1 /* $Cambridge: exim/src/src/spool_in.c,v 1.9 2005/03/08 15:32:02 tom Exp $ */
3 /*************************************************
4 * Exim - an Internet mail transport agent *
5 *************************************************/
7 /* Copyright (c) University of Cambridge 1995 - 2005 */
8 /* See the file NOTICE for conditions of use and distribution. */
10 /* Functions for reading spool files. When compiling for a utility (eximon),
11 not all are needed, and some functionality can be cut out. */
18 #ifndef COMPILE_UTILITY
19 /*************************************************
20 * Open and lock data file *
21 *************************************************/
23 /* The data file is the one that is used for locking, because the header file
24 can get replaced during delivery because of header rewriting. The file has
25 to opened with write access so that we can get an exclusive lock, but in
26 fact it won't be written to. Just in case there's a major disaster (e.g.
27 overwriting some other file descriptor with the value of this one), open it
30 Argument: the id of the message
31 Returns: TRUE if file successfully opened and locked
33 Side effect: deliver_datafile is set to the fd of the open file.
37 spool_open_datafile(uschar *id)
42 uschar spoolname[256];
44 /* If split_spool_directory is set, first look for the file in the appropriate
45 sub-directory of the input directory. If it is not found there, try the input
46 directory itself, to pick up leftovers from before the splitting. If split_
47 spool_directory is not set, first look in the main input directory. If it is
48 not found there, try the split sub-directory, in case it is left over from a
51 for (i = 0; i < 2; i++)
54 message_subdir[0] = (split_spool_directory == (i == 0))? id[5] : 0;
55 sprintf(CS spoolname, "%s/input/%s/%s-D", spool_directory, message_subdir, id);
56 deliver_datafile = Uopen(spoolname, O_RDWR | O_APPEND, 0);
57 if (deliver_datafile >= 0) break;
63 log_write(0, LOG_MAIN, "Spool file %s-D not found", id);
65 else log_write(0, LOG_MAIN, "Spool error for %s: %s", spoolname,
71 /* File is open and message_subdir is set. Set the close-on-exec flag, and lock
72 the file. We lock only the first line of the file (containing the message ID)
73 because this apparently is needed for running Exim under Cygwin. If the entire
74 file is locked in one process, a sub-process cannot access it, even when passed
75 an open file descriptor (at least, I think that's the Cygwin story). On real
76 Unix systems it doesn't make any difference as long as Exim is consistent in
79 fcntl(deliver_datafile, F_SETFD, fcntl(deliver_datafile, F_GETFD) |
82 lock_data.l_type = F_WRLCK;
83 lock_data.l_whence = SEEK_SET;
84 lock_data.l_start = 0;
85 lock_data.l_len = SPOOL_DATA_START_OFFSET;
87 if (fcntl(deliver_datafile, F_SETLK, &lock_data) < 0)
89 log_write(L_skip_delivery,
91 "Spool file is locked (another process is handling this message)");
92 close(deliver_datafile);
93 deliver_datafile = -1;
98 /* Get the size of the data; don't include the leading filename line
99 in the count, but add one for the newline before the data. */
101 if (fstat(deliver_datafile, &statbuf) == 0)
103 message_body_size = statbuf.st_size - SPOOL_DATA_START_OFFSET;
104 message_size = message_body_size + 1;
109 #endif /* COMPILE_UTILITY */
113 /*************************************************
114 * Read non-recipients tree from spool file *
115 *************************************************/
117 /* The tree of non-recipients is written to the spool file in a form that
118 makes it easy to read back into a tree. The format is as follows:
120 . Each node is preceded by two letter(Y/N) indicating whether it has left
121 or right children. There's one space after the two flags, before the name.
123 . The left subtree (if any) then follows, then the right subtree (if any).
125 This function is entered with the next input line in the buffer. Note we must
126 save the right flag before recursing with the same buffer.
128 Once the tree is read, we re-construct the balance fields by scanning the tree.
129 I forgot to write them out originally, and the compatible fix is to do it this
130 way. This initial local recursing function does the necessary.
135 Returns: maximum depth below the node, including the node itself
139 count_below(tree_node *node)
142 if (node == NULL) return 0;
143 nleft = count_below(node->left);
144 nright = count_below(node->right);
145 node->balance = (nleft > nright)? 1 : ((nright > nleft)? 2 : 0);
146 return 1 + ((nleft > nright)? nleft : nright);
149 /* This is the real function...
152 connect pointer to the root of the tree
153 f FILE to read data from
154 buffer contains next input line; further lines read into it
155 buffer_size size of the buffer
157 Returns: FALSE on format error
161 read_nonrecipients_tree(tree_node **connect, FILE *f, uschar *buffer,
165 int n = Ustrlen(buffer);
166 BOOL right = buffer[1] == 'Y';
168 if (n < 5) return FALSE; /* malformed line */
169 buffer[n-1] = 0; /* Remove \n */
170 node = store_get(sizeof(tree_node) + n - 3);
172 Ustrcpy(node->name, buffer + 3);
173 node->data.ptr = NULL;
175 if (buffer[0] == 'Y')
177 if (Ufgets(buffer, buffer_size, f) == NULL ||
178 !read_nonrecipients_tree(&node->left, f, buffer, buffer_size))
181 else node->left = NULL;
185 if (Ufgets(buffer, buffer_size, f) == NULL ||
186 !read_nonrecipients_tree(&node->right, f, buffer, buffer_size))
189 else node->right = NULL;
191 (void) count_below(*connect);
198 /*************************************************
199 * Read spool header file *
200 *************************************************/
202 /* This function reads a spool header file and places the data into the
203 appropriate global variables. The header portion is always read, but header
204 structures are built only if read_headers is set true. It isn't, for example,
205 while generating -bp output.
207 It may be possible for blocks of nulls (binary zeroes) to get written on the
208 end of a file if there is a system crash during writing. It was observed on an
209 earlier version of Exim that omitted to fsync() the files - this is thought to
210 have been the cause of that incident, but in any case, this code must be robust
211 against such an event, and if such a file is encountered, it must be treated as
215 name name of the header file, including the -H
216 read_headers TRUE if in-store header structures are to be built
217 subdir_set TRUE is message_subdir is already set
219 Returns: spool_read_OK success
220 spool_read_notopen open failed
221 spool_read_enverror error in the envelope portion
222 spool_read_hdrdrror error in the header portion
226 spool_read_header(uschar *name, BOOL read_headers, BOOL subdir_set)
232 BOOL inheader = FALSE;
235 /* Reset all the global variables to their default values. However, there is
236 one exception. DO NOT change the default value of dont_deliver, because it may
237 be forced by an external setting. */
239 for (n = 0; n < ACL_C_MAX + ACL_M_MAX; n++) acl_var[n] = NULL;
241 authenticated_id = NULL;
242 authenticated_sender = NULL;
243 allow_unqualified_recipient = FALSE;
244 allow_unqualified_sender = FALSE;
247 deliver_firsttime = FALSE;
248 deliver_freeze = FALSE;
249 deliver_frozen_at = 0;
250 deliver_manual_thaw = FALSE;
251 /* dont_deliver must NOT be reset */
252 header_list = header_last = NULL;
253 host_lookup_deferred = FALSE;
254 host_lookup_failed = FALSE;
255 interface_address = NULL;
257 local_error_message = FALSE;
258 local_scan_data = NULL;
259 message_linecount = 0;
260 received_protocol = NULL;
262 recipients_list = NULL;
263 sender_address = NULL;
264 sender_fullhost = NULL;
265 sender_helo_name = NULL;
266 sender_host_address = NULL;
267 sender_host_name = NULL;
268 sender_host_port = 0;
269 sender_host_authenticated = NULL;
271 sender_local = FALSE;
272 sender_set_untrusted = FALSE;
273 smtp_active_hostname = primary_hostname;
274 tree_nonrecipients = NULL;
276 #ifdef EXPERIMENTAL_BRIGHTMAIL
281 #ifdef EXPERIMENTAL_DOMAINKEYS
286 tls_certificate_verified = FALSE;
291 #ifdef WITH_CONTENT_SCAN
292 spam_score_int = NULL;
295 /* Generate the full name and open the file. If message_subdir is already
296 set, just look in the given directory. Otherwise, look in both the split
297 and unsplit directories, as for the data file above. */
299 for (n = 0; n < 2; n++)
302 message_subdir[0] = (split_spool_directory == (n == 0))? name[5] : 0;
303 sprintf(CS big_buffer, "%s/input/%s/%s", spool_directory, message_subdir,
305 f = Ufopen(big_buffer, "rb");
306 if (f != NULL) break;
307 if (n != 0 || subdir_set || errno != ENOENT) return spool_read_notopen;
312 #ifndef COMPILE_UTILITY
313 DEBUG(D_deliver) debug_printf("reading spool file %s\n", name);
314 #endif /* COMPILE_UTILITY */
316 /* The first line of a spool file contains the message id followed by -H (i.e.
317 the file name), in order to make the file self-identifying. */
319 if (Ufgets(big_buffer, big_buffer_size, f) == NULL) goto SPOOL_READ_ERROR;
320 if (Ustrlen(big_buffer) != MESSAGE_ID_LENGTH + 3 ||
321 Ustrncmp(big_buffer, name, MESSAGE_ID_LENGTH + 2) != 0)
322 goto SPOOL_FORMAT_ERROR;
324 /* The next three lines in the header file are in a fixed format. The first
325 contains the login, uid, and gid of the user who caused the file to be written.
326 The second contains the mail address of the message's sender, enclosed in <>.
327 The third contains the time the message was received, and the number of warning
328 messages for delivery delays that have been sent. */
330 if (Ufgets(big_buffer, big_buffer_size, f) == NULL) goto SPOOL_READ_ERROR;
332 p = big_buffer + Ustrlen(big_buffer);
333 while (p > big_buffer && isspace(p[-1])) p--;
335 if (!isdigit(p[-1])) goto SPOOL_FORMAT_ERROR;
336 while (p > big_buffer && isdigit(p[-1])) p--;
338 if (p <= big_buffer || *(--p) != ' ') goto SPOOL_FORMAT_ERROR;
340 if (!isdigit(p[-1])) goto SPOOL_FORMAT_ERROR;
341 while (p > big_buffer && isdigit(p[-1])) p--;
343 if (p <= big_buffer || *(--p) != ' ') goto SPOOL_FORMAT_ERROR;
346 originator_login = string_copy(big_buffer);
347 originator_uid = (uid_t)uid;
348 originator_gid = (gid_t)gid;
350 if (Ufgets(big_buffer, big_buffer_size, f) == NULL) goto SPOOL_READ_ERROR;
351 n = Ustrlen(big_buffer);
352 if (n < 3 || big_buffer[0] != '<' || big_buffer[n-2] != '>')
353 goto SPOOL_FORMAT_ERROR;
355 sender_address = store_get(n-2);
356 Ustrncpy(sender_address, big_buffer+1, n-3);
357 sender_address[n-3] = 0;
359 if (Ufgets(big_buffer, big_buffer_size, f) == NULL) goto SPOOL_READ_ERROR;
360 if (sscanf(CS big_buffer, "%d %d", &received_time, &warning_count) != 2)
361 goto SPOOL_FORMAT_ERROR;
363 message_age = time(NULL) - received_time;
365 #ifndef COMPILE_UTILITY
366 DEBUG(D_deliver) debug_printf("user=%s uid=%ld gid=%ld sender=%s\n",
367 originator_login, (long int)originator_uid, (long int)originator_gid,
369 #endif /* COMPILE_UTILITY */
371 /* Now there may be a number of optional lines, each starting with "-".
372 If you add a new setting here, make sure you set the default above. */
376 if (Ufgets(big_buffer, big_buffer_size, f) == NULL) goto SPOOL_READ_ERROR;
377 if (big_buffer[0] != '-') break;
379 big_buffer[Ustrlen(big_buffer) - 1] = 0;
380 if (Ustrncmp(big_buffer, "-acl ", 5) == 0)
383 if (sscanf(CS big_buffer + 5, "%d %d", &index, &count) != 2)
384 goto SPOOL_FORMAT_ERROR;
385 /* Ignore if index too big - might be if a later release with more
386 variables built this spool file. */
387 if (index < ACL_C_MAX + ACL_M_MAX)
389 acl_var[index] = store_get(count + 1);
390 if (fread(acl_var[index], 1, count+1, f) < count) goto SPOOL_READ_ERROR;
391 acl_var[index][count] = 0;
394 else if (Ustrcmp(big_buffer, "-local") == 0) sender_local = TRUE;
395 else if (Ustrcmp(big_buffer, "-localerror") == 0)
396 local_error_message = TRUE;
397 else if (Ustrncmp(big_buffer, "-local_scan ", 12) == 0)
398 local_scan_data = string_copy(big_buffer + 12);
399 #ifdef WITH_CONTENT_SCAN
400 else if (Ustrncmp(big_buffer, "-spam_score_int ", 16) == 0)
401 spam_score_int = string_copy(big_buffer + 16);
403 #ifdef EXPERIMENTAL_BRIGHTMAIL
404 else if (Ustrncmp(big_buffer, "-bmi_verdicts ", 14) == 0)
405 bmi_verdicts = string_copy(big_buffer + 14);
407 else if (Ustrcmp(big_buffer, "-host_lookup_deferred") == 0)
408 host_lookup_deferred = TRUE;
409 else if (Ustrcmp(big_buffer, "-host_lookup_failed") == 0)
410 host_lookup_failed = TRUE;
411 else if (Ustrncmp(big_buffer, "-body_linecount", 15) == 0)
412 body_linecount = Uatoi(big_buffer + 15);
413 else if (Ustrncmp(big_buffer, "-body_zerocount", 15) == 0)
414 body_zerocount = Uatoi(big_buffer + 15);
415 else if (Ustrncmp(big_buffer, "-frozen", 7) == 0)
417 deliver_freeze = TRUE;
418 deliver_frozen_at = Uatoi(big_buffer + 7);
420 else if (Ustrcmp(big_buffer, "-allow_unqualified_recipient") == 0)
421 allow_unqualified_recipient = TRUE;
422 else if (Ustrcmp(big_buffer, "-allow_unqualified_sender") == 0)
423 allow_unqualified_sender = TRUE;
424 else if (Ustrcmp(big_buffer, "-deliver_firsttime") == 0)
425 deliver_firsttime = TRUE;
426 else if (Ustrcmp(big_buffer, "-manual_thaw") == 0)
427 deliver_manual_thaw = TRUE;
428 else if (Ustrncmp(big_buffer, "-auth_id", 8) == 0)
429 authenticated_id = string_copy(big_buffer + 9);
430 else if (Ustrncmp(big_buffer, "-auth_sender", 12) == 0)
431 authenticated_sender = string_copy(big_buffer + 13);
432 else if (Ustrncmp(big_buffer, "-sender_set_untrusted", 21) == 0)
433 sender_set_untrusted = TRUE;
436 else if (Ustrncmp(big_buffer, "-tls_certificate_verified", 25) == 0)
437 tls_certificate_verified = TRUE;
438 else if (Ustrncmp(big_buffer, "-tls_cipher", 11) == 0)
439 tls_cipher = string_copy(big_buffer + 12);
440 else if (Ustrncmp(big_buffer, "-tls_peerdn", 11) == 0)
441 tls_peerdn = string_copy(big_buffer + 12);
444 /* We now record the port number after the address, separated by a
445 dot. For compatibility during upgrading, do nothing if there
446 isn't a value (it gets left at zero). */
448 else if (Ustrncmp(big_buffer, "-host_address", 13) == 0)
450 sender_host_port = host_extract_port(big_buffer + 14);
451 sender_host_address = string_copy(big_buffer + 14);
454 else if (Ustrncmp(big_buffer, "-interface_address", 18) == 0)
456 interface_port = host_extract_port(big_buffer + 19);
457 interface_address = string_copy(big_buffer + 19);
460 else if (Ustrncmp(big_buffer, "-active_hostname", 16) == 0)
461 smtp_active_hostname = string_copy(big_buffer + 17);
462 else if (Ustrncmp(big_buffer, "-host_auth", 10) == 0)
463 sender_host_authenticated = string_copy(big_buffer + 11);
464 else if (Ustrncmp(big_buffer, "-host_name", 10) == 0)
465 sender_host_name = string_copy(big_buffer + 11);
466 else if (Ustrncmp(big_buffer, "-helo_name", 10) == 0)
467 sender_helo_name = string_copy(big_buffer + 11);
468 else if (Ustrncmp(big_buffer, "-ident", 6) == 0)
469 sender_ident = string_copy(big_buffer + 7);
470 else if (Ustrncmp(big_buffer, "-received_protocol", 18) == 0)
471 received_protocol = string_copy(big_buffer + 19);
472 else if (Ustrncmp(big_buffer, "-N", 2) == 0)
475 /* To allow new versions of Exim that add additional flags to interwork
476 with older versions that do not understand them, just ignore any flagged
477 lines that we don't recognize. Otherwise it wouldn't be possible to back
478 off a new version that left new-style flags written on the spool. That's
479 why the following line is commented out. */
481 /* else goto SPOOL_FORMAT_ERROR; */
484 /* Build sender_fullhost if required */
486 #ifndef COMPILE_UTILITY
487 host_build_sender_fullhost();
488 #endif /* COMPILE_UTILITY */
490 #ifndef COMPILE_UTILITY
492 debug_printf("sender_local=%d ident=%s\n", sender_local,
493 (sender_ident == NULL)? US"unset" : sender_ident);
494 #endif /* COMPILE_UTILITY */
496 /* We now have the tree of addresses NOT to deliver to, or a line
497 containing "XX", indicating no tree. */
499 if (Ustrncmp(big_buffer, "XX\n", 3) != 0 &&
500 !read_nonrecipients_tree(&tree_nonrecipients, f, big_buffer, big_buffer_size))
501 goto SPOOL_FORMAT_ERROR;
503 #ifndef COMPILE_UTILITY
506 debug_printf("Non-recipients:\n");
507 debug_print_tree(tree_nonrecipients);
509 #endif /* COMPILE_UTILITY */
511 /* After reading the tree, the next line has not yet been read into the
512 buffer. It contains the count of recipients which follow on separate lines. */
514 if (Ufgets(big_buffer, big_buffer_size, f) == NULL) goto SPOOL_READ_ERROR;
515 if (sscanf(CS big_buffer, "%d", &rcount) != 1) goto SPOOL_FORMAT_ERROR;
517 #ifndef COMPILE_UTILITY
518 DEBUG(D_deliver) debug_printf("recipients_count=%d\n", rcount);
519 #endif /* COMPILE_UTILITY */
521 recipients_list_max = rcount;
522 recipients_list = store_get(rcount * sizeof(recipient_item));
524 for (recipients_count = 0; recipients_count < rcount; recipients_count++)
528 uschar *errors_to = NULL;
531 if (Ufgets(big_buffer, big_buffer_size, f) == NULL) goto SPOOL_READ_ERROR;
532 nn = Ustrlen(big_buffer);
533 if (nn < 2) goto SPOOL_FORMAT_ERROR;
535 /* Remove the newline; this terminates the address if there is no additional
538 p = big_buffer + nn - 1;
541 /* Look back from the end of the line for digits and special terminators.
542 Since an address must end with a domain, we can tell that extra data is
543 present by the presence of the terminator, which is always some character
544 that cannot exist in a domain. (If I'd thought of the need for additional
545 data early on, I'd have put it at the start, with the address at the end. As
546 it is, we have to operate backwards. Addresses are permitted to contain
549 This code has to cope with various versions of this data that have evolved
550 over time. In all cases, the line might just contain an address, with no
551 additional data. Otherwise, the possibilities are as follows:
553 Exim 3 type: <address><space><digits>,<digits>,<digits>
555 The second set of digits is the parent number for one_time addresses. The
556 other values were remnants of earlier experiments that were abandoned.
558 Exim 4 first type: <address><space><digits>
560 The digits are the parent number for one_time addresses.
562 Exim 4 new type: <address><space><data>#<type bits>
564 The type bits indicate what the contents of the data are.
566 Bit 01 indicates that, reading from right to left, the data
567 ends with <errors_to address><space><len>,<pno> where pno is
568 the parent number for one_time addresses, and len is the length
569 of the errors_to address (zero meaning none).
572 while (isdigit(*p)) p--;
574 /* Handle Exim 3 spool files */
579 while (isdigit(*(--p)) || *p == ',');
583 sscanf(CS p, "%d,%d", &dummy, &pno);
587 /* Handle early Exim 4 spool files */
592 sscanf(CS p, "%d", &pno);
595 /* Handle current format Exim 4 spool files */
600 sscanf(CS p+1, "%d", &flags);
602 if ((flags & 0x01) != 0) /* one_time data exists */
605 while (isdigit(*(--p)) || *p == ',' || *p == '-');
606 sscanf(CS p+1, "%d,%d", &len, &pno);
611 errors_to = string_copy(p);
615 *(--p) = 0; /* Terminate address */
618 recipients_list[recipients_count].address = string_copy(big_buffer);
619 recipients_list[recipients_count].pno = pno;
620 recipients_list[recipients_count].errors_to = errors_to;
623 /* The remainder of the spool header file contains the headers for the message,
624 separated off from the previous data by a blank line. Each header is preceded
625 by a count of its length and either a certain letter (for various identified
626 headers), space (for a miscellaneous live header) or an asterisk (for a header
627 that has been rewritten). Count the Received: headers. We read the headers
628 always, in order to check on the format of the file, but only create a header
629 list if requested to do so. */
632 if (Ufgets(big_buffer, big_buffer_size, f) == NULL) goto SPOOL_READ_ERROR;
633 if (big_buffer[0] != '\n') goto SPOOL_FORMAT_ERROR;
635 while ((n = fgetc(f)) != EOF)
641 if (!isdigit(n)) goto SPOOL_FORMAT_ERROR;
643 fscanf(f, "%d%c ", &n, flag);
644 if (flag[0] != '*') message_size += n; /* Omit non-transmitted headers */
648 h = store_get(sizeof(header_line));
652 h->text = store_get(n+1);
654 if (h->type == htype_received) received_count++;
656 if (header_list == NULL) header_list = h;
657 else header_last->next = h;
660 for (i = 0; i < n; i++)
663 if (c == 0 || c == EOF) goto SPOOL_FORMAT_ERROR;
664 if (c == '\n' && h->type != htype_old) message_linecount++;
670 /* Not requiring header data, just skip through the bytes */
672 else for (i = 0; i < n; i++)
675 if (c == 0 || c == EOF) goto SPOOL_FORMAT_ERROR;
679 /* We have successfully read the data in the header file. Update the message
680 line count by adding the body linecount to the header linecount. Close the file
681 and give a positive response. */
683 #ifndef COMPILE_UTILITY
684 DEBUG(D_deliver) debug_printf("body_linecount=%d message_linecount=%d\n",
685 body_linecount, message_linecount);
686 #endif /* COMPILE_UTILITY */
688 message_linecount += body_linecount;
691 return spool_read_OK;
694 /* There was an error reading the spool or there was missing data,
695 or there was a format error. A "read error" with no errno means an
696 unexpected EOF, which we treat as a format error. */
703 #ifndef COMPILE_UTILITY
704 DEBUG(D_any) debug_printf("Error while reading spool file %s\n", name);
705 #endif /* COMPILE_UTILITY */
709 return inheader? spool_read_hdrerror : spool_read_enverror;
714 #ifndef COMPILE_UTILITY
715 DEBUG(D_any) debug_printf("Format error in spool file %s\n", name);
716 #endif /* COMPILE_UTILITY */
719 errno = ERRNO_SPOOLFORMAT;
720 return inheader? spool_read_hdrerror : spool_read_enverror;
723 /* End of spool_in.c */