1 /* $Cambridge: exim/src/src/spool_in.c,v 1.12 2005/06/27 14:29:44 ph10 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 (void)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 (void)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 There are known cases where a negative gid is used, so we allow for both
327 negative uids and gids. The second contains the mail address of the message's
328 sender, enclosed in <>. The third contains the time the message was received,
329 and the number of warning messages for delivery delays that have been sent. */
331 if (Ufgets(big_buffer, big_buffer_size, f) == NULL) goto SPOOL_READ_ERROR;
333 p = big_buffer + Ustrlen(big_buffer);
334 while (p > big_buffer && isspace(p[-1])) p--;
336 if (!isdigit(p[-1])) goto SPOOL_FORMAT_ERROR;
337 while (p > big_buffer && (isdigit(p[-1]) || '-' == p[-1])) p--;
339 if (p <= big_buffer || *(--p) != ' ') goto SPOOL_FORMAT_ERROR;
341 if (!isdigit(p[-1])) goto SPOOL_FORMAT_ERROR;
342 while (p > big_buffer && (isdigit(p[-1]) || '-' == p[-1])) p--;
344 if (p <= big_buffer || *(--p) != ' ') goto SPOOL_FORMAT_ERROR;
347 originator_login = string_copy(big_buffer);
348 originator_uid = (uid_t)uid;
349 originator_gid = (gid_t)gid;
351 if (Ufgets(big_buffer, big_buffer_size, f) == NULL) goto SPOOL_READ_ERROR;
352 n = Ustrlen(big_buffer);
353 if (n < 3 || big_buffer[0] != '<' || big_buffer[n-2] != '>')
354 goto SPOOL_FORMAT_ERROR;
356 sender_address = store_get(n-2);
357 Ustrncpy(sender_address, big_buffer+1, n-3);
358 sender_address[n-3] = 0;
360 if (Ufgets(big_buffer, big_buffer_size, f) == NULL) goto SPOOL_READ_ERROR;
361 if (sscanf(CS big_buffer, "%d %d", &received_time, &warning_count) != 2)
362 goto SPOOL_FORMAT_ERROR;
364 message_age = time(NULL) - received_time;
366 #ifndef COMPILE_UTILITY
367 DEBUG(D_deliver) debug_printf("user=%s uid=%ld gid=%ld sender=%s\n",
368 originator_login, (long int)originator_uid, (long int)originator_gid,
370 #endif /* COMPILE_UTILITY */
372 /* Now there may be a number of optional lines, each starting with "-".
373 If you add a new setting here, make sure you set the default above. */
377 if (Ufgets(big_buffer, big_buffer_size, f) == NULL) goto SPOOL_READ_ERROR;
378 if (big_buffer[0] != '-') break;
380 big_buffer[Ustrlen(big_buffer) - 1] = 0;
381 if (Ustrncmp(big_buffer, "-acl ", 5) == 0)
384 if (sscanf(CS big_buffer + 5, "%d %d", &index, &count) != 2)
385 goto SPOOL_FORMAT_ERROR;
386 /* Ignore if index too big - might be if a later release with more
387 variables built this spool file. */
388 if (index < ACL_C_MAX + ACL_M_MAX)
390 acl_var[index] = store_get(count + 1);
391 if (fread(acl_var[index], 1, count+1, f) < count) goto SPOOL_READ_ERROR;
392 acl_var[index][count] = 0;
395 else if (Ustrcmp(big_buffer, "-local") == 0) sender_local = TRUE;
396 else if (Ustrcmp(big_buffer, "-localerror") == 0)
397 local_error_message = TRUE;
398 else if (Ustrncmp(big_buffer, "-local_scan ", 12) == 0)
399 local_scan_data = string_copy(big_buffer + 12);
400 #ifdef WITH_CONTENT_SCAN
401 else if (Ustrncmp(big_buffer, "-spam_score_int ", 16) == 0)
402 spam_score_int = string_copy(big_buffer + 16);
404 #ifdef EXPERIMENTAL_BRIGHTMAIL
405 else if (Ustrncmp(big_buffer, "-bmi_verdicts ", 14) == 0)
406 bmi_verdicts = string_copy(big_buffer + 14);
408 else if (Ustrcmp(big_buffer, "-host_lookup_deferred") == 0)
409 host_lookup_deferred = TRUE;
410 else if (Ustrcmp(big_buffer, "-host_lookup_failed") == 0)
411 host_lookup_failed = TRUE;
412 else if (Ustrncmp(big_buffer, "-body_linecount", 15) == 0)
413 body_linecount = Uatoi(big_buffer + 15);
414 else if (Ustrncmp(big_buffer, "-body_zerocount", 15) == 0)
415 body_zerocount = Uatoi(big_buffer + 15);
416 else if (Ustrncmp(big_buffer, "-frozen", 7) == 0)
418 deliver_freeze = TRUE;
419 deliver_frozen_at = Uatoi(big_buffer + 7);
421 else if (Ustrcmp(big_buffer, "-allow_unqualified_recipient") == 0)
422 allow_unqualified_recipient = TRUE;
423 else if (Ustrcmp(big_buffer, "-allow_unqualified_sender") == 0)
424 allow_unqualified_sender = TRUE;
425 else if (Ustrcmp(big_buffer, "-deliver_firsttime") == 0)
426 deliver_firsttime = TRUE;
427 else if (Ustrcmp(big_buffer, "-manual_thaw") == 0)
428 deliver_manual_thaw = TRUE;
429 else if (Ustrncmp(big_buffer, "-auth_id", 8) == 0)
430 authenticated_id = string_copy(big_buffer + 9);
431 else if (Ustrncmp(big_buffer, "-auth_sender", 12) == 0)
432 authenticated_sender = string_copy(big_buffer + 13);
433 else if (Ustrncmp(big_buffer, "-sender_set_untrusted", 21) == 0)
434 sender_set_untrusted = TRUE;
437 else if (Ustrncmp(big_buffer, "-tls_certificate_verified", 25) == 0)
438 tls_certificate_verified = TRUE;
439 else if (Ustrncmp(big_buffer, "-tls_cipher", 11) == 0)
440 tls_cipher = string_copy(big_buffer + 12);
441 else if (Ustrncmp(big_buffer, "-tls_peerdn", 11) == 0)
442 tls_peerdn = string_copy(big_buffer + 12);
445 /* We now record the port number after the address, separated by a
446 dot. For compatibility during upgrading, do nothing if there
447 isn't a value (it gets left at zero). */
449 else if (Ustrncmp(big_buffer, "-host_address", 13) == 0)
451 sender_host_port = host_extract_port(big_buffer + 14);
452 sender_host_address = string_copy(big_buffer + 14);
455 else if (Ustrncmp(big_buffer, "-interface_address", 18) == 0)
457 interface_port = host_extract_port(big_buffer + 19);
458 interface_address = string_copy(big_buffer + 19);
461 else if (Ustrncmp(big_buffer, "-active_hostname", 16) == 0)
462 smtp_active_hostname = string_copy(big_buffer + 17);
463 else if (Ustrncmp(big_buffer, "-host_auth", 10) == 0)
464 sender_host_authenticated = string_copy(big_buffer + 11);
465 else if (Ustrncmp(big_buffer, "-host_name", 10) == 0)
466 sender_host_name = string_copy(big_buffer + 11);
467 else if (Ustrncmp(big_buffer, "-helo_name", 10) == 0)
468 sender_helo_name = string_copy(big_buffer + 11);
469 else if (Ustrncmp(big_buffer, "-ident", 6) == 0)
470 sender_ident = string_copy(big_buffer + 7);
471 else if (Ustrncmp(big_buffer, "-received_protocol", 18) == 0)
472 received_protocol = string_copy(big_buffer + 19);
473 else if (Ustrncmp(big_buffer, "-N", 2) == 0)
476 /* To allow new versions of Exim that add additional flags to interwork
477 with older versions that do not understand them, just ignore any flagged
478 lines that we don't recognize. Otherwise it wouldn't be possible to back
479 off a new version that left new-style flags written on the spool. That's
480 why the following line is commented out. */
482 /* else goto SPOOL_FORMAT_ERROR; */
485 /* Build sender_fullhost if required */
487 #ifndef COMPILE_UTILITY
488 host_build_sender_fullhost();
489 #endif /* COMPILE_UTILITY */
491 #ifndef COMPILE_UTILITY
493 debug_printf("sender_local=%d ident=%s\n", sender_local,
494 (sender_ident == NULL)? US"unset" : sender_ident);
495 #endif /* COMPILE_UTILITY */
497 /* We now have the tree of addresses NOT to deliver to, or a line
498 containing "XX", indicating no tree. */
500 if (Ustrncmp(big_buffer, "XX\n", 3) != 0 &&
501 !read_nonrecipients_tree(&tree_nonrecipients, f, big_buffer, big_buffer_size))
502 goto SPOOL_FORMAT_ERROR;
504 #ifndef COMPILE_UTILITY
507 debug_printf("Non-recipients:\n");
508 debug_print_tree(tree_nonrecipients);
510 #endif /* COMPILE_UTILITY */
512 /* After reading the tree, the next line has not yet been read into the
513 buffer. It contains the count of recipients which follow on separate lines. */
515 if (Ufgets(big_buffer, big_buffer_size, f) == NULL) goto SPOOL_READ_ERROR;
516 if (sscanf(CS big_buffer, "%d", &rcount) != 1) goto SPOOL_FORMAT_ERROR;
518 #ifndef COMPILE_UTILITY
519 DEBUG(D_deliver) debug_printf("recipients_count=%d\n", rcount);
520 #endif /* COMPILE_UTILITY */
522 recipients_list_max = rcount;
523 recipients_list = store_get(rcount * sizeof(recipient_item));
525 for (recipients_count = 0; recipients_count < rcount; recipients_count++)
529 uschar *errors_to = NULL;
532 if (Ufgets(big_buffer, big_buffer_size, f) == NULL) goto SPOOL_READ_ERROR;
533 nn = Ustrlen(big_buffer);
534 if (nn < 2) goto SPOOL_FORMAT_ERROR;
536 /* Remove the newline; this terminates the address if there is no additional
539 p = big_buffer + nn - 1;
542 /* Look back from the end of the line for digits and special terminators.
543 Since an address must end with a domain, we can tell that extra data is
544 present by the presence of the terminator, which is always some character
545 that cannot exist in a domain. (If I'd thought of the need for additional
546 data early on, I'd have put it at the start, with the address at the end. As
547 it is, we have to operate backwards. Addresses are permitted to contain
550 This code has to cope with various versions of this data that have evolved
551 over time. In all cases, the line might just contain an address, with no
552 additional data. Otherwise, the possibilities are as follows:
554 Exim 3 type: <address><space><digits>,<digits>,<digits>
556 The second set of digits is the parent number for one_time addresses. The
557 other values were remnants of earlier experiments that were abandoned.
559 Exim 4 first type: <address><space><digits>
561 The digits are the parent number for one_time addresses.
563 Exim 4 new type: <address><space><data>#<type bits>
565 The type bits indicate what the contents of the data are.
567 Bit 01 indicates that, reading from right to left, the data
568 ends with <errors_to address><space><len>,<pno> where pno is
569 the parent number for one_time addresses, and len is the length
570 of the errors_to address (zero meaning none).
573 while (isdigit(*p)) p--;
575 /* Handle Exim 3 spool files */
580 while (isdigit(*(--p)) || *p == ',');
584 (void)sscanf(CS p, "%d,%d", &dummy, &pno);
588 /* Handle early Exim 4 spool files */
593 (void)sscanf(CS p, "%d", &pno);
596 /* Handle current format Exim 4 spool files */
601 (void)sscanf(CS p+1, "%d", &flags);
603 if ((flags & 0x01) != 0) /* one_time data exists */
606 while (isdigit(*(--p)) || *p == ',' || *p == '-');
607 (void)sscanf(CS p+1, "%d,%d", &len, &pno);
612 errors_to = string_copy(p);
616 *(--p) = 0; /* Terminate address */
619 recipients_list[recipients_count].address = string_copy(big_buffer);
620 recipients_list[recipients_count].pno = pno;
621 recipients_list[recipients_count].errors_to = errors_to;
624 /* The remainder of the spool header file contains the headers for the message,
625 separated off from the previous data by a blank line. Each header is preceded
626 by a count of its length and either a certain letter (for various identified
627 headers), space (for a miscellaneous live header) or an asterisk (for a header
628 that has been rewritten). Count the Received: headers. We read the headers
629 always, in order to check on the format of the file, but only create a header
630 list if requested to do so. */
633 if (Ufgets(big_buffer, big_buffer_size, f) == NULL) goto SPOOL_READ_ERROR;
634 if (big_buffer[0] != '\n') goto SPOOL_FORMAT_ERROR;
636 while ((n = fgetc(f)) != EOF)
642 if (!isdigit(n)) goto SPOOL_FORMAT_ERROR;
644 (void)fscanf(f, "%d%c ", &n, flag);
645 if (flag[0] != '*') message_size += n; /* Omit non-transmitted headers */
649 h = store_get(sizeof(header_line));
653 h->text = store_get(n+1);
655 if (h->type == htype_received) received_count++;
657 if (header_list == NULL) header_list = h;
658 else header_last->next = h;
661 for (i = 0; i < n; i++)
664 if (c == 0 || c == EOF) goto SPOOL_FORMAT_ERROR;
665 if (c == '\n' && h->type != htype_old) message_linecount++;
671 /* Not requiring header data, just skip through the bytes */
673 else for (i = 0; i < n; i++)
676 if (c == 0 || c == EOF) goto SPOOL_FORMAT_ERROR;
680 /* We have successfully read the data in the header file. Update the message
681 line count by adding the body linecount to the header linecount. Close the file
682 and give a positive response. */
684 #ifndef COMPILE_UTILITY
685 DEBUG(D_deliver) debug_printf("body_linecount=%d message_linecount=%d\n",
686 body_linecount, message_linecount);
687 #endif /* COMPILE_UTILITY */
689 message_linecount += body_linecount;
692 return spool_read_OK;
695 /* There was an error reading the spool or there was missing data,
696 or there was a format error. A "read error" with no errno means an
697 unexpected EOF, which we treat as a format error. */
704 #ifndef COMPILE_UTILITY
705 DEBUG(D_any) debug_printf("Error while reading spool file %s\n", name);
706 #endif /* COMPILE_UTILITY */
710 return inheader? spool_read_hdrerror : spool_read_enverror;
715 #ifndef COMPILE_UTILITY
716 DEBUG(D_any) debug_printf("Format error in spool file %s\n", name);
717 #endif /* COMPILE_UTILITY */
720 errno = ERRNO_SPOOLFORMAT;
721 return inheader? spool_read_hdrerror : spool_read_enverror;
724 /* End of spool_in.c */