More checks on header line length during reception
[exim.git] / src / src / receive.c
1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
4
5 /* Copyright (c) University of Cambridge 1995 - 2018 */
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 #include <setjmp.h>
12
13 #ifdef EXPERIMENTAL_DCC
14 extern int dcc_ok;
15 #endif
16
17 #ifdef EXPERIMENTAL_DMARC
18 # include "dmarc.h"
19 #endif /* EXPERIMENTAL_DMARC */
20
21 /*************************************************
22 * Local static variables *
23 *************************************************/
24
25 static int data_fd = -1;
26 static uschar *spool_name = US"";
27
28 enum CH_STATE {LF_SEEN, MID_LINE, CR_SEEN};
29
30 #ifdef HAVE_LOCAL_SCAN
31 jmp_buf local_scan_env; /* error-handling context for local_scan */
32 unsigned had_local_scan_crash;
33 unsigned had_local_scan_timeout;
34 #endif
35
36
37 /*************************************************
38 * Non-SMTP character reading functions *
39 *************************************************/
40
41 /* These are the default functions that are set up in the variables such as
42 receive_getc initially. They just call the standard functions, passing stdin as
43 the file. (When SMTP input is occurring, different functions are used by
44 changing the pointer variables.) */
45
46 int
47 stdin_getc(unsigned lim)
48 {
49 int c = getc(stdin);
50
51 if (had_data_timeout)
52 {
53 fprintf(stderr, "exim: timed out while reading - message abandoned\n");
54 log_write(L_lost_incoming_connection,
55 LOG_MAIN, "timed out while reading local message");
56 receive_bomb_out(US"data-timeout", NULL); /* Does not return */
57 }
58 if (had_data_sigint)
59 {
60 if (filter_test == FTEST_NONE)
61 {
62 fprintf(stderr, "\nexim: %s received - message abandoned\n",
63 had_data_sigint == SIGTERM ? "SIGTERM" : "SIGINT");
64 log_write(0, LOG_MAIN, "%s received while reading local message",
65 had_data_sigint == SIGTERM ? "SIGTERM" : "SIGINT");
66 }
67 receive_bomb_out(US"signal-exit", NULL); /* Does not return */
68 }
69 return c;
70 }
71
72 int
73 stdin_ungetc(int c)
74 {
75 return ungetc(c, stdin);
76 }
77
78 int
79 stdin_feof(void)
80 {
81 return feof(stdin);
82 }
83
84 int
85 stdin_ferror(void)
86 {
87 return ferror(stdin);
88 }
89
90
91
92
93 /*************************************************
94 * Check that a set sender is allowed *
95 *************************************************/
96
97 /* This function is called when a local caller sets an explicit sender address.
98 It checks whether this is permitted, which it is for trusted callers.
99 Otherwise, it must match the pattern(s) in untrusted_set_sender.
100
101 Arguments: the proposed sender address
102 Returns: TRUE for a trusted caller
103 TRUE if the address has been set, untrusted_set_sender has been
104 set, and the address matches something in the list
105 FALSE otherwise
106 */
107
108 BOOL
109 receive_check_set_sender(uschar *newsender)
110 {
111 uschar *qnewsender;
112 if (f.trusted_caller) return TRUE;
113 if (!newsender || !untrusted_set_sender) return FALSE;
114 qnewsender = Ustrchr(newsender, '@')
115 ? newsender : string_sprintf("%s@%s", newsender, qualify_domain_sender);
116 return match_address_list_basic(qnewsender, CUSS &untrusted_set_sender, 0) == OK;
117 }
118
119
120
121
122 /*************************************************
123 * Read space info for a partition *
124 *************************************************/
125
126 /* This function is called by receive_check_fs() below, and also by string
127 expansion for variables such as $spool_space. The field names for the statvfs
128 structure are macros, because not all OS have F_FAVAIL and it seems tidier to
129 have macros for F_BAVAIL and F_FILES as well. Some kinds of file system do not
130 have inodes, and they return -1 for the number available.
131
132 Later: It turns out that some file systems that do not have the concept of
133 inodes return 0 rather than -1. Such systems should also return 0 for the total
134 number of inodes, so we require that to be greater than zero before returning
135 an inode count.
136
137 Arguments:
138 isspool TRUE for spool partition, FALSE for log partition
139 inodeptr address of int to receive inode count; -1 if there isn't one
140
141 Returns: available on-root space, in kilobytes
142 -1 for log partition if there isn't one
143
144 All values are -1 if the STATFS functions are not available.
145 */
146
147 int_eximarith_t
148 receive_statvfs(BOOL isspool, int *inodeptr)
149 {
150 #ifdef HAVE_STATFS
151 struct STATVFS statbuf;
152 struct stat dummy;
153 uschar *path;
154 uschar *name;
155 uschar buffer[1024];
156
157 /* The spool directory must always exist. */
158
159 if (isspool)
160 {
161 path = spool_directory;
162 name = US"spool";
163 }
164
165 /* Need to cut down the log file path to the directory, and to ignore any
166 appearance of "syslog" in it. */
167
168 else
169 {
170 int sep = ':'; /* Not variable - outside scripts use */
171 const uschar *p = log_file_path;
172 name = US"log";
173
174 /* An empty log_file_path means "use the default". This is the same as an
175 empty item in a list. */
176
177 if (*p == 0) p = US":";
178 while ((path = string_nextinlist(&p, &sep, buffer, sizeof(buffer))))
179 if (Ustrcmp(path, "syslog") != 0)
180 break;
181
182 if (path == NULL) /* No log files */
183 {
184 *inodeptr = -1;
185 return -1;
186 }
187
188 /* An empty string means use the default, which is in the spool directory.
189 But don't just use the spool directory, as it is possible that the log
190 subdirectory has been symbolically linked elsewhere. */
191
192 if (path[0] == 0)
193 {
194 sprintf(CS buffer, CS"%s/log", CS spool_directory);
195 path = buffer;
196 }
197 else
198 {
199 uschar *cp;
200 if ((cp = Ustrrchr(path, '/')) != NULL) *cp = 0;
201 }
202 }
203
204 /* We now have the path; do the business */
205
206 memset(&statbuf, 0, sizeof(statbuf));
207
208 if (STATVFS(CS path, &statbuf) != 0)
209 if (stat(CS path, &dummy) == -1 && errno == ENOENT)
210 { /* Can happen on first run after installation */
211 *inodeptr = -1;
212 return -1;
213 }
214 else
215 {
216 log_write(0, LOG_MAIN|LOG_PANIC, "cannot accept message: failed to stat "
217 "%s directory %s: %s", name, path, strerror(errno));
218 smtp_closedown(US"spool or log directory problem");
219 exim_exit(EXIT_FAILURE, NULL);
220 }
221
222 *inodeptr = (statbuf.F_FILES > 0)? statbuf.F_FAVAIL : -1;
223
224 /* Disks are getting huge. Take care with computing the size in kilobytes. */
225
226 return (int_eximarith_t)(((double)statbuf.F_BAVAIL * (double)statbuf.F_FRSIZE)/1024.0);
227
228 #else
229 /* Unable to find partition sizes in this environment. */
230
231 *inodeptr = -1;
232 return -1;
233 #endif
234 }
235
236
237
238
239 /*************************************************
240 * Check space on spool and log partitions *
241 *************************************************/
242
243 /* This function is called before accepting a message; if any thresholds are
244 set, it checks them. If a message_size is supplied, it checks that there is
245 enough space for that size plus the threshold - i.e. that the message won't
246 reduce the space to the threshold. Not all OS have statvfs(); for those that
247 don't, this function always returns TRUE. For some OS the old function and
248 struct name statfs is used; that is handled by a macro, defined in exim.h.
249
250 Arguments:
251 msg_size the (estimated) size of an incoming message
252
253 Returns: FALSE if there isn't enough space, or if the information cannot
254 be obtained
255 TRUE if no check was done or there is enough space
256 */
257
258 BOOL
259 receive_check_fs(int msg_size)
260 {
261 int_eximarith_t space;
262 int inodes;
263
264 if (check_spool_space > 0 || msg_size > 0 || check_spool_inodes > 0)
265 {
266 space = receive_statvfs(TRUE, &inodes);
267
268 DEBUG(D_receive)
269 debug_printf("spool directory space = " PR_EXIM_ARITH "K inodes = %d "
270 "check_space = " PR_EXIM_ARITH "K inodes = %d msg_size = %d\n",
271 space, inodes, check_spool_space, check_spool_inodes, msg_size);
272
273 if ((space >= 0 && space < check_spool_space) ||
274 (inodes >= 0 && inodes < check_spool_inodes))
275 {
276 log_write(0, LOG_MAIN, "spool directory space check failed: space="
277 PR_EXIM_ARITH " inodes=%d", space, inodes);
278 return FALSE;
279 }
280 }
281
282 if (check_log_space > 0 || check_log_inodes > 0)
283 {
284 space = receive_statvfs(FALSE, &inodes);
285
286 DEBUG(D_receive)
287 debug_printf("log directory space = " PR_EXIM_ARITH "K inodes = %d "
288 "check_space = " PR_EXIM_ARITH "K inodes = %d\n",
289 space, inodes, check_log_space, check_log_inodes);
290
291 if ( space >= 0 && space < check_log_space
292 || inodes >= 0 && inodes < check_log_inodes)
293 {
294 log_write(0, LOG_MAIN, "log directory space check failed: space=" PR_EXIM_ARITH
295 " inodes=%d", space, inodes);
296 return FALSE;
297 }
298 }
299
300 return TRUE;
301 }
302
303
304
305 /*************************************************
306 * Bomb out while reading a message *
307 *************************************************/
308
309 /* The common case of wanting to bomb out is if a SIGTERM or SIGINT is
310 received, or if there is a timeout. A rarer case might be if the log files are
311 screwed up and Exim can't open them to record a message's arrival. Handling
312 that case is done by setting a flag to cause the log functions to call this
313 function if there is an ultimate disaster. That is why it is globally
314 accessible.
315
316 Arguments:
317 reason text reason to pass to the not-quit ACL
318 msg default SMTP response to give if in an SMTP session
319 Returns: it doesn't
320 */
321
322 void
323 receive_bomb_out(uschar *reason, uschar *msg)
324 {
325 static BOOL already_bombing_out;
326 /* The smtp_notquit_exit() below can call ACLs which can trigger recursive
327 timeouts, if someone has something slow in their quit ACL. Since the only
328 things we should be doing are to close down cleanly ASAP, on the second
329 pass we also close down stuff that might be opened again, before bypassing
330 the ACL call and exiting. */
331
332 /* If spool_name is set, it contains the name of the data file that is being
333 written. Unlink it before closing so that it cannot be picked up by a delivery
334 process. Ensure that any header file is also removed. */
335
336 if (spool_name[0] != '\0')
337 {
338 Uunlink(spool_name);
339 spool_name[Ustrlen(spool_name) - 1] = 'H';
340 Uunlink(spool_name);
341 spool_name[0] = '\0';
342 }
343
344 /* Now close the file if it is open, either as a fd or a stream. */
345
346 if (spool_data_file)
347 {
348 (void)fclose(spool_data_file);
349 spool_data_file = NULL;
350 }
351 else if (data_fd >= 0)
352 {
353 (void)close(data_fd);
354 data_fd = -1;
355 }
356
357 /* Attempt to close down an SMTP connection tidily. For non-batched SMTP, call
358 smtp_notquit_exit(), which runs the NOTQUIT ACL, if present, and handles the
359 SMTP response. */
360
361 if (!already_bombing_out)
362 {
363 already_bombing_out = TRUE;
364 if (smtp_input)
365 {
366 if (smtp_batched_input)
367 moan_smtp_batch(NULL, "421 %s - message abandoned", msg); /* No return */
368 smtp_notquit_exit(reason, US"421", US"%s %s - closing connection.",
369 smtp_active_hostname, msg);
370 }
371 }
372
373 /* Exit from the program (non-BSMTP cases) */
374
375 exim_exit(EXIT_FAILURE, NULL);
376 }
377
378
379 /*************************************************
380 * Data read timeout *
381 *************************************************/
382
383 /* Handler function for timeouts that occur while reading the data that
384 comprises a message.
385
386 Argument: the signal number
387 Returns: nothing
388 */
389
390 static void
391 data_timeout_handler(int sig)
392 {
393 had_data_timeout = sig;
394 }
395
396
397
398 #ifdef HAVE_LOCAL_SCAN
399 /*************************************************
400 * local_scan() timeout *
401 *************************************************/
402
403 /* Handler function for timeouts that occur while running a local_scan()
404 function. Posix recommends against calling longjmp() from a signal-handler,
405 but the GCC manual says you can so we will, and trust that it's better than
406 calling probably non-signal-safe funxtions during logging from within the
407 handler, even with other compilers.
408
409 See also https://cwe.mitre.org/data/definitions/745.html which also lists
410 it as unsafe.
411
412 This is all because we have no control over what might be written for a
413 local-scan function, so cannot sprinkle had-signal checks after each
414 call-site. At least with the default "do-nothing" function we won't
415 ever get here.
416
417 Argument: the signal number
418 Returns: nothing
419 */
420
421 static void
422 local_scan_timeout_handler(int sig)
423 {
424 had_local_scan_timeout = sig;
425 siglongjmp(local_scan_env, 1);
426 }
427
428
429
430 /*************************************************
431 * local_scan() crashed *
432 *************************************************/
433
434 /* Handler function for signals that occur while running a local_scan()
435 function.
436
437 Argument: the signal number
438 Returns: nothing
439 */
440
441 static void
442 local_scan_crash_handler(int sig)
443 {
444 had_local_scan_crash = sig;
445 siglongjmp(local_scan_env, 1);
446 }
447
448 #endif /*HAVE_LOCAL_SCAN*/
449
450
451 /*************************************************
452 * SIGTERM or SIGINT received *
453 *************************************************/
454
455 /* Handler for SIGTERM or SIGINT signals that occur while reading the
456 data that comprises a message.
457
458 Argument: the signal number
459 Returns: nothing
460 */
461
462 static void
463 data_sigterm_sigint_handler(int sig)
464 {
465 had_data_sigint = sig;
466 }
467
468
469
470 /*************************************************
471 * Add new recipient to list *
472 *************************************************/
473
474 /* This function builds a list of recipient addresses in argc/argv
475 format.
476
477 Arguments:
478 recipient the next address to add to recipients_list
479 pno parent number for fixed aliases; -1 otherwise
480
481 Returns: nothing
482 */
483
484 void
485 receive_add_recipient(uschar *recipient, int pno)
486 {
487 if (recipients_count >= recipients_list_max)
488 {
489 recipient_item *oldlist = recipients_list;
490 int oldmax = recipients_list_max;
491 recipients_list_max = recipients_list_max ? 2*recipients_list_max : 50;
492 recipients_list = store_get(recipients_list_max * sizeof(recipient_item));
493 if (oldlist != NULL)
494 memcpy(recipients_list, oldlist, oldmax * sizeof(recipient_item));
495 }
496
497 recipients_list[recipients_count].address = recipient;
498 recipients_list[recipients_count].pno = pno;
499 #ifdef EXPERIMENTAL_BRIGHTMAIL
500 recipients_list[recipients_count].bmi_optin = bmi_current_optin;
501 /* reset optin string pointer for next recipient */
502 bmi_current_optin = NULL;
503 #endif
504 recipients_list[recipients_count].orcpt = NULL;
505 recipients_list[recipients_count].dsn_flags = 0;
506 recipients_list[recipients_count++].errors_to = NULL;
507 }
508
509
510
511
512 /*************************************************
513 * Send user response message *
514 *************************************************/
515
516 /* This function is passed a default response code and a user message. It calls
517 smtp_message_code() to check and possibly modify the response code, and then
518 calls smtp_respond() to transmit the response. I put this into a function
519 just to avoid a lot of repetition.
520
521 Arguments:
522 code the response code
523 user_msg the user message
524
525 Returns: nothing
526 */
527
528 #ifndef DISABLE_PRDR
529 static void
530 smtp_user_msg(uschar *code, uschar *user_msg)
531 {
532 int len = 3;
533 smtp_message_code(&code, &len, &user_msg, NULL, TRUE);
534 smtp_respond(code, len, TRUE, user_msg);
535 }
536 #endif
537
538
539
540
541
542 /*************************************************
543 * Remove a recipient from the list *
544 *************************************************/
545
546 /* This function is provided for local_scan() to use.
547
548 Argument:
549 recipient address to remove
550
551 Returns: TRUE if it did remove something; FALSE otherwise
552 */
553
554 BOOL
555 receive_remove_recipient(uschar *recipient)
556 {
557 int count;
558 DEBUG(D_receive) debug_printf("receive_remove_recipient(\"%s\") called\n",
559 recipient);
560 for (count = 0; count < recipients_count; count++)
561 {
562 if (Ustrcmp(recipients_list[count].address, recipient) == 0)
563 {
564 if ((--recipients_count - count) > 0)
565 memmove(recipients_list + count, recipients_list + count + 1,
566 (recipients_count - count)*sizeof(recipient_item));
567 return TRUE;
568 }
569 }
570 return FALSE;
571 }
572
573
574
575
576
577 /*************************************************
578 * Read data portion of a non-SMTP message *
579 *************************************************/
580
581 /* This function is called to read the remainder of a message (following the
582 header) when the input is not from SMTP - we are receiving a local message on
583 a standard input stream. The message is always terminated by EOF, and is also
584 terminated by a dot on a line by itself if the flag dot_ends is TRUE. Split the
585 two cases for maximum efficiency.
586
587 Ensure that the body ends with a newline. This will naturally be the case when
588 the termination is "\n.\n" but may not be otherwise. The RFC defines messages
589 as "sequences of lines" - this of course strictly applies only to SMTP, but
590 deliveries into BSD-type mailbox files also require it. Exim used to have a
591 flag for doing this at delivery time, but as it was always set for all
592 transports, I decided to simplify things by putting the check here instead.
593
594 There is at least one MUA (dtmail) that sends CRLF via this interface, and
595 other programs are known to do this as well. Exim used to have a option for
596 dealing with this: in July 2003, after much discussion, the code has been
597 changed to default to treat any of LF, CRLF, and bare CR as line terminators.
598
599 However, for the case when a dot on a line by itself terminates a message, the
600 only recognized terminating sequences before and after the dot are LF and CRLF.
601 Otherwise, having read EOL . CR, you don't know whether to read another
602 character or not.
603
604 Internally, in messages stored in Exim's spool files, LF is used as the line
605 terminator. Under the new regime, bare CRs will no longer appear in these
606 files.
607
608 Arguments:
609 fout a FILE to which to write the message
610
611 Returns: One of the END_xxx values indicating why it stopped reading
612 */
613
614 static int
615 read_message_data(FILE *fout)
616 {
617 int ch_state;
618 register int ch;
619 register int linelength = 0;
620
621 /* Handle the case when only EOF terminates the message */
622
623 if (!f.dot_ends)
624 {
625 register int last_ch = '\n';
626
627 for (; (ch = (receive_getc)(GETC_BUFFER_UNLIMITED)) != EOF; last_ch = ch)
628 {
629 if (ch == 0) body_zerocount++;
630 if (last_ch == '\r' && ch != '\n')
631 {
632 if (linelength > max_received_linelength)
633 max_received_linelength = linelength;
634 linelength = 0;
635 if (fputc('\n', fout) == EOF) return END_WERROR;
636 message_size++;
637 body_linecount++;
638 }
639 if (ch == '\r') continue;
640
641 if (fputc(ch, fout) == EOF) return END_WERROR;
642 if (ch == '\n')
643 {
644 if (linelength > max_received_linelength)
645 max_received_linelength = linelength;
646 linelength = 0;
647 body_linecount++;
648 }
649 else linelength++;
650 if (++message_size > thismessage_size_limit) return END_SIZE;
651 }
652
653 if (last_ch != '\n')
654 {
655 if (linelength > max_received_linelength)
656 max_received_linelength = linelength;
657 if (fputc('\n', fout) == EOF) return END_WERROR;
658 message_size++;
659 body_linecount++;
660 }
661
662 return END_EOF;
663 }
664
665 /* Handle the case when a dot on a line on its own, or EOF, terminates. */
666
667 ch_state = 1;
668
669 while ((ch = (receive_getc)(GETC_BUFFER_UNLIMITED)) != EOF)
670 {
671 if (ch == 0) body_zerocount++;
672 switch (ch_state)
673 {
674 case 0: /* Normal state (previous char written) */
675 if (ch == '\n')
676 {
677 body_linecount++;
678 if (linelength > max_received_linelength)
679 max_received_linelength = linelength;
680 linelength = -1;
681 ch_state = 1;
682 }
683 else if (ch == '\r')
684 { ch_state = 2; continue; }
685 break;
686
687 case 1: /* After written "\n" */
688 if (ch == '.') { ch_state = 3; continue; }
689 if (ch == '\r') { ch_state = 2; continue; }
690 if (ch == '\n') { body_linecount++; linelength = -1; }
691 else ch_state = 0;
692 break;
693
694 case 2:
695 body_linecount++; /* After unwritten "\r" */
696 if (linelength > max_received_linelength)
697 max_received_linelength = linelength;
698 if (ch == '\n')
699 {
700 ch_state = 1;
701 linelength = -1;
702 }
703 else
704 {
705 if (message_size++, fputc('\n', fout) == EOF) return END_WERROR;
706 if (ch == '\r') continue;
707 ch_state = 0;
708 linelength = 0;
709 }
710 break;
711
712 case 3: /* After "\n." (\n written, dot not) */
713 if (ch == '\n') return END_DOT;
714 if (ch == '\r') { ch_state = 4; continue; }
715 message_size++;
716 linelength++;
717 if (fputc('.', fout) == EOF) return END_WERROR;
718 ch_state = 0;
719 break;
720
721 case 4: /* After "\n.\r" (\n written, rest not) */
722 if (ch == '\n') return END_DOT;
723 message_size += 2;
724 body_linecount++;
725 if (fputs(".\n", fout) == EOF) return END_WERROR;
726 if (ch == '\r') { ch_state = 2; continue; }
727 ch_state = 0;
728 break;
729 }
730
731 linelength++;
732 if (fputc(ch, fout) == EOF) return END_WERROR;
733 if (++message_size > thismessage_size_limit) return END_SIZE;
734 }
735
736 /* Get here if EOF read. Unless we have just written "\n", we need to ensure
737 the message ends with a newline, and we must also write any characters that
738 were saved up while testing for an ending dot. */
739
740 if (ch_state != 1)
741 {
742 static uschar *ends[] = { US"\n", NULL, US"\n", US".\n", US".\n" };
743 if (fputs(CS ends[ch_state], fout) == EOF) return END_WERROR;
744 message_size += Ustrlen(ends[ch_state]);
745 body_linecount++;
746 }
747
748 return END_EOF;
749 }
750
751
752
753
754 /*************************************************
755 * Read data portion of an SMTP message *
756 *************************************************/
757
758 /* This function is called to read the remainder of an SMTP message (after the
759 headers), or to skip over it when an error has occurred. In this case, the
760 output file is passed as NULL.
761
762 If any line begins with a dot, that character is skipped. The input should only
763 be successfully terminated by CR LF . CR LF unless it is local (non-network)
764 SMTP, in which case the CRs are optional, but...
765
766 FUDGE: It seems that sites on the net send out messages with just LF
767 terminators, despite the warnings in the RFCs, and other MTAs handle this. So
768 we make the CRs optional in all cases.
769
770 July 2003: Bare CRs cause trouble. We now treat them as line terminators as
771 well, so that there are no CRs in spooled messages. However, the message
772 terminating dot is not recognized between two bare CRs.
773
774 Arguments:
775 fout a FILE to which to write the message; NULL if skipping
776
777 Returns: One of the END_xxx values indicating why it stopped reading
778 */
779
780 static int
781 read_message_data_smtp(FILE *fout)
782 {
783 int ch_state = 0;
784 int ch;
785 int linelength = 0;
786
787 while ((ch = (receive_getc)(GETC_BUFFER_UNLIMITED)) != EOF)
788 {
789 if (ch == 0) body_zerocount++;
790 switch (ch_state)
791 {
792 case 0: /* After LF or CRLF */
793 if (ch == '.')
794 {
795 ch_state = 3;
796 continue; /* Don't ever write . after LF */
797 }
798 ch_state = 1;
799
800 /* Else fall through to handle as normal uschar. */
801
802 case 1: /* Normal state */
803 if (ch == '\n')
804 {
805 ch_state = 0;
806 body_linecount++;
807 if (linelength > max_received_linelength)
808 max_received_linelength = linelength;
809 linelength = -1;
810 }
811 else if (ch == '\r')
812 {
813 ch_state = 2;
814 continue;
815 }
816 break;
817
818 case 2: /* After (unwritten) CR */
819 body_linecount++;
820 if (linelength > max_received_linelength)
821 max_received_linelength = linelength;
822 linelength = -1;
823 if (ch == '\n')
824 {
825 ch_state = 0;
826 }
827 else
828 {
829 message_size++;
830 if (fout != NULL && fputc('\n', fout) == EOF) return END_WERROR;
831 cutthrough_data_put_nl();
832 if (ch != '\r') ch_state = 1; else continue;
833 }
834 break;
835
836 case 3: /* After [CR] LF . */
837 if (ch == '\n')
838 return END_DOT;
839 if (ch == '\r')
840 {
841 ch_state = 4;
842 continue;
843 }
844 /* The dot was removed at state 3. For a doubled dot, here, reinstate
845 it to cutthrough. The current ch, dot or not, is passed both to cutthrough
846 and to file below. */
847 if (ch == '.')
848 {
849 uschar c= ch;
850 cutthrough_data_puts(&c, 1);
851 }
852 ch_state = 1;
853 break;
854
855 case 4: /* After [CR] LF . CR */
856 if (ch == '\n') return END_DOT;
857 message_size++;
858 body_linecount++;
859 if (fout != NULL && fputc('\n', fout) == EOF) return END_WERROR;
860 cutthrough_data_put_nl();
861 if (ch == '\r')
862 {
863 ch_state = 2;
864 continue;
865 }
866 ch_state = 1;
867 break;
868 }
869
870 /* Add the character to the spool file, unless skipping; then loop for the
871 next. */
872
873 message_size++;
874 linelength++;
875 if (fout)
876 {
877 if (fputc(ch, fout) == EOF) return END_WERROR;
878 if (message_size > thismessage_size_limit) return END_SIZE;
879 }
880 if(ch == '\n')
881 cutthrough_data_put_nl();
882 else
883 {
884 uschar c = ch;
885 cutthrough_data_puts(&c, 1);
886 }
887 }
888
889 /* Fall through here if EOF encountered. This indicates some kind of error,
890 since a correct message is terminated by [CR] LF . [CR] LF. */
891
892 return END_EOF;
893 }
894
895
896
897
898 /* Variant of the above read_message_data_smtp() specialised for RFC 3030
899 CHUNKING. Accept input lines separated by either CRLF or CR or LF and write
900 LF-delimited spoolfile. Until we have wireformat spoolfiles, we need the
901 body_linecount accounting for proper re-expansion for the wire, so use
902 a cut-down version of the state-machine above; we don't need to do leading-dot
903 detection and unstuffing.
904
905 Arguments:
906 fout a FILE to which to write the message; NULL if skipping;
907 must be open for both writing and reading.
908
909 Returns: One of the END_xxx values indicating why it stopped reading
910 */
911
912 static int
913 read_message_bdat_smtp(FILE *fout)
914 {
915 int linelength = 0, ch;
916 enum CH_STATE ch_state = LF_SEEN;
917 BOOL fix_nl = FALSE;
918
919 for(;;)
920 {
921 switch ((ch = bdat_getc(GETC_BUFFER_UNLIMITED)))
922 {
923 case EOF: return END_EOF;
924 case ERR: return END_PROTOCOL;
925 case EOD:
926 /* Nothing to get from the sender anymore. We check the last
927 character written to the spool.
928
929 RFC 3030 states, that BDAT chunks are normal text, terminated by CRLF.
930 If we would be strict, we would refuse such broken messages.
931 But we are liberal, so we fix it. It would be easy just to append
932 the "\n" to the spool.
933
934 But there are some more things (line counting, message size calculation and such),
935 that would need to be duplicated here. So we simply do some ungetc
936 trickery.
937 */
938 if (fout)
939 {
940 if (fseek(fout, -1, SEEK_CUR) < 0) return END_PROTOCOL;
941 if (fgetc(fout) == '\n') return END_DOT;
942 }
943
944 if (linelength == -1) /* \r already seen (see below) */
945 {
946 DEBUG(D_receive) debug_printf("Add missing LF\n");
947 bdat_ungetc('\n');
948 continue;
949 }
950 DEBUG(D_receive) debug_printf("Add missing CRLF\n");
951 bdat_ungetc('\r'); /* not even \r was seen */
952 fix_nl = TRUE;
953
954 continue;
955 case '\0': body_zerocount++; break;
956 }
957 switch (ch_state)
958 {
959 case LF_SEEN: /* After LF or CRLF */
960 ch_state = MID_LINE;
961 /* fall through to handle as normal uschar. */
962
963 case MID_LINE: /* Mid-line state */
964 if (ch == '\n')
965 {
966 ch_state = LF_SEEN;
967 body_linecount++;
968 if (linelength > max_received_linelength)
969 max_received_linelength = linelength;
970 linelength = -1;
971 }
972 else if (ch == '\r')
973 {
974 ch_state = CR_SEEN;
975 if (fix_nl) bdat_ungetc('\n');
976 continue; /* don't write CR */
977 }
978 break;
979
980 case CR_SEEN: /* After (unwritten) CR */
981 body_linecount++;
982 if (linelength > max_received_linelength)
983 max_received_linelength = linelength;
984 linelength = -1;
985 if (ch == '\n')
986 ch_state = LF_SEEN;
987 else
988 {
989 message_size++;
990 if (fout && fputc('\n', fout) == EOF) return END_WERROR;
991 cutthrough_data_put_nl();
992 if (ch == '\r') continue; /* don't write CR */
993 ch_state = MID_LINE;
994 }
995 break;
996 }
997
998 /* Add the character to the spool file, unless skipping */
999
1000 message_size++;
1001 linelength++;
1002 if (fout)
1003 {
1004 if (fputc(ch, fout) == EOF) return END_WERROR;
1005 if (message_size > thismessage_size_limit) return END_SIZE;
1006 }
1007 if(ch == '\n')
1008 cutthrough_data_put_nl();
1009 else
1010 {
1011 uschar c = ch;
1012 cutthrough_data_puts(&c, 1);
1013 }
1014 }
1015 /*NOTREACHED*/
1016 }
1017
1018 static int
1019 read_message_bdat_smtp_wire(FILE *fout)
1020 {
1021 int ch;
1022
1023 /* Remember that this message uses wireformat. */
1024
1025 DEBUG(D_receive) debug_printf("CHUNKING: %s\n",
1026 fout ? "writing spoolfile in wire format" : "flushing input");
1027 f.spool_file_wireformat = TRUE;
1028
1029 for (;;)
1030 {
1031 if (chunking_data_left > 0)
1032 {
1033 unsigned len = MAX(chunking_data_left, thismessage_size_limit - message_size + 1);
1034 uschar * buf = bdat_getbuf(&len);
1035
1036 if (!buf) return END_EOF;
1037 message_size += len;
1038 if (fout && fwrite(buf, len, 1, fout) != 1) return END_WERROR;
1039 }
1040 else switch (ch = bdat_getc(GETC_BUFFER_UNLIMITED))
1041 {
1042 case EOF: return END_EOF;
1043 case EOD: return END_DOT;
1044 case ERR: return END_PROTOCOL;
1045
1046 default:
1047 message_size++;
1048 /*XXX not done:
1049 linelength
1050 max_received_linelength
1051 body_linecount
1052 body_zerocount
1053 */
1054 if (fout && fputc(ch, fout) == EOF) return END_WERROR;
1055 break;
1056 }
1057 if (message_size > thismessage_size_limit) return END_SIZE;
1058 }
1059 /*NOTREACHED*/
1060 }
1061
1062
1063
1064
1065 /*************************************************
1066 * Swallow SMTP message *
1067 *************************************************/
1068
1069 /* This function is called when there has been some kind of error while reading
1070 an SMTP message, and the remaining data may need to be swallowed. It is global
1071 because it is called from smtp_closedown() to shut down an incoming call
1072 tidily.
1073
1074 Argument: a FILE from which to read the message
1075 Returns: nothing
1076 */
1077
1078 void
1079 receive_swallow_smtp(void)
1080 {
1081 if (message_ended >= END_NOTENDED)
1082 message_ended = chunking_state <= CHUNKING_OFFERED
1083 ? read_message_data_smtp(NULL)
1084 : read_message_bdat_smtp_wire(NULL);
1085 }
1086
1087
1088
1089 /*************************************************
1090 * Handle lost SMTP connection *
1091 *************************************************/
1092
1093 /* This function logs connection loss incidents and generates an appropriate
1094 SMTP response.
1095
1096 Argument: additional data for the message
1097 Returns: the SMTP response
1098 */
1099
1100 static uschar *
1101 handle_lost_connection(uschar *s)
1102 {
1103 log_write(L_lost_incoming_connection | L_smtp_connection, LOG_MAIN,
1104 "%s lost while reading message data%s", smtp_get_connection_info(), s);
1105 smtp_notquit_exit(US"connection-lost", NULL, NULL);
1106 return US"421 Lost incoming connection";
1107 }
1108
1109
1110
1111
1112 /*************************************************
1113 * Handle a non-smtp reception error *
1114 *************************************************/
1115
1116 /* This function is called for various errors during the reception of non-SMTP
1117 messages. It either sends a message to the sender of the problem message, or it
1118 writes to the standard error stream.
1119
1120 Arguments:
1121 errcode code for moan_to_sender(), identifying the error
1122 text1 first message text, passed to moan_to_sender()
1123 text2 second message text, used only for stderrr
1124 error_rc code to pass to exim_exit if no problem
1125 f FILE containing body of message (may be stdin)
1126 hptr pointer to instore headers or NULL
1127
1128 Returns: calls exim_exit(), which does not return
1129 */
1130
1131 static void
1132 give_local_error(int errcode, uschar *text1, uschar *text2, int error_rc,
1133 FILE *f, header_line *hptr)
1134 {
1135 if (error_handling == ERRORS_SENDER)
1136 {
1137 error_block eblock;
1138 eblock.next = NULL;
1139 eblock.text1 = text1;
1140 eblock.text2 = US"";
1141 if (!moan_to_sender(errcode, &eblock, hptr, f, FALSE))
1142 error_rc = EXIT_FAILURE;
1143 }
1144 else
1145 fprintf(stderr, "exim: %s%s\n", text2, text1); /* Sic */
1146 (void)fclose(f);
1147 exim_exit(error_rc, US"");
1148 }
1149
1150
1151
1152 /*************************************************
1153 * Add header lines set up by ACL *
1154 *************************************************/
1155
1156 /* This function is called to add the header lines that were set up by
1157 statements in an ACL to the list of headers in memory. It is done in two stages
1158 like this, because when the ACL for RCPT is running, the other headers have not
1159 yet been received. This function is called twice; once just before running the
1160 DATA ACL, and once after. This is so that header lines added by MAIL or RCPT
1161 are visible to the DATA ACL.
1162
1163 Originally these header lines were added at the end. Now there is support for
1164 three different places: top, bottom, and after the Received: header(s). There
1165 will always be at least one Received: header, even if it is marked deleted, and
1166 even if something else has been put in front of it.
1167
1168 Arguments:
1169 acl_name text to identify which ACL
1170
1171 Returns: nothing
1172 */
1173
1174 static void
1175 add_acl_headers(int where, uschar *acl_name)
1176 {
1177 header_line *h, *next;
1178 header_line *last_received = NULL;
1179
1180 switch(where)
1181 {
1182 case ACL_WHERE_DKIM:
1183 case ACL_WHERE_MIME:
1184 case ACL_WHERE_DATA:
1185 if ( cutthrough.cctx.sock >= 0 && cutthrough.delivery
1186 && (acl_removed_headers || acl_added_headers))
1187 {
1188 log_write(0, LOG_MAIN|LOG_PANIC, "Header modification in data ACLs"
1189 " will not take effect on cutthrough deliveries");
1190 return;
1191 }
1192 }
1193
1194 if (acl_removed_headers)
1195 {
1196 DEBUG(D_receive|D_acl) debug_printf_indent(">>Headers removed by %s ACL:\n", acl_name);
1197
1198 for (h = header_list; h; h = h->next) if (h->type != htype_old)
1199 {
1200 const uschar * list = acl_removed_headers;
1201 int sep = ':'; /* This is specified as a colon-separated list */
1202 uschar *s;
1203 uschar buffer[128];
1204
1205 while ((s = string_nextinlist(&list, &sep, buffer, sizeof(buffer))))
1206 if (header_testname(h, s, Ustrlen(s), FALSE))
1207 {
1208 h->type = htype_old;
1209 DEBUG(D_receive|D_acl) debug_printf_indent(" %s", h->text);
1210 }
1211 }
1212 acl_removed_headers = NULL;
1213 DEBUG(D_receive|D_acl) debug_printf_indent(">>\n");
1214 }
1215
1216 if (!acl_added_headers) return;
1217 DEBUG(D_receive|D_acl) debug_printf_indent(">>Headers added by %s ACL:\n", acl_name);
1218
1219 for (h = acl_added_headers; h; h = next)
1220 {
1221 next = h->next;
1222
1223 switch(h->type)
1224 {
1225 case htype_add_top:
1226 h->next = header_list;
1227 header_list = h;
1228 DEBUG(D_receive|D_acl) debug_printf_indent(" (at top)");
1229 break;
1230
1231 case htype_add_rec:
1232 if (!last_received)
1233 {
1234 last_received = header_list;
1235 while (!header_testname(last_received, US"Received", 8, FALSE))
1236 last_received = last_received->next;
1237 while (last_received->next &&
1238 header_testname(last_received->next, US"Received", 8, FALSE))
1239 last_received = last_received->next;
1240 }
1241 h->next = last_received->next;
1242 last_received->next = h;
1243 DEBUG(D_receive|D_acl) debug_printf_indent(" (after Received:)");
1244 break;
1245
1246 case htype_add_rfc:
1247 /* add header before any header which is NOT Received: or Resent- */
1248 last_received = header_list;
1249 while ( last_received->next &&
1250 ( (header_testname(last_received->next, US"Received", 8, FALSE)) ||
1251 (header_testname_incomplete(last_received->next, US"Resent-", 7, FALSE)) ) )
1252 last_received = last_received->next;
1253 /* last_received now points to the last Received: or Resent-* header
1254 in an uninterrupted chain of those header types (seen from the beginning
1255 of all headers. Our current header must follow it. */
1256 h->next = last_received->next;
1257 last_received->next = h;
1258 DEBUG(D_receive|D_acl) debug_printf_indent(" (before any non-Received: or Resent-*: header)");
1259 break;
1260
1261 default:
1262 h->next = NULL;
1263 header_last->next = h;
1264 DEBUG(D_receive|D_acl) debug_printf_indent(" ");
1265 break;
1266 }
1267
1268 if (!h->next) header_last = h;
1269
1270 /* Check for one of the known header types (From:, To:, etc.) though in
1271 practice most added headers are going to be "other". Lower case
1272 identification letters are never stored with the header; they are used
1273 for existence tests when messages are received. So discard any lower case
1274 flag values. */
1275
1276 h->type = header_checkname(h, FALSE);
1277 if (h->type >= 'a') h->type = htype_other;
1278
1279 DEBUG(D_receive|D_acl) debug_printf("%s", h->text);
1280 }
1281
1282 acl_added_headers = NULL;
1283 DEBUG(D_receive|D_acl) debug_printf_indent(">>\n");
1284 }
1285
1286
1287
1288 /*************************************************
1289 * Add host information for log line *
1290 *************************************************/
1291
1292 /* Called for acceptance and rejecting log lines. This adds information about
1293 the calling host to a string that is being built dynamically.
1294
1295 Arguments:
1296 s the dynamic string
1297
1298 Returns: the extended string
1299 */
1300
1301 static gstring *
1302 add_host_info_for_log(gstring * g)
1303 {
1304 if (sender_fullhost)
1305 {
1306 if (LOGGING(dnssec) && sender_host_dnssec) /*XXX sender_helo_dnssec? */
1307 g = string_catn(g, US" DS", 3);
1308 g = string_append(g, 2, US" H=", sender_fullhost);
1309 if (LOGGING(incoming_interface) && interface_address)
1310 g = string_fmt_append(g, " I=[%s]:%d", interface_address, interface_port);
1311 }
1312 if (f.tcp_in_fastopen && !f.tcp_in_fastopen_logged)
1313 {
1314 g = string_catn(g, US" TFO*", f.tcp_in_fastopen_data ? 5 : 4);
1315 f.tcp_in_fastopen_logged = TRUE;
1316 }
1317 if (sender_ident)
1318 g = string_append(g, 2, US" U=", sender_ident);
1319 if (received_protocol)
1320 g = string_append(g, 2, US" P=", received_protocol);
1321 if (LOGGING(pipelining) && f.smtp_in_pipelining_advertised)
1322 {
1323 g = string_catn(g, US" L", 2);
1324 #ifdef EXPERIMENTAL_PIPE_CONNECT
1325 if (f.smtp_in_early_pipe_used)
1326 g = string_catn(g, US"*", 1);
1327 else if (f.smtp_in_early_pipe_advertised)
1328 g = string_catn(g, US".", 1);
1329 #endif
1330 if (!f.smtp_in_pipelining_used)
1331 g = string_catn(g, US"-", 1);
1332 }
1333 return g;
1334 }
1335
1336
1337
1338 #ifdef WITH_CONTENT_SCAN
1339
1340 /*************************************************
1341 * Run the MIME ACL on a message *
1342 *************************************************/
1343
1344 /* This code is in a subroutine so that it can be used for both SMTP
1345 and non-SMTP messages. It is called with a non-NULL ACL pointer.
1346
1347 Arguments:
1348 acl The ACL to run (acl_smtp_mime or acl_not_smtp_mime)
1349 smtp_yield_ptr Set FALSE to kill messages after dropped connection
1350 smtp_reply_ptr Where SMTP reply is being built
1351 blackholed_by_ptr Where "blackholed by" message is being built
1352
1353 Returns: TRUE to carry on; FALSE to abandon the message
1354 */
1355
1356 static BOOL
1357 run_mime_acl(uschar *acl, BOOL *smtp_yield_ptr, uschar **smtp_reply_ptr,
1358 uschar **blackholed_by_ptr)
1359 {
1360 FILE *mbox_file;
1361 uschar * rfc822_file_path = NULL;
1362 unsigned long mbox_size;
1363 header_line *my_headerlist;
1364 uschar *user_msg, *log_msg;
1365 int mime_part_count_buffer = -1;
1366 uschar * mbox_filename;
1367 int rc = OK;
1368
1369 /* check if it is a MIME message */
1370
1371 for (my_headerlist = header_list; my_headerlist; my_headerlist = my_headerlist->next)
1372 if ( my_headerlist->type != '*' /* skip deleted headers */
1373 && strncmpic(my_headerlist->text, US"Content-Type:", 13) == 0
1374 )
1375 {
1376 DEBUG(D_receive) debug_printf("Found Content-Type: header - executing acl_smtp_mime.\n");
1377 goto DO_MIME_ACL;
1378 }
1379
1380 DEBUG(D_receive) debug_printf("No Content-Type: header - presumably not a MIME message.\n");
1381 return TRUE;
1382
1383 DO_MIME_ACL:
1384
1385 /* make sure the eml mbox file is spooled up */
1386 if (!(mbox_file = spool_mbox(&mbox_size, NULL, &mbox_filename)))
1387 { /* error while spooling */
1388 log_write(0, LOG_MAIN|LOG_PANIC,
1389 "acl_smtp_mime: error while creating mbox spool file, message temporarily rejected.");
1390 Uunlink(spool_name);
1391 unspool_mbox();
1392 #ifdef EXPERIMENTAL_DCC
1393 dcc_ok = 0;
1394 #endif
1395 smtp_respond(US"451", 3, TRUE, US"temporary local problem");
1396 message_id[0] = 0; /* Indicate no message accepted */
1397 *smtp_reply_ptr = US""; /* Indicate reply already sent */
1398 return FALSE; /* Indicate skip to end of receive function */
1399 }
1400
1401 mime_is_rfc822 = 0;
1402
1403 MIME_ACL_CHECK:
1404 mime_part_count = -1;
1405 rc = mime_acl_check(acl, mbox_file, NULL, &user_msg, &log_msg);
1406 (void)fclose(mbox_file);
1407
1408 if (rfc822_file_path)
1409 {
1410 mime_part_count = mime_part_count_buffer;
1411
1412 if (unlink(CS rfc822_file_path) == -1)
1413 {
1414 log_write(0, LOG_PANIC,
1415 "acl_smtp_mime: can't unlink RFC822 spool file, skipping.");
1416 goto END_MIME_ACL;
1417 }
1418 rfc822_file_path = NULL;
1419 }
1420
1421 /* check if we must check any message/rfc822 attachments */
1422 if (rc == OK)
1423 {
1424 uschar * scandir = string_copyn(mbox_filename,
1425 Ustrrchr(mbox_filename, '/') - mbox_filename);
1426 struct dirent * entry;
1427 DIR * tempdir;
1428
1429 for (tempdir = opendir(CS scandir); entry = readdir(tempdir); )
1430 if (strncmpic(US entry->d_name, US"__rfc822_", 9) == 0)
1431 {
1432 rfc822_file_path = string_sprintf("%s/%s", scandir, entry->d_name);
1433 DEBUG(D_receive)
1434 debug_printf("RFC822 attachment detected: running MIME ACL for '%s'\n",
1435 rfc822_file_path);
1436 break;
1437 }
1438 closedir(tempdir);
1439
1440 if (rfc822_file_path)
1441 {
1442 if ((mbox_file = Ufopen(rfc822_file_path, "rb")))
1443 {
1444 /* set RFC822 expansion variable */
1445 mime_is_rfc822 = 1;
1446 mime_part_count_buffer = mime_part_count;
1447 goto MIME_ACL_CHECK;
1448 }
1449 log_write(0, LOG_PANIC,
1450 "acl_smtp_mime: can't open RFC822 spool file, skipping.");
1451 unlink(CS rfc822_file_path);
1452 }
1453 }
1454
1455 END_MIME_ACL:
1456 add_acl_headers(ACL_WHERE_MIME, US"MIME");
1457 if (rc == DISCARD)
1458 {
1459 recipients_count = 0;
1460 *blackholed_by_ptr = US"MIME ACL";
1461 cancel_cutthrough_connection(TRUE, US"mime acl discard");
1462 }
1463 else if (rc != OK)
1464 {
1465 Uunlink(spool_name);
1466 cancel_cutthrough_connection(TRUE, US"mime acl not ok");
1467 unspool_mbox();
1468 #ifdef EXPERIMENTAL_DCC
1469 dcc_ok = 0;
1470 #endif
1471 if (smtp_input)
1472 {
1473 if (smtp_handle_acl_fail(ACL_WHERE_MIME, rc, user_msg, log_msg) != 0)
1474 *smtp_yield_ptr = FALSE; /* No more messages after dropped connection */
1475 *smtp_reply_ptr = US""; /* Indicate reply already sent */
1476 }
1477 message_id[0] = 0; /* Indicate no message accepted */
1478 return FALSE; /* Cause skip to end of receive function */
1479 }
1480
1481 return TRUE;
1482 }
1483
1484 #endif /* WITH_CONTENT_SCAN */
1485
1486
1487
1488 void
1489 received_header_gen(void)
1490 {
1491 uschar *received;
1492 uschar *timestamp;
1493 header_line *received_header= header_list;
1494
1495 timestamp = expand_string(US"${tod_full}");
1496 if (recipients_count == 1) received_for = recipients_list[0].address;
1497 received = expand_string(received_header_text);
1498 received_for = NULL;
1499
1500 if (!received)
1501 {
1502 if(spool_name[0] != 0)
1503 Uunlink(spool_name); /* Lose the data file */
1504 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Expansion of \"%s\" "
1505 "(received_header_text) failed: %s", string_printing(received_header_text),
1506 expand_string_message);
1507 }
1508
1509 /* The first element on the header chain is reserved for the Received header,
1510 so all we have to do is fill in the text pointer, and set the type. However, if
1511 the result of the expansion is an empty string, we leave the header marked as
1512 "old" so as to refrain from adding a Received header. */
1513
1514 if (received[0] == 0)
1515 {
1516 received_header->text = string_sprintf("Received: ; %s\n", timestamp);
1517 received_header->type = htype_old;
1518 }
1519 else
1520 {
1521 received_header->text = string_sprintf("%s; %s\n", received, timestamp);
1522 received_header->type = htype_received;
1523 }
1524
1525 received_header->slen = Ustrlen(received_header->text);
1526
1527 DEBUG(D_receive) debug_printf(">>Generated Received: header line\n%c %s",
1528 received_header->type, received_header->text);
1529 }
1530
1531
1532
1533 /*************************************************
1534 * Receive message *
1535 *************************************************/
1536
1537 /* Receive a message on the given input, and put it into a pair of spool files.
1538 Either a non-null list of recipients, or the extract flag will be true, or
1539 both. The flag sender_local is true for locally generated messages. The flag
1540 submission_mode is true if an ACL has obeyed "control = submission". The flag
1541 suppress_local_fixups is true if an ACL has obeyed "control =
1542 suppress_local_fixups" or -G was passed on the command-line.
1543 The flag smtp_input is true if the message is to be
1544 handled using SMTP conventions about termination and lines starting with dots.
1545 For non-SMTP messages, dot_ends is true for dot-terminated messages.
1546
1547 If a message was successfully read, message_id[0] will be non-zero.
1548
1549 The general actions of this function are:
1550
1551 . Read the headers of the message (if any) into a chain of store
1552 blocks.
1553
1554 . If there is a "sender:" header and the message is locally originated,
1555 throw it away, unless the caller is trusted, or unless
1556 active_local_sender_retain is set - which can only happen if
1557 active_local_from_check is false.
1558
1559 . If recipients are to be extracted from the message, build the
1560 recipients list from the headers, removing any that were on the
1561 original recipients list (unless extract_addresses_remove_arguments is
1562 false), and at the same time, remove any bcc header that may be present.
1563
1564 . Get the spool file for the data, sort out its unique name, open
1565 and lock it (but don't give it the name yet).
1566
1567 . Generate a "Message-Id" header if the message doesn't have one, for
1568 locally-originated messages.
1569
1570 . Generate a "Received" header.
1571
1572 . Ensure the recipients list is fully qualified and rewritten if necessary.
1573
1574 . If there are any rewriting rules, apply them to the sender address
1575 and also to the headers.
1576
1577 . If there is no from: header, generate one, for locally-generated messages
1578 and messages in "submission mode" only.
1579
1580 . If the sender is local, check that from: is correct, and if not, generate
1581 a Sender: header, unless message comes from a trusted caller, or this
1582 feature is disabled by active_local_from_check being false.
1583
1584 . If there is no "date" header, generate one, for locally-originated
1585 or submission mode messages only.
1586
1587 . Copy the rest of the input, or up to a terminating "." if in SMTP or
1588 dot_ends mode, to the data file. Leave it open, to hold the lock.
1589
1590 . Write the envelope and the headers to a new file.
1591
1592 . Set the name for the header file; close it.
1593
1594 . Set the name for the data file; close it.
1595
1596 Because this function can potentially be called many times in a single
1597 SMTP connection, all store should be got by store_get(), so that it will be
1598 automatically retrieved after the message is accepted.
1599
1600 FUDGE: It seems that sites on the net send out messages with just LF
1601 terminators, despite the warnings in the RFCs, and other MTAs handle this. So
1602 we make the CRs optional in all cases.
1603
1604 July 2003: Bare CRs in messages, especially in header lines, cause trouble. A
1605 new regime is now in place in which bare CRs in header lines are turned into LF
1606 followed by a space, so as not to terminate the header line.
1607
1608 February 2004: A bare LF in a header line in a message whose first line was
1609 terminated by CRLF is treated in the same way as a bare CR.
1610
1611 Arguments:
1612 extract_recip TRUE if recipients are to be extracted from the message's
1613 headers
1614
1615 Returns: TRUE there are more messages to be read (SMTP input)
1616 FALSE there are no more messages to be read (non-SMTP input
1617 or SMTP connection collapsed, or other failure)
1618
1619 When reading a message for filter testing, the returned value indicates
1620 whether the headers (which is all that is read) were terminated by '.' or
1621 not. */
1622
1623 BOOL
1624 receive_msg(BOOL extract_recip)
1625 {
1626 int i;
1627 int rc = FAIL;
1628 int msg_size = 0;
1629 int process_info_len = Ustrlen(process_info);
1630 int error_rc = error_handling == ERRORS_SENDER
1631 ? errors_sender_rc : EXIT_FAILURE;
1632 int header_size = 256;
1633 int start, end, domain;
1634 int id_resolution = 0;
1635 int had_zero = 0;
1636 int prevlines_length = 0;
1637
1638 int ptr = 0;
1639
1640 BOOL contains_resent_headers = FALSE;
1641 BOOL extracted_ignored = FALSE;
1642 BOOL first_line_ended_crlf = TRUE_UNSET;
1643 BOOL smtp_yield = TRUE;
1644 BOOL yield = FALSE;
1645
1646 BOOL resents_exist = FALSE;
1647 uschar *resent_prefix = US"";
1648 uschar *blackholed_by = NULL;
1649 uschar *blackhole_log_msg = US"";
1650 enum {NOT_TRIED, TMP_REJ, PERM_REJ, ACCEPTED} cutthrough_done = NOT_TRIED;
1651
1652 flock_t lock_data;
1653 error_block *bad_addresses = NULL;
1654
1655 uschar *frozen_by = NULL;
1656 uschar *queued_by = NULL;
1657
1658 uschar *errmsg;
1659 gstring * g;
1660 struct stat statbuf;
1661
1662 /* Final message to give to SMTP caller, and messages from ACLs */
1663
1664 uschar *smtp_reply = NULL;
1665 uschar *user_msg, *log_msg;
1666
1667 /* Working header pointers */
1668
1669 header_line *h, *next;
1670
1671 /* Flags for noting the existence of certain headers (only one left) */
1672
1673 BOOL date_header_exists = FALSE;
1674
1675 /* Pointers to receive the addresses of headers whose contents we need. */
1676
1677 header_line *from_header = NULL;
1678 header_line *subject_header = NULL;
1679 header_line *msgid_header = NULL;
1680 header_line *received_header;
1681
1682 #ifdef EXPERIMENTAL_DMARC
1683 int dmarc_up = 0;
1684 #endif /* EXPERIMENTAL_DMARC */
1685
1686 /* Variables for use when building the Received: header. */
1687
1688 uschar *timestamp;
1689 int tslen;
1690
1691
1692 /* Release any open files that might have been cached while preparing to
1693 accept the message - e.g. by verifying addresses - because reading a message
1694 might take a fair bit of real time. */
1695
1696 search_tidyup();
1697
1698 /* Extracting the recipient list from an input file is incompatible with
1699 cutthrough delivery with the no-spool option. It shouldn't be possible
1700 to set up the combination, but just in case kill any ongoing connection. */
1701 if (extract_recip || !smtp_input)
1702 cancel_cutthrough_connection(TRUE, US"not smtp input");
1703
1704 /* Initialize the chain of headers by setting up a place-holder for Received:
1705 header. Temporarily mark it as "old", i.e. not to be used. We keep header_last
1706 pointing to the end of the chain to make adding headers simple. */
1707
1708 received_header = header_list = header_last = store_get(sizeof(header_line));
1709 header_list->next = NULL;
1710 header_list->type = htype_old;
1711 header_list->text = NULL;
1712 header_list->slen = 0;
1713
1714 /* Control block for the next header to be read. */
1715
1716 next = store_get(sizeof(header_line));
1717 next->text = store_get(header_size);
1718
1719 /* Initialize message id to be null (indicating no message read), and the
1720 header names list to be the normal list. Indicate there is no data file open
1721 yet, initialize the size and warning count, and deal with no size limit. */
1722
1723 message_id[0] = 0;
1724 spool_data_file = NULL;
1725 data_fd = -1;
1726 spool_name = US"";
1727 message_size = 0;
1728 warning_count = 0;
1729 received_count = 1; /* For the one we will add */
1730
1731 if (thismessage_size_limit <= 0) thismessage_size_limit = INT_MAX;
1732
1733 /* While reading the message, the following counts are computed. */
1734
1735 message_linecount = body_linecount = body_zerocount =
1736 max_received_linelength = 0;
1737
1738 #ifndef DISABLE_DKIM
1739 /* Call into DKIM to set up the context. In CHUNKING mode
1740 we clear the dot-stuffing flag */
1741 if (smtp_input && !smtp_batched_input && !f.dkim_disable_verify)
1742 dkim_exim_verify_init(chunking_state <= CHUNKING_OFFERED);
1743 #endif
1744
1745 #ifdef EXPERIMENTAL_DMARC
1746 /* initialize libopendmarc */
1747 dmarc_up = dmarc_init();
1748 #endif
1749
1750 /* Remember the time of reception. Exim uses time+pid for uniqueness of message
1751 ids, and fractions of a second are required. See the comments that precede the
1752 message id creation below. */
1753
1754 (void)gettimeofday(&message_id_tv, NULL);
1755
1756 /* For other uses of the received time we can operate with granularity of one
1757 second, and for that we use the global variable received_time. This is for
1758 things like ultimate message timeouts. */
1759
1760 received_time = message_id_tv;
1761
1762 /* If SMTP input, set the special handler for timeouts. The alarm() calls
1763 happen in the smtp_getc() function when it refills its buffer. */
1764
1765 had_data_timeout = 0;
1766 if (smtp_input)
1767 os_non_restarting_signal(SIGALRM, data_timeout_handler);
1768
1769 /* If not SMTP input, timeout happens only if configured, and we just set a
1770 single timeout for the whole message. */
1771
1772 else if (receive_timeout > 0)
1773 {
1774 os_non_restarting_signal(SIGALRM, data_timeout_handler);
1775 ALARM(receive_timeout);
1776 }
1777
1778 /* SIGTERM and SIGINT are caught always. */
1779
1780 had_data_sigint = 0;
1781 signal(SIGTERM, data_sigterm_sigint_handler);
1782 signal(SIGINT, data_sigterm_sigint_handler);
1783
1784 /* Header lines in messages are not supposed to be very long, though when
1785 unfolded, to: and cc: headers can take up a lot of store. We must also cope
1786 with the possibility of junk being thrown at us. Start by getting 256 bytes for
1787 storing the header, and extend this as necessary using string_cat().
1788
1789 To cope with total lunacies, impose an upper limit on the length of the header
1790 section of the message, as otherwise the store will fill up. We must also cope
1791 with the possibility of binary zeros in the data. Hence we cannot use fgets().
1792 Folded header lines are joined into one string, leaving the '\n' characters
1793 inside them, so that writing them out reproduces the input.
1794
1795 Loop for each character of each header; the next structure for chaining the
1796 header is set up already, with ptr the offset of the next character in
1797 next->text. */
1798
1799 for (;;)
1800 {
1801 int ch = (receive_getc)(GETC_BUFFER_UNLIMITED);
1802
1803 /* If we hit EOF on a SMTP connection, it's an error, since incoming
1804 SMTP must have a correct "." terminator. */
1805
1806 if (ch == EOF && smtp_input /* && !smtp_batched_input */)
1807 {
1808 smtp_reply = handle_lost_connection(US" (header)");
1809 smtp_yield = FALSE;
1810 goto TIDYUP; /* Skip to end of function */
1811 }
1812
1813 /* See if we are at the current header's size limit - there must be at least
1814 four bytes left. This allows for the new character plus a zero, plus two for
1815 extra insertions when we are playing games with dots and carriage returns. If
1816 we are at the limit, extend the text buffer. This could have been done
1817 automatically using string_cat() but because this is a tightish loop storing
1818 only one character at a time, we choose to do it inline. Normally
1819 store_extend() will be able to extend the block; only at the end of a big
1820 store block will a copy be needed. To handle the case of very long headers
1821 (and sometimes lunatic messages can have ones that are 100s of K long) we
1822 call store_release() for strings that have been copied - if the string is at
1823 the start of a block (and therefore the only thing in it, because we aren't
1824 doing any other gets), the block gets freed. We can only do this release if
1825 there were no allocations since the once that we want to free. */
1826
1827 if (ptr >= header_size - 4)
1828 {
1829 int oldsize = header_size;
1830
1831 if (header_size >= INT_MAX/2)
1832 goto OVERSIZE;
1833 header_size *= 2;
1834
1835 if (!store_extend(next->text, oldsize, header_size))
1836 next->text = store_newblock(next->text, header_size, ptr);
1837 }
1838
1839 /* Cope with receiving a binary zero. There is dispute about whether
1840 these should be allowed in RFC 822 messages. The middle view is that they
1841 should not be allowed in headers, at least. Exim takes this attitude at
1842 the moment. We can't just stomp on them here, because we don't know that
1843 this line is a header yet. Set a flag to cause scanning later. */
1844
1845 if (ch == 0) had_zero++;
1846
1847 /* Test for termination. Lines in remote SMTP are terminated by CRLF, while
1848 those from data files use just LF. Treat LF in local SMTP input as a
1849 terminator too. Treat EOF as a line terminator always. */
1850
1851 if (ch == EOF) goto EOL;
1852
1853 /* FUDGE: There are sites out there that don't send CRs before their LFs, and
1854 other MTAs accept this. We are therefore forced into this "liberalisation"
1855 too, so we accept LF as a line terminator whatever the source of the message.
1856 However, if the first line of the message ended with a CRLF, we treat a bare
1857 LF specially by inserting a white space after it to ensure that the header
1858 line is not terminated. */
1859
1860 if (ch == '\n')
1861 {
1862 if (first_line_ended_crlf == TRUE_UNSET) first_line_ended_crlf = FALSE;
1863 else if (first_line_ended_crlf) receive_ungetc(' ');
1864 goto EOL;
1865 }
1866
1867 /* This is not the end of the line. If this is SMTP input and this is
1868 the first character in the line and it is a "." character, ignore it.
1869 This implements the dot-doubling rule, though header lines starting with
1870 dots aren't exactly common. They are legal in RFC 822, though. If the
1871 following is CRLF or LF, this is the line that that terminates the
1872 entire message. We set message_ended to indicate this has happened (to
1873 prevent further reading), and break out of the loop, having freed the
1874 empty header, and set next = NULL to indicate no data line. */
1875
1876 if (ptr == 0 && ch == '.' && f.dot_ends)
1877 {
1878 ch = (receive_getc)(GETC_BUFFER_UNLIMITED);
1879 if (ch == '\r')
1880 {
1881 ch = (receive_getc)(GETC_BUFFER_UNLIMITED);
1882 if (ch != '\n')
1883 {
1884 receive_ungetc(ch);
1885 ch = '\r'; /* Revert to CR */
1886 }
1887 }
1888 if (ch == '\n')
1889 {
1890 message_ended = END_DOT;
1891 store_reset(next);
1892 next = NULL;
1893 break; /* End character-reading loop */
1894 }
1895
1896 /* For non-SMTP input, the dot at the start of the line was really a data
1897 character. What is now in ch is the following character. We guaranteed
1898 enough space for this above. */
1899
1900 if (!smtp_input)
1901 {
1902 next->text[ptr++] = '.';
1903 message_size++;
1904 }
1905 }
1906
1907 /* If CR is immediately followed by LF, end the line, ignoring the CR, and
1908 remember this case if this is the first line ending. */
1909
1910 if (ch == '\r')
1911 {
1912 ch = (receive_getc)(GETC_BUFFER_UNLIMITED);
1913 if (ch == '\n')
1914 {
1915 if (first_line_ended_crlf == TRUE_UNSET) first_line_ended_crlf = TRUE;
1916 goto EOL;
1917 }
1918
1919 /* Otherwise, put back the character after CR, and turn the bare CR
1920 into LF SP. */
1921
1922 ch = (receive_ungetc)(ch);
1923 next->text[ptr++] = '\n';
1924 message_size++;
1925 ch = ' ';
1926 }
1927
1928 /* We have a data character for the header line. */
1929
1930 next->text[ptr++] = ch; /* Add to buffer */
1931 message_size++; /* Total message size so far */
1932
1933 /* Handle failure due to a humungously long header section. The >= allows
1934 for the terminating \n. Add what we have so far onto the headers list so
1935 that it gets reflected in any error message, and back up the just-read
1936 character. */
1937
1938 if (message_size >= header_maxsize)
1939 {
1940 OVERSIZE:
1941 next->text[ptr] = 0;
1942 next->slen = ptr;
1943 next->type = htype_other;
1944 next->next = NULL;
1945 header_last->next = next;
1946 header_last = next;
1947
1948 log_write(0, LOG_MAIN, "ridiculously long message header received from "
1949 "%s (more than %d characters): message abandoned",
1950 f.sender_host_unknown ? sender_ident : sender_fullhost, header_maxsize);
1951
1952 if (smtp_input)
1953 {
1954 smtp_reply = US"552 Message header is ridiculously long";
1955 receive_swallow_smtp();
1956 goto TIDYUP; /* Skip to end of function */
1957 }
1958
1959 else
1960 {
1961 give_local_error(ERRMESS_VLONGHEADER,
1962 string_sprintf("message header longer than %d characters received: "
1963 "message not accepted", header_maxsize), US"", error_rc, stdin,
1964 header_list->next);
1965 /* Does not return */
1966 }
1967 }
1968
1969 continue; /* With next input character */
1970
1971 /* End of header line reached */
1972
1973 EOL:
1974
1975 /* Keep track of lines for BSMTP errors and overall message_linecount. */
1976
1977 receive_linecount++;
1978 message_linecount++;
1979
1980 /* Keep track of maximum line length */
1981
1982 if (ptr - prevlines_length > max_received_linelength)
1983 max_received_linelength = ptr - prevlines_length;
1984 prevlines_length = ptr + 1;
1985
1986 /* Now put in the terminating newline. There is always space for
1987 at least two more characters. */
1988
1989 next->text[ptr++] = '\n';
1990 message_size++;
1991
1992 /* A blank line signals the end of the headers; release the unwanted
1993 space and set next to NULL to indicate this. */
1994
1995 if (ptr == 1)
1996 {
1997 store_reset(next);
1998 next = NULL;
1999 break;
2000 }
2001
2002 /* There is data in the line; see if the next input character is a
2003 whitespace character. If it is, we have a continuation of this header line.
2004 There is always space for at least one character at this point. */
2005
2006 if (ch != EOF)
2007 {
2008 int nextch = (receive_getc)(GETC_BUFFER_UNLIMITED);
2009 if (nextch == ' ' || nextch == '\t')
2010 {
2011 next->text[ptr++] = nextch;
2012 if (++message_size >= header_maxsize)
2013 goto OVERSIZE;
2014 continue; /* Iterate the loop */
2015 }
2016 else if (nextch != EOF) (receive_ungetc)(nextch); /* For next time */
2017 else ch = EOF; /* Cause main loop to exit at end */
2018 }
2019
2020 /* We have got to the real line end. Terminate the string and release store
2021 beyond it. If it turns out to be a real header, internal binary zeros will
2022 be squashed later. */
2023
2024 next->text[ptr] = 0;
2025 next->slen = ptr;
2026 store_reset(next->text + ptr + 1);
2027
2028 /* Check the running total size against the overall message size limit. We
2029 don't expect to fail here, but if the overall limit is set less than MESSAGE_
2030 MAXSIZE and a big header is sent, we want to catch it. Just stop reading
2031 headers - the code to read the body will then also hit the buffer. */
2032
2033 if (message_size > thismessage_size_limit) break;
2034
2035 /* A line that is not syntactically correct for a header also marks
2036 the end of the headers. In this case, we leave next containing the
2037 first data line. This might actually be several lines because of the
2038 continuation logic applied above, but that doesn't matter.
2039
2040 It turns out that smail, and presumably sendmail, accept leading lines
2041 of the form
2042
2043 From ph10 Fri Jan 5 12:35 GMT 1996
2044
2045 in messages. The "mail" command on Solaris 2 sends such lines. I cannot
2046 find any documentation of this, but for compatibility it had better be
2047 accepted. Exim restricts it to the case of non-smtp messages, and
2048 treats it as an alternative to the -f command line option. Thus it is
2049 ignored except for trusted users or filter testing. Otherwise it is taken
2050 as the sender address, unless -f was used (sendmail compatibility).
2051
2052 It further turns out that some UUCPs generate the From_line in a different
2053 format, e.g.
2054
2055 From ph10 Fri, 7 Jan 97 14:00:00 GMT
2056
2057 The regex for matching these things is now capable of recognizing both
2058 formats (including 2- and 4-digit years in the latter). In fact, the regex
2059 is now configurable, as is the expansion string to fish out the sender.
2060
2061 Even further on it has been discovered that some broken clients send
2062 these lines in SMTP messages. There is now an option to ignore them from
2063 specified hosts or networks. Sigh. */
2064
2065 if ( header_last == header_list
2066 && ( !smtp_input
2067 || ( sender_host_address
2068 && verify_check_host(&ignore_fromline_hosts) == OK
2069 )
2070 || (!sender_host_address && ignore_fromline_local)
2071 )
2072 && regex_match_and_setup(regex_From, next->text, 0, -1)
2073 )
2074 {
2075 if (!f.sender_address_forced)
2076 {
2077 uschar *uucp_sender = expand_string(uucp_from_sender);
2078 if (!uucp_sender)
2079 log_write(0, LOG_MAIN|LOG_PANIC,
2080 "expansion of \"%s\" failed after matching "
2081 "\"From \" line: %s", uucp_from_sender, expand_string_message);
2082 else
2083 {
2084 int start, end, domain;
2085 uschar *errmess;
2086 uschar *newsender = parse_extract_address(uucp_sender, &errmess,
2087 &start, &end, &domain, TRUE);
2088 if (newsender)
2089 {
2090 if (domain == 0 && newsender[0] != 0)
2091 newsender = rewrite_address_qualify(newsender, FALSE);
2092
2093 if (filter_test != FTEST_NONE || receive_check_set_sender(newsender))
2094 {
2095 sender_address = newsender;
2096
2097 if (f.trusted_caller || filter_test != FTEST_NONE)
2098 {
2099 authenticated_sender = NULL;
2100 originator_name = US"";
2101 f.sender_local = FALSE;
2102 }
2103
2104 if (filter_test != FTEST_NONE)
2105 printf("Sender taken from \"From \" line\n");
2106 }
2107 }
2108 }
2109 }
2110 }
2111
2112 /* Not a leading "From " line. Check to see if it is a valid header line.
2113 Header names may contain any non-control characters except space and colon,
2114 amazingly. */
2115
2116 else
2117 {
2118 uschar *p = next->text;
2119
2120 /* If not a valid header line, break from the header reading loop, leaving
2121 next != NULL, indicating that it holds the first line of the body. */
2122
2123 if (isspace(*p)) break;
2124 while (mac_isgraph(*p) && *p != ':') p++;
2125 while (isspace(*p)) p++;
2126 if (*p != ':')
2127 {
2128 body_zerocount = had_zero;
2129 break;
2130 }
2131
2132 /* We have a valid header line. If there were any binary zeroes in
2133 the line, stomp on them here. */
2134
2135 if (had_zero > 0)
2136 for (p = next->text; p < next->text + ptr; p++) if (*p == 0) *p = '?';
2137
2138 /* It is perfectly legal to have an empty continuation line
2139 at the end of a header, but it is confusing to humans
2140 looking at such messages, since it looks like a blank line.
2141 Reduce confusion by removing redundant white space at the
2142 end. We know that there is at least one printing character
2143 (the ':' tested for above) so there is no danger of running
2144 off the end. */
2145
2146 p = next->text + ptr - 2;
2147 for (;;)
2148 {
2149 while (*p == ' ' || *p == '\t') p--;
2150 if (*p != '\n') break;
2151 ptr = (p--) - next->text + 1;
2152 message_size -= next->slen - ptr;
2153 next->text[ptr] = 0;
2154 next->slen = ptr;
2155 }
2156
2157 /* Add the header to the chain */
2158
2159 next->type = htype_other;
2160 next->next = NULL;
2161 header_last->next = next;
2162 header_last = next;
2163
2164 /* Check the limit for individual line lengths. This comes after adding to
2165 the chain so that the failing line is reflected if a bounce is generated
2166 (for a local message). */
2167
2168 if (header_line_maxsize > 0 && next->slen > header_line_maxsize)
2169 {
2170 log_write(0, LOG_MAIN, "overlong message header line received from "
2171 "%s (more than %d characters): message abandoned",
2172 f.sender_host_unknown ? sender_ident : sender_fullhost,
2173 header_line_maxsize);
2174
2175 if (smtp_input)
2176 {
2177 smtp_reply = US"552 A message header line is too long";
2178 receive_swallow_smtp();
2179 goto TIDYUP; /* Skip to end of function */
2180 }
2181
2182 else
2183 give_local_error(ERRMESS_VLONGHDRLINE,
2184 string_sprintf("message header line longer than %d characters "
2185 "received: message not accepted", header_line_maxsize), US"",
2186 error_rc, stdin, header_list->next);
2187 /* Does not return */
2188 }
2189
2190 /* Note if any resent- fields exist. */
2191
2192 if (!resents_exist && strncmpic(next->text, US"resent-", 7) == 0)
2193 {
2194 resents_exist = TRUE;
2195 resent_prefix = US"Resent-";
2196 }
2197 }
2198
2199 /* Reject CHUNKING messages that do not CRLF their first header line */
2200
2201 if (!first_line_ended_crlf && chunking_state > CHUNKING_OFFERED)
2202 {
2203 log_write(L_size_reject, LOG_MAIN|LOG_REJECT, "rejected from <%s>%s%s%s%s: "
2204 "Non-CRLF-terminated header, under CHUNKING: message abandoned",
2205 sender_address,
2206 sender_fullhost ? " H=" : "", sender_fullhost ? sender_fullhost : US"",
2207 sender_ident ? " U=" : "", sender_ident ? sender_ident : US"");
2208 smtp_printf("552 Message header not CRLF terminated\r\n", FALSE);
2209 bdat_flush_data();
2210 smtp_reply = US"";
2211 goto TIDYUP; /* Skip to end of function */
2212 }
2213
2214 /* The line has been handled. If we have hit EOF, break out of the loop,
2215 indicating no pending data line. */
2216
2217 if (ch == EOF) { next = NULL; break; }
2218
2219 /* Set up for the next header */
2220
2221 header_size = 256;
2222 next = store_get(sizeof(header_line));
2223 next->text = store_get(header_size);
2224 ptr = 0;
2225 had_zero = 0;
2226 prevlines_length = 0;
2227 } /* Continue, starting to read the next header */
2228
2229 /* At this point, we have read all the headers into a data structure in main
2230 store. The first header is still the dummy placeholder for the Received: header
2231 we are going to generate a bit later on. If next != NULL, it contains the first
2232 data line - which terminated the headers before reaching a blank line (not the
2233 normal case). */
2234
2235 DEBUG(D_receive)
2236 {
2237 debug_printf(">>Headers received:\n");
2238 for (h = header_list->next; h; h = h->next)
2239 debug_printf("%s", h->text);
2240 debug_printf("\n");
2241 }
2242
2243 /* End of file on any SMTP connection is an error. If an incoming SMTP call
2244 is dropped immediately after valid headers, the next thing we will see is EOF.
2245 We must test for this specially, as further down the reading of the data is
2246 skipped if already at EOF. */
2247
2248 if (smtp_input && (receive_feof)())
2249 {
2250 smtp_reply = handle_lost_connection(US" (after header)");
2251 smtp_yield = FALSE;
2252 goto TIDYUP; /* Skip to end of function */
2253 }
2254
2255 /* If this is a filter test run and no headers were read, output a warning
2256 in case there is a mistake in the test message. */
2257
2258 if (filter_test != FTEST_NONE && header_list->next == NULL)
2259 printf("Warning: no message headers read\n");
2260
2261
2262 /* Scan the headers to identify them. Some are merely marked for later
2263 processing; some are dealt with here. */
2264
2265 for (h = header_list->next; h; h = h->next)
2266 {
2267 BOOL is_resent = strncmpic(h->text, US"resent-", 7) == 0;
2268 if (is_resent) contains_resent_headers = TRUE;
2269
2270 switch (header_checkname(h, is_resent))
2271 {
2272 case htype_bcc:
2273 h->type = htype_bcc; /* Both Bcc: and Resent-Bcc: */
2274 break;
2275
2276 case htype_cc:
2277 h->type = htype_cc; /* Both Cc: and Resent-Cc: */
2278 break;
2279
2280 /* Record whether a Date: or Resent-Date: header exists, as appropriate. */
2281
2282 case htype_date:
2283 if (!resents_exist || is_resent) date_header_exists = TRUE;
2284 break;
2285
2286 /* Same comments as about Return-Path: below. */
2287
2288 case htype_delivery_date:
2289 if (delivery_date_remove) h->type = htype_old;
2290 break;
2291
2292 /* Same comments as about Return-Path: below. */
2293
2294 case htype_envelope_to:
2295 if (envelope_to_remove) h->type = htype_old;
2296 break;
2297
2298 /* Mark all "From:" headers so they get rewritten. Save the one that is to
2299 be used for Sender: checking. For Sendmail compatibility, if the "From:"
2300 header consists of just the login id of the user who called Exim, rewrite
2301 it with the gecos field first. Apply this rule to Resent-From: if there
2302 are resent- fields. */
2303
2304 case htype_from:
2305 h->type = htype_from;
2306 if (!resents_exist || is_resent)
2307 {
2308 from_header = h;
2309 if (!smtp_input)
2310 {
2311 int len;
2312 uschar *s = Ustrchr(h->text, ':') + 1;
2313 while (isspace(*s)) s++;
2314 len = h->slen - (s - h->text) - 1;
2315 if (Ustrlen(originator_login) == len &&
2316 strncmpic(s, originator_login, len) == 0)
2317 {
2318 uschar *name = is_resent? US"Resent-From" : US"From";
2319 header_add(htype_from, "%s: %s <%s@%s>\n", name, originator_name,
2320 originator_login, qualify_domain_sender);
2321 from_header = header_last;
2322 h->type = htype_old;
2323 DEBUG(D_receive|D_rewrite)
2324 debug_printf("rewrote \"%s:\" header using gecos\n", name);
2325 }
2326 }
2327 }
2328 break;
2329
2330 /* Identify the Message-id: header for generating "in-reply-to" in the
2331 autoreply transport. For incoming logging, save any resent- value. In both
2332 cases, take just the first of any multiples. */
2333
2334 case htype_id:
2335 if (!msgid_header && (!resents_exist || is_resent))
2336 {
2337 msgid_header = h;
2338 h->type = htype_id;
2339 }
2340 break;
2341
2342 /* Flag all Received: headers */
2343
2344 case htype_received:
2345 h->type = htype_received;
2346 received_count++;
2347 break;
2348
2349 /* "Reply-to:" is just noted (there is no resent-reply-to field) */
2350
2351 case htype_reply_to:
2352 h->type = htype_reply_to;
2353 break;
2354
2355 /* The Return-path: header is supposed to be added to messages when
2356 they leave the SMTP system. We shouldn't receive messages that already
2357 contain Return-path. However, since Exim generates Return-path: on
2358 local delivery, resent messages may well contain it. We therefore
2359 provide an option (which defaults on) to remove any Return-path: headers
2360 on input. Removal actually means flagging as "old", which prevents the
2361 header being transmitted with the message. */
2362
2363 case htype_return_path:
2364 if (return_path_remove) h->type = htype_old;
2365
2366 /* If we are testing a mail filter file, use the value of the
2367 Return-Path: header to set up the return_path variable, which is not
2368 otherwise set. However, remove any <> that surround the address
2369 because the variable doesn't have these. */
2370
2371 if (filter_test != FTEST_NONE)
2372 {
2373 uschar *start = h->text + 12;
2374 uschar *end = start + Ustrlen(start);
2375 while (isspace(*start)) start++;
2376 while (end > start && isspace(end[-1])) end--;
2377 if (*start == '<' && end[-1] == '>')
2378 {
2379 start++;
2380 end--;
2381 }
2382 return_path = string_copyn(start, end - start);
2383 printf("Return-path taken from \"Return-path:\" header line\n");
2384 }
2385 break;
2386
2387 /* If there is a "Sender:" header and the message is locally originated,
2388 and from an untrusted caller and suppress_local_fixups is not set, or if we
2389 are in submission mode for a remote message, mark it "old" so that it will
2390 not be transmitted with the message, unless active_local_sender_retain is
2391 set. (This can only be true if active_local_from_check is false.) If there
2392 are any resent- headers in the message, apply this rule to Resent-Sender:
2393 instead of Sender:. Messages with multiple resent- header sets cannot be
2394 tidily handled. (For this reason, at least one MUA - Pine - turns old
2395 resent- headers into X-resent- headers when resending, leaving just one
2396 set.) */
2397
2398 case htype_sender:
2399 h->type = !f.active_local_sender_retain
2400 && ( f.sender_local && !f.trusted_caller && !f.suppress_local_fixups
2401 || f.submission_mode
2402 )
2403 && (!resents_exist || is_resent)
2404 ? htype_old : htype_sender;
2405 break;
2406
2407 /* Remember the Subject: header for logging. There is no Resent-Subject */
2408
2409 case htype_subject:
2410 subject_header = h;
2411 break;
2412
2413 /* "To:" gets flagged, and the existence of a recipient header is noted,
2414 whether it's resent- or not. */
2415
2416 case htype_to:
2417 h->type = htype_to;
2418 /****
2419 to_or_cc_header_exists = TRUE;
2420 ****/
2421 break;
2422 }
2423 }
2424
2425 /* Extract recipients from the headers if that is required (the -t option).
2426 Note that this is documented as being done *before* any address rewriting takes
2427 place. There are two possibilities:
2428
2429 (1) According to sendmail documentation for Solaris, IRIX, and HP-UX, any
2430 recipients already listed are to be REMOVED from the message. Smail 3 works
2431 like this. We need to build a non-recipients tree for that list, because in
2432 subsequent processing this data is held in a tree and that's what the
2433 spool_write_header() function expects. Make sure that non-recipient addresses
2434 are fully qualified and rewritten if necessary.
2435
2436 (2) According to other sendmail documentation, -t ADDS extracted recipients to
2437 those in the command line arguments (and it is rumoured some other MTAs do
2438 this). Therefore, there is an option to make Exim behave this way.
2439
2440 *** Notes on "Resent-" header lines ***
2441
2442 The presence of resent-headers in the message makes -t horribly ambiguous.
2443 Experiments with sendmail showed that it uses recipients for all resent-
2444 headers, totally ignoring the concept of "sets of resent- headers" as described
2445 in RFC 2822 section 3.6.6. Sendmail also amalgamates them into a single set
2446 with all the addresses in one instance of each header.
2447
2448 This seems to me not to be at all sensible. Before release 4.20, Exim 4 gave an
2449 error for -t if there were resent- headers in the message. However, after a
2450 discussion on the mailing list, I've learned that there are MUAs that use
2451 resent- headers with -t, and also that the stuff about sets of resent- headers
2452 and their ordering in RFC 2822 is generally ignored. An MUA that submits a
2453 message with -t and resent- header lines makes sure that only *its* resent-
2454 headers are present; previous ones are often renamed as X-resent- for example.
2455
2456 Consequently, Exim has been changed so that, if any resent- header lines are
2457 present, the recipients are taken from all of the appropriate resent- lines,
2458 and not from the ordinary To:, Cc:, etc. */
2459
2460 if (extract_recip)
2461 {
2462 int rcount = 0;
2463 error_block **bnext = &bad_addresses;
2464
2465 if (extract_addresses_remove_arguments)
2466 {
2467 while (recipients_count-- > 0)
2468 {
2469 uschar *s = rewrite_address(recipients_list[recipients_count].address,
2470 TRUE, TRUE, global_rewrite_rules, rewrite_existflags);
2471 tree_add_nonrecipient(s);
2472 }
2473 recipients_list = NULL;
2474 recipients_count = recipients_list_max = 0;
2475 }
2476
2477 /* Now scan the headers */
2478
2479 for (h = header_list->next; h; h = h->next)
2480 {
2481 if ((h->type == htype_to || h->type == htype_cc || h->type == htype_bcc) &&
2482 (!contains_resent_headers || strncmpic(h->text, US"resent-", 7) == 0))
2483 {
2484 uschar *s = Ustrchr(h->text, ':') + 1;
2485 while (isspace(*s)) s++;
2486
2487 f.parse_allow_group = TRUE; /* Allow address group syntax */
2488
2489 while (*s != 0)
2490 {
2491 uschar *ss = parse_find_address_end(s, FALSE);
2492 uschar *recipient, *errmess, *p, *pp;
2493 int start, end, domain;
2494
2495 /* Check on maximum */
2496
2497 if (recipients_max > 0 && ++rcount > recipients_max)
2498 give_local_error(ERRMESS_TOOMANYRECIP, US"too many recipients",
2499 US"message rejected: ", error_rc, stdin, NULL);
2500 /* Does not return */
2501
2502 /* Make a copy of the address, and remove any internal newlines. These
2503 may be present as a result of continuations of the header line. The
2504 white space that follows the newline must not be removed - it is part
2505 of the header. */
2506
2507 pp = recipient = store_get(ss - s + 1);
2508 for (p = s; p < ss; p++) if (*p != '\n') *pp++ = *p;
2509 *pp = 0;
2510
2511 #ifdef SUPPORT_I18N
2512 {
2513 BOOL b = allow_utf8_domains;
2514 allow_utf8_domains = TRUE;
2515 #endif
2516 recipient = parse_extract_address(recipient, &errmess, &start, &end,
2517 &domain, FALSE);
2518
2519 #ifdef SUPPORT_I18N
2520 if (string_is_utf8(recipient))
2521 message_smtputf8 = TRUE;
2522 else
2523 allow_utf8_domains = b;
2524 }
2525 #endif
2526
2527 /* Keep a list of all the bad addresses so we can send a single
2528 error message at the end. However, an empty address is not an error;
2529 just ignore it. This can come from an empty group list like
2530
2531 To: Recipients of list:;
2532
2533 If there are no recipients at all, an error will occur later. */
2534
2535 if (recipient == NULL && Ustrcmp(errmess, "empty address") != 0)
2536 {
2537 int len = Ustrlen(s);
2538 error_block *b = store_get(sizeof(error_block));
2539 while (len > 0 && isspace(s[len-1])) len--;
2540 b->next = NULL;
2541 b->text1 = string_printing(string_copyn(s, len));
2542 b->text2 = errmess;
2543 *bnext = b;
2544 bnext = &(b->next);
2545 }
2546
2547 /* If the recipient is already in the nonrecipients tree, it must
2548 have appeared on the command line with the option extract_addresses_
2549 remove_arguments set. Do not add it to the recipients, and keep a note
2550 that this has happened, in order to give a better error if there are
2551 no recipients left. */
2552
2553 else if (recipient != NULL)
2554 {
2555 if (tree_search(tree_nonrecipients, recipient) == NULL)
2556 receive_add_recipient(recipient, -1);
2557 else
2558 extracted_ignored = TRUE;
2559 }
2560
2561 /* Move on past this address */
2562
2563 s = ss + (*ss? 1:0);
2564 while (isspace(*s)) s++;
2565 } /* Next address */
2566
2567 f.parse_allow_group = FALSE; /* Reset group syntax flags */
2568 f.parse_found_group = FALSE;
2569
2570 /* If this was the bcc: header, mark it "old", which means it
2571 will be kept on the spool, but not transmitted as part of the
2572 message. */
2573
2574 if (h->type == htype_bcc) h->type = htype_old;
2575 } /* For appropriate header line */
2576 } /* For each header line */
2577
2578 }
2579
2580 /* Now build the unique message id. This has changed several times over the
2581 lifetime of Exim. This description was rewritten for Exim 4.14 (February 2003).
2582 Retaining all the history in the comment has become too unwieldy - read
2583 previous release sources if you want it.
2584
2585 The message ID has 3 parts: tttttt-pppppp-ss. Each part is a number in base 62.
2586 The first part is the current time, in seconds. The second part is the current
2587 pid. Both are large enough to hold 32-bit numbers in base 62. The third part
2588 can hold a number in the range 0-3843. It used to be a computed sequence
2589 number, but is now the fractional component of the current time in units of
2590 1/2000 of a second (i.e. a value in the range 0-1999). After a message has been
2591 received, Exim ensures that the timer has ticked at the appropriate level
2592 before proceeding, to avoid duplication if the pid happened to be re-used
2593 within the same time period. It seems likely that most messages will take at
2594 least half a millisecond to be received, so no delay will normally be
2595 necessary. At least for some time...
2596
2597 There is a modification when localhost_number is set. Formerly this was allowed
2598 to be as large as 255. Now it is restricted to the range 0-16, and the final
2599 component of the message id becomes (localhost_number * 200) + fractional time
2600 in units of 1/200 of a second (i.e. a value in the range 0-3399).
2601
2602 Some not-really-Unix operating systems use case-insensitive file names (Darwin,
2603 Cygwin). For these, we have to use base 36 instead of base 62. Luckily, this
2604 still allows the tttttt field to hold a large enough number to last for some
2605 more decades, and the final two-digit field can hold numbers up to 1295, which
2606 is enough for milliseconds (instead of 1/2000 of a second).
2607
2608 However, the pppppp field cannot hold a 32-bit pid, but it can hold a 31-bit
2609 pid, so it is probably safe because pids have to be positive. The
2610 localhost_number is restricted to 0-10 for these hosts, and when it is set, the
2611 final field becomes (localhost_number * 100) + fractional time in centiseconds.
2612
2613 Note that string_base62() returns its data in a static storage block, so it
2614 must be copied before calling string_base62() again. It always returns exactly
2615 6 characters.
2616
2617 There doesn't seem to be anything in the RFC which requires a message id to
2618 start with a letter, but Smail was changed to ensure this. The external form of
2619 the message id (as supplied by string expansion) therefore starts with an
2620 additional leading 'E'. The spool file names do not include this leading
2621 letter and it is not used internally.
2622
2623 NOTE: If ever the format of message ids is changed, the regular expression for
2624 checking that a string is in this format must be updated in a corresponding
2625 way. It appears in the initializing code in exim.c. The macro MESSAGE_ID_LENGTH
2626 must also be changed to reflect the correct string length. The queue-sort code
2627 needs to know the layout. Then, of course, other programs that rely on the
2628 message id format will need updating too. */
2629
2630 Ustrncpy(message_id, string_base62((long int)(message_id_tv.tv_sec)), 6);
2631 message_id[6] = '-';
2632 Ustrncpy(message_id + 7, string_base62((long int)getpid()), 6);
2633
2634 /* Deal with the case where the host number is set. The value of the number was
2635 checked when it was read, to ensure it isn't too big. The timing granularity is
2636 left in id_resolution so that an appropriate wait can be done after receiving
2637 the message, if necessary (we hope it won't be). */
2638
2639 if (host_number_string)
2640 {
2641 id_resolution = BASE_62 == 62 ? 5000 : 10000;
2642 sprintf(CS(message_id + MESSAGE_ID_LENGTH - 3), "-%2s",
2643 string_base62((long int)(
2644 host_number * (1000000/id_resolution) +
2645 message_id_tv.tv_usec/id_resolution)) + 4);
2646 }
2647
2648 /* Host number not set: final field is just the fractional time at an
2649 appropriate resolution. */
2650
2651 else
2652 {
2653 id_resolution = BASE_62 == 62 ? 500 : 1000;
2654 sprintf(CS(message_id + MESSAGE_ID_LENGTH - 3), "-%2s",
2655 string_base62((long int)(message_id_tv.tv_usec/id_resolution)) + 4);
2656 }
2657
2658 /* Add the current message id onto the current process info string if
2659 it will fit. */
2660
2661 (void)string_format(process_info + process_info_len,
2662 PROCESS_INFO_SIZE - process_info_len, " id=%s", message_id);
2663
2664 /* If we are using multiple input directories, set up the one for this message
2665 to be the least significant base-62 digit of the time of arrival. Otherwise
2666 ensure that it is an empty string. */
2667
2668 message_subdir[0] = split_spool_directory ? message_id[5] : 0;
2669
2670 /* Now that we have the message-id, if there is no message-id: header, generate
2671 one, but only for local (without suppress_local_fixups) or submission mode
2672 messages. This can be user-configured if required, but we had better flatten
2673 any illegal characters therein. */
2674
2675 if ( !msgid_header
2676 && ((!sender_host_address && !f.suppress_local_fixups) || f.submission_mode))
2677 {
2678 uschar *p;
2679 uschar *id_text = US"";
2680 uschar *id_domain = primary_hostname;
2681
2682 /* Permit only letters, digits, dots, and hyphens in the domain */
2683
2684 if (message_id_domain)
2685 {
2686 uschar *new_id_domain = expand_string(message_id_domain);
2687 if (!new_id_domain)
2688 {
2689 if (!f.expand_string_forcedfail)
2690 log_write(0, LOG_MAIN|LOG_PANIC,
2691 "expansion of \"%s\" (message_id_header_domain) "
2692 "failed: %s", message_id_domain, expand_string_message);
2693 }
2694 else if (*new_id_domain)
2695 {
2696 id_domain = new_id_domain;
2697 for (p = id_domain; *p; p++)
2698 if (!isalnum(*p) && *p != '.') *p = '-'; /* No need to test '-' ! */
2699 }
2700 }
2701
2702 /* Permit all characters except controls and RFC 2822 specials in the
2703 additional text part. */
2704
2705 if (message_id_text)
2706 {
2707 uschar *new_id_text = expand_string(message_id_text);
2708 if (!new_id_text)
2709 {
2710 if (!f.expand_string_forcedfail)
2711 log_write(0, LOG_MAIN|LOG_PANIC,
2712 "expansion of \"%s\" (message_id_header_text) "
2713 "failed: %s", message_id_text, expand_string_message);
2714 }
2715 else if (*new_id_text)
2716 {
2717 id_text = new_id_text;
2718 for (p = id_text; *p; p++) if (mac_iscntrl_or_special(*p)) *p = '-';
2719 }
2720 }
2721
2722 /* Add the header line
2723 * Resent-* headers are prepended, per RFC 5322 3.6.6. Non-Resent-* are
2724 * appended, to preserve classical expectations of header ordering. */
2725
2726 header_add_at_position(!resents_exist, NULL, FALSE, htype_id,
2727 "%sMessage-Id: <%s%s%s@%s>\n", resent_prefix, message_id_external,
2728 (*id_text == 0)? "" : ".", id_text, id_domain);
2729 }
2730
2731 /* If we are to log recipients, keep a copy of the raw ones before any possible
2732 rewriting. Must copy the count, because later ACLs and the local_scan()
2733 function may mess with the real recipients. */
2734
2735 if (LOGGING(received_recipients))
2736 {
2737 raw_recipients = store_get(recipients_count * sizeof(uschar *));
2738 for (i = 0; i < recipients_count; i++)
2739 raw_recipients[i] = string_copy(recipients_list[i].address);
2740 raw_recipients_count = recipients_count;
2741 }
2742
2743 /* Ensure the recipients list is fully qualified and rewritten. Unqualified
2744 recipients will get here only if the conditions were right (allow_unqualified_
2745 recipient is TRUE). */
2746
2747 for (i = 0; i < recipients_count; i++)
2748 recipients_list[i].address =
2749 rewrite_address(recipients_list[i].address, TRUE, TRUE,
2750 global_rewrite_rules, rewrite_existflags);
2751
2752 /* If there is no From: header, generate one for local (without
2753 suppress_local_fixups) or submission_mode messages. If there is no sender
2754 address, but the sender is local or this is a local delivery error, use the
2755 originator login. This shouldn't happen for genuine bounces, but might happen
2756 for autoreplies. The addition of From: must be done *before* checking for the
2757 possible addition of a Sender: header, because untrusted_set_sender allows an
2758 untrusted user to set anything in the envelope (which might then get info
2759 From:) but we still want to ensure a valid Sender: if it is required. */
2760
2761 if ( !from_header
2762 && ((!sender_host_address && !f.suppress_local_fixups) || f.submission_mode))
2763 {
2764 uschar *oname = US"";
2765
2766 /* Use the originator_name if this is a locally submitted message and the
2767 caller is not trusted. For trusted callers, use it only if -F was used to
2768 force its value or if we have a non-SMTP message for which -f was not used
2769 to set the sender. */
2770
2771 if (!sender_host_address)
2772 {
2773 if (!f.trusted_caller || f.sender_name_forced ||
2774 (!smtp_input && !f.sender_address_forced))
2775 oname = originator_name;
2776 }
2777
2778 /* For non-locally submitted messages, the only time we use the originator
2779 name is when it was forced by the /name= option on control=submission. */
2780
2781 else if (submission_name) oname = submission_name;
2782
2783 /* Envelope sender is empty */
2784
2785 if (!*sender_address)
2786 {
2787 uschar *fromstart, *fromend;
2788
2789 fromstart = string_sprintf("%sFrom: %s%s",
2790 resent_prefix, oname, *oname ? " <" : "");
2791 fromend = *oname ? US">" : US"";
2792
2793 if (f.sender_local || f.local_error_message)
2794 header_add(htype_from, "%s%s@%s%s\n", fromstart,
2795 local_part_quote(originator_login), qualify_domain_sender,
2796 fromend);
2797
2798 else if (f.submission_mode && authenticated_id)
2799 {
2800 if (!submission_domain)
2801 header_add(htype_from, "%s%s@%s%s\n", fromstart,
2802 local_part_quote(authenticated_id), qualify_domain_sender,
2803 fromend);
2804
2805 else if (!*submission_domain) /* empty => whole address set */
2806 header_add(htype_from, "%s%s%s\n", fromstart, authenticated_id,
2807 fromend);
2808
2809 else
2810 header_add(htype_from, "%s%s@%s%s\n", fromstart,
2811 local_part_quote(authenticated_id), submission_domain, fromend);
2812
2813 from_header = header_last; /* To get it checked for Sender: */
2814 }
2815 }
2816
2817 /* There is a non-null envelope sender. Build the header using the original
2818 sender address, before any rewriting that might have been done while
2819 verifying it. */
2820
2821 else
2822 {
2823 header_add(htype_from, "%sFrom: %s%s%s%s\n", resent_prefix,
2824 oname,
2825 *oname ? " <" : "",
2826 sender_address_unrewritten ? sender_address_unrewritten : sender_address,
2827 *oname ? ">" : "");
2828
2829 from_header = header_last; /* To get it checked for Sender: */
2830 }
2831 }
2832
2833
2834 /* If the sender is local (without suppress_local_fixups), or if we are in
2835 submission mode and there is an authenticated_id, check that an existing From:
2836 is correct, and if not, generate a Sender: header, unless disabled. Any
2837 previously-existing Sender: header was removed above. Note that sender_local,
2838 as well as being TRUE if the caller of exim is not trusted, is also true if a
2839 trusted caller did not supply a -f argument for non-smtp input. To allow
2840 trusted callers to forge From: without supplying -f, we have to test explicitly
2841 here. If the From: header contains more than one address, then the call to
2842 parse_extract_address fails, and a Sender: header is inserted, as required. */
2843
2844 if ( from_header
2845 && ( f.active_local_from_check
2846 && ( f.sender_local && !f.trusted_caller && !f.suppress_local_fixups
2847 || f.submission_mode && authenticated_id
2848 ) ) )
2849 {
2850 BOOL make_sender = TRUE;
2851 int start, end, domain;
2852 uschar *errmess;
2853 uschar *from_address =
2854 parse_extract_address(Ustrchr(from_header->text, ':') + 1, &errmess,
2855 &start, &end, &domain, FALSE);
2856 uschar *generated_sender_address;
2857
2858 generated_sender_address = f.submission_mode
2859 ? !submission_domain
2860 ? string_sprintf("%s@%s",
2861 local_part_quote(authenticated_id), qualify_domain_sender)
2862 : !*submission_domain /* empty => full address */
2863 ? string_sprintf("%s", authenticated_id)
2864 : string_sprintf("%s@%s",
2865 local_part_quote(authenticated_id), submission_domain)
2866 : string_sprintf("%s@%s",
2867 local_part_quote(originator_login), qualify_domain_sender);
2868
2869 /* Remove permitted prefixes and suffixes from the local part of the From:
2870 address before doing the comparison with the generated sender. */
2871
2872 if (from_address)
2873 {
2874 int slen;
2875 uschar *at = domain ? from_address + domain - 1 : NULL;
2876
2877 if (at) *at = 0;
2878 from_address += route_check_prefix(from_address, local_from_prefix);
2879 slen = route_check_suffix(from_address, local_from_suffix);
2880 if (slen > 0)
2881 {
2882 memmove(from_address+slen, from_address, Ustrlen(from_address)-slen);
2883 from_address += slen;
2884 }
2885 if (at) *at = '@';
2886
2887 if ( strcmpic(generated_sender_address, from_address) == 0
2888 || (!domain && strcmpic(from_address, originator_login) == 0))
2889 make_sender = FALSE;
2890 }
2891
2892 /* We have to cause the Sender header to be rewritten if there are
2893 appropriate rewriting rules. */
2894
2895 if (make_sender)
2896 if (f.submission_mode && !submission_name)
2897 header_add(htype_sender, "%sSender: %s\n", resent_prefix,
2898 generated_sender_address);
2899 else
2900 header_add(htype_sender, "%sSender: %s <%s>\n",
2901 resent_prefix,
2902 f.submission_mode ? submission_name : originator_name,
2903 generated_sender_address);
2904
2905 /* Ensure that a non-null envelope sender address corresponds to the
2906 submission mode sender address. */
2907
2908 if (f.submission_mode && *sender_address)
2909 {
2910 if (!sender_address_unrewritten)
2911 sender_address_unrewritten = sender_address;
2912 sender_address = generated_sender_address;
2913 if (Ustrcmp(sender_address_unrewritten, generated_sender_address) != 0)
2914 log_write(L_address_rewrite, LOG_MAIN,
2915 "\"%s\" from env-from rewritten as \"%s\" by submission mode",
2916 sender_address_unrewritten, generated_sender_address);
2917 }
2918 }
2919
2920 /* If there are any rewriting rules, apply them to the sender address, unless
2921 it has already been rewritten as part of verification for SMTP input. */
2922
2923 if (global_rewrite_rules && !sender_address_unrewritten && *sender_address)
2924 {
2925 sender_address = rewrite_address(sender_address, FALSE, TRUE,
2926 global_rewrite_rules, rewrite_existflags);
2927 DEBUG(D_receive|D_rewrite)
2928 debug_printf("rewritten sender = %s\n", sender_address);
2929 }
2930
2931
2932 /* The headers must be run through rewrite_header(), because it ensures that
2933 addresses are fully qualified, as well as applying any rewriting rules that may
2934 exist.
2935
2936 Qualification of header addresses in a message from a remote host happens only
2937 if the host is in sender_unqualified_hosts or recipient_unqualified hosts, as
2938 appropriate. For local messages, qualification always happens, unless -bnq is
2939 used to explicitly suppress it. No rewriting is done for an unqualified address
2940 that is left untouched.
2941
2942 We start at the second header, skipping our own Received:. This rewriting is
2943 documented as happening *after* recipient addresses are taken from the headers
2944 by the -t command line option. An added Sender: gets rewritten here. */
2945
2946 for (h = header_list->next; h; h = h->next)
2947 {
2948 header_line *newh = rewrite_header(h, NULL, NULL, global_rewrite_rules,
2949 rewrite_existflags, TRUE);
2950 if (newh) h = newh;
2951 }
2952
2953
2954 /* An RFC 822 (sic) message is not legal unless it has at least one of "to",
2955 "cc", or "bcc". Note that although the minimal examples in RFC 822 show just
2956 "to" or "bcc", the full syntax spec allows "cc" as well. If any resent- header
2957 exists, this applies to the set of resent- headers rather than the normal set.
2958
2959 The requirement for a recipient header has been removed in RFC 2822. At this
2960 point in the code, earlier versions of Exim added a To: header for locally
2961 submitted messages, and an empty Bcc: header for others. In the light of the
2962 changes in RFC 2822, this was dropped in November 2003. */
2963
2964
2965 /* If there is no date header, generate one if the message originates locally
2966 (i.e. not over TCP/IP) and suppress_local_fixups is not set, or if the
2967 submission mode flag is set. Messages without Date: are not valid, but it seems
2968 to be more confusing if Exim adds one to all remotely-originated messages.
2969 As per Message-Id, we prepend if resending, else append.
2970 */
2971
2972 if ( !date_header_exists
2973 && ((!sender_host_address && !f.suppress_local_fixups) || f.submission_mode))
2974 header_add_at_position(!resents_exist, NULL, FALSE, htype_other,
2975 "%sDate: %s\n", resent_prefix, tod_stamp(tod_full));
2976
2977 search_tidyup(); /* Free any cached resources */
2978
2979 /* Show the complete set of headers if debugging. Note that the first one (the
2980 new Received:) has not yet been set. */
2981
2982 DEBUG(D_receive)
2983 {
2984 debug_printf(">>Headers after rewriting and local additions:\n");
2985 for (h = header_list->next; h; h = h->next)
2986 debug_printf("%c %s", h->type, h->text);
2987 debug_printf("\n");
2988 }
2989
2990 /* The headers are now complete in store. If we are running in filter
2991 testing mode, that is all this function does. Return TRUE if the message
2992 ended with a dot. */
2993
2994 if (filter_test != FTEST_NONE)
2995 {
2996 process_info[process_info_len] = 0;
2997 return message_ended == END_DOT;
2998 }
2999
3000 /*XXX CHUNKING: need to cancel cutthrough under BDAT, for now. In future,
3001 think more if it could be handled. Cannot do onward CHUNKING unless
3002 inbound is, but inbound chunking ought to be ok with outbound plain.
3003 Could we do onward CHUNKING given inbound CHUNKING?
3004 */
3005 if (chunking_state > CHUNKING_OFFERED)
3006 cancel_cutthrough_connection(FALSE, US"chunking active");
3007
3008 /* Cutthrough delivery:
3009 We have to create the Received header now rather than at the end of reception,
3010 so the timestamp behaviour is a change to the normal case.
3011 Having created it, send the headers to the destination. */
3012
3013 if (cutthrough.cctx.sock >= 0 && cutthrough.delivery)
3014 {
3015 if (received_count > received_headers_max)
3016 {
3017 cancel_cutthrough_connection(TRUE, US"too many headers");
3018 if (smtp_input) receive_swallow_smtp(); /* Swallow incoming SMTP */
3019 log_write(0, LOG_MAIN|LOG_REJECT, "rejected from <%s>%s%s%s%s: "
3020 "Too many \"Received\" headers",
3021 sender_address,
3022 sender_fullhost ? "H=" : "", sender_fullhost ? sender_fullhost : US"",
3023 sender_ident ? "U=" : "", sender_ident ? sender_ident : US"");
3024 message_id[0] = 0; /* Indicate no message accepted */
3025 smtp_reply = US"550 Too many \"Received\" headers - suspected mail loop";
3026 goto TIDYUP; /* Skip to end of function */
3027 }
3028 received_header_gen();
3029 add_acl_headers(ACL_WHERE_RCPT, US"MAIL or RCPT");
3030 (void) cutthrough_headers_send();
3031 }
3032
3033
3034 /* Open a new spool file for the data portion of the message. We need
3035 to access it both via a file descriptor and a stream. Try to make the
3036 directory if it isn't there. */
3037
3038 spool_name = spool_fname(US"input", message_subdir, message_id, US"-D");
3039 DEBUG(D_receive) debug_printf("Data file name: %s\n", spool_name);
3040
3041 if ((data_fd = Uopen(spool_name, O_RDWR|O_CREAT|O_EXCL, SPOOL_MODE)) < 0)
3042 {
3043 if (errno == ENOENT)
3044 {
3045 (void) directory_make(spool_directory,
3046 spool_sname(US"input", message_subdir),
3047 INPUT_DIRECTORY_MODE, TRUE);
3048 data_fd = Uopen(spool_name, O_RDWR|O_CREAT|O_EXCL, SPOOL_MODE);
3049 }
3050 if (data_fd < 0)
3051 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Failed to create spool file %s: %s",
3052 spool_name, strerror(errno));
3053 }
3054
3055 /* Make sure the file's group is the Exim gid, and double-check the mode
3056 because the group setting doesn't always get set automatically. */
3057
3058 if (fchown(data_fd, exim_uid, exim_gid))
3059 log_write(0, LOG_MAIN|LOG_PANIC_DIE,
3060 "Failed setting ownership on spool file %s: %s",
3061 spool_name, strerror(errno));
3062 (void)fchmod(data_fd, SPOOL_MODE);
3063
3064 /* We now have data file open. Build a stream for it and lock it. We lock only
3065 the first line of the file (containing the message ID) because otherwise there
3066 are problems when Exim is run under Cygwin (I'm told). See comments in
3067 spool_in.c, where the same locking is done. */
3068
3069 spool_data_file = fdopen(data_fd, "w+");
3070 lock_data.l_type = F_WRLCK;
3071 lock_data.l_whence = SEEK_SET;
3072 lock_data.l_start = 0;
3073 lock_data.l_len = SPOOL_DATA_START_OFFSET;
3074
3075 if (fcntl(data_fd, F_SETLK, &lock_data) < 0)
3076 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Cannot lock %s (%d): %s", spool_name,
3077 errno, strerror(errno));
3078
3079 /* We have an open, locked data file. Write the message id to it to make it
3080 self-identifying. Then read the remainder of the input of this message and
3081 write it to the data file. If the variable next != NULL, it contains the first
3082 data line (which was read as a header but then turned out not to have the right
3083 format); write it (remembering that it might contain binary zeros). The result
3084 of fwrite() isn't inspected; instead we call ferror() below. */
3085
3086 fprintf(spool_data_file, "%s-D\n", message_id);
3087 if (next)
3088 {
3089 uschar *s = next->text;
3090 int len = next->slen;
3091 if (fwrite(s, 1, len, spool_data_file) == len) /* "if" for compiler quietening */
3092 body_linecount++; /* Assumes only 1 line */
3093 }
3094
3095 /* Note that we might already be at end of file, or the logical end of file
3096 (indicated by '.'), or might have encountered an error while writing the
3097 message id or "next" line. */
3098
3099 if (!ferror(spool_data_file) && !(receive_feof)() && message_ended != END_DOT)
3100 {
3101 if (smtp_input)
3102 {
3103 message_ended = chunking_state <= CHUNKING_OFFERED
3104 ? read_message_data_smtp(spool_data_file)
3105 : spool_wireformat
3106 ? read_message_bdat_smtp_wire(spool_data_file)
3107 : read_message_bdat_smtp(spool_data_file);
3108 receive_linecount++; /* The terminating "." line */
3109 }
3110 else
3111 message_ended = read_message_data(spool_data_file);
3112
3113 receive_linecount += body_linecount; /* For BSMTP errors mainly */
3114 message_linecount += body_linecount;
3115
3116 switch (message_ended)
3117 {
3118 /* Handle premature termination of SMTP */
3119
3120 case END_EOF:
3121 if (smtp_input)
3122 {
3123 Uunlink(spool_name); /* Lose data file when closed */
3124 cancel_cutthrough_connection(TRUE, US"sender closed connection");
3125 message_id[0] = 0; /* Indicate no message accepted */
3126 smtp_reply = handle_lost_connection(US"");
3127 smtp_yield = FALSE;
3128 goto TIDYUP; /* Skip to end of function */
3129 }
3130 break;
3131
3132 /* Handle message that is too big. Don't use host_or_ident() in the log
3133 message; we want to see the ident value even for non-remote messages. */
3134
3135 case END_SIZE:
3136 Uunlink(spool_name); /* Lose the data file when closed */
3137 cancel_cutthrough_connection(TRUE, US"mail too big");
3138 if (smtp_input) receive_swallow_smtp(); /* Swallow incoming SMTP */
3139
3140 log_write(L_size_reject, LOG_MAIN|LOG_REJECT, "rejected from <%s>%s%s%s%s: "
3141 "message too big: read=%d max=%d",
3142 sender_address,
3143 sender_fullhost ? " H=" : "",
3144 sender_fullhost ? sender_fullhost : US"",
3145 sender_ident ? " U=" : "",
3146 sender_ident ? sender_ident : US"",
3147 message_size,
3148 thismessage_size_limit);
3149
3150 if (smtp_input)
3151 {
3152 smtp_reply = US"552 Message size exceeds maximum permitted";
3153 message_id[0] = 0; /* Indicate no message accepted */
3154 goto TIDYUP; /* Skip to end of function */
3155 }
3156 else
3157 {
3158 fseek(spool_data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3159 give_local_error(ERRMESS_TOOBIG,
3160 string_sprintf("message too big (max=%d)", thismessage_size_limit),
3161 US"message rejected: ", error_rc, spool_data_file, header_list);
3162 /* Does not return */
3163 }
3164 break;
3165
3166 /* Handle bad BDAT protocol sequence */
3167
3168 case END_PROTOCOL:
3169 Uunlink(spool_name); /* Lose the data file when closed */
3170 cancel_cutthrough_connection(TRUE, US"sender protocol error");
3171 smtp_reply = US""; /* Response already sent */
3172 message_id[0] = 0; /* Indicate no message accepted */
3173 goto TIDYUP; /* Skip to end of function */
3174 }
3175 }
3176
3177 /* Restore the standard SIGALRM handler for any subsequent processing. (For
3178 example, there may be some expansion in an ACL that uses a timer.) */
3179
3180 os_non_restarting_signal(SIGALRM, sigalrm_handler);
3181
3182 /* The message body has now been read into the data file. Call fflush() to
3183 empty the buffers in C, and then call fsync() to get the data written out onto
3184 the disk, as fflush() doesn't do this (or at least, it isn't documented as
3185 having to do this). If there was an I/O error on either input or output,
3186 attempt to send an error message, and unlink the spool file. For non-SMTP input
3187 we can then give up. Note that for SMTP input we must swallow the remainder of
3188 the input in cases of output errors, since the far end doesn't expect to see
3189 anything until the terminating dot line is sent. */
3190
3191 if (fflush(spool_data_file) == EOF || ferror(spool_data_file) ||
3192 EXIMfsync(fileno(spool_data_file)) < 0 || (receive_ferror)())
3193 {
3194 uschar *msg_errno = US strerror(errno);
3195 BOOL input_error = (receive_ferror)() != 0;
3196 uschar *msg = string_sprintf("%s error (%s) while receiving message from %s",
3197 input_error? "Input read" : "Spool write",
3198 msg_errno,
3199 sender_fullhost ? sender_fullhost : sender_ident);
3200
3201 log_write(0, LOG_MAIN, "Message abandoned: %s", msg);
3202 Uunlink(spool_name); /* Lose the data file */
3203 cancel_cutthrough_connection(TRUE, US"error writing spoolfile");
3204
3205 if (smtp_input)
3206 {
3207 if (input_error)
3208 smtp_reply = US"451 Error while reading input data";
3209 else
3210 {
3211 smtp_reply = US"451 Error while writing spool file";
3212 receive_swallow_smtp();
3213 }
3214 message_id[0] = 0; /* Indicate no message accepted */
3215 goto TIDYUP; /* Skip to end of function */
3216 }
3217
3218 else
3219 {
3220 fseek(spool_data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3221 give_local_error(ERRMESS_IOERR, msg, US"", error_rc, spool_data_file,
3222 header_list);
3223 /* Does not return */
3224 }
3225 }
3226
3227
3228 /* No I/O errors were encountered while writing the data file. */
3229
3230 DEBUG(D_receive) debug_printf("Data file written for message %s\n", message_id);
3231 if (LOGGING(receive_time)) timesince(&received_time_taken, &received_time);
3232
3233
3234 /* If there were any bad addresses extracted by -t, or there were no recipients
3235 left after -t, send a message to the sender of this message, or write it to
3236 stderr if the error handling option is set that way. Note that there may
3237 legitimately be no recipients for an SMTP message if they have all been removed
3238 by "discard".
3239
3240 We need to rewind the data file in order to read it. In the case of no
3241 recipients or stderr error writing, throw the data file away afterwards, and
3242 exit. (This can't be SMTP, which always ensures there's at least one
3243 syntactically good recipient address.) */
3244
3245 if (extract_recip && (bad_addresses || recipients_count == 0))
3246 {
3247 DEBUG(D_receive)
3248 {
3249 if (recipients_count == 0) debug_printf("*** No recipients\n");
3250 if (bad_addresses)
3251 {
3252 error_block * eblock;
3253 debug_printf("*** Bad address(es)\n");
3254 for (eblock = bad_addresses; eblock; eblock = eblock->next)
3255 debug_printf(" %s: %s\n", eblock->text1, eblock->text2);
3256 }
3257 }
3258
3259 log_write(0, LOG_MAIN|LOG_PANIC, "%s %s found in headers",
3260 message_id, bad_addresses ? "bad addresses" : "no recipients");
3261
3262 fseek(spool_data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3263
3264 /* If configured to send errors to the sender, but this fails, force
3265 a failure error code. We use a special one for no recipients so that it
3266 can be detected by the autoreply transport. Otherwise error_rc is set to
3267 errors_sender_rc, which is EXIT_FAILURE unless -oee was given, in which case
3268 it is EXIT_SUCCESS. */
3269
3270 if (error_handling == ERRORS_SENDER)
3271 {
3272 if (!moan_to_sender(
3273 bad_addresses
3274 ? recipients_list ? ERRMESS_BADADDRESS : ERRMESS_BADNOADDRESS
3275 : extracted_ignored ? ERRMESS_IGADDRESS : ERRMESS_NOADDRESS,
3276 bad_addresses, header_list, spool_data_file, FALSE
3277 ) )
3278 error_rc = bad_addresses ? EXIT_FAILURE : EXIT_NORECIPIENTS;
3279 }
3280 else
3281 {
3282 if (!bad_addresses)
3283 if (extracted_ignored)
3284 fprintf(stderr, "exim: all -t recipients overridden by command line\n");
3285 else
3286 fprintf(stderr, "exim: no recipients in message\n");
3287 else
3288 {
3289 fprintf(stderr, "exim: invalid address%s",
3290 bad_addresses->next ? "es:\n" : ":");
3291 for ( ; bad_addresses; bad_addresses = bad_addresses->next)
3292 fprintf(stderr, " %s: %s\n", bad_addresses->text1,
3293 bad_addresses->text2);
3294 }
3295 }
3296
3297 if (recipients_count == 0 || error_handling == ERRORS_STDERR)
3298 {
3299 Uunlink(spool_name);
3300 (void)fclose(spool_data_file);
3301 exim_exit(error_rc, US"receiving");
3302 }
3303 }
3304
3305 /* Data file successfully written. Generate text for the Received: header by
3306 expanding the configured string, and adding a timestamp. By leaving this
3307 operation till now, we ensure that the timestamp is the time that message
3308 reception was completed. However, this is deliberately done before calling the
3309 data ACL and local_scan().
3310
3311 This Received: header may therefore be inspected by the data ACL and by code in
3312 the local_scan() function. When they have run, we update the timestamp to be
3313 the final time of reception.
3314
3315 If there is just one recipient, set up its value in the $received_for variable
3316 for use when we generate the Received: header.
3317
3318 Note: the checking for too many Received: headers is handled by the delivery
3319 code. */
3320 /*XXX eventually add excess Received: check for cutthrough case back when classifying them */
3321
3322 if (!received_header->text) /* Non-cutthrough case */
3323 {
3324 received_header_gen();
3325
3326 /* Set the value of message_body_size for the DATA ACL and for local_scan() */
3327
3328 message_body_size = (fstat(data_fd, &statbuf) == 0)?
3329 statbuf.st_size - SPOOL_DATA_START_OFFSET : -1;
3330
3331 /* If an ACL from any RCPT commands set up any warning headers to add, do so
3332 now, before running the DATA ACL. */
3333
3334 add_acl_headers(ACL_WHERE_RCPT, US"MAIL or RCPT");
3335 }
3336 else
3337 message_body_size = (fstat(data_fd, &statbuf) == 0)?
3338 statbuf.st_size - SPOOL_DATA_START_OFFSET : -1;
3339
3340 /* If an ACL is specified for checking things at this stage of reception of a
3341 message, run it, unless all the recipients were removed by "discard" in earlier
3342 ACLs. That is the only case in which recipients_count can be zero at this
3343 stage. Set deliver_datafile to point to the data file so that $message_body and
3344 $message_body_end can be extracted if needed. Allow $recipients in expansions.
3345 */
3346
3347 deliver_datafile = data_fd;
3348 user_msg = NULL;
3349
3350 f.enable_dollar_recipients = TRUE;
3351
3352 if (recipients_count == 0)
3353 blackholed_by = f.recipients_discarded ? US"MAIL ACL" : US"RCPT ACL";
3354
3355 else
3356 {
3357 /* Handle interactive SMTP messages */
3358
3359 if (smtp_input && !smtp_batched_input)
3360 {
3361
3362 #ifndef DISABLE_DKIM
3363 if (!f.dkim_disable_verify)
3364 {
3365 /* Finish verification */
3366 dkim_exim_verify_finish();
3367
3368 /* Check if we must run the DKIM ACL */
3369 if (acl_smtp_dkim && dkim_verify_signers && *dkim_verify_signers)
3370 {
3371 uschar * dkim_verify_signers_expanded =
3372 expand_string(dkim_verify_signers);
3373 gstring * results = NULL;
3374 int signer_sep = 0;
3375 const uschar * ptr;
3376 uschar * item;
3377 gstring * seen_items = NULL;
3378 int old_pool = store_pool;
3379
3380 store_pool = POOL_PERM; /* Allow created variables to live to data ACL */
3381
3382 if (!(ptr = dkim_verify_signers_expanded))
3383 log_write(0, LOG_MAIN|LOG_PANIC,
3384 "expansion of dkim_verify_signers option failed: %s",
3385 expand_string_message);
3386
3387 /* Default to OK when no items are present */
3388 rc = OK;
3389 while ((item = string_nextinlist(&ptr, &signer_sep, NULL, 0)))
3390 {
3391 /* Prevent running ACL for an empty item */
3392 if (!item || !*item) continue;
3393
3394 /* Only run ACL once for each domain or identity,
3395 no matter how often it appears in the expanded list. */
3396 if (seen_items)
3397 {
3398 uschar * seen_item;
3399 const uschar * seen_items_list = string_from_gstring(seen_items);
3400 int seen_sep = ':';
3401 BOOL seen_this_item = FALSE;
3402
3403 while ((seen_item = string_nextinlist(&seen_items_list, &seen_sep,
3404 NULL, 0)))
3405 if (Ustrcmp(seen_item,item) == 0)
3406 {
3407 seen_this_item = TRUE;
3408 break;
3409 }
3410
3411 if (seen_this_item)
3412 {
3413 DEBUG(D_receive)
3414 debug_printf("acl_smtp_dkim: skipping signer %s, "
3415 "already seen\n", item);
3416 continue;
3417 }
3418
3419 seen_items = string_catn(seen_items, US":", 1);
3420 }
3421 seen_items = string_cat(seen_items, item);
3422
3423 rc = dkim_exim_acl_run(item, &results, &user_msg, &log_msg);
3424 if (rc != OK)
3425 {
3426 DEBUG(D_receive)
3427 debug_printf("acl_smtp_dkim: acl_check returned %d on %s, "
3428 "skipping remaining items\n", rc, item);
3429 cancel_cutthrough_connection(TRUE, US"dkim acl not ok");
3430 break;
3431 }
3432 }
3433 dkim_verify_status = string_from_gstring(results);
3434 store_pool = old_pool;
3435 add_acl_headers(ACL_WHERE_DKIM, US"DKIM");
3436 if (rc == DISCARD)
3437 {
3438 recipients_count = 0;
3439 blackholed_by = US"DKIM ACL";
3440 if (log_msg)
3441 blackhole_log_msg = string_sprintf(": %s", log_msg);
3442 }
3443 else if (rc != OK)
3444 {
3445 Uunlink(spool_name);
3446 if (smtp_handle_acl_fail(ACL_WHERE_DKIM, rc, user_msg, log_msg) != 0)
3447 smtp_yield = FALSE; /* No more messages after dropped connection */
3448 smtp_reply = US""; /* Indicate reply already sent */
3449 message_id[0] = 0; /* Indicate no message accepted */
3450 goto TIDYUP; /* Skip to end of function */
3451 }
3452 }
3453 else
3454 dkim_exim_verify_log_all();
3455 }
3456 #endif /* DISABLE_DKIM */
3457
3458 #ifdef WITH_CONTENT_SCAN
3459 if ( recipients_count > 0
3460 && acl_smtp_mime
3461 && !run_mime_acl(acl_smtp_mime, &smtp_yield, &smtp_reply, &blackholed_by)
3462 )
3463 goto TIDYUP;
3464 #endif /* WITH_CONTENT_SCAN */
3465
3466 #ifdef EXPERIMENTAL_DMARC
3467 dmarc_up = dmarc_store_data(from_header);
3468 #endif /* EXPERIMENTAL_DMARC */
3469
3470 #ifndef DISABLE_PRDR
3471 if (prdr_requested && recipients_count > 1 && acl_smtp_data_prdr)
3472 {
3473 unsigned int c;
3474 int all_pass = OK;
3475 int all_fail = FAIL;
3476
3477 smtp_printf("353 PRDR content analysis beginning\r\n", TRUE);
3478 /* Loop through recipients, responses must be in same order received */
3479 for (c = 0; recipients_count > c; c++)
3480 {
3481 uschar * addr= recipients_list[c].address;
3482 uschar * msg= US"PRDR R=<%s> %s";
3483 uschar * code;
3484 DEBUG(D_receive)
3485 debug_printf("PRDR processing recipient %s (%d of %d)\n",
3486 addr, c+1, recipients_count);
3487 rc = acl_check(ACL_WHERE_PRDR, addr,
3488 acl_smtp_data_prdr, &user_msg, &log_msg);
3489
3490 /* If any recipient rejected content, indicate it in final message */
3491 all_pass |= rc;
3492 /* If all recipients rejected, indicate in final message */
3493 all_fail &= rc;
3494
3495 switch (rc)
3496 {
3497 case OK: case DISCARD: code = US"250"; break;
3498 case DEFER: code = US"450"; break;