1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
5 /* Copyright (c) University of Cambridge 1995 - 2018 */
6 /* See the file NOTICE for conditions of use and distribution. */
8 /* Functions concerned with rewriting headers */
13 /* Names for testing rewriting */
15 static const char *rrname[] = {
26 /* Structure and table for finding source of address for debug printing */
28 typedef struct where_list_block {
33 static where_list_block where_list[] = {
34 { rewrite_sender, CUS"sender:" },
35 { rewrite_from, CUS"from:" },
36 { rewrite_to, CUS"to:" },
37 { rewrite_cc, CUS"cc:" },
38 { rewrite_bcc, CUS"bcc:" },
39 { rewrite_replyto, CUS"reply-to:" },
40 { rewrite_envfrom, CUS"env-from" },
41 { rewrite_envto, CUS"env-to" },
42 { rewrite_smtp, CUS"smtp recipient" },
43 { rewrite_smtp|rewrite_smtp_sender, CUS"smtp sender" }
46 static int where_list_size = sizeof(where_list)/sizeof(where_list_block);
50 /*************************************************
51 * Ensure an address is qualified *
52 *************************************************/
57 is_recipient TRUE if a recipient address; FALSE if a sender address
59 Returns: fully-qualified address
63 rewrite_address_qualify(uschar *s, BOOL is_recipient)
65 return (parse_find_at(s) != NULL)? s :
66 string_sprintf("%s@%s", s,
67 is_recipient? qualify_domain_recipient : qualify_domain_sender);
72 /*************************************************
73 * Rewrite a single address *
74 *************************************************/
76 /* The yield is the input address if there is no rewriting to be done. Assume
77 the input is a valid address, except in the case of SMTP-time rewriting, which
78 is handled specially. When this function is called while processing filter and
79 forward files, the uid may be that of the user. Ensure it is reset while
80 expanding a replacement, in case that involves file lookups.
84 flag indicates where this address comes from; it must match the
85 flags in the rewriting rule
86 whole if not NULL, set TRUE if any rewriting rule contained the
87 "whole" bit and it is a header that is being rewritten
88 add_header if TRUE and rewriting occurs, add an "X-rewrote-xxx" header
89 if headers are in existence; this should be TRUE only when
90 a message is being received, not during delivery
91 name name of header, for use when adding X-rewrote-xxxx
92 rewrite_rules chain of rewriting rules
94 Returns: new address if rewritten; the input address if no change;
95 for a header rewrite, if the "whole" bit is set, the entire
96 rewritten address is returned, not just the active bit.
100 rewrite_one(uschar *s, int flag, BOOL *whole, BOOL add_header, uschar *name,
101 rewrite_rule *rewrite_rules)
105 uschar *domain = NULL;
108 int yield_start = 0, yield_end = 0;
110 if (whole) *whole = FALSE;
112 /* Scan the rewriting rules, ignoring any without matching flag */
114 for (rewrite_rule * rule = rewrite_rules;
116 rule_number++, rule = rule->next) if (rule->flags & flag)
118 int start, end, pdomain;
120 uschar *save_localpart;
121 const uschar *save_domain;
122 uschar *error, *new, *newparsed;
124 /* Come back here for a repeat after a successful rewrite. We do this
125 only so many times. */
129 /* If this is an SMTP-time rewrite, the pattern must be a regex and
130 the subject may have any structure. No local part or domain variables
131 can be set for the expansion. We expand the pattern in order to be consistent
132 with the other kinds of rewrite, where expansion happens inside
133 match_address_list(). */
135 if (flag & rewrite_smtp)
137 uschar *key = expand_string(rule->key);
140 if (!f.expand_string_forcedfail)
141 log_write(0, LOG_MAIN|LOG_PANIC, "failed to expand \"%s\" while "
142 "checking for SMTP rewriting: %s", rule->key, expand_string_message);
145 if (match_check_string(subject, key, 0, TRUE, FALSE, FALSE, NULL) != OK)
147 new = expand_string(rule->replacement);
150 /* All other rewrites expect the input to be a valid address, so local part
151 and domain variables can be set for expansion. For the first rule, to be
152 applied to this address, domain will be NULL and needs to be set. */
156 if (!domain) domain = Ustrrchr(subject, '@') + 1;
158 /* Use the general function for matching an address against a list (here
159 just one item, so use the "impossible value" separator UCHAR_MAX+1). */
161 if (match_address_list(subject, FALSE, TRUE, CUSS &(rule->key), NULL, 0,
162 UCHAR_MAX + 1, NULL) != OK)
165 /* The source address matches, and numerical variables have been
166 set up. If the replacement string consists of precisely "*" then no
167 rewriting is required for this address - the behaviour is as for "fail"
168 in the replacement expansion, but assuming the quit flag. */
170 if (Ustrcmp(rule->replacement, "*") == 0) break;
172 /* Otherwise, expand the replacement string. Set $local_part and $domain to
173 the appropriate values, restoring whatever value they previously had
176 save_localpart = deliver_localpart;
177 save_domain = deliver_domain;
179 /* We have subject pointing to "localpart@domain" and domain pointing to
180 the domain. Temporarily terminate the local part so that it can be
181 set up as an expansion variable */
184 deliver_localpart = subject;
185 deliver_domain = domain;
187 new = expand_string(rule->replacement);
190 deliver_localpart = save_localpart;
191 deliver_domain = save_domain;
194 /* If the expansion failed with the "forcedfail" flag, don't generate
195 an error - just give up on this rewriting rule. If the "q" flag is set,
196 give up altogether. For other expansion failures we have a configuration
201 if (f.expand_string_forcedfail)
202 { if (rule->flags & rewrite_quit) break; else continue; }
204 expand_string_message = expand_hide_passwords(expand_string_message);
206 log_write(0, LOG_MAIN|LOG_PANIC, "Expansion of %s failed while rewriting: "
207 "%s", rule->replacement, expand_string_message);
211 /* Check the what has been generated is a valid RFC 2822 address. Only
212 envelope from or SMTP sender is permitted to be rewritten as <>.*/
214 newparsed = parse_extract_address(new, &error, &start, &end, &pdomain,
215 flag == rewrite_envfrom || flag == (rewrite_smtp|rewrite_smtp_sender));
219 log_write(0, LOG_MAIN|LOG_PANIC, "Rewrite of %s yielded unparseable "
220 "address: %s in address %s", subject, error, new);
221 break; /* Give up on this address */
224 /* A non-null unqualified address can be qualified if requested. Otherwise,
225 this is an error unless it's the empty address in circumstances where that is
228 if (pdomain == 0 && (*newparsed != 0 ||
229 (flag != rewrite_envfrom && flag != (rewrite_smtp|rewrite_smtp_sender))))
231 if (rule->flags & rewrite_qualify)
233 newparsed = rewrite_address_qualify(newparsed, TRUE);
234 new = string_sprintf("%.*s%s%.*s", start, new, newparsed,
235 Ustrlen(new) - end, new + end);
236 end = start + Ustrlen(newparsed);
240 log_write(0, LOG_MAIN|LOG_PANIC, "Rewrite of %s yielded unqualified "
241 "address \"%s\"", subject, new);
242 break; /* Give up on this address */
246 /* We have a validly rewritten address */
248 if (LOGGING(address_rewrite) || (debug_selector & D_rewrite) != 0)
250 const uschar *where = CUS"?";
252 for (int i = 0; i < where_list_size; i++)
253 if (flag == where_list[i].bit)
255 where = where_list[i].string;
258 log_write(L_address_rewrite,
259 LOG_MAIN, "\"%s\" from %s rewritten as \"%s\" by rule %d",
260 yield, where, new, rule_number);
263 /* A header will only actually be added if header_last is non-NULL,
264 i.e. during message reception or delivery, but add_header should not
265 be set TRUE during delivery, as otherwise multiple instances of the header
266 can fill up the -H file and make it embarrassingly large. We don't need
267 to set header_rewritten because the -H file always gets written at the end
268 of message reception. */
271 header_add(htype_old, "X-rewrote-%s: %s\n", name, subject);
273 /* Handle the case when replacement of the whole address is possible.
274 This happens only when whole is not NULL and we are rewriting a header.
275 If *whole is already TRUE it means that a previous rule had the w
276 flag set and so we must preserve the non-active portion of the current
277 subject unless the current rule also has the w flag set. */
279 if (whole && (flag & rewrite_all_headers))
281 /* Current rule has the w flag set. We must ensure the phrase parts
282 are syntactically valid if they are present. */
284 if (rule->flags & rewrite_whole)
286 if (start > 0 && new[start-1] == '<')
288 uschar *p1 = new + start - 1;
289 uschar *p2 = new + end + 1;
290 const uschar *pf1, *pf2;
292 while (p1 > new && p1[-1] == ' ') p1--;
293 pf1 = parse_fix_phrase(new, p1 - new);
294 while (*p2 == ' ') p2++;
295 pf2 = parse_fix_phrase(p2, Ustrlen(p2));
297 start = Ustrlen(pf1) + start + new - p1;
298 end = start + Ustrlen(newparsed);
299 new = string_sprintf("%s%.*s%s", pf1, (int)(p2 - p1), p1, pf2);
302 /* Now accept the whole thing */
311 /* Current rule does not have the w flag set; if not previously
312 done any whole rewriting, behave in non-whole manner. */
314 else if (!*whole) goto NEVER_WHOLE;
316 /* Current rule does not have the w flag set, but a previous
317 rule did rewrite the whole address. Thus yield and subject will be
318 different. Preserve the previous non-active part of the address. */
323 new = string_sprintf("%.*s%s%n%s",
324 yield_start, yield, subject, &end, yield + yield_end);
330 /* Rule just rewrites active part, or handling an envelope. This
331 code is obeyed only when all rules so far have not done "whole"
337 subject = yield = newparsed;
340 domain = NULL; /* Reset for next rule */
342 /* If no further rewrites are to be done, set the done flag. This allows
343 repeats of the current rule if configured before breaking the loop. */
345 if (rule->flags & rewrite_quit) done = TRUE;
347 /* Allow the current rule to be applied up to 10 times if
350 if (rule->flags & rewrite_repeat)
352 if (count++ < 10) goto REPEAT_RULE;
353 log_write(0, LOG_MAIN|LOG_PANIC, "rewrite rule repeat ignored after 10 "
358 /* Unset expansion numeric variables, and that's it. */
366 /*************************************************
367 * Ensure qualification and rewrite *
368 *************************************************/
370 /* This function is called for envelope addresses, the boolean specifying
371 whether a recipient or a sender. It must first of all ensure the address is
372 fully qualified, and then apply any relevant re-writing rules. The add-header
373 flag causes a header to be added, recording the old address. This is marked
374 "old", so that it is never transported anywhere; it exists for local checking
375 and debugging purposes.
378 s the address to be considered
379 is_recipient TRUE for recipient addresses; FALSE otherwise
380 add_header add "X-rewrote-xxx" header when rewriting; this is
381 set TRUE only for calls from the reception functions
382 rewrite_rules points to chain of rewrite rules
383 existflags bits indicating which headers there are rewrites for
384 (just an optimisation)
386 Returns: possibly rewritten address
390 rewrite_address(uschar *s, BOOL is_recipient, BOOL add_header,
391 rewrite_rule *rewrite_rules, int existflags)
393 int flag = is_recipient? rewrite_envto : rewrite_envfrom;
394 s = rewrite_address_qualify(s, is_recipient);
395 if ((existflags & flag) != 0)
397 uschar *new = rewrite_one(s, flag, NULL, add_header, is_recipient?
398 US"original-recipient" : US"sender", rewrite_rules);
399 if (new != s) s = new;
406 /*************************************************
407 * Qualify and possibly rewrite one header *
408 *************************************************/
410 /* This is called only from rewrite_header() below, either when reading a
411 message. or when routing, in order to rewrite addresses that get changed by a
412 router. This is normally the addition of full qualification to a partial
413 domain. The first rewriting rule in this case is "change routed_old into
414 routed_new", and it applies to all header lines that contain addresses. Then
415 header-specific rewriting rules are applied.
417 Before rewriting can be done, addresses without domains have to be qualified.
418 This should only be done for messages from "local" senders. This is a difficult
419 concept to pin down, what with the use of SMTP both as a submission and as a
420 transmission protocol. Exim normally requires incoming SMTP to contain fully-
421 qualified addresses, but there are options to permit unqualified ones from
422 certain hosts. For those hosts only, addresses in headers can also be
423 qualified. For other hosts, unqualified addresses in headers do not get touched
424 in any way. For locally sourced messages, unqualified addresses always get
425 qualified, except when -bnq is used to explicitly suppress this.
428 h pointer to header line block
429 flag indicates which header this is
430 routed_old if not NULL, this is a rewrite caused by a router, changing
431 this domain into routed_new
432 routed_new new routed domain if routed_old is not NULL
433 rewrite_rules points to chain of rewriting rules
434 existflags bits indicating which rewrites exist
435 replace if TRUE, insert the new header in the chain after the old
436 one, and mark the old one "replaced"
438 Returns: NULL if header unchanged; otherwise the rewritten header
442 rewrite_one_header(header_line *h, int flag,
443 const uschar *routed_old, const uschar *routed_new,
444 rewrite_rule *rewrite_rules, int existflags, BOOL replace)
447 header_line *newh = NULL;
448 rmark function_reset_point = store_mark();
449 uschar *s = Ustrchr(h->text, ':') + 1;
451 while (isspace(*s)) s++;
454 debug_printf("rewrite_one_header: type=%c:\n %s", h->type, h->text);
456 f.parse_allow_group = TRUE; /* Allow group syntax */
458 /* Loop for multiple addresses in the header. We have to go through them all
459 in case any need qualifying, even if there's no rewriting. Pathological headers
460 may have thousands of addresses in them, so cause the store to be reset for
461 any that don't actually get rewritten. We also play silly games for those that
462 _are_ rewritten so as to avoid runaway store usage for these kinds of header.
463 We want to avoid keeping store for any intermediate versions. */
468 uschar *ss = parse_find_address_end(s, FALSE);
469 uschar *recipient, *new, *errmess;
470 rmark loop_reset_point = store_mark();
471 BOOL changed = FALSE;
472 int terminator = *ss;
473 int start, end, domain;
475 /* Temporarily terminate the string at this point, and extract the
476 operative address within. Then put back the terminator and prepare for
477 the next address, saving the start of the old one. */
480 recipient = parse_extract_address(s, &errmess, &start, &end, &domain, FALSE);
483 s = ss + (terminator ? 1 :0);
484 while (isspace(*s)) s++;
486 /* There isn't much we can do for syntactic disasters at this stage.
487 Pro tem (possibly for ever) ignore them. */
491 loop_reset_point = store_reset(loop_reset_point);
495 /* If routed_old is not NULL, this is a rewrite caused by a router,
496 consisting of changing routed_old into routed_new, and applying to all
497 headers. If the header address has no domain, it is excluded, since a router
498 rewrite affects domains only. The new value should always be fully qualified,
499 but it may be something that has an explicit re-write rule set, so we need to
500 check the configured rules subsequently as well. (Example: there's an
501 explicit rewrite turning *.foo.com into foo.com, and an address is supplied
502 as abc@xyz, which the DNS lookup turns into abc@xyz.foo.com). However, if no
503 change is made here, don't bother carrying on. */
507 if (domain <= 0 || strcmpic(recipient+domain, routed_old) != 0) continue;
508 recipient[domain-1] = 0;
509 new = string_sprintf("%s@%s", recipient, routed_new);
512 recipient[domain-1] = '@';
513 debug_printf("%s rewritten by router as %s\n", recipient, new);
519 /* This is not a router-inspired rewrite. Ensure the address is fully
520 qualified if that is permitted. If an unqualified address was received
521 from a host that isn't listed, do not continue rewriting this address.
522 Sender, From or Reply-To headers are treated as senders, the rest as
523 recipients. This matters only when there are different qualify strings. */
528 (flag & (rewrite_sender | rewrite_from | rewrite_replyto)) == 0;
529 new = rewrite_address_qualify(recipient, is_recipient);
530 changed = (new != recipient);
533 /* Can only qualify if permitted; if not, no rewrite. */
535 if (changed && ((is_recipient && !f.allow_unqualified_recipient) ||
536 (!is_recipient && !f.allow_unqualified_sender)))
538 loop_reset_point = store_reset(loop_reset_point);
543 /* If there are rewrite rules for this type of header, apply
544 them. This test is just for efficiency, to save scanning the rules
545 in cases when nothing is going to change. If any rewrite rule had the
546 "whole" flag set, adjust the pointers so that the whole address gets
547 replaced, except possibly a final \n. */
549 if (existflags & flag)
552 new = rewrite_one(recipient, flag, &whole, FALSE, NULL, rewrite_rules);
553 if (new != recipient)
560 if (sprev[end-1] == '\n') end--;
565 /* If nothing has changed, lose all dynamic store obtained in this loop, and
566 move on to the next address. We can't reset to the function start store
567 point, because we may have a rewritten line from a previous time round the
570 if (!changed) loop_reset_point = store_reset(loop_reset_point);
572 /* If the address has changed, create a new header containing the
573 rewritten address. We do not need to set the chain pointers at this
574 stage. We want to avoid using more and more memory if the header is very long
575 and contains lots and lots of rewritten addresses. Therefore, we build the
576 new text string in malloc store, then at the end we reset dynamic store
577 before copying the new header to a new block (and then freeing the malloc
578 block). The header must end up in dynamic store so that it's freed at the end
579 of receiving a message. */
584 int newlen = Ustrlen(new);
585 int oldlen = end - start;
587 header_line * prev = newh ? newh : h;
588 uschar * newt = store_get_perm(prev->slen - oldlen + newlen + 4, TRUE);
589 uschar * newtstart = newt;
591 int type = prev->type;
592 int slen = prev->slen - oldlen + newlen;
594 /* Build the new header text by copying the old and putting in the
595 replacement. This process may make the header substantially longer
596 than it was before - qualification of a list of bare addresses can
597 often do this - so we stick in a newline after the re-written address
598 if it has increased in length and ends more than 40 characters in. In
599 fact, the code is not perfect, since it does not scan for existing
600 newlines in the header, but it doesn't seem worth going to that
601 amount of trouble. */
603 Ustrncpy(newt, prev->text, sprev - prev->text + start);
604 newt += sprev - prev->text + start;
608 remlen = s - (sprev + end);
611 Ustrncpy(newt, sprev + end, remlen);
616 /* Must check that there isn't a newline here anyway; in particular, there
617 will be one at the very end of the header, where we DON'T want to insert
618 another one! The pointer s has been skipped over white space, so just
619 look back to see if the last non-space-or-tab was a newline. */
621 if (newlen > oldlen && newt - newtstart - lastnewline > 40)
624 while (p >= prev->text && (*p == ' ' || *p == '\t')) p--;
627 lastnewline = newt - newtstart;
628 Ustrcat(newt, US"\n\t");
633 /* Finally, the remaining unprocessed addresses, if any. */
637 DEBUG(D_rewrite) debug_printf("newlen=%d newtype=%c newtext:\n%s",
638 slen, type, newtstart);
640 /* Compute the length of the rest of the header line before we possibly
641 flatten a previously rewritten copy. */
643 remlen = (s - prev->text) - oldlen + newlen;
645 /* We have the new text in a malloc block. That enables us to release all
646 the memory that has been used, back to the point at which the function was
647 entered. Then set up a new header in dynamic store. This will override a
648 rewritten copy from a previous time round this loop. */
650 store_reset(function_reset_point);
651 function_reset_point = store_mark();
652 newh = store_get(sizeof(header_line), FALSE);
655 newh->text = string_copyn(newtstart, slen);
657 /* Set up for scanning the rest of the header */
659 s = newh->text + remlen;
660 DEBUG(D_rewrite) debug_printf("remainder: %s", *s ? s : US"\n");
664 f.parse_allow_group = FALSE; /* Reset group flags */
665 f.parse_found_group = FALSE;
667 /* If a rewrite happened and "replace" is true, put the new header into the
668 chain following the old one, and mark the old one as replaced. */
672 newh->next = h->next;
673 if (!newh->next) header_last = newh;
684 /*************************************************
685 * Rewrite a header line *
686 *************************************************/
688 /* This function may be passed any old header line. It must detect those which
689 contain addresses, then then apply any rewriting rules that apply. If
690 routed_old is NULL, only the configured rewriting rules are consulted.
691 Otherwise, the rewriting rule is "change routed_old into routed_new", and it
692 applies to all header lines that contain addresses. Then header-specific
693 rewriting rules are applied.
695 The old header line is flagged as "old". Old headers are saved on the spool for
696 debugging but are never sent to any recipients.
699 h header line to rewrite
700 routed_old if not NULL, this is a rewrite caused by a router, changing
701 this domain into routed_new
702 routed_new new routed domain if routed_old is not NULL
703 rewrite_rules points to chain of rewrite rules
704 existflags bits indicating which rewrites exist
705 replace if TRUE, the new header is inserted into the header chain
706 after the old one, and the old one is marked replaced
708 Returns: NULL if header unchanged; otherwise the rewritten header
712 rewrite_header(header_line *h,
713 const uschar *routed_old, const uschar *routed_new,
714 rewrite_rule *rewrite_rules, int existflags, BOOL replace)
719 case htype_sender: flag = rewrite_sender; break;
720 case htype_from: flag = rewrite_from; break;
721 case htype_to: flag = rewrite_to; break;
722 case htype_cc: flag = rewrite_cc; break;
723 case htype_bcc: flag = rewrite_bcc; break;
724 case htype_reply_to: flag = rewrite_replyto; break;
725 default: return NULL;
727 return rewrite_one_header(h, flag, routed_old, routed_new,
728 rewrite_rules, existflags, replace);
733 /************************************************
734 * Test rewriting rules *
735 ************************************************/
737 /* Called from the mainline as a result of the -brw option. Test the
738 address for all possible cases.
740 Argument: the address to test
744 void rewrite_test(uschar *s)
746 uschar *recipient, *error;
747 int start, end, domain;
748 BOOL done_smtp = FALSE;
750 if (rewrite_existflags == 0)
752 printf("No rewrite rules are defined\n");
756 /* Do SMTP rewrite only if a rule with the S flag exists. Allow <> by
757 pretending it is a sender. */
759 if ((rewrite_existflags & rewrite_smtp) != 0)
761 uschar *new = rewrite_one(s, rewrite_smtp|rewrite_smtp_sender, NULL, FALSE,
762 US"", global_rewrite_rules);
766 printf(" SMTP: <>\n");
768 printf(" SMTP: %s\n", new);
773 /* Do the other rewrites only if a rule without the S flag exists */
775 if ((rewrite_existflags & ~rewrite_smtp) == 0) return;
777 /* Qualify if necessary before extracting the address */
779 if (parse_find_at(s) == NULL)
780 s = string_sprintf("%s@%s", s, qualify_domain_recipient);
782 recipient = parse_extract_address(s, &error, &start, &end, &domain, FALSE);
787 printf("Syntax error in %s\n%c%s\n", s, toupper(error[0]), error+1);
791 for (int i = 0; i < 8; i++)
795 uschar *new = rewrite_one(recipient, flag, &whole, FALSE, US"",
796 global_rewrite_rules);
797 printf("%s: ", rrname[i]);
800 else if (whole || (flag & rewrite_all_headers) == 0)
801 printf("%s\n", CS new);
802 else printf("%.*s%s%s\n", start, s, new, s+end);
806 /* End of rewrite.c */