Logging-only patch for 8BITMIME; bug 817.
[exim.git] / src / src / receive.c
1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
4
5 /* Copyright (c) University of Cambridge 1995 - 2012 */
6 /* See the file NOTICE for conditions of use and distribution. */
7
8 /* Code for receiving a message and setting up spool files. */
9
10 #include "exim.h"
11
12 #ifdef EXPERIMENTAL_DCC
13 extern int dcc_ok;
14 #endif
15
16 /*************************************************
17 * Local static variables *
18 *************************************************/
19
20 static FILE *data_file = NULL;
21 static int data_fd = -1;
22 static uschar spool_name[256];
23
24
25
26 /*************************************************
27 * Non-SMTP character reading functions *
28 *************************************************/
29
30 /* These are the default functions that are set up in the variables such as
31 receive_getc initially. They just call the standard functions, passing stdin as
32 the file. (When SMTP input is occurring, different functions are used by
33 changing the pointer variables.) */
34
35 int
36 stdin_getc(void)
37 {
38 return getc(stdin);
39 }
40
41 int
42 stdin_ungetc(int c)
43 {
44 return ungetc(c, stdin);
45 }
46
47 int
48 stdin_feof(void)
49 {
50 return feof(stdin);
51 }
52
53 int
54 stdin_ferror(void)
55 {
56 return ferror(stdin);
57 }
58
59
60
61
62 /*************************************************
63 * Check that a set sender is allowed *
64 *************************************************/
65
66 /* This function is called when a local caller sets an explicit sender address.
67 It checks whether this is permitted, which it is for trusted callers.
68 Otherwise, it must match the pattern(s) in untrusted_set_sender.
69
70 Arguments: the proposed sender address
71 Returns: TRUE for a trusted caller
72 TRUE if the address has been set, untrusted_set_sender has been
73 set, and the address matches something in the list
74 FALSE otherwise
75 */
76
77 BOOL
78 receive_check_set_sender(uschar *newsender)
79 {
80 uschar *qnewsender;
81 if (trusted_caller) return TRUE;
82 if (newsender == NULL || untrusted_set_sender == NULL) return FALSE;
83 qnewsender = (Ustrchr(newsender, '@') != NULL)?
84 newsender : string_sprintf("%s@%s", newsender, qualify_domain_sender);
85 return
86 match_address_list(qnewsender, TRUE, TRUE, &untrusted_set_sender, NULL, -1,
87 0, NULL) == OK;
88 }
89
90
91
92
93 /*************************************************
94 * Read space info for a partition *
95 *************************************************/
96
97 /* This function is called by receive_check_fs() below, and also by string
98 expansion for variables such as $spool_space. The field names for the statvfs
99 structure are macros, because not all OS have F_FAVAIL and it seems tidier to
100 have macros for F_BAVAIL and F_FILES as well. Some kinds of file system do not
101 have inodes, and they return -1 for the number available.
102
103 Later: It turns out that some file systems that do not have the concept of
104 inodes return 0 rather than -1. Such systems should also return 0 for the total
105 number of inodes, so we require that to be greater than zero before returning
106 an inode count.
107
108 Arguments:
109 isspool TRUE for spool partition, FALSE for log partition
110 inodeptr address of int to receive inode count; -1 if there isn't one
111
112 Returns: available on-root space, in kilobytes
113 -1 for log partition if there isn't one
114
115 All values are -1 if the STATFS functions are not available.
116 */
117
118 int
119 receive_statvfs(BOOL isspool, int *inodeptr)
120 {
121 #ifdef HAVE_STATFS
122 struct STATVFS statbuf;
123 uschar *path;
124 uschar *name;
125 uschar buffer[1024];
126
127 /* The spool directory must always exist. */
128
129 if (isspool)
130 {
131 path = spool_directory;
132 name = US"spool";
133 }
134
135 /* Need to cut down the log file path to the directory, and to ignore any
136 appearance of "syslog" in it. */
137
138 else
139 {
140 int sep = ':'; /* Not variable - outside scripts use */
141 uschar *p = log_file_path;
142 name = US"log";
143
144 /* An empty log_file_path means "use the default". This is the same as an
145 empty item in a list. */
146
147 if (*p == 0) p = US":";
148 while ((path = string_nextinlist(&p, &sep, buffer, sizeof(buffer))) != NULL)
149 {
150 if (Ustrcmp(path, "syslog") != 0) break;
151 }
152
153 if (path == NULL) /* No log files */
154 {
155 *inodeptr = -1;
156 return -1;
157 }
158
159 /* An empty string means use the default, which is in the spool directory.
160 But don't just use the spool directory, as it is possible that the log
161 subdirectory has been symbolically linked elsewhere. */
162
163 if (path[0] == 0)
164 {
165 sprintf(CS buffer, CS"%s/log", CS spool_directory);
166 path = buffer;
167 }
168 else
169 {
170 uschar *cp;
171 if ((cp = Ustrrchr(path, '/')) != NULL) *cp = 0;
172 }
173 }
174
175 /* We now have the path; do the business */
176
177 memset(&statbuf, 0, sizeof(statbuf));
178
179 if (STATVFS(CS path, &statbuf) != 0)
180 {
181 log_write(0, LOG_MAIN|LOG_PANIC, "cannot accept message: failed to stat "
182 "%s directory %s: %s", name, spool_directory, strerror(errno));
183 smtp_closedown(US"spool or log directory problem");
184 exim_exit(EXIT_FAILURE);
185 }
186
187 *inodeptr = (statbuf.F_FILES > 0)? statbuf.F_FAVAIL : -1;
188
189 /* Disks are getting huge. Take care with computing the size in kilobytes. */
190
191 return (int)(((double)statbuf.F_BAVAIL * (double)statbuf.F_FRSIZE)/1024.0);
192
193 /* Unable to find partition sizes in this environment. */
194
195 #else
196 *inodeptr = -1;
197 return -1;
198 #endif
199 }
200
201
202
203
204 /*************************************************
205 * Check space on spool and log partitions *
206 *************************************************/
207
208 /* This function is called before accepting a message; if any thresholds are
209 set, it checks them. If a message_size is supplied, it checks that there is
210 enough space for that size plus the threshold - i.e. that the message won't
211 reduce the space to the threshold. Not all OS have statvfs(); for those that
212 don't, this function always returns TRUE. For some OS the old function and
213 struct name statfs is used; that is handled by a macro, defined in exim.h.
214
215 Arguments:
216 msg_size the (estimated) size of an incoming message
217
218 Returns: FALSE if there isn't enough space, or if the information cannot
219 be obtained
220 TRUE if no check was done or there is enough space
221 */
222
223 BOOL
224 receive_check_fs(int msg_size)
225 {
226 int space, inodes;
227
228 if (check_spool_space > 0 || msg_size > 0 || check_spool_inodes > 0)
229 {
230 space = receive_statvfs(TRUE, &inodes);
231
232 DEBUG(D_receive)
233 debug_printf("spool directory space = %dK inodes = %d "
234 "check_space = %dK inodes = %d msg_size = %d\n",
235 space, inodes, check_spool_space, check_spool_inodes, msg_size);
236
237 if ((space >= 0 && space < check_spool_space) ||
238 (inodes >= 0 && inodes < check_spool_inodes))
239 {
240 log_write(0, LOG_MAIN, "spool directory space check failed: space=%d "
241 "inodes=%d", space, inodes);
242 return FALSE;
243 }
244 }
245
246 if (check_log_space > 0 || check_log_inodes > 0)
247 {
248 space = receive_statvfs(FALSE, &inodes);
249
250 DEBUG(D_receive)
251 debug_printf("log directory space = %dK inodes = %d "
252 "check_space = %dK inodes = %d\n",
253 space, inodes, check_log_space, check_log_inodes);
254
255 if ((space >= 0 && space < check_log_space) ||
256 (inodes >= 0 && inodes < check_log_inodes))
257 {
258 log_write(0, LOG_MAIN, "log directory space check failed: space=%d "
259 "inodes=%d", space, inodes);
260 return FALSE;
261 }
262 }
263
264 return TRUE;
265 }
266
267
268
269 /*************************************************
270 * Bomb out while reading a message *
271 *************************************************/
272
273 /* The common case of wanting to bomb out is if a SIGTERM or SIGINT is
274 received, or if there is a timeout. A rarer case might be if the log files are
275 screwed up and Exim can't open them to record a message's arrival. Handling
276 that case is done by setting a flag to cause the log functions to call this
277 function if there is an ultimate disaster. That is why it is globally
278 accessible.
279
280 Arguments:
281 reason text reason to pass to the not-quit ACL
282 msg default SMTP response to give if in an SMTP session
283 Returns: it doesn't
284 */
285
286 void
287 receive_bomb_out(uschar *reason, uschar *msg)
288 {
289 /* If spool_name is set, it contains the name of the data file that is being
290 written. Unlink it before closing so that it cannot be picked up by a delivery
291 process. Ensure that any header file is also removed. */
292
293 if (spool_name[0] != 0)
294 {
295 Uunlink(spool_name);
296 spool_name[Ustrlen(spool_name) - 1] = 'H';
297 Uunlink(spool_name);
298 }
299
300 /* Now close the file if it is open, either as a fd or a stream. */
301
302 if (data_file != NULL) (void)fclose(data_file);
303 else if (data_fd >= 0) (void)close(data_fd);
304
305 /* Attempt to close down an SMTP connection tidily. For non-batched SMTP, call
306 smtp_notquit_exit(), which runs the NOTQUIT ACL, if present, and handles the
307 SMTP response. */
308
309 if (smtp_input)
310 {
311 if (smtp_batched_input)
312 moan_smtp_batch(NULL, "421 %s - message abandoned", msg); /* No return */
313 smtp_notquit_exit(reason, US"421", US"%s %s - closing connection.",
314 smtp_active_hostname, msg);
315 }
316
317 /* Exit from the program (non-BSMTP cases) */
318
319 exim_exit(EXIT_FAILURE);
320 }
321
322
323 /*************************************************
324 * Data read timeout *
325 *************************************************/
326
327 /* Handler function for timeouts that occur while reading the data that
328 comprises a message.
329
330 Argument: the signal number
331 Returns: nothing
332 */
333
334 static void
335 data_timeout_handler(int sig)
336 {
337 uschar *msg = NULL;
338
339 sig = sig; /* Keep picky compilers happy */
340
341 if (smtp_input)
342 {
343 msg = US"SMTP incoming data timeout";
344 log_write(L_lost_incoming_connection,
345 LOG_MAIN, "SMTP data timeout (message abandoned) on connection "
346 "from %s F=<%s>",
347 (sender_fullhost != NULL)? sender_fullhost : US"local process",
348 sender_address);
349 }
350 else
351 {
352 fprintf(stderr, "exim: timed out while reading - message abandoned\n");
353 log_write(L_lost_incoming_connection,
354 LOG_MAIN, "timed out while reading local message");
355 }
356
357 receive_bomb_out(US"data-timeout", msg); /* Does not return */
358 }
359
360
361
362 /*************************************************
363 * local_scan() timeout *
364 *************************************************/
365
366 /* Handler function for timeouts that occur while running a local_scan()
367 function.
368
369 Argument: the signal number
370 Returns: nothing
371 */
372
373 static void
374 local_scan_timeout_handler(int sig)
375 {
376 sig = sig; /* Keep picky compilers happy */
377 log_write(0, LOG_MAIN|LOG_REJECT, "local_scan() function timed out - "
378 "message temporarily rejected (size %d)", message_size);
379 /* Does not return */
380 receive_bomb_out(US"local-scan-timeout", US"local verification problem");
381 }
382
383
384
385 /*************************************************
386 * local_scan() crashed *
387 *************************************************/
388
389 /* Handler function for signals that occur while running a local_scan()
390 function.
391
392 Argument: the signal number
393 Returns: nothing
394 */
395
396 static void
397 local_scan_crash_handler(int sig)
398 {
399 log_write(0, LOG_MAIN|LOG_REJECT, "local_scan() function crashed with "
400 "signal %d - message temporarily rejected (size %d)", sig, message_size);
401 /* Does not return */
402 receive_bomb_out(US"local-scan-error", US"local verification problem");
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 == FTEST_NONE)
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(US"signal-exit", 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 #ifdef EXPERIMENTAL_BRIGHTMAIL
474 recipients_list[recipients_count].bmi_optin = bmi_current_optin;
475 /* reset optin string pointer for next recipient */
476 bmi_current_optin = NULL;
477 #endif
478 recipients_list[recipients_count++].errors_to = NULL;
479 }
480
481
482
483
484 /*************************************************
485 * Remove a recipient from the list *
486 *************************************************/
487
488 /* This function is provided for local_scan() to use.
489
490 Argument:
491 recipient address to remove
492
493 Returns: TRUE if it did remove something; FALSE otherwise
494 */
495
496 BOOL
497 receive_remove_recipient(uschar *recipient)
498 {
499 int count;
500 DEBUG(D_receive) debug_printf("receive_remove_recipient(\"%s\") called\n",
501 recipient);
502 for (count = 0; count < recipients_count; count++)
503 {
504 if (Ustrcmp(recipients_list[count].address, recipient) == 0)
505 {
506 if ((--recipients_count - count) > 0)
507 memmove(recipients_list + count, recipients_list + count + 1,
508 (recipients_count - count)*sizeof(recipient_item));
509 return TRUE;
510 }
511 }
512 return FALSE;
513 }
514
515
516
517
518
519 /*************************************************
520 * Read data portion of a non-SMTP message *
521 *************************************************/
522
523 /* This function is called to read the remainder of a message (following the
524 header) when the input is not from SMTP - we are receiving a local message on
525 a standard input stream. The message is always terminated by EOF, and is also
526 terminated by a dot on a line by itself if the flag dot_ends is TRUE. Split the
527 two cases for maximum efficiency.
528
529 Ensure that the body ends with a newline. This will naturally be the case when
530 the termination is "\n.\n" but may not be otherwise. The RFC defines messages
531 as "sequences of lines" - this of course strictly applies only to SMTP, but
532 deliveries into BSD-type mailbox files also require it. Exim used to have a
533 flag for doing this at delivery time, but as it was always set for all
534 transports, I decided to simplify things by putting the check here instead.
535
536 There is at least one MUA (dtmail) that sends CRLF via this interface, and
537 other programs are known to do this as well. Exim used to have a option for
538 dealing with this: in July 2003, after much discussion, the code has been
539 changed to default to treat any of LF, CRLF, and bare CR as line terminators.
540
541 However, for the case when a dot on a line by itself terminates a message, the
542 only recognized terminating sequences before and after the dot are LF and CRLF.
543 Otherwise, having read EOL . CR, you don't know whether to read another
544 character or not.
545
546 Internally, in messages stored in Exim's spool files, LF is used as the line
547 terminator. Under the new regime, bare CRs will no longer appear in these
548 files.
549
550 Arguments:
551 fout a FILE to which to write the message
552
553 Returns: One of the END_xxx values indicating why it stopped reading
554 */
555
556 static int
557 read_message_data(FILE *fout)
558 {
559 int ch_state;
560 register int ch;
561 register int linelength = 0;
562
563 /* Handle the case when only EOF terminates the message */
564
565 if (!dot_ends)
566 {
567 register int last_ch = '\n';
568
569 for (; (ch = (receive_getc)()) != EOF; last_ch = ch)
570 {
571 if (ch == 0) body_zerocount++;
572 if (last_ch == '\r' && ch != '\n')
573 {
574 if (linelength > max_received_linelength)
575 max_received_linelength = linelength;
576 linelength = 0;
577 if (fputc('\n', fout) == EOF) return END_WERROR;
578 message_size++;
579 body_linecount++;
580 }
581 if (ch == '\r') continue;
582
583 if (fputc(ch, fout) == EOF) return END_WERROR;
584 if (ch == '\n')
585 {
586 if (linelength > max_received_linelength)
587 max_received_linelength = linelength;
588 linelength = 0;
589 body_linecount++;
590 }
591 else linelength++;
592 if (++message_size > thismessage_size_limit) return END_SIZE;
593 }
594
595 if (last_ch != '\n')
596 {
597 if (linelength > max_received_linelength)
598 max_received_linelength = linelength;
599 if (fputc('\n', fout) == EOF) return END_WERROR;
600 message_size++;
601 body_linecount++;
602 }
603
604 return END_EOF;
605 }
606
607 /* Handle the case when a dot on a line on its own, or EOF, terminates. */
608
609 ch_state = 1;
610
611 while ((ch = (receive_getc)()) != EOF)
612 {
613 if (ch == 0) body_zerocount++;
614 switch (ch_state)
615 {
616 case 0: /* Normal state (previous char written) */
617 if (ch == '\n')
618 {
619 body_linecount++;
620 if (linelength > max_received_linelength)
621 max_received_linelength = linelength;
622 linelength = -1;
623 ch_state = 1;
624 }
625 else if (ch == '\r')
626 { ch_state = 2; continue; }
627 break;
628
629 case 1: /* After written "\n" */
630 if (ch == '.') { ch_state = 3; continue; }
631 if (ch != '\n') ch_state = 0; else linelength = -1;
632 break;
633
634 case 2:
635 body_linecount++; /* After unwritten "\r" */
636 if (linelength > max_received_linelength)
637 max_received_linelength = linelength;
638 if (ch == '\n')
639 {
640 ch_state = 1;
641 linelength = -1;
642 }
643 else
644 {
645 if (message_size++, fputc('\n', fout) == EOF) return END_WERROR;
646 if (ch == '\r') continue;
647 ch_state = 0;
648 linelength = 0;
649 }
650 break;
651
652 case 3: /* After "\n." (\n written, dot not) */
653 if (ch == '\n') return END_DOT;
654 if (ch == '\r') { ch_state = 4; continue; }
655 message_size++;
656 linelength++;
657 if (fputc('.', fout) == EOF) return END_WERROR;
658 ch_state = 0;
659 break;
660
661 case 4: /* After "\n.\r" (\n written, rest not) */
662 if (ch == '\n') return END_DOT;
663 message_size += 2;
664 body_linecount++;
665 if (fputs(".\n", fout) == EOF) return END_WERROR;
666 if (ch == '\r') { ch_state = 2; continue; }
667 ch_state = 0;
668 break;
669 }
670
671 linelength++;
672 if (fputc(ch, fout) == EOF) return END_WERROR;
673 if (++message_size > thismessage_size_limit) return END_SIZE;
674 }
675
676 /* Get here if EOF read. Unless we have just written "\n", we need to ensure
677 the message ends with a newline, and we must also write any characters that
678 were saved up while testing for an ending dot. */
679
680 if (ch_state != 1)
681 {
682 static uschar *ends[] = { US"\n", NULL, US"\n", US".\n", US".\n" };
683 if (fputs(CS ends[ch_state], fout) == EOF) return END_WERROR;
684 message_size += Ustrlen(ends[ch_state]);
685 body_linecount++;
686 }
687
688 return END_EOF;
689 }
690
691
692
693
694 /*************************************************
695 * Read data portion of an SMTP message *
696 *************************************************/
697
698 /* This function is called to read the remainder of an SMTP message (after the
699 headers), or to skip over it when an error has occurred. In this case, the
700 output file is passed as NULL.
701
702 If any line begins with a dot, that character is skipped. The input should only
703 be successfully terminated by CR LF . CR LF unless it is local (non-network)
704 SMTP, in which case the CRs are optional, but...
705
706 FUDGE: It seems that sites on the net send out messages with just LF
707 terminators, despite the warnings in the RFCs, and other MTAs handle this. So
708 we make the CRs optional in all cases.
709
710 July 2003: Bare CRs cause trouble. We now treat them as line terminators as
711 well, so that there are no CRs in spooled messages. However, the message
712 terminating dot is not recognized between two bare CRs.
713
714 Arguments:
715 fout a FILE to which to write the message; NULL if skipping
716
717 Returns: One of the END_xxx values indicating why it stopped reading
718 */
719
720 static int
721 read_message_data_smtp(FILE *fout)
722 {
723 int ch_state = 0;
724 int ch;
725 register int linelength = 0;
726
727 while ((ch = (receive_getc)()) != EOF)
728 {
729 if (ch == 0) body_zerocount++;
730 switch (ch_state)
731 {
732 case 0: /* After LF or CRLF */
733 if (ch == '.')
734 {
735 ch_state = 3;
736 continue; /* Don't ever write . after LF */
737 }
738 ch_state = 1;
739
740 /* Else fall through to handle as normal uschar. */
741
742 case 1: /* Normal state */
743 if (ch == '\n')
744 {
745 ch_state = 0;
746 body_linecount++;
747 if (linelength > max_received_linelength)
748 max_received_linelength = linelength;
749 linelength = -1;
750 }
751 else if (ch == '\r')
752 {
753 ch_state = 2;
754 continue;
755 }
756 break;
757
758 case 2: /* After (unwritten) CR */
759 body_linecount++;
760 if (linelength > max_received_linelength)
761 max_received_linelength = linelength;
762 linelength = -1;
763 if (ch == '\n')
764 {
765 ch_state = 0;
766 }
767 else
768 {
769 message_size++;
770 if (fout != NULL && fputc('\n', fout) == EOF) return END_WERROR;
771 (void) cutthrough_put_nl();
772 if (ch != '\r') ch_state = 1; else continue;
773 }
774 break;
775
776 case 3: /* After [CR] LF . */
777 if (ch == '\n')
778 return END_DOT;
779 if (ch == '\r')
780 {
781 ch_state = 4;
782 continue;
783 }
784 ch_state = 1; /* The dot itself is removed */
785 break;
786
787 case 4: /* After [CR] LF . CR */
788 if (ch == '\n') return END_DOT;
789 message_size++;
790 body_linecount++;
791 if (fout != NULL && fputc('\n', fout) == EOF) return END_WERROR;
792 (void) cutthrough_put_nl();
793 if (ch == '\r')
794 {
795 ch_state = 2;
796 continue;
797 }
798 ch_state = 1;
799 break;
800 }
801
802 /* Add the character to the spool file, unless skipping; then loop for the
803 next. */
804
805 message_size++;
806 linelength++;
807 if (fout != NULL)
808 {
809 if (fputc(ch, fout) == EOF) return END_WERROR;
810 if (message_size > thismessage_size_limit) return END_SIZE;
811 }
812 if(ch == '\n')
813 (void) cutthrough_put_nl();
814 else
815 {
816 uschar c= ch;
817 (void) cutthrough_puts(&c, 1);
818 }
819 }
820
821 /* Fall through here if EOF encountered. This indicates some kind of error,
822 since a correct message is terminated by [CR] LF . [CR] LF. */
823
824 return END_EOF;
825 }
826
827
828
829
830 /*************************************************
831 * Swallow SMTP message *
832 *************************************************/
833
834 /* This function is called when there has been some kind of error while reading
835 an SMTP message, and the remaining data may need to be swallowed. It is global
836 because it is called from smtp_closedown() to shut down an incoming call
837 tidily.
838
839 Argument: a FILE from which to read the message
840 Returns: nothing
841 */
842
843 void
844 receive_swallow_smtp(void)
845 {
846 if (message_ended >= END_NOTENDED)
847 message_ended = read_message_data_smtp(NULL);
848 }
849
850
851
852 /*************************************************
853 * Handle lost SMTP connection *
854 *************************************************/
855
856 /* This function logs connection loss incidents and generates an appropriate
857 SMTP response.
858
859 Argument: additional data for the message
860 Returns: the SMTP response
861 */
862
863 static uschar *
864 handle_lost_connection(uschar *s)
865 {
866 log_write(L_lost_incoming_connection | L_smtp_connection, LOG_MAIN,
867 "%s lost while reading message data%s", smtp_get_connection_info(), s);
868 return US"421 Lost incoming connection";
869 }
870
871
872
873
874 /*************************************************
875 * Handle a non-smtp reception error *
876 *************************************************/
877
878 /* This function is called for various errors during the reception of non-SMTP
879 messages. It either sends a message to the sender of the problem message, or it
880 writes to the standard error stream.
881
882 Arguments:
883 errcode code for moan_to_sender(), identifying the error
884 text1 first message text, passed to moan_to_sender()
885 text2 second message text, used only for stderrr
886 error_rc code to pass to exim_exit if no problem
887 f FILE containing body of message (may be stdin)
888 hptr pointer to instore headers or NULL
889
890 Returns: calls exim_exit(), which does not return
891 */
892
893 static void
894 give_local_error(int errcode, uschar *text1, uschar *text2, int error_rc,
895 FILE *f, header_line *hptr)
896 {
897 if (error_handling == ERRORS_SENDER)
898 {
899 error_block eblock;
900 eblock.next = NULL;
901 eblock.text1 = text1;
902 if (!moan_to_sender(errcode, &eblock, hptr, f, FALSE))
903 error_rc = EXIT_FAILURE;
904 }
905 else fprintf(stderr, "exim: %s%s\n", text2, text1); /* Sic */
906 (void)fclose(f);
907 exim_exit(error_rc);
908 }
909
910
911
912 /*************************************************
913 * Add header lines set up by ACL *
914 *************************************************/
915
916 /* This function is called to add the header lines that were set up by
917 statements in an ACL to the list of headers in memory. It is done in two stages
918 like this, because when the ACL for RCPT is running, the other headers have not
919 yet been received. This function is called twice; once just before running the
920 DATA ACL, and once after. This is so that header lines added by MAIL or RCPT
921 are visible to the DATA ACL.
922
923 Originally these header lines were added at the end. Now there is support for
924 three different places: top, bottom, and after the Received: header(s). There
925 will always be at least one Received: header, even if it is marked deleted, and
926 even if something else has been put in front of it.
927
928 Arguments:
929 acl_name text to identify which ACL
930
931 Returns: nothing
932 */
933
934 static void
935 add_acl_headers(uschar *acl_name)
936 {
937 header_line *h, *next;
938 header_line *last_received = NULL;
939 int sep = ':';
940
941 if (acl_removed_headers != NULL)
942 {
943 DEBUG(D_receive|D_acl) debug_printf(">>Headers removed by %s ACL:\n", acl_name);
944
945 for (h = header_list; h != NULL; h = h->next)
946 {
947 int i;
948 uschar *list;
949 BOOL include_header;
950
951 if (h->type == htype_old) continue;
952
953 include_header = TRUE;
954 list = acl_removed_headers;
955
956 int sep = ':'; /* This is specified as a colon-separated list */
957 uschar *s;
958 uschar buffer[128];
959 while ((s = string_nextinlist(&list, &sep, buffer, sizeof(buffer)))
960 != NULL)
961 {
962 int len = Ustrlen(s);
963 if (header_testname(h, s, len, FALSE))
964 {
965 h->type = htype_old;
966 DEBUG(D_receive|D_acl) debug_printf(" %s", h->text);
967 }
968 }
969 }
970 acl_removed_headers = NULL;
971 DEBUG(D_receive|D_acl) debug_printf(">>\n");
972 }
973
974 if (acl_added_headers == NULL) return;
975 DEBUG(D_receive|D_acl) debug_printf(">>Headers added by %s ACL:\n", acl_name);
976
977 for (h = acl_added_headers; h != NULL; h = next)
978 {
979 next = h->next;
980
981 switch(h->type)
982 {
983 case htype_add_top:
984 h->next = header_list;
985 header_list = h;
986 DEBUG(D_receive|D_acl) debug_printf(" (at top)");
987 break;
988
989 case htype_add_rec:
990 if (last_received == NULL)
991 {
992 last_received = header_list;
993 while (!header_testname(last_received, US"Received", 8, FALSE))
994 last_received = last_received->next;
995 while (last_received->next != NULL &&
996 header_testname(last_received->next, US"Received", 8, FALSE))
997 last_received = last_received->next;
998 }
999 h->next = last_received->next;
1000 last_received->next = h;
1001 DEBUG(D_receive|D_acl) debug_printf(" (after Received:)");
1002 break;
1003
1004 case htype_add_rfc:
1005 /* add header before any header which is NOT Received: or Resent- */
1006 last_received = header_list;
1007 while ( (last_received->next != NULL) &&
1008 ( (header_testname(last_received->next, US"Received", 8, FALSE)) ||
1009 (header_testname_incomplete(last_received->next, US"Resent-", 7, FALSE)) ) )
1010 last_received = last_received->next;
1011 /* last_received now points to the last Received: or Resent-* header
1012 in an uninterrupted chain of those header types (seen from the beginning
1013 of all headers. Our current header must follow it. */
1014 h->next = last_received->next;
1015 last_received->next = h;
1016 DEBUG(D_receive|D_acl) debug_printf(" (before any non-Received: or Resent-*: header)");
1017 break;
1018
1019 default:
1020 h->next = NULL;
1021 header_last->next = h;
1022 break;
1023 }
1024
1025 if (h->next == NULL) header_last = h;
1026
1027 /* Check for one of the known header types (From:, To:, etc.) though in
1028 practice most added headers are going to be "other". Lower case
1029 identification letters are never stored with the header; they are used
1030 for existence tests when messages are received. So discard any lower case
1031 flag values. */
1032
1033 h->type = header_checkname(h, FALSE);
1034 if (h->type >= 'a') h->type = htype_other;
1035
1036 DEBUG(D_receive|D_acl) debug_printf(" %s", header_last->text);
1037 }
1038
1039 acl_added_headers = NULL;
1040 DEBUG(D_receive|D_acl) debug_printf(">>\n");
1041 }
1042
1043
1044
1045 /*************************************************
1046 * Add host information for log line *
1047 *************************************************/
1048
1049 /* Called for acceptance and rejecting log lines. This adds information about
1050 the calling host to a string that is being built dynamically.
1051
1052 Arguments:
1053 s the dynamic string
1054 sizeptr points to the size variable
1055 ptrptr points to the pointer variable
1056
1057 Returns: the extended string
1058 */
1059
1060 static uschar *
1061 add_host_info_for_log(uschar *s, int *sizeptr, int *ptrptr)
1062 {
1063 if (sender_fullhost != NULL)
1064 {
1065 s = string_append(s, sizeptr, ptrptr, 2, US" H=", sender_fullhost);
1066 if ((log_extra_selector & LX_incoming_interface) != 0 &&
1067 interface_address != NULL)
1068 {
1069 uschar *ss = string_sprintf(" I=[%s]:%d", interface_address,
1070 interface_port);
1071 s = string_cat(s, sizeptr, ptrptr, ss, Ustrlen(ss));
1072 }
1073 }
1074 if (sender_ident != NULL)
1075 s = string_append(s, sizeptr, ptrptr, 2, US" U=", sender_ident);
1076 if (received_protocol != NULL)
1077 s = string_append(s, sizeptr, ptrptr, 2, US" P=", received_protocol);
1078 return s;
1079 }
1080
1081
1082
1083 #ifdef WITH_CONTENT_SCAN
1084
1085 /*************************************************
1086 * Run the MIME ACL on a message *
1087 *************************************************/
1088
1089 /* This code is in a subroutine so that it can be used for both SMTP
1090 and non-SMTP messages. It is called with a non-NULL ACL pointer.
1091
1092 Arguments:
1093 acl The ACL to run (acl_smtp_mime or acl_not_smtp_mime)
1094 smtp_yield_ptr Set FALSE to kill messages after dropped connection
1095 smtp_reply_ptr Where SMTP reply is being built
1096 blackholed_by_ptr Where "blackholed by" message is being built
1097
1098 Returns: TRUE to carry on; FALSE to abandon the message
1099 */
1100
1101 static BOOL
1102 run_mime_acl(uschar *acl, BOOL *smtp_yield_ptr, uschar **smtp_reply_ptr,
1103 uschar **blackholed_by_ptr)
1104 {
1105 FILE *mbox_file;
1106 uschar rfc822_file_path[2048];
1107 unsigned long mbox_size;
1108 header_line *my_headerlist;
1109 uschar *user_msg, *log_msg;
1110 int mime_part_count_buffer = -1;
1111 int rc = OK;
1112
1113 memset(CS rfc822_file_path,0,2048);
1114
1115 /* check if it is a MIME message */
1116 my_headerlist = header_list;
1117 while (my_headerlist != NULL)
1118 {
1119 /* skip deleted headers */
1120 if (my_headerlist->type == '*')
1121 {
1122 my_headerlist = my_headerlist->next;
1123 continue;
1124 }
1125 if (strncmpic(my_headerlist->text, US"Content-Type:", 13) == 0)
1126 {
1127 DEBUG(D_receive) debug_printf("Found Content-Type: header - executing acl_smtp_mime.\n");
1128 goto DO_MIME_ACL;
1129 }
1130 my_headerlist = my_headerlist->next;
1131 }
1132
1133 DEBUG(D_receive) debug_printf("No Content-Type: header - presumably not a MIME message.\n");
1134 return TRUE;
1135
1136 DO_MIME_ACL:
1137 /* make sure the eml mbox file is spooled up */
1138 mbox_file = spool_mbox(&mbox_size, NULL);
1139 if (mbox_file == NULL) {
1140 /* error while spooling */
1141 log_write(0, LOG_MAIN|LOG_PANIC,
1142 "acl_smtp_mime: error while creating mbox spool file, message temporarily rejected.");
1143 Uunlink(spool_name);
1144 unspool_mbox();
1145 #ifdef EXPERIMENTAL_DCC
1146 dcc_ok = 0;
1147 #endif
1148 smtp_respond(US"451", 3, TRUE, US"temporary local problem");
1149 message_id[0] = 0; /* Indicate no message accepted */
1150 *smtp_reply_ptr = US""; /* Indicate reply already sent */
1151 return FALSE; /* Indicate skip to end of receive function */
1152 };
1153
1154 mime_is_rfc822 = 0;
1155
1156 MIME_ACL_CHECK:
1157 mime_part_count = -1;
1158 rc = mime_acl_check(acl, mbox_file, NULL, &user_msg, &log_msg);
1159 (void)fclose(mbox_file);
1160
1161 if (Ustrlen(rfc822_file_path) > 0)
1162 {
1163 mime_part_count = mime_part_count_buffer;
1164
1165 if (unlink(CS rfc822_file_path) == -1)
1166 {
1167 log_write(0, LOG_PANIC,
1168 "acl_smtp_mime: can't unlink RFC822 spool file, skipping.");
1169 goto END_MIME_ACL;
1170 }
1171 }
1172
1173 /* check if we must check any message/rfc822 attachments */
1174 if (rc == OK)
1175 {
1176 uschar temp_path[1024];
1177 int n;
1178 struct dirent *entry;
1179 DIR *tempdir;
1180
1181 (void)string_format(temp_path, 1024, "%s/scan/%s", spool_directory,
1182 message_id);
1183
1184 tempdir = opendir(CS temp_path);
1185 n = 0;
1186 do
1187 {
1188 entry = readdir(tempdir);
1189 if (entry == NULL) break;
1190 if (strncmpic(US entry->d_name,US"__rfc822_",9) == 0)
1191 {
1192 (void)string_format(rfc822_file_path, 2048,"%s/scan/%s/%s", spool_directory, message_id, entry->d_name);
1193 debug_printf("RFC822 attachment detected: running MIME ACL for '%s'\n", rfc822_file_path);
1194 break;
1195 }
1196 } while (1);
1197 closedir(tempdir);
1198
1199 if (entry != NULL)
1200 {
1201 mbox_file = Ufopen(rfc822_file_path,"rb");
1202 if (mbox_file == NULL)
1203 {
1204 log_write(0, LOG_PANIC,
1205 "acl_smtp_mime: can't open RFC822 spool file, skipping.");
1206 unlink(CS rfc822_file_path);
1207 goto END_MIME_ACL;
1208 }
1209 /* set RFC822 expansion variable */
1210 mime_is_rfc822 = 1;
1211 mime_part_count_buffer = mime_part_count;
1212 goto MIME_ACL_CHECK;
1213 }
1214 }
1215
1216 END_MIME_ACL:
1217 add_acl_headers(US"MIME");
1218 if (rc == DISCARD)
1219 {
1220 recipients_count = 0;
1221 *blackholed_by_ptr = US"MIME ACL";
1222 }
1223 else if (rc != OK)
1224 {
1225 Uunlink(spool_name);
1226 unspool_mbox();
1227 #ifdef EXPERIMENTAL_DCC
1228 dcc_ok = 0;
1229 #endif
1230 if (smtp_handle_acl_fail(ACL_WHERE_MIME, rc, user_msg, log_msg) != 0)
1231 *smtp_yield_ptr = FALSE; /* No more messsages after dropped connection */
1232 *smtp_reply_ptr = US""; /* Indicate reply already sent */
1233 message_id[0] = 0; /* Indicate no message accepted */
1234 return FALSE; /* Cause skip to end of receive function */
1235 }
1236
1237 return TRUE;
1238 }
1239
1240 #endif /* WITH_CONTENT_SCAN */
1241
1242
1243
1244 void
1245 received_header_gen(void)
1246 {
1247 uschar *received;
1248 uschar *timestamp;
1249 header_line *received_header= header_list;
1250
1251 timestamp = expand_string(US"${tod_full}");
1252 if (recipients_count == 1) received_for = recipients_list[0].address;
1253 received = expand_string(received_header_text);
1254 received_for = NULL;
1255
1256 if (received == NULL)
1257 {
1258 if(spool_name[0] != 0)
1259 Uunlink(spool_name); /* Lose the data file */
1260 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Expansion of \"%s\" "
1261 "(received_header_text) failed: %s", string_printing(received_header_text),
1262 expand_string_message);
1263 }
1264
1265 /* The first element on the header chain is reserved for the Received header,
1266 so all we have to do is fill in the text pointer, and set the type. However, if
1267 the result of the expansion is an empty string, we leave the header marked as
1268 "old" so as to refrain from adding a Received header. */
1269
1270 if (received[0] == 0)
1271 {
1272 received_header->text = string_sprintf("Received: ; %s\n", timestamp);
1273 received_header->type = htype_old;
1274 }
1275 else
1276 {
1277 received_header->text = string_sprintf("%s; %s\n", received, timestamp);
1278 received_header->type = htype_received;
1279 }
1280
1281 received_header->slen = Ustrlen(received_header->text);
1282
1283 DEBUG(D_receive) debug_printf(">>Generated Received: header line\n%c %s",
1284 received_header->type, received_header->text);
1285 }
1286
1287
1288
1289 /*************************************************
1290 * Receive message *
1291 *************************************************/
1292
1293 /* Receive a message on the given input, and put it into a pair of spool files.
1294 Either a non-null list of recipients, or the extract flag will be true, or
1295 both. The flag sender_local is true for locally generated messages. The flag
1296 submission_mode is true if an ACL has obeyed "control = submission". The flag
1297 suppress_local_fixups is true if an ACL has obeyed "control =
1298 suppress_local_fixups" or -G was passed on the command-line.
1299 The flag smtp_input is true if the message is to be
1300 handled using SMTP conventions about termination and lines starting with dots.
1301 For non-SMTP messages, dot_ends is true for dot-terminated messages.
1302
1303 If a message was successfully read, message_id[0] will be non-zero.
1304
1305 The general actions of this function are:
1306
1307 . Read the headers of the message (if any) into a chain of store
1308 blocks.
1309
1310 . If there is a "sender:" header and the message is locally originated,
1311 throw it away, unless the caller is trusted, or unless
1312 active_local_sender_retain is set - which can only happen if
1313 active_local_from_check is false.
1314
1315 . If recipients are to be extracted from the message, build the
1316 recipients list from the headers, removing any that were on the
1317 original recipients list (unless extract_addresses_remove_arguments is
1318 false), and at the same time, remove any bcc header that may be present.
1319
1320 . Get the spool file for the data, sort out its unique name, open
1321 and lock it (but don't give it the name yet).
1322
1323 . Generate a "Message-Id" header if the message doesn't have one, for
1324 locally-originated messages.
1325
1326 . Generate a "Received" header.
1327
1328 . Ensure the recipients list is fully qualified and rewritten if necessary.
1329
1330 . If there are any rewriting rules, apply them to the sender address
1331 and also to the headers.
1332
1333 . If there is no from: header, generate one, for locally-generated messages
1334 and messages in "submission mode" only.
1335
1336 . If the sender is local, check that from: is correct, and if not, generate
1337 a Sender: header, unless message comes from a trusted caller, or this
1338 feature is disabled by active_local_from_check being false.
1339
1340 . If there is no "date" header, generate one, for locally-originated
1341 or submission mode messages only.
1342
1343 . Copy the rest of the input, or up to a terminating "." if in SMTP or
1344 dot_ends mode, to the data file. Leave it open, to hold the lock.
1345
1346 . Write the envelope and the headers to a new file.
1347
1348 . Set the name for the header file; close it.
1349
1350 . Set the name for the data file; close it.
1351
1352 Because this function can potentially be called many times in a single
1353 SMTP connection, all store should be got by store_get(), so that it will be
1354 automatically retrieved after the message is accepted.
1355
1356 FUDGE: It seems that sites on the net send out messages with just LF
1357 terminators, despite the warnings in the RFCs, and other MTAs handle this. So
1358 we make the CRs optional in all cases.
1359
1360 July 2003: Bare CRs in messages, especially in header lines, cause trouble. A
1361 new regime is now in place in which bare CRs in header lines are turned into LF
1362 followed by a space, so as not to terminate the header line.
1363
1364 February 2004: A bare LF in a header line in a message whose first line was
1365 terminated by CRLF is treated in the same way as a bare CR.
1366
1367 Arguments:
1368 extract_recip TRUE if recipients are to be extracted from the message's
1369 headers
1370
1371 Returns: TRUE there are more messages to be read (SMTP input)
1372 FALSE there are no more messages to be read (non-SMTP input
1373 or SMTP connection collapsed, or other failure)
1374
1375 When reading a message for filter testing, the returned value indicates
1376 whether the headers (which is all that is read) were terminated by '.' or
1377 not. */
1378
1379 BOOL
1380 receive_msg(BOOL extract_recip)
1381 {
1382 int i;
1383 int rc = FAIL;
1384 int msg_size = 0;
1385 int process_info_len = Ustrlen(process_info);
1386 int error_rc = (error_handling == ERRORS_SENDER)?
1387 errors_sender_rc : EXIT_FAILURE;
1388 int header_size = 256;
1389 int start, end, domain, size, sptr;
1390 int id_resolution;
1391 int had_zero = 0;
1392 int prevlines_length = 0;
1393
1394 register int ptr = 0;
1395
1396 BOOL contains_resent_headers = FALSE;
1397 BOOL extracted_ignored = FALSE;
1398 BOOL first_line_ended_crlf = TRUE_UNSET;
1399 BOOL smtp_yield = TRUE;
1400 BOOL yield = FALSE;
1401
1402 BOOL resents_exist = FALSE;
1403 uschar *resent_prefix = US"";
1404 uschar *blackholed_by = NULL;
1405 uschar *blackhole_log_msg = US"";
1406 int cutthrough_done;
1407
1408 flock_t lock_data;
1409 error_block *bad_addresses = NULL;
1410
1411 uschar *frozen_by = NULL;
1412 uschar *queued_by = NULL;
1413
1414 uschar *errmsg, *s;
1415 struct stat statbuf;
1416
1417 /* Final message to give to SMTP caller, and messages from ACLs */
1418
1419 uschar *smtp_reply = NULL;
1420 uschar *user_msg, *log_msg;
1421
1422 /* Working header pointers */
1423
1424 header_line *h, *next;
1425
1426 /* Flags for noting the existence of certain headers (only one left) */
1427
1428 BOOL date_header_exists = FALSE;
1429
1430 /* Pointers to receive the addresses of headers whose contents we need. */
1431
1432 header_line *from_header = NULL;
1433 header_line *subject_header = NULL;
1434 header_line *msgid_header = NULL;
1435 header_line *received_header;
1436
1437 /* Variables for use when building the Received: header. */
1438
1439 uschar *timestamp;
1440 int tslen;
1441
1442 /* Release any open files that might have been cached while preparing to
1443 accept the message - e.g. by verifying addresses - because reading a message
1444 might take a fair bit of real time. */
1445
1446 search_tidyup();
1447
1448 /* Extracting the recipient list from an input file is incompatible with
1449 cutthrough delivery with the no-spool option. It shouldn't be possible
1450 to set up the combination, but just in case kill any ongoing connection. */
1451 if (extract_recip || !smtp_input)
1452 cancel_cutthrough_connection("not smtp input");
1453
1454 /* Initialize the chain of headers by setting up a place-holder for Received:
1455 header. Temporarily mark it as "old", i.e. not to be used. We keep header_last
1456 pointing to the end of the chain to make adding headers simple. */
1457
1458 received_header = header_list = header_last = store_get(sizeof(header_line));
1459 header_list->next = NULL;
1460 header_list->type = htype_old;
1461 header_list->text = NULL;
1462 header_list->slen = 0;
1463
1464 /* Control block for the next header to be read. */
1465
1466 next = store_get(sizeof(header_line));
1467 next->text = store_get(header_size);
1468
1469 /* Initialize message id to be null (indicating no message read), and the
1470 header names list to be the normal list. Indicate there is no data file open
1471 yet, initialize the size and warning count, and deal with no size limit. */
1472
1473 message_id[0] = 0;
1474 data_file = NULL;
1475 data_fd = -1;
1476 spool_name[0] = 0;
1477 message_size = 0;
1478 warning_count = 0;
1479 received_count = 1; /* For the one we will add */
1480
1481 if (thismessage_size_limit <= 0) thismessage_size_limit = INT_MAX;
1482
1483 /* While reading the message, the following counts are computed. */
1484
1485 message_linecount = body_linecount = body_zerocount =
1486 max_received_linelength = 0;
1487
1488 #ifndef DISABLE_DKIM
1489 /* Call into DKIM to set up the context. */
1490 if (smtp_input && !smtp_batched_input && !dkim_disable_verify) dkim_exim_verify_init();
1491 #endif
1492
1493 /* Remember the time of reception. Exim uses time+pid for uniqueness of message
1494 ids, and fractions of a second are required. See the comments that precede the
1495 message id creation below. */
1496
1497 (void)gettimeofday(&message_id_tv, NULL);
1498
1499 /* For other uses of the received time we can operate with granularity of one
1500 second, and for that we use the global variable received_time. This is for
1501 things like ultimate message timeouts. */
1502
1503 received_time = message_id_tv.tv_sec;
1504
1505 /* If SMTP input, set the special handler for timeouts. The alarm() calls
1506 happen in the smtp_getc() function when it refills its buffer. */
1507
1508 if (smtp_input) os_non_restarting_signal(SIGALRM, data_timeout_handler);
1509
1510 /* If not SMTP input, timeout happens only if configured, and we just set a
1511 single timeout for the whole message. */
1512
1513 else if (receive_timeout > 0)
1514 {
1515 os_non_restarting_signal(SIGALRM, data_timeout_handler);
1516 alarm(receive_timeout);
1517 }
1518
1519 /* SIGTERM and SIGINT are caught always. */
1520
1521 signal(SIGTERM, data_sigterm_sigint_handler);
1522 signal(SIGINT, data_sigterm_sigint_handler);
1523
1524 /* Header lines in messages are not supposed to be very long, though when
1525 unfolded, to: and cc: headers can take up a lot of store. We must also cope
1526 with the possibility of junk being thrown at us. Start by getting 256 bytes for
1527 storing the header, and extend this as necessary using string_cat().
1528
1529 To cope with total lunacies, impose an upper limit on the length of the header
1530 section of the message, as otherwise the store will fill up. We must also cope
1531 with the possibility of binary zeros in the data. Hence we cannot use fgets().
1532 Folded header lines are joined into one string, leaving the '\n' characters
1533 inside them, so that writing them out reproduces the input.
1534
1535 Loop for each character of each header; the next structure for chaining the
1536 header is set up already, with ptr the offset of the next character in
1537 next->text. */
1538
1539 for (;;)
1540 {
1541 int ch = (receive_getc)();
1542
1543 /* If we hit EOF on a SMTP connection, it's an error, since incoming
1544 SMTP must have a correct "." terminator. */
1545
1546 if (ch == EOF && smtp_input /* && !smtp_batched_input */)
1547 {
1548 smtp_reply = handle_lost_connection(US" (header)");
1549 smtp_yield = FALSE;
1550 goto TIDYUP; /* Skip to end of function */
1551 }
1552
1553 /* See if we are at the current header's size limit - there must be at least
1554 four bytes left. This allows for the new character plus a zero, plus two for
1555 extra insertions when we are playing games with dots and carriage returns. If
1556 we are at the limit, extend the text buffer. This could have been done
1557 automatically using string_cat() but because this is a tightish loop storing
1558 only one character at a time, we choose to do it inline. Normally
1559 store_extend() will be able to extend the block; only at the end of a big
1560 store block will a copy be needed. To handle the case of very long headers
1561 (and sometimes lunatic messages can have ones that are 100s of K long) we
1562 call store_release() for strings that have been copied - if the string is at
1563 the start of a block (and therefore the only thing in it, because we aren't
1564 doing any other gets), the block gets freed. We can only do this because we
1565 know there are no other calls to store_get() going on. */
1566
1567 if (ptr >= header_size - 4)
1568 {
1569 int oldsize = header_size;
1570 /* header_size += 256; */
1571 header_size *= 2;
1572 if (!store_extend(next->text, oldsize, header_size))
1573 {
1574 uschar *newtext = store_get(header_size);
1575 memcpy(newtext, next->text, ptr);
1576 store_release(next->text);
1577 next->text = newtext;
1578 }
1579 }
1580
1581 /* Cope with receiving a binary zero. There is dispute about whether
1582 these should be allowed in RFC 822 messages. The middle view is that they
1583 should not be allowed in headers, at least. Exim takes this attitude at
1584 the moment. We can't just stomp on them here, because we don't know that
1585 this line is a header yet. Set a flag to cause scanning later. */
1586
1587 if (ch == 0) had_zero++;
1588
1589 /* Test for termination. Lines in remote SMTP are terminated by CRLF, while
1590 those from data files use just LF. Treat LF in local SMTP input as a
1591 terminator too. Treat EOF as a line terminator always. */
1592
1593 if (ch == EOF) goto EOL;
1594
1595 /* FUDGE: There are sites out there that don't send CRs before their LFs, and
1596 other MTAs accept this. We are therefore forced into this "liberalisation"
1597 too, so we accept LF as a line terminator whatever the source of the message.
1598 However, if the first line of the message ended with a CRLF, we treat a bare
1599 LF specially by inserting a white space after it to ensure that the header
1600 line is not terminated. */
1601
1602 if (ch == '\n')
1603 {
1604 if (first_line_ended_crlf == TRUE_UNSET) first_line_ended_crlf = FALSE;
1605 else if (first_line_ended_crlf) receive_ungetc(' ');
1606 goto EOL;
1607 }
1608
1609 /* This is not the end of the line. If this is SMTP input and this is
1610 the first character in the line and it is a "." character, ignore it.
1611 This implements the dot-doubling rule, though header lines starting with
1612 dots aren't exactly common. They are legal in RFC 822, though. If the
1613 following is CRLF or LF, this is the line that that terminates the
1614 entire message. We set message_ended to indicate this has happened (to
1615 prevent further reading), and break out of the loop, having freed the
1616 empty header, and set next = NULL to indicate no data line. */
1617
1618 if (ptr == 0 && ch == '.' && (smtp_input || dot_ends))
1619 {
1620 ch = (receive_getc)();
1621 if (ch == '\r')
1622 {
1623 ch = (receive_getc)();
1624 if (ch != '\n')
1625 {
1626 receive_ungetc(ch);
1627 ch = '\r'; /* Revert to CR */
1628 }
1629 }
1630 if (ch == '\n')
1631 {
1632 message_ended = END_DOT;
1633 store_reset(next);
1634 next = NULL;
1635 break; /* End character-reading loop */
1636 }
1637
1638 /* For non-SMTP input, the dot at the start of the line was really a data
1639 character. What is now in ch is the following character. We guaranteed
1640 enough space for this above. */
1641
1642 if (!smtp_input)
1643 {
1644 next->text[ptr++] = '.';
1645 message_size++;
1646 }
1647 }
1648
1649 /* If CR is immediately followed by LF, end the line, ignoring the CR, and
1650 remember this case if this is the first line ending. */
1651
1652 if (ch == '\r')
1653 {
1654 ch = (receive_getc)();
1655 if (ch == '\n')
1656 {
1657 if (first_line_ended_crlf == TRUE_UNSET) first_line_ended_crlf = TRUE;
1658 goto EOL;
1659 }
1660
1661 /* Otherwise, put back the character after CR, and turn the bare CR
1662 into LF SP. */
1663
1664 ch = (receive_ungetc)(ch);
1665 next->text[ptr++] = '\n';
1666 message_size++;
1667 ch = ' ';
1668 }
1669
1670 /* We have a data character for the header line. */
1671
1672 next->text[ptr++] = ch; /* Add to buffer */
1673 message_size++; /* Total message size so far */
1674
1675 /* Handle failure due to a humungously long header section. The >= allows
1676 for the terminating \n. Add what we have so far onto the headers list so
1677 that it gets reflected in any error message, and back up the just-read
1678 character. */
1679
1680 if (message_size >= header_maxsize)
1681 {
1682 next->text[ptr] = 0;
1683 next->slen = ptr;
1684 next->type = htype_other;
1685 next->next = NULL;
1686 header_last->next = next;
1687 header_last = next;
1688
1689 log_write(0, LOG_MAIN, "ridiculously long message header received from "
1690 "%s (more than %d characters): message abandoned",
1691 sender_host_unknown? sender_ident : sender_fullhost, header_maxsize);
1692
1693 if (smtp_input)
1694 {
1695 smtp_reply = US"552 Message header is ridiculously long";
1696 receive_swallow_smtp();
1697 goto TIDYUP; /* Skip to end of function */
1698 }
1699
1700 else
1701 {
1702 give_local_error(ERRMESS_VLONGHEADER,
1703 string_sprintf("message header longer than %d characters received: "
1704 "message not accepted", header_maxsize), US"", error_rc, stdin,
1705 header_list->next);
1706 /* Does not return */
1707 }
1708 }
1709
1710 continue; /* With next input character */
1711
1712 /* End of header line reached */
1713
1714 EOL:
1715
1716 /* Keep track of lines for BSMTP errors and overall message_linecount. */
1717
1718 receive_linecount++;
1719 message_linecount++;
1720
1721 /* Keep track of maximum line length */
1722
1723 if (ptr - prevlines_length > max_received_linelength)
1724 max_received_linelength = ptr - prevlines_length;
1725 prevlines_length = ptr + 1;
1726
1727 /* Now put in the terminating newline. There is always space for
1728 at least two more characters. */
1729
1730 next->text[ptr++] = '\n';
1731 message_size++;
1732
1733 /* A blank line signals the end of the headers; release the unwanted
1734 space and set next to NULL to indicate this. */
1735
1736 if (ptr == 1)
1737 {
1738 store_reset(next);
1739 next = NULL;
1740 break;
1741 }
1742
1743 /* There is data in the line; see if the next input character is a
1744 whitespace character. If it is, we have a continuation of this header line.
1745 There is always space for at least one character at this point. */
1746
1747 if (ch != EOF)
1748 {
1749 int nextch = (receive_getc)();
1750 if (nextch == ' ' || nextch == '\t')
1751 {
1752 next->text[ptr++] = nextch;
1753 message_size++;
1754 continue; /* Iterate the loop */
1755 }
1756 else if (nextch != EOF) (receive_ungetc)(nextch); /* For next time */
1757 else ch = EOF; /* Cause main loop to exit at end */
1758 }
1759
1760 /* We have got to the real line end. Terminate the string and release store
1761 beyond it. If it turns out to be a real header, internal binary zeros will
1762 be squashed later. */
1763
1764 next->text[ptr] = 0;
1765 next->slen = ptr;
1766 store_reset(next->text + ptr + 1);
1767
1768 /* Check the running total size against the overall message size limit. We
1769 don't expect to fail here, but if the overall limit is set less than MESSAGE_
1770 MAXSIZE and a big header is sent, we want to catch it. Just stop reading
1771 headers - the code to read the body will then also hit the buffer. */
1772
1773 if (message_size > thismessage_size_limit) break;
1774
1775 /* A line that is not syntactically correct for a header also marks
1776 the end of the headers. In this case, we leave next containing the
1777 first data line. This might actually be several lines because of the
1778 continuation logic applied above, but that doesn't matter.
1779
1780 It turns out that smail, and presumably sendmail, accept leading lines
1781 of the form
1782
1783 From ph10 Fri Jan 5 12:35 GMT 1996
1784
1785 in messages. The "mail" command on Solaris 2 sends such lines. I cannot
1786 find any documentation of this, but for compatibility it had better be
1787 accepted. Exim restricts it to the case of non-smtp messages, and
1788 treats it as an alternative to the -f command line option. Thus it is
1789 ignored except for trusted users or filter testing. Otherwise it is taken
1790 as the sender address, unless -f was used (sendmail compatibility).
1791
1792 It further turns out that some UUCPs generate the From_line in a different
1793 format, e.g.
1794
1795 From ph10 Fri, 7 Jan 97 14:00:00 GMT
1796
1797 The regex for matching these things is now capable of recognizing both
1798 formats (including 2- and 4-digit years in the latter). In fact, the regex
1799 is now configurable, as is the expansion string to fish out the sender.
1800
1801 Even further on it has been discovered that some broken clients send
1802 these lines in SMTP messages. There is now an option to ignore them from
1803 specified hosts or networks. Sigh. */
1804
1805 if (header_last == header_list &&
1806 (!smtp_input
1807 ||
1808 (sender_host_address != NULL &&
1809 verify_check_host(&ignore_fromline_hosts) == OK)
1810 ||
1811 (sender_host_address == NULL && ignore_fromline_local)
1812 ) &&
1813 regex_match_and_setup(regex_From, next->text, 0, -1))
1814 {
1815 if (!sender_address_forced)
1816 {
1817 uschar *uucp_sender = expand_string(uucp_from_sender);
1818 if (uucp_sender == NULL)
1819 {
1820 log_write(0, LOG_MAIN|LOG_PANIC,
1821 "expansion of \"%s\" failed after matching "
1822 "\"From \" line: %s", uucp_from_sender, expand_string_message);
1823 }
1824 else
1825 {
1826 int start, end, domain;
1827 uschar *errmess;
1828 uschar *newsender = parse_extract_address(uucp_sender, &errmess,
1829 &start, &end, &domain, TRUE);
1830 if (newsender != NULL)
1831 {
1832 if (domain == 0 && newsender[0] != 0)
1833 newsender = rewrite_address_qualify(newsender, FALSE);
1834
1835 if (filter_test != FTEST_NONE || receive_check_set_sender(newsender))
1836 {
1837 sender_address = newsender;
1838
1839 if (trusted_caller || filter_test != FTEST_NONE)
1840 {
1841 authenticated_sender = NULL;
1842 originator_name = US"";
1843 sender_local = FALSE;
1844 }
1845
1846 if (filter_test != FTEST_NONE)
1847 printf("Sender taken from \"From \" line\n");
1848 }
1849 }
1850 }
1851 }
1852 }
1853
1854 /* Not a leading "From " line. Check to see if it is a valid header line.
1855 Header names may contain any non-control characters except space and colon,
1856 amazingly. */
1857
1858 else
1859 {
1860 uschar *p = next->text;
1861
1862 /* If not a valid header line, break from the header reading loop, leaving
1863 next != NULL, indicating that it holds the first line of the body. */
1864
1865 if (isspace(*p)) break;
1866 while (mac_isgraph(*p) && *p != ':') p++;
1867 while (isspace(*p)) p++;
1868 if (*p != ':')
1869 {
1870 body_zerocount = had_zero;
1871 break;
1872 }
1873
1874 /* We have a valid header line. If there were any binary zeroes in
1875 the line, stomp on them here. */
1876
1877 if (had_zero > 0)
1878 for (p = next->text; p < next->text + ptr; p++) if (*p == 0) *p = '?';
1879
1880 /* It is perfectly legal to have an empty continuation line
1881 at the end of a header, but it is confusing to humans
1882 looking at such messages, since it looks like a blank line.
1883 Reduce confusion by removing redundant white space at the
1884 end. We know that there is at least one printing character
1885 (the ':' tested for above) so there is no danger of running
1886 off the end. */
1887
1888 p = next->text + ptr - 2;
1889 for (;;)
1890 {
1891 while (*p == ' ' || *p == '\t') p--;
1892 if (*p != '\n') break;
1893 ptr = (p--) - next->text + 1;
1894 message_size -= next->slen - ptr;
1895 next->text[ptr] = 0;
1896 next->slen = ptr;
1897 }
1898
1899 /* Add the header to the chain */
1900
1901 next->type = htype_other;
1902 next->next = NULL;
1903 header_last->next = next;
1904 header_last = next;
1905
1906 /* Check the limit for individual line lengths. This comes after adding to
1907 the chain so that the failing line is reflected if a bounce is generated
1908 (for a local message). */
1909
1910 if (header_line_maxsize > 0 && next->slen > header_line_maxsize)
1911 {
1912 log_write(0, LOG_MAIN, "overlong message header line received from "
1913 "%s (more than %d characters): message abandoned",
1914 sender_host_unknown? sender_ident : sender_fullhost,
1915 header_line_maxsize);
1916
1917 if (smtp_input)
1918 {
1919 smtp_reply = US"552 A message header line is too long";
1920 receive_swallow_smtp();
1921 goto TIDYUP; /* Skip to end of function */
1922 }
1923
1924 else
1925 {
1926 give_local_error(ERRMESS_VLONGHDRLINE,
1927 string_sprintf("message header line longer than %d characters "
1928 "received: message not accepted", header_line_maxsize), US"",
1929 error_rc, stdin, header_list->next);
1930 /* Does not return */
1931 }
1932 }
1933
1934 /* Note if any resent- fields exist. */
1935
1936 if (!resents_exist && strncmpic(next->text, US"resent-", 7) == 0)
1937 {
1938 resents_exist = TRUE;
1939 resent_prefix = US"Resent-";
1940 }
1941 }
1942
1943 /* The line has been handled. If we have hit EOF, break out of the loop,
1944 indicating no pending data line. */
1945
1946 if (ch == EOF) { next = NULL; break; }
1947
1948 /* Set up for the next header */
1949
1950 header_size = 256;
1951 next = store_get(sizeof(header_line));
1952 next->text = store_get(header_size);
1953 ptr = 0;
1954 had_zero = 0;
1955 prevlines_length = 0;
1956 } /* Continue, starting to read the next header */
1957
1958 /* At this point, we have read all the headers into a data structure in main
1959 store. The first header is still the dummy placeholder for the Received: header
1960 we are going to generate a bit later on. If next != NULL, it contains the first
1961 data line - which terminated the headers before reaching a blank line (not the
1962 normal case). */
1963
1964 DEBUG(D_receive)
1965 {
1966 debug_printf(">>Headers received:\n");
1967 for (h = header_list->next; h != NULL; h = h->next)
1968 debug_printf("%s", h->text);
1969 debug_printf("\n");
1970 }
1971
1972 /* End of file on any SMTP connection is an error. If an incoming SMTP call
1973 is dropped immediately after valid headers, the next thing we will see is EOF.
1974 We must test for this specially, as further down the reading of the data is
1975 skipped if already at EOF. */
1976
1977 if (smtp_input && (receive_feof)())
1978 {
1979 smtp_reply = handle_lost_connection(US" (after header)");
1980 smtp_yield = FALSE;
1981 goto TIDYUP; /* Skip to end of function */
1982 }
1983
1984 /* If this is a filter test run and no headers were read, output a warning
1985 in case there is a mistake in the test message. */
1986
1987 if (filter_test != FTEST_NONE && header_list->next == NULL)
1988 printf("Warning: no message headers read\n");
1989
1990
1991 /* Scan the headers to identify them. Some are merely marked for later
1992 processing; some are dealt with here. */
1993
1994 for (h = header_list->next; h != NULL; h = h->next)
1995 {
1996 BOOL is_resent = strncmpic(h->text, US"resent-", 7) == 0;
1997 if (is_resent) contains_resent_headers = TRUE;
1998
1999 switch (header_checkname(h, is_resent))
2000 {
2001 case htype_bcc:
2002 h->type = htype_bcc; /* Both Bcc: and Resent-Bcc: */
2003 break;
2004
2005 case htype_cc:
2006 h->type = htype_cc; /* Both Cc: and Resent-Cc: */
2007 break;
2008
2009 /* Record whether a Date: or Resent-Date: header exists, as appropriate. */
2010
2011 case htype_date:
2012 if (!resents_exist || is_resent) date_header_exists = TRUE;
2013 break;
2014
2015 /* Same comments as about Return-Path: below. */
2016
2017 case htype_delivery_date:
2018 if (delivery_date_remove) h->type = htype_old;
2019 break;
2020
2021 /* Same comments as about Return-Path: below. */
2022
2023 case htype_envelope_to:
2024 if (envelope_to_remove) h->type = htype_old;
2025 break;
2026
2027 /* Mark all "From:" headers so they get rewritten. Save the one that is to
2028 be used for Sender: checking. For Sendmail compatibility, if the "From:"
2029 header consists of just the login id of the user who called Exim, rewrite
2030 it with the gecos field first. Apply this rule to Resent-From: if there
2031 are resent- fields. */
2032
2033 case htype_from:
2034 h->type = htype_from;
2035 if (!resents_exist || is_resent)
2036 {
2037 from_header = h;
2038 if (!smtp_input)
2039 {
2040 uschar *s = Ustrchr(h->text, ':') + 1;
2041 while (isspace(*s)) s++;
2042 if (strncmpic(s, originator_login, h->slen - (s - h->text) - 1) == 0)
2043 {
2044 uschar *name = is_resent? US"Resent-From" : US"From";
2045 header_add(htype_from, "%s: %s <%s@%s>\n", name, originator_name,
2046 originator_login, qualify_domain_sender);
2047 from_header = header_last;
2048 h->type = htype_old;
2049 DEBUG(D_receive|D_rewrite)
2050 debug_printf("rewrote \"%s:\" header using gecos\n", name);
2051 }
2052 }
2053 }
2054 break;
2055
2056 /* Identify the Message-id: header for generating "in-reply-to" in the
2057 autoreply transport. For incoming logging, save any resent- value. In both
2058 cases, take just the first of any multiples. */
2059
2060 case htype_id:
2061 if (msgid_header == NULL && (!resents_exist || is_resent))
2062 {
2063 msgid_header = h;
2064 h->type = htype_id;
2065 }
2066 break;
2067
2068 /* Flag all Received: headers */
2069
2070 case htype_received:
2071 h->type = htype_received;
2072 received_count++;
2073 break;
2074
2075 /* "Reply-to:" is just noted (there is no resent-reply-to field) */
2076
2077 case htype_reply_to:
2078 h->type = htype_reply_to;
2079 break;
2080
2081 /* The Return-path: header is supposed to be added to messages when
2082 they leave the SMTP system. We shouldn't receive messages that already
2083 contain Return-path. However, since Exim generates Return-path: on
2084 local delivery, resent messages may well contain it. We therefore
2085 provide an option (which defaults on) to remove any Return-path: headers
2086 on input. Removal actually means flagging as "old", which prevents the
2087 header being transmitted with the message. */
2088
2089 case htype_return_path:
2090 if (return_path_remove) h->type = htype_old;
2091
2092 /* If we are testing a mail filter file, use the value of the
2093 Return-Path: header to set up the return_path variable, which is not
2094 otherwise set. However, remove any <> that surround the address
2095 because the variable doesn't have these. */
2096
2097 if (filter_test != FTEST_NONE)
2098 {
2099 uschar *start = h->text + 12;
2100 uschar *end = start + Ustrlen(start);
2101 while (isspace(*start)) start++;
2102 while (end > start && isspace(end[-1])) end--;
2103 if (*start == '<' && end[-1] == '>')
2104 {
2105 start++;
2106 end--;
2107 }
2108 return_path = string_copyn(start, end - start);
2109 printf("Return-path taken from \"Return-path:\" header line\n");
2110 }
2111 break;
2112
2113 /* If there is a "Sender:" header and the message is locally originated,
2114 and from an untrusted caller and suppress_local_fixups is not set, or if we
2115 are in submission mode for a remote message, mark it "old" so that it will
2116 not be transmitted with the message, unless active_local_sender_retain is
2117 set. (This can only be true if active_local_from_check is false.) If there
2118 are any resent- headers in the message, apply this rule to Resent-Sender:
2119 instead of Sender:. Messages with multiple resent- header sets cannot be
2120 tidily handled. (For this reason, at least one MUA - Pine - turns old
2121 resent- headers into X-resent- headers when resending, leaving just one
2122 set.) */
2123
2124 case htype_sender:
2125 h->type = ((!active_local_sender_retain &&
2126 (
2127 (sender_local && !trusted_caller && !suppress_local_fixups)
2128 || submission_mode
2129 )
2130 ) &&
2131 (!resents_exist||is_resent))?
2132 htype_old : htype_sender;
2133 break;
2134
2135 /* Remember the Subject: header for logging. There is no Resent-Subject */
2136
2137 case htype_subject:
2138 subject_header = h;
2139 break;
2140
2141 /* "To:" gets flagged, and the existence of a recipient header is noted,
2142 whether it's resent- or not. */
2143
2144 case htype_to:
2145 h->type = htype_to;
2146 /****
2147 to_or_cc_header_exists = TRUE;
2148 ****/
2149 break;
2150 }
2151 }
2152
2153 /* Extract recipients from the headers if that is required (the -t option).
2154 Note that this is documented as being done *before* any address rewriting takes
2155 place. There are two possibilities:
2156
2157 (1) According to sendmail documentation for Solaris, IRIX, and HP-UX, any
2158 recipients already listed are to be REMOVED from the message. Smail 3 works
2159 like this. We need to build a non-recipients tree for that list, because in
2160 subsequent processing this data is held in a tree and that's what the
2161 spool_write_header() function expects. Make sure that non-recipient addresses
2162 are fully qualified and rewritten if necessary.
2163
2164 (2) According to other sendmail documentation, -t ADDS extracted recipients to
2165 those in the command line arguments (and it is rumoured some other MTAs do
2166 this). Therefore, there is an option to make Exim behave this way.
2167
2168 *** Notes on "Resent-" header lines ***
2169
2170 The presence of resent-headers in the message makes -t horribly ambiguous.
2171 Experiments with sendmail showed that it uses recipients for all resent-
2172 headers, totally ignoring the concept of "sets of resent- headers" as described
2173 in RFC 2822 section 3.6.6. Sendmail also amalgamates them into a single set
2174 with all the addresses in one instance of each header.
2175
2176 This seems to me not to be at all sensible. Before release 4.20, Exim 4 gave an
2177 error for -t if there were resent- headers in the message. However, after a
2178 discussion on the mailing list, I've learned that there are MUAs that use
2179 resent- headers with -t, and also that the stuff about sets of resent- headers
2180 and their ordering in RFC 2822 is generally ignored. An MUA that submits a
2181 message with -t and resent- header lines makes sure that only *its* resent-
2182 headers are present; previous ones are often renamed as X-resent- for example.
2183
2184 Consequently, Exim has been changed so that, if any resent- header lines are
2185 present, the recipients are taken from all of the appropriate resent- lines,
2186 and not from the ordinary To:, Cc:, etc. */
2187
2188 if (extract_recip)
2189 {
2190 int rcount = 0;
2191 error_block **bnext = &bad_addresses;
2192
2193 if (extract_addresses_remove_arguments)
2194 {
2195 while (recipients_count-- > 0)
2196 {
2197 uschar *s = rewrite_address(recipients_list[recipients_count].address,
2198 TRUE, TRUE, global_rewrite_rules, rewrite_existflags);
2199 tree_add_nonrecipient(s);
2200 }
2201 recipients_list = NULL;
2202 recipients_count = recipients_list_max = 0;
2203 }
2204
2205 /* Now scan the headers */
2206
2207 for (h = header_list->next; h != NULL; h = h->next)
2208 {
2209 if ((h->type == htype_to || h->type == htype_cc || h->type == htype_bcc) &&
2210 (!contains_resent_headers || strncmpic(h->text, US"resent-", 7) == 0))
2211 {
2212 uschar *s = Ustrchr(h->text, ':') + 1;
2213 while (isspace(*s)) s++;
2214
2215 parse_allow_group = TRUE; /* Allow address group syntax */
2216
2217 while (*s != 0)
2218 {
2219 uschar *ss = parse_find_address_end(s, FALSE);
2220 uschar *recipient, *errmess, *p, *pp;
2221 int start, end, domain;
2222
2223 /* Check on maximum */
2224
2225 if (recipients_max > 0 && ++rcount > recipients_max)
2226 {
2227 give_local_error(ERRMESS_TOOMANYRECIP, US"too many recipients",
2228 US"message rejected: ", error_rc, stdin, NULL);
2229 /* Does not return */
2230 }
2231
2232 /* Make a copy of the address, and remove any internal newlines. These
2233 may be present as a result of continuations of the header line. The
2234 white space that follows the newline must not be removed - it is part
2235 of the header. */
2236
2237 pp = recipient = store_get(ss - s + 1);
2238 for (p = s; p < ss; p++) if (*p != '\n') *pp++ = *p;
2239 *pp = 0;
2240 recipient = parse_extract_address(recipient, &errmess, &start, &end,
2241 &domain, FALSE);
2242
2243 /* Keep a list of all the bad addresses so we can send a single
2244 error message at the end. However, an empty address is not an error;
2245 just ignore it. This can come from an empty group list like
2246
2247 To: Recipients of list:;
2248
2249 If there are no recipients at all, an error will occur later. */
2250
2251 if (recipient == NULL && Ustrcmp(errmess, "empty address") != 0)
2252 {
2253 int len = Ustrlen(s);
2254 error_block *b = store_get(sizeof(error_block));
2255 while (len > 0 && isspace(s[len-1])) len--;
2256 b->next = NULL;
2257 b->text1 = string_printing(string_copyn(s, len));
2258 b->text2 = errmess;
2259 *bnext = b;
2260 bnext = &(b->next);
2261 }
2262
2263 /* If the recipient is already in the nonrecipients tree, it must
2264 have appeared on the command line with the option extract_addresses_
2265 remove_arguments set. Do not add it to the recipients, and keep a note
2266 that this has happened, in order to give a better error if there are
2267 no recipients left. */
2268
2269 else if (recipient != NULL)
2270 {
2271 if (tree_search(tree_nonrecipients, recipient) == NULL)
2272 receive_add_recipient(recipient, -1);
2273 else
2274 extracted_ignored = TRUE;
2275 }
2276
2277 /* Move on past this address */
2278
2279 s = ss + (*ss? 1:0);
2280 while (isspace(*s)) s++;
2281 } /* Next address */
2282
2283 parse_allow_group = FALSE; /* Reset group syntax flags */
2284 parse_found_group = FALSE;
2285
2286 /* If this was the bcc: header, mark it "old", which means it
2287 will be kept on the spool, but not transmitted as part of the
2288 message. */
2289
2290 if (h->type == htype_bcc) h->type = htype_old;
2291 } /* For appropriate header line */
2292 } /* For each header line */
2293
2294 }
2295
2296 /* Now build the unique message id. This has changed several times over the
2297 lifetime of Exim. This description was rewritten for Exim 4.14 (February 2003).
2298 Retaining all the history in the comment has become too unwieldy - read
2299 previous release sources if you want it.
2300
2301 The message ID has 3 parts: tttttt-pppppp-ss. Each part is a number in base 62.
2302 The first part is the current time, in seconds. The second part is the current
2303 pid. Both are large enough to hold 32-bit numbers in base 62. The third part
2304 can hold a number in the range 0-3843. It used to be a computed sequence
2305 number, but is now the fractional component of the current time in units of
2306 1/2000 of a second (i.e. a value in the range 0-1999). After a message has been
2307 received, Exim ensures that the timer has ticked at the appropriate level
2308 before proceeding, to avoid duplication if the pid happened to be re-used
2309 within the same time period. It seems likely that most messages will take at
2310 least half a millisecond to be received, so no delay will normally be
2311 necessary. At least for some time...
2312
2313 There is a modification when localhost_number is set. Formerly this was allowed
2314 to be as large as 255. Now it is restricted to the range 0-16, and the final
2315 component of the message id becomes (localhost_number * 200) + fractional time
2316 in units of 1/200 of a second (i.e. a value in the range 0-3399).
2317
2318 Some not-really-Unix operating systems use case-insensitive file names (Darwin,
2319 Cygwin). For these, we have to use base 36 instead of base 62. Luckily, this
2320 still allows the tttttt field to hold a large enough number to last for some
2321 more decades, and the final two-digit field can hold numbers up to 1295, which
2322 is enough for milliseconds (instead of 1/2000 of a second).
2323
2324 However, the pppppp field cannot hold a 32-bit pid, but it can hold a 31-bit
2325 pid, so it is probably safe because pids have to be positive. The
2326 localhost_number is restricted to 0-10 for these hosts, and when it is set, the
2327 final field becomes (localhost_number * 100) + fractional time in centiseconds.
2328
2329 Note that string_base62() returns its data in a static storage block, so it
2330 must be copied before calling string_base62() again. It always returns exactly
2331 6 characters.
2332
2333 There doesn't seem to be anything in the RFC which requires a message id to
2334 start with a letter, but Smail was changed to ensure this. The external form of
2335 the message id (as supplied by string expansion) therefore starts with an
2336 additional leading 'E'. The spool file names do not include this leading
2337 letter and it is not used internally.
2338
2339 NOTE: If ever the format of message ids is changed, the regular expression for
2340 checking that a string is in this format must be updated in a corresponding
2341 way. It appears in the initializing code in exim.c. The macro MESSAGE_ID_LENGTH
2342 must also be changed to reflect the correct string length. Then, of course,
2343 other programs that rely on the message id format will need updating too. */
2344
2345 Ustrncpy(message_id, string_base62((long int)(message_id_tv.tv_sec)), 6);
2346 message_id[6] = '-';
2347 Ustrncpy(message_id + 7, string_base62((long int)getpid()), 6);
2348
2349 /* Deal with the case where the host number is set. The value of the number was
2350 checked when it was read, to ensure it isn't too big. The timing granularity is
2351 left in id_resolution so that an appropriate wait can be done after receiving
2352 the message, if necessary (we hope it won't be). */
2353
2354 if (host_number_string != NULL)
2355 {
2356 id_resolution = (BASE_62 == 62)? 5000 : 10000;
2357 sprintf(CS(message_id + MESSAGE_ID_LENGTH - 3), "-%2s",
2358 string_base62((long int)(
2359 host_number * (1000000/id_resolution) +
2360 message_id_tv.tv_usec/id_resolution)) + 4);
2361 }
2362
2363 /* Host number not set: final field is just the fractional time at an
2364 appropriate resolution. */
2365
2366 else
2367 {
2368 id_resolution = (BASE_62 == 62)? 500 : 1000;
2369 sprintf(CS(message_id + MESSAGE_ID_LENGTH - 3), "-%2s",
2370 string_base62((long int)(message_id_tv.tv_usec/id_resolution)) + 4);
2371 }
2372
2373 /* Add the current message id onto the current process info string if
2374 it will fit. */
2375
2376 (void)string_format(process_info + process_info_len,
2377 PROCESS_INFO_SIZE - process_info_len, " id=%s", message_id);
2378
2379 /* If we are using multiple input directories, set up the one for this message
2380 to be the least significant base-62 digit of the time of arrival. Otherwise
2381 ensure that it is an empty string. */
2382
2383 message_subdir[0] = split_spool_directory? message_id[5] : 0;
2384
2385 /* Now that we have the message-id, if there is no message-id: header, generate
2386 one, but only for local (without suppress_local_fixups) or submission mode
2387 messages. This can be user-configured if required, but we had better flatten
2388 any illegal characters therein. */
2389
2390 if (msgid_header == NULL &&
2391 ((sender_host_address == NULL && !suppress_local_fixups)
2392 || submission_mode))
2393 {
2394 uschar *p;
2395 uschar *id_text = US"";
2396 uschar *id_domain = primary_hostname;
2397
2398 /* Permit only letters, digits, dots, and hyphens in the domain */
2399
2400 if (message_id_domain != NULL)
2401 {
2402 uschar *new_id_domain = expand_string(message_id_domain);
2403 if (new_id_domain == NULL)
2404 {
2405 if (!expand_string_forcedfail)
2406 log_write(0, LOG_MAIN|LOG_PANIC,
2407 "expansion of \"%s\" (message_id_header_domain) "
2408 "failed: %s", message_id_domain, expand_string_message);
2409 }
2410 else if (*new_id_domain != 0)
2411 {
2412 id_domain = new_id_domain;
2413 for (p = id_domain; *p != 0; p++)
2414 if (!isalnum(*p) && *p != '.') *p = '-'; /* No need to test '-' ! */
2415 }
2416 }
2417
2418 /* Permit all characters except controls and RFC 2822 specials in the
2419 additional text part. */
2420
2421 if (message_id_text != NULL)
2422 {
2423 uschar *new_id_text = expand_string(message_id_text);
2424 if (new_id_text == NULL)
2425 {
2426 if (!expand_string_forcedfail)
2427 log_write(0, LOG_MAIN|LOG_PANIC,
2428 "expansion of \"%s\" (message_id_header_text) "
2429 "failed: %s", message_id_text, expand_string_message);
2430 }
2431 else if (*new_id_text != 0)
2432 {
2433 id_text = new_id_text;
2434 for (p = id_text; *p != 0; p++)
2435 if (mac_iscntrl_or_special(*p)) *p = '-';
2436 }
2437 }
2438
2439 /* Add the header line
2440 * Resent-* headers are prepended, per RFC 5322 3.6.6. Non-Resent-* are
2441 * appended, to preserve classical expectations of header ordering. */
2442
2443 header_add_at_position(!resents_exist, NULL, FALSE, htype_id,
2444 "%sMessage-Id: <%s%s%s@%s>\n", resent_prefix, message_id_external,
2445 (*id_text == 0)? "" : ".", id_text, id_domain);
2446 }
2447
2448 /* If we are to log recipients, keep a copy of the raw ones before any possible
2449 rewriting. Must copy the count, because later ACLs and the local_scan()
2450 function may mess with the real recipients. */
2451
2452 if ((log_extra_selector & LX_received_recipients) != 0)
2453 {
2454 raw_recipients = store_get(recipients_count * sizeof(uschar *));
2455 for (i = 0; i < recipients_count; i++)
2456 raw_recipients[i] = string_copy(recipients_list[i].address);
2457 raw_recipients_count = recipients_count;
2458 }
2459
2460 /* Ensure the recipients list is fully qualified and rewritten. Unqualified
2461 recipients will get here only if the conditions were right (allow_unqualified_
2462 recipient is TRUE). */
2463
2464 for (i = 0; i < recipients_count; i++)
2465 recipients_list[i].address =
2466 rewrite_address(recipients_list[i].address, TRUE, TRUE,
2467 global_rewrite_rules, rewrite_existflags);
2468
2469 /* If there is no From: header, generate one for local (without
2470 suppress_local_fixups) or submission_mode messages. If there is no sender
2471 address, but the sender is local or this is a local delivery error, use the
2472 originator login. This shouldn't happen for genuine bounces, but might happen
2473 for autoreplies. The addition of From: must be done *before* checking for the
2474 possible addition of a Sender: header, because untrusted_set_sender allows an
2475 untrusted user to set anything in the envelope (which might then get info
2476 From:) but we still want to ensure a valid Sender: if it is required. */
2477
2478 if (from_header == NULL &&
2479 ((sender_host_address == NULL && !suppress_local_fixups)
2480 || submission_mode))
2481 {
2482 uschar *oname = US"";
2483
2484 /* Use the originator_name if this is a locally submitted message and the
2485 caller is not trusted. For trusted callers, use it only if -F was used to
2486 force its value or if we have a non-SMTP message for which -f was not used
2487 to set the sender. */
2488
2489 if (sender_host_address == NULL)
2490 {
2491 if (!trusted_caller || sender_name_forced ||
2492 (!smtp_input && !sender_address_forced))
2493 oname = originator_name;
2494 }
2495
2496 /* For non-locally submitted messages, the only time we use the originator
2497 name is when it was forced by the /name= option on control=submission. */
2498
2499 else
2500 {
2501 if (submission_name != NULL) oname = submission_name;
2502 }
2503
2504 /* Envelope sender is empty */
2505
2506 if (sender_address[0] == 0)
2507 {
2508 uschar *fromstart, *fromend;
2509
2510 fromstart = string_sprintf("%sFrom: %s%s", resent_prefix,
2511 oname, (oname[0] == 0)? "" : " <");
2512 fromend = (oname[0] == 0)? US"" : US">";
2513
2514 if (sender_local || local_error_message)
2515 {
2516 header_add(htype_from, "%s%s@%s%s\n", fromstart,
2517 local_part_quote(originator_login), qualify_domain_sender,
2518 fromend);
2519 }
2520 else if (submission_mode && authenticated_id != NULL)
2521 {
2522 if (submission_domain == NULL)
2523 {
2524 header_add(htype_from, "%s%s@%s%s\n", fromstart,
2525 local_part_quote(authenticated_id), qualify_domain_sender,
2526 fromend);
2527 }
2528 else if (submission_domain[0] == 0) /* empty => whole address set */
2529 {
2530 header_add(htype_from, "%s%s%s\n", fromstart, authenticated_id,
2531 fromend);
2532 }
2533 else
2534 {
2535 header_add(htype_from, "%s%s@%s%s\n", fromstart,
2536 local_part_quote(authenticated_id), submission_domain,
2537 fromend);
2538 }
2539 from_header = header_last; /* To get it checked for Sender: */
2540 }
2541 }
2542
2543 /* There is a non-null envelope sender. Build the header using the original
2544 sender address, before any rewriting that might have been done while
2545 verifying it. */
2546
2547 else
2548 {
2549 header_add(htype_from, "%sFrom: %s%s%s%s\n", resent_prefix,
2550 oname,
2551 (oname[0] == 0)? "" : " <",
2552 (sender_address_unrewritten == NULL)?
2553 sender_address : sender_address_unrewritten,
2554 (oname[0] == 0)? "" : ">");
2555
2556 from_header = header_last; /* To get it checked for Sender: */
2557 }
2558 }
2559
2560
2561 /* If the sender is local (without suppress_local_fixups), or if we are in
2562 submission mode and there is an authenticated_id, check that an existing From:
2563 is correct, and if not, generate a Sender: header, unless disabled. Any
2564 previously-existing Sender: header was removed above. Note that sender_local,
2565 as well as being TRUE if the caller of exim is not trusted, is also true if a
2566 trusted caller did not supply a -f argument for non-smtp input. To allow
2567 trusted callers to forge From: without supplying -f, we have to test explicitly
2568 here. If the From: header contains more than one address, then the call to
2569 parse_extract_address fails, and a Sender: header is inserted, as required. */
2570
2571 if (from_header != NULL &&
2572 (active_local_from_check &&
2573 ((sender_local && !trusted_caller && !suppress_local_fixups) ||
2574 (submission_mode && authenticated_id != NULL))
2575 ))
2576 {
2577 BOOL make_sender = TRUE;
2578 int start, end, domain;
2579 uschar *errmess;
2580 uschar *from_address =
2581 parse_extract_address(Ustrchr(from_header->text, ':') + 1, &errmess,
2582 &start, &end, &domain, FALSE);
2583 uschar *generated_sender_address;
2584
2585 if (submission_mode)
2586 {
2587 if (submission_domain == NULL)
2588 {
2589 generated_sender_address = string_sprintf("%s@%s",
2590 local_part_quote(authenticated_id), qualify_domain_sender);
2591 }
2592 else if (submission_domain[0] == 0) /* empty => full address */
2593 {
2594 generated_sender_address = string_sprintf("%s",
2595 authenticated_id);
2596 }
2597 else
2598 {
2599 generated_sender_address = string_sprintf("%s@%s",
2600 local_part_quote(authenticated_id), submission_domain);
2601 }
2602 }
2603 else
2604 generated_sender_address = string_sprintf("%s@%s",
2605 local_part_quote(originator_login), qualify_domain_sender);
2606
2607 /* Remove permitted prefixes and suffixes from the local part of the From:
2608 address before doing the comparison with the generated sender. */
2609
2610 if (from_address != NULL)
2611 {
2612 int slen;
2613 uschar *at = (domain == 0)? NULL : from_address + domain - 1;
2614
2615 if (at != NULL) *at = 0;
2616 from_address += route_check_prefix(from_address, local_from_prefix);
2617 slen = route_check_suffix(from_address, local_from_suffix);
2618 if (slen > 0)
2619 {
2620 memmove(from_address+slen, from_address, Ustrlen(from_address)-slen);
2621 from_address += slen;
2622 }
2623 if (at != NULL) *at = '@';
2624
2625 if (strcmpic(generated_sender_address, from_address) == 0 ||
2626 (domain == 0 && strcmpic(from_address, originator_login) == 0))
2627 make_sender = FALSE;
2628 }
2629
2630 /* We have to cause the Sender header to be rewritten if there are
2631 appropriate rewriting rules. */
2632
2633 if (make_sender)
2634 {
2635 if (submission_mode && submission_name == NULL)
2636 header_add(htype_sender, "%sSender: %s\n", resent_prefix,
2637 generated_sender_address);
2638 else
2639 header_add(htype_sender, "%sSender: %s <%s>\n",
2640 resent_prefix,
2641 submission_mode? submission_name : originator_name,
2642 generated_sender_address);
2643 }
2644
2645 /* Ensure that a non-null envelope sender address corresponds to the
2646 submission mode sender address. */
2647
2648 if (submission_mode && sender_address[0] != 0)
2649 {
2650 if (sender_address_unrewritten == NULL)
2651 sender_address_unrewritten = sender_address;
2652 sender_address = generated_sender_address;
2653 if (Ustrcmp(sender_address_unrewritten, generated_sender_address) != 0)
2654 log_write(L_address_rewrite, LOG_MAIN,
2655 "\"%s\" from env-from rewritten as \"%s\" by submission mode",
2656 sender_address_unrewritten, generated_sender_address);
2657 }
2658 }
2659
2660
2661 /* If there are any rewriting rules, apply them to the sender address, unless
2662 it has already been rewritten as part of verification for SMTP input. */
2663
2664 if (global_rewrite_rules != NULL && sender_address_unrewritten == NULL &&
2665 sender_address[0] != 0)
2666 {
2667 sender_address = rewrite_address(sender_address, FALSE, TRUE,
2668 global_rewrite_rules, rewrite_existflags);
2669 DEBUG(D_receive|D_rewrite)
2670 debug_printf("rewritten sender = %s\n", sender_address);
2671 }
2672
2673
2674 /* The headers must be run through rewrite_header(), because it ensures that
2675 addresses are fully qualified, as well as applying any rewriting rules that may
2676 exist.
2677
2678 Qualification of header addresses in a message from a remote host happens only
2679 if the host is in sender_unqualified_hosts or recipient_unqualified hosts, as
2680 appropriate. For local messages, qualification always happens, unless -bnq is
2681 used to explicitly suppress it. No rewriting is done for an unqualified address
2682 that is left untouched.
2683
2684 We start at the second header, skipping our own Received:. This rewriting is
2685 documented as happening *after* recipient addresses are taken from the headers
2686 by the -t command line option. An added Sender: gets rewritten here. */
2687
2688 for (h = header_list->next; h != NULL; h = h->next)
2689 {
2690 header_line *newh = rewrite_header(h, NULL, NULL, global_rewrite_rules,
2691 rewrite_existflags, TRUE);
2692 if (newh != NULL) h = newh;
2693 }
2694
2695
2696 /* An RFC 822 (sic) message is not legal unless it has at least one of "to",
2697 "cc", or "bcc". Note that although the minimal examples in RFC 822 show just
2698 "to" or "bcc", the full syntax spec allows "cc" as well. If any resent- header
2699 exists, this applies to the set of resent- headers rather than the normal set.
2700
2701 The requirement for a recipient header has been removed in RFC 2822. At this
2702 point in the code, earlier versions of Exim added a To: header for locally
2703 submitted messages, and an empty Bcc: header for others. In the light of the
2704 changes in RFC 2822, this was dropped in November 2003. */
2705
2706
2707 /* If there is no date header, generate one if the message originates locally
2708 (i.e. not over TCP/IP) and suppress_local_fixups is not set, or if the
2709 submission mode flag is set. Messages without Date: are not valid, but it seems
2710 to be more confusing if Exim adds one to all remotely-originated messages.
2711 As per Message-Id, we prepend if resending, else append.
2712 */
2713
2714 if (!date_header_exists &&
2715 ((sender_host_address == NULL && !suppress_local_fixups)
2716 || submission_mode))
2717 header_add_at_position(!resents_exist, NULL, FALSE, htype_other,
2718 "%sDate: %s\n", resent_prefix, tod_stamp(tod_full));
2719
2720 search_tidyup(); /* Free any cached resources */
2721
2722 /* Show the complete set of headers if debugging. Note that the first one (the
2723 new Received:) has not yet been set. */
2724
2725 DEBUG(D_receive)
2726 {
2727 debug_printf(">>Headers after rewriting and local additions:\n");
2728 for (h = header_list->next; h != NULL; h = h->next)
2729 debug_printf("%c %s", h->type, h->text);
2730 debug_printf("\n");
2731 }
2732
2733 /* The headers are now complete in store. If we are running in filter
2734 testing mode, that is all this function does. Return TRUE if the message
2735 ended with a dot. */
2736
2737 if (filter_test != FTEST_NONE)
2738 {
2739 process_info[process_info_len] = 0;
2740 return message_ended == END_DOT;
2741 }
2742
2743 /* Cutthrough delivery:
2744 We have to create the Received header now rather than at the end of reception,
2745 so the timestamp behaviour is a change to the normal case.
2746 XXX Ensure this gets documented XXX.
2747 Having created it, send the headers to the destination.
2748 */
2749 if (cutthrough_fd >= 0)
2750 {
2751 if (received_count > received_headers_max)
2752 {
2753 cancel_cutthrough_connection("too many headers");
2754 if (smtp_input) receive_swallow_smtp(); /* Swallow incoming SMTP */
2755 log_write(0, LOG_MAIN|LOG_REJECT, "rejected from <%s>%s%s%s%s: "
2756 "Too many \"Received\" headers",
2757 sender_address,
2758 (sender_fullhost == NULL)? "" : " H=",
2759 (sender_fullhost == NULL)? US"" : sender_fullhost,
2760 (sender_ident == NULL)? "" : " U=",
2761 (sender_ident == NULL)? US"" : sender_ident);
2762 message_id[0] = 0; /* Indicate no message accepted */
2763 smtp_reply = US"550 Too many \"Received\" headers - suspected mail loop";
2764 goto TIDYUP; /* Skip to end of function */
2765 }
2766 received_header_gen();
2767 add_acl_headers(US"MAIL or RCPT");
2768 (void) cutthrough_headers_send();
2769 }
2770
2771
2772 /* Open a new spool file for the data portion of the message. We need
2773 to access it both via a file descriptor and a stream. Try to make the
2774 directory if it isn't there. Note re use of sprintf: spool_directory
2775 is checked on input to be < 200 characters long. */
2776
2777 sprintf(CS spool_name, "%s/input/%s/%s-D", spool_directory, message_subdir,
2778 message_id);
2779 data_fd = Uopen(spool_name, O_RDWR|O_CREAT|O_EXCL, SPOOL_MODE);
2780 if (data_fd < 0)
2781 {
2782 if (errno == ENOENT)
2783 {
2784 uschar temp[16];
2785 sprintf(CS temp, "input/%s", message_subdir);
2786 if (message_subdir[0] == 0) temp[5] = 0;
2787 (void)directory_make(spool_directory, temp, INPUT_DIRECTORY_MODE, TRUE);
2788 data_fd = Uopen(spool_name, O_RDWR|O_CREAT|O_EXCL, SPOOL_MODE);
2789 }
2790 if (data_fd < 0)
2791 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Failed to create spool file %s: %s",
2792 spool_name, strerror(errno));
2793 }
2794
2795 /* Make sure the file's group is the Exim gid, and double-check the mode
2796 because the group setting doesn't always get set automatically. */
2797
2798 (void)fchown(data_fd, exim_uid, exim_gid);
2799 (void)fchmod(data_fd, SPOOL_MODE);
2800
2801 /* We now have data file open. Build a stream for it and lock it. We lock only
2802 the first line of the file (containing the message ID) because otherwise there
2803 are problems when Exim is run under Cygwin (I'm told). See comments in
2804 spool_in.c, where the same locking is done. */
2805
2806 data_file = fdopen(data_fd, "w+");
2807 lock_data.l_type = F_WRLCK;
2808 lock_data.l_whence = SEEK_SET;
2809 lock_data.l_start = 0;
2810 lock_data.l_len = SPOOL_DATA_START_OFFSET;
2811
2812 if (fcntl(data_fd, F_SETLK, &lock_data) < 0)
2813 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Cannot lock %s (%d): %s", spool_name,
2814 errno, strerror(errno));
2815
2816 /* We have an open, locked data file. Write the message id to it to make it
2817 self-identifying. Then read the remainder of the input of this message and
2818 write it to the data file. If the variable next != NULL, it contains the first
2819 data line (which was read as a header but then turned out not to have the right
2820 format); write it (remembering that it might contain binary zeros). The result
2821 of fwrite() isn't inspected; instead we call ferror() below. */
2822
2823 fprintf(data_file, "%s-D\n", message_id);
2824 if (next != NULL)
2825 {
2826 uschar *s = next->text;
2827 int len = next->slen;
2828 (void)fwrite(s, 1, len, data_file);
2829 body_linecount++; /* Assumes only 1 line */
2830 }
2831
2832 /* Note that we might already be at end of file, or the logical end of file
2833 (indicated by '.'), or might have encountered an error while writing the
2834 message id or "next" line. */
2835
2836 if (!ferror(data_file) && !(receive_feof)() && message_ended != END_DOT)
2837 {
2838 if (smtp_input)
2839 {
2840 message_ended = read_message_data_smtp(data_file);
2841 receive_linecount++; /* The terminating "." line */
2842 }
2843 else message_ended = read_message_data(data_file);
2844
2845 receive_linecount += body_linecount; /* For BSMTP errors mainly */
2846 message_linecount += body_linecount;
2847
2848 /* Handle premature termination of SMTP */
2849
2850 if (smtp_input && message_ended == END_EOF)
2851 {
2852 Uunlink(spool_name); /* Lose data file when closed */
2853 cancel_cutthrough_connection("sender closed connection");
2854 message_id[0] = 0; /* Indicate no message accepted */
2855 smtp_reply = handle_lost_connection(US"");
2856 smtp_yield = FALSE;
2857 goto TIDYUP; /* Skip to end of function */
2858 }
2859
2860 /* Handle message that is too big. Don't use host_or_ident() in the log
2861 message; we want to see the ident value even for non-remote messages. */
2862
2863 if (message_ended == END_SIZE)
2864 {
2865 Uunlink(spool_name); /* Lose the data file when closed */
2866 cancel_cutthrough_connection("mail too big");
2867 if (smtp_input) receive_swallow_smtp(); /* Swallow incoming SMTP */
2868
2869 log_write(L_size_reject, LOG_MAIN|LOG_REJECT, "rejected from <%s>%s%s%s%s: "
2870 "message too big: read=%d max=%d",
2871 sender_address,
2872 (sender_fullhost == NULL)? "" : " H=",
2873 (sender_fullhost == NULL)? US"" : sender_fullhost,
2874 (sender_ident == NULL)? "" : " U=",
2875 (sender_ident == NULL)? US"" : sender_ident,
2876 message_size,
2877 thismessage_size_limit);
2878
2879 if (smtp_input)
2880 {
2881 smtp_reply = US"552 Message size exceeds maximum permitted";
2882 message_id[0] = 0; /* Indicate no message accepted */
2883 goto TIDYUP; /* Skip to end of function */
2884 }
2885 else
2886 {
2887 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
2888 give_local_error(ERRMESS_TOOBIG,
2889 string_sprintf("message too big (max=%d)", thismessage_size_limit),
2890 US"message rejected: ", error_rc, data_file, header_list);
2891 /* Does not return */
2892 }
2893 }
2894 }
2895
2896 /* Restore the standard SIGALRM handler for any subsequent processing. (For
2897 example, there may be some expansion in an ACL that uses a timer.) */
2898
2899 os_non_restarting_signal(SIGALRM, sigalrm_handler);
2900
2901 /* The message body has now been read into the data file. Call fflush() to
2902 empty the buffers in C, and then call fsync() to get the data written out onto
2903 the disk, as fflush() doesn't do this (or at least, it isn't documented as
2904 having to do this). If there was an I/O error on either input or output,
2905 attempt to send an error message, and unlink the spool file. For non-SMTP input
2906 we can then give up. Note that for SMTP input we must swallow the remainder of
2907 the input in cases of output errors, since the far end doesn't expect to see
2908 anything until the terminating dot line is sent. */
2909
2910 if (fflush(data_file) == EOF || ferror(data_file) ||
2911 EXIMfsync(fileno(data_file)) < 0 || (receive_ferror)())
2912 {
2913 uschar *msg_errno = US strerror(errno);
2914 BOOL input_error = (receive_ferror)() != 0;
2915 uschar *msg = string_sprintf("%s error (%s) while receiving message from %s",
2916 input_error? "Input read" : "Spool write",
2917 msg_errno,
2918 (sender_fullhost != NULL)? sender_fullhost : sender_ident);
2919
2920 log_write(0, LOG_MAIN, "Message abandoned: %s", msg);
2921 Uunlink(spool_name); /* Lose the data file */
2922 cancel_cutthrough_connection("error writing spoolfile");
2923
2924 if (smtp_input)
2925 {
2926 if (input_error)
2927 smtp_reply = US"451 Error while reading input data";
2928 else
2929 {
2930 smtp_reply = US"451 Error while writing spool file";
2931 receive_swallow_smtp();
2932 }
2933 message_id[0] = 0; /* Indicate no message accepted */
2934 goto TIDYUP; /* Skip to end of function */
2935 }
2936
2937 else
2938 {
2939 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
2940 give_local_error(ERRMESS_IOERR, msg, US"", error_rc, data_file,
2941 header_list);
2942 /* Does not return */
2943 }
2944 }
2945
2946
2947 /* No I/O errors were encountered while writing the data file. */
2948
2949 DEBUG(D_receive) debug_printf("Data file written for message %s\n", message_id);
2950
2951
2952 /* If there were any bad addresses extracted by -t, or there were no recipients
2953 left after -t, send a message to the sender of this message, or write it to
2954 stderr if the error handling option is set that way. Note that there may
2955 legitimately be no recipients for an SMTP message if they have all been removed
2956 by "discard".
2957
2958 We need to rewind the data file in order to read it. In the case of no
2959 recipients or stderr error writing, throw the data file away afterwards, and
2960 exit. (This can't be SMTP, which always ensures there's at least one
2961 syntactically good recipient address.) */
2962
2963 if (extract_recip && (bad_addresses != NULL || recipients_count == 0))
2964 {
2965 DEBUG(D_receive)
2966 {
2967 if (recipients_count == 0) debug_printf("*** No recipients\n");
2968 if (bad_addresses != NULL)
2969 {
2970 error_block *eblock = bad_addresses;
2971 debug_printf("*** Bad address(es)\n");
2972 while (eblock != NULL)
2973 {
2974 debug_printf(" %s: %s\n", eblock->text1, eblock->text2);
2975 eblock = eblock->next;
2976 }
2977 }
2978 }
2979
2980 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
2981
2982 /* If configured to send errors to the sender, but this fails, force
2983 a failure error code. We use a special one for no recipients so that it
2984 can be detected by the autoreply transport. Otherwise error_rc is set to
2985 errors_sender_rc, which is EXIT_FAILURE unless -oee was given, in which case
2986 it is EXIT_SUCCESS. */
2987
2988 if (error_handling == ERRORS_SENDER)
2989 {
2990 if (!moan_to_sender(
2991 (bad_addresses == NULL)?
2992 (extracted_ignored? ERRMESS_IGADDRESS : ERRMESS_NOADDRESS) :
2993 (recipients_list == NULL)? ERRMESS_BADNOADDRESS : ERRMESS_BADADDRESS,
2994 bad_addresses, header_list, data_file, FALSE))
2995 error_rc = (bad_addresses == NULL)? EXIT_NORECIPIENTS : EXIT_FAILURE;
2996 }
2997 else
2998 {
2999 if (bad_addresses == NULL)
3000 {
3001 if (extracted_ignored)
3002 fprintf(stderr, "exim: all -t recipients overridden by command line\n");
3003 else
3004 fprintf(stderr, "exim: no recipients in message\n");
3005 }
3006 else
3007 {
3008 fprintf(stderr, "exim: invalid address%s",
3009 (bad_addresses->next == NULL)? ":" : "es:\n");
3010 while (bad_addresses != NULL)
3011 {
3012 fprintf(stderr, " %s: %s\n", bad_addresses->text1,
3013 bad_addresses->text2);
3014 bad_addresses = bad_addresses->next;
3015 }
3016 }
3017 }
3018
3019 if (recipients_count == 0 || error_handling == ERRORS_STDERR)
3020 {
3021 Uunlink(spool_name);
3022 (void)fclose(data_file);
3023 exim_exit(error_rc);
3024 }
3025 }
3026
3027 /* Data file successfully written. Generate text for the Received: header by
3028 expanding the configured string, and adding a timestamp. By leaving this
3029 operation till now, we ensure that the timestamp is the time that message
3030 reception was completed. However, this is deliberately done before calling the
3031 data ACL and local_scan().
3032
3033 This Received: header may therefore be inspected by the data ACL and by code in
3034 the local_scan() function. When they have run, we update the timestamp to be
3035 the final time of reception.
3036
3037 If there is just one recipient, set up its value in the $received_for variable
3038 for use when we generate the Received: header.
3039
3040 Note: the checking for too many Received: headers is handled by the delivery
3041 code. */
3042 /*XXX eventually add excess Received: check for cutthrough case back when classifying them */
3043
3044 if (received_header->text == NULL) /* Non-cutthrough case */
3045 {
3046 received_header_gen();
3047
3048 /* Set the value of message_body_size for the DATA ACL and for local_scan() */
3049
3050 message_body_size = (fstat(data_fd, &statbuf) == 0)?
3051 statbuf.st_size - SPOOL_DATA_START_OFFSET : -1;
3052
3053 /* If an ACL from any RCPT commands set up any warning headers to add, do so
3054 now, before running the DATA ACL. */
3055
3056 add_acl_headers(US"MAIL or RCPT");
3057 }
3058 else
3059 message_body_size = (fstat(data_fd, &statbuf) == 0)?
3060 statbuf.st_size - SPOOL_DATA_START_OFFSET : -1;
3061
3062 /* If an ACL is specified for checking things at this stage of reception of a
3063 message, run it, unless all the recipients were removed by "discard" in earlier
3064 ACLs. That is the only case in which recipients_count can be zero at this
3065 stage. Set deliver_datafile to point to the data file so that $message_body and
3066 $message_body_end can be extracted if needed. Allow $recipients in expansions.
3067 */
3068
3069 deliver_datafile = data_fd;
3070 user_msg = NULL;
3071
3072 enable_dollar_recipients = TRUE;
3073
3074 if (recipients_count == 0)
3075 {
3076 blackholed_by = recipients_discarded? US"MAIL ACL" : US"RCPT ACL";
3077 }
3078 else
3079 {
3080 /* Handle interactive SMTP messages */
3081
3082 if (smtp_input && !smtp_batched_input)
3083 {
3084
3085 #ifndef DISABLE_DKIM
3086 if (!dkim_disable_verify)
3087 {
3088 /* Finish verification, this will log individual signature results to
3089 the mainlog */
3090 dkim_exim_verify_finish();
3091
3092 /* Check if we must run the DKIM ACL */
3093 if ((acl_smtp_dkim != NULL) &&
3094 (dkim_verify_signers != NULL) &&
3095 (dkim_verify_signers[0] != '\0'))
3096 {
3097 uschar *dkim_verify_signers_expanded =
3098 expand_string(dkim_verify_signers);
3099 if (dkim_verify_signers_expanded == NULL)
3100 {
3101 log_write(0, LOG_MAIN|LOG_PANIC,
3102 "expansion of dkim_verify_signers option failed: %s",
3103 expand_string_message);
3104 }
3105 else
3106 {
3107 int sep = 0;
3108 uschar *ptr = dkim_verify_signers_expanded;
3109 uschar *item = NULL;
3110 uschar *seen_items = NULL;
3111 int seen_items_size = 0;
3112 int seen_items_offset = 0;
3113 uschar itembuf[256];
3114 /* Default to OK when no items are present */
3115 rc = OK;
3116 while ((item = string_nextinlist(&ptr, &sep,
3117 itembuf,
3118 sizeof(itembuf))) != NULL)
3119 {
3120 /* Prevent running ACL for an empty item */
3121 if (!item || (item[0] == '\0')) continue;
3122 /* Only run ACL once for each domain or identity, no matter how often it
3123 appears in the expanded list. */
3124 if (seen_items != NULL)
3125 {
3126 uschar *seen_item = NULL;
3127 uschar seen_item_buf[256];
3128 uschar *seen_items_list = seen_items;
3129 int seen_this_item = 0;
3130
3131 while ((seen_item = string_nextinlist(&seen_items_list, &sep,
3132 seen_item_buf,
3133 sizeof(seen_item_buf))) != NULL)
3134 {
3135 if (Ustrcmp(seen_item,item) == 0)
3136 {
3137 seen_this_item = 1;
3138 break;
3139 }
3140 }
3141
3142 if (seen_this_item > 0)
3143 {
3144 DEBUG(D_receive)
3145 debug_printf("acl_smtp_dkim: skipping signer %s, already seen\n", item);
3146 continue;
3147 }
3148
3149 seen_items = string_append(seen_items,&seen_items_size,&seen_items_offset,1,":");
3150 }
3151
3152 seen_items = string_append(seen_items,&seen_items_size,&seen_items_offset,1,item);
3153 seen_items[seen_items_offset] = '\0';
3154
3155 DEBUG(D_receive)
3156 debug_printf("calling acl_smtp_dkim for dkim_cur_signer=%s\n", item);
3157
3158 dkim_exim_acl_setup(item);
3159 rc = acl_check(ACL_WHERE_DKIM, NULL, acl_smtp_dkim, &user_msg, &log_msg);
3160
3161 if (rc != OK)
3162 {
3163 DEBUG(D_receive)
3164 debug_printf("acl_smtp_dkim: acl_check returned %d on %s, skipping remaining items\n", rc, item);
3165 cancel_cutthrough_connection("dkim acl not ok");
3166 break;
3167 }
3168 }
3169 add_acl_headers(US"DKIM");
3170 if (rc == DISCARD)
3171 {
3172 recipients_count = 0;
3173 blackholed_by = US"DKIM ACL";
3174 if (log_msg != NULL)
3175 blackhole_log_msg = string_sprintf(": %s", log_msg);
3176 }
3177 else if (rc != OK)
3178 {
3179 Uunlink(spool_name);
3180 if (smtp_handle_acl_fail(ACL_WHERE_DKIM, rc, user_msg, log_msg) != 0)
3181 smtp_yield = FALSE; /* No more messsages after dropped connection */
3182 smtp_reply = US""; /* Indicate reply already sent */
3183 message_id[0] = 0; /* Indicate no message accepted */
3184 goto TIDYUP; /* Skip to end of function */
3185 }
3186 }
3187 }
3188 }
3189 #endif /* DISABLE_DKIM */
3190
3191 #ifdef WITH_CONTENT_SCAN
3192 if (recipients_count > 0 &&
3193 acl_smtp_mime != NULL &&
3194 !run_mime_acl(acl_smtp_mime, &smtp_yield, &smtp_reply, &blackholed_by))
3195 goto TIDYUP;
3196 #endif /* WITH_CONTENT_SCAN */
3197
3198 /* Check the recipients count again, as the MIME ACL might have changed
3199 them. */
3200
3201 if (acl_smtp_data != NULL && recipients_count > 0)
3202 {
3203 rc = acl_check(ACL_WHERE_DATA, NULL, acl_smtp_data, &user_msg, &log_msg);
3204 add_acl_headers(US"DATA");
3205 if (rc == DISCARD)
3206 {
3207 recipients_count = 0;
3208 blackholed_by = US"DATA ACL";
3209 if (log_msg != NULL)
3210 blackhole_log_msg = string_sprintf(": %s", log_msg);
3211 cancel_cutthrough_connection("data acl discard");
3212 }
3213 else if (rc != OK)
3214 {
3215 Uunlink(spool_name);
3216 cancel_cutthrough_connection("data acl not ok");
3217 #ifdef WITH_CONTENT_SCAN
3218 unspool_mbox();
3219 #endif
3220 #ifdef EXPERIMENTAL_DCC
3221 dcc_ok = 0;
3222 #endif
3223 if (smtp_handle_acl_fail(ACL_WHERE_DATA, rc, user_msg, log_msg) != 0)
3224 smtp_yield = FALSE; /* No more messsages after dropped connection */
3225 smtp_reply = US""; /* Indicate reply already sent */
3226 message_id[0] = 0; /* Indicate no message accepted */
3227 goto TIDYUP; /* Skip to end of function */
3228 }
3229 }
3230 }
3231
3232 /* Handle non-SMTP and batch SMTP (i.e. non-interactive) messages. Note that
3233 we cannot take different actions for permanent and temporary rejections. */
3234
3235 else
3236 {
3237
3238 #ifdef WITH_CONTENT_SCAN
3239 if (acl_not_smtp_mime != NULL &&
3240 !run_mime_acl(acl_not_smtp_mime, &smtp_yield, &smtp_reply,
3241 &blackholed_by))
3242 goto TIDYUP;
3243 #endif /* WITH_CONTENT_SCAN */
3244
3245 if (acl_not_smtp != NULL)
3246 {
3247 uschar *user_msg, *log_msg;
3248 rc = acl_check(ACL_WHERE_NOTSMTP, NULL, acl_not_smtp, &user_msg, &log_msg);
3249 if (rc == DISCARD)
3250 {
3251 recipients_count = 0;
3252 blackholed_by = US"non-SMTP ACL";
3253 if (log_msg != NULL)
3254 blackhole_log_msg = string_sprintf(": %s", log_msg);
3255 }
3256 else if (rc != OK)
3257 {
3258 Uunlink(spool_name);
3259 #ifdef WITH_CONTENT_SCAN
3260 unspool_mbox();
3261 #endif
3262 #ifdef EXPERIMENTAL_DCC
3263 dcc_ok = 0;
3264 #endif
3265 /* The ACL can specify where rejections are to be logged, possibly
3266 nowhere. The default is main and reject logs. */
3267
3268 if (log_reject_target != 0)
3269 log_write(0, log_reject_target, "F=<%s> rejected by non-SMTP ACL: %s",
3270 sender_address, log_msg);
3271
3272 if (user_msg == NULL) user_msg = US"local configuration problem";
3273 if (smtp_batched_input)
3274 {
3275 moan_smtp_batch(NULL, "%d %s", 550, user_msg);
3276 /* Does not return */
3277 }
3278 else
3279 {
3280 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3281 give_local_error(ERRMESS_LOCAL_ACL, user_msg,
3282 US"message rejected by non-SMTP ACL: ", error_rc, data_file,
3283 header_list);
3284 /* Does not return */
3285 }
3286 }
3287 add_acl_headers(US"non-SMTP");
3288 }
3289 }
3290
3291 /* The applicable ACLs have been run */
3292
3293 if (deliver_freeze) frozen_by = US"ACL"; /* for later logging */
3294 if (queue_only_policy) queued_by = US"ACL";
3295 }
3296
3297 #ifdef WITH_CONTENT_SCAN
3298 unspool_mbox();
3299 #endif
3300
3301 #ifdef EXPERIMENTAL_DCC
3302 dcc_ok = 0;
3303 #endif
3304
3305
3306 /* The final check on the message is to run the scan_local() function. The
3307 version supplied with Exim always accepts, but this is a hook for sysadmins to
3308 supply their own checking code. The local_scan() function is run even when all
3309 the recipients have been discarded. */
3310
3311 lseek(data_fd, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3312
3313 /* Arrange to catch crashes in local_scan(), so that the -D file gets
3314 deleted, and the incident gets logged. */
3315
3316 os_non_restarting_signal(SIGSEGV, local_scan_crash_handler);
3317 os_non_restarting_signal(SIGFPE, local_scan_crash_handler);
3318 os_non_restarting_signal(SIGILL, local_scan_crash_handler);
3319 os_non_restarting_signal(SIGBUS, local_scan_crash_handler);
3320
3321 DEBUG(D_receive) debug_printf("calling local_scan(); timeout=%d\n",
3322 local_scan_timeout);
3323 local_scan_data = NULL;
3324
3325 os_non_restarting_signal(SIGALRM, local_scan_timeout_handler);
3326 if (local_scan_timeout > 0) alarm(local_scan_timeout);
3327 rc = local_scan(data_fd, &local_scan_data);
3328 alarm(0);
3329 os_non_restarting_signal(SIGALRM, sigalrm_handler);
3330
3331 enable_dollar_recipients = FALSE;
3332
3333 store_pool = POOL_MAIN; /* In case changed */
3334 DEBUG(D_receive) debug_printf("local_scan() returned %d %s\n", rc,
3335 local_scan_data);
3336
3337 os_non_restarting_signal(SIGSEGV, SIG_DFL);
3338 os_non_restarting_signal(SIGFPE, SIG_DFL);
3339 os_non_restarting_signal(SIGILL, SIG_DFL);
3340 os_non_restarting_signal(SIGBUS, SIG_DFL);
3341
3342 /* The length check is paranoia against some runaway code, and also because
3343 (for a success return) lines in the spool file are read into big_buffer. */
3344
3345 if (local_scan_data != NULL)
3346 {
3347 int len = Ustrlen(local_scan_data);
3348 if (len > LOCAL_SCAN_MAX_RETURN) len = LOCAL_SCAN_MAX_RETURN;
3349 local_scan_data = string_copyn(local_scan_data, len);
3350 }
3351
3352 if (rc == LOCAL_SCAN_ACCEPT_FREEZE)
3353 {
3354 if (!deliver_freeze) /* ACL might have already frozen */
3355 {
3356 deliver_freeze = TRUE;
3357 deliver_frozen_at = time(NULL);
3358 frozen_by = US"local_scan()";
3359 }
3360 rc = LOCAL_SCAN_ACCEPT;
3361 }
3362 else if (rc == LOCAL_SCAN_ACCEPT_QUEUE)
3363 {
3364 if (!queue_only_policy) /* ACL might have already queued */
3365 {
3366 queue_only_policy = TRUE;
3367 queued_by = US"local_scan()";
3368 }
3369 rc = LOCAL_SCAN_ACCEPT;
3370 }
3371
3372 /* Message accepted: remove newlines in local_scan_data because otherwise
3373 the spool file gets corrupted. Ensure that all recipients are qualified. */
3374
3375 if (rc == LOCAL_SCAN_ACCEPT)
3376 {
3377 if (local_scan_data != NULL)
3378 {
3379 uschar *s;
3380 for (s = local_scan_data; *s != 0; s++) if (*s == '\n') *s = ' ';
3381 }
3382 for (i = 0; i < recipients_count; i++)
3383 {
3384 recipient_item *r = recipients_list + i;
3385 r->address = rewrite_address_qualify(r->address, TRUE);
3386 if (r->errors_to != NULL)
3387 r->errors_to = rewrite_address_qualify(r->errors_to, TRUE);
3388 }
3389 if (recipients_count == 0 && blackholed_by == NULL)
3390 blackholed_by = US"local_scan";
3391 }
3392
3393 /* Message rejected: newlines permitted in local_scan_data to generate
3394 multiline SMTP responses. */
3395
3396 else
3397 {
3398 uschar *istemp = US"";
3399 uschar *s = NULL;
3400 uschar *smtp_code;
3401 int size = 0;
3402 int sptr = 0;
3403
3404 errmsg = local_scan_data;
3405
3406 Uunlink(spool_name); /* Cancel this message */
3407 switch(rc)
3408 {
3409 default:
3410 log_write(0, LOG_MAIN, "invalid return %d from local_scan(). Temporary "
3411 "rejection given", rc);
3412 goto TEMPREJECT;
3413
3414 case LOCAL_SCAN_REJECT_NOLOGHDR:
3415 log_extra_selector &= ~LX_rejected_header;
3416 /* Fall through */
3417
3418 case LOCAL_SCAN_REJECT:
3419 smtp_code = US"550";
3420 if (errmsg == NULL) errmsg = US"Administrative prohibition";
3421 break;
3422
3423 case LOCAL_SCAN_TEMPREJECT_NOLOGHDR:
3424 log_extra_selector &= ~LX_rejected_header;
3425 /* Fall through */
3426
3427 case LOCAL_SCAN_TEMPREJECT:
3428 TEMPREJECT:
3429 smtp_code = US"451";
3430 if (errmsg == NULL) errmsg = US"Temporary local problem";
3431 istemp = US"temporarily ";
3432 break;
3433 }
3434
3435 s = string_append(s, &size, &sptr, 2, US"F=",
3436 (sender_address[0] == 0)? US"<>" : sender_address);
3437 s = add_host_info_for_log(s, &size, &sptr);
3438 s[sptr] = 0;
3439
3440 log_write(0, LOG_MAIN|LOG_REJECT, "%s %srejected by local_scan(): %.256s",
3441 s, istemp, string_printing(errmsg));
3442
3443 if (smtp_input)
3444 {
3445 if (!smtp_batched_input)
3446 {
3447 smtp_respond(smtp_code, 3, TRUE, errmsg);
3448 message_id[0] = 0; /* Indicate no message accepted */
3449 smtp_reply = US""; /* Indicate reply already sent */
3450 goto TIDYUP; /* Skip to end of function */
3451 }
3452 else
3453 {
3454 moan_smtp_batch(NULL, "%s %s", smtp_code, errmsg);
3455 /* Does not return */
3456 }
3457 }
3458 else
3459 {
3460 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3461 give_local_error(ERRMESS_LOCAL_SCAN, errmsg,
3462 US"message rejected by local scan code: ", error_rc, data_file,
3463 header_list);
3464 /* Does not return */
3465 }
3466 }
3467
3468 /* Reset signal handlers to ignore signals that previously would have caused
3469 the message to be abandoned. */
3470
3471 signal(SIGTERM, SIG_IGN);
3472 signal(SIGINT, SIG_IGN);
3473
3474
3475 /* Ensure the first time flag is set in the newly-received message. */
3476
3477 deliver_firsttime = TRUE;
3478
3479 #ifdef EXPERIMENTAL_BRIGHTMAIL
3480 if (bmi_run == 1) {
3481 /* rewind data file */
3482 lseek(data_fd, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3483 bmi_verdicts = bmi_process_message(header_list, data_fd);
3484 };
3485 #endif
3486
3487 /* Update the timstamp in our Received: header to account for any time taken by
3488 an ACL or by local_scan(). The new time is the time that all reception
3489 processing is complete. */
3490
3491 timestamp = expand_string(US"${tod_full}");
3492 tslen = Ustrlen(timestamp);
3493
3494 memcpy(received_header->text + received_header->slen - tslen - 1,
3495 timestamp, tslen);
3496
3497 /* In MUA wrapper mode, ignore queueing actions set by ACL or local_scan() */
3498
3499 if (mua_wrapper)
3500 {
3501 deliver_freeze = FALSE;
3502 queue_only_policy = FALSE;
3503 }
3504
3505 /* Keep the data file open until we have written the header file, in order to
3506 hold onto the lock. In a -bh run, or if the message is to be blackholed, we
3507 don't write the header file, and we unlink the data file. If writing the header
3508 file fails, we have failed to accept this message. */
3509
3510 if (host_checking || blackholed_by != NULL)
3511 {
3512 header_line *h;
3513 Uunlink(spool_name);
3514 msg_size = 0; /* Compute size for log line */
3515 for (h = header_list; h != NULL; h = h->next)
3516 if (h->type != '*') msg_size += h->slen;
3517 }
3518
3519 /* Write the -H file */
3520
3521 else
3522 {
3523 if ((msg_size = spool_write_header(message_id, SW_RECEIVING, &errmsg)) < 0)
3524 {
3525 log_write(0, LOG_MAIN, "Message abandoned: %s", errmsg);
3526 Uunlink(spool_name); /* Lose the data file */
3527
3528 if (smtp_input)
3529 {
3530 smtp_reply = US"451 Error in writing spool file";
3531 message_id[0] = 0; /* Indicate no message accepted */
3532 goto TIDYUP;
3533 }
3534 else
3535 {
3536 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3537 give_local_error(ERRMESS_IOERR, errmsg, US"", error_rc, data_file,
3538 header_list);
3539 /* Does not return */
3540 }
3541 }
3542 }
3543
3544
3545 /* The message has now been successfully received. */
3546
3547 receive_messagecount++;
3548
3549 /* In SMTP sessions we may receive several in one connection. After each one,
3550 we wait for the clock to tick at the level of message-id granularity. This is
3551 so that the combination of time+pid is unique, even on systems where the pid
3552 can be re-used within our time interval. We can't shorten the interval without
3553 re-designing the message-id. See comments above where the message id is
3554 created. This is Something For The Future. */
3555
3556 message_id_tv.tv_usec = (message_id_tv.tv_usec/id_resolution) * id_resolution;
3557 exim_wait_tick(&message_id_tv, id_resolution);
3558
3559 /* Add data size to written header size. We do not count the initial file name
3560 that is in the file, but we do add one extra for the notional blank line that
3561 precedes the data. This total differs from message_size in that it include the
3562 added Received: header and any other headers that got created locally. */
3563
3564 fflush(data_file);
3565 fstat(data_fd, &statbuf);
3566
3567 msg_size += statbuf.st_size - SPOOL_DATA_START_OFFSET + 1;
3568
3569 /* Generate a "message received" log entry. We do this by building up a dynamic
3570 string as required. Since we commonly want to add two items at a time, use a
3571 macro to simplify the coding. We log the arrival of a new message while the
3572 file is still locked, just in case the machine is *really* fast, and delivers
3573 it first! Include any message id that is in the message - since the syntax of a
3574 message id is actually an addr-spec, we can use the parse routine to canonicize
3575 it. */
3576
3577 size = 256;
3578 sptr = 0;
3579 s = store_get(size);
3580
3581 s = string_append(s, &size, &sptr, 2, US"<= ",
3582 (sender_address[0] == 0)? US"<>" : sender_address);
3583 if (message_reference != NULL)
3584 s = string_append(s, &size, &sptr, 2, US" R=", message_reference);
3585
3586 s = add_host_info_for_log(s, &size, &sptr);
3587
3588 #ifdef SUPPORT_TLS
3589 if ((log_extra_selector & LX_tls_cipher) != 0 && tls_in.cipher != NULL)
3590 s = string_append(s, &size, &sptr, 2, US" X=", tls_in.cipher);
3591 if ((log_extra_selector & LX_tls_certificate_verified) != 0 &&
3592 tls_in.cipher != NULL)
3593 s = string_append(s, &size, &sptr, 2, US" CV=",
3594 tls_in.certificate_verified? "yes":"no");
3595 if ((log_extra_selector & LX_tls_peerdn) != 0 && tls_in.peerdn != NULL)
3596 s = string_append(s, &size, &sptr, 3, US" DN=\"",
3597 string_printing(tls_in.peerdn), US"\"");
3598 if ((log_extra_selector & LX_tls_sni) != 0 && tls_in.sni != NULL)
3599 s = string_append(s, &size, &sptr, 3, US" SNI=\"",
3600 string_printing(tls_in.sni), US"\"");
3601 #endif
3602
3603 if (sender_host_authenticated != NULL)
3604 {
3605 s = string_append(s, &size, &sptr, 2, US" A=", sender_host_authenticated);
3606 if (authenticated_id != NULL)
3607 s = string_append(s, &size, &sptr, 2, US":", authenticated_id);
3608 }
3609
3610 sprintf(CS big_buffer, "%d", msg_size);
3611 s = string_append(s, &size, &sptr, 2, US" S=", big_buffer);
3612
3613 /* log 8BITMIME mode announced in MAIL_FROM
3614 0 ... no BODY= used
3615 7 ... 7BIT
3616 8 ... 8BITMIME */
3617 if (log_extra_selector & LX_8bitmime) {
3618 sprintf(CS big_buffer, "%d", body_8bitmime);
3619 s = string_append(s, &size, &sptr, 2, US" M8S=", big_buffer);
3620 }
3621
3622 /* If an addr-spec in a message-id contains a quoted string, it can contain
3623 any characters except " \ and CR and so in particular it can contain NL!
3624 Therefore, make sure we use a printing-characters only version for the log.
3625 Also, allow for domain literals in the message id. */
3626
3627 if (msgid_header != NULL)
3628 {
3629 uschar *old_id;
3630 BOOL save_allow_domain_literals = allow_domain_literals;
3631 allow_domain_literals = TRUE;
3632 old_id = parse_extract_address(Ustrchr(msgid_header->text, ':') + 1,
3633 &errmsg, &start, &end, &domain, FALSE);
3634 allow_domain_literals = save_allow_domain_literals;
3635 if (old_id != NULL)
3636 s = string_append(s, &size, &sptr, 2, US" id=", string_printing(old_id));
3637 }
3638
3639 /* If subject logging is turned on, create suitable printing-character
3640 text. By expanding $h_subject: we make use of the MIME decoding. */
3641
3642 if ((log_extra_selector & LX_subject) != 0 && subject_header != NULL)
3643 {
3644 int i;
3645 uschar *p = big_buffer;
3646 uschar *ss = expand_string(US"$h_subject:");
3647
3648 /* Backslash-quote any double quotes or backslashes so as to make a
3649 a C-like string, and turn any non-printers into escape sequences. */
3650
3651 *p++ = '\"';
3652 if (*ss != 0) for (i = 0; i < 100 && ss[i] != 0; i++)
3653 {
3654 if (ss[i] == '\"' || ss[i] == '\\') *p++ = '\\';
3655 *p++ = ss[i];
3656 }
3657 *p++ = '\"';
3658 *p = 0;
3659 s = string_append(s, &size, &sptr, 2, US" T=", string_printing(big_buffer));
3660 }
3661
3662 /* Terminate the string: string_cat() and string_append() leave room, but do
3663 not put the zero in. */
3664
3665 s[sptr] = 0;
3666
3667 /* Create a message log file if message logs are being used and this message is
3668 not blackholed. Write the reception stuff to it. We used to leave message log
3669 creation until the first delivery, but this has proved confusing for some
3670 people. */
3671
3672 if (message_logs && blackholed_by == NULL)
3673 {
3674 int fd;
3675
3676 sprintf(CS spool_name, "%s/msglog/%s/%s", spool_directory, message_subdir,
3677 message_id);
3678 fd = Uopen(spool_name, O_WRONLY|O_APPEND|O_CREAT, SPOOL_MODE);
3679
3680 if (fd < 0 && errno == ENOENT)
3681 {
3682 uschar temp[16];
3683 sprintf(CS temp, "msglog/%s", message_subdir);
3684 if (message_subdir[0] == 0) temp[6] = 0;
3685 (void)directory_make(spool_directory, temp, MSGLOG_DIRECTORY_MODE, TRUE);
3686 fd = Uopen(spool_name, O_WRONLY|O_APPEND|O_CREAT, SPOOL_MODE);
3687 }
3688
3689 if (fd < 0)
3690 {
3691 log_write(0, LOG_MAIN|LOG_PANIC, "Couldn't open message log %s: %s",
3692 spool_name, strerror(errno));
3693 }
3694
3695 else
3696 {
3697 FILE *message_log = fdopen(fd, "a");
3698 if (message_log == NULL)
3699 {
3700 log_write(0, LOG_MAIN|LOG_PANIC, "Couldn't fdopen message log %s: %s",
3701 spool_name, strerror(errno));
3702 (void)close(fd);
3703 }
3704 else
3705 {
3706 uschar *now = tod_stamp(tod_log);
3707 fprintf(message_log, "%s Received from %s\n", now, s+3);
3708 if (deliver_freeze) fprintf(message_log, "%s frozen by %s\n", now,
3709 frozen_by);
3710 if (queue_only_policy) fprintf(message_log,
3711 "%s no immediate delivery: queued by %s\n", now, queued_by);
3712 (void)fclose(message_log);
3713 }
3714 }
3715 }
3716
3717 /* Everything has now been done for a successful message except logging its
3718 arrival, and outputting an SMTP response. While writing to the log, set a flag
3719 to cause a call to receive_bomb_out() if the log cannot be opened. */
3720
3721 receive_call_bombout = TRUE;
3722
3723 /* Before sending an SMTP response in a TCP/IP session, we check to see if the
3724 connection has gone away. This can only be done if there is no unconsumed input
3725 waiting in the local input buffer. We can test for this by calling
3726 receive_smtp_buffered(). RFC 2920 (pipelining) explicitly allows for additional
3727 input to be sent following the final dot, so the presence of following input is
3728 not an error.
3729
3730 If the connection is still present, but there is no unread input for the
3731 socket, the result of a select() call will be zero. If, however, the connection
3732 has gone away, or if there is pending input, the result of select() will be
3733 non-zero. The two cases can be distinguished by trying to read the next input
3734 character. If we succeed, we can unread it so that it remains in the local
3735 buffer for handling later. If not, the connection has been lost.
3736
3737 Of course, since TCP/IP is asynchronous, there is always a chance that the
3738 connection will vanish between the time of this test and the sending of the
3739 response, but the chance of this happening should be small. */
3740
3741 if (smtp_input && sender_host_address != NULL && !sender_host_notsocket &&
3742 !receive_smtp_buffered())
3743 {
3744 struct timeval tv;
3745 fd_set select_check;
3746 FD_ZERO(&select_check);
3747 FD_SET(fileno(smtp_in), &select_check);
3748 tv.tv_sec = 0;
3749 tv.tv_usec = 0;
3750
3751 if (select(fileno(smtp_in) + 1, &select_check, NULL, NULL, &tv) != 0)
3752 {
3753 int c = (receive_getc)();
3754 if (c != EOF) (receive_ungetc)(c); else
3755 {
3756 uschar *msg = US"SMTP connection lost after final dot";
3757 smtp_reply = US""; /* No attempt to send a response */
3758 smtp_yield = FALSE; /* Nothing more on this connection */
3759
3760 /* Re-use the log line workspace */
3761
3762 sptr = 0;
3763 s = string_cat(s, &size, &sptr, msg, Ustrlen(msg));
3764 s = add_host_info_for_log(s, &size, &sptr);
3765 s[sptr] = 0;
3766 log_write(0, LOG_MAIN, "%s", s);
3767
3768 /* Delete the files for this aborted message. */
3769
3770 sprintf(CS spool_name, "%s/input/%s/%s-D", spool_directory,
3771 message_subdir, message_id);
3772 Uunlink(spool_name);
3773
3774 sprintf(CS spool_name, "%s/input/%s/%s-H", spool_directory,
3775 message_subdir, message_id);
3776 Uunlink(spool_name);
3777
3778 sprintf(CS spool_name, "%s/msglog/%s/%s", spool_directory,
3779 message_subdir, message_id);
3780 Uunlink(spool_name);
3781
3782 goto TIDYUP;
3783 }
3784 }
3785 }
3786
3787 /* The connection has not gone away; we really are going to take responsibility
3788 for this message. */
3789
3790 /* Cutthrough - had sender last-dot; assume we've sent (or bufferred) all
3791 data onward by now.
3792
3793 Send dot onward. If accepted, wipe the spooled files, log as delivered and accept
3794 the sender's dot (below).
3795 If rejected: copy response to sender, wipe the spooled files, log approriately.
3796 If temp-reject: accept to sender, keep the spooled files.
3797
3798 Having the normal spool files lets us do data-filtering, and store/forward on temp-reject.
3799
3800 XXX We do not handle queue-only, freezing, or blackholes.
3801 */
3802 if(cutthrough_fd >= 0)
3803 {
3804 uschar * msg= cutthrough_finaldot(); /* Ask the target system to accept the messsage */
3805 /* Logging was done in finaldot() */
3806 switch(msg[0])
3807 {
3808 case '2': /* Accept. Do the same to the source; dump any spoolfiles. */
3809 cutthrough_done = 3;
3810 break; /* message_id needed for SMTP accept below */
3811
3812 default: /* Unknown response, or error. Treat as temp-reject. */
3813 case '4': /* Temp-reject. Keep spoolfiles and accept. */
3814 cutthrough_done = 1; /* Avoid the usual immediate delivery attempt */
3815 break; /* message_id needed for SMTP accept below */
3816
3817 case '5': /* Perm-reject. Do the same to the source. Dump any spoolfiles */
3818 smtp_reply= msg; /* Pass on the exact error */
3819 cutthrough_done = 2;
3820 break;
3821 }
3822 }
3823 else
3824 cutthrough_done = 0;
3825
3826 if(smtp_reply == NULL)
3827 {
3828 log_write(0, LOG_MAIN |
3829 (((log_extra_selector & LX_received_recipients) != 0)? LOG_RECIPIENTS : 0) |
3830 (((log_extra_selector & LX_received_sender) != 0)? LOG_SENDER : 0),
3831 "%s", s);
3832
3833 /* Log any control actions taken by an ACL or local_scan(). */
3834
3835 if (deliver_freeze) log_write(0, LOG_MAIN, "frozen by %s", frozen_by);
3836 if (queue_only_policy) log_write(L_delay_delivery, LOG_MAIN,
3837 "no immediate delivery: queued by %s", queued_by);
3838 }
3839 receive_call_bombout = FALSE;
3840
3841 store_reset(s); /* The store for the main log message can be reused */
3842
3843 /* If the message is frozen, and freeze_tell is set, do the telling. */
3844
3845 if (deliver_freeze && freeze_tell != NULL && freeze_tell[0] != 0)
3846 {
3847 moan_tell_someone(freeze_tell, NULL, US"Message frozen on arrival",
3848 "Message %s was frozen on arrival by %s.\nThe sender is <%s>.\n",
3849 message_id, frozen_by, sender_address);
3850 }
3851
3852
3853 /* Either a message has been successfully received and written to the two spool
3854 files, or an error in writing the spool has occurred for an SMTP message, or
3855 an SMTP message has been rejected for policy reasons. (For a non-SMTP message
3856 we will have already given up because there's no point in carrying on!) In
3857 either event, we must now close (and thereby unlock) the data file. In the
3858 successful case, this leaves the message on the spool, ready for delivery. In
3859 the error case, the spool file will be deleted. Then tidy up store, interact
3860 with an SMTP call if necessary, and return.
3861
3862 A fflush() was done earlier in the expectation that any write errors on the
3863 data file will be flushed(!) out thereby. Nevertheless, it is theoretically
3864 possible for fclose() to fail - but what to do? What has happened to the lock
3865 if this happens? */
3866
3867
3868 TIDYUP:
3869 process_info[process_info_len] = 0; /* Remove message id */
3870 if (data_file != NULL) (void)fclose(data_file); /* Frees the lock */
3871
3872 /* Now reset signal handlers to their defaults */
3873
3874 signal(SIGTERM, SIG_DFL);
3875 signal(SIGINT, SIG_DFL);
3876
3877 /* Tell an SMTP caller the state of play, and arrange to return the SMTP return
3878 value, which defaults TRUE - meaning there may be more incoming messages from
3879 this connection. For non-SMTP callers (where there is only ever one message),
3880 the default is FALSE. */
3881
3882 if (smtp_input)
3883 {
3884 yield = smtp_yield;
3885
3886 /* Handle interactive SMTP callers. After several kinds of error, smtp_reply
3887 is set to the response that should be sent. When it is NULL, we generate
3888 default responses. After an ACL error or local_scan() error, the response has
3889 already been sent, and smtp_reply is an empty string to indicate this. */
3890
3891 if (!smtp_batched_input)
3892 {
3893 if (smtp_reply == NULL)
3894 {
3895 if (fake_response != OK)
3896 smtp_respond((fake_response == DEFER)? US"450" : US"550", 3, TRUE,
3897 fake_response_text);
3898
3899 /* An OK response is required; use "message" text if present. */
3900
3901 else if (user_msg != NULL)
3902 {
3903 uschar *code = US"250";
3904 int len = 3;
3905 smtp_message_code(&code, &len, &user_msg, NULL);
3906 smtp_respond(code, len, TRUE, user_msg);
3907 }
3908
3909 /* Default OK response */
3910
3911 else
3912 smtp_printf("250 OK id=%s\r\n", message_id);
3913 if (host_checking)
3914 fprintf(stdout,
3915 "\n**** SMTP testing: that is not a real message id!\n\n");
3916 }
3917
3918 /* smtp_reply is set non-empty */
3919
3920 else if (smtp_reply[0] != 0)
3921 {
3922 if (fake_response != OK && (smtp_reply[0] == '2'))
3923 smtp_respond((fake_response == DEFER)? US"450" : US"550", 3, TRUE,
3924 fake_response_text);
3925 else
3926 smtp_printf("%.1024s\r\n", smtp_reply);
3927 }
3928
3929 switch (cutthrough_done)
3930 {
3931 case 3: log_write(0, LOG_MAIN, "Completed"); /* Delivery was done */
3932 case 2: { /* Delete spool files */
3933 sprintf(CS spool_name, "%s/input/%s/%s-D", spool_directory,
3934 message_subdir, message_id);
3935 Uunlink(spool_name);
3936 sprintf(CS spool_name, "%s/input/%s/%s-H", spool_directory,
3937 message_subdir, message_id);
3938 Uunlink(spool_name);
3939 sprintf(CS spool_name, "%s/msglog/%s/%s", spool_directory,
3940 message_subdir, message_id);
3941 Uunlink(spool_name);
3942 }
3943 case 1: message_id[0] = 0; /* Prevent a delivery from starting */
3944 default:break;
3945 }
3946 cutthrough_delivery = FALSE;
3947 }
3948
3949 /* For batched SMTP, generate an error message on failure, and do
3950 nothing on success. The function moan_smtp_batch() does not return -
3951 it exits from the program with a non-zero return code. */
3952
3953 else if (smtp_reply != NULL) moan_smtp_batch(NULL, "%s", smtp_reply);
3954 }
3955
3956
3957 /* If blackholing, we can immediately log this message's sad fate. The data
3958 file has already been unlinked, and the header file was never written to disk.
3959 We must now indicate that nothing was received, to prevent a delivery from
3960 starting. */
3961
3962 if (blackholed_by != NULL)
3963 {
3964 uschar *detail = (local_scan_data != NULL)?
3965 string_printing(local_scan_data) :
3966 string_sprintf("(%s discarded recipients)", blackholed_by);
3967 log_write(0, LOG_MAIN, "=> blackhole %s%s", detail, blackhole_log_msg);
3968 log_write(0, LOG_MAIN, "Completed");
3969 message_id[0] = 0;
3970 }
3971
3972 /* Reset headers so that logging of rejects for a subsequent message doesn't
3973 include them. It is also important to set header_last = NULL before exiting
3974 from this function, as this prevents certain rewrites that might happen during
3975 subsequent verifying (of another incoming message) from trying to add headers
3976 when they shouldn't. */
3977
3978 header_list = header_last = NULL;
3979
3980 return yield; /* TRUE if more messages (SMTP only) */
3981 }
3982
3983 /* End of receive.c */