Basic cutthrough delivery.
[exim.git] / src / src / verify.c
1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
4
5 /* Copyright (c) University of Cambridge 1995 - 2009 */
6 /* See the file NOTICE for conditions of use and distribution. */
7
8 /* Functions concerned with verifying things. The original code for callout
9 caching was contributed by Kevin Fleming (but I hacked it around a bit). */
10
11
12 #include "exim.h"
13
14 #define CUTTHROUGH_CMD_TIMEOUT 30 /* timeout for cutthrough-routing calls */
15 #define CUTTHROUGH_DATA_TIMEOUT 60 /* timeout for cutthrough-routing calls */
16 address_item cutthrough_addr;
17
18 /* Structure for caching DNSBL lookups */
19
20 typedef struct dnsbl_cache_block {
21 dns_address *rhs;
22 uschar *text;
23 int rc;
24 BOOL text_set;
25 } dnsbl_cache_block;
26
27
28 /* Anchor for DNSBL cache */
29
30 static tree_node *dnsbl_cache = NULL;
31
32
33 /* Bits for match_type in one_check_dnsbl() */
34
35 #define MT_NOT 1
36 #define MT_ALL 2
37
38
39
40 /*************************************************
41 * Retrieve a callout cache record *
42 *************************************************/
43
44 /* If a record exists, check whether it has expired.
45
46 Arguments:
47 dbm_file an open hints file
48 key the record key
49 type "address" or "domain"
50 positive_expire expire time for positive records
51 negative_expire expire time for negative records
52
53 Returns: the cache record if a non-expired one exists, else NULL
54 */
55
56 static dbdata_callout_cache *
57 get_callout_cache_record(open_db *dbm_file, uschar *key, uschar *type,
58 int positive_expire, int negative_expire)
59 {
60 BOOL negative;
61 int length, expire;
62 time_t now;
63 dbdata_callout_cache *cache_record;
64
65 cache_record = dbfn_read_with_length(dbm_file, key, &length);
66
67 if (cache_record == NULL)
68 {
69 HDEBUG(D_verify) debug_printf("callout cache: no %s record found\n", type);
70 return NULL;
71 }
72
73 /* We treat a record as "negative" if its result field is not positive, or if
74 it is a domain record and the postmaster field is negative. */
75
76 negative = cache_record->result != ccache_accept ||
77 (type[0] == 'd' && cache_record->postmaster_result == ccache_reject);
78 expire = negative? negative_expire : positive_expire;
79 now = time(NULL);
80
81 if (now - cache_record->time_stamp > expire)
82 {
83 HDEBUG(D_verify) debug_printf("callout cache: %s record expired\n", type);
84 return NULL;
85 }
86
87 /* If this is a non-reject domain record, check for the obsolete format version
88 that doesn't have the postmaster and random timestamps, by looking at the
89 length. If so, copy it to a new-style block, replicating the record's
90 timestamp. Then check the additional timestamps. (There's no point wasting
91 effort if connections are rejected.) */
92
93 if (type[0] == 'd' && cache_record->result != ccache_reject)
94 {
95 if (length == sizeof(dbdata_callout_cache_obs))
96 {
97 dbdata_callout_cache *new = store_get(sizeof(dbdata_callout_cache));
98 memcpy(new, cache_record, length);
99 new->postmaster_stamp = new->random_stamp = new->time_stamp;
100 cache_record = new;
101 }
102
103 if (now - cache_record->postmaster_stamp > expire)
104 cache_record->postmaster_result = ccache_unknown;
105
106 if (now - cache_record->random_stamp > expire)
107 cache_record->random_result = ccache_unknown;
108 }
109
110 HDEBUG(D_verify) debug_printf("callout cache: found %s record\n", type);
111 return cache_record;
112 }
113
114
115
116 /*************************************************
117 * Do callout verification for an address *
118 *************************************************/
119
120 /* This function is called from verify_address() when the address has routed to
121 a host list, and a callout has been requested. Callouts are expensive; that is
122 why a cache is used to improve the efficiency.
123
124 Arguments:
125 addr the address that's been routed
126 host_list the list of hosts to try
127 tf the transport feedback block
128
129 ifstring "interface" option from transport, or NULL
130 portstring "port" option from transport, or NULL
131 protocolstring "protocol" option from transport, or NULL
132 callout the per-command callout timeout
133 callout_overall the overall callout timeout (if < 0 use 4*callout)
134 callout_connect the callout connection timeout (if < 0 use callout)
135 options the verification options - these bits are used:
136 vopt_is_recipient => this is a recipient address
137 vopt_callout_no_cache => don't use callout cache
138 vopt_callout_fullpm => if postmaster check, do full one
139 vopt_callout_random => do the "random" thing
140 vopt_callout_recipsender => use real sender for recipient
141 vopt_callout_recippmaster => use postmaster for recipient
142 se_mailfrom MAIL FROM address for sender verify; NULL => ""
143 pm_mailfrom if non-NULL, do the postmaster check with this sender
144
145 Returns: OK/FAIL/DEFER
146 */
147
148 static int
149 do_callout(address_item *addr, host_item *host_list, transport_feedback *tf,
150 int callout, int callout_overall, int callout_connect, int options,
151 uschar *se_mailfrom, uschar *pm_mailfrom)
152 {
153 BOOL is_recipient = (options & vopt_is_recipient) != 0;
154 BOOL callout_no_cache = (options & vopt_callout_no_cache) != 0;
155 BOOL callout_random = (options & vopt_callout_random) != 0;
156
157 int yield = OK;
158 int old_domain_cache_result = ccache_accept;
159 BOOL done = FALSE;
160 uschar *address_key;
161 uschar *from_address;
162 uschar *random_local_part = NULL;
163 uschar *save_deliver_domain = deliver_domain;
164 uschar **failure_ptr = is_recipient?
165 &recipient_verify_failure : &sender_verify_failure;
166 open_db dbblock;
167 open_db *dbm_file = NULL;
168 dbdata_callout_cache new_domain_record;
169 dbdata_callout_cache_address new_address_record;
170 host_item *host;
171 time_t callout_start_time;
172
173 new_domain_record.result = ccache_unknown;
174 new_domain_record.postmaster_result = ccache_unknown;
175 new_domain_record.random_result = ccache_unknown;
176
177 memset(&new_address_record, 0, sizeof(new_address_record));
178
179 /* For a recipient callout, the key used for the address cache record must
180 include the sender address if we are using the real sender in the callout,
181 because that may influence the result of the callout. */
182
183 address_key = addr->address;
184 from_address = US"";
185
186 if (is_recipient)
187 {
188 if ((options & vopt_callout_recipsender) != 0)
189 {
190 address_key = string_sprintf("%s/<%s>", addr->address, sender_address);
191 from_address = sender_address;
192 }
193 else if ((options & vopt_callout_recippmaster) != 0)
194 {
195 address_key = string_sprintf("%s/<postmaster@%s>", addr->address,
196 qualify_domain_sender);
197 from_address = string_sprintf("postmaster@%s", qualify_domain_sender);
198 }
199 }
200
201 /* For a sender callout, we must adjust the key if the mailfrom address is not
202 empty. */
203
204 else
205 {
206 from_address = (se_mailfrom == NULL)? US"" : se_mailfrom;
207 if (from_address[0] != 0)
208 address_key = string_sprintf("%s/<%s>", addr->address, from_address);
209 }
210
211 /* Open the callout cache database, it it exists, for reading only at this
212 stage, unless caching has been disabled. */
213
214 if (callout_no_cache)
215 {
216 HDEBUG(D_verify) debug_printf("callout cache: disabled by no_cache\n");
217 }
218 else if ((dbm_file = dbfn_open(US"callout", O_RDWR, &dbblock, FALSE)) == NULL)
219 {
220 HDEBUG(D_verify) debug_printf("callout cache: not available\n");
221 }
222
223 /* If a cache database is available see if we can avoid the need to do an
224 actual callout by making use of previously-obtained data. */
225
226 if (dbm_file != NULL)
227 {
228 dbdata_callout_cache_address *cache_address_record;
229 dbdata_callout_cache *cache_record = get_callout_cache_record(dbm_file,
230 addr->domain, US"domain",
231 callout_cache_domain_positive_expire,
232 callout_cache_domain_negative_expire);
233
234 /* If an unexpired cache record was found for this domain, see if the callout
235 process can be short-circuited. */
236
237 if (cache_record != NULL)
238 {
239 /* In most cases, if an early command (up to and including MAIL FROM:<>)
240 was rejected, there is no point carrying on. The callout fails. However, if
241 we are doing a recipient verification with use_sender or use_postmaster
242 set, a previous failure of MAIL FROM:<> doesn't count, because this time we
243 will be using a non-empty sender. We have to remember this situation so as
244 not to disturb the cached domain value if this whole verification succeeds
245 (we don't want it turning into "accept"). */
246
247 old_domain_cache_result = cache_record->result;
248
249 if (cache_record->result == ccache_reject ||
250 (*from_address == 0 && cache_record->result == ccache_reject_mfnull))
251 {
252 setflag(addr, af_verify_nsfail);
253 HDEBUG(D_verify)
254 debug_printf("callout cache: domain gave initial rejection, or "
255 "does not accept HELO or MAIL FROM:<>\n");
256 setflag(addr, af_verify_nsfail);
257 addr->user_message = US"(result of an earlier callout reused).";
258 yield = FAIL;
259 *failure_ptr = US"mail";
260 goto END_CALLOUT;
261 }
262
263 /* If a previous check on a "random" local part was accepted, we assume
264 that the server does not do any checking on local parts. There is therefore
265 no point in doing the callout, because it will always be successful. If a
266 random check previously failed, arrange not to do it again, but preserve
267 the data in the new record. If a random check is required but hasn't been
268 done, skip the remaining cache processing. */
269
270 if (callout_random) switch(cache_record->random_result)
271 {
272 case ccache_accept:
273 HDEBUG(D_verify)
274 debug_printf("callout cache: domain accepts random addresses\n");
275 goto END_CALLOUT; /* Default yield is OK */
276
277 case ccache_reject:
278 HDEBUG(D_verify)
279 debug_printf("callout cache: domain rejects random addresses\n");
280 callout_random = FALSE;
281 new_domain_record.random_result = ccache_reject;
282 new_domain_record.random_stamp = cache_record->random_stamp;
283 break;
284
285 default:
286 HDEBUG(D_verify)
287 debug_printf("callout cache: need to check random address handling "
288 "(not cached or cache expired)\n");
289 goto END_CACHE;
290 }
291
292 /* If a postmaster check is requested, but there was a previous failure,
293 there is again no point in carrying on. If a postmaster check is required,
294 but has not been done before, we are going to have to do a callout, so skip
295 remaining cache processing. */
296
297 if (pm_mailfrom != NULL)
298 {
299 if (cache_record->postmaster_result == ccache_reject)
300 {
301 setflag(addr, af_verify_pmfail);
302 HDEBUG(D_verify)
303 debug_printf("callout cache: domain does not accept "
304 "RCPT TO:<postmaster@domain>\n");
305 yield = FAIL;
306 *failure_ptr = US"postmaster";
307 setflag(addr, af_verify_pmfail);
308 addr->user_message = US"(result of earlier verification reused).";
309 goto END_CALLOUT;
310 }
311 if (cache_record->postmaster_result == ccache_unknown)
312 {
313 HDEBUG(D_verify)
314 debug_printf("callout cache: need to check RCPT "
315 "TO:<postmaster@domain> (not cached or cache expired)\n");
316 goto END_CACHE;
317 }
318
319 /* If cache says OK, set pm_mailfrom NULL to prevent a redundant
320 postmaster check if the address itself has to be checked. Also ensure
321 that the value in the cache record is preserved (with its old timestamp).
322 */
323
324 HDEBUG(D_verify) debug_printf("callout cache: domain accepts RCPT "
325 "TO:<postmaster@domain>\n");
326 pm_mailfrom = NULL;
327 new_domain_record.postmaster_result = ccache_accept;
328 new_domain_record.postmaster_stamp = cache_record->postmaster_stamp;
329 }
330 }
331
332 /* We can't give a result based on information about the domain. See if there
333 is an unexpired cache record for this specific address (combined with the
334 sender address if we are doing a recipient callout with a non-empty sender).
335 */
336
337 cache_address_record = (dbdata_callout_cache_address *)
338 get_callout_cache_record(dbm_file,
339 address_key, US"address",
340 callout_cache_positive_expire,
341 callout_cache_negative_expire);
342
343 if (cache_address_record != NULL)
344 {
345 if (cache_address_record->result == ccache_accept)
346 {
347 HDEBUG(D_verify)
348 debug_printf("callout cache: address record is positive\n");
349 }
350 else
351 {
352 HDEBUG(D_verify)
353 debug_printf("callout cache: address record is negative\n");
354 addr->user_message = US"Previous (cached) callout verification failure";
355 *failure_ptr = US"recipient";
356 yield = FAIL;
357 }
358 goto END_CALLOUT;
359 }
360
361 /* Close the cache database while we actually do the callout for real. */
362
363 END_CACHE:
364 dbfn_close(dbm_file);
365 dbm_file = NULL;
366 }
367
368 /* The information wasn't available in the cache, so we have to do a real
369 callout and save the result in the cache for next time, unless no_cache is set,
370 or unless we have a previously cached negative random result. If we are to test
371 with a random local part, ensure that such a local part is available. If not,
372 log the fact, but carry on without randomming. */
373
374 if (callout_random && callout_random_local_part != NULL)
375 {
376 random_local_part = expand_string(callout_random_local_part);
377 if (random_local_part == NULL)
378 log_write(0, LOG_MAIN|LOG_PANIC, "failed to expand "
379 "callout_random_local_part: %s", expand_string_message);
380 }
381
382 /* Default the connect and overall callout timeouts if not set, and record the
383 time we are starting so that we can enforce it. */
384
385 if (callout_overall < 0) callout_overall = 4 * callout;
386 if (callout_connect < 0) callout_connect = callout;
387 callout_start_time = time(NULL);
388
389 /* Before doing a real callout, if this is an SMTP connection, flush the SMTP
390 output because a callout might take some time. When PIPELINING is active and
391 there are many recipients, the total time for doing lots of callouts can add up
392 and cause the client to time out. So in this case we forgo the PIPELINING
393 optimization. */
394
395 if (smtp_out != NULL && !disable_callout_flush) mac_smtp_fflush();
396
397 /* Now make connections to the hosts and do real callouts. The list of hosts
398 is passed in as an argument. */
399
400 for (host = host_list; host != NULL && !done; host = host->next)
401 {
402 smtp_inblock inblock;
403 smtp_outblock outblock;
404 int host_af;
405 int port = 25;
406 BOOL send_quit = TRUE;
407 uschar *active_hostname = smtp_active_hostname;
408 uschar *helo = US"HELO";
409 uschar *interface = NULL; /* Outgoing interface to use; NULL => any */
410 uschar inbuffer[4096];
411 uschar outbuffer[1024];
412 uschar responsebuffer[4096];
413
414 clearflag(addr, af_verify_pmfail); /* postmaster callout flag */
415 clearflag(addr, af_verify_nsfail); /* null sender callout flag */
416
417 /* Skip this host if we don't have an IP address for it. */
418
419 if (host->address == NULL)
420 {
421 DEBUG(D_verify) debug_printf("no IP address for host name %s: skipping\n",
422 host->name);
423 continue;
424 }
425
426 /* Check the overall callout timeout */
427
428 if (time(NULL) - callout_start_time >= callout_overall)
429 {
430 HDEBUG(D_verify) debug_printf("overall timeout for callout exceeded\n");
431 break;
432 }
433
434 /* Set IPv4 or IPv6 */
435
436 host_af = (Ustrchr(host->address, ':') == NULL)? AF_INET:AF_INET6;
437
438 /* Expand and interpret the interface and port strings. The latter will not
439 be used if there is a host-specific port (e.g. from a manualroute router).
440 This has to be delayed till now, because they may expand differently for
441 different hosts. If there's a failure, log it, but carry on with the
442 defaults. */
443
444 deliver_host = host->name;
445 deliver_host_address = host->address;
446 deliver_domain = addr->domain;
447
448 if (!smtp_get_interface(tf->interface, host_af, addr, NULL, &interface,
449 US"callout") ||
450 !smtp_get_port(tf->port, addr, &port, US"callout"))
451 log_write(0, LOG_MAIN|LOG_PANIC, "<%s>: %s", addr->address,
452 addr->message);
453
454 /* Set HELO string according to the protocol */
455
456 if (Ustrcmp(tf->protocol, "lmtp") == 0) helo = US"LHLO";
457
458 HDEBUG(D_verify) debug_printf("interface=%s port=%d\n", interface, port);
459
460 /* Set up the buffer for reading SMTP response packets. */
461
462 inblock.buffer = inbuffer;
463 inblock.buffersize = sizeof(inbuffer);
464 inblock.ptr = inbuffer;
465 inblock.ptrend = inbuffer;
466
467 /* Set up the buffer for holding SMTP commands while pipelining */
468
469 outblock.buffer = outbuffer;
470 outblock.buffersize = sizeof(outbuffer);
471 outblock.ptr = outbuffer;
472 outblock.cmd_count = 0;
473 outblock.authenticating = FALSE;
474
475 /* Connect to the host; on failure, just loop for the next one, but we
476 set the error for the last one. Use the callout_connect timeout. */
477
478 inblock.sock = outblock.sock =
479 smtp_connect(host, host_af, port, interface, callout_connect, TRUE, NULL);
480 /* reconsider DSCP here */
481 if (inblock.sock < 0)
482 {
483 addr->message = string_sprintf("could not connect to %s [%s]: %s",
484 host->name, host->address, strerror(errno));
485 deliver_host = deliver_host_address = NULL;
486 deliver_domain = save_deliver_domain;
487 continue;
488 }
489
490 /* Expand the helo_data string to find the host name to use. */
491
492 if (tf->helo_data != NULL)
493 {
494 uschar *s = expand_string(tf->helo_data);
495 if (s == NULL)
496 log_write(0, LOG_MAIN|LOG_PANIC, "<%s>: failed to expand transport's "
497 "helo_data value for callout: %s", addr->address,
498 expand_string_message);
499 else active_hostname = s;
500 }
501
502 deliver_host = deliver_host_address = NULL;
503 deliver_domain = save_deliver_domain;
504
505 /* Wait for initial response, and send HELO. The smtp_write_command()
506 function leaves its command in big_buffer. This is used in error responses.
507 Initialize it in case the connection is rejected. */
508
509 Ustrcpy(big_buffer, "initial connection");
510
511 done =
512 smtp_read_response(&inblock, responsebuffer, sizeof(responsebuffer),
513 '2', callout) &&
514 smtp_write_command(&outblock, FALSE, "%s %s\r\n", helo,
515 active_hostname) >= 0 &&
516 smtp_read_response(&inblock, responsebuffer, sizeof(responsebuffer),
517 '2', callout);
518
519 /* Failure to accept HELO is cached; this blocks the whole domain for all
520 senders. I/O errors and defer responses are not cached. */
521
522 if (!done)
523 {
524 *failure_ptr = US"mail"; /* At or before MAIL */
525 if (errno == 0 && responsebuffer[0] == '5')
526 {
527 setflag(addr, af_verify_nsfail);
528 new_domain_record.result = ccache_reject;
529 }
530 }
531
532 /* Send the MAIL command */
533
534 else done =
535 smtp_write_command(&outblock, FALSE, "MAIL FROM:<%s>\r\n",
536 from_address) >= 0 &&
537 smtp_read_response(&inblock, responsebuffer, sizeof(responsebuffer),
538 '2', callout);
539
540 /* If the host does not accept MAIL FROM:<>, arrange to cache this
541 information, but again, don't record anything for an I/O error or a defer. Do
542 not cache rejections of MAIL when a non-empty sender has been used, because
543 that blocks the whole domain for all senders. */
544
545 if (!done)
546 {
547 *failure_ptr = US"mail"; /* At or before MAIL */
548 if (errno == 0 && responsebuffer[0] == '5')
549 {
550 setflag(addr, af_verify_nsfail);
551 if (from_address[0] == 0)
552 new_domain_record.result = ccache_reject_mfnull;
553 }
554 }
555
556 /* Otherwise, proceed to check a "random" address (if required), then the
557 given address, and the postmaster address (if required). Between each check,
558 issue RSET, because some servers accept only one recipient after MAIL
559 FROM:<>.
560
561 Before doing this, set the result in the domain cache record to "accept",
562 unless its previous value was ccache_reject_mfnull. In that case, the domain
563 rejects MAIL FROM:<> and we want to continue to remember that. When that is
564 the case, we have got here only in the case of a recipient verification with
565 a non-null sender. */
566
567 else
568 {
569 new_domain_record.result =
570 (old_domain_cache_result == ccache_reject_mfnull)?
571 ccache_reject_mfnull: ccache_accept;
572
573 /* Do the random local part check first */
574
575 if (random_local_part != NULL)
576 {
577 uschar randombuffer[1024];
578 BOOL random_ok =
579 smtp_write_command(&outblock, FALSE,
580 "RCPT TO:<%.1000s@%.1000s>\r\n", random_local_part,
581 addr->domain) >= 0 &&
582 smtp_read_response(&inblock, randombuffer,
583 sizeof(randombuffer), '2', callout);
584
585 /* Remember when we last did a random test */
586
587 new_domain_record.random_stamp = time(NULL);
588
589 /* If accepted, we aren't going to do any further tests below. */
590
591 if (random_ok)
592 {
593 new_domain_record.random_result = ccache_accept;
594 }
595
596 /* Otherwise, cache a real negative response, and get back to the right
597 state to send RCPT. Unless there's some problem such as a dropped
598 connection, we expect to succeed, because the commands succeeded above. */
599
600 else if (errno == 0)
601 {
602 if (randombuffer[0] == '5')
603 new_domain_record.random_result = ccache_reject;
604
605 done =
606 smtp_write_command(&outblock, FALSE, "RSET\r\n") >= 0 &&
607 smtp_read_response(&inblock, responsebuffer, sizeof(responsebuffer),
608 '2', callout) &&
609
610 smtp_write_command(&outblock, FALSE, "MAIL FROM:<%s>\r\n",
611 from_address) >= 0 &&
612 smtp_read_response(&inblock, responsebuffer, sizeof(responsebuffer),
613 '2', callout);
614 }
615 else done = FALSE; /* Some timeout/connection problem */
616 } /* Random check */
617
618 /* If the host is accepting all local parts, as determined by the "random"
619 check, we don't need to waste time doing any further checking. */
620
621 if (new_domain_record.random_result != ccache_accept && done)
622 {
623 /* Get the rcpt_include_affixes flag from the transport if there is one,
624 but assume FALSE if there is not. */
625
626 done =
627 smtp_write_command(&outblock, FALSE, "RCPT TO:<%.1000s>\r\n",
628 transport_rcpt_address(addr,
629 (addr->transport == NULL)? FALSE :
630 addr->transport->rcpt_include_affixes)) >= 0 &&
631 smtp_read_response(&inblock, responsebuffer, sizeof(responsebuffer),
632 '2', callout);
633
634 if (done)
635 new_address_record.result = ccache_accept;
636 else if (errno == 0 && responsebuffer[0] == '5')
637 {
638 *failure_ptr = US"recipient";
639 new_address_record.result = ccache_reject;
640 }
641
642 /* Do postmaster check if requested; if a full check is required, we
643 check for RCPT TO:<postmaster> (no domain) in accordance with RFC 821. */
644
645 if (done && pm_mailfrom != NULL)
646 {
647 /*XXX not suitable for cutthrough - sequencing problems */
648 cutthrough_delivery= FALSE;
649
650 done =
651 smtp_write_command(&outblock, FALSE, "RSET\r\n") >= 0 &&
652 smtp_read_response(&inblock, responsebuffer,
653 sizeof(responsebuffer), '2', callout) &&
654
655 smtp_write_command(&outblock, FALSE,
656 "MAIL FROM:<%s>\r\n", pm_mailfrom) >= 0 &&
657 smtp_read_response(&inblock, responsebuffer,
658 sizeof(responsebuffer), '2', callout) &&
659
660 /* First try using the current domain */
661
662 ((
663 smtp_write_command(&outblock, FALSE,
664 "RCPT TO:<postmaster@%.1000s>\r\n", addr->domain) >= 0 &&
665 smtp_read_response(&inblock, responsebuffer,
666 sizeof(responsebuffer), '2', callout)
667 )
668
669 ||
670
671 /* If that doesn't work, and a full check is requested,
672 try without the domain. */
673
674 (
675 (options & vopt_callout_fullpm) != 0 &&
676 smtp_write_command(&outblock, FALSE,
677 "RCPT TO:<postmaster>\r\n") >= 0 &&
678 smtp_read_response(&inblock, responsebuffer,
679 sizeof(responsebuffer), '2', callout)
680 ));
681
682 /* Sort out the cache record */
683
684 new_domain_record.postmaster_stamp = time(NULL);
685
686 if (done)
687 new_domain_record.postmaster_result = ccache_accept;
688 else if (errno == 0 && responsebuffer[0] == '5')
689 {
690 *failure_ptr = US"postmaster";
691 setflag(addr, af_verify_pmfail);
692 new_domain_record.postmaster_result = ccache_reject;
693 }
694 }
695 } /* Random not accepted */
696 } /* MAIL FROM: accepted */
697
698 /* For any failure of the main check, other than a negative response, we just
699 close the connection and carry on. We can identify a negative response by the
700 fact that errno is zero. For I/O errors it will be non-zero
701
702 Set up different error texts for logging and for sending back to the caller
703 as an SMTP response. Log in all cases, using a one-line format. For sender
704 callouts, give a full response to the caller, but for recipient callouts,
705 don't give the IP address because this may be an internal host whose identity
706 is not to be widely broadcast. */
707
708 if (!done)
709 {
710 if (errno == ETIMEDOUT)
711 {
712 HDEBUG(D_verify) debug_printf("SMTP timeout\n");
713 send_quit = FALSE;
714 }
715 else if (errno == 0)
716 {
717 if (*responsebuffer == 0) Ustrcpy(responsebuffer, US"connection dropped");
718
719 addr->message =
720 string_sprintf("response to \"%s\" from %s [%s] was: %s",
721 big_buffer, host->name, host->address,
722 string_printing(responsebuffer));
723
724 addr->user_message = is_recipient?
725 string_sprintf("Callout verification failed:\n%s", responsebuffer)
726 :
727 string_sprintf("Called: %s\nSent: %s\nResponse: %s",
728 host->address, big_buffer, responsebuffer);
729
730 /* Hard rejection ends the process */
731
732 if (responsebuffer[0] == '5') /* Address rejected */
733 {
734 yield = FAIL;
735 done = TRUE;
736 }
737 }
738 }
739
740 /* End the SMTP conversation and close the connection. */
741
742 /*XXX cutthrough - if "done"
743 and "yeild" is OK
744 and we have no cutthrough conn so far
745 here is where we want to leave the conn open */
746 /* and leave some form of marker for it */
747 /*XXX in fact for simplicity we should abandon cutthrough as soon as more than one address
748 comes into play */
749 /*XXX what about TLS? */
750 if ( cutthrough_delivery
751 && done
752 && yield == OK
753 && cutthrough_fd < 0
754 && (options & (vopt_callout_recipsender|vopt_callout_recippmaster)) == vopt_callout_recipsender
755 && !random_local_part
756 && !pm_mailfrom
757 )
758 {
759 cutthrough_fd= outblock.sock; /* We assume no buffer in use in the outblock */
760 cutthrough_addr= *addr; /* Save the address_item for later logging */
761 }
762 else
763 {
764 if (send_quit) (void)smtp_write_command(&outblock, FALSE, "QUIT\r\n");
765 (void)close(inblock.sock);
766 }
767
768 } /* Loop through all hosts, while !done */
769
770 /* If we get here with done == TRUE, a successful callout happened, and yield
771 will be set OK or FAIL according to the response to the RCPT command.
772 Otherwise, we looped through the hosts but couldn't complete the business.
773 However, there may be domain-specific information to cache in both cases.
774
775 The value of the result field in the new_domain record is ccache_unknown if
776 there was an error before or with MAIL FROM:, and errno was not zero,
777 implying some kind of I/O error. We don't want to write the cache in that case.
778 Otherwise the value is ccache_accept, ccache_reject, or ccache_reject_mfnull. */
779
780 if (!callout_no_cache && new_domain_record.result != ccache_unknown)
781 {
782 if ((dbm_file = dbfn_open(US"callout", O_RDWR|O_CREAT, &dbblock, FALSE))
783 == NULL)
784 {
785 HDEBUG(D_verify) debug_printf("callout cache: not available\n");
786 }
787 else
788 {
789 (void)dbfn_write(dbm_file, addr->domain, &new_domain_record,
790 (int)sizeof(dbdata_callout_cache));
791 HDEBUG(D_verify) debug_printf("wrote callout cache domain record:\n"
792 " result=%d postmaster=%d random=%d\n",
793 new_domain_record.result,
794 new_domain_record.postmaster_result,
795 new_domain_record.random_result);
796 }
797 }
798
799 /* If a definite result was obtained for the callout, cache it unless caching
800 is disabled. */
801
802 if (done)
803 {
804 if (!callout_no_cache && new_address_record.result != ccache_unknown)
805 {
806 if (dbm_file == NULL)
807 dbm_file = dbfn_open(US"callout", O_RDWR|O_CREAT, &dbblock, FALSE);
808 if (dbm_file == NULL)
809 {
810 HDEBUG(D_verify) debug_printf("no callout cache available\n");
811 }
812 else
813 {
814 (void)dbfn_write(dbm_file, address_key, &new_address_record,
815 (int)sizeof(dbdata_callout_cache_address));
816 HDEBUG(D_verify) debug_printf("wrote %s callout cache address record\n",
817 (new_address_record.result == ccache_accept)? "positive" : "negative");
818 }
819 }
820 } /* done */
821
822 /* Failure to connect to any host, or any response other than 2xx or 5xx is a
823 temporary error. If there was only one host, and a response was received, leave
824 it alone if supplying details. Otherwise, give a generic response. */
825
826 else /* !done */
827 {
828 uschar *dullmsg = string_sprintf("Could not complete %s verify callout",
829 is_recipient? "recipient" : "sender");
830 yield = DEFER;
831
832 if (host_list->next != NULL || addr->message == NULL) addr->message = dullmsg;
833
834 addr->user_message = (!smtp_return_error_details)? dullmsg :
835 string_sprintf("%s for <%s>.\n"
836 "The mail server(s) for the domain may be temporarily unreachable, or\n"
837 "they may be permanently unreachable from this server. In the latter case,\n%s",
838 dullmsg, addr->address,
839 is_recipient?
840 "the address will never be accepted."
841 :
842 "you need to change the address or create an MX record for its domain\n"
843 "if it is supposed to be generally accessible from the Internet.\n"
844 "Talk to your mail administrator for details.");
845
846 /* Force a specific error code */
847
848 addr->basic_errno = ERRNO_CALLOUTDEFER;
849 }
850
851 /* Come here from within the cache-reading code on fast-track exit. */
852
853 END_CALLOUT:
854 if (dbm_file != NULL) dbfn_close(dbm_file);
855 return yield;
856 }
857
858
859
860 void
861 open_cutthrough_connection( address_item * addr )
862 {
863 address_item addr2;
864
865 /* Use a recipient-verify-callout to set up the cutthrough connection. */
866 /* We must use a copy of the address for verification, because it might
867 get rewritten. */
868
869 addr2 = *addr;
870 HDEBUG(D_acl) debug_printf("----------- start cutthrough setup ------------\n");
871 (void) verify_address(&addr2, NULL,
872 vopt_is_recipient | vopt_callout_recipsender | vopt_callout_no_cache,
873 CUTTHROUGH_CMD_TIMEOUT, -1, -1,
874 NULL, NULL, NULL);
875 HDEBUG(D_acl) debug_printf("----------- end cutthrough setup ------------\n");
876 return;
877 }
878
879
880 static smtp_outblock ctblock;
881 uschar ctbuffer[8192];
882
883
884 void
885 cancel_cutthrough_connection( void )
886 {
887 ctblock.ptr = ctbuffer;
888 cutthrough_delivery= FALSE;
889 if(cutthrough_fd >= 0) /*XXX get that initialised, also at RSET */
890 {
891 int rc;
892
893 /* We could be sending this after a bunch of data, but that is ok as
894 the only way to cancel the transfer in dataphase is to drop the tcp
895 conn before the final dot.
896 */
897 HDEBUG(D_transport|D_acl|D_v) debug_printf(" SMTP>> QUIT\n");
898 rc= send(cutthrough_fd, "QUIT\r\n", 6, 0);
899 /*XXX error handling? TLS? See flush_buffer() in smtp_out.c */
900
901 (void)close(cutthrough_fd);
902 cutthrough_fd= -1;
903 HDEBUG(D_acl) debug_printf("----------- cutthrough shutdown ------------\n");
904 }
905 }
906
907
908
909 /* Buffered output counted data block. Return boolean success */
910 BOOL
911 cutthrough_puts(uschar * cp, int n)
912 {
913 if(cutthrough_fd >= 0)
914 while(n--)
915 {
916 /*XXX TLS? See flush_buffer() in smtp_out.c */
917
918 if(ctblock.ptr >= ctblock.buffer+ctblock.buffersize)
919 {
920 if(send(cutthrough_fd, ctblock.buffer, ctblock.buffersize, 0) < 0)
921 goto bad;
922 transport_count += ctblock.buffersize;
923 ctblock.ptr= ctblock.buffer;
924 }
925
926 *ctblock.ptr++ = *cp++;
927 }
928 return TRUE;
929
930 bad:
931 cancel_cutthrough_connection();
932 return FALSE;
933 }
934
935 BOOL
936 cutthrough_flush_send( void )
937 {
938 if(cutthrough_fd >= 0)
939 {
940 if(send(cutthrough_fd, ctblock.buffer, ctblock.ptr-ctblock.buffer, 0) < 0)
941 goto bad;
942 transport_count += ctblock.ptr-ctblock.buffer;
943 ctblock.ptr= ctblock.buffer;
944 }
945 return TRUE;
946
947 bad:
948 cancel_cutthrough_connection();
949 return FALSE;
950 }
951
952
953 BOOL
954 cutthrough_put_nl( void )
955 {
956 return cutthrough_puts(US"\r\n", 2);
957 }
958
959
960 /* Get and check response from cutthrough target */
961 static uschar
962 cutthrough_response(char expect, uschar ** copy)
963 {
964 smtp_inblock inblock;
965 uschar inbuffer[4096];
966 uschar responsebuffer[4096];
967
968 inblock.buffer = inbuffer;
969 inblock.buffersize = sizeof(inbuffer);
970 inblock.ptr = inbuffer;
971 inblock.ptrend = inbuffer;
972 inblock.sock = cutthrough_fd;
973 if(!smtp_read_response(&inblock, responsebuffer, sizeof(responsebuffer), expect, CUTTHROUGH_DATA_TIMEOUT))
974 cancel_cutthrough_connection();
975
976 if(copy != NULL)
977 {
978 uschar * cp;
979 *copy= cp= string_copy(responsebuffer);
980 /* Trim the trailing end of line */
981 cp += Ustrlen(responsebuffer);
982 if(cp > *copy && cp[-1] == '\n') *--cp = '\0';
983 if(cp > *copy && cp[-1] == '\r') *--cp = '\0';
984 }
985
986 return responsebuffer[0];
987 }
988
989
990 /* Negotiate dataphase with the cutthrough target, returning success boolean */
991 BOOL
992 cutthrough_predata( void )
993 {
994 int rc;
995
996 if(cutthrough_fd < 0)
997 return FALSE;
998
999 HDEBUG(D_transport|D_acl|D_v) debug_printf(" SMTP>> DATA\n");
1000 rc= send(cutthrough_fd, "DATA\r\n", 6, 0);
1001 if (rc <= 0)
1002 {
1003 HDEBUG(D_transport|D_acl) debug_printf("send failed: %s\n", strerror(errno));
1004 cancel_cutthrough_connection();
1005 return FALSE;
1006 }
1007 /*XXX error handling? TLS? See flush_buffer() in smtp_out.c */
1008
1009 /* Assume nothing buffered. If it was it gets ignored. */
1010 return cutthrough_response('3', NULL) == '3';
1011 }
1012
1013
1014 /* Buffered send of headers. Return success boolean. */
1015 /* Also sends header-terminating blank line. */
1016 /* Sets up the "ctblock" buffer as a side-effect. */
1017 BOOL
1018 cutthrough_headers_send( void )
1019 {
1020 header_line * h;
1021
1022 if(cutthrough_fd < 0)
1023 return FALSE;
1024
1025 ctblock.buffer = ctbuffer;
1026 ctblock.buffersize = sizeof(ctbuffer);
1027 ctblock.ptr = ctbuffer;
1028 /* ctblock.cmd_count = 0; ctblock.authenticating = FALSE; */
1029 ctblock.sock = cutthrough_fd;
1030
1031 for(h= header_list; h != NULL; h= h->next)
1032 if(h->type != htype_old && h->text != NULL)
1033 if(!cutthrough_puts(h->text, h->slen))
1034 return FALSE;
1035
1036 if(!cutthrough_put_nl())
1037 return TRUE;
1038 }
1039
1040
1041 /* Have senders final-dot. Send one to cutthrough target, and grab the response.
1042 Log an OK response as a transmission.
1043 Return smtp response-class digit.
1044 XXX where do fail responses from target get logged?
1045 */
1046 uschar *
1047 cutthrough_finaldot( void )
1048 {
1049 HDEBUG(D_transport|D_acl|D_v) debug_printf(" SMTP>> .\n");
1050
1051 /* Assume data finshed with new-line */
1052 if(!cutthrough_puts(US".", 1) || !cutthrough_put_nl()
1053 || !cutthrough_flush_send()
1054 || cutthrough_response('2', &cutthrough_addr.message) != '2')
1055 return cutthrough_addr.message;
1056
1057 (void)close(cutthrough_fd);
1058 cutthrough_fd= -1;
1059 HDEBUG(D_acl) debug_printf("----------- cutthrough close ------------\n");
1060
1061 delivery_log(&cutthrough_addr, (int)'>');
1062 /* C= ok */
1063 /* QT ok */
1064 /* DT always 0? */
1065 /* delivery S= zero! (transport_count) */
1066 /* not TLS yet hence no X, CV, DN */
1067
1068 return cutthrough_addr.message;
1069 }
1070
1071
1072 /*************************************************
1073 * Copy error to toplevel address *
1074 *************************************************/
1075
1076 /* This function is used when a verify fails or defers, to ensure that the
1077 failure or defer information is in the original toplevel address. This applies
1078 when an address is redirected to a single new address, and the failure or
1079 deferral happens to the child address.
1080
1081 Arguments:
1082 vaddr the verify address item
1083 addr the final address item
1084 yield FAIL or DEFER
1085
1086 Returns: the value of YIELD
1087 */
1088
1089 static int
1090 copy_error(address_item *vaddr, address_item *addr, int yield)
1091 {
1092 if (addr != vaddr)
1093 {
1094 vaddr->message = addr->message;
1095 vaddr->user_message = addr->user_message;
1096 vaddr->basic_errno = addr->basic_errno;
1097 vaddr->more_errno = addr->more_errno;
1098 vaddr->p.address_data = addr->p.address_data;
1099 copyflag(vaddr, addr, af_pass_message);
1100 }
1101 return yield;
1102 }
1103
1104
1105
1106
1107 /**************************************************
1108 * printf that automatically handles TLS if needed *
1109 ***************************************************/
1110
1111 /* This function is used by verify_address() as a substitute for all fprintf()
1112 calls; a direct fprintf() will not produce output in a TLS SMTP session, such
1113 as a response to an EXPN command. smtp_in.c makes smtp_printf available but
1114 that assumes that we always use the smtp_out FILE* when not using TLS or the
1115 ssl buffer when we are. Instead we take a FILE* parameter and check to see if
1116 that is smtp_out; if so, smtp_printf() with TLS support, otherwise regular
1117 fprintf().
1118
1119 Arguments:
1120 f the candidate FILE* to write to
1121 format format string
1122 ... optional arguments
1123
1124 Returns:
1125 nothing
1126 */
1127
1128 static void PRINTF_FUNCTION(2,3)
1129 respond_printf(FILE *f, const char *format, ...)
1130 {
1131 va_list ap;
1132
1133 va_start(ap, format);
1134 if (smtp_out && (f == smtp_out))
1135 smtp_vprintf(format, ap);
1136 else
1137 vfprintf(f, format, ap);
1138 va_end(ap);
1139 }
1140
1141
1142
1143 /*************************************************
1144 * Verify an email address *
1145 *************************************************/
1146
1147 /* This function is used both for verification (-bv and at other times) and
1148 address testing (-bt), which is indicated by address_test_mode being set.
1149
1150 Arguments:
1151 vaddr contains the address to verify; the next field in this block
1152 must be NULL
1153 f if not NULL, write the result to this file
1154 options various option bits:
1155 vopt_fake_sender => this sender verify is not for the real
1156 sender (it was verify=sender=xxxx or an address from a
1157 header line) - rewriting must not change sender_address
1158 vopt_is_recipient => this is a recipient address, otherwise
1159 it's a sender address - this affects qualification and
1160 rewriting and messages from callouts
1161 vopt_qualify => qualify an unqualified address; else error
1162 vopt_expn => called from SMTP EXPN command
1163 vopt_success_on_redirect => when a new address is generated
1164 the verification instantly succeeds
1165
1166 These ones are used by do_callout() -- the options variable
1167 is passed to it.
1168
1169 vopt_callout_fullpm => if postmaster check, do full one
1170 vopt_callout_no_cache => don't use callout cache
1171 vopt_callout_random => do the "random" thing
1172 vopt_callout_recipsender => use real sender for recipient
1173 vopt_callout_recippmaster => use postmaster for recipient
1174
1175 callout if > 0, specifies that callout is required, and gives timeout
1176 for individual commands
1177 callout_overall if > 0, gives overall timeout for the callout function;
1178 if < 0, a default is used (see do_callout())
1179 callout_connect the connection timeout for callouts
1180 se_mailfrom when callout is requested to verify a sender, use this
1181 in MAIL FROM; NULL => ""
1182 pm_mailfrom when callout is requested, if non-NULL, do the postmaster
1183 thing and use this as the sender address (may be "")
1184
1185 routed if not NULL, set TRUE if routing succeeded, so we can
1186 distinguish between routing failed and callout failed
1187
1188 Returns: OK address verified
1189 FAIL address failed to verify
1190 DEFER can't tell at present
1191 */
1192
1193 int
1194 verify_address(address_item *vaddr, FILE *f, int options, int callout,
1195 int callout_overall, int callout_connect, uschar *se_mailfrom,
1196 uschar *pm_mailfrom, BOOL *routed)
1197 {
1198 BOOL allok = TRUE;
1199 BOOL full_info = (f == NULL)? FALSE : (debug_selector != 0);
1200 BOOL is_recipient = (options & vopt_is_recipient) != 0;
1201 BOOL expn = (options & vopt_expn) != 0;
1202 BOOL success_on_redirect = (options & vopt_success_on_redirect) != 0;
1203 int i;
1204 int yield = OK;
1205 int verify_type = expn? v_expn :
1206 address_test_mode? v_none :
1207 is_recipient? v_recipient : v_sender;
1208 address_item *addr_list;
1209 address_item *addr_new = NULL;
1210 address_item *addr_remote = NULL;
1211 address_item *addr_local = NULL;
1212 address_item *addr_succeed = NULL;
1213 uschar **failure_ptr = is_recipient?
1214 &recipient_verify_failure : &sender_verify_failure;
1215 uschar *ko_prefix, *cr;
1216 uschar *address = vaddr->address;
1217 uschar *save_sender;
1218 uschar null_sender[] = { 0 }; /* Ensure writeable memory */
1219
1220 /* Clear, just in case */
1221
1222 *failure_ptr = NULL;
1223
1224 /* Set up a prefix and suffix for error message which allow us to use the same
1225 output statements both in EXPN mode (where an SMTP response is needed) and when
1226 debugging with an output file. */
1227
1228 if (expn)
1229 {
1230 ko_prefix = US"553 ";
1231 cr = US"\r";
1232 }
1233 else ko_prefix = cr = US"";
1234
1235 /* Add qualify domain if permitted; otherwise an unqualified address fails. */
1236
1237 if (parse_find_at(address) == NULL)
1238 {
1239 if ((options & vopt_qualify) == 0)
1240 {
1241 if (f != NULL)
1242 respond_printf(f, "%sA domain is required for \"%s\"%s\n",
1243 ko_prefix, address, cr);
1244 *failure_ptr = US"qualify";
1245 return FAIL;
1246 }
1247 address = rewrite_address_qualify(address, is_recipient);
1248 }
1249
1250 DEBUG(D_verify)
1251 {
1252 debug_printf(">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>\n");
1253 debug_printf("%s %s\n", address_test_mode? "Testing" : "Verifying", address);
1254 }
1255
1256 /* Rewrite and report on it. Clear the domain and local part caches - these
1257 may have been set by domains and local part tests during an ACL. */
1258
1259 if (global_rewrite_rules != NULL)
1260 {
1261 uschar *old = address;
1262 address = rewrite_address(address, is_recipient, FALSE,
1263 global_rewrite_rules, rewrite_existflags);
1264 if (address != old)
1265 {
1266 for (i = 0; i < (MAX_NAMED_LIST * 2)/32; i++) vaddr->localpart_cache[i] = 0;
1267 for (i = 0; i < (MAX_NAMED_LIST * 2)/32; i++) vaddr->domain_cache[i] = 0;
1268 if (f != NULL && !expn) fprintf(f, "Address rewritten as: %s\n", address);
1269 }
1270 }
1271
1272 /* If this is the real sender address, we must update sender_address at
1273 this point, because it may be referred to in the routers. */
1274
1275 if ((options & (vopt_fake_sender|vopt_is_recipient)) == 0)
1276 sender_address = address;
1277
1278 /* If the address was rewritten to <> no verification can be done, and we have
1279 to return OK. This rewriting is permitted only for sender addresses; for other
1280 addresses, such rewriting fails. */
1281
1282 if (address[0] == 0) return OK;
1283
1284 /* Save a copy of the sender address for re-instating if we change it to <>
1285 while verifying a sender address (a nice bit of self-reference there). */
1286
1287 save_sender = sender_address;
1288
1289 /* Update the address structure with the possibly qualified and rewritten
1290 address. Set it up as the starting address on the chain of new addresses. */
1291
1292 vaddr->address = address;
1293 addr_new = vaddr;
1294
1295 /* We need a loop, because an address can generate new addresses. We must also
1296 cope with generated pipes and files at the top level. (See also the code and
1297 comment in deliver.c.) However, it is usually the case that the router for
1298 user's .forward files has its verify flag turned off.
1299
1300 If an address generates more than one child, the loop is used only when
1301 full_info is set, and this can only be set locally. Remote enquiries just get
1302 information about the top level address, not anything that it generated. */
1303
1304 while (addr_new != NULL)
1305 {
1306 int rc;
1307 address_item *addr = addr_new;
1308
1309 addr_new = addr->next;
1310 addr->next = NULL;
1311
1312 DEBUG(D_verify)
1313 {
1314 debug_printf(">>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>\n");
1315 debug_printf("Considering %s\n", addr->address);
1316 }
1317
1318 /* Handle generated pipe, file or reply addresses. We don't get these
1319 when handling EXPN, as it does only one level of expansion. */
1320
1321 if (testflag(addr, af_pfr))
1322 {
1323 allok = FALSE;
1324 if (f != NULL)
1325 {
1326 BOOL allow;
1327
1328 if (addr->address[0] == '>')
1329 {
1330 allow = testflag(addr, af_allow_reply);
1331 fprintf(f, "%s -> mail %s", addr->parent->address, addr->address + 1);
1332 }
1333 else
1334 {
1335 allow = (addr->address[0] == '|')?
1336 testflag(addr, af_allow_pipe) : testflag(addr, af_allow_file);
1337 fprintf(f, "%s -> %s", addr->parent->address, addr->address);
1338 }
1339
1340 if (addr->basic_errno == ERRNO_BADTRANSPORT)
1341 fprintf(f, "\n*** Error in setting up pipe, file, or autoreply:\n"
1342 "%s\n", addr->message);
1343 else if (allow)
1344 fprintf(f, "\n transport = %s\n", addr->transport->name);
1345 else
1346 fprintf(f, " *** forbidden ***\n");
1347 }
1348 continue;
1349 }
1350
1351 /* Just in case some router parameter refers to it. */
1352
1353 return_path = (addr->p.errors_address != NULL)?
1354 addr->p.errors_address : sender_address;
1355
1356 /* Split the address into domain and local part, handling the %-hack if
1357 necessary, and then route it. While routing a sender address, set
1358 $sender_address to <> because that is what it will be if we were trying to
1359 send a bounce to the sender. */
1360
1361 if (routed != NULL) *routed = FALSE;
1362 if ((rc = deliver_split_address(addr)) == OK)
1363 {
1364 if (!is_recipient) sender_address = null_sender;
1365 rc = route_address(addr, &addr_local, &addr_remote, &addr_new,
1366 &addr_succeed, verify_type);
1367 sender_address = save_sender; /* Put back the real sender */
1368 }
1369
1370 /* If routing an address succeeded, set the flag that remembers, for use when
1371 an ACL cached a sender verify (in case a callout fails). Then if routing set
1372 up a list of hosts or the transport has a host list, and the callout option
1373 is set, and we aren't in a host checking run, do the callout verification,
1374 and set another flag that notes that a callout happened. */
1375
1376 if (rc == OK)
1377 {
1378 if (routed != NULL) *routed = TRUE;
1379 if (callout > 0)
1380 {
1381 host_item *host_list = addr->host_list;
1382
1383 /* Make up some data for use in the case where there is no remote
1384 transport. */
1385
1386 transport_feedback tf = {
1387 NULL, /* interface (=> any) */
1388 US"smtp", /* port */
1389 US"smtp", /* protocol */
1390 NULL, /* hosts */
1391 US"$smtp_active_hostname", /* helo_data */
1392 FALSE, /* hosts_override */
1393 FALSE, /* hosts_randomize */
1394 FALSE, /* gethostbyname */
1395 TRUE, /* qualify_single */
1396 FALSE /* search_parents */
1397 };
1398
1399 /* If verification yielded a remote transport, we want to use that
1400 transport's options, so as to mimic what would happen if we were really
1401 sending a message to this address. */
1402
1403 if (addr->transport != NULL && !addr->transport->info->local)
1404 {
1405 (void)(addr->transport->setup)(addr->transport, addr, &tf, 0, 0, NULL);
1406
1407 /* If the transport has hosts and the router does not, or if the
1408 transport is configured to override the router's hosts, we must build a
1409 host list of the transport's hosts, and find the IP addresses */
1410
1411 if (tf.hosts != NULL && (host_list == NULL || tf.hosts_override))
1412 {
1413 uschar *s;
1414 uschar *save_deliver_domain = deliver_domain;
1415 uschar *save_deliver_localpart = deliver_localpart;
1416
1417 host_list = NULL; /* Ignore the router's hosts */
1418
1419 deliver_domain = addr->domain;
1420 deliver_localpart = addr->local_part;
1421 s = expand_string(tf.hosts);
1422 deliver_domain = save_deliver_domain;
1423 deliver_localpart = save_deliver_localpart;
1424
1425 if (s == NULL)
1426 {
1427 log_write(0, LOG_MAIN|LOG_PANIC, "failed to expand list of hosts "
1428 "\"%s\" in %s transport for callout: %s", tf.hosts,
1429 addr->transport->name, expand_string_message);
1430 }
1431 else
1432 {
1433 int flags;
1434 uschar *canonical_name;
1435 host_item *host, *nexthost;
1436 host_build_hostlist(&host_list, s, tf.hosts_randomize);
1437
1438 /* Just ignore failures to find a host address. If we don't manage
1439 to find any addresses, the callout will defer. Note that more than
1440 one address may be found for a single host, which will result in
1441 additional host items being inserted into the chain. Hence we must
1442 save the next host first. */
1443
1444 flags = HOST_FIND_BY_A;
1445 if (tf.qualify_single) flags |= HOST_FIND_QUALIFY_SINGLE;
1446 if (tf.search_parents) flags |= HOST_FIND_SEARCH_PARENTS;
1447
1448 for (host = host_list; host != NULL; host = nexthost)
1449 {
1450 nexthost = host->next;
1451 if (tf.gethostbyname ||
1452 string_is_ip_address(host->name, NULL) != 0)
1453 (void)host_find_byname(host, NULL, flags, &canonical_name, TRUE);
1454 else
1455 (void)host_find_bydns(host, NULL, flags, NULL, NULL, NULL,
1456 &canonical_name, NULL);
1457 }
1458 }
1459 }
1460 }
1461
1462 /* Can only do a callout if we have at least one host! If the callout
1463 fails, it will have set ${sender,recipient}_verify_failure. */
1464
1465 if (host_list != NULL)
1466 {
1467 HDEBUG(D_verify) debug_printf("Attempting full verification using callout\n");
1468 if (host_checking && !host_checking_callout)
1469 {
1470 HDEBUG(D_verify)
1471 debug_printf("... callout omitted by default when host testing\n"
1472 "(Use -bhc if you want the callouts to happen.)\n");
1473 }
1474 else
1475 {
1476 rc = do_callout(addr, host_list, &tf, callout, callout_overall,
1477 callout_connect, options, se_mailfrom, pm_mailfrom);
1478 }
1479 }
1480 else
1481 {
1482 HDEBUG(D_verify) debug_printf("Cannot do callout: neither router nor "
1483 "transport provided a host list\n");
1484 }
1485 }
1486 }
1487
1488 /* Otherwise, any failure is a routing failure */
1489
1490 else *failure_ptr = US"route";
1491
1492 /* A router may return REROUTED if it has set up a child address as a result
1493 of a change of domain name (typically from widening). In this case we always
1494 want to continue to verify the new child. */
1495
1496 if (rc == REROUTED) continue;
1497
1498 /* Handle hard failures */
1499
1500 if (rc == FAIL)
1501 {
1502 allok = FALSE;
1503 if (f != NULL)
1504 {
1505 address_item *p = addr->parent;
1506
1507 respond_printf(f, "%s%s %s", ko_prefix,
1508 full_info? addr->address : address,
1509 address_test_mode? "is undeliverable" : "failed to verify");
1510 if (!expn && admin_user)
1511 {
1512 if (addr->basic_errno > 0)
1513 respond_printf(f, ": %s", strerror(addr->basic_errno));
1514 if (addr->message != NULL)
1515 respond_printf(f, ": %s", addr->message);
1516 }
1517
1518 /* Show parents iff doing full info */
1519
1520 if (full_info) while (p != NULL)
1521 {
1522 respond_printf(f, "%s\n <-- %s", cr, p->address);
1523 p = p->parent;
1524 }
1525 respond_printf(f, "%s\n", cr);
1526 }
1527 cancel_cutthrough_connection();
1528
1529 if (!full_info) return copy_error(vaddr, addr, FAIL);
1530 else yield = FAIL;
1531 }
1532
1533 /* Soft failure */
1534
1535 else if (rc == DEFER)
1536 {
1537 allok = FALSE;
1538 if (f != NULL)
1539 {
1540 address_item *p = addr->parent;
1541 respond_printf(f, "%s%s cannot be resolved at this time", ko_prefix,
1542 full_info? addr->address : address);
1543 if (!expn && admin_user)
1544 {
1545 if (addr->basic_errno > 0)
1546 respond_printf(f, ": %s", strerror(addr->basic_errno));
1547 if (addr->message != NULL)
1548 respond_printf(f, ": %s", addr->message);
1549 else if (addr->basic_errno <= 0)
1550 respond_printf(f, ": unknown error");
1551 }
1552
1553 /* Show parents iff doing full info */
1554
1555 if (full_info) while (p != NULL)
1556 {
1557 respond_printf(f, "%s\n <-- %s", cr, p->address);
1558 p = p->parent;
1559 }
1560 respond_printf(f, "%s\n", cr);
1561 }
1562 cancel_cutthrough_connection();
1563
1564 if (!full_info) return copy_error(vaddr, addr, DEFER);
1565 else if (yield == OK) yield = DEFER;
1566 }
1567
1568 /* If we are handling EXPN, we do not want to continue to route beyond
1569 the top level (whose address is in "address"). */
1570
1571 else if (expn)
1572 {
1573 uschar *ok_prefix = US"250-";
1574 if (addr_new == NULL)
1575 {
1576 if (addr_local == NULL && addr_remote == NULL)
1577 respond_printf(f, "250 mail to <%s> is discarded\r\n", address);
1578 else
1579 respond_printf(f, "250 <%s>\r\n", address);
1580 }
1581 else while (addr_new != NULL)
1582 {
1583 address_item *addr2 = addr_new;
1584 addr_new = addr2->next;
1585 if (addr_new == NULL) ok_prefix = US"250 ";
1586 respond_printf(f, "%s<%s>\r\n", ok_prefix, addr2->address);
1587 }
1588 return OK;
1589 }
1590
1591 /* Successful routing other than EXPN. */
1592
1593 else
1594 {
1595 /* Handle successful routing when short info wanted. Otherwise continue for
1596 other (generated) addresses. Short info is the operational case. Full info
1597 can be requested only when debug_selector != 0 and a file is supplied.
1598
1599 There is a conflict between the use of aliasing as an alternate email
1600 address, and as a sort of mailing list. If an alias turns the incoming
1601 address into just one address (e.g. J.Caesar->jc44) you may well want to
1602 carry on verifying the generated address to ensure it is valid when
1603 checking incoming mail. If aliasing generates multiple addresses, you
1604 probably don't want to do this. Exim therefore treats the generation of
1605 just a single new address as a special case, and continues on to verify the
1606 generated address. */
1607
1608 if (!full_info && /* Stop if short info wanted AND */
1609 (((addr_new == NULL || /* No new address OR */
1610 addr_new->next != NULL || /* More than one new address OR */
1611 testflag(addr_new, af_pfr))) /* New address is pfr */
1612 || /* OR */
1613 (addr_new != NULL && /* At least one new address AND */
1614 success_on_redirect))) /* success_on_redirect is set */
1615 {
1616 if (f != NULL) fprintf(f, "%s %s\n", address,
1617 address_test_mode? "is deliverable" : "verified");
1618
1619 /* If we have carried on to verify a child address, we want the value
1620 of $address_data to be that of the child */
1621
1622 vaddr->p.address_data = addr->p.address_data;
1623 return OK;
1624 }
1625 }
1626 } /* Loop for generated addresses */
1627
1628 /* Display the full results of the successful routing, including any generated
1629 addresses. Control gets here only when full_info is set, which requires f not
1630 to be NULL, and this occurs only when a top-level verify is called with the
1631 debugging switch on.
1632
1633 If there are no local and no remote addresses, and there were no pipes, files,
1634 or autoreplies, and there were no errors or deferments, the message is to be
1635 discarded, usually because of the use of :blackhole: in an alias file. */
1636
1637 if (allok && addr_local == NULL && addr_remote == NULL)
1638 {
1639 fprintf(f, "mail to %s is discarded\n", address);
1640 return yield;
1641 }
1642
1643 for (addr_list = addr_local, i = 0; i < 2; addr_list = addr_remote, i++)
1644 {
1645 while (addr_list != NULL)
1646 {
1647 address_item *addr = addr_list;
1648 address_item *p = addr->parent;
1649 addr_list = addr->next;
1650
1651 fprintf(f, "%s", CS addr->address);
1652 #ifdef EXPERIMENTAL_SRS
1653 if(addr->p.srs_sender)
1654 fprintf(f, " [srs = %s]", addr->p.srs_sender);
1655 #endif
1656
1657 /* If the address is a duplicate, show something about it. */
1658
1659 if (!testflag(addr, af_pfr))
1660 {
1661 tree_node *tnode;
1662 if ((tnode = tree_search(tree_duplicates, addr->unique)) != NULL)
1663 fprintf(f, " [duplicate, would not be delivered]");
1664 else tree_add_duplicate(addr->unique, addr);
1665 }
1666
1667 /* Now show its parents */
1668
1669 while (p != NULL)
1670 {
1671 fprintf(f, "\n <-- %s", p->address);
1672 p = p->parent;
1673 }
1674 fprintf(f, "\n ");
1675
1676 /* Show router, and transport */
1677
1678 fprintf(f, "router = %s, ", addr->router->name);
1679 fprintf(f, "transport = %s\n", (addr->transport == NULL)? US"unset" :
1680 addr->transport->name);
1681
1682 /* Show any hosts that are set up by a router unless the transport
1683 is going to override them; fiddle a bit to get a nice format. */
1684
1685 if (addr->host_list != NULL && addr->transport != NULL &&
1686 !addr->transport->overrides_hosts)
1687 {
1688 host_item *h;
1689 int maxlen = 0;
1690 int maxaddlen = 0;
1691 for (h = addr->host_list; h != NULL; h = h->next)
1692 {
1693 int len = Ustrlen(h->name);
1694 if (len > maxlen) maxlen = len;
1695 len = (h->address != NULL)? Ustrlen(h->address) : 7;
1696 if (len > maxaddlen) maxaddlen = len;
1697 }
1698 for (h = addr->host_list; h != NULL; h = h->next)
1699 {
1700 int len = Ustrlen(h->name);
1701 fprintf(f, " host %s ", h->name);
1702 while (len++ < maxlen) fprintf(f, " ");
1703 if (h->address != NULL)
1704 {
1705 fprintf(f, "[%s] ", h->address);
1706 len = Ustrlen(h->address);
1707 }
1708 else if (!addr->transport->info->local) /* Omit [unknown] for local */
1709 {
1710 fprintf(f, "[unknown] ");
1711 len = 7;
1712 }
1713 else len = -3;
1714 while (len++ < maxaddlen) fprintf(f," ");
1715 if (h->mx >= 0) fprintf(f, "MX=%d", h->mx);
1716 if (h->port != PORT_NONE) fprintf(f, " port=%d", h->port);
1717 if (h->status == hstatus_unusable) fprintf(f, " ** unusable **");
1718 fprintf(f, "\n");
1719 }
1720 }
1721 }
1722 }
1723
1724 /* Will be DEFER or FAIL if any one address has, only for full_info (which is
1725 the -bv or -bt case). */
1726
1727 return yield;
1728 }
1729
1730
1731
1732
1733 /*************************************************
1734 * Check headers for syntax errors *
1735 *************************************************/
1736
1737 /* This function checks those header lines that contain addresses, and verifies
1738 that all the addresses therein are syntactially correct.
1739
1740 Arguments:
1741 msgptr where to put an error message
1742
1743 Returns: OK
1744 FAIL
1745 */
1746
1747 int
1748 verify_check_headers(uschar **msgptr)
1749 {
1750 header_line *h;
1751 uschar *colon, *s;
1752 int yield = OK;
1753
1754 for (h = header_list; h != NULL && yield == OK; h = h->next)
1755 {
1756 if (h->type != htype_from &&
1757 h->type != htype_reply_to &&
1758 h->type != htype_sender &&
1759 h->type != htype_to &&
1760 h->type != htype_cc &&
1761 h->type != htype_bcc)
1762 continue;
1763
1764 colon = Ustrchr(h->text, ':');
1765 s = colon + 1;
1766 while (isspace(*s)) s++;
1767
1768 /* Loop for multiple addresses in the header, enabling group syntax. Note
1769 that we have to reset this after the header has been scanned. */
1770
1771 parse_allow_group = TRUE;
1772
1773 while (*s != 0)
1774 {
1775 uschar *ss = parse_find_address_end(s, FALSE);
1776 uschar *recipient, *errmess;
1777 int terminator = *ss;
1778 int start, end, domain;
1779
1780 /* Temporarily terminate the string at this point, and extract the
1781 operative address within, allowing group syntax. */
1782
1783 *ss = 0;
1784 recipient = parse_extract_address(s,&errmess,&start,&end,&domain,FALSE);
1785 *ss = terminator;
1786
1787 /* Permit an unqualified address only if the message is local, or if the
1788 sending host is configured to be permitted to send them. */
1789
1790 if (recipient != NULL && domain == 0)
1791 {
1792 if (h->type == htype_from || h->type == htype_sender)
1793 {
1794 if (!allow_unqualified_sender) recipient = NULL;
1795 }
1796 else
1797 {
1798 if (!allow_unqualified_recipient) recipient = NULL;
1799 }
1800 if (recipient == NULL) errmess = US"unqualified address not permitted";
1801 }
1802
1803 /* It's an error if no address could be extracted, except for the special
1804 case of an empty address. */
1805
1806 if (recipient == NULL && Ustrcmp(errmess, "empty address") != 0)
1807 {
1808 uschar *verb = US"is";
1809 uschar *t = ss;
1810 uschar *tt = colon;
1811 int len;
1812
1813 /* Arrange not to include any white space at the end in the
1814 error message or the header name. */
1815
1816 while (t > s && isspace(t[-1])) t--;
1817 while (tt > h->text && isspace(tt[-1])) tt--;
1818
1819 /* Add the address that failed to the error message, since in a
1820 header with very many addresses it is sometimes hard to spot
1821 which one is at fault. However, limit the amount of address to
1822 quote - cases have been seen where, for example, a missing double
1823 quote in a humungous To: header creates an "address" that is longer
1824 than string_sprintf can handle. */
1825
1826 len = t - s;
1827 if (len > 1024)
1828 {
1829 len = 1024;
1830 verb = US"begins";
1831 }
1832
1833 *msgptr = string_printing(
1834 string_sprintf("%s: failing address in \"%.*s:\" header %s: %.*s",
1835 errmess, tt - h->text, h->text, verb, len, s));
1836
1837 yield = FAIL;
1838 break; /* Out of address loop */
1839 }
1840
1841 /* Advance to the next address */
1842
1843 s = ss + (terminator? 1:0);
1844 while (isspace(*s)) s++;
1845 } /* Next address */
1846
1847 parse_allow_group = FALSE;
1848 parse_found_group = FALSE;
1849 } /* Next header unless yield has been set FALSE */
1850
1851 return yield;
1852 }
1853
1854
1855
1856 /*************************************************
1857 * Check for blind recipients *
1858 *************************************************/
1859
1860 /* This function checks that every (envelope) recipient is mentioned in either
1861 the To: or Cc: header lines, thus detecting blind carbon copies.
1862
1863 There are two ways of scanning that could be used: either scan the header lines
1864 and tick off the recipients, or scan the recipients and check the header lines.
1865 The original proposed patch did the former, but I have chosen to do the latter,
1866 because (a) it requires no memory and (b) will use fewer resources when there
1867 are many addresses in To: and/or Cc: and only one or two envelope recipients.
1868
1869 Arguments: none
1870 Returns: OK if there are no blind recipients
1871 FAIL if there is at least one blind recipient
1872 */
1873
1874 int
1875 verify_check_notblind(void)
1876 {
1877 int i;
1878 for (i = 0; i < recipients_count; i++)
1879 {
1880 header_line *h;
1881 BOOL found = FALSE;
1882 uschar *address = recipients_list[i].address;
1883
1884 for (h = header_list; !found && h != NULL; h = h->next)
1885 {
1886 uschar *colon, *s;
1887
1888 if (h->type != htype_to && h->type != htype_cc) continue;
1889
1890 colon = Ustrchr(h->text, ':');
1891 s = colon + 1;
1892 while (isspace(*s)) s++;
1893
1894 /* Loop for multiple addresses in the header, enabling group syntax. Note
1895 that we have to reset this after the header has been scanned. */
1896
1897 parse_allow_group = TRUE;
1898
1899 while (*s != 0)
1900 {
1901 uschar *ss = parse_find_address_end(s, FALSE);
1902 uschar *recipient,*errmess;
1903 int terminator = *ss;
1904 int start, end, domain;
1905
1906 /* Temporarily terminate the string at this point, and extract the
1907 operative address within, allowing group syntax. */
1908
1909 *ss = 0;
1910 recipient = parse_extract_address(s,&errmess,&start,&end,&domain,FALSE);
1911 *ss = terminator;
1912
1913 /* If we found a valid recipient that has a domain, compare it with the
1914 envelope recipient. Local parts are compared case-sensitively, domains
1915 case-insensitively. By comparing from the start with length "domain", we
1916 include the "@" at the end, which ensures that we are comparing the whole
1917 local part of each address. */
1918
1919 if (recipient != NULL && domain != 0)
1920 {
1921 found = Ustrncmp(recipient, address, domain) == 0 &&
1922 strcmpic(recipient + domain, address + domain) == 0;
1923 if (found) break;
1924 }
1925
1926 /* Advance to the next address */
1927
1928 s = ss + (terminator? 1:0);
1929 while (isspace(*s)) s++;
1930 } /* Next address */
1931
1932 parse_allow_group = FALSE;
1933 parse_found_group = FALSE;
1934 } /* Next header (if found is false) */
1935
1936 if (!found) return FAIL;
1937 } /* Next recipient */
1938
1939 return OK;
1940 }
1941
1942
1943
1944 /*************************************************
1945 * Find if verified sender *
1946 *************************************************/
1947
1948 /* Usually, just a single address is verified as the sender of the message.
1949 However, Exim can be made to verify other addresses as well (often related in
1950 some way), and this is useful in some environments. There may therefore be a
1951 chain of such addresses that have previously been tested. This function finds
1952 whether a given address is on the chain.
1953
1954 Arguments: the address to be verified
1955 Returns: pointer to an address item, or NULL
1956 */
1957
1958 address_item *
1959 verify_checked_sender(uschar *sender)
1960 {
1961 address_item *addr;
1962 for (addr = sender_verified_list; addr != NULL; addr = addr->next)
1963 if (Ustrcmp(sender, addr->address) == 0) break;
1964 return addr;
1965 }
1966
1967
1968
1969
1970
1971 /*************************************************
1972 * Get valid header address *
1973 *************************************************/
1974
1975 /* Scan the originator headers of the message, looking for an address that
1976 verifies successfully. RFC 822 says:
1977
1978 o The "Sender" field mailbox should be sent notices of
1979 any problems in transport or delivery of the original
1980 messages. If there is no "Sender" field, then the
1981 "From" field mailbox should be used.
1982
1983 o If the "Reply-To" field exists, then the reply should
1984 go to the addresses indicated in that field and not to
1985 the address(es) indicated in the "From" field.
1986
1987 So we check a Sender field if there is one, else a Reply_to field, else a From
1988 field. As some strange messages may have more than one of these fields,
1989 especially if they are resent- fields, check all of them if there is more than
1990 one.
1991
1992 Arguments:
1993 user_msgptr points to where to put a user error message
1994 log_msgptr points to where to put a log error message
1995 callout timeout for callout check (passed to verify_address())
1996 callout_overall overall callout timeout (ditto)
1997 callout_connect connect callout timeout (ditto)
1998 se_mailfrom mailfrom for verify; NULL => ""
1999 pm_mailfrom sender for pm callout check (passed to verify_address())
2000 options callout options (passed to verify_address())
2001 verrno where to put the address basic_errno
2002
2003 If log_msgptr is set to something without setting user_msgptr, the caller
2004 normally uses log_msgptr for both things.
2005
2006 Returns: result of the verification attempt: OK, FAIL, or DEFER;
2007 FAIL is given if no appropriate headers are found
2008 */
2009
2010 int
2011 verify_check_header_address(uschar **user_msgptr, uschar **log_msgptr,
2012 int callout, int callout_overall, int callout_connect, uschar *se_mailfrom,
2013 uschar *pm_mailfrom, int options, int *verrno)
2014 {
2015 static int header_types[] = { htype_sender, htype_reply_to, htype_from };
2016 BOOL done = FALSE;
2017 int yield = FAIL;
2018 int i;
2019
2020 for (i = 0; i < 3 && !done; i++)
2021 {
2022 header_line *h;
2023 for (h = header_list; h != NULL && !done; h = h->next)
2024 {
2025 int terminator, new_ok;
2026 uschar *s, *ss, *endname;
2027
2028 if (h->type != header_types[i]) continue;
2029 s = endname = Ustrchr(h->text, ':') + 1;
2030
2031 /* Scan the addresses in the header, enabling group syntax. Note that we
2032 have to reset this after the header has been scanned. */
2033
2034 parse_allow_group = TRUE;
2035
2036 while (*s != 0)
2037 {
2038 address_item *vaddr;
2039
2040 while (isspace(*s) || *s == ',') s++;
2041 if (*s == 0) break; /* End of header */
2042
2043 ss = parse_find_address_end(s, FALSE);
2044
2045 /* The terminator is a comma or end of header, but there may be white
2046 space preceding it (including newline for the last address). Move back
2047 past any white space so we can check against any cached envelope sender
2048 address verifications. */
2049
2050 while (isspace(ss[-1])) ss--;
2051 terminator = *ss;
2052 *ss = 0;
2053
2054 HDEBUG(D_verify) debug_printf("verifying %.*s header address %s\n",
2055 (int)(endname - h->text), h->text, s);
2056
2057 /* See if we have already verified this address as an envelope sender,
2058 and if so, use the previous answer. */
2059
2060 vaddr = verify_checked_sender(s);
2061
2062 if (vaddr != NULL && /* Previously checked */
2063 (callout <= 0 || /* No callout needed; OR */
2064 vaddr->special_action > 256)) /* Callout was done */
2065 {
2066 new_ok = vaddr->special_action & 255;
2067 HDEBUG(D_verify) debug_printf("previously checked as envelope sender\n");
2068 *ss = terminator; /* Restore shortened string */
2069 }
2070
2071 /* Otherwise we run the verification now. We must restore the shortened
2072 string before running the verification, so the headers are correct, in
2073 case there is any rewriting. */
2074
2075 else
2076 {
2077 int start, end, domain;
2078 uschar *address = parse_extract_address(s, log_msgptr, &start, &end,
2079 &domain, FALSE);
2080
2081 *ss = terminator;
2082
2083 /* If we found an empty address, just carry on with the next one, but
2084 kill the message. */
2085
2086 if (address == NULL && Ustrcmp(*log_msgptr, "empty address") == 0)
2087 {
2088 *log_msgptr = NULL;
2089 s = ss;
2090 continue;
2091 }
2092
2093 /* If verification failed because of a syntax error, fail this
2094 function, and ensure that the failing address gets added to the error
2095 message. */
2096
2097 if (address == NULL)
2098 {
2099 new_ok = FAIL;
2100 while (ss > s && isspace(ss[-1])) ss--;
2101 *log_msgptr = string_sprintf("syntax error in '%.*s' header when "
2102 "scanning for sender: %s in \"%.*s\"",
2103 endname - h->text, h->text, *log_msgptr, ss - s, s);
2104 yield = FAIL;
2105 done = TRUE;
2106 break;
2107 }
2108
2109 /* Else go ahead with the sender verification. But it isn't *the*
2110 sender of the message, so set vopt_fake_sender to stop sender_address
2111 being replaced after rewriting or qualification. */
2112
2113 else
2114 {
2115 vaddr = deliver_make_addr(address, FALSE);
2116 new_ok = verify_address(vaddr, NULL, options | vopt_fake_sender,
2117 callout, callout_overall, callout_connect, se_mailfrom,
2118 pm_mailfrom, NULL);
2119 }
2120 }
2121
2122 /* We now have the result, either newly found, or cached. If we are
2123 giving out error details, set a specific user error. This means that the
2124 last of these will be returned to the user if all three fail. We do not
2125 set a log message - the generic one below will be used. */
2126
2127 if (new_ok != OK)
2128 {
2129 *verrno = vaddr->basic_errno;
2130 if (smtp_return_error_details)
2131 {
2132 *user_msgptr = string_sprintf("Rejected after DATA: "
2133 "could not verify \"%.*s\" header address\n%s: %s",
2134 endname - h->text, h->text, vaddr->address, vaddr->message);
2135 }
2136 }
2137
2138 /* Success or defer */
2139
2140 if (new_ok == OK)
2141 {
2142 yield = OK;
2143 done = TRUE;
2144 break;
2145 }
2146
2147 if (new_ok == DEFER) yield = DEFER;
2148
2149 /* Move on to any more addresses in the header */
2150
2151 s = ss;
2152 } /* Next address */
2153
2154 parse_allow_group = FALSE;
2155 parse_found_group = FALSE;
2156 } /* Next header, unless done */
2157 } /* Next header type unless done */
2158
2159 if (yield == FAIL && *log_msgptr == NULL)
2160 *log_msgptr = US"there is no valid sender in any header line";
2161
2162 if (yield == DEFER && *log_msgptr == NULL)
2163 *log_msgptr = US"all attempts to verify a sender in a header line deferred";
2164
2165 return yield;
2166 }
2167
2168
2169
2170
2171 /*************************************************
2172 * Get RFC 1413 identification *
2173 *************************************************/
2174
2175 /* Attempt to get an id from the sending machine via the RFC 1413 protocol. If
2176 the timeout is set to zero, then the query is not done. There may also be lists
2177 of hosts and nets which are exempt. To guard against malefactors sending
2178 non-printing characters which could, for example, disrupt a message's headers,
2179 make sure the string consists of printing characters only.
2180
2181 Argument:
2182 port the port to connect to; usually this is IDENT_PORT (113), but when
2183 running in the test harness with -bh a different value is used.
2184
2185 Returns: nothing
2186
2187 Side effect: any received ident value is put in sender_ident (NULL otherwise)
2188 */
2189
2190 void
2191 verify_get_ident(int port)
2192 {
2193 int sock, host_af, qlen;
2194 int received_sender_port, received_interface_port, n;
2195 uschar *p;
2196 uschar buffer[2048];
2197
2198 /* Default is no ident. Check whether we want to do an ident check for this
2199 host. */
2200
2201 sender_ident = NULL;
2202 if (rfc1413_query_timeout <= 0 || verify_check_host(&rfc1413_hosts) != OK)
2203 return;
2204
2205 DEBUG(D_ident) debug_printf("doing ident callback\n");
2206
2207 /* Set up a connection to the ident port of the remote host. Bind the local end
2208 to the incoming interface address. If the sender host address is an IPv6
2209 address, the incoming interface address will also be IPv6. */
2210
2211 host_af = (Ustrchr(sender_host_address, ':') == NULL)? AF_INET : AF_INET6;
2212 sock = ip_socket(SOCK_STREAM, host_af);
2213 if (sock < 0) return;
2214
2215 if (ip_bind(sock, host_af, interface_address, 0) < 0)
2216 {
2217 DEBUG(D_ident) debug_printf("bind socket for ident failed: %s\n",
2218 strerror(errno));
2219 goto END_OFF;
2220 }
2221
2222 if (ip_connect(sock, host_af, sender_host_address, port, rfc1413_query_timeout)
2223 < 0)
2224 {
2225 if (errno == ETIMEDOUT && (log_extra_selector & LX_ident_timeout) != 0)
2226 {
2227 log_write(0, LOG_MAIN, "ident connection to %s timed out",
2228 sender_host_address);
2229 }
2230 else
2231 {
2232 DEBUG(D_ident) debug_printf("ident connection to %s failed: %s\n",
2233 sender_host_address, strerror(errno));
2234 }
2235 goto END_OFF;
2236 }
2237
2238 /* Construct and send the query. */
2239
2240 sprintf(CS buffer, "%d , %d\r\n", sender_host_port, interface_port);
2241 qlen = Ustrlen(buffer);
2242 if (send(sock, buffer, qlen, 0) < 0)
2243 {
2244 DEBUG(D_ident) debug_printf("ident send failed: %s\n", strerror(errno));
2245 goto END_OFF;
2246 }
2247
2248 /* Read a response line. We put it into the rest of the buffer, using several
2249 recv() calls if necessary. */
2250
2251 p = buffer + qlen;
2252
2253 for (;;)
2254 {
2255 uschar *pp;
2256 int count;
2257 int size = sizeof(buffer) - (p - buffer);
2258
2259 if (size <= 0) goto END_OFF; /* Buffer filled without seeing \n. */
2260 count = ip_recv(sock, p, size, rfc1413_query_timeout);
2261 if (count <= 0) goto END_OFF; /* Read error or EOF */
2262
2263 /* Scan what we just read, to see if we have reached the terminating \r\n. Be
2264 generous, and accept a plain \n terminator as well. The only illegal
2265 character is 0. */
2266
2267 for (pp = p; pp < p + count; pp++)
2268 {
2269 if (*pp == 0) goto END_OFF; /* Zero octet not allowed */
2270 if (*pp == '\n')
2271 {
2272 if (pp[-1] == '\r') pp--;
2273 *pp = 0;
2274 goto GOT_DATA; /* Break out of both loops */
2275 }
2276 }
2277
2278 /* Reached the end of the data without finding \n. Let the loop continue to
2279 read some more, if there is room. */
2280
2281 p = pp;
2282 }
2283
2284 GOT_DATA:
2285
2286 /* We have received a line of data. Check it carefully. It must start with the
2287 same two port numbers that we sent, followed by data as defined by the RFC. For
2288 example,
2289
2290 12345 , 25 : USERID : UNIX :root
2291
2292 However, the amount of white space may be different to what we sent. In the
2293 "osname" field there may be several sub-fields, comma separated. The data we
2294 actually want to save follows the third colon. Some systems put leading spaces
2295 in it - we discard those. */
2296
2297 if (sscanf(CS buffer + qlen, "%d , %d%n", &received_sender_port,
2298 &received_interface_port, &n) != 2 ||
2299 received_sender_port != sender_host_port ||
2300 received_interface_port != interface_port)
2301 goto END_OFF;
2302
2303 p = buffer + qlen + n;
2304 while(isspace(*p)) p++;
2305 if (*p++ != ':') goto END_OFF;
2306 while(isspace(*p)) p++;
2307 if (Ustrncmp(p, "USERID", 6) != 0) goto END_OFF;
2308 p += 6;
2309 while(isspace(*p)) p++;
2310 if (*p++ != ':') goto END_OFF;
2311 while (*p != 0 && *p != ':') p++;
2312 if (*p++ == 0) goto END_OFF;
2313 while(isspace(*p)) p++;
2314 if (*p == 0) goto END_OFF;
2315
2316 /* The rest of the line is the data we want. We turn it into printing
2317 characters when we save it, so that it cannot mess up the format of any logging
2318 or Received: lines into which it gets inserted. We keep a maximum of 127
2319 characters. */
2320
2321 sender_ident = string_printing(string_copyn(p, 127));
2322 DEBUG(D_ident) debug_printf("sender_ident = %s\n", sender_ident);
2323
2324 END_OFF:
2325 (void)close(sock);
2326 return;
2327 }
2328
2329
2330
2331
2332 /*************************************************
2333 * Match host to a single host-list item *
2334 *************************************************/
2335
2336 /* This function compares a host (name or address) against a single item
2337 from a host list. The host name gets looked up if it is needed and is not
2338 already known. The function is called from verify_check_this_host() via
2339 match_check_list(), which is why most of its arguments are in a single block.
2340
2341 Arguments:
2342 arg the argument block (see below)
2343 ss the host-list item
2344 valueptr where to pass back looked up data, or NULL
2345 error for error message when returning ERROR
2346
2347 The block contains:
2348 host_name (a) the host name, or
2349 (b) NULL, implying use sender_host_name and
2350 sender_host_aliases, looking them up if required, or
2351 (c) the empty string, meaning that only IP address matches
2352 are permitted
2353 host_address the host address
2354 host_ipv4 the IPv4 address taken from an IPv6 one
2355
2356 Returns: OK matched
2357 FAIL did not match
2358 DEFER lookup deferred
2359 ERROR (a) failed to find the host name or IP address, or
2360 (b) unknown lookup type specified, or
2361 (c) host name encountered when only IP addresses are
2362 being matched
2363 */
2364
2365 int
2366 check_host(void *arg, uschar *ss, uschar **valueptr, uschar **error)
2367 {
2368 check_host_block *cb = (check_host_block *)arg;
2369 int mlen = -1;
2370 int maskoffset;
2371 BOOL iplookup = FALSE;
2372 BOOL isquery = FALSE;
2373 BOOL isiponly = cb->host_name != NULL && cb->host_name[0] == 0;
2374 uschar *t;
2375 uschar *semicolon;
2376 uschar **aliases;
2377
2378 /* Optimize for the special case when the pattern is "*". */
2379
2380 if (*ss == '*' && ss[1] == 0) return OK;
2381
2382 /* If the pattern is empty, it matches only in the case when there is no host -
2383 this can occur in ACL checking for SMTP input using the -bs option. In this
2384 situation, the host address is the empty string. */
2385
2386 if (cb->host_address[0] == 0) return (*ss == 0)? OK : FAIL;
2387 if (*ss == 0) return FAIL;
2388
2389 /* If the pattern is precisely "@" then match against the primary host name,
2390 provided that host name matching is permitted; if it's "@[]" match against the
2391 local host's IP addresses. */
2392
2393 if (*ss == '@')
2394 {
2395 if (ss[1] == 0)
2396 {
2397 if (isiponly) return ERROR;
2398 ss = primary_hostname;
2399 }
2400 else if (Ustrcmp(ss, "@[]") == 0)
2401 {
2402 ip_address_item *ip;
2403 for (ip = host_find_interfaces(); ip != NULL; ip = ip->next)
2404 if (Ustrcmp(ip->address, cb->host_address) == 0) return OK;
2405 return FAIL;
2406 }
2407 }
2408
2409 /* If the pattern is an IP address, optionally followed by a bitmask count, do
2410 a (possibly masked) comparision with the current IP address. */
2411
2412 if (string_is_ip_address(ss, &maskoffset) != 0)
2413 return (host_is_in_net(cb->host_address, ss, maskoffset)? OK : FAIL);
2414
2415 /* The pattern is not an IP address. A common error that people make is to omit
2416 one component of an IPv4 address, either by accident, or believing that, for
2417 example, 1.2.3/24 is the same as 1.2.3.0/24, or 1.2.3 is the same as 1.2.3.0,
2418 which it isn't. (Those applications that do accept 1.2.3 as an IP address
2419 interpret it as 1.2.0.3 because the final component becomes 16-bit - this is an
2420 ancient specification.) To aid in debugging these cases, we give a specific
2421 error if the pattern contains only digits and dots or contains a slash preceded
2422 only by digits and dots (a slash at the start indicates a file name and of
2423 course slashes may be present in lookups, but not preceded only by digits and
2424 dots). */
2425
2426 for (t = ss; isdigit(*t) || *t == '.'; t++);
2427 if (*t == 0 || (*t == '/' && t != ss))
2428 {
2429 *error = US"malformed IPv4 address or address mask";
2430 return ERROR;
2431 }
2432
2433 /* See if there is a semicolon in the pattern */
2434
2435 semicolon = Ustrchr(ss, ';');
2436
2437 /* If we are doing an IP address only match, then all lookups must be IP
2438 address lookups, even if there is no "net-". */
2439
2440 if (isiponly)
2441 {
2442 iplookup = semicolon != NULL;
2443 }
2444
2445 /* Otherwise, if the item is of the form net[n]-lookup;<file|query> then it is
2446 a lookup on a masked IP network, in textual form. We obey this code even if we
2447 have already set iplookup, so as to skip over the "net-" prefix and to set the
2448 mask length. The net- stuff really only applies to single-key lookups where the
2449 key is implicit. For query-style lookups the key is specified in the query.
2450 From release 4.30, the use of net- for query style is no longer needed, but we
2451 retain it for backward compatibility. */
2452
2453 if (Ustrncmp(ss, "net", 3) == 0 && semicolon != NULL)
2454 {
2455 mlen = 0;
2456 for (t = ss + 3; isdigit(*t); t++) mlen = mlen * 10 + *t - '0';
2457 if (mlen == 0 && t == ss+3) mlen = -1; /* No mask supplied */
2458 iplookup = (*t++ == '-');
2459 }
2460 else t = ss;
2461
2462 /* Do the IP address lookup if that is indeed what we have */
2463
2464 if (iplookup)
2465 {
2466 int insize;
2467 int search_type;
2468 int incoming[4];
2469 void *handle;
2470 uschar *filename, *key, *result;
2471 uschar buffer[64];
2472
2473 /* Find the search type */
2474
2475 search_type = search_findtype(t, semicolon - t);
2476
2477 if (search_type < 0) log_write(0, LOG_MAIN|LOG_PANIC_DIE, "%s",
2478 search_error_message);
2479
2480 /* Adjust parameters for the type of lookup. For a query-style lookup, there
2481 is no file name, and the "key" is just the query. For query-style with a file
2482 name, we have to fish the file off the start of the query. For a single-key
2483 lookup, the key is the current IP address, masked appropriately, and
2484 reconverted to text form, with the mask appended. For IPv6 addresses, specify
2485 dot separators instead of colons, except when the lookup type is "iplsearch".
2486 */
2487
2488 if (mac_islookup(search_type, lookup_absfilequery))
2489 {
2490 filename = semicolon + 1;
2491 key = filename;
2492 while (*key != 0 && !isspace(*key)) key++;
2493 filename = string_copyn(filename, key - filename);
2494 while (isspace(*key)) key++;
2495 }
2496 else if (mac_islookup(search_type, lookup_querystyle))
2497 {
2498 filename = NULL;
2499 key = semicolon + 1;
2500 }
2501 else /* Single-key style */
2502 {
2503 int sep = (Ustrcmp(lookup_list[search_type]->name, "iplsearch") == 0)?
2504 ':' : '.';
2505 insize = host_aton(cb->host_address, incoming);
2506 host_mask(insize, incoming, mlen);
2507 (void)host_nmtoa(insize, incoming, mlen, buffer, sep);
2508 key = buffer;
2509 filename = semicolon + 1;
2510 }
2511
2512 /* Now do the actual lookup; note that there is no search_close() because
2513 of the caching arrangements. */
2514
2515 handle = search_open(filename, search_type, 0, NULL, NULL);
2516 if (handle == NULL) log_write(0, LOG_MAIN|LOG_PANIC_DIE, "%s",
2517 search_error_message);
2518 result = search_find(handle, filename, key, -1, NULL, 0, 0, NULL);
2519 if (valueptr != NULL) *valueptr = result;
2520 return (result != NULL)? OK : search_find_defer? DEFER: FAIL;
2521 }
2522
2523 /* The pattern is not an IP address or network reference of any kind. That is,
2524 it is a host name pattern. If this is an IP only match, there's an error in the
2525 host list. */
2526
2527 if (isiponly)
2528 {
2529 *error = US"cannot match host name in match_ip list";
2530 return ERROR;
2531 }
2532
2533 /* Check the characters of the pattern to see if they comprise only letters,
2534 digits, full stops, and hyphens (the constituents of domain names). Allow
2535 underscores, as they are all too commonly found. Sigh. Also, if
2536 allow_utf8_domains is set, allow top-bit characters. */
2537
2538 for (t = ss; *t != 0; t++)
2539 if (!isalnum(*t) && *t != '.' && *t != '-' && *t != '_' &&
2540 (!allow_utf8_domains || *t < 128)) break;
2541
2542 /* If the pattern is a complete domain name, with no fancy characters, look up
2543 its IP address and match against that. Note that a multi-homed host will add
2544 items to the chain. */
2545
2546 if (*t == 0)
2547 {
2548 int rc;
2549 host_item h;
2550 h.next = NULL;
2551 h.name = ss;
2552 h.address = NULL;
2553 h.mx = MX_NONE;
2554
2555 rc = host_find_byname(&h, NULL, HOST_FIND_QUALIFY_SINGLE, NULL, FALSE);
2556 if (rc == HOST_FOUND || rc == HOST_FOUND_LOCAL)
2557 {
2558 host_item *hh;
2559 for (hh = &h; hh != NULL; hh = hh->next)
2560 {
2561 if (host_is_in_net(hh->address, cb->host_address, 0)) return OK;
2562 }
2563 return FAIL;
2564 }
2565 if (rc == HOST_FIND_AGAIN) return DEFER;
2566 *error = string_sprintf("failed to find IP address for %s", ss);
2567 return ERROR;
2568 }
2569
2570 /* Almost all subsequent comparisons require the host name, and can be done
2571 using the general string matching function. When this function is called for
2572 outgoing hosts, the name is always given explicitly. If it is NULL, it means we
2573 must use sender_host_name and its aliases, looking them up if necessary. */
2574
2575 if (cb->host_name != NULL) /* Explicit host name given */
2576 return match_check_string(cb->host_name, ss, -1, TRUE, TRUE, TRUE,
2577 valueptr);
2578
2579 /* Host name not given; in principle we need the sender host name and its
2580 aliases. However, for query-style lookups, we do not need the name if the
2581 query does not contain $sender_host_name. From release 4.23, a reference to
2582 $sender_host_name causes it to be looked up, so we don't need to do the lookup
2583 on spec. */
2584
2585 if ((semicolon = Ustrchr(ss, ';')) != NULL)
2586 {
2587 uschar *affix;
2588 int partial, affixlen, starflags, id;
2589
2590 *semicolon = 0;
2591 id = search_findtype_partial(ss, &partial, &affix, &affixlen, &starflags);
2592 *semicolon=';';
2593
2594 if (id < 0) /* Unknown lookup type */
2595 {
2596 log_write(0, LOG_MAIN|LOG_PANIC, "%s in host list item \"%s\"",
2597 search_error_message, ss);
2598 return DEFER;
2599 }
2600 isquery = mac_islookup(id, lookup_querystyle|lookup_absfilequery);
2601 }
2602
2603 if (isquery)
2604 {
2605 switch(match_check_string(US"", ss, -1, TRUE, TRUE, TRUE, valueptr))
2606 {
2607 case OK: return OK;
2608 case DEFER: return DEFER;
2609 default: return FAIL;
2610 }
2611 }
2612
2613 /* Not a query-style lookup; must ensure the host name is present, and then we
2614 do a check on the name and all its aliases. */
2615
2616 if (sender_host_name == NULL)
2617 {
2618 HDEBUG(D_host_lookup)
2619 debug_printf("sender host name required, to match against %s\n", ss);
2620 if (host_lookup_failed || host_name_lookup() != OK)
2621 {
2622 *error = string_sprintf("failed to find host name for %s",
2623 sender_host_address);;
2624 return ERROR;
2625 }
2626 host_build_sender_fullhost();
2627 }
2628
2629 /* Match on the sender host name, using the general matching function */
2630
2631 switch(match_check_string(sender_host_name, ss, -1, TRUE, TRUE, TRUE,
2632 valueptr))
2633 {
2634 case OK: return OK;
2635 case DEFER: return DEFER;
2636 }
2637
2638 /* If there are aliases, try matching on them. */
2639
2640 aliases = sender_host_aliases;
2641 while (*aliases != NULL)
2642 {
2643 switch(match_check_string(*aliases++, ss, -1, TRUE, TRUE, TRUE, valueptr))
2644 {
2645 case OK: return OK;
2646 case DEFER: return DEFER;
2647 }
2648 }
2649 return FAIL;
2650 }
2651
2652
2653
2654
2655 /*************************************************
2656 * Check a specific host matches a host list *
2657 *************************************************/
2658
2659 /* This function is passed a host list containing items in a number of
2660 different formats and the identity of a host. Its job is to determine whether
2661 the given host is in the set of hosts defined by the list. The host name is
2662 passed as a pointer so that it can be looked up if needed and not already
2663 known. This is commonly the case when called from verify_check_host() to check
2664 an incoming connection. When called from elsewhere the host name should usually
2665 be set.
2666
2667 This function is now just a front end to match_check_list(), which runs common
2668 code for scanning a list. We pass it the check_host() function to perform a
2669 single test.
2670
2671 Arguments:
2672 listptr pointer to the host list
2673 cache_bits pointer to cache for named lists, or NULL
2674 host_name the host name or NULL, implying use sender_host_name and
2675 sender_host_aliases, looking them up if required
2676 host_address the IP address
2677 valueptr if not NULL, data from a lookup is passed back here
2678
2679 Returns: OK if the host is in the defined set
2680 FAIL if the host is not in the defined set,
2681 DEFER if a data lookup deferred (not a host lookup)
2682
2683 If the host name was needed in order to make a comparison, and could not be
2684 determined from the IP address, the result is FAIL unless the item
2685 "+allow_unknown" was met earlier in the list, in which case OK is returned. */
2686
2687 int
2688 verify_check_this_host(uschar **listptr, unsigned int *cache_bits,
2689 uschar *host_name, uschar *host_address, uschar **valueptr)
2690 {
2691 int rc;
2692 unsigned int *local_cache_bits = cache_bits;
2693 uschar *save_host_address = deliver_host_address;
2694 check_host_block cb;
2695 cb.host_name = host_name;
2696 cb.host_address = host_address;
2697
2698 if (valueptr != NULL) *valueptr = NULL;
2699
2700 /* If the host address starts off ::ffff: it is an IPv6 address in
2701 IPv4-compatible mode. Find the IPv4 part for checking against IPv4
2702 addresses. */
2703
2704 cb.host_ipv4 = (Ustrncmp(host_address, "::ffff:", 7) == 0)?
2705 host_address + 7 : host_address;
2706
2707 /* During the running of the check, put the IP address into $host_address. In
2708 the case of calls from the smtp transport, it will already be there. However,
2709 in other calls (e.g. when testing ignore_target_hosts), it won't. Just to be on
2710 the safe side, any existing setting is preserved, though as I write this
2711 (November 2004) I can't see any cases where it is actually needed. */
2712
2713 deliver_host_address = host_address;
2714 rc = match_check_list(
2715 listptr, /* the list */
2716 0, /* separator character */
2717 &hostlist_anchor, /* anchor pointer */
2718 &local_cache_bits, /* cache pointer */
2719 check_host, /* function for testing */
2720 &cb, /* argument for function */
2721 MCL_HOST, /* type of check */
2722 (host_address == sender_host_address)?
2723 US"host" : host_address, /* text for debugging */
2724 valueptr); /* where to pass back data */
2725 deliver_host_address = save_host_address;
2726 return rc;
2727 }
2728
2729
2730
2731
2732 /*************************************************
2733 * Check the remote host matches a list *
2734 *************************************************/
2735
2736 /* This is a front end to verify_check_this_host(), created because checking
2737 the remote host is a common occurrence. With luck, a good compiler will spot
2738 the tail recursion and optimize it. If there's no host address, this is
2739 command-line SMTP input - check against an empty string for the address.
2740
2741 Arguments:
2742 listptr pointer to the host list
2743
2744 Returns: the yield of verify_check_this_host(),
2745 i.e. OK, FAIL, or DEFER
2746 */
2747
2748 int
2749 verify_check_host(uschar **listptr)
2750 {
2751 return verify_check_this_host(listptr, sender_host_cache, NULL,
2752 (sender_host_address == NULL)? US"" : sender_host_address, NULL);
2753 }
2754
2755
2756
2757
2758
2759 /*************************************************
2760 * Invert an IP address *
2761 *************************************************/
2762
2763 /* Originally just used for DNS xBL lists, now also used for the
2764 reverse_ip expansion operator.
2765
2766 Arguments:
2767 buffer where to put the answer
2768 address the address to invert
2769 */
2770
2771 void
2772 invert_address(uschar *buffer, uschar *address)
2773 {
2774 int bin[4];
2775 uschar *bptr = buffer;
2776
2777 /* If this is an IPv4 address mapped into IPv6 format, adjust the pointer
2778 to the IPv4 part only. */
2779
2780 if (Ustrncmp(address, "::ffff:", 7) == 0) address += 7;
2781
2782 /* Handle IPv4 address: when HAVE_IPV6 is false, the result of host_aton() is
2783 always 1. */
2784
2785 if (host_aton(address, bin) == 1)
2786 {
2787 int i;
2788 int x = bin[0];
2789 for (i = 0; i < 4; i++)
2790 {
2791 sprintf(CS bptr, "%d.", x & 255);
2792 while (*bptr) bptr++;
2793 x >>= 8;
2794 }
2795 }
2796
2797 /* Handle IPv6 address. Actually, as far as I know, there are no IPv6 addresses
2798 in any DNS black lists, and the format in which they will be looked up is
2799 unknown. This is just a guess. */
2800
2801 #if HAVE_IPV6
2802 else
2803 {
2804 int i, j;
2805 for (j = 3; j >= 0; j--)
2806 {
2807 int x = bin[j];
2808 for (i = 0; i < 8; i++)
2809 {
2810 sprintf(CS bptr, "%x.", x & 15);
2811 while (*bptr) bptr++;
2812 x >>= 4;
2813 }
2814 }
2815 }
2816 #endif
2817
2818 /* Remove trailing period -- this is needed so that both arbitrary
2819 dnsbl keydomains and inverted addresses may be combined with the
2820 same format string, "%s.%s" */
2821
2822 *(--bptr) = 0;
2823 }
2824
2825
2826
2827 /*************************************************
2828 * Perform a single dnsbl lookup *
2829 *************************************************/
2830
2831 /* This function is called from verify_check_dnsbl() below. It is also called
2832 recursively from within itself when domain and domain_txt are different
2833 pointers, in order to get the TXT record from the alternate domain.
2834
2835 Arguments:
2836 domain the outer dnsbl domain
2837 domain_txt alternate domain to lookup TXT record on success; when the
2838 same domain is to be used, domain_txt == domain (that is,
2839 the pointers must be identical, not just the text)
2840 keydomain the current keydomain (for debug message)
2841 prepend subdomain to lookup (like keydomain, but
2842 reversed if IP address)
2843 iplist the list of matching IP addresses, or NULL for "any"
2844 bitmask true if bitmask matching is wanted
2845 match_type condition for 'succeed' result
2846 0 => Any RR in iplist (=)
2847 1 => No RR in iplist (!=)
2848 2 => All RRs in iplist (==)
2849 3 => Some RRs not in iplist (!==)
2850 the two bits are defined as MT_NOT and MT_ALL
2851 defer_return what to return for a defer
2852
2853 Returns: OK if lookup succeeded
2854 FAIL if not
2855 */
2856
2857 static int
2858 one_check_dnsbl(uschar *domain, uschar *domain_txt, uschar *keydomain,
2859 uschar *prepend, uschar *iplist, BOOL bitmask, int match_type,
2860 int defer_return)
2861 {
2862 dns_answer dnsa;
2863 dns_scan dnss;
2864 tree_node *t;
2865 dnsbl_cache_block *cb;
2866 int old_pool = store_pool;
2867 uschar query[256]; /* DNS domain max length */
2868
2869 /* Construct the specific query domainname */
2870
2871 if (!string_format(query, sizeof(query), "%s.%s", prepend, domain))
2872 {
2873 log_write(0, LOG_MAIN|LOG_PANIC, "dnslist query is too long "
2874 "(ignored): %s...", query);
2875 return FAIL;
2876 }
2877
2878 /* Look for this query in the cache. */
2879
2880 t = tree_search(dnsbl_cache, query);
2881
2882 /* If not cached from a previous lookup, we must do a DNS lookup, and
2883 cache the result in permanent memory. */
2884
2885 if (t == NULL)
2886 {
2887 store_pool = POOL_PERM;
2888
2889 /* Set up a tree entry to cache the lookup */
2890
2891 t = store_get(sizeof(tree_node) + Ustrlen(query));
2892 Ustrcpy(t->name, query);
2893 t->data.ptr = cb = store_get(sizeof(dnsbl_cache_block));
2894 (void)tree_insertnode(&dnsbl_cache, t);
2895
2896 /* Do the DNS loopup . */
2897
2898 HDEBUG(D_dnsbl) debug_printf("new DNS lookup for %s\n", query);
2899 cb->rc = dns_basic_lookup(&dnsa, query, T_A);
2900 cb->text_set = FALSE;
2901 cb->text = NULL;
2902 cb->rhs = NULL;
2903
2904 /* If the lookup succeeded, cache the RHS address. The code allows for
2905 more than one address - this was for complete generality and the possible
2906 use of A6 records. However, A6 records have been reduced to experimental
2907 status (August 2001) and may die out. So they may never get used at all,
2908 let alone in dnsbl records. However, leave the code here, just in case.
2909
2910 Quite apart from one A6 RR generating multiple addresses, there are DNS
2911 lists that return more than one A record, so we must handle multiple
2912 addresses generated in that way as well. */
2913
2914 if (cb->rc == DNS_SUCCEED)
2915 {
2916 dns_record *rr;
2917 dns_address **addrp = &(cb->rhs);
2918 for (rr = dns_next_rr(&dnsa, &dnss, RESET_ANSWERS);
2919 rr != NULL;
2920 rr = dns_next_rr(&dnsa, &dnss, RESET_NEXT))
2921 {
2922 if (rr->type == T_A)
2923 {
2924 dns_address *da = dns_address_from_rr(&dnsa, rr);
2925 if (da != NULL)
2926 {
2927 *addrp = da;
2928 while (da->next != NULL) da = da->next;
2929 addrp = &(da->next);
2930 }
2931 }
2932 }
2933
2934 /* If we didn't find any A records, change the return code. This can
2935 happen when there is a CNAME record but there are no A records for what
2936 it points to. */
2937
2938 if (cb->rhs == NULL) cb->rc = DNS_NODATA;
2939 }
2940
2941 store_pool = old_pool;
2942 }
2943
2944 /* Previous lookup was cached */
2945
2946 else
2947 {
2948 HDEBUG(D_dnsbl) debug_printf("using result of previous DNS lookup\n");
2949 cb = t->data.ptr;
2950 }
2951
2952 /* We now have the result of the DNS lookup, either newly done, or cached
2953 from a previous call. If the lookup succeeded, check against the address
2954 list if there is one. This may be a positive equality list (introduced by
2955 "="), a negative equality list (introduced by "!="), a positive bitmask
2956 list (introduced by "&"), or a negative bitmask list (introduced by "!&").*/
2957
2958 if (cb->rc == DNS_SUCCEED)
2959 {
2960 dns_address *da = NULL;
2961 uschar *addlist = cb->rhs->address;
2962
2963 /* For A and AAAA records, there may be multiple addresses from multiple
2964 records. For A6 records (currently not expected to be used) there may be
2965 multiple addresses from a single record. */
2966
2967 for (da = cb->rhs->next; da != NULL; da = da->next)
2968 addlist = string_sprintf("%s, %s", addlist, da->address);
2969
2970 HDEBUG(D_dnsbl) debug_printf("DNS lookup for %s succeeded (yielding %s)\n",
2971 query, addlist);
2972
2973 /* Address list check; this can be either for equality, or via a bitmask.
2974 In the latter case, all the bits must match. */
2975
2976 if (iplist != NULL)
2977 {
2978 for (da = cb->rhs; da != NULL; da = da->next)
2979 {
2980 int ipsep = ',';
2981 uschar ip[46];
2982 uschar *ptr = iplist;
2983 uschar *res;
2984
2985 /* Handle exact matching */
2986
2987 if (!bitmask)
2988 {
2989 while ((res = string_nextinlist(&ptr, &ipsep, ip, sizeof(ip))) != NULL)
2990 {
2991 if (Ustrcmp(CS da->address, ip) == 0) break;
2992 }
2993 }
2994
2995 /* Handle bitmask matching */
2996
2997 else
2998 {
2999 int address[4];
3000 int mask = 0;
3001
3002 /* At present, all known DNS blocking lists use A records, with
3003 IPv4 addresses on the RHS encoding the information they return. I
3004 wonder if this will linger on as the last vestige of IPv4 when IPv6
3005 is ubiquitous? Anyway, for now we use paranoia code to completely
3006 ignore IPv6 addresses. The default mask is 0, which always matches.
3007 We change this only for IPv4 addresses in the list. */
3008
3009 if (host_aton(da->address, address) == 1) mask = address[0];
3010
3011 /* Scan the returned addresses, skipping any that are IPv6 */
3012
3013 while ((res = string_nextinlist(&ptr, &ipsep, ip, sizeof(ip))) != NULL)
3014 {
3015 if (host_aton(ip, address) != 1) continue;
3016 if ((address[0] & mask) == address[0]) break;
3017 }
3018 }
3019
3020 /* If either
3021
3022 (a) An IP address in an any ('=') list matched, or
3023 (b) No IP address in an all ('==') list matched
3024
3025 then we're done searching. */
3026
3027 if (((match_type & MT_ALL) != 0) == (res == NULL)) break;
3028 }
3029
3030 /* If da == NULL, either
3031
3032 (a) No IP address in an any ('=') list matched, or
3033 (b) An IP address in an all ('==') list didn't match
3034
3035 so behave as if the DNSBL lookup had not succeeded, i.e. the host is not on
3036 the list. */
3037
3038 if ((match_type == MT_NOT || match_type == MT_ALL) != (da == NULL))
3039 {
3040 HDEBUG(D_dnsbl)
3041 {
3042 uschar *res = NULL;
3043 switch(match_type)
3044 {
3045 case 0:
3046 res = US"was no match";
3047 break;
3048 case MT_NOT:
3049 res = US"was an exclude match";
3050 break;
3051 case MT_ALL:
3052 res = US"was an IP address that did not match";
3053 break;
3054 case MT_NOT|MT_ALL:
3055 res = US"were no IP addresses that did not match";
3056 break;
3057 }
3058 debug_printf("=> but we are not accepting this block class because\n");
3059 debug_printf("=> there %s for %s%c%s\n",
3060 res,
3061 ((match_type & MT_ALL) == 0)? "" : "=",
3062 bitmask? '&' : '=', iplist);
3063 }
3064 return FAIL;
3065 }
3066 }
3067
3068 /* Either there was no IP list, or the record matched, implying that the
3069 domain is on the list. We now want to find a corresponding TXT record. If an
3070 alternate domain is specified for the TXT record, call this function
3071 recursively to look that up; this has the side effect of re-checking that
3072 there is indeed an A record at the alternate domain. */
3073
3074 if (domain_txt != domain)
3075 return one_check_dnsbl(domain_txt, domain_txt, keydomain, prepend, NULL,
3076 FALSE, match_type, defer_return);
3077
3078 /* If there is no alternate domain, look up a TXT record in the main domain
3079 if it has not previously been cached. */
3080
3081 if (!cb->text_set)
3082 {
3083 cb->text_set = TRUE;
3084 if (dns_basic_lookup(&dnsa, query, T_TXT) == DNS_SUCCEED)
3085 {
3086 dns_record *rr;
3087 for (rr = dns_next_rr(&dnsa, &dnss, RESET_ANSWERS);
3088 rr != NULL;
3089 rr = dns_next_rr(&dnsa, &dnss, RESET_NEXT))
3090 if (rr->type == T_TXT) break;
3091 if (rr != NULL)
3092 {
3093 int len = (rr->data)[0];
3094 if (len > 511) len = 127;
3095 store_pool = POOL_PERM;
3096 cb->text = string_sprintf("%.*s", len, (const uschar *)(rr->data+1));
3097 store_pool = old_pool;
3098 }
3099 }
3100 }
3101
3102 dnslist_value = addlist;
3103 dnslist_text = cb->text;
3104 return OK;
3105 }
3106
3107 /* There was a problem with the DNS lookup */
3108
3109 if (cb->rc != DNS_NOMATCH && cb->rc != DNS_NODATA)
3110 {
3111 log_write(L_dnslist_defer, LOG_MAIN,
3112 "DNS list lookup defer (probably timeout) for %s: %s", query,
3113 (defer_return == OK)? US"assumed in list" :
3114 (defer_return == FAIL)? US"assumed not in list" :
3115 US"returned DEFER");
3116 return defer_return;
3117 }
3118
3119 /* No entry was found in the DNS; continue for next domain */
3120
3121 HDEBUG(D_dnsbl)
3122 {
3123 debug_printf("DNS lookup for %s failed\n", query);
3124 debug_printf("=> that means %s is not listed at %s\n",
3125 keydomain, domain);
3126 }
3127
3128 return FAIL;
3129 }
3130
3131
3132
3133
3134 /*************************************************
3135 * Check host against DNS black lists *
3136 *************************************************/
3137
3138 /* This function runs checks against a list of DNS black lists, until one
3139 matches. Each item on the list can be of the form
3140
3141 domain=ip-address/key
3142
3143 The domain is the right-most domain that is used for the query, for example,
3144 blackholes.mail-abuse.org. If the IP address is present, there is a match only
3145 if the DNS lookup returns a matching IP address. Several addresses may be
3146 given, comma-separated, for example: x.y.z=127.0.0.1,127.0.0.2.
3147
3148 If no key is given, what is looked up in the domain is the inverted IP address
3149 of the current client host. If a key is given, it is used to construct the
3150 domain for the lookup. For example:
3151
3152 dsn.rfc-ignorant.org/$sender_address_domain
3153
3154 After finding a match in the DNS, the domain is placed in $dnslist_domain, and
3155 then we check for a TXT record for an error message, and if found, save its
3156 value in $dnslist_text. We also cache everything in a tree, to optimize
3157 multiple lookups.
3158
3159 The TXT record is normally looked up in the same domain as the A record, but
3160 when many lists are combined in a single DNS domain, this will not be a very
3161 specific message. It is possible to specify a different domain for looking up
3162 TXT records; this is given before the main domain, comma-separated. For
3163 example:
3164
3165 dnslists = http.dnsbl.sorbs.net,dnsbl.sorbs.net=127.0.0.2 : \
3166 socks.dnsbl.sorbs.net,dnsbl.sorbs.net=127.0.0.3
3167
3168 The caching ensures that only one lookup in dnsbl.sorbs.net is done.
3169
3170 Note: an address for testing RBL is 192.203.178.39
3171 Note: an address for testing DUL is 192.203.178.4
3172 Note: a domain for testing RFCI is example.tld.dsn.rfc-ignorant.org
3173
3174 Arguments:
3175 listptr the domain/address/data list
3176
3177 Returns: OK successful lookup (i.e. the address is on the list), or
3178 lookup deferred after +include_unknown
3179 FAIL name not found, or no data found for the given type, or
3180 lookup deferred after +exclude_unknown (default)
3181 DEFER lookup failure, if +defer_unknown was set
3182 */
3183
3184 int
3185 verify_check_dnsbl(uschar **listptr)
3186 {
3187 int sep = 0;
3188 int defer_return = FAIL;
3189 uschar *list = *listptr;
3190 uschar *domain;
3191 uschar *s;
3192 uschar buffer[1024];
3193 uschar revadd[128]; /* Long enough for IPv6 address */
3194
3195 /* Indicate that the inverted IP address is not yet set up */
3196
3197 revadd[0] = 0;
3198
3199 /* In case this is the first time the DNS resolver is being used. */
3200
3201 dns_init(FALSE, FALSE);
3202
3203 /* Loop through all the domains supplied, until something matches */
3204
3205 while ((domain = string_nextinlist(&list, &sep, buffer, sizeof(buffer))) != NULL)
3206 {
3207 int rc;
3208 BOOL bitmask = FALSE;
3209 int match_type = 0;
3210 uschar *domain_txt;
3211 uschar *comma;
3212 uschar *iplist;
3213 uschar *key;
3214
3215 HDEBUG(D_dnsbl) debug_printf("DNS list check: %s\n", domain);
3216
3217 /* Deal with special values that change the behaviour on defer */
3218
3219 if (domain[0] == '+')
3220 {
3221 if (strcmpic(domain, US"+include_unknown") == 0) defer_return = OK;
3222 else if (strcmpic(domain, US"+exclude_unknown") == 0) defer_return = FAIL;
3223 else if (strcmpic(domain, US"+defer_unknown") == 0) defer_return = DEFER;
3224 else
3225 log_write(0, LOG_MAIN|LOG_PANIC, "unknown item in dnslist (ignored): %s",
3226 domain);
3227 continue;
3228 }
3229
3230 /* See if there's explicit data to be looked up */
3231
3232 key = Ustrchr(domain, '/');
3233 if (key != NULL) *key++ = 0;
3234
3235 /* See if there's a list of addresses supplied after the domain name. This is
3236 introduced by an = or a & character; if preceded by = we require all matches
3237 and if preceded by ! we invert the result. */
3238
3239 iplist = Ustrchr(domain, '=');
3240 if (iplist == NULL)
3241 {
3242 bitmask = TRUE;
3243 iplist = Ustrchr(domain, '&');
3244 }
3245
3246 if (iplist != NULL) /* Found either = or & */
3247 {
3248 if (iplist > domain && iplist[-1] == '!') /* Handle preceding ! */
3249 {
3250 match_type |= MT_NOT;
3251 iplist[-1] = 0;
3252 }
3253
3254 *iplist++ = 0; /* Terminate domain, move on */
3255
3256 /* If we found = (bitmask == FALSE), check for == or =& */
3257
3258 if (!bitmask && (*iplist == '=' || *iplist == '&'))
3259 {
3260 bitmask = *iplist++ == '&';
3261 match_type |= MT_ALL;
3262 }
3263 }
3264
3265 /* If there is a comma in the domain, it indicates that a second domain for
3266 looking up TXT records is provided, before the main domain. Otherwise we must
3267 set domain_txt == domain. */
3268
3269 domain_txt = domain;
3270 comma = Ustrchr(domain, ',');
3271 if (comma != NULL)
3272 {
3273 *comma++ = 0;
3274 domain = comma;
3275 }
3276
3277 /* Check that what we have left is a sensible domain name. There is no reason
3278 why these domains should in fact use the same syntax as hosts and email
3279 domains, but in practice they seem to. However, there is little point in
3280 actually causing an error here, because that would no doubt hold up incoming
3281 mail. Instead, I'll just log it. */
3282
3283 for (s = domain; *s != 0; s++)
3284 {
3285 if (!isalnum(*s) && *s != '-' && *s != '.' && *s != '_')
3286 {
3287 log_write(0, LOG_MAIN, "dnslists domain \"%s\" contains "
3288 "strange characters - is this right?", domain);
3289 break;
3290 }
3291 }
3292
3293 /* Check the alternate domain if present */
3294
3295 if (domain_txt != domain) for (s = domain_txt; *s != 0; s++)
3296 {
3297 if (!isalnum(*s) && *s != '-' && *s != '.' && *s != '_')
3298 {
3299 log_write(0, LOG_MAIN, "dnslists domain \"%s\" contains "
3300 "strange characters - is this right?", domain_txt);
3301 break;
3302 }
3303 }
3304
3305 /* If there is no key string, construct the query by adding the domain name
3306 onto the inverted host address, and perform a single DNS lookup. */
3307
3308 if (key == NULL)
3309 {
3310 if (sender_host_address == NULL) return FAIL; /* can never match */
3311 if (revadd[0] == 0) invert_address(revadd, sender_host_address);
3312 rc = one_check_dnsbl(domain, domain_txt, sender_host_address, revadd,
3313 iplist, bitmask, match_type, defer_return);
3314 if (rc == OK)
3315 {
3316 dnslist_domain = string_copy(domain_txt);
3317 dnslist_matched = string_copy(sender_host_address);
3318 HDEBUG(D_dnsbl) debug_printf("=> that means %s is listed at %s\n",
3319 sender_host_address, dnslist_domain);
3320 }
3321 if (rc != FAIL) return rc; /* OK or DEFER */
3322 }
3323
3324 /* If there is a key string, it can be a list of domains or IP addresses to
3325 be concatenated with the main domain. */
3326
3327 else
3328 {
3329 int keysep = 0;
3330 BOOL defer = FALSE;
3331 uschar *keydomain;
3332 uschar keybuffer[256];
3333 uschar keyrevadd[128];
3334
3335 while ((keydomain = string_nextinlist(&key, &keysep, keybuffer,
3336 sizeof(keybuffer))) != NULL)
3337 {
3338 uschar *prepend = keydomain;
3339
3340 if (string_is_ip_address(keydomain, NULL) != 0)
3341 {
3342 invert_address(keyrevadd, keydomain);
3343 prepend = keyrevadd;
3344 }
3345
3346 rc = one_check_dnsbl(domain, domain_txt, keydomain, prepend, iplist,
3347 bitmask, match_type, defer_return);
3348
3349 if (rc == OK)
3350 {
3351 dnslist_domain = string_copy(domain_txt);
3352 dnslist_matched = string_copy(keydomain);
3353 HDEBUG(D_dnsbl) debug_printf("=> that means %s is listed at %s\n",
3354 keydomain, dnslist_domain);
3355 return OK;
3356 }
3357
3358 /* If the lookup deferred, remember this fact. We keep trying the rest
3359 of the list to see if we get a useful result, and if we don't, we return
3360 DEFER at the end. */
3361
3362 if (rc == DEFER) defer = TRUE;
3363 } /* continue with next keystring domain/address */
3364
3365 if (defer) return DEFER;
3366 }
3367 } /* continue with next dnsdb outer domain */
3368
3369 return FAIL;
3370 }
3371
3372 /* End of verify.c */