the required polling at the socket
55 * layer of the connection. Polling state is not guaranteed when entering
56 * these handlers, so any handshake handler which does not complete its
Willy Tarreaud6e999b2013-11-25 08:41:15 +010057 * work must explicitly disable events it's not interested in. Error
58 * handling is also performed here in order to reduce the number of tests
59 * around.
Willy Tarreauf9dabec2012-08-17 17:33:53 +020060 */
Willy Tarreaud6e999b2013-11-25 08:41:15 +010061 while (unlikely(conn->flags & (CO_FL_HANDSHAKE | CO_FL_ERROR))) {
Willy Tarreaud9de7ca2012-09-02 18:48:46 +020062 if (unlikely(conn->flags & (CO_FL_ERROR|CO_FL_WAIT_RD|CO_FL_WAIT_WR)))
Willy Tarreau2c6be842012-07-06 17:12:34 +020063 goto leave;
Willy Tarreau59f98392012-07-06 14:13:49 +020064
Willy Tarreau22cda212012-08-31 17:43:29 +020065 if (conn->flags & CO_FL_ACCEPT_PROXY)
66 if (!conn_recv_proxy(conn, CO_FL_ACCEPT_PROXY))
67 goto leave;
68
Willy Tarreauc76ae332012-07-12 15:32:13 +020069 if (conn->flags & CO_FL_SI_SEND_PROXY)
Willy Tarreauafad0e02012-08-09 14:45:22 +020070 if (!conn_si_send_proxy(conn, CO_FL_SI_SEND_PROXY))
Willy Tarreauc76ae332012-07-12 15:32:13 +020071 goto leave;
Willy Tarreau5f1504f2012-10-04 23:55:57 +020072
73 if (conn->flags & CO_FL_LOCAL_SPROXY)
74 if (!conn_local_send_proxy(conn, CO_FL_LOCAL_SPROXY))
75 goto leave;
Emeric Brun46591952012-05-18 15:47:34 +020076#ifdef USE_OPENSSL
77 if (conn->flags & CO_FL_SSL_WAIT_HS)
78 if (!ssl_sock_handshake(conn, CO_FL_SSL_WAIT_HS))
79 goto leave;
80#endif
Willy Tarreauc76ae332012-07-12 15:32:13 +020081 }
82
Willy Tarreauf9dabec2012-08-17 17:33:53 +020083 /* Once we're purely in the data phase, we disable handshake polling */
84 if (!(conn->flags & CO_FL_POLL_SOCK))
85 __conn_sock_stop_both(conn);
Willy Tarreauc76ae332012-07-12 15:32:13 +020086
Willy Tarreau071e1372012-10-03 01:39:48 +020087 /* The data layer might not be ready yet (eg: when using embryonic
88 * sessions). If we're about to move data, we must initialize it first.
89 * The function may fail and cause the connection to be destroyed, thus
Willy Tarreau2542b532012-08-31 16:01:23 +020090 * we must not use it anymore and should immediately leave instead.
91 */
Willy Tarreau071e1372012-10-03 01:39:48 +020092 if ((conn->flags & CO_FL_INIT_DATA) && conn->data->init(conn) < 0)
Willy Tarreau2542b532012-08-31 16:01:23 +020093 return 0;
94
Willy Tarreau153c3ca2012-10-22 22:47:55 +020095 /* The data transfer starts here and stops on error and handshakes. Note
96 * that we must absolutely test conn->xprt at each step in case it suddenly
97 * changes due to a quick unexpected close().
98 */
Willy Tarreaud9de7ca2012-09-02 18:48:46 +020099 if ((fdtab[fd].ev & (FD_POLL_IN | FD_POLL_HUP | FD_POLL_ERR)) &&
Willy Tarreau153c3ca2012-10-22 22:47:55 +0200100 conn->xprt &&
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200101 !(conn->flags & (CO_FL_WAIT_RD|CO_FL_WAIT_ROOM|CO_FL_ERROR|CO_FL_HANDSHAKE))) {
Willy Tarreau3b5bc662012-10-05 21:29:37 +0200102 /* force detection of a flag change : it's impossible to have both
103 * CONNECTED and WAIT_CONN so we're certain to trigger a change.
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200104 */
Willy Tarreau3b5bc662012-10-05 21:29:37 +0200105 flags = CO_FL_WAIT_L4_CONN | CO_FL_CONNECTED;
Willy Tarreau74beec32012-10-03 00:41:04 +0200106 conn->data->recv(conn);
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200107 }
Willy Tarreau59f98392012-07-06 14:13:49 +0200108
Willy Tarreaud9de7ca2012-09-02 18:48:46 +0200109 if ((fdtab[fd].ev & (FD_POLL_OUT | FD_POLL_ERR)) &&
Willy Tarreau153c3ca2012-10-22 22:47:55 +0200110 conn->xprt &&
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200111 !(conn->flags & (CO_FL_WAIT_WR|CO_FL_WAIT_DATA|CO_FL_ERROR|CO_FL_HANDSHAKE))) {
Willy Tarreau3b5bc662012-10-05 21:29:37 +0200112 /* force detection of a flag change : it's impossible to have both
113 * CONNECTED and WAIT_CONN so we're certain to trigger a change.
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200114 */
Willy Tarreau3b5bc662012-10-05 21:29:37 +0200115 flags = CO_FL_WAIT_L4_CONN | CO_FL_CONNECTED;
Willy Tarreau74beec32012-10-03 00:41:04 +0200116 conn->data->send(conn);
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200117 }
Willy Tarreau2da156f2012-07-23 15:07:23 +0200118
Willy Tarreauc76ae332012-07-12 15:32:13 +0200119 /* It may happen during the data phase that a handshake is
120 * enabled again (eg: SSL)
121 */
Willy Tarreaud6e999b2013-11-25 08:41:15 +0100122 if (unlikely(conn->flags & (CO_FL_HANDSHAKE | CO_FL_ERROR)))
Willy Tarreauc76ae332012-07-12 15:32:13 +0200123 goto process_handshake;
124
Willy Tarreauf8deb0c2012-09-01 17:59:22 +0200125 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN) && !(conn->flags & CO_FL_WAIT_WR)) {
126 /* still waiting for a connection to establish and nothing was
127 * attempted yet to probe the connection. Then let's retry the
128 * connect().
Willy Tarreau2da156f2012-07-23 15:07:23 +0200129 */
Willy Tarreau239d7182012-07-23 18:53:03 +0200130 if (!tcp_connect_probe(conn))
Willy Tarreauafad0e02012-08-09 14:45:22 +0200131 goto leave;
Willy Tarreau2da156f2012-07-23 15:07:23 +0200132 }
133
Willy Tarreau2c6be842012-07-06 17:12:34 +0200134 leave:
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200135 /* The wake callback may be used to process a critical error and abort the
136 * connection. If so, we don't want to go further as the connection will
137 * have been released and the FD destroyed.
138 */
139 if ((conn->flags & CO_FL_WAKE_DATA) &&
140 ((conn->flags ^ flags) & CO_FL_CONN_STATE) &&
141 conn->data->wake(conn) < 0)
142 return 0;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200143
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200144 /* Last check, verify if the connection just established */
Willy Tarreauc76ae332012-07-12 15:32:13 +0200145 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED))))
Willy Tarreau8f8c92f2012-07-23 19:45:44 +0200146 conn->flags |= CO_FL_CONNECTED;
147
Willy Tarreau61ace1b2012-07-23 12:14:26 +0200148 /* remove the events before leaving */
Willy Tarreau26d7cfc2012-12-07 00:09:43 +0100149 fdtab[fd].ev &= FD_POLL_STICKY;
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200150
151 /* commit polling changes */
152 conn_cond_update_polling(conn);
Willy Tarreauafad0e02012-08-09 14:45:22 +0200153 return 0;
Willy Tarreau59f98392012-07-06 14:13:49 +0200154}
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200155
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200156/* Update polling on connection <c>'s file descriptor depending on its current
157 * state as reported in the connection's CO_FL_CURR_* flags, reports of EAGAIN
158 * in CO_FL_WAIT_*, and the data layer expectations indicated by CO_FL_DATA_*.
159 * The connection flags are updated with the new flags at the end of the
Willy Tarreau0ffde2c2012-10-04 22:21:15 +0200160 * operation. Polling is totally disabled if an error was reported.
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200161 */
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200162void conn_update_data_polling(struct connection *c)
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200163{
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200164 unsigned int f = c->flags;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200165
166 /* update read status if needed */
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100167 if (unlikely((f & (CO_FL_DATA_RD_ENA|CO_FL_WAIT_RD)) == (CO_FL_DATA_RD_ENA|CO_FL_WAIT_RD))) {
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200168 fd_poll_recv(c->t.sock.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100169 f |= CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200170 }
171 else if (unlikely((f & (CO_FL_CURR_RD_ENA|CO_FL_DATA_RD_ENA)) == CO_FL_DATA_RD_ENA)) {
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200172 fd_want_recv(c->t.sock.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100173 f |= CO_FL_CURR_RD_ENA;
174 }
175 else if (unlikely((f & (CO_FL_CURR_RD_ENA|CO_FL_DATA_RD_ENA)) == CO_FL_CURR_RD_ENA)) {
176 fd_stop_recv(c->t.sock.fd);
177 f &= ~CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200178 }
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200179
180 /* update write status if needed */
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100181 if (unlikely((f & (CO_FL_DATA_WR_ENA|CO_FL_WAIT_WR)) == (CO_FL_DATA_WR_ENA|CO_FL_WAIT_WR))) {
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200182 fd_poll_send(c->t.sock.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100183 f |= CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200184 }
185 else if (unlikely((f & (CO_FL_CURR_WR_ENA|CO_FL_DATA_WR_ENA)) == CO_FL_DATA_WR_ENA)) {
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200186 fd_want_send(c->t.sock.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100187 f |= CO_FL_CURR_WR_ENA;
188 }
189 else if (unlikely((f & (CO_FL_CURR_WR_ENA|CO_FL_DATA_WR_ENA)) == CO_FL_CURR_WR_ENA)) {
190 fd_stop_send(c->t.sock.fd);
191 f &= ~CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200192 }
Willy Tarreauc9f78042012-11-05 20:00:43 +0100193 c->flags = f & ~(CO_FL_WAIT_RD | CO_FL_WAIT_WR);
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200194}
195
196/* 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
198 * in CO_FL_WAIT_*, and the sock layer expectations indicated by CO_FL_SOCK_*.
199 * 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 Tarreaue9dfa792012-09-01 17:26:16 +0200201 */
202void conn_update_sock_polling(struct connection *c)
203{
204 unsigned int f = c->flags;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200205
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200206 /* update read status if needed */
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100207 if (unlikely((f & (CO_FL_SOCK_RD_ENA|CO_FL_WAIT_RD)) == (CO_FL_SOCK_RD_ENA|CO_FL_WAIT_RD))) {
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200208 fd_poll_recv(c->t.sock.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100209 f |= CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200210 }
211 else if (unlikely((f & (CO_FL_CURR_RD_ENA|CO_FL_SOCK_RD_ENA)) == CO_FL_SOCK_RD_ENA)) {
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200212 fd_want_recv(c->t.sock.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100213 f |= CO_FL_CURR_RD_ENA;
214 }
215 else if (unlikely((f & (CO_FL_CURR_RD_ENA|CO_FL_SOCK_RD_ENA)) == CO_FL_CURR_RD_ENA)) {
216 fd_stop_recv(c->t.sock.fd);
217 f &= ~CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200218 }
219
220 /* update write status if needed */
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100221 if (unlikely((f & (CO_FL_SOCK_WR_ENA|CO_FL_WAIT_WR)) == (CO_FL_SOCK_WR_ENA|CO_FL_WAIT_WR))) {
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200222 fd_poll_send(c->t.sock.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100223 f |= CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200224 }
225 else if (unlikely((f & (CO_FL_CURR_WR_ENA|CO_FL_SOCK_WR_ENA)) == CO_FL_SOCK_WR_ENA)) {
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200226 fd_want_send(c->t.sock.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100227 f |= CO_FL_CURR_WR_ENA;
228 }
229 else if (unlikely((f & (CO_FL_CURR_WR_ENA|CO_FL_SOCK_WR_ENA)) == CO_FL_CURR_WR_ENA)) {
230 fd_stop_send(c->t.sock.fd);
231 f &= ~CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200232 }
Willy Tarreauc9f78042012-11-05 20:00:43 +0100233 c->flags = f & ~(CO_FL_WAIT_RD | CO_FL_WAIT_WR);
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200234}
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200235
236/* This handshake handler waits a PROXY protocol header at the beginning of the
237 * raw data stream. The header looks like this :
238 *
239 * "PROXY" <SP> PROTO <SP> SRC3 <SP> DST3 <SP> SRC4 <SP> <DST4> "\r\n"
240 *
241 * There must be exactly one space between each field. Fields are :
242 * - PROTO : layer 4 protocol, which must be "TCP4" or "TCP6".
243 * - SRC3 : layer 3 (eg: IP) source address in standard text form
244 * - DST3 : layer 3 (eg: IP) destination address in standard text form
245 * - SRC4 : layer 4 (eg: TCP port) source address in standard text form
246 * - DST4 : layer 4 (eg: TCP port) destination address in standard text form
247 *
248 * This line MUST be at the beginning of the buffer and MUST NOT wrap.
249 *
250 * The header line is small and in all cases smaller than the smallest normal
251 * TCP MSS. So it MUST always be delivered as one segment, which ensures we
252 * can safely use MSG_PEEK and avoid buffering.
253 *
254 * Once the data is fetched, the values are set in the connection's address
255 * fields, and data are removed from the socket's buffer. The function returns
256 * zero if it needs to wait for more data or if it fails, or 1 if it completed
257 * and removed itself.
258 */
259int conn_recv_proxy(struct connection *conn, int flag)
260{
261 char *line, *end;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200262
263 /* we might have been called just after an asynchronous shutr */
264 if (conn->flags & CO_FL_SOCK_RD_SH)
265 goto fail;
266
267 do {
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100268 trash.len = recv(conn->t.sock.fd, trash.str, trash.size, MSG_PEEK);
269 if (trash.len < 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200270 if (errno == EINTR)
271 continue;
272 if (errno == EAGAIN) {
Willy Tarreaud486ef52012-12-10 17:03:52 +0100273 __conn_sock_poll_recv(conn);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200274 return 0;
275 }
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100276 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200277 }
278 } while (0);
279
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100280 if (!trash.len) {
281 /* client shutdown */
282 conn->err_code = CO_ER_PRX_EMPTY;
283 goto fail;
284 }
285
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100286 if (trash.len < 6)
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200287 goto missing;
288
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100289 line = trash.str;
290 end = trash.str + trash.len;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200291
292 /* Decode a possible proxy request, fail early if it does not match */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100293 if (strncmp(line, "PROXY ", 6) != 0) {
294 conn->err_code = CO_ER_PRX_NOT_HDR;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200295 goto fail;
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100296 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200297
298 line += 6;
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100299 if (trash.len < 18) /* shortest possible line */
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200300 goto missing;
301
302 if (!memcmp(line, "TCP4 ", 5) != 0) {
303 u32 src3, dst3, sport, dport;
304
305 line += 5;
306
307 src3 = inetaddr_host_lim_ret(line, end, &line);
308 if (line == end)
309 goto missing;
310 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100311 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200312
313 dst3 = inetaddr_host_lim_ret(line, end, &line);
314 if (line == end)
315 goto missing;
316 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100317 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200318
319 sport = read_uint((const char **)&line, end);
320 if (line == end)
321 goto missing;
322 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100323 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200324
325 dport = read_uint((const char **)&line, end);
326 if (line > end - 2)
327 goto missing;
328 if (*line++ != '\r')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100329 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200330 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100331 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200332
333 /* update the session's addresses and mark them set */
334 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
335 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = htonl(src3);
336 ((struct sockaddr_in *)&conn->addr.from)->sin_port = htons(sport);
337
338 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
339 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = htonl(dst3);
340 ((struct sockaddr_in *)&conn->addr.to)->sin_port = htons(dport);
341 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
342 }
343 else if (!memcmp(line, "TCP6 ", 5) != 0) {
344 u32 sport, dport;
345 char *src_s;
346 char *dst_s, *sport_s, *dport_s;
347 struct in6_addr src3, dst3;
348
349 line += 5;
350
351 src_s = line;
352 dst_s = sport_s = dport_s = NULL;
353 while (1) {
354 if (line > end - 2) {
355 goto missing;
356 }
357 else if (*line == '\r') {
358 *line = 0;
359 line++;
360 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100361 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200362 break;
363 }
364
365 if (*line == ' ') {
366 *line = 0;
367 if (!dst_s)
368 dst_s = line + 1;
369 else if (!sport_s)
370 sport_s = line + 1;
371 else if (!dport_s)
372 dport_s = line + 1;
373 }
374 line++;
375 }
376
377 if (!dst_s || !sport_s || !dport_s)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100378 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200379
380 sport = read_uint((const char **)&sport_s,dport_s - 1);
381 if (*sport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100382 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200383
384 dport = read_uint((const char **)&dport_s,line - 2);
385 if (*dport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100386 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200387
388 if (inet_pton(AF_INET6, src_s, (void *)&src3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100389 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200390
391 if (inet_pton(AF_INET6, dst_s, (void *)&dst3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100392 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200393
394 /* update the session's addresses and mark them set */
395 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
396 memcpy(&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr, &src3, sizeof(struct in6_addr));
397 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = htons(sport);
398
399 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
400 memcpy(&((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr, &dst3, sizeof(struct in6_addr));
401 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = htons(dport);
402 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
403 }
404 else {
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100405 /* The protocol does not match something known (TCP4/TCP6) */
406 conn->err_code = CO_ER_PRX_BAD_PROTO;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200407 goto fail;
408 }
409
410 /* remove the PROXY line from the request. For this we re-read the
411 * exact line at once. If we don't get the exact same result, we
412 * fail.
413 */
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100414 trash.len = line - trash.str;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200415 do {
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100416 int len2 = recv(conn->t.sock.fd, trash.str, trash.len, 0);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200417 if (len2 < 0 && errno == EINTR)
418 continue;
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100419 if (len2 != trash.len)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100420 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200421 } while (0);
422
423 conn->flags &= ~flag;
424 return 1;
425
426 missing:
427 /* Missing data. Since we're using MSG_PEEK, we can only poll again if
428 * we have not read anything. Otherwise we need to fail because we won't
429 * be able to poll anymore.
430 */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100431 conn->err_code = CO_ER_PRX_TRUNCATED;
432 goto fail;
433
434 bad_header:
435 /* This is not a valid proxy protocol header */
436 conn->err_code = CO_ER_PRX_BAD_HDR;
437 goto fail;
438
439 recv_abort:
440 conn->err_code = CO_ER_PRX_ABORT;
441 goto fail;
442
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200443 fail:
Willy Tarreaud486ef52012-12-10 17:03:52 +0100444 __conn_sock_stop_both(conn);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200445 conn->flags |= CO_FL_ERROR;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200446 return 0;
447}
448
449/* Makes a PROXY protocol line from the two addresses. The output is sent to
450 * buffer <buf> for a maximum size of <buf_len> (including the trailing zero).
451 * It returns the number of bytes composing this line (including the trailing
452 * LF), or zero in case of failure (eg: not enough space). It supports TCP4,
453 * TCP6 and "UNKNOWN" formats.
454 */
455int make_proxy_line(char *buf, int buf_len, struct sockaddr_storage *src, struct sockaddr_storage *dst)
456{
457 int ret = 0;
458
459 if (src->ss_family == dst->ss_family && src->ss_family == AF_INET) {
460 ret = snprintf(buf + ret, buf_len - ret, "PROXY TCP4 ");
461 if (ret >= buf_len)
462 return 0;
463
464 /* IPv4 src */
465 if (!inet_ntop(src->ss_family, &((struct sockaddr_in *)src)->sin_addr, buf + ret, buf_len - ret))
466 return 0;
467
468 ret += strlen(buf + ret);
469 if (ret >= buf_len)
470 return 0;
471
472 buf[ret++] = ' ';
473
474 /* IPv4 dst */
475 if (!inet_ntop(dst->ss_family, &((struct sockaddr_in *)dst)->sin_addr, buf + ret, buf_len - ret))
476 return 0;
477
478 ret += strlen(buf + ret);
479 if (ret >= buf_len)
480 return 0;
481
482 /* source and destination ports */
483 ret += snprintf(buf + ret, buf_len - ret, " %u %u\r\n",
484 ntohs(((struct sockaddr_in *)src)->sin_port),
485 ntohs(((struct sockaddr_in *)dst)->sin_port));
486 if (ret >= buf_len)
487 return 0;
488 }
489 else if (src->ss_family == dst->ss_family && src->ss_family == AF_INET6) {
490 ret = snprintf(buf + ret, buf_len - ret, "PROXY TCP6 ");
491 if (ret >= buf_len)
492 return 0;
493
494 /* IPv6 src */
495 if (!inet_ntop(src->ss_family, &((struct sockaddr_in6 *)src)->sin6_addr, buf + ret, buf_len - ret))
496 return 0;
497
498 ret += strlen(buf + ret);
499 if (ret >= buf_len)
500 return 0;
501
502 buf[ret++] = ' ';
503
504 /* IPv6 dst */
505 if (!inet_ntop(dst->ss_family, &((struct sockaddr_in6 *)dst)->sin6_addr, buf + ret, buf_len - ret))
506 return 0;
507
508 ret += strlen(buf + ret);
509 if (ret >= buf_len)
510 return 0;
511
512 /* source and destination ports */
513 ret += snprintf(buf + ret, buf_len - ret, " %u %u\r\n",
514 ntohs(((struct sockaddr_in6 *)src)->sin6_port),
515 ntohs(((struct sockaddr_in6 *)dst)->sin6_port));
516 if (ret >= buf_len)
517 return 0;
518 }
519 else {
520 /* unknown family combination */
521 ret = snprintf(buf, buf_len, "PROXY UNKNOWN\r\n");
522 if (ret >= buf_len)
523 return 0;
524 }
525 return ret;
526}
Willy Tarreau5f1504f2012-10-04 23:55:57 +0200527
528/* This callback is used to send a valid PROXY protocol line to a socket being
529 * established from the local machine. It sets the protocol addresses to the
530 * local and remote address. This is typically used with health checks or when
531 * it is not possible to determine the other end's address. It returns 0 if it
532 * fails in a fatal way or needs to poll to go further, otherwise it returns
533 * non-zero and removes itself from the connection's flags (the bit is provided
534 * in <flag> by the caller). It is designed to be called by the connection
535 * handler and relies on it to commit polling changes. Note that this function
536 * expects to be able to send the whole line at once, which should always be
537 * possible since it is supposed to start at the first byte of the outgoing
538 * data segment.
539 */
540int conn_local_send_proxy(struct connection *conn, unsigned int flag)
541{
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100542 int ret;
Willy Tarreau5f1504f2012-10-04 23:55:57 +0200543
544 /* we might have been called just after an asynchronous shutw */
545 if (conn->flags & CO_FL_SOCK_WR_SH)
546 goto out_error;
547
Willy Tarreaue3635ed2012-11-24 11:23:04 +0100548 /* The target server expects a PROXY line to be sent first. Retrieving
549 * local or remote addresses may fail until the connection is established.
550 */
Willy Tarreau5f1504f2012-10-04 23:55:57 +0200551 conn_get_from_addr(conn);
552 if (!(conn->flags & CO_FL_ADDR_FROM_SET))
Willy Tarreaue3635ed2012-11-24 11:23:04 +0100553 goto out_wait;
Willy Tarreau5f1504f2012-10-04 23:55:57 +0200554
555 conn_get_to_addr(conn);
556 if (!(conn->flags & CO_FL_ADDR_TO_SET))
Willy Tarreaue3635ed2012-11-24 11:23:04 +0100557 goto out_wait;
Willy Tarreau5f1504f2012-10-04 23:55:57 +0200558
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100559 trash.len = make_proxy_line(trash.str, trash.size, &conn->addr.from, &conn->addr.to);
560 if (!trash.len)
Willy Tarreau5f1504f2012-10-04 23:55:57 +0200561 goto out_error;
562
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100563 /* we have to send the whole trash. If the data layer has a
Willy Tarreau5f1504f2012-10-04 23:55:57 +0200564 * pending write, we'll also set MSG_MORE.
565 */
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100566 ret = send(conn->t.sock.fd, trash.str, trash.len, (conn->flags & CO_FL_DATA_WR_ENA) ? MSG_MORE : 0);
Willy Tarreau5f1504f2012-10-04 23:55:57 +0200567
568 if (ret == 0)
569 goto out_wait;
570
571 if (ret < 0) {
Willy Tarreau95742a42013-09-03 09:02:11 +0200572 if (errno == EAGAIN || errno == ENOTCONN)
Willy Tarreau5f1504f2012-10-04 23:55:57 +0200573 goto out_wait;
574 goto out_error;
575 }
576
Willy Tarreau19d14ef2012-10-29 16:51:55 +0100577 if (ret != trash.len)
Willy Tarreau5f1504f2012-10-04 23:55:57 +0200578 goto out_error;
579
580 /* The connection is ready now, simply return and let the connection
581 * handler notify upper layers if needed.
582 */
583 if (conn->flags & CO_FL_WAIT_L4_CONN)
584 conn->flags &= ~CO_FL_WAIT_L4_CONN;
585 conn->flags &= ~flag;
586 return 1;
587
588 out_error:
589 /* Write error on the file descriptor */
590 conn->flags |= CO_FL_ERROR;
Willy Tarreau5f1504f2012-10-04 23:55:57 +0200591 return 0;
592
593 out_wait:
594 __conn_sock_stop_recv(conn);
595 __conn_sock_poll_send(conn);
596 return 0;
597}