Build: try another dummy for Solaris 10 compile
[exim.git] / src / src / ip.c
... / ...
CommitLineData
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
9got messier, so that it's worth pulling out the code into separate functions
10that other parts of Exim can call, especially as there are now several
11different places in the code where sockets are used. */
12
13
14#include "exim.h"
15
16
17#if defined(TCP_FASTOPEN)
18# if defined(MSG_FASTOPEN) || defined(EXIM_TFO_CONNECTX) || defined(EXIM_TFO_FREEBSD)
19# define EXIM_SUPPORT_TFO
20# endif
21#endif
22
23/*************************************************
24* Create a socket *
25*************************************************/
26
27/* Socket creation happens in a number of places so it's packaged here for
28convenience.
29
30Arguments:
31 type SOCK_DGRAM or SOCK_STREAM
32 af AF_INET or AF_INET6
33
34Returns: socket number or -1 on failure
35*/
36
37int
38ip_socket(int type, int af)
39{
40int sock = socket(af, type, 0);
41if (sock < 0)
42 log_write(0, LOG_MAIN, "IPv%c socket creation failed: %s",
43 (af == AF_INET6)? '6':'4', strerror(errno));
44return sock;
45}
46
47
48
49
50#if HAVE_IPV6
51/*************************************************
52* Convert printing address to numeric *
53*************************************************/
54
55/* This function converts the textual form of an IP address into a numeric form
56in an appropriate structure in an IPv6 environment. The getaddrinfo() function
57can (apparently) handle more complicated addresses (e.g. those containing
58scopes) than inet_pton() in some environments. We use hints to tell it that the
59input must be a numeric address.
60
61However, apparently some operating systems (or libraries) don't support
62getaddrinfo(), so there is a build-time option to revert to inet_pton() (which
63does not support scopes).
64
65Arguments:
66 address textual form of the address
67 addr where to copy back the answer
68
69Returns: nothing - failure provokes a panic-die
70*/
71
72static void
73ip_addrinfo(const uschar *address, struct sockaddr_in6 *saddr)
74{
75#ifdef IPV6_USE_INET_PTON
76
77 if (inet_pton(AF_INET6, CCS address, &saddr->sin6_addr) != 1)
78 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "unable to parse \"%s\" as an "
79 "IP address", address);
80 saddr->sin6_family = AF_INET6;
81
82#else
83
84 int rc;
85 struct addrinfo hints, *res;
86 memset(&hints, 0, sizeof(hints));
87 hints.ai_family = AF_INET6;
88 hints.ai_socktype = SOCK_STREAM;
89 hints.ai_flags = AI_NUMERICHOST;
90 if ((rc = getaddrinfo(CCS address, NULL, &hints, &res)) != 0 || res == NULL)
91 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "unable to parse \"%s\" as an "
92 "IP address: %s", address,
93 (rc == 0)? "NULL result returned" : gai_strerror(rc));
94 memcpy(saddr, res->ai_addr, res->ai_addrlen);
95 freeaddrinfo(res);
96
97#endif
98}
99#endif /* HAVE_IPV6 */
100
101
102/*************************************************
103* Bind socket to interface and port *
104*************************************************/
105
106int
107ip_addr(void * sin_, int af, const uschar * address, int port)
108{
109union sockaddr_46 * sin = sin_;
110memset(sin, 0, sizeof(*sin));
111
112/* Setup code when using an IPv6 socket. The wildcard address is ":", to
113ensure an IPv6 socket is used. */
114
115#if HAVE_IPV6
116if (af == AF_INET6)
117 {
118 if (address[0] == ':' && address[1] == 0)
119 {
120 sin->v6.sin6_family = AF_INET6;
121 sin->v6.sin6_addr = in6addr_any;
122 }
123 else
124 ip_addrinfo(address, &sin->v6); /* Panic-dies on error */
125 sin->v6.sin6_port = htons(port);
126 return sizeof(sin->v6);
127 }
128else
129#else /* HAVE_IPv6 */
130af = af; /* Avoid compiler warning */
131#endif /* HAVE_IPV6 */
132
133/* Setup code when using IPv4 socket. The wildcard address is "". */
134
135 {
136 sin->v4.sin_family = AF_INET;
137 sin->v4.sin_port = htons(port);
138 sin->v4.sin_addr.s_addr = address[0] == 0
139 ? (S_ADDR_TYPE)INADDR_ANY
140 : (S_ADDR_TYPE)inet_addr(CS address);
141 return sizeof(sin->v4);
142 }
143}
144
145
146
147/* This function binds a socket to a local interface address and port. For a
148wildcard IPv6 bind, the address is ":".
149
150Arguments:
151 sock the socket
152 af AF_INET or AF_INET6 - the socket type
153 address the IP address, in text form
154 port the IP port (host order)
155
156Returns: the result of bind()
157*/
158
159int
160ip_bind(int sock, int af, uschar *address, int port)
161{
162union sockaddr_46 sin;
163int s_len = ip_addr(&sin, af, address, port);
164return bind(sock, (struct sockaddr *)&sin, s_len);
165}
166
167
168
169/*************************************************
170* Connect socket to remote host *
171*************************************************/
172
173/* This function connects a socket to a remote address and port. The socket may
174or may not have previously been bound to a local interface. The socket is not
175closed, even in cases of error. It is expected that the calling function, which
176created the socket, will be the one that closes it.
177
178Arguments:
179 sock the socket
180 af AF_INET6 or AF_INET for the socket type
181 address the remote address, in text form
182 port the remote port
183 timeout a timeout (zero for indefinite timeout)
184 fastopen_blob non-null iff TCP_FASTOPEN can be used; may indicate early-data to
185 be sent in SYN segment. Any such data must be idempotent.
186
187Returns: 0 on success; -1 on failure, with errno set
188*/
189
190int
191ip_connect(int sock, int af, const uschar *address, int port, int timeout,
192 const blob * fastopen_blob)
193{
194struct sockaddr_in s_in4;
195struct sockaddr *s_ptr;
196int s_len, rc, save_errno;
197
198/* For an IPv6 address, use an IPv6 sockaddr structure. */
199
200#if HAVE_IPV6
201struct sockaddr_in6 s_in6;
202if (af == AF_INET6)
203 {
204 memset(&s_in6, 0, sizeof(s_in6));
205 ip_addrinfo(address, &s_in6); /* Panic-dies on error */
206 s_in6.sin6_port = htons(port);
207 s_ptr = (struct sockaddr *)&s_in6;
208 s_len = sizeof(s_in6);
209 }
210else
211#else /* HAVE_IPV6 */
212af = af; /* Avoid compiler warning */
213#endif /* HAVE_IPV6 */
214
215/* For an IPv4 address, use an IPv4 sockaddr structure, even on a system with
216IPv6 support. */
217
218 {
219 memset(&s_in4, 0, sizeof(s_in4));
220 s_in4.sin_family = AF_INET;
221 s_in4.sin_port = htons(port);
222 s_in4.sin_addr.s_addr = (S_ADDR_TYPE)inet_addr(CCS address);
223 s_ptr = (struct sockaddr *)&s_in4;
224 s_len = sizeof(s_in4);
225 }
226
227/* If no connection timeout is set, just call connect() without setting a
228timer, thereby allowing the inbuilt OS timeout to operate. */
229
230callout_address = string_sprintf("[%s]:%d", address, port);
231sigalrm_seen = FALSE;
232if (timeout > 0) ALARM(timeout);
233
234#ifdef EXIM_SUPPORT_TFO
235/* TCP Fast Open, if the system has a cookie from a previous call to
236this peer, can send data in the SYN packet. The peer can send data
237before it gets our ACK of its SYN,ACK - the latter is useful for
238the SMTP banner. Other (than SMTP) cases of TCP connections can
239possibly use the data-on-syn, so support that too. */
240
241if (fastopen_blob && f.tcp_fastopen_ok)
242 {
243# ifdef MSG_FASTOPEN
244 /* This is a Linux implementation. */
245
246 if ((rc = sendto(sock, fastopen_blob->data, fastopen_blob->len,
247 MSG_FASTOPEN | MSG_DONTWAIT, s_ptr, s_len)) >= 0)
248 /* seen for with-data, experimental TFO option, with-cookie case */
249 /* seen for with-data, proper TFO opt, with-cookie case */
250 {
251 DEBUG(D_transport|D_v)
252 debug_printf(" TFO mode connection attempt to %s, %lu data\n",
253 address, (unsigned long)fastopen_blob->len);
254 /*XXX also seen on successful TFO, sigh */
255 tcp_out_fastopen = fastopen_blob->len > 0 ? TFO_ATTEMPTED_DATA : TFO_ATTEMPTED_NODATA;
256 }
257 else if (errno == EINPROGRESS) /* expected if we had no cookie for peer */
258 /* seen for no-data, proper TFO option, both cookie-request and with-cookie cases */
259 /* apparently no visibility of the diffference at this point */
260 /* seen for with-data, proper TFO opt, cookie-req */
261 /* with netwk delay, post-conn tcp_info sees unacked 1 for R, 2 for C; code in smtp_out.c */
262 /* ? older Experimental TFO option behaviour ? */
263 { /* queue unsent data */
264 DEBUG(D_transport|D_v) debug_printf(" TFO mode sendto, %s data: EINPROGRESS\n",
265 fastopen_blob->len > 0 ? "with" : "no");
266 if (!fastopen_blob->data)
267 {
268 tcp_out_fastopen = TFO_ATTEMPTED_NODATA; /* we tried; unknown if useful yet */
269 rc = 0;
270 }
271 else
272 rc = send(sock, fastopen_blob->data, fastopen_blob->len, 0);
273 }
274 else if(errno == EOPNOTSUPP)
275 {
276 DEBUG(D_transport)
277 debug_printf("Tried TCP Fast Open but apparently not enabled by sysctl\n");
278 goto legacy_connect;
279 }
280
281# elif defined(EXIM_TFO_FREEBSD)
282 /* Re: https://people.freebsd.org/~pkelsey/tfo-tools/tfo-client.c */
283
284 if (setsockopt(sock, IPPROTO_TCP, TCP_FASTOPEN, &on, sizeof(on)) < 0)
285 {
286 DEBUG(D_transport)
287 debug_printf("Tried TCP Fast Open but apparently not enabled by sysctl\n");
288 goto legacy_connect;
289 }
290 if ((rc = sendto(sock, fastopen_blob->data, fastopen_blob->len, 0,
291 s_ptr, s_len)) >= 0)
292 {
293 DEBUG(D_transport|D_v)
294 debug_printf(" TFO mode connection attempt to %s, %lu data\n",
295 address, (unsigned long)fastopen_blob->len);
296 tcp_out_fastopen = fastopen_blob->len > 0 ? TFO_ATTEMPTED_DATA : TFO_ATTEMPTED_NODATA;
297 }
298
299# elif defined(EXIM_TFO_CONNECTX)
300 /* MacOS */
301 sa_endpoints_t ends = {
302 .sae_srcif = 0, .sae_srcaddr = NULL, .sae_srcaddrlen = 0,
303 .sae_dstaddr = s_ptr, .sae_dstaddrlen = s_len };
304 struct iovec iov = {
305 .iov_base = fastopen_blob->data, .iov_len = fastopen_blob->len };
306 size_t len;
307
308 if ((rc = connectx(sock, &ends, SAE_ASSOCID_ANY,
309 CONNECT_DATA_IDEMPOTENT, &iov, 1, &len, NULL)) == 0)
310 {
311 DEBUG(D_transport|D_v)
312 debug_printf(" TFO mode connection attempt to %s, %lu data\n",
313 address, (unsigned long)fastopen_blob->len);
314 tcp_out_fastopen = fastopen_blob->len > 0 ? TFO_ATTEMPTED_DATA : TFO_ATTEMPTED_NODATA;
315
316 if (len != fastopen_blob->len)
317 DEBUG(D_transport|D_v)
318 debug_printf(" only queued %lu data!\n", (unsigned long)len);
319 }
320 else if (errno == EINPROGRESS)
321 {
322 DEBUG(D_transport|D_v) debug_printf(" TFO mode connectx, %s data: EINPROGRESS\n",
323 fastopen_blob->len > 0 ? "with" : "no");
324 if (!fastopen_blob->data)
325 {
326 tcp_out_fastopen = TFO_ATTEMPTED_NODATA; /* we tried; unknown if useful yet */
327 rc = 0;
328 }
329 else /* assume that no data was queued; block in send */
330 rc = send(sock, fastopen_blob->data, fastopen_blob->len, 0);
331 }
332# endif
333 }
334else
335#endif /*EXIM_SUPPORT_TFO*/
336 {
337#if defined(EXIM_SUPPORT_TFO) && !defined(EXIM_TFO_CONNECTX)
338legacy_connect:
339#endif
340
341 DEBUG(D_transport|D_v) if (fastopen_blob)
342 debug_printf(" non-TFO mode connection attempt to %s, %lu data\n",
343 address, (unsigned long)fastopen_blob->len);
344 if ((rc = connect(sock, s_ptr, s_len)) >= 0)
345 if ( fastopen_blob && fastopen_blob->data && fastopen_blob->len
346 && send(sock, fastopen_blob->data, fastopen_blob->len, 0) < 0)
347 rc = -1;
348 }
349
350save_errno = errno;
351ALARM_CLR(0);
352
353/* There is a testing facility for simulating a connection timeout, as I
354can't think of any other way of doing this. It converts a connection refused
355into a timeout if the timeout is set to 999999. */
356
357if (f.running_in_test_harness && save_errno == ECONNREFUSED && timeout == 999999)
358 {
359 rc = -1;
360 save_errno = EINTR;
361 sigalrm_seen = TRUE;
362 }
363
364/* Success */
365
366if (rc >= 0)
367 return 0;
368
369/* A failure whose error code is "Interrupted system call" is in fact
370an externally applied timeout if the signal handler has been run. */
371
372errno = save_errno == EINTR && sigalrm_seen ? ETIMEDOUT : save_errno;
373return -1;
374}
375
376
377
378/*************************************************
379* Create connected socket to remote host *
380*************************************************/
381
382/* Create a socket and connect to host (name or number, ipv6 ok)
383 at one of port-range.
384
385Arguments:
386 type SOCK_DGRAM or SOCK_STREAM
387 af AF_INET6 or AF_INET for the socket type
388 hostname host name, or ip address (as text)
389 portlo,porthi the remote port range
390 timeout a timeout
391 connhost if not NULL, host_item to be filled in with connection details
392 errstr pointer for allocated string on error
393 fastopen_blob with SOCK_STREAM, if non-null, request TCP Fast Open.
394 Additionally, optional idempotent early-data to send
395
396Return:
397 socket fd, or -1 on failure (having allocated an error string)
398*/
399int
400ip_connectedsocket(int type, const uschar * hostname, int portlo, int porthi,
401 int timeout, host_item * connhost, uschar ** errstr, const blob * fastopen_blob)
402{
403int namelen;
404host_item shost;
405int af = 0, fd, fd4 = -1, fd6 = -1;
406
407shost.next = NULL;
408shost.address = NULL;
409shost.port = portlo;
410shost.mx = -1;
411
412namelen = Ustrlen(hostname);
413
414/* Anything enclosed in [] must be an IP address. */
415
416if (hostname[0] == '[' &&
417 hostname[namelen - 1] == ']')
418 {
419 uschar * host = string_copyn(hostname+1, namelen-2);
420 if (string_is_ip_address(host, NULL) == 0)
421 {
422 *errstr = string_sprintf("malformed IP address \"%s\"", hostname);
423 return -1;
424 }
425 shost.name = shost.address = host;
426 }
427
428/* Otherwise check for an unadorned IP address */
429
430else if (string_is_ip_address(hostname, NULL) != 0)
431 shost.name = shost.address = string_copyn(hostname, namelen);
432
433/* Otherwise lookup IP address(es) from the name */
434
435else
436 {
437 shost.name = string_copyn(hostname, namelen);
438 if (host_find_byname(&shost, NULL, HOST_FIND_QUALIFY_SINGLE,
439 NULL, FALSE) != HOST_FOUND)
440 {
441 *errstr = string_sprintf("no IP address found for host %s", shost.name);
442 return -1;
443 }
444 }
445
446/* Try to connect to the server - test each IP till one works */
447
448for (host_item * h = &shost; h; h = h->next)
449 {
450 fd = Ustrchr(h->address, ':') != 0
451 ? fd6 < 0 ? (fd6 = ip_socket(type, af = AF_INET6)) : fd6
452 : fd4 < 0 ? (fd4 = ip_socket(type, af = AF_INET )) : fd4;
453
454 if (fd < 0)
455 {
456 *errstr = string_sprintf("failed to create socket: %s", strerror(errno));
457 goto bad;
458 }
459
460 for (int port = portlo; port <= porthi; port++)
461 if (ip_connect(fd, af, h->address, port, timeout, fastopen_blob) == 0)
462 {
463 if (fd != fd6) close(fd6);
464 if (fd != fd4) close(fd4);
465 if (connhost)
466 {
467 h->port = port;
468 *connhost = *h;
469 connhost->next = NULL;
470 }
471 return fd;
472 }
473 }
474
475*errstr = string_sprintf("failed to connect to any address for %s: %s",
476 hostname, strerror(errno));
477
478bad:
479 close(fd4); close(fd6); return -1;
480}
481
482
483/*XXX TFO? */
484int
485ip_tcpsocket(const uschar * hostport, uschar ** errstr, int tmo)
486{
487int scan;
488uschar hostname[256];
489unsigned int portlow, porthigh;
490
491/* extract host and port part */
492scan = sscanf(CS hostport, "%255s %u-%u", hostname, &portlow, &porthigh);
493if (scan != 3)
494 {
495 if (scan != 2)
496 {
497 *errstr = string_sprintf("invalid socket '%s'", hostport);
498 return -1;
499 }
500 porthigh = portlow;
501 }
502
503return ip_connectedsocket(SOCK_STREAM, hostname, portlow, porthigh,
504 tmo, NULL, errstr, NULL);
505}
506
507int
508ip_unixsocket(const uschar * path, uschar ** errstr)
509{
510int sock;
511struct sockaddr_un server;
512
513if ((sock = socket(AF_UNIX, SOCK_STREAM, 0)) < 0)
514 {
515 *errstr = US"can't open UNIX socket.";
516 return -1;
517 }
518
519callout_address = string_copy(path);
520server.sun_family = AF_UNIX;
521Ustrncpy(US server.sun_path, path, sizeof(server.sun_path)-1);
522server.sun_path[sizeof(server.sun_path)-1] = '\0';
523if (connect(sock, (struct sockaddr *) &server, sizeof(server)) < 0)
524 {
525 int err = errno;
526 (void)close(sock);
527 *errstr = string_sprintf("unable to connect to UNIX socket (%s): %s",
528 path, strerror(err));
529 return -1;
530 }
531return sock;
532}
533
534/* spec is either an absolute path (with a leading /), or
535a host (name or IP) and port (whitespace-separated).
536The port can be a range, dash-separated, or a single number.
537*/
538int
539ip_streamsocket(const uschar * spec, uschar ** errstr, int tmo)
540{
541return *spec == '/'
542 ? ip_unixsocket(spec, errstr) : ip_tcpsocket(spec, errstr, tmo);
543}
544
545/*************************************************
546* Set keepalive on a socket *
547*************************************************/
548
549/* Can be called for both incoming and outgoing sockets.
550
551Arguments:
552 sock the socket
553 address the remote host address, for failure logging
554 torf true for outgoing connection, false for incoming
555
556Returns: nothing
557*/
558
559void
560ip_keepalive(int sock, const uschar *address, BOOL torf)
561{
562int fodder = 1;
563if (setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
564 US (&fodder), sizeof(fodder)) != 0)
565 log_write(0, LOG_MAIN, "setsockopt(SO_KEEPALIVE) on connection %s %s "
566 "failed: %s", torf? "to":"from", address, strerror(errno));
567}
568
569
570
571/*************************************************
572* Receive from a socket with timeout *
573*************************************************/
574
575/*
576Arguments:
577 fd the file descriptor
578 timelimit the timeout endpoint, seconds-since-epoch
579Returns: TRUE => ready for i/o
580 FALSE => timed out, or other error
581*/
582BOOL
583fd_ready(int fd, time_t timelimit)
584{
585fd_set select_inset;
586int time_left = timelimit - time(NULL);
587int rc;
588
589if (time_left <= 0)
590 {
591 errno = ETIMEDOUT;
592 return FALSE;
593 }
594/* Wait until the socket is ready */
595
596do
597 {
598 struct timeval tv = { .tv_sec = time_left, .tv_usec = 0 };
599 FD_ZERO (&select_inset);
600 FD_SET (fd, &select_inset);
601
602 /*DEBUG(D_transport) debug_printf("waiting for data on fd\n");*/
603 rc = select(fd + 1, (SELECT_ARG2_TYPE *)&select_inset, NULL, NULL, &tv);
604
605 /* If some interrupt arrived, just retry. We presume this to be rare,
606 but it can happen (e.g. the SIGUSR1 signal sent by exiwhat causes
607 select() to exit).
608
609 Aug 2004: Somebody set up a cron job that ran exiwhat every 2 minutes, making
610 the interrupt not at all rare. Since the timeout is typically more than 2
611 minutes, the effect was to block the timeout completely. To prevent this
612 happening again, we do an explicit time test and adjust the timeout
613 accordingly */
614
615 if (rc < 0 && errno == EINTR)
616 {
617 DEBUG(D_transport) debug_printf("EINTR while waiting for socket data\n");
618
619 /* Watch out, 'continue' jumps to the condition, not to the loops top */
620 if ((time_left = timelimit - time(NULL)) > 0) continue;
621 }
622
623 if (rc <= 0)
624 {
625 errno = ETIMEDOUT;
626 return FALSE;
627 }
628
629 /* Checking the FD_ISSET is not enough, if we're interrupted, the
630 select_inset may still contain the 'input'. */
631 }
632while (rc < 0 || !FD_ISSET(fd, &select_inset));
633return TRUE;
634}
635
636/* The timeout is implemented using select(), and we loop to cover select()
637getting interrupted, and the possibility of select() returning with a positive
638result but no ready descriptor. Is this in fact possible?
639
640Arguments:
641 cctx the connection context (socket fd, possibly TLS context)
642 buffer to read into
643 bufsize the buffer size
644 timelimit the timeout endpoint, seconds-since-epoch
645
646Returns: > 0 => that much data read
647 <= 0 on error or EOF; errno set - zero for EOF
648*/
649
650int
651ip_recv(client_conn_ctx * cctx, uschar * buffer, int buffsize, time_t timelimit)
652{
653int rc;
654
655if (!fd_ready(cctx->sock, timelimit))
656 return -1;
657
658/* The socket is ready, read from it (via TLS if it's active). On EOF (i.e.
659close down of the connection), set errno to zero; otherwise leave it alone. */
660
661#ifndef DISABLE_TLS
662if (cctx->tls_ctx) /* client TLS */
663 rc = tls_read(cctx->tls_ctx, buffer, buffsize);
664else if (tls_in.active.sock == cctx->sock) /* server TLS */
665 rc = tls_read(NULL, buffer, buffsize);
666else
667#endif
668 rc = recv(cctx->sock, buffer, buffsize, 0);
669
670if (rc > 0) return rc;
671if (rc == 0) errno = 0;
672return -1;
673}
674
675
676
677
678/*************************************************
679* Lookup address family of potential socket *
680*************************************************/
681
682/* Given a file-descriptor, check to see if it's a socket and, if so,
683return the address family; detects IPv4 vs IPv6. If not a socket then
684return -1.
685
686The value 0 is typically AF_UNSPEC, which should not be seen on a connected
687fd. If the return is -1, the errno will be from getsockname(); probably
688ENOTSOCK or ECONNRESET.
689
690Arguments: socket-or-not fd
691Returns: address family or -1
692*/
693
694int
695ip_get_address_family(int fd)
696{
697struct sockaddr_storage ss;
698socklen_t sslen = sizeof(ss);
699
700if (getsockname(fd, (struct sockaddr *) &ss, &sslen) < 0)
701 return -1;
702
703return (int) ss.ss_family;
704}
705
706
707
708
709/*************************************************
710* Lookup DSCP settings for a socket *
711*************************************************/
712
713struct dscp_name_tableentry {
714 const uschar *name;
715 int value;
716};
717/* Keep both of these tables sorted! */
718static struct dscp_name_tableentry dscp_table[] = {
719#ifdef IPTOS_DSCP_AF11
720 { CUS"af11", IPTOS_DSCP_AF11 },
721 { CUS"af12", IPTOS_DSCP_AF12 },
722 { CUS"af13", IPTOS_DSCP_AF13 },
723 { CUS"af21", IPTOS_DSCP_AF21 },
724 { CUS"af22", IPTOS_DSCP_AF22 },
725 { CUS"af23", IPTOS_DSCP_AF23 },
726 { CUS"af31", IPTOS_DSCP_AF31 },
727 { CUS"af32", IPTOS_DSCP_AF32 },
728 { CUS"af33", IPTOS_DSCP_AF33 },
729 { CUS"af41", IPTOS_DSCP_AF41 },
730 { CUS"af42", IPTOS_DSCP_AF42 },
731 { CUS"af43", IPTOS_DSCP_AF43 },
732 { CUS"ef", IPTOS_DSCP_EF },
733#endif
734#ifdef IPTOS_LOWCOST
735 { CUS"lowcost", IPTOS_LOWCOST },
736#endif
737 { CUS"lowdelay", IPTOS_LOWDELAY },
738#ifdef IPTOS_MINCOST
739 { CUS"mincost", IPTOS_MINCOST },
740#endif
741 { CUS"reliability", IPTOS_RELIABILITY },
742 { CUS"throughput", IPTOS_THROUGHPUT }
743};
744static int dscp_table_size =
745 sizeof(dscp_table) / sizeof(struct dscp_name_tableentry);
746
747/* DSCP values change by protocol family, and so do the options used for
748setsockopt(); this utility does all the lookups. It takes an unexpanded
749option string, expands it, strips off affix whitespace, then checks if it's
750a number. If all of what's left is a number, then that's how the option will
751be parsed and success/failure is a range check. If it's not all a number,
752then it must be a supported keyword.
753
754Arguments:
755 dscp_name a string, so far unvalidated
756 af address_family in use
757 level setsockopt level to use
758 optname setsockopt name to use
759 dscp_value value for dscp_name
760
761Returns: TRUE if okay to setsockopt(), else FALSE
762
763*level and *optname may be set even if FALSE is returned
764*/
765
766BOOL
767dscp_lookup(const uschar *dscp_name, int af,
768 int *level, int *optname, int *dscp_value)
769{
770uschar *dscp_lookup, *p;
771int first, last;
772long rawlong;
773
774if (af == AF_INET)
775 {
776 *level = IPPROTO_IP;
777 *optname = IP_TOS;
778 }
779#if HAVE_IPV6 && defined(IPV6_TCLASS)
780else if (af == AF_INET6)
781 {
782 *level = IPPROTO_IPV6;
783 *optname = IPV6_TCLASS;
784 }
785#endif
786else
787 {
788 DEBUG(D_transport)
789 debug_printf("Unhandled address family %d in dscp_lookup()\n", af);
790 return FALSE;
791 }
792if (!dscp_name)
793 {
794 DEBUG(D_transport)
795 debug_printf("[empty DSCP]\n");
796 return FALSE;
797 }
798dscp_lookup = expand_string(US dscp_name);
799if (dscp_lookup == NULL || *dscp_lookup == '\0')
800 return FALSE;
801
802p = dscp_lookup + Ustrlen(dscp_lookup) - 1;
803while (isspace(*p)) *p-- = '\0';
804while (isspace(*dscp_lookup) && dscp_lookup < p) dscp_lookup++;
805if (*dscp_lookup == '\0')
806 return FALSE;
807
808rawlong = Ustrtol(dscp_lookup, &p, 0);
809if (p != dscp_lookup && *p == '\0')
810 {
811 /* We have six bits available, which will end up shifted to fit in 0xFC mask.
812 RFC 2597 defines the values unshifted. */
813 if (rawlong < 0 || rawlong > 0x3F)
814 {
815 DEBUG(D_transport)
816 debug_printf("DSCP value %ld out of range, ignored.\n", rawlong);
817 return FALSE;
818 }
819 *dscp_value = rawlong << 2;
820 return TRUE;
821 }
822
823first = 0;
824last = dscp_table_size;
825while (last > first)
826 {
827 int middle = (first + last)/2;
828 int c = Ustrcmp(dscp_lookup, dscp_table[middle].name);
829 if (c == 0)
830 {
831 *dscp_value = dscp_table[middle].value;
832 return TRUE;
833 }
834 else if (c > 0)
835 first = middle + 1;
836 else
837 last = middle;
838 }
839return FALSE;
840}
841
842void
843dscp_list_to_stream(FILE *stream)
844{
845for (int i = 0; i < dscp_table_size; ++i)
846 fprintf(stream, "%s\n", dscp_table[i].name);
847}
848
849
850/* End of ip.c */
851/* vi: aw ai sw=2
852*/