Fix retry key bug for pipe, file, or autoreply deliveries.
[exim.git] / src / src / retry.c
1 /* $Cambridge: exim/src/src/retry.c,v 1.6 2006/02/08 14:28:51 ph10 Exp $ */
2
3 /*************************************************
4 * Exim - an Internet mail transport agent *
5 *************************************************/
6
7 /* Copyright (c) University of Cambridge 1995 - 2006 */
8 /* See the file NOTICE for conditions of use and distribution. */
9
10 /* Functions concerned with retrying unsuccessful deliveries. */
11
12
13 #include "exim.h"
14
15
16
17 /*************************************************
18 * Check the ultimate address timeout *
19 *************************************************/
20
21 /* This function tests whether a message has been on the queue longer than
22 the maximum retry time for a particular host.
23
24 Arguments:
25 host_key the key to look up a host retry rule
26 domain the domain to look up a domain retry rule
27 basic_errno a specific error number, or zero if none
28 more_errno additional data for the error
29 now the time
30
31 Returns: TRUE if the ultimate timeout has been reached
32 */
33
34 static BOOL
35 ultimate_address_timeout(uschar *host_key, uschar *domain, int basic_errno,
36 int more_errno, time_t now)
37 {
38 BOOL address_timeout = TRUE; /* no rule => timed out */
39
40 retry_config *retry =
41 retry_find_config(host_key+2, domain, basic_errno, more_errno);
42
43 if (retry != NULL && retry->rules != NULL)
44 {
45 retry_rule *last_rule;
46 for (last_rule = retry->rules;
47 last_rule->next != NULL;
48 last_rule = last_rule->next);
49 DEBUG(D_transport|D_retry)
50 debug_printf("now=%d received_time=%d diff=%d timeout=%d\n",
51 (int)now, received_time, (int)(now - received_time),
52 last_rule->timeout);
53 address_timeout = (now - received_time > last_rule->timeout);
54 }
55 else
56 {
57 DEBUG(D_transport|D_retry)
58 debug_printf("no retry rule found: assume timed out\n");
59 }
60
61 return address_timeout;
62 }
63
64
65
66 /*************************************************
67 * Set status of a host+address item *
68 *************************************************/
69
70 /* This function is passed a host_item which contains a host name and an
71 IP address string. Its job is to set the status of the address if it is not
72 already set (indicated by hstatus_unknown). The possible values are:
73
74 hstatus_usable the address is not listed in the unusable tree, and does
75 not have a retry record, OR the time is past the next
76 try time, OR the message has been on the queue for more
77 than the maximum retry time for a failing host
78
79 hstatus_unusable the address is listed in the unusable tree, or does have
80 a retry record, and the time is not yet at the next retry
81 time.
82
83 hstatus_unusable_expired as above, but also the retry time has expired
84 for this address.
85
86 The reason a delivery is permitted when a message has been around for a very
87 long time is to allow the ultimate address timeout to operate after a delivery
88 failure. Otherwise some messages may stick around without being tried for too
89 long.
90
91 If a host retry record is retrieved from the hints database, the time of last
92 trying is filled into the last_try field of the host block. If a host is
93 generally usable, a check is made to see if there is a retry delay on this
94 specific message at this host.
95
96 If a non-standard port is being used, it is added to the retry key.
97
98 Arguments:
99 domain the address domain
100 host pointer to a host item
101 portstring "" for standard port, ":xxxx" for a non-standard port
102 include_ip_address TRUE to include the address in the key - this is
103 usual, but sometimes is not wanted
104 retry_host_key where to put a pointer to the key for the host-specific
105 retry record, if one is read and the host is usable
106 retry_message_key where to put a pointer to the key for the message+host
107 retry record, if one is read and the host is usable
108
109 Returns: TRUE if the host has expired but is usable because
110 its retry time has come
111 */
112
113 BOOL
114 retry_check_address(uschar *domain, host_item *host, uschar *portstring,
115 BOOL include_ip_address, uschar **retry_host_key, uschar **retry_message_key)
116 {
117 BOOL yield = FALSE;
118 time_t now = time(NULL);
119 uschar *host_key, *message_key;
120 open_db dbblock;
121 open_db *dbm_file;
122 tree_node *node;
123 dbdata_retry *host_retry_record, *message_retry_record;
124
125 *retry_host_key = *retry_message_key = NULL;
126
127 DEBUG(D_transport|D_retry) debug_printf("checking status of %s\n", host->name);
128
129 /* Do nothing if status already set; otherwise initialize status as usable. */
130
131 if (host->status != hstatus_unknown) return FALSE;
132 host->status = hstatus_usable;
133
134 /* Generate the host key for the unusable tree and the retry database. Ensure
135 host names are lower cased (that's what %S does). */
136
137 host_key = include_ip_address?
138 string_sprintf("T:%S:%s%s", host->name, host->address, portstring) :
139 string_sprintf("T:%S%s", host->name, portstring);
140
141 /* Generate the message-specific key */
142
143 message_key = string_sprintf("%s:%s", host_key, message_id);
144
145 /* Search the tree of unusable IP addresses. This is filled in when deliveries
146 fail, because the retry database itself is not updated until the end of all
147 deliveries (so as to do it all in one go). The tree records addresses that have
148 become unusable during this delivery process (i.e. those that will get put into
149 the retry database when it is updated). */
150
151 node = tree_search(tree_unusable, host_key);
152 if (node != NULL)
153 {
154 DEBUG(D_transport|D_retry) debug_printf("found in tree of unusables\n");
155 host->status = (node->data.val > 255)?
156 hstatus_unusable_expired : hstatus_unusable;
157 host->why = node->data.val & 255;
158 return FALSE;
159 }
160
161 /* Open the retry database, giving up if there isn't one. Otherwise, search for
162 the retry records, and then close the database again. */
163
164 if ((dbm_file = dbfn_open(US"retry", O_RDONLY, &dbblock, FALSE)) == NULL)
165 {
166 DEBUG(D_deliver|D_retry|D_hints_lookup)
167 debug_printf("no retry data available\n");
168 return FALSE;
169 }
170 host_retry_record = dbfn_read(dbm_file, host_key);
171 message_retry_record = dbfn_read(dbm_file, message_key);
172 dbfn_close(dbm_file);
173
174 /* Ignore the data if it is too old - too long since it was written */
175
176 if (host_retry_record == NULL)
177 {
178 DEBUG(D_transport|D_retry) debug_printf("no host retry record\n");
179 }
180 else if (now - host_retry_record->time_stamp > retry_data_expire)
181 {
182 host_retry_record = NULL;
183 DEBUG(D_transport|D_retry) debug_printf("host retry record too old\n");
184 }
185
186 if (message_retry_record == NULL)
187 {
188 DEBUG(D_transport|D_retry) debug_printf("no message retry record\n");
189 }
190 else if (now - message_retry_record->time_stamp > retry_data_expire)
191 {
192 message_retry_record = NULL;
193 DEBUG(D_transport|D_retry) debug_printf("message retry record too old\n");
194 }
195
196 /* If there's a host-specific retry record, check for reaching the retry
197 time (or forcing). If not, and the host is not expired, check for the message
198 having been around for longer than the maximum retry time for this host or
199 address. Allow the delivery if it has. Otherwise set the appropriate unusable
200 flag and return FALSE. Otherwise arrange to return TRUE if this is an expired
201 host. */
202
203 if (host_retry_record != NULL)
204 {
205 *retry_host_key = host_key;
206
207 /* We have not reached the next try time. Check for the ultimate address
208 timeout if the host has not expired. */
209
210 if (now < host_retry_record->next_try && !deliver_force)
211 {
212 DEBUG(D_transport|D_retry)
213 debug_printf("host retry time not reached: checking ultimate address "
214 "timeout\n");
215
216 if (!host_retry_record->expired &&
217 ultimate_address_timeout(host_key, domain,
218 host_retry_record->basic_errno, host_retry_record->more_errno, now))
219 {
220 DEBUG(D_transport|D_retry)
221 debug_printf("on queue longer than maximum retry for "
222 "address - allowing delivery\n");
223 return FALSE;
224 }
225
226 /* We have not hit the ultimate address timeout; host is unusable. */
227
228 host->status = (host_retry_record->expired)?
229 hstatus_unusable_expired : hstatus_unusable;
230 host->why = hwhy_retry;
231 host->last_try = host_retry_record->last_try;
232 return FALSE;
233 }
234
235 /* Host is usable; set return TRUE if expired. */
236
237 yield = host_retry_record->expired;
238 }
239
240 /* It's OK to try the host. If there's a message-specific retry record, check
241 for reaching its retry time (or forcing). If not, mark the host unusable,
242 unless the ultimate address timeout has been reached. */
243
244 if (message_retry_record != NULL)
245 {
246 *retry_message_key = message_key;
247 if (now < message_retry_record->next_try && !deliver_force)
248 {
249 DEBUG(D_transport|D_retry)
250 debug_printf("host+message retry time not reached: checking ultimate "
251 "address timeout\n");
252 if (!ultimate_address_timeout(host_key, domain, 0, 0, now))
253 {
254 host->status = hstatus_unusable;
255 host->why = hwhy_retry;
256 }
257 else
258 {
259 DEBUG(D_transport|D_retry)
260 debug_printf("on queue longer than maximum retry for "
261 "address - allowing delivery\n");
262 }
263 return FALSE;
264 }
265 }
266
267 return yield;
268 }
269
270
271
272
273 /*************************************************
274 * Add a retry item to an address *
275 *************************************************/
276
277 /* Retry items are chained onto an address when it is deferred either by router
278 or by a transport, or if it succeeds or fails and there was a previous retry
279 item that now needs to be deleted. Sometimes there can be both kinds of item:
280 for example, if routing was deferred but then succeeded, and delivery then
281 deferred. In that case there is a delete item for the routing retry, and an
282 updating item for the delivery.
283
284 (But note that that is only visible at the outer level, because in remote
285 delivery subprocesses, the address starts "clean", with no retry items carried
286 in.)
287
288 These items are used at the end of a delivery attempt to update the retry
289 database. The keys start R: for routing delays and T: for transport delays.
290
291 Arguments:
292 addr the address block onto which to hang the item
293 key the retry key
294 flags delete, host, and message flags, copied into the block
295
296 Returns: nothing
297 */
298
299 void
300 retry_add_item(address_item *addr, uschar *key, int flags)
301 {
302 retry_item *rti = store_get(sizeof(retry_item));
303 rti->next = addr->retries;
304 addr->retries = rti;
305 rti->key = key;
306 rti->basic_errno = addr->basic_errno;
307 rti->more_errno = addr->more_errno;
308 rti->message = addr->message;
309 rti->flags = flags;
310
311 DEBUG(D_transport|D_retry)
312 {
313 int letter = rti->more_errno & 255;
314 debug_printf("added retry item for %s: errno=%d more_errno=", rti->key,
315 rti->basic_errno);
316 if (letter == 'A' || letter == 'M')
317 debug_printf("%d,%c", (rti->more_errno >> 8) & 255, letter);
318 else
319 debug_printf("%d", rti->more_errno);
320 debug_printf(" flags=%d\n", flags);
321 }
322 }
323
324
325
326 /*************************************************
327 * Find retry configuration data *
328 *************************************************/
329
330 /* Search the in-store retry information for the first retry item that applies
331 to a given destination. If the key contains an @ we are probably handling a
332 local delivery and have a complete address to search for; this happens when
333 retry_use_local_part is set on a router. Otherwise, the key is likely to be a
334 host name for a remote delivery, or a domain name for a local delivery. We
335 prepend *@ on the front of it so that it will match a retry item whose address
336 item pattern is independent of the local part. The alternate key, if set, is
337 always just a domain, so we treat it likewise.
338
339 Arguments:
340 key key for which retry info is wanted
341 alternate alternative key, always just a domain
342 basic_errno specific error predicate on the retry rule, or zero
343 more_errno additional data for errno predicate
344
345 Returns: pointer to retry rule, or NULL
346 */
347
348 retry_config *
349 retry_find_config(uschar *key, uschar *alternate, int basic_errno,
350 int more_errno)
351 {
352 int replace = 0;
353 uschar *use_key, *use_alternate;
354 uschar *colon = Ustrchr(key, ':');
355 retry_config *yield;
356
357 /* If there's a colon in the key, there are two possibilities:
358
359 (1) This is a key for a host, ip address, and possibly port, in the format
360
361 hostname:ip+port
362
363 In this case, we temporarily replace the colon with a zero, to terminate
364 the string after the host name.
365
366 (2) This is a key for a pipe, file, or autoreply delivery, in the format
367
368 pipe-or-file-or-auto:x@y
369
370 where x@y is the original address that provoked the delivery. The pipe or
371 file or auto will start with | or / or >, whereas a host name will start
372 with a letter or a digit. In this case we want to use the original address
373 to search for a retry rule. */
374
375 if (colon != NULL)
376 {
377 if (isalnum(*key))
378 replace = ':';
379 else
380 key = Ustrrchr(key, ':') + 1; /* Take from the last colon */
381 }
382
383 if (replace == 0) colon = key + Ustrlen(key);
384 *colon = 0;
385
386 /* Sort out the keys */
387
388 use_key = (Ustrchr(key, '@') != NULL)? key : string_sprintf("*@%s", key);
389 use_alternate = (alternate == NULL)? NULL : string_sprintf("*@%s", alternate);
390
391 /* Scan the configured retry items. */
392
393 for (yield = retries; yield != NULL; yield = yield->next)
394 {
395 uschar *plist = yield->pattern;
396 uschar *slist = yield->senders;
397
398 /* If a specific error is set for this item, check that we are handling that
399 specific error, and if so, check any additional error information if
400 required. */
401
402 if (yield->basic_errno != 0)
403 {
404 /* Special code is required for quota errors, as these can either be system
405 quota errors, or Exim's own quota imposition, which has a different error
406 number. Full partitions are also treated in the same way as quota errors.
407 */
408
409 if (yield->basic_errno == ERRNO_EXIMQUOTA)
410 {
411 if ((basic_errno != ERRNO_EXIMQUOTA && basic_errno != errno_quota &&
412 basic_errno != ENOSPC) ||
413 (yield->more_errno != 0 && yield->more_errno > more_errno))
414 continue;
415 }
416
417 /* Handle 4xx responses to RCPT. The code that was received is in the 2nd
418 least significant byte of more_errno (with 400 subtracted). The required
419 value is coded in the 2nd least significant byte of the yield->more_errno
420 field as follows:
421
422 255 => any 4xx code
423 >= 100 => the decade must match the value less 100
424 < 100 => the exact value must match
425 */
426
427 else if (yield->basic_errno == ERRNO_RCPT4XX)
428 {
429 int wanted;
430 if (basic_errno != ERRNO_RCPT4XX) continue;
431 wanted = (yield->more_errno >> 8) & 255;
432 if (wanted != 255)
433 {
434 int evalue = (more_errno >> 8) & 255;
435 if (wanted >= 100)
436 {
437 if ((evalue/10)*10 != wanted - 100) continue;
438 }
439 else if (evalue != wanted) continue;
440 }
441 }
442
443 /* There are some special cases for timeouts */
444
445 else if (yield->basic_errno == ETIMEDOUT)
446 {
447 if (basic_errno != ETIMEDOUT) continue;
448
449 /* Just RTEF_CTOUT in the rule => don't care about 'A'/'M' addresses */
450 if (yield->more_errno == RTEF_CTOUT)
451 {
452 if ((more_errno & RTEF_CTOUT) == 0) continue;
453 }
454
455 else if (yield->more_errno != 0)
456 {
457 int cf_errno = more_errno;
458 if ((yield->more_errno & RTEF_CTOUT) == 0) cf_errno &= ~RTEF_CTOUT;
459 if (yield->more_errno != cf_errno) continue;
460 }
461 }
462
463 /* Default checks for exact match */
464
465 else
466 {
467 if (yield->basic_errno != basic_errno ||
468 (yield->more_errno != 0 && yield->more_errno != more_errno))
469 continue;
470 }
471 }
472
473 /* If the "senders" condition is set, check it. Note that sender_address may
474 be null during -brt checking, in which case we do not use this rule. */
475
476 if (slist != NULL && (sender_address == NULL ||
477 match_address_list(sender_address, TRUE, TRUE, &slist, NULL, -1, 0,
478 NULL) != OK))
479 continue;
480
481 /* Check for a match between the address list item at the start of this retry
482 rule and either the main or alternate keys. */
483
484 if (match_address_list(use_key, TRUE, TRUE, &plist, NULL, -1, UCHAR_MAX+1,
485 NULL) == OK ||
486 (use_alternate != NULL &&
487 match_address_list(use_alternate, TRUE, TRUE, &plist, NULL, -1,
488 UCHAR_MAX+1, NULL) == OK))
489 break;
490 }
491
492 *colon = replace;
493 return yield;
494 }
495
496
497
498
499 /*************************************************
500 * Update retry database *
501 *************************************************/
502
503 /* Update the retry data for any directing/routing/transporting that was
504 deferred, or delete it for those that succeeded after a previous defer. This is
505 done all in one go to minimize opening/closing/locking of the database file.
506
507 Note that, because SMTP delivery involves a list of destinations to try, there
508 may be defer-type retry information for some of them even when the message was
509 successfully delivered. Likewise if it eventually failed.
510
511 This function may move addresses from the defer to the failed queue if the
512 ultimate retry time has expired.
513
514 Arguments:
515 addr_defer queue of deferred addresses
516 addr_failed queue of failed addresses
517 addr_succeed queue of successful addresses
518
519 Returns: nothing
520 */
521
522 void
523 retry_update(address_item **addr_defer, address_item **addr_failed,
524 address_item **addr_succeed)
525 {
526 open_db dbblock;
527 open_db *dbm_file = NULL;
528 time_t now = time(NULL);
529 int i;
530
531 DEBUG(D_retry) debug_printf("Processing retry items\n");
532
533 /* Three-times loop to handle succeeded, failed, and deferred addresses.
534 Deferred addresses must be handled after failed ones, because some may be moved
535 to the failed chain if they have timed out. */
536
537 for (i = 0; i < 3; i++)
538 {
539 address_item *endaddr, *addr;
540 address_item *last_first = NULL;
541 address_item **paddr = (i==0)? addr_succeed :
542 (i==1)? addr_failed : addr_defer;
543 address_item **saved_paddr = NULL;
544
545 DEBUG(D_retry) debug_printf("%s addresses:\n", (i == 0)? "Succeeded" :
546 (i == 1)? "Failed" : "Deferred");
547
548 /* Loop for each address on the chain. For deferred addresses, the whole
549 address times out unless one of its retry addresses has a retry rule that
550 hasn't yet timed out. Deferred addresses should not be requesting deletion
551 of retry items, but just in case they do by accident, treat that case
552 as "not timed out".
553
554 As well as handling the addresses themselves, we must also process any
555 retry items for any parent addresses - these are typically "delete" items,
556 because the parent must have succeeded in order to generate the child. */
557
558 while ((endaddr = *paddr) != NULL)
559 {
560 BOOL timed_out = FALSE;
561 retry_item *rti;
562
563 for (addr = endaddr; addr != NULL; addr = addr->parent)
564 {
565 int update_count = 0;
566 int timedout_count = 0;
567
568 DEBUG(D_retry) debug_printf("%s%s\n", addr->address, (addr->retries == NULL)?
569 ": no retry items" : "");
570
571 /* Loop for each retry item. */
572
573 for (rti = addr->retries; rti != NULL; rti = rti->next)
574 {
575 uschar *message;
576 int message_length, message_space, failing_interval, next_try;
577 retry_rule *rule, *final_rule;
578 retry_config *retry;
579 dbdata_retry *retry_record;
580
581 /* Open the retry database if it is not already open; failure to open
582 the file is logged, but otherwise ignored - deferred addresses will
583 get retried at the next opportunity. Not opening earlier than this saves
584 opening if no addresses have retry items - common when none have yet
585 reached their retry next try time. */
586
587 if (dbm_file == NULL)
588 dbm_file = dbfn_open(US"retry", O_RDWR, &dbblock, TRUE);
589
590 if (dbm_file == NULL)
591 {
592 DEBUG(D_deliver|D_retry|D_hints_lookup)
593 debug_printf("retry database not available for updating\n");
594 return;
595 }
596
597 /* If there are no deferred addresses, that is, if this message is
598 completing, and the retry item is for a message-specific SMTP error,
599 force it to be deleted, because there's no point in keeping data for
600 no-longer-existing messages. This situation can occur when a domain has
601 two hosts and a message-specific error occurs for the first of them,
602 but the address gets delivered to the second one. This optimization
603 doesn't succeed in cleaning out all the dead entries, but it helps. */
604
605 if (*addr_defer == NULL && (rti->flags & rf_message) != 0)
606 rti->flags |= rf_delete;
607
608 /* Handle the case of a request to delete the retry info for this
609 destination. */
610
611 if ((rti->flags & rf_delete) != 0)
612 {
613 (void)dbfn_delete(dbm_file, rti->key);
614 DEBUG(D_retry)
615 debug_printf("deleted retry information for %s\n", rti->key);
616 continue;
617 }
618
619 /* Count the number of non-delete retry items. This is so that we
620 can compare it to the count of timed_out ones, to check whether
621 all are timed out. */
622
623 update_count++;
624
625 /* Get the retry information for this destination and error code, if
626 any. If this item is for a remote host with ip address, then pass
627 the domain name as an alternative to search for. If no retry
628 information is found, we can't generate a retry time, so there is
629 no point updating the database. This retry item is timed out. */
630
631 if ((retry = retry_find_config(rti->key + 2,
632 ((rti->flags & rf_host) != 0)? addr->domain : NULL,
633 rti->basic_errno, rti->more_errno)) == NULL)
634 {
635 DEBUG(D_retry) debug_printf("No configured retry item for %s%s%s\n",
636 rti->key,
637 ((rti->flags & rf_host) != 0)? US" or " : US"",
638 ((rti->flags & rf_host) != 0)? addr->domain : US"");
639 if (addr == endaddr) timedout_count++;
640 continue;
641 }
642
643 DEBUG(D_retry)
644 {
645 if ((rti->flags & rf_host) != 0)
646 debug_printf("retry for %s (%s) = %s %d %d\n", rti->key,
647 addr->domain, retry->pattern, retry->basic_errno,
648 retry->more_errno);
649 else
650 debug_printf("retry for %s = %s %d %d\n", rti->key, retry->pattern,
651 retry->basic_errno, retry->more_errno);
652 }
653
654 /* Set up the message for the database retry record. Because DBM
655 records have a maximum data length, we enforce a limit. There isn't
656 much point in keeping a huge message here, anyway. */
657
658 message = (rti->basic_errno > 0)? US strerror(rti->basic_errno) :
659 (rti->message == NULL)?
660 US"unknown error" : string_printing(rti->message);
661 message_length = Ustrlen(message);
662 if (message_length > 150) message_length = 150;
663
664 /* Read a retry record from the database or construct a new one.
665 Ignore an old one if it is too old since it was last updated. */
666
667 retry_record = dbfn_read(dbm_file, rti->key);
668 if (retry_record != NULL &&
669 now - retry_record->time_stamp > retry_data_expire)
670 retry_record = NULL;
671
672 if (retry_record == NULL)
673 {
674 retry_record = store_get(sizeof(dbdata_retry) + message_length);
675 message_space = message_length;
676 retry_record->first_failed = now;
677 retry_record->last_try = now;
678 retry_record->next_try = now;
679 retry_record->expired = FALSE;
680 retry_record->text[0] = 0; /* just in case */
681 }
682 else message_space = Ustrlen(retry_record->text);
683
684 /* Compute how long this destination has been failing */
685
686 failing_interval = now - retry_record->first_failed;
687
688 /* Search for the current retry rule. The cutoff time of the
689 last rule is handled differently to the others. The rule continues
690 to operate for ever (the global maximum interval will eventually
691 limit the gaps) but its cutoff time determines when an individual
692 destination times out. If there are no retry rules, the destination
693 always times out, but we can't compute a retry time. */
694
695 final_rule = NULL;
696 for (rule = retry->rules; rule != NULL; rule = rule->next)
697 {
698 if (failing_interval <= rule->timeout) break;
699 final_rule = rule;
700 }
701
702 /* If there's an un-timed out rule, the destination has not
703 yet timed out, so the address as a whole has not timed out (but we are
704 interested in this only for the end address). Make sure the expired
705 flag is false (can be forced via fixdb from outside, but ensure it is
706 consistent with the rules whenever we go through here). */
707
708 if (rule != NULL)
709 {
710 retry_record->expired = FALSE;
711 }
712
713 /* Otherwise, set the retry timeout expired, and set the final rule
714 as the one from which to compute the next retry time. Subsequent
715 messages will fail immediately until the retry time is reached (unless
716 there are other, still active, retries). */
717
718 else
719 {
720 rule = final_rule;
721 retry_record->expired = TRUE;
722 if (addr == endaddr) timedout_count++;
723 }
724
725 /* There is a special case to consider when some messages get through
726 to a destination and others don't. This can happen locally when a
727 large message pushes a user over quota, and it can happen remotely
728 when a machine is on a dodgy Internet connection. The messages that
729 get through wipe the retry information, causing those that don't to
730 stay on the queue longer than the final retry time. In order to
731 avoid this, we check, using the time of arrival of the message, to
732 see if it has been on the queue for more than the final cutoff time,
733 and if so, cause this retry item to time out, and the retry time to
734 be set to "now" so that any subsequent messages in the same condition
735 also get tried. We search for the last rule onwards from the one that
736 is in use. If there are no retry rules for the item, rule will be null
737 and timedout_count will already have been updated.
738
739 This implements "timeout this rule if EITHER the host (or routing or
740 directing) has been failing for more than the maximum time, OR if the
741 message has been on the queue for more than the maximum time." */
742
743 if (received_time <= retry_record->first_failed &&
744 addr == endaddr && !retry_record->expired && rule != NULL)
745 {
746 retry_rule *last_rule;
747 for (last_rule = rule;
748 last_rule->next != NULL;
749 last_rule = last_rule->next);
750 if (now - received_time > last_rule->timeout)
751 {
752 DEBUG(D_retry) debug_printf("on queue longer than maximum retry\n");
753 timedout_count++;
754 rule = NULL;
755 }
756 }
757
758 /* Compute the next try time from the rule, subject to the global
759 maximum, and update the retry database. If rule == NULL it means
760 there were no rules at all (and the timeout will be set expired),
761 or we have a message that is older than the final timeout. In this
762 case set the next retry time to now, so that one delivery attempt
763 happens for subsequent messages. */
764
765 if (rule == NULL) next_try = now; else
766 {
767 if (rule->rule == 'F') next_try = now + rule->p1;
768 else /* rule = 'G' or 'H' */
769 {
770 int last_predicted_gap =
771 retry_record->next_try - retry_record->last_try;
772 int last_actual_gap = now - retry_record->last_try;
773 int lastgap = (last_predicted_gap < last_actual_gap)?
774 last_predicted_gap : last_actual_gap;
775 int next_gap = (lastgap * rule->p2)/1000;
776 if (rule->rule == 'G')
777 {
778 next_try = now + ((lastgap < rule->p1)? rule->p1 : next_gap);
779 }
780 else /* The 'H' rule */
781 {
782 next_try = now + rule->p1;
783 if (next_gap > rule->p1)
784 next_try += random_number(next_gap - rule->p1);
785 }
786 }
787 }
788
789 /* Impose a global retry max */
790
791 if (next_try - now > retry_interval_max)
792 next_try = now + retry_interval_max;
793
794 /* If the new message length is greater than the previous one, we
795 have to copy the record first. */
796
797 if (message_length > message_space)
798 {
799 dbdata_retry *newr = store_get(sizeof(dbdata_retry) + message_length);
800 memcpy(newr, retry_record, sizeof(dbdata_retry));
801 retry_record = newr;
802 }
803
804 /* Set up the retry record; message_length may be less than the string
805 length for very long error strings. */
806
807 retry_record->last_try = now;
808 retry_record->next_try = next_try;
809 retry_record->basic_errno = rti->basic_errno;
810 retry_record->more_errno = rti->more_errno;
811 Ustrncpy(retry_record->text, message, message_length);
812 retry_record->text[message_length] = 0;
813
814 DEBUG(D_retry)
815 {
816 int letter = retry_record->more_errno & 255;
817 debug_printf("Writing retry data for %s\n", rti->key);
818 debug_printf(" first failed=%d last try=%d next try=%d expired=%d\n",
819 (int)retry_record->first_failed, (int)retry_record->last_try,
820 (int)retry_record->next_try, retry_record->expired);
821 debug_printf(" errno=%d more_errno=", retry_record->basic_errno);
822 if (letter == 'A' || letter == 'M')
823 debug_printf("%d,%c", (retry_record->more_errno >> 8) & 255,
824 letter);
825 else
826 debug_printf("%d", retry_record->more_errno);
827 debug_printf(" %s\n", retry_record->text);
828 }
829
830 (void)dbfn_write(dbm_file, rti->key, retry_record,
831 sizeof(dbdata_retry) + message_length);
832 } /* Loop for each retry item */
833
834 /* If all the non-delete retry items are timed out, the address is
835 timed out, provided that we didn't skip any hosts because their retry
836 time was not reached (or because of hosts_max_try). */
837
838 if (update_count > 0 && update_count == timedout_count)
839 {
840 if (!testflag(endaddr, af_retry_skipped))
841 {
842 DEBUG(D_retry) debug_printf("timed out: all retries expired\n");
843 timed_out = TRUE;
844 }
845 else
846 {
847 DEBUG(D_retry)
848 debug_printf("timed out but some hosts were skipped\n");
849 }
850 }
851 } /* Loop for an address and its parents */
852
853 /* If this is a deferred address, and retry processing was requested by
854 means of one or more retry items, and they all timed out, move the address
855 to the failed queue, and restart this loop without updating paddr.
856
857 If there were several addresses batched in the same remote delivery, only
858 the original top one will have host retry items attached to it, but we want
859 to handle all the same. Each will have a pointer back to its "top" address,
860 and they will now precede the item with the retries because addresses are
861 inverted when added to these final queues. We have saved information about
862 them in passing (below) so they can all be cut out at once. */
863
864 if (i == 2) /* Handling defers */
865 {
866 if (endaddr->retries != NULL && timed_out)
867 {
868 if (last_first == endaddr) paddr = saved_paddr;
869 addr = *paddr;
870 *paddr = endaddr->next;
871
872 endaddr->next = *addr_failed;
873 *addr_failed = addr;
874
875 for (;; addr = addr->next)
876 {
877 setflag(addr, af_retry_timedout);
878 addr->message = (addr->message == NULL)? US"retry timeout exceeded" :
879 string_sprintf("%s: retry timeout exceeded", addr->message);
880 addr->user_message = (addr->user_message == NULL)?
881 US"retry timeout exceeded" :
882 string_sprintf("%s: retry timeout exceeded", addr->user_message);
883 log_write(0, LOG_MAIN, "** %s%s%s%s: retry timeout exceeded",
884 addr->address,
885 (addr->parent == NULL)? US"" : US" <",
886 (addr->parent == NULL)? US"" : addr->parent->address,
887 (addr->parent == NULL)? US"" : US">");
888
889 if (addr == endaddr) break;
890 }
891
892 continue; /* Restart from changed *paddr */
893 }
894
895 /* This address is to remain on the defer chain. If it has a "first"
896 pointer, save the pointer to it in case we want to fail the set of
897 addresses when we get to the first one. */
898
899 if (endaddr->first != last_first)
900 {
901 last_first = endaddr->first;
902 saved_paddr = paddr;
903 }
904 }
905
906 /* All cases (succeed, fail, defer left on queue) */
907
908 paddr = &(endaddr->next); /* Advance to next address */
909 } /* Loop for all addresses */
910 } /* Loop for succeed, fail, defer */
911
912 /* Close and unlock the database */
913
914 if (dbm_file != NULL) dbfn_close(dbm_file);
915
916 DEBUG(D_retry) debug_printf("end of retry processing\n");
917 }
918
919 /* End of retry.c */