tidying
[exim.git] / src / src / retry.c
1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
4
5 /* Copyright (c) University of Cambridge 1995 - 2015 */
6 /* See the file NOTICE for conditions of use and distribution. */
7
8 /* Functions concerned with retrying unsuccessful deliveries. */
9
10
11 #include "exim.h"
12
13
14
15 /*************************************************
16 * Check the ultimate address timeout *
17 *************************************************/
18
19 /* This function tests whether a message has been on the queue longer than
20 the maximum retry time for a particular host or address.
21
22 Arguments:
23 retry_key the key to look up a retry rule
24 domain the domain to look up a domain retry rule
25 retry_record contains error information for finding rule
26 now the time
27
28 Returns: TRUE if the ultimate timeout has been reached
29 */
30
31 BOOL
32 retry_ultimate_address_timeout(uschar *retry_key, const uschar *domain,
33 dbdata_retry *retry_record, time_t now)
34 {
35 BOOL address_timeout;
36
37 DEBUG(D_retry)
38 {
39 debug_printf("retry time not reached: checking ultimate address timeout\n");
40 debug_printf(" now=%d first_failed=%d next_try=%d expired=%d\n",
41 (int)now, (int)retry_record->first_failed,
42 (int)retry_record->next_try, retry_record->expired);
43 }
44
45 retry_config *retry =
46 retry_find_config(retry_key+2, domain,
47 retry_record->basic_errno, retry_record->more_errno);
48
49 if (retry != NULL && retry->rules != NULL)
50 {
51 retry_rule *last_rule;
52 for (last_rule = retry->rules;
53 last_rule->next != NULL;
54 last_rule = last_rule->next);
55 DEBUG(D_retry)
56 debug_printf(" received_time=%d diff=%d timeout=%d\n",
57 (int)received_time.tv_sec, (int)(now - received_time.tv_sec), last_rule->timeout);
58 address_timeout = (now - received_time.tv_sec > last_rule->timeout);
59 }
60 else
61 {
62 DEBUG(D_retry)
63 debug_printf("no retry rule found: assume timed out\n");
64 address_timeout = TRUE;
65 }
66
67 DEBUG(D_retry)
68 if (address_timeout)
69 debug_printf("on queue longer than maximum retry for address - "
70 "allowing delivery\n");
71
72 return address_timeout;
73 }
74
75
76
77 /*************************************************
78 * Set status of a host+address item *
79 *************************************************/
80
81 /* This function is passed a host_item which contains a host name and an
82 IP address string. Its job is to set the status of the address if it is not
83 already set (indicated by hstatus_unknown). The possible values are:
84
85 hstatus_usable the address is not listed in the unusable tree, and does
86 not have a retry record, OR the time is past the next
87 try time, OR the message has been on the queue for more
88 than the maximum retry time for a failing host
89
90 hstatus_unusable the address is listed in the unusable tree, or does have
91 a retry record, and the time is not yet at the next retry
92 time.
93
94 hstatus_unusable_expired as above, but also the retry time has expired
95 for this address.
96
97 The reason a delivery is permitted when a message has been around for a very
98 long time is to allow the ultimate address timeout to operate after a delivery
99 failure. Otherwise some messages may stick around without being tried for too
100 long.
101
102 If a host retry record is retrieved from the hints database, the time of last
103 trying is filled into the last_try field of the host block. If a host is
104 generally usable, a check is made to see if there is a retry delay on this
105 specific message at this host.
106
107 If a non-standard port is being used, it is added to the retry key.
108
109 Arguments:
110 domain the address domain
111 host pointer to a host item
112 portstring "" for standard port, ":xxxx" for a non-standard port
113 include_ip_address TRUE to include the address in the key - this is
114 usual, but sometimes is not wanted
115 retry_host_key where to put a pointer to the key for the host-specific
116 retry record, if one is read and the host is usable
117 retry_message_key where to put a pointer to the key for the message+host
118 retry record, if one is read and the host is usable
119
120 Returns: TRUE if the host has expired but is usable because
121 its retry time has come
122 */
123
124 BOOL
125 retry_check_address(const uschar *domain, host_item *host, uschar *portstring,
126 BOOL include_ip_address, uschar **retry_host_key, uschar **retry_message_key)
127 {
128 BOOL yield = FALSE;
129 time_t now = time(NULL);
130 uschar *host_key, *message_key;
131 open_db dbblock;
132 open_db *dbm_file;
133 tree_node *node;
134 dbdata_retry *host_retry_record, *message_retry_record;
135
136 *retry_host_key = *retry_message_key = NULL;
137
138 DEBUG(D_transport|D_retry) debug_printf("checking status of %s\n", host->name);
139
140 /* Do nothing if status already set; otherwise initialize status as usable. */
141
142 if (host->status != hstatus_unknown) return FALSE;
143 host->status = hstatus_usable;
144
145 /* Generate the host key for the unusable tree and the retry database. Ensure
146 host names are lower cased (that's what %S does). */
147
148 host_key = include_ip_address?
149 string_sprintf("T:%S:%s%s", host->name, host->address, portstring) :
150 string_sprintf("T:%S%s", host->name, portstring);
151
152 /* Generate the message-specific key */
153
154 message_key = string_sprintf("%s:%s", host_key, message_id);
155
156 /* Search the tree of unusable IP addresses. This is filled in when deliveries
157 fail, because the retry database itself is not updated until the end of all
158 deliveries (so as to do it all in one go). The tree records addresses that have
159 become unusable during this delivery process (i.e. those that will get put into
160 the retry database when it is updated). */
161
162 if ((node = tree_search(tree_unusable, host_key)))
163 {
164 DEBUG(D_transport|D_retry) debug_printf("found in tree of unusables\n");
165 host->status = (node->data.val > 255)?
166 hstatus_unusable_expired : hstatus_unusable;
167 host->why = node->data.val & 255;
168 return FALSE;
169 }
170
171 /* Open the retry database, giving up if there isn't one. Otherwise, search for
172 the retry records, and then close the database again. */
173
174 if (!(dbm_file = dbfn_open(US"retry", O_RDONLY, &dbblock, FALSE)))
175 {
176 DEBUG(D_deliver|D_retry|D_hints_lookup)
177 debug_printf("no retry data available\n");
178 return FALSE;
179 }
180 host_retry_record = dbfn_read(dbm_file, host_key);
181 message_retry_record = dbfn_read(dbm_file, message_key);
182 dbfn_close(dbm_file);
183
184 /* Ignore the data if it is too old - too long since it was written */
185
186 if (!host_retry_record)
187 {
188 DEBUG(D_transport|D_retry) debug_printf("no host retry record\n");
189 }
190 else if (now - host_retry_record->time_stamp > retry_data_expire)
191 {
192 host_retry_record = NULL;
193 DEBUG(D_transport|D_retry) debug_printf("host retry record too old\n");
194 }
195
196 if (!message_retry_record)
197 {
198 DEBUG(D_transport|D_retry) debug_printf("no message retry record\n");
199 }
200 else if (now - message_retry_record->time_stamp > retry_data_expire)
201 {
202 message_retry_record = NULL;
203 DEBUG(D_transport|D_retry) debug_printf("message retry record too old\n");
204 }
205
206 /* If there's a host-specific retry record, check for reaching the retry
207 time (or forcing). If not, and the host is not expired, check for the message
208 having been around for longer than the maximum retry time for this host or
209 address. Allow the delivery if it has. Otherwise set the appropriate unusable
210 flag and return FALSE. Otherwise arrange to return TRUE if this is an expired
211 host. */
212
213 if (host_retry_record)
214 {
215 *retry_host_key = host_key;
216
217 /* We have not reached the next try time. Check for the ultimate address
218 timeout if the host has not expired. */
219
220 if (now < host_retry_record->next_try && !deliver_force)
221 {
222 if (!host_retry_record->expired &&
223 retry_ultimate_address_timeout(host_key, domain,
224 host_retry_record, now))
225 return FALSE;
226
227 /* We have not hit the ultimate address timeout; host is unusable. */
228
229 host->status = (host_retry_record->expired)?
230 hstatus_unusable_expired : hstatus_unusable;
231 host->why = hwhy_retry;
232 host->last_try = host_retry_record->last_try;
233 return FALSE;
234 }
235
236 /* Host is usable; set return TRUE if expired. */
237
238 yield = host_retry_record->expired;
239 }
240
241 /* It's OK to try the host. If there's a message-specific retry record, check
242 for reaching its retry time (or forcing). If not, mark the host unusable,
243 unless the ultimate address timeout has been reached. */
244
245 if (message_retry_record)
246 {
247 *retry_message_key = message_key;
248 if (now < message_retry_record->next_try && !deliver_force)
249 {
250 if (!retry_ultimate_address_timeout(host_key, domain,
251 message_retry_record, now))
252 {
253 host->status = hstatus_unusable;
254 host->why = hwhy_retry;
255 }
256 return FALSE;
257 }
258 }
259
260 return yield;
261 }
262
263
264
265
266 /*************************************************
267 * Add a retry item to an address *
268 *************************************************/
269
270 /* Retry items are chained onto an address when it is deferred either by router
271 or by a transport, or if it succeeds or fails and there was a previous retry
272 item that now needs to be deleted. Sometimes there can be both kinds of item:
273 for example, if routing was deferred but then succeeded, and delivery then
274 deferred. In that case there is a delete item for the routing retry, and an
275 updating item for the delivery.
276
277 (But note that that is only visible at the outer level, because in remote
278 delivery subprocesses, the address starts "clean", with no retry items carried
279 in.)
280
281 These items are used at the end of a delivery attempt to update the retry
282 database. The keys start R: for routing delays and T: for transport delays.
283
284 Arguments:
285 addr the address block onto which to hang the item
286 key the retry key
287 flags delete, host, and message flags, copied into the block
288
289 Returns: nothing
290 */
291
292 void
293 retry_add_item(address_item *addr, uschar *key, int flags)
294 {
295 retry_item *rti = store_get(sizeof(retry_item));
296 host_item * host = addr->host_used;
297
298 rti->next = addr->retries;
299 addr->retries = rti;
300 rti->key = key;
301 rti->basic_errno = addr->basic_errno;
302 rti->more_errno = addr->more_errno;
303 rti->message = host
304 ? string_sprintf("H=%s [%s]: %s", host->name, host->address, addr->message)
305 : addr->message;
306 rti->flags = flags;
307
308 DEBUG(D_transport|D_retry)
309 {
310 int letter = rti->more_errno & 255;
311 debug_printf("added retry item for %s: errno=%d more_errno=", rti->key,
312 rti->basic_errno);
313 if (letter == 'A' || letter == 'M')
314 debug_printf("%d,%c", (rti->more_errno >> 8) & 255, letter);
315 else
316 debug_printf("%d", rti->more_errno);
317 debug_printf(" flags=%d\n", flags);
318 }
319 }
320
321
322
323 /*************************************************
324 * Find retry configuration data *
325 *************************************************/
326
327 /* Search the in-store retry information for the first retry item that applies
328 to a given destination. If the key contains an @ we are probably handling a
329 local delivery and have a complete address to search for; this happens when
330 retry_use_local_part is set on a router. Otherwise, the key is likely to be a
331 host name for a remote delivery, or a domain name for a local delivery. We
332 prepend *@ on the front of it so that it will match a retry item whose address
333 item pattern is independent of the local part. The alternate key, if set, is
334 always just a domain, so we treat it likewise.
335
336 Arguments:
337 key key for which retry info is wanted
338 alternate alternative key, always just a domain
339 basic_errno specific error predicate on the retry rule, or zero
340 more_errno additional data for errno predicate
341
342 Returns: pointer to retry rule, or NULL
343 */
344
345 retry_config *
346 retry_find_config(const uschar *key, const uschar *alternate, int basic_errno,
347 int more_errno)
348 {
349 const uschar *colon = Ustrchr(key, ':');
350 retry_config *yield;
351
352 /* If there's a colon in the key, there are two possibilities:
353
354 (1) This is a key for a host, ip address, and possibly port, in the format
355
356 hostname:ip+port
357
358 In this case, we copy the host name.
359
360 (2) This is a key for a pipe, file, or autoreply delivery, in the format
361
362 pipe-or-file-or-auto:x@y
363
364 where x@y is the original address that provoked the delivery. The pipe or
365 file or auto will start with | or / or >, whereas a host name will start
366 with a letter or a digit. In this case we want to use the original address
367 to search for a retry rule. */
368
369 if (colon)
370 key = isalnum(*key)
371 ? string_copyn(key, colon-key) /* the hostname */
372 : Ustrrchr(key, ':') + 1; /* Take from the last colon */
373
374 /* Sort out the keys */
375
376 if (!Ustrchr(key, '@')) key = string_sprintf("*@%s", key);
377 if (alternate) alternate = string_sprintf("*@%s", alternate);
378
379 /* Scan the configured retry items. */
380
381 for (yield = retries; yield; yield = yield->next)
382 {
383 const uschar *plist = yield->pattern;
384 const uschar *slist = yield->senders;
385
386 /* If a specific error is set for this item, check that we are handling that
387 specific error, and if so, check any additional error information if
388 required. */
389
390 if (yield->basic_errno != 0)
391 {
392 /* Special code is required for quota errors, as these can either be system
393 quota errors, or Exim's own quota imposition, which has a different error
394 number. Full partitions are also treated in the same way as quota errors.
395 */
396
397 if (yield->basic_errno == ERRNO_EXIMQUOTA)
398 {
399 if ((basic_errno != ERRNO_EXIMQUOTA && basic_errno != errno_quota &&
400 basic_errno != ENOSPC) ||
401 (yield->more_errno != 0 && yield->more_errno > more_errno))
402 continue;
403 }
404
405 /* The TLSREQUIRED error also covers TLSFAILURE. These are subtly different
406 errors, but not worth separating at this level. */
407
408 else if (yield->basic_errno == ERRNO_TLSREQUIRED)
409 {
410 if (basic_errno != ERRNO_TLSREQUIRED && basic_errno != ERRNO_TLSFAILURE)
411 continue;
412 }
413
414 /* Handle 4xx responses to MAIL, RCPT, or DATA. The code that was received
415 is in the 2nd least significant byte of more_errno (with 400 subtracted).
416 The required value is coded in the 2nd least significant byte of the
417 yield->more_errno field as follows:
418
419 255 => any 4xx code
420 >= 100 => the decade must match the value less 100
421 < 100 => the exact value must match
422 */
423
424 else if (yield->basic_errno == ERRNO_MAIL4XX ||
425 yield->basic_errno == ERRNO_RCPT4XX ||
426 yield->basic_errno == ERRNO_DATA4XX)
427 {
428 int wanted;
429 if (basic_errno != yield->basic_errno) continue;
430 wanted = (yield->more_errno >> 8) & 255;
431 if (wanted != 255)
432 {
433 int evalue = (more_errno >> 8) & 255;
434 if (wanted >= 100)
435 {
436 if ((evalue/10)*10 != wanted - 100) continue;
437 }
438 else if (evalue != wanted) continue;
439 }
440 }
441
442 /* There are some special cases for timeouts */
443
444 else if (yield->basic_errno == ETIMEDOUT)
445 {
446 if (basic_errno != ETIMEDOUT) continue;
447
448 /* Just RTEF_CTOUT in the rule => don't care about 'A'/'M' addresses */
449 if (yield->more_errno == RTEF_CTOUT)
450 {
451 if ((more_errno & RTEF_CTOUT) == 0) continue;
452 }
453
454 else if (yield->more_errno != 0)
455 {
456 int cf_errno = more_errno;
457 if ((yield->more_errno & RTEF_CTOUT) == 0) cf_errno &= ~RTEF_CTOUT;
458 if (yield->more_errno != cf_errno) continue;
459 }
460 }
461
462 /* Default checks for exact match */
463
464 else
465 {
466 if (yield->basic_errno != basic_errno ||
467 (yield->more_errno != 0 && yield->more_errno != more_errno))
468 continue;
469 }
470 }
471
472 /* If the "senders" condition is set, check it. Note that sender_address may
473 be null during -brt checking, in which case we do not use this rule. */
474
475 if ( slist
476 && ( !sender_address
477 || match_address_list_basic(sender_address, &slist, 0) != OK
478 ) )
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_basic(key, &plist, UCHAR_MAX+1) == OK
485 || ( alternate
486 && match_address_list_basic(alternate, &plist, UCHAR_MAX+1) == OK
487 ) )
488 break;
489 }
490
491 return yield;
492 }
493
494
495
496
497 /*************************************************
498 * Update retry database *
499 *************************************************/
500
501 /* Update the retry data for any directing/routing/transporting that was
502 deferred, or delete it for those that succeeded after a previous defer. This is
503 done all in one go to minimize opening/closing/locking of the database file.
504
505 Note that, because SMTP delivery involves a list of destinations to try, there
506 may be defer-type retry information for some of them even when the message was
507 successfully delivered. Likewise if it eventually failed.
508
509 This function may move addresses from the defer to the failed queue if the
510 ultimate retry time has expired.
511
512 Arguments:
513 addr_defer queue of deferred addresses
514 addr_failed queue of failed addresses
515 addr_succeed queue of successful addresses
516
517 Returns: nothing
518 */
519
520 void
521 retry_update(address_item **addr_defer, address_item **addr_failed,
522 address_item **addr_succeed)
523 {
524 open_db dbblock;
525 open_db *dbm_file = NULL;
526 time_t now = time(NULL);
527 int i;
528
529 DEBUG(D_retry) debug_printf("Processing retry items\n");
530
531 /* Three-times loop to handle succeeded, failed, and deferred addresses.
532 Deferred addresses must be handled after failed ones, because some may be moved
533 to the failed chain if they have timed out. */
534
535 for (i = 0; i < 3; i++)
536 {
537 address_item *endaddr, *addr;
538 address_item *last_first = NULL;
539 address_item **paddr = i==0 ? addr_succeed :
540 i==1 ? addr_failed : addr_defer;
541 address_item **saved_paddr = NULL;
542
543 DEBUG(D_retry) debug_printf("%s addresses:\n",
544 i == 0 ? "Succeeded" : i == 1 ? "Failed" : "Deferred");
545
546 /* Loop for each address on the chain. For deferred addresses, the whole
547 address times out unless one of its retry addresses has a retry rule that
548 hasn't yet timed out. Deferred addresses should not be requesting deletion
549 of retry items, but just in case they do by accident, treat that case
550 as "not timed out".
551
552 As well as handling the addresses themselves, we must also process any
553 retry items for any parent addresses - these are typically "delete" items,
554 because the parent must have succeeded in order to generate the child. */
555
556 while ((endaddr = *paddr))
557 {
558 BOOL timed_out = FALSE;
559 retry_item *rti;
560
561 for (addr = endaddr; addr; addr = addr->parent)
562 {
563 int update_count = 0;
564 int timedout_count = 0;
565
566 DEBUG(D_retry) debug_printf(" %s%s\n", addr->address,
567 addr->retries ? "" : ": no retry items");
568
569 /* Loop for each retry item. */
570
571 for (rti = addr->retries; rti; rti = rti->next)
572 {
573 uschar *message;
574 int message_length, message_space, failing_interval, next_try;
575 retry_rule *rule, *final_rule;
576 retry_config *retry;
577 dbdata_retry *retry_record;
578
579 /* Open the retry database if it is not already open; failure to open
580 the file is logged, but otherwise ignored - deferred addresses will
581 get retried at the next opportunity. Not opening earlier than this saves
582 opening if no addresses have retry items - common when none have yet
583 reached their retry next try time. */
584
585 if (!dbm_file)
586 dbm_file = dbfn_open(US"retry", O_RDWR, &dbblock, TRUE);
587
588 if (!dbm_file)
589 {
590 DEBUG(D_deliver|D_retry|D_hints_lookup)
591 debug_printf("retry database not available for updating\n");
592 return;
593 }
594
595 /* If there are no deferred addresses, that is, if this message is
596 completing, and the retry item is for a message-specific SMTP error,
597 force it to be deleted, because there's no point in keeping data for
598 no-longer-existing messages. This situation can occur when a domain has
599 two hosts and a message-specific error occurs for the first of them,
600 but the address gets delivered to the second one. This optimization
601 doesn't succeed in cleaning out all the dead entries, but it helps. */
602
603 if (!*addr_defer && rti->flags & rf_message)
604 rti->flags |= rf_delete;
605
606 /* Handle the case of a request to delete the retry info for this
607 destination. */
608
609 if (rti->flags & rf_delete)
610 {
611 (void)dbfn_delete(dbm_file, rti->key);
612 DEBUG(D_retry)
613 debug_printf("deleted retry information for %s\n", rti->key);
614 continue;
615 }
616
617 /* Count the number of non-delete retry items. This is so that we
618 can compare it to the count of timed_out ones, to check whether
619 all are timed out. */
620
621 update_count++;
622
623 /* Get the retry information for this destination and error code, if
624 any. If this item is for a remote host with ip address, then pass
625 the domain name as an alternative to search for. If no retry
626 information is found, we can't generate a retry time, so there is
627 no point updating the database. This retry item is timed out. */
628
629 if (!(retry = retry_find_config(rti->key + 2,
630 rti->flags & rf_host ? addr->domain : NULL,
631 rti->basic_errno, rti->more_errno)))
632 {
633 DEBUG(D_retry) debug_printf("No configured retry item for %s%s%s\n",
634 rti->key,
635 rti->flags & rf_host ? US" or " : US"",
636 rti->flags & rf_host ? addr->domain : US"");
637 if (addr == endaddr) timedout_count++;
638 continue;
639 }
640
641 DEBUG(D_retry)
642 if (rti->flags & rf_host)
643 debug_printf("retry for %s (%s) = %s %d %d\n", rti->key,
644 addr->domain, retry->pattern, retry->basic_errno,
645 retry->more_errno);
646 else
647 debug_printf("retry for %s = %s %d %d\n", rti->key, retry->pattern,
648 retry->basic_errno, retry->more_errno);
649
650 /* Set up the message for the database retry record. Because DBM
651 records have a maximum data length, we enforce a limit. There isn't
652 much point in keeping a huge message here, anyway. */
653
654 message = rti->basic_errno > 0
655 ? US strerror(rti->basic_errno)
656 : rti->message
657 ? US string_printing(rti->message)
658 : US"unknown error";
659 message_length = Ustrlen(message);
660 if (message_length > 150) message_length = 150;
661
662 /* Read a retry record from the database or construct a new one.
663 Ignore an old one if it is too old since it was last updated. */
664
665 retry_record = dbfn_read(dbm_file, rti->key);
666 if ( retry_record
667 && now - retry_record->time_stamp > retry_data_expire)
668 retry_record = NULL;
669
670 if (!retry_record)
671 {
672 retry_record = store_get(sizeof(dbdata_retry) + message_length);
673 message_space = message_length;
674 retry_record->first_failed = now;
675 retry_record->last_try = now;
676 retry_record->next_try = now;
677 retry_record->expired = FALSE;
678 retry_record->text[0] = 0; /* just in case */
679 }
680 else message_space = Ustrlen(retry_record->text);
681
682 /* Compute how long this destination has been failing */
683
684 failing_interval = now - retry_record->first_failed;
685 DEBUG(D_retry) debug_printf("failing_interval=%d message_age=%d\n",
686 failing_interval, message_age);
687
688 /* For a non-host error, if the message has been on the queue longer
689 than the recorded time of failure, use the message's age instead. This
690 can happen when some messages can be delivered and others cannot; a
691 successful delivery will reset the first_failed time, and this can lead
692 to a failing message being retried too often. */
693
694 if (!(rti->flags & rf_host) && message_age > failing_interval)
695 failing_interval = message_age;
696
697 /* Search for the current retry rule. The cutoff time of the
698 last rule is handled differently to the others. The rule continues
699 to operate for ever (the global maximum interval will eventually
700 limit the gaps) but its cutoff time determines when an individual
701 destination times out. If there are no retry rules, the destination
702 always times out, but we can't compute a retry time. */
703
704 final_rule = NULL;
705 for (rule = retry->rules; rule; rule = rule->next)
706 {
707 if (failing_interval <= rule->timeout) break;
708 final_rule = rule;
709 }
710
711 /* If there's an un-timed out rule, the destination has not
712 yet timed out, so the address as a whole has not timed out (but we are
713 interested in this only for the end address). Make sure the expired
714 flag is false (can be forced via fixdb from outside, but ensure it is
715 consistent with the rules whenever we go through here). */
716
717 if (rule)
718 retry_record->expired = FALSE;
719
720 /* Otherwise, set the retry timeout expired, and set the final rule
721 as the one from which to compute the next retry time. Subsequent
722 messages will fail immediately until the retry time is reached (unless
723 there are other, still active, retries). */
724
725 else
726 {
727 rule = final_rule;
728 retry_record->expired = TRUE;
729 if (addr == endaddr) timedout_count++;
730 }
731
732 /* There is a special case to consider when some messages get through
733 to a destination and others don't. This can happen locally when a
734 large message pushes a user over quota, and it can happen remotely
735 when a machine is on a dodgy Internet connection. The messages that
736 get through wipe the retry information, causing those that don't to
737 stay on the queue longer than the final retry time. In order to
738 avoid this, we check, using the time of arrival of the message, to
739 see if it has been on the queue for more than the final cutoff time,
740 and if so, cause this retry item to time out, and the retry time to
741 be set to "now" so that any subsequent messages in the same condition
742 also get tried. We search for the last rule onwards from the one that
743 is in use. If there are no retry rules for the item, rule will be null
744 and timedout_count will already have been updated.
745
746 This implements "timeout this rule if EITHER the host (or routing or
747 directing) has been failing for more than the maximum time, OR if the
748 message has been on the queue for more than the maximum time."
749
750 February 2006: It is possible that this code is no longer needed
751 following the change to the retry calculation to use the message age if
752 it is larger than the time since first failure. It may be that the
753 expired flag is always set when the other conditions are met. However,
754 this is a small bit of code, and it does no harm to leave it in place,
755 just in case. */
756
757 if ( received_time.tv_sec <= retry_record->first_failed
758 && addr == endaddr
759 && !retry_record->expired
760 && rule)
761 {
762 retry_rule *last_rule;
763 for (last_rule = rule; last_rule->next; last_rule = last_rule->next)
764 ;
765 if (now - received_time.tv_sec > last_rule->timeout)
766 {
767 DEBUG(D_retry) debug_printf("on queue longer than maximum retry\n");
768 timedout_count++;
769 rule = NULL;
770 }
771 }
772
773 /* Compute the next try time from the rule, subject to the global
774 maximum, and update the retry database. If rule == NULL it means
775 there were no rules at all (and the timeout will be set expired),
776 or we have a message that is older than the final timeout. In this
777 case set the next retry time to now, so that one delivery attempt
778 happens for subsequent messages. */
779
780 if (!rule)
781 next_try = now;
782 else
783 {
784 if (rule->rule == 'F')
785 next_try = now + rule->p1;
786 else /* rule = 'G' or 'H' */
787 {
788 int last_predicted_gap =
789 retry_record->next_try - retry_record->last_try;
790 int last_actual_gap = now - retry_record->last_try;
791 int lastgap = (last_predicted_gap < last_actual_gap)?
792 last_predicted_gap : last_actual_gap;
793 int next_gap = (lastgap * rule->p2)/1000;
794 if (rule->rule == 'G')
795 next_try = now + ((lastgap < rule->p1)? rule->p1 : next_gap);
796 else /* The 'H' rule */
797 {
798 next_try = now + rule->p1;
799 if (next_gap > rule->p1)
800 next_try += random_number(next_gap - rule->p1)/2 +
801 (next_gap - rule->p1)/2;
802 }
803 }
804 }
805
806 /* Impose a global retry max */
807
808 if (next_try - now > retry_interval_max)
809 next_try = now + retry_interval_max;
810
811 /* If the new message length is greater than the previous one, we
812 have to copy the record first. */
813
814 if (message_length > message_space)
815 {
816 dbdata_retry *newr = store_get(sizeof(dbdata_retry) + message_length);
817 memcpy(newr, retry_record, sizeof(dbdata_retry));
818 retry_record = newr;
819 }
820
821 /* Set up the retry record; message_length may be less than the string
822 length for very long error strings. */
823
824 retry_record->last_try = now;
825 retry_record->next_try = next_try;
826 retry_record->basic_errno = rti->basic_errno;
827 retry_record->more_errno = rti->more_errno;
828 Ustrncpy(retry_record->text, message, message_length);
829 retry_record->text[message_length] = 0;
830
831 DEBUG(D_retry)
832 {
833 int letter = retry_record->more_errno & 255;
834 debug_printf("Writing retry data for %s\n", rti->key);
835 debug_printf(" first failed=%d last try=%d next try=%d expired=%d\n",
836 (int)retry_record->first_failed, (int)retry_record->last_try,
837 (int)retry_record->next_try, retry_record->expired);
838 debug_printf(" errno=%d more_errno=", retry_record->basic_errno);
839 if (letter == 'A' || letter == 'M')
840 debug_printf("%d,%c", (retry_record->more_errno >> 8) & 255,
841 letter);
842 else
843 debug_printf("%d", retry_record->more_errno);
844 debug_printf(" %s\n", retry_record->text);
845 }
846
847 (void)dbfn_write(dbm_file, rti->key, retry_record,
848 sizeof(dbdata_retry) + message_length);
849 } /* Loop for each retry item */
850
851 /* If all the non-delete retry items are timed out, the address is
852 timed out, provided that we didn't skip any hosts because their retry
853 time was not reached (or because of hosts_max_try). */
854
855 if (update_count > 0 && update_count == timedout_count)
856 if (!testflag(endaddr, af_retry_skipped))
857 {
858 DEBUG(D_retry) debug_printf("timed out: all retries expired\n");
859 timed_out = TRUE;
860 }
861 else
862 DEBUG(D_retry)
863 debug_printf("timed out but some hosts were skipped\n");
864 } /* Loop for an address and its parents */
865
866 /* If this is a deferred address, and retry processing was requested by
867 means of one or more retry items, and they all timed out, move the address
868 to the failed queue, and restart this loop without updating paddr.
869
870 If there were several addresses batched in the same remote delivery, only
871 the original top one will have host retry items attached to it, but we want
872 to handle all the same. Each will have a pointer back to its "top" address,
873 and they will now precede the item with the retries because addresses are
874 inverted when added to these final queues. We have saved information about
875 them in passing (below) so they can all be cut out at once. */
876
877 if (i == 2) /* Handling defers */
878 {
879 if (endaddr->retries && timed_out)
880 {
881 if (last_first == endaddr) paddr = saved_paddr;
882 addr = *paddr;
883 *paddr = endaddr->next;
884
885 endaddr->next = *addr_failed;
886 *addr_failed = addr;
887
888 for (;; addr = addr->next)
889 {
890 setflag(addr, af_retry_timedout);
891 addr->message = (addr->message == NULL)? US"retry timeout exceeded" :
892 string_sprintf("%s: retry timeout exceeded", addr->message);
893 addr->user_message = (addr->user_message == NULL)?
894 US"retry timeout exceeded" :
895 string_sprintf("%s: retry timeout exceeded", addr->user_message);
896 log_write(0, LOG_MAIN, "** %s%s%s%s: retry timeout exceeded",
897 addr->address,
898 (addr->parent == NULL)? US"" : US" <",
899 (addr->parent == NULL)? US"" : addr->parent->address,
900 (addr->parent == NULL)? US"" : US">");
901
902 if (addr == endaddr) break;
903 }
904
905 continue; /* Restart from changed *paddr */
906 }
907
908 /* This address is to remain on the defer chain. If it has a "first"
909 pointer, save the pointer to it in case we want to fail the set of
910 addresses when we get to the first one. */
911
912 if (endaddr->first != last_first)
913 {
914 last_first = endaddr->first;
915 saved_paddr = paddr;
916 }
917 }
918
919 /* All cases (succeed, fail, defer left on queue) */
920
921 paddr = &(endaddr->next); /* Advance to next address */
922 } /* Loop for all addresses */
923 } /* Loop for succeed, fail, defer */
924
925 /* Close and unlock the database */
926
927 if (dbm_file) dbfn_close(dbm_file);
928
929 DEBUG(D_retry) debug_printf("end of retry processing\n");
930 }
931
932 /* End of retry.c */