Commit | Line | Data |
---|---|---|
8e669ac1 | 1 | /* $Cambridge: exim/src/src/parse.c,v 1.4 2005/02/17 11:58:26 ph10 Exp $ */ |
059ec3d9 PH |
2 | |
3 | /************************************************* | |
4 | * Exim - an Internet mail transport agent * | |
5 | *************************************************/ | |
6 | ||
c988f1f4 | 7 | /* Copyright (c) University of Cambridge 1995 - 2005 */ |
059ec3d9 PH |
8 | /* See the file NOTICE for conditions of use and distribution. */ |
9 | ||
10 | /* Functions for parsing addresses */ | |
11 | ||
12 | ||
13 | #include "exim.h" | |
14 | ||
15 | ||
16 | static uschar *last_comment_position; | |
17 | ||
18 | ||
19 | ||
20 | /* In stand-alone mode, provide a replacement for deliver_make_addr() | |
21 | and rewrite_address[_qualify]() so as to avoid having to drag in too much | |
22 | redundant apparatus. */ | |
23 | ||
24 | #ifdef STAND_ALONE | |
25 | ||
26 | address_item *deliver_make_addr(uschar *address, BOOL copy) | |
27 | { | |
28 | address_item *addr = store_get(sizeof(address_item)); | |
29 | addr->next = NULL; | |
30 | addr->parent = NULL; | |
31 | addr->address = address; | |
32 | return addr; | |
33 | } | |
34 | ||
35 | uschar *rewrite_address(uschar *recipient, BOOL dummy1, BOOL dummy2, rewrite_rule | |
36 | *dummy3, int dummy4) | |
37 | { | |
38 | return recipient; | |
39 | } | |
40 | ||
41 | uschar *rewrite_address_qualify(uschar *recipient, BOOL dummy1) | |
42 | { | |
43 | return recipient; | |
44 | } | |
45 | ||
46 | #endif | |
47 | ||
48 | ||
49 | ||
50 | ||
51 | /************************************************* | |
52 | * Find the end of an address * | |
53 | *************************************************/ | |
54 | ||
55 | /* Scan over a string looking for the termination of an address at a comma, | |
56 | or end of the string. It's the source-routed addresses which cause much pain | |
57 | here. Although Exim ignores source routes, it must recognize such addresses, so | |
58 | we cannot get rid of this logic. | |
59 | ||
60 | Argument: | |
61 | s pointer to the start of an address | |
62 | nl_ends if TRUE, '\n' terminates an address | |
63 | ||
64 | Returns: pointer past the end of the address | |
65 | (i.e. points to null or comma) | |
66 | */ | |
67 | ||
68 | uschar * | |
69 | parse_find_address_end(uschar *s, BOOL nl_ends) | |
70 | { | |
71 | BOOL source_routing = *s == '@'; | |
72 | int no_term = source_routing? 1 : 0; | |
73 | ||
74 | while (*s != 0 && (*s != ',' || no_term > 0) && (*s != '\n' || !nl_ends)) | |
75 | { | |
76 | /* Skip single quoted characters. Strictly these should not occur outside | |
77 | quoted strings in RFC 822 addresses, but they can in RFC 821 addresses. Pity | |
78 | about the lack of consistency, isn't it? */ | |
79 | ||
80 | if (*s == '\\' && s[1] != 0) s += 2; | |
81 | ||
82 | /* Skip quoted items that are not inside brackets. Note that | |
83 | quoted pairs are allowed inside quoted strings. */ | |
84 | ||
85 | else if (*s == '\"') | |
86 | { | |
87 | while (*(++s) != 0 && (*s != '\n' || !nl_ends)) | |
88 | { | |
89 | if (*s == '\\' && s[1] != 0) s++; | |
90 | else if (*s == '\"') { s++; break; } | |
91 | } | |
92 | } | |
93 | ||
94 | /* Skip comments, which may include nested brackets, but quotes | |
95 | are not recognized inside comments, though quoted pairs are. */ | |
96 | ||
97 | else if (*s == '(') | |
98 | { | |
99 | int level = 1; | |
100 | while (*(++s) != 0 && (*s != '\n' || !nl_ends)) | |
101 | { | |
102 | if (*s == '\\' && s[1] != 0) s++; | |
103 | else if (*s == '(') level++; | |
104 | else if (*s == ')' && --level <= 0) { s++; break; } | |
105 | } | |
106 | } | |
107 | ||
108 | /* Non-special character; just advance. Passing the colon in a source | |
109 | routed address means that any subsequent comma or colon may terminate unless | |
110 | inside angle brackets. */ | |
111 | ||
112 | else | |
113 | { | |
114 | if (*s == '<') | |
115 | { | |
116 | source_routing = s[1] == '@'; | |
117 | no_term = source_routing? 2 : 1; | |
118 | } | |
119 | else if (*s == '>') no_term--; | |
120 | else if (source_routing && *s == ':') no_term--; | |
121 | s++; | |
122 | } | |
123 | } | |
124 | ||
125 | return s; | |
126 | } | |
127 | ||
128 | ||
129 | ||
130 | /************************************************* | |
131 | * Find last @ in an address * | |
132 | *************************************************/ | |
133 | ||
134 | /* This function is used when we have something that may not qualified. If we | |
135 | know it's qualified, searching for the rightmost '@' is sufficient. Here we | |
136 | have to be a bit more clever than just a plain search, in order to handle | |
137 | unqualified local parts like "thing@thong" correctly. Since quotes may not | |
138 | legally be part of a domain name, we can give up on hitting the first quote | |
139 | when searching from the right. Now that the parsing also permits the RFC 821 | |
140 | form of address, where quoted-pairs are allowed in unquoted local parts, we | |
141 | must take care to handle that too. | |
142 | ||
143 | Argument: pointer to an address, possibly unqualified | |
144 | Returns: pointer to the last @ in an address, or NULL if none | |
145 | */ | |
146 | ||
147 | uschar * | |
148 | parse_find_at(uschar *s) | |
149 | { | |
150 | uschar *t = s + Ustrlen(s); | |
151 | while (--t >= s) | |
152 | { | |
153 | if (*t == '@') | |
154 | { | |
155 | int backslash_count = 0; | |
156 | uschar *tt = t - 1; | |
157 | while (tt > s && *tt-- == '\\') backslash_count++; | |
158 | if ((backslash_count & 1) == 0) return t; | |
159 | } | |
160 | else if (*t == '\"') return NULL; | |
161 | } | |
162 | return NULL; | |
163 | } | |
164 | ||
165 | ||
166 | ||
167 | ||
168 | /*************************************************************************** | |
169 | * In all the functions below that read a particular object type from * | |
170 | * the input, return the new value of the pointer s (the first argument), * | |
171 | * and put the object into the store pointed to by t (the second argument), * | |
172 | * adding a terminating zero. If no object is found, t will point to zero * | |
173 | * on return. * | |
174 | ***************************************************************************/ | |
175 | ||
176 | ||
177 | /************************************************* | |
178 | * Skip white space and comment * | |
179 | *************************************************/ | |
180 | ||
181 | /* Algorithm: | |
182 | (1) Skip spaces. | |
183 | (2) If uschar not '(', return. | |
184 | (3) Skip till matching ')', not counting any characters | |
185 | escaped with '\'. | |
186 | (4) Move past ')' and goto (1). | |
187 | ||
188 | The start of the last potential comment position is remembered to | |
189 | make it possible to ignore comments at the end of compound items. | |
190 | ||
191 | Argument: current character pointer | |
192 | Regurns: new character pointer | |
193 | */ | |
194 | ||
195 | static uschar * | |
196 | skip_comment(uschar *s) | |
197 | { | |
198 | last_comment_position = s; | |
199 | while (*s) | |
200 | { | |
201 | int c, level; | |
202 | while (isspace(*s)) s++; | |
203 | if (*s != '(') break; | |
204 | level = 1; | |
205 | while((c = *(++s)) != 0) | |
206 | { | |
207 | if (c == '(') level++; | |
208 | else if (c == ')') { if (--level <= 0) { s++; break; } } | |
209 | else if (c == '\\' && s[1] != 0) s++; | |
210 | } | |
211 | } | |
212 | return s; | |
213 | } | |
214 | ||
215 | ||
216 | ||
217 | /************************************************* | |
218 | * Read a domain * | |
219 | *************************************************/ | |
220 | ||
221 | /* A domain is a sequence of subdomains, separated by dots. See comments below | |
222 | for detailed syntax of the subdomains. | |
223 | ||
224 | If allow_domain_literals is TRUE, a "domain" may also be an IP address enclosed | |
225 | in []. Make sure the output is set to the null string if there is a syntax | |
226 | error as well as if there is no domain at all. | |
227 | ||
228 | Arguments: | |
229 | s current character pointer | |
230 | t where to put the domain | |
231 | errorptr put error message here on failure (*t will be 0 on exit) | |
232 | ||
233 | Returns: new character pointer | |
234 | */ | |
235 | ||
236 | static uschar * | |
237 | read_domain(uschar *s, uschar *t, uschar **errorptr) | |
238 | { | |
239 | uschar *tt = t; | |
240 | s = skip_comment(s); | |
241 | ||
242 | /* Handle domain literals if permitted. An RFC 822 domain literal may contain | |
243 | any character except [ ] \, including linear white space, and may contain | |
244 | quoted characters. However, RFC 821 restricts literals to being dot-separated | |
245 | 3-digit numbers, and we make the obvious extension for IPv6. Go for a sequence | |
246 | of digits and dots (hex digits and colons for IPv6) here; later this will be | |
247 | checked for being a syntactically valid IP address if it ever gets to a router. | |
248 | ||
249 | If IPv6 is supported, allow both the formal form, with IPV6: at the start, and | |
250 | the informal form without it, and accept IPV4: as well, 'cause someone will use | |
251 | it sooner or later. */ | |
252 | ||
253 | if (*s == '[') | |
254 | { | |
255 | *t++ = *s++; | |
256 | ||
257 | #if HAVE_IPV6 | |
258 | if (strncmpic(s, US"IPv6:", 5) == 0 || strncmpic(s, US"IPv4:", 5) == 0) | |
259 | { | |
260 | memcpy(t, s, 5); | |
261 | t += 5; | |
262 | s += 5; | |
263 | } | |
264 | while (*s == '.' || *s == ':' || isxdigit(*s)) *t++ = *s++; | |
265 | ||
266 | #else | |
267 | while (*s == '.' || isdigit(*s)) *t++ = *s++; | |
268 | #endif | |
269 | ||
270 | if (*s == ']') *t++ = *s++; else | |
271 | { | |
272 | *errorptr = US"malformed domain literal"; | |
273 | *tt = 0; | |
274 | } | |
275 | ||
276 | if (!allow_domain_literals) | |
277 | { | |
278 | *errorptr = US"domain literals not allowed"; | |
279 | *tt = 0; | |
280 | } | |
281 | *t = 0; | |
282 | return skip_comment(s); | |
283 | } | |
284 | ||
285 | /* Handle a proper domain, which is a sequence of dot-separated atoms. Remove | |
286 | trailing dots if strip_trailing_dot is set. A subdomain is an atom. | |
287 | ||
288 | An atom is a sequence of any characters except specials, space, and controls. | |
289 | The specials are ( ) < > @ , ; : \ " . [ and ]. This is the rule for RFC 822 | |
290 | and its successor (RFC 2822). However, RFC 821 and its successor (RFC 2821) is | |
291 | tighter, allowing only letters, digits, and hyphens, not starting with a | |
292 | hyphen. | |
293 | ||
294 | There used to be a global flag that got set when checking addresses that came | |
295 | in over SMTP and which should therefore should be checked according to the | |
296 | stricter rule. However, it seems silly to make the distinction, because I don't | |
297 | suppose anybody ever uses local domains that are 822-compliant and not | |
298 | 821-compliant. Furthermore, Exim now has additional data on the spool file line | |
299 | after an address (after "one_time" processing), and it makes use of a # | |
300 | character to delimit it. When I wrote that code, I forgot about this 822-domain | |
301 | stuff, and assumed # could never appear in a domain. | |
302 | ||
303 | So the old code is now cut out for Release 4.11 onwards, on 09-Aug-02. In a few | |
304 | years, when we are sure this isn't actually causing trouble, throw it away. | |
305 | ||
306 | March 2003: the story continues: There is a camp that is arguing for the use of | |
307 | UTF-8 in domain names as the way to internationalization, and other MTAs | |
308 | support this. Therefore, we now have a flag that permits the use of characters | |
309 | with values greater than 127, encoded in UTF-8, in subdomains, so that Exim can | |
310 | be used experimentally in this way. */ | |
311 | ||
312 | for (;;) | |
313 | { | |
314 | uschar *tsave = t; | |
315 | ||
316 | /********************* | |
317 | if (rfc821_domains) | |
318 | { | |
319 | if (*s != '-') while (isalnum(*s) || *s == '-') *t++ = *s++; | |
320 | } | |
321 | else | |
322 | while (!mac_iscntrl_or_special(*s)) *t++ = *s++; | |
323 | *********************/ | |
324 | ||
325 | if (*s != '-') | |
326 | { | |
327 | /* Only letters, digits, and hyphens */ | |
328 | ||
329 | if (!allow_utf8_domains) | |
330 | { | |
331 | while (isalnum(*s) || *s == '-') *t++ = *s++; | |
332 | } | |
333 | ||
334 | /* Permit legal UTF-8 characters to be included */ | |
335 | ||
336 | else for(;;) | |
337 | { | |
338 | int i, d; | |
339 | if (isalnum(*s) || *s == '-') /* legal ascii characters */ | |
340 | { | |
341 | *t++ = *s++; | |
342 | continue; | |
343 | } | |
344 | if ((*s & 0xc0) != 0xc0) break; /* not start of UTF-8 character */ | |
345 | d = *s << 2; | |
346 | for (i = 1; i < 6; i++) /* i is the number of additional bytes */ | |
347 | { | |
348 | if ((d & 0x80) == 0) break; | |
349 | d <<= 1; | |
350 | } | |
351 | if (i == 6) goto BAD_UTF8; /* invalid UTF-8 */ | |
352 | *t++ = *s++; /* leading UTF-8 byte */ | |
353 | while (i-- > 0) /* copy and check remainder */ | |
354 | { | |
355 | if ((*s & 0xc0) != 0x80) | |
356 | { | |
357 | BAD_UTF8: | |
358 | *errorptr = US"invalid UTF-8 byte sequence"; | |
359 | *tt = 0; | |
360 | return s; | |
361 | } | |
362 | *t++ = *s++; | |
363 | } | |
364 | } /* End of loop for UTF-8 character */ | |
365 | } /* End of subdomain */ | |
366 | ||
367 | s = skip_comment(s); | |
368 | *t = 0; | |
369 | ||
370 | if (t == tsave) /* empty component */ | |
371 | { | |
372 | if (strip_trailing_dot && t > tt && *s != '.') t[-1] = 0; else | |
373 | { | |
374 | *errorptr = US"domain missing or malformed"; | |
375 | *tt = 0; | |
376 | } | |
377 | return s; | |
378 | } | |
379 | ||
380 | if (*s != '.') break; | |
381 | *t++ = *s++; | |
382 | s = skip_comment(s); | |
383 | } | |
384 | ||
385 | return s; | |
386 | } | |
387 | ||
388 | ||
389 | ||
390 | /************************************************* | |
391 | * Read a local-part * | |
392 | *************************************************/ | |
393 | ||
394 | /* A local-part is a sequence of words, separated by periods. A null word | |
395 | between dots is not strictly allowed but apparently many mailers permit it, | |
396 | so, sigh, better be compatible. Even accept a trailing dot... | |
397 | ||
398 | A <word> is either a quoted string, or an <atom>, which is a sequence | |
399 | of any characters except specials, space, and controls. The specials are | |
400 | ( ) < > @ , ; : \ " . [ and ]. In RFC 822, a single quoted character, (a | |
401 | quoted-pair) is not allowed in a word. However, in RFC 821, it is permitted in | |
402 | the local part of an address. Rather than have separate parsing functions for | |
403 | the different cases, take the liberal attitude always. At least one MUA is | |
404 | happy to recognize this case; I don't know how many other programs do. | |
405 | ||
406 | Arguments: | |
407 | s current character pointer | |
408 | t where to put the local part | |
409 | error where to point error text | |
410 | allow_null TRUE if an empty local part is not an error | |
411 | ||
412 | Returns: new character pointer | |
413 | */ | |
414 | ||
415 | static uschar * | |
416 | read_local_part(uschar *s, uschar *t, uschar **error, BOOL allow_null) | |
417 | { | |
418 | uschar *tt = t; | |
419 | *error = NULL; | |
420 | for (;;) | |
421 | { | |
422 | int c; | |
423 | uschar *tsave = t; | |
424 | s = skip_comment(s); | |
425 | ||
426 | /* Handle a quoted string */ | |
427 | ||
428 | if (*s == '\"') | |
429 | { | |
430 | *t++ = '\"'; | |
431 | while ((c = *(++s)) != 0 && c != '\"') | |
432 | { | |
433 | *t++ = c; | |
434 | if (c == '\\' && s[1] != 0) *t++ = *(++s); | |
435 | } | |
436 | if (c == '\"') | |
437 | { | |
438 | s++; | |
439 | *t++ = '\"'; | |
440 | } | |
441 | else | |
442 | { | |
443 | *error = US"unmatched doublequote in local part"; | |
444 | return s; | |
445 | } | |
446 | } | |
447 | ||
448 | /* Handle an atom, but allow quoted pairs within it. */ | |
449 | ||
450 | else while (!mac_iscntrl_or_special(*s) || *s == '\\') | |
451 | { | |
452 | c = *t++ = *s++; | |
453 | if (c == '\\' && *s != 0) *t++ = *s++; | |
454 | } | |
455 | ||
456 | /* Terminate the word and skip subsequent comment */ | |
457 | ||
458 | *t = 0; | |
459 | s = skip_comment(s); | |
460 | ||
461 | /* If we have read a null component at this point, give an error unless it is | |
462 | terminated by a dot - an extension to RFC 822 - or if it is the first | |
463 | component of the local part and an empty local part is permitted, in which | |
464 | case just return normally. */ | |
465 | ||
466 | if (t == tsave && *s != '.') | |
467 | { | |
468 | if (t == tt && !allow_null) | |
469 | *error = US"missing or malformed local part"; | |
470 | return s; | |
471 | } | |
472 | ||
473 | /* Anything other than a dot terminates the local part. Treat multiple dots | |
474 | as a single dot, as this seems to be a common extension. */ | |
475 | ||
476 | if (*s != '.') break; | |
477 | do { *t++ = *s++; } while (*s == '.'); | |
478 | } | |
479 | ||
480 | return s; | |
481 | } | |
482 | ||
483 | ||
484 | /************************************************* | |
485 | * Read route part of route-addr * | |
486 | *************************************************/ | |
487 | ||
488 | /* The pointer is at the initial "@" on entry. Return it following the | |
489 | terminating colon. Exim no longer supports the use of source routes, but it is | |
490 | required to accept the syntax. | |
491 | ||
492 | Arguments: | |
493 | s current character pointer | |
494 | t where to put the route | |
495 | errorptr where to put an error message | |
496 | ||
497 | Returns: new character pointer | |
498 | */ | |
499 | ||
500 | static uschar * | |
501 | read_route(uschar *s, uschar *t, uschar **errorptr) | |
502 | { | |
503 | BOOL commas = FALSE; | |
504 | *errorptr = NULL; | |
505 | ||
506 | while (*s == '@') | |
507 | { | |
508 | *t++ = '@'; | |
509 | s = read_domain(s+1, t, errorptr); | |
510 | if (*t == 0) return s; | |
511 | t += Ustrlen((const uschar *)t); | |
512 | if (*s != ',') break; | |
513 | *t++ = *s++; | |
514 | commas = TRUE; | |
515 | s = skip_comment(s); | |
516 | } | |
517 | ||
518 | if (*s == ':') *t++ = *s++; | |
519 | ||
520 | /* If there is no colon, and there were no commas, the most likely error | |
521 | is in fact a missing local part in the address rather than a missing colon | |
522 | after the route. */ | |
523 | ||
524 | else *errorptr = commas? | |
525 | US"colon expected after route list" : | |
526 | US"no local part"; | |
527 | ||
528 | /* Terminate the route and return */ | |
529 | ||
530 | *t = 0; | |
531 | return skip_comment(s); | |
532 | } | |
533 | ||
534 | ||
535 | ||
536 | /************************************************* | |
537 | * Read addr-spec * | |
538 | *************************************************/ | |
539 | ||
540 | /* Addr-spec is local-part@domain. We make the domain optional - | |
541 | the expected terminator for the whole thing is passed to check this. | |
542 | This function is called only when we know we have a route-addr. | |
543 | ||
544 | Arguments: | |
545 | s current character pointer | |
546 | t where to put the addr-spec | |
547 | term expected terminator (0 or >) | |
548 | errorptr where to put an error message | |
549 | domainptr set to point to the start of the domain | |
550 | ||
551 | Returns: new character pointer | |
552 | */ | |
553 | ||
554 | static uschar * | |
555 | read_addr_spec(uschar *s, uschar *t, int term, uschar **errorptr, | |
556 | uschar **domainptr) | |
557 | { | |
558 | s = read_local_part(s, t, errorptr, FALSE); | |
559 | if (*errorptr == NULL) | |
560 | { | |
561 | if (*s != term) | |
562 | { | |
563 | if (*s != '@') | |
564 | *errorptr = string_sprintf("\"@\" or \".\" expected after \"%s\"", t); | |
565 | else | |
566 | { | |
567 | t += Ustrlen((const uschar *)t); | |
568 | *t++ = *s++; | |
569 | *domainptr = t; | |
570 | s = read_domain(s, t, errorptr); | |
571 | } | |
572 | } | |
573 | } | |
574 | return s; | |
575 | } | |
576 | ||
577 | ||
578 | ||
579 | /************************************************* | |
580 | * Extract operative address * | |
581 | *************************************************/ | |
582 | ||
583 | /* This function extracts an operative address from a full RFC822 mailbox and | |
584 | returns it in a piece of dynamic store. We take the easy way and get a piece | |
585 | of store the same size as the input, and then copy into it whatever is | |
586 | necessary. If we cannot find a valid address (syntax error), return NULL, and | |
587 | point the error pointer to the reason. The arguments "start" and "end" are used | |
588 | to return the offsets of the first and one past the last characters in the | |
589 | original mailbox of the address that has been extracted, to aid in re-writing. | |
590 | The argument "domain" is set to point to the first character after "@" in the | |
591 | final part of the returned address, or zero if there is no @. | |
592 | ||
593 | Exim no longer supports the use of source routed addresses (those of the form | |
594 | @domain,...:route_addr). It recognizes the syntax, but collapses such addresses | |
595 | down to their final components. Formerly, collapse_source_routes had to be set | |
596 | to achieve this effect. RFC 1123 allows collapsing with MAY, while the revision | |
597 | of RFC 821 had increased this to SHOULD, so I've gone for it, because it makes | |
598 | a lot of code elsewhere in Exim much simpler. | |
599 | ||
600 | There are some special fudges here for handling RFC 822 group address notation | |
601 | which may appear in certain headers. If the flag parse_allow_group is set | |
602 | TRUE and parse_found_group is FALSE when this function is called, an address | |
603 | which is the start of a group (i.e. preceded by a phrase and a colon) is | |
604 | recognized; the phrase is ignored and the flag parse_found_group is set. If | |
605 | this flag is TRUE at the end of an address, then if an extraneous semicolon is | |
606 | found, it is ignored and the flag is cleared. This logic is used only when | |
607 | scanning through addresses in headers, either to fulfil the -t option or for | |
608 | rewriting or checking header syntax. | |
609 | ||
610 | Arguments: | |
611 | mailbox points to the RFC822 mailbox | |
612 | errorptr where to point an error message | |
613 | start set to start offset in mailbox | |
614 | end set to end offset in mailbox | |
615 | domain set to domain offset in result, or 0 if no domain present | |
616 | allow_null allow <> if TRUE | |
617 | ||
618 | Returns: points to the extracted address, or NULL on error | |
619 | */ | |
620 | ||
621 | #define FAILED(s) { *errorptr = s; goto PARSE_FAILED; } | |
622 | ||
623 | uschar * | |
624 | parse_extract_address(uschar *mailbox, uschar **errorptr, int *start, int *end, | |
625 | int *domain, BOOL allow_null) | |
626 | { | |
627 | uschar *yield = store_get(Ustrlen(mailbox) + 1); | |
628 | uschar *startptr, *endptr; | |
629 | uschar *s = (uschar *)mailbox; | |
630 | uschar *t = (uschar *)yield; | |
631 | ||
632 | *domain = 0; | |
633 | ||
634 | /* At the start of the string we expect either an addr-spec or a phrase | |
635 | preceding a <route-addr>. If groups are allowed, we might also find a phrase | |
636 | preceding a colon and an address. If we find an initial word followed by | |
637 | a dot, strict interpretation of the RFC would cause it to be taken | |
638 | as the start of an addr-spec. However, many mailers break the rules | |
639 | and use addresses of the form "a.n.other <ano@somewhere>" and so we | |
640 | allow this case. */ | |
641 | ||
642 | RESTART: /* Come back here after passing a group name */ | |
643 | ||
644 | s = skip_comment(s); | |
645 | startptr = s; /* In case addr-spec */ | |
646 | s = read_local_part(s, t, errorptr, TRUE); /* Dot separated words */ | |
647 | if (*errorptr != NULL) goto PARSE_FAILED; | |
648 | ||
649 | /* If the terminator is neither < nor @ then the format of the address | |
650 | must either be a bare local-part (we are now at the end), or a phrase | |
651 | followed by a route-addr (more words must follow). */ | |
652 | ||
653 | if (*s != '@' && *s != '<') | |
654 | { | |
655 | if (*s == 0 || *s == ';') | |
656 | { | |
657 | if (*t == 0) FAILED(US"empty address"); | |
658 | endptr = last_comment_position; | |
659 | goto PARSE_SUCCEEDED; /* Bare local part */ | |
660 | } | |
661 | ||
662 | /* Expect phrase route-addr, or phrase : if groups permitted, but allow | |
663 | dots in the phrase; complete the loop only when '<' or ':' is encountered - | |
664 | end of string will produce a null local_part and therefore fail. We don't | |
665 | need to keep updating t, as the phrase isn't to be kept. */ | |
666 | ||
667 | while (*s != '<' && (!parse_allow_group || *s != ':')) | |
668 | { | |
669 | s = read_local_part(s, t, errorptr, FALSE); | |
670 | if (*errorptr != NULL) | |
671 | { | |
672 | *errorptr = string_sprintf("%s (expected word or \"<\")", *errorptr); | |
673 | goto PARSE_FAILED; | |
674 | } | |
675 | } | |
676 | ||
677 | if (*s == ':') | |
678 | { | |
679 | parse_found_group = TRUE; | |
680 | parse_allow_group = FALSE; | |
681 | s++; | |
682 | goto RESTART; | |
683 | } | |
684 | ||
685 | /* Assert *s == '<' */ | |
686 | } | |
687 | ||
688 | /* At this point the next character is either '@' or '<'. If it is '@', only a | |
689 | single local-part has previously been read. An angle bracket signifies the | |
690 | start of an <addr-spec>. Throw away anything we have saved so far before | |
691 | processing it. Note that this is "if" rather than "else if" because it's also | |
692 | used after reading a preceding phrase. | |
693 | ||
694 | There are a lot of broken sendmails out there that put additional pairs of <> | |
695 | round <route-addr>s. If strip_excess_angle_brackets is set, allow any number of | |
696 | them, as long as they match. */ | |
697 | ||
698 | if (*s == '<') | |
699 | { | |
700 | uschar *domainptr = yield; | |
701 | BOOL source_routed = FALSE; | |
702 | int bracket_count = 1; | |
703 | ||
704 | s++; | |
705 | if (strip_excess_angle_brackets) | |
706 | while (*s == '<') { bracket_count++; s++; } | |
707 | ||
708 | t = yield; | |
709 | startptr = s; | |
710 | s = skip_comment(s); | |
711 | ||
712 | /* Read an optional series of routes, each of which is a domain. They | |
713 | are separated by commas and terminated by a colon. However, we totally ignore | |
714 | such routes (RFC 1123 says we MAY, and the revision of RFC 821 says we | |
715 | SHOULD). */ | |
716 | ||
717 | if (*s == '@') | |
718 | { | |
719 | s = read_route(s, t, errorptr); | |
720 | if (*errorptr != NULL) goto PARSE_FAILED; | |
721 | *t = 0; /* Ensure route is ignored - probably overkill */ | |
722 | source_routed = TRUE; | |
723 | } | |
724 | ||
725 | /* Now an addr-spec, terminated by '>'. If there is no preceding route, | |
726 | we must allow an empty addr-spec if allow_null is TRUE, to permit the | |
727 | address "<>" in some circumstances. A source-routed address MUST have | |
728 | a domain in the final part. */ | |
729 | ||
730 | if (allow_null && !source_routed && *s == '>') | |
731 | { | |
732 | *t = 0; | |
733 | *errorptr = NULL; | |
734 | } | |
735 | else | |
736 | { | |
737 | s = read_addr_spec(s, t, '>', errorptr, &domainptr); | |
738 | if (*errorptr != NULL) goto PARSE_FAILED; | |
739 | *domain = domainptr - yield; | |
740 | if (source_routed && *domain == 0) | |
741 | FAILED(US"domain missing in source-routed address"); | |
742 | } | |
743 | ||
744 | endptr = s; | |
745 | if (*errorptr != NULL) goto PARSE_FAILED; | |
746 | while (bracket_count-- > 0) if (*s++ != '>') | |
747 | { | |
748 | *errorptr = (s[-1] == 0)? US"'>' missing at end of address" : | |
749 | string_sprintf("malformed address: %.32s may not follow %.*s", | |
750 | s-1, s - (uschar *)mailbox - 1, mailbox); | |
751 | goto PARSE_FAILED; | |
752 | } | |
753 | ||
754 | s = skip_comment(s); | |
755 | } | |
756 | ||
757 | /* Hitting '@' after the first local-part means we have definitely got an | |
758 | addr-spec, on a strict reading of the RFC, and the rest of the string | |
759 | should be the domain. However, for flexibility we allow for a route-address | |
760 | not enclosed in <> as well, which is indicated by an empty first local | |
761 | part preceding '@'. The source routing is, however, ignored. */ | |
762 | ||
763 | else if (*t == 0) | |
764 | { | |
765 | uschar *domainptr = yield; | |
766 | s = read_route(s, t, errorptr); | |
767 | if (*errorptr != NULL) goto PARSE_FAILED; | |
768 | *t = 0; /* Ensure route is ignored - probably overkill */ | |
769 | s = read_addr_spec(s, t, 0, errorptr, &domainptr); | |
770 | if (*errorptr != NULL) goto PARSE_FAILED; | |
771 | *domain = domainptr - yield; | |
772 | endptr = last_comment_position; | |
773 | if (*domain == 0) FAILED(US"domain missing in source-routed address"); | |
774 | } | |
775 | ||
776 | /* This is the strict case of local-part@domain. */ | |
777 | ||
778 | else | |
779 | { | |
780 | t += Ustrlen((const uschar *)t); | |
781 | *t++ = *s++; | |
782 | *domain = t - yield; | |
783 | s = read_domain(s, t, errorptr); | |
784 | if (*t == 0) goto PARSE_FAILED; | |
785 | endptr = last_comment_position; | |
786 | } | |
787 | ||
788 | /* Use goto to get here from the bare local part case. Arrive by falling | |
789 | through for other cases. Endptr may have been moved over whitespace, so | |
790 | move it back past white space if necessary. */ | |
791 | ||
792 | PARSE_SUCCEEDED: | |
793 | if (*s != 0) | |
794 | { | |
795 | if (parse_found_group && *s == ';') | |
796 | { | |
797 | parse_found_group = FALSE; | |
798 | parse_allow_group = TRUE; | |
799 | } | |
800 | else | |
801 | { | |
802 | *errorptr = string_sprintf("malformed address: %.32s may not follow %.*s", | |
803 | s, s - (uschar *)mailbox, mailbox); | |
804 | goto PARSE_FAILED; | |
805 | } | |
806 | } | |
807 | *start = startptr - (uschar *)mailbox; /* Return offsets */ | |
808 | while (isspace(endptr[-1])) endptr--; | |
809 | *end = endptr - (uschar *)mailbox; | |
810 | ||
811 | /* Although this code has no limitation on the length of address extracted, | |
812 | other parts of Exim may have limits, and in any case, RFC 2821 limits local | |
813 | parts to 64 and domains to 255, so we do a check here, giving an error if the | |
814 | address is ridiculously long. */ | |
815 | ||
816 | if (*end - *start > ADDRESS_MAXLENGTH) | |
817 | { | |
818 | *errorptr = string_sprintf("address is ridiculously long: %.64s...", yield); | |
819 | return NULL; | |
820 | } | |
821 | ||
822 | return (uschar *)yield; | |
823 | ||
824 | /* Use goto (via the macro FAILED) to get to here from a variety of places. | |
825 | We might have an empty address in a group - the caller can choose to ignore | |
826 | this. We must, however, keep the flags correct. */ | |
827 | ||
828 | PARSE_FAILED: | |
829 | if (parse_found_group && *s == ';') | |
830 | { | |
831 | parse_found_group = FALSE; | |
832 | parse_allow_group = TRUE; | |
833 | } | |
834 | return NULL; | |
835 | } | |
836 | ||
837 | #undef FAILED | |
838 | ||
839 | ||
840 | ||
841 | /************************************************* | |
842 | * Quote according to RFC 2047 * | |
843 | *************************************************/ | |
844 | ||
845 | /* This function is used for quoting text in headers according to RFC 2047. | |
846 | If the only characters that strictly need quoting are spaces, we return the | |
847 | original string, unmodified. If a quoted string is too long for the buffer, it | |
848 | is truncated. (This shouldn't happen: this is normally handling short strings.) | |
849 | ||
8e669ac1 PH |
850 | Hmmph. As always, things get perverted for other uses. This function was |
851 | originally for the "phrase" part of addresses. Now it is being used for much | |
852 | longer texts in ACLs and via the ${rfc2047: expansion item. This means we have | |
14702f5b PH |
853 | to check for overlong "encoded-word"s and split them. November 2004. |
854 | ||
059ec3d9 PH |
855 | Arguments: |
856 | string the string to quote - already checked to contain non-printing | |
857 | chars | |
858 | len the length of the string | |
859 | charset the name of the character set; NULL => iso-8859-1 | |
860 | buffer the buffer to put the answer in | |
861 | buffer_size the size of the buffer | |
862 | ||
863 | Returns: pointer to the original string, if no quoting needed, or | |
864 | pointer to buffer containing the quoted string, or | |
865 | a pointer to "String too long" if the buffer can't even hold | |
866 | the introduction | |
867 | */ | |
868 | ||
869 | uschar * | |
870 | parse_quote_2047(uschar *string, int len, uschar *charset, uschar *buffer, | |
871 | int buffer_size) | |
872 | { | |
873 | uschar *s = string; | |
14702f5b PH |
874 | uschar *p, *t; |
875 | int hlen; | |
059ec3d9 PH |
876 | BOOL coded = FALSE; |
877 | ||
878 | if (charset == NULL) charset = US"iso-8859-1"; | |
879 | ||
880 | /* We don't expect this to fail! */ | |
881 | ||
882 | if (!string_format(buffer, buffer_size, "=?%s?Q?", charset)) | |
883 | return US"String too long"; | |
884 | ||
14702f5b PH |
885 | hlen = Ustrlen(buffer); |
886 | t = buffer + hlen; | |
887 | p = buffer; | |
888 | ||
059ec3d9 PH |
889 | for (; len > 0; len--) |
890 | { | |
891 | int ch = *s++; | |
14702f5b | 892 | if (t > buffer + buffer_size - hlen - 8) break; |
8e669ac1 | 893 | |
14702f5b PH |
894 | if (t - p > 70) |
895 | { | |
896 | *t++ = '?'; | |
897 | *t++ = '='; | |
898 | *t++ = ' '; | |
899 | p = t; | |
900 | Ustrncpy(p, buffer, hlen); | |
901 | t += hlen; | |
8e669ac1 PH |
902 | } |
903 | ||
059ec3d9 PH |
904 | if (ch < 33 || ch > 126 || |
905 | Ustrchr("?=()<>@,;:\\\".[]_", ch) != NULL) | |
906 | { | |
907 | if (ch == ' ') *t++ = '_'; else | |
908 | { | |
909 | sprintf(CS t, "=%02X", ch); | |
910 | while (*t != 0) t++; | |
911 | coded = TRUE; | |
912 | } | |
913 | } | |
914 | else *t++ = ch; | |
915 | } | |
8e669ac1 | 916 | |
14702f5b | 917 | *t++ = '?'; |
8e669ac1 | 918 | *t++ = '='; |
14702f5b | 919 | *t = 0; |
8e669ac1 | 920 | |
059ec3d9 PH |
921 | return coded? buffer : string; |
922 | } | |
923 | ||
924 | ||
925 | ||
926 | ||
927 | /************************************************* | |
928 | * Fix up an RFC 822 "phrase" * | |
929 | *************************************************/ | |
930 | ||
931 | /* This function is called to repair any syntactic defects in the "phrase" part | |
932 | of an RFC822 address. In particular, it is applied to the user's name as read | |
933 | from the passwd file when accepting a local message, and to the data from the | |
934 | -F option. | |
935 | ||
936 | If the string contains existing quoted strings or comments containing | |
937 | freestanding quotes, then we just quote those bits that need quoting - | |
938 | otherwise it would get awfully messy and probably not look good. If not, we | |
939 | quote the whole thing if necessary. Thus | |
940 | ||
941 | John Q. Smith => "John Q. Smith" | |
942 | John "Jack" Smith => John "Jack" Smith | |
943 | John "Jack" Q. Smith => John "Jack" "Q." Smith | |
944 | John (Jack) Q. Smith => "John (Jack) Q. Smith" | |
945 | John ("Jack") Q. Smith => John ("Jack") "Q." Smith | |
946 | but | |
947 | John (\"Jack\") Q. Smith => "John (\"Jack\") Q. Smith" | |
948 | ||
949 | Sheesh! This is tedious code. It is a great pity that the syntax of RFC822 is | |
950 | the way it is... | |
951 | ||
952 | August 2000: Additional code added: | |
953 | ||
954 | Previously, non-printing characters were turned into question marks, which do | |
955 | not need to be quoted. | |
956 | ||
957 | Now, a different tactic is used if there are any non-printing ASCII | |
958 | characters. The encoding method from RFC 2047 is used, assuming iso-8859-1 as | |
959 | the character set. | |
960 | ||
961 | We *could* use this for all cases, getting rid of the messy original code, | |
962 | but leave it for now. It would complicate simple cases like "John Q. Smith". | |
963 | ||
964 | The result is passed back in the buffer; it is usually going to be added to | |
965 | some other string. In order to be sure there is going to be no overflow, | |
966 | restrict the length of the input to 1/4 of the buffer size - this allows for | |
967 | every single character to be quoted or encoded without overflowing, and that | |
968 | wouldn't happen because of amalgamation. If the phrase is too long, return a | |
969 | fixed string. | |
970 | ||
971 | Arguments: | |
972 | phrase an RFC822 phrase | |
973 | len the length of the phrase | |
974 | buffer a buffer to put the result in | |
975 | buffer_size the size of the buffer | |
976 | ||
977 | Returns: the fixed RFC822 phrase | |
978 | */ | |
979 | ||
980 | uschar * | |
981 | parse_fix_phrase(uschar *phrase, int len, uschar *buffer, int buffer_size) | |
982 | { | |
983 | int ch, i; | |
984 | BOOL quoted = FALSE; | |
985 | uschar *s, *t, *end, *yield; | |
986 | ||
987 | while (len > 0 && isspace(*phrase)) { phrase++; len--; } | |
988 | if (len > buffer_size/4) return US"Name too long"; | |
989 | ||
990 | /* See if there are any non-printing characters, and if so, use the RFC 2047 | |
991 | encoding for the whole thing. */ | |
992 | ||
993 | for (i = 0, s = phrase; i < len; i++, s++) | |
994 | if ((*s < 32 && *s != '\t') || *s > 126) break; | |
995 | ||
996 | if (i < len) return parse_quote_2047(phrase, len, headers_charset, buffer, | |
997 | buffer_size); | |
998 | ||
999 | /* No non-printers; use the RFC 822 quoting rules */ | |
1000 | ||
1001 | s = phrase; | |
1002 | end = s + len; | |
1003 | yield = t = buffer + 1; | |
1004 | ||
1005 | while (s < end) | |
1006 | { | |
1007 | ch = *s++; | |
1008 | ||
1009 | /* Copy over quoted strings, remembering we encountered one */ | |
1010 | ||
1011 | if (ch == '\"') | |
1012 | { | |
1013 | *t++ = '\"'; | |
1014 | while (s < end && (ch = *s++) != '\"') | |
1015 | { | |
1016 | *t++ = ch; | |
1017 | if (ch == '\\' && s < end) *t++ = *s++; | |
1018 | } | |
1019 | *t++ = '\"'; | |
1020 | if (s >= end) break; | |
1021 | quoted = TRUE; | |
1022 | } | |
1023 | ||
1024 | /* Copy over comments, noting if they contain freestanding quote | |
1025 | characters */ | |
1026 | ||
1027 | else if (ch == '(') | |
1028 | { | |
1029 | int level = 1; | |
1030 | *t++ = '('; | |
1031 | while (s < end) | |
1032 | { | |
1033 | ch = *s++; | |
1034 | *t++ = ch; | |
1035 | if (ch == '(') level++; | |
1036 | else if (ch == ')') { if (--level <= 0) break; } | |
1037 | else if (ch == '\\' && s < end) *t++ = *s++ & 127; | |
1038 | else if (ch == '\"') quoted = TRUE; | |
1039 | } | |
1040 | if (ch == 0) | |
1041 | { | |
1042 | while (level--) *t++ = ')'; | |
1043 | break; | |
1044 | } | |
1045 | } | |
1046 | ||
1047 | /* Handle special characters that need to be quoted */ | |
1048 | ||
1049 | else if (Ustrchr(")<>@,;:\\.[]", ch) != NULL) | |
1050 | { | |
1051 | /* If hit previous quotes just make one quoted "word" */ | |
1052 | ||
1053 | if (quoted) | |
1054 | { | |
1055 | uschar *tt = t++; | |
1056 | while (*(--tt) != ' ' && *tt != '\"' && *tt != ')') tt[1] = *tt; | |
1057 | tt[1] = '\"'; | |
1058 | *t++ = ch; | |
1059 | while (s < end) | |
1060 | { | |
1061 | ch = *s++; | |
1062 | if (ch == ' ' || ch == '\"') { s--; break; } else *t++ = ch; | |
1063 | } | |
1064 | *t++ = '\"'; | |
1065 | } | |
1066 | ||
1067 | /* Else quote the whole string so far, and the rest up to any following | |
1068 | quotes. We must treat anything following a backslash as a literal. */ | |
1069 | ||
1070 | else | |
1071 | { | |
1072 | BOOL escaped = (ch == '\\'); | |
1073 | *(--yield) = '\"'; | |
1074 | *t++ = ch; | |
1075 | ||
1076 | /* Now look for the end or a quote */ | |
1077 | ||
1078 | while (s < end) | |
1079 | { | |
1080 | ch = *s++; | |
1081 | ||
1082 | /* Handle escaped pairs */ | |
1083 | ||
1084 | if (escaped) | |
1085 | { | |
1086 | *t++ = ch; | |
1087 | escaped = FALSE; | |
1088 | } | |
1089 | ||
1090 | else if (ch == '\\') | |
1091 | { | |
1092 | *t++ = ch; | |
1093 | escaped = TRUE; | |
1094 | } | |
1095 | ||
1096 | /* If hit subsequent quotes, insert our quote before any trailing | |
1097 | spaces and back up to re-handle the quote in the outer loop. */ | |
1098 | ||
1099 | else if (ch == '\"') | |
1100 | { | |
1101 | int count = 0; | |
1102 | while (t[-1] == ' ') { t--; count++; } | |
1103 | *t++ = '\"'; | |
1104 | while (count-- > 0) *t++ = ' '; | |
1105 | s--; | |
1106 | break; | |
1107 | } | |
1108 | ||
1109 | /* If hit a subsequent comment, check it for unescaped quotes, | |
1110 | and if so, end our quote before it. */ | |
1111 | ||
1112 | else if (ch == '(') | |
1113 | { | |
1114 | uschar *ss = s; /* uschar after '(' */ | |
1115 | int level = 1; | |
1116 | while(ss < end) | |
1117 | { | |
1118 | ch = *ss++; | |
1119 | if (ch == '(') level++; | |
1120 | else if (ch == ')') { if (--level <= 0) break; } | |
1121 | else if (ch == '\\' && ss+1 < end) ss++; | |
1122 | else if (ch == '\"') { quoted = TRUE; break; } | |
1123 | } | |
1124 | ||
1125 | /* Comment contains unescaped quotes; end our quote before | |
1126 | the start of the comment. */ | |
1127 | ||
1128 | if (quoted) | |
1129 | { | |
1130 | int count = 0; | |
1131 | while (t[-1] == ' ') { t--; count++; } | |
1132 | *t++ = '\"'; | |
1133 | while (count-- > 0) *t++ = ' '; | |
1134 | break; | |
1135 | } | |
1136 | ||
1137 | /* Comment does not contain unescaped quotes; include it in | |
1138 | our quote. */ | |
1139 | ||
1140 | else | |
1141 | { | |
1142 | if (ss >= end) ss--; | |
1143 | *t++ = '('; | |
1144 | Ustrncpy(t, s, ss-s); | |
1145 | t += ss-s; | |
1146 | s = ss; | |
1147 | } | |
1148 | } | |
1149 | ||
1150 | /* Not a comment or quote; include this character in our quotes. */ | |
1151 | ||
1152 | else *t++ = ch; | |
1153 | } | |
1154 | } | |
1155 | ||
1156 | /* Add a final quote if we hit the end of the string. */ | |
1157 | ||
1158 | if (s >= end) *t++ = '\"'; | |
1159 | } | |
1160 | ||
1161 | /* Non-special character; just copy it over */ | |
1162 | ||
1163 | else *t++ = ch; | |
1164 | } | |
1165 | ||
1166 | *t = 0; | |
1167 | return yield; | |
1168 | } | |
1169 | ||
1170 | ||
1171 | /************************************************* | |
1172 | * Extract addresses from a list * | |
1173 | *************************************************/ | |
1174 | ||
1175 | /* This function is called by the redirect router to scan a string containing a | |
1176 | list of addresses separated by commas (with optional white space) or by | |
1177 | newlines, and to generate a chain of address items from them. In other words, | |
1178 | to unpick data from an alias or .forward file. | |
1179 | ||
1180 | The SunOS5 documentation for alias files is not very clear on the syntax; it | |
1181 | does not say that either a comma or a newline can be used for separation. | |
1182 | However, that is the way Smail does it, so we follow suit. | |
1183 | ||
1184 | If a # character is encountered in a white space position, then characters from | |
1185 | there to the next newline are skipped. | |
1186 | ||
1187 | If an unqualified address begins with '\', just skip that character. This gives | |
1188 | compatibility with Sendmail's use of \ to prevent looping. Exim has its own | |
1189 | loop prevention scheme which handles other cases too - see the code in | |
1190 | route_address(). | |
1191 | ||
1192 | An "address" can be a specification of a file or a pipe; the latter may often | |
1193 | need to be quoted because it may contain spaces, but we don't want to retain | |
1194 | the quotes. Quotes may appear in normal addresses too, and should be retained. | |
1195 | We can distinguish between these cases, because in addresses, quotes are used | |
1196 | only for parts of the address, not the whole thing. Therefore, we remove quotes | |
1197 | from items when they entirely enclose them, but not otherwise. | |
1198 | ||
1199 | An "address" can also be of the form :include:pathname to include a list of | |
1200 | addresses contained in the specified file. | |
1201 | ||
1202 | Any unqualified addresses are qualified with and rewritten if necessary, via | |
1203 | the rewrite_address() function. | |
1204 | ||
1205 | Arguments: | |
1206 | s the list of addresses (typically a complete | |
1207 | .forward file or a list of entries in an alias file) | |
1208 | options option bits for permitting or denying various special cases; | |
1209 | not all bits are relevant here - some are for filter | |
1210 | files; those we use here are: | |
1211 | RDO_DEFER | |
1212 | RDO_FREEZE | |
1213 | RDO_FAIL | |
1214 | RDO_BLACKHOLE | |
1215 | RDO_REWRITE | |
1216 | RDO_INCLUDE | |
1217 | anchor where to hang the chain of newly-created addresses. This | |
1218 | should be initialized to NULL. | |
1219 | error where to return an error text | |
1220 | incoming domain domain of the incoming address; used to qualify unqualified | |
1221 | local parts preceded by \ | |
1222 | directory if NULL, no checks are done on :include: files | |
1223 | otherwise, included file names must start with the given | |
1224 | directory | |
1225 | syntax_errors if not NULL, it carries on after syntax errors in addresses, | |
1226 | building up a list of errors as error blocks chained on | |
1227 | here. | |
1228 | ||
1229 | Returns: FF_DELIVERED addresses extracted | |
1230 | FF_NOTDELIVERED no addresses extracted, but no errors | |
1231 | FF_BLACKHOLE :blackhole: | |
1232 | FF_DEFER :defer: | |
1233 | FF_FAIL :fail: | |
1234 | FF_INCLUDEFAIL some problem with :include:; *error set | |
1235 | FF_ERROR other problems; *error is set | |
1236 | */ | |
1237 | ||
1238 | int | |
1239 | parse_forward_list(uschar *s, int options, address_item **anchor, | |
1240 | uschar **error, uschar *incoming_domain, uschar *directory, | |
1241 | error_block **syntax_errors) | |
1242 | { | |
1243 | int count = 0; | |
1244 | ||
1245 | DEBUG(D_route) debug_printf("parse_forward_list: %s\n", s); | |
1246 | ||
1247 | for (;;) | |
1248 | { | |
1249 | int len; | |
1250 | int special = 0; | |
1251 | int specopt = 0; | |
1252 | int specbit = 0; | |
1253 | uschar *ss, *nexts; | |
1254 | address_item *addr; | |
1255 | BOOL inquote = FALSE; | |
1256 | ||
1257 | for (;;) | |
1258 | { | |
1259 | while (isspace(*s) || *s == ',') s++; | |
1260 | if (*s == '#') { while (*s != 0 && *s != '\n') s++; } else break; | |
1261 | } | |
1262 | ||
1263 | /* When we reach the end of the list, we return FF_DELIVERED if any child | |
1264 | addresses have been generated. If nothing has been generated, there are two | |
1265 | possibilities: either the list is really empty, or there were syntax errors | |
1266 | that are being skipped. (If syntax errors are not being skipped, an FF_ERROR | |
1267 | return is generated on hitting a syntax error and we don't get here.) For a | |
1268 | truly empty list we return FF_NOTDELIVERED so that the router can decline. | |
1269 | However, if the list is empty only because syntax errors were skipped, we | |
1270 | return FF_DELIVERED. */ | |
1271 | ||
1272 | if (*s == 0) | |
1273 | { | |
1274 | return (count > 0 || (syntax_errors != NULL && *syntax_errors != NULL))? | |
1275 | FF_DELIVERED : FF_NOTDELIVERED; | |
1276 | ||
1277 | /* This previous code returns FF_ERROR if nothing is generated but a | |
1278 | syntax error has been skipped. I now think it is the wrong approach, but | |
1279 | have left this here just in case, and for the record. */ | |
1280 | ||
1281 | #ifdef NEVER | |
1282 | if (count > 0) return FF_DELIVERED; /* Something was generated */ | |
1283 | ||
1284 | if (syntax_errors == NULL || /* Not skipping syntax errors, or */ | |
1285 | *syntax_errors == NULL) /* we didn't actually skip any */ | |
1286 | return FF_NOTDELIVERED; | |
1287 | ||
1288 | *error = string_sprintf("no addresses generated: syntax error in %s: %s", | |
1289 | (*syntax_errors)->text2, (*syntax_errors)->text1); | |
1290 | return FF_ERROR; | |
1291 | #endif | |
1292 | ||
1293 | } | |
1294 | ||
1295 | /* Find the end of the next address. Quoted strings in addresses may contain | |
1296 | escaped characters; I haven't found a proper specification of .forward or | |
1297 | alias files that mentions the quoting properties, but it seems right to do | |
1298 | the escaping thing in all cases, so use the function that finds the end of an | |
1299 | address. However, don't let a quoted string extend over the end of a line. */ | |
1300 | ||
1301 | ss = parse_find_address_end(s, TRUE); | |
1302 | ||
1303 | /* Remember where we finished, for starting the next one. */ | |
1304 | ||
1305 | nexts = ss; | |
1306 | ||
1307 | /* Remove any trailing spaces; we know there's at least one non-space. */ | |
1308 | ||
1309 | while (isspace((ss[-1]))) ss--; | |
1310 | ||
1311 | /* We now have s->start and ss->end of the next address. Remove quotes | |
1312 | if they completely enclose, remembering the address started with a quote | |
1313 | for handling pipes and files. Another round of removal of leading and | |
1314 | trailing spaces is then required. */ | |
1315 | ||
1316 | if (*s == '\"' && ss[-1] == '\"') | |
1317 | { | |
1318 | s++; | |
1319 | ss--; | |
1320 | inquote = TRUE; | |
1321 | while (s < ss && isspace(*s)) s++; | |
1322 | while (ss > s && isspace((ss[-1]))) ss--; | |
1323 | } | |
1324 | ||
1325 | /* Set up the length of the address. */ | |
1326 | ||
1327 | len = ss - s; | |
1328 | ||
1329 | DEBUG(D_route) | |
1330 | { | |
1331 | int save = s[len]; | |
1332 | s[len] = 0; | |
1333 | debug_printf("extract item: %s\n", s); | |
1334 | s[len] = save; | |
1335 | } | |
1336 | ||
1337 | /* Handle special addresses if permitted. If the address is :unknown: | |
1338 | ignore it - this is for backward compatibility with old alias files. You | |
1339 | don't need to use it nowadays - just generate an empty string. For :defer:, | |
1340 | :blackhole:, or :fail: we have to set up the error message and give up right | |
1341 | away. */ | |
1342 | ||
1343 | if (Ustrncmp(s, ":unknown:", len) == 0) | |
1344 | { | |
1345 | s = nexts; | |
1346 | continue; | |
1347 | } | |
1348 | ||
1349 | if (Ustrncmp(s, ":defer:", 7) == 0) | |
1350 | { special = FF_DEFER; specopt = RDO_DEFER; } /* specbit is 0 */ | |
1351 | else if (Ustrncmp(s, ":blackhole:", 11) == 0) | |
1352 | { special = FF_BLACKHOLE; specopt = specbit = RDO_BLACKHOLE; } | |
1353 | else if (Ustrncmp(s, ":fail:", 6) == 0) | |
1354 | { special = FF_FAIL; specopt = RDO_FAIL; } /* specbit is 0 */ | |
1355 | ||
1356 | if (special != 0) | |
1357 | { | |
1358 | uschar *ss = Ustrchr(s+1, ':') + 1; | |
1359 | if ((options & specopt) == specbit) | |
1360 | { | |
1361 | *error = string_sprintf("\"%.*s\" is not permitted", len, s); | |
1362 | return FF_ERROR; | |
1363 | } | |
1364 | while (*ss != 0 && isspace(*ss)) ss++; | |
1365 | while (s[len] != 0 && s[len] != '\n') len++; | |
1366 | s[len] = 0; | |
1367 | *error = string_copy(ss); | |
1368 | return special; | |
1369 | } | |
1370 | ||
1371 | /* If the address is of the form :include:pathname, read the file, and call | |
1372 | this function recursively to extract the addresses from it. If directory is | |
1373 | NULL, do no checks. Otherwise, insist that the file name starts with the | |
1374 | given directory and is a regular file. */ | |
1375 | ||
1376 | if (Ustrncmp(s, ":include:", 9) == 0) | |
1377 | { | |
1378 | uschar *filebuf; | |
1379 | uschar filename[256]; | |
1380 | uschar *t = s+9; | |
1381 | int flen = len - 9; | |
1382 | int frc; | |
1383 | struct stat statbuf; | |
1384 | address_item *last; | |
1385 | FILE *f; | |
1386 | ||
1387 | while (flen > 0 && isspace(*t)) { t++; flen--; } | |
1388 | ||
1389 | if (flen <= 0) | |
1390 | { | |
1391 | *error = string_sprintf("file name missing after :include:"); | |
1392 | return FF_ERROR; | |
1393 | } | |
1394 | ||
1395 | if (flen > 255) | |
1396 | { | |
1397 | *error = string_sprintf("included file name \"%s\" is too long", t); | |
1398 | return FF_ERROR; | |
1399 | } | |
1400 | ||
1401 | Ustrncpy(filename, t, flen); | |
1402 | filename[flen] = 0; | |
1403 | ||
1404 | /* Insist on absolute path */ | |
1405 | ||
1406 | if (filename[0]!= '/') | |
1407 | { | |
1408 | *error = string_sprintf("included file \"%s\" is not an absolute path", | |
1409 | filename); | |
1410 | return FF_ERROR; | |
1411 | } | |
1412 | ||
1413 | /* Check if include is permitted */ | |
1414 | ||
1415 | if ((options & RDO_INCLUDE) != 0) | |
1416 | { | |
1417 | *error = US"included files not permitted"; | |
1418 | return FF_ERROR; | |
1419 | } | |
1420 | ||
1421 | /* Check file name if required */ | |
1422 | ||
1423 | if (directory != NULL) | |
1424 | { | |
1425 | int len = Ustrlen(directory); | |
1426 | uschar *p = filename + len; | |
1427 | ||
1428 | if (Ustrncmp(filename, directory, len) != 0 || *p != '/') | |
1429 | { | |
1430 | *error = string_sprintf("included file %s is not in directory %s", | |
1431 | filename, directory); | |
1432 | return FF_ERROR; | |
1433 | } | |
1434 | ||
1435 | /* It is necessary to check that every component inside the directory | |
1436 | is NOT a symbolic link, in order to keep the file inside the directory. | |
1437 | This is mighty tedious. It is also not totally foolproof in that it | |
1438 | leaves the possibility of a race attack, but I don't know how to do | |
1439 | any better. */ | |
1440 | ||
1441 | while (*p != 0) | |
1442 | { | |
1443 | int temp; | |
1444 | while (*(++p) != 0 && *p != '/'); | |
1445 | temp = *p; | |
1446 | *p = 0; | |
1447 | if (Ulstat(filename, &statbuf) != 0) | |
1448 | { | |
1449 | *error = string_sprintf("failed to stat %s (component of included " | |
1450 | "file)", filename); | |
1451 | *p = temp; | |
1452 | return FF_ERROR; | |
1453 | } | |
1454 | ||
1455 | *p = temp; | |
1456 | ||
1457 | if ((statbuf.st_mode & S_IFMT) == S_IFLNK) | |
1458 | { | |
1459 | *error = string_sprintf("included file %s in the %s directory " | |
1460 | "involves a symbolic link", filename, directory); | |
1461 | return FF_ERROR; | |
1462 | } | |
1463 | } | |
1464 | } | |
1465 | ||
1466 | /* Open and stat the file */ | |
1467 | ||
1468 | if ((f = Ufopen(filename, "rb")) == NULL) | |
1469 | { | |
1470 | *error = string_open_failed(errno, "included file %s", filename); | |
1471 | return FF_INCLUDEFAIL; | |
1472 | } | |
1473 | ||
1474 | if (fstat(fileno(f), &statbuf) != 0) | |
1475 | { | |
1476 | *error = string_sprintf("failed to stat included file %s: %s", | |
1477 | filename, strerror(errno)); | |
1478 | fclose(f); | |
1479 | return FF_INCLUDEFAIL; | |
1480 | } | |
1481 | ||
1482 | /* If directory was checked, double check that we opened a regular file */ | |
1483 | ||
1484 | if (directory != NULL && (statbuf.st_mode & S_IFMT) != S_IFREG) | |
1485 | { | |
1486 | *error = string_sprintf("included file %s is not a regular file in " | |
1487 | "the %s directory", filename, directory); | |
1488 | return FF_ERROR; | |
1489 | } | |
1490 | ||
1491 | /* Get a buffer and read the contents */ | |
1492 | ||
1493 | if (statbuf.st_size > MAX_INCLUDE_SIZE) | |
1494 | { | |
1495 | *error = string_sprintf("included file %s is too big (max %d)", | |
1496 | filename, MAX_INCLUDE_SIZE); | |
1497 | return FF_ERROR; | |
1498 | } | |
1499 | ||
1500 | filebuf = store_get(statbuf.st_size + 1); | |
1501 | if (fread(filebuf, 1, statbuf.st_size, f) != statbuf.st_size) | |
1502 | { | |
1503 | *error = string_sprintf("error while reading included file %s: %s", | |
1504 | filename, strerror(errno)); | |
1505 | fclose(f); | |
1506 | return FF_ERROR; | |
1507 | } | |
1508 | filebuf[statbuf.st_size] = 0; | |
1509 | fclose(f); | |
1510 | ||
1511 | addr = NULL; | |
1512 | frc = parse_forward_list(filebuf, options, &addr, | |
1513 | error, incoming_domain, directory, syntax_errors); | |
1514 | if (frc != FF_DELIVERED && frc != FF_NOTDELIVERED) return frc; | |
1515 | ||
1516 | if (addr != NULL) | |
1517 | { | |
1518 | last = addr; | |
1519 | while (last->next != NULL) { count++; last = last->next; } | |
1520 | last->next = *anchor; | |
1521 | *anchor = addr; | |
1522 | count++; | |
1523 | } | |
1524 | } | |
1525 | ||
1526 | /* Else (not :include:) ensure address is syntactically correct and fully | |
1527 | qualified if not a pipe or a file, removing a leading \ if present on an | |
1528 | unqualified address. For pipes and files we must handle quoting. It's | |
1529 | not quite clear exactly what to do for partially quoted things, but the | |
1530 | common case of having the whole thing in quotes is straightforward. If this | |
1531 | was the case, inquote will have been set TRUE above and the quotes removed. | |
1532 | ||
1533 | There is a possible ambiguity over addresses whose local parts start with | |
1534 | a vertical bar or a slash, and the latter do in fact occur, thanks to X.400. | |
1535 | Consider a .forward file that contains the line | |
1536 | ||
1537 | /X=xxx/Y=xxx/OU=xxx/@some.gate.way | |
1538 | ||
1539 | Is this a file or an X.400 address? Does it make any difference if it is in | |
1540 | quotes? On the grounds that file names of this type are rare, Exim treats | |
1541 | something that parses as an RFC 822 address and has a domain as an address | |
1542 | rather than a file or a pipe. This is also how an address such as the above | |
1543 | would be treated if it came in from outside. */ | |
1544 | ||
1545 | else | |
1546 | { | |
1547 | int start, end, domain; | |
1548 | uschar *recipient = NULL; | |
1549 | int save = s[len]; | |
1550 | s[len] = 0; | |
1551 | ||
1552 | /* If it starts with \ and the rest of it parses as a valid mail address | |
1553 | without a domain, carry on with that address, but qualify it with the | |
1554 | incoming domain. Otherwise arrange for the address to fall through, | |
1555 | causing an error message on the re-parse. */ | |
1556 | ||
1557 | if (*s == '\\') | |
1558 | { | |
1559 | recipient = | |
1560 | parse_extract_address(s+1, error, &start, &end, &domain, FALSE); | |
1561 | if (recipient != NULL) | |
1562 | recipient = (domain != 0)? NULL : | |
1563 | string_sprintf("%s@%s", recipient, incoming_domain); | |
1564 | } | |
1565 | ||
1566 | /* Try parsing the item as an address. */ | |
1567 | ||
1568 | if (recipient == NULL) recipient = | |
1569 | parse_extract_address(s, error, &start, &end, &domain, FALSE); | |
1570 | ||
1571 | /* If item starts with / or | and is not a valid address, or there | |
1572 | is no domain, treat it as a file or pipe. If it was a quoted item, | |
1573 | remove the quoting occurrences of \ within it. */ | |
1574 | ||
1575 | if ((*s == '|' || *s == '/') && (recipient == NULL || domain == 0)) | |
1576 | { | |
1577 | uschar *t = store_get(Ustrlen(s) + 1); | |
1578 | uschar *p = t; | |
1579 | uschar *q = s; | |
1580 | while (*q != 0) | |
1581 | { | |
1582 | if (inquote) | |
1583 | { | |
1584 | *p++ = (*q == '\\')? *(++q) : *q; | |
1585 | q++; | |
1586 | } | |
1587 | else *p++ = *q++; | |
1588 | } | |
1589 | *p = 0; | |
1590 | addr = deliver_make_addr(t, TRUE); | |
1591 | setflag(addr, af_pfr); /* indicates pipe/file/reply */ | |
1592 | if (*s != '|') setflag(addr, af_file); /* indicates file */ | |
1593 | } | |
1594 | ||
1595 | /* Item must be an address. Complain if not, else qualify, rewrite and set | |
1596 | up the control block. It appears that people are in the habit of using | |
1597 | empty addresses but with comments as a way of putting comments into | |
1598 | alias and forward files. Therefore, ignore the error "empty address". | |
1599 | Mailing lists might want to tolerate syntax errors; there is therefore | |
1600 | an option to do so. */ | |
1601 | ||
1602 | else | |
1603 | { | |
1604 | if (recipient == NULL) | |
1605 | { | |
1606 | if (Ustrcmp(*error, "empty address") == 0) | |
1607 | { | |
1608 | *error = NULL; | |
1609 | s[len] = save; | |
1610 | s = nexts; | |
1611 | continue; | |
1612 | } | |
1613 | ||
1614 | if (syntax_errors != NULL) | |
1615 | { | |
1616 | error_block *e = store_get(sizeof(error_block)); | |
1617 | error_block *last = *syntax_errors; | |
1618 | if (last == NULL) *syntax_errors = e; else | |
1619 | { | |
1620 | while (last->next != NULL) last = last->next; | |
1621 | last->next = e; | |
1622 | } | |
1623 | e->next = NULL; | |
1624 | e->text1 = *error; | |
1625 | e->text2 = string_copy(s); | |
1626 | s[len] = save; | |
1627 | s = nexts; | |
1628 | continue; | |
1629 | } | |
1630 | else | |
1631 | { | |
1632 | *error = string_sprintf("%s in \"%s\"", *error, s); | |
1633 | s[len] = save; /* _after_ using it for *error */ | |
1634 | return FF_ERROR; | |
1635 | } | |
1636 | } | |
1637 | ||
1638 | /* Address was successfully parsed. Rewrite, and then make an address | |
1639 | block. */ | |
1640 | ||
1641 | recipient = ((options & RDO_REWRITE) != 0)? | |
1642 | rewrite_address(recipient, TRUE, FALSE, global_rewrite_rules, | |
1643 | rewrite_existflags) : | |
1644 | rewrite_address_qualify(recipient, TRUE); | |
1645 | addr = deliver_make_addr(recipient, TRUE); /* TRUE => copy recipient */ | |
1646 | } | |
1647 | ||
1648 | /* Restore the final character in the original data, and add to the | |
1649 | output chain. */ | |
1650 | ||
1651 | s[len] = save; | |
1652 | addr->next = *anchor; | |
1653 | *anchor = addr; | |
1654 | count++; | |
1655 | } | |
1656 | ||
1657 | /* Advance pointer for the next address */ | |
1658 | ||
1659 | s = nexts; | |
1660 | } | |
1661 | } | |
1662 | ||
1663 | ||
1664 | /************************************************* | |
1665 | ************************************************** | |
1666 | * Stand-alone test program * | |
1667 | ************************************************** | |
1668 | *************************************************/ | |
1669 | ||
1670 | #if defined STAND_ALONE | |
1671 | int main(void) | |
1672 | { | |
1673 | int start, end, domain; | |
1674 | uschar buffer[1024]; | |
1675 | uschar outbuff[1024]; | |
1676 | ||
1677 | big_buffer = store_malloc(big_buffer_size); | |
1678 | ||
1679 | /* strip_trailing_dot = TRUE; */ | |
1680 | allow_domain_literals = TRUE; | |
1681 | ||
1682 | printf("Testing parse_fix_phrase\n"); | |
1683 | ||
1684 | while (Ufgets(buffer, sizeof(buffer), stdin) != NULL) | |
1685 | { | |
1686 | buffer[Ustrlen(buffer)-1] = 0; | |
1687 | if (buffer[0] == 0) break; | |
1688 | printf("%s\n", CS parse_fix_phrase(buffer, Ustrlen(buffer), outbuff, | |
1689 | sizeof(outbuff))); | |
1690 | } | |
1691 | ||
1692 | printf("Testing parse_extract_address without group syntax and without UTF-8\n"); | |
1693 | ||
1694 | while (Ufgets(buffer, sizeof(buffer), stdin) != NULL) | |
1695 | { | |
1696 | uschar *out; | |
1697 | uschar *errmess; | |
1698 | buffer[Ustrlen(buffer) - 1] = 0; | |
1699 | if (buffer[0] == 0) break; | |
1700 | out = parse_extract_address(buffer, &errmess, &start, &end, &domain, FALSE); | |
1701 | if (out == NULL) printf("*** bad address: %s\n", errmess); else | |
1702 | { | |
1703 | uschar extract[1024]; | |
1704 | Ustrncpy(extract, buffer+start, end-start); | |
1705 | extract[end-start] = 0; | |
1706 | printf("%s %d %d %d \"%s\"\n", out, start, end, domain, extract); | |
1707 | } | |
1708 | } | |
1709 | ||
1710 | printf("Testing parse_extract_address without group syntax but with UTF-8\n"); | |
1711 | ||
1712 | allow_utf8_domains = TRUE; | |
1713 | while (Ufgets(buffer, sizeof(buffer), stdin) != NULL) | |
1714 | { | |
1715 | uschar *out; | |
1716 | uschar *errmess; | |
1717 | buffer[Ustrlen(buffer) - 1] = 0; | |
1718 | if (buffer[0] == 0) break; | |
1719 | out = parse_extract_address(buffer, &errmess, &start, &end, &domain, FALSE); | |
1720 | if (out == NULL) printf("*** bad address: %s\n", errmess); else | |
1721 | { | |
1722 | uschar extract[1024]; | |
1723 | Ustrncpy(extract, buffer+start, end-start); | |
1724 | extract[end-start] = 0; | |
1725 | printf("%s %d %d %d \"%s\"\n", out, start, end, domain, extract); | |
1726 | } | |
1727 | } | |
1728 | allow_utf8_domains = FALSE; | |
1729 | ||
1730 | printf("Testing parse_extract_address with group syntax\n"); | |
1731 | ||
1732 | parse_allow_group = TRUE; | |
1733 | while (Ufgets(buffer, sizeof(buffer), stdin) != NULL) | |
1734 | { | |
1735 | uschar *out; | |
1736 | uschar *errmess; | |
1737 | uschar *s; | |
1738 | buffer[Ustrlen(buffer) - 1] = 0; | |
1739 | if (buffer[0] == 0) break; | |
1740 | s = buffer; | |
1741 | while (*s != 0) | |
1742 | { | |
1743 | uschar *ss = parse_find_address_end(s, FALSE); | |
1744 | int terminator = *ss; | |
1745 | *ss = 0; | |
1746 | out = parse_extract_address(buffer, &errmess, &start, &end, &domain, FALSE); | |
1747 | *ss = terminator; | |
1748 | ||
1749 | if (out == NULL) printf("*** bad address: %s\n", errmess); else | |
1750 | { | |
1751 | uschar extract[1024]; | |
1752 | Ustrncpy(extract, buffer+start, end-start); | |
1753 | extract[end-start] = 0; | |
1754 | printf("%s %d %d %d \"%s\"\n", out, start, end, domain, extract); | |
1755 | } | |
1756 | ||
1757 | s = ss + (terminator? 1:0); | |
1758 | while (isspace(*s)) s++; | |
1759 | } | |
1760 | } | |
1761 | ||
1762 | printf("Testing parse_find_at\n"); | |
1763 | ||
1764 | while (Ufgets(buffer, sizeof(buffer), stdin) != NULL) | |
1765 | { | |
1766 | uschar *s; | |
1767 | buffer[Ustrlen(buffer)-1] = 0; | |
1768 | if (buffer[0] == 0) break; | |
1769 | s = parse_find_at(buffer); | |
1770 | if (s == NULL) printf("no @ found\n"); | |
1771 | else printf("offset = %d\n", s - buffer); | |
1772 | } | |
1773 | ||
1774 | printf("Testing parse_extract_addresses\n"); | |
1775 | ||
1776 | while (Ufgets(buffer, sizeof(buffer), stdin) != NULL) | |
1777 | { | |
1778 | uschar *errmess; | |
1779 | int extracted; | |
1780 | address_item *anchor = NULL; | |
1781 | buffer[Ustrlen(buffer) - 1] = 0; | |
1782 | if (buffer[0] == 0) break; | |
1783 | if ((extracted = parse_forward_list(buffer, -1, &anchor, | |
1784 | &errmess, US"incoming.domain", NULL, NULL)) == FF_DELIVERED) | |
1785 | { | |
1786 | while (anchor != NULL) | |
1787 | { | |
1788 | address_item *addr = anchor; | |
1789 | anchor = anchor->next; | |
1790 | printf("%d %s\n", testflag(addr, af_pfr), addr->address); | |
1791 | } | |
1792 | } | |
1793 | else printf("Failed: %d %s\n", extracted, errmess); | |
1794 | } | |
1795 | ||
1796 | return 0; | |
1797 | } | |
1798 | ||
1799 | #endif | |
1800 | ||
1801 | /* End of parse.c */ |