fb42f005107e442e92545b6bc806c1ec26024c3e
[exim.git] / src / src / ip.c
1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
4
5 /* Copyright (c) University of Cambridge 1995 - 2018 */
6 /* See the file NOTICE for conditions of use and distribution. */
7
8 /* Functions for doing things with sockets. With the advent of IPv6 this has
9 got messier, so that it's worth pulling out the code into separate functions
10 that other parts of Exim can call, especially as there are now several
11 different places in the code where sockets are used. */
12
13
14 #include "exim.h"
15
16
17 /*************************************************
18 * Create a socket *
19 *************************************************/
20
21 /* Socket creation happens in a number of places so it's packaged here for
22 convenience.
23
24 Arguments:
25 type SOCK_DGRAM or SOCK_STREAM
26 af AF_INET or AF_INET6
27
28 Returns: socket number or -1 on failure
29 */
30
31 int
32 ip_socket(int type, int af)
33 {
34 int sock = socket(af, type, 0);
35 if (sock < 0)
36 log_write(0, LOG_MAIN, "IPv%c socket creation failed: %s",
37 (af == AF_INET6)? '6':'4', strerror(errno));
38 return sock;
39 }
40
41
42
43
44 #if HAVE_IPV6
45 /*************************************************
46 * Convert printing address to numeric *
47 *************************************************/
48
49 /* This function converts the textual form of an IP address into a numeric form
50 in an appropriate structure in an IPv6 environment. The getaddrinfo() function
51 can (apparently) handle more complicated addresses (e.g. those containing
52 scopes) than inet_pton() in some environments. We use hints to tell it that the
53 input must be a numeric address.
54
55 However, apparently some operating systems (or libraries) don't support
56 getaddrinfo(), so there is a build-time option to revert to inet_pton() (which
57 does not support scopes).
58
59 Arguments:
60 address textual form of the address
61 addr where to copy back the answer
62
63 Returns: nothing - failure provokes a panic-die
64 */
65
66 static void
67 ip_addrinfo(const uschar *address, struct sockaddr_in6 *saddr)
68 {
69 #ifdef IPV6_USE_INET_PTON
70
71 if (inet_pton(AF_INET6, CCS address, &saddr->sin6_addr) != 1)
72 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "unable to parse \"%s\" as an "
73 "IP address", address);
74 saddr->sin6_family = AF_INET6;
75
76 #else
77
78 int rc;
79 struct addrinfo hints, *res;
80 memset(&hints, 0, sizeof(hints));
81 hints.ai_family = AF_INET6;
82 hints.ai_socktype = SOCK_STREAM;
83 hints.ai_flags = AI_NUMERICHOST;
84 if ((rc = getaddrinfo(CCS address, NULL, &hints, &res)) != 0 || res == NULL)
85 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "unable to parse \"%s\" as an "
86 "IP address: %s", address,
87 (rc == 0)? "NULL result returned" : gai_strerror(rc));
88 memcpy(saddr, res->ai_addr, res->ai_addrlen);
89 freeaddrinfo(res);
90
91 #endif
92 }
93 #endif /* HAVE_IPV6 */
94
95
96 /*************************************************
97 * Bind socket to interface and port *
98 *************************************************/
99
100 int
101 ip_addr(void * sin_, int af, const uschar * address, int port)
102 {
103 union sockaddr_46 * sin = sin_;
104 memset(sin, 0, sizeof(*sin));
105
106 /* Setup code when using an IPv6 socket. The wildcard address is ":", to
107 ensure an IPv6 socket is used. */
108
109 #if HAVE_IPV6
110 if (af == AF_INET6)
111 {
112 if (address[0] == ':' && address[1] == 0)
113 {
114 sin->v6.sin6_family = AF_INET6;
115 sin->v6.sin6_addr = in6addr_any;
116 }
117 else
118 ip_addrinfo(address, &sin->v6); /* Panic-dies on error */
119 sin->v6.sin6_port = htons(port);
120 return sizeof(sin->v6);
121 }
122 else
123 #else /* HAVE_IPv6 */
124 af = af; /* Avoid compiler warning */
125 #endif /* HAVE_IPV6 */
126
127 /* Setup code when using IPv4 socket. The wildcard address is "". */
128
129 {
130 sin->v4.sin_family = AF_INET;
131 sin->v4.sin_port = htons(port);
132 sin->v4.sin_addr.s_addr = address[0] == 0
133 ? (S_ADDR_TYPE)INADDR_ANY
134 : (S_ADDR_TYPE)inet_addr(CS address);
135 return sizeof(sin->v4);
136 }
137 }
138
139
140
141 /* This function binds a socket to a local interface address and port. For a
142 wildcard IPv6 bind, the address is ":".
143
144 Arguments:
145 sock the socket
146 af AF_INET or AF_INET6 - the socket type
147 address the IP address, in text form
148 port the IP port (host order)
149
150 Returns: the result of bind()
151 */
152
153 int
154 ip_bind(int sock, int af, uschar *address, int port)
155 {
156 union sockaddr_46 sin;
157 int s_len = ip_addr(&sin, af, address, port);
158 return bind(sock, (struct sockaddr *)&sin, s_len);
159 }
160
161
162
163 /*************************************************
164 *************************************************/
165
166 #ifdef EXIM_TFO_PROBE
167 void
168 tfo_probe(void)
169 {
170 # ifdef TCP_FASTOPEN
171 int sock, backlog = 5;
172
173 if ( (sock = socket(SOCK_STREAM, AF_INET, 0)) < 0
174 && setsockopt(sock, IPPROTO_TCP, TCP_FASTOPEN, &backlog, sizeof(backlog))
175 )
176 f.tcp_fastopen_ok = TRUE;
177 close(sock);
178 # endif
179 }
180 #endif
181
182
183 /*************************************************
184 * Connect socket to remote host *
185 *************************************************/
186
187 /* This function connects a socket to a remote address and port. The socket may
188 or may not have previously been bound to a local interface. The socket is not
189 closed, even in cases of error. It is expected that the calling function, which
190 created the socket, will be the one that closes it.
191
192 Arguments:
193 sock the socket
194 af AF_INET6 or AF_INET for the socket type
195 address the remote address, in text form
196 port the remote port
197 timeout a timeout (zero for indefinite timeout)
198 fastopen_blob non-null iff TCP_FASTOPEN can be used; may indicate early-data to
199 be sent in SYN segment. Any such data must be idempotent.
200
201 Returns: 0 on success; -1 on failure, with errno set
202 */
203
204 int
205 ip_connect(int sock, int af, const uschar *address, int port, int timeout,
206 const blob * fastopen_blob)
207 {
208 struct sockaddr_in s_in4;
209 struct sockaddr *s_ptr;
210 int s_len, rc, save_errno;
211
212 /* For an IPv6 address, use an IPv6 sockaddr structure. */
213
214 #if HAVE_IPV6
215 struct sockaddr_in6 s_in6;
216 if (af == AF_INET6)
217 {
218 memset(&s_in6, 0, sizeof(s_in6));
219 ip_addrinfo(address, &s_in6); /* Panic-dies on error */
220 s_in6.sin6_port = htons(port);
221 s_ptr = (struct sockaddr *)&s_in6;
222 s_len = sizeof(s_in6);
223 }
224 else
225 #else /* HAVE_IPV6 */
226 af = af; /* Avoid compiler warning */
227 #endif /* HAVE_IPV6 */
228
229 /* For an IPv4 address, use an IPv4 sockaddr structure, even on a system with
230 IPv6 support. */
231
232 {
233 memset(&s_in4, 0, sizeof(s_in4));
234 s_in4.sin_family = AF_INET;
235 s_in4.sin_port = htons(port);
236 s_in4.sin_addr.s_addr = (S_ADDR_TYPE)inet_addr(CCS address);
237 s_ptr = (struct sockaddr *)&s_in4;
238 s_len = sizeof(s_in4);
239 }
240
241 /* If no connection timeout is set, just call connect() without setting a
242 timer, thereby allowing the inbuilt OS timeout to operate. */
243
244 callout_address = string_sprintf("[%s]:%d", address, port);
245 sigalrm_seen = FALSE;
246 if (timeout > 0) ALARM(timeout);
247
248 #ifdef TCP_FASTOPEN
249 /* TCP Fast Open, if the system has a cookie from a previous call to
250 this peer, can send data in the SYN packet. The peer can send data
251 before it gets our ACK of its SYN,ACK - the latter is useful for
252 the SMTP banner. Other (than SMTP) cases of TCP connections can
253 possibly use the data-on-syn, so support that too. */
254
255 if (fastopen_blob && f.tcp_fastopen_ok)
256 {
257 # ifdef MSG_FASTOPEN
258 /* This is a Linux implementation. It might be useable on FreeBSD; I have
259 not checked. */
260
261 if ((rc = sendto(sock, fastopen_blob->data, fastopen_blob->len,
262 MSG_FASTOPEN | MSG_DONTWAIT, s_ptr, s_len)) >= 0)
263 /* seen for with-data, experimental TFO option, with-cookie case */
264 /* seen for with-data, proper TFO opt, with-cookie case */
265 {
266 DEBUG(D_transport|D_v)
267 debug_printf("TFO mode connection attempt to %s, %lu data\n",
268 address, (unsigned long)fastopen_blob->len);
269 /*XXX also seen on successful TFO, sigh */
270 tcp_out_fastopen = fastopen_blob->len > 0 ? TFO_ATTEMPTED_DATA : TFO_ATTEMPTED_NODATA;
271 }
272 else if (errno == EINPROGRESS) /* expected if we had no cookie for peer */
273 /* seen for no-data, proper TFO option, both cookie-request and with-cookie cases */
274 /* apparently no visibility of the diffference at this point */
275 /* seen for with-data, proper TFO opt, cookie-req */
276 /* with netwk delay, post-conn tcp_info sees unacked 1 for R, 2 for C; code in smtp_out.c */
277 /* ? older Experimental TFO option behaviour ? */
278 { /* queue unsent data */
279 DEBUG(D_transport|D_v) debug_printf("TFO mode sendto, %s data: EINPROGRESS\n",
280 fastopen_blob->len > 0 ? "with" : "no");
281 if (!fastopen_blob->data)
282 {
283 tcp_out_fastopen = TFO_ATTEMPTED_NODATA; /* we tried; unknown if useful yet */
284 rc = 0;
285 }
286 else
287 rc = send(sock, fastopen_blob->data, fastopen_blob->len, 0);
288 }
289 else if(errno == EOPNOTSUPP)
290 {
291 DEBUG(D_transport)
292 debug_printf("Tried TCP Fast Open but apparently not enabled by sysctl\n");
293 goto legacy_connect;
294 }
295 # endif
296 # ifdef EXIM_TFO_CONNECTX
297 /* MacOS */
298 sa_endpoints_t ends = {
299 .sae_srcif = 0, .sae_srcaddr = NULL, .sae_srcaddrlen = 0,
300 .sae_dstaddr = s_ptr, .sae_dstaddrlen = s_len };
301 struct iovec iov = {
302 .iov_base = fastopen_blob->data, .iov_len = fastopen_blob->len };
303 size_t len;
304
305 if ((rc = connectx(sock, &ends, SAE_ASSOCID_ANY,
306 CONNECT_DATA_IDEMPOTENT, &iov, 1, &len, NULL)) == 0)
307 {
308 DEBUG(D_transport|D_v)
309 debug_printf("TFO mode connection attempt to %s, %lu data\n",
310 address, (unsigned long)fastopen_blob->len);
311 tcp_out_fastopen = fastopen_blob->len > 0 ? TFO_ATTEMPTED_DATA : TFO_ATTEMPTED_NODATA;
312
313 if (len != fastopen_blob->len)
314 DEBUG(D_transport|D_v)
315 debug_printf(" only queued %lu data!\n", (unsigned long)len);
316 }
317 else if (errno == EINPROGRESS)
318 {
319 DEBUG(D_transport|D_v) debug_printf("TFO mode sendto, %s data: EINPROGRESS\n",
320 fastopen_blob->len > 0 ? "with" : "no");
321 if (!fastopen_blob->data)
322 {
323 tcp_out_fastopen = TFO_ATTEMPTED_NODATA; /* we tried; unknown if useful yet */
324 rc = 0;
325 }
326 else /* assume that no data was queued; block in send */
327 rc = send(sock, fastopen_blob->data, fastopen_blob->len, 0);
328 }
329 # endif
330 }
331 else
332 #endif /*TCP_FASTOPEN*/
333 {
334 #if defined(TCP_FASTOPEN) && defined(MSG_FASTOPEN)
335 legacy_connect:
336 #endif
337
338 DEBUG(D_transport|D_v) if (fastopen_blob)
339 debug_printf("non-TFO mode connection attempt to %s, %lu data\n",
340 address, (unsigned long)fastopen_blob->len);
341 if ((rc = connect(sock, s_ptr, s_len)) >= 0)
342 if ( fastopen_blob && fastopen_blob->data && fastopen_blob->len
343 && send(sock, fastopen_blob->data, fastopen_blob->len, 0) < 0)
344 rc = -1;
345 }
346
347 save_errno = errno;
348 ALARM_CLR(0);
349
350 /* There is a testing facility for simulating a connection timeout, as I
351 can't think of any other way of doing this. It converts a connection refused
352 into a timeout if the timeout is set to 999999. */
353
354 if (f.running_in_test_harness && save_errno == ECONNREFUSED && timeout == 999999)
355 {
356 rc = -1;
357 save_errno = EINTR;
358 sigalrm_seen = TRUE;
359 }
360
361 /* Success */
362
363 if (rc >= 0)
364 return 0;
365
366 /* A failure whose error code is "Interrupted system call" is in fact
367 an externally applied timeout if the signal handler has been run. */
368
369 errno = save_errno == EINTR && sigalrm_seen ? ETIMEDOUT : save_errno;
370 return -1;
371 }
372
373
374
375 /*************************************************
376 * Create connected socket to remote host *
377 *************************************************/
378
379 /* Create a socket and connect to host (name or number, ipv6 ok)
380 at one of port-range.
381
382 Arguments:
383 type SOCK_DGRAM or SOCK_STREAM
384 af AF_INET6 or AF_INET for the socket type
385 hostname host name, or ip address (as text)
386 portlo,porthi the remote port range
387 timeout a timeout
388 connhost if not NULL, host_item to be filled in with connection details
389 errstr pointer for allocated string on error
390 fastopen_blob with SOCK_STREAM, if non-null, request TCP Fast Open.
391 Additionally, optional idempotent early-data to send
392
393 Return:
394 socket fd, or -1 on failure (having allocated an error string)
395 */
396 int
397 ip_connectedsocket(int type, const uschar * hostname, int portlo, int porthi,
398 int timeout, host_item * connhost, uschar ** errstr, const blob * fastopen_blob)
399 {
400 int namelen;
401 host_item shost;
402 int af = 0, fd, fd4 = -1, fd6 = -1;
403
404 shost.next = NULL;
405 shost.address = NULL;
406 shost.port = portlo;
407 shost.mx = -1;
408
409 namelen = Ustrlen(hostname);
410
411 /* Anything enclosed in [] must be an IP address. */
412
413 if (hostname[0] == '[' &&
414 hostname[namelen - 1] == ']')
415 {
416 uschar * host = string_copyn(hostname+1, namelen-2);
417 if (string_is_ip_address(host, NULL) == 0)
418 {
419 *errstr = string_sprintf("malformed IP address \"%s\"", hostname);
420 return -1;
421 }
422 shost.name = shost.address = host;
423 }
424
425 /* Otherwise check for an unadorned IP address */
426
427 else if (string_is_ip_address(hostname, NULL) != 0)
428 shost.name = shost.address = string_copyn(hostname, namelen);
429
430 /* Otherwise lookup IP address(es) from the name */
431
432 else
433 {
434 shost.name = string_copyn(hostname, namelen);
435 if (host_find_byname(&shost, NULL, HOST_FIND_QUALIFY_SINGLE,
436 NULL, FALSE) != HOST_FOUND)
437 {
438 *errstr = string_sprintf("no IP address found for host %s", shost.name);
439 return -1;
440 }
441 }
442
443 /* Try to connect to the server - test each IP till one works */
444
445 for (host_item * h = &shost; h; h = h->next)
446 {
447 fd = Ustrchr(h->address, ':') != 0
448 ? fd6 < 0 ? (fd6 = ip_socket(type, af = AF_INET6)) : fd6
449 : fd4 < 0 ? (fd4 = ip_socket(type, af = AF_INET )) : fd4;
450
451 if (fd < 0)
452 {
453 *errstr = string_sprintf("failed to create socket: %s", strerror(errno));
454 goto bad;
455 }
456
457 for (int port = portlo; port <= porthi; port++)
458 if (ip_connect(fd, af, h->address, port, timeout, fastopen_blob) == 0)
459 {
460 if (fd != fd6) close(fd6);
461 if (fd != fd4) close(fd4);
462 if (connhost)
463 {
464 h->port = port;
465 *connhost = *h;
466 connhost->next = NULL;
467 }
468 return fd;
469 }
470 }
471
472 *errstr = string_sprintf("failed to connect to any address for %s: %s",
473 hostname, strerror(errno));
474
475 bad:
476 close(fd4); close(fd6); return -1;
477 }
478
479
480 /*XXX TFO? */
481 int
482 ip_tcpsocket(const uschar * hostport, uschar ** errstr, int tmo)
483 {
484 int scan;
485 uschar hostname[256];
486 unsigned int portlow, porthigh;
487
488 /* extract host and port part */
489 scan = sscanf(CS hostport, "%255s %u-%u", hostname, &portlow, &porthigh);
490 if (scan != 3)
491 {
492 if (scan != 2)
493 {
494 *errstr = string_sprintf("invalid socket '%s'", hostport);
495 return -1;
496 }
497 porthigh = portlow;
498 }
499
500 return ip_connectedsocket(SOCK_STREAM, hostname, portlow, porthigh,
501 tmo, NULL, errstr, NULL);
502 }
503
504 int
505 ip_unixsocket(const uschar * path, uschar ** errstr)
506 {
507 int sock;
508 struct sockaddr_un server;
509
510 if ((sock = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
511 {
512 *errstr = US"can't open UNIX socket.";
513 return -1;
514 }
515
516 callout_address = string_copy(path);
517 server.sun_family = AF_UNIX;
518 Ustrncpy(server.sun_path, path, sizeof(server.sun_path)-1);
519 server.sun_path[sizeof(server.sun_path)-1] = '\0';
520 if (connect(sock, (struct sockaddr *) &server, sizeof(server)) < 0)
521 {
522 int err = errno;
523 (void)close(sock);
524 *errstr = string_sprintf("unable to connect to UNIX socket (%s): %s",
525 path, strerror(err));
526 return -1;
527 }
528 return sock;
529 }
530
531 int
532 ip_streamsocket(const uschar * spec, uschar ** errstr, int tmo)
533 {
534 return *spec == '/'
535 ? ip_unixsocket(spec, errstr) : ip_tcpsocket(spec, errstr, tmo);
536 }
537
538 /*************************************************
539 * Set keepalive on a socket *
540 *************************************************/
541
542 /* Can be called for both incoming and outgoing sockets.
543
544 Arguments:
545 sock the socket
546 address the remote host address, for failure logging
547 torf true for outgoing connection, false for incoming
548
549 Returns: nothing
550 */
551
552 void
553 ip_keepalive(int sock, const uschar *address, BOOL torf)
554 {
555 int fodder = 1;
556 if (setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
557 US (&fodder), sizeof(fodder)) != 0)
558 log_write(0, LOG_MAIN, "setsockopt(SO_KEEPALIVE) on connection %s %s "
559 "failed: %s", torf? "to":"from", address, strerror(errno));
560 }
561
562
563
564 /*************************************************
565 * Receive from a socket with timeout *
566 *************************************************/
567
568 /*
569 Arguments:
570 fd the file descriptor
571 timeout the timeout, seconds
572 Returns: TRUE => ready for i/o
573 FALSE => timed out, or other error
574 */
575 BOOL
576 fd_ready(int fd, int timeout)
577 {
578 fd_set select_inset;
579 time_t start_recv = time(NULL);
580 int time_left = timeout;
581 int rc;
582
583 if (time_left <= 0)
584 {
585 errno = ETIMEDOUT;
586 return FALSE;
587 }
588 /* Wait until the socket is ready */
589
590 do
591 {
592 struct timeval tv = { .tv_sec = time_left, .tv_usec = 0 };
593 FD_ZERO (&select_inset);
594 FD_SET (fd, &select_inset);
595
596 /*DEBUG(D_transport) debug_printf("waiting for data on fd\n");*/
597 rc = select(fd + 1, (SELECT_ARG2_TYPE *)&select_inset, NULL, NULL, &tv);
598
599 /* If some interrupt arrived, just retry. We presume this to be rare,
600 but it can happen (e.g. the SIGUSR1 signal sent by exiwhat causes
601 select() to exit).
602
603 Aug 2004: Somebody set up a cron job that ran exiwhat every 2 minutes, making
604 the interrupt not at all rare. Since the timeout is typically more than 2
605 minutes, the effect was to block the timeout completely. To prevent this
606 happening again, we do an explicit time test and adjust the timeout
607 accordingly */
608
609 if (rc < 0 && errno == EINTR)
610 {
611 DEBUG(D_transport) debug_printf("EINTR while waiting for socket data\n");
612
613 /* Watch out, 'continue' jumps to the condition, not to the loops top */
614 time_left = timeout - (time(NULL) - start_recv);
615 if (time_left > 0) continue;
616 }
617
618 if (rc <= 0)
619 {
620 errno = ETIMEDOUT;
621 return FALSE;
622 }
623
624 /* Checking the FD_ISSET is not enough, if we're interrupted, the
625 select_inset may still contain the 'input'. */
626 }
627 while (rc < 0 || !FD_ISSET(fd, &select_inset));
628 return TRUE;
629 }
630
631 /* The timeout is implemented using select(), and we loop to cover select()
632 getting interrupted, and the possibility of select() returning with a positive
633 result but no ready descriptor. Is this in fact possible?
634
635 Arguments:
636 cctx the connection context (socket fd, possibly TLS context)
637 buffer to read into
638 bufsize the buffer size
639 timeout the timeout
640
641 Returns: > 0 => that much data read
642 <= 0 on error or EOF; errno set - zero for EOF
643 */
644
645 int
646 ip_recv(client_conn_ctx * cctx, uschar * buffer, int buffsize, int timeout)
647 {
648 int rc;
649
650 if (!fd_ready(cctx->sock, timeout))
651 return -1;
652
653 /* The socket is ready, read from it (via TLS if it's active). On EOF (i.e.
654 close down of the connection), set errno to zero; otherwise leave it alone. */
655
656 #ifndef DISABLE_TLS
657 if (cctx->tls_ctx) /* client TLS */
658 rc = tls_read(cctx->tls_ctx, buffer, buffsize);
659 else if (tls_in.active.sock == cctx->sock) /* server TLS */
660 rc = tls_read(NULL, buffer, buffsize);
661 else
662 #endif
663 rc = recv(cctx->sock, buffer, buffsize, 0);
664
665 if (rc > 0) return rc;
666 if (rc == 0) errno = 0;
667 return -1;
668 }
669
670
671
672
673 /*************************************************
674 * Lookup address family of potential socket *
675 *************************************************/
676
677 /* Given a file-descriptor, check to see if it's a socket and, if so,
678 return the address family; detects IPv4 vs IPv6. If not a socket then
679 return -1.
680
681 The value 0 is typically AF_UNSPEC, which should not be seen on a connected
682 fd. If the return is -1, the errno will be from getsockname(); probably
683 ENOTSOCK or ECONNRESET.
684
685 Arguments: socket-or-not fd
686 Returns: address family or -1
687 */
688
689 int
690 ip_get_address_family(int fd)
691 {
692 struct sockaddr_storage ss;
693 socklen_t sslen = sizeof(ss);
694
695 if (getsockname(fd, (struct sockaddr *) &ss, &sslen) < 0)
696 return -1;
697
698 return (int) ss.ss_family;
699 }
700
701
702
703
704 /*************************************************
705 * Lookup DSCP settings for a socket *
706 *************************************************/
707
708 struct dscp_name_tableentry {
709 const uschar *name;
710 int value;
711 };
712 /* Keep both of these tables sorted! */
713 static struct dscp_name_tableentry dscp_table[] = {
714 #ifdef IPTOS_DSCP_AF11
715 { CUS"af11", IPTOS_DSCP_AF11 },
716 { CUS"af12", IPTOS_DSCP_AF12 },
717 { CUS"af13", IPTOS_DSCP_AF13 },
718 { CUS"af21", IPTOS_DSCP_AF21 },
719 { CUS"af22", IPTOS_DSCP_AF22 },
720 { CUS"af23", IPTOS_DSCP_AF23 },
721 { CUS"af31", IPTOS_DSCP_AF31 },
722 { CUS"af32", IPTOS_DSCP_AF32 },
723 { CUS"af33", IPTOS_DSCP_AF33 },
724 { CUS"af41", IPTOS_DSCP_AF41 },
725 { CUS"af42", IPTOS_DSCP_AF42 },
726 { CUS"af43", IPTOS_DSCP_AF43 },
727 { CUS"ef", IPTOS_DSCP_EF },
728 #endif
729 #ifdef IPTOS_LOWCOST
730 { CUS"lowcost", IPTOS_LOWCOST },
731 #endif
732 { CUS"lowdelay", IPTOS_LOWDELAY },
733 #ifdef IPTOS_MINCOST
734 { CUS"mincost", IPTOS_MINCOST },
735 #endif
736 { CUS"reliability", IPTOS_RELIABILITY },
737 { CUS"throughput", IPTOS_THROUGHPUT }
738 };
739 static int dscp_table_size =
740 sizeof(dscp_table) / sizeof(struct dscp_name_tableentry);
741
742 /* DSCP values change by protocol family, and so do the options used for
743 setsockopt(); this utility does all the lookups. It takes an unexpanded
744 option string, expands it, strips off affix whitespace, then checks if it's
745 a number. If all of what's left is a number, then that's how the option will
746 be parsed and success/failure is a range check. If it's not all a number,
747 then it must be a supported keyword.
748
749 Arguments:
750 dscp_name a string, so far unvalidated
751 af address_family in use
752 level setsockopt level to use
753 optname setsockopt name to use
754 dscp_value value for dscp_name
755
756 Returns: TRUE if okay to setsockopt(), else FALSE
757
758 *level and *optname may be set even if FALSE is returned
759 */
760
761 BOOL
762 dscp_lookup(const uschar *dscp_name, int af,
763 int *level, int *optname, int *dscp_value)
764 {
765 uschar *dscp_lookup, *p;
766 int first, last;
767 long rawlong;
768
769 if (af == AF_INET)
770 {
771 *level = IPPROTO_IP;
772 *optname = IP_TOS;
773 }
774 #if HAVE_IPV6 && defined(IPV6_TCLASS)
775 else if (af == AF_INET6)
776 {
777 *level = IPPROTO_IPV6;
778 *optname = IPV6_TCLASS;
779 }
780 #endif
781 else
782 {
783 DEBUG(D_transport)
784 debug_printf("Unhandled address family %d in dscp_lookup()\n", af);
785 return FALSE;
786 }
787 if (!dscp_name)
788 {
789 DEBUG(D_transport)
790 debug_printf("[empty DSCP]\n");
791 return FALSE;
792 }
793 dscp_lookup = expand_string(US dscp_name);
794 if (dscp_lookup == NULL || *dscp_lookup == '\0')
795 return FALSE;
796
797 p = dscp_lookup + Ustrlen(dscp_lookup) - 1;
798 while (isspace(*p)) *p-- = '\0';
799 while (isspace(*dscp_lookup) && dscp_lookup < p) dscp_lookup++;
800 if (*dscp_lookup == '\0')
801 return FALSE;
802
803 rawlong = Ustrtol(dscp_lookup, &p, 0);
804 if (p != dscp_lookup && *p == '\0')
805 {
806 /* We have six bits available, which will end up shifted to fit in 0xFC mask.
807 RFC 2597 defines the values unshifted. */
808 if (rawlong < 0 || rawlong > 0x3F)
809 {
810 DEBUG(D_transport)
811 debug_printf("DSCP value %ld out of range, ignored.\n", rawlong);
812 return FALSE;
813 }
814 *dscp_value = rawlong << 2;
815 return TRUE;
816 }
817
818 first = 0;
819 last = dscp_table_size;
820 while (last > first)
821 {
822 int middle = (first + last)/2;
823 int c = Ustrcmp(dscp_lookup, dscp_table[middle].name);
824 if (c == 0)
825 {
826 *dscp_value = dscp_table[middle].value;
827 return TRUE;
828 }
829 else if (c > 0)
830 first = middle + 1;
831 else
832 last = middle;
833 }
834 return FALSE;
835 }
836
837 void
838 dscp_list_to_stream(FILE *stream)
839 {
840 for (int i = 0; i < dscp_table_size; ++i)
841 fprintf(stream, "%s\n", dscp_table[i].name);
842 }
843
844
845 /* End of ip.c */
846 /* vi: aw ai sw=2
847 */