blob: adf4e230c0158f4cb870a8921396b5a1cfe6c5c2 [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
Willy Tarreaufb14edc2009-06-14 15:24:37 +0200215 if (fcntl(fd, F_SETFL, O_NONBLOCK) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +0100216 err |= ERR_FATAL | ERR_ALERT;
217 msg = "cannot make socket non-blocking";
218 goto tcp_close_return;
219 }
220
221 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (char *) &one, sizeof(one)) == -1) {
222 /* not fatal but should be reported */
223 msg = "cannot do so_reuseaddr";
224 err |= ERR_ALERT;
225 }
226
227 if (listener->options & LI_O_NOLINGER)
228 setsockopt(fd, SOL_SOCKET, SO_LINGER, (struct linger *) &nolinger, sizeof(struct linger));
Willy Tarreauedcf6682008-11-30 23:15:34 +0100229
Willy Tarreaue6b98942007-10-29 01:09:36 +0100230#ifdef SO_REUSEPORT
231 /* OpenBSD supports this. As it's present in old libc versions of Linux,
232 * it might return an error that we will silently ignore.
233 */
234 setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, (char *) &one, sizeof(one));
235#endif
Willy Tarreaub1e52e82008-01-13 14:49:51 +0100236#ifdef CONFIG_HAP_LINUX_TPROXY
Willy Tarreauedcf6682008-11-30 23:15:34 +0100237 if ((listener->options & LI_O_FOREIGN)
Willy Tarreau0a459892008-01-13 17:37:16 +0100238 && (setsockopt(fd, SOL_IP, IP_TRANSPARENT, (char *) &one, sizeof(one)) == -1)
239 && (setsockopt(fd, SOL_IP, IP_FREEBIND, (char *) &one, sizeof(one)) == -1)) {
Willy Tarreaub1e52e82008-01-13 14:49:51 +0100240 msg = "cannot make listening socket transparent";
241 err |= ERR_ALERT;
242 }
243#endif
Willy Tarreau5e6e2042009-02-04 17:19:29 +0100244#ifdef SO_BINDTODEVICE
245 /* Note: this might fail if not CAP_NET_RAW */
246 if (listener->interface) {
247 if (setsockopt(fd, SOL_SOCKET, SO_BINDTODEVICE,
Willy Tarreau604e8302009-03-06 00:48:23 +0100248 listener->interface, strlen(listener->interface) + 1) == -1) {
Willy Tarreau5e6e2042009-02-04 17:19:29 +0100249 msg = "cannot bind listener to device";
250 err |= ERR_WARN;
251 }
252 }
253#endif
Willy Tarreaube1b9182009-06-14 18:48:19 +0200254#ifdef TCP_MAXSEG
255 if (listener->maxseg) {
256 if (setsockopt(fd, SOL_TCP, TCP_MAXSEG,
257 &listener->maxseg, sizeof(listener->maxseg)) == -1) {
258 msg = "cannot set MSS";
259 err |= ERR_WARN;
260 }
261 }
262#endif
Willy Tarreaue6b98942007-10-29 01:09:36 +0100263 if (bind(fd, (struct sockaddr *)&listener->addr, listener->proto->sock_addrlen) == -1) {
264 err |= ERR_RETRYABLE | ERR_ALERT;
265 msg = "cannot bind socket";
266 goto tcp_close_return;
267 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100268
Willy Tarreauc73ce2b2008-01-06 10:55:10 +0100269 if (listen(fd, listener->backlog ? listener->backlog : listener->maxconn) == -1) {
Willy Tarreaue6b98942007-10-29 01:09:36 +0100270 err |= ERR_RETRYABLE | ERR_ALERT;
271 msg = "cannot listen to socket";
272 goto tcp_close_return;
273 }
Willy Tarreauedcf6682008-11-30 23:15:34 +0100274
Willy Tarreau9ea05a72009-06-14 12:07:01 +0200275#ifdef TCP_QUICKACK
276 if (listener->options & LI_O_NOQUICKACK)
277 setsockopt(fd, SOL_TCP, TCP_QUICKACK, (char *) &zero, sizeof(zero));
278#endif
279
Willy Tarreaue6b98942007-10-29 01:09:36 +0100280 /* the socket is ready */
281 listener->fd = fd;
282 listener->state = LI_LISTEN;
283
284 /* the function for the accept() event */
285 fd_insert(fd);
286 fdtab[fd].cb[DIR_RD].f = listener->accept;
287 fdtab[fd].cb[DIR_WR].f = NULL; /* never called */
288 fdtab[fd].cb[DIR_RD].b = fdtab[fd].cb[DIR_WR].b = NULL;
Willy Tarreaueabf3132008-08-29 23:36:51 +0200289 fdtab[fd].owner = listener; /* reference the listener instead of a task */
Willy Tarreaue6b98942007-10-29 01:09:36 +0100290 fdtab[fd].state = FD_STLISTEN;
Willy Tarreaufb14edc2009-06-14 15:24:37 +0200291 fdtab[fd].flags = FD_FL_TCP;
Willy Tarreaue6b98942007-10-29 01:09:36 +0100292 fdtab[fd].peeraddr = NULL;
293 fdtab[fd].peerlen = 0;
Willy Tarreaue6b98942007-10-29 01:09:36 +0100294 tcp_return:
295 if (msg && errlen)
296 strlcpy2(errmsg, msg, errlen);
297 return err;
298
299 tcp_close_return:
300 close(fd);
301 goto tcp_return;
302}
303
304/* This function creates all TCP sockets bound to the protocol entry <proto>.
305 * It is intended to be used as the protocol's bind_all() function.
306 * The sockets will be registered but not added to any fd_set, in order not to
307 * loose them across the fork(). A call to enable_all_listeners() is needed
308 * to complete initialization. The return value is composed from ERR_*.
309 */
310static int tcp_bind_listeners(struct protocol *proto)
311{
312 struct listener *listener;
313 int err = ERR_NONE;
314
315 list_for_each_entry(listener, &proto->listeners, proto_list) {
316 err |= tcp_bind_listener(listener, NULL, 0);
317 if ((err & ERR_CODE) == ERR_ABORT)
318 break;
319 }
320
321 return err;
322}
323
324/* Add listener to the list of tcpv4 listeners. The listener's state
325 * is automatically updated from LI_INIT to LI_ASSIGNED. The number of
326 * listeners is updated. This is the function to use to add a new listener.
327 */
328void tcpv4_add_listener(struct listener *listener)
329{
330 if (listener->state != LI_INIT)
331 return;
332 listener->state = LI_ASSIGNED;
333 listener->proto = &proto_tcpv4;
334 LIST_ADDQ(&proto_tcpv4.listeners, &listener->proto_list);
335 proto_tcpv4.nb_listeners++;
336}
337
338/* Add listener to the list of tcpv4 listeners. The listener's state
339 * is automatically updated from LI_INIT to LI_ASSIGNED. The number of
340 * listeners is updated. This is the function to use to add a new listener.
341 */
342void tcpv6_add_listener(struct listener *listener)
343{
344 if (listener->state != LI_INIT)
345 return;
346 listener->state = LI_ASSIGNED;
347 listener->proto = &proto_tcpv6;
348 LIST_ADDQ(&proto_tcpv6.listeners, &listener->proto_list);
349 proto_tcpv6.nb_listeners++;
350}
351
Willy Tarreauedcf6682008-11-30 23:15:34 +0100352/* This function performs the TCP request analysis on the current request. It
353 * returns 1 if the processing can continue on next analysers, or zero if it
354 * needs more data, encounters an error, or wants to immediately abort the
355 * request. It relies on buffers flags, and updates s->req->analysers. Its
356 * behaviour is rather simple:
Willy Tarreau86ef7dc2009-03-15 22:55:47 +0100357 * - the analyser should check for errors and timeouts, and react as expected.
358 * It does not have to close anything upon error, the caller will. Note that
359 * the caller also knows how to report errors and timeouts.
360 * - if the analyser does not have enough data, it must return 0 without calling
Willy Tarreauedcf6682008-11-30 23:15:34 +0100361 * other ones. It should also probably do a buffer_write_dis() to ensure
362 * that unprocessed data will not be forwarded. But that probably depends on
363 * the protocol.
364 * - if an analyser has enough data, it just has to pass on to the next
365 * analyser without using buffer_write_dis() (enabled by default).
366 * - if an analyser thinks it has no added value anymore staying here, it must
367 * reset its bit from the analysers flags in order not to be called anymore.
368 *
369 * In the future, analysers should be able to indicate that they want to be
370 * called after XXX bytes have been received (or transfered), and the min of
371 * all's wishes will be used to ring back (unless a special condition occurs).
372 */
373int tcp_inspect_request(struct session *s, struct buffer *req)
374{
375 struct tcp_rule *rule;
376 int partial;
377
378 DPRINTF(stderr,"[%u] %s: session=%p b=%p, exp(r,w)=%u,%u bf=%08x bl=%d analysers=%02x\n",
379 now_ms, __FUNCTION__,
380 s,
381 req,
382 req->rex, req->wex,
383 req->flags,
384 req->l,
385 req->analysers);
386
Willy Tarreauedcf6682008-11-30 23:15:34 +0100387 /* We don't know whether we have enough data, so must proceed
388 * this way :
389 * - iterate through all rules in their declaration order
390 * - if one rule returns MISS, it means the inspect delay is
391 * not over yet, then return immediately, otherwise consider
392 * it as a non-match.
393 * - if one rule returns OK, then return OK
394 * - if one rule returns KO, then return KO
395 */
396
Willy Tarreaud869b242009-03-15 14:43:58 +0100397 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 +0100398 partial = 0;
399 else
400 partial = ACL_PARTIAL;
401
402 list_for_each_entry(rule, &s->fe->tcp_req.inspect_rules, list) {
403 int ret = ACL_PAT_PASS;
404
405 if (rule->cond) {
406 ret = acl_exec_cond(rule->cond, s->fe, s, NULL, ACL_DIR_REQ | partial);
407 if (ret == ACL_PAT_MISS) {
408 buffer_write_dis(req);
409 /* just set the request timeout once at the beginning of the request */
Willy Tarreaud869b242009-03-15 14:43:58 +0100410 if (!tick_isset(req->analyse_exp) && s->fe->tcp_req.inspect_delay)
Willy Tarreauedcf6682008-11-30 23:15:34 +0100411 req->analyse_exp = tick_add_ifset(now_ms, s->fe->tcp_req.inspect_delay);
412 return 0;
413 }
414
415 ret = acl_pass(ret);
416 if (rule->cond->pol == ACL_COND_UNLESS)
417 ret = !ret;
418 }
419
420 if (ret) {
421 /* we have a matching rule. */
422 if (rule->action == TCP_ACT_REJECT) {
423 buffer_abort(req);
424 buffer_abort(s->rep);
425 req->analysers = 0;
426 s->fe->failed_req++;
427 if (!(s->flags & SN_ERR_MASK))
428 s->flags |= SN_ERR_PRXCOND;
429 if (!(s->flags & SN_FINST_MASK))
430 s->flags |= SN_FINST_R;
431 return 0;
432 }
433 /* otherwise accept */
434 break;
435 }
436 }
437
438 /* if we get there, it means we have no rule which matches, or
439 * we have an explicit accept, so we apply the default accept.
440 */
441 req->analysers &= ~AN_REQ_INSPECT;
442 req->analyse_exp = TICK_ETERNITY;
443 return 1;
444}
445
446
Willy Tarreaub6866442008-07-14 23:54:42 +0200447/* This function should be called to parse a line starting with the "tcp-request"
448 * keyword.
449 */
450static int tcp_parse_tcp_req(char **args, int section_type, struct proxy *curpx,
451 struct proxy *defpx, char *err, int errlen)
452{
453 const char *ptr = NULL;
Willy Tarreauc7e961e2008-08-17 17:13:47 +0200454 unsigned int val;
Willy Tarreaub6866442008-07-14 23:54:42 +0200455 int retlen;
456
457 if (!*args[1]) {
458 snprintf(err, errlen, "missing argument for '%s' in %s '%s'",
459 args[0], proxy_type_str(proxy), curpx->id);
460 return -1;
461 }
462
463 if (!strcmp(args[1], "inspect-delay")) {
464 if (curpx == defpx) {
465 snprintf(err, errlen, "%s %s is not allowed in 'defaults' sections",
466 args[0], args[1]);
467 return -1;
468 }
469
470 if (!(curpx->cap & PR_CAP_FE)) {
471 snprintf(err, errlen, "%s %s will be ignored because %s '%s' has no %s capability",
472 args[0], args[1], proxy_type_str(proxy), curpx->id,
473 "frontend");
474 return 1;
475 }
476
477 if (!*args[2] || (ptr = parse_time_err(args[2], &val, TIME_UNIT_MS))) {
478 retlen = snprintf(err, errlen,
479 "'%s %s' expects a positive delay in milliseconds, in %s '%s'",
480 args[0], args[1], proxy_type_str(proxy), curpx->id);
481 if (ptr && retlen < errlen)
482 retlen += snprintf(err+retlen, errlen - retlen,
483 " (unexpected character '%c')", *ptr);
484 return -1;
485 }
486
487 if (curpx->tcp_req.inspect_delay) {
488 snprintf(err, errlen, "ignoring %s %s (was already defined) in %s '%s'",
489 args[0], args[1], proxy_type_str(proxy), curpx->id);
490 return 1;
491 }
492 curpx->tcp_req.inspect_delay = val;
493 return 0;
494 }
495
496 if (!strcmp(args[1], "content")) {
497 int action;
Willy Tarreaudd64f8d2008-07-27 22:02:32 +0200498 int warn = 0;
Willy Tarreaub6866442008-07-14 23:54:42 +0200499 int pol = ACL_COND_NONE;
500 struct acl_cond *cond;
501 struct tcp_rule *rule;
502
503 if (curpx == defpx) {
504 snprintf(err, errlen, "%s %s is not allowed in 'defaults' sections",
505 args[0], args[1]);
506 return -1;
507 }
508
509 if (!strcmp(args[2], "accept"))
510 action = TCP_ACT_ACCEPT;
511 else if (!strcmp(args[2], "reject"))
512 action = TCP_ACT_REJECT;
513 else {
514 retlen = snprintf(err, errlen,
515 "'%s %s' expects 'accept' or 'reject', in %s '%s' (was '%s')",
516 args[0], args[1], proxy_type_str(curpx), curpx->id, args[2]);
517 return -1;
518 }
519
520 pol = ACL_COND_NONE;
521 cond = NULL;
522
523 if (!strcmp(args[3], "if"))
524 pol = ACL_COND_IF;
525 else if (!strcmp(args[3], "unless"))
526 pol = ACL_COND_UNLESS;
527
528 /* Note: we consider "if TRUE" when there is no condition */
529 if (pol != ACL_COND_NONE &&
530 (cond = parse_acl_cond((const char **)args+4, &curpx->acl, pol)) == NULL) {
531 retlen = snprintf(err, errlen,
Willy Tarreaudd64f8d2008-07-27 22:02:32 +0200532 "error detected in %s '%s' while parsing '%s' condition",
Willy Tarreaub6866442008-07-14 23:54:42 +0200533 proxy_type_str(curpx), curpx->id, args[3]);
534 return -1;
535 }
536
Willy Tarreaudd64f8d2008-07-27 22:02:32 +0200537 // FIXME: how to set this ?
538 // cond->line = linenum;
Willy Tarreau8e80e0b2009-05-10 12:05:46 +0200539 if (cond && cond->requires & (ACL_USE_RTR_ANY | ACL_USE_L7_ANY)) {
Willy Tarreaudd64f8d2008-07-27 22:02:32 +0200540 struct acl *acl;
541 const char *name;
542
543 acl = cond_find_require(cond, ACL_USE_RTR_ANY|ACL_USE_L7_ANY);
544 name = acl ? acl->name : "(unknown)";
545
546 retlen = snprintf(err, errlen,
547 "acl '%s' involves some %s criteria which will be ignored.",
548 name,
549 (acl->requires & ACL_USE_RTR_ANY) ? "response-only" : "layer 7");
550 warn++;
551 }
Willy Tarreaub6866442008-07-14 23:54:42 +0200552 rule = (struct tcp_rule *)calloc(1, sizeof(*rule));
553 rule->cond = cond;
554 rule->action = action;
555 LIST_INIT(&rule->list);
556 LIST_ADDQ(&curpx->tcp_req.inspect_rules, &rule->list);
Willy Tarreaudd64f8d2008-07-27 22:02:32 +0200557 return warn;
Willy Tarreaub6866442008-07-14 23:54:42 +0200558 }
559
560 snprintf(err, errlen, "unknown argument '%s' after '%s' in %s '%s'",
561 args[1], args[0], proxy_type_str(proxy), curpx->id);
562 return -1;
563}
564
565/* return the number of bytes in the request buffer */
566static int
567acl_fetch_req_len(struct proxy *px, struct session *l4, void *l7, int dir,
568 struct acl_expr *expr, struct acl_test *test)
569{
570 if (!l4 || !l4->req)
571 return 0;
572
573 test->i = l4->req->l;
574 test->flags = ACL_TEST_F_VOLATILE | ACL_TEST_F_MAY_CHANGE;
575 return 1;
576}
577
Willy Tarreau655e26a2008-07-15 18:58:05 +0200578/* Return the version of the SSL protocol in the request. It supports both
579 * SSLv3 (TLSv1) header format for any message, and SSLv2 header format for
580 * the hello message. The SSLv3 format is described in RFC 2246 p49, and the
581 * SSLv2 format is described here, and completed p67 of RFC 2246 :
582 * http://wp.netscape.com/eng/security/SSL_2.html
583 *
584 * Note: this decoder only works with non-wrapping data.
585 */
586static int
587acl_fetch_req_ssl_ver(struct proxy *px, struct session *l4, void *l7, int dir,
588 struct acl_expr *expr, struct acl_test *test)
589{
590 int version, bleft, msg_len;
591 const unsigned char *data;
592
593 if (!l4 || !l4->req)
594 return 0;
595
596 msg_len = 0;
597 bleft = l4->req->l;
598 if (!bleft)
599 goto too_short;
600
Willy Tarreauc7e961e2008-08-17 17:13:47 +0200601 data = (const unsigned char *)l4->req->w;
Willy Tarreau655e26a2008-07-15 18:58:05 +0200602 if ((*data >= 0x14 && *data <= 0x17) || (*data == 0xFF)) {
603 /* SSLv3 header format */
604 if (bleft < 5)
605 goto too_short;
606
607 version = (data[1] << 16) + data[2]; /* version: major, minor */
608 msg_len = (data[3] << 8) + data[4]; /* record length */
609
610 /* format introduced with SSLv3 */
611 if (version < 0x00030000)
612 goto not_ssl;
613
614 /* message length between 1 and 2^14 + 2048 */
615 if (msg_len < 1 || msg_len > ((1<<14) + 2048))
616 goto not_ssl;
617
618 bleft -= 5; data += 5;
619 } else {
620 /* SSLv2 header format, only supported for hello (msg type 1) */
621 int rlen, plen, cilen, silen, chlen;
622
623 if (*data & 0x80) {
624 if (bleft < 3)
625 goto too_short;
626 /* short header format : 15 bits for length */
627 rlen = ((data[0] & 0x7F) << 8) | data[1];
628 plen = 0;
629 bleft -= 2; data += 2;
630 } else {
631 if (bleft < 4)
632 goto too_short;
633 /* long header format : 14 bits for length + pad length */
634 rlen = ((data[0] & 0x3F) << 8) | data[1];
635 plen = data[2];
636 bleft -= 3; data += 2;
637 }
638
639 if (*data != 0x01)
640 goto not_ssl;
641 bleft--; data++;
642
643 if (bleft < 8)
644 goto too_short;
645 version = (data[0] << 16) + data[1]; /* version: major, minor */
646 cilen = (data[2] << 8) + data[3]; /* cipher len, multiple of 3 */
647 silen = (data[4] << 8) + data[5]; /* session_id_len: 0 or 16 */
648 chlen = (data[6] << 8) + data[7]; /* 16<=challenge length<=32 */
649
650 bleft -= 8; data += 8;
651 if (cilen % 3 != 0)
652 goto not_ssl;
653 if (silen && silen != 16)
654 goto not_ssl;
655 if (chlen < 16 || chlen > 32)
656 goto not_ssl;
657 if (rlen != 9 + cilen + silen + chlen)
658 goto not_ssl;
659
660 /* focus on the remaining data length */
661 msg_len = cilen + silen + chlen + plen;
662 }
663 /* We could recursively check that the buffer ends exactly on an SSL
664 * fragment boundary and that a possible next segment is still SSL,
665 * but that's a bit pointless. However, we could still check that
666 * all the part of the request which fits in a buffer is already
667 * there.
668 */
Willy Tarreau03d60bb2009-01-09 11:13:00 +0100669 if (msg_len > l4->req->max_len + l4->req->data - l4->req->w)
670 msg_len = l4->req->max_len + l4->req->data - l4->req->w;
Willy Tarreau655e26a2008-07-15 18:58:05 +0200671
672 if (bleft < msg_len)
673 goto too_short;
674
675 /* OK that's enough. We have at least the whole message, and we have
676 * the protocol version.
677 */
678 test->i = version;
679 test->flags = ACL_TEST_F_VOLATILE;
680 return 1;
681
682 too_short:
683 test->flags = ACL_TEST_F_MAY_CHANGE;
684 not_ssl:
685 return 0;
686}
687
Willy Tarreaub6866442008-07-14 23:54:42 +0200688
689static struct cfg_kw_list cfg_kws = {{ },{
690 { CFG_LISTEN, "tcp-request", tcp_parse_tcp_req },
691 { 0, NULL, NULL },
692}};
693
694static struct acl_kw_list acl_kws = {{ },{
Willy Tarreau0ceba5a2008-07-25 19:31:03 +0200695 { "req_len", acl_parse_int, acl_fetch_req_len, acl_match_int, ACL_USE_L4REQ_VOLATILE },
696 { "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 +0200697 { NULL, NULL, NULL, NULL },
698}};
699
Willy Tarreaue6b98942007-10-29 01:09:36 +0100700__attribute__((constructor))
701static void __tcp_protocol_init(void)
702{
703 protocol_register(&proto_tcpv4);
704 protocol_register(&proto_tcpv6);
Willy Tarreaub6866442008-07-14 23:54:42 +0200705 cfg_register_keywords(&cfg_kws);
706 acl_register_keywords(&acl_kws);
Willy Tarreaue6b98942007-10-29 01:09:36 +0100707}
708
709
710/*
711 * Local variables:
712 * c-indent-level: 8
713 * c-basic-offset: 8
714 * End:
715 */