Check syscall return values.
[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
940 if (acl_removed_headers != NULL)
941 {
942 DEBUG(D_receive|D_acl) debug_printf(">>Headers removed by %s ACL:\n", acl_name);
943
944 for (h = header_list; h != NULL; h = h->next)
945 {
946 uschar *list;
947 BOOL include_header;
948
949 if (h->type == htype_old) continue;
950
951 include_header = TRUE;
952 list = acl_removed_headers;
953
954 int sep = ':'; /* This is specified as a colon-separated list */
955 uschar *s;
956 uschar buffer[128];
957 while ((s = string_nextinlist(&list, &sep, buffer, sizeof(buffer)))
958 != NULL)
959 {
960 int len = Ustrlen(s);
961 if (header_testname(h, s, len, FALSE))
962 {
963 h->type = htype_old;
964 DEBUG(D_receive|D_acl) debug_printf(" %s", h->text);
965 }
966 }
967 }
968 acl_removed_headers = NULL;
969 DEBUG(D_receive|D_acl) debug_printf(">>\n");
970 }
971
972 if (acl_added_headers == NULL) return;
973 DEBUG(D_receive|D_acl) debug_printf(">>Headers added by %s ACL:\n", acl_name);
974
975 for (h = acl_added_headers; h != NULL; h = next)
976 {
977 next = h->next;
978
979 switch(h->type)
980 {
981 case htype_add_top:
982 h->next = header_list;
983 header_list = h;
984 DEBUG(D_receive|D_acl) debug_printf(" (at top)");
985 break;
986
987 case htype_add_rec:
988 if (last_received == NULL)
989 {
990 last_received = header_list;
991 while (!header_testname(last_received, US"Received", 8, FALSE))
992 last_received = last_received->next;
993 while (last_received->next != NULL &&
994 header_testname(last_received->next, US"Received", 8, FALSE))
995 last_received = last_received->next;
996 }
997 h->next = last_received->next;
998 last_received->next = h;
999 DEBUG(D_receive|D_acl) debug_printf(" (after Received:)");
1000 break;
1001
1002 case htype_add_rfc:
1003 /* add header before any header which is NOT Received: or Resent- */
1004 last_received = header_list;
1005 while ( (last_received->next != NULL) &&
1006 ( (header_testname(last_received->next, US"Received", 8, FALSE)) ||
1007 (header_testname_incomplete(last_received->next, US"Resent-", 7, FALSE)) ) )
1008 last_received = last_received->next;
1009 /* last_received now points to the last Received: or Resent-* header
1010 in an uninterrupted chain of those header types (seen from the beginning
1011 of all headers. Our current header must follow it. */
1012 h->next = last_received->next;
1013 last_received->next = h;
1014 DEBUG(D_receive|D_acl) debug_printf(" (before any non-Received: or Resent-*: header)");
1015 break;
1016
1017 default:
1018 h->next = NULL;
1019 header_last->next = h;
1020 break;
1021 }
1022
1023 if (h->next == NULL) header_last = h;
1024
1025 /* Check for one of the known header types (From:, To:, etc.) though in
1026 practice most added headers are going to be "other". Lower case
1027 identification letters are never stored with the header; they are used
1028 for existence tests when messages are received. So discard any lower case
1029 flag values. */
1030
1031 h->type = header_checkname(h, FALSE);
1032 if (h->type >= 'a') h->type = htype_other;
1033
1034 DEBUG(D_receive|D_acl) debug_printf(" %s", header_last->text);
1035 }
1036
1037 acl_added_headers = NULL;
1038 DEBUG(D_receive|D_acl) debug_printf(">>\n");
1039 }
1040
1041
1042
1043 /*************************************************
1044 * Add host information for log line *
1045 *************************************************/
1046
1047 /* Called for acceptance and rejecting log lines. This adds information about
1048 the calling host to a string that is being built dynamically.
1049
1050 Arguments:
1051 s the dynamic string
1052 sizeptr points to the size variable
1053 ptrptr points to the pointer variable
1054
1055 Returns: the extended string
1056 */
1057
1058 static uschar *
1059 add_host_info_for_log(uschar *s, int *sizeptr, int *ptrptr)
1060 {
1061 if (sender_fullhost != NULL)
1062 {
1063 s = string_append(s, sizeptr, ptrptr, 2, US" H=", sender_fullhost);
1064 if ((log_extra_selector & LX_incoming_interface) != 0 &&
1065 interface_address != NULL)
1066 {
1067 uschar *ss = string_sprintf(" I=[%s]:%d", interface_address,
1068 interface_port);
1069 s = string_cat(s, sizeptr, ptrptr, ss, Ustrlen(ss));
1070 }
1071 }
1072 if (sender_ident != NULL)
1073 s = string_append(s, sizeptr, ptrptr, 2, US" U=", sender_ident);
1074 if (received_protocol != NULL)
1075 s = string_append(s, sizeptr, ptrptr, 2, US" P=", received_protocol);
1076 return s;
1077 }
1078
1079
1080
1081 #ifdef WITH_CONTENT_SCAN
1082
1083 /*************************************************
1084 * Run the MIME ACL on a message *
1085 *************************************************/
1086
1087 /* This code is in a subroutine so that it can be used for both SMTP
1088 and non-SMTP messages. It is called with a non-NULL ACL pointer.
1089
1090 Arguments:
1091 acl The ACL to run (acl_smtp_mime or acl_not_smtp_mime)
1092 smtp_yield_ptr Set FALSE to kill messages after dropped connection
1093 smtp_reply_ptr Where SMTP reply is being built
1094 blackholed_by_ptr Where "blackholed by" message is being built
1095
1096 Returns: TRUE to carry on; FALSE to abandon the message
1097 */
1098
1099 static BOOL
1100 run_mime_acl(uschar *acl, BOOL *smtp_yield_ptr, uschar **smtp_reply_ptr,
1101 uschar **blackholed_by_ptr)
1102 {
1103 FILE *mbox_file;
1104 uschar rfc822_file_path[2048];
1105 unsigned long mbox_size;
1106 header_line *my_headerlist;
1107 uschar *user_msg, *log_msg;
1108 int mime_part_count_buffer = -1;
1109 int rc = OK;
1110
1111 memset(CS rfc822_file_path,0,2048);
1112
1113 /* check if it is a MIME message */
1114 my_headerlist = header_list;
1115 while (my_headerlist != NULL)
1116 {
1117 /* skip deleted headers */
1118 if (my_headerlist->type == '*')
1119 {
1120 my_headerlist = my_headerlist->next;
1121 continue;
1122 }
1123 if (strncmpic(my_headerlist->text, US"Content-Type:", 13) == 0)
1124 {
1125 DEBUG(D_receive) debug_printf("Found Content-Type: header - executing acl_smtp_mime.\n");
1126 goto DO_MIME_ACL;
1127 }
1128 my_headerlist = my_headerlist->next;
1129 }
1130
1131 DEBUG(D_receive) debug_printf("No Content-Type: header - presumably not a MIME message.\n");
1132 return TRUE;
1133
1134 DO_MIME_ACL:
1135 /* make sure the eml mbox file is spooled up */
1136 mbox_file = spool_mbox(&mbox_size, NULL);
1137 if (mbox_file == NULL) {
1138 /* error while spooling */
1139 log_write(0, LOG_MAIN|LOG_PANIC,
1140 "acl_smtp_mime: error while creating mbox spool file, message temporarily rejected.");
1141 Uunlink(spool_name);
1142 unspool_mbox();
1143 #ifdef EXPERIMENTAL_DCC
1144 dcc_ok = 0;
1145 #endif
1146 smtp_respond(US"451", 3, TRUE, US"temporary local problem");
1147 message_id[0] = 0; /* Indicate no message accepted */
1148 *smtp_reply_ptr = US""; /* Indicate reply already sent */
1149 return FALSE; /* Indicate skip to end of receive function */
1150 };
1151
1152 mime_is_rfc822 = 0;
1153
1154 MIME_ACL_CHECK:
1155 mime_part_count = -1;
1156 rc = mime_acl_check(acl, mbox_file, NULL, &user_msg, &log_msg);
1157 (void)fclose(mbox_file);
1158
1159 if (Ustrlen(rfc822_file_path) > 0)
1160 {
1161 mime_part_count = mime_part_count_buffer;
1162
1163 if (unlink(CS rfc822_file_path) == -1)
1164 {
1165 log_write(0, LOG_PANIC,
1166 "acl_smtp_mime: can't unlink RFC822 spool file, skipping.");
1167 goto END_MIME_ACL;
1168 }
1169 }
1170
1171 /* check if we must check any message/rfc822 attachments */
1172 if (rc == OK)
1173 {
1174 uschar temp_path[1024];
1175 int n;
1176 struct dirent *entry;
1177 DIR *tempdir;
1178
1179 (void)string_format(temp_path, 1024, "%s/scan/%s", spool_directory,
1180 message_id);
1181
1182 tempdir = opendir(CS temp_path);
1183 n = 0;
1184 do
1185 {
1186 entry = readdir(tempdir);
1187 if (entry == NULL) break;
1188 if (strncmpic(US entry->d_name,US"__rfc822_",9) == 0)
1189 {
1190 (void)string_format(rfc822_file_path, 2048,"%s/scan/%s/%s", spool_directory, message_id, entry->d_name);
1191 debug_printf("RFC822 attachment detected: running MIME ACL for '%s'\n", rfc822_file_path);
1192 break;
1193 }
1194 } while (1);
1195 closedir(tempdir);
1196
1197 if (entry != NULL)
1198 {
1199 mbox_file = Ufopen(rfc822_file_path,"rb");
1200 if (mbox_file == NULL)
1201 {
1202 log_write(0, LOG_PANIC,
1203 "acl_smtp_mime: can't open RFC822 spool file, skipping.");
1204 unlink(CS rfc822_file_path);
1205 goto END_MIME_ACL;
1206 }
1207 /* set RFC822 expansion variable */
1208 mime_is_rfc822 = 1;
1209 mime_part_count_buffer = mime_part_count;
1210 goto MIME_ACL_CHECK;
1211 }
1212 }
1213
1214 END_MIME_ACL:
1215 add_acl_headers(US"MIME");
1216 if (rc == DISCARD)
1217 {
1218 recipients_count = 0;
1219 *blackholed_by_ptr = US"MIME ACL";
1220 }
1221 else if (rc != OK)
1222 {
1223 Uunlink(spool_name);
1224 unspool_mbox();
1225 #ifdef EXPERIMENTAL_DCC
1226 dcc_ok = 0;
1227 #endif
1228 if (smtp_handle_acl_fail(ACL_WHERE_MIME, rc, user_msg, log_msg) != 0)
1229 *smtp_yield_ptr = FALSE; /* No more messsages after dropped connection */
1230 *smtp_reply_ptr = US""; /* Indicate reply already sent */
1231 message_id[0] = 0; /* Indicate no message accepted */
1232 return FALSE; /* Cause skip to end of receive function */
1233 }
1234
1235 return TRUE;
1236 }
1237
1238 #endif /* WITH_CONTENT_SCAN */
1239
1240
1241
1242 void
1243 received_header_gen(void)
1244 {
1245 uschar *received;
1246 uschar *timestamp;
1247 header_line *received_header= header_list;
1248
1249 timestamp = expand_string(US"${tod_full}");
1250 if (recipients_count == 1) received_for = recipients_list[0].address;
1251 received = expand_string(received_header_text);
1252 received_for = NULL;
1253
1254 if (received == NULL)
1255 {
1256 if(spool_name[0] != 0)
1257 Uunlink(spool_name); /* Lose the data file */
1258 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Expansion of \"%s\" "
1259 "(received_header_text) failed: %s", string_printing(received_header_text),
1260 expand_string_message);
1261 }
1262
1263 /* The first element on the header chain is reserved for the Received header,
1264 so all we have to do is fill in the text pointer, and set the type. However, if
1265 the result of the expansion is an empty string, we leave the header marked as
1266 "old" so as to refrain from adding a Received header. */
1267
1268 if (received[0] == 0)
1269 {
1270 received_header->text = string_sprintf("Received: ; %s\n", timestamp);
1271 received_header->type = htype_old;
1272 }
1273 else
1274 {
1275 received_header->text = string_sprintf("%s; %s\n", received, timestamp);
1276 received_header->type = htype_received;
1277 }
1278
1279 received_header->slen = Ustrlen(received_header->text);
1280
1281 DEBUG(D_receive) debug_printf(">>Generated Received: header line\n%c %s",
1282 received_header->type, received_header->text);
1283 }
1284
1285
1286
1287 /*************************************************
1288 * Receive message *
1289 *************************************************/
1290
1291 /* Receive a message on the given input, and put it into a pair of spool files.
1292 Either a non-null list of recipients, or the extract flag will be true, or
1293 both. The flag sender_local is true for locally generated messages. The flag
1294 submission_mode is true if an ACL has obeyed "control = submission". The flag
1295 suppress_local_fixups is true if an ACL has obeyed "control =
1296 suppress_local_fixups" or -G was passed on the command-line.
1297 The flag smtp_input is true if the message is to be
1298 handled using SMTP conventions about termination and lines starting with dots.
1299 For non-SMTP messages, dot_ends is true for dot-terminated messages.
1300
1301 If a message was successfully read, message_id[0] will be non-zero.
1302
1303 The general actions of this function are:
1304
1305 . Read the headers of the message (if any) into a chain of store
1306 blocks.
1307
1308 . If there is a "sender:" header and the message is locally originated,
1309 throw it away, unless the caller is trusted, or unless
1310 active_local_sender_retain is set - which can only happen if
1311 active_local_from_check is false.
1312
1313 . If recipients are to be extracted from the message, build the
1314 recipients list from the headers, removing any that were on the
1315 original recipients list (unless extract_addresses_remove_arguments is
1316 false), and at the same time, remove any bcc header that may be present.
1317
1318 . Get the spool file for the data, sort out its unique name, open
1319 and lock it (but don't give it the name yet).
1320
1321 . Generate a "Message-Id" header if the message doesn't have one, for
1322 locally-originated messages.
1323
1324 . Generate a "Received" header.
1325
1326 . Ensure the recipients list is fully qualified and rewritten if necessary.
1327
1328 . If there are any rewriting rules, apply them to the sender address
1329 and also to the headers.
1330
1331 . If there is no from: header, generate one, for locally-generated messages
1332 and messages in "submission mode" only.
1333
1334 . If the sender is local, check that from: is correct, and if not, generate
1335 a Sender: header, unless message comes from a trusted caller, or this
1336 feature is disabled by active_local_from_check being false.
1337
1338 . If there is no "date" header, generate one, for locally-originated
1339 or submission mode messages only.
1340
1341 . Copy the rest of the input, or up to a terminating "." if in SMTP or
1342 dot_ends mode, to the data file. Leave it open, to hold the lock.
1343
1344 . Write the envelope and the headers to a new file.
1345
1346 . Set the name for the header file; close it.
1347
1348 . Set the name for the data file; close it.
1349
1350 Because this function can potentially be called many times in a single
1351 SMTP connection, all store should be got by store_get(), so that it will be
1352 automatically retrieved after the message is accepted.
1353
1354 FUDGE: It seems that sites on the net send out messages with just LF
1355 terminators, despite the warnings in the RFCs, and other MTAs handle this. So
1356 we make the CRs optional in all cases.
1357
1358 July 2003: Bare CRs in messages, especially in header lines, cause trouble. A
1359 new regime is now in place in which bare CRs in header lines are turned into LF
1360 followed by a space, so as not to terminate the header line.
1361
1362 February 2004: A bare LF in a header line in a message whose first line was
1363 terminated by CRLF is treated in the same way as a bare CR.
1364
1365 Arguments:
1366 extract_recip TRUE if recipients are to be extracted from the message's
1367 headers
1368
1369 Returns: TRUE there are more messages to be read (SMTP input)
1370 FALSE there are no more messages to be read (non-SMTP input
1371 or SMTP connection collapsed, or other failure)
1372
1373 When reading a message for filter testing, the returned value indicates
1374 whether the headers (which is all that is read) were terminated by '.' or
1375 not. */
1376
1377 BOOL
1378 receive_msg(BOOL extract_recip)
1379 {
1380 int i;
1381 int rc = FAIL;
1382 int msg_size = 0;
1383 int process_info_len = Ustrlen(process_info);
1384 int error_rc = (error_handling == ERRORS_SENDER)?
1385 errors_sender_rc : EXIT_FAILURE;
1386 int header_size = 256;
1387 int start, end, domain, size, sptr;
1388 int id_resolution;
1389 int had_zero = 0;
1390 int prevlines_length = 0;
1391
1392 register int ptr = 0;
1393
1394 BOOL contains_resent_headers = FALSE;
1395 BOOL extracted_ignored = FALSE;
1396 BOOL first_line_ended_crlf = TRUE_UNSET;
1397 BOOL smtp_yield = TRUE;
1398 BOOL yield = FALSE;
1399
1400 BOOL resents_exist = FALSE;
1401 uschar *resent_prefix = US"";
1402 uschar *blackholed_by = NULL;
1403 uschar *blackhole_log_msg = US"";
1404 int cutthrough_done = 0;
1405
1406 flock_t lock_data;
1407 error_block *bad_addresses = NULL;
1408
1409 uschar *frozen_by = NULL;
1410 uschar *queued_by = NULL;
1411
1412 uschar *errmsg, *s;
1413 struct stat statbuf;
1414
1415 /* Final message to give to SMTP caller, and messages from ACLs */
1416
1417 uschar *smtp_reply = NULL;
1418 uschar *user_msg, *log_msg;
1419
1420 /* Working header pointers */
1421
1422 header_line *h, *next;
1423
1424 /* Flags for noting the existence of certain headers (only one left) */
1425
1426 BOOL date_header_exists = FALSE;
1427
1428 /* Pointers to receive the addresses of headers whose contents we need. */
1429
1430 header_line *from_header = NULL;
1431 header_line *subject_header = NULL;
1432 header_line *msgid_header = NULL;
1433 header_line *received_header;
1434
1435 /* Variables for use when building the Received: header. */
1436
1437 uschar *timestamp;
1438 int tslen;
1439
1440 /* Release any open files that might have been cached while preparing to
1441 accept the message - e.g. by verifying addresses - because reading a message
1442 might take a fair bit of real time. */
1443
1444 search_tidyup();
1445
1446 /* Extracting the recipient list from an input file is incompatible with
1447 cutthrough delivery with the no-spool option. It shouldn't be possible
1448 to set up the combination, but just in case kill any ongoing connection. */
1449 if (extract_recip || !smtp_input)
1450 cancel_cutthrough_connection("not smtp input");
1451
1452 /* Initialize the chain of headers by setting up a place-holder for Received:
1453 header. Temporarily mark it as "old", i.e. not to be used. We keep header_last
1454 pointing to the end of the chain to make adding headers simple. */
1455
1456 received_header = header_list = header_last = store_get(sizeof(header_line));
1457 header_list->next = NULL;
1458 header_list->type = htype_old;
1459 header_list->text = NULL;
1460 header_list->slen = 0;
1461
1462 /* Control block for the next header to be read. */
1463
1464 next = store_get(sizeof(header_line));
1465 next->text = store_get(header_size);
1466
1467 /* Initialize message id to be null (indicating no message read), and the
1468 header names list to be the normal list. Indicate there is no data file open
1469 yet, initialize the size and warning count, and deal with no size limit. */
1470
1471 message_id[0] = 0;
1472 data_file = NULL;
1473 data_fd = -1;
1474 spool_name[0] = 0;
1475 message_size = 0;
1476 warning_count = 0;
1477 received_count = 1; /* For the one we will add */
1478
1479 if (thismessage_size_limit <= 0) thismessage_size_limit = INT_MAX;
1480
1481 /* While reading the message, the following counts are computed. */
1482
1483 message_linecount = body_linecount = body_zerocount =
1484 max_received_linelength = 0;
1485
1486 #ifndef DISABLE_DKIM
1487 /* Call into DKIM to set up the context. */
1488 if (smtp_input && !smtp_batched_input && !dkim_disable_verify) dkim_exim_verify_init();
1489 #endif
1490
1491 /* Remember the time of reception. Exim uses time+pid for uniqueness of message
1492 ids, and fractions of a second are required. See the comments that precede the
1493 message id creation below. */
1494
1495 (void)gettimeofday(&message_id_tv, NULL);
1496
1497 /* For other uses of the received time we can operate with granularity of one
1498 second, and for that we use the global variable received_time. This is for
1499 things like ultimate message timeouts. */
1500
1501 received_time = message_id_tv.tv_sec;
1502
1503 /* If SMTP input, set the special handler for timeouts. The alarm() calls
1504 happen in the smtp_getc() function when it refills its buffer. */
1505
1506 if (smtp_input) os_non_restarting_signal(SIGALRM, data_timeout_handler);
1507
1508 /* If not SMTP input, timeout happens only if configured, and we just set a
1509 single timeout for the whole message. */
1510
1511 else if (receive_timeout > 0)
1512 {
1513 os_non_restarting_signal(SIGALRM, data_timeout_handler);
1514 alarm(receive_timeout);
1515 }
1516
1517 /* SIGTERM and SIGINT are caught always. */
1518
1519 signal(SIGTERM, data_sigterm_sigint_handler);
1520 signal(SIGINT, data_sigterm_sigint_handler);
1521
1522 /* Header lines in messages are not supposed to be very long, though when
1523 unfolded, to: and cc: headers can take up a lot of store. We must also cope
1524 with the possibility of junk being thrown at us. Start by getting 256 bytes for
1525 storing the header, and extend this as necessary using string_cat().
1526
1527 To cope with total lunacies, impose an upper limit on the length of the header
1528 section of the message, as otherwise the store will fill up. We must also cope
1529 with the possibility of binary zeros in the data. Hence we cannot use fgets().
1530 Folded header lines are joined into one string, leaving the '\n' characters
1531 inside them, so that writing them out reproduces the input.
1532
1533 Loop for each character of each header; the next structure for chaining the
1534 header is set up already, with ptr the offset of the next character in
1535 next->text. */
1536
1537 for (;;)
1538 {
1539 int ch = (receive_getc)();
1540
1541 /* If we hit EOF on a SMTP connection, it's an error, since incoming
1542 SMTP must have a correct "." terminator. */
1543
1544 if (ch == EOF && smtp_input /* && !smtp_batched_input */)
1545 {
1546 smtp_reply = handle_lost_connection(US" (header)");
1547 smtp_yield = FALSE;
1548 goto TIDYUP; /* Skip to end of function */
1549 }
1550
1551 /* See if we are at the current header's size limit - there must be at least
1552 four bytes left. This allows for the new character plus a zero, plus two for
1553 extra insertions when we are playing games with dots and carriage returns. If
1554 we are at the limit, extend the text buffer. This could have been done
1555 automatically using string_cat() but because this is a tightish loop storing
1556 only one character at a time, we choose to do it inline. Normally
1557 store_extend() will be able to extend the block; only at the end of a big
1558 store block will a copy be needed. To handle the case of very long headers
1559 (and sometimes lunatic messages can have ones that are 100s of K long) we
1560 call store_release() for strings that have been copied - if the string is at
1561 the start of a block (and therefore the only thing in it, because we aren't
1562 doing any other gets), the block gets freed. We can only do this because we
1563 know there are no other calls to store_get() going on. */
1564
1565 if (ptr >= header_size - 4)
1566 {
1567 int oldsize = header_size;
1568 /* header_size += 256; */
1569 header_size *= 2;
1570 if (!store_extend(next->text, oldsize, header_size))
1571 {
1572 uschar *newtext = store_get(header_size);
1573 memcpy(newtext, next->text, ptr);
1574 store_release(next->text);
1575 next->text = newtext;
1576 }
1577 }
1578
1579 /* Cope with receiving a binary zero. There is dispute about whether
1580 these should be allowed in RFC 822 messages. The middle view is that they
1581 should not be allowed in headers, at least. Exim takes this attitude at
1582 the moment. We can't just stomp on them here, because we don't know that
1583 this line is a header yet. Set a flag to cause scanning later. */
1584
1585 if (ch == 0) had_zero++;
1586
1587 /* Test for termination. Lines in remote SMTP are terminated by CRLF, while
1588 those from data files use just LF. Treat LF in local SMTP input as a
1589 terminator too. Treat EOF as a line terminator always. */
1590
1591 if (ch == EOF) goto EOL;
1592
1593 /* FUDGE: There are sites out there that don't send CRs before their LFs, and
1594 other MTAs accept this. We are therefore forced into this "liberalisation"
1595 too, so we accept LF as a line terminator whatever the source of the message.
1596 However, if the first line of the message ended with a CRLF, we treat a bare
1597 LF specially by inserting a white space after it to ensure that the header
1598 line is not terminated. */
1599
1600 if (ch == '\n')
1601 {
1602 if (first_line_ended_crlf == TRUE_UNSET) first_line_ended_crlf = FALSE;
1603 else if (first_line_ended_crlf) receive_ungetc(' ');
1604 goto EOL;
1605 }
1606
1607 /* This is not the end of the line. If this is SMTP input and this is
1608 the first character in the line and it is a "." character, ignore it.
1609 This implements the dot-doubling rule, though header lines starting with
1610 dots aren't exactly common. They are legal in RFC 822, though. If the
1611 following is CRLF or LF, this is the line that that terminates the
1612 entire message. We set message_ended to indicate this has happened (to
1613 prevent further reading), and break out of the loop, having freed the
1614 empty header, and set next = NULL to indicate no data line. */
1615
1616 if (ptr == 0 && ch == '.' && (smtp_input || dot_ends))
1617 {
1618 ch = (receive_getc)();
1619 if (ch == '\r')
1620 {
1621 ch = (receive_getc)();
1622 if (ch != '\n')
1623 {
1624 receive_ungetc(ch);
1625 ch = '\r'; /* Revert to CR */
1626 }
1627 }
1628 if (ch == '\n')
1629 {
1630 message_ended = END_DOT;
1631 store_reset(next);
1632 next = NULL;
1633 break; /* End character-reading loop */
1634 }
1635
1636 /* For non-SMTP input, the dot at the start of the line was really a data
1637 character. What is now in ch is the following character. We guaranteed
1638 enough space for this above. */
1639
1640 if (!smtp_input)
1641 {
1642 next->text[ptr++] = '.';
1643 message_size++;
1644 }
1645 }
1646
1647 /* If CR is immediately followed by LF, end the line, ignoring the CR, and
1648 remember this case if this is the first line ending. */
1649
1650 if (ch == '\r')
1651 {
1652 ch = (receive_getc)();
1653 if (ch == '\n')
1654 {
1655 if (first_line_ended_crlf == TRUE_UNSET) first_line_ended_crlf = TRUE;
1656 goto EOL;
1657 }
1658
1659 /* Otherwise, put back the character after CR, and turn the bare CR
1660 into LF SP. */
1661
1662 ch = (receive_ungetc)(ch);
1663 next->text[ptr++] = '\n';
1664 message_size++;
1665 ch = ' ';
1666 }
1667
1668 /* We have a data character for the header line. */
1669
1670 next->text[ptr++] = ch; /* Add to buffer */
1671 message_size++; /* Total message size so far */
1672
1673 /* Handle failure due to a humungously long header section. The >= allows
1674 for the terminating \n. Add what we have so far onto the headers list so
1675 that it gets reflected in any error message, and back up the just-read
1676 character. */
1677
1678 if (message_size >= header_maxsize)
1679 {
1680 next->text[ptr] = 0;
1681 next->slen = ptr;
1682 next->type = htype_other;
1683 next->next = NULL;
1684 header_last->next = next;
1685 header_last = next;
1686
1687 log_write(0, LOG_MAIN, "ridiculously long message header received from "
1688 "%s (more than %d characters): message abandoned",
1689 sender_host_unknown? sender_ident : sender_fullhost, header_maxsize);
1690
1691 if (smtp_input)
1692 {
1693 smtp_reply = US"552 Message header is ridiculously long";
1694 receive_swallow_smtp();
1695 goto TIDYUP; /* Skip to end of function */
1696 }
1697
1698 else
1699 {
1700 give_local_error(ERRMESS_VLONGHEADER,
1701 string_sprintf("message header longer than %d characters received: "
1702 "message not accepted", header_maxsize), US"", error_rc, stdin,
1703 header_list->next);
1704 /* Does not return */
1705 }
1706 }
1707
1708 continue; /* With next input character */
1709
1710 /* End of header line reached */
1711
1712 EOL:
1713
1714 /* Keep track of lines for BSMTP errors and overall message_linecount. */
1715
1716 receive_linecount++;
1717 message_linecount++;
1718
1719 /* Keep track of maximum line length */
1720
1721 if (ptr - prevlines_length > max_received_linelength)
1722 max_received_linelength = ptr - prevlines_length;
1723 prevlines_length = ptr + 1;
1724
1725 /* Now put in the terminating newline. There is always space for
1726 at least two more characters. */
1727
1728 next->text[ptr++] = '\n';
1729 message_size++;
1730
1731 /* A blank line signals the end of the headers; release the unwanted
1732 space and set next to NULL to indicate this. */
1733
1734 if (ptr == 1)
1735 {
1736 store_reset(next);
1737 next = NULL;
1738 break;
1739 }
1740
1741 /* There is data in the line; see if the next input character is a
1742 whitespace character. If it is, we have a continuation of this header line.
1743 There is always space for at least one character at this point. */
1744
1745 if (ch != EOF)
1746 {
1747 int nextch = (receive_getc)();
1748 if (nextch == ' ' || nextch == '\t')
1749 {
1750 next->text[ptr++] = nextch;
1751 message_size++;
1752 continue; /* Iterate the loop */
1753 }
1754 else if (nextch != EOF) (receive_ungetc)(nextch); /* For next time */
1755 else ch = EOF; /* Cause main loop to exit at end */
1756 }
1757
1758 /* We have got to the real line end. Terminate the string and release store
1759 beyond it. If it turns out to be a real header, internal binary zeros will
1760 be squashed later. */
1761
1762 next->text[ptr] = 0;
1763 next->slen = ptr;
1764 store_reset(next->text + ptr + 1);
1765
1766 /* Check the running total size against the overall message size limit. We
1767 don't expect to fail here, but if the overall limit is set less than MESSAGE_
1768 MAXSIZE and a big header is sent, we want to catch it. Just stop reading
1769 headers - the code to read the body will then also hit the buffer. */
1770
1771 if (message_size > thismessage_size_limit) break;
1772
1773 /* A line that is not syntactically correct for a header also marks
1774 the end of the headers. In this case, we leave next containing the
1775 first data line. This might actually be several lines because of the
1776 continuation logic applied above, but that doesn't matter.
1777
1778 It turns out that smail, and presumably sendmail, accept leading lines
1779 of the form
1780
1781 From ph10 Fri Jan 5 12:35 GMT 1996
1782
1783 in messages. The "mail" command on Solaris 2 sends such lines. I cannot
1784 find any documentation of this, but for compatibility it had better be
1785 accepted. Exim restricts it to the case of non-smtp messages, and
1786 treats it as an alternative to the -f command line option. Thus it is
1787 ignored except for trusted users or filter testing. Otherwise it is taken
1788 as the sender address, unless -f was used (sendmail compatibility).
1789
1790 It further turns out that some UUCPs generate the From_line in a different
1791 format, e.g.
1792
1793 From ph10 Fri, 7 Jan 97 14:00:00 GMT
1794
1795 The regex for matching these things is now capable of recognizing both
1796 formats (including 2- and 4-digit years in the latter). In fact, the regex
1797 is now configurable, as is the expansion string to fish out the sender.
1798
1799 Even further on it has been discovered that some broken clients send
1800 these lines in SMTP messages. There is now an option to ignore them from
1801 specified hosts or networks. Sigh. */
1802
1803 if (header_last == header_list &&
1804 (!smtp_input
1805 ||
1806 (sender_host_address != NULL &&
1807 verify_check_host(&ignore_fromline_hosts) == OK)
1808 ||
1809 (sender_host_address == NULL && ignore_fromline_local)
1810 ) &&
1811 regex_match_and_setup(regex_From, next->text, 0, -1))
1812 {
1813 if (!sender_address_forced)
1814 {
1815 uschar *uucp_sender = expand_string(uucp_from_sender);
1816 if (uucp_sender == NULL)
1817 {
1818 log_write(0, LOG_MAIN|LOG_PANIC,
1819 "expansion of \"%s\" failed after matching "
1820 "\"From \" line: %s", uucp_from_sender, expand_string_message);
1821 }
1822 else
1823 {
1824 int start, end, domain;
1825 uschar *errmess;
1826 uschar *newsender = parse_extract_address(uucp_sender, &errmess,
1827 &start, &end, &domain, TRUE);
1828 if (newsender != NULL)
1829 {
1830 if (domain == 0 && newsender[0] != 0)
1831 newsender = rewrite_address_qualify(newsender, FALSE);
1832
1833 if (filter_test != FTEST_NONE || receive_check_set_sender(newsender))
1834 {
1835 sender_address = newsender;
1836
1837 if (trusted_caller || filter_test != FTEST_NONE)
1838 {
1839 authenticated_sender = NULL;
1840 originator_name = US"";
1841 sender_local = FALSE;
1842 }
1843
1844 if (filter_test != FTEST_NONE)
1845 printf("Sender taken from \"From \" line\n");
1846 }
1847 }
1848 }
1849 }
1850 }
1851
1852 /* Not a leading "From " line. Check to see if it is a valid header line.
1853 Header names may contain any non-control characters except space and colon,
1854 amazingly. */
1855
1856 else
1857 {
1858 uschar *p = next->text;
1859
1860 /* If not a valid header line, break from the header reading loop, leaving
1861 next != NULL, indicating that it holds the first line of the body. */
1862
1863 if (isspace(*p)) break;
1864 while (mac_isgraph(*p) && *p != ':') p++;
1865 while (isspace(*p)) p++;
1866 if (*p != ':')
1867 {
1868 body_zerocount = had_zero;
1869 break;
1870 }
1871
1872 /* We have a valid header line. If there were any binary zeroes in
1873 the line, stomp on them here. */
1874
1875 if (had_zero > 0)
1876 for (p = next->text; p < next->text + ptr; p++) if (*p == 0) *p = '?';
1877
1878 /* It is perfectly legal to have an empty continuation line
1879 at the end of a header, but it is confusing to humans
1880 looking at such messages, since it looks like a blank line.
1881 Reduce confusion by removing redundant white space at the
1882 end. We know that there is at least one printing character
1883 (the ':' tested for above) so there is no danger of running
1884 off the end. */
1885
1886 p = next->text + ptr - 2;
1887 for (;;)
1888 {
1889 while (*p == ' ' || *p == '\t') p--;
1890 if (*p != '\n') break;
1891 ptr = (p--) - next->text + 1;
1892 message_size -= next->slen - ptr;
1893 next->text[ptr] = 0;
1894 next->slen = ptr;
1895 }
1896
1897 /* Add the header to the chain */
1898
1899 next->type = htype_other;
1900 next->next = NULL;
1901 header_last->next = next;
1902 header_last = next;
1903
1904 /* Check the limit for individual line lengths. This comes after adding to
1905 the chain so that the failing line is reflected if a bounce is generated
1906 (for a local message). */
1907
1908 if (header_line_maxsize > 0 && next->slen > header_line_maxsize)
1909 {
1910 log_write(0, LOG_MAIN, "overlong message header line received from "
1911 "%s (more than %d characters): message abandoned",
1912 sender_host_unknown? sender_ident : sender_fullhost,
1913 header_line_maxsize);
1914
1915 if (smtp_input)
1916 {
1917 smtp_reply = US"552 A message header line is too long";
1918 receive_swallow_smtp();
1919 goto TIDYUP; /* Skip to end of function */
1920 }
1921
1922 else
1923 {
1924 give_local_error(ERRMESS_VLONGHDRLINE,
1925 string_sprintf("message header line longer than %d characters "
1926 "received: message not accepted", header_line_maxsize), US"",
1927 error_rc, stdin, header_list->next);
1928 /* Does not return */
1929 }
1930 }
1931
1932 /* Note if any resent- fields exist. */
1933
1934 if (!resents_exist && strncmpic(next->text, US"resent-", 7) == 0)
1935 {
1936 resents_exist = TRUE;
1937 resent_prefix = US"Resent-";
1938 }
1939 }
1940
1941 /* The line has been handled. If we have hit EOF, break out of the loop,
1942 indicating no pending data line. */
1943
1944 if (ch == EOF) { next = NULL; break; }
1945
1946 /* Set up for the next header */
1947
1948 header_size = 256;
1949 next = store_get(sizeof(header_line));
1950 next->text = store_get(header_size);
1951 ptr = 0;
1952 had_zero = 0;
1953 prevlines_length = 0;
1954 } /* Continue, starting to read the next header */
1955
1956 /* At this point, we have read all the headers into a data structure in main
1957 store. The first header is still the dummy placeholder for the Received: header
1958 we are going to generate a bit later on. If next != NULL, it contains the first
1959 data line - which terminated the headers before reaching a blank line (not the
1960 normal case). */
1961
1962 DEBUG(D_receive)
1963 {
1964 debug_printf(">>Headers received:\n");
1965 for (h = header_list->next; h != NULL; h = h->next)
1966 debug_printf("%s", h->text);
1967 debug_printf("\n");
1968 }
1969
1970 /* End of file on any SMTP connection is an error. If an incoming SMTP call
1971 is dropped immediately after valid headers, the next thing we will see is EOF.
1972 We must test for this specially, as further down the reading of the data is
1973 skipped if already at EOF. */
1974
1975 if (smtp_input && (receive_feof)())
1976 {
1977 smtp_reply = handle_lost_connection(US" (after header)");
1978 smtp_yield = FALSE;
1979 goto TIDYUP; /* Skip to end of function */
1980 }
1981
1982 /* If this is a filter test run and no headers were read, output a warning
1983 in case there is a mistake in the test message. */
1984
1985 if (filter_test != FTEST_NONE && header_list->next == NULL)
1986 printf("Warning: no message headers read\n");
1987
1988
1989 /* Scan the headers to identify them. Some are merely marked for later
1990 processing; some are dealt with here. */
1991
1992 for (h = header_list->next; h != NULL; h = h->next)
1993 {
1994 BOOL is_resent = strncmpic(h->text, US"resent-", 7) == 0;
1995 if (is_resent) contains_resent_headers = TRUE;
1996
1997 switch (header_checkname(h, is_resent))
1998 {
1999 case htype_bcc:
2000 h->type = htype_bcc; /* Both Bcc: and Resent-Bcc: */
2001 break;
2002
2003 case htype_cc:
2004 h->type = htype_cc; /* Both Cc: and Resent-Cc: */
2005 break;
2006
2007 /* Record whether a Date: or Resent-Date: header exists, as appropriate. */
2008
2009 case htype_date:
2010 if (!resents_exist || is_resent) date_header_exists = TRUE;
2011 break;
2012
2013 /* Same comments as about Return-Path: below. */
2014
2015 case htype_delivery_date:
2016 if (delivery_date_remove) h->type = htype_old;
2017 break;
2018
2019 /* Same comments as about Return-Path: below. */
2020
2021 case htype_envelope_to:
2022 if (envelope_to_remove) h->type = htype_old;
2023 break;
2024
2025 /* Mark all "From:" headers so they get rewritten. Save the one that is to
2026 be used for Sender: checking. For Sendmail compatibility, if the "From:"
2027 header consists of just the login id of the user who called Exim, rewrite
2028 it with the gecos field first. Apply this rule to Resent-From: if there
2029 are resent- fields. */
2030
2031 case htype_from:
2032 h->type = htype_from;
2033 if (!resents_exist || is_resent)
2034 {
2035 from_header = h;
2036 if (!smtp_input)
2037 {
2038 uschar *s = Ustrchr(h->text, ':') + 1;
2039 while (isspace(*s)) s++;
2040 if (strncmpic(s, originator_login, h->slen - (s - h->text) - 1) == 0)
2041 {
2042 uschar *name = is_resent? US"Resent-From" : US"From";
2043 header_add(htype_from, "%s: %s <%s@%s>\n", name, originator_name,
2044 originator_login, qualify_domain_sender);
2045 from_header = header_last;
2046 h->type = htype_old;
2047 DEBUG(D_receive|D_rewrite)
2048 debug_printf("rewrote \"%s:\" header using gecos\n", name);
2049 }
2050 }
2051 }
2052 break;
2053
2054 /* Identify the Message-id: header for generating "in-reply-to" in the
2055 autoreply transport. For incoming logging, save any resent- value. In both
2056 cases, take just the first of any multiples. */
2057
2058 case htype_id:
2059 if (msgid_header == NULL && (!resents_exist || is_resent))
2060 {
2061 msgid_header = h;
2062 h->type = htype_id;
2063 }
2064 break;
2065
2066 /* Flag all Received: headers */
2067
2068 case htype_received:
2069 h->type = htype_received;
2070 received_count++;
2071 break;
2072
2073 /* "Reply-to:" is just noted (there is no resent-reply-to field) */
2074
2075 case htype_reply_to:
2076 h->type = htype_reply_to;
2077 break;
2078
2079 /* The Return-path: header is supposed to be added to messages when
2080 they leave the SMTP system. We shouldn't receive messages that already
2081 contain Return-path. However, since Exim generates Return-path: on
2082 local delivery, resent messages may well contain it. We therefore
2083 provide an option (which defaults on) to remove any Return-path: headers
2084 on input. Removal actually means flagging as "old", which prevents the
2085 header being transmitted with the message. */
2086
2087 case htype_return_path:
2088 if (return_path_remove) h->type = htype_old;
2089
2090 /* If we are testing a mail filter file, use the value of the
2091 Return-Path: header to set up the return_path variable, which is not
2092 otherwise set. However, remove any <> that surround the address
2093 because the variable doesn't have these. */
2094
2095 if (filter_test != FTEST_NONE)
2096 {
2097 uschar *start = h->text + 12;
2098 uschar *end = start + Ustrlen(start);
2099 while (isspace(*start)) start++;
2100 while (end > start && isspace(end[-1])) end--;
2101 if (*start == '<' && end[-1] == '>')
2102 {
2103 start++;
2104 end--;
2105 }
2106 return_path = string_copyn(start, end - start);
2107 printf("Return-path taken from \"Return-path:\" header line\n");
2108 }
2109 break;
2110
2111 /* If there is a "Sender:" header and the message is locally originated,
2112 and from an untrusted caller and suppress_local_fixups is not set, or if we
2113 are in submission mode for a remote message, mark it "old" so that it will
2114 not be transmitted with the message, unless active_local_sender_retain is
2115 set. (This can only be true if active_local_from_check is false.) If there
2116 are any resent- headers in the message, apply this rule to Resent-Sender:
2117 instead of Sender:. Messages with multiple resent- header sets cannot be
2118 tidily handled. (For this reason, at least one MUA - Pine - turns old
2119 resent- headers into X-resent- headers when resending, leaving just one
2120 set.) */
2121
2122 case htype_sender:
2123 h->type = ((!active_local_sender_retain &&
2124 (
2125 (sender_local && !trusted_caller && !suppress_local_fixups)
2126 || submission_mode
2127 )
2128 ) &&
2129 (!resents_exist||is_resent))?
2130 htype_old : htype_sender;
2131 break;
2132
2133 /* Remember the Subject: header for logging. There is no Resent-Subject */
2134
2135 case htype_subject:
2136 subject_header = h;
2137 break;
2138
2139 /* "To:" gets flagged, and the existence of a recipient header is noted,
2140 whether it's resent- or not. */
2141
2142 case htype_to:
2143 h->type = htype_to;
2144 /****
2145 to_or_cc_header_exists = TRUE;
2146 ****/
2147 break;
2148 }
2149 }
2150
2151 /* Extract recipients from the headers if that is required (the -t option).
2152 Note that this is documented as being done *before* any address rewriting takes
2153 place. There are two possibilities:
2154
2155 (1) According to sendmail documentation for Solaris, IRIX, and HP-UX, any
2156 recipients already listed are to be REMOVED from the message. Smail 3 works
2157 like this. We need to build a non-recipients tree for that list, because in
2158 subsequent processing this data is held in a tree and that's what the
2159 spool_write_header() function expects. Make sure that non-recipient addresses
2160 are fully qualified and rewritten if necessary.
2161
2162 (2) According to other sendmail documentation, -t ADDS extracted recipients to
2163 those in the command line arguments (and it is rumoured some other MTAs do
2164 this). Therefore, there is an option to make Exim behave this way.
2165
2166 *** Notes on "Resent-" header lines ***
2167
2168 The presence of resent-headers in the message makes -t horribly ambiguous.
2169 Experiments with sendmail showed that it uses recipients for all resent-
2170 headers, totally ignoring the concept of "sets of resent- headers" as described
2171 in RFC 2822 section 3.6.6. Sendmail also amalgamates them into a single set
2172 with all the addresses in one instance of each header.
2173
2174 This seems to me not to be at all sensible. Before release 4.20, Exim 4 gave an
2175 error for -t if there were resent- headers in the message. However, after a
2176 discussion on the mailing list, I've learned that there are MUAs that use
2177 resent- headers with -t, and also that the stuff about sets of resent- headers
2178 and their ordering in RFC 2822 is generally ignored. An MUA that submits a
2179 message with -t and resent- header lines makes sure that only *its* resent-
2180 headers are present; previous ones are often renamed as X-resent- for example.
2181
2182 Consequently, Exim has been changed so that, if any resent- header lines are
2183 present, the recipients are taken from all of the appropriate resent- lines,
2184 and not from the ordinary To:, Cc:, etc. */
2185
2186 if (extract_recip)
2187 {
2188 int rcount = 0;
2189 error_block **bnext = &bad_addresses;
2190
2191 if (extract_addresses_remove_arguments)
2192 {
2193 while (recipients_count-- > 0)
2194 {
2195 uschar *s = rewrite_address(recipients_list[recipients_count].address,
2196 TRUE, TRUE, global_rewrite_rules, rewrite_existflags);
2197 tree_add_nonrecipient(s);
2198 }
2199 recipients_list = NULL;
2200 recipients_count = recipients_list_max = 0;
2201 }
2202
2203 /* Now scan the headers */
2204
2205 for (h = header_list->next; h != NULL; h = h->next)
2206 {
2207 if ((h->type == htype_to || h->type == htype_cc || h->type == htype_bcc) &&
2208 (!contains_resent_headers || strncmpic(h->text, US"resent-", 7) == 0))
2209 {
2210 uschar *s = Ustrchr(h->text, ':') + 1;
2211 while (isspace(*s)) s++;
2212
2213 parse_allow_group = TRUE; /* Allow address group syntax */
2214
2215 while (*s != 0)
2216 {
2217 uschar *ss = parse_find_address_end(s, FALSE);
2218 uschar *recipient, *errmess, *p, *pp;
2219 int start, end, domain;
2220
2221 /* Check on maximum */
2222
2223 if (recipients_max > 0 && ++rcount > recipients_max)
2224 {
2225 give_local_error(ERRMESS_TOOMANYRECIP, US"too many recipients",
2226 US"message rejected: ", error_rc, stdin, NULL);
2227 /* Does not return */
2228 }
2229
2230 /* Make a copy of the address, and remove any internal newlines. These
2231 may be present as a result of continuations of the header line. The
2232 white space that follows the newline must not be removed - it is part
2233 of the header. */
2234
2235 pp = recipient = store_get(ss - s + 1);
2236 for (p = s; p < ss; p++) if (*p != '\n') *pp++ = *p;
2237 *pp = 0;
2238 recipient = parse_extract_address(recipient, &errmess, &start, &end,
2239 &domain, FALSE);
2240
2241 /* Keep a list of all the bad addresses so we can send a single
2242 error message at the end. However, an empty address is not an error;
2243 just ignore it. This can come from an empty group list like
2244
2245 To: Recipients of list:;
2246
2247 If there are no recipients at all, an error will occur later. */
2248
2249 if (recipient == NULL && Ustrcmp(errmess, "empty address") != 0)
2250 {
2251 int len = Ustrlen(s);
2252 error_block *b = store_get(sizeof(error_block));
2253 while (len > 0 && isspace(s[len-1])) len--;
2254 b->next = NULL;
2255 b->text1 = string_printing(string_copyn(s, len));
2256 b->text2 = errmess;
2257 *bnext = b;
2258 bnext = &(b->next);
2259 }
2260
2261 /* If the recipient is already in the nonrecipients tree, it must
2262 have appeared on the command line with the option extract_addresses_
2263 remove_arguments set. Do not add it to the recipients, and keep a note
2264 that this has happened, in order to give a better error if there are
2265 no recipients left. */
2266
2267 else if (recipient != NULL)
2268 {
2269 if (tree_search(tree_nonrecipients, recipient) == NULL)
2270 receive_add_recipient(recipient, -1);
2271 else
2272 extracted_ignored = TRUE;
2273 }
2274
2275 /* Move on past this address */
2276
2277 s = ss + (*ss? 1:0);
2278 while (isspace(*s)) s++;
2279 } /* Next address */
2280
2281 parse_allow_group = FALSE; /* Reset group syntax flags */
2282 parse_found_group = FALSE;
2283
2284 /* If this was the bcc: header, mark it "old", which means it
2285 will be kept on the spool, but not transmitted as part of the
2286 message. */
2287
2288 if (h->type == htype_bcc) h->type = htype_old;
2289 } /* For appropriate header line */
2290 } /* For each header line */
2291
2292 }
2293
2294 /* Now build the unique message id. This has changed several times over the
2295 lifetime of Exim. This description was rewritten for Exim 4.14 (February 2003).
2296 Retaining all the history in the comment has become too unwieldy - read
2297 previous release sources if you want it.
2298
2299 The message ID has 3 parts: tttttt-pppppp-ss. Each part is a number in base 62.
2300 The first part is the current time, in seconds. The second part is the current
2301 pid. Both are large enough to hold 32-bit numbers in base 62. The third part
2302 can hold a number in the range 0-3843. It used to be a computed sequence
2303 number, but is now the fractional component of the current time in units of
2304 1/2000 of a second (i.e. a value in the range 0-1999). After a message has been
2305 received, Exim ensures that the timer has ticked at the appropriate level
2306 before proceeding, to avoid duplication if the pid happened to be re-used
2307 within the same time period. It seems likely that most messages will take at
2308 least half a millisecond to be received, so no delay will normally be
2309 necessary. At least for some time...
2310
2311 There is a modification when localhost_number is set. Formerly this was allowed
2312 to be as large as 255. Now it is restricted to the range 0-16, and the final
2313 component of the message id becomes (localhost_number * 200) + fractional time
2314 in units of 1/200 of a second (i.e. a value in the range 0-3399).
2315
2316 Some not-really-Unix operating systems use case-insensitive file names (Darwin,
2317 Cygwin). For these, we have to use base 36 instead of base 62. Luckily, this
2318 still allows the tttttt field to hold a large enough number to last for some
2319 more decades, and the final two-digit field can hold numbers up to 1295, which
2320 is enough for milliseconds (instead of 1/2000 of a second).
2321
2322 However, the pppppp field cannot hold a 32-bit pid, but it can hold a 31-bit
2323 pid, so it is probably safe because pids have to be positive. The
2324 localhost_number is restricted to 0-10 for these hosts, and when it is set, the
2325 final field becomes (localhost_number * 100) + fractional time in centiseconds.
2326
2327 Note that string_base62() returns its data in a static storage block, so it
2328 must be copied before calling string_base62() again. It always returns exactly
2329 6 characters.
2330
2331 There doesn't seem to be anything in the RFC which requires a message id to
2332 start with a letter, but Smail was changed to ensure this. The external form of
2333 the message id (as supplied by string expansion) therefore starts with an
2334 additional leading 'E'. The spool file names do not include this leading
2335 letter and it is not used internally.
2336
2337 NOTE: If ever the format of message ids is changed, the regular expression for
2338 checking that a string is in this format must be updated in a corresponding
2339 way. It appears in the initializing code in exim.c. The macro MESSAGE_ID_LENGTH
2340 must also be changed to reflect the correct string length. Then, of course,
2341 other programs that rely on the message id format will need updating too. */
2342
2343 Ustrncpy(message_id, string_base62((long int)(message_id_tv.tv_sec)), 6);
2344 message_id[6] = '-';
2345 Ustrncpy(message_id + 7, string_base62((long int)getpid()), 6);
2346
2347 /* Deal with the case where the host number is set. The value of the number was
2348 checked when it was read, to ensure it isn't too big. The timing granularity is
2349 left in id_resolution so that an appropriate wait can be done after receiving
2350 the message, if necessary (we hope it won't be). */
2351
2352 if (host_number_string != NULL)
2353 {
2354 id_resolution = (BASE_62 == 62)? 5000 : 10000;
2355 sprintf(CS(message_id + MESSAGE_ID_LENGTH - 3), "-%2s",
2356 string_base62((long int)(
2357 host_number * (1000000/id_resolution) +
2358 message_id_tv.tv_usec/id_resolution)) + 4);
2359 }
2360
2361 /* Host number not set: final field is just the fractional time at an
2362 appropriate resolution. */
2363
2364 else
2365 {
2366 id_resolution = (BASE_62 == 62)? 500 : 1000;
2367 sprintf(CS(message_id + MESSAGE_ID_LENGTH - 3), "-%2s",
2368 string_base62((long int)(message_id_tv.tv_usec/id_resolution)) + 4);
2369 }
2370
2371 /* Add the current message id onto the current process info string if
2372 it will fit. */
2373
2374 (void)string_format(process_info + process_info_len,
2375 PROCESS_INFO_SIZE - process_info_len, " id=%s", message_id);
2376
2377 /* If we are using multiple input directories, set up the one for this message
2378 to be the least significant base-62 digit of the time of arrival. Otherwise
2379 ensure that it is an empty string. */
2380
2381 message_subdir[0] = split_spool_directory? message_id[5] : 0;
2382
2383 /* Now that we have the message-id, if there is no message-id: header, generate
2384 one, but only for local (without suppress_local_fixups) or submission mode
2385 messages. This can be user-configured if required, but we had better flatten
2386 any illegal characters therein. */
2387
2388 if (msgid_header == NULL &&
2389 ((sender_host_address == NULL && !suppress_local_fixups)
2390 || submission_mode))
2391 {
2392 uschar *p;
2393 uschar *id_text = US"";
2394 uschar *id_domain = primary_hostname;
2395
2396 /* Permit only letters, digits, dots, and hyphens in the domain */
2397
2398 if (message_id_domain != NULL)
2399 {
2400 uschar *new_id_domain = expand_string(message_id_domain);
2401 if (new_id_domain == NULL)
2402 {
2403 if (!expand_string_forcedfail)
2404 log_write(0, LOG_MAIN|LOG_PANIC,
2405 "expansion of \"%s\" (message_id_header_domain) "
2406 "failed: %s", message_id_domain, expand_string_message);
2407 }
2408 else if (*new_id_domain != 0)
2409 {
2410 id_domain = new_id_domain;
2411 for (p = id_domain; *p != 0; p++)
2412 if (!isalnum(*p) && *p != '.') *p = '-'; /* No need to test '-' ! */
2413 }
2414 }
2415
2416 /* Permit all characters except controls and RFC 2822 specials in the
2417 additional text part. */
2418
2419 if (message_id_text != NULL)
2420 {
2421 uschar *new_id_text = expand_string(message_id_text);
2422 if (new_id_text == NULL)
2423 {
2424 if (!expand_string_forcedfail)
2425 log_write(0, LOG_MAIN|LOG_PANIC,
2426 "expansion of \"%s\" (message_id_header_text) "
2427 "failed: %s", message_id_text, expand_string_message);
2428 }
2429 else if (*new_id_text != 0)
2430 {
2431 id_text = new_id_text;
2432 for (p = id_text; *p != 0; p++)
2433 if (mac_iscntrl_or_special(*p)) *p = '-';
2434 }
2435 }
2436
2437 /* Add the header line
2438 * Resent-* headers are prepended, per RFC 5322 3.6.6. Non-Resent-* are
2439 * appended, to preserve classical expectations of header ordering. */
2440
2441 header_add_at_position(!resents_exist, NULL, FALSE, htype_id,
2442 "%sMessage-Id: <%s%s%s@%s>\n", resent_prefix, message_id_external,
2443 (*id_text == 0)? "" : ".", id_text, id_domain);
2444 }
2445
2446 /* If we are to log recipients, keep a copy of the raw ones before any possible
2447 rewriting. Must copy the count, because later ACLs and the local_scan()
2448 function may mess with the real recipients. */
2449
2450 if ((log_extra_selector & LX_received_recipients) != 0)
2451 {
2452 raw_recipients = store_get(recipients_count * sizeof(uschar *));
2453 for (i = 0; i < recipients_count; i++)
2454 raw_recipients[i] = string_copy(recipients_list[i].address);
2455 raw_recipients_count = recipients_count;
2456 }
2457
2458 /* Ensure the recipients list is fully qualified and rewritten. Unqualified
2459 recipients will get here only if the conditions were right (allow_unqualified_
2460 recipient is TRUE). */
2461
2462 for (i = 0; i < recipients_count; i++)
2463 recipients_list[i].address =
2464 rewrite_address(recipients_list[i].address, TRUE, TRUE,
2465 global_rewrite_rules, rewrite_existflags);
2466
2467 /* If there is no From: header, generate one for local (without
2468 suppress_local_fixups) or submission_mode messages. If there is no sender
2469 address, but the sender is local or this is a local delivery error, use the
2470 originator login. This shouldn't happen for genuine bounces, but might happen
2471 for autoreplies. The addition of From: must be done *before* checking for the
2472 possible addition of a Sender: header, because untrusted_set_sender allows an
2473 untrusted user to set anything in the envelope (which might then get info
2474 From:) but we still want to ensure a valid Sender: if it is required. */
2475
2476 if (from_header == NULL &&
2477 ((sender_host_address == NULL && !suppress_local_fixups)
2478 || submission_mode))
2479 {
2480 uschar *oname = US"";
2481
2482 /* Use the originator_name if this is a locally submitted message and the
2483 caller is not trusted. For trusted callers, use it only if -F was used to
2484 force its value or if we have a non-SMTP message for which -f was not used
2485 to set the sender. */
2486
2487 if (sender_host_address == NULL)
2488 {
2489 if (!trusted_caller || sender_name_forced ||
2490 (!smtp_input && !sender_address_forced))
2491 oname = originator_name;
2492 }
2493
2494 /* For non-locally submitted messages, the only time we use the originator
2495 name is when it was forced by the /name= option on control=submission. */
2496
2497 else
2498 {
2499 if (submission_name != NULL) oname = submission_name;
2500 }
2501
2502 /* Envelope sender is empty */
2503
2504 if (sender_address[0] == 0)
2505 {
2506 uschar *fromstart, *fromend;
2507
2508 fromstart = string_sprintf("%sFrom: %s%s", resent_prefix,
2509 oname, (oname[0] == 0)? "" : " <");
2510 fromend = (oname[0] == 0)? US"" : US">";
2511
2512 if (sender_local || local_error_message)
2513 {
2514 header_add(htype_from, "%s%s@%s%s\n", fromstart,
2515 local_part_quote(originator_login), qualify_domain_sender,
2516 fromend);
2517 }
2518 else if (submission_mode && authenticated_id != NULL)
2519 {
2520 if (submission_domain == NULL)
2521 {
2522 header_add(htype_from, "%s%s@%s%s\n", fromstart,
2523 local_part_quote(authenticated_id), qualify_domain_sender,
2524 fromend);
2525 }
2526 else if (submission_domain[0] == 0) /* empty => whole address set */
2527 {
2528 header_add(htype_from, "%s%s%s\n", fromstart, authenticated_id,
2529 fromend);
2530 }
2531 else
2532 {
2533 header_add(htype_from, "%s%s@%s%s\n", fromstart,
2534 local_part_quote(authenticated_id), submission_domain,
2535 fromend);
2536 }
2537 from_header = header_last; /* To get it checked for Sender: */
2538 }
2539 }
2540
2541 /* There is a non-null envelope sender. Build the header using the original
2542 sender address, before any rewriting that might have been done while
2543 verifying it. */
2544
2545 else
2546 {
2547 header_add(htype_from, "%sFrom: %s%s%s%s\n", resent_prefix,
2548 oname,
2549 (oname[0] == 0)? "" : " <",
2550 (sender_address_unrewritten == NULL)?
2551 sender_address : sender_address_unrewritten,
2552 (oname[0] == 0)? "" : ">");
2553
2554 from_header = header_last; /* To get it checked for Sender: */
2555 }
2556 }
2557
2558
2559 /* If the sender is local (without suppress_local_fixups), or if we are in
2560 submission mode and there is an authenticated_id, check that an existing From:
2561 is correct, and if not, generate a Sender: header, unless disabled. Any
2562 previously-existing Sender: header was removed above. Note that sender_local,
2563 as well as being TRUE if the caller of exim is not trusted, is also true if a
2564 trusted caller did not supply a -f argument for non-smtp input. To allow
2565 trusted callers to forge From: without supplying -f, we have to test explicitly
2566 here. If the From: header contains more than one address, then the call to
2567 parse_extract_address fails, and a Sender: header is inserted, as required. */
2568
2569 if (from_header != NULL &&
2570 (active_local_from_check &&
2571 ((sender_local && !trusted_caller && !suppress_local_fixups) ||
2572 (submission_mode && authenticated_id != NULL))
2573 ))
2574 {
2575 BOOL make_sender = TRUE;
2576 int start, end, domain;
2577 uschar *errmess;
2578 uschar *from_address =
2579 parse_extract_address(Ustrchr(from_header->text, ':') + 1, &errmess,
2580 &start, &end, &domain, FALSE);
2581 uschar *generated_sender_address;
2582
2583 if (submission_mode)
2584 {
2585 if (submission_domain == NULL)
2586 {
2587 generated_sender_address = string_sprintf("%s@%s",
2588 local_part_quote(authenticated_id), qualify_domain_sender);
2589 }
2590 else if (submission_domain[0] == 0) /* empty => full address */
2591 {
2592 generated_sender_address = string_sprintf("%s",
2593 authenticated_id);
2594 }
2595 else
2596 {
2597 generated_sender_address = string_sprintf("%s@%s",
2598 local_part_quote(authenticated_id), submission_domain);
2599 }
2600 }
2601 else
2602 generated_sender_address = string_sprintf("%s@%s",
2603 local_part_quote(originator_login), qualify_domain_sender);
2604
2605 /* Remove permitted prefixes and suffixes from the local part of the From:
2606 address before doing the comparison with the generated sender. */
2607
2608 if (from_address != NULL)
2609 {
2610 int slen;
2611 uschar *at = (domain == 0)? NULL : from_address + domain - 1;
2612
2613 if (at != NULL) *at = 0;
2614 from_address += route_check_prefix(from_address, local_from_prefix);
2615 slen = route_check_suffix(from_address, local_from_suffix);
2616 if (slen > 0)
2617 {
2618 memmove(from_address+slen, from_address, Ustrlen(from_address)-slen);
2619 from_address += slen;
2620 }
2621 if (at != NULL) *at = '@';
2622
2623 if (strcmpic(generated_sender_address, from_address) == 0 ||
2624 (domain == 0 && strcmpic(from_address, originator_login) == 0))
2625 make_sender = FALSE;
2626 }
2627
2628 /* We have to cause the Sender header to be rewritten if there are
2629 appropriate rewriting rules. */
2630
2631 if (make_sender)
2632 {
2633 if (submission_mode && submission_name == NULL)
2634 header_add(htype_sender, "%sSender: %s\n", resent_prefix,
2635 generated_sender_address);
2636 else
2637 header_add(htype_sender, "%sSender: %s <%s>\n",
2638 resent_prefix,
2639 submission_mode? submission_name : originator_name,
2640 generated_sender_address);
2641 }
2642
2643 /* Ensure that a non-null envelope sender address corresponds to the
2644 submission mode sender address. */
2645
2646 if (submission_mode && sender_address[0] != 0)
2647 {
2648 if (sender_address_unrewritten == NULL)
2649 sender_address_unrewritten = sender_address;
2650 sender_address = generated_sender_address;
2651 if (Ustrcmp(sender_address_unrewritten, generated_sender_address) != 0)
2652 log_write(L_address_rewrite, LOG_MAIN,
2653 "\"%s\" from env-from rewritten as \"%s\" by submission mode",
2654 sender_address_unrewritten, generated_sender_address);
2655 }
2656 }
2657
2658
2659 /* If there are any rewriting rules, apply them to the sender address, unless
2660 it has already been rewritten as part of verification for SMTP input. */
2661
2662 if (global_rewrite_rules != NULL && sender_address_unrewritten == NULL &&
2663 sender_address[0] != 0)
2664 {
2665 sender_address = rewrite_address(sender_address, FALSE, TRUE,
2666 global_rewrite_rules, rewrite_existflags);
2667 DEBUG(D_receive|D_rewrite)
2668 debug_printf("rewritten sender = %s\n", sender_address);
2669 }
2670
2671
2672 /* The headers must be run through rewrite_header(), because it ensures that
2673 addresses are fully qualified, as well as applying any rewriting rules that may
2674 exist.
2675
2676 Qualification of header addresses in a message from a remote host happens only
2677 if the host is in sender_unqualified_hosts or recipient_unqualified hosts, as
2678 appropriate. For local messages, qualification always happens, unless -bnq is
2679 used to explicitly suppress it. No rewriting is done for an unqualified address
2680 that is left untouched.
2681
2682 We start at the second header, skipping our own Received:. This rewriting is
2683 documented as happening *after* recipient addresses are taken from the headers
2684 by the -t command line option. An added Sender: gets rewritten here. */
2685
2686 for (h = header_list->next; h != NULL; h = h->next)
2687 {
2688 header_line *newh = rewrite_header(h, NULL, NULL, global_rewrite_rules,
2689 rewrite_existflags, TRUE);
2690 if (newh != NULL) h = newh;
2691 }
2692
2693
2694 /* An RFC 822 (sic) message is not legal unless it has at least one of "to",
2695 "cc", or "bcc". Note that although the minimal examples in RFC 822 show just
2696 "to" or "bcc", the full syntax spec allows "cc" as well. If any resent- header
2697 exists, this applies to the set of resent- headers rather than the normal set.
2698
2699 The requirement for a recipient header has been removed in RFC 2822. At this
2700 point in the code, earlier versions of Exim added a To: header for locally
2701 submitted messages, and an empty Bcc: header for others. In the light of the
2702 changes in RFC 2822, this was dropped in November 2003. */
2703
2704
2705 /* If there is no date header, generate one if the message originates locally
2706 (i.e. not over TCP/IP) and suppress_local_fixups is not set, or if the
2707 submission mode flag is set. Messages without Date: are not valid, but it seems
2708 to be more confusing if Exim adds one to all remotely-originated messages.
2709 As per Message-Id, we prepend if resending, else append.
2710 */
2711
2712 if (!date_header_exists &&
2713 ((sender_host_address == NULL && !suppress_local_fixups)
2714 || submission_mode))
2715 header_add_at_position(!resents_exist, NULL, FALSE, htype_other,
2716 "%sDate: %s\n", resent_prefix, tod_stamp(tod_full));
2717
2718 search_tidyup(); /* Free any cached resources */
2719
2720 /* Show the complete set of headers if debugging. Note that the first one (the
2721 new Received:) has not yet been set. */
2722
2723 DEBUG(D_receive)
2724 {
2725 debug_printf(">>Headers after rewriting and local additions:\n");
2726 for (h = header_list->next; h != NULL; h = h->next)
2727 debug_printf("%c %s", h->type, h->text);
2728 debug_printf("\n");
2729 }
2730
2731 /* The headers are now complete in store. If we are running in filter
2732 testing mode, that is all this function does. Return TRUE if the message
2733 ended with a dot. */
2734
2735 if (filter_test != FTEST_NONE)
2736 {
2737 process_info[process_info_len] = 0;
2738 return message_ended == END_DOT;
2739 }
2740
2741 /* Cutthrough delivery:
2742 We have to create the Received header now rather than at the end of reception,
2743 so the timestamp behaviour is a change to the normal case.
2744 XXX Ensure this gets documented XXX.
2745 Having created it, send the headers to the destination.
2746 */
2747 if (cutthrough_fd >= 0)
2748 {
2749 if (received_count > received_headers_max)
2750 {
2751 cancel_cutthrough_connection("too many headers");
2752 if (smtp_input) receive_swallow_smtp(); /* Swallow incoming SMTP */
2753 log_write(0, LOG_MAIN|LOG_REJECT, "rejected from <%s>%s%s%s%s: "
2754 "Too many \"Received\" headers",
2755 sender_address,
2756 (sender_fullhost == NULL)? "" : " H=",
2757 (sender_fullhost == NULL)? US"" : sender_fullhost,
2758 (sender_ident == NULL)? "" : " U=",
2759 (sender_ident == NULL)? US"" : sender_ident);
2760 message_id[0] = 0; /* Indicate no message accepted */
2761 smtp_reply = US"550 Too many \"Received\" headers - suspected mail loop";
2762 goto TIDYUP; /* Skip to end of function */
2763 }
2764 received_header_gen();
2765 add_acl_headers(US"MAIL or RCPT");
2766 (void) cutthrough_headers_send();
2767 }
2768
2769
2770 /* Open a new spool file for the data portion of the message. We need
2771 to access it both via a file descriptor and a stream. Try to make the
2772 directory if it isn't there. Note re use of sprintf: spool_directory
2773 is checked on input to be < 200 characters long. */
2774
2775 sprintf(CS spool_name, "%s/input/%s/%s-D", spool_directory, message_subdir,
2776 message_id);
2777 data_fd = Uopen(spool_name, O_RDWR|O_CREAT|O_EXCL, SPOOL_MODE);
2778 if (data_fd < 0)
2779 {
2780 if (errno == ENOENT)
2781 {
2782 uschar temp[16];
2783 sprintf(CS temp, "input/%s", message_subdir);
2784 if (message_subdir[0] == 0) temp[5] = 0;
2785 (void)directory_make(spool_directory, temp, INPUT_DIRECTORY_MODE, TRUE);
2786 data_fd = Uopen(spool_name, O_RDWR|O_CREAT|O_EXCL, SPOOL_MODE);
2787 }
2788 if (data_fd < 0)
2789 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Failed to create spool file %s: %s",
2790 spool_name, strerror(errno));
2791 }
2792
2793 /* Make sure the file's group is the Exim gid, and double-check the mode
2794 because the group setting doesn't always get set automatically. */
2795
2796 if (fchown(data_fd, exim_uid, exim_gid))
2797 log_write(0, LOG_MAIN|LOG_PANIC_DIE,
2798 "Failed setting ownership on spool file %s: %s",
2799 spool_name, strerror(errno));
2800 (void)fchmod(data_fd, SPOOL_MODE);
2801
2802 /* We now have data file open. Build a stream for it and lock it. We lock only
2803 the first line of the file (containing the message ID) because otherwise there
2804 are problems when Exim is run under Cygwin (I'm told). See comments in
2805 spool_in.c, where the same locking is done. */
2806
2807 data_file = fdopen(data_fd, "w+");
2808 lock_data.l_type = F_WRLCK;
2809 lock_data.l_whence = SEEK_SET;
2810 lock_data.l_start = 0;
2811 lock_data.l_len = SPOOL_DATA_START_OFFSET;
2812
2813 if (fcntl(data_fd, F_SETLK, &lock_data) < 0)
2814 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Cannot lock %s (%d): %s", spool_name,
2815 errno, strerror(errno));
2816
2817 /* We have an open, locked data file. Write the message id to it to make it
2818 self-identifying. Then read the remainder of the input of this message and
2819 write it to the data file. If the variable next != NULL, it contains the first
2820 data line (which was read as a header but then turned out not to have the right
2821 format); write it (remembering that it might contain binary zeros). The result
2822 of fwrite() isn't inspected; instead we call ferror() below. */
2823
2824 fprintf(data_file, "%s-D\n", message_id);
2825 if (next != NULL)
2826 {
2827 uschar *s = next->text;
2828 int len = next->slen;
2829 len = fwrite(s, 1, len, data_file); len = len; /* compiler quietening */
2830 body_linecount++; /* Assumes only 1 line */
2831 }
2832
2833 /* Note that we might already be at end of file, or the logical end of file
2834 (indicated by '.'), or might have encountered an error while writing the
2835 message id or "next" line. */
2836
2837 if (!ferror(data_file) && !(receive_feof)() && message_ended != END_DOT)
2838 {
2839 if (smtp_input)
2840 {
2841 message_ended = read_message_data_smtp(data_file);
2842 receive_linecount++; /* The terminating "." line */
2843 }
2844 else message_ended = read_message_data(data_file);
2845
2846 receive_linecount += body_linecount; /* For BSMTP errors mainly */
2847 message_linecount += body_linecount;
2848
2849 /* Handle premature termination of SMTP */
2850
2851 if (smtp_input && message_ended == END_EOF)
2852 {
2853 Uunlink(spool_name); /* Lose data file when closed */
2854 cancel_cutthrough_connection("sender closed connection");
2855 message_id[0] = 0; /* Indicate no message accepted */
2856 smtp_reply = handle_lost_connection(US"");
2857 smtp_yield = FALSE;
2858 goto TIDYUP; /* Skip to end of function */
2859 }
2860
2861 /* Handle message that is too big. Don't use host_or_ident() in the log
2862 message; we want to see the ident value even for non-remote messages. */
2863
2864 if (message_ended == END_SIZE)
2865 {
2866 Uunlink(spool_name); /* Lose the data file when closed */
2867 cancel_cutthrough_connection("mail too big");
2868 if (smtp_input) receive_swallow_smtp(); /* Swallow incoming SMTP */
2869
2870 log_write(L_size_reject, LOG_MAIN|LOG_REJECT, "rejected from <%s>%s%s%s%s: "
2871 "message too big: read=%d max=%d",
2872 sender_address,
2873 (sender_fullhost == NULL)? "" : " H=",
2874 (sender_fullhost == NULL)? US"" : sender_fullhost,
2875 (sender_ident == NULL)? "" : " U=",
2876 (sender_ident == NULL)? US"" : sender_ident,
2877 message_size,
2878 thismessage_size_limit);
2879
2880 if (smtp_input)
2881 {
2882 smtp_reply = US"552 Message size exceeds maximum permitted";
2883 message_id[0] = 0; /* Indicate no message accepted */
2884 goto TIDYUP; /* Skip to end of function */
2885 }
2886 else
2887 {
2888 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
2889 give_local_error(ERRMESS_TOOBIG,
2890 string_sprintf("message too big (max=%d)", thismessage_size_limit),
2891 US"message rejected: ", error_rc, data_file, header_list);
2892 /* Does not return */
2893 }
2894 }
2895 }
2896
2897 /* Restore the standard SIGALRM handler for any subsequent processing. (For
2898 example, there may be some expansion in an ACL that uses a timer.) */
2899
2900 os_non_restarting_signal(SIGALRM, sigalrm_handler);
2901
2902 /* The message body has now been read into the data file. Call fflush() to
2903 empty the buffers in C, and then call fsync() to get the data written out onto
2904 the disk, as fflush() doesn't do this (or at least, it isn't documented as
2905 having to do this). If there was an I/O error on either input or output,
2906 attempt to send an error message, and unlink the spool file. For non-SMTP input
2907 we can then give up. Note that for SMTP input we must swallow the remainder of
2908 the input in cases of output errors, since the far end doesn't expect to see
2909 anything until the terminating dot line is sent. */
2910
2911 if (fflush(data_file) == EOF || ferror(data_file) ||
2912 EXIMfsync(fileno(data_file)) < 0 || (receive_ferror)())
2913 {
2914 uschar *msg_errno = US strerror(errno);
2915 BOOL input_error = (receive_ferror)() != 0;
2916 uschar *msg = string_sprintf("%s error (%s) while receiving message from %s",
2917 input_error? "Input read" : "Spool write",
2918 msg_errno,
2919 (sender_fullhost != NULL)? sender_fullhost : sender_ident);
2920
2921 log_write(0, LOG_MAIN, "Message abandoned: %s", msg);
2922 Uunlink(spool_name); /* Lose the data file */
2923 cancel_cutthrough_connection("error writing spoolfile");
2924
2925 if (smtp_input)
2926 {
2927 if (input_error)
2928 smtp_reply = US"451 Error while reading input data";
2929 else
2930 {
2931 smtp_reply = US"451 Error while writing spool file";
2932 receive_swallow_smtp();
2933 }
2934 message_id[0] = 0; /* Indicate no message accepted */
2935 goto TIDYUP; /* Skip to end of function */
2936 }
2937
2938 else
2939 {
2940 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
2941 give_local_error(ERRMESS_IOERR, msg, US"", error_rc, data_file,
2942 header_list);
2943 /* Does not return */
2944 }
2945 }
2946
2947
2948 /* No I/O errors were encountered while writing the data file. */
2949
2950 DEBUG(D_receive) debug_printf("Data file written for message %s\n", message_id);
2951
2952
2953 /* If there were any bad addresses extracted by -t, or there were no recipients
2954 left after -t, send a message to the sender of this message, or write it to
2955 stderr if the error handling option is set that way. Note that there may
2956 legitimately be no recipients for an SMTP message if they have all been removed
2957 by "discard".
2958
2959 We need to rewind the data file in order to read it. In the case of no
2960 recipients or stderr error writing, throw the data file away afterwards, and
2961 exit. (This can't be SMTP, which always ensures there's at least one
2962 syntactically good recipient address.) */
2963
2964 if (extract_recip && (bad_addresses != NULL || recipients_count == 0))
2965 {
2966 DEBUG(D_receive)
2967 {
2968 if (recipients_count == 0) debug_printf("*** No recipients\n");
2969 if (bad_addresses != NULL)
2970 {
2971 error_block *eblock = bad_addresses;
2972 debug_printf("*** Bad address(es)\n");
2973 while (eblock != NULL)
2974 {
2975 debug_printf(" %s: %s\n", eblock->text1, eblock->text2);
2976 eblock = eblock->next;
2977 }
2978 }
2979 }
2980
2981 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
2982
2983 /* If configured to send errors to the sender, but this fails, force
2984 a failure error code. We use a special one for no recipients so that it
2985 can be detected by the autoreply transport. Otherwise error_rc is set to
2986 errors_sender_rc, which is EXIT_FAILURE unless -oee was given, in which case
2987 it is EXIT_SUCCESS. */
2988
2989 if (error_handling == ERRORS_SENDER)
2990 {
2991 if (!moan_to_sender(
2992 (bad_addresses == NULL)?
2993 (extracted_ignored? ERRMESS_IGADDRESS : ERRMESS_NOADDRESS) :
2994 (recipients_list == NULL)? ERRMESS_BADNOADDRESS : ERRMESS_BADADDRESS,
2995 bad_addresses, header_list, data_file, FALSE))
2996 error_rc = (bad_addresses == NULL)? EXIT_NORECIPIENTS : EXIT_FAILURE;
2997 }
2998 else
2999 {
3000 if (bad_addresses == NULL)
3001 {
3002 if (extracted_ignored)
3003 fprintf(stderr, "exim: all -t recipients overridden by command line\n");
3004 else
3005 fprintf(stderr, "exim: no recipients in message\n");
3006 }
3007 else
3008 {
3009 fprintf(stderr, "exim: invalid address%s",
3010 (bad_addresses->next == NULL)? ":" : "es:\n");
3011 while (bad_addresses != NULL)
3012 {
3013 fprintf(stderr, " %s: %s\n", bad_addresses->text1,
3014 bad_addresses->text2);
3015 bad_addresses = bad_addresses->next;
3016 }
3017 }
3018 }
3019
3020 if (recipients_count == 0 || error_handling == ERRORS_STDERR)
3021 {
3022 Uunlink(spool_name);
3023 (void)fclose(data_file);
3024 exim_exit(error_rc);
3025 }
3026 }
3027
3028 /* Data file successfully written. Generate text for the Received: header by
3029 expanding the configured string, and adding a timestamp. By leaving this
3030 operation till now, we ensure that the timestamp is the time that message
3031 reception was completed. However, this is deliberately done before calling the
3032 data ACL and local_scan().
3033
3034 This Received: header may therefore be inspected by the data ACL and by code in
3035 the local_scan() function. When they have run, we update the timestamp to be
3036 the final time of reception.
3037
3038 If there is just one recipient, set up its value in the $received_for variable
3039 for use when we generate the Received: header.
3040
3041 Note: the checking for too many Received: headers is handled by the delivery
3042 code. */
3043 /*XXX eventually add excess Received: check for cutthrough case back when classifying them */
3044
3045 if (received_header->text == NULL) /* Non-cutthrough case */
3046 {
3047 received_header_gen();
3048
3049 /* Set the value of message_body_size for the DATA ACL and for local_scan() */
3050
3051 message_body_size = (fstat(data_fd, &statbuf) == 0)?
3052 statbuf.st_size - SPOOL_DATA_START_OFFSET : -1;
3053
3054 /* If an ACL from any RCPT commands set up any warning headers to add, do so
3055 now, before running the DATA ACL. */
3056
3057 add_acl_headers(US"MAIL or RCPT");
3058 }
3059 else
3060 message_body_size = (fstat(data_fd, &statbuf) == 0)?
3061 statbuf.st_size - SPOOL_DATA_START_OFFSET : -1;
3062
3063 /* If an ACL is specified for checking things at this stage of reception of a
3064 message, run it, unless all the recipients were removed by "discard" in earlier
3065 ACLs. That is the only case in which recipients_count can be zero at this
3066 stage. Set deliver_datafile to point to the data file so that $message_body and
3067 $message_body_end can be extracted if needed. Allow $recipients in expansions.
3068 */
3069
3070 deliver_datafile = data_fd;
3071 user_msg = NULL;
3072
3073 enable_dollar_recipients = TRUE;
3074
3075 if (recipients_count == 0)
3076 {
3077 blackholed_by = recipients_discarded? US"MAIL ACL" : US"RCPT ACL";
3078 }
3079 else
3080 {
3081 /* Handle interactive SMTP messages */
3082
3083 if (smtp_input && !smtp_batched_input)
3084 {
3085
3086 #ifndef DISABLE_DKIM
3087 if (!dkim_disable_verify)
3088 {
3089 /* Finish verification, this will log individual signature results to
3090 the mainlog */
3091 dkim_exim_verify_finish();
3092
3093 /* Check if we must run the DKIM ACL */
3094 if ((acl_smtp_dkim != NULL) &&
3095 (dkim_verify_signers != NULL) &&
3096 (dkim_verify_signers[0] != '\0'))
3097 {
3098 uschar *dkim_verify_signers_expanded =
3099 expand_string(dkim_verify_signers);
3100 if (dkim_verify_signers_expanded == NULL)
3101 {
3102 log_write(0, LOG_MAIN|LOG_PANIC,
3103 "expansion of dkim_verify_signers option failed: %s",
3104 expand_string_message);
3105 }
3106 else
3107 {
3108 int sep = 0;
3109 uschar *ptr = dkim_verify_signers_expanded;
3110 uschar *item = NULL;
3111 uschar *seen_items = NULL;
3112 int seen_items_size = 0;
3113 int seen_items_offset = 0;
3114 uschar itembuf[256];
3115 /* Default to OK when no items are present */
3116 rc = OK;
3117 while ((item = string_nextinlist(&ptr, &sep,
3118 itembuf,
3119 sizeof(itembuf))) != NULL)
3120 {
3121 /* Prevent running ACL for an empty item */
3122 if (!item || (item[0] == '\0')) continue;
3123 /* Only run ACL once for each domain or identity, no matter how often it
3124 appears in the expanded list. */
3125 if (seen_items != NULL)
3126 {
3127 uschar *seen_item = NULL;
3128 uschar seen_item_buf[256];
3129 uschar *seen_items_list = seen_items;
3130 int seen_this_item = 0;
3131
3132 while ((seen_item = string_nextinlist(&seen_items_list, &sep,
3133 seen_item_buf,
3134 sizeof(seen_item_buf))) != NULL)
3135 {
3136 if (Ustrcmp(seen_item,item) == 0)
3137 {
3138 seen_this_item = 1;
3139 break;
3140 }
3141 }
3142
3143 if (seen_this_item > 0)
3144 {
3145 DEBUG(D_receive)
3146 debug_printf("acl_smtp_dkim: skipping signer %s, already seen\n", item);
3147 continue;
3148 }
3149
3150 seen_items = string_append(seen_items,&seen_items_size,&seen_items_offset,1,":");
3151 }
3152
3153 seen_items = string_append(seen_items,&seen_items_size,&seen_items_offset,1,item);
3154 seen_items[seen_items_offset] = '\0';
3155
3156 DEBUG(D_receive)
3157 debug_printf("calling acl_smtp_dkim for dkim_cur_signer=%s\n", item);
3158
3159 dkim_exim_acl_setup(item);
3160 rc = acl_check(ACL_WHERE_DKIM, NULL, acl_smtp_dkim, &user_msg, &log_msg);
3161
3162 if (rc != OK)
3163 {
3164 DEBUG(D_receive)
3165 debug_printf("acl_smtp_dkim: acl_check returned %d on %s, skipping remaining items\n", rc, item);
3166 cancel_cutthrough_connection("dkim acl not ok");
3167 break;
3168 }
3169 }
3170 add_acl_headers(US"DKIM");
3171 if (rc == DISCARD)
3172 {
3173 recipients_count = 0;
3174 blackholed_by = US"DKIM ACL";
3175 if (log_msg != NULL)
3176 blackhole_log_msg = string_sprintf(": %s", log_msg);
3177 }
3178 else if (rc != OK)
3179 {
3180 Uunlink(spool_name);
3181 if (smtp_handle_acl_fail(ACL_WHERE_DKIM, rc, user_msg, log_msg) != 0)
3182 smtp_yield = FALSE; /* No more messsages after dropped connection */
3183 smtp_reply = US""; /* Indicate reply already sent */
3184 message_id[0] = 0; /* Indicate no message accepted */
3185 goto TIDYUP; /* Skip to end of function */
3186 }
3187 }
3188 }
3189 }
3190 #endif /* DISABLE_DKIM */
3191
3192 #ifdef WITH_CONTENT_SCAN
3193 if (recipients_count > 0 &&
3194 acl_smtp_mime != NULL &&
3195 !run_mime_acl(acl_smtp_mime, &smtp_yield, &smtp_reply, &blackholed_by))
3196 goto TIDYUP;
3197 #endif /* WITH_CONTENT_SCAN */
3198
3199 /* Check the recipients count again, as the MIME ACL might have changed
3200 them. */
3201
3202 if (acl_smtp_data != NULL && recipients_count > 0)
3203 {
3204 rc = acl_check(ACL_WHERE_DATA, NULL, acl_smtp_data, &user_msg, &log_msg);
3205 add_acl_headers(US"DATA");
3206 if (rc == DISCARD)
3207 {
3208 recipients_count = 0;
3209 blackholed_by = US"DATA ACL";
3210 if (log_msg != NULL)
3211 blackhole_log_msg = string_sprintf(": %s", log_msg);
3212 cancel_cutthrough_connection("data acl discard");
3213 }
3214 else if (rc != OK)
3215 {
3216 Uunlink(spool_name);
3217 cancel_cutthrough_connection("data acl not ok");
3218 #ifdef WITH_CONTENT_SCAN
3219 unspool_mbox();
3220 #endif
3221 #ifdef EXPERIMENTAL_DCC
3222 dcc_ok = 0;
3223 #endif
3224 if (smtp_handle_acl_fail(ACL_WHERE_DATA, rc, user_msg, log_msg) != 0)
3225 smtp_yield = FALSE; /* No more messsages after dropped connection */
3226 smtp_reply = US""; /* Indicate reply already sent */
3227 message_id[0] = 0; /* Indicate no message accepted */
3228 goto TIDYUP; /* Skip to end of function */
3229 }
3230 }
3231 }
3232
3233 /* Handle non-SMTP and batch SMTP (i.e. non-interactive) messages. Note that
3234 we cannot take different actions for permanent and temporary rejections. */
3235
3236 else
3237 {
3238
3239 #ifdef WITH_CONTENT_SCAN
3240 if (acl_not_smtp_mime != NULL &&
3241 !run_mime_acl(acl_not_smtp_mime, &smtp_yield, &smtp_reply,
3242 &blackholed_by))
3243 goto TIDYUP;
3244 #endif /* WITH_CONTENT_SCAN */
3245
3246 if (acl_not_smtp != NULL)
3247 {
3248 uschar *user_msg, *log_msg;
3249 rc = acl_check(ACL_WHERE_NOTSMTP, NULL, acl_not_smtp, &user_msg, &log_msg);
3250 if (rc == DISCARD)
3251 {
3252 recipients_count = 0;
3253 blackholed_by = US"non-SMTP ACL";
3254 if (log_msg != NULL)
3255 blackhole_log_msg = string_sprintf(": %s", log_msg);
3256 }
3257 else if (rc != OK)
3258 {
3259 Uunlink(spool_name);
3260 #ifdef WITH_CONTENT_SCAN
3261 unspool_mbox();
3262 #endif
3263 #ifdef EXPERIMENTAL_DCC
3264 dcc_ok = 0;
3265 #endif
3266 /* The ACL can specify where rejections are to be logged, possibly
3267 nowhere. The default is main and reject logs. */
3268
3269 if (log_reject_target != 0)
3270 log_write(0, log_reject_target, "F=<%s> rejected by non-SMTP ACL: %s",
3271 sender_address, log_msg);
3272
3273 if (user_msg == NULL) user_msg = US"local configuration problem";
3274 if (smtp_batched_input)
3275 {
3276 moan_smtp_batch(NULL, "%d %s", 550, user_msg);
3277 /* Does not return */
3278 }
3279 else
3280 {
3281 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3282 give_local_error(ERRMESS_LOCAL_ACL, user_msg,
3283 US"message rejected by non-SMTP ACL: ", error_rc, data_file,
3284 header_list);
3285 /* Does not return */
3286 }
3287 }
3288 add_acl_headers(US"non-SMTP");
3289 }
3290 }
3291
3292 /* The applicable ACLs have been run */
3293
3294 if (deliver_freeze) frozen_by = US"ACL"; /* for later logging */
3295 if (queue_only_policy) queued_by = US"ACL";
3296 }
3297
3298 #ifdef WITH_CONTENT_SCAN
3299 unspool_mbox();
3300 #endif
3301
3302 #ifdef EXPERIMENTAL_DCC
3303 dcc_ok = 0;
3304 #endif
3305
3306
3307 /* The final check on the message is to run the scan_local() function. The
3308 version supplied with Exim always accepts, but this is a hook for sysadmins to
3309 supply their own checking code. The local_scan() function is run even when all
3310 the recipients have been discarded. */
3311
3312 lseek(data_fd, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3313
3314 /* Arrange to catch crashes in local_scan(), so that the -D file gets
3315 deleted, and the incident gets logged. */
3316
3317 os_non_restarting_signal(SIGSEGV, local_scan_crash_handler);
3318 os_non_restarting_signal(SIGFPE, local_scan_crash_handler);
3319 os_non_restarting_signal(SIGILL, local_scan_crash_handler);
3320 os_non_restarting_signal(SIGBUS, local_scan_crash_handler);
3321
3322 DEBUG(D_receive) debug_printf("calling local_scan(); timeout=%d\n",
3323 local_scan_timeout);
3324 local_scan_data = NULL;
3325
3326 os_non_restarting_signal(SIGALRM, local_scan_timeout_handler);
3327 if (local_scan_timeout > 0) alarm(local_scan_timeout);
3328 rc = local_scan(data_fd, &local_scan_data);
3329 alarm(0);
3330 os_non_restarting_signal(SIGALRM, sigalrm_handler);
3331
3332 enable_dollar_recipients = FALSE;
3333
3334 store_pool = POOL_MAIN; /* In case changed */
3335 DEBUG(D_receive) debug_printf("local_scan() returned %d %s\n", rc,
3336 local_scan_data);
3337
3338 os_non_restarting_signal(SIGSEGV, SIG_DFL);
3339 os_non_restarting_signal(SIGFPE, SIG_DFL);
3340 os_non_restarting_signal(SIGILL, SIG_DFL);
3341 os_non_restarting_signal(SIGBUS, SIG_DFL);
3342
3343 /* The length check is paranoia against some runaway code, and also because
3344 (for a success return) lines in the spool file are read into big_buffer. */
3345
3346 if (local_scan_data != NULL)
3347 {
3348 int len = Ustrlen(local_scan_data);
3349 if (len > LOCAL_SCAN_MAX_RETURN) len = LOCAL_SCAN_MAX_RETURN;
3350 local_scan_data = string_copyn(local_scan_data, len);
3351 }
3352
3353 if (rc == LOCAL_SCAN_ACCEPT_FREEZE)
3354 {
3355 if (!deliver_freeze) /* ACL might have already frozen */
3356 {
3357 deliver_freeze = TRUE;
3358 deliver_frozen_at = time(NULL);
3359 frozen_by = US"local_scan()";
3360 }
3361 rc = LOCAL_SCAN_ACCEPT;
3362 }
3363 else if (rc == LOCAL_SCAN_ACCEPT_QUEUE)
3364 {
3365 if (!queue_only_policy) /* ACL might have already queued */
3366 {
3367 queue_only_policy = TRUE;
3368 queued_by = US"local_scan()";
3369 }
3370 rc = LOCAL_SCAN_ACCEPT;
3371 }
3372
3373 /* Message accepted: remove newlines in local_scan_data because otherwise
3374 the spool file gets corrupted. Ensure that all recipients are qualified. */
3375
3376 if (rc == LOCAL_SCAN_ACCEPT)
3377 {
3378 if (local_scan_data != NULL)
3379 {
3380 uschar *s;
3381 for (s = local_scan_data; *s != 0; s++) if (*s == '\n') *s = ' ';
3382 }
3383 for (i = 0; i < recipients_count; i++)
3384 {
3385 recipient_item *r = recipients_list + i;
3386 r->address = rewrite_address_qualify(r->address, TRUE);
3387 if (r->errors_to != NULL)
3388 r->errors_to = rewrite_address_qualify(r->errors_to, TRUE);
3389 }
3390 if (recipients_count == 0 && blackholed_by == NULL)
3391 blackholed_by = US"local_scan";
3392 }
3393
3394 /* Message rejected: newlines permitted in local_scan_data to generate
3395 multiline SMTP responses. */
3396
3397 else
3398 {
3399 uschar *istemp = US"";
3400 uschar *s = NULL;
3401 uschar *smtp_code;
3402 int size = 0;
3403 int sptr = 0;
3404
3405 errmsg = local_scan_data;
3406
3407 Uunlink(spool_name); /* Cancel this message */
3408 switch(rc)
3409 {
3410 default:
3411 log_write(0, LOG_MAIN, "invalid return %d from local_scan(). Temporary "
3412 "rejection given", rc);
3413 goto TEMPREJECT;
3414
3415 case LOCAL_SCAN_REJECT_NOLOGHDR:
3416 log_extra_selector &= ~LX_rejected_header;
3417 /* Fall through */
3418
3419 case LOCAL_SCAN_REJECT:
3420 smtp_code = US"550";
3421 if (errmsg == NULL) errmsg = US"Administrative prohibition";
3422 break;
3423
3424 case LOCAL_SCAN_TEMPREJECT_NOLOGHDR:
3425 log_extra_selector &= ~LX_rejected_header;
3426 /* Fall through */
3427
3428 case LOCAL_SCAN_TEMPREJECT:
3429 TEMPREJECT:
3430 smtp_code = US"451";
3431 if (errmsg == NULL) errmsg = US"Temporary local problem";
3432 istemp = US"temporarily ";
3433 break;
3434 }
3435
3436 s = string_append(s, &size, &sptr, 2, US"F=",
3437 (sender_address[0] == 0)? US"<>" : sender_address);
3438 s = add_host_info_for_log(s, &size, &sptr);
3439 s[sptr] = 0;
3440
3441 log_write(0, LOG_MAIN|LOG_REJECT, "%s %srejected by local_scan(): %.256s",
3442 s, istemp, string_printing(errmsg));
3443
3444 if (smtp_input)
3445 {
3446 if (!smtp_batched_input)
3447 {
3448 smtp_respond(smtp_code, 3, TRUE, errmsg);
3449 message_id[0] = 0; /* Indicate no message accepted */
3450 smtp_reply = US""; /* Indicate reply already sent */
3451 goto TIDYUP; /* Skip to end of function */
3452 }
3453 else
3454 {
3455 moan_smtp_batch(NULL, "%s %s", smtp_code, errmsg);
3456 /* Does not return */
3457 }
3458 }
3459 else
3460 {
3461 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3462 give_local_error(ERRMESS_LOCAL_SCAN, errmsg,
3463 US"message rejected by local scan code: ", error_rc, data_file,
3464 header_list);
3465 /* Does not return */
3466 }
3467 }
3468
3469 /* Reset signal handlers to ignore signals that previously would have caused
3470 the message to be abandoned. */
3471
3472 signal(SIGTERM, SIG_IGN);
3473 signal(SIGINT, SIG_IGN);
3474
3475
3476 /* Ensure the first time flag is set in the newly-received message. */
3477
3478 deliver_firsttime = TRUE;
3479
3480 #ifdef EXPERIMENTAL_BRIGHTMAIL
3481 if (bmi_run == 1) {
3482 /* rewind data file */
3483 lseek(data_fd, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3484 bmi_verdicts = bmi_process_message(header_list, data_fd);
3485 };
3486 #endif
3487
3488 /* Update the timstamp in our Received: header to account for any time taken by
3489 an ACL or by local_scan(). The new time is the time that all reception
3490 processing is complete. */
3491
3492 timestamp = expand_string(US"${tod_full}");
3493 tslen = Ustrlen(timestamp);
3494
3495 memcpy(received_header->text + received_header->slen - tslen - 1,
3496 timestamp, tslen);
3497
3498 /* In MUA wrapper mode, ignore queueing actions set by ACL or local_scan() */
3499
3500 if (mua_wrapper)
3501 {
3502 deliver_freeze = FALSE;
3503 queue_only_policy = FALSE;
3504 }
3505
3506 /* Keep the data file open until we have written the header file, in order to
3507 hold onto the lock. In a -bh run, or if the message is to be blackholed, we
3508 don't write the header file, and we unlink the data file. If writing the header
3509 file fails, we have failed to accept this message. */
3510
3511 if (host_checking || blackholed_by != NULL)
3512 {
3513 header_line *h;
3514 Uunlink(spool_name);
3515 msg_size = 0; /* Compute size for log line */
3516 for (h = header_list; h != NULL; h = h->next)
3517 if (h->type != '*') msg_size += h->slen;
3518 }
3519
3520 /* Write the -H file */
3521
3522 else
3523 {
3524 if ((msg_size = spool_write_header(message_id, SW_RECEIVING, &errmsg)) < 0)
3525 {
3526 log_write(0, LOG_MAIN, "Message abandoned: %s", errmsg);
3527 Uunlink(spool_name); /* Lose the data file */
3528
3529 if (smtp_input)
3530 {
3531 smtp_reply = US"451 Error in writing spool file";
3532 message_id[0] = 0; /* Indicate no message accepted */
3533 goto TIDYUP;
3534 }
3535 else
3536 {
3537 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3538 give_local_error(ERRMESS_IOERR, errmsg, US"", error_rc, data_file,
3539 header_list);
3540 /* Does not return */
3541 }
3542 }
3543 }
3544
3545
3546 /* The message has now been successfully received. */
3547
3548 receive_messagecount++;
3549
3550 /* In SMTP sessions we may receive several in one connection. After each one,
3551 we wait for the clock to tick at the level of message-id granularity. This is
3552 so that the combination of time+pid is unique, even on systems where the pid
3553 can be re-used within our time interval. We can't shorten the interval without
3554 re-designing the message-id. See comments above where the message id is
3555 created. This is Something For The Future. */
3556
3557 message_id_tv.tv_usec = (message_id_tv.tv_usec/id_resolution) * id_resolution;
3558 exim_wait_tick(&message_id_tv, id_resolution);
3559
3560 /* Add data size to written header size. We do not count the initial file name
3561 that is in the file, but we do add one extra for the notional blank line that
3562 precedes the data. This total differs from message_size in that it include the
3563 added Received: header and any other headers that got created locally. */
3564
3565 fflush(data_file);
3566 fstat(data_fd, &statbuf);
3567
3568 msg_size += statbuf.st_size - SPOOL_DATA_START_OFFSET + 1;
3569
3570 /* Generate a "message received" log entry. We do this by building up a dynamic
3571 string as required. Since we commonly want to add two items at a time, use a
3572 macro to simplify the coding. We log the arrival of a new message while the
3573 file is still locked, just in case the machine is *really* fast, and delivers
3574 it first! Include any message id that is in the message - since the syntax of a
3575 message id is actually an addr-spec, we can use the parse routine to canonicize
3576 it. */
3577
3578 size = 256;
3579 sptr = 0;
3580 s = store_get(size);
3581
3582 s = string_append(s, &size, &sptr, 2, US"<= ",
3583 (sender_address[0] == 0)? US"<>" : sender_address);
3584 if (message_reference != NULL)
3585 s = string_append(s, &size, &sptr, 2, US" R=", message_reference);
3586
3587 s = add_host_info_for_log(s, &size, &sptr);
3588
3589 #ifdef SUPPORT_TLS
3590 if ((log_extra_selector & LX_tls_cipher) != 0 && tls_in.cipher != NULL)
3591 s = string_append(s, &size, &sptr, 2, US" X=", tls_in.cipher);
3592 if ((log_extra_selector & LX_tls_certificate_verified) != 0 &&
3593 tls_in.cipher != NULL)
3594 s = string_append(s, &size, &sptr, 2, US" CV=",
3595 tls_in.certificate_verified? "yes":"no");
3596 if ((log_extra_selector & LX_tls_peerdn) != 0 && tls_in.peerdn != NULL)
3597 s = string_append(s, &size, &sptr, 3, US" DN=\"",
3598 string_printing(tls_in.peerdn), US"\"");
3599 if ((log_extra_selector & LX_tls_sni) != 0 && tls_in.sni != NULL)
3600 s = string_append(s, &size, &sptr, 3, US" SNI=\"",
3601 string_printing(tls_in.sni), US"\"");
3602 #endif
3603
3604 if (sender_host_authenticated != NULL)
3605 {
3606 s = string_append(s, &size, &sptr, 2, US" A=", sender_host_authenticated);
3607 if (authenticated_id != NULL)
3608 {
3609 s = string_append(s, &size, &sptr, 2, US":", authenticated_id);
3610 if (log_extra_selector & LX_smtp_mailauth && authenticated_sender != NULL)
3611 s = string_append(s, &size, &sptr, 2, US":", authenticated_sender);
3612 }
3613 }
3614
3615 sprintf(CS big_buffer, "%d", msg_size);
3616 s = string_append(s, &size, &sptr, 2, US" S=", big_buffer);
3617
3618 /* log 8BITMIME mode announced in MAIL_FROM
3619 0 ... no BODY= used
3620 7 ... 7BIT
3621 8 ... 8BITMIME */
3622 if (log_extra_selector & LX_8bitmime)
3623 {
3624 sprintf(CS big_buffer, "%d", body_8bitmime);
3625 s = string_append(s, &size, &sptr, 2, US" M8S=", big_buffer);
3626 }
3627
3628 /* If an addr-spec in a message-id contains a quoted string, it can contain
3629 any characters except " \ and CR and so in particular it can contain NL!
3630 Therefore, make sure we use a printing-characters only version for the log.
3631 Also, allow for domain literals in the message id. */
3632
3633 if (msgid_header != NULL)
3634 {
3635 uschar *old_id;
3636 BOOL save_allow_domain_literals = allow_domain_literals;
3637 allow_domain_literals = TRUE;
3638 old_id = parse_extract_address(Ustrchr(msgid_header->text, ':') + 1,
3639 &errmsg, &start, &end, &domain, FALSE);
3640 allow_domain_literals = save_allow_domain_literals;
3641 if (old_id != NULL)
3642 s = string_append(s, &size, &sptr, 2, US" id=", string_printing(old_id));
3643 }
3644
3645 /* If subject logging is turned on, create suitable printing-character
3646 text. By expanding $h_subject: we make use of the MIME decoding. */
3647
3648 if ((log_extra_selector & LX_subject) != 0 && subject_header != NULL)
3649 {
3650 int i;
3651 uschar *p = big_buffer;
3652 uschar *ss = expand_string(US"$h_subject:");
3653
3654 /* Backslash-quote any double quotes or backslashes so as to make a
3655 a C-like string, and turn any non-printers into escape sequences. */
3656
3657 *p++ = '\"';
3658 if (*ss != 0) for (i = 0; i < 100 && ss[i] != 0; i++)
3659 {
3660 if (ss[i] == '\"' || ss[i] == '\\') *p++ = '\\';
3661 *p++ = ss[i];
3662 }
3663 *p++ = '\"';
3664 *p = 0;
3665 s = string_append(s, &size, &sptr, 2, US" T=", string_printing(big_buffer));
3666 }
3667
3668 /* Terminate the string: string_cat() and string_append() leave room, but do
3669 not put the zero in. */
3670
3671 s[sptr] = 0;
3672
3673 /* Create a message log file if message logs are being used and this message is
3674 not blackholed. Write the reception stuff to it. We used to leave message log
3675 creation until the first delivery, but this has proved confusing for some
3676 people. */
3677
3678 if (message_logs && blackholed_by == NULL)
3679 {
3680 int fd;
3681
3682 sprintf(CS spool_name, "%s/msglog/%s/%s", spool_directory, message_subdir,
3683 message_id);
3684 fd = Uopen(spool_name, O_WRONLY|O_APPEND|O_CREAT, SPOOL_MODE);
3685
3686 if (fd < 0 && errno == ENOENT)
3687 {
3688 uschar temp[16];
3689 sprintf(CS temp, "msglog/%s", message_subdir);
3690 if (message_subdir[0] == 0) temp[6] = 0;
3691 (void)directory_make(spool_directory, temp, MSGLOG_DIRECTORY_MODE, TRUE);
3692 fd = Uopen(spool_name, O_WRONLY|O_APPEND|O_CREAT, SPOOL_MODE);
3693 }
3694
3695 if (fd < 0)
3696 {
3697 log_write(0, LOG_MAIN|LOG_PANIC, "Couldn't open message log %s: %s",
3698 spool_name, strerror(errno));
3699 }
3700
3701 else
3702 {
3703 FILE *message_log = fdopen(fd, "a");
3704 if (message_log == NULL)
3705 {
3706 log_write(0, LOG_MAIN|LOG_PANIC, "Couldn't fdopen message log %s: %s",
3707 spool_name, strerror(errno));
3708 (void)close(fd);
3709 }
3710 else
3711 {
3712 uschar *now = tod_stamp(tod_log);
3713 fprintf(message_log, "%s Received from %s\n", now, s+3);
3714 if (deliver_freeze) fprintf(message_log, "%s frozen by %s\n", now,
3715 frozen_by);
3716 if (queue_only_policy) fprintf(message_log,
3717 "%s no immediate delivery: queued by %s\n", now, queued_by);
3718 (void)fclose(message_log);
3719 }
3720 }
3721 }
3722
3723 /* Everything has now been done for a successful message except logging its
3724 arrival, and outputting an SMTP response. While writing to the log, set a flag
3725 to cause a call to receive_bomb_out() if the log cannot be opened. */
3726
3727 receive_call_bombout = TRUE;
3728
3729 /* Before sending an SMTP response in a TCP/IP session, we check to see if the
3730 connection has gone away. This can only be done if there is no unconsumed input
3731 waiting in the local input buffer. We can test for this by calling
3732 receive_smtp_buffered(). RFC 2920 (pipelining) explicitly allows for additional
3733 input to be sent following the final dot, so the presence of following input is
3734 not an error.
3735
3736 If the connection is still present, but there is no unread input for the
3737 socket, the result of a select() call will be zero. If, however, the connection
3738 has gone away, or if there is pending input, the result of select() will be
3739 non-zero. The two cases can be distinguished by trying to read the next input
3740 character. If we succeed, we can unread it so that it remains in the local
3741 buffer for handling later. If not, the connection has been lost.
3742
3743 Of course, since TCP/IP is asynchronous, there is always a chance that the
3744 connection will vanish between the time of this test and the sending of the
3745 response, but the chance of this happening should be small. */
3746
3747 if (smtp_input && sender_host_address != NULL && !sender_host_notsocket &&
3748 !receive_smtp_buffered())
3749 {
3750 struct timeval tv;
3751 fd_set select_check;
3752 FD_ZERO(&select_check);
3753 FD_SET(fileno(smtp_in), &select_check);
3754 tv.tv_sec = 0;
3755 tv.tv_usec = 0;
3756
3757 if (select(fileno(smtp_in) + 1, &select_check, NULL, NULL, &tv) != 0)
3758 {
3759 int c = (receive_getc)();
3760 if (c != EOF) (receive_ungetc)(c); else
3761 {
3762 uschar *msg = US"SMTP connection lost after final dot";
3763 smtp_reply = US""; /* No attempt to send a response */
3764 smtp_yield = FALSE; /* Nothing more on this connection */
3765
3766 /* Re-use the log line workspace */
3767
3768 sptr = 0;
3769 s = string_cat(s, &size, &sptr, msg, Ustrlen(msg));
3770 s = add_host_info_for_log(s, &size, &sptr);
3771 s[sptr] = 0;
3772 log_write(0, LOG_MAIN, "%s", s);
3773
3774 /* Delete the files for this aborted message. */
3775
3776 sprintf(CS spool_name, "%s/input/%s/%s-D", spool_directory,
3777 message_subdir, message_id);
3778 Uunlink(spool_name);
3779
3780 sprintf(CS spool_name, "%s/input/%s/%s-H", spool_directory,
3781 message_subdir, message_id);
3782 Uunlink(spool_name);
3783
3784 sprintf(CS spool_name, "%s/msglog/%s/%s", spool_directory,
3785 message_subdir, message_id);
3786 Uunlink(spool_name);
3787
3788 goto TIDYUP;
3789 }
3790 }
3791 }
3792
3793 /* The connection has not gone away; we really are going to take responsibility
3794 for this message. */
3795
3796 /* Cutthrough - had sender last-dot; assume we've sent (or bufferred) all
3797 data onward by now.
3798
3799 Send dot onward. If accepted, wipe the spooled files, log as delivered and accept
3800 the sender's dot (below).
3801 If rejected: copy response to sender, wipe the spooled files, log approriately.
3802 If temp-reject: accept to sender, keep the spooled files.
3803
3804 Having the normal spool files lets us do data-filtering, and store/forward on temp-reject.
3805
3806 XXX We do not handle queue-only, freezing, or blackholes.
3807 */
3808 cutthrough_done = 0;
3809 if(cutthrough_fd >= 0)
3810 {
3811 uschar * msg= cutthrough_finaldot(); /* Ask the target system to accept the messsage */
3812 /* Logging was done in finaldot() */
3813 switch(msg[0])
3814 {
3815 case '2': /* Accept. Do the same to the source; dump any spoolfiles. */
3816 cutthrough_done = 3;
3817 break; /* message_id needed for SMTP accept below */
3818
3819 default: /* Unknown response, or error. Treat as temp-reject. */
3820 case '4': /* Temp-reject. Keep spoolfiles and accept. */
3821 cutthrough_done = 1; /* Avoid the usual immediate delivery attempt */
3822 break; /* message_id needed for SMTP accept below */
3823
3824 case '5': /* Perm-reject. Do the same to the source. Dump any spoolfiles */
3825 smtp_reply= msg; /* Pass on the exact error */
3826 cutthrough_done = 2;
3827 break;
3828 }
3829 }
3830
3831 if(smtp_reply == NULL)
3832 {
3833 log_write(0, LOG_MAIN |
3834 (((log_extra_selector & LX_received_recipients) != 0)? LOG_RECIPIENTS : 0) |
3835 (((log_extra_selector & LX_received_sender) != 0)? LOG_SENDER : 0),
3836 "%s", s);
3837
3838 /* Log any control actions taken by an ACL or local_scan(). */
3839
3840 if (deliver_freeze) log_write(0, LOG_MAIN, "frozen by %s", frozen_by);
3841 if (queue_only_policy) log_write(L_delay_delivery, LOG_MAIN,
3842 "no immediate delivery: queued by %s", queued_by);
3843 }
3844 receive_call_bombout = FALSE;
3845
3846 store_reset(s); /* The store for the main log message can be reused */
3847
3848 /* If the message is frozen, and freeze_tell is set, do the telling. */
3849
3850 if (deliver_freeze && freeze_tell != NULL && freeze_tell[0] != 0)
3851 {
3852 moan_tell_someone(freeze_tell, NULL, US"Message frozen on arrival",
3853 "Message %s was frozen on arrival by %s.\nThe sender is <%s>.\n",
3854 message_id, frozen_by, sender_address);
3855 }
3856
3857
3858 /* Either a message has been successfully received and written to the two spool
3859 files, or an error in writing the spool has occurred for an SMTP message, or
3860 an SMTP message has been rejected for policy reasons. (For a non-SMTP message
3861 we will have already given up because there's no point in carrying on!) In
3862 either event, we must now close (and thereby unlock) the data file. In the
3863 successful case, this leaves the message on the spool, ready for delivery. In
3864 the error case, the spool file will be deleted. Then tidy up store, interact
3865 with an SMTP call if necessary, and return.
3866
3867 A fflush() was done earlier in the expectation that any write errors on the
3868 data file will be flushed(!) out thereby. Nevertheless, it is theoretically
3869 possible for fclose() to fail - but what to do? What has happened to the lock
3870 if this happens? */
3871
3872
3873 TIDYUP:
3874 process_info[process_info_len] = 0; /* Remove message id */
3875 if (data_file != NULL) (void)fclose(data_file); /* Frees the lock */
3876
3877 /* Now reset signal handlers to their defaults */
3878
3879 signal(SIGTERM, SIG_DFL);
3880 signal(SIGINT, SIG_DFL);
3881
3882 /* Tell an SMTP caller the state of play, and arrange to return the SMTP return
3883 value, which defaults TRUE - meaning there may be more incoming messages from
3884 this connection. For non-SMTP callers (where there is only ever one message),
3885 the default is FALSE. */
3886
3887 if (smtp_input)
3888 {
3889 yield = smtp_yield;
3890
3891 /* Handle interactive SMTP callers. After several kinds of error, smtp_reply
3892 is set to the response that should be sent. When it is NULL, we generate
3893 default responses. After an ACL error or local_scan() error, the response has
3894 already been sent, and smtp_reply is an empty string to indicate this. */
3895
3896 if (!smtp_batched_input)
3897 {
3898 if (smtp_reply == NULL)
3899 {
3900 if (fake_response != OK)
3901 smtp_respond((fake_response == DEFER)? US"450" : US"550", 3, TRUE,
3902 fake_response_text);
3903
3904 /* An OK response is required; use "message" text if present. */
3905
3906 else if (user_msg != NULL)
3907 {
3908 uschar *code = US"250";
3909 int len = 3;
3910 smtp_message_code(&code, &len, &user_msg, NULL);
3911 smtp_respond(code, len, TRUE, user_msg);
3912 }
3913
3914 /* Default OK response */
3915
3916 else
3917 smtp_printf("250 OK id=%s\r\n", message_id);
3918 if (host_checking)
3919 fprintf(stdout,
3920 "\n**** SMTP testing: that is not a real message id!\n\n");
3921 }
3922
3923 /* smtp_reply is set non-empty */
3924
3925 else if (smtp_reply[0] != 0)
3926 {
3927 if (fake_response != OK && (smtp_reply[0] == '2'))
3928 smtp_respond((fake_response == DEFER)? US"450" : US"550", 3, TRUE,
3929 fake_response_text);
3930 else
3931 smtp_printf("%.1024s\r\n", smtp_reply);
3932 }
3933
3934 switch (cutthrough_done)
3935 {
3936 case 3: log_write(0, LOG_MAIN, "Completed"); /* Delivery was done */
3937 case 2: { /* Delete spool files */
3938 sprintf(CS spool_name, "%s/input/%s/%s-D", spool_directory,
3939 message_subdir, message_id);
3940 Uunlink(spool_name);
3941 sprintf(CS spool_name, "%s/input/%s/%s-H", spool_directory,
3942 message_subdir, message_id);
3943 Uunlink(spool_name);
3944 sprintf(CS spool_name, "%s/msglog/%s/%s", spool_directory,
3945 message_subdir, message_id);
3946 Uunlink(spool_name);
3947 }
3948 case 1: message_id[0] = 0; /* Prevent a delivery from starting */
3949 default:break;
3950 }
3951 cutthrough_delivery = FALSE;
3952 }
3953
3954 /* For batched SMTP, generate an error message on failure, and do
3955 nothing on success. The function moan_smtp_batch() does not return -
3956 it exits from the program with a non-zero return code. */
3957
3958 else if (smtp_reply != NULL) moan_smtp_batch(NULL, "%s", smtp_reply);
3959 }
3960
3961
3962 /* If blackholing, we can immediately log this message's sad fate. The data
3963 file has already been unlinked, and the header file was never written to disk.
3964 We must now indicate that nothing was received, to prevent a delivery from
3965 starting. */
3966
3967 if (blackholed_by != NULL)
3968 {
3969 uschar *detail = (local_scan_data != NULL)?
3970 string_printing(local_scan_data) :
3971 string_sprintf("(%s discarded recipients)", blackholed_by);
3972 log_write(0, LOG_MAIN, "=> blackhole %s%s", detail, blackhole_log_msg);
3973 log_write(0, LOG_MAIN, "Completed");
3974 message_id[0] = 0;
3975 }
3976
3977 /* Reset headers so that logging of rejects for a subsequent message doesn't
3978 include them. It is also important to set header_last = NULL before exiting
3979 from this function, as this prevents certain rewrites that might happen during
3980 subsequent verifying (of another incoming message) from trying to add headers
3981 when they shouldn't. */
3982
3983 header_list = header_last = NULL;
3984
3985 return yield; /* TRUE if more messages (SMTP only) */
3986 }
3987
3988 /* End of receive.c */