Bugzilla 1041: pull patch id=425, DCC fixes.
[exim.git] / src / src / receive.c
CommitLineData
8544e77a 1/* $Cambridge: exim/src/src/receive.c,v 1.55 2010/06/05 11:13:30 pdp Exp $ */
059ec3d9
PH
2
3/*************************************************
4* Exim - an Internet mail transport agent *
5*************************************************/
6
0a49a7a4 7/* Copyright (c) University of Cambridge 1995 - 2009 */
059ec3d9
PH
8/* See the file NOTICE for conditions of use and distribution. */
9
10/* Code for receiving a message and setting up spool files. */
11
059ec3d9
PH
12#include "exim.h"
13
6a8f9482
TK
14#ifdef EXPERIMENTAL_DCC
15extern int dcc_ok;
16#endif
17
059ec3d9
PH
18/*************************************************
19* Local static variables *
20*************************************************/
21
22static FILE *data_file = NULL;
23static int data_fd = -1;
24static uschar spool_name[256];
25
26
27
28/*************************************************
29* Non-SMTP character reading functions *
30*************************************************/
31
32/* These are the default functions that are set up in the variables such as
33receive_getc initially. They just call the standard functions, passing stdin as
34the file. (When SMTP input is occurring, different functions are used by
35changing the pointer variables.) */
36
37int
38stdin_getc(void)
39{
40return getc(stdin);
41}
42
43int
44stdin_ungetc(int c)
45{
46return ungetc(c, stdin);
47}
48
49int
50stdin_feof(void)
51{
52return feof(stdin);
53}
54
55int
56stdin_ferror(void)
57{
58return ferror(stdin);
59}
60
61
62
63
64/*************************************************
65* Check that a set sender is allowed *
66*************************************************/
67
68/* This function is called when a local caller sets an explicit sender address.
69It checks whether this is permitted, which it is for trusted callers.
70Otherwise, it must match the pattern(s) in untrusted_set_sender.
71
72Arguments: the proposed sender address
73Returns: TRUE for a trusted caller
74 TRUE if the address has been set, untrusted_set_sender has been
75 set, and the address matches something in the list
76 FALSE otherwise
77*/
78
79BOOL
80receive_check_set_sender(uschar *newsender)
81{
82uschar *qnewsender;
83if (trusted_caller) return TRUE;
84if (newsender == NULL || untrusted_set_sender == NULL) return FALSE;
85qnewsender = (Ustrchr(newsender, '@') != NULL)?
86 newsender : string_sprintf("%s@%s", newsender, qualify_domain_sender);
87return
88 match_address_list(qnewsender, TRUE, TRUE, &untrusted_set_sender, NULL, -1,
89 0, NULL) == OK;
90}
91
92
93
94
95/*************************************************
5cb8cbc6 96* Read space info for a partition *
059ec3d9
PH
97*************************************************/
98
8e669ac1
PH
99/* This function is called by receive_check_fs() below, and also by string
100expansion for variables such as $spool_space. The field names for the statvfs
5cb8cbc6
PH
101structure are macros, because not all OS have F_FAVAIL and it seems tidier to
102have macros for F_BAVAIL and F_FILES as well. Some kinds of file system do not
103have inodes, and they return -1 for the number available.
059ec3d9 104
5cb8cbc6
PH
105Later: It turns out that some file systems that do not have the concept of
106inodes return 0 rather than -1. Such systems should also return 0 for the total
8e669ac1 107number of inodes, so we require that to be greater than zero before returning
5cb8cbc6 108an inode count.
059ec3d9 109
5cb8cbc6
PH
110Arguments:
111 isspool TRUE for spool partition, FALSE for log partition
112 inodeptr address of int to receive inode count; -1 if there isn't one
8e669ac1 113
5cb8cbc6 114Returns: available on-root space, in kilobytes
8e669ac1
PH
115 -1 for log partition if there isn't one
116
117All values are -1 if the STATFS functions are not available.
059ec3d9
PH
118*/
119
8e669ac1 120int
5cb8cbc6 121receive_statvfs(BOOL isspool, int *inodeptr)
059ec3d9
PH
122{
123#ifdef HAVE_STATFS
059ec3d9 124struct STATVFS statbuf;
5cb8cbc6
PH
125uschar *path;
126uschar *name;
127uschar buffer[1024];
059ec3d9 128
5cb8cbc6 129/* The spool directory must always exist. */
059ec3d9 130
5cb8cbc6 131if (isspool)
059ec3d9 132 {
8e669ac1
PH
133 path = spool_directory;
134 name = US"spool";
135 }
136
059ec3d9
PH
137/* Need to cut down the log file path to the directory, and to ignore any
138appearance of "syslog" in it. */
139
5cb8cbc6 140else
059ec3d9 141 {
059ec3d9 142 int sep = ':'; /* Not variable - outside scripts use */
059ec3d9 143 uschar *p = log_file_path;
8e669ac1 144 name = US"log";
059ec3d9
PH
145
146 /* An empty log_file_path means "use the default". This is the same as an
147 empty item in a list. */
148
149 if (*p == 0) p = US":";
150 while ((path = string_nextinlist(&p, &sep, buffer, sizeof(buffer))) != NULL)
151 {
152 if (Ustrcmp(path, "syslog") != 0) break;
153 }
154
5cb8cbc6
PH
155 if (path == NULL) /* No log files */
156 {
8e669ac1
PH
157 *inodeptr = -1;
158 return -1;
159 }
059ec3d9 160
8e669ac1
PH
161 /* An empty string means use the default, which is in the spool directory.
162 But don't just use the spool directory, as it is possible that the log
5cb8cbc6 163 subdirectory has been symbolically linked elsewhere. */
059ec3d9 164
8e669ac1 165 if (path[0] == 0)
059ec3d9 166 {
5cb8cbc6
PH
167 sprintf(CS buffer, CS"%s/log", CS spool_directory);
168 path = buffer;
8e669ac1
PH
169 }
170 else
059ec3d9 171 {
8e669ac1 172 uschar *cp;
5cb8cbc6 173 if ((cp = Ustrrchr(path, '/')) != NULL) *cp = 0;
8e669ac1 174 }
5cb8cbc6 175 }
8e669ac1 176
8f128379 177/* We now have the path; do the business */
5cb8cbc6
PH
178
179memset(&statbuf, 0, sizeof(statbuf));
180
181if (STATVFS(CS path, &statbuf) != 0)
182 {
183 log_write(0, LOG_MAIN|LOG_PANIC, "cannot accept message: failed to stat "
184 "%s directory %s: %s", name, spool_directory, strerror(errno));
185 smtp_closedown(US"spool or log directory problem");
186 exim_exit(EXIT_FAILURE);
187 }
8e669ac1 188
5cb8cbc6
PH
189*inodeptr = (statbuf.F_FILES > 0)? statbuf.F_FAVAIL : -1;
190
191/* Disks are getting huge. Take care with computing the size in kilobytes. */
8e669ac1 192
5cb8cbc6
PH
193return (int)(((double)statbuf.F_BAVAIL * (double)statbuf.F_FRSIZE)/1024.0);
194
195/* Unable to find partition sizes in this environment. */
196
197#else
198*inodeptr = -1;
199return -1;
200#endif
201}
202
059ec3d9 203
059ec3d9 204
5cb8cbc6
PH
205
206/*************************************************
207* Check space on spool and log partitions *
208*************************************************/
209
210/* This function is called before accepting a message; if any thresholds are
211set, it checks them. If a message_size is supplied, it checks that there is
212enough space for that size plus the threshold - i.e. that the message won't
213reduce the space to the threshold. Not all OS have statvfs(); for those that
214don't, this function always returns TRUE. For some OS the old function and
215struct name statfs is used; that is handled by a macro, defined in exim.h.
216
217Arguments:
218 msg_size the (estimated) size of an incoming message
219
220Returns: FALSE if there isn't enough space, or if the information cannot
221 be obtained
222 TRUE if no check was done or there is enough space
223*/
224
225BOOL
226receive_check_fs(int msg_size)
227{
228int space, inodes;
229
230if (check_spool_space > 0 || msg_size > 0 || check_spool_inodes > 0)
231 {
8e669ac1
PH
232 space = receive_statvfs(TRUE, &inodes);
233
059ec3d9 234 DEBUG(D_receive)
5cb8cbc6
PH
235 debug_printf("spool directory space = %dK inodes = %d "
236 "check_space = %dK inodes = %d msg_size = %d\n",
237 space, inodes, check_spool_space, check_spool_inodes, msg_size);
8e669ac1
PH
238
239 if ((space >= 0 && space < check_spool_space) ||
5cb8cbc6 240 (inodes >= 0 && inodes < check_spool_inodes))
8e669ac1 241 {
5cb8cbc6
PH
242 log_write(0, LOG_MAIN, "spool directory space check failed: space=%d "
243 "inodes=%d", space, inodes);
059ec3d9
PH
244 return FALSE;
245 }
246 }
247
5cb8cbc6
PH
248if (check_log_space > 0 || check_log_inodes > 0)
249 {
8e669ac1
PH
250 space = receive_statvfs(FALSE, &inodes);
251
5cb8cbc6
PH
252 DEBUG(D_receive)
253 debug_printf("log directory space = %dK inodes = %d "
254 "check_space = %dK inodes = %d\n",
255 space, inodes, check_log_space, check_log_inodes);
8e669ac1
PH
256
257 if ((space >= 0 && space < check_log_space) ||
5cb8cbc6 258 (inodes >= 0 && inodes < check_log_inodes))
8e669ac1 259 {
5cb8cbc6
PH
260 log_write(0, LOG_MAIN, "log directory space check failed: space=%d "
261 "inodes=%d", space, inodes);
262 return FALSE;
263 }
8e669ac1
PH
264 }
265
059ec3d9
PH
266return TRUE;
267}
268
269
270
271/*************************************************
272* Bomb out while reading a message *
273*************************************************/
274
275/* The common case of wanting to bomb out is if a SIGTERM or SIGINT is
276received, or if there is a timeout. A rarer case might be if the log files are
277screwed up and Exim can't open them to record a message's arrival. Handling
278that case is done by setting a flag to cause the log functions to call this
279function if there is an ultimate disaster. That is why it is globally
280accessible.
281
8f128379
PH
282Arguments:
283 reason text reason to pass to the not-quit ACL
284 msg default SMTP response to give if in an SMTP session
059ec3d9
PH
285Returns: it doesn't
286*/
287
288void
8f128379 289receive_bomb_out(uschar *reason, uschar *msg)
059ec3d9
PH
290{
291/* If spool_name is set, it contains the name of the data file that is being
292written. Unlink it before closing so that it cannot be picked up by a delivery
293process. Ensure that any header file is also removed. */
294
295if (spool_name[0] != 0)
296 {
297 Uunlink(spool_name);
298 spool_name[Ustrlen(spool_name) - 1] = 'H';
299 Uunlink(spool_name);
300 }
301
302/* Now close the file if it is open, either as a fd or a stream. */
303
f1e894f3
PH
304if (data_file != NULL) (void)fclose(data_file);
305 else if (data_fd >= 0) (void)close(data_fd);
059ec3d9 306
8f128379
PH
307/* Attempt to close down an SMTP connection tidily. For non-batched SMTP, call
308smtp_notquit_exit(), which runs the NOTQUIT ACL, if present, and handles the
309SMTP response. */
059ec3d9
PH
310
311if (smtp_input)
312 {
8f128379
PH
313 if (smtp_batched_input)
314 moan_smtp_batch(NULL, "421 %s - message abandoned", msg); /* No return */
315 smtp_notquit_exit(reason, US"421", US"%s %s - closing connection.",
316 smtp_active_hostname, msg);
059ec3d9
PH
317 }
318
319/* Exit from the program (non-BSMTP cases) */
320
321exim_exit(EXIT_FAILURE);
322}
323
324
325/*************************************************
326* Data read timeout *
327*************************************************/
328
329/* Handler function for timeouts that occur while reading the data that
330comprises a message.
331
332Argument: the signal number
333Returns: nothing
334*/
335
336static void
337data_timeout_handler(int sig)
338{
339uschar *msg = NULL;
340
341sig = sig; /* Keep picky compilers happy */
342
343if (smtp_input)
344 {
345 msg = US"SMTP incoming data timeout";
346 log_write(L_lost_incoming_connection,
347 LOG_MAIN, "SMTP data timeout (message abandoned) on connection "
fed77020
PH
348 "from %s F=<%s>",
349 (sender_fullhost != NULL)? sender_fullhost : US"local process",
350 sender_address);
059ec3d9
PH
351 }
352else
353 {
354 fprintf(stderr, "exim: timed out while reading - message abandoned\n");
355 log_write(L_lost_incoming_connection,
356 LOG_MAIN, "timed out while reading local message");
357 }
358
8f128379 359receive_bomb_out(US"data-timeout", msg); /* Does not return */
059ec3d9
PH
360}
361
362
363
364/*************************************************
365* local_scan() timeout *
366*************************************************/
367
368/* Handler function for timeouts that occur while running a local_scan()
369function.
370
371Argument: the signal number
372Returns: nothing
373*/
374
375static void
376local_scan_timeout_handler(int sig)
377{
378sig = sig; /* Keep picky compilers happy */
379log_write(0, LOG_MAIN|LOG_REJECT, "local_scan() function timed out - "
380 "message temporarily rejected (size %d)", message_size);
8f128379
PH
381/* Does not return */
382receive_bomb_out(US"local-scan-timeout", US"local verification problem");
059ec3d9
PH
383}
384
385
386
387/*************************************************
388* local_scan() crashed *
389*************************************************/
390
391/* Handler function for signals that occur while running a local_scan()
392function.
393
394Argument: the signal number
395Returns: nothing
396*/
397
398static void
399local_scan_crash_handler(int sig)
400{
401log_write(0, LOG_MAIN|LOG_REJECT, "local_scan() function crashed with "
402 "signal %d - message temporarily rejected (size %d)", sig, message_size);
8f128379
PH
403/* Does not return */
404receive_bomb_out(US"local-scan-error", US"local verification problem");
059ec3d9
PH
405}
406
407
408/*************************************************
409* SIGTERM or SIGINT received *
410*************************************************/
411
412/* Handler for SIGTERM or SIGINT signals that occur while reading the
413data that comprises a message.
414
415Argument: the signal number
416Returns: nothing
417*/
418
419static void
420data_sigterm_sigint_handler(int sig)
421{
422uschar *msg = NULL;
423
424if (smtp_input)
425 {
426 msg = US"Service not available - SIGTERM or SIGINT received";
427 log_write(0, LOG_MAIN, "%s closed after %s", smtp_get_connection_info(),
428 (sig == SIGTERM)? "SIGTERM" : "SIGINT");
429 }
430else
431 {
f05da2e8 432 if (filter_test == FTEST_NONE)
059ec3d9
PH
433 {
434 fprintf(stderr, "\nexim: %s received - message abandoned\n",
435 (sig == SIGTERM)? "SIGTERM" : "SIGINT");
436 log_write(0, LOG_MAIN, "%s received while reading local message",
437 (sig == SIGTERM)? "SIGTERM" : "SIGINT");
438 }
439 }
440
8f128379 441receive_bomb_out(US"signal-exit", msg); /* Does not return */
059ec3d9
PH
442}
443
444
445
446/*************************************************
447* Add new recipient to list *
448*************************************************/
449
450/* This function builds a list of recipient addresses in argc/argv
451format.
452
453Arguments:
454 recipient the next address to add to recipients_list
455 pno parent number for fixed aliases; -1 otherwise
456
457Returns: nothing
458*/
459
460void
461receive_add_recipient(uschar *recipient, int pno)
462{
463if (recipients_count >= recipients_list_max)
464 {
465 recipient_item *oldlist = recipients_list;
466 int oldmax = recipients_list_max;
467 recipients_list_max = recipients_list_max? 2*recipients_list_max : 50;
468 recipients_list = store_get(recipients_list_max * sizeof(recipient_item));
469 if (oldlist != NULL)
470 memcpy(recipients_list, oldlist, oldmax * sizeof(recipient_item));
471 }
472
473recipients_list[recipients_count].address = recipient;
474recipients_list[recipients_count].pno = pno;
8523533c
TK
475#ifdef EXPERIMENTAL_BRIGHTMAIL
476recipients_list[recipients_count].bmi_optin = bmi_current_optin;
477/* reset optin string pointer for next recipient */
478bmi_current_optin = NULL;
479#endif
059ec3d9
PH
480recipients_list[recipients_count++].errors_to = NULL;
481}
482
483
484
485
486/*************************************************
487* Remove a recipient from the list *
488*************************************************/
489
490/* This function is provided for local_scan() to use.
491
492Argument:
493 recipient address to remove
494
495Returns: TRUE if it did remove something; FALSE otherwise
496*/
497
498BOOL
499receive_remove_recipient(uschar *recipient)
500{
501int count;
502DEBUG(D_receive) debug_printf("receive_remove_recipient(\"%s\") called\n",
503 recipient);
504for (count = 0; count < recipients_count; count++)
505 {
506 if (Ustrcmp(recipients_list[count].address, recipient) == 0)
507 {
508 if ((--recipients_count - count) > 0)
509 memmove(recipients_list + count, recipients_list + count + 1,
54cdb463 510 (recipients_count - count)*sizeof(recipient_item));
059ec3d9
PH
511 return TRUE;
512 }
513 }
514return FALSE;
515}
516
517
518
519
520
521/*************************************************
522* Read data portion of a non-SMTP message *
523*************************************************/
524
525/* This function is called to read the remainder of a message (following the
526header) when the input is not from SMTP - we are receiving a local message on
527a standard input stream. The message is always terminated by EOF, and is also
528terminated by a dot on a line by itself if the flag dot_ends is TRUE. Split the
529two cases for maximum efficiency.
530
531Ensure that the body ends with a newline. This will naturally be the case when
532the termination is "\n.\n" but may not be otherwise. The RFC defines messages
533as "sequences of lines" - this of course strictly applies only to SMTP, but
534deliveries into BSD-type mailbox files also require it. Exim used to have a
535flag for doing this at delivery time, but as it was always set for all
536transports, I decided to simplify things by putting the check here instead.
537
538There is at least one MUA (dtmail) that sends CRLF via this interface, and
539other programs are known to do this as well. Exim used to have a option for
540dealing with this: in July 2003, after much discussion, the code has been
541changed to default to treat any of LF, CRLF, and bare CR as line terminators.
542
543However, for the case when a dot on a line by itself terminates a message, the
544only recognized terminating sequences before and after the dot are LF and CRLF.
545Otherwise, having read EOL . CR, you don't know whether to read another
546character or not.
547
548Internally, in messages stored in Exim's spool files, LF is used as the line
549terminator. Under the new regime, bare CRs will no longer appear in these
550files.
551
552Arguments:
553 fout a FILE to which to write the message
554
555Returns: One of the END_xxx values indicating why it stopped reading
556*/
557
558static int
559read_message_data(FILE *fout)
560{
561int ch_state;
562register int ch;
d677b2f2 563register int linelength = 0;
059ec3d9
PH
564
565/* Handle the case when only EOF terminates the message */
566
567if (!dot_ends)
568 {
569 register int last_ch = '\n';
570
80a47a2c 571 for (; (ch = (receive_getc)()) != EOF; last_ch = ch)
059ec3d9
PH
572 {
573 if (ch == 0) body_zerocount++;
574 if (last_ch == '\r' && ch != '\n')
575 {
d677b2f2
PH
576 if (linelength > max_received_linelength)
577 max_received_linelength = linelength;
578 linelength = 0;
059ec3d9
PH
579 if (fputc('\n', fout) == EOF) return END_WERROR;
580 message_size++;
581 body_linecount++;
582 }
583 if (ch == '\r') continue;
584
585 if (fputc(ch, fout) == EOF) return END_WERROR;
d677b2f2
PH
586 if (ch == '\n')
587 {
588 if (linelength > max_received_linelength)
589 max_received_linelength = linelength;
590 linelength = 0;
591 body_linecount++;
592 }
593 else linelength++;
059ec3d9
PH
594 if (++message_size > thismessage_size_limit) return END_SIZE;
595 }
596
597 if (last_ch != '\n')
598 {
d677b2f2
PH
599 if (linelength > max_received_linelength)
600 max_received_linelength = linelength;
059ec3d9
PH
601 if (fputc('\n', fout) == EOF) return END_WERROR;
602 message_size++;
603 body_linecount++;
604 }
605
606 return END_EOF;
607 }
608
609/* Handle the case when a dot on a line on its own, or EOF, terminates. */
610
611ch_state = 1;
612
80a47a2c 613while ((ch = (receive_getc)()) != EOF)
059ec3d9
PH
614 {
615 if (ch == 0) body_zerocount++;
616 switch (ch_state)
617 {
618 case 0: /* Normal state (previous char written) */
619 if (ch == '\n')
d677b2f2
PH
620 {
621 body_linecount++;
622 if (linelength > max_received_linelength)
623 max_received_linelength = linelength;
624 linelength = -1;
625 ch_state = 1;
626 }
059ec3d9
PH
627 else if (ch == '\r')
628 { ch_state = 2; continue; }
629 break;
630
631 case 1: /* After written "\n" */
632 if (ch == '.') { ch_state = 3; continue; }
d677b2f2 633 if (ch != '\n') ch_state = 0; else linelength = -1;
059ec3d9
PH
634 break;
635
636 case 2:
637 body_linecount++; /* After unwritten "\r" */
d677b2f2
PH
638 if (linelength > max_received_linelength)
639 max_received_linelength = linelength;
059ec3d9 640 if (ch == '\n')
d677b2f2
PH
641 {
642 ch_state = 1;
643 linelength = -1;
644 }
059ec3d9
PH
645 else
646 {
647 if (message_size++, fputc('\n', fout) == EOF) return END_WERROR;
648 if (ch == '\r') continue;
649 ch_state = 0;
d677b2f2 650 linelength = 0;
059ec3d9
PH
651 }
652 break;
653
654 case 3: /* After "\n." (\n written, dot not) */
655 if (ch == '\n') return END_DOT;
656 if (ch == '\r') { ch_state = 4; continue; }
657 message_size++;
d677b2f2 658 linelength++;
059ec3d9
PH
659 if (fputc('.', fout) == EOF) return END_WERROR;
660 ch_state = 0;
661 break;
662
663 case 4: /* After "\n.\r" (\n written, rest not) */
664 if (ch == '\n') return END_DOT;
665 message_size += 2;
666 body_linecount++;
667 if (fputs(".\n", fout) == EOF) return END_WERROR;
668 if (ch == '\r') { ch_state = 2; continue; }
669 ch_state = 0;
670 break;
671 }
672
d677b2f2 673 linelength++;
059ec3d9
PH
674 if (fputc(ch, fout) == EOF) return END_WERROR;
675 if (++message_size > thismessage_size_limit) return END_SIZE;
676 }
677
678/* Get here if EOF read. Unless we have just written "\n", we need to ensure
679the message ends with a newline, and we must also write any characters that
680were saved up while testing for an ending dot. */
681
682if (ch_state != 1)
683 {
684 static uschar *ends[] = { US"\n", NULL, US"\n", US".\n", US".\n" };
685 if (fputs(CS ends[ch_state], fout) == EOF) return END_WERROR;
686 message_size += Ustrlen(ends[ch_state]);
687 body_linecount++;
688 }
689
690return END_EOF;
691}
692
693
694
695
696/*************************************************
697* Read data portion of an SMTP message *
698*************************************************/
699
700/* This function is called to read the remainder of an SMTP message (after the
701headers), or to skip over it when an error has occurred. In this case, the
702output file is passed as NULL.
703
704If any line begins with a dot, that character is skipped. The input should only
705be successfully terminated by CR LF . CR LF unless it is local (non-network)
706SMTP, in which case the CRs are optional, but...
707
708FUDGE: It seems that sites on the net send out messages with just LF
709terminators, despite the warnings in the RFCs, and other MTAs handle this. So
710we make the CRs optional in all cases.
711
712July 2003: Bare CRs cause trouble. We now treat them as line terminators as
713well, so that there are no CRs in spooled messages. However, the message
714terminating dot is not recognized between two bare CRs.
715
716Arguments:
717 fout a FILE to which to write the message; NULL if skipping
718
719Returns: One of the END_xxx values indicating why it stopped reading
720*/
721
722static int
723read_message_data_smtp(FILE *fout)
724{
725int ch_state = 0;
726register int ch;
1f5497b2 727register int linelength = 0;
059ec3d9 728
80a47a2c 729while ((ch = (receive_getc)()) != EOF)
059ec3d9
PH
730 {
731 if (ch == 0) body_zerocount++;
732 switch (ch_state)
733 {
734 case 0: /* After LF or CRLF */
735 if (ch == '.')
736 {
737 ch_state = 3;
738 continue; /* Don't ever write . after LF */
739 }
740 ch_state = 1;
741
742 /* Else fall through to handle as normal uschar. */
743
744 case 1: /* Normal state */
745 if (ch == '\n')
746 {
747 ch_state = 0;
748 body_linecount++;
1f5497b2
PH
749 if (linelength > max_received_linelength)
750 max_received_linelength = linelength;
751 linelength = -1;
059ec3d9
PH
752 }
753 else if (ch == '\r')
754 {
755 ch_state = 2;
756 continue;
757 }
758 break;
759
760 case 2: /* After (unwritten) CR */
761 body_linecount++;
1f5497b2
PH
762 if (linelength > max_received_linelength)
763 max_received_linelength = linelength;
764 linelength = -1;
059ec3d9
PH
765 if (ch == '\n')
766 {
767 ch_state = 0;
768 }
769 else
770 {
771 message_size++;
772 if (fout != NULL && fputc('\n', fout) == EOF) return END_WERROR;
773 if (ch != '\r') ch_state = 1; else continue;
774 }
775 break;
776
777 case 3: /* After [CR] LF . */
778 if (ch == '\n')
779 return END_DOT;
780 if (ch == '\r')
781 {
782 ch_state = 4;
783 continue;
784 }
785 ch_state = 1; /* The dot itself is removed */
786 break;
787
788 case 4: /* After [CR] LF . CR */
789 if (ch == '\n') return END_DOT;
790 message_size++;
791 body_linecount++;
792 if (fout != NULL && fputc('\n', fout) == EOF) return END_WERROR;
793 if (ch == '\r')
794 {
795 ch_state = 2;
796 continue;
797 }
798 ch_state = 1;
799 break;
800 }
801
802 /* Add the character to the spool file, unless skipping; then loop for the
803 next. */
804
805 message_size++;
1f5497b2 806 linelength++;
059ec3d9
PH
807 if (fout != NULL)
808 {
809 if (fputc(ch, fout) == EOF) return END_WERROR;
810 if (message_size > thismessage_size_limit) return END_SIZE;
811 }
812 }
813
814/* Fall through here if EOF encountered. This indicates some kind of error,
815since a correct message is terminated by [CR] LF . [CR] LF. */
816
817return END_EOF;
818}
819
820
821
822
823/*************************************************
824* Swallow SMTP message *
825*************************************************/
826
827/* This function is called when there has been some kind of error while reading
828an SMTP message, and the remaining data may need to be swallowed. It is global
829because it is called from smtp_closedown() to shut down an incoming call
830tidily.
831
832Argument: a FILE from which to read the message
833Returns: nothing
834*/
835
836void
837receive_swallow_smtp(void)
838{
839if (message_ended >= END_NOTENDED)
840 message_ended = read_message_data_smtp(NULL);
841}
842
843
844
845/*************************************************
846* Handle lost SMTP connection *
847*************************************************/
848
849/* This function logs connection loss incidents and generates an appropriate
850SMTP response.
851
852Argument: additional data for the message
853Returns: the SMTP response
854*/
855
856static uschar *
857handle_lost_connection(uschar *s)
858{
859log_write(L_lost_incoming_connection | L_smtp_connection, LOG_MAIN,
860 "%s lost while reading message data%s", smtp_get_connection_info(), s);
861return US"421 Lost incoming connection";
862}
863
864
865
866
867/*************************************************
868* Handle a non-smtp reception error *
869*************************************************/
870
871/* This function is called for various errors during the reception of non-SMTP
872messages. It either sends a message to the sender of the problem message, or it
873writes to the standard error stream.
874
875Arguments:
876 errcode code for moan_to_sender(), identifying the error
877 text1 first message text, passed to moan_to_sender()
878 text2 second message text, used only for stderrr
879 error_rc code to pass to exim_exit if no problem
880 f FILE containing body of message (may be stdin)
881 hptr pointer to instore headers or NULL
882
883Returns: calls exim_exit(), which does not return
884*/
885
886static void
887give_local_error(int errcode, uschar *text1, uschar *text2, int error_rc,
888 FILE *f, header_line *hptr)
889{
890if (error_handling == ERRORS_SENDER)
891 {
892 error_block eblock;
893 eblock.next = NULL;
894 eblock.text1 = text1;
895 if (!moan_to_sender(errcode, &eblock, hptr, f, FALSE))
896 error_rc = EXIT_FAILURE;
897 }
898else fprintf(stderr, "exim: %s%s\n", text2, text1); /* Sic */
f1e894f3 899(void)fclose(f);
059ec3d9
PH
900exim_exit(error_rc);
901}
902
903
904
905/*************************************************
906* Add header lines set up by ACL *
907*************************************************/
908
850635b6
PH
909/* This function is called to add the header lines that were set up by
910statements in an ACL to the list of headers in memory. It is done in two stages
911like this, because when the ACL for RCPT is running, the other headers have not
912yet been received. This function is called twice; once just before running the
913DATA ACL, and once after. This is so that header lines added by MAIL or RCPT
914are visible to the DATA ACL.
059ec3d9
PH
915
916Originally these header lines were added at the end. Now there is support for
917three different places: top, bottom, and after the Received: header(s). There
918will always be at least one Received: header, even if it is marked deleted, and
919even if something else has been put in front of it.
920
921Arguments:
922 acl_name text to identify which ACL
923
924Returns: nothing
925*/
926
927static void
928add_acl_headers(uschar *acl_name)
929{
930header_line *h, *next;
931header_line *last_received = NULL;
932
71fafd95 933if (acl_added_headers == NULL) return;
059ec3d9
PH
934DEBUG(D_receive|D_acl) debug_printf(">>Headers added by %s ACL:\n", acl_name);
935
71fafd95 936for (h = acl_added_headers; h != NULL; h = next)
059ec3d9
PH
937 {
938 next = h->next;
939
940 switch(h->type)
941 {
942 case htype_add_top:
943 h->next = header_list;
944 header_list = h;
945 DEBUG(D_receive|D_acl) debug_printf(" (at top)");
946 break;
947
948 case htype_add_rec:
949 if (last_received == NULL)
950 {
951 last_received = header_list;
952 while (!header_testname(last_received, US"Received", 8, FALSE))
953 last_received = last_received->next;
954 while (last_received->next != NULL &&
955 header_testname(last_received->next, US"Received", 8, FALSE))
956 last_received = last_received->next;
957 }
958 h->next = last_received->next;
959 last_received->next = h;
960 DEBUG(D_receive|D_acl) debug_printf(" (after Received:)");
961 break;
962
8523533c
TK
963 case htype_add_rfc:
964 /* add header before any header which is NOT Received: or Resent- */
965 last_received = header_list;
966 while ( (last_received->next != NULL) &&
967 ( (header_testname(last_received->next, US"Received", 8, FALSE)) ||
968 (header_testname_incomplete(last_received->next, US"Resent-", 7, FALSE)) ) )
969 last_received = last_received->next;
970 /* last_received now points to the last Received: or Resent-* header
971 in an uninterrupted chain of those header types (seen from the beginning
972 of all headers. Our current header must follow it. */
973 h->next = last_received->next;
974 last_received->next = h;
8e669ac1 975 DEBUG(D_receive|D_acl) debug_printf(" (before any non-Received: or Resent-*: header)");
8523533c
TK
976 break;
977
059ec3d9
PH
978 default:
979 h->next = NULL;
980 header_last->next = h;
981 break;
982 }
983
984 if (h->next == NULL) header_last = h;
985
986 /* Check for one of the known header types (From:, To:, etc.) though in
987 practice most added headers are going to be "other". Lower case
988 identification letters are never stored with the header; they are used
989 for existence tests when messages are received. So discard any lower case
990 flag values. */
991
992 h->type = header_checkname(h, FALSE);
993 if (h->type >= 'a') h->type = htype_other;
994
995 DEBUG(D_receive|D_acl) debug_printf(" %s", header_last->text);
996 }
997
71fafd95 998acl_added_headers = NULL;
059ec3d9
PH
999DEBUG(D_receive|D_acl) debug_printf(">>\n");
1000}
1001
1002
1003
1004/*************************************************
1005* Add host information for log line *
1006*************************************************/
1007
1008/* Called for acceptance and rejecting log lines. This adds information about
1009the calling host to a string that is being built dynamically.
1010
1011Arguments:
1012 s the dynamic string
1013 sizeptr points to the size variable
1014 ptrptr points to the pointer variable
1015
1016Returns: the extended string
1017*/
1018
1019static uschar *
1020add_host_info_for_log(uschar *s, int *sizeptr, int *ptrptr)
1021{
1022if (sender_fullhost != NULL)
1023 {
1024 s = string_append(s, sizeptr, ptrptr, 2, US" H=", sender_fullhost);
1025 if ((log_extra_selector & LX_incoming_interface) != 0 &&
1026 interface_address != NULL)
1027 {
1028 uschar *ss = string_sprintf(" I=[%s]:%d", interface_address,
1029 interface_port);
1030 s = string_cat(s, sizeptr, ptrptr, ss, Ustrlen(ss));
1031 }
1032 }
1033if (sender_ident != NULL)
1034 s = string_append(s, sizeptr, ptrptr, 2, US" U=", sender_ident);
1035if (received_protocol != NULL)
1036 s = string_append(s, sizeptr, ptrptr, 2, US" P=", received_protocol);
1037return s;
1038}
1039
1040
1041
63955bf2 1042#ifdef WITH_CONTENT_SCAN
059ec3d9
PH
1043
1044/*************************************************
54cdb463
PH
1045* Run the MIME ACL on a message *
1046*************************************************/
1047
1048/* This code is in a subroutine so that it can be used for both SMTP
1049and non-SMTP messages. It is called with a non-NULL ACL pointer.
1050
1051Arguments:
1052 acl The ACL to run (acl_smtp_mime or acl_not_smtp_mime)
1053 smtp_yield_ptr Set FALSE to kill messages after dropped connection
1054 smtp_reply_ptr Where SMTP reply is being built
1055 blackholed_by_ptr Where "blackholed by" message is being built
1056
1057Returns: TRUE to carry on; FALSE to abandon the message
1058*/
1059
1060static BOOL
1061run_mime_acl(uschar *acl, BOOL *smtp_yield_ptr, uschar **smtp_reply_ptr,
1062 uschar **blackholed_by_ptr)
1063{
1064FILE *mbox_file;
1065uschar rfc822_file_path[2048];
1066unsigned long mbox_size;
1067header_line *my_headerlist;
1068uschar *user_msg, *log_msg;
1069int mime_part_count_buffer = -1;
7156b1ef 1070int rc = OK;
54cdb463
PH
1071
1072memset(CS rfc822_file_path,0,2048);
1073
1074/* check if it is a MIME message */
1075my_headerlist = header_list;
4e88a19f
PH
1076while (my_headerlist != NULL)
1077 {
54cdb463 1078 /* skip deleted headers */
4e88a19f
PH
1079 if (my_headerlist->type == '*')
1080 {
54cdb463
PH
1081 my_headerlist = my_headerlist->next;
1082 continue;
4e88a19f
PH
1083 }
1084 if (strncmpic(my_headerlist->text, US"Content-Type:", 13) == 0)
1085 {
54cdb463
PH
1086 DEBUG(D_receive) debug_printf("Found Content-Type: header - executing acl_smtp_mime.\n");
1087 goto DO_MIME_ACL;
4e88a19f 1088 }
54cdb463 1089 my_headerlist = my_headerlist->next;
4e88a19f 1090 }
54cdb463
PH
1091
1092DEBUG(D_receive) debug_printf("No Content-Type: header - presumably not a MIME message.\n");
1093return TRUE;
1094
1095DO_MIME_ACL:
1096/* make sure the eml mbox file is spooled up */
8544e77a 1097mbox_file = spool_mbox(&mbox_size, NULL);
54cdb463
PH
1098if (mbox_file == NULL) {
1099 /* error while spooling */
1100 log_write(0, LOG_MAIN|LOG_PANIC,
1101 "acl_smtp_mime: error while creating mbox spool file, message temporarily rejected.");
1102 Uunlink(spool_name);
1103 unspool_mbox();
6f0c431a
PP
1104#ifdef EXPERIMENTAL_DCC
1105 dcc_ok = 0;
1106#endif
a5bd321b 1107 smtp_respond(US"451", 3, TRUE, US"temporary local problem");
54cdb463
PH
1108 message_id[0] = 0; /* Indicate no message accepted */
1109 *smtp_reply_ptr = US""; /* Indicate reply already sent */
1110 return FALSE; /* Indicate skip to end of receive function */
1111};
1112
1113mime_is_rfc822 = 0;
1114
1115MIME_ACL_CHECK:
1116mime_part_count = -1;
1117rc = mime_acl_check(acl, mbox_file, NULL, &user_msg, &log_msg);
f1e894f3 1118(void)fclose(mbox_file);
54cdb463 1119
4e88a19f
PH
1120if (Ustrlen(rfc822_file_path) > 0)
1121 {
54cdb463
PH
1122 mime_part_count = mime_part_count_buffer;
1123
4e88a19f
PH
1124 if (unlink(CS rfc822_file_path) == -1)
1125 {
54cdb463
PH
1126 log_write(0, LOG_PANIC,
1127 "acl_smtp_mime: can't unlink RFC822 spool file, skipping.");
1128 goto END_MIME_ACL;
4e88a19f
PH
1129 }
1130 }
54cdb463
PH
1131
1132/* check if we must check any message/rfc822 attachments */
4e88a19f
PH
1133if (rc == OK)
1134 {
54cdb463
PH
1135 uschar temp_path[1024];
1136 int n;
1137 struct dirent *entry;
1138 DIR *tempdir;
1139
b07e6aa3
PH
1140 (void)string_format(temp_path, 1024, "%s/scan/%s", spool_directory,
1141 message_id);
54cdb463 1142
4e88a19f
PH
1143 tempdir = opendir(CS temp_path);
1144 n = 0;
1145 do
1146 {
1147 entry = readdir(tempdir);
1148 if (entry == NULL) break;
1149 if (strncmpic(US entry->d_name,US"__rfc822_",9) == 0)
1150 {
b07e6aa3 1151 (void)string_format(rfc822_file_path, 2048,"%s/scan/%s/%s", spool_directory, message_id, entry->d_name);
4e88a19f
PH
1152 debug_printf("RFC822 attachment detected: running MIME ACL for '%s'\n", rfc822_file_path);
1153 break;
1154 }
1155 } while (1);
1156 closedir(tempdir);
54cdb463 1157
4e88a19f
PH
1158 if (entry != NULL)
1159 {
c58b88df 1160 mbox_file = Ufopen(rfc822_file_path,"rb");
4e88a19f
PH
1161 if (mbox_file == NULL)
1162 {
54cdb463
PH
1163 log_write(0, LOG_PANIC,
1164 "acl_smtp_mime: can't open RFC822 spool file, skipping.");
1165 unlink(CS rfc822_file_path);
1166 goto END_MIME_ACL;
4e88a19f 1167 }
54cdb463
PH
1168 /* set RFC822 expansion variable */
1169 mime_is_rfc822 = 1;
1170 mime_part_count_buffer = mime_part_count;
1171 goto MIME_ACL_CHECK;
4e88a19f
PH
1172 }
1173 }
54cdb463
PH
1174
1175END_MIME_ACL:
1176add_acl_headers(US"MIME");
1177if (rc == DISCARD)
1178 {
1179 recipients_count = 0;
1180 *blackholed_by_ptr = US"MIME ACL";
1181 }
1182else if (rc != OK)
1183 {
1184 Uunlink(spool_name);
1185 unspool_mbox();
6f0c431a
PP
1186#ifdef EXPERIMENTAL_DCC
1187 dcc_ok = 0;
1188#endif
54cdb463
PH
1189 if (smtp_handle_acl_fail(ACL_WHERE_MIME, rc, user_msg, log_msg) != 0)
1190 *smtp_yield_ptr = FALSE; /* No more messsages after dropped connection */
1191 *smtp_reply_ptr = US""; /* Indicate reply already sent */
1192 message_id[0] = 0; /* Indicate no message accepted */
1193 return FALSE; /* Cause skip to end of receive function */
4e88a19f 1194 }
54cdb463
PH
1195
1196return TRUE;
1197}
1198
63955bf2 1199#endif /* WITH_CONTENT_SCAN */
54cdb463
PH
1200
1201
1202/*************************************************
059ec3d9
PH
1203* Receive message *
1204*************************************************/
1205
1206/* Receive a message on the given input, and put it into a pair of spool files.
1207Either a non-null list of recipients, or the extract flag will be true, or
1208both. The flag sender_local is true for locally generated messages. The flag
1209submission_mode is true if an ACL has obeyed "control = submission". The flag
8800895a
PH
1210suppress_local_fixups is true if an ACL has obeyed "control =
1211suppress_local_fixups". The flag smtp_input is true if the message is to be
1212handled using SMTP conventions about termination and lines starting with dots.
1213For non-SMTP messages, dot_ends is true for dot-terminated messages.
059ec3d9
PH
1214
1215If a message was successfully read, message_id[0] will be non-zero.
1216
1217The general actions of this function are:
1218
1219 . Read the headers of the message (if any) into a chain of store
1220 blocks.
1221
1222 . If there is a "sender:" header and the message is locally originated,
69358f02
PH
1223 throw it away, unless the caller is trusted, or unless
1224 active_local_sender_retain is set - which can only happen if
1225 active_local_from_check is false.
059ec3d9
PH
1226
1227 . If recipients are to be extracted from the message, build the
1228 recipients list from the headers, removing any that were on the
1229 original recipients list (unless extract_addresses_remove_arguments is
1230 false), and at the same time, remove any bcc header that may be present.
1231
1232 . Get the spool file for the data, sort out its unique name, open
1233 and lock it (but don't give it the name yet).
1234
1235 . Generate a "Message-Id" header if the message doesn't have one, for
1236 locally-originated messages.
1237
1238 . Generate a "Received" header.
1239
1240 . Ensure the recipients list is fully qualified and rewritten if necessary.
1241
1242 . If there are any rewriting rules, apply them to the sender address
1243 and also to the headers.
1244
1245 . If there is no from: header, generate one, for locally-generated messages
1246 and messages in "submission mode" only.
1247
1248 . If the sender is local, check that from: is correct, and if not, generate
1249 a Sender: header, unless message comes from a trusted caller, or this
69358f02 1250 feature is disabled by active_local_from_check being false.
059ec3d9
PH
1251
1252 . If there is no "date" header, generate one, for locally-originated
1253 or submission mode messages only.
1254
1255 . Copy the rest of the input, or up to a terminating "." if in SMTP or
1256 dot_ends mode, to the data file. Leave it open, to hold the lock.
1257
1258 . Write the envelope and the headers to a new file.
1259
1260 . Set the name for the header file; close it.
1261
1262 . Set the name for the data file; close it.
1263
1264Because this function can potentially be called many times in a single
1265SMTP connection, all store should be got by store_get(), so that it will be
1266automatically retrieved after the message is accepted.
1267
1268FUDGE: It seems that sites on the net send out messages with just LF
1269terminators, despite the warnings in the RFCs, and other MTAs handle this. So
1270we make the CRs optional in all cases.
1271
1272July 2003: Bare CRs in messages, especially in header lines, cause trouble. A
1273new regime is now in place in which bare CRs in header lines are turned into LF
1274followed by a space, so as not to terminate the header line.
1275
1276February 2004: A bare LF in a header line in a message whose first line was
1277terminated by CRLF is treated in the same way as a bare CR.
1278
1279Arguments:
1280 extract_recip TRUE if recipients are to be extracted from the message's
1281 headers
1282
1283Returns: TRUE there are more messages to be read (SMTP input)
1284 FALSE there are no more messages to be read (non-SMTP input
1285 or SMTP connection collapsed, or other failure)
1286
1287When reading a message for filter testing, the returned value indicates
1288whether the headers (which is all that is read) were terminated by '.' or
1289not. */
1290
1291BOOL
1292receive_msg(BOOL extract_recip)
1293{
7156b1ef
NM
1294int i;
1295int rc = FAIL;
059ec3d9
PH
1296int msg_size = 0;
1297int process_info_len = Ustrlen(process_info);
1298int error_rc = (error_handling == ERRORS_SENDER)?
1299 errors_sender_rc : EXIT_FAILURE;
1300int header_size = 256;
1301int start, end, domain, size, sptr;
1302int id_resolution;
1303int had_zero = 0;
d677b2f2 1304int prevlines_length = 0;
059ec3d9
PH
1305
1306register int ptr = 0;
1307
1308BOOL contains_resent_headers = FALSE;
1309BOOL extracted_ignored = FALSE;
1310BOOL first_line_ended_crlf = TRUE_UNSET;
1311BOOL smtp_yield = TRUE;
1312BOOL yield = FALSE;
1313
1314BOOL resents_exist = FALSE;
1315uschar *resent_prefix = US"";
1316uschar *blackholed_by = NULL;
04f7d5b9 1317uschar *blackhole_log_msg = US"";
059ec3d9
PH
1318
1319flock_t lock_data;
1320error_block *bad_addresses = NULL;
1321
1322uschar *frozen_by = NULL;
1323uschar *queued_by = NULL;
1324
1325uschar *errmsg, *s;
1326struct stat statbuf;
1327
4e88a19f 1328/* Final message to give to SMTP caller, and messages from ACLs */
059ec3d9
PH
1329
1330uschar *smtp_reply = NULL;
4e88a19f 1331uschar *user_msg, *log_msg;
059ec3d9
PH
1332
1333/* Working header pointers */
1334
1335header_line *h, *next;
1336
2cbb4081 1337/* Flags for noting the existence of certain headers (only one left) */
059ec3d9
PH
1338
1339BOOL date_header_exists = FALSE;
1340
1341/* Pointers to receive the addresses of headers whose contents we need. */
1342
1343header_line *from_header = NULL;
1344header_line *subject_header = NULL;
1345header_line *msgid_header = NULL;
1346header_line *received_header;
1347
1348/* Variables for use when building the Received: header. */
1349
1350uschar *received;
1351uschar *timestamp;
1352int tslen;
1353
1354/* Release any open files that might have been cached while preparing to
1355accept the message - e.g. by verifying addresses - because reading a message
1356might take a fair bit of real time. */
1357
1358search_tidyup();
1359
1360/* Initialize the chain of headers by setting up a place-holder for Received:
1361header. Temporarily mark it as "old", i.e. not to be used. We keep header_last
1362pointing to the end of the chain to make adding headers simple. */
1363
1364received_header = header_list = header_last = store_get(sizeof(header_line));
1365header_list->next = NULL;
1366header_list->type = htype_old;
1367header_list->text = NULL;
1368header_list->slen = 0;
1369
1370/* Control block for the next header to be read. */
1371
1372next = store_get(sizeof(header_line));
1373next->text = store_get(header_size);
1374
1375/* Initialize message id to be null (indicating no message read), and the
1376header names list to be the normal list. Indicate there is no data file open
1377yet, initialize the size and warning count, and deal with no size limit. */
1378
1379message_id[0] = 0;
1380data_file = NULL;
1381data_fd = -1;
1382spool_name[0] = 0;
1383message_size = 0;
1384warning_count = 0;
d677b2f2 1385received_count = 1; /* For the one we will add */
059ec3d9
PH
1386
1387if (thismessage_size_limit <= 0) thismessage_size_limit = INT_MAX;
1388
2e0c1448 1389/* While reading the message, the following counts are computed. */
059ec3d9 1390
d677b2f2
PH
1391message_linecount = body_linecount = body_zerocount =
1392 max_received_linelength = 0;
059ec3d9 1393
80a47a2c
TK
1394#ifndef DISABLE_DKIM
1395/* Call into DKIM to set up the context. */
1396if (smtp_input && !smtp_batched_input && !dkim_disable_verify) dkim_exim_verify_init();
fb2274d4
TK
1397#endif
1398
059ec3d9
PH
1399/* Remember the time of reception. Exim uses time+pid for uniqueness of message
1400ids, and fractions of a second are required. See the comments that precede the
1401message id creation below. */
1402
1403(void)gettimeofday(&message_id_tv, NULL);
1404
1405/* For other uses of the received time we can operate with granularity of one
1406second, and for that we use the global variable received_time. This is for
1407things like ultimate message timeouts. */
1408
1409received_time = message_id_tv.tv_sec;
1410
1411/* If SMTP input, set the special handler for timeouts. The alarm() calls
1412happen in the smtp_getc() function when it refills its buffer. */
1413
1414if (smtp_input) os_non_restarting_signal(SIGALRM, data_timeout_handler);
1415
1416/* If not SMTP input, timeout happens only if configured, and we just set a
1417single timeout for the whole message. */
1418
1419else if (receive_timeout > 0)
1420 {
1421 os_non_restarting_signal(SIGALRM, data_timeout_handler);
1422 alarm(receive_timeout);
1423 }
1424
1425/* SIGTERM and SIGINT are caught always. */
1426
1427signal(SIGTERM, data_sigterm_sigint_handler);
1428signal(SIGINT, data_sigterm_sigint_handler);
1429
1430/* Header lines in messages are not supposed to be very long, though when
1431unfolded, to: and cc: headers can take up a lot of store. We must also cope
1432with the possibility of junk being thrown at us. Start by getting 256 bytes for
1433storing the header, and extend this as necessary using string_cat().
1434
1435To cope with total lunacies, impose an upper limit on the length of the header
1436section of the message, as otherwise the store will fill up. We must also cope
1437with the possibility of binary zeros in the data. Hence we cannot use fgets().
1438Folded header lines are joined into one string, leaving the '\n' characters
1439inside them, so that writing them out reproduces the input.
1440
1441Loop for each character of each header; the next structure for chaining the
1442header is set up already, with ptr the offset of the next character in
1443next->text. */
1444
1445for (;;)
1446 {
80a47a2c 1447 int ch = (receive_getc)();
059ec3d9
PH
1448
1449 /* If we hit EOF on a SMTP connection, it's an error, since incoming
1450 SMTP must have a correct "." terminator. */
1451
1452 if (ch == EOF && smtp_input /* && !smtp_batched_input */)
1453 {
1454 smtp_reply = handle_lost_connection(US" (header)");
1455 smtp_yield = FALSE;
1456 goto TIDYUP; /* Skip to end of function */
1457 }
1458
1459 /* See if we are at the current header's size limit - there must be at least
1460 four bytes left. This allows for the new character plus a zero, plus two for
1461 extra insertions when we are playing games with dots and carriage returns. If
1462 we are at the limit, extend the text buffer. This could have been done
1463 automatically using string_cat() but because this is a tightish loop storing
1464 only one character at a time, we choose to do it inline. Normally
1465 store_extend() will be able to extend the block; only at the end of a big
1466 store block will a copy be needed. To handle the case of very long headers
1467 (and sometimes lunatic messages can have ones that are 100s of K long) we
1468 call store_release() for strings that have been copied - if the string is at
1469 the start of a block (and therefore the only thing in it, because we aren't
1470 doing any other gets), the block gets freed. We can only do this because we
1471 know there are no other calls to store_get() going on. */
1472
1473 if (ptr >= header_size - 4)
1474 {
1475 int oldsize = header_size;
1476 /* header_size += 256; */
1477 header_size *= 2;
1478 if (!store_extend(next->text, oldsize, header_size))
1479 {
1480 uschar *newtext = store_get(header_size);
1481 memcpy(newtext, next->text, ptr);
1482 store_release(next->text);
1483 next->text = newtext;
1484 }
1485 }
1486
1487 /* Cope with receiving a binary zero. There is dispute about whether
1488 these should be allowed in RFC 822 messages. The middle view is that they
1489 should not be allowed in headers, at least. Exim takes this attitude at
1490 the moment. We can't just stomp on them here, because we don't know that
1491 this line is a header yet. Set a flag to cause scanning later. */
1492
1493 if (ch == 0) had_zero++;
1494
1495 /* Test for termination. Lines in remote SMTP are terminated by CRLF, while
1496 those from data files use just LF. Treat LF in local SMTP input as a
1497 terminator too. Treat EOF as a line terminator always. */
1498
1499 if (ch == EOF) goto EOL;
1500
1501 /* FUDGE: There are sites out there that don't send CRs before their LFs, and
1502 other MTAs accept this. We are therefore forced into this "liberalisation"
1503 too, so we accept LF as a line terminator whatever the source of the message.
1504 However, if the first line of the message ended with a CRLF, we treat a bare
1505 LF specially by inserting a white space after it to ensure that the header
1506 line is not terminated. */
1507
1508 if (ch == '\n')
1509 {
1510 if (first_line_ended_crlf == TRUE_UNSET) first_line_ended_crlf = FALSE;
80a47a2c 1511 else if (first_line_ended_crlf) receive_ungetc(' ');
059ec3d9
PH
1512 goto EOL;
1513 }
1514
1515 /* This is not the end of the line. If this is SMTP input and this is
1516 the first character in the line and it is a "." character, ignore it.
1517 This implements the dot-doubling rule, though header lines starting with
1518 dots aren't exactly common. They are legal in RFC 822, though. If the
1519 following is CRLF or LF, this is the line that that terminates the
1520 entire message. We set message_ended to indicate this has happened (to
1521 prevent further reading), and break out of the loop, having freed the
1522 empty header, and set next = NULL to indicate no data line. */
1523
1524 if (ptr == 0 && ch == '.' && (smtp_input || dot_ends))
1525 {
80a47a2c 1526 ch = (receive_getc)();
059ec3d9
PH
1527 if (ch == '\r')
1528 {
80a47a2c 1529 ch = (receive_getc)();
059ec3d9
PH
1530 if (ch != '\n')
1531 {
80a47a2c 1532 receive_ungetc(ch);
059ec3d9
PH
1533 ch = '\r'; /* Revert to CR */
1534 }
1535 }
1536 if (ch == '\n')
1537 {
1538 message_ended = END_DOT;
1539 store_reset(next);
1540 next = NULL;
1541 break; /* End character-reading loop */
1542 }
1543
1544 /* For non-SMTP input, the dot at the start of the line was really a data
1545 character. What is now in ch is the following character. We guaranteed
1546 enough space for this above. */
1547
1548 if (!smtp_input)
1549 {
1550 next->text[ptr++] = '.';
1551 message_size++;
1552 }
1553 }
1554
1555 /* If CR is immediately followed by LF, end the line, ignoring the CR, and
1556 remember this case if this is the first line ending. */
1557
1558 if (ch == '\r')
1559 {
80a47a2c 1560 ch = (receive_getc)();
059ec3d9
PH
1561 if (ch == '\n')
1562 {
1563 if (first_line_ended_crlf == TRUE_UNSET) first_line_ended_crlf = TRUE;
1564 goto EOL;
1565 }
1566
1567 /* Otherwise, put back the character after CR, and turn the bare CR
1568 into LF SP. */
1569
80a47a2c 1570 ch = (receive_ungetc)(ch);
059ec3d9
PH
1571 next->text[ptr++] = '\n';
1572 message_size++;
1573 ch = ' ';
1574 }
1575
1576 /* We have a data character for the header line. */
1577
1578 next->text[ptr++] = ch; /* Add to buffer */
1579 message_size++; /* Total message size so far */
1580
1581 /* Handle failure due to a humungously long header section. The >= allows
1582 for the terminating \n. Add what we have so far onto the headers list so
1583 that it gets reflected in any error message, and back up the just-read
1584 character. */
1585
1586 if (message_size >= header_maxsize)
1587 {
1588 next->text[ptr] = 0;
1589 next->slen = ptr;
1590 next->type = htype_other;
1591 next->next = NULL;
1592 header_last->next = next;
1593 header_last = next;
1594
1595 log_write(0, LOG_MAIN, "ridiculously long message header received from "
1596 "%s (more than %d characters): message abandoned",
1597 sender_host_unknown? sender_ident : sender_fullhost, header_maxsize);
1598
1599 if (smtp_input)
1600 {
1601 smtp_reply = US"552 Message header is ridiculously long";
1602 receive_swallow_smtp();
1603 goto TIDYUP; /* Skip to end of function */
1604 }
1605
1606 else
1607 {
1608 give_local_error(ERRMESS_VLONGHEADER,
1609 string_sprintf("message header longer than %d characters received: "
1610 "message not accepted", header_maxsize), US"", error_rc, stdin,
1611 header_list->next);
1612 /* Does not return */
1613 }
1614 }
1615
1616 continue; /* With next input character */
1617
1618 /* End of header line reached */
1619
1620 EOL:
2e0c1448
PH
1621
1622 /* Keep track of lines for BSMTP errors and overall message_linecount. */
1623
1624 receive_linecount++;
1625 message_linecount++;
059ec3d9 1626
d677b2f2
PH
1627 /* Keep track of maximum line length */
1628
1629 if (ptr - prevlines_length > max_received_linelength)
1630 max_received_linelength = ptr - prevlines_length;
1631 prevlines_length = ptr + 1;
1632
059ec3d9
PH
1633 /* Now put in the terminating newline. There is always space for
1634 at least two more characters. */
1635
1636 next->text[ptr++] = '\n';
1637 message_size++;
1638
1639 /* A blank line signals the end of the headers; release the unwanted
1640 space and set next to NULL to indicate this. */
1641
1642 if (ptr == 1)
1643 {
1644 store_reset(next);
1645 next = NULL;
1646 break;
1647 }
1648
1649 /* There is data in the line; see if the next input character is a
1650 whitespace character. If it is, we have a continuation of this header line.
1651 There is always space for at least one character at this point. */
1652
1653 if (ch != EOF)
1654 {
80a47a2c 1655 int nextch = (receive_getc)();
059ec3d9
PH
1656 if (nextch == ' ' || nextch == '\t')
1657 {
1658 next->text[ptr++] = nextch;
1659 message_size++;
1660 continue; /* Iterate the loop */
1661 }
80a47a2c 1662 else if (nextch != EOF) (receive_ungetc)(nextch); /* For next time */
059ec3d9
PH
1663 else ch = EOF; /* Cause main loop to exit at end */
1664 }
1665
1666 /* We have got to the real line end. Terminate the string and release store
1667 beyond it. If it turns out to be a real header, internal binary zeros will
1668 be squashed later. */
1669
1670 next->text[ptr] = 0;
1671 next->slen = ptr;
1672 store_reset(next->text + ptr + 1);
1673
1674 /* Check the running total size against the overall message size limit. We
1675 don't expect to fail here, but if the overall limit is set less than MESSAGE_
1676 MAXSIZE and a big header is sent, we want to catch it. Just stop reading
1677 headers - the code to read the body will then also hit the buffer. */
1678
1679 if (message_size > thismessage_size_limit) break;
1680
1681 /* A line that is not syntactically correct for a header also marks
1682 the end of the headers. In this case, we leave next containing the
1683 first data line. This might actually be several lines because of the
1684 continuation logic applied above, but that doesn't matter.
1685
1686 It turns out that smail, and presumably sendmail, accept leading lines
1687 of the form
1688
1689 From ph10 Fri Jan 5 12:35 GMT 1996
1690
1691 in messages. The "mail" command on Solaris 2 sends such lines. I cannot
1692 find any documentation of this, but for compatibility it had better be
1693 accepted. Exim restricts it to the case of non-smtp messages, and
1694 treats it as an alternative to the -f command line option. Thus it is
1695 ignored except for trusted users or filter testing. Otherwise it is taken
1696 as the sender address, unless -f was used (sendmail compatibility).
1697
1698 It further turns out that some UUCPs generate the From_line in a different
1699 format, e.g.
1700
1701 From ph10 Fri, 7 Jan 97 14:00:00 GMT
1702
1703 The regex for matching these things is now capable of recognizing both
1704 formats (including 2- and 4-digit years in the latter). In fact, the regex
1705 is now configurable, as is the expansion string to fish out the sender.
1706
1707 Even further on it has been discovered that some broken clients send
1708 these lines in SMTP messages. There is now an option to ignore them from
1709 specified hosts or networks. Sigh. */
1710
1711 if (header_last == header_list &&
1712 (!smtp_input
1713 ||
1714 (sender_host_address != NULL &&
1715 verify_check_host(&ignore_fromline_hosts) == OK)
1716 ||
1717 (sender_host_address == NULL && ignore_fromline_local)
1718 ) &&
1719 regex_match_and_setup(regex_From, next->text, 0, -1))
1720 {
1721 if (!sender_address_forced)
1722 {
1723 uschar *uucp_sender = expand_string(uucp_from_sender);
1724 if (uucp_sender == NULL)
1725 {
1726 log_write(0, LOG_MAIN|LOG_PANIC,
1727 "expansion of \"%s\" failed after matching "
1728 "\"From \" line: %s", uucp_from_sender, expand_string_message);
1729 }
1730 else
1731 {
1732 int start, end, domain;
1733 uschar *errmess;
1734 uschar *newsender = parse_extract_address(uucp_sender, &errmess,
1735 &start, &end, &domain, TRUE);
1736 if (newsender != NULL)
1737 {
1738 if (domain == 0 && newsender[0] != 0)
1739 newsender = rewrite_address_qualify(newsender, FALSE);
1740
f05da2e8 1741 if (filter_test != FTEST_NONE || receive_check_set_sender(newsender))
059ec3d9
PH
1742 {
1743 sender_address = newsender;
1744
f05da2e8 1745 if (trusted_caller || filter_test != FTEST_NONE)
059ec3d9
PH
1746 {
1747 authenticated_sender = NULL;
1748 originator_name = US"";
1749 sender_local = FALSE;
1750 }
1751
f05da2e8 1752 if (filter_test != FTEST_NONE)
059ec3d9
PH
1753 printf("Sender taken from \"From \" line\n");
1754 }
1755 }
1756 }
1757 }
1758 }
1759
1760 /* Not a leading "From " line. Check to see if it is a valid header line.
1761 Header names may contain any non-control characters except space and colon,
1762 amazingly. */
1763
1764 else
1765 {
1766 uschar *p = next->text;
1767
1768 /* If not a valid header line, break from the header reading loop, leaving
1769 next != NULL, indicating that it holds the first line of the body. */
1770
1771 if (isspace(*p)) break;
1772 while (mac_isgraph(*p) && *p != ':') p++;
1773 while (isspace(*p)) p++;
1774 if (*p != ':')
1775 {
1776 body_zerocount = had_zero;
1777 break;
1778 }
1779
1780 /* We have a valid header line. If there were any binary zeroes in
1781 the line, stomp on them here. */
1782
1783 if (had_zero > 0)
1784 for (p = next->text; p < next->text + ptr; p++) if (*p == 0) *p = '?';
1785
1786 /* It is perfectly legal to have an empty continuation line
1787 at the end of a header, but it is confusing to humans
1788 looking at such messages, since it looks like a blank line.
1789 Reduce confusion by removing redundant white space at the
1790 end. We know that there is at least one printing character
1791 (the ':' tested for above) so there is no danger of running
1792 off the end. */
1793
1794 p = next->text + ptr - 2;
1795 for (;;)
1796 {
1797 while (*p == ' ' || *p == '\t') p--;
1798 if (*p != '\n') break;
1799 ptr = (p--) - next->text + 1;
1800 message_size -= next->slen - ptr;
1801 next->text[ptr] = 0;
1802 next->slen = ptr;
1803 }
1804
1805 /* Add the header to the chain */
1806
1807 next->type = htype_other;
1808 next->next = NULL;
1809 header_last->next = next;
1810 header_last = next;
1811
1812 /* Check the limit for individual line lengths. This comes after adding to
1813 the chain so that the failing line is reflected if a bounce is generated
1814 (for a local message). */
1815
1816 if (header_line_maxsize > 0 && next->slen > header_line_maxsize)
1817 {
1818 log_write(0, LOG_MAIN, "overlong message header line received from "
1819 "%s (more than %d characters): message abandoned",
1820 sender_host_unknown? sender_ident : sender_fullhost,
1821 header_line_maxsize);
1822
1823 if (smtp_input)
1824 {
1825 smtp_reply = US"552 A message header line is too long";
1826 receive_swallow_smtp();
1827 goto TIDYUP; /* Skip to end of function */
1828 }
1829
1830 else
1831 {
1832 give_local_error(ERRMESS_VLONGHDRLINE,
1833 string_sprintf("message header line longer than %d characters "
1834 "received: message not accepted", header_line_maxsize), US"",
1835 error_rc, stdin, header_list->next);
1836 /* Does not return */
1837 }
1838 }
1839
1840 /* Note if any resent- fields exist. */
1841
1842 if (!resents_exist && strncmpic(next->text, US"resent-", 7) == 0)
1843 {
1844 resents_exist = TRUE;
1845 resent_prefix = US"Resent-";
1846 }
1847 }
1848
1849 /* The line has been handled. If we have hit EOF, break out of the loop,
1850 indicating no pending data line. */
1851
1852 if (ch == EOF) { next = NULL; break; }
1853
1854 /* Set up for the next header */
1855
1856 header_size = 256;
1857 next = store_get(sizeof(header_line));
1858 next->text = store_get(header_size);
1859 ptr = 0;
1860 had_zero = 0;
d677b2f2 1861 prevlines_length = 0;
059ec3d9
PH
1862 } /* Continue, starting to read the next header */
1863
1864/* At this point, we have read all the headers into a data structure in main
1865store. The first header is still the dummy placeholder for the Received: header
1866we are going to generate a bit later on. If next != NULL, it contains the first
1867data line - which terminated the headers before reaching a blank line (not the
1868normal case). */
1869
1870DEBUG(D_receive)
1871 {
1872 debug_printf(">>Headers received:\n");
1873 for (h = header_list->next; h != NULL; h = h->next)
1874 debug_printf("%s", h->text);
1875 debug_printf("\n");
1876 }
1877
1878/* End of file on any SMTP connection is an error. If an incoming SMTP call
1879is dropped immediately after valid headers, the next thing we will see is EOF.
1880We must test for this specially, as further down the reading of the data is
1881skipped if already at EOF. */
1882
1883if (smtp_input && (receive_feof)())
1884 {
1885 smtp_reply = handle_lost_connection(US" (after header)");
1886 smtp_yield = FALSE;
1887 goto TIDYUP; /* Skip to end of function */
1888 }
1889
1890/* If this is a filter test run and no headers were read, output a warning
1891in case there is a mistake in the test message. */
1892
f05da2e8 1893if (filter_test != FTEST_NONE && header_list->next == NULL)
059ec3d9
PH
1894 printf("Warning: no message headers read\n");
1895
1896
1897/* Scan the headers to identify them. Some are merely marked for later
1898processing; some are dealt with here. */
1899
1900for (h = header_list->next; h != NULL; h = h->next)
1901 {
1902 BOOL is_resent = strncmpic(h->text, US"resent-", 7) == 0;
1903 if (is_resent) contains_resent_headers = TRUE;
1904
1905 switch (header_checkname(h, is_resent))
1906 {
059ec3d9 1907 case htype_bcc:
2cbb4081 1908 h->type = htype_bcc; /* Both Bcc: and Resent-Bcc: */
059ec3d9
PH
1909 break;
1910
059ec3d9 1911 case htype_cc:
2cbb4081 1912 h->type = htype_cc; /* Both Cc: and Resent-Cc: */
059ec3d9
PH
1913 break;
1914
1915 /* Record whether a Date: or Resent-Date: header exists, as appropriate. */
1916
1917 case htype_date:
4c69d561 1918 if (!resents_exist || is_resent) date_header_exists = TRUE;
059ec3d9
PH
1919 break;
1920
1921 /* Same comments as about Return-Path: below. */
1922
1923 case htype_delivery_date:
1924 if (delivery_date_remove) h->type = htype_old;
1925 break;
1926
1927 /* Same comments as about Return-Path: below. */
1928
1929 case htype_envelope_to:
1930 if (envelope_to_remove) h->type = htype_old;
1931 break;
1932
1933 /* Mark all "From:" headers so they get rewritten. Save the one that is to
1934 be used for Sender: checking. For Sendmail compatibility, if the "From:"
1935 header consists of just the login id of the user who called Exim, rewrite
1936 it with the gecos field first. Apply this rule to Resent-From: if there
1937 are resent- fields. */
1938
1939 case htype_from:
1940 h->type = htype_from;
1941 if (!resents_exist || is_resent)
1942 {
1943 from_header = h;
1944 if (!smtp_input)
1945 {
1946 uschar *s = Ustrchr(h->text, ':') + 1;
1947 while (isspace(*s)) s++;
1948 if (strncmpic(s, originator_login, h->slen - (s - h->text) - 1) == 0)
1949 {
1950 uschar *name = is_resent? US"Resent-From" : US"From";
1951 header_add(htype_from, "%s: %s <%s@%s>\n", name, originator_name,
1952 originator_login, qualify_domain_sender);
1953 from_header = header_last;
1954 h->type = htype_old;
1955 DEBUG(D_receive|D_rewrite)
1956 debug_printf("rewrote \"%s:\" header using gecos\n", name);
1957 }
1958 }
1959 }
1960 break;
1961
1962 /* Identify the Message-id: header for generating "in-reply-to" in the
1963 autoreply transport. For incoming logging, save any resent- value. In both
1964 cases, take just the first of any multiples. */
1965
1966 case htype_id:
1967 if (msgid_header == NULL && (!resents_exist || is_resent))
1968 {
1969 msgid_header = h;
1970 h->type = htype_id;
1971 }
1972 break;
1973
1974 /* Flag all Received: headers */
1975
1976 case htype_received:
1977 h->type = htype_received;
1978 received_count++;
1979 break;
1980
1981 /* "Reply-to:" is just noted (there is no resent-reply-to field) */
1982
1983 case htype_reply_to:
1984 h->type = htype_reply_to;
1985 break;
1986
1987 /* The Return-path: header is supposed to be added to messages when
1988 they leave the SMTP system. We shouldn't receive messages that already
1989 contain Return-path. However, since Exim generates Return-path: on
1990 local delivery, resent messages may well contain it. We therefore
1991 provide an option (which defaults on) to remove any Return-path: headers
1992 on input. Removal actually means flagging as "old", which prevents the
1993 header being transmitted with the message. */
1994
1995 case htype_return_path:
1996 if (return_path_remove) h->type = htype_old;
1997
1998 /* If we are testing a mail filter file, use the value of the
1999 Return-Path: header to set up the return_path variable, which is not
2000 otherwise set. However, remove any <> that surround the address
2001 because the variable doesn't have these. */
2002
f05da2e8 2003 if (filter_test != FTEST_NONE)
059ec3d9
PH
2004 {
2005 uschar *start = h->text + 12;
2006 uschar *end = start + Ustrlen(start);
2007 while (isspace(*start)) start++;
2008 while (end > start && isspace(end[-1])) end--;
2009 if (*start == '<' && end[-1] == '>')
2010 {
2011 start++;
2012 end--;
2013 }
2014 return_path = string_copyn(start, end - start);
2015 printf("Return-path taken from \"Return-path:\" header line\n");
2016 }
2017 break;
2018
2019 /* If there is a "Sender:" header and the message is locally originated,
8800895a
PH
2020 and from an untrusted caller and suppress_local_fixups is not set, or if we
2021 are in submission mode for a remote message, mark it "old" so that it will
2022 not be transmitted with the message, unless active_local_sender_retain is
2023 set. (This can only be true if active_local_from_check is false.) If there
2024 are any resent- headers in the message, apply this rule to Resent-Sender:
2025 instead of Sender:. Messages with multiple resent- header sets cannot be
2026 tidily handled. (For this reason, at least one MUA - Pine - turns old
2027 resent- headers into X-resent- headers when resending, leaving just one
2028 set.) */
059ec3d9
PH
2029
2030 case htype_sender:
69358f02 2031 h->type = ((!active_local_sender_retain &&
8800895a
PH
2032 (
2033 (sender_local && !trusted_caller && !suppress_local_fixups)
2034 || submission_mode
2035 )
059ec3d9
PH
2036 ) &&
2037 (!resents_exist||is_resent))?
2038 htype_old : htype_sender;
2039 break;
2040
2041 /* Remember the Subject: header for logging. There is no Resent-Subject */
2042
2043 case htype_subject:
2044 subject_header = h;
2045 break;
2046
2047 /* "To:" gets flagged, and the existence of a recipient header is noted,
2048 whether it's resent- or not. */
2049
2050 case htype_to:
2051 h->type = htype_to;
2052 /****
2053 to_or_cc_header_exists = TRUE;
2054 ****/
2055 break;
2056 }
2057 }
2058
2059/* Extract recipients from the headers if that is required (the -t option).
2060Note that this is documented as being done *before* any address rewriting takes
2061place. There are two possibilities:
2062
2063(1) According to sendmail documentation for Solaris, IRIX, and HP-UX, any
2064recipients already listed are to be REMOVED from the message. Smail 3 works
2065like this. We need to build a non-recipients tree for that list, because in
2066subsequent processing this data is held in a tree and that's what the
2067spool_write_header() function expects. Make sure that non-recipient addresses
2068are fully qualified and rewritten if necessary.
2069
2070(2) According to other sendmail documentation, -t ADDS extracted recipients to
2071those in the command line arguments (and it is rumoured some other MTAs do
2072this). Therefore, there is an option to make Exim behave this way.
2073
2074*** Notes on "Resent-" header lines ***
2075
2076The presence of resent-headers in the message makes -t horribly ambiguous.
2077Experiments with sendmail showed that it uses recipients for all resent-
2078headers, totally ignoring the concept of "sets of resent- headers" as described
2079in RFC 2822 section 3.6.6. Sendmail also amalgamates them into a single set
2080with all the addresses in one instance of each header.
2081
2082This seems to me not to be at all sensible. Before release 4.20, Exim 4 gave an
2083error for -t if there were resent- headers in the message. However, after a
2084discussion on the mailing list, I've learned that there are MUAs that use
2085resent- headers with -t, and also that the stuff about sets of resent- headers
2086and their ordering in RFC 2822 is generally ignored. An MUA that submits a
2087message with -t and resent- header lines makes sure that only *its* resent-
2088headers are present; previous ones are often renamed as X-resent- for example.
2089
2090Consequently, Exim has been changed so that, if any resent- header lines are
2091present, the recipients are taken from all of the appropriate resent- lines,
2092and not from the ordinary To:, Cc:, etc. */
2093
2094if (extract_recip)
2095 {
2096 int rcount = 0;
2097 error_block **bnext = &bad_addresses;
2098
2099 if (extract_addresses_remove_arguments)
2100 {
2101 while (recipients_count-- > 0)
2102 {
2103 uschar *s = rewrite_address(recipients_list[recipients_count].address,
2104 TRUE, TRUE, global_rewrite_rules, rewrite_existflags);
2105 tree_add_nonrecipient(s);
2106 }
2107 recipients_list = NULL;
2108 recipients_count = recipients_list_max = 0;
2109 }
2110
059ec3d9
PH
2111 /* Now scan the headers */
2112
2113 for (h = header_list->next; h != NULL; h = h->next)
2114 {
2115 if ((h->type == htype_to || h->type == htype_cc || h->type == htype_bcc) &&
2116 (!contains_resent_headers || strncmpic(h->text, US"resent-", 7) == 0))
2117 {
2118 uschar *s = Ustrchr(h->text, ':') + 1;
2119 while (isspace(*s)) s++;
2120
1eccaa59
PH
2121 parse_allow_group = TRUE; /* Allow address group syntax */
2122
059ec3d9
PH
2123 while (*s != 0)
2124 {
2125 uschar *ss = parse_find_address_end(s, FALSE);
2126 uschar *recipient, *errmess, *p, *pp;
2127 int start, end, domain;
2128
2129 /* Check on maximum */
2130
2131 if (recipients_max > 0 && ++rcount > recipients_max)
2132 {
2133 give_local_error(ERRMESS_TOOMANYRECIP, US"too many recipients",
2134 US"message rejected: ", error_rc, stdin, NULL);
2135 /* Does not return */
2136 }
2137
2138 /* Make a copy of the address, and remove any internal newlines. These
2139 may be present as a result of continuations of the header line. The
2140 white space that follows the newline must not be removed - it is part
2141 of the header. */
2142
2143 pp = recipient = store_get(ss - s + 1);
2144 for (p = s; p < ss; p++) if (*p != '\n') *pp++ = *p;
2145 *pp = 0;
2146 recipient = parse_extract_address(recipient, &errmess, &start, &end,
2147 &domain, FALSE);
2148
2149 /* Keep a list of all the bad addresses so we can send a single
2150 error message at the end. However, an empty address is not an error;
2151 just ignore it. This can come from an empty group list like
2152
2153 To: Recipients of list:;
2154
2155 If there are no recipients at all, an error will occur later. */
2156
2157 if (recipient == NULL && Ustrcmp(errmess, "empty address") != 0)
2158 {
2159 int len = Ustrlen(s);
2160 error_block *b = store_get(sizeof(error_block));
2161 while (len > 0 && isspace(s[len-1])) len--;
2162 b->next = NULL;
2163 b->text1 = string_printing(string_copyn(s, len));
2164 b->text2 = errmess;
2165 *bnext = b;
2166 bnext = &(b->next);
2167 }
2168
2169 /* If the recipient is already in the nonrecipients tree, it must
2170 have appeared on the command line with the option extract_addresses_
2171 remove_arguments set. Do not add it to the recipients, and keep a note
2172 that this has happened, in order to give a better error if there are
2173 no recipients left. */
2174
2175 else if (recipient != NULL)
2176 {
2177 if (tree_search(tree_nonrecipients, recipient) == NULL)
2178 receive_add_recipient(recipient, -1);
2179 else
2180 extracted_ignored = TRUE;
2181 }
2182
2183 /* Move on past this address */
2184
2185 s = ss + (*ss? 1:0);
2186 while (isspace(*s)) s++;
1eccaa59
PH
2187 } /* Next address */
2188
2189 parse_allow_group = FALSE; /* Reset group syntax flags */
2190 parse_found_group = FALSE;
059ec3d9
PH
2191
2192 /* If this was the bcc: header, mark it "old", which means it
2193 will be kept on the spool, but not transmitted as part of the
2194 message. */
2195
2cbb4081 2196 if (h->type == htype_bcc) h->type = htype_old;
059ec3d9
PH
2197 } /* For appropriate header line */
2198 } /* For each header line */
2199
059ec3d9
PH
2200 }
2201
2202/* Now build the unique message id. This has changed several times over the
2203lifetime of Exim. This description was rewritten for Exim 4.14 (February 2003).
2204Retaining all the history in the comment has become too unwieldy - read
2205previous release sources if you want it.
2206
2207The message ID has 3 parts: tttttt-pppppp-ss. Each part is a number in base 62.
2208The first part is the current time, in seconds. The second part is the current
2209pid. Both are large enough to hold 32-bit numbers in base 62. The third part
2210can hold a number in the range 0-3843. It used to be a computed sequence
2211number, but is now the fractional component of the current time in units of
22121/2000 of a second (i.e. a value in the range 0-1999). After a message has been
2213received, Exim ensures that the timer has ticked at the appropriate level
2214before proceeding, to avoid duplication if the pid happened to be re-used
2215within the same time period. It seems likely that most messages will take at
2216least half a millisecond to be received, so no delay will normally be
2217necessary. At least for some time...
2218
2219There is a modification when localhost_number is set. Formerly this was allowed
2220to be as large as 255. Now it is restricted to the range 0-16, and the final
2221component of the message id becomes (localhost_number * 200) + fractional time
2222in units of 1/200 of a second (i.e. a value in the range 0-3399).
2223
2224Some not-really-Unix operating systems use case-insensitive file names (Darwin,
2225Cygwin). For these, we have to use base 36 instead of base 62. Luckily, this
2226still allows the tttttt field to hold a large enough number to last for some
2227more decades, and the final two-digit field can hold numbers up to 1295, which
2228is enough for milliseconds (instead of 1/2000 of a second).
2229
2230However, the pppppp field cannot hold a 32-bit pid, but it can hold a 31-bit
2231pid, so it is probably safe because pids have to be positive. The
2232localhost_number is restricted to 0-10 for these hosts, and when it is set, the
2233final field becomes (localhost_number * 100) + fractional time in centiseconds.
2234
2235Note that string_base62() returns its data in a static storage block, so it
2236must be copied before calling string_base62() again. It always returns exactly
22376 characters.
2238
2239There doesn't seem to be anything in the RFC which requires a message id to
2240start with a letter, but Smail was changed to ensure this. The external form of
2241the message id (as supplied by string expansion) therefore starts with an
2242additional leading 'E'. The spool file names do not include this leading
2243letter and it is not used internally.
2244
2245NOTE: If ever the format of message ids is changed, the regular expression for
2246checking that a string is in this format must be updated in a corresponding
2247way. It appears in the initializing code in exim.c. The macro MESSAGE_ID_LENGTH
2248must also be changed to reflect the correct string length. Then, of course,
2249other programs that rely on the message id format will need updating too. */
2250
2251Ustrncpy(message_id, string_base62((long int)(message_id_tv.tv_sec)), 6);
2252message_id[6] = '-';
2253Ustrncpy(message_id + 7, string_base62((long int)getpid()), 6);
2254
2255/* Deal with the case where the host number is set. The value of the number was
2256checked when it was read, to ensure it isn't too big. The timing granularity is
2257left in id_resolution so that an appropriate wait can be done after receiving
2258the message, if necessary (we hope it won't be). */
2259
2260if (host_number_string != NULL)
2261 {
2262 id_resolution = (BASE_62 == 62)? 5000 : 10000;
2263 sprintf(CS(message_id + MESSAGE_ID_LENGTH - 3), "-%2s",
2264 string_base62((long int)(
2265 host_number * (1000000/id_resolution) +
2266 message_id_tv.tv_usec/id_resolution)) + 4);
2267 }
2268
2269/* Host number not set: final field is just the fractional time at an
2270appropriate resolution. */
2271
2272else
2273 {
2274 id_resolution = (BASE_62 == 62)? 500 : 1000;
2275 sprintf(CS(message_id + MESSAGE_ID_LENGTH - 3), "-%2s",
2276 string_base62((long int)(message_id_tv.tv_usec/id_resolution)) + 4);
2277 }
2278
2279/* Add the current message id onto the current process info string if
2280it will fit. */
2281
2282(void)string_format(process_info + process_info_len,
2283 PROCESS_INFO_SIZE - process_info_len, " id=%s", message_id);
2284
2285/* If we are using multiple input directories, set up the one for this message
2286to be the least significant base-62 digit of the time of arrival. Otherwise
2287ensure that it is an empty string. */
2288
2289message_subdir[0] = split_spool_directory? message_id[5] : 0;
2290
2291/* Now that we have the message-id, if there is no message-id: header, generate
8800895a
PH
2292one, but only for local (without suppress_local_fixups) or submission mode
2293messages. This can be user-configured if required, but we had better flatten
2294any illegal characters therein. */
059ec3d9 2295
8800895a
PH
2296if (msgid_header == NULL &&
2297 ((sender_host_address == NULL && !suppress_local_fixups)
2298 || submission_mode))
059ec3d9
PH
2299 {
2300 uschar *p;
2301 uschar *id_text = US"";
2302 uschar *id_domain = primary_hostname;
2303
2304 /* Permit only letters, digits, dots, and hyphens in the domain */
2305
2306 if (message_id_domain != NULL)
2307 {
2308 uschar *new_id_domain = expand_string(message_id_domain);
2309 if (new_id_domain == NULL)
2310 {
2311 if (!expand_string_forcedfail)
2312 log_write(0, LOG_MAIN|LOG_PANIC,
2313 "expansion of \"%s\" (message_id_header_domain) "
2314 "failed: %s", message_id_domain, expand_string_message);
2315 }
2316 else if (*new_id_domain != 0)
2317 {
2318 id_domain = new_id_domain;
2319 for (p = id_domain; *p != 0; p++)
2320 if (!isalnum(*p) && *p != '.') *p = '-'; /* No need to test '-' ! */
2321 }
2322 }
2323
2324 /* Permit all characters except controls and RFC 2822 specials in the
2325 additional text part. */
2326
2327 if (message_id_text != NULL)
2328 {
2329 uschar *new_id_text = expand_string(message_id_text);
2330 if (new_id_text == NULL)
2331 {
2332 if (!expand_string_forcedfail)
2333 log_write(0, LOG_MAIN|LOG_PANIC,
2334 "expansion of \"%s\" (message_id_header_text) "
2335 "failed: %s", message_id_text, expand_string_message);
2336 }
2337 else if (*new_id_text != 0)
2338 {
2339 id_text = new_id_text;
2340 for (p = id_text; *p != 0; p++)
2341 if (mac_iscntrl_or_special(*p)) *p = '-';
2342 }
2343 }
2344
e7e680d6
PP
2345 /* Add the header line
2346 * Resent-* headers are prepended, per RFC 5322 3.6.6. Non-Resent-* are
2347 * appended, to preserve classical expectations of header ordering. */
059ec3d9 2348
e7e680d6 2349 header_add_at_position(!resents_exist, NULL, FALSE, htype_id,
5eb690a1
NM
2350 "%sMessage-Id: <%s%s%s@%s>\n", resent_prefix, message_id_external,
2351 (*id_text == 0)? "" : ".", id_text, id_domain);
059ec3d9
PH
2352 }
2353
2354/* If we are to log recipients, keep a copy of the raw ones before any possible
2355rewriting. Must copy the count, because later ACLs and the local_scan()
2356function may mess with the real recipients. */
2357
2358if ((log_extra_selector & LX_received_recipients) != 0)
2359 {
2360 raw_recipients = store_get(recipients_count * sizeof(uschar *));
2361 for (i = 0; i < recipients_count; i++)
2362 raw_recipients[i] = string_copy(recipients_list[i].address);
2363 raw_recipients_count = recipients_count;
2364 }
2365
2366/* Ensure the recipients list is fully qualified and rewritten. Unqualified
2367recipients will get here only if the conditions were right (allow_unqualified_
2368recipient is TRUE). */
2369
2370for (i = 0; i < recipients_count; i++)
2371 recipients_list[i].address =
2372 rewrite_address(recipients_list[i].address, TRUE, TRUE,
2373 global_rewrite_rules, rewrite_existflags);
2374
8800895a
PH
2375/* If there is no From: header, generate one for local (without
2376suppress_local_fixups) or submission_mode messages. If there is no sender
2377address, but the sender is local or this is a local delivery error, use the
2378originator login. This shouldn't happen for genuine bounces, but might happen
2379for autoreplies. The addition of From: must be done *before* checking for the
2380possible addition of a Sender: header, because untrusted_set_sender allows an
2381untrusted user to set anything in the envelope (which might then get info
2382From:) but we still want to ensure a valid Sender: if it is required. */
2383
2384if (from_header == NULL &&
2385 ((sender_host_address == NULL && !suppress_local_fixups)
2386 || submission_mode))
059ec3d9 2387 {
2fe1a124
PH
2388 uschar *oname = US"";
2389
2390 /* Use the originator_name if this is a locally submitted message and the
2391 caller is not trusted. For trusted callers, use it only if -F was used to
2392 force its value or if we have a non-SMTP message for which -f was not used
2393 to set the sender. */
2394
2395 if (sender_host_address == NULL)
2396 {
2397 if (!trusted_caller || sender_name_forced ||
2398 (!smtp_input && !sender_address_forced))
2399 oname = originator_name;
2400 }
2401
2402 /* For non-locally submitted messages, the only time we use the originator
2403 name is when it was forced by the /name= option on control=submission. */
2404
2405 else
2406 {
2407 if (submission_name != NULL) oname = submission_name;
2408 }
2409
059ec3d9
PH
2410 /* Envelope sender is empty */
2411
2412 if (sender_address[0] == 0)
2413 {
87ba3f5f
PH
2414 uschar *fromstart, *fromend;
2415
2416 fromstart = string_sprintf("%sFrom: %s%s", resent_prefix,
2fe1a124
PH
2417 oname, (oname[0] == 0)? "" : " <");
2418 fromend = (oname[0] == 0)? US"" : US">";
87ba3f5f 2419
059ec3d9
PH
2420 if (sender_local || local_error_message)
2421 {
87ba3f5f
PH
2422 header_add(htype_from, "%s%s@%s%s\n", fromstart,
2423 local_part_quote(originator_login), qualify_domain_sender,
2424 fromend);
059ec3d9
PH
2425 }
2426 else if (submission_mode && authenticated_id != NULL)
2427 {
2428 if (submission_domain == NULL)
2429 {
87ba3f5f
PH
2430 header_add(htype_from, "%s%s@%s%s\n", fromstart,
2431 local_part_quote(authenticated_id), qualify_domain_sender,
2432 fromend);
059ec3d9
PH
2433 }
2434 else if (submission_domain[0] == 0) /* empty => whole address set */
2435 {
87ba3f5f
PH
2436 header_add(htype_from, "%s%s%s\n", fromstart, authenticated_id,
2437 fromend);
059ec3d9
PH
2438 }
2439 else
2440 {
87ba3f5f
PH
2441 header_add(htype_from, "%s%s@%s%s\n", fromstart,
2442 local_part_quote(authenticated_id), submission_domain,
2443 fromend);
059ec3d9
PH
2444 }
2445 from_header = header_last; /* To get it checked for Sender: */
2446 }
2447 }
2448
2449 /* There is a non-null envelope sender. Build the header using the original
2450 sender address, before any rewriting that might have been done while
2451 verifying it. */
2452
2453 else
2454 {
87ba3f5f 2455 header_add(htype_from, "%sFrom: %s%s%s%s\n", resent_prefix,
2fe1a124
PH
2456 oname,
2457 (oname[0] == 0)? "" : " <",
87ba3f5f
PH
2458 (sender_address_unrewritten == NULL)?
2459 sender_address : sender_address_unrewritten,
2fe1a124 2460 (oname[0] == 0)? "" : ">");
059ec3d9
PH
2461
2462 from_header = header_last; /* To get it checked for Sender: */
2463 }
2464 }
2465
2466
8800895a
PH
2467/* If the sender is local (without suppress_local_fixups), or if we are in
2468submission mode and there is an authenticated_id, check that an existing From:
2469is correct, and if not, generate a Sender: header, unless disabled. Any
2470previously-existing Sender: header was removed above. Note that sender_local,
2471as well as being TRUE if the caller of exim is not trusted, is also true if a
2472trusted caller did not supply a -f argument for non-smtp input. To allow
2473trusted callers to forge From: without supplying -f, we have to test explicitly
2474here. If the From: header contains more than one address, then the call to
2475parse_extract_address fails, and a Sender: header is inserted, as required. */
059ec3d9
PH
2476
2477if (from_header != NULL &&
69358f02 2478 (active_local_from_check &&
8800895a 2479 ((sender_local && !trusted_caller && !suppress_local_fixups) ||
69358f02 2480 (submission_mode && authenticated_id != NULL))
059ec3d9
PH
2481 ))
2482 {
2483 BOOL make_sender = TRUE;
2484 int start, end, domain;
2485 uschar *errmess;
2486 uschar *from_address =
2487 parse_extract_address(Ustrchr(from_header->text, ':') + 1, &errmess,
2488 &start, &end, &domain, FALSE);
2489 uschar *generated_sender_address;
2490
2491 if (submission_mode)
2492 {
2493 if (submission_domain == NULL)
2494 {
2495 generated_sender_address = string_sprintf("%s@%s",
2496 local_part_quote(authenticated_id), qualify_domain_sender);
2497 }
2498 else if (submission_domain[0] == 0) /* empty => full address */
2499 {
2500 generated_sender_address = string_sprintf("%s",
2501 authenticated_id);
2502 }
2503 else
2504 {
2505 generated_sender_address = string_sprintf("%s@%s",
2506 local_part_quote(authenticated_id), submission_domain);
2507 }
2508 }
2509 else
2510 generated_sender_address = string_sprintf("%s@%s",
2511 local_part_quote(originator_login), qualify_domain_sender);
2512
2513 /* Remove permitted prefixes and suffixes from the local part of the From:
2514 address before doing the comparison with the generated sender. */
2515
2516 if (from_address != NULL)
2517 {
2518 int slen;
2519 uschar *at = (domain == 0)? NULL : from_address + domain - 1;
2520
2521 if (at != NULL) *at = 0;
2522 from_address += route_check_prefix(from_address, local_from_prefix);
2523 slen = route_check_suffix(from_address, local_from_suffix);
2524 if (slen > 0)
2525 {
2526 memmove(from_address+slen, from_address, Ustrlen(from_address)-slen);
2527 from_address += slen;
2528 }
2529 if (at != NULL) *at = '@';
2530
2531 if (strcmpic(generated_sender_address, from_address) == 0 ||
2532 (domain == 0 && strcmpic(from_address, originator_login) == 0))
2533 make_sender = FALSE;
2534 }
2535
2536 /* We have to cause the Sender header to be rewritten if there are
2537 appropriate rewriting rules. */
2538
2539 if (make_sender)
2540 {
2fe1a124 2541 if (submission_mode && submission_name == NULL)
059ec3d9
PH
2542 header_add(htype_sender, "%sSender: %s\n", resent_prefix,
2543 generated_sender_address);
2544 else
2545 header_add(htype_sender, "%sSender: %s <%s>\n",
2fe1a124
PH
2546 resent_prefix,
2547 submission_mode? submission_name : originator_name,
2548 generated_sender_address);
059ec3d9 2549 }
87ba3f5f
PH
2550
2551 /* Ensure that a non-null envelope sender address corresponds to the
2552 submission mode sender address. */
2553
2554 if (submission_mode && sender_address[0] != 0)
2555 {
2556 if (sender_address_unrewritten == NULL)
2557 sender_address_unrewritten = sender_address;
2558 sender_address = generated_sender_address;
089793a4
TF
2559 if (Ustrcmp(sender_address_unrewritten, generated_sender_address) != 0)
2560 log_write(L_address_rewrite, LOG_MAIN,
2561 "\"%s\" from env-from rewritten as \"%s\" by submission mode",
2562 sender_address_unrewritten, generated_sender_address);
87ba3f5f 2563 }
059ec3d9
PH
2564 }
2565
2566
2567/* If there are any rewriting rules, apply them to the sender address, unless
2568it has already been rewritten as part of verification for SMTP input. */
2569
2570if (global_rewrite_rules != NULL && sender_address_unrewritten == NULL &&
2571 sender_address[0] != 0)
2572 {
2573 sender_address = rewrite_address(sender_address, FALSE, TRUE,
2574 global_rewrite_rules, rewrite_existflags);
2575 DEBUG(D_receive|D_rewrite)
2576 debug_printf("rewritten sender = %s\n", sender_address);
2577 }
2578
2579
2580/* The headers must be run through rewrite_header(), because it ensures that
2581addresses are fully qualified, as well as applying any rewriting rules that may
2582exist.
2583
2584Qualification of header addresses in a message from a remote host happens only
2585if the host is in sender_unqualified_hosts or recipient_unqualified hosts, as
2586appropriate. For local messages, qualification always happens, unless -bnq is
2587used to explicitly suppress it. No rewriting is done for an unqualified address
2588that is left untouched.
2589
2590We start at the second header, skipping our own Received:. This rewriting is
2591documented as happening *after* recipient addresses are taken from the headers
2592by the -t command line option. An added Sender: gets rewritten here. */
2593
2594for (h = header_list->next; h != NULL; h = h->next)
2595 {
2596 header_line *newh = rewrite_header(h, NULL, NULL, global_rewrite_rules,
2597 rewrite_existflags, TRUE);
2598 if (newh != NULL) h = newh;
2599 }
2600
2601
2602/* An RFC 822 (sic) message is not legal unless it has at least one of "to",
2cbb4081 2603"cc", or "bcc". Note that although the minimal examples in RFC 822 show just
059ec3d9
PH
2604"to" or "bcc", the full syntax spec allows "cc" as well. If any resent- header
2605exists, this applies to the set of resent- headers rather than the normal set.
2606
2cbb4081
PH
2607The requirement for a recipient header has been removed in RFC 2822. At this
2608point in the code, earlier versions of Exim added a To: header for locally
2609submitted messages, and an empty Bcc: header for others. In the light of the
2610changes in RFC 2822, this was dropped in November 2003. */
059ec3d9 2611
059ec3d9
PH
2612
2613/* If there is no date header, generate one if the message originates locally
8800895a
PH
2614(i.e. not over TCP/IP) and suppress_local_fixups is not set, or if the
2615submission mode flag is set. Messages without Date: are not valid, but it seems
e7e680d6
PP
2616to be more confusing if Exim adds one to all remotely-originated messages.
2617As per Message-Id, we prepend if resending, else append.
2618*/
059ec3d9 2619
8800895a
PH
2620if (!date_header_exists &&
2621 ((sender_host_address == NULL && !suppress_local_fixups)
2622 || submission_mode))
e7e680d6
PP
2623 header_add_at_position(!resents_exist, NULL, FALSE, htype_other,
2624 "%sDate: %s\n", resent_prefix, tod_stamp(tod_full));
059ec3d9
PH
2625
2626search_tidyup(); /* Free any cached resources */
2627
2628/* Show the complete set of headers if debugging. Note that the first one (the
2629new Received:) has not yet been set. */
2630
2631DEBUG(D_receive)
2632 {
2633 debug_printf(">>Headers after rewriting and local additions:\n");
2634 for (h = header_list->next; h != NULL; h = h->next)
2635 debug_printf("%c %s", h->type, h->text);
2636 debug_printf("\n");
2637 }
2638
2639/* The headers are now complete in store. If we are running in filter
2640testing mode, that is all this function does. Return TRUE if the message
2641ended with a dot. */
2642
f05da2e8 2643if (filter_test != FTEST_NONE)
059ec3d9
PH
2644 {
2645 process_info[process_info_len] = 0;
2646 return message_ended == END_DOT;
2647 }
2648
2649/* Open a new spool file for the data portion of the message. We need
2650to access it both via a file descriptor and a stream. Try to make the
2651directory if it isn't there. Note re use of sprintf: spool_directory
2652is checked on input to be < 200 characters long. */
2653
2654sprintf(CS spool_name, "%s/input/%s/%s-D", spool_directory, message_subdir,
2655 message_id);
2656data_fd = Uopen(spool_name, O_RDWR|O_CREAT|O_EXCL, SPOOL_MODE);
2657if (data_fd < 0)
2658 {
2659 if (errno == ENOENT)
2660 {
2661 uschar temp[16];
2662 sprintf(CS temp, "input/%s", message_subdir);
2663 if (message_subdir[0] == 0) temp[5] = 0;
2664 (void)directory_make(spool_directory, temp, INPUT_DIRECTORY_MODE, TRUE);
2665 data_fd = Uopen(spool_name, O_RDWR|O_CREAT|O_EXCL, SPOOL_MODE);
2666 }
2667 if (data_fd < 0)
2668 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Failed to create spool file %s: %s",
2669 spool_name, strerror(errno));
2670 }
2671
2672/* Make sure the file's group is the Exim gid, and double-check the mode
2673because the group setting doesn't always get set automatically. */
2674
ff790e47
PH
2675(void)fchown(data_fd, exim_uid, exim_gid);
2676(void)fchmod(data_fd, SPOOL_MODE);
059ec3d9
PH
2677
2678/* We now have data file open. Build a stream for it and lock it. We lock only
2679the first line of the file (containing the message ID) because otherwise there
2680are problems when Exim is run under Cygwin (I'm told). See comments in
2681spool_in.c, where the same locking is done. */
2682
2683data_file = fdopen(data_fd, "w+");
2684lock_data.l_type = F_WRLCK;
2685lock_data.l_whence = SEEK_SET;
2686lock_data.l_start = 0;
2687lock_data.l_len = SPOOL_DATA_START_OFFSET;
2688
2689if (fcntl(data_fd, F_SETLK, &lock_data) < 0)
2690 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Cannot lock %s (%d): %s", spool_name,
2691 errno, strerror(errno));
2692
2693/* We have an open, locked data file. Write the message id to it to make it
2694self-identifying. Then read the remainder of the input of this message and
2695write it to the data file. If the variable next != NULL, it contains the first
2696data line (which was read as a header but then turned out not to have the right
2697format); write it (remembering that it might contain binary zeros). The result
2698of fwrite() isn't inspected; instead we call ferror() below. */
2699
2700fprintf(data_file, "%s-D\n", message_id);
2701if (next != NULL)
2702 {
2703 uschar *s = next->text;
2704 int len = next->slen;
f1e894f3 2705 (void)fwrite(s, 1, len, data_file);
059ec3d9
PH
2706 body_linecount++; /* Assumes only 1 line */
2707 }
2708
2709/* Note that we might already be at end of file, or the logical end of file
2710(indicated by '.'), or might have encountered an error while writing the
2711message id or "next" line. */
2712
2713if (!ferror(data_file) && !(receive_feof)() && message_ended != END_DOT)
2714 {
2715 if (smtp_input)
2716 {
2717 message_ended = read_message_data_smtp(data_file);
2718 receive_linecount++; /* The terminating "." line */
2719 }
2720 else message_ended = read_message_data(data_file);
2721
2722 receive_linecount += body_linecount; /* For BSMTP errors mainly */
2e0c1448 2723 message_linecount += body_linecount;
059ec3d9
PH
2724
2725 /* Handle premature termination of SMTP */
2726
2727 if (smtp_input && message_ended == END_EOF)
2728 {
2729 Uunlink(spool_name); /* Lose data file when closed */
2730 message_id[0] = 0; /* Indicate no message accepted */
2731 smtp_reply = handle_lost_connection(US"");
2732 smtp_yield = FALSE;
2733 goto TIDYUP; /* Skip to end of function */
2734 }
2735
2736 /* Handle message that is too big. Don't use host_or_ident() in the log
2737 message; we want to see the ident value even for non-remote messages. */
2738
2739 if (message_ended == END_SIZE)
2740 {
2741 Uunlink(spool_name); /* Lose the data file when closed */
2742 if (smtp_input) receive_swallow_smtp(); /* Swallow incoming SMTP */
2743
2744 log_write(L_size_reject, LOG_MAIN|LOG_REJECT, "rejected from <%s>%s%s%s%s: "
2745 "message too big: read=%d max=%d",
2746 sender_address,
2747 (sender_fullhost == NULL)? "" : " H=",
2748 (sender_fullhost == NULL)? US"" : sender_fullhost,
2749 (sender_ident == NULL)? "" : " U=",
2750 (sender_ident == NULL)? US"" : sender_ident,
2751 message_size,
2752 thismessage_size_limit);
2753
2754 if (smtp_input)
2755 {
2756 smtp_reply = US"552 Message size exceeds maximum permitted";
2757 message_id[0] = 0; /* Indicate no message accepted */
2758 goto TIDYUP; /* Skip to end of function */
2759 }
2760 else
2761 {
2762 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
2763 give_local_error(ERRMESS_TOOBIG,
2764 string_sprintf("message too big (max=%d)", thismessage_size_limit),
2765 US"message rejected: ", error_rc, data_file, header_list);
2766 /* Does not return */
2767 }
2768 }
2769 }
2770
2771/* Restore the standard SIGALRM handler for any subsequent processing. (For
2772example, there may be some expansion in an ACL that uses a timer.) */
2773
2774os_non_restarting_signal(SIGALRM, sigalrm_handler);
2775
2776/* The message body has now been read into the data file. Call fflush() to
2777empty the buffers in C, and then call fsync() to get the data written out onto
2778the disk, as fflush() doesn't do this (or at least, it isn't documented as
2779having to do this). If there was an I/O error on either input or output,
2780attempt to send an error message, and unlink the spool file. For non-SMTP input
2781we can then give up. Note that for SMTP input we must swallow the remainder of
2782the input in cases of output errors, since the far end doesn't expect to see
2783anything until the terminating dot line is sent. */
2784
2785if (fflush(data_file) == EOF || ferror(data_file) ||
54fc8428 2786 EXIMfsync(fileno(data_file)) < 0 || (receive_ferror)())
059ec3d9
PH
2787 {
2788 uschar *msg_errno = US strerror(errno);
2789 BOOL input_error = (receive_ferror)() != 0;
2790 uschar *msg = string_sprintf("%s error (%s) while receiving message from %s",
2791 input_error? "Input read" : "Spool write",
2792 msg_errno,
2793 (sender_fullhost != NULL)? sender_fullhost : sender_ident);
2794
2795 log_write(0, LOG_MAIN, "Message abandoned: %s", msg);
2796 Uunlink(spool_name); /* Lose the data file */
2797
2798 if (smtp_input)
2799 {
2800 if (input_error)
2801 smtp_reply = US"451 Error while reading input data";
2802 else
2803 {
2804 smtp_reply = US"451 Error while writing spool file";
2805 receive_swallow_smtp();
2806 }
2807 message_id[0] = 0; /* Indicate no message accepted */
2808 goto TIDYUP; /* Skip to end of function */
2809 }
2810
2811 else
2812 {
2813 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
2814 give_local_error(ERRMESS_IOERR, msg, US"", error_rc, data_file,
2815 header_list);
2816 /* Does not return */
2817 }
2818 }
2819
2820
2821/* No I/O errors were encountered while writing the data file. */
2822
2823DEBUG(D_receive) debug_printf("Data file written for message %s\n", message_id);
2824
2825
2826/* If there were any bad addresses extracted by -t, or there were no recipients
2827left after -t, send a message to the sender of this message, or write it to
2828stderr if the error handling option is set that way. Note that there may
2829legitimately be no recipients for an SMTP message if they have all been removed
2830by "discard".
2831
2832We need to rewind the data file in order to read it. In the case of no
2833recipients or stderr error writing, throw the data file away afterwards, and
2834exit. (This can't be SMTP, which always ensures there's at least one
2835syntactically good recipient address.) */
2836
2837if (extract_recip && (bad_addresses != NULL || recipients_count == 0))
2838 {
2839 DEBUG(D_receive)
2840 {
2841 if (recipients_count == 0) debug_printf("*** No recipients\n");
2842 if (bad_addresses != NULL)
2843 {
2844 error_block *eblock = bad_addresses;
2845 debug_printf("*** Bad address(es)\n");
2846 while (eblock != NULL)
2847 {
2848 debug_printf(" %s: %s\n", eblock->text1, eblock->text2);
2849 eblock = eblock->next;
2850 }
2851 }
2852 }
2853
2854 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
2855
2856 /* If configured to send errors to the sender, but this fails, force
2857 a failure error code. We use a special one for no recipients so that it
2858 can be detected by the autoreply transport. Otherwise error_rc is set to
2859 errors_sender_rc, which is EXIT_FAILURE unless -oee was given, in which case
2860 it is EXIT_SUCCESS. */
2861
2862 if (error_handling == ERRORS_SENDER)
2863 {
2864 if (!moan_to_sender(
2865 (bad_addresses == NULL)?
2866 (extracted_ignored? ERRMESS_IGADDRESS : ERRMESS_NOADDRESS) :
2867 (recipients_list == NULL)? ERRMESS_BADNOADDRESS : ERRMESS_BADADDRESS,
2868 bad_addresses, header_list, data_file, FALSE))
2869 error_rc = (bad_addresses == NULL)? EXIT_NORECIPIENTS : EXIT_FAILURE;
2870 }
2871 else
2872 {
2873 if (bad_addresses == NULL)
2874 {
2875 if (extracted_ignored)
2876 fprintf(stderr, "exim: all -t recipients overridden by command line\n");
2877 else
2878 fprintf(stderr, "exim: no recipients in message\n");
2879 }
2880 else
2881 {
2882 fprintf(stderr, "exim: invalid address%s",
2883 (bad_addresses->next == NULL)? ":" : "es:\n");
2884 while (bad_addresses != NULL)
2885 {
2886 fprintf(stderr, " %s: %s\n", bad_addresses->text1,
2887 bad_addresses->text2);
2888 bad_addresses = bad_addresses->next;
2889 }
2890 }
2891 }
2892
2893 if (recipients_count == 0 || error_handling == ERRORS_STDERR)
2894 {
2895 Uunlink(spool_name);
f1e894f3 2896 (void)fclose(data_file);
059ec3d9
PH
2897 exim_exit(error_rc);
2898 }
2899 }
2900
2901/* Data file successfully written. Generate text for the Received: header by
2902expanding the configured string, and adding a timestamp. By leaving this
2903operation till now, we ensure that the timestamp is the time that message
2904reception was completed. However, this is deliberately done before calling the
2905data ACL and local_scan().
2906
2907This Received: header may therefore be inspected by the data ACL and by code in
2908the local_scan() function. When they have run, we update the timestamp to be
2909the final time of reception.
2910
2911If there is just one recipient, set up its value in the $received_for variable
2912for use when we generate the Received: header.
2913
2914Note: the checking for too many Received: headers is handled by the delivery
2915code. */
2916
2917timestamp = expand_string(US"${tod_full}");
2918if (recipients_count == 1) received_for = recipients_list[0].address;
2919received = expand_string(received_header_text);
2920received_for = NULL;
2921
2922if (received == NULL)
2923 {
2924 Uunlink(spool_name); /* Lose the data file */
2925 log_write(0, LOG_MAIN|LOG_PANIC_DIE, "Expansion of \"%s\" "
2926 "(received_header_text) failed: %s", string_printing(received_header_text),
2927 expand_string_message);
2928 }
2929
2930/* The first element on the header chain is reserved for the Received header,
2931so all we have to do is fill in the text pointer, and set the type. However, if
2932the result of the expansion is an empty string, we leave the header marked as
2933"old" so as to refrain from adding a Received header. */
2934
2935if (received[0] == 0)
2936 {
2937 received_header->text = string_sprintf("Received: ; %s\n", timestamp);
2938 received_header->type = htype_old;
2939 }
2940else
2941 {
2942 received_header->text = string_sprintf("%s; %s\n", received, timestamp);
2943 received_header->type = htype_received;
2944 }
2945
2946received_header->slen = Ustrlen(received_header->text);
2947
2948DEBUG(D_receive) debug_printf(">>Generated Received: header line\n%c %s",
2949 received_header->type, received_header->text);
2950
2951/* Set the value of message_body_size for the DATA ACL and for local_scan() */
2952
2953message_body_size = (fstat(data_fd, &statbuf) == 0)?
2954 statbuf.st_size - SPOOL_DATA_START_OFFSET : -1;
2955
2956/* If an ACL from any RCPT commands set up any warning headers to add, do so
2957now, before running the DATA ACL. */
2958
2959add_acl_headers(US"MAIL or RCPT");
2960
2961/* If an ACL is specified for checking things at this stage of reception of a
2962message, run it, unless all the recipients were removed by "discard" in earlier
2963ACLs. That is the only case in which recipients_count can be zero at this
2964stage. Set deliver_datafile to point to the data file so that $message_body and
2965$message_body_end can be extracted if needed. Allow $recipients in expansions.
2966*/
2967
2968deliver_datafile = data_fd;
4e88a19f 2969user_msg = NULL;
059ec3d9 2970
0e20aff9
MH
2971enable_dollar_recipients = TRUE;
2972
059ec3d9
PH
2973if (recipients_count == 0)
2974 {
2975 blackholed_by = recipients_discarded? US"MAIL ACL" : US"RCPT ACL";
2976 }
2977else
2978 {
059ec3d9
PH
2979 /* Handle interactive SMTP messages */
2980
2981 if (smtp_input && !smtp_batched_input)
2982 {
8523533c 2983
80a47a2c
TK
2984#ifndef DISABLE_DKIM
2985 if (!dkim_disable_verify)
2986 {
2987 /* Finish verification, this will log individual signature results to
2988 the mainlog */
2989 dkim_exim_verify_finish();
2990
2991 /* Check if we must run the DKIM ACL */
2992 if ((acl_smtp_dkim != NULL) &&
2993 (dkim_verify_signers != NULL) &&
2994 (dkim_verify_signers[0] != '\0'))
2995 {
2996 uschar *dkim_verify_signers_expanded =
2997 expand_string(dkim_verify_signers);
2998 if (dkim_verify_signers_expanded == NULL)
2999 {
3000 log_write(0, LOG_MAIN|LOG_PANIC,
3001 "expansion of dkim_verify_signers option failed: %s",
3002 expand_string_message);
3003 }
3004 else
3005 {
3006 int sep = 0;
3007 uschar *ptr = dkim_verify_signers_expanded;
3008 uschar *item = NULL;
9e5d6b55
TK
3009 uschar *seen_items = NULL;
3010 int seen_items_size = 0;
3011 int seen_items_offset = 0;
80a47a2c 3012 uschar itembuf[256];
9122af94
TK
3013 /* Default to OK when no items are present */
3014 rc = OK;
80a47a2c
TK
3015 while ((item = string_nextinlist(&ptr, &sep,
3016 itembuf,
3017 sizeof(itembuf))) != NULL)
3018 {
6119d1ea
TK
3019 /* Prevent running ACL for an empty item */
3020 if (!item || (item[0] == '\0')) continue;
9e5d6b55
TK
3021 /* Only run ACL once for each domain or identity, no matter how often it
3022 appears in the expanded list. */
6119d1ea
TK
3023 if (seen_items != NULL)
3024 {
4a73449b 3025 uschar *seen_items_list = seen_items;
9e5d6b55 3026 if (match_isinlist(item,
4a73449b 3027 &seen_items_list,0,NULL,NULL,MCL_STRING,TRUE,NULL) == OK)
6119d1ea
TK
3028 {
3029 DEBUG(D_receive)
3030 debug_printf("acl_smtp_dkim: skipping signer %s, already seen\n", item);
3031 continue;
3032 }
4a73449b 3033 seen_items = string_append(seen_items,&seen_items_size,&seen_items_offset,1,":");
6119d1ea
TK
3034 }
3035
4a73449b
TK
3036 seen_items = string_append(seen_items,&seen_items_size,&seen_items_offset,1,item);
3037 seen_items[seen_items_offset] = '\0';
6119d1ea
TK
3038
3039 DEBUG(D_receive)
3040 debug_printf("calling acl_smtp_dkim for dkim_cur_signer=%s\n", item);
3041
80a47a2c
TK
3042 dkim_exim_acl_setup(item);
3043 rc = acl_check(ACL_WHERE_DKIM, NULL, acl_smtp_dkim, &user_msg, &log_msg);
6119d1ea
TK
3044
3045 if (rc != OK)
3046 {
3047 DEBUG(D_receive)
3048 debug_printf("acl_smtp_dkim: acl_check returned %d on %s, skipping remaining items\n", rc, item);
3049 break;
3050 }
80a47a2c
TK
3051 }
3052 add_acl_headers(US"DKIM");
3053 if (rc == DISCARD)
3054 {
3055 recipients_count = 0;
3056 blackholed_by = US"DKIM ACL";
3057 if (log_msg != NULL)
3058 blackhole_log_msg = string_sprintf(": %s", log_msg);
3059 }
3060 else if (rc != OK)
3061 {
3062 Uunlink(spool_name);
3063 if (smtp_handle_acl_fail(ACL_WHERE_DKIM, rc, user_msg, log_msg) != 0)
3064 smtp_yield = FALSE; /* No more messsages after dropped connection */
3065 smtp_reply = US""; /* Indicate reply already sent */
3066 message_id[0] = 0; /* Indicate no message accepted */
3067 goto TIDYUP; /* Skip to end of function */
3068 }
3069 }
3070 }
3071 }
3072#endif /* DISABLE_DKIM */
fb2274d4 3073
8523533c 3074#ifdef WITH_CONTENT_SCAN
80a47a2c
TK
3075 if (recipients_count > 0 &&
3076 acl_smtp_mime != NULL &&
54cdb463
PH
3077 !run_mime_acl(acl_smtp_mime, &smtp_yield, &smtp_reply, &blackholed_by))
3078 goto TIDYUP;
8523533c
TK
3079#endif /* WITH_CONTENT_SCAN */
3080
54cdb463
PH
3081 /* Check the recipients count again, as the MIME ACL might have changed
3082 them. */
8523533c 3083
059ec3d9
PH
3084 if (acl_smtp_data != NULL && recipients_count > 0)
3085 {
059ec3d9
PH
3086 rc = acl_check(ACL_WHERE_DATA, NULL, acl_smtp_data, &user_msg, &log_msg);
3087 add_acl_headers(US"DATA");
3088 if (rc == DISCARD)
3089 {
3090 recipients_count = 0;
3091 blackholed_by = US"DATA ACL";
8e669ac1
PH
3092 if (log_msg != NULL)
3093 blackhole_log_msg = string_sprintf(": %s", log_msg);
059ec3d9
PH
3094 }
3095 else if (rc != OK)
3096 {
3097 Uunlink(spool_name);
8523533c
TK
3098#ifdef WITH_CONTENT_SCAN
3099 unspool_mbox();
3100#endif
6f0c431a
PP
3101#ifdef EXPERIMENTAL_DCC
3102 dcc_ok = 0;
3103#endif
059ec3d9
PH
3104 if (smtp_handle_acl_fail(ACL_WHERE_DATA, rc, user_msg, log_msg) != 0)
3105 smtp_yield = FALSE; /* No more messsages after dropped connection */
3106 smtp_reply = US""; /* Indicate reply already sent */
3107 message_id[0] = 0; /* Indicate no message accepted */
3108 goto TIDYUP; /* Skip to end of function */
3109 }
3110 }
3111 }
3112
3113 /* Handle non-SMTP and batch SMTP (i.e. non-interactive) messages. Note that
3114 we cannot take different actions for permanent and temporary rejections. */
3115
54cdb463 3116 else
059ec3d9 3117 {
54cdb463
PH
3118
3119#ifdef WITH_CONTENT_SCAN
3120 if (acl_not_smtp_mime != NULL &&
3121 !run_mime_acl(acl_not_smtp_mime, &smtp_yield, &smtp_reply,
3122 &blackholed_by))
3123 goto TIDYUP;
3124#endif /* WITH_CONTENT_SCAN */
3125
3126 if (acl_not_smtp != NULL)
059ec3d9 3127 {
54cdb463
PH
3128 uschar *user_msg, *log_msg;
3129 rc = acl_check(ACL_WHERE_NOTSMTP, NULL, acl_not_smtp, &user_msg, &log_msg);
3130 if (rc == DISCARD)
059ec3d9 3131 {
54cdb463
PH
3132 recipients_count = 0;
3133 blackholed_by = US"non-SMTP ACL";
3134 if (log_msg != NULL)
3135 blackhole_log_msg = string_sprintf(": %s", log_msg);
059ec3d9 3136 }
54cdb463 3137 else if (rc != OK)
059ec3d9 3138 {
54cdb463
PH
3139 Uunlink(spool_name);
3140#ifdef WITH_CONTENT_SCAN
3141 unspool_mbox();
3142#endif
6f0c431a
PP
3143#ifdef EXPERIMENTAL_DCC
3144 dcc_ok = 0;
3145#endif
6ea85e9a
PH
3146 /* The ACL can specify where rejections are to be logged, possibly
3147 nowhere. The default is main and reject logs. */
3148
3149 if (log_reject_target != 0)
3150 log_write(0, log_reject_target, "F=<%s> rejected by non-SMTP ACL: %s",
3151 sender_address, log_msg);
3152
54cdb463
PH
3153 if (user_msg == NULL) user_msg = US"local configuration problem";
3154 if (smtp_batched_input)
3155 {
3156 moan_smtp_batch(NULL, "%d %s", 550, user_msg);
3157 /* Does not return */
3158 }
3159 else
3160 {
3161 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3162 give_local_error(ERRMESS_LOCAL_ACL, user_msg,
3163 US"message rejected by non-SMTP ACL: ", error_rc, data_file,
3164 header_list);
3165 /* Does not return */
3166 }
059ec3d9 3167 }
54cdb463 3168 add_acl_headers(US"non-SMTP");
059ec3d9 3169 }
059ec3d9
PH
3170 }
3171
54cdb463
PH
3172 /* The applicable ACLs have been run */
3173
059ec3d9
PH
3174 if (deliver_freeze) frozen_by = US"ACL"; /* for later logging */
3175 if (queue_only_policy) queued_by = US"ACL";
059ec3d9
PH
3176 }
3177
8523533c
TK
3178#ifdef WITH_CONTENT_SCAN
3179unspool_mbox();
3180#endif
3181
6a8f9482
TK
3182#ifdef EXPERIMENTAL_DCC
3183dcc_ok = 0;
3184#endif
3185
3186
059ec3d9
PH
3187/* The final check on the message is to run the scan_local() function. The
3188version supplied with Exim always accepts, but this is a hook for sysadmins to
3189supply their own checking code. The local_scan() function is run even when all
3190the recipients have been discarded. */
3191
3192lseek(data_fd, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3193
3194/* Arrange to catch crashes in local_scan(), so that the -D file gets
3195deleted, and the incident gets logged. */
3196
3197os_non_restarting_signal(SIGSEGV, local_scan_crash_handler);
3198os_non_restarting_signal(SIGFPE, local_scan_crash_handler);
3199os_non_restarting_signal(SIGILL, local_scan_crash_handler);
3200os_non_restarting_signal(SIGBUS, local_scan_crash_handler);
3201
3202DEBUG(D_receive) debug_printf("calling local_scan(); timeout=%d\n",
3203 local_scan_timeout);
3204local_scan_data = NULL;
3205
3206os_non_restarting_signal(SIGALRM, local_scan_timeout_handler);
3207if (local_scan_timeout > 0) alarm(local_scan_timeout);
3208rc = local_scan(data_fd, &local_scan_data);
3209alarm(0);
3210os_non_restarting_signal(SIGALRM, sigalrm_handler);
3211
0e20aff9
MH
3212enable_dollar_recipients = FALSE;
3213
059ec3d9
PH
3214store_pool = POOL_MAIN; /* In case changed */
3215DEBUG(D_receive) debug_printf("local_scan() returned %d %s\n", rc,
3216 local_scan_data);
3217
3218os_non_restarting_signal(SIGSEGV, SIG_DFL);
3219os_non_restarting_signal(SIGFPE, SIG_DFL);
3220os_non_restarting_signal(SIGILL, SIG_DFL);
3221os_non_restarting_signal(SIGBUS, SIG_DFL);
3222
3223/* The length check is paranoia against some runaway code, and also because
3224(for a success return) lines in the spool file are read into big_buffer. */
3225
3226if (local_scan_data != NULL)
3227 {
3228 int len = Ustrlen(local_scan_data);
3229 if (len > LOCAL_SCAN_MAX_RETURN) len = LOCAL_SCAN_MAX_RETURN;
3230 local_scan_data = string_copyn(local_scan_data, len);
3231 }
3232
3233if (rc == LOCAL_SCAN_ACCEPT_FREEZE)
3234 {
58eb016e 3235 if (!deliver_freeze) /* ACL might have already frozen */
059ec3d9
PH
3236 {
3237 deliver_freeze = TRUE;
3238 deliver_frozen_at = time(NULL);
3239 frozen_by = US"local_scan()";
3240 }
3241 rc = LOCAL_SCAN_ACCEPT;
3242 }
3243else if (rc == LOCAL_SCAN_ACCEPT_QUEUE)
3244 {
3245 if (!queue_only_policy) /* ACL might have already queued */
3246 {
3247 queue_only_policy = TRUE;
3248 queued_by = US"local_scan()";
3249 }
3250 rc = LOCAL_SCAN_ACCEPT;
3251 }
3252
3253/* Message accepted: remove newlines in local_scan_data because otherwise
3254the spool file gets corrupted. Ensure that all recipients are qualified. */
3255
3256if (rc == LOCAL_SCAN_ACCEPT)
3257 {
3258 if (local_scan_data != NULL)
3259 {
3260 uschar *s;
3261 for (s = local_scan_data; *s != 0; s++) if (*s == '\n') *s = ' ';
3262 }
3263 for (i = 0; i < recipients_count; i++)
3264 {
3265 recipient_item *r = recipients_list + i;
3266 r->address = rewrite_address_qualify(r->address, TRUE);
3267 if (r->errors_to != NULL)
3268 r->errors_to = rewrite_address_qualify(r->errors_to, TRUE);
3269 }
3270 if (recipients_count == 0 && blackholed_by == NULL)
3271 blackholed_by = US"local_scan";
3272 }
3273
3274/* Message rejected: newlines permitted in local_scan_data to generate
3275multiline SMTP responses. */
3276
3277else
3278 {
3279 uschar *istemp = US"";
3280 uschar *s = NULL;
a5bd321b 3281 uschar *smtp_code;
059ec3d9
PH
3282 int size = 0;
3283 int sptr = 0;
059ec3d9
PH
3284
3285 errmsg = local_scan_data;
3286
3287 Uunlink(spool_name); /* Cancel this message */
3288 switch(rc)
3289 {
3290 default:
3291 log_write(0, LOG_MAIN, "invalid return %d from local_scan(). Temporary "
3292 "rejection given", rc);
3293 goto TEMPREJECT;
3294
3295 case LOCAL_SCAN_REJECT_NOLOGHDR:
3296 log_extra_selector &= ~LX_rejected_header;
3297 /* Fall through */
3298
3299 case LOCAL_SCAN_REJECT:
a5bd321b 3300 smtp_code = US"550";
059ec3d9
PH
3301 if (errmsg == NULL) errmsg = US"Administrative prohibition";
3302 break;
3303
3304 case LOCAL_SCAN_TEMPREJECT_NOLOGHDR:
3305 log_extra_selector &= ~LX_rejected_header;
3306 /* Fall through */
3307
3308 case LOCAL_SCAN_TEMPREJECT:
3309 TEMPREJECT:
a5bd321b 3310 smtp_code = US"451";
059ec3d9
PH
3311 if (errmsg == NULL) errmsg = US"Temporary local problem";
3312 istemp = US"temporarily ";
3313 break;
3314 }
3315
3316 s = string_append(s, &size, &sptr, 2, US"F=",
3317 (sender_address[0] == 0)? US"<>" : sender_address);
3318 s = add_host_info_for_log(s, &size, &sptr);
3319 s[sptr] = 0;
3320
3321 log_write(0, LOG_MAIN|LOG_REJECT, "%s %srejected by local_scan(): %.256s",
3322 s, istemp, string_printing(errmsg));
3323
3324 if (smtp_input)
3325 {
3326 if (!smtp_batched_input)
3327 {
a5bd321b 3328 smtp_respond(smtp_code, 3, TRUE, errmsg);
059ec3d9
PH
3329 message_id[0] = 0; /* Indicate no message accepted */
3330 smtp_reply = US""; /* Indicate reply already sent */
3331 goto TIDYUP; /* Skip to end of function */
3332 }
3333 else
3334 {
a5bd321b 3335 moan_smtp_batch(NULL, "%s %s", smtp_code, errmsg);
059ec3d9
PH
3336 /* Does not return */
3337 }
3338 }
3339 else
3340 {
3341 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3342 give_local_error(ERRMESS_LOCAL_SCAN, errmsg,
3343 US"message rejected by local scan code: ", error_rc, data_file,
3344 header_list);
3345 /* Does not return */
3346 }
3347 }
3348
3349/* Reset signal handlers to ignore signals that previously would have caused
3350the message to be abandoned. */
3351
3352signal(SIGTERM, SIG_IGN);
3353signal(SIGINT, SIG_IGN);
3354
3355/* Ensure the first time flag is set in the newly-received message. */
3356
3357deliver_firsttime = TRUE;
3358
8523533c
TK
3359#ifdef EXPERIMENTAL_BRIGHTMAIL
3360if (bmi_run == 1) {
3361 /* rewind data file */
3362 lseek(data_fd, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3363 bmi_verdicts = bmi_process_message(header_list, data_fd);
3364};
3365#endif
3366
059ec3d9
PH
3367/* Update the timstamp in our Received: header to account for any time taken by
3368an ACL or by local_scan(). The new time is the time that all reception
3369processing is complete. */
3370
3371timestamp = expand_string(US"${tod_full}");
3372tslen = Ustrlen(timestamp);
3373
3374memcpy(received_header->text + received_header->slen - tslen - 1,
3375 timestamp, tslen);
3376
3377/* In MUA wrapper mode, ignore queueing actions set by ACL or local_scan() */
3378
3379if (mua_wrapper)
3380 {
3381 deliver_freeze = FALSE;
3382 queue_only_policy = FALSE;
3383 }
3384
3385/* Keep the data file open until we have written the header file, in order to
3386hold onto the lock. In a -bh run, or if the message is to be blackholed, we
3387don't write the header file, and we unlink the data file. If writing the header
3388file fails, we have failed to accept this message. */
3389
3390if (host_checking || blackholed_by != NULL)
3391 {
3392 header_line *h;
3393 Uunlink(spool_name);
3394 msg_size = 0; /* Compute size for log line */
3395 for (h = header_list; h != NULL; h = h->next)
3396 if (h->type != '*') msg_size += h->slen;
3397 }
3398
3399/* Write the -H file */
3400
3401else
3402 {
3403 if ((msg_size = spool_write_header(message_id, SW_RECEIVING, &errmsg)) < 0)
3404 {
3405 log_write(0, LOG_MAIN, "Message abandoned: %s", errmsg);
3406 Uunlink(spool_name); /* Lose the data file */
3407
3408 if (smtp_input)
3409 {
3410 smtp_reply = US"451 Error in writing spool file";
3411 message_id[0] = 0; /* Indicate no message accepted */
3412 goto TIDYUP;
3413 }
3414 else
3415 {
3416 fseek(data_file, (long int)SPOOL_DATA_START_OFFSET, SEEK_SET);
3417 give_local_error(ERRMESS_IOERR, errmsg, US"", error_rc, data_file,
3418 header_list);
3419 /* Does not return */
3420 }
3421 }
3422 }
3423
3424
3425/* The message has now been successfully received. */
3426
3427receive_messagecount++;
3428
3429/* In SMTP sessions we may receive several in one connection. After each one,
3430we wait for the clock to tick at the level of message-id granularity. This is
3431so that the combination of time+pid is unique, even on systems where the pid
3432can be re-used within our time interval. We can't shorten the interval without
3433re-designing the message-id. See comments above where the message id is
3434created. This is Something For The Future. */
3435
3436message_id_tv.tv_usec = (message_id_tv.tv_usec/id_resolution) * id_resolution;
3437exim_wait_tick(&message_id_tv, id_resolution);
3438
3439/* Add data size to written header size. We do not count the initial file name
3440that is in the file, but we do add one extra for the notional blank line that
3441precedes the data. This total differs from message_size in that it include the
3442added Received: header and any other headers that got created locally. */
3443
3444fflush(data_file);
3445fstat(data_fd, &statbuf);
3446
3447msg_size += statbuf.st_size - SPOOL_DATA_START_OFFSET + 1;
3448
3449/* Generate a "message received" log entry. We do this by building up a dynamic
3450string as required. Since we commonly want to add two items at a time, use a
3451macro to simplify the coding. We log the arrival of a new message while the
3452file is still locked, just in case the machine is *really* fast, and delivers
3453it first! Include any message id that is in the message - since the syntax of a
3454message id is actually an addr-spec, we can use the parse routine to canonicize
3455it. */
3456
3457size = 256;
3458sptr = 0;
3459s = store_get(size);
3460
3461s = string_append(s, &size, &sptr, 2, US"<= ",
3462 (sender_address[0] == 0)? US"<>" : sender_address);
3463if (message_reference != NULL)
3464 s = string_append(s, &size, &sptr, 2, US" R=", message_reference);
3465
3466s = add_host_info_for_log(s, &size, &sptr);
3467
3468#ifdef SUPPORT_TLS
3469if ((log_extra_selector & LX_tls_cipher) != 0 && tls_cipher != NULL)
3470 s = string_append(s, &size, &sptr, 2, US" X=", tls_cipher);
3471if ((log_extra_selector & LX_tls_certificate_verified) != 0 &&
3472 tls_cipher != NULL)
3473 s = string_append(s, &size, &sptr, 2, US" CV=",
3474 tls_certificate_verified? "yes":"no");
3475if ((log_extra_selector & LX_tls_peerdn) != 0 && tls_peerdn != NULL)
48ed62d9
PH
3476 s = string_append(s, &size, &sptr, 3, US" DN=\"",
3477 string_printing(tls_peerdn), US"\"");
059ec3d9
PH
3478#endif
3479
3480if (sender_host_authenticated != NULL)
3481 {
3482 s = string_append(s, &size, &sptr, 2, US" A=", sender_host_authenticated);
3483 if (authenticated_id != NULL)
3484 s = string_append(s, &size, &sptr, 2, US":", authenticated_id);
3485 }
3486
3487sprintf(CS big_buffer, "%d", msg_size);
3488s = string_append(s, &size, &sptr, 2, US" S=", big_buffer);
3489
3490/* If an addr-spec in a message-id contains a quoted string, it can contain
3491any characters except " \ and CR and so in particular it can contain NL!
3492Therefore, make sure we use a printing-characters only version for the log.
3493Also, allow for domain literals in the message id. */
3494
3495if (msgid_header != NULL)
3496 {
3497 uschar *old_id;
3498 BOOL save_allow_domain_literals = allow_domain_literals;
3499 allow_domain_literals = TRUE;
3500 old_id = parse_extract_address(Ustrchr(msgid_header->text, ':') + 1,
3501 &errmsg, &start, &end, &domain, FALSE);
3502 allow_domain_literals = save_allow_domain_literals;
3503 if (old_id != NULL)
3504 s = string_append(s, &size, &sptr, 2, US" id=", string_printing(old_id));
3505 }
3506
3507/* If subject logging is turned on, create suitable printing-character
3508text. By expanding $h_subject: we make use of the MIME decoding. */
3509
3510if ((log_extra_selector & LX_subject) != 0 && subject_header != NULL)
3511 {
3512 int i;
3513 uschar *p = big_buffer;
3514 uschar *ss = expand_string(US"$h_subject:");
3515
3516 /* Backslash-quote any double quotes or backslashes so as to make a
3517 a C-like string, and turn any non-printers into escape sequences. */
3518
3519 *p++ = '\"';
3520 if (*ss != 0) for (i = 0; i < 100 && ss[i] != 0; i++)
3521 {
3522 if (ss[i] == '\"' || ss[i] == '\\') *p++ = '\\';
3523 *p++ = ss[i];
3524 }
3525 *p++ = '\"';
3526 *p = 0;
3527 s = string_append(s, &size, &sptr, 2, US" T=", string_printing(big_buffer));
3528 }
3529
3530/* Terminate the string: string_cat() and string_append() leave room, but do
3531not put the zero in. */
3532
3533s[sptr] = 0;
3534
059ec3d9
PH
3535/* Create a message log file if message logs are being used and this message is
3536not blackholed. Write the reception stuff to it. We used to leave message log
3537creation until the first delivery, but this has proved confusing for somep
3538people. */
3539
3540if (message_logs && blackholed_by == NULL)
3541 {
3542 int fd;
3543
3544 sprintf(CS spool_name, "%s/msglog/%s/%s", spool_directory, message_subdir,
3545 message_id);
3546 fd = Uopen(spool_name, O_WRONLY|O_APPEND|O_CREAT, SPOOL_MODE);
3547
3548 if (fd < 0 && errno == ENOENT)
3549 {
3550 uschar temp[16];
3551 sprintf(CS temp, "msglog/%s", message_subdir);
3552 if (message_subdir[0] == 0) temp[6] = 0;
3553 (void)directory_make(spool_directory, temp, MSGLOG_DIRECTORY_MODE, TRUE);
3554 fd = Uopen(spool_name, O_WRONLY|O_APPEND|O_CREAT, SPOOL_MODE);
3555 }
3556
3557 if (fd < 0)
3558 {
3559 log_write(0, LOG_MAIN|LOG_PANIC, "Couldn't open message log %s: %s",
3560 spool_name, strerror(errno));
3561 }
3562
3563 else
3564 {
3565 FILE *message_log = fdopen(fd, "a");
3566 if (message_log == NULL)
3567 {
3568 log_write(0, LOG_MAIN|LOG_PANIC, "Couldn't fdopen message log %s: %s",
3569 spool_name, strerror(errno));
f1e894f3 3570 (void)close(fd);
059ec3d9
PH
3571 }
3572 else
3573 {
3574 uschar *now = tod_stamp(tod_log);
3575 fprintf(message_log, "%s Received from %s\n", now, s+3);
3576 if (deliver_freeze) fprintf(message_log, "%s frozen by %s\n", now,
3577 frozen_by);
3578 if (queue_only_policy) fprintf(message_log,
3579 "%s no immediate delivery: queued by %s\n", now, queued_by);
f1e894f3 3580 (void)fclose(message_log);
059ec3d9
PH
3581 }
3582 }
3583 }
3584
58eb016e
PH
3585/* Everything has now been done for a successful message except logging its
3586arrival, and outputting an SMTP response. While writing to the log, set a flag
3587to cause a call to receive_bomb_out() if the log cannot be opened. */
3588
3589receive_call_bombout = TRUE;
3590
563b63fa
PH
3591/* Before sending an SMTP response in a TCP/IP session, we check to see if the
3592connection has gone away. This can only be done if there is no unconsumed input
3593waiting in the local input buffer. We can test for this by calling
3594receive_smtp_buffered(). RFC 2920 (pipelining) explicitly allows for additional
3595input to be sent following the final dot, so the presence of following input is
3596not an error.
58eb016e 3597
563b63fa
PH
3598If the connection is still present, but there is no unread input for the
3599socket, the result of a select() call will be zero. If, however, the connection
3600has gone away, or if there is pending input, the result of select() will be
3601non-zero. The two cases can be distinguished by trying to read the next input
3602character. If we succeed, we can unread it so that it remains in the local
3603buffer for handling later. If not, the connection has been lost.
58eb016e 3604
563b63fa
PH
3605Of course, since TCP/IP is asynchronous, there is always a chance that the
3606connection will vanish between the time of this test and the sending of the
3607response, but the chance of this happening should be small. */
3608
3609if (smtp_input && sender_host_address != NULL && !sender_host_notsocket &&
3610 !receive_smtp_buffered())
58eb016e
PH
3611 {
3612 struct timeval tv;
3613 fd_set select_check;
3614 FD_ZERO(&select_check);
3615 FD_SET(fileno(smtp_in), &select_check);
3616 tv.tv_sec = 0;
3617 tv.tv_usec = 0;
3618
563b63fa 3619 if (select(fileno(smtp_in) + 1, &select_check, NULL, NULL, &tv) != 0)
58eb016e 3620 {
80a47a2c
TK
3621 int c = (receive_getc)();
3622 if (c != EOF) (receive_ungetc)(c); else
58eb016e 3623 {
563b63fa
PH
3624 uschar *msg = US"SMTP connection lost after final dot";
3625 smtp_reply = US""; /* No attempt to send a response */
3626 smtp_yield = FALSE; /* Nothing more on this connection */
58eb016e 3627
563b63fa 3628 /* Re-use the log line workspace */
58eb016e 3629
563b63fa
PH
3630 sptr = 0;
3631 s = string_cat(s, &size, &sptr, msg, Ustrlen(msg));
3632 s = add_host_info_for_log(s, &size, &sptr);
3633 s[sptr] = 0;
3634 log_write(0, LOG_MAIN, "%s", s);
58eb016e 3635
563b63fa 3636 /* Delete the files for this aborted message. */
58eb016e 3637
563b63fa
PH
3638 sprintf(CS spool_name, "%s/input/%s/%s-D", spool_directory,
3639 message_subdir, message_id);
3640 Uunlink(spool_name);
58eb016e 3641
563b63fa
PH
3642 sprintf(CS spool_name, "%s/input/%s/%s-H", spool_directory,
3643 message_subdir, message_id);
3644 Uunlink(spool_name);
58eb016e 3645
563b63fa
PH
3646 sprintf(CS spool_name, "%s/msglog/%s/%s", spool_directory,
3647 message_subdir, message_id);
3648 Uunlink(spool_name);
58eb016e 3649
563b63fa
PH
3650 goto TIDYUP;
3651 }
58eb016e
PH
3652 }
3653 }
3654
3655/* The connection has not gone away; we really are going to take responsibility
3656for this message. */
3657
3658log_write(0, LOG_MAIN |
3659 (((log_extra_selector & LX_received_recipients) != 0)? LOG_RECIPIENTS : 0) |
3660 (((log_extra_selector & LX_received_sender) != 0)? LOG_SENDER : 0),
3661 "%s", s);
3662receive_call_bombout = FALSE;
3663
3664/* Log any control actions taken by an ACL or local_scan(). */
3665
3666if (deliver_freeze) log_write(0, LOG_MAIN, "frozen by %s", frozen_by);
3667if (queue_only_policy) log_write(L_delay_delivery, LOG_MAIN,
3668 "no immediate delivery: queued by %s", queued_by);
3669
059ec3d9
PH
3670store_reset(s); /* The store for the main log message can be reused */
3671
3672/* If the message is frozen, and freeze_tell is set, do the telling. */
3673
3674if (deliver_freeze && freeze_tell != NULL && freeze_tell[0] != 0)
3675 {
3676 moan_tell_someone(freeze_tell, NULL, US"Message frozen on arrival",
3677 "Message %s was frozen on arrival by %s.\nThe sender is <%s>.\n",
3678 message_id, frozen_by, sender_address);
3679 }
3680
3681
3682/* Either a message has been successfully received and written to the two spool
3683files, or an error in writing the spool has occurred for an SMTP message, or
58eb016e
PH
3684an SMTP message has been rejected for policy reasons. (For a non-SMTP message
3685we will have already given up because there's no point in carrying on!) In
3686either event, we must now close (and thereby unlock) the data file. In the
059ec3d9
PH
3687successful case, this leaves the message on the spool, ready for delivery. In
3688the error case, the spool file will be deleted. Then tidy up store, interact
3689with an SMTP call if necessary, and return.
3690
3691A fflush() was done earlier in the expectation that any write errors on the
3692data file will be flushed(!) out thereby. Nevertheless, it is theoretically
3693possible for fclose() to fail - but what to do? What has happened to the lock
3694if this happens? */
3695
3696TIDYUP:
f1e894f3
PH
3697process_info[process_info_len] = 0; /* Remove message id */
3698if (data_file != NULL) (void)fclose(data_file); /* Frees the lock */
059ec3d9
PH
3699
3700/* Now reset signal handlers to their defaults */
3701
3702signal(SIGTERM, SIG_DFL);
3703signal(SIGINT, SIG_DFL);
3704
3705/* Tell an SMTP caller the state of play, and arrange to return the SMTP return
3706value, which defaults TRUE - meaning there may be more incoming messages from
3707this connection. For non-SMTP callers (where there is only ever one message),
3708the default is FALSE. */
3709
3710if (smtp_input)
3711 {
3712 yield = smtp_yield;
3713
3714 /* Handle interactive SMTP callers. After several kinds of error, smtp_reply
58eb016e
PH
3715 is set to the response that should be sent. When it is NULL, we generate
3716 default responses. After an ACL error or local_scan() error, the response has
3717 already been sent, and smtp_reply is an empty string to indicate this. */
059ec3d9
PH
3718
3719 if (!smtp_batched_input)
3720 {
3721 if (smtp_reply == NULL)
3722 {
29aba418 3723 if (fake_response != OK)
a5bd321b
PH
3724 smtp_respond((fake_response == DEFER)? US"450" : US"550", 3, TRUE,
3725 fake_response_text);
4e88a19f
PH
3726
3727 /* An OK response is required; use "message" text if present. */
3728
3729 else if (user_msg != NULL)
3730 {
3731 uschar *code = US"250";
3732 int len = 3;
3733 smtp_message_code(&code, &len, &user_msg, NULL);
3734 smtp_respond(code, len, TRUE, user_msg);
3735 }
3736
3737 /* Default OK response */
3738
8e669ac1
PH
3739 else
3740 smtp_printf("250 OK id=%s\r\n", message_id);
059ec3d9
PH
3741 if (host_checking)
3742 fprintf(stdout,
3743 "\n**** SMTP testing: that is not a real message id!\n\n");
3744 }
4e88a19f 3745
58eb016e 3746 /* smtp_reply is set non-empty */
4e88a19f 3747
8523533c
TK
3748 else if (smtp_reply[0] != 0)
3749 {
29aba418 3750 if (fake_response != OK && (smtp_reply[0] == '2'))
a5bd321b
PH
3751 smtp_respond((fake_response == DEFER)? US"450" : US"550", 3, TRUE,
3752 fake_response_text);
8e669ac1 3753 else
6951ac6c
PH
3754 smtp_printf("%.1024s\r\n", smtp_reply);
3755 }
059ec3d9
PH
3756 }
3757
3758 /* For batched SMTP, generate an error message on failure, and do
3759 nothing on success. The function moan_smtp_batch() does not return -
3760 it exits from the program with a non-zero return code. */
3761
3762 else if (smtp_reply != NULL) moan_smtp_batch(NULL, "%s", smtp_reply);
3763 }
3764
3765
3766/* If blackholing, we can immediately log this message's sad fate. The data
3767file has already been unlinked, and the header file was never written to disk.
3768We must now indicate that nothing was received, to prevent a delivery from
3769starting. */
3770
3771if (blackholed_by != NULL)
3772 {
3773 uschar *detail = (local_scan_data != NULL)?
3774 string_printing(local_scan_data) :
3775 string_sprintf("(%s discarded recipients)", blackholed_by);
04f7d5b9 3776 log_write(0, LOG_MAIN, "=> blackhole %s%s", detail, blackhole_log_msg);
059ec3d9
PH
3777 log_write(0, LOG_MAIN, "Completed");
3778 message_id[0] = 0;
3779 }
3780
3781/* Reset headers so that logging of rejects for a subsequent message doesn't
3782include them. It is also important to set header_last = NULL before exiting
3783from this function, as this prevents certain rewrites that might happen during
3784subsequent verifying (of another incoming message) from trying to add headers
3785when they shouldn't. */
3786
3787header_list = header_last = NULL;
3788
3789return yield; /* TRUE if more messages (SMTP only) */
3790}
3791
3792/* End of receive.c */