Fix parsing of cmdline -os & -pr options. Bug 2538
[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 SUPPORT_DMARC
18 # include "dmarc.h"
19 #endif
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), FALSE);
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 DEBUG(D_receive) debug_printf("receive_remove_recipient(\"%s\") called\n",
558 recipient);
559 for (int count = 0; count < recipients_count; count++)
560 if (Ustrcmp(recipients_list[count].address, recipient) == 0)
561 {
562 if ((--recipients_count - count) > 0)
563 memmove(recipients_list + count, recipients_list + count + 1,
564 (recipients_count - count)*sizeof(recipient_item));
565 return TRUE;
566 }
567 return FALSE;
568 }
569
570
571
572
573
574 /* Pause for a while waiting for input. If none received in that time,
575 close the logfile, if we had one open; then if we wait for a long-running
576 datasource (months, in one use-case) log rotation will not leave us holding
577 the file copy. */
578
579 static void
580 log_close_chk(void)
581 {
582 if (!receive_timeout)
583 {
584 struct timeval t;
585 timesince(&t, &received_time);
586 if (t.tv_sec > 30*60)
587 mainlog_close();
588 else
589 {
590 fd_set r;
591 FD_ZERO(&r); FD_SET(0, &r);
592 t.tv_sec = 30*60 - t.tv_sec; t.tv_usec = 0;
593 if (select(1, &r, NULL, NULL, &t) == 0) mainlog_close();
594 }
595 }
596 }
597
598 /*************************************************
599 * Read data portion of a non-SMTP message *
600 *************************************************/
601
602 /* This function is called to read the remainder of a message (following the
603 header) when the input is not from SMTP - we are receiving a local message on
604 a standard input stream. The message is always terminated by EOF, and is also
605 terminated by a dot on a line by itself if the flag dot_ends is TRUE. Split the
606 two cases for maximum efficiency.
607
608 Ensure that the body ends with a newline. This will naturally be the case when
609 the termination is "\n.\n" but may not be otherwise. The RFC defines messages
610 as "sequences of lines" - this of course strictly applies only to SMTP, but
611 deliveries into BSD-type mailbox files also require it. Exim used to have a
612 flag for doing this at delivery time, but as it was always set for all
613 transports, I decided to simplify things by putting the check here instead.
614
615 There is at least one MUA (dtmail) that sends CRLF via this interface, and
616 other programs are known to do this as well. Exim used to have a option for
617 dealing with this: in July 2003, after much discussion, the code has been
618 changed to default to treat any of LF, CRLF, and bare CR as line terminators.
619
620 However, for the case when a dot on a line by itself terminates a message, the
621 only recognized terminating sequences before and after the dot are LF and CRLF.
622 Otherwise, having read EOL . CR, you don't know whether to read another
623 character or not.
624
625 Internally, in messages stored in Exim's spool files, LF is used as the line
626 terminator. Under the new regime, bare CRs will no longer appear in these
627 files.
628
629 Arguments:
630 fout a FILE to which to write the message
631
632 Returns: One of the END_xxx values indicating why it stopped reading
633 */
634
635 static int
636 read_message_data(FILE *fout)
637 {
638 int ch_state;
639 register int ch;
640 register int linelength = 0;
641
642 /* Handle the case when only EOF terminates the message */
643
644 if (!f.dot_ends)
645 {
646 int last_ch = '\n';
647
648 /*XXX we do a gettimeofday before checking for every received char,
649 which is hardly clever. The function-indirection doesn't help, but
650 an additional function to check for nonempty read buffer would help.
651 See stdin_getc() / smtp_getc() / tls_getc() / bdat_getc(). */
652
653 for ( ;
654 log_close_chk(), (ch = (receive_getc)(GETC_BUFFER_UNLIMITED)) != EOF;
655 last_ch = ch)
656 {
657 if (ch == 0) body_zerocount++;
658 if (last_ch == '\r' && ch != '\n')
659 {
660 if (linelength > max_received_linelength)
661 max_received_linelength = linelength;
662 linelength = 0;
663 if (fputc('\n', fout) == EOF) return END_WERROR;
664 message_size++;
665 body_linecount++;
666 }
667 if (ch == '\r') continue;
668
669 if (fputc(ch, fout) == EOF) return END_WERROR;
670 if (ch == '\n')
671 {
672 if (linelength > max_received_linelength)
673 max_received_linelength = linelength;
674 linelength = 0;
675 body_linecount++;
676 }
677 else linelength++;
678 if (++message_size > thismessage_size_limit) return END_SIZE;
679 }
680
681 if (last_ch != '\n')
682 {
683 if (linelength > max_received_linelength)
684 max_received_linelength = linelength;
685 if (fputc('\n', fout) == EOF) return END_WERROR;
686 message_size++;
687 body_linecount++;
688 }
689
690 return END_EOF;
691 }
692
693 /* Handle the case when a dot on a line on its own, or EOF, terminates. */
694
695 ch_state = 1;
696
697 while (log_close_chk(), (ch = (receive_getc)(GETC_BUFFER_UNLIMITED)) != EOF)
698 {
699 if (ch == 0) body_zerocount++;
700 switch (ch_state)
701 {
702 case 0: /* Normal state (previous char written) */
703 if (ch == '\n')
704 {
705 body_linecount++;
706 if (linelength > max_received_linelength)
707 max_received_linelength = linelength;
708 linelength = -1;
709 ch_state = 1;
710 }
711 else if (ch == '\r')
712 { ch_state = 2; continue; }
713 break;
714
715 case 1: /* After written "\n" */
716 if (ch == '.') { ch_state = 3; continue; }
717 if (ch == '\r') { ch_state = 2; continue; }
718 if (ch == '\n') { body_linecount++; linelength = -1; }
719 else ch_state = 0;
720 break;
721
722 case 2:
723 body_linecount++; /* After unwritten "\r" */
724 if (linelength > max_received_linelength)
725 max_received_linelength = linelength;
726 if (ch == '\n')
727 {
728 ch_state = 1;
729 linelength = -1;
730 }
731 else
732 {
733 if (message_size++, fputc('\n', fout) == EOF) return END_WERROR;
734 if (ch == '\r') continue;
735 ch_state = 0;
736 linelength = 0;
737 }
738 break;
739
740 case 3: /* After "\n." (\n written, dot not) */
741 if (ch == '\n') return END_DOT;
742 if (ch == '\r') { ch_state = 4; continue; }
743 message_size++;
744 linelength++;
745 if (fputc('.', fout) == EOF) return END_WERROR;
746 ch_state = 0;
747 break;
748
749 case 4: /* After "\n.\r" (\n written, rest not) */
750 if (ch == '\n') return END_DOT;
751 message_size += 2;
752 body_linecount++;
753 if (fputs(".\n", fout) == EOF) return END_WERROR;
754 if (ch == '\r') { ch_state = 2; continue; }
755 ch_state = 0;
756 break;
757 }
758
759 linelength++;
760 if (fputc(ch, fout) == EOF) return END_WERROR;
761 if (++message_size > thismessage_size_limit) return END_SIZE;
762 }
763
764 /* Get here if EOF read. Unless we have just written "\n", we need to ensure
765 the message ends with a newline, and we must also write any characters that
766 were saved up while testing for an ending dot. */
767
768 if (ch_state != 1)
769 {
770 static uschar *ends[] = { US"\n", NULL, US"\n", US".\n", US".\n" };
771 if (fputs(CS ends[ch_state], fout) == EOF) return END_WERROR;
772 message_size += Ustrlen(ends[ch_state]);
773 body_linecount++;
774 }
775
776 return END_EOF;
777 }
778
779
780
781
782 /*************************************************
783 * Read data portion of an SMTP message *
784 *************************************************/
785
786 /* This function is called to read the remainder of an SMTP message (after the
787 headers), or to skip over it when an error has occurred. In this case, the
788 output file is passed as NULL.
789
790 If any line begins with a dot, that character is skipped. The input should only
791 be successfully terminated by CR LF . CR LF unless it is local (non-network)
792 SMTP, in which case the CRs are optional, but...
793
794 FUDGE: It seems that sites on the net send out messages with just LF
795 terminators, despite the warnings in the RFCs, and other MTAs handle this. So
796 we make the CRs optional in all cases.
797
798 July 2003: Bare CRs cause trouble. We now treat them as line terminators as
799 well, so that there are no CRs in spooled messages. However, the message
800 terminating dot is not recognized between two bare CRs.
801
802 Arguments:
803 fout a FILE to which to write the message; NULL if skipping
804
805 Returns: One of the END_xxx values indicating why it stopped reading
806 */
807
808 static int
809 read_message_data_smtp(FILE *fout)
810 {
811 int ch_state = 0;
812 int ch;
813 int linelength = 0;
814
815 while ((ch = (receive_getc)(GETC_BUFFER_UNLIMITED)) != EOF)
816 {
817 if (ch == 0) body_zerocount++;
818 switch (ch_state)
819 {
820 case 0: /* After LF or CRLF */
821 if (ch == '.')
822 {
823 ch_state = 3;
824 continue; /* Don't ever write . after LF */
825 }
826 ch_state = 1;
827
828 /* Else fall through to handle as normal uschar. */
829
830 case 1: /* Normal state */
831 if (ch == '\n')
832 {
833 ch_state = 0;
834 body_linecount++;
835 if (linelength > max_received_linelength)
836 max_received_linelength = linelength;
837 linelength = -1;
838 }
839 else if (ch == '\r')
840 {
841 ch_state = 2;
842 continue;
843 }
844 break;
845
846 case 2: /* After (unwritten) CR */
847 body_linecount++;
848 if (linelength > max_received_linelength)
849 max_received_linelength = linelength;
850 linelength = -1;
851 if (ch == '\n')
852 {
853 ch_state = 0;
854 }
855 else
856 {
857 message_size++;
858 if (fout != NULL && fputc('\n', fout) == EOF) return END_WERROR;
859 cutthrough_data_put_nl();
860 if (ch != '\r') ch_state = 1; else continue;
861 }
862 break;
863
864 case 3: /* After [CR] LF . */
865 if (ch == '\n')
866 return END_DOT;
867 if (ch == '\r')
868 {
869 ch_state = 4;
870 continue;
871 }
872 /* The dot was removed at state 3. For a doubled dot, here, reinstate
873 it to cutthrough. The current ch, dot or not, is passed both to cutthrough
874 and to file below. */
875 if (ch == '.')
876 {
877 uschar c= ch;
878 cutthrough_data_puts(&c, 1);
879 }
880 ch_state = 1;
881 break;
882
883 case 4: /* After [CR] LF . CR */
884 if (ch == '\n') return END_DOT;
885 message_size++;
886 body_linecount++;
887 if (fout != NULL && fputc('\n', fout) == EOF) return END_WERROR;
888 cutthrough_data_put_nl();
889 if (ch == '\r')
890 {
891 ch_state = 2;
892 continue;
893 }
894 ch_state = 1;
895 break;
896 }
897
898 /* Add the character to the spool file, unless skipping; then loop for the
899 next. */
900
901 message_size++;
902 linelength++;
903 if (fout)
904 {
905 if (fputc(ch, fout) == EOF) return END_WERROR;
906 if (message_size > thismessage_size_limit) return END_SIZE;
907 }
908 if(ch == '\n')
909 cutthrough_data_put_nl();
910 else
911 {
912 uschar c = ch;
913 cutthrough_data_puts(&c, 1);
914 }
915 }
916
917 /* Fall through here if EOF encountered. This indicates some kind of error,
918 since a correct message is terminated by [CR] LF . [CR] LF. */
919
920 return END_EOF;
921 }
922
923
924
925
926 /* Variant of the above read_message_data_smtp() specialised for RFC 3030
927 CHUNKING. Accept input lines separated by either CRLF or CR or LF and write
928 LF-delimited spoolfile. Until we have wireformat spoolfiles, we need the
929 body_linecount accounting for proper re-expansion for the wire, so use
930 a cut-down version of the state-machine above; we don't need to do leading-dot
931 detection and unstuffing.
932
933 Arguments:
934 fout a FILE to which to write the message; NULL if skipping;
935 must be open for both writing and reading.
936
937 Returns: One of the END_xxx values indicating why it stopped reading
938 */
939
940 static int
941 read_message_bdat_smtp(FILE *fout)
942 {
943 int linelength = 0, ch;
944 enum CH_STATE ch_state = LF_SEEN;
945 BOOL fix_nl = FALSE;
946
947 for(;;)
948 {
949 switch ((ch = bdat_getc(GETC_BUFFER_UNLIMITED)))
950 {
951 case EOF: return END_EOF;
952 case ERR: return END_PROTOCOL;
953 case EOD:
954 /* Nothing to get from the sender anymore. We check the last
955 character written to the spool.
956
957 RFC 3030 states, that BDAT chunks are normal text, terminated by CRLF.
958 If we would be strict, we would refuse such broken messages.
959 But we are liberal, so we fix it. It would be easy just to append
960 the "\n" to the spool.
961
962 But there are some more things (line counting, message size calculation and such),
963 that would need to be duplicated here. So we simply do some ungetc
964 trickery.
965 */
966 if (fout)
967 {
968 if (fseek(fout, -1, SEEK_CUR) < 0) return END_PROTOCOL;
969 if (fgetc(fout) == '\n') return END_DOT;
970 }
971
972 if (linelength == -1) /* \r already seen (see below) */
973 {
974 DEBUG(D_receive) debug_printf("Add missing LF\n");
975 bdat_ungetc('\n');
976 continue;
977 }
978 DEBUG(D_receive) debug_printf("Add missing CRLF\n");
979 bdat_ungetc('\r'); /* not even \r was seen */
980 fix_nl = TRUE;
981
982 continue;
983 case '\0': body_zerocount++; break;
984 }
985 switch (ch_state)
986 {
987 case LF_SEEN: /* After LF or CRLF */
988 ch_state = MID_LINE;
989 /* fall through to handle as normal uschar. */
990
991 case MID_LINE: /* Mid-line state */
992 if (ch == '\n')
993 {
994 ch_state = LF_SEEN;
995 body_linecount++;
996 if (linelength > max_received_linelength)
997 max_received_linelength = linelength;
998 linelength = -1;
999 }
1000 else if (ch == '\r')
1001 {
1002 ch_state = CR_SEEN;
1003 if (fix_nl) bdat_ungetc('\n');
1004 continue; /* don't write CR */
1005 }
1006 break;
1007
1008 case CR_SEEN: /* After (unwritten) CR */
1009 body_linecount++;
1010 if (linelength > max_received_linelength)
1011 max_received_linelength = linelength;
1012 linelength = -1;
1013 if (ch == '\n')
1014 ch_state = LF_SEEN;
1015 else
1016 {
1017 message_size++;
1018 if (fout && fputc('\n', fout) == EOF) return END_WERROR;
1019 cutthrough_data_put_nl();
1020 if (ch == '\r') continue; /* don't write CR */
1021 ch_state = MID_LINE;
1022 }
1023 break;
1024 }
1025
1026 /* Add the character to the spool file, unless skipping */
1027
1028 message_size++;
1029 linelength++;
1030 if (fout)
1031 {
1032 if (fputc(ch, fout) == EOF) return END_WERROR;
1033 if (message_size > thismessage_size_limit) return END_SIZE;
1034 }
1035 if(ch == '\n')
1036 cutthrough_data_put_nl();
1037 else
1038 {
1039 uschar c = ch;
1040 cutthrough_data_puts(&c, 1);
1041 }
1042 }
1043 /*NOTREACHED*/
1044 }
1045
1046 static int
1047 read_message_bdat_smtp_wire(FILE *fout)
1048 {
1049 int ch;
1050
1051 /* Remember that this message uses wireformat. */
1052
1053 DEBUG(D_receive) debug_printf("CHUNKING: %s\n",
1054 fout ? "writing spoolfile in wire format" : "flushing input");
1055 f.spool_file_wireformat = TRUE;
1056
1057 for (;;)
1058 {
1059 if (chunking_data_left > 0)
1060 {
1061 unsigned len = MAX(chunking_data_left, thismessage_size_limit - message_size + 1);
1062 uschar * buf = bdat_getbuf(&len);
1063
1064 if (!buf) return END_EOF;
1065 message_size += len;
1066 if (fout && fwrite(buf, len, 1, fout) != 1) return END_WERROR;
1067 }
1068 else switch (ch = bdat_getc(GETC_BUFFER_UNLIMITED))
1069 {
1070 case EOF: return END_EOF;
1071 case EOD: return END_DOT;
1072 case ERR: return END_PROTOCOL;
1073
1074 default:
1075 message_size++;
1076 /*XXX not done:
1077 linelength
1078 max_received_linelength
1079 body_linecount
1080 body_zerocount
1081 */
1082 if (fout && fputc(ch, fout) == EOF) return END_WERROR;
1083 break;
1084 }
1085 if (message_size > thismessage_size_limit) return END_SIZE;
1086 }
1087 /*NOTREACHED*/
1088 }
1089
1090
1091
1092
1093 /*************************************************
1094 * Swallow SMTP message *
1095 *************************************************/
1096
1097 /* This function is called when there has been some kind of error while reading
1098 an SMTP message, and the remaining data may need to be swallowed. It is global
1099 because it is called from smtp_closedown() to shut down an incoming call
1100 tidily.
1101
1102 Argument: a FILE from which to read the message
1103 Returns: nothing
1104 */
1105
1106 void
1107 receive_swallow_smtp(void)
1108 {
1109 if (message_ended >= END_NOTENDED)
1110 message_ended = chunking_state <= CHUNKING_OFFERED
1111 ? read_message_data_smtp(NULL)
1112 : read_message_bdat_smtp_wire(NULL);
1113 }
1114
1115
1116
1117 /*************************************************
1118 * Handle lost SMTP connection *
1119 *************************************************/
1120
1121 /* This function logs connection loss incidents and generates an appropriate
1122 SMTP response.
1123
1124 Argument: additional data for the message
1125 Returns: the SMTP response
1126 */
1127
1128 static uschar *
1129 handle_lost_connection(uschar *s)
1130 {
1131 log_write(L_lost_incoming_connection | L_smtp_connection, LOG_MAIN,
1132 "%s lost while reading message data%s", smtp_get_connection_info(), s);
1133 smtp_notquit_exit(US"connection-lost", NULL, NULL);
1134 return US"421 Lost incoming connection";
1135 }
1136
1137
1138
1139
1140 /*************************************************
1141 * Handle a non-smtp reception error *
1142 *************************************************/
1143
1144 /* This function is called for various errors during the reception of non-SMTP
1145 messages. It either sends a message to the sender of the problem message, or it
1146 writes to the standard error stream.
1147
1148 Arguments:
1149 errcode code for moan_to_sender(), identifying the error
1150 text1 first message text, passed to moan_to_sender()
1151 text2 second message text, used only for stderrr
1152 error_rc code to pass to exim_exit if no problem
1153 f FILE containing body of message (may be stdin)
1154 hptr pointer to instore headers or NULL
1155
1156 Returns: calls exim_exit(), which does not return
1157 */
1158
1159 static void
1160 give_local_error(int errcode, uschar *text1, uschar *text2, int error_rc,
1161 FILE *f, header_line *hptr)
1162 {
1163 if (error_handling == ERRORS_SENDER)
1164 {
1165 error_block eblock;
1166 eblock.next = NULL;
1167 eblock.text1 = text1;
1168 eblock.text2 = US"";
1169 if (!moan_to_sender(errcode, &eblock, hptr, f, FALSE))
1170 error_rc = EXIT_FAILURE;
1171 }
1172 else
1173 fprintf(stderr, "exim: %s%s\n", text2, text1); /* Sic */
1174 (void)fclose(f);
1175 exim_exit(error_rc, US"");
1176 }
1177
1178
1179
1180 /*************************************************
1181 * Add header lines set up by ACL *
1182 *************************************************/
1183
1184 /* This function is called to add the header lines that were set up by
1185 statements in an ACL to the list of headers in memory. It is done in two stages
1186 like this, because when the ACL for RCPT is running, the other headers have not
1187 yet been received. This function is called twice; once just before running the
1188 DATA ACL, and once after. This is so that header lines added by MAIL or RCPT
1189 are visible to the DATA ACL.
1190
1191 Originally these header lines were added at the end. Now there is support for
1192 three different places: top, bottom, and after the Received: header(s). There
1193 will always be at least one Received: header, even if it is marked deleted, and
1194 even if something else has been put in front of it.
1195
1196 Arguments:
1197 acl_name text to identify which ACL
1198
1199 Returns: nothing
1200 */
1201
1202 static void
1203 add_acl_headers(int where, uschar *acl_name)
1204 {
1205 header_line *last_received = NULL;
1206
1207 switch(where)
1208 {
1209 case ACL_WHERE_DKIM:
1210 case ACL_WHERE_MIME:
1211 case ACL_WHERE_DATA:
1212 if ( cutthrough.cctx.sock >= 0 && cutthrough.delivery
1213 && (acl_removed_headers || acl_added_headers))
1214 {
1215 log_write(0, LOG_MAIN|LOG_PANIC, "Header modification in data ACLs"
1216 " will not take effect on cutthrough deliveries");
1217 return;
1218 }
1219 }
1220
1221 if (acl_removed_headers)
1222 {
1223 DEBUG(D_receive|D_acl) debug_printf_indent(">>Headers removed by %s ACL:\n", acl_name);
1224
1225 for (header_line * h = header_list; h; h = h->next) if (h->type != htype_old)
1226 {
1227 const uschar * list = acl_removed_headers;
1228 int sep = ':'; /* This is specified as a colon-separated list */
1229 uschar *s;
1230 uschar buffer[128];
1231
1232 while ((s = string_nextinlist(&list, &sep, buffer, sizeof(buffer))))
1233 if (header_testname(h, s, Ustrlen(s), FALSE))
1234 {
1235 h->type = htype_old;
1236 DEBUG(D_receive|D_acl) debug_printf_indent(" %s", h->text);
1237 }
1238 }
1239 acl_removed_headers = NULL;
1240 DEBUG(D_receive|D_acl) debug_printf_indent(">>\n");
1241 }
1242
1243 if (!acl_added_headers) return;
1244 DEBUG(D_receive|D_acl) debug_printf_indent(">>Headers added by %s ACL:\n", acl_name);
1245
1246 for (header_line * h = acl_added_headers, * next; h; h = next)
1247 {
1248 next = h->next;
1249
1250 switch(h->type)
1251 {
1252 case htype_add_top:
1253 h->next = header_list;
1254 header_list = h;
1255 DEBUG(D_receive|D_acl) debug_printf_indent(" (at top)");
1256 break;
1257
1258 case htype_add_rec:
1259 if (!last_received)
1260 {
1261 last_received = header_list;
1262 while (!header_testname(last_received, US"Received", 8, FALSE))
1263 last_received = last_received->next;
1264 while (last_received->next &&
1265 header_testname(last_received->next, US"Received", 8, FALSE))
1266 last_received = last_received->next;
1267 }
1268 h->next = last_received->next;
1269 last_received->next = h;
1270 DEBUG(D_receive|D_acl) debug_printf_indent(" (after Received:)");
1271 break;
1272
1273 case htype_add_rfc:
1274 /* add header before any header which is NOT Received: or Resent- */
1275 last_received = header_list;
1276 while ( last_received->next &&
1277 ( (header_testname(last_received->next, US"Received", 8, FALSE)) ||
1278 (header_testname_incomplete(last_received->next, US"Resent-", 7, FALSE)) ) )
1279 last_received = last_received->next;
1280 /* last_received now points to the last Received: or Resent-* header
1281 in an uninterrupted chain of those header types (seen from the beginning
1282 of all headers. Our current header must follow it. */
1283 h->next = last_received->next;
1284 last_received->next = h;
1285 DEBUG(D_receive|D_acl) debug_printf_indent(" (before any non-Received: or Resent-*: header)");
1286 break;
1287
1288 default:
1289 h->next = NULL;
1290 header_last->next = h;
1291 DEBUG(D_receive|D_acl) debug_printf_indent(" ");
1292 break;
1293 }
1294
1295 if (!h->next) header_last = h;
1296
1297 /* Check for one of the known header types (From:, To:, etc.) though in
1298 practice most added headers are going to be "other". Lower case
1299 identification letters are never stored with the header; they are used
1300 for existence tests when messages are received. So discard any lower case
1301 flag values. */
1302
1303 h->type = header_checkname(h, FALSE);
1304 if (h->type >= 'a') h->type = htype_other;
1305
1306 DEBUG(D_receive|D_acl) debug_printf("%s", h->text);
1307 }
1308
1309 acl_added_headers = NULL;
1310 DEBUG(D_receive|D_acl) debug_printf_indent(">>\n");
1311 }
1312
1313
1314
1315 /*************************************************
1316 * Add host information for log line *
1317 *************************************************/
1318
1319 /* Called for acceptance and rejecting log lines. This adds information about
1320 the calling host to a string that is being built dynamically.
1321
1322 Arguments:
1323 s the dynamic string
1324
1325 Returns: the extended string
1326 */
1327
1328 static gstring *
1329 add_host_info_for_log(gstring * g)
1330 {
1331 if (sender_fullhost)
1332 {
1333 if (LOGGING(dnssec) && sender_host_dnssec) /*XXX sender_helo_dnssec? */
1334 g = string_catn(g, US" DS", 3);
1335 g = string_append(g, 2, US" H=", sender_fullhost);
1336 if (LOGGING(incoming_interface) && interface_address)
1337 g = string_fmt_append(g, " I=[%s]:%d", interface_address, interface_port);
1338 }
1339 if (f.tcp_in_fastopen && !f.tcp_in_fastopen_logged)
1340 {
1341 g = string_catn(g, US" TFO*", f.tcp_in_fastopen_data ? 5 : 4);
1342 f.tcp_in_fastopen_logged = TRUE;
1343 }
1344 if (sender_ident)
1345 g = string_append(g, 2, US" U=", sender_ident);
1346 if (received_protocol)
1347 g = string_append(g, 2, US" P=", received_protocol);
1348 if (LOGGING(pipelining) && f.smtp_in_pipelining_advertised)
1349 {
1350 g = string_catn(g, US" L", 2);
1351 #ifndef DISABLE_PIPE_CONNECT
1352 if (f.smtp_in_early_pipe_used)
1353 g = string_catn(g, US"*", 1);
1354 else if (f.smtp_in_early_pipe_advertised)
1355 g = string_catn(g, US".", 1);
1356 #endif
1357 if (!f.smtp_in_pipelining_used)
1358 g = string_catn(g, US"-", 1);
1359 }
1360 return g;
1361 }
1362
1363
1364
1365 #ifdef WITH_CONTENT_SCAN
1366
1367 /*************************************************
1368 * Run the MIME ACL on a message *
1369 *************************************************/
1370
1371 /* This code is in a subroutine so that it can be used for both SMTP
1372 and non-SMTP messages. It is called with a non-NULL ACL pointer.
1373
1374 Arguments:
1375 acl The ACL to run (acl_smtp_mime or acl_not_smtp_mime)
1376 smtp_yield_ptr Set FALSE to kill messages after dropped connection
1377 smtp_reply_ptr Where SMTP reply is being built
1378 blackholed_by_ptr Where "blackholed by" message is being built
1379
1380 Returns: TRUE to carry on; FALSE to abandon the message
1381 */
1382
1383 static BOOL
1384 run_mime_acl(uschar *acl, BOOL *smtp_yield_ptr, uschar **smtp_reply_ptr,
1385 uschar **blackholed_by_ptr)
1386 {
1387 FILE *mbox_file;
1388 uschar * rfc822_file_path = NULL;
1389 unsigned long mbox_size;
1390 uschar *user_msg, *log_msg;
1391 int mime_part_count_buffer = -1;
1392 uschar * mbox_filename;
1393 int rc = OK;
1394
1395 /* check if it is a MIME message */
1396
1397 for (header_line * my_headerlist = header_list; my_headerlist;
1398 my_headerlist = my_headerlist->next)
1399 if ( my_headerlist->type != '*' /* skip deleted headers */
1400 && strncmpic(my_headerlist->text, US"Content-Type:", 13) == 0
1401 )
1402 {
1403 DEBUG(D_receive) debug_printf("Found Content-Type: header - executing acl_smtp_mime.\n");
1404 goto DO_MIME_ACL;
1405 }
1406
1407 mime_part_count = -1;
1408 DEBUG(D_receive) debug_printf("No Content-Type: header - presumably not a MIME message.\n");
1409 return TRUE;
1410
1411 DO_MIME_ACL:
1412
1413 /* make sure the eml mbox file is spooled up */
1414 if (!(mbox_file = spool_mbox(&mbox_size, NULL, &mbox_filename)))
1415 { /* error while spooling */
1416 log_write(0, LOG_MAIN|LOG_PANIC,
1417 "acl_smtp_mime: error while creating mbox spool file, message temporarily rejected.");
1418 Uunlink(spool_name);
1419 unspool_mbox();
1420 #ifdef EXPERIMENTAL_DCC
1421 dcc_ok = 0;
1422 #endif
1423 smtp_respond(US"451", 3, TRUE, US"temporary local problem");
1424 message_id[0] = 0; /* Indicate no message accepted */
1425 *smtp_reply_ptr = US""; /* Indicate reply already sent */
1426 return FALSE; /* Indicate skip to end of receive function */
1427 }
1428
1429 mime_is_rfc822 = 0;
1430
1431 MIME_ACL_CHECK:
1432 mime_part_count = -1;
1433 rc = mime_acl_check(acl, mbox_file, NULL, &user_msg, &log_msg);
1434 (void)fclose(mbox_file);
1435
1436 if (rfc822_file_path)
1437 {
1438 mime_part_count = mime_part_count_buffer;
1439
1440 if (unlink(CS rfc822_file_path) == -1)
1441 {
1442 log_write(0, LOG_PANIC,
1443 "acl_smtp_mime: can't unlink RFC822 spool file, skipping.");
1444 goto END_MIME_ACL;
1445 }
1446 rfc822_file_path = NULL;
1447 }
1448
1449 /* check if we must check any message/rfc822 attachments */
1450 if (rc == OK)
1451 {
1452 uschar * scandir = string_copyn(mbox_filename,
1453 Ustrrchr(mbox_filename, '/') - mbox_filename);
1454 struct dirent * entry;
1455 DIR * tempdir;
1456
1457 for (tempdir = opendir(CS scandir); entry = readdir(tempdir); )
1458 if (strncmpic(US entry->d_name, US"__rfc822_", 9) == 0)
1459 {
1460 rfc822_file_path = string_sprintf("%s/%s", scandir, entry->d_name);
1461 DEBUG(D_receive)
1462 debug_printf("RFC822 attachment detected: running MIME ACL for '%s'\n",
1463 rfc822_file_path);
1464 break;
1465 }
1466 closedir(tempdir);
1467
1468 if (rfc822_file_path)
1469 {
1470 if ((mbox_file = Ufopen(rfc822_file_path, "rb")))
1471 {
1472 /* set RFC822 expansion variable */
1473 mime_is_rfc822 = 1;
1474 mime_part_count_buffer = mime_part_count;
1475 goto MIME_ACL_CHECK;
1476 }
1477 log_write(0, LOG_PANIC,
1478 "acl_smtp_mime: can't open RFC822 spool file, skipping.");
1479 unlink(CS rfc822_file_path);
1480 }
1481 }
1482
1483 END_MIME_ACL:
1484 add_acl_headers(ACL_WHERE_MIME, US"MIME");
1485 if (rc == DISCARD)
1486 {
1487 recipients_count = 0;
1488 *blackholed_by_ptr = US"MIME ACL";
1489 cancel_cutthrough_connection(TRUE, US"mime acl discard");
1490 }
1491 else if (rc != OK)
1492 {
1493 Uunlink(spool_name);
1494 cancel_cutthrough_connection(TRUE, US"mime acl not ok");
1495 unspool_mbox();
1496 #ifdef EXPERIMENTAL_DCC
1497 dcc_ok = 0;
1498 #endif
1499 if (smtp_input)
1500 {
1501 if (smtp_handle_acl_fail(ACL_WHERE_MIME, rc, user_msg, log_msg) != 0)
1502 *smtp_yield_ptr = FALSE; /* No more messages after dropped connection */
1503 *smtp_reply_ptr = US""; /* Indicate reply already sent */
1504 }
1505 message_id[0] = 0; /* Indicate no message accepted */
1506 return FALSE; /* Cause skip to end of receive function */
1507 }
1508
1509 return TRUE;
1510 }
1511
1512 #endif /* WITH_CONTENT_SCAN */
1513
1514
1515
1516 void
1517 received_header_gen(void)
1518 {
1519 uschar *received;
1520 uschar *timestamp;
1521 header_line *received_header= header_list;
1522
1523 timestamp = expand_string(US"${tod_full}");
1524 if (recipients_count == 1) received_for = recipients_list[0].address;
1525 received = expand_string(received_header_text);
1526 received_for = NULL;
1527
1528 if (!received)
1529 {
1530 if(spool_name[0] != 0)
1531 Uunlink(spool_name); /* Lose the data file */
1532 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Expansion of \"%s\" "
1533 "(received_header_text) failed: %s", string_printing(received_header_text),
1534 expand_string_message);
1535 }
1536
1537 /* The first element on the header chain is reserved for the Received header,
1538 so all we have to do is fill in the text pointer, and set the type. However, if
1539 the result of the expansion is an empty string, we leave the header marked as
1540 "old" so as to refrain from adding a Received header. */
1541
1542 if (received[0] == 0)
1543 {
1544 received_header->text = string_sprintf("Received: ; %s\n", timestamp);
1545 received_header->type = htype_old;
1546 }
1547 else
1548 {
1549 received_header->text = string_sprintf("%s; %s\n", received, timestamp);
1550 received_header->type = htype_received;
1551 }
1552
1553 received_header->slen = Ustrlen(received_header->text);
1554
1555 DEBUG(D_receive) debug_printf(">>Generated Received: header line\n%c %s",
1556 received_header->type, received_header->text);
1557 }
1558
1559
1560
1561 /*************************************************
1562 * Receive message *
1563 *************************************************/
1564
1565 /* Receive a message on the given input, and put it into a pair of spool files.
1566 Either a non-null list of recipients, or the extract flag will be true, or
1567 both. The flag sender_local is true for locally generated messages. The flag
1568 submission_mode is true if an ACL has obeyed "control = submission". The flag
1569 suppress_local_fixups is true if an ACL has obeyed "control =
1570 suppress_local_fixups" or -G was passed on the command-line.
1571 The flag smtp_input is true if the message is to be
1572 handled using SMTP conventions about termination and lines starting with dots.
1573 For non-SMTP messages, dot_ends is true for dot-terminated messages.
1574
1575 If a message was successfully read, message_id[0] will be non-zero.
1576
1577 The general actions of this function are:
1578
1579 . Read the headers of the message (if any) into a chain of store
1580 blocks.
1581
1582 . If there is a "sender:" header and the message is locally originated,
1583 throw it away, unless the caller is trusted, or unless
1584 active_local_sender_retain is set - which can only happen if
1585 active_local_from_check is false.
1586
1587 . If recipients are to be extracted from the message, build the
1588 recipients list from the headers, removing any that were on the
1589 original recipients list (unless extract_addresses_remove_arguments is
1590 false), and at the same time, remove any bcc header that may be present.
1591
1592 . Get the spool file for the data, sort out its unique name, open
1593 and lock it (but don't give it the name yet).
1594
1595 . Generate a "Message-Id" header if the message doesn't have one, for
1596 locally-originated messages.
1597
1598 . Generate a "Received" header.
1599
1600 . Ensure the recipients list is fully qualified and rewritten if necessary.
1601
1602 . If there are any rewriting rules, apply them to the sender address
1603 and also to the headers.
1604
1605 . If there is no from: header, generate one, for locally-generated messages
1606 and messages in "submission mode" only.
1607
1608 . If the sender is local, check that from: is correct, and if not, generate
1609 a Sender: header, unless message comes from a trusted caller, or this
1610 feature is disabled by active_local_from_check being false.
1611
1612 . If there is no "date" header, generate one, for locally-originated
1613 or submission mode messages only.
1614
1615 . Copy the rest of the input, or up to a terminating "." if in SMTP or
1616 dot_ends mode, to the data file. Leave it open, to hold the lock.
1617
1618 . Write the envelope and the headers to a new file.
1619
1620 . Set the name for the header file; close it.
1621
1622 . Set the name for the data file; close it.
1623
1624 Because this function can potentially be called many times in a single
1625 SMTP connection, all store should be got by store_get(), so that it will be
1626 automatically retrieved after the message is accepted.
1627
1628 FUDGE: It seems that sites on the net send out messages with just LF
1629 terminators, despite the warnings in the RFCs, and other MTAs handle this. So
1630 we make the CRs optional in all cases.
1631
1632 July 2003: Bare CRs in messages, especially in header lines, cause trouble. A
1633 new regime is now in place in which bare CRs in header lines are turned into LF
1634 followed by a space, so as not to terminate the header line.
1635
1636 February 2004: A bare LF in a header line in a message whose first line was
1637 terminated by CRLF is treated in the same way as a bare CR.
1638
1639 Arguments:
1640 extract_recip TRUE if recipients are to be extracted from the message's
1641 headers
1642
1643 Returns: TRUE there are more messages to be read (SMTP input)
1644 FALSE there are no more messages to be read (non-SMTP input
1645 or SMTP connection collapsed, or other failure)
1646
1647 When reading a message for filter testing, the returned value indicates
1648 whether the headers (which is all that is read) were terminated by '.' or
1649 not. */
1650
1651 BOOL
1652 receive_msg(BOOL extract_recip)
1653 {
1654 int rc = FAIL;
1655 int msg_size = 0;
1656 int process_info_len = Ustrlen(process_info);
1657 int error_rc = error_handling == ERRORS_SENDER
1658 ? errors_sender_rc : EXIT_FAILURE;
1659 int header_size = 256;
1660 int start, end, domain;
1661 int id_resolution = 0;
1662 int had_zero = 0;
1663 int prevlines_length = 0;
1664
1665 int ptr = 0;
1666
1667 BOOL contains_resent_headers = FALSE;
1668 BOOL extracted_ignored = FALSE;
1669 BOOL first_line_ended_crlf = TRUE_UNSET;
1670 BOOL smtp_yield = TRUE;
1671 BOOL yield = FALSE;
1672
1673 BOOL resents_exist = FALSE;
1674 uschar *resent_prefix = US"";
1675 uschar *blackholed_by = NULL;
1676 uschar *blackhole_log_msg = US"";
1677 enum {NOT_TRIED, TMP_REJ, PERM_REJ, ACCEPTED} cutthrough_done = NOT_TRIED;
1678
1679 flock_t lock_data;
1680 error_block *bad_addresses = NULL;
1681
1682 uschar *frozen_by = NULL;
1683 uschar *queued_by = NULL;
1684
1685 uschar *errmsg;
1686 rmark rcvd_log_reset_point;
1687 gstring * g;
1688 struct stat statbuf;
1689
1690 /* Final message to give to SMTP caller, and messages from ACLs */
1691
1692 uschar *smtp_reply = NULL;
1693 uschar *user_msg, *log_msg;
1694
1695 /* Working header pointers */
1696
1697 rmark reset_point;
1698 header_line *next;
1699
1700 /* Flags for noting the existence of certain headers (only one left) */
1701
1702 BOOL date_header_exists = FALSE;
1703
1704 /* Pointers to receive the addresses of headers whose contents we need. */
1705
1706 header_line *from_header = NULL;
1707 header_line *subject_header = NULL;
1708 header_line *msgid_header = NULL;
1709 header_line *received_header;
1710 BOOL msgid_header_newly_created = FALSE;
1711
1712 /* Variables for use when building the Received: header. */
1713
1714 uschar *timestamp;
1715 int tslen;
1716
1717
1718 /* Release any open files that might have been cached while preparing to
1719 accept the message - e.g. by verifying addresses - because reading a message
1720 might take a fair bit of real time. */
1721
1722 search_tidyup();
1723
1724 /* Extracting the recipient list from an input file is incompatible with
1725 cutthrough delivery with the no-spool option. It shouldn't be possible
1726 to set up the combination, but just in case kill any ongoing connection. */
1727 if (extract_recip || !smtp_input)
1728 cancel_cutthrough_connection(TRUE, US"not smtp input");
1729
1730 /* Initialize the chain of headers by setting up a place-holder for Received:
1731 header. Temporarily mark it as "old", i.e. not to be used. We keep header_last
1732 pointing to the end of the chain to make adding headers simple. */
1733
1734 received_header = header_list = header_last = store_get(sizeof(header_line), FALSE);
1735 header_list->next = NULL;
1736 header_list->type = htype_old;
1737 header_list->text = NULL;
1738 header_list->slen = 0;
1739
1740 /* Control block for the next header to be read. */
1741
1742 reset_point = store_mark();
1743 next = store_get(sizeof(header_line), FALSE); /* not tainted */
1744 next->text = store_get(header_size, TRUE); /* tainted */
1745
1746 /* Initialize message id to be null (indicating no message read), and the
1747 header names list to be the normal list. Indicate there is no data file open
1748 yet, initialize the size and warning count, and deal with no size limit. */
1749
1750 message_id[0] = 0;
1751 spool_data_file = NULL;
1752 data_fd = -1;
1753 spool_name = US"";
1754 message_size = 0;
1755 warning_count = 0;
1756 received_count = 1; /* For the one we will add */
1757
1758 if (thismessage_size_limit <= 0) thismessage_size_limit = INT_MAX;
1759
1760 /* While reading the message, the following counts are computed. */
1761
1762 message_linecount = body_linecount = body_zerocount =
1763 max_received_linelength = 0;
1764
1765 #ifdef WITH_CONTENT_SCAN
1766 /* reset non-per-part mime variables */
1767 mime_is_coverletter = 0;
1768 mime_is_rfc822 = 0;
1769 mime_part_count = -1;
1770 #endif
1771
1772 #ifndef DISABLE_DKIM
1773 /* Call into DKIM to set up the context. In CHUNKING mode
1774 we clear the dot-stuffing flag */
1775 if (smtp_input && !smtp_batched_input && !f.dkim_disable_verify)
1776 dkim_exim_verify_init(chunking_state <= CHUNKING_OFFERED);
1777 #endif
1778
1779 #ifdef SUPPORT_DMARC
1780 if (sender_host_address) dmarc_init(); /* initialize libopendmarc */
1781 #endif
1782
1783 /* Remember the time of reception. Exim uses time+pid for uniqueness of message
1784 ids, and fractions of a second are required. See the comments that precede the
1785 message id creation below. */
1786
1787 (void)gettimeofday(&message_id_tv, NULL);
1788
1789 /* For other uses of the received time we can operate with granularity of one
1790 second, and for that we use the global variable received_time. This is for
1791 things like ultimate message timeouts. */
1792
1793 received_time = message_id_tv;
1794
1795 /* If SMTP input, set the special handler for timeouts. The alarm() calls
1796 happen in the smtp_getc() function when it refills its buffer. */
1797
1798 had_data_timeout = 0;
1799 if (smtp_input)
1800 os_non_restarting_signal(SIGALRM, data_timeout_handler);
1801
1802 /* If not SMTP input, timeout happens only if configured, and we just set a
1803 single timeout for the whole message. */
1804
1805 else if (receive_timeout > 0)
1806 {
1807 os_non_restarting_signal(SIGALRM, data_timeout_handler);
1808 ALARM(receive_timeout);
1809 }
1810
1811 /* SIGTERM and SIGINT are caught always. */
1812
1813 had_data_sigint = 0;
1814 signal(SIGTERM, data_sigterm_sigint_handler);
1815 signal(SIGINT, data_sigterm_sigint_handler);
1816
1817 /* Header lines in messages are not supposed to be very long, though when
1818 unfolded, to: and cc: headers can take up a lot of store. We must also cope
1819 with the possibility of junk being thrown at us. Start by getting 256 bytes for
1820 storing the header, and extend this as necessary using string_cat().
1821
1822 To cope with total lunacies, impose an upper limit on the length of the header
1823 section of the message, as otherwise the store will fill up. We must also cope
1824 with the possibility of binary zeros in the data. Hence we cannot use fgets().
1825 Folded header lines are joined into one string, leaving the '\n' characters
1826 inside them, so that writing them out reproduces the input.
1827
1828 Loop for each character of each header; the next structure for chaining the
1829 header is set up already, with ptr the offset of the next character in
1830 next->text. */
1831
1832 for (;;)
1833 {
1834 int ch = (receive_getc)(GETC_BUFFER_UNLIMITED);
1835
1836 /* If we hit EOF on a SMTP connection, it's an error, since incoming
1837 SMTP must have a correct "." terminator. */
1838
1839 if (ch == EOF && smtp_input /* && !smtp_batched_input */)
1840 {
1841 smtp_reply = handle_lost_connection(US" (header)");
1842 smtp_yield = FALSE;
1843 goto TIDYUP; /* Skip to end of function */
1844 }
1845
1846 /* See if we are at the current header's size limit - there must be at least
1847 four bytes left. This allows for the new character plus a zero, plus two for
1848 extra insertions when we are playing games with dots and carriage returns. If
1849 we are at the limit, extend the text buffer. This could have been done
1850 automatically using string_cat() but because this is a tightish loop storing
1851 only one character at a time, we choose to do it inline. Normally
1852 store_extend() will be able to extend the block; only at the end of a big
1853 store block will a copy be needed. To handle the case of very long headers
1854 (and sometimes lunatic messages can have ones that are 100s of K long) we
1855 call store_release() for strings that have been copied - if the string is at
1856 the start of a block (and therefore the only thing in it, because we aren't
1857 doing any other gets), the block gets freed. We can only do this release if
1858 there were no allocations since the once that we want to free. */
1859
1860 if (ptr >= header_size - 4)
1861 {
1862 int oldsize = header_size;
1863
1864 if (header_size >= INT_MAX/2)
1865 goto OVERSIZE;
1866 header_size *= 2;
1867
1868 /* The data came from the message, so is tainted. */
1869
1870 if (!store_extend(next->text, TRUE, oldsize, header_size))
1871 next->text = store_newblock(next->text, TRUE, header_size, ptr);
1872 }
1873
1874 /* Cope with receiving a binary zero. There is dispute about whether
1875 these should be allowed in RFC 822 messages. The middle view is that they
1876 should not be allowed in headers, at least. Exim takes this attitude at
1877 the moment. We can't just stomp on them here, because we don't know that
1878 this line is a header yet. Set a flag to cause scanning later. */
1879
1880 if (ch == 0) had_zero++;
1881
1882 /* Test for termination. Lines in remote SMTP are terminated by CRLF, while
1883 those from data files use just LF. Treat LF in local SMTP input as a
1884 terminator too. Treat EOF as a line terminator always. */
1885
1886 if (ch == EOF) goto EOL;
1887
1888 /* FUDGE: There are sites out there that don't send CRs before their LFs, and
1889 other MTAs accept this. We are therefore forced into this "liberalisation"
1890 too, so we accept LF as a line terminator whatever the source of the message.
1891 However, if the first line of the message ended with a CRLF, we treat a bare
1892 LF specially by inserting a white space after it to ensure that the header
1893 line is not terminated. */
1894
1895 if (ch == '\n')
1896 {
1897 if (first_line_ended_crlf == TRUE_UNSET) first_line_ended_crlf = FALSE;
1898 else if (first_line_ended_crlf) receive_ungetc(' ');
1899 goto EOL;
1900 }
1901
1902 /* This is not the end of the line. If this is SMTP input and this is
1903 the first character in the line and it is a "." character, ignore it.
1904 This implements the dot-doubling rule, though header lines starting with
1905 dots aren't exactly common. They are legal in RFC 822, though. If the
1906 following is CRLF or LF, this is the line that that terminates the
1907 entire message. We set message_ended to indicate this has happened (to
1908 prevent further reading), and break out of the loop, having freed the
1909 empty header, and set next = NULL to indicate no data line. */
1910
1911 if (ptr == 0 && ch == '.' && f.dot_ends)
1912 {
1913 ch = (receive_getc)(GETC_BUFFER_UNLIMITED);
1914 if (ch == '\r')
1915 {
1916 ch = (receive_getc)(GETC_BUFFER_UNLIMITED);
1917 if (ch != '\n')
1918 {
1919 receive_ungetc(ch);
1920 ch = '\r'; /* Revert to CR */
1921 }
1922 }
1923 if (ch == '\n')
1924 {
1925 message_ended = END_DOT;
1926 reset_point = store_reset(reset_point);
1927 next = NULL;
1928 break; /* End character-reading loop */
1929 }
1930
1931 /* For non-SMTP input, the dot at the start of the line was really a data
1932 character. What is now in ch is the following character. We guaranteed
1933 enough space for this above. */
1934
1935 if (!smtp_input)
1936 {
1937 next->text[ptr++] = '.';
1938 message_size++;
1939 }
1940 }
1941
1942 /* If CR is immediately followed by LF, end the line, ignoring the CR, and
1943 remember this case if this is the first line ending. */
1944
1945 if (ch == '\r')
1946 {
1947 ch = (receive_getc)(GETC_BUFFER_UNLIMITED);
1948 if (ch == '\n')
1949 {
1950 if (first_line_ended_crlf == TRUE_UNSET) first_line_ended_crlf = TRUE;
1951 goto EOL;
1952 }
1953
1954 /* Otherwise, put back the character after CR, and turn the bare CR
1955 into LF SP. */
1956
1957 ch = (receive_ungetc)(ch);
1958 next->text[ptr++] = '\n';
1959 message_size++;
1960 ch = ' ';
1961 }
1962
1963 /* We have a data character for the header line. */
1964
1965 next->text[ptr++] = ch; /* Add to buffer */
1966 message_size++; /* Total message size so far */
1967
1968 /* Handle failure due to a humungously long header section. The >= allows
1969 for the terminating \n. Add what we have so far onto the headers list so
1970 that it gets reflected in any error message, and back up the just-read
1971 character. */
1972
1973 if (message_size >= header_maxsize)
1974 {
1975 OVERSIZE:
1976 next->text[ptr] = 0;
1977 next->slen = ptr;
1978 next->type = htype_other;
1979 next->next = NULL;
1980 header_last->next = next;
1981 header_last = next;
1982
1983 log_write(0, LOG_MAIN, "ridiculously long message header received from "
1984 "%s (more than %d characters): message abandoned",
1985 f.sender_host_unknown ? sender_ident : sender_fullhost, header_maxsize);
1986
1987 if (smtp_input)
1988 {
1989 smtp_reply = US"552 Message header is ridiculously long";
1990 receive_swallow_smtp();
1991 goto TIDYUP; /* Skip to end of function */
1992 }
1993
1994 else
1995 {
1996 give_local_error(ERRMESS_VLONGHEADER,
1997 string_sprintf("message header longer than %d characters received: "
1998 "message not accepted", header_maxsize), US"", error_rc, stdin,
1999 header_list->next);
2000 /* Does not return */
2001 }
2002 }
2003
2004 continue; /* With next input character */
2005
2006 /* End of header line reached */
2007
2008 EOL:
2009
2010 /* Keep track of lines for BSMTP errors and overall message_linecount. */
2011
2012 receive_linecount++;
2013 message_linecount++;
2014
2015 /* Keep track of maximum line length */
2016
2017 if (ptr - prevlines_length > max_received_linelength)
2018 max_received_linelength = ptr - prevlines_length;
2019 prevlines_length = ptr + 1;
2020
2021 /* Now put in the terminating newline. There is always space for
2022 at least two more characters. */
2023
2024 next->text[ptr++] = '\n';
2025 message_size++;
2026
2027 /* A blank line signals the end of the headers; release the unwanted
2028 space and set next to NULL to indicate this. */
2029
2030 if (ptr == 1)
2031 {
2032 reset_point = store_reset(reset_point);
2033 next = NULL;
2034 break;
2035 }
2036
2037 /* There is data in the line; see if the next input character is a
2038 whitespace character. If it is, we have a continuation of this header line.
2039 There is always space for at least one character at this point. */
2040
2041 if (ch != EOF)
2042 {
2043 int nextch = (receive_getc)(GETC_BUFFER_UNLIMITED);
2044 if (nextch == ' ' || nextch == '\t')
2045 {
2046 next->text[ptr++] = nextch;
2047 if (++message_size >= header_maxsize)
2048 goto OVERSIZE;
2049 continue; /* Iterate the loop */
2050 }
2051 else if (nextch != EOF) (receive_ungetc)(nextch); /* For next time */
2052 else ch = EOF; /* Cause main loop to exit at end */
2053 }
2054
2055 /* We have got to the real line end. Terminate the string and release store
2056 beyond it. If it turns out to be a real header, internal binary zeros will
2057 be squashed later. */
2058
2059 next->text[ptr] = 0;
2060 next->slen = ptr;
2061 store_release_above(next->text + ptr + 1);
2062
2063 /* Check the running total size against the overall message size limit. We
2064 don't expect to fail here, but if the overall limit is set less than MESSAGE_
2065 MAXSIZE and a big header is sent, we want to catch it. Just stop reading
2066 headers - the code to read the body will then also hit the buffer. */
2067
2068 if (message_size > thismessage_size_limit) break;
2069
2070 /* A line that is not syntactically correct for a header also marks
2071 the end of the headers. In this case, we leave next containing the
2072 first data line. This might actually be several lines because of the
2073 continuation logic applied above, but that doesn't matter.
2074
2075 It turns out that smail, and presumably sendmail, accept leading lines
2076 of the form
2077
2078 From ph10 Fri Jan 5 12:35 GMT 1996
2079
2080 in messages. The "mail" command on Solaris 2 sends such lines. I cannot
2081 find any documentation of this, but for compatibility it had better be
2082 accepted. Exim restricts it to the case of non-smtp messages, and
2083 treats it as an alternative to the -f command line option. Thus it is
2084 ignored except for trusted users or filter testing. Otherwise it is taken
2085 as the sender address, unless -f was used (sendmail compatibility).
2086
2087 It further turns out that some UUCPs generate the From_line in a different
2088 format, e.g.
2089
2090 From ph10 Fri, 7 Jan 97 14:00:00 GMT
2091
2092 The regex for matching these things is now capable of recognizing both
2093 formats (including 2- and 4-digit years in the latter). In fact, the regex
2094 is now configurable, as is the expansion string to fish out the sender.
2095
2096 Even further on it has been discovered that some broken clients send
2097 these lines in SMTP messages. There is now an option to ignore them from
2098 specified hosts or networks. Sigh. */
2099
2100 if ( header_last == header_list
2101 && ( !smtp_input
2102 || ( sender_host_address
2103 && verify_check_host(&ignore_fromline_hosts) == OK
2104 )
2105 || (!sender_host_address && ignore_fromline_local)
2106 )
2107 && regex_match_and_setup(regex_From, next->text, 0, -1)
2108 )
2109 {
2110 if (!f.sender_address_forced)
2111 {
2112 uschar *uucp_sender = expand_string(uucp_from_sender);
2113 if (!uucp_sender)
2114 log_write(0, LOG_MAIN|LOG_PANIC,
2115 "expansion of \"%s\" failed after matching "
2116 "\"From \" line: %s", uucp_from_sender, expand_string_message);
2117 else
2118 {
2119 int start, end, domain;
2120 uschar *errmess;
2121 uschar *newsender = parse_extract_address(uucp_sender, &errmess,
2122 &start, &end, &domain, TRUE);
2123 if (newsender)
2124 {
2125 if (domain == 0 && newsender[0] != 0)
2126 newsender = rewrite_address_qualify(newsender, FALSE);
2127
2128 if (filter_test != FTEST_NONE || receive_check_set_sender(newsender))
2129 {
2130 sender_address = newsender;
2131
2132 if (f.trusted_caller || filter_test != FTEST_NONE)
2133 {
2134 authenticated_sender = NULL;
2135 originator_name = US"";
2136 f.sender_local = FALSE;
2137 }
2138
2139 if (filter_test != FTEST_NONE)
2140 printf("Sender taken from \"From \" line\n");
2141 }
2142 }
2143 }
2144 }
2145 }
2146
2147 /* Not a leading "From " line. Check to see if it is a valid header line.
2148 Header names may contain any non-control characters except space and colon,
2149 amazingly. */
2150
2151 else
2152 {
2153 uschar *p = next->text;
2154
2155 /* If not a valid header line, break from the header reading loop, leaving
2156 next != NULL, indicating that it holds the first line of the body. */
2157
2158 if (isspace(*p)) break;
2159 while (mac_isgraph(*p) && *p != ':') p++;
2160 while (isspace(*p)) p++;
2161 if (*p != ':')
2162 {
2163 body_zerocount = had_zero;
2164 break;
2165 }
2166
2167 /* We have a valid header line. If there were any binary zeroes in
2168 the line, stomp on them here. */
2169
2170 if (had_zero > 0)
2171 for (uschar * p = next->text; p < next->text + ptr; p++) if (*p == 0)
2172 *p = '?';
2173
2174 /* It is perfectly legal to have an empty continuation line
2175 at the end of a header, but it is confusing to humans
2176 looking at such messages, since it looks like a blank line.
2177 Reduce confusion by removing redundant white space at the
2178 end. We know that there is at least one printing character
2179 (the ':' tested for above) so there is no danger of running
2180 off the end. */
2181
2182 p = next->text + ptr - 2;
2183 for (;;)
2184 {
2185 while (*p == ' ' || *p == '\t') p--;
2186 if (*p != '\n') break;
2187 ptr = (p--) - next->text + 1;
2188 message_size -= next->slen - ptr;
2189 next->text[ptr] = 0;
2190 next->slen = ptr;
2191 }
2192
2193 /* Add the header to the chain */
2194
2195 next->type = htype_other;
2196 next->next = NULL;
2197 header_last->next = next;
2198 header_last = next;
2199
2200 /* Check the limit for individual line lengths. This comes after adding to
2201 the chain so that the failing line is reflected if a bounce is generated
2202 (for a local message). */
2203
2204 if (header_line_maxsize > 0 && next->slen > header_line_maxsize)
2205 {
2206 log_write(0, LOG_MAIN, "overlong message header line received from "
2207 "%s (more than %d characters): message abandoned",
2208 f.sender_host_unknown ? sender_ident : sender_fullhost,
2209 header_line_maxsize);
2210
2211 if (smtp_input)
2212 {
2213 smtp_reply = US"552 A message header line is too long";
2214 receive_swallow_smtp();
2215 goto TIDYUP; /* Skip to end of function */
2216 }
2217
2218 else
2219 give_local_error(ERRMESS_VLONGHDRLINE,
2220 string_sprintf("message header line longer than %d characters "
2221 "received: message not accepted", header_line_maxsize), US"",
2222 error_rc, stdin, header_list->next);
2223 /* Does not return */
2224 }
2225
2226 /* Note if any resent- fields exist. */
2227
2228 if (!resents_exist && strncmpic(next->text, US"resent-", 7) == 0)
2229 {
2230 resents_exist = TRUE;
2231 resent_prefix = US"Resent-";
2232 }
2233 }
2234
2235 /* Reject CHUNKING messages that do not CRLF their first header line */
2236
2237 if (!first_line_ended_crlf && chunking_state > CHUNKING_OFFERED)
2238 {
2239 log_write(L_size_reject, LOG_MAIN|LOG_REJECT, "rejected from <%s>%s%s%s%s: "
2240 "Non-CRLF-terminated header, under CHUNKING: message abandoned",
2241 sender_address,
2242 sender_fullhost ? " H=" : "", sender_fullhost ? sender_fullhost : US"",
2243 sender_ident ? " U=" : "", sender_ident ? sender_ident : US"");
2244 smtp_printf("552 Message header not CRLF terminated\r\n", FALSE);
2245 bdat_flush_data();
2246 smtp_reply = US"";
2247 goto TIDYUP; /* Skip to end of function */
2248 }
2249
2250 /* The line has been handled. If we have hit EOF, break out of the loop,
2251 indicating no pending data line. */
2252
2253 if (ch == EOF) { next = NULL; break; }
2254
2255 /* Set up for the next header */
2256
2257 reset_point = store_mark();
2258 header_size = 256;
2259 next = store_get(sizeof(header_line), FALSE);
2260 next->text = store_get(header_size, TRUE);
2261 ptr = 0;
2262 had_zero = 0;
2263 prevlines_length = 0;
2264 } /* Continue, starting to read the next header */
2265
2266 /* At this point, we have read all the headers into a data structure in main
2267 store. The first header is still the dummy placeholder for the Received: header
2268 we are going to generate a bit later on. If next != NULL, it contains the first
2269 data line - which terminated the headers before reaching a blank line (not the
2270 normal case). */
2271
2272 DEBUG(D_receive)
2273 {
2274 debug_printf(">>Headers received:\n");
2275 for (header_line * h = header_list->next; h; h = h->next)
2276 debug_printf("%s", h->text);
2277 debug_printf("\n");
2278 }
2279
2280 /* End of file on any SMTP connection is an error. If an incoming SMTP call
2281 is dropped immediately after valid headers, the next thing we will see is EOF.
2282 We must test for this specially, as further down the reading of the data is
2283 skipped if already at EOF. */
2284
2285 if (smtp_input && (receive_feof)())
2286 {
2287 smtp_reply = handle_lost_connection(US" (after header)");
2288 smtp_yield = FALSE;
2289 goto TIDYUP; /* Skip to end of function */
2290 }
2291
2292 /* If this is a filter test run and no headers were read, output a warning
2293 in case there is a mistake in the test message. */
2294
2295 if (filter_test != FTEST_NONE && header_list->next == NULL)
2296 printf("Warning: no message headers read\n");
2297
2298
2299 /* Scan the headers to identify them. Some are merely marked for later
2300 processing; some are dealt with here. */
2301
2302 for (header_line * h = header_list->next; h; h = h->next)
2303 {
2304 BOOL is_resent = strncmpic(h->text, US"resent-", 7) == 0;
2305 if (is_resent) contains_resent_headers = TRUE;
2306
2307 switch (header_checkname(h, is_resent))
2308 {
2309 case htype_bcc:
2310 h->type = htype_bcc; /* Both Bcc: and Resent-Bcc: */
2311 break;
2312
2313 case htype_cc:
2314 h->type = htype_cc; /* Both Cc: and Resent-Cc: */
2315 break;
2316
2317 /* Record whether a Date: or Resent-Date: header exists, as appropriate. */
2318
2319 case htype_date:
2320 if (!resents_exist || is_resent) date_header_exists = TRUE;
2321 break;
2322
2323 /* Same comments as about Return-Path: below. */
2324
2325 case htype_delivery_date:
2326 if (delivery_date_remove) h->type = htype_old;
2327 break;
2328
2329 /* Same comments as about Return-Path: below. */
2330
2331 case htype_envelope_to:
2332 if (envelope_to_remove) h->type = htype_old;
2333 break;
2334
2335 /* Mark all "From:" headers so they get rewritten. Save the one that is to
2336 be used for Sender: checking. For Sendmail compatibility, if the "From:"
2337 header consists of just the login id of the user who called Exim, rewrite
2338 it with the gecos field first. Apply this rule to Resent-From: if there
2339 are resent- fields. */
2340
2341 case htype_from:
2342 h->type = htype_from;
2343 if (!resents_exist || is_resent)
2344 {
2345 from_header = h;
2346 if (!smtp_input)
2347 {
2348 int len;
2349 uschar *s = Ustrchr(h->text, ':') + 1;
2350 while (isspace(*s)) s++;
2351 len = h->slen - (s - h->text) - 1;
2352 if (Ustrlen(originator_login) == len &&
2353 strncmpic(s, originator_login, len) == 0)
2354 {
2355 uschar *name = is_resent? US"Resent-From" : US"From";
2356 header_add(htype_from, "%s: %s <%s@%s>\n", name, originator_name,
2357 originator_login, qualify_domain_sender);
2358 from_header = header_last;
2359 h->type = htype_old;
2360 DEBUG(D_receive|D_rewrite)
2361 debug_printf("rewrote \"%s:\" header using gecos\n", name);
2362 }
2363 }
2364 }
2365 break;
2366
2367 /* Identify the Message-id: header for generating "in-reply-to" in the
2368 autoreply transport. For incoming logging, save any resent- value. In both
2369 cases, take just the first of any multiples. */
2370
2371 case htype_id:
2372 if (!msgid_header && (!resents_exist || is_resent))
2373 {
2374 msgid_header = h;
2375 h->type = htype_id;
2376 }
2377 break;
2378
2379 /* Flag all Received: headers */
2380
2381 case htype_received:
2382 h->type = htype_received;
2383 received_count++;
2384 break;
2385
2386 /* "Reply-to:" is just noted (there is no resent-reply-to field) */
2387
2388 case htype_reply_to:
2389 h->type = htype_reply_to;
2390 break;
2391
2392 /* The Return-path: header is supposed to be added to messages when
2393 they leave the SMTP system. We shouldn't receive messages that already
2394 contain Return-path. However, since Exim generates Return-path: on
2395 local delivery, resent messages may well contain it. We therefore
2396 provide an option (which defaults on) to remove any Return-path: headers
2397 on input. Removal actually means flagging as "old", which prevents the
2398 header being transmitted with the message. */
2399
2400 case htype_return_path:
2401 if (return_path_remove) h->type = htype_old;
2402
2403 /* If we are testing a mail filter file, use the value of the
2404 Return-Path: header to set up the return_path variable, which is not
2405 otherwise set. However, remove any <> that surround the address
2406 because the variable doesn't have these. */
2407
2408 if (filter_test != FTEST_NONE)
2409 {
2410 uschar *start = h->text + 12;
2411 uschar *end = start + Ustrlen(start);
2412 while (isspace(*start)) start++;
2413 while (end > start && isspace(end[-1])) end--;
2414 if (*start == '<' && end[-1] == '>')
2415 {
2416 start++;
2417 end--;
2418 }
2419 return_path = string_copyn(start, end - start);
2420 printf("Return-path taken from \"Return-path:\" header line\n");
2421 }
2422 break;
2423
2424 /* If there is a "Sender:" header and the message is locally originated,
2425 and from an untrusted caller and suppress_local_fixups is not set, or if we
2426 are in submission mode for a remote message, mark it "old" so that it will
2427 not be transmitted with the message, unless active_local_sender_retain is
2428 set. (This can only be true if active_local_from_check is false.) If there
2429 are any resent- headers in the message, apply this rule to Resent-Sender:
2430 instead of Sender:. Messages with multiple resent- header sets cannot be
2431 tidily handled. (For this reason, at least one MUA - Pine - turns old
2432 resent- headers into X-resent- headers when resending, leaving just one
2433 set.) */
2434
2435 case htype_sender:
2436 h->type = !f.active_local_sender_retain
2437 && ( f.sender_local && !f.trusted_caller && !f.suppress_local_fixups
2438 || f.submission_mode
2439 )
2440 && (!resents_exist || is_resent)
2441 ? htype_old : htype_sender;
2442 break;
2443
2444 /* Remember the Subject: header for logging. There is no Resent-Subject */
2445
2446 case htype_subject:
2447 subject_header = h;
2448 break;
2449
2450 /* "To:" gets flagged, and the existence of a recipient header is noted,
2451 whether it's resent- or not. */
2452
2453 case htype_to:
2454 h->type = htype_to;
2455 /****
2456 to_or_cc_header_exists = TRUE;
2457 ****/
2458 break;
2459 }
2460 }
2461
2462 /* Extract recipients from the headers if that is required (the -t option).
2463 Note that this is documented as being done *before* any address rewriting takes
2464 place. There are two possibilities:
2465
2466 (1) According to sendmail documentation for Solaris, IRIX, and HP-UX, any
2467 recipients already listed are to be REMOVED from the message. Smail 3 works
2468 like this. We need to build a non-recipients tree for that list, because in
2469 subsequent processing this data is held in a tree and that's what the
2470 spool_write_header() function expects. Make sure that non-recipient addresses
2471 are fully qualified and rewritten if necessary.
2472
2473 (2) According to other sendmail documentation, -t ADDS extracted recipients to
2474 those in the command line arguments (and it is rumoured some other MTAs do
2475 this). Therefore, there is an option to make Exim behave this way.
2476
2477 *** Notes on "Resent-" header lines ***
2478
2479 The presence of resent-headers in the message makes -t horribly ambiguous.
2480 Experiments with sendmail showed that it uses recipients for all resent-
2481 headers, totally ignoring the concept of "sets of resent- headers" as described
2482 in RFC 2822 section 3.6.6. Sendmail also amalgamates them into a single set
2483 with all the addresses in one instance of each header.
2484
2485 This seems to me not to be at all sensible. Before release 4.20, Exim 4 gave an
2486 error for -t if there were resent- headers in the message. However, after a
2487 discussion on the mailing list, I've learned that there are MUAs that use
2488 resent- headers with -t, and also that the stuff about sets of resent- headers
2489 and their ordering in RFC 2822 is generally ignored. An MUA that submits a
2490 message with -t and resent- header lines makes sure that only *its* resent-
2491 headers are present; previous ones are often renamed as X-resent- for example.
2492
2493 Consequently, Exim has been changed so that, if any resent- header lines are
2494 present, the recipients are taken from all of the appropriate resent- lines,
2495 and not from the ordinary To:, Cc:, etc. */
2496
2497 if (extract_recip)
2498 {
2499 int rcount = 0;
2500 error_block **bnext = &bad_addresses;
2501
2502 if (extract_addresses_remove_arguments)
2503 {
2504 while (recipients_count-- > 0)
2505 {
2506 uschar *s = rewrite_address(recipients_list[recipients_count].address,
2507 TRUE, TRUE, global_rewrite_rules, rewrite_existflags);
2508 tree_add_nonrecipient(s);
2509 }
2510 recipients_list = NULL;
2511 recipients_count = recipients_list_max = 0;
2512 }
2513
2514 /* Now scan the headers */
2515
2516 for (header_line * h = header_list->next; h; h = h->next)
2517 {
2518 if ((h->type == htype_to || h->type == htype_cc || h->type == htype_bcc) &&
2519 (!contains_resent_headers || strncmpic(h->text, US"resent-", 7) == 0))
2520 {
2521 uschar *s = Ustrchr(h->text, ':') + 1;
2522 while (isspace(*s)) s++;
2523
2524 f.parse_allow_group = TRUE; /* Allow address group syntax */
2525
2526 while (*s != 0)
2527 {
2528 uschar *ss = parse_find_address_end(s, FALSE);
2529 uschar *recipient, *errmess, *pp;
2530 int start, end, domain;
2531
2532 /* Check on maximum */
2533
2534 if (recipients_max > 0 && ++rcount > recipients_max)
2535 give_local_error(ERRMESS_TOOMANYRECIP, US"too many recipients",
2536 US"message rejected: ", error_rc, stdin, NULL);
2537 /* Does not return */
2538
2539 /* Make a copy of the address, and remove any internal newlines. These
2540 may be present as a result of continuations of the header line. The
2541 white space that follows the newline must not be removed - it is part
2542 of the header. */
2543
2544 pp = recipient = store_get(ss - s + 1, is_tainted(s));
2545 for (uschar * p = s; p < ss; p++) if (*p != '\n') *pp++ = *p;
2546 *pp = 0;
2547
2548 #ifdef SUPPORT_I18N
2549 {
2550 BOOL b = allow_utf8_domains;
2551 allow_utf8_domains = TRUE;
2552 #endif
2553 recipient = parse_extract_address(recipient, &errmess, &start, &end,
2554 &domain, FALSE);
2555
2556 #ifdef SUPPORT_I18N
2557 if (string_is_utf8(recipient))
2558 message_smtputf8 = TRUE;
2559 else
2560 allow_utf8_domains = b;
2561 }
2562 #endif
2563
2564 /* Keep a list of all the bad addresses so we can send a single
2565 error message at the end. However, an empty address is not an error;
2566 just ignore it. This can come from an empty group list like
2567
2568 To: Recipients of list:;
2569
2570 If there are no recipients at all, an error will occur later. */
2571
2572 if (recipient == NULL && Ustrcmp(errmess, "empty address") != 0)
2573 {
2574 int len = Ustrlen(s);
2575 error_block *b = store_get(sizeof(error_block), FALSE);
2576 while (len > 0 && isspace(s[len-1])) len--;
2577 b->next = NULL;
2578 b->text1 = string_printing(string_copyn(s, len));
2579 b->text2 = errmess;
2580 *bnext = b;
2581 bnext = &(b->next);
2582 }
2583
2584 /* If the recipient is already in the nonrecipients tree, it must
2585 have appeared on the command line with the option extract_addresses_
2586 remove_arguments set. Do not add it to the recipients, and keep a note
2587 that this has happened, in order to give a better error if there are
2588 no recipients left. */
2589
2590 else if (recipient != NULL)
2591 {
2592 if (tree_search(tree_nonrecipients, recipient) == NULL)
2593 receive_add_recipient(recipient, -1);
2594 else
2595 extracted_ignored = TRUE;
2596 }
2597
2598 /* Move on past this address */
2599
2600 s = ss + (*ss? 1:0);
2601 while (isspace(*s)) s++;
2602 } /* Next address */
2603
2604 f.parse_allow_group = FALSE; /* Reset group syntax flags */
2605 f.parse_found_group = FALSE;
2606
2607 /* If this was the bcc: header, mark it "old", which means it
2608 will be kept on the spool, but not transmitted as part of the
2609 message. */
2610
2611 if (h->type == htype_bcc) h->type = htype_old;
2612 } /* For appropriate header line */
2613 } /* For each header line */
2614
2615 }
2616
2617 /* Now build the unique message id. This has changed several times over the
2618 lifetime of Exim. This description was rewritten for Exim 4.14 (February 2003).
2619 Retaining all the history in the comment has become too unwieldy - read
2620 previous release sources if you want it.
2621
2622 The message ID has 3 parts: tttttt-pppppp-ss. Each part is a number in base 62.
2623 The first part is the current time, in seconds. The second part is the current
2624 pid. Both are large enough to hold 32-bit numbers in base 62. The third part
2625 can hold a number in the range 0-3843. It used to be a computed sequence
2626 number, but is now the fractional component of the current time in units of
2627 1/2000 of a second (i.e. a value in the range 0-1999). After a message has been
2628 received, Exim ensures that the timer has ticked at the appropriate level
2629 before proceeding, to avoid duplication if the pid happened to be re-used
2630 within the same time period. It seems likely that most messages will take at
2631 least half a millisecond to be received, so no delay will normally be
2632 necessary. At least for some time...
2633
2634 There is a modification when localhost_number is set. Formerly this was allowed
2635 to be as large as 255. Now it is restricted to the range 0-16, and the final
2636 component of the message id becomes (localhost_number * 200) + fractional time
2637 in units of 1/200 of a second (i.e. a value in the range 0-3399).
2638
2639 Some not-really-Unix operating systems use case-insensitive file names (Darwin,
2640 Cygwin). For these, we have to use base 36 instead of base 62. Luckily, this
2641 still allows the tttttt field to hold a large enough number to last for some
2642 more decades, and the final two-digit field can hold numbers up to 1295, which
2643 is enough for milliseconds (instead of 1/2000 of a second).
2644
2645 However, the pppppp field cannot hold a 32-bit pid, but it can hold a 31-bit
2646 pid, so it is probably safe because pids have to be positive. The
2647 localhost_number is restricted to 0-10 for these hosts, and when it is set, the
2648 final field becomes (localhost_number * 100) + fractional time in centiseconds.
2649
2650 Note that string_base62() returns its data in a static storage block, so it
2651 must be copied before calling string_base62() again. It always returns exactly
2652 6 characters.
2653
2654 There doesn't seem to be anything in the RFC which requires a message id to
2655 start with a letter, but Smail was changed to ensure this. The external form of
2656 the message id (as supplied by string expansion) therefore starts with an
2657 additional leading 'E'. The spool file names do not include this leading
2658 letter and it is not used internally.
2659
2660 NOTE: If ever the format of message ids is changed, the regular expression for
2661 checking that a string is in this format must be updated in a corresponding
2662 way. It appears in the initializing code in exim.c. The macro MESSAGE_ID_LENGTH
2663 must also be changed to reflect the correct string length. The queue-sort code
2664 needs to know the layout. Then, of course, other programs that rely on the
2665 message id format will need updating too. */
2666
2667 Ustrncpy(message_id, string_base62((long int)(message_id_tv.tv_sec)), 6);
2668 message_id[6] = '-';
2669 Ustrncpy(message_id + 7, string_base62((long int)getpid()), 6);
2670
2671 /* Deal with the case where the host number is set. The value of the number was
2672 checked when it was read, to ensure it isn't too big. The timing granularity is
2673 left in id_resolution so that an appropriate wait can be done after receiving
2674 the message, if necessary (we hope it won't be). */
2675
2676 if (host_number_string)
2677 {
2678 id_resolution = BASE_62 == 62 ? 5000 : 10000;
2679 sprintf(CS(message_id + MESSAGE_ID_LENGTH - 3), "-%2s",
2680 string_base62((long int)(
2681 host_number * (1000000/id_resolution) +
2682 message_id_tv.tv_usec/id_resolution)) + 4);
2683 }
2684
2685 /* Host number not set: final field is just the fractional time at an
2686 appropriate resolution. */
2687
2688 else
2689 {
2690 id_resolution = BASE_62 == 62 ? 500 : 1000;
2691 sprintf(CS(message_id + MESSAGE_ID_LENGTH - 3), "-%2s",
2692 string_base62((long int)(message_id_tv.tv_usec/id_resolution)) + 4);
2693 }
2694
2695 /* Add the current message id onto the current process info string if
2696 it will fit. */
2697
2698 (void)string_format(process_info + process_info_len,
2699 PROCESS_INFO_SIZE - process_info_len, " id=%s", message_id);
2700
2701 /* If we are using multiple input directories, set up the one for this message
2702 to be the least significant base-62 digit of the time of arrival. Otherwise
2703 ensure that it is an empty string. */
2704
2705 set_subdir_str(message_subdir, message_id, 0);
2706
2707 /* Now that we have the message-id, if there is no message-id: header, generate
2708 one, but only for local (without suppress_local_fixups) or submission mode
2709 messages. This can be user-configured if required, but we had better flatten
2710 any illegal characters therein. */
2711
2712 if ( !msgid_header
2713 && ((!sender_host_address && !f.suppress_local_fixups) || f.submission_mode))
2714 {
2715 uschar *id_text = US"";
2716 uschar *id_domain = primary_hostname;
2717 header_line * h;
2718
2719 /* Permit only letters, digits, dots, and hyphens in the domain */
2720
2721 if (message_id_domain)
2722 {
2723 uschar *new_id_domain = expand_string(message_id_domain);
2724 if (!new_id_domain)
2725 {
2726 if (!f.expand_string_forcedfail)
2727 log_write(0, LOG_MAIN|LOG_PANIC,
2728 "expansion of \"%s\" (message_id_header_domain) "
2729 "failed: %s", message_id_domain, expand_string_message);
2730 }
2731 else if (*new_id_domain)
2732 {
2733 id_domain = new_id_domain;
2734 for (uschar * p = id_domain; *p; p++)
2735 if (!isalnum(*p) && *p != '.') *p = '-'; /* No need to test '-' ! */
2736 }
2737 }
2738
2739 /* Permit all characters except controls and RFC 2822 specials in the
2740 additional text part. */
2741
2742 if (message_id_text)
2743 {
2744 uschar *new_id_text = expand_string(message_id_text);
2745 if (!new_id_text)
2746 {
2747 if (!f.expand_string_forcedfail)
2748 log_write(0, LOG_MAIN|LOG_PANIC,
2749 "expansion of \"%s\" (message_id_header_text) "
2750 "failed: %s", message_id_text, expand_string_message);
2751 }
2752 else if (*new_id_text)
2753 {
2754 id_text = new_id_text;
2755 for (uschar * p = id_text; *p; p++) if (mac_iscntrl_or_special(*p)) *p = '-';
2756 }
2757 }
2758
2759 /* Add the header line.
2760 Resent-* headers are prepended, per RFC 5322 3.6.6. Non-Resent-* are
2761 appended, to preserve classical expectations of header ordering. */
2762
2763 h = header_add_at_position_internal(!resents_exist, NULL, FALSE, htype_id,
2764 "%sMessage-Id: <%s%s%s@%s>\n", resent_prefix, message_id_external,
2765 *id_text == 0 ? "" : ".", id_text, id_domain);
2766
2767 /* Arrange for newly-created Message-Id to be logged */
2768
2769 if (!resents_exist)
2770 {
2771 msgid_header_newly_created = TRUE;
2772 msgid_header = h;
2773 }
2774 }
2775
2776 /* If we are to log recipients, keep a copy of the raw ones before any possible
2777 rewriting. Must copy the count, because later ACLs and the local_scan()
2778 function may mess with the real recipients. */
2779
2780 if (LOGGING(received_recipients))
2781 {
2782 raw_recipients = store_get(recipients_count * sizeof(uschar *), FALSE);
2783 for (int i = 0; i < recipients_count; i++)
2784 raw_recipients[i] = string_copy(recipients_list[i].address);
2785 raw_recipients_count = recipients_count;
2786 }
2787
2788 /* Ensure the recipients list is fully qualified and rewritten. Unqualified
2789 recipients will get here only if the conditions were right (allow_unqualified_
2790 recipient is TRUE). */
2791
2792 for (int i = 0; i < recipients_count; i++)
2793 recipients_list[i].address =
2794 rewrite_address(recipients_list[i].address, TRUE, TRUE,
2795 global_rewrite_rules, rewrite_existflags);
2796
2797 /* If there is no From: header, generate one for local (without
2798 suppress_local_fixups) or submission_mode messages. If there is no sender
2799 address, but the sender is local or this is a local delivery error, use the
2800 originator login. This shouldn't happen for genuine bounces, but might happen
2801 for autoreplies. The addition of From: must be done *before* checking for the
2802 possible addition of a Sender: header, because untrusted_set_sender allows an
2803 untrusted user to set anything in the envelope (which might then get info
2804 From:) but we still want to ensure a valid Sender: if it is required. */
2805
2806 if ( !from_header
2807 && ((!sender_host_address && !f.suppress_local_fixups) || f.submission_mode))
2808 {
2809 uschar *oname = US"";
2810
2811 /* Use the originator_name if this is a locally submitted message and the
2812 caller is not trusted. For trusted callers, use it only if -F was used to
2813 force its value or if we have a non-SMTP message for which -f was not used
2814 to set the sender. */
2815
2816 if (!sender_host_address)
2817 {
2818 if (!f.trusted_caller || f.sender_name_forced ||
2819 (!smtp_input && !f.sender_address_forced))
2820 oname = originator_name;
2821 }
2822
2823 /* For non-locally submitted messages, the only time we use the originator
2824 name is when it was forced by the /name= option on control=submission. */
2825
2826 else if (submission_name) oname = submission_name;
2827
2828 /* Envelope sender is empty */
2829
2830 if (!*sender_address)
2831 {
2832 uschar *fromstart, *fromend;
2833
2834 fromstart = string_sprintf("%sFrom: %s%s",
2835 resent_prefix, oname, *oname ? " <" : "");
2836 fromend = *oname ? US">" : US"";
2837
2838 if (f.sender_local || f.local_error_message)
2839 header_add(htype_from, "%s%s@%s%s\n", fromstart,
2840 local_part_quote(originator_login), qualify_domain_sender,
2841 fromend);
2842
2843 else if (f.submission_mode && authenticated_id)
2844 {
2845 if (!submission_domain)
2846 header_add(htype_from, "%s%s@%s%s\n", fromstart,
2847 local_part_quote(authenticated_id), qualify_domain_sender,
2848 fromend);
2849
2850 else if (!*submission_domain) /* empty => whole address set */
2851 header_add(htype_from, "%s%s%s\n", fromstart, authenticated_id,
2852 fromend);
2853
2854 else
2855 header_add(htype_from, "%s%s@%s%s\n", fromstart,
2856 local_part_quote(authenticated_id), submission_domain, fromend);
2857
2858 from_header = header_last; /* To get it checked for Sender: */
2859 }
2860 }
2861
2862 /* There is a non-null envelope sender. Build the header using the original
2863 sender address, before any rewriting that might have been done while
2864 verifying it. */
2865
2866 else
2867 {
2868 header_add(htype_from, "%sFrom: %s%s%s%s\n", resent_prefix,
2869 oname,
2870 *oname ? " <" : "",
2871 sender_address_unrewritten ? sender_address_unrewritten : sender_address,
2872 *oname ? ">" : "");
2873
2874 from_header = header_last; /* To get it checked for Sender: */
2875 }
2876 }
2877
2878
2879 /* If the sender is local (without suppress_local_fixups), or if we are in
2880 submission mode and there is an authenticated_id, check that an existing From:
2881 is correct, and if not, generate a Sender: header, unless disabled. Any
2882 previously-existing Sender: header was removed above. Note that sender_local,
2883 as well as being TRUE if the caller of exim is not trusted, is also true if a
2884 trusted caller did not supply a -f argument for non-smtp input. To allow
2885 trusted callers to forge From: without supplying -f, we have to test explicitly
2886 here. If the From: header contains more than one address, then the call to
2887 parse_extract_address fails, and a Sender: header is inserted, as required. */
2888
2889 if ( from_header
2890 && ( f.active_local_from_check
2891 && ( f.sender_local && !f.trusted_caller && !f.suppress_local_fixups
2892 || f.submission_mode && authenticated_id
2893 ) ) )
2894 {
2895 BOOL make_sender = TRUE;
2896 int start, end, domain;
2897 uschar *errmess;
2898 uschar *from_address =
2899 parse_extract_address(Ustrchr(from_header->text, ':') + 1, &errmess,
2900 &start, &end, &domain, FALSE);
2901 uschar *generated_sender_address;
2902
2903 generated_sender_address = f.submission_mode
2904 ? !submission_domain
2905 ? string_sprintf("%s@%s",
2906 local_part_quote(authenticated_id), qualify_domain_sender)
2907 : !*submission_domain /* empty => full address */
2908 ? string_sprintf("%s", authenticated_id)
2909 : string_sprintf("%s@%s",
2910 local_part_quote(authenticated_id), submission_domain)
2911 : string_sprintf("%s@%s",
2912 local_part_quote(originator_login), qualify_domain_sender);
2913
2914 /* Remove permitted prefixes and suffixes from the local part of the From:
2915 address before doing the comparison with the generated sender. */
2916
2917 if (from_address)
2918 {
2919 int slen;
2920 uschar *at = domain ? from_address + domain - 1 : NULL;
2921
2922 if (at) *at = 0;
2923 from_address += route_check_prefix(from_address, local_from_prefix, NULL);
2924 if ((slen = route_check_suffix(from_address, local_from_suffix, NULL)) > 0)
2925 {
2926 memmove(from_address+slen, from_address, Ustrlen(from_address)-slen);
2927 from_address += slen;
2928 }
2929 if (at) *at = '@';
2930
2931 if ( strcmpic(generated_sender_address, from_address) == 0
2932 || (!domain && strcmpic(from_address, originator_login) == 0))
2933 make_sender = FALSE;
2934 }
2935
2936 /* We have to cause the Sender header to be rewritten if there are
2937 appropriate rewriting rules. */
2938
2939 if (make_sender)
2940 if (f.submission_mode && !submission_name)
2941 header_add(htype_sender, "%sSender: %s\n", resent_prefix,
2942 generated_sender_address);
2943 else
2944 header_add(htype_sender, "%sSender: %s <%s>\n",
2945 resent_prefix,
2946 f.submission_mode ? submission_name : originator_name,
2947 generated_sender_address);
2948
2949 /* Ensure that a non-null envelope sender address corresponds to the
2950 submission mode sender address. */
2951
2952 if (f.submission_mode && *sender_address)
2953 {
2954 if (!sender_address_unrewritten)
2955 sender_address_unrewritten = sender_address;
2956 sender_address = generated_sender_address;
2957 if (Ustrcmp(sender_address_unrewritten, generated_sender_address) != 0)
2958 log_write(L_address_rewrite, LOG_MAIN,
2959 "\"%s\" from env-from rewritten as \"%s\" by submission mode",
2960 sender_address_unrewritten, generated_sender_address);
2961 }
2962 }
2963
2964 /* If there are any rewriting rules, apply them to the sender address, unless
2965 it has already been rewritten as part of verification for SMTP input. */
2966
2967 if (global_rewrite_rules && !sender_address_unrewritten && *sender_address)
2968 {
2969 sender_address = rewrite_address(sender_address, FALSE, TRUE,
2970 global_rewrite_rules, rewrite_existflags);
2971 DEBUG(D_receive|D_rewrite)
2972 debug_printf("rewritten sender = %s\n", sender_address);
2973 }
2974
2975
2976 /* The headers must be run through rewrite_header(), because it ensures that
2977 addresses are fully qualified, as well as applying any rewriting rules that may
2978 exist.
2979
2980 Qualification of header addresses in a message from a remote host happens only
2981 if the host is in sender_unqualified_hosts or recipient_unqualified hosts, as
2982 appropriate. For local messages, qualification always happens, unless -bnq is
2983 used to explicitly suppress it. No rewriting is done for an unqualified address
2984 that is left untouched.
2985
2986 We start at the second header, skipping our own Received:. This rewriting is
2987 documented as happening *after* recipient addresses are taken from the headers
2988 by the -t command line option. An added Sender: gets rewritten here. */
2989
2990 for (header_line * h = header_list->next; h; h = h->next)
2991 {
2992 header_line *newh = rewrite_header(h, NULL, NULL, global_rewrite_rules,
2993 rewrite_existflags, TRUE);
2994 if (newh) h = newh;
2995 }
2996
2997
2998 /* An RFC 822 (sic) message is not legal unless it has at least one of "to",
2999 "cc", or "bcc". Note that although the minimal examples in RFC 822 show just
3000 "to" or "bcc", the full syntax spec allows "cc" as well. If any resent- header
3001 exists, this applies to the set of resent- headers rather than the normal set.
3002
3003 The requirement for a recipient header has been removed in RFC 2822. At this
3004 point in the code, earlier versions of Exim added a To: header for locally
3005 submitted messages, and an empty Bcc: header for others. In the light of the
3006 changes in RFC 2822, this was dropped in November 2003. */
3007
3008
3009 /* If there is no date header, generate one if the message originates locally
3010 (i.e. not over TCP/IP) and suppress_local_fixups is not set, or if the
3011 submission mode flag is set. Messages without Date: are not valid, but it seems
3012 to be more confusing if Exim adds one to all remotely-originated messages.
3013 As per Message-Id, we prepend if resending, else append.
3014 */
3015
3016 if ( !date_header_exists
3017 && ((!sender_host_address && !f.suppress_local_fixups) || f.submission_mode))
3018 header_add_at_position(!resents_exist, NULL, FALSE, htype_other,
3019 "%sDate: %s\n", resent_prefix, tod_stamp(tod_full));
3020
3021 search_tidyup(); /* Free any cached resources */
3022
3023 /* Show the complete set of headers if debugging. Note that the first one (the
3024 new Received:) has not yet been set. */
3025
3026 DEBUG(D_receive)
3027 {
3028 debug_printf(">>Headers after rewriting and local additions:\n");
3029 for (header_line * h = header_list->next; h; h = h->next)
3030 debug_printf("%c %s", h->type, h->text);
3031 debug_printf("\n");
3032 }
3033
3034 /* The headers are now complete in store. If we are running in filter
3035 testing mode, that is all this function does. Return TRUE if the message
3036 ended with a dot. */
3037
3038 if (filter_test != FTEST_NONE)
3039 {
3040 process_info[process_info_len] = 0;
3041 return message_ended == END_DOT;
3042 }
3043
3044 /*XXX CHUNKING: need to cancel cutthrough under BDAT, for now. In future,
3045 think more if it could be handled. Cannot do onward CHUNKING unless
3046 inbound is, but inbound chunking ought to be ok with outbound plain.
3047 Could we do onward CHUNKING given inbound CHUNKING?
3048 */
3049 if (chunking_state > CHUNKING_OFFERED)
3050 cancel_cutthrough_connection(FALSE, US"chunking active");
3051
3052 /* Cutthrough delivery:
3053 We have to create the Received header now rather than at the end of reception,
3054 so the timestamp behaviour is a change to the normal case.
3055 Having created it, send the headers to the destination. */
3056
3057 if (cutthrough.cctx.sock >= 0 && cutthrough.delivery)
3058 {
3059 if (received_count > received_headers_max)
3060 {
3061 cancel_cutthrough_connection(TRUE, US"too many headers");
3062 if (smtp_input) receive_swallow_smtp(); /* Swallow incoming SMTP */
3063 log_write(0, LOG_MAIN|LOG_REJECT, "rejected from <%s>%s%s%s%s: "
3064 "Too many \"Received\" headers",
3065 sender_address,
3066 sender_fullhost ? "H=" : "", sender_fullhost ? sender_fullhost : US"",
3067 sender_ident ? "U=" : "", sender_ident ? sender_ident : US"");
3068 message_id[0] = 0; /* Indicate no message accepted */
3069 smtp_reply = US"550 Too many \"Received\" headers - suspected mail loop";
3070 goto TIDYUP; /* Skip to end of function */
3071 }
3072 received_header_gen();
3073 add_acl_headers(ACL_WHERE_RCPT, US"MAIL or RCPT");
3074 (void) cutthrough_headers_send();
3075 }
3076
3077
3078 /* Open a new spool file for the data portion of the message. We need
3079 to access it both via a file descriptor and a stream. Try to make the
3080 directory if it isn't there. */
3081
3082 spool_name = spool_fname(US"input", message_subdir, message_id, US"-D");
3083 DEBUG(D_receive) debug_printf("Data file name: %s\n", spool_name);
3084
3085 if ((data_fd = Uopen(spool_name, O_RDWR|O_CREAT|O_EXCL, SPOOL_MODE)) < 0)
3086 {
3087 if (errno == ENOENT)
3088 {
3089 (void) directory_make(spool_directory,
3090 spool_sname(US"input", message_subdir),
3091 INPUT_DIRECTORY_MODE, TRUE);
3092 data_fd = Uopen(spool_name, O_RDWR|O_CREAT|O_EXCL, SPOOL_MODE);
3093 }
3094 if (data_fd < 0)
3095 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Failed to create spool file %s: %s",
3096 spool_name, strerror(errno));
3097 }
3098
3099 /* Make sure the file's group is the Exim gid, and double-check the mode
3100 because the group setting doesn't always get set automatically. */
3101
3102 if (0 != exim_fchown(data_fd, exim_uid, exim_gid, spool_name))
3103 log_write(0, LOG_MAIN|LOG_PANIC_DIE,
3104 "Failed setting ownership on spool file %s: %s",
3105 spool_name, strerror(errno));
3106 (void)fchmod(data_fd, SPOOL_MODE);
3107
3108 /* We now have data file open. Build a stream for it and lock it. We lock only
3109 the first line of the file (containing the message ID) because otherwise there
3110 are problems when Exim is run under Cygwin (I'm told). See comments in
3111 spool_in.c, where the same locking is done. */
3112
3113 spool_data_file = fdopen(data_fd, "w+");
3114 lock_data.l_type = F_WRLCK;
3115 lock_data.l_whence = SEEK_SET;
3116 lock_data.l_start = 0;
3117 lock_data.l_len = SPOOL_DATA_START_OFFSET;
3118
3119 if (fcntl(data_fd, F_SETLK, &lock_data) < 0)
3120 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Cannot lock %s (%d): %s", spool_name,
3121 errno, strerror(errno));
3122
3123 /* We have an open, locked data file. Write the message id to it to make it
3124 self-identifying. Then read the remainder of the input of this message and
3125 write it to the data file. If the variable next != NULL, it contains the first
3126 data line (which was read as a header but then turned out not to have the right
3127 format); write it (remembering that it might contain binary zeros). The result
3128 of fwrite() isn't inspected; instead we call ferror() below. */
3129
3130 fprintf(spool_data_file, "%s-D\n", message_id);
3131 if (next)
3132 {
3133 uschar *s = next->text;
3134 int len = next->slen;
3135 if (fwrite(s, 1, len, spool_data_file) == len) /* "if" for compiler quietening */
3136 body_linecount++; /* Assumes only 1 line */
3137 }
3138
3139 /* Note that we might already be at end of file, or the logical end of file
3140 (indicated by '.'), or might have encountered an error while writing the
3141 message id or "next" line. */
3142
3143 if (!ferror(spool_data_file) && !(receive_feof)() && message_ended != END_DOT)
3144 {
3145 if (smtp_input)
3146 {
3147 message_ended = chunking_state <= CHUNKING_OFFERED
3148 ? read_message_data_smtp(spool_data_file)
3149 : spool_wireformat
3150 ? read_message_bdat_smtp_wire(spool_data_file)
3151 : read_message_bdat_smtp(spool_data_file);
3152 receive_linecount++; /* The terminating "." line */
3153 }
3154 else
3155 message_ended = read_message_data(spool_data_file);
3156
3157 receive_linecount += body_linecount; /* For BSMTP errors mainly */
3158 message_linecount += body_linecount;
3159
3160 switch (message_ended)
3161 {
3162 /* Handle premature termination of SMTP */
3163
3164 case END_EOF:
3165 if (smtp_input)
3166 {
3167 Uunlink(spool_name); /* Lose data file when closed */
3168 cancel_cutthrough_connection(TRUE, US"sender closed connection");
3169 message_id[0] = 0; /* Indicate no message accepted */
3170 smtp_reply = handle_lost_connection(US"");
3171 smtp_yield = FALSE;
3172 goto TIDYUP; /* Skip to end of function */
3173 }
3174 break;
3175
3176 /* Handle message that is too big. Don't use host_or_ident() in the log
3177 message; we want to see the ident value even for non-remote messages. */
3178
3179 case END_SIZE:
3180 Uunlink(spool_name); /* Lose the data file when closed */
3181 cancel_cutthrough_connection(TRUE, US"mail too big");
3182 if (smtp_input) receive_swallow_smtp(); /* Swallow incoming SMTP */
3183
3184 log_write(L_size_reject, LOG_MAIN|LOG_REJECT, "rejected from <%s>%s%s%s%s: "
3185 "message too big: read=%d max=%d",
3186 sender_address,
3187 sender_fullhost ? " H=" : "",
3188 sender_fullhost ? sender_fullhost : US"",
3189 sender_ident ? " U=" : "",
3190 sender_ident ? sender_ident : US"",
3191 message_size,
3192 thismessage_size_limit);
3193
3194 if (smtp_input)
3195 {
3196 smtp_reply = US"552 Message size exceeds maximum permitted";
3197 message_id[0] = 0; /* Indicate no message accepted */
3198 goto TIDYUP; /* Skip to end of function */
3199 }
3200 else
3201 {
3202 fseek(spool_data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3203 give_local_error(ERRMESS_TOOBIG,
3204 string_sprintf("message too big (max=%d)", thismessage_size_limit),
3205 US"message rejected: ", error_rc, spool_data_file, header_list);
3206 /* Does not return */
3207 }
3208 break;
3209
3210 /* Handle bad BDAT protocol sequence */
3211
3212 case END_PROTOCOL:
3213 Uunlink(spool_name); /* Lose the data file when closed */
3214 cancel_cutthrough_connection(TRUE, US"sender protocol error");
3215 smtp_reply = US""; /* Response already sent */
3216 message_id[0] = 0; /* Indicate no message accepted */
3217 goto TIDYUP; /* Skip to end of function */
3218 }
3219 }
3220
3221 /* Restore the standard SIGALRM handler for any subsequent processing. (For
3222 example, there may be some expansion in an ACL that uses a timer.) */
3223
3224 os_non_restarting_signal(SIGALRM, sigalrm_handler);
3225
3226 /* The message body has now been read into the data file. Call fflush() to
3227 empty the buffers in C, and then call fsync() to get the data written out onto
3228 the disk, as fflush() doesn't do this (or at least, it isn't documented as
3229 having to do this). If there was an I/O error on either input or output,
3230 attempt to send an error message, and unlink the spool file. For non-SMTP input
3231 we can then give up. Note that for SMTP input we must swallow the remainder of
3232 the input in cases of output errors, since the far end doesn't expect to see
3233 anything until the terminating dot line is sent. */
3234
3235 if (fflush(spool_data_file) == EOF || ferror(spool_data_file) ||
3236 EXIMfsync(fileno(spool_data_file)) < 0 || (receive_ferror)())
3237 {
3238 uschar *msg_errno = US strerror(errno);
3239 BOOL input_error = (receive_ferror)() != 0;
3240 uschar *msg = string_sprintf("%s error (%s) while receiving message from %s",
3241 input_error? "Input read" : "Spool write",
3242 msg_errno,
3243 sender_fullhost ? sender_fullhost : sender_ident);
3244
3245 log_write(0, LOG_MAIN, "Message abandoned: %s", msg);
3246 Uunlink(spool_name); /* Lose the data file */
3247 cancel_cutthrough_connection(TRUE, US"error writing spoolfile");
3248
3249 if (smtp_input)
3250 {
3251 if (input_error)
3252 smtp_reply = US"451 Error while reading input data";
3253 else
3254 {
3255 smtp_reply = US"451 Error while writing spool file";
3256 receive_swallow_smtp();
3257 }
3258 message_id[0] = 0; /* Indicate no message accepted */
3259 goto TIDYUP; /* Skip to end of function */
3260 }
3261
3262 else
3263 {
3264 fseek(spool_data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3265 give_local_error(ERRMESS_IOERR, msg, US"", error_rc, spool_data_file,
3266 header_list);
3267 /* Does not return */
3268 }
3269 }
3270
3271
3272 /* No I/O errors were encountered while writing the data file. */
3273
3274 DEBUG(D_receive) debug_printf("Data file written for message %s\n", message_id);
3275 if (LOGGING(receive_time)) timesince(&received_time_taken, &received_time);
3276
3277
3278 /* If there were any bad addresses extracted by -t, or there were no recipients
3279 left after -t, send a message to the sender of this message, or write it to
3280 stderr if the error handling option is set that way. Note that there may
3281 legitimately be no recipients for an SMTP message if they have all been removed
3282 by "discard".
3283
3284 We need to rewind the data file in order to read it. In the case of no
3285 recipients or stderr error writing, throw the data file away afterwards, and
3286 exit. (This can't be SMTP, which always ensures there's at least one
3287 syntactically good recipient address.) */
3288
3289 if (extract_recip && (bad_addresses || recipients_count == 0))
3290 {
3291 DEBUG(D_receive)
3292 {
3293 if (recipients_count == 0) debug_printf("*** No recipients\n");
3294 if (bad_addresses)
3295 {
3296 debug_printf("*** Bad address(es)\n");
3297 for (error_block * eblock = bad_addresses; eblock; eblock = eblock->next)
3298 debug_printf(" %s: %s\n", eblock->text1, eblock->text2);
3299 }
3300 }
3301
3302 log_write(0, LOG_MAIN|LOG_PANIC, "%s %s found in headers",
3303 message_id, bad_addresses ? "bad addresses" : "no recipients");
3304
3305 fseek(spool_data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3306
3307 /* If configured to send errors to the sender, but this fails, force
3308 a failure error code. We use a special one for no recipients so that it
3309 can be detected by the autoreply transport. Otherwise error_rc is set to
3310 errors_sender_rc, which is EXIT_FAILURE unless -oee was given, in which case
3311 it is EXIT_SUCCESS. */
3312
3313 if (error_handling == ERRORS_SENDER)
3314 {
3315 if (!moan_to_sender(
3316 bad_addresses
3317 ? recipients_list ? ERRMESS_BADADDRESS : ERRMESS_BADNOADDRESS
3318 : extracted_ignored ? ERRMESS_IGADDRESS : ERRMESS_NOADDRESS,
3319 bad_addresses, header_list, spool_data_file, FALSE
3320 ) )
3321 error_rc = bad_addresses ? EXIT_FAILURE : EXIT_NORECIPIENTS;
3322 }
3323 else
3324 {
3325 if (!bad_addresses)
3326 if (extracted_ignored)
3327 fprintf(stderr, "exim: all -t recipients overridden by command line\n");
3328 else
3329 fprintf(stderr, "exim: no recipients in message\n");
3330 else
3331 {
3332 fprintf(stderr, "exim: invalid address%s",
3333 bad_addresses->next ? "es:\n" : ":");
3334 for ( ; bad_addresses; bad_addresses = bad_addresses->next)
3335 fprintf(stderr, " %s: %s\n", bad_addresses->text1,
3336 bad_addresses->text2);
3337 }
3338 }
3339
3340 if (recipients_count == 0 || error_handling == ERRORS_STDERR)
3341 {
3342 Uunlink(spool_name);
3343 (void)fclose(spool_data_file);
3344 exim_exit(error_rc, US"receiving");
3345 }
3346 }
3347
3348 /* Data file successfully written. Generate text for the Received: header by
3349 expanding the configured string, and adding a timestamp. By leaving this
3350 operation till now, we ensure that the timestamp is the time that message
3351 reception was completed. However, this is deliberately done before calling the
3352 data ACL and local_scan().
3353
3354 This Received: header may therefore be inspected by the data ACL and by code in
3355 the local_scan() function. When they have run, we update the timestamp to be
3356 the final time of reception.
3357
3358 If there is just one recipient, set up its value in the $received_for variable
3359 for use when we generate the Received: header.
3360
3361 Note: the checking for too many Received: headers is handled by the delivery
3362 code. */
3363 /*XXX eventually add excess Received: check for cutthrough case back when classifying them */
3364
3365 if (!received_header->text) /* Non-cutthrough case */
3366 {
3367 received_header_gen();
3368
3369 /* Set the value of message_body_size for the DATA ACL and for local_scan() */
3370
3371 message_body_size = (fstat(data_fd, &statbuf) == 0)?
3372 statbuf.st_size - SPOOL_DATA_START_OFFSET : -1;
3373
3374 /* If an ACL from any RCPT commands set up any warning headers to add, do so
3375 now, before running the DATA ACL. */
3376
3377 add_acl_headers(ACL_WHERE_RCPT, US"MAIL or RCPT");
3378 }
3379 else
3380 message_body_size = (fstat(data_fd, &statbuf) == 0)?
3381 statbuf.st_size - SPOOL_DATA_START_OFFSET : -1;
3382
3383 /* If an ACL is specified for checking things at this stage of reception of a
3384 message, run it, unless all the recipients were removed by "discard" in earlier
3385 ACLs. That is the only case in which recipients_count can be zero at this
3386 stage. Set deliver_datafile to point to the data file so that $message_body and
3387 $message_body_end can be extracted if needed. Allow $recipients in expansions.
3388 */
3389
3390 deliver_datafile = data_fd;
3391 user_msg = NULL;
3392
3393 f.enable_dollar_recipients = TRUE;
3394
3395 if (recipients_count == 0)
3396 blackholed_by = f.recipients_discarded ? US"MAIL ACL" : US"RCPT ACL";
3397
3398 else
3399 {
3400 /* Handle interactive SMTP messages */
3401
3402 if (smtp_input && !smtp_batched_input)
3403 {
3404
3405 #ifndef DISABLE_DKIM
3406 if (!f.dkim_disable_verify)
3407 {
3408 /* Finish verification */
3409 dkim_exim_verify_finish();
3410
3411 /* Check if we must run the DKIM ACL */
3412 if (acl_smtp_dkim && dkim_verify_signers && *dkim_verify_signers)
3413 {
3414 uschar * dkim_verify_signers_expanded =
3415 expand_string(dkim_verify_signers);
3416 gstring * results = NULL;
3417 int signer_sep = 0;
3418 const uschar * ptr;
3419 uschar * item;
3420 gstring * seen_items = NULL;
3421 int old_pool = store_pool;
3422
3423 store_pool = POOL_PERM; /* Allow created variables to live to data ACL */
3424
3425 if (!(ptr = dkim_verify_signers_expanded))
3426 log_write(0, LOG_MAIN|LOG_PANIC,
3427 "expansion of dkim_verify_signers option failed: %s",
3428 expand_string_message);
3429
3430 /* Default to OK when no items are present */
3431 rc = OK;
3432 while ((item = string_nextinlist(&ptr, &signer_sep, NULL, 0)))
3433 {
3434 /* Prevent running ACL for an empty item */
3435 if (!item || !*item) continue;
3436
3437 /* Only run ACL once for each domain or identity,
3438 no matter how often it appears in the expanded list. */
3439 if (seen_items)
3440 {
3441 uschar * seen_item;
3442 const uschar * seen_items_list = string_from_gstring(seen_items);
3443 int seen_sep = ':';
3444 BOOL seen_this_item = FALSE;
3445
3446 while ((seen_item = string_nextinlist(&seen_items_list, &seen_sep,
3447 NULL, 0)))
3448 if (Ustrcmp(seen_item,item) == 0)
3449 {
3450 seen_this_item = TRUE;
3451 break;
3452 }
3453
3454 if (seen_this_item)
3455 {
3456 DEBUG(D_receive)
3457 debug_printf("acl_smtp_dkim: skipping signer %s, "
3458 "already seen\n", item);
3459 continue;
3460 }
3461
3462 seen_items = string_catn(seen_items, US":", 1);
3463 }
3464 seen_items = string_cat(seen_items, item);
3465
3466 rc = dkim_exim_acl_run(item, &results, &user_msg, &log_msg);
3467 if (rc != OK)
3468 {
3469 DEBUG(D_receive)
3470 debug_printf("acl_smtp_dkim: acl_check returned %d on %s, "
3471 "skipping remaining items\n", rc, item);
3472 cancel_cutthrough_connection(TRUE, US"dkim acl not ok");
3473 break;
3474 }
3475 }
3476 dkim_verify_status = string_from_gstring(results);
3477 store_pool = old_pool;
3478 add_acl_headers(ACL_WHERE_DKIM, US"DKIM");
3479 if (rc == DISCARD)
3480 {
3481 recipients_count = 0;
3482 blackholed_by = US"DKIM ACL";
3483 if (log_msg)
3484 blackhole_log_msg = string_sprintf(": %s", log_msg);
3485 }
3486 else if (rc != OK)
3487 {
3488 Uunlink(spool_name);
3489 if (smtp_handle_acl_fail(ACL_WHERE_DKIM, rc, user_msg, log_msg) != 0)
3490 smtp_yield = FALSE; /* No more messages after dropped connection */
3491 smtp_reply = US""; /* Indicate reply already sent */
3492 message_id[0] = 0; /* Indicate no message accepted */
3493 goto TIDYUP; /* Skip to end of function */
3494 }
3495 }
3496 else
3497 dkim_exim_verify_log_all();
3498 }
3499 #endif /* DISABLE_DKIM */
3500
3501 #ifdef WITH_CONTENT_SCAN
3502 if ( recipients_count > 0
3503 && acl_smtp_mime
3504 && !run_mime_acl(acl_smtp_mime, &smtp_yield, &smtp_reply, &blackholed_by)
3505 )
3506 goto TIDYUP;
3507 #endif /* WITH_CONTENT_SCAN */
3508
3509 #ifdef SUPPORT_DMARC
3510 dmarc_store_data(from_header);
3511 #endif
3512
3513 #ifndef DISABLE_PRDR
3514 if (prdr_requested && recipients_count > 1 && acl_smtp_data_prdr)
3515 {
3516 int all_pass = OK;
3517 int all_fail = FAIL;
3518
3519 smtp_printf("353 PRDR content analysis beginning\r\n", TRUE);
3520 /* Loop through recipients, responses must be in same order received */
3521 for (unsigned int c = 0; recipients_count > c; c++)
3522 {
3523 uschar * addr= recipients_list[c].address;
3524 uschar * msg= US"PRDR R=<%s> %s";
3525 uschar * code;
3526 DEBUG(D_receive)
3527 debug_printf("PRDR processing recipient %s (%d of %d)\n",
3528 addr, c+1, recipients_count);
3529 rc = acl_check(ACL_WHERE_PRDR, addr,
3530 acl_smtp_data_prdr, &user_msg, &log_msg);
3531
3532 /* If any recipient rejected content, indicate it in final message */
3533 all_pass |= rc;
3534 /* If all recipients rejected, indicate in final message */
3535 all_fail &= rc;
3536
3537 switch (rc)
3538 {
3539 case OK: case DISCARD: code = US"250"; break;
3540 case DEFER: code = US"450"; break;
3541 default: code = US"550"; break;
3542 }
3543 if (user_msg != NULL)
3544 smtp_user_msg(code, user_msg);
3545 else
3546 {
3547 switch (rc)
3548 {
3549 case OK: case DISCARD:
3550 msg = string_sprintf(CS msg, addr, "acceptance"); break;
3551 case DEFER:
3552 msg = string_sprintf(CS msg, addr, "temporary refusal"); break;
3553 default:
3554 msg = string_sprintf(CS msg, addr, "refusal"); break;
3555 }
3556 smtp_user_msg(code, msg);
3557 }
3558 if (log_msg) log_write(0, LOG_MAIN, "PRDR %s %s", addr, log_msg);
3559 else if (user_msg) log_write(0, LOG_MAIN, "PRDR %s %s", addr, user_msg);
3560 else log_write(0, LOG_MAIN, "%s", CS msg);
3561
3562 if (rc != OK) { receive_remove_recipient(addr); c--; }
3563 }
3564 /* Set up final message, used if data acl gives OK */
3565 smtp_reply = string_sprintf("%s id=%s message %s",
3566 all_fail == FAIL ? US"550" : US"250",
3567 message_id,
3568 all_fail == FAIL
3569 ? US"rejected for all recipients"
3570 : all_pass == OK
3571 ? US"accepted"
3572 : US"accepted for some recipients");
3573 if (recipients_count == 0)
3574 {
3575 message_id[0] = 0; /* Indicate no message accepted */
3576 goto TIDYUP;
3577 }
3578 }
3579 else
3580 prdr_requested = FALSE;
3581 #endif /* !DISABLE_PRDR */
3582
3583 /* Check the recipients count again, as the MIME ACL might have changed
3584 them. */
3585
3586 if (acl_smtp_data != NULL && recipients_count > 0)
3587 {
3588 rc = acl_check(ACL_WHERE_DATA, NULL, acl_smtp_data, &user_msg, &log_msg);
3589 add_acl_headers(ACL_WHERE_DATA, US"DATA");
3590 if (rc == DISCARD)
3591 {
3592 recipients_count = 0;
3593 blackholed_by = US"DATA ACL";
3594 if (log_msg)
3595 blackhole_log_msg = string_sprintf(": %s", log_msg);
3596 cancel_cutthrough_connection(TRUE, US"data acl discard");
3597 }
3598 else if (rc != OK)
3599 {
3600 Uunlink(spool_name);
3601 cancel_cutthrough_connection(TRUE, US"data acl not ok");
3602 #ifdef WITH_CONTENT_SCAN
3603 unspool_mbox();
3604 #endif
3605 #ifdef EXPERIMENTAL_DCC
3606 dcc_ok = 0;
3607 #endif
3608 if (smtp_handle_acl_fail(ACL_WHERE_DATA, rc, user_msg, log_msg) != 0)
3609 smtp_yield = FALSE; /* No more messages after dropped connection */
3610 smtp_reply = US""; /* Indicate reply already sent */
3611 message_id[0] = 0; /* Indicate no message accepted */
3612 goto TIDYUP; /* Skip to end of function */
3613 }
3614 }
3615 }
3616
3617 /* Handle non-SMTP and batch SMTP (i.e. non-interactive) messages. Note that
3618 we cannot take different actions for permanent and temporary rejections. */
3619
3620 else
3621 {
3622
3623 #ifdef WITH_CONTENT_SCAN
3624 if ( acl_not_smtp_mime
3625 && !run_mime_acl(acl_not_smtp_mime, &smtp_yield, &smtp_reply,
3626 &blackholed_by)
3627 )
3628 goto TIDYUP;
3629 #endif /* WITH_CONTENT_SCAN */
3630
3631 if (acl_not_smtp)
3632 {
3633 uschar *user_msg, *log_msg;
3634 f.authentication_local = TRUE;
3635 rc = acl_check(ACL_WHERE_NOTSMTP, NULL, acl_not_smtp, &user_msg, &log_msg);
3636 if (rc == DISCARD)
3637 {
3638 recipients_count = 0;
3639 blackholed_by = US"non-SMTP ACL";
3640 if (log_msg)
3641 blackhole_log_msg = string_sprintf(": %s", log_msg);
3642 }
3643 else if (rc != OK)
3644 {
3645 Uunlink(spool_name);
3646 #ifdef WITH_CONTENT_SCAN
3647 unspool_mbox();
3648 #endif
3649 #ifdef EXPERIMENTAL_DCC
3650 dcc_ok = 0;
3651 #endif
3652 /* The ACL can specify where rejections are to be logged, possibly
3653 nowhere. The default is main and reject logs. */
3654
3655 if (log_reject_target)
3656 log_write(0, log_reject_target, "F=<%s> rejected by non-SMTP ACL: %s",
3657 sender_address, log_msg);
3658
3659 if (!user_msg) user_msg = US"local configuration problem";
3660 if (smtp_batched_input)
3661 moan_smtp_batch(NULL, "%d %s", 550, user_msg);
3662 /* Does not return */
3663 else
3664 {
3665 fseek(spool_data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3666 give_local_error(ERRMESS_LOCAL_ACL, user_msg,
3667 US"message rejected by non-SMTP ACL: ", error_rc, spool_data_file,
3668 header_list);
3669 /* Does not return */
3670 }
3671 }
3672 add_acl_headers(ACL_WHERE_NOTSMTP, US"non-SMTP");
3673 }
3674 }
3675
3676 /* The applicable ACLs have been run */
3677
3678 if (f.deliver_freeze) frozen_by = US"ACL"; /* for later logging */
3679 if (f.queue_only_policy) queued_by = US"ACL";
3680 }
3681
3682 #ifdef WITH_CONTENT_SCAN
3683 unspool_mbox();
3684 #endif
3685
3686 #ifdef EXPERIMENTAL_DCC
3687 dcc_ok = 0;
3688 #endif
3689
3690
3691 #ifdef HAVE_LOCAL_SCAN
3692 /* The final check on the message is to run the scan_local() function. The
3693 version supplied with Exim always accepts, but this is a hook for sysadmins to
3694 supply their own checking code. The local_scan() function is run even when all
3695 the recipients have been discarded. */
3696
3697 lseek(data_fd, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3698
3699 /* Arrange to catch crashes in local_scan(), so that the -D file gets
3700 deleted, and the incident gets logged. */
3701
3702 if (sigsetjmp(local_scan_env, 1) == 0)
3703 {
3704 had_local_scan_crash = 0;
3705 os_non_restarting_signal(SIGSEGV, local_scan_crash_handler);
3706 os_non_restarting_signal(SIGFPE, local_scan_crash_handler);
3707 os_non_restarting_signal(SIGILL, local_scan_crash_handler);
3708 os_non_restarting_signal(SIGBUS, local_scan_crash_handler);
3709
3710 DEBUG(D_receive) debug_printf("calling local_scan(); timeout=%d\n",
3711 local_scan_timeout);
3712 local_scan_data = NULL;
3713
3714 had_local_scan_timeout = 0;
3715 os_non_restarting_signal(SIGALRM, local_scan_timeout_handler);
3716 if (local_scan_timeout > 0) ALARM(local_scan_timeout);
3717 rc = local_scan(data_fd, &local_scan_data);
3718 ALARM_CLR(0);
3719 os_non_restarting_signal(SIGALRM, sigalrm_handler);
3720
3721 f.enable_dollar_recipients = FALSE;
3722
3723 store_pool = POOL_MAIN; /* In case changed */
3724 DEBUG(D_receive) debug_printf("local_scan() returned %d %s\n", rc,
3725 local_scan_data);
3726
3727 os_non_restarting_signal(SIGSEGV, SIG_DFL);
3728 os_non_restarting_signal(SIGFPE, SIG_DFL);
3729 os_non_restarting_signal(SIGILL, SIG_DFL);
3730 os_non_restarting_signal(SIGBUS, SIG_DFL);
3731 }
3732 else
3733 {
3734 if (had_local_scan_crash)
3735 {
3736 log_write(0, LOG_MAIN|LOG_REJECT, "local_scan() function crashed with "
3737 "signal %d - message temporarily rejected (size %d)",
3738 had_local_scan_crash, message_size);
3739 receive_bomb_out(US"local-scan-error", US"local verification problem");
3740 /* Does not return */
3741 }
3742 if (had_local_scan_timeout)
3743 {
3744 log_write(0, LOG_MAIN|LOG_REJECT, "local_scan() function timed out - "
3745 "message temporarily rejected (size %d)", message_size);
3746 receive_bomb_out(US"local-scan-timeout", US"local verification problem");
3747 /* Does not return */
3748 }
3749 }
3750
3751 /* The length check is paranoia against some runaway code, and also because
3752 (for a success return) lines in the spool file are read into big_buffer. */
3753
3754 if (local_scan_data)
3755 {
3756 int len = Ustrlen(local_scan_data);
3757 if (len > LOCAL_SCAN_MAX_RETURN) len = LOCAL_SCAN_MAX_RETURN;
3758 local_scan_data = string_copyn(local_scan_data, len);
3759 }
3760
3761 if (rc == LOCAL_SCAN_ACCEPT_FREEZE)
3762 {
3763 if (!f.deliver_freeze) /* ACL might have already frozen */
3764 {
3765 f.deliver_freeze = TRUE;
3766 deliver_frozen_at = time(NULL);
3767 frozen_by = US"local_scan()";
3768 }
3769 rc = LOCAL_SCAN_ACCEPT;
3770 }
3771 else if (rc == LOCAL_SCAN_ACCEPT_QUEUE)
3772 {
3773 if (!f.queue_only_policy) /* ACL might have already queued */
3774 {
3775 f.queue_only_policy = TRUE;
3776 queued_by = US"local_scan()";
3777 }
3778 rc = LOCAL_SCAN_ACCEPT;
3779 }
3780
3781 /* Message accepted: remove newlines in local_scan_data because otherwise
3782 the spool file gets corrupted. Ensure that all recipients are qualified. */
3783
3784 if (rc == LOCAL_SCAN_ACCEPT)
3785 {
3786 if (local_scan_data)
3787 for (uschar * s = local_scan_data; *s != 0; s++) if (*s == '\n') *s = ' ';
3788 for (int i = 0; i < recipients_count; i++)
3789 {
3790 recipient_item *r = recipients_list + i;
3791 r->address = rewrite_address_qualify(r->address, TRUE);
3792 if (r->errors_to)
3793 r->errors_to = rewrite_address_qualify(r->errors_to, TRUE);
3794 }
3795 if (recipients_count == 0 && !blackholed_by)
3796 blackholed_by = US"local_scan";
3797 }
3798
3799 /* Message rejected: newlines permitted in local_scan_data to generate
3800 multiline SMTP responses. */
3801
3802 else
3803 {
3804 uschar *istemp = US"";
3805 uschar *smtp_code;
3806 gstring * g;
3807
3808 errmsg = local_scan_data;
3809
3810 Uunlink(spool_name); /* Cancel this message */
3811 switch(rc)
3812 {
3813 default:
3814 log_write(0, LOG_MAIN, "invalid return %d from local_scan(). Temporary "
3815 "rejection given", rc);
3816 goto TEMPREJECT;
3817
3818 case LOCAL_SCAN_REJECT_NOLOGHDR:
3819 BIT_CLEAR(log_selector, log_selector_size, Li_rejected_header);
3820 /* Fall through */
3821
3822 case LOCAL_SCAN_REJECT:
3823 smtp_code = US"550";
3824 if (!errmsg) errmsg = US"Administrative prohibition";
3825 break;
3826
3827 case LOCAL_SCAN_TEMPREJECT_NOLOGHDR:
3828 BIT_CLEAR(log_selector, log_selector_size, Li_rejected_header);
3829 /* Fall through */
3830
3831 case LOCAL_SCAN_TEMPREJECT:
3832 TEMPREJECT:
3833 smtp_code = US"451";
3834 if (!errmsg) errmsg = US"Temporary local problem";
3835 istemp = US"temporarily ";
3836 break;
3837 }
3838
3839 g = string_append(NULL, 2, US"F=",
3840 sender_address[0] == 0 ? US"<>" : sender_address);
3841 g = add_host_info_for_log(g);
3842
3843 log_write(0, LOG_MAIN|LOG_REJECT, "%s %srejected by local_scan(): %.256s",
3844 string_from_gstring(g), istemp, string_printing(errmsg));
3845
3846 if (smtp_input)
3847 if (!smtp_batched_input)
3848 {
3849 smtp_respond(smtp_code, 3, TRUE, errmsg);
3850 message_id[0] = 0; /* Indicate no message accepted */
3851 smtp_reply = US""; /* Indicate reply already sent */
3852 goto TIDYUP; /* Skip to end of function */
3853 }
3854 else
3855 moan_smtp_batch(NULL, "%s %s", smtp_code, errmsg);
3856 /* Does not return */
3857 else
3858 {
3859 fseek(spool_data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3860 give_local_error(ERRMESS_LOCAL_SCAN, errmsg,
3861 US"message rejected by local scan code: ", error_rc, spool_data_file,
3862 header_list);
3863 /* Does not return */
3864 }
3865 }
3866
3867 /* Reset signal handlers to ignore signals that previously would have caused
3868 the message to be abandoned. */
3869
3870 signal(SIGTERM, SIG_IGN);
3871 signal(SIGINT, SIG_IGN);
3872 #endif /* HAVE_LOCAL_SCAN */