Added the mxh lookup type for dnsdb lookups.
[exim.git] / src / src / receive.c
1 /* $Cambridge: exim/src/src/receive.c,v 1.4 2004/11/17 14:32:25 ph10 Exp $ */
2
3 /*************************************************
4 * Exim - an Internet mail transport agent *
5 *************************************************/
6
7 /* Copyright (c) University of Cambridge 1995 - 2004 */
8 /* See the file NOTICE for conditions of use and distribution. */
9
10 /* Code for receiving a message and setting up spool files. */
11
12
13 #include "exim.h"
14
15
16
17 /*************************************************
18 * Local static variables *
19 *************************************************/
20
21 static FILE *data_file = NULL;
22 static int data_fd = -1;
23 static uschar spool_name[256];
24
25
26
27 /*************************************************
28 * Non-SMTP character reading functions *
29 *************************************************/
30
31 /* These are the default functions that are set up in the variables such as
32 receive_getc initially. They just call the standard functions, passing stdin as
33 the file. (When SMTP input is occurring, different functions are used by
34 changing the pointer variables.) */
35
36 int
37 stdin_getc(void)
38 {
39 return getc(stdin);
40 }
41
42 int
43 stdin_ungetc(int c)
44 {
45 return ungetc(c, stdin);
46 }
47
48 int
49 stdin_feof(void)
50 {
51 return feof(stdin);
52 }
53
54 int
55 stdin_ferror(void)
56 {
57 return ferror(stdin);
58 }
59
60
61
62
63 /*************************************************
64 * Check that a set sender is allowed *
65 *************************************************/
66
67 /* This function is called when a local caller sets an explicit sender address.
68 It checks whether this is permitted, which it is for trusted callers.
69 Otherwise, it must match the pattern(s) in untrusted_set_sender.
70
71 Arguments: the proposed sender address
72 Returns: TRUE for a trusted caller
73 TRUE if the address has been set, untrusted_set_sender has been
74 set, and the address matches something in the list
75 FALSE otherwise
76 */
77
78 BOOL
79 receive_check_set_sender(uschar *newsender)
80 {
81 uschar *qnewsender;
82 if (trusted_caller) return TRUE;
83 if (newsender == NULL || untrusted_set_sender == NULL) return FALSE;
84 qnewsender = (Ustrchr(newsender, '@') != NULL)?
85 newsender : string_sprintf("%s@%s", newsender, qualify_domain_sender);
86 return
87 match_address_list(qnewsender, TRUE, TRUE, &untrusted_set_sender, NULL, -1,
88 0, NULL) == OK;
89 }
90
91
92
93
94 /*************************************************
95 * Read space info for a partition *
96 *************************************************/
97
98 /* This function is called by receive_check_fs() below, and also by string
99 expansion for variables such as $spool_space. The field names for the statvfs
100 structure are macros, because not all OS have F_FAVAIL and it seems tidier to
101 have macros for F_BAVAIL and F_FILES as well. Some kinds of file system do not
102 have inodes, and they return -1 for the number available.
103
104 Later: It turns out that some file systems that do not have the concept of
105 inodes return 0 rather than -1. Such systems should also return 0 for the total
106 number of inodes, so we require that to be greater than zero before returning
107 an inode count.
108
109 Arguments:
110 isspool TRUE for spool partition, FALSE for log partition
111 inodeptr address of int to receive inode count; -1 if there isn't one
112
113 Returns: available on-root space, in kilobytes
114 -1 for log partition if there isn't one
115
116 All values are -1 if the STATFS functions are not available.
117 */
118
119 int
120 receive_statvfs(BOOL isspool, int *inodeptr)
121 {
122 #ifdef HAVE_STATFS
123 struct STATVFS statbuf;
124 uschar *path;
125 uschar *name;
126 uschar buffer[1024];
127
128 /* The spool directory must always exist. */
129
130 if (isspool)
131 {
132 path = spool_directory;
133 name = US"spool";
134 }
135
136 /* Need to cut down the log file path to the directory, and to ignore any
137 appearance of "syslog" in it. */
138
139 else
140 {
141 int sep = ':'; /* Not variable - outside scripts use */
142 uschar *p = log_file_path;
143 name = US"log";
144
145 /* An empty log_file_path means "use the default". This is the same as an
146 empty item in a list. */
147
148 if (*p == 0) p = US":";
149 while ((path = string_nextinlist(&p, &sep, buffer, sizeof(buffer))) != NULL)
150 {
151 if (Ustrcmp(path, "syslog") != 0) break;
152 }
153
154 if (path == NULL) /* No log files */
155 {
156 *inodeptr = -1;
157 return -1;
158 }
159
160 /* An empty string means use the default, which is in the spool directory.
161 But don't just use the spool directory, as it is possible that the log
162 subdirectory has been symbolically linked elsewhere. */
163
164 if (path[0] == 0)
165 {
166 sprintf(CS buffer, CS"%s/log", CS spool_directory);
167 path = buffer;
168 }
169 else
170 {
171 uschar *cp;
172 if ((cp = Ustrrchr(path, '/')) != NULL) *cp = 0;
173 }
174 }
175
176 /* We now have the patch; do the business */
177
178 memset(&statbuf, 0, sizeof(statbuf));
179
180 if (STATVFS(CS path, &statbuf) != 0)
181 {
182 log_write(0, LOG_MAIN|LOG_PANIC, "cannot accept message: failed to stat "
183 "%s directory %s: %s", name, spool_directory, strerror(errno));
184 smtp_closedown(US"spool or log directory problem");
185 exim_exit(EXIT_FAILURE);
186 }
187
188 *inodeptr = (statbuf.F_FILES > 0)? statbuf.F_FAVAIL : -1;
189
190 /* Disks are getting huge. Take care with computing the size in kilobytes. */
191
192 return (int)(((double)statbuf.F_BAVAIL * (double)statbuf.F_FRSIZE)/1024.0);
193
194 /* Unable to find partition sizes in this environment. */
195
196 #else
197 *inodeptr = -1;
198 return -1;
199 #endif
200 }
201
202
203
204
205 /*************************************************
206 * Check space on spool and log partitions *
207 *************************************************/
208
209 /* This function is called before accepting a message; if any thresholds are
210 set, it checks them. If a message_size is supplied, it checks that there is
211 enough space for that size plus the threshold - i.e. that the message won't
212 reduce the space to the threshold. Not all OS have statvfs(); for those that
213 don't, this function always returns TRUE. For some OS the old function and
214 struct name statfs is used; that is handled by a macro, defined in exim.h.
215
216 Arguments:
217 msg_size the (estimated) size of an incoming message
218
219 Returns: FALSE if there isn't enough space, or if the information cannot
220 be obtained
221 TRUE if no check was done or there is enough space
222 */
223
224 BOOL
225 receive_check_fs(int msg_size)
226 {
227 int space, inodes;
228
229 if (check_spool_space > 0 || msg_size > 0 || check_spool_inodes > 0)
230 {
231 space = receive_statvfs(TRUE, &inodes);
232
233 DEBUG(D_receive)
234 debug_printf("spool directory space = %dK inodes = %d "
235 "check_space = %dK inodes = %d msg_size = %d\n",
236 space, inodes, check_spool_space, check_spool_inodes, msg_size);
237
238 if ((space >= 0 && space < check_spool_space) ||
239 (inodes >= 0 && inodes < check_spool_inodes))
240 {
241 log_write(0, LOG_MAIN, "spool directory space check failed: space=%d "
242 "inodes=%d", space, inodes);
243 return FALSE;
244 }
245 }
246
247 if (check_log_space > 0 || check_log_inodes > 0)
248 {
249 space = receive_statvfs(FALSE, &inodes);
250
251 DEBUG(D_receive)
252 debug_printf("log directory space = %dK inodes = %d "
253 "check_space = %dK inodes = %d\n",
254 space, inodes, check_log_space, check_log_inodes);
255
256 if ((space >= 0 && space < check_log_space) ||
257 (inodes >= 0 && inodes < check_log_inodes))
258 {
259 log_write(0, LOG_MAIN, "log directory space check failed: space=%d "
260 "inodes=%d", space, inodes);
261 return FALSE;
262 }
263 }
264
265 return TRUE;
266 }
267
268
269
270 /*************************************************
271 * Bomb out while reading a message *
272 *************************************************/
273
274 /* The common case of wanting to bomb out is if a SIGTERM or SIGINT is
275 received, or if there is a timeout. A rarer case might be if the log files are
276 screwed up and Exim can't open them to record a message's arrival. Handling
277 that case is done by setting a flag to cause the log functions to call this
278 function if there is an ultimate disaster. That is why it is globally
279 accessible.
280
281 Arguments: SMTP response to give if in an SMTP session
282 Returns: it doesn't
283 */
284
285 void
286 receive_bomb_out(uschar *msg)
287 {
288 /* If spool_name is set, it contains the name of the data file that is being
289 written. Unlink it before closing so that it cannot be picked up by a delivery
290 process. Ensure that any header file is also removed. */
291
292 if (spool_name[0] != 0)
293 {
294 Uunlink(spool_name);
295 spool_name[Ustrlen(spool_name) - 1] = 'H';
296 Uunlink(spool_name);
297 }
298
299 /* Now close the file if it is open, either as a fd or a stream. */
300
301 if (data_file != NULL) fclose(data_file);
302 else if (data_fd >= 0) close(data_fd);
303
304 /* Attempt to close down an SMTP connection tidily. */
305
306 if (smtp_input)
307 {
308 if (!smtp_batched_input)
309 {
310 smtp_printf("421 %s %s - closing connection.\r\n", smtp_active_hostname,
311 msg);
312 mac_smtp_fflush();
313 }
314
315 /* Control does not return from moan_smtp_batch(). */
316
317 else moan_smtp_batch(NULL, "421 %s - message abandoned", msg);
318 }
319
320 /* Exit from the program (non-BSMTP cases) */
321
322 exim_exit(EXIT_FAILURE);
323 }
324
325
326 /*************************************************
327 * Data read timeout *
328 *************************************************/
329
330 /* Handler function for timeouts that occur while reading the data that
331 comprises a message.
332
333 Argument: the signal number
334 Returns: nothing
335 */
336
337 static void
338 data_timeout_handler(int sig)
339 {
340 uschar *msg = NULL;
341
342 sig = sig; /* Keep picky compilers happy */
343
344 if (smtp_input)
345 {
346 msg = US"SMTP incoming data timeout";
347 log_write(L_lost_incoming_connection,
348 LOG_MAIN, "SMTP data timeout (message abandoned) on connection "
349 "from %s",
350 (sender_fullhost != NULL)? sender_fullhost : US"local process");
351 }
352 else
353 {
354 fprintf(stderr, "exim: timed out while reading - message abandoned\n");
355 log_write(L_lost_incoming_connection,
356 LOG_MAIN, "timed out while reading local message");
357 }
358
359 receive_bomb_out(msg); /* Does not return */
360 }
361
362
363
364 /*************************************************
365 * local_scan() timeout *
366 *************************************************/
367
368 /* Handler function for timeouts that occur while running a local_scan()
369 function.
370
371 Argument: the signal number
372 Returns: nothing
373 */
374
375 static void
376 local_scan_timeout_handler(int sig)
377 {
378 sig = sig; /* Keep picky compilers happy */
379 log_write(0, LOG_MAIN|LOG_REJECT, "local_scan() function timed out - "
380 "message temporarily rejected (size %d)", message_size);
381 receive_bomb_out(US"local verification problem"); /* Does not return */
382 }
383
384
385
386 /*************************************************
387 * local_scan() crashed *
388 *************************************************/
389
390 /* Handler function for signals that occur while running a local_scan()
391 function.
392
393 Argument: the signal number
394 Returns: nothing
395 */
396
397 static void
398 local_scan_crash_handler(int sig)
399 {
400 log_write(0, LOG_MAIN|LOG_REJECT, "local_scan() function crashed with "
401 "signal %d - message temporarily rejected (size %d)", sig, message_size);
402 receive_bomb_out(US"local verification problem"); /* Does not return */
403 }
404
405
406 /*************************************************
407 * SIGTERM or SIGINT received *
408 *************************************************/
409
410 /* Handler for SIGTERM or SIGINT signals that occur while reading the
411 data that comprises a message.
412
413 Argument: the signal number
414 Returns: nothing
415 */
416
417 static void
418 data_sigterm_sigint_handler(int sig)
419 {
420 uschar *msg = NULL;
421
422 if (smtp_input)
423 {
424 msg = US"Service not available - SIGTERM or SIGINT received";
425 log_write(0, LOG_MAIN, "%s closed after %s", smtp_get_connection_info(),
426 (sig == SIGTERM)? "SIGTERM" : "SIGINT");
427 }
428 else
429 {
430 if (filter_test == NULL)
431 {
432 fprintf(stderr, "\nexim: %s received - message abandoned\n",
433 (sig == SIGTERM)? "SIGTERM" : "SIGINT");
434 log_write(0, LOG_MAIN, "%s received while reading local message",
435 (sig == SIGTERM)? "SIGTERM" : "SIGINT");
436 }
437 }
438
439 receive_bomb_out(msg); /* Does not return */
440 }
441
442
443
444 /*************************************************
445 * Add new recipient to list *
446 *************************************************/
447
448 /* This function builds a list of recipient addresses in argc/argv
449 format.
450
451 Arguments:
452 recipient the next address to add to recipients_list
453 pno parent number for fixed aliases; -1 otherwise
454
455 Returns: nothing
456 */
457
458 void
459 receive_add_recipient(uschar *recipient, int pno)
460 {
461 if (recipients_count >= recipients_list_max)
462 {
463 recipient_item *oldlist = recipients_list;
464 int oldmax = recipients_list_max;
465 recipients_list_max = recipients_list_max? 2*recipients_list_max : 50;
466 recipients_list = store_get(recipients_list_max * sizeof(recipient_item));
467 if (oldlist != NULL)
468 memcpy(recipients_list, oldlist, oldmax * sizeof(recipient_item));
469 }
470
471 recipients_list[recipients_count].address = recipient;
472 recipients_list[recipients_count].pno = pno;
473 recipients_list[recipients_count++].errors_to = NULL;
474 }
475
476
477
478
479 /*************************************************
480 * Remove a recipient from the list *
481 *************************************************/
482
483 /* This function is provided for local_scan() to use.
484
485 Argument:
486 recipient address to remove
487
488 Returns: TRUE if it did remove something; FALSE otherwise
489 */
490
491 BOOL
492 receive_remove_recipient(uschar *recipient)
493 {
494 int count;
495 DEBUG(D_receive) debug_printf("receive_remove_recipient(\"%s\") called\n",
496 recipient);
497 for (count = 0; count < recipients_count; count++)
498 {
499 if (Ustrcmp(recipients_list[count].address, recipient) == 0)
500 {
501 if ((--recipients_count - count) > 0)
502 memmove(recipients_list + count, recipients_list + count + 1,
503 (recipients_count - count)*sizeof(recipient_item));
504 return TRUE;
505 }
506 }
507 return FALSE;
508 }
509
510
511
512
513
514 /*************************************************
515 * Read data portion of a non-SMTP message *
516 *************************************************/
517
518 /* This function is called to read the remainder of a message (following the
519 header) when the input is not from SMTP - we are receiving a local message on
520 a standard input stream. The message is always terminated by EOF, and is also
521 terminated by a dot on a line by itself if the flag dot_ends is TRUE. Split the
522 two cases for maximum efficiency.
523
524 Ensure that the body ends with a newline. This will naturally be the case when
525 the termination is "\n.\n" but may not be otherwise. The RFC defines messages
526 as "sequences of lines" - this of course strictly applies only to SMTP, but
527 deliveries into BSD-type mailbox files also require it. Exim used to have a
528 flag for doing this at delivery time, but as it was always set for all
529 transports, I decided to simplify things by putting the check here instead.
530
531 There is at least one MUA (dtmail) that sends CRLF via this interface, and
532 other programs are known to do this as well. Exim used to have a option for
533 dealing with this: in July 2003, after much discussion, the code has been
534 changed to default to treat any of LF, CRLF, and bare CR as line terminators.
535
536 However, for the case when a dot on a line by itself terminates a message, the
537 only recognized terminating sequences before and after the dot are LF and CRLF.
538 Otherwise, having read EOL . CR, you don't know whether to read another
539 character or not.
540
541 Internally, in messages stored in Exim's spool files, LF is used as the line
542 terminator. Under the new regime, bare CRs will no longer appear in these
543 files.
544
545 Arguments:
546 fout a FILE to which to write the message
547
548 Returns: One of the END_xxx values indicating why it stopped reading
549 */
550
551 static int
552 read_message_data(FILE *fout)
553 {
554 int ch_state;
555 register int ch;
556
557 /* Handle the case when only EOF terminates the message */
558
559 if (!dot_ends)
560 {
561 register int last_ch = '\n';
562
563 for (; (ch = (receive_getc)()) != EOF; last_ch = ch)
564 {
565 if (ch == 0) body_zerocount++;
566 if (last_ch == '\r' && ch != '\n')
567 {
568 if (fputc('\n', fout) == EOF) return END_WERROR;
569 message_size++;
570 body_linecount++;
571 }
572 if (ch == '\r') continue;
573
574 if (fputc(ch, fout) == EOF) return END_WERROR;
575 if (ch == '\n') body_linecount++;
576 if (++message_size > thismessage_size_limit) return END_SIZE;
577 }
578
579 if (last_ch != '\n')
580 {
581 if (fputc('\n', fout) == EOF) return END_WERROR;
582 message_size++;
583 body_linecount++;
584 }
585
586 return END_EOF;
587 }
588
589 /* Handle the case when a dot on a line on its own, or EOF, terminates. */
590
591 ch_state = 1;
592
593 while ((ch = (receive_getc)()) != EOF)
594 {
595 if (ch == 0) body_zerocount++;
596 switch (ch_state)
597 {
598 case 0: /* Normal state (previous char written) */
599 if (ch == '\n')
600 { body_linecount++; ch_state = 1; }
601 else if (ch == '\r')
602 { ch_state = 2; continue; }
603 break;
604
605 case 1: /* After written "\n" */
606 if (ch == '.') { ch_state = 3; continue; }
607 if (ch != '\n') ch_state = 0;
608 break;
609
610 case 2:
611 body_linecount++; /* After unwritten "\r" */
612 if (ch == '\n')
613 { ch_state = 1; }
614 else
615 {
616 if (message_size++, fputc('\n', fout) == EOF) return END_WERROR;
617 if (ch == '\r') continue;
618 ch_state = 0;
619 }
620 break;
621
622 case 3: /* After "\n." (\n written, dot not) */
623 if (ch == '\n') return END_DOT;
624 if (ch == '\r') { ch_state = 4; continue; }
625 message_size++;
626 if (fputc('.', fout) == EOF) return END_WERROR;
627 ch_state = 0;
628 break;
629
630 case 4: /* After "\n.\r" (\n written, rest not) */
631 if (ch == '\n') return END_DOT;
632 message_size += 2;
633 body_linecount++;
634 if (fputs(".\n", fout) == EOF) return END_WERROR;
635 if (ch == '\r') { ch_state = 2; continue; }
636 ch_state = 0;
637 break;
638 }
639
640 if (fputc(ch, fout) == EOF) return END_WERROR;
641 if (++message_size > thismessage_size_limit) return END_SIZE;
642 }
643
644 /* Get here if EOF read. Unless we have just written "\n", we need to ensure
645 the message ends with a newline, and we must also write any characters that
646 were saved up while testing for an ending dot. */
647
648 if (ch_state != 1)
649 {
650 static uschar *ends[] = { US"\n", NULL, US"\n", US".\n", US".\n" };
651 if (fputs(CS ends[ch_state], fout) == EOF) return END_WERROR;
652 message_size += Ustrlen(ends[ch_state]);
653 body_linecount++;
654 }
655
656 return END_EOF;
657 }
658
659
660
661
662 /*************************************************
663 * Read data portion of an SMTP message *
664 *************************************************/
665
666 /* This function is called to read the remainder of an SMTP message (after the
667 headers), or to skip over it when an error has occurred. In this case, the
668 output file is passed as NULL.
669
670 If any line begins with a dot, that character is skipped. The input should only
671 be successfully terminated by CR LF . CR LF unless it is local (non-network)
672 SMTP, in which case the CRs are optional, but...
673
674 FUDGE: It seems that sites on the net send out messages with just LF
675 terminators, despite the warnings in the RFCs, and other MTAs handle this. So
676 we make the CRs optional in all cases.
677
678 July 2003: Bare CRs cause trouble. We now treat them as line terminators as
679 well, so that there are no CRs in spooled messages. However, the message
680 terminating dot is not recognized between two bare CRs.
681
682 Arguments:
683 fout a FILE to which to write the message; NULL if skipping
684
685 Returns: One of the END_xxx values indicating why it stopped reading
686 */
687
688 static int
689 read_message_data_smtp(FILE *fout)
690 {
691 int ch_state = 0;
692 register int ch;
693
694 while ((ch = (receive_getc)()) != EOF)
695 {
696 if (ch == 0) body_zerocount++;
697 switch (ch_state)
698 {
699 case 0: /* After LF or CRLF */
700 if (ch == '.')
701 {
702 ch_state = 3;
703 continue; /* Don't ever write . after LF */
704 }
705 ch_state = 1;
706
707 /* Else fall through to handle as normal uschar. */
708
709 case 1: /* Normal state */
710 if (ch == '\n')
711 {
712 ch_state = 0;
713 body_linecount++;
714 }
715 else if (ch == '\r')
716 {
717 ch_state = 2;
718 continue;
719 }
720 break;
721
722 case 2: /* After (unwritten) CR */
723 body_linecount++;
724 if (ch == '\n')
725 {
726 ch_state = 0;
727 }
728 else
729 {
730 message_size++;
731 if (fout != NULL && fputc('\n', fout) == EOF) return END_WERROR;
732 if (ch != '\r') ch_state = 1; else continue;
733 }
734 break;
735
736 case 3: /* After [CR] LF . */
737 if (ch == '\n')
738 return END_DOT;
739 if (ch == '\r')
740 {
741 ch_state = 4;
742 continue;
743 }
744 ch_state = 1; /* The dot itself is removed */
745 break;
746
747 case 4: /* After [CR] LF . CR */
748 if (ch == '\n') return END_DOT;
749 message_size++;
750 body_linecount++;
751 if (fout != NULL && fputc('\n', fout) == EOF) return END_WERROR;
752 if (ch == '\r')
753 {
754 ch_state = 2;
755 continue;
756 }
757 ch_state = 1;
758 break;
759 }
760
761 /* Add the character to the spool file, unless skipping; then loop for the
762 next. */
763
764 message_size++;
765 if (fout != NULL)
766 {
767 if (fputc(ch, fout) == EOF) return END_WERROR;
768 if (message_size > thismessage_size_limit) return END_SIZE;
769 }
770 }
771
772 /* Fall through here if EOF encountered. This indicates some kind of error,
773 since a correct message is terminated by [CR] LF . [CR] LF. */
774
775 return END_EOF;
776 }
777
778
779
780
781 /*************************************************
782 * Swallow SMTP message *
783 *************************************************/
784
785 /* This function is called when there has been some kind of error while reading
786 an SMTP message, and the remaining data may need to be swallowed. It is global
787 because it is called from smtp_closedown() to shut down an incoming call
788 tidily.
789
790 Argument: a FILE from which to read the message
791 Returns: nothing
792 */
793
794 void
795 receive_swallow_smtp(void)
796 {
797 if (message_ended >= END_NOTENDED)
798 message_ended = read_message_data_smtp(NULL);
799 }
800
801
802
803 /*************************************************
804 * Handle lost SMTP connection *
805 *************************************************/
806
807 /* This function logs connection loss incidents and generates an appropriate
808 SMTP response.
809
810 Argument: additional data for the message
811 Returns: the SMTP response
812 */
813
814 static uschar *
815 handle_lost_connection(uschar *s)
816 {
817 log_write(L_lost_incoming_connection | L_smtp_connection, LOG_MAIN,
818 "%s lost while reading message data%s", smtp_get_connection_info(), s);
819 return US"421 Lost incoming connection";
820 }
821
822
823
824
825 /*************************************************
826 * Handle a non-smtp reception error *
827 *************************************************/
828
829 /* This function is called for various errors during the reception of non-SMTP
830 messages. It either sends a message to the sender of the problem message, or it
831 writes to the standard error stream.
832
833 Arguments:
834 errcode code for moan_to_sender(), identifying the error
835 text1 first message text, passed to moan_to_sender()
836 text2 second message text, used only for stderrr
837 error_rc code to pass to exim_exit if no problem
838 f FILE containing body of message (may be stdin)
839 hptr pointer to instore headers or NULL
840
841 Returns: calls exim_exit(), which does not return
842 */
843
844 static void
845 give_local_error(int errcode, uschar *text1, uschar *text2, int error_rc,
846 FILE *f, header_line *hptr)
847 {
848 if (error_handling == ERRORS_SENDER)
849 {
850 error_block eblock;
851 eblock.next = NULL;
852 eblock.text1 = text1;
853 if (!moan_to_sender(errcode, &eblock, hptr, f, FALSE))
854 error_rc = EXIT_FAILURE;
855 }
856 else fprintf(stderr, "exim: %s%s\n", text2, text1); /* Sic */
857 fclose(f);
858 exim_exit(error_rc);
859 }
860
861
862
863 /*************************************************
864 * Add header lines set up by ACL *
865 *************************************************/
866
867 /* This function is called to add the header lines that were set up by "warn"
868 statements in an ACL onto the list of headers in memory. It is done in two
869 stages like this, because when the ACL for RCPT is running, the other headers
870 have not yet been received. This function is called twice; once just before
871 running the DATA ACL, and once after. This is so that header lines added by
872 MAIL or RCPT are visible to the DATA ACL.
873
874 Originally these header lines were added at the end. Now there is support for
875 three different places: top, bottom, and after the Received: header(s). There
876 will always be at least one Received: header, even if it is marked deleted, and
877 even if something else has been put in front of it.
878
879 Arguments:
880 acl_name text to identify which ACL
881
882 Returns: nothing
883 */
884
885 static void
886 add_acl_headers(uschar *acl_name)
887 {
888 header_line *h, *next;
889 header_line *last_received = NULL;
890
891 if (acl_warn_headers == NULL) return;
892 DEBUG(D_receive|D_acl) debug_printf(">>Headers added by %s ACL:\n", acl_name);
893
894 for (h = acl_warn_headers; h != NULL; h = next)
895 {
896 next = h->next;
897
898 switch(h->type)
899 {
900 case htype_add_top:
901 h->next = header_list;
902 header_list = h;
903 DEBUG(D_receive|D_acl) debug_printf(" (at top)");
904 break;
905
906 case htype_add_rec:
907 if (last_received == NULL)
908 {
909 last_received = header_list;
910 while (!header_testname(last_received, US"Received", 8, FALSE))
911 last_received = last_received->next;
912 while (last_received->next != NULL &&
913 header_testname(last_received->next, US"Received", 8, FALSE))
914 last_received = last_received->next;
915 }
916 h->next = last_received->next;
917 last_received->next = h;
918 DEBUG(D_receive|D_acl) debug_printf(" (after Received:)");
919 break;
920
921 default:
922 h->next = NULL;
923 header_last->next = h;
924 break;
925 }
926
927 if (h->next == NULL) header_last = h;
928
929 /* Check for one of the known header types (From:, To:, etc.) though in
930 practice most added headers are going to be "other". Lower case
931 identification letters are never stored with the header; they are used
932 for existence tests when messages are received. So discard any lower case
933 flag values. */
934
935 h->type = header_checkname(h, FALSE);
936 if (h->type >= 'a') h->type = htype_other;
937
938 DEBUG(D_receive|D_acl) debug_printf(" %s", header_last->text);
939 }
940
941 acl_warn_headers = NULL;
942 DEBUG(D_receive|D_acl) debug_printf(">>\n");
943 }
944
945
946
947 /*************************************************
948 * Add host information for log line *
949 *************************************************/
950
951 /* Called for acceptance and rejecting log lines. This adds information about
952 the calling host to a string that is being built dynamically.
953
954 Arguments:
955 s the dynamic string
956 sizeptr points to the size variable
957 ptrptr points to the pointer variable
958
959 Returns: the extended string
960 */
961
962 static uschar *
963 add_host_info_for_log(uschar *s, int *sizeptr, int *ptrptr)
964 {
965 if (sender_fullhost != NULL)
966 {
967 s = string_append(s, sizeptr, ptrptr, 2, US" H=", sender_fullhost);
968 if ((log_extra_selector & LX_incoming_interface) != 0 &&
969 interface_address != NULL)
970 {
971 uschar *ss = string_sprintf(" I=[%s]:%d", interface_address,
972 interface_port);
973 s = string_cat(s, sizeptr, ptrptr, ss, Ustrlen(ss));
974 }
975 }
976 if (sender_ident != NULL)
977 s = string_append(s, sizeptr, ptrptr, 2, US" U=", sender_ident);
978 if (received_protocol != NULL)
979 s = string_append(s, sizeptr, ptrptr, 2, US" P=", received_protocol);
980 return s;
981 }
982
983
984
985
986 /*************************************************
987 * Receive message *
988 *************************************************/
989
990 /* Receive a message on the given input, and put it into a pair of spool files.
991 Either a non-null list of recipients, or the extract flag will be true, or
992 both. The flag sender_local is true for locally generated messages. The flag
993 submission_mode is true if an ACL has obeyed "control = submission". The flag
994 smtp_input is true if the message is to be handled using SMTP conventions about
995 termination and lines starting with dots. For non-SMTP messages, dot_ends is
996 true for dot-terminated messages.
997
998 If a message was successfully read, message_id[0] will be non-zero.
999
1000 The general actions of this function are:
1001
1002 . Read the headers of the message (if any) into a chain of store
1003 blocks.
1004
1005 . If there is a "sender:" header and the message is locally originated,
1006 throw it away, unless the caller is trusted, or unless
1007 active_local_sender_retain is set - which can only happen if
1008 active_local_from_check is false.
1009
1010 . If recipients are to be extracted from the message, build the
1011 recipients list from the headers, removing any that were on the
1012 original recipients list (unless extract_addresses_remove_arguments is
1013 false), and at the same time, remove any bcc header that may be present.
1014
1015 . Get the spool file for the data, sort out its unique name, open
1016 and lock it (but don't give it the name yet).
1017
1018 . Generate a "Message-Id" header if the message doesn't have one, for
1019 locally-originated messages.
1020
1021 . Generate a "Received" header.
1022
1023 . Ensure the recipients list is fully qualified and rewritten if necessary.
1024
1025 . If there are any rewriting rules, apply them to the sender address
1026 and also to the headers.
1027
1028 . If there is no from: header, generate one, for locally-generated messages
1029 and messages in "submission mode" only.
1030
1031 . If the sender is local, check that from: is correct, and if not, generate
1032 a Sender: header, unless message comes from a trusted caller, or this
1033 feature is disabled by active_local_from_check being false.
1034
1035 . If there is no "date" header, generate one, for locally-originated
1036 or submission mode messages only.
1037
1038 . Copy the rest of the input, or up to a terminating "." if in SMTP or
1039 dot_ends mode, to the data file. Leave it open, to hold the lock.
1040
1041 . Write the envelope and the headers to a new file.
1042
1043 . Set the name for the header file; close it.
1044
1045 . Set the name for the data file; close it.
1046
1047 Because this function can potentially be called many times in a single
1048 SMTP connection, all store should be got by store_get(), so that it will be
1049 automatically retrieved after the message is accepted.
1050
1051 FUDGE: It seems that sites on the net send out messages with just LF
1052 terminators, despite the warnings in the RFCs, and other MTAs handle this. So
1053 we make the CRs optional in all cases.
1054
1055 July 2003: Bare CRs in messages, especially in header lines, cause trouble. A
1056 new regime is now in place in which bare CRs in header lines are turned into LF
1057 followed by a space, so as not to terminate the header line.
1058
1059 February 2004: A bare LF in a header line in a message whose first line was
1060 terminated by CRLF is treated in the same way as a bare CR.
1061
1062 Arguments:
1063 extract_recip TRUE if recipients are to be extracted from the message's
1064 headers
1065
1066 Returns: TRUE there are more messages to be read (SMTP input)
1067 FALSE there are no more messages to be read (non-SMTP input
1068 or SMTP connection collapsed, or other failure)
1069
1070 When reading a message for filter testing, the returned value indicates
1071 whether the headers (which is all that is read) were terminated by '.' or
1072 not. */
1073
1074 BOOL
1075 receive_msg(BOOL extract_recip)
1076 {
1077 int i, rc;
1078 int msg_size = 0;
1079 int process_info_len = Ustrlen(process_info);
1080 int error_rc = (error_handling == ERRORS_SENDER)?
1081 errors_sender_rc : EXIT_FAILURE;
1082 int header_size = 256;
1083 int start, end, domain, size, sptr;
1084 int id_resolution;
1085 int had_zero = 0;
1086
1087 register int ptr = 0;
1088
1089 BOOL contains_resent_headers = FALSE;
1090 BOOL extracted_ignored = FALSE;
1091 BOOL first_line_ended_crlf = TRUE_UNSET;
1092 BOOL smtp_yield = TRUE;
1093 BOOL yield = FALSE;
1094
1095 BOOL resents_exist = FALSE;
1096 uschar *resent_prefix = US"";
1097 uschar *blackholed_by = NULL;
1098
1099 flock_t lock_data;
1100 error_block *bad_addresses = NULL;
1101
1102 uschar *frozen_by = NULL;
1103 uschar *queued_by = NULL;
1104
1105 uschar *errmsg, *s;
1106 struct stat statbuf;
1107
1108 /* Final message to give to SMTP caller */
1109
1110 uschar *smtp_reply = NULL;
1111
1112 /* Working header pointers */
1113
1114 header_line *h, *next;
1115
1116 /* Flags for noting the existence of certain headers */
1117
1118 /**** No longer check for these (Nov 2003)
1119 BOOL to_or_cc_header_exists = FALSE;
1120 BOOL bcc_header_exists = FALSE;
1121 ****/
1122
1123 BOOL date_header_exists = FALSE;
1124
1125 /* Pointers to receive the addresses of headers whose contents we need. */
1126
1127 header_line *from_header = NULL;
1128 header_line *subject_header = NULL;
1129 header_line *msgid_header = NULL;
1130 header_line *received_header;
1131
1132 /* Variables for use when building the Received: header. */
1133
1134 uschar *received;
1135 uschar *timestamp;
1136 int tslen;
1137
1138 /* Release any open files that might have been cached while preparing to
1139 accept the message - e.g. by verifying addresses - because reading a message
1140 might take a fair bit of real time. */
1141
1142 search_tidyup();
1143
1144 /* Initialize the chain of headers by setting up a place-holder for Received:
1145 header. Temporarily mark it as "old", i.e. not to be used. We keep header_last
1146 pointing to the end of the chain to make adding headers simple. */
1147
1148 received_header = header_list = header_last = store_get(sizeof(header_line));
1149 header_list->next = NULL;
1150 header_list->type = htype_old;
1151 header_list->text = NULL;
1152 header_list->slen = 0;
1153
1154 /* Control block for the next header to be read. */
1155
1156 next = store_get(sizeof(header_line));
1157 next->text = store_get(header_size);
1158
1159 /* Initialize message id to be null (indicating no message read), and the
1160 header names list to be the normal list. Indicate there is no data file open
1161 yet, initialize the size and warning count, and deal with no size limit. */
1162
1163 message_id[0] = 0;
1164 data_file = NULL;
1165 data_fd = -1;
1166 spool_name[0] = 0;
1167 message_size = 0;
1168 warning_count = 0;
1169 received_count = 1; /* For the one we will add */
1170
1171 if (thismessage_size_limit <= 0) thismessage_size_limit = INT_MAX;
1172
1173 /* While reading the message, body_linecount and body_zerocount is computed.
1174 The full message_ linecount is set up only when the headers are read back in
1175 from the spool for delivery. */
1176
1177 body_linecount = body_zerocount = 0;
1178
1179 /* Remember the time of reception. Exim uses time+pid for uniqueness of message
1180 ids, and fractions of a second are required. See the comments that precede the
1181 message id creation below. */
1182
1183 (void)gettimeofday(&message_id_tv, NULL);
1184
1185 /* For other uses of the received time we can operate with granularity of one
1186 second, and for that we use the global variable received_time. This is for
1187 things like ultimate message timeouts. */
1188
1189 received_time = message_id_tv.tv_sec;
1190
1191 /* If SMTP input, set the special handler for timeouts. The alarm() calls
1192 happen in the smtp_getc() function when it refills its buffer. */
1193
1194 if (smtp_input) os_non_restarting_signal(SIGALRM, data_timeout_handler);
1195
1196 /* If not SMTP input, timeout happens only if configured, and we just set a
1197 single timeout for the whole message. */
1198
1199 else if (receive_timeout > 0)
1200 {
1201 os_non_restarting_signal(SIGALRM, data_timeout_handler);
1202 alarm(receive_timeout);
1203 }
1204
1205 /* SIGTERM and SIGINT are caught always. */
1206
1207 signal(SIGTERM, data_sigterm_sigint_handler);
1208 signal(SIGINT, data_sigterm_sigint_handler);
1209
1210 /* Header lines in messages are not supposed to be very long, though when
1211 unfolded, to: and cc: headers can take up a lot of store. We must also cope
1212 with the possibility of junk being thrown at us. Start by getting 256 bytes for
1213 storing the header, and extend this as necessary using string_cat().
1214
1215 To cope with total lunacies, impose an upper limit on the length of the header
1216 section of the message, as otherwise the store will fill up. We must also cope
1217 with the possibility of binary zeros in the data. Hence we cannot use fgets().
1218 Folded header lines are joined into one string, leaving the '\n' characters
1219 inside them, so that writing them out reproduces the input.
1220
1221 Loop for each character of each header; the next structure for chaining the
1222 header is set up already, with ptr the offset of the next character in
1223 next->text. */
1224
1225 for (;;)
1226 {
1227 int ch = (receive_getc)();
1228
1229 /* If we hit EOF on a SMTP connection, it's an error, since incoming
1230 SMTP must have a correct "." terminator. */
1231
1232 if (ch == EOF && smtp_input /* && !smtp_batched_input */)
1233 {
1234 smtp_reply = handle_lost_connection(US" (header)");
1235 smtp_yield = FALSE;
1236 goto TIDYUP; /* Skip to end of function */
1237 }
1238
1239 /* See if we are at the current header's size limit - there must be at least
1240 four bytes left. This allows for the new character plus a zero, plus two for
1241 extra insertions when we are playing games with dots and carriage returns. If
1242 we are at the limit, extend the text buffer. This could have been done
1243 automatically using string_cat() but because this is a tightish loop storing
1244 only one character at a time, we choose to do it inline. Normally
1245 store_extend() will be able to extend the block; only at the end of a big
1246 store block will a copy be needed. To handle the case of very long headers
1247 (and sometimes lunatic messages can have ones that are 100s of K long) we
1248 call store_release() for strings that have been copied - if the string is at
1249 the start of a block (and therefore the only thing in it, because we aren't
1250 doing any other gets), the block gets freed. We can only do this because we
1251 know there are no other calls to store_get() going on. */
1252
1253 if (ptr >= header_size - 4)
1254 {
1255 int oldsize = header_size;
1256 /* header_size += 256; */
1257 header_size *= 2;
1258 if (!store_extend(next->text, oldsize, header_size))
1259 {
1260 uschar *newtext = store_get(header_size);
1261 memcpy(newtext, next->text, ptr);
1262 store_release(next->text);
1263 next->text = newtext;
1264 }
1265 }
1266
1267 /* Cope with receiving a binary zero. There is dispute about whether
1268 these should be allowed in RFC 822 messages. The middle view is that they
1269 should not be allowed in headers, at least. Exim takes this attitude at
1270 the moment. We can't just stomp on them here, because we don't know that
1271 this line is a header yet. Set a flag to cause scanning later. */
1272
1273 if (ch == 0) had_zero++;
1274
1275 /* Test for termination. Lines in remote SMTP are terminated by CRLF, while
1276 those from data files use just LF. Treat LF in local SMTP input as a
1277 terminator too. Treat EOF as a line terminator always. */
1278
1279 if (ch == EOF) goto EOL;
1280
1281 /* FUDGE: There are sites out there that don't send CRs before their LFs, and
1282 other MTAs accept this. We are therefore forced into this "liberalisation"
1283 too, so we accept LF as a line terminator whatever the source of the message.
1284 However, if the first line of the message ended with a CRLF, we treat a bare
1285 LF specially by inserting a white space after it to ensure that the header
1286 line is not terminated. */
1287
1288 if (ch == '\n')
1289 {
1290 if (first_line_ended_crlf == TRUE_UNSET) first_line_ended_crlf = FALSE;
1291 else if (first_line_ended_crlf) receive_ungetc(' ');
1292 goto EOL;
1293 }
1294
1295 /* This is not the end of the line. If this is SMTP input and this is
1296 the first character in the line and it is a "." character, ignore it.
1297 This implements the dot-doubling rule, though header lines starting with
1298 dots aren't exactly common. They are legal in RFC 822, though. If the
1299 following is CRLF or LF, this is the line that that terminates the
1300 entire message. We set message_ended to indicate this has happened (to
1301 prevent further reading), and break out of the loop, having freed the
1302 empty header, and set next = NULL to indicate no data line. */
1303
1304 if (ptr == 0 && ch == '.' && (smtp_input || dot_ends))
1305 {
1306 ch = (receive_getc)();
1307 if (ch == '\r')
1308 {
1309 ch = (receive_getc)();
1310 if (ch != '\n')
1311 {
1312 receive_ungetc(ch);
1313 ch = '\r'; /* Revert to CR */
1314 }
1315 }
1316 if (ch == '\n')
1317 {
1318 message_ended = END_DOT;
1319 store_reset(next);
1320 next = NULL;
1321 break; /* End character-reading loop */
1322 }
1323
1324 /* For non-SMTP input, the dot at the start of the line was really a data
1325 character. What is now in ch is the following character. We guaranteed
1326 enough space for this above. */
1327
1328 if (!smtp_input)
1329 {
1330 next->text[ptr++] = '.';
1331 message_size++;
1332 }
1333 }
1334
1335 /* If CR is immediately followed by LF, end the line, ignoring the CR, and
1336 remember this case if this is the first line ending. */
1337
1338 if (ch == '\r')
1339 {
1340 ch = (receive_getc)();
1341 if (ch == '\n')
1342 {
1343 if (first_line_ended_crlf == TRUE_UNSET) first_line_ended_crlf = TRUE;
1344 goto EOL;
1345 }
1346
1347 /* Otherwise, put back the character after CR, and turn the bare CR
1348 into LF SP. */
1349
1350 ch = (receive_ungetc)(ch);
1351 next->text[ptr++] = '\n';
1352 message_size++;
1353 ch = ' ';
1354 }
1355
1356 /* We have a data character for the header line. */
1357
1358 next->text[ptr++] = ch; /* Add to buffer */
1359 message_size++; /* Total message size so far */
1360
1361 /* Handle failure due to a humungously long header section. The >= allows
1362 for the terminating \n. Add what we have so far onto the headers list so
1363 that it gets reflected in any error message, and back up the just-read
1364 character. */
1365
1366 if (message_size >= header_maxsize)
1367 {
1368 next->text[ptr] = 0;
1369 next->slen = ptr;
1370 next->type = htype_other;
1371 next->next = NULL;
1372 header_last->next = next;
1373 header_last = next;
1374
1375 log_write(0, LOG_MAIN, "ridiculously long message header received from "
1376 "%s (more than %d characters): message abandoned",
1377 sender_host_unknown? sender_ident : sender_fullhost, header_maxsize);
1378
1379 if (smtp_input)
1380 {
1381 smtp_reply = US"552 Message header is ridiculously long";
1382 receive_swallow_smtp();
1383 goto TIDYUP; /* Skip to end of function */
1384 }
1385
1386 else
1387 {
1388 give_local_error(ERRMESS_VLONGHEADER,
1389 string_sprintf("message header longer than %d characters received: "
1390 "message not accepted", header_maxsize), US"", error_rc, stdin,
1391 header_list->next);
1392 /* Does not return */
1393 }
1394 }
1395
1396 continue; /* With next input character */
1397
1398 /* End of header line reached */
1399
1400 EOL:
1401 receive_linecount++; /* For BSMTP errors */
1402
1403 /* Now put in the terminating newline. There is always space for
1404 at least two more characters. */
1405
1406 next->text[ptr++] = '\n';
1407 message_size++;
1408
1409 /* A blank line signals the end of the headers; release the unwanted
1410 space and set next to NULL to indicate this. */
1411
1412 if (ptr == 1)
1413 {
1414 store_reset(next);
1415 next = NULL;
1416 break;
1417 }
1418
1419 /* There is data in the line; see if the next input character is a
1420 whitespace character. If it is, we have a continuation of this header line.
1421 There is always space for at least one character at this point. */
1422
1423 if (ch != EOF)
1424 {
1425 int nextch = (receive_getc)();
1426 if (nextch == ' ' || nextch == '\t')
1427 {
1428 next->text[ptr++] = nextch;
1429 message_size++;
1430 continue; /* Iterate the loop */
1431 }
1432 else if (nextch != EOF) (receive_ungetc)(nextch); /* For next time */
1433 else ch = EOF; /* Cause main loop to exit at end */
1434 }
1435
1436 /* We have got to the real line end. Terminate the string and release store
1437 beyond it. If it turns out to be a real header, internal binary zeros will
1438 be squashed later. */
1439
1440 next->text[ptr] = 0;
1441 next->slen = ptr;
1442 store_reset(next->text + ptr + 1);
1443
1444 /* Check the running total size against the overall message size limit. We
1445 don't expect to fail here, but if the overall limit is set less than MESSAGE_
1446 MAXSIZE and a big header is sent, we want to catch it. Just stop reading
1447 headers - the code to read the body will then also hit the buffer. */
1448
1449 if (message_size > thismessage_size_limit) break;
1450
1451 /* A line that is not syntactically correct for a header also marks
1452 the end of the headers. In this case, we leave next containing the
1453 first data line. This might actually be several lines because of the
1454 continuation logic applied above, but that doesn't matter.
1455
1456 It turns out that smail, and presumably sendmail, accept leading lines
1457 of the form
1458
1459 From ph10 Fri Jan 5 12:35 GMT 1996
1460
1461 in messages. The "mail" command on Solaris 2 sends such lines. I cannot
1462 find any documentation of this, but for compatibility it had better be
1463 accepted. Exim restricts it to the case of non-smtp messages, and
1464 treats it as an alternative to the -f command line option. Thus it is
1465 ignored except for trusted users or filter testing. Otherwise it is taken
1466 as the sender address, unless -f was used (sendmail compatibility).
1467
1468 It further turns out that some UUCPs generate the From_line in a different
1469 format, e.g.
1470
1471 From ph10 Fri, 7 Jan 97 14:00:00 GMT
1472
1473 The regex for matching these things is now capable of recognizing both
1474 formats (including 2- and 4-digit years in the latter). In fact, the regex
1475 is now configurable, as is the expansion string to fish out the sender.
1476
1477 Even further on it has been discovered that some broken clients send
1478 these lines in SMTP messages. There is now an option to ignore them from
1479 specified hosts or networks. Sigh. */
1480
1481 if (header_last == header_list &&
1482 (!smtp_input
1483 ||
1484 (sender_host_address != NULL &&
1485 verify_check_host(&ignore_fromline_hosts) == OK)
1486 ||
1487 (sender_host_address == NULL && ignore_fromline_local)
1488 ) &&
1489 regex_match_and_setup(regex_From, next->text, 0, -1))
1490 {
1491 if (!sender_address_forced)
1492 {
1493 uschar *uucp_sender = expand_string(uucp_from_sender);
1494 if (uucp_sender == NULL)
1495 {
1496 log_write(0, LOG_MAIN|LOG_PANIC,
1497 "expansion of \"%s\" failed after matching "
1498 "\"From \" line: %s", uucp_from_sender, expand_string_message);
1499 }
1500 else
1501 {
1502 int start, end, domain;
1503 uschar *errmess;
1504 uschar *newsender = parse_extract_address(uucp_sender, &errmess,
1505 &start, &end, &domain, TRUE);
1506 if (newsender != NULL)
1507 {
1508 if (domain == 0 && newsender[0] != 0)
1509 newsender = rewrite_address_qualify(newsender, FALSE);
1510
1511 if (filter_test != NULL || receive_check_set_sender(newsender))
1512 {
1513 sender_address = newsender;
1514
1515 if (trusted_caller || filter_test != NULL)
1516 {
1517 authenticated_sender = NULL;
1518 originator_name = US"";
1519 sender_local = FALSE;
1520 }
1521
1522 if (filter_test != NULL)
1523 printf("Sender taken from \"From \" line\n");
1524 }
1525 }
1526 }
1527 }
1528 }
1529
1530 /* Not a leading "From " line. Check to see if it is a valid header line.
1531 Header names may contain any non-control characters except space and colon,
1532 amazingly. */
1533
1534 else
1535 {
1536 uschar *p = next->text;
1537
1538 /* If not a valid header line, break from the header reading loop, leaving
1539 next != NULL, indicating that it holds the first line of the body. */
1540
1541 if (isspace(*p)) break;
1542 while (mac_isgraph(*p) && *p != ':') p++;
1543 while (isspace(*p)) p++;
1544 if (*p != ':')
1545 {
1546 body_zerocount = had_zero;
1547 break;
1548 }
1549
1550 /* We have a valid header line. If there were any binary zeroes in
1551 the line, stomp on them here. */
1552
1553 if (had_zero > 0)
1554 for (p = next->text; p < next->text + ptr; p++) if (*p == 0) *p = '?';
1555
1556 /* It is perfectly legal to have an empty continuation line
1557 at the end of a header, but it is confusing to humans
1558 looking at such messages, since it looks like a blank line.
1559 Reduce confusion by removing redundant white space at the
1560 end. We know that there is at least one printing character
1561 (the ':' tested for above) so there is no danger of running
1562 off the end. */
1563
1564 p = next->text + ptr - 2;
1565 for (;;)
1566 {
1567 while (*p == ' ' || *p == '\t') p--;
1568 if (*p != '\n') break;
1569 ptr = (p--) - next->text + 1;
1570 message_size -= next->slen - ptr;
1571 next->text[ptr] = 0;
1572 next->slen = ptr;
1573 }
1574
1575 /* Add the header to the chain */
1576
1577 next->type = htype_other;
1578 next->next = NULL;
1579 header_last->next = next;
1580 header_last = next;
1581
1582 /* Check the limit for individual line lengths. This comes after adding to
1583 the chain so that the failing line is reflected if a bounce is generated
1584 (for a local message). */
1585
1586 if (header_line_maxsize > 0 && next->slen > header_line_maxsize)
1587 {
1588 log_write(0, LOG_MAIN, "overlong message header line received from "
1589 "%s (more than %d characters): message abandoned",
1590 sender_host_unknown? sender_ident : sender_fullhost,
1591 header_line_maxsize);
1592
1593 if (smtp_input)
1594 {
1595 smtp_reply = US"552 A message header line is too long";
1596 receive_swallow_smtp();
1597 goto TIDYUP; /* Skip to end of function */
1598 }
1599
1600 else
1601 {
1602 give_local_error(ERRMESS_VLONGHDRLINE,
1603 string_sprintf("message header line longer than %d characters "
1604 "received: message not accepted", header_line_maxsize), US"",
1605 error_rc, stdin, header_list->next);
1606 /* Does not return */
1607 }
1608 }
1609
1610 /* Note if any resent- fields exist. */
1611
1612 if (!resents_exist && strncmpic(next->text, US"resent-", 7) == 0)
1613 {
1614 resents_exist = TRUE;
1615 resent_prefix = US"Resent-";
1616 }
1617 }
1618
1619 /* The line has been handled. If we have hit EOF, break out of the loop,
1620 indicating no pending data line. */
1621
1622 if (ch == EOF) { next = NULL; break; }
1623
1624 /* Set up for the next header */
1625
1626 header_size = 256;
1627 next = store_get(sizeof(header_line));
1628 next->text = store_get(header_size);
1629 ptr = 0;
1630 had_zero = 0;
1631 } /* Continue, starting to read the next header */
1632
1633 /* At this point, we have read all the headers into a data structure in main
1634 store. The first header is still the dummy placeholder for the Received: header
1635 we are going to generate a bit later on. If next != NULL, it contains the first
1636 data line - which terminated the headers before reaching a blank line (not the
1637 normal case). */
1638
1639 DEBUG(D_receive)
1640 {
1641 debug_printf(">>Headers received:\n");
1642 for (h = header_list->next; h != NULL; h = h->next)
1643 debug_printf("%s", h->text);
1644 debug_printf("\n");
1645 }
1646
1647 /* End of file on any SMTP connection is an error. If an incoming SMTP call
1648 is dropped immediately after valid headers, the next thing we will see is EOF.
1649 We must test for this specially, as further down the reading of the data is
1650 skipped if already at EOF. */
1651
1652 if (smtp_input && (receive_feof)())
1653 {
1654 smtp_reply = handle_lost_connection(US" (after header)");
1655 smtp_yield = FALSE;
1656 goto TIDYUP; /* Skip to end of function */
1657 }
1658
1659 /* If this is a filter test run and no headers were read, output a warning
1660 in case there is a mistake in the test message. */
1661
1662 if (filter_test != NULL && header_list->next == NULL)
1663 printf("Warning: no message headers read\n");
1664
1665
1666 /* Scan the headers to identify them. Some are merely marked for later
1667 processing; some are dealt with here. */
1668
1669 for (h = header_list->next; h != NULL; h = h->next)
1670 {
1671 BOOL is_resent = strncmpic(h->text, US"resent-", 7) == 0;
1672 if (is_resent) contains_resent_headers = TRUE;
1673
1674 switch (header_checkname(h, is_resent))
1675 {
1676 /* "Bcc:" gets flagged, and its existence noted, whether it's resent- or
1677 not. */
1678
1679 case htype_bcc:
1680 h->type = htype_bcc;
1681 /****
1682 bcc_header_exists = TRUE;
1683 ****/
1684 break;
1685
1686 /* "Cc:" gets flagged, and the existence of a recipient header is noted,
1687 whether it's resent- or not. */
1688
1689 case htype_cc:
1690 h->type = htype_cc;
1691 /****
1692 to_or_cc_header_exists = TRUE;
1693 ****/
1694 break;
1695
1696 /* Record whether a Date: or Resent-Date: header exists, as appropriate. */
1697
1698 case htype_date:
1699 date_header_exists = !resents_exist || is_resent;
1700 break;
1701
1702 /* Same comments as about Return-Path: below. */
1703
1704 case htype_delivery_date:
1705 if (delivery_date_remove) h->type = htype_old;
1706 break;
1707
1708 /* Same comments as about Return-Path: below. */
1709
1710 case htype_envelope_to:
1711 if (envelope_to_remove) h->type = htype_old;
1712 break;
1713
1714 /* Mark all "From:" headers so they get rewritten. Save the one that is to
1715 be used for Sender: checking. For Sendmail compatibility, if the "From:"
1716 header consists of just the login id of the user who called Exim, rewrite
1717 it with the gecos field first. Apply this rule to Resent-From: if there
1718 are resent- fields. */
1719
1720 case htype_from:
1721 h->type = htype_from;
1722 if (!resents_exist || is_resent)
1723 {
1724 from_header = h;
1725 if (!smtp_input)
1726 {
1727 uschar *s = Ustrchr(h->text, ':') + 1;
1728 while (isspace(*s)) s++;
1729 if (strncmpic(s, originator_login, h->slen - (s - h->text) - 1) == 0)
1730 {
1731 uschar *name = is_resent? US"Resent-From" : US"From";
1732 header_add(htype_from, "%s: %s <%s@%s>\n", name, originator_name,
1733 originator_login, qualify_domain_sender);
1734 from_header = header_last;
1735 h->type = htype_old;
1736 DEBUG(D_receive|D_rewrite)
1737 debug_printf("rewrote \"%s:\" header using gecos\n", name);
1738 }
1739 }
1740 }
1741 break;
1742
1743 /* Identify the Message-id: header for generating "in-reply-to" in the
1744 autoreply transport. For incoming logging, save any resent- value. In both
1745 cases, take just the first of any multiples. */
1746
1747 case htype_id:
1748 if (msgid_header == NULL && (!resents_exist || is_resent))
1749 {
1750 msgid_header = h;
1751 h->type = htype_id;
1752 }
1753 break;
1754
1755 /* Flag all Received: headers */
1756
1757 case htype_received:
1758 h->type = htype_received;
1759 received_count++;
1760 break;
1761
1762 /* "Reply-to:" is just noted (there is no resent-reply-to field) */
1763
1764 case htype_reply_to:
1765 h->type = htype_reply_to;
1766 break;
1767
1768 /* The Return-path: header is supposed to be added to messages when
1769 they leave the SMTP system. We shouldn't receive messages that already
1770 contain Return-path. However, since Exim generates Return-path: on
1771 local delivery, resent messages may well contain it. We therefore
1772 provide an option (which defaults on) to remove any Return-path: headers
1773 on input. Removal actually means flagging as "old", which prevents the
1774 header being transmitted with the message. */
1775
1776 case htype_return_path:
1777 if (return_path_remove) h->type = htype_old;
1778
1779 /* If we are testing a mail filter file, use the value of the
1780 Return-Path: header to set up the return_path variable, which is not
1781 otherwise set. However, remove any <> that surround the address
1782 because the variable doesn't have these. */
1783
1784 if (filter_test != NULL)
1785 {
1786 uschar *start = h->text + 12;
1787 uschar *end = start + Ustrlen(start);
1788 while (isspace(*start)) start++;
1789 while (end > start && isspace(end[-1])) end--;
1790 if (*start == '<' && end[-1] == '>')
1791 {
1792 start++;
1793 end--;
1794 }
1795 return_path = string_copyn(start, end - start);
1796 printf("Return-path taken from \"Return-path:\" header line\n");
1797 }
1798 break;
1799
1800 /* If there is a "Sender:" header and the message is locally originated,
1801 and from an untrusted caller, or if we are in submission mode for a remote
1802 message, mark it "old" so that it will not be transmitted with the message,
1803 unless active_local_sender_retain is set. (This can only be true if
1804 active_local_from_check is false.) If there are any resent- headers in the
1805 message, apply this rule to Resent-Sender: instead of Sender:. Messages
1806 with multiple resent- header sets cannot be tidily handled. (For this
1807 reason, at least one MUA - Pine - turns old resent- headers into X-resent-
1808 headers when resending, leaving just one set.) */
1809
1810 case htype_sender:
1811 h->type = ((!active_local_sender_retain &&
1812 ((sender_local && !trusted_caller) || submission_mode)
1813 ) &&
1814 (!resents_exist||is_resent))?
1815 htype_old : htype_sender;
1816 break;
1817
1818 /* Remember the Subject: header for logging. There is no Resent-Subject */
1819
1820 case htype_subject:
1821 subject_header = h;
1822 break;
1823
1824 /* "To:" gets flagged, and the existence of a recipient header is noted,
1825 whether it's resent- or not. */
1826
1827 case htype_to:
1828 h->type = htype_to;
1829 /****
1830 to_or_cc_header_exists = TRUE;
1831 ****/
1832 break;
1833 }
1834 }
1835
1836 /* Extract recipients from the headers if that is required (the -t option).
1837 Note that this is documented as being done *before* any address rewriting takes
1838 place. There are two possibilities:
1839
1840 (1) According to sendmail documentation for Solaris, IRIX, and HP-UX, any
1841 recipients already listed are to be REMOVED from the message. Smail 3 works
1842 like this. We need to build a non-recipients tree for that list, because in
1843 subsequent processing this data is held in a tree and that's what the
1844 spool_write_header() function expects. Make sure that non-recipient addresses
1845 are fully qualified and rewritten if necessary.
1846
1847 (2) According to other sendmail documentation, -t ADDS extracted recipients to
1848 those in the command line arguments (and it is rumoured some other MTAs do
1849 this). Therefore, there is an option to make Exim behave this way.
1850
1851 *** Notes on "Resent-" header lines ***
1852
1853 The presence of resent-headers in the message makes -t horribly ambiguous.
1854 Experiments with sendmail showed that it uses recipients for all resent-
1855 headers, totally ignoring the concept of "sets of resent- headers" as described
1856 in RFC 2822 section 3.6.6. Sendmail also amalgamates them into a single set
1857 with all the addresses in one instance of each header.
1858
1859 This seems to me not to be at all sensible. Before release 4.20, Exim 4 gave an
1860 error for -t if there were resent- headers in the message. However, after a
1861 discussion on the mailing list, I've learned that there are MUAs that use
1862 resent- headers with -t, and also that the stuff about sets of resent- headers
1863 and their ordering in RFC 2822 is generally ignored. An MUA that submits a
1864 message with -t and resent- header lines makes sure that only *its* resent-
1865 headers are present; previous ones are often renamed as X-resent- for example.
1866
1867 Consequently, Exim has been changed so that, if any resent- header lines are
1868 present, the recipients are taken from all of the appropriate resent- lines,
1869 and not from the ordinary To:, Cc:, etc. */
1870
1871 if (extract_recip)
1872 {
1873 int rcount = 0;
1874 error_block **bnext = &bad_addresses;
1875
1876 if (extract_addresses_remove_arguments)
1877 {
1878 while (recipients_count-- > 0)
1879 {
1880 uschar *s = rewrite_address(recipients_list[recipients_count].address,
1881 TRUE, TRUE, global_rewrite_rules, rewrite_existflags);
1882 tree_add_nonrecipient(s);
1883 }
1884 recipients_list = NULL;
1885 recipients_count = recipients_list_max = 0;
1886 }
1887
1888 parse_allow_group = TRUE; /* Allow address group syntax */
1889
1890 /* Now scan the headers */
1891
1892 for (h = header_list->next; h != NULL; h = h->next)
1893 {
1894 if ((h->type == htype_to || h->type == htype_cc || h->type == htype_bcc) &&
1895 (!contains_resent_headers || strncmpic(h->text, US"resent-", 7) == 0))
1896 {
1897 uschar *s = Ustrchr(h->text, ':') + 1;
1898 while (isspace(*s)) s++;
1899
1900 while (*s != 0)
1901 {
1902 uschar *ss = parse_find_address_end(s, FALSE);
1903 uschar *recipient, *errmess, *p, *pp;
1904 int start, end, domain;
1905
1906 /* Check on maximum */
1907
1908 if (recipients_max > 0 && ++rcount > recipients_max)
1909 {
1910 give_local_error(ERRMESS_TOOMANYRECIP, US"too many recipients",
1911 US"message rejected: ", error_rc, stdin, NULL);
1912 /* Does not return */
1913 }
1914
1915 /* Make a copy of the address, and remove any internal newlines. These
1916 may be present as a result of continuations of the header line. The
1917 white space that follows the newline must not be removed - it is part
1918 of the header. */
1919
1920 pp = recipient = store_get(ss - s + 1);
1921 for (p = s; p < ss; p++) if (*p != '\n') *pp++ = *p;
1922 *pp = 0;
1923 recipient = parse_extract_address(recipient, &errmess, &start, &end,
1924 &domain, FALSE);
1925
1926 /* Keep a list of all the bad addresses so we can send a single
1927 error message at the end. However, an empty address is not an error;
1928 just ignore it. This can come from an empty group list like
1929
1930 To: Recipients of list:;
1931
1932 If there are no recipients at all, an error will occur later. */
1933
1934 if (recipient == NULL && Ustrcmp(errmess, "empty address") != 0)
1935 {
1936 int len = Ustrlen(s);
1937 error_block *b = store_get(sizeof(error_block));
1938 while (len > 0 && isspace(s[len-1])) len--;
1939 b->next = NULL;
1940 b->text1 = string_printing(string_copyn(s, len));
1941 b->text2 = errmess;
1942 *bnext = b;
1943 bnext = &(b->next);
1944 }
1945
1946 /* If the recipient is already in the nonrecipients tree, it must
1947 have appeared on the command line with the option extract_addresses_
1948 remove_arguments set. Do not add it to the recipients, and keep a note
1949 that this has happened, in order to give a better error if there are
1950 no recipients left. */
1951
1952 else if (recipient != NULL)
1953 {
1954 if (tree_search(tree_nonrecipients, recipient) == NULL)
1955 receive_add_recipient(recipient, -1);
1956 else
1957 extracted_ignored = TRUE;
1958 }
1959
1960 /* Move on past this address */
1961
1962 s = ss + (*ss? 1:0);
1963 while (isspace(*s)) s++;
1964 }
1965
1966 /* If this was the bcc: header, mark it "old", which means it
1967 will be kept on the spool, but not transmitted as part of the
1968 message. */
1969
1970 if (h->type == htype_bcc)
1971 {
1972 h->type = htype_old;
1973 /****
1974 bcc_header_exists = FALSE;
1975 ****/
1976 }
1977 } /* For appropriate header line */
1978 } /* For each header line */
1979
1980 parse_allow_group = FALSE; /* Reset group syntax flags */
1981 parse_found_group = FALSE;
1982 }
1983
1984 /* Now build the unique message id. This has changed several times over the
1985 lifetime of Exim. This description was rewritten for Exim 4.14 (February 2003).
1986 Retaining all the history in the comment has become too unwieldy - read
1987 previous release sources if you want it.
1988
1989 The message ID has 3 parts: tttttt-pppppp-ss. Each part is a number in base 62.
1990 The first part is the current time, in seconds. The second part is the current
1991 pid. Both are large enough to hold 32-bit numbers in base 62. The third part
1992 can hold a number in the range 0-3843. It used to be a computed sequence
1993 number, but is now the fractional component of the current time in units of
1994 1/2000 of a second (i.e. a value in the range 0-1999). After a message has been
1995 received, Exim ensures that the timer has ticked at the appropriate level
1996 before proceeding, to avoid duplication if the pid happened to be re-used
1997 within the same time period. It seems likely that most messages will take at
1998 least half a millisecond to be received, so no delay will normally be
1999 necessary. At least for some time...
2000
2001 There is a modification when localhost_number is set. Formerly this was allowed
2002 to be as large as 255. Now it is restricted to the range 0-16, and the final
2003 component of the message id becomes (localhost_number * 200) + fractional time
2004 in units of 1/200 of a second (i.e. a value in the range 0-3399).
2005
2006 Some not-really-Unix operating systems use case-insensitive file names (Darwin,
2007 Cygwin). For these, we have to use base 36 instead of base 62. Luckily, this
2008 still allows the tttttt field to hold a large enough number to last for some
2009 more decades, and the final two-digit field can hold numbers up to 1295, which
2010 is enough for milliseconds (instead of 1/2000 of a second).
2011
2012 However, the pppppp field cannot hold a 32-bit pid, but it can hold a 31-bit
2013 pid, so it is probably safe because pids have to be positive. The
2014 localhost_number is restricted to 0-10 for these hosts, and when it is set, the
2015 final field becomes (localhost_number * 100) + fractional time in centiseconds.
2016
2017 Note that string_base62() returns its data in a static storage block, so it
2018 must be copied before calling string_base62() again. It always returns exactly
2019 6 characters.
2020
2021 There doesn't seem to be anything in the RFC which requires a message id to
2022 start with a letter, but Smail was changed to ensure this. The external form of
2023 the message id (as supplied by string expansion) therefore starts with an
2024 additional leading 'E'. The spool file names do not include this leading
2025 letter and it is not used internally.
2026
2027 NOTE: If ever the format of message ids is changed, the regular expression for
2028 checking that a string is in this format must be updated in a corresponding
2029 way. It appears in the initializing code in exim.c. The macro MESSAGE_ID_LENGTH
2030 must also be changed to reflect the correct string length. Then, of course,
2031 other programs that rely on the message id format will need updating too. */
2032
2033 Ustrncpy(message_id, string_base62((long int)(message_id_tv.tv_sec)), 6);
2034 message_id[6] = '-';
2035 Ustrncpy(message_id + 7, string_base62((long int)getpid()), 6);
2036
2037 /* Deal with the case where the host number is set. The value of the number was
2038 checked when it was read, to ensure it isn't too big. The timing granularity is
2039 left in id_resolution so that an appropriate wait can be done after receiving
2040 the message, if necessary (we hope it won't be). */
2041
2042 if (host_number_string != NULL)
2043 {
2044 id_resolution = (BASE_62 == 62)? 5000 : 10000;
2045 sprintf(CS(message_id + MESSAGE_ID_LENGTH - 3), "-%2s",
2046 string_base62((long int)(
2047 host_number * (1000000/id_resolution) +
2048 message_id_tv.tv_usec/id_resolution)) + 4);
2049 }
2050
2051 /* Host number not set: final field is just the fractional time at an
2052 appropriate resolution. */
2053
2054 else
2055 {
2056 id_resolution = (BASE_62 == 62)? 500 : 1000;
2057 sprintf(CS(message_id + MESSAGE_ID_LENGTH - 3), "-%2s",
2058 string_base62((long int)(message_id_tv.tv_usec/id_resolution)) + 4);
2059 }
2060
2061 /* Add the current message id onto the current process info string if
2062 it will fit. */
2063
2064 (void)string_format(process_info + process_info_len,
2065 PROCESS_INFO_SIZE - process_info_len, " id=%s", message_id);
2066
2067 /* If we are using multiple input directories, set up the one for this message
2068 to be the least significant base-62 digit of the time of arrival. Otherwise
2069 ensure that it is an empty string. */
2070
2071 message_subdir[0] = split_spool_directory? message_id[5] : 0;
2072
2073 /* Now that we have the message-id, if there is no message-id: header, generate
2074 one, but only for local or submission mode messages. This can be
2075 user-configured if required, but we had better flatten any illegal characters
2076 therein. */
2077
2078 if (msgid_header == NULL && (sender_host_address == NULL || submission_mode))
2079 {
2080 uschar *p;
2081 uschar *id_text = US"";
2082 uschar *id_domain = primary_hostname;
2083
2084 /* Permit only letters, digits, dots, and hyphens in the domain */
2085
2086 if (message_id_domain != NULL)
2087 {
2088 uschar *new_id_domain = expand_string(message_id_domain);
2089 if (new_id_domain == NULL)
2090 {
2091 if (!expand_string_forcedfail)
2092 log_write(0, LOG_MAIN|LOG_PANIC,
2093 "expansion of \"%s\" (message_id_header_domain) "
2094 "failed: %s", message_id_domain, expand_string_message);
2095 }
2096 else if (*new_id_domain != 0)
2097 {
2098 id_domain = new_id_domain;
2099 for (p = id_domain; *p != 0; p++)
2100 if (!isalnum(*p) && *p != '.') *p = '-'; /* No need to test '-' ! */
2101 }
2102 }
2103
2104 /* Permit all characters except controls and RFC 2822 specials in the
2105 additional text part. */
2106
2107 if (message_id_text != NULL)
2108 {
2109 uschar *new_id_text = expand_string(message_id_text);
2110 if (new_id_text == NULL)
2111 {
2112 if (!expand_string_forcedfail)
2113 log_write(0, LOG_MAIN|LOG_PANIC,
2114 "expansion of \"%s\" (message_id_header_text) "
2115 "failed: %s", message_id_text, expand_string_message);
2116 }
2117 else if (*new_id_text != 0)
2118 {
2119 id_text = new_id_text;
2120 for (p = id_text; *p != 0; p++)
2121 if (mac_iscntrl_or_special(*p)) *p = '-';
2122 }
2123 }
2124
2125 /* Add the header line */
2126
2127 header_add(htype_id, "%sMessage-Id: <%s%s%s@%s>\n", resent_prefix,
2128 message_id_external, (*id_text == 0)? "" : ".", id_text, id_domain);
2129 }
2130
2131 /* If we are to log recipients, keep a copy of the raw ones before any possible
2132 rewriting. Must copy the count, because later ACLs and the local_scan()
2133 function may mess with the real recipients. */
2134
2135 if ((log_extra_selector & LX_received_recipients) != 0)
2136 {
2137 raw_recipients = store_get(recipients_count * sizeof(uschar *));
2138 for (i = 0; i < recipients_count; i++)
2139 raw_recipients[i] = string_copy(recipients_list[i].address);
2140 raw_recipients_count = recipients_count;
2141 }
2142
2143 /* Ensure the recipients list is fully qualified and rewritten. Unqualified
2144 recipients will get here only if the conditions were right (allow_unqualified_
2145 recipient is TRUE). */
2146
2147 for (i = 0; i < recipients_count; i++)
2148 recipients_list[i].address =
2149 rewrite_address(recipients_list[i].address, TRUE, TRUE,
2150 global_rewrite_rules, rewrite_existflags);
2151
2152 /* If there is no From: header, generate one for local or submission_mode
2153 messages. If there is no sender address, but the sender is local or this is a
2154 local delivery error, use the originator login. This shouldn't happen for
2155 genuine bounces, but might happen for autoreplies. The addition of From: must
2156 be done *before* checking for the possible addition of a Sender: header,
2157 because untrusted_set_sender allows an untrusted user to set anything in the
2158 envelope (which might then get info From:) but we still want to ensure a valid
2159 Sender: if it is required. */
2160
2161 if (from_header == NULL && (sender_host_address == NULL || submission_mode))
2162 {
2163 /* Envelope sender is empty */
2164
2165 if (sender_address[0] == 0)
2166 {
2167 if (sender_local || local_error_message)
2168 {
2169 header_add(htype_from, "%sFrom: %s%s%s@%s%s\n", resent_prefix,
2170 originator_name,
2171 (originator_name[0] == 0)? "" : " <",
2172 local_part_quote(originator_login),
2173 qualify_domain_sender,
2174 (originator_name[0] == 0)? "" : ">");
2175 }
2176 else if (submission_mode && authenticated_id != NULL)
2177 {
2178 if (submission_domain == NULL)
2179 {
2180 header_add(htype_from, "%sFrom: %s@%s\n", resent_prefix,
2181 local_part_quote(authenticated_id), qualify_domain_sender);
2182 }
2183 else if (submission_domain[0] == 0) /* empty => whole address set */
2184 {
2185 header_add(htype_from, "%sFrom: %s\n", resent_prefix,
2186 authenticated_id);
2187 }
2188 else
2189 {
2190 header_add(htype_from, "%sFrom: %s@%s\n", resent_prefix,
2191 local_part_quote(authenticated_id), submission_domain);
2192 }
2193 from_header = header_last; /* To get it checked for Sender: */
2194 }
2195 }
2196
2197 /* There is a non-null envelope sender. Build the header using the original
2198 sender address, before any rewriting that might have been done while
2199 verifying it. */
2200
2201 else
2202 {
2203 if (!smtp_input || sender_local)
2204 header_add(htype_from, "%sFrom: %s%s%s%s\n",
2205 resent_prefix, originator_name,
2206 (originator_name[0] == 0)? "" : " <",
2207 (sender_address_unrewritten == NULL)?
2208 sender_address : sender_address_unrewritten,
2209 (originator_name[0] == 0)? "" : ">");
2210 else
2211 header_add(htype_from, "%sFrom: %s\n", resent_prefix, sender_address);
2212
2213 from_header = header_last; /* To get it checked for Sender: */
2214 }
2215 }
2216
2217
2218 /* If the sender is local, or if we are in submission mode and there is an
2219 authenticated_id, check that an existing From: is correct, and if not, generate
2220 a Sender: header, unless disabled. Any previously-existing Sender: header was
2221 removed above. Note that sender_local, as well as being TRUE if the caller of
2222 exim is not trusted, is also true if a trusted caller did not supply a -f
2223 argument for non-smtp input. To allow trusted callers to forge From: without
2224 supplying -f, we have to test explicitly here. If the From: header contains
2225 more than one address, then the call to parse_extract_address fails, and a
2226 Sender: header is inserted, as required. */
2227
2228 if (from_header != NULL &&
2229 (active_local_from_check &&
2230 ((sender_local && !trusted_caller) ||
2231 (submission_mode && authenticated_id != NULL))
2232 ))
2233 {
2234 BOOL make_sender = TRUE;
2235 int start, end, domain;
2236 uschar *errmess;
2237 uschar *from_address =
2238 parse_extract_address(Ustrchr(from_header->text, ':') + 1, &errmess,
2239 &start, &end, &domain, FALSE);
2240 uschar *generated_sender_address;
2241
2242 if (submission_mode)
2243 {
2244 if (submission_domain == NULL)
2245 {
2246 generated_sender_address = string_sprintf("%s@%s",
2247 local_part_quote(authenticated_id), qualify_domain_sender);
2248 }
2249 else if (submission_domain[0] == 0) /* empty => full address */
2250 {
2251 generated_sender_address = string_sprintf("%s",
2252 authenticated_id);
2253 }
2254 else
2255 {
2256 generated_sender_address = string_sprintf("%s@%s",
2257 local_part_quote(authenticated_id), submission_domain);
2258 }
2259 }
2260 else
2261 generated_sender_address = string_sprintf("%s@%s",
2262 local_part_quote(originator_login), qualify_domain_sender);
2263
2264 /* Remove permitted prefixes and suffixes from the local part of the From:
2265 address before doing the comparison with the generated sender. */
2266
2267 if (from_address != NULL)
2268 {
2269 int slen;
2270 uschar *at = (domain == 0)? NULL : from_address + domain - 1;
2271
2272 if (at != NULL) *at = 0;
2273 from_address += route_check_prefix(from_address, local_from_prefix);
2274 slen = route_check_suffix(from_address, local_from_suffix);
2275 if (slen > 0)
2276 {
2277 memmove(from_address+slen, from_address, Ustrlen(from_address)-slen);
2278 from_address += slen;
2279 }
2280 if (at != NULL) *at = '@';
2281
2282 if (strcmpic(generated_sender_address, from_address) == 0 ||
2283 (domain == 0 && strcmpic(from_address, originator_login) == 0))
2284 make_sender = FALSE;
2285 }
2286
2287 /* We have to cause the Sender header to be rewritten if there are
2288 appropriate rewriting rules. */
2289
2290 if (make_sender)
2291 {
2292 if (submission_mode)
2293 header_add(htype_sender, "%sSender: %s\n", resent_prefix,
2294 generated_sender_address);
2295 else
2296 header_add(htype_sender, "%sSender: %s <%s>\n",
2297 resent_prefix, originator_name, generated_sender_address);
2298 }
2299 }
2300
2301
2302 /* If there are any rewriting rules, apply them to the sender address, unless
2303 it has already been rewritten as part of verification for SMTP input. */
2304
2305 if (global_rewrite_rules != NULL && sender_address_unrewritten == NULL &&
2306 sender_address[0] != 0)
2307 {
2308 sender_address = rewrite_address(sender_address, FALSE, TRUE,
2309 global_rewrite_rules, rewrite_existflags);
2310 DEBUG(D_receive|D_rewrite)
2311 debug_printf("rewritten sender = %s\n", sender_address);
2312 }
2313
2314
2315 /* The headers must be run through rewrite_header(), because it ensures that
2316 addresses are fully qualified, as well as applying any rewriting rules that may
2317 exist.
2318
2319 Qualification of header addresses in a message from a remote host happens only
2320 if the host is in sender_unqualified_hosts or recipient_unqualified hosts, as
2321 appropriate. For local messages, qualification always happens, unless -bnq is
2322 used to explicitly suppress it. No rewriting is done for an unqualified address
2323 that is left untouched.
2324
2325 We start at the second header, skipping our own Received:. This rewriting is
2326 documented as happening *after* recipient addresses are taken from the headers
2327 by the -t command line option. An added Sender: gets rewritten here. */
2328
2329 for (h = header_list->next; h != NULL; h = h->next)
2330 {
2331 header_line *newh = rewrite_header(h, NULL, NULL, global_rewrite_rules,
2332 rewrite_existflags, TRUE);
2333 if (newh != NULL) h = newh;
2334 }
2335
2336
2337 /* An RFC 822 (sic) message is not legal unless it has at least one of "to",
2338 "cc", or "bcc". Note that although the minimal examples in RFC822 show just
2339 "to" or "bcc", the full syntax spec allows "cc" as well. If any resent- header
2340 exists, this applies to the set of resent- headers rather than the normal set.
2341
2342 The requirement for a recipient header has been removed in RFC 2822. Earlier
2343 versions of Exim added a To: header for locally submitted messages, and an
2344 empty Bcc: header for others or when always_bcc was set. In the light of the
2345 changes in RFC 2822, we now always add Bcc: just in case there are still MTAs
2346 out there that insist on the RFC 822 syntax.
2347
2348 November 2003: While generally revising what Exim does to fix up headers, it
2349 seems like a good time to remove this altogether. */
2350
2351 /******
2352 if (!to_or_cc_header_exists && !bcc_header_exists)
2353 header_add(htype_bcc, "Bcc:\n");
2354 ******/
2355
2356 /* If there is no date header, generate one if the message originates locally
2357 (i.e. not over TCP/IP) or the submission mode flag is set. Messages without
2358 Date: are not valid, but it seems to be more confusing if Exim adds one to
2359 all remotely-originated messages. */
2360
2361 if (!date_header_exists && (sender_host_address == NULL || submission_mode))
2362 header_add(htype_other, "%sDate: %s\n", resent_prefix, tod_stamp(tod_full));
2363
2364 search_tidyup(); /* Free any cached resources */
2365
2366 /* Show the complete set of headers if debugging. Note that the first one (the
2367 new Received:) has not yet been set. */
2368
2369 DEBUG(D_receive)
2370 {
2371 debug_printf(">>Headers after rewriting and local additions:\n");
2372 for (h = header_list->next; h != NULL; h = h->next)
2373 debug_printf("%c %s", h->type, h->text);
2374 debug_printf("\n");
2375 }
2376
2377 /* The headers are now complete in store. If we are running in filter
2378 testing mode, that is all this function does. Return TRUE if the message
2379 ended with a dot. */
2380
2381 if (filter_test != NULL)
2382 {
2383 process_info[process_info_len] = 0;
2384 return message_ended == END_DOT;
2385 }
2386
2387 /* Open a new spool file for the data portion of the message. We need
2388 to access it both via a file descriptor and a stream. Try to make the
2389 directory if it isn't there. Note re use of sprintf: spool_directory
2390 is checked on input to be < 200 characters long. */
2391
2392 sprintf(CS spool_name, "%s/input/%s/%s-D", spool_directory, message_subdir,
2393 message_id);
2394 data_fd = Uopen(spool_name, O_RDWR|O_CREAT|O_EXCL, SPOOL_MODE);
2395 if (data_fd < 0)
2396 {
2397 if (errno == ENOENT)
2398 {
2399 uschar temp[16];
2400 sprintf(CS temp, "input/%s", message_subdir);
2401 if (message_subdir[0] == 0) temp[5] = 0;
2402 (void)directory_make(spool_directory, temp, INPUT_DIRECTORY_MODE, TRUE);
2403 data_fd = Uopen(spool_name, O_RDWR|O_CREAT|O_EXCL, SPOOL_MODE);
2404 }
2405 if (data_fd < 0)
2406 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Failed to create spool file %s: %s",
2407 spool_name, strerror(errno));
2408 }
2409
2410 /* Make sure the file's group is the Exim gid, and double-check the mode
2411 because the group setting doesn't always get set automatically. */
2412
2413 fchown(data_fd, exim_uid, exim_gid);
2414 fchmod(data_fd, SPOOL_MODE);
2415
2416 /* We now have data file open. Build a stream for it and lock it. We lock only
2417 the first line of the file (containing the message ID) because otherwise there
2418 are problems when Exim is run under Cygwin (I'm told). See comments in
2419 spool_in.c, where the same locking is done. */
2420
2421 data_file = fdopen(data_fd, "w+");
2422 lock_data.l_type = F_WRLCK;
2423 lock_data.l_whence = SEEK_SET;
2424 lock_data.l_start = 0;
2425 lock_data.l_len = SPOOL_DATA_START_OFFSET;
2426
2427 if (fcntl(data_fd, F_SETLK, &lock_data) < 0)
2428 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Cannot lock %s (%d): %s", spool_name,
2429 errno, strerror(errno));
2430
2431 /* We have an open, locked data file. Write the message id to it to make it
2432 self-identifying. Then read the remainder of the input of this message and
2433 write it to the data file. If the variable next != NULL, it contains the first
2434 data line (which was read as a header but then turned out not to have the right
2435 format); write it (remembering that it might contain binary zeros). The result
2436 of fwrite() isn't inspected; instead we call ferror() below. */
2437
2438 fprintf(data_file, "%s-D\n", message_id);
2439 if (next != NULL)
2440 {
2441 uschar *s = next->text;
2442 int len = next->slen;
2443 fwrite(s, 1, len, data_file);
2444 body_linecount++; /* Assumes only 1 line */
2445 }
2446
2447 /* Note that we might already be at end of file, or the logical end of file
2448 (indicated by '.'), or might have encountered an error while writing the
2449 message id or "next" line. */
2450
2451 if (!ferror(data_file) && !(receive_feof)() && message_ended != END_DOT)
2452 {
2453 if (smtp_input)
2454 {
2455 message_ended = read_message_data_smtp(data_file);
2456 receive_linecount++; /* The terminating "." line */
2457 }
2458 else message_ended = read_message_data(data_file);
2459
2460 receive_linecount += body_linecount; /* For BSMTP errors mainly */
2461
2462 /* Handle premature termination of SMTP */
2463
2464 if (smtp_input && message_ended == END_EOF)
2465 {
2466 Uunlink(spool_name); /* Lose data file when closed */
2467 message_id[0] = 0; /* Indicate no message accepted */
2468 smtp_reply = handle_lost_connection(US"");
2469 smtp_yield = FALSE;
2470 goto TIDYUP; /* Skip to end of function */
2471 }
2472
2473 /* Handle message that is too big. Don't use host_or_ident() in the log
2474 message; we want to see the ident value even for non-remote messages. */
2475
2476 if (message_ended == END_SIZE)
2477 {
2478 Uunlink(spool_name); /* Lose the data file when closed */
2479 if (smtp_input) receive_swallow_smtp(); /* Swallow incoming SMTP */
2480
2481 log_write(L_size_reject, LOG_MAIN|LOG_REJECT, "rejected from <%s>%s%s%s%s: "
2482 "message too big: read=%d max=%d",
2483 sender_address,
2484 (sender_fullhost == NULL)? "" : " H=",
2485 (sender_fullhost == NULL)? US"" : sender_fullhost,
2486 (sender_ident == NULL)? "" : " U=",
2487 (sender_ident == NULL)? US"" : sender_ident,
2488 message_size,
2489 thismessage_size_limit);
2490
2491 if (smtp_input)
2492 {
2493 smtp_reply = US"552 Message size exceeds maximum permitted";
2494 message_id[0] = 0; /* Indicate no message accepted */
2495 goto TIDYUP; /* Skip to end of function */
2496 }
2497 else
2498 {
2499 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
2500 give_local_error(ERRMESS_TOOBIG,
2501 string_sprintf("message too big (max=%d)", thismessage_size_limit),
2502 US"message rejected: ", error_rc, data_file, header_list);
2503 /* Does not return */
2504 }
2505 }
2506 }
2507
2508 /* Restore the standard SIGALRM handler for any subsequent processing. (For
2509 example, there may be some expansion in an ACL that uses a timer.) */
2510
2511 os_non_restarting_signal(SIGALRM, sigalrm_handler);
2512
2513 /* The message body has now been read into the data file. Call fflush() to
2514 empty the buffers in C, and then call fsync() to get the data written out onto
2515 the disk, as fflush() doesn't do this (or at least, it isn't documented as
2516 having to do this). If there was an I/O error on either input or output,
2517 attempt to send an error message, and unlink the spool file. For non-SMTP input
2518 we can then give up. Note that for SMTP input we must swallow the remainder of
2519 the input in cases of output errors, since the far end doesn't expect to see
2520 anything until the terminating dot line is sent. */
2521
2522 if (fflush(data_file) == EOF || ferror(data_file) ||
2523 fsync(fileno(data_file)) < 0 || (receive_ferror)())
2524 {
2525 uschar *msg_errno = US strerror(errno);
2526 BOOL input_error = (receive_ferror)() != 0;
2527 uschar *msg = string_sprintf("%s error (%s) while receiving message from %s",
2528 input_error? "Input read" : "Spool write",
2529 msg_errno,
2530 (sender_fullhost != NULL)? sender_fullhost : sender_ident);
2531
2532 log_write(0, LOG_MAIN, "Message abandoned: %s", msg);
2533 Uunlink(spool_name); /* Lose the data file */
2534
2535 if (smtp_input)
2536 {
2537 if (input_error)
2538 smtp_reply = US"451 Error while reading input data";
2539 else
2540 {
2541 smtp_reply = US"451 Error while writing spool file";
2542 receive_swallow_smtp();
2543 }
2544 message_id[0] = 0; /* Indicate no message accepted */
2545 goto TIDYUP; /* Skip to end of function */
2546 }
2547
2548 else
2549 {
2550 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
2551 give_local_error(ERRMESS_IOERR, msg, US"", error_rc, data_file,
2552 header_list);
2553 /* Does not return */
2554 }
2555 }
2556
2557
2558 /* No I/O errors were encountered while writing the data file. */
2559
2560 DEBUG(D_receive) debug_printf("Data file written for message %s\n", message_id);
2561
2562
2563 /* If there were any bad addresses extracted by -t, or there were no recipients
2564 left after -t, send a message to the sender of this message, or write it to
2565 stderr if the error handling option is set that way. Note that there may
2566 legitimately be no recipients for an SMTP message if they have all been removed
2567 by "discard".
2568
2569 We need to rewind the data file in order to read it. In the case of no
2570 recipients or stderr error writing, throw the data file away afterwards, and
2571 exit. (This can't be SMTP, which always ensures there's at least one
2572 syntactically good recipient address.) */
2573
2574 if (extract_recip && (bad_addresses != NULL || recipients_count == 0))
2575 {
2576 DEBUG(D_receive)
2577 {
2578 if (recipients_count == 0) debug_printf("*** No recipients\n");
2579 if (bad_addresses != NULL)
2580 {
2581 error_block *eblock = bad_addresses;
2582 debug_printf("*** Bad address(es)\n");
2583 while (eblock != NULL)
2584 {
2585 debug_printf(" %s: %s\n", eblock->text1, eblock->text2);
2586 eblock = eblock->next;
2587 }
2588 }
2589 }
2590
2591 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
2592
2593 /* If configured to send errors to the sender, but this fails, force
2594 a failure error code. We use a special one for no recipients so that it
2595 can be detected by the autoreply transport. Otherwise error_rc is set to
2596 errors_sender_rc, which is EXIT_FAILURE unless -oee was given, in which case
2597 it is EXIT_SUCCESS. */
2598
2599 if (error_handling == ERRORS_SENDER)
2600 {
2601 if (!moan_to_sender(
2602 (bad_addresses == NULL)?
2603 (extracted_ignored? ERRMESS_IGADDRESS : ERRMESS_NOADDRESS) :
2604 (recipients_list == NULL)? ERRMESS_BADNOADDRESS : ERRMESS_BADADDRESS,
2605 bad_addresses, header_list, data_file, FALSE))
2606 error_rc = (bad_addresses == NULL)? EXIT_NORECIPIENTS : EXIT_FAILURE;
2607 }
2608 else
2609 {
2610 if (bad_addresses == NULL)
2611 {
2612 if (extracted_ignored)
2613 fprintf(stderr, "exim: all -t recipients overridden by command line\n");
2614 else
2615 fprintf(stderr, "exim: no recipients in message\n");
2616 }
2617 else
2618 {
2619 fprintf(stderr, "exim: invalid address%s",
2620 (bad_addresses->next == NULL)? ":" : "es:\n");
2621 while (bad_addresses != NULL)
2622 {
2623 fprintf(stderr, " %s: %s\n", bad_addresses->text1,
2624 bad_addresses->text2);
2625 bad_addresses = bad_addresses->next;
2626 }
2627 }
2628 }
2629
2630 if (recipients_count == 0 || error_handling == ERRORS_STDERR)
2631 {
2632 Uunlink(spool_name);
2633 fclose(data_file);
2634 exim_exit(error_rc);
2635 }
2636 }
2637
2638 /* Data file successfully written. Generate text for the Received: header by
2639 expanding the configured string, and adding a timestamp. By leaving this
2640 operation till now, we ensure that the timestamp is the time that message
2641 reception was completed. However, this is deliberately done before calling the
2642 data ACL and local_scan().
2643
2644 This Received: header may therefore be inspected by the data ACL and by code in
2645 the local_scan() function. When they have run, we update the timestamp to be
2646 the final time of reception.
2647
2648 If there is just one recipient, set up its value in the $received_for variable
2649 for use when we generate the Received: header.
2650
2651 Note: the checking for too many Received: headers is handled by the delivery
2652 code. */
2653
2654 timestamp = expand_string(US"${tod_full}");
2655 if (recipients_count == 1) received_for = recipients_list[0].address;
2656 received = expand_string(received_header_text);
2657 received_for = NULL;
2658
2659 if (received == NULL)
2660 {
2661 Uunlink(spool_name); /* Lose the data file */
2662 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Expansion of \"%s\" "
2663 "(received_header_text) failed: %s", string_printing(received_header_text),
2664 expand_string_message);
2665 }
2666
2667 /* The first element on the header chain is reserved for the Received header,
2668 so all we have to do is fill in the text pointer, and set the type. However, if
2669 the result of the expansion is an empty string, we leave the header marked as
2670 "old" so as to refrain from adding a Received header. */
2671
2672 if (received[0] == 0)
2673 {
2674 received_header->text = string_sprintf("Received: ; %s\n", timestamp);
2675 received_header->type = htype_old;
2676 }
2677 else
2678 {
2679 received_header->text = string_sprintf("%s; %s\n", received, timestamp);
2680 received_header->type = htype_received;
2681 }
2682
2683 received_header->slen = Ustrlen(received_header->text);
2684
2685 DEBUG(D_receive) debug_printf(">>Generated Received: header line\n%c %s",
2686 received_header->type, received_header->text);
2687
2688 /* Set the value of message_body_size for the DATA ACL and for local_scan() */
2689
2690 message_body_size = (fstat(data_fd, &statbuf) == 0)?
2691 statbuf.st_size - SPOOL_DATA_START_OFFSET : -1;
2692
2693 /* If an ACL from any RCPT commands set up any warning headers to add, do so
2694 now, before running the DATA ACL. */
2695
2696 add_acl_headers(US"MAIL or RCPT");
2697
2698 /* If an ACL is specified for checking things at this stage of reception of a
2699 message, run it, unless all the recipients were removed by "discard" in earlier
2700 ACLs. That is the only case in which recipients_count can be zero at this
2701 stage. Set deliver_datafile to point to the data file so that $message_body and
2702 $message_body_end can be extracted if needed. Allow $recipients in expansions.
2703 */
2704
2705 deliver_datafile = data_fd;
2706
2707 if (recipients_count == 0)
2708 {
2709 blackholed_by = recipients_discarded? US"MAIL ACL" : US"RCPT ACL";
2710 }
2711 else
2712 {
2713 enable_dollar_recipients = TRUE;
2714
2715 /* Handle interactive SMTP messages */
2716
2717 if (smtp_input && !smtp_batched_input)
2718 {
2719 if (acl_smtp_data != NULL && recipients_count > 0)
2720 {
2721 uschar *user_msg, *log_msg;
2722 rc = acl_check(ACL_WHERE_DATA, NULL, acl_smtp_data, &user_msg, &log_msg);
2723 add_acl_headers(US"DATA");
2724 if (rc == DISCARD)
2725 {
2726 recipients_count = 0;
2727 blackholed_by = US"DATA ACL";
2728 }
2729 else if (rc != OK)
2730 {
2731 Uunlink(spool_name);
2732 if (smtp_handle_acl_fail(ACL_WHERE_DATA, rc, user_msg, log_msg) != 0)
2733 smtp_yield = FALSE; /* No more messsages after dropped connection */
2734 smtp_reply = US""; /* Indicate reply already sent */
2735 message_id[0] = 0; /* Indicate no message accepted */
2736 goto TIDYUP; /* Skip to end of function */
2737 }
2738 }
2739 }
2740
2741 /* Handle non-SMTP and batch SMTP (i.e. non-interactive) messages. Note that
2742 we cannot take different actions for permanent and temporary rejections. */
2743
2744 else if (acl_not_smtp != NULL)
2745 {
2746 uschar *user_msg, *log_msg;
2747 rc = acl_check(ACL_WHERE_NOTSMTP, NULL, acl_not_smtp, &user_msg, &log_msg);
2748 if (rc == DISCARD)
2749 {
2750 recipients_count = 0;
2751 blackholed_by = US"non-SMTP ACL";
2752 }
2753 else if (rc != OK)
2754 {
2755 Uunlink(spool_name);
2756 log_write(0, LOG_MAIN|LOG_REJECT, "F=<%s> rejected by non-SMTP ACL: %s",
2757 sender_address, log_msg);
2758 if (user_msg == NULL) user_msg = US"local configuration problem";
2759 if (smtp_batched_input)
2760 {
2761 moan_smtp_batch(NULL, "%d %s", 550, user_msg);
2762 /* Does not return */
2763 }
2764 else
2765 {
2766 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
2767 give_local_error(ERRMESS_LOCAL_ACL, user_msg,
2768 US"message rejected by non-SMTP ACL: ", error_rc, data_file,
2769 header_list);
2770 /* Does not return */
2771 }
2772 }
2773 add_acl_headers(US"non-SMTP");
2774 }
2775
2776 if (deliver_freeze) frozen_by = US"ACL"; /* for later logging */
2777 if (queue_only_policy) queued_by = US"ACL";
2778
2779 enable_dollar_recipients = FALSE;
2780 }
2781
2782 /* The final check on the message is to run the scan_local() function. The
2783 version supplied with Exim always accepts, but this is a hook for sysadmins to
2784 supply their own checking code. The local_scan() function is run even when all
2785 the recipients have been discarded. */
2786
2787 lseek(data_fd, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
2788
2789 /* Arrange to catch crashes in local_scan(), so that the -D file gets
2790 deleted, and the incident gets logged. */
2791
2792 os_non_restarting_signal(SIGSEGV, local_scan_crash_handler);
2793 os_non_restarting_signal(SIGFPE, local_scan_crash_handler);
2794 os_non_restarting_signal(SIGILL, local_scan_crash_handler);
2795 os_non_restarting_signal(SIGBUS, local_scan_crash_handler);
2796
2797 DEBUG(D_receive) debug_printf("calling local_scan(); timeout=%d\n",
2798 local_scan_timeout);
2799 local_scan_data = NULL;
2800
2801 os_non_restarting_signal(SIGALRM, local_scan_timeout_handler);
2802 if (local_scan_timeout > 0) alarm(local_scan_timeout);
2803 rc = local_scan(data_fd, &local_scan_data);
2804 alarm(0);
2805 os_non_restarting_signal(SIGALRM, sigalrm_handler);
2806
2807 store_pool = POOL_MAIN; /* In case changed */
2808 DEBUG(D_receive) debug_printf("local_scan() returned %d %s\n", rc,
2809 local_scan_data);
2810
2811 os_non_restarting_signal(SIGSEGV, SIG_DFL);
2812 os_non_restarting_signal(SIGFPE, SIG_DFL);
2813 os_non_restarting_signal(SIGILL, SIG_DFL);
2814 os_non_restarting_signal(SIGBUS, SIG_DFL);
2815
2816 /* The length check is paranoia against some runaway code, and also because
2817 (for a success return) lines in the spool file are read into big_buffer. */
2818
2819 if (local_scan_data != NULL)
2820 {
2821 int len = Ustrlen(local_scan_data);
2822 if (len > LOCAL_SCAN_MAX_RETURN) len = LOCAL_SCAN_MAX_RETURN;
2823 local_scan_data = string_copyn(local_scan_data, len);
2824 }
2825
2826 if (rc == LOCAL_SCAN_ACCEPT_FREEZE)
2827 {
2828 if (!deliver_freeze) /* ACL might have already frozen */
2829 {
2830 deliver_freeze = TRUE;
2831 deliver_frozen_at = time(NULL);
2832 frozen_by = US"local_scan()";
2833 }
2834 rc = LOCAL_SCAN_ACCEPT;
2835 }
2836 else if (rc == LOCAL_SCAN_ACCEPT_QUEUE)
2837 {
2838 if (!queue_only_policy) /* ACL might have already queued */
2839 {
2840 queue_only_policy = TRUE;
2841 queued_by = US"local_scan()";
2842 }
2843 rc = LOCAL_SCAN_ACCEPT;
2844 }
2845
2846 /* Message accepted: remove newlines in local_scan_data because otherwise
2847 the spool file gets corrupted. Ensure that all recipients are qualified. */
2848
2849 if (rc == LOCAL_SCAN_ACCEPT)
2850 {
2851 if (local_scan_data != NULL)
2852 {
2853 uschar *s;
2854 for (s = local_scan_data; *s != 0; s++) if (*s == '\n') *s = ' ';
2855 }
2856 for (i = 0; i < recipients_count; i++)
2857 {
2858 recipient_item *r = recipients_list + i;
2859 r->address = rewrite_address_qualify(r->address, TRUE);
2860 if (r->errors_to != NULL)
2861 r->errors_to = rewrite_address_qualify(r->errors_to, TRUE);
2862 }
2863 if (recipients_count == 0 && blackholed_by == NULL)
2864 blackholed_by = US"local_scan";
2865 }
2866
2867 /* Message rejected: newlines permitted in local_scan_data to generate
2868 multiline SMTP responses. */
2869
2870 else
2871 {
2872 uschar *istemp = US"";
2873 uschar *s = NULL;
2874 int size = 0;
2875 int sptr = 0;
2876 int code;
2877
2878 errmsg = local_scan_data;
2879
2880 Uunlink(spool_name); /* Cancel this message */
2881 switch(rc)
2882 {
2883 default:
2884 log_write(0, LOG_MAIN, "invalid return %d from local_scan(). Temporary "
2885 "rejection given", rc);
2886 goto TEMPREJECT;
2887
2888 case LOCAL_SCAN_REJECT_NOLOGHDR:
2889 log_extra_selector &= ~LX_rejected_header;
2890 /* Fall through */
2891
2892 case LOCAL_SCAN_REJECT:
2893 code = 550;
2894 if (errmsg == NULL) errmsg = US"Administrative prohibition";
2895 break;
2896
2897 case LOCAL_SCAN_TEMPREJECT_NOLOGHDR:
2898 log_extra_selector &= ~LX_rejected_header;
2899 /* Fall through */
2900
2901 case LOCAL_SCAN_TEMPREJECT:
2902 TEMPREJECT:
2903 code = 451;
2904 if (errmsg == NULL) errmsg = US"Temporary local problem";
2905 istemp = US"temporarily ";
2906 break;
2907 }
2908
2909 s = string_append(s, &size, &sptr, 2, US"F=",
2910 (sender_address[0] == 0)? US"<>" : sender_address);
2911 s = add_host_info_for_log(s, &size, &sptr);
2912 s[sptr] = 0;
2913
2914 log_write(0, LOG_MAIN|LOG_REJECT, "%s %srejected by local_scan(): %.256s",
2915 s, istemp, string_printing(errmsg));
2916
2917 if (smtp_input)
2918 {
2919 if (!smtp_batched_input)
2920 {
2921 smtp_respond(code, TRUE, errmsg);
2922 message_id[0] = 0; /* Indicate no message accepted */
2923 smtp_reply = US""; /* Indicate reply already sent */
2924 goto TIDYUP; /* Skip to end of function */
2925 }
2926 else
2927 {
2928 moan_smtp_batch(NULL, "%d %s", code, errmsg);
2929 /* Does not return */
2930 }
2931 }
2932 else
2933 {
2934 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
2935 give_local_error(ERRMESS_LOCAL_SCAN, errmsg,
2936 US"message rejected by local scan code: ", error_rc, data_file,
2937 header_list);
2938 /* Does not return */
2939 }
2940 }
2941
2942 /* Reset signal handlers to ignore signals that previously would have caused
2943 the message to be abandoned. */
2944
2945 signal(SIGTERM, SIG_IGN);
2946 signal(SIGINT, SIG_IGN);
2947
2948 /* Ensure the first time flag is set in the newly-received message. */
2949
2950 deliver_firsttime = TRUE;
2951
2952 /* Update the timstamp in our Received: header to account for any time taken by
2953 an ACL or by local_scan(). The new time is the time that all reception
2954 processing is complete. */
2955
2956 timestamp = expand_string(US"${tod_full}");
2957 tslen = Ustrlen(timestamp);
2958
2959 memcpy(received_header->text + received_header->slen - tslen - 1,
2960 timestamp, tslen);
2961
2962 /* In MUA wrapper mode, ignore queueing actions set by ACL or local_scan() */
2963
2964 if (mua_wrapper)
2965 {
2966 deliver_freeze = FALSE;
2967 queue_only_policy = FALSE;
2968 }
2969
2970 /* Keep the data file open until we have written the header file, in order to
2971 hold onto the lock. In a -bh run, or if the message is to be blackholed, we
2972 don't write the header file, and we unlink the data file. If writing the header
2973 file fails, we have failed to accept this message. */
2974
2975 if (host_checking || blackholed_by != NULL)
2976 {
2977 header_line *h;
2978 Uunlink(spool_name);
2979 msg_size = 0; /* Compute size for log line */
2980 for (h = header_list; h != NULL; h = h->next)
2981 if (h->type != '*') msg_size += h->slen;
2982 }
2983
2984 /* Write the -H file */
2985
2986 else
2987 {
2988 if ((msg_size = spool_write_header(message_id, SW_RECEIVING, &errmsg)) < 0)
2989 {
2990 log_write(0, LOG_MAIN, "Message abandoned: %s", errmsg);
2991 Uunlink(spool_name); /* Lose the data file */
2992
2993 if (smtp_input)
2994 {
2995 smtp_reply = US"451 Error in writing spool file";
2996 message_id[0] = 0; /* Indicate no message accepted */
2997 goto TIDYUP;
2998 }
2999 else
3000 {
3001 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3002 give_local_error(ERRMESS_IOERR, errmsg, US"", error_rc, data_file,
3003 header_list);
3004 /* Does not return */
3005 }
3006 }
3007 }
3008
3009
3010 /* The message has now been successfully received. */
3011
3012 receive_messagecount++;
3013
3014 /* In SMTP sessions we may receive several in one connection. After each one,
3015 we wait for the clock to tick at the level of message-id granularity. This is
3016 so that the combination of time+pid is unique, even on systems where the pid
3017 can be re-used within our time interval. We can't shorten the interval without
3018 re-designing the message-id. See comments above where the message id is
3019 created. This is Something For The Future. */
3020
3021 message_id_tv.tv_usec = (message_id_tv.tv_usec/id_resolution) * id_resolution;
3022 exim_wait_tick(&message_id_tv, id_resolution);
3023
3024 /* Add data size to written header size. We do not count the initial file name
3025 that is in the file, but we do add one extra for the notional blank line that
3026 precedes the data. This total differs from message_size in that it include the
3027 added Received: header and any other headers that got created locally. */
3028
3029 fflush(data_file);
3030 fstat(data_fd, &statbuf);
3031
3032 msg_size += statbuf.st_size - SPOOL_DATA_START_OFFSET + 1;
3033
3034 /* Generate a "message received" log entry. We do this by building up a dynamic
3035 string as required. Since we commonly want to add two items at a time, use a
3036 macro to simplify the coding. We log the arrival of a new message while the
3037 file is still locked, just in case the machine is *really* fast, and delivers
3038 it first! Include any message id that is in the message - since the syntax of a
3039 message id is actually an addr-spec, we can use the parse routine to canonicize
3040 it. */
3041
3042 size = 256;
3043 sptr = 0;
3044 s = store_get(size);
3045
3046 s = string_append(s, &size, &sptr, 2, US"<= ",
3047 (sender_address[0] == 0)? US"<>" : sender_address);
3048 if (message_reference != NULL)
3049 s = string_append(s, &size, &sptr, 2, US" R=", message_reference);
3050
3051 s = add_host_info_for_log(s, &size, &sptr);
3052
3053 #ifdef SUPPORT_TLS
3054 if ((log_extra_selector & LX_tls_cipher) != 0 && tls_cipher != NULL)
3055 s = string_append(s, &size, &sptr, 2, US" X=", tls_cipher);
3056 if ((log_extra_selector & LX_tls_certificate_verified) != 0 &&
3057 tls_cipher != NULL)
3058 s = string_append(s, &size, &sptr, 2, US" CV=",
3059 tls_certificate_verified? "yes":"no");
3060 if ((log_extra_selector & LX_tls_peerdn) != 0 && tls_peerdn != NULL)
3061 s = string_append(s, &size, &sptr, 3, US" DN=\"", tls_peerdn, US"\"");
3062 #endif
3063
3064 if (sender_host_authenticated != NULL)
3065 {
3066 s = string_append(s, &size, &sptr, 2, US" A=", sender_host_authenticated);
3067 if (authenticated_id != NULL)
3068 s = string_append(s, &size, &sptr, 2, US":", authenticated_id);
3069 }
3070
3071 sprintf(CS big_buffer, "%d", msg_size);
3072 s = string_append(s, &size, &sptr, 2, US" S=", big_buffer);
3073
3074 /* If an addr-spec in a message-id contains a quoted string, it can contain
3075 any characters except " \ and CR and so in particular it can contain NL!
3076 Therefore, make sure we use a printing-characters only version for the log.
3077 Also, allow for domain literals in the message id. */
3078
3079 if (msgid_header != NULL)
3080 {
3081 uschar *old_id;
3082 BOOL save_allow_domain_literals = allow_domain_literals;
3083 allow_domain_literals = TRUE;
3084 old_id = parse_extract_address(Ustrchr(msgid_header->text, ':') + 1,
3085 &errmsg, &start, &end, &domain, FALSE);
3086 allow_domain_literals = save_allow_domain_literals;
3087 if (old_id != NULL)
3088 s = string_append(s, &size, &sptr, 2, US" id=", string_printing(old_id));
3089 }
3090
3091 /* If subject logging is turned on, create suitable printing-character
3092 text. By expanding $h_subject: we make use of the MIME decoding. */
3093
3094 if ((log_extra_selector & LX_subject) != 0 && subject_header != NULL)
3095 {
3096 int i;
3097 uschar *p = big_buffer;
3098 uschar *ss = expand_string(US"$h_subject:");
3099
3100 /* Backslash-quote any double quotes or backslashes so as to make a
3101 a C-like string, and turn any non-printers into escape sequences. */
3102
3103 *p++ = '\"';
3104 if (*ss != 0) for (i = 0; i < 100 && ss[i] != 0; i++)
3105 {
3106 if (ss[i] == '\"' || ss[i] == '\\') *p++ = '\\';
3107 *p++ = ss[i];
3108 }
3109 *p++ = '\"';
3110 *p = 0;
3111 s = string_append(s, &size, &sptr, 2, US" T=", string_printing(big_buffer));
3112 }
3113
3114 /* Terminate the string: string_cat() and string_append() leave room, but do
3115 not put the zero in. */
3116
3117 s[sptr] = 0;
3118
3119 /* While writing to the log, set a flag to cause a call to receive_bomb_out()
3120 if the log cannot be opened. */
3121
3122 receive_call_bombout = TRUE;
3123 log_write(0, LOG_MAIN |
3124 (((log_extra_selector & LX_received_recipients) != 0)? LOG_RECIPIENTS : 0) |
3125 (((log_extra_selector & LX_received_sender) != 0)? LOG_SENDER : 0),
3126 "%s", s);
3127 receive_call_bombout = FALSE;
3128
3129 /* Log any control actions taken by an ACL or local_scan(). */
3130
3131 if (deliver_freeze) log_write(0, LOG_MAIN, "frozen by %s", frozen_by);
3132 if (queue_only_policy) log_write(L_delay_delivery, LOG_MAIN,
3133 "no immediate delivery: queued by %s", queued_by);
3134
3135 /* Create a message log file if message logs are being used and this message is
3136 not blackholed. Write the reception stuff to it. We used to leave message log
3137 creation until the first delivery, but this has proved confusing for somep
3138 people. */
3139
3140 if (message_logs && blackholed_by == NULL)
3141 {
3142 int fd;
3143
3144 sprintf(CS spool_name, "%s/msglog/%s/%s", spool_directory, message_subdir,
3145 message_id);
3146 fd = Uopen(spool_name, O_WRONLY|O_APPEND|O_CREAT, SPOOL_MODE);
3147
3148 if (fd < 0 && errno == ENOENT)
3149 {
3150 uschar temp[16];
3151 sprintf(CS temp, "msglog/%s", message_subdir);
3152 if (message_subdir[0] == 0) temp[6] = 0;
3153 (void)directory_make(spool_directory, temp, MSGLOG_DIRECTORY_MODE, TRUE);
3154 fd = Uopen(spool_name, O_WRONLY|O_APPEND|O_CREAT, SPOOL_MODE);
3155 }
3156
3157 if (fd < 0)
3158 {
3159 log_write(0, LOG_MAIN|LOG_PANIC, "Couldn't open message log %s: %s",
3160 spool_name, strerror(errno));
3161 }
3162
3163 else
3164 {
3165 FILE *message_log = fdopen(fd, "a");
3166 if (message_log == NULL)
3167 {
3168 log_write(0, LOG_MAIN|LOG_PANIC, "Couldn't fdopen message log %s: %s",
3169 spool_name, strerror(errno));
3170 close(fd);
3171 }
3172 else
3173 {
3174 uschar *now = tod_stamp(tod_log);
3175 fprintf(message_log, "%s Received from %s\n", now, s+3);
3176 if (deliver_freeze) fprintf(message_log, "%s frozen by %s\n", now,
3177 frozen_by);
3178 if (queue_only_policy) fprintf(message_log,
3179 "%s no immediate delivery: queued by %s\n", now, queued_by);
3180 fclose(message_log);
3181 }
3182 }
3183 }
3184
3185 store_reset(s); /* The store for the main log message can be reused */
3186
3187 /* If the message is frozen, and freeze_tell is set, do the telling. */
3188
3189 if (deliver_freeze && freeze_tell != NULL && freeze_tell[0] != 0)
3190 {
3191 moan_tell_someone(freeze_tell, NULL, US"Message frozen on arrival",
3192 "Message %s was frozen on arrival by %s.\nThe sender is <%s>.\n",
3193 message_id, frozen_by, sender_address);
3194 }
3195
3196
3197 /* Either a message has been successfully received and written to the two spool
3198 files, or an error in writing the spool has occurred for an SMTP message, or
3199 an SMTP message has been rejected because of a bad sender. (For a non-SMTP
3200 message we will have already given up because there's no point in carrying on!)
3201 In either event, we must now close (and thereby unlock) the data file. In the
3202 successful case, this leaves the message on the spool, ready for delivery. In
3203 the error case, the spool file will be deleted. Then tidy up store, interact
3204 with an SMTP call if necessary, and return.
3205
3206 A fflush() was done earlier in the expectation that any write errors on the
3207 data file will be flushed(!) out thereby. Nevertheless, it is theoretically
3208 possible for fclose() to fail - but what to do? What has happened to the lock
3209 if this happens? */
3210
3211 TIDYUP:
3212 process_info[process_info_len] = 0; /* Remove message id */
3213 if (data_file != NULL) fclose(data_file); /* Frees the lock */
3214
3215 /* Now reset signal handlers to their defaults */
3216
3217 signal(SIGTERM, SIG_DFL);
3218 signal(SIGINT, SIG_DFL);
3219
3220 /* Tell an SMTP caller the state of play, and arrange to return the SMTP return
3221 value, which defaults TRUE - meaning there may be more incoming messages from
3222 this connection. For non-SMTP callers (where there is only ever one message),
3223 the default is FALSE. */
3224
3225 if (smtp_input)
3226 {
3227 yield = smtp_yield;
3228
3229 /* Handle interactive SMTP callers. After several kinds of error, smtp_reply
3230 is set to the response. However, after an ACL error or local_scan() error,
3231 the response has already been sent, and smtp_reply is an empty string to
3232 indicate this. */
3233
3234 if (!smtp_batched_input)
3235 {
3236 if (smtp_reply == NULL)
3237 {
3238 smtp_printf("250 OK id=%s\r\n", message_id);
3239 if (host_checking)
3240 fprintf(stdout,
3241 "\n**** SMTP testing: that is not a real message id!\n\n");
3242 }
3243 else if (smtp_reply[0] != 0) smtp_printf("%.1024s\r\n", smtp_reply);
3244 }
3245
3246 /* For batched SMTP, generate an error message on failure, and do
3247 nothing on success. The function moan_smtp_batch() does not return -
3248 it exits from the program with a non-zero return code. */
3249
3250 else if (smtp_reply != NULL) moan_smtp_batch(NULL, "%s", smtp_reply);
3251 }
3252
3253
3254 /* If blackholing, we can immediately log this message's sad fate. The data
3255 file has already been unlinked, and the header file was never written to disk.
3256 We must now indicate that nothing was received, to prevent a delivery from
3257 starting. */
3258
3259 if (blackholed_by != NULL)
3260 {
3261 uschar *detail = (local_scan_data != NULL)?
3262 string_printing(local_scan_data) :
3263 string_sprintf("(%s discarded recipients)", blackholed_by);
3264 log_write(0, LOG_MAIN, "=> blackhole %s", detail);
3265 log_write(0, LOG_MAIN, "Completed");
3266 message_id[0] = 0;
3267 }
3268
3269 /* Reset headers so that logging of rejects for a subsequent message doesn't
3270 include them. It is also important to set header_last = NULL before exiting
3271 from this function, as this prevents certain rewrites that might happen during
3272 subsequent verifying (of another incoming message) from trying to add headers
3273 when they shouldn't. */
3274
3275 header_list = header_last = NULL;
3276
3277 return yield; /* TRUE if more messages (SMTP only) */
3278 }
3279
3280 /* End of receive.c */