blob: ed0812c20d69e0c7bb30577b04f0be6d2ea1510d [file] [log] [blame]
Willy Tarreaue6b98942007-10-29 01:09:36 +01001/*
2 * AF_INET/AF_INET6 SOCK_STREAM protocol layer (tcp)
3 *
Willy Tarreaue8c66af2008-01-13 18:40:14 +01004 * Copyright 2000-2008 Willy Tarreau <w@1wt.eu>
Willy Tarreaue6b98942007-10-29 01:09:36 +01005 *
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
13#include <ctype.h>
14#include <errno.h>
15#include <fcntl.h>
16#include <stdio.h>
17#include <stdlib.h>
18#include <string.h>
19#include <time.h>
20
Willy Tarreau9ea05a72009-06-14 12:07:01 +020021#include <netinet/tcp.h>
22
Willy Tarreaue6b98942007-10-29 01:09:36 +010023#include <sys/param.h>
24#include <sys/socket.h>
25#include <sys/stat.h>
26#include <sys/types.h>
27#include <sys/un.h>
28
Willy Tarreaub6866442008-07-14 23:54:42 +020029#include <common/cfgparse.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010030#include <common/compat.h>
31#include <common/config.h>
32#include <common/debug.h>
33#include <common/errors.h>
34#include <common/memory.h>
35#include <common/mini-clist.h>
36#include <common/standard.h>
37#include <common/time.h>
38#include <common/version.h>
39
Willy Tarreaue6b98942007-10-29 01:09:36 +010040#include <types/global.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010041
42#include <proto/acl.h>
43#include <proto/backend.h>
44#include <proto/buffers.h>
45#include <proto/fd.h>
46#include <proto/protocols.h>
47#include <proto/proto_tcp.h>
Willy Tarreaub6866442008-07-14 23:54:42 +020048#include <proto/proxy.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010049#include <proto/queue.h>
Willy Tarreaue6b98942007-10-29 01:09:36 +010050#include <proto/session.h>
51#include <proto/stream_sock.h>
52#include <proto/task.h>
53
Willy Tarreaue8c66af2008-01-13 18:40:14 +010054#ifdef CONFIG_HAP_CTTPROXY
55#include <import/ip_tproxy.h>
56#endif
57
Willy Tarreaue6b98942007-10-29 01:09:36 +010058static int tcp_bind_listeners(struct protocol *proto);
59
60/* Note: must not be declared <const> as its list will be overwritten */
61static struct protocol proto_tcpv4 = {
62 .name = "tcpv4",
63 .sock_domain = AF_INET,
64 .sock_type = SOCK_STREAM,
65 .sock_prot = IPPROTO_TCP,
66 .sock_family = AF_INET,
67 .sock_addrlen = sizeof(struct sockaddr_in),
68 .l3_addrlen = 32/8,
69 .read = &stream_sock_read,
70 .write = &stream_sock_write,
71 .bind_all = tcp_bind_listeners,
72 .unbind_all = unbind_all_listeners,
73 .enable_all = enable_all_listeners,
74 .listeners = LIST_HEAD_INIT(proto_tcpv4.listeners),
75 .nb_listeners = 0,
76};
77
78/* Note: must not be declared <const> as its list will be overwritten */
79static struct protocol proto_tcpv6 = {
80 .name = "tcpv6",
81 .sock_domain = AF_INET6,
82 .sock_type = SOCK_STREAM,
83 .sock_prot = IPPROTO_TCP,
84 .sock_family = AF_INET6,
85 .sock_addrlen = sizeof(struct sockaddr_in6),
86 .l3_addrlen = 128/8,
87 .read = &stream_sock_read,
88 .write = &stream_sock_write,
89 .bind_all = tcp_bind_listeners,
90 .unbind_all = unbind_all_listeners,
91 .enable_all = enable_all_listeners,
92 .listeners = LIST_HEAD_INIT(proto_tcpv6.listeners),
93 .nb_listeners = 0,
94};
95
Willy Tarreaue8c66af2008-01-13 18:40:14 +010096
97/* Binds ipv4 address <local> to socket <fd>, unless <flags> is set, in which
98 * case we try to bind <remote>. <flags> is a 2-bit field consisting of :
99 * - 0 : ignore remote address (may even be a NULL pointer)
100 * - 1 : use provided address
101 * - 2 : use provided port
102 * - 3 : use both
103 *
104 * The function supports multiple foreign binding methods :
105 * - linux_tproxy: we directly bind to the foreign address
106 * - cttproxy: we bind to a local address then nat.
107 * The second one can be used as a fallback for the first one.
108 * This function returns 0 when everything's OK, 1 if it could not bind, to the
109 * local address, 2 if it could not bind to the foreign address.
110 */
111int tcpv4_bind_socket(int fd, int flags, struct sockaddr_in *local, struct sockaddr_in *remote)
112{
113 struct sockaddr_in bind_addr;
114 int foreign_ok = 0;
115 int ret;
116
117#ifdef CONFIG_HAP_LINUX_TPROXY
118 static int ip_transp_working = 1;
119 if (flags && ip_transp_working) {
120 if (setsockopt(fd, SOL_IP, IP_TRANSPARENT, (char *) &one, sizeof(one)) == 0
121 || setsockopt(fd, SOL_IP, IP_FREEBIND, (char *) &one, sizeof(one)) == 0)
122 foreign_ok = 1;
123 else
124 ip_transp_working = 0;
125 }
126#endif
127 if (flags) {
128 memset(&bind_addr, 0, sizeof(bind_addr));
129 if (flags & 1)
130 bind_addr.sin_addr = remote->sin_addr;
131 if (flags & 2)
132 bind_addr.sin_port = remote->sin_port;
133 }
134
135 setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (char *) &one, sizeof(one));
136 if (foreign_ok) {
137 ret = bind(fd, (struct sockaddr *)&bind_addr, sizeof(bind_addr));
138 if (ret < 0)
139 return 2;
140 }
141 else {
142 ret = bind(fd, (struct sockaddr *)local, sizeof(*local));
143 if (ret < 0)
144 return 1;
145 }
146
147 if (!flags)
148 return 0;
149
150#ifdef CONFIG_HAP_CTTPROXY
151 if (!foreign_ok) {
152 struct in_tproxy itp1, itp2;
153 memset(&itp1, 0, sizeof(itp1));
154
155 itp1.op = TPROXY_ASSIGN;
156 itp1.v.addr.faddr = bind_addr.sin_addr;
157 itp1.v.addr.fport = bind_addr.sin_port;
158
159 /* set connect flag on socket */
160 itp2.op = TPROXY_FLAGS;
161 itp2.v.flags = ITP_CONNECT | ITP_ONCE;
162
163 if (setsockopt(fd, SOL_IP, IP_TPROXY, &itp1, sizeof(itp1)) != -1 &&
164 setsockopt(fd, SOL_IP, IP_TPROXY, &itp2, sizeof(itp2)) != -1) {
165 foreign_ok = 1;
166 }
167 }
168#endif
169 if (!foreign_ok)
170 /* we could not bind to a foreign address */
171 return 2;
172
173 return 0;
174}
Willy Tarreaue6b98942007-10-29 01:09:36 +0100175
176/* This function tries to bind a TCPv4/v6 listener. It may return a warning or
177 * an error message in <err> if the message is at most <errlen> bytes long
178 * (including '\0'). The return value is composed from ERR_ABORT, ERR_WARN,
179 * ERR_ALERT, ERR_RETRYABLE and ERR_FATAL. ERR_NONE indicates that everything
180 * was alright and that no message was returned. ERR_RETRYABLE means that an
181 * error occurred but that it may vanish after a retry (eg: port in use), and
182 * ERR_FATAL indicates a non-fixable error.ERR_WARN and ERR_ALERT do not alter
183 * the meaning of the error, but just indicate that a message is present which
184 * should be displayed with the respective level. Last, ERR_ABORT indicates
185 * that it's pointless to try to start other listeners. No error message is
186 * returned if errlen is NULL.
187 */
188int tcp_bind_listener(struct listener *listener, char *errmsg, int errlen)
189{
190 __label__ tcp_return, tcp_close_return;
191 int fd, err;
192 const char *msg = NULL;
193
194 /* ensure we never return garbage */
195 if (errmsg && errlen)
196 *errmsg = 0;
197
198 if (listener->state != LI_ASSIGNED)
199 return ERR_NONE; /* already bound */
200
201 err = ERR_NONE;
202
203 if ((fd = socket(listener->addr.ss_family, SOCK_STREAM, IPPROTO_TCP)) == -1) {
204 err |= ERR_RETRYABLE | ERR_ALERT;
205 msg = "cannot create listening socket";
206 goto tcp_return;
207 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100208
Willy Tarreaue6b98942007-10-29 01:09:36 +0100209 if (fd >= global.maxsock) {
210 err |= ERR_FATAL | ERR_ABORT | ERR_ALERT;
211 msg = "not enough free sockets (raise '-n' parameter)";
212 goto tcp_close_return;
213 }
214
215 if ((fcntl(fd, F_SETFL, O_NONBLOCK) == -1) ||
216 (setsockopt(fd, IPPROTO_TCP, TCP_NODELAY,
217 (char *) &one, sizeof(one)) == -1)) {
218 err |= ERR_FATAL | ERR_ALERT;
219 msg = "cannot make socket non-blocking";
220 goto tcp_close_return;
221 }
222
223 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (char *) &one, sizeof(one)) == -1) {
224 /* not fatal but should be reported */
225 msg = "cannot do so_reuseaddr";
226 err |= ERR_ALERT;
227 }
228
229 if (listener->options & LI_O_NOLINGER)
230 setsockopt(fd, SOL_SOCKET, SO_LINGER, (struct linger *) &nolinger, sizeof(struct linger));
Willy Tarreauedcf6682008-11-30 23:15:34 +0100231
Willy Tarreaue6b98942007-10-29 01:09:36 +0100232#ifdef SO_REUSEPORT
233 /* OpenBSD supports this. As it's present in old libc versions of Linux,
234 * it might return an error that we will silently ignore.
235 */
236 setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, (char *) &one, sizeof(one));
237#endif
Willy Tarreaub1e52e82008-01-13 14:49:51 +0100238#ifdef CONFIG_HAP_LINUX_TPROXY
Willy Tarreauedcf6682008-11-30 23:15:34 +0100239 if ((listener->options & LI_O_FOREIGN)
Willy Tarreau0a459892008-01-13 17:37:16 +0100240 && (setsockopt(fd, SOL_IP, IP_TRANSPARENT, (char *) &one, sizeof(one)) == -1)
241 && (setsockopt(fd, SOL_IP, IP_FREEBIND, (char *) &one, sizeof(one)) == -1)) {
Willy Tarreaub1e52e82008-01-13 14:49:51 +0100242 msg = "cannot make listening socket transparent";
243 err |= ERR_ALERT;
244 }
245#endif
Willy Tarreau5e6e2042009-02-04 17:19:29 +0100246#ifdef SO_BINDTODEVICE
247 /* Note: this might fail if not CAP_NET_RAW */
248 if (listener->interface) {
249 if (setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE,
Willy Tarreau604e8302009-03-06 00:48:23 +0100250 listener->interface, strlen(listener->interface) + 1) == -1) {
Willy Tarreau5e6e2042009-02-04 17:19:29 +0100251 msg = "cannot bind listener to device";
252 err |= ERR_WARN;
253 }
254 }
255#endif
Willy Tarreaue6b98942007-10-29 01:09:36 +0100256 if (bind(fd, (struct sockaddr *)&listener->addr, listener->proto->sock_addrlen) == -1) {
257 err |= ERR_RETRYABLE | ERR_ALERT;
258 msg = "cannot bind socket";
259 goto tcp_close_return;
260 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100261
Willy Tarreauc73ce2b2008-01-06 10:55:10 +0100262 if (listen(fd, listener->backlog ? listener->backlog : listener->maxconn) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +0100263 err |= ERR_RETRYABLE | ERR_ALERT;
264 msg = "cannot listen to socket";
265 goto tcp_close_return;
266 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100267
Willy Tarreau9ea05a72009-06-14 12:07:01 +0200268#ifdef TCP_QUICKACK
269 if (listener->options & LI_O_NOQUICKACK)
270 setsockopt(fd, SOL_TCP, TCP_QUICKACK, (char *) &zero, sizeof(zero));
271#endif
272
Willy Tarreaue6b98942007-10-29 01:09:36 +0100273 /* the socket is ready */
274 listener->fd = fd;
275 listener->state = LI_LISTEN;
276
277 /* the function for the accept() event */
278 fd_insert(fd);
279 fdtab[fd].cb[DIR_RD].f = listener->accept;
280 fdtab[fd].cb[DIR_WR].f = NULL; /* never called */
281 fdtab[fd].cb[DIR_RD].b = fdtab[fd].cb[DIR_WR].b = NULL;
Willy Tarreaueabf3132008-08-29 23:36:51 +0200282 fdtab[fd].owner = listener; /* reference the listener instead of a task */
Willy Tarreaue6b98942007-10-29 01:09:36 +0100283 fdtab[fd].state = FD_STLISTEN;
284 fdtab[fd].peeraddr = NULL;
285 fdtab[fd].peerlen = 0;
Willy Tarreaue6b98942007-10-29 01:09:36 +0100286 tcp_return:
287 if (msg && errlen)
288 strlcpy2(errmsg, msg, errlen);
289 return err;
290
291 tcp_close_return:
292 close(fd);
293 goto tcp_return;
294}
295
296/* This function creates all TCP sockets bound to the protocol entry <proto>.
297 * It is intended to be used as the protocol's bind_all() function.
298 * The sockets will be registered but not added to any fd_set, in order not to
299 * loose them across the fork(). A call to enable_all_listeners() is needed
300 * to complete initialization. The return value is composed from ERR_*.
301 */
302static int tcp_bind_listeners(struct protocol *proto)
303{
304 struct listener *listener;
305 int err = ERR_NONE;
306
307 list_for_each_entry(listener, &proto->listeners, proto_list) {
308 err |= tcp_bind_listener(listener, NULL, 0);
309 if ((err & ERR_CODE) == ERR_ABORT)
310 break;
311 }
312
313 return err;
314}
315
316/* Add listener to the list of tcpv4 listeners. The listener's state
317 * is automatically updated from LI_INIT to LI_ASSIGNED. The number of
318 * listeners is updated. This is the function to use to add a new listener.
319 */
320void tcpv4_add_listener(struct listener *listener)
321{
322 if (listener->state != LI_INIT)
323 return;
324 listener->state = LI_ASSIGNED;
325 listener->proto = &proto_tcpv4;
326 LIST_ADDQ(&proto_tcpv4.listeners, &listener->proto_list);
327 proto_tcpv4.nb_listeners++;
328}
329
330/* Add listener to the list of tcpv4 listeners. The listener's state
331 * is automatically updated from LI_INIT to LI_ASSIGNED. The number of
332 * listeners is updated. This is the function to use to add a new listener.
333 */
334void tcpv6_add_listener(struct listener *listener)
335{
336 if (listener->state != LI_INIT)
337 return;
338 listener->state = LI_ASSIGNED;
339 listener->proto = &proto_tcpv6;
340 LIST_ADDQ(&proto_tcpv6.listeners, &listener->proto_list);
341 proto_tcpv6.nb_listeners++;
342}
343
Willy Tarreauedcf6682008-11-30 23:15:34 +0100344/* This function performs the TCP request analysis on the current request. It
345 * returns 1 if the processing can continue on next analysers, or zero if it
346 * needs more data, encounters an error, or wants to immediately abort the
347 * request. It relies on buffers flags, and updates s->req->analysers. Its
348 * behaviour is rather simple:
Willy Tarreau86ef7dc2009-03-15 22:55:47 +0100349 * - the analyser should check for errors and timeouts, and react as expected.
350 * It does not have to close anything upon error, the caller will. Note that
351 * the caller also knows how to report errors and timeouts.
352 * - if the analyser does not have enough data, it must return 0 without calling
Willy Tarreauedcf6682008-11-30 23:15:34 +0100353 * other ones. It should also probably do a buffer_write_dis() to ensure
354 * that unprocessed data will not be forwarded. But that probably depends on
355 * the protocol.
356 * - if an analyser has enough data, it just has to pass on to the next
357 * analyser without using buffer_write_dis() (enabled by default).
358 * - if an analyser thinks it has no added value anymore staying here, it must
359 * reset its bit from the analysers flags in order not to be called anymore.
360 *
361 * In the future, analysers should be able to indicate that they want to be
362 * called after XXX bytes have been received (or transfered), and the min of
363 * all's wishes will be used to ring back (unless a special condition occurs).
364 */
365int tcp_inspect_request(struct session *s, struct buffer *req)
366{
367 struct tcp_rule *rule;
368 int partial;
369
370 DPRINTF(stderr,"[%u] %s: session=%p b=%p, exp(r,w)=%u,%u bf=%08x bl=%d analysers=%02x\n",
371 now_ms, __FUNCTION__,
372 s,
373 req,
374 req->rex, req->wex,
375 req->flags,
376 req->l,
377 req->analysers);
378
Willy Tarreauedcf6682008-11-30 23:15:34 +0100379 /* We don't know whether we have enough data, so must proceed
380 * this way :
381 * - iterate through all rules in their declaration order
382 * - if one rule returns MISS, it means the inspect delay is
383 * not over yet, then return immediately, otherwise consider
384 * it as a non-match.
385 * - if one rule returns OK, then return OK
386 * - if one rule returns KO, then return KO
387 */
388
Willy Tarreaud869b242009-03-15 14:43:58 +0100389 if (req->flags & BF_SHUTR || !s->fe->tcp_req.inspect_delay || tick_is_expired(req->analyse_exp, now_ms))
Willy Tarreauedcf6682008-11-30 23:15:34 +0100390 partial = 0;
391 else
392 partial = ACL_PARTIAL;
393
394 list_for_each_entry(rule, &s->fe->tcp_req.inspect_rules, list) {
395 int ret = ACL_PAT_PASS;
396
397 if (rule->cond) {
398 ret = acl_exec_cond(rule->cond, s->fe, s, NULL, ACL_DIR_REQ | partial);
399 if (ret == ACL_PAT_MISS) {
400 buffer_write_dis(req);
401 /* just set the request timeout once at the beginning of the request */
Willy Tarreaud869b242009-03-15 14:43:58 +0100402 if (!tick_isset(req->analyse_exp) && s->fe->tcp_req.inspect_delay)
Willy Tarreauedcf6682008-11-30 23:15:34 +0100403 req->analyse_exp = tick_add_ifset(now_ms, s->fe->tcp_req.inspect_delay);
404 return 0;
405 }
406
407 ret = acl_pass(ret);
408 if (rule->cond->pol == ACL_COND_UNLESS)
409 ret = !ret;
410 }
411
412 if (ret) {
413 /* we have a matching rule. */
414 if (rule->action == TCP_ACT_REJECT) {
415 buffer_abort(req);
416 buffer_abort(s->rep);
417 req->analysers = 0;
418 s->fe->failed_req++;
419 if (!(s->flags & SN_ERR_MASK))
420 s->flags |= SN_ERR_PRXCOND;
421 if (!(s->flags & SN_FINST_MASK))
422 s->flags |= SN_FINST_R;
423 return 0;
424 }
425 /* otherwise accept */
426 break;
427 }
428 }
429
430 /* if we get there, it means we have no rule which matches, or
431 * we have an explicit accept, so we apply the default accept.
432 */
433 req->analysers &= ~AN_REQ_INSPECT;
434 req->analyse_exp = TICK_ETERNITY;
435 return 1;
436}
437
438
Willy Tarreaub6866442008-07-14 23:54:42 +0200439/* This function should be called to parse a line starting with the "tcp-request"
440 * keyword.
441 */
442static int tcp_parse_tcp_req(char **args, int section_type, struct proxy *curpx,
443 struct proxy *defpx, char *err, int errlen)
444{
445 const char *ptr = NULL;
Willy Tarreauc7e961e2008-08-17 17:13:47 +0200446 unsigned int val;
Willy Tarreaub6866442008-07-14 23:54:42 +0200447 int retlen;
448
449 if (!*args[1]) {
450 snprintf(err, errlen, "missing argument for '%s' in %s '%s'",
451 args[0], proxy_type_str(proxy), curpx->id);
452 return -1;
453 }
454
455 if (!strcmp(args[1], "inspect-delay")) {
456 if (curpx == defpx) {
457 snprintf(err, errlen, "%s %s is not allowed in 'defaults' sections",
458 args[0], args[1]);
459 return -1;
460 }
461
462 if (!(curpx->cap & PR_CAP_FE)) {
463 snprintf(err, errlen, "%s %s will be ignored because %s '%s' has no %s capability",
464 args[0], args[1], proxy_type_str(proxy), curpx->id,
465 "frontend");
466 return 1;
467 }
468
469 if (!*args[2] || (ptr = parse_time_err(args[2], &val, TIME_UNIT_MS))) {
470 retlen = snprintf(err, errlen,
471 "'%s %s' expects a positive delay in milliseconds, in %s '%s'",
472 args[0], args[1], proxy_type_str(proxy), curpx->id);
473 if (ptr && retlen < errlen)
474 retlen += snprintf(err+retlen, errlen - retlen,
475 " (unexpected character '%c')", *ptr);
476 return -1;
477 }
478
479 if (curpx->tcp_req.inspect_delay) {
480 snprintf(err, errlen, "ignoring %s %s (was already defined) in %s '%s'",
481 args[0], args[1], proxy_type_str(proxy), curpx->id);
482 return 1;
483 }
484 curpx->tcp_req.inspect_delay = val;
485 return 0;
486 }
487
488 if (!strcmp(args[1], "content")) {
489 int action;
Willy Tarreaudd64f8d2008-07-27 22:02:32 +0200490 int warn = 0;
Willy Tarreaub6866442008-07-14 23:54:42 +0200491 int pol = ACL_COND_NONE;
492 struct acl_cond *cond;
493 struct tcp_rule *rule;
494
495 if (curpx == defpx) {
496 snprintf(err, errlen, "%s %s is not allowed in 'defaults' sections",
497 args[0], args[1]);
498 return -1;
499 }
500
501 if (!strcmp(args[2], "accept"))
502 action = TCP_ACT_ACCEPT;
503 else if (!strcmp(args[2], "reject"))
504 action = TCP_ACT_REJECT;
505 else {
506 retlen = snprintf(err, errlen,
507 "'%s %s' expects 'accept' or 'reject', in %s '%s' (was '%s')",
508 args[0], args[1], proxy_type_str(curpx), curpx->id, args[2]);
509 return -1;
510 }
511
512 pol = ACL_COND_NONE;
513 cond = NULL;
514
515 if (!strcmp(args[3], "if"))
516 pol = ACL_COND_IF;
517 else if (!strcmp(args[3], "unless"))
518 pol = ACL_COND_UNLESS;
519
520 /* Note: we consider "if TRUE" when there is no condition */
521 if (pol != ACL_COND_NONE &&
522 (cond = parse_acl_cond((const char **)args+4, &curpx->acl, pol)) == NULL) {
523 retlen = snprintf(err, errlen,
Willy Tarreaudd64f8d2008-07-27 22:02:32 +0200524 "error detected in %s '%s' while parsing '%s' condition",
Willy Tarreaub6866442008-07-14 23:54:42 +0200525 proxy_type_str(curpx), curpx->id, args[3]);
526 return -1;
527 }
528
Willy Tarreaudd64f8d2008-07-27 22:02:32 +0200529 // FIXME: how to set this ?
530 // cond->line = linenum;
Willy Tarreau8e80e0b2009-05-10 12:05:46 +0200531 if (cond && cond->requires & (ACL_USE_RTR_ANY | ACL_USE_L7_ANY)) {
Willy Tarreaudd64f8d2008-07-27 22:02:32 +0200532 struct acl *acl;
533 const char *name;
534
535 acl = cond_find_require(cond, ACL_USE_RTR_ANY|ACL_USE_L7_ANY);
536 name = acl ? acl->name : "(unknown)";
537
538 retlen = snprintf(err, errlen,
539 "acl '%s' involves some %s criteria which will be ignored.",
540 name,
541 (acl->requires & ACL_USE_RTR_ANY) ? "response-only" : "layer 7");
542 warn++;
543 }
Willy Tarreaub6866442008-07-14 23:54:42 +0200544 rule = (struct tcp_rule *)calloc(1, sizeof(*rule));
545 rule->cond = cond;
546 rule->action = action;
547 LIST_INIT(&rule->list);
548 LIST_ADDQ(&curpx->tcp_req.inspect_rules, &rule->list);
Willy Tarreaudd64f8d2008-07-27 22:02:32 +0200549 return warn;
Willy Tarreaub6866442008-07-14 23:54:42 +0200550 }
551
552 snprintf(err, errlen, "unknown argument '%s' after '%s' in %s '%s'",
553 args[1], args[0], proxy_type_str(proxy), curpx->id);
554 return -1;
555}
556
557/* return the number of bytes in the request buffer */
558static int
559acl_fetch_req_len(struct proxy *px, struct session *l4, void *l7, int dir,
560 struct acl_expr *expr, struct acl_test *test)
561{
562 if (!l4 || !l4->req)
563 return 0;
564
565 test->i = l4->req->l;
566 test->flags = ACL_TEST_F_VOLATILE | ACL_TEST_F_MAY_CHANGE;
567 return 1;
568}
569
Willy Tarreau655e26a2008-07-15 18:58:05 +0200570/* Return the version of the SSL protocol in the request. It supports both
571 * SSLv3 (TLSv1) header format for any message, and SSLv2 header format for
572 * the hello message. The SSLv3 format is described in RFC 2246 p49, and the
573 * SSLv2 format is described here, and completed p67 of RFC 2246 :
574 * http://wp.netscape.com/eng/security/SSL_2.html
575 *
576 * Note: this decoder only works with non-wrapping data.
577 */
578static int
579acl_fetch_req_ssl_ver(struct proxy *px, struct session *l4, void *l7, int dir,
580 struct acl_expr *expr, struct acl_test *test)
581{
582 int version, bleft, msg_len;
583 const unsigned char *data;
584
585 if (!l4 || !l4->req)
586 return 0;
587
588 msg_len = 0;
589 bleft = l4->req->l;
590 if (!bleft)
591 goto too_short;
592
Willy Tarreauc7e961e2008-08-17 17:13:47 +0200593 data = (const unsigned char *)l4->req->w;
Willy Tarreau655e26a2008-07-15 18:58:05 +0200594 if ((*data >= 0x14 && *data <= 0x17) || (*data == 0xFF)) {
595 /* SSLv3 header format */
596 if (bleft < 5)
597 goto too_short;
598
599 version = (data[1] << 16) + data[2]; /* version: major, minor */
600 msg_len = (data[3] << 8) + data[4]; /* record length */
601
602 /* format introduced with SSLv3 */
603 if (version < 0x00030000)
604 goto not_ssl;
605
606 /* message length between 1 and 2^14 + 2048 */
607 if (msg_len < 1 || msg_len > ((1<<14) + 2048))
608 goto not_ssl;
609
610 bleft -= 5; data += 5;
611 } else {
612 /* SSLv2 header format, only supported for hello (msg type 1) */
613 int rlen, plen, cilen, silen, chlen;
614
615 if (*data & 0x80) {
616 if (bleft < 3)
617 goto too_short;
618 /* short header format : 15 bits for length */
619 rlen = ((data[0] & 0x7F) << 8) | data[1];
620 plen = 0;
621 bleft -= 2; data += 2;
622 } else {
623 if (bleft < 4)
624 goto too_short;
625 /* long header format : 14 bits for length + pad length */
626 rlen = ((data[0] & 0x3F) << 8) | data[1];
627 plen = data[2];
628 bleft -= 3; data += 2;
629 }
630
631 if (*data != 0x01)
632 goto not_ssl;
633 bleft--; data++;
634
635 if (bleft < 8)
636 goto too_short;
637 version = (data[0] << 16) + data[1]; /* version: major, minor */
638 cilen = (data[2] << 8) + data[3]; /* cipher len, multiple of 3 */
639 silen = (data[4] << 8) + data[5]; /* session_id_len: 0 or 16 */
640 chlen = (data[6] << 8) + data[7]; /* 16<=challenge length<=32 */
641
642 bleft -= 8; data += 8;
643 if (cilen % 3 != 0)
644 goto not_ssl;
645 if (silen && silen != 16)
646 goto not_ssl;
647 if (chlen < 16 || chlen > 32)
648 goto not_ssl;
649 if (rlen != 9 + cilen + silen + chlen)
650 goto not_ssl;
651
652 /* focus on the remaining data length */
653 msg_len = cilen + silen + chlen + plen;
654 }
655 /* We could recursively check that the buffer ends exactly on an SSL
656 * fragment boundary and that a possible next segment is still SSL,
657 * but that's a bit pointless. However, we could still check that
658 * all the part of the request which fits in a buffer is already
659 * there.
660 */
Willy Tarreau03d60bb2009-01-09 11:13:00 +0100661 if (msg_len > l4->req->max_len + l4->req->data - l4->req->w)
662 msg_len = l4->req->max_len + l4->req->data - l4->req->w;
Willy Tarreau655e26a2008-07-15 18:58:05 +0200663
664 if (bleft < msg_len)
665 goto too_short;
666
667 /* OK that's enough. We have at least the whole message, and we have
668 * the protocol version.
669 */
670 test->i = version;
671 test->flags = ACL_TEST_F_VOLATILE;
672 return 1;
673
674 too_short:
675 test->flags = ACL_TEST_F_MAY_CHANGE;
676 not_ssl:
677 return 0;
678}
679
Willy Tarreaub6866442008-07-14 23:54:42 +0200680
681static struct cfg_kw_list cfg_kws = {{ },{
682 { CFG_LISTEN, "tcp-request", tcp_parse_tcp_req },
683 { 0, NULL, NULL },
684}};
685
686static struct acl_kw_list acl_kws = {{ },{
Willy Tarreau0ceba5a2008-07-25 19:31:03 +0200687 { "req_len", acl_parse_int, acl_fetch_req_len, acl_match_int, ACL_USE_L4REQ_VOLATILE },
688 { "req_ssl_ver", acl_parse_dotted_ver, acl_fetch_req_ssl_ver, acl_match_int, ACL_USE_L4REQ_VOLATILE },
Willy Tarreaub6866442008-07-14 23:54:42 +0200689 { NULL, NULL, NULL, NULL },
690}};
691
Willy Tarreaue6b98942007-10-29 01:09:36 +0100692__attribute__((constructor))
693static void __tcp_protocol_init(void)
694{
695 protocol_register(&proto_tcpv4);
696 protocol_register(&proto_tcpv6);
Willy Tarreaub6866442008-07-14 23:54:42 +0200697 cfg_register_keywords(&cfg_kws);
698 acl_register_keywords(&acl_kws);
Willy Tarreaue6b98942007-10-29 01:09:36 +0100699}
700
701
702/*
703 * Local variables:
704 * c-indent-level: 8
705 * c-basic-offset: 8
706 * End:
707 */