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(const uschar *s, BOOL is_recipient)
65 return parse_find_at(s)
66 ? s : 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(const uschar *s, int flag, BOOL *whole, BOOL add_header, uschar *name,
101 rewrite_rule *rewrite_rules)
103 const uschar *yield = s;
104 const uschar *subject = s;
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;
123 const uschar * newparsed;
125 /* Come back here for a repeat after a successful rewrite. We do this
126 only so many times. */
130 /* If this is an SMTP-time rewrite, the pattern must be a regex and
131 the subject may have any structure. No local part or domain variables
132 can be set for the expansion. We expand the pattern in order to be consistent
133 with the other kinds of rewrite, where expansion happens inside
134 match_address_list(). */
136 if (flag & rewrite_smtp)
138 uschar *key = expand_string(rule->key);
141 if (!f.expand_string_forcedfail)
142 log_write(0, LOG_MAIN|LOG_PANIC, "failed to expand \"%s\" while "
143 "checking for SMTP rewriting: %s", rule->key, expand_string_message);
146 if (match_check_string(subject, key, 0, TRUE, FALSE, FALSE, NULL) != OK)
148 new = expand_string(rule->replacement);
151 /* All other rewrites expect the input to be a valid address, so local part
152 and domain variables can be set for expansion. For the first rule, to be
153 applied to this address, domain will be NULL and needs to be set. */
157 if (!domain) domain = Ustrrchr(subject, '@') + 1;
159 /* Use the general function for matching an address against a list (here
160 just one item, so use the "impossible value" separator UCHAR_MAX+1). */
162 if (match_address_list(subject, FALSE, TRUE, CUSS &(rule->key), NULL, 0,
163 UCHAR_MAX + 1, NULL) != OK)
166 /* The source address matches, and numerical variables have been
167 set up. If the replacement string consists of precisely "*" then no
168 rewriting is required for this address - the behaviour is as for "fail"
169 in the replacement expansion, but assuming the quit flag. */
171 if (Ustrcmp(rule->replacement, "*") == 0) break;
173 /* Otherwise, expand the replacement string. Set $local_part and $domain to
174 the appropriate values, restoring whatever value they previously had
177 save_localpart = deliver_localpart;
178 save_domain = deliver_domain;
180 /* We have subject pointing to "localpart@domain" and domain pointing to
181 the domain. Temporarily terminate the local part so that it can be
182 set up as an expansion variable */
185 deliver_localpart = US subject;
186 deliver_domain = domain;
188 new = expand_string(rule->replacement);
191 deliver_localpart = save_localpart;
192 deliver_domain = save_domain;
195 /* If the expansion failed with the "forcedfail" flag, don't generate
196 an error - just give up on this rewriting rule. If the "q" flag is set,
197 give up altogether. For other expansion failures we have a configuration
202 if (f.expand_string_forcedfail)
203 { if (rule->flags & rewrite_quit) break; else continue; }
205 expand_string_message = expand_hide_passwords(expand_string_message);
207 log_write(0, LOG_MAIN|LOG_PANIC, "Expansion of %s failed while rewriting: "
208 "%s", rule->replacement, expand_string_message);
212 /* Check the what has been generated is a valid RFC 2822 address. Only
213 envelope from or SMTP sender is permitted to be rewritten as <>.*/
215 newparsed = parse_extract_address(new, &error, &start, &end, &pdomain,
216 flag == rewrite_envfrom || flag == (rewrite_smtp|rewrite_smtp_sender));
220 log_write(0, LOG_MAIN|LOG_PANIC, "Rewrite of %s yielded unparseable "
221 "address: %s in address %s", subject, error, new);
222 break; /* Give up on this address */
225 /* A non-null unqualified address can be qualified if requested. Otherwise,
226 this is an error unless it's the empty address in circumstances where that is
229 if (pdomain == 0 && (*newparsed != 0 ||
230 (flag != rewrite_envfrom && flag != (rewrite_smtp|rewrite_smtp_sender))))
232 if (rule->flags & rewrite_qualify)
234 newparsed = rewrite_address_qualify(newparsed, TRUE);
235 new = string_sprintf("%.*s%s%.*s", start, new, newparsed,
236 Ustrlen(new) - end, new + end);
237 end = start + Ustrlen(newparsed);
241 log_write(0, LOG_MAIN|LOG_PANIC, "Rewrite of %s yielded unqualified "
242 "address \"%s\"", subject, new);
243 break; /* Give up on this address */
247 /* We have a validly rewritten address */
249 if (LOGGING(address_rewrite) || (debug_selector & D_rewrite) != 0)
251 const uschar *where = CUS"?";
253 for (int i = 0; i < where_list_size; i++)
254 if (flag == where_list[i].bit)
256 where = where_list[i].string;
259 log_write(L_address_rewrite,
260 LOG_MAIN, "\"%s\" from %s rewritten as \"%s\" by rule %d",
261 yield, where, new, rule_number);
264 /* A header will only actually be added if header_last is non-NULL,
265 i.e. during message reception or delivery, but add_header should not
266 be set TRUE during delivery, as otherwise multiple instances of the header
267 can fill up the -H file and make it embarrassingly large. We don't need
268 to set header_rewritten because the -H file always gets written at the end
269 of message reception. */
272 header_add(htype_old, "X-rewrote-%s: %s\n", name, subject);
274 /* Handle the case when replacement of the whole address is possible.
275 This happens only when whole is not NULL and we are rewriting a header.
276 If *whole is already TRUE it means that a previous rule had the w
277 flag set and so we must preserve the non-active portion of the current
278 subject unless the current rule also has the w flag set. */
280 if (whole && (flag & rewrite_all_headers))
282 /* Current rule has the w flag set. We must ensure the phrase parts
283 are syntactically valid if they are present. */
285 if (rule->flags & rewrite_whole)
287 if (start > 0 && new[start-1] == '<')
289 uschar *p1 = new + start - 1;
290 uschar *p2 = new + end + 1;
291 const uschar *pf1, *pf2;
293 while (p1 > new && p1[-1] == ' ') p1--;
294 pf1 = parse_fix_phrase(new, p1 - new);
295 while (*p2 == ' ') p2++;
296 pf2 = parse_fix_phrase(p2, Ustrlen(p2));
298 start = Ustrlen(pf1) + start + new - p1;
299 end = start + Ustrlen(newparsed);
300 new = string_sprintf("%s%.*s%s", pf1, (int)(p2 - p1), p1, pf2);
303 /* Now accept the whole thing */
312 /* Current rule does not have the w flag set; if not previously
313 done any whole rewriting, behave in non-whole manner. */
315 else if (!*whole) goto NEVER_WHOLE;
317 /* Current rule does not have the w flag set, but a previous
318 rule did rewrite the whole address. Thus yield and subject will be
319 different. Preserve the previous non-active part of the address. */
324 new = string_sprintf("%.*s%s%n%s",
325 yield_start, yield, subject, &end, yield + yield_end);
331 /* Rule just rewrites active part, or handling an envelope. This
332 code is obeyed only when all rules so far have not done "whole"
338 subject = yield = newparsed;
341 domain = NULL; /* Reset for next rule */
343 /* If no further rewrites are to be done, set the done flag. This allows
344 repeats of the current rule if configured before breaking the loop. */
346 if (rule->flags & rewrite_quit) done = TRUE;
348 /* Allow the current rule to be applied up to 10 times if
351 if (rule->flags & rewrite_repeat)
353 if (count++ < 10) goto REPEAT_RULE;
354 log_write(0, LOG_MAIN|LOG_PANIC, "rewrite rule repeat ignored after 10 "
359 /* Unset expansion numeric variables, and that's it. */
367 /*************************************************
368 * Ensure qualification and rewrite *
369 *************************************************/
371 /* This function is called for envelope addresses, the boolean specifying
372 whether a recipient or a sender. It must first of all ensure the address is
373 fully qualified, and then apply any relevant re-writing rules. The add-header
374 flag causes a header to be added, recording the old address. This is marked
375 "old", so that it is never transported anywhere; it exists for local checking
376 and debugging purposes.
379 s the address to be considered
380 is_recipient TRUE for recipient addresses; FALSE otherwise
381 add_header add "X-rewrote-xxx" header when rewriting; this is
382 set TRUE only for calls from the reception functions
383 rewrite_rules points to chain of rewrite rules
384 existflags bits indicating which headers there are rewrites for
385 (just an optimisation)
387 Returns: possibly rewritten address
391 rewrite_address(const uschar *s, BOOL is_recipient, BOOL add_header,
392 rewrite_rule *rewrite_rules, int existflags)
394 int flag = is_recipient ? rewrite_envto : rewrite_envfrom;
396 s = rewrite_address_qualify(s, is_recipient);
397 if (existflags & flag)
399 const uschar *new = rewrite_one(s, flag, NULL, add_header, is_recipient?
400 US"original-recipient" : US"sender", rewrite_rules);
401 if (new != s) s = new;
408 /*************************************************
409 * Qualify and possibly rewrite one header *
410 *************************************************/
412 /* This is called only from rewrite_header() below, either when reading a
413 message. or when routing, in order to rewrite addresses that get changed by a
414 router. This is normally the addition of full qualification to a partial
415 domain. The first rewriting rule in this case is "change routed_old into
416 routed_new", and it applies to all header lines that contain addresses. Then
417 header-specific rewriting rules are applied.
419 Before rewriting can be done, addresses without domains have to be qualified.
420 This should only be done for messages from "local" senders. This is a difficult
421 concept to pin down, what with the use of SMTP both as a submission and as a
422 transmission protocol. Exim normally requires incoming SMTP to contain fully-
423 qualified addresses, but there are options to permit unqualified ones from
424 certain hosts. For those hosts only, addresses in headers can also be
425 qualified. For other hosts, unqualified addresses in headers do not get touched
426 in any way. For locally sourced messages, unqualified addresses always get
427 qualified, except when -bnq is used to explicitly suppress this.
430 h pointer to header line block
431 flag indicates which header this is
432 routed_old if not NULL, this is a rewrite caused by a router, changing
433 this domain into routed_new
434 routed_new new routed domain if routed_old is not NULL
435 rewrite_rules points to chain of rewriting rules
436 existflags bits indicating which rewrites exist
437 replace if TRUE, insert the new header in the chain after the old
438 one, and mark the old one "replaced"
440 Returns: NULL if header unchanged; otherwise the rewritten header
444 rewrite_one_header(header_line *h, int flag,
445 const uschar *routed_old, const uschar *routed_new,
446 rewrite_rule *rewrite_rules, int existflags, BOOL replace)
449 header_line *newh = NULL;
450 rmark function_reset_point = store_mark();
451 uschar *s = Ustrchr(h->text, ':') + 1;
453 while (isspace(*s)) s++;
456 debug_printf("rewrite_one_header: type=%c:\n %s", h->type, h->text);
458 f.parse_allow_group = TRUE; /* Allow group syntax */
460 /* Loop for multiple addresses in the header. We have to go through them all
461 in case any need qualifying, even if there's no rewriting. Pathological headers
462 may have thousands of addresses in them, so cause the store to be reset for
463 any that don't actually get rewritten. We also play silly games for those that
464 _are_ rewritten so as to avoid runaway store usage for these kinds of header.
465 We want to avoid keeping store for any intermediate versions. */
470 uschar *ss = parse_find_address_end(s, FALSE);
471 uschar *recipient, *new, *errmess;
472 rmark loop_reset_point = store_mark();
473 BOOL changed = FALSE;
474 int terminator = *ss;
475 int start, end, domain;
477 /* Temporarily terminate the string at this point, and extract the
478 operative address within. Then put back the terminator and prepare for
479 the next address, saving the start of the old one. */
482 recipient = parse_extract_address(s, &errmess, &start, &end, &domain, FALSE);
485 s = ss + (terminator ? 1 :0);
486 while (isspace(*s)) s++;
488 /* There isn't much we can do for syntactic disasters at this stage.
489 Pro tem (possibly for ever) ignore them. */
493 loop_reset_point = store_reset(loop_reset_point);
497 /* If routed_old is not NULL, this is a rewrite caused by a router,
498 consisting of changing routed_old into routed_new, and applying to all
499 headers. If the header address has no domain, it is excluded, since a router
500 rewrite affects domains only. The new value should always be fully qualified,
501 but it may be something that has an explicit re-write rule set, so we need to
502 check the configured rules subsequently as well. (Example: there's an
503 explicit rewrite turning *.foo.com into foo.com, and an address is supplied
504 as abc@xyz, which the DNS lookup turns into abc@xyz.foo.com). However, if no
505 change is made here, don't bother carrying on. */
509 if (domain <= 0 || strcmpic(recipient+domain, routed_old) != 0) continue;
510 recipient[domain-1] = 0;
511 new = string_sprintf("%s@%s", recipient, routed_new);
514 recipient[domain-1] = '@';
515 debug_printf("%s rewritten by router as %s\n", recipient, new);
521 /* This is not a router-inspired rewrite. Ensure the address is fully
522 qualified if that is permitted. If an unqualified address was received
523 from a host that isn't listed, do not continue rewriting this address.
524 Sender, From or Reply-To headers are treated as senders, the rest as
525 recipients. This matters only when there are different qualify strings. */
530 (flag & (rewrite_sender | rewrite_from | rewrite_replyto)) == 0;
531 /* deconst ok as recipient was notconst */
532 new = US rewrite_address_qualify(recipient, is_recipient);
533 changed = (new != recipient);
536 /* Can only qualify if permitted; if not, no rewrite. */
538 if (changed && ((is_recipient && !f.allow_unqualified_recipient) ||
539 (!is_recipient && !f.allow_unqualified_sender)))
541 loop_reset_point = store_reset(loop_reset_point);
546 /* If there are rewrite rules for this type of header, apply
547 them. This test is just for efficiency, to save scanning the rules
548 in cases when nothing is going to change. If any rewrite rule had the
549 "whole" flag set, adjust the pointers so that the whole address gets
550 replaced, except possibly a final \n. */
552 if (existflags & flag)
555 /* deconst ok as recipient was notconst */
556 new = US rewrite_one(recipient, flag, &whole, FALSE, NULL, rewrite_rules);
557 if (new != recipient)
564 if (sprev[end-1] == '\n') end--;
569 /* If nothing has changed, lose all dynamic store obtained in this loop, and
570 move on to the next address. We can't reset to the function start store
571 point, because we may have a rewritten line from a previous time round the
574 if (!changed) loop_reset_point = store_reset(loop_reset_point);
576 /* If the address has changed, create a new header containing the
577 rewritten address. We do not need to set the chain pointers at this
578 stage. We want to avoid using more and more memory if the header is very long
579 and contains lots and lots of rewritten addresses. Therefore, we build the
580 new text string in malloc store, then at the end we reset dynamic store
581 before copying the new header to a new block (and then freeing the malloc
582 block). The header must end up in dynamic store so that it's freed at the end
583 of receiving a message. */
588 int newlen = Ustrlen(new);
589 int oldlen = end - start;
591 header_line * prev = newh ? newh : h;
592 uschar * newt = store_get_perm(prev->slen - oldlen + newlen + 4, TRUE);
593 uschar * newtstart = newt;
595 int type = prev->type;
596 int slen = prev->slen - oldlen + newlen;
598 /* Build the new header text by copying the old and putting in the
599 replacement. This process may make the header substantially longer
600 than it was before - qualification of a list of bare addresses can
601 often do this - so we stick in a newline after the re-written address
602 if it has increased in length and ends more than 40 characters in. In
603 fact, the code is not perfect, since it does not scan for existing
604 newlines in the header, but it doesn't seem worth going to that
605 amount of trouble. */
607 Ustrncpy(newt, prev->text, sprev - prev->text + start);
608 newt += sprev - prev->text + start;
612 remlen = s - (sprev + end);
615 Ustrncpy(newt, sprev + end, remlen);
620 /* Must check that there isn't a newline here anyway; in particular, there
621 will be one at the very end of the header, where we DON'T want to insert
622 another one! The pointer s has been skipped over white space, so just
623 look back to see if the last non-space-or-tab was a newline. */
625 if (newlen > oldlen && newt - newtstart - lastnewline > 40)
628 while (p >= prev->text && (*p == ' ' || *p == '\t')) p--;
631 lastnewline = newt - newtstart;
632 Ustrcat(newt, US"\n\t");
637 /* Finally, the remaining unprocessed addresses, if any. */
641 DEBUG(D_rewrite) debug_printf("newlen=%d newtype=%c newtext:\n%s",
642 slen, type, newtstart);
644 /* Compute the length of the rest of the header line before we possibly
645 flatten a previously rewritten copy. */
647 remlen = (s - prev->text) - oldlen + newlen;
649 /* We have the new text in a malloc block. That enables us to release all
650 the memory that has been used, back to the point at which the function was
651 entered. Then set up a new header in dynamic store. This will override a
652 rewritten copy from a previous time round this loop. */
654 store_reset(function_reset_point);
655 function_reset_point = store_mark();
656 newh = store_get(sizeof(header_line), FALSE);
659 newh->text = string_copyn(newtstart, slen);
661 /* Set up for scanning the rest of the header */
663 s = newh->text + remlen;
664 DEBUG(D_rewrite) debug_printf("remainder: %s", *s ? s : US"\n");
668 f.parse_allow_group = FALSE; /* Reset group flags */
669 f.parse_found_group = FALSE;
671 /* If a rewrite happened and "replace" is true, put the new header into the
672 chain following the old one, and mark the old one as replaced. */
676 newh->next = h->next;
677 if (!newh->next) header_last = newh;
688 /*************************************************
689 * Rewrite a header line *
690 *************************************************/
692 /* This function may be passed any old header line. It must detect those which
693 contain addresses, then then apply any rewriting rules that apply. If
694 routed_old is NULL, only the configured rewriting rules are consulted.
695 Otherwise, the rewriting rule is "change routed_old into routed_new", and it
696 applies to all header lines that contain addresses. Then header-specific
697 rewriting rules are applied.
699 The old header line is flagged as "old". Old headers are saved on the spool for
700 debugging but are never sent to any recipients.
703 h header line to rewrite
704 routed_old if not NULL, this is a rewrite caused by a router, changing
705 this domain into routed_new
706 routed_new new routed domain if routed_old is not NULL
707 rewrite_rules points to chain of rewrite rules
708 existflags bits indicating which rewrites exist
709 replace if TRUE, the new header is inserted into the header chain
710 after the old one, and the old one is marked replaced
712 Returns: NULL if header unchanged; otherwise the rewritten header
716 rewrite_header(header_line *h,
717 const uschar *routed_old, const uschar *routed_new,
718 rewrite_rule *rewrite_rules, int existflags, BOOL replace)
723 case htype_sender: flag = rewrite_sender; break;
724 case htype_from: flag = rewrite_from; break;
725 case htype_to: flag = rewrite_to; break;
726 case htype_cc: flag = rewrite_cc; break;
727 case htype_bcc: flag = rewrite_bcc; break;
728 case htype_reply_to: flag = rewrite_replyto; break;
729 default: return NULL;
731 return rewrite_one_header(h, flag, routed_old, routed_new,
732 rewrite_rules, existflags, replace);
737 /************************************************
738 * Test rewriting rules *
739 ************************************************/
741 /* Called from the mainline as a result of the -brw option. Test the
742 address for all possible cases.
744 Argument: the address to test
749 rewrite_test(const uschar *s)
751 uschar *recipient, *error;
752 int start, end, domain;
753 BOOL done_smtp = FALSE;
755 if (rewrite_existflags == 0)
757 printf("No rewrite rules are defined\n");
761 /* Do SMTP rewrite only if a rule with the S flag exists. Allow <> by
762 pretending it is a sender. */
764 if ((rewrite_existflags & rewrite_smtp) != 0)
766 const uschar * new = rewrite_one(s, rewrite_smtp|rewrite_smtp_sender, NULL,
767 FALSE, US"", global_rewrite_rules);
771 printf(" SMTP: <>\n");
773 printf(" SMTP: %s\n", new);
778 /* Do the other rewrites only if a rule without the S flag exists */
780 if ((rewrite_existflags & ~rewrite_smtp) == 0) return;
782 /* Qualify if necessary before extracting the address */
784 if (parse_find_at(s) == NULL)
785 s = string_sprintf("%s@%s", s, qualify_domain_recipient);
787 recipient = parse_extract_address(s, &error, &start, &end, &domain, FALSE);
792 printf("Syntax error in %s\n%c%s\n", s, toupper(error[0]), error+1);
796 for (int i = 0; i < 8; i++)
800 const uschar * new = rewrite_one(recipient, flag, &whole, FALSE, US"",
801 global_rewrite_rules);
802 printf("%s: ", rrname[i]);
805 else if (whole || (flag & rewrite_all_headers) == 0)
806 printf("%s\n", CS new);
807 else printf("%.*s%s%s\n", start, s, new, s+end);
811 /* End of rewrite.c */