dc39813fba731687cfc301e7b62ae8fbf577d213
[exim.git] / src / src / retry.c
1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
4
5 /* Copyright (c) University of Cambridge 1995 - 2018 */
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 retry_config * retry;
37
38 DEBUG(D_retry)
39 {
40 debug_printf("retry time not reached: checking ultimate address timeout\n");
41 debug_printf(" now=" TIME_T_FMT " first_failed=" TIME_T_FMT
42 " next_try=" TIME_T_FMT " expired=%c\n",
43 now, retry_record->first_failed,
44 retry_record->next_try, retry_record->expired ? 'T' : 'F');
45 }
46
47 retry = retry_find_config(retry_key+2, domain,
48 retry_record->basic_errno, retry_record->more_errno);
49
50 if (retry && retry->rules)
51 {
52 retry_rule *last_rule;
53 for (last_rule = retry->rules; last_rule->next; last_rule = last_rule->next) ;
54 DEBUG(D_retry)
55 debug_printf(" received_time=" TIME_T_FMT " diff=%d timeout=%d\n",
56 received_time.tv_sec, (int)(now - received_time.tv_sec), last_rule->timeout);
57 address_timeout = (now - received_time.tv_sec > last_rule->timeout);
58 }
59 else
60 {
61 DEBUG(D_retry)
62 debug_printf("no retry rule found: assume timed out\n");
63 address_timeout = TRUE;
64 }
65
66 DEBUG(D_retry)
67 if (address_timeout)
68 debug_printf("on queue longer than maximum retry for address - "
69 "allowing delivery\n");
70
71 return address_timeout;
72 }
73
74
75
76 /*************************************************
77 * Set status of a host+address item *
78 *************************************************/
79
80 /* This function is passed a host_item which contains a host name and an
81 IP address string. Its job is to set the status of the address if it is not
82 already set (indicated by hstatus_unknown). The possible values are:
83
84 hstatus_usable the address is not listed in the unusable tree, and does
85 not have a retry record, OR the time is past the next
86 try time, OR the message has been on the queue for more
87 than the maximum retry time for a failing host
88
89 hstatus_unusable the address is listed in the unusable tree, or does have
90 a retry record, and the time is not yet at the next retry
91 time.
92
93 hstatus_unusable_expired as above, but also the retry time has expired
94 for this address.
95
96 The reason a delivery is permitted when a message has been around for a very
97 long time is to allow the ultimate address timeout to operate after a delivery
98 failure. Otherwise some messages may stick around without being tried for too
99 long.
100
101 If a host retry record is retrieved from the hints database, the time of last
102 trying is filled into the last_try field of the host block. If a host is
103 generally usable, a check is made to see if there is a retry delay on this
104 specific message at this host.
105
106 If a non-standard port is being used, it is added to the retry key.
107
108 Arguments:
109 domain the address domain
110 host pointer to a host item
111 portstring "" for standard port, ":xxxx" for a non-standard port
112 include_ip_address TRUE to include the address in the key - this is
113 usual, but sometimes is not wanted
114 retry_host_key where to put a pointer to the key for the host-specific
115 retry record, if one is read and the host is usable
116 retry_message_key where to put a pointer to the key for the message+host
117 retry record, if one is read and the host is usable
118
119 Returns: TRUE if the host has expired but is usable because
120 its retry time has come
121 */
122
123 BOOL
124 retry_check_address(const uschar *domain, host_item *host, uschar *portstring,
125 BOOL include_ip_address, uschar **retry_host_key, uschar **retry_message_key)
126 {
127 BOOL yield = FALSE;
128 time_t now = time(NULL);
129 uschar *host_key, *message_key;
130 open_db dbblock;
131 open_db *dbm_file;
132 tree_node *node;
133 dbdata_retry *host_retry_record, *message_retry_record;
134
135 *retry_host_key = *retry_message_key = NULL;
136
137 DEBUG(D_transport|D_retry) debug_printf("checking status of %s\n", host->name);
138
139 /* Do nothing if status already set; otherwise initialize status as usable. */
140
141 if (host->status != hstatus_unknown) return FALSE;
142 host->status = hstatus_usable;
143
144 /* Generate the host key for the unusable tree and the retry database. Ensure
145 host names are lower cased (that's what %S does). */
146
147 host_key = include_ip_address?
148 string_sprintf("T:%S:%s%s", host->name, host->address, portstring) :
149 string_sprintf("T:%S%s", host->name, portstring);
150
151 /* Generate the message-specific key */
152
153 message_key = string_sprintf("%s:%s", host_key, message_id);
154
155 /* Search the tree of unusable IP addresses. This is filled in when deliveries
156 fail, because the retry database itself is not updated until the end of all
157 deliveries (so as to do it all in one go). The tree records addresses that have
158 become unusable during this delivery process (i.e. those that will get put into
159 the retry database when it is updated). */
160
161 if ((node = tree_search(tree_unusable, host_key)))
162 {
163 DEBUG(D_transport|D_retry) debug_printf("found in tree of unusables\n");
164 host->status = (node->data.val > 255)?
165 hstatus_unusable_expired : hstatus_unusable;
166 host->why = node->data.val & 255;
167 return FALSE;
168 }
169
170 /* Open the retry database, giving up if there isn't one. Otherwise, search for
171 the retry records, and then close the database again. */
172
173 if (!(dbm_file = dbfn_open(US"retry", O_RDONLY, &dbblock, FALSE)))
174 {
175 DEBUG(D_deliver|D_retry|D_hints_lookup)
176 debug_printf("no retry data available\n");
177 return FALSE;
178 }
179 host_retry_record = dbfn_read(dbm_file, host_key);
180 message_retry_record = dbfn_read(dbm_file, message_key);
181 dbfn_close(dbm_file);
182
183 /* Ignore the data if it is too old - too long since it was written */
184
185 if (!host_retry_record)
186 {
187 DEBUG(D_transport|D_retry) debug_printf("no host retry record\n");
188 }
189 else if (now - host_retry_record->time_stamp > retry_data_expire)
190 {
191 host_retry_record = NULL;
192 DEBUG(D_transport|D_retry) debug_printf("host retry record too old\n");
193 }
194
195 if (!message_retry_record)
196 {
197 DEBUG(D_transport|D_retry) debug_printf("no message retry record\n");
198 }
199 else if (now - message_retry_record->time_stamp > retry_data_expire)
200 {
201 message_retry_record = NULL;
202 DEBUG(D_transport|D_retry) debug_printf("message retry record too old\n");
203 }
204
205 /* If there's a host-specific retry record, check for reaching the retry
206 time (or forcing). If not, and the host is not expired, check for the message
207 having been around for longer than the maximum retry time for this host or
208 address. Allow the delivery if it has. Otherwise set the appropriate unusable
209 flag and return FALSE. Otherwise arrange to return TRUE if this is an expired
210 host. */
211
212 if (host_retry_record)
213 {
214 *retry_host_key = host_key;
215
216 /* We have not reached the next try time. Check for the ultimate address
217 timeout if the host has not expired. */
218
219 if (now < host_retry_record->next_try && !f.deliver_force)
220 {
221 if (!host_retry_record->expired &&
222 retry_ultimate_address_timeout(host_key, domain,
223 host_retry_record, now))
224 return FALSE;
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)
245 {
246 *retry_message_key = message_key;
247 if (now < message_retry_record->next_try && !f.deliver_force)
248 {
249 if (!retry_ultimate_address_timeout(host_key, domain,
250 message_retry_record, now))
251 {
252 host->status = hstatus_unusable;
253 host->why = hwhy_retry;
254 }
255 return FALSE;
256 }
257 }
258
259 return yield;
260 }
261
262
263
264
265 /*************************************************
266 * Add a retry item to an address *
267 *************************************************/
268
269 /* Retry items are chained onto an address when it is deferred either by router
270 or by a transport, or if it succeeds or fails and there was a previous retry
271 item that now needs to be deleted. Sometimes there can be both kinds of item:
272 for example, if routing was deferred but then succeeded, and delivery then
273 deferred. In that case there is a delete item for the routing retry, and an
274 updating item for the delivery.
275
276 (But note that that is only visible at the outer level, because in remote
277 delivery subprocesses, the address starts "clean", with no retry items carried
278 in.)
279
280 These items are used at the end of a delivery attempt to update the retry
281 database. The keys start R: for routing delays and T: for transport delays.
282
283 Arguments:
284 addr the address block onto which to hang the item
285 key the retry key
286 flags delete, host, and message flags, copied into the block
287
288 Returns: nothing
289 */
290
291 void
292 retry_add_item(address_item *addr, uschar *key, int flags)
293 {
294 retry_item *rti = store_get(sizeof(retry_item));
295 host_item * host = addr->host_used;
296
297 rti->next = addr->retries;
298 addr->retries = rti;
299 rti->key = key;
300 rti->basic_errno = addr->basic_errno;
301 rti->more_errno = addr->more_errno;
302 rti->message = host
303 ? string_sprintf("H=%s [%s]: %s", host->name, host->address, addr->message)
304 : addr->message;
305 rti->flags = flags;
306
307 DEBUG(D_transport|D_retry)
308 {
309 int letter = rti->more_errno & 255;
310 debug_printf("added retry item for %s: errno=%d more_errno=", rti->key,
311 rti->basic_errno);
312 if (letter == 'A' || letter == 'M')
313 debug_printf("%d,%c", (rti->more_errno >> 8) & 255, letter);
314 else
315 debug_printf("%d", rti->more_errno);
316 debug_printf(" flags=%d\n", flags);
317 }
318 }
319
320
321
322 /*************************************************
323 * Find retry configuration data *
324 *************************************************/
325
326 /* Search the in-store retry information for the first retry item that applies
327 to a given destination. If the key contains an @ we are probably handling a
328 local delivery and have a complete address to search for; this happens when
329 retry_use_local_part is set on a router. Otherwise, the key is likely to be a
330 host name for a remote delivery, or a domain name for a local delivery. We
331 prepend *@ on the front of it so that it will match a retry item whose address
332 item pattern is independent of the local part. The alternate key, if set, is
333 always just a domain, so we treat it likewise.
334
335 Arguments:
336 key key for which retry info is wanted
337 alternate alternative key, always just a domain
338 basic_errno specific error predicate on the retry rule, or zero
339 more_errno additional data for errno predicate
340
341 Returns: pointer to retry rule, or NULL
342 */
343
344 retry_config *
345 retry_find_config(const uschar *key, const uschar *alternate, int basic_errno,
346 int more_errno)
347 {
348 const uschar *colon = Ustrchr(key, ':');
349 retry_config *yield;
350
351 /* If there's a colon in the key, there are two possibilities:
352
353 (1) This is a key for a host, ip address, and possibly port, in the format
354
355 hostname:ip+port
356
357 In this case, we copy the host name.
358
359 (2) This is a key for a pipe, file, or autoreply delivery, in the format
360
361 pipe-or-file-or-auto:x@y
362
363 where x@y is the original address that provoked the delivery. The pipe or
364 file or auto will start with | or / or >, whereas a host name will start
365 with a letter or a digit. In this case we want to use the original address
366 to search for a retry rule. */
367
368 if (colon)
369 key = isalnum(*key)
370 ? string_copyn(key, colon-key) /* the hostname */
371 : Ustrrchr(key, ':') + 1; /* Take from the last colon */
372
373 /* Sort out the keys */
374
375 if (!Ustrchr(key, '@')) key = string_sprintf("*@%s", key);
376 if (alternate) alternate = string_sprintf("*@%s", alternate);
377
378 /* Scan the configured retry items. */
379
380 for (yield = retries; yield; yield = yield->next)
381 {
382 const uschar *plist = yield->pattern;
383 const uschar *slist = yield->senders;
384
385 /* If a specific error is set for this item, check that we are handling that
386 specific error, and if so, check any additional error information if
387 required. */
388
389 if (yield->basic_errno != 0)
390 {
391 /* Special code is required for quota errors, as these can either be system
392 quota errors, or Exim's own quota imposition, which has a different error
393 number. Full partitions are also treated in the same way as quota errors.
394 */
395
396 if (yield->basic_errno == ERRNO_EXIMQUOTA)
397 {
398 if ((basic_errno != ERRNO_EXIMQUOTA && basic_errno != errno_quota &&
399 basic_errno != ENOSPC) ||
400 (yield->more_errno != 0 && yield->more_errno > more_errno))
401 continue;
402 }
403
404 /* The TLSREQUIRED error also covers TLSFAILURE. These are subtly different
405 errors, but not worth separating at this level. */
406
407 else if (yield->basic_errno == ERRNO_TLSREQUIRED)
408 {
409 if (basic_errno != ERRNO_TLSREQUIRED && basic_errno != ERRNO_TLSFAILURE)
410 continue;
411 }
412
413 /* Handle 4xx responses to MAIL, RCPT, or DATA. The code that was received
414 is in the 2nd least significant byte of more_errno (with 400 subtracted).
415 The required value is coded in the 2nd least significant byte of the
416 yield->more_errno field as follows:
417
418 255 => any 4xx code
419 >= 100 => the decade must match the value less 100
420 < 100 => the exact value must match
421 */
422
423 else if (yield->basic_errno == ERRNO_MAIL4XX ||
424 yield->basic_errno == ERRNO_RCPT4XX ||
425 yield->basic_errno == ERRNO_DATA4XX)
426 {
427 int wanted;
428 if (basic_errno != yield->basic_errno) continue;
429 wanted = (yield->more_errno >> 8) & 255;
430 if (wanted != 255)
431 {
432 int evalue = (more_errno >> 8) & 255;
433 if (wanted >= 100)
434 {
435 if ((evalue/10)*10 != wanted - 100) continue;
436 }
437 else if (evalue != wanted) continue;
438 }
439 }
440
441 /* There are some special cases for timeouts */
442
443 else if (yield->basic_errno == ETIMEDOUT)
444 {
445 if (basic_errno != ETIMEDOUT) continue;
446
447 /* Just RTEF_CTOUT in the rule => don't care about 'A'/'M' addresses */
448 if (yield->more_errno == RTEF_CTOUT)
449 {
450 if ((more_errno & RTEF_CTOUT) == 0) continue;
451 }
452
453 else if (yield->more_errno != 0)
454 {
455 int cf_errno = more_errno;
456 if ((yield->more_errno & RTEF_CTOUT) == 0) cf_errno &= ~RTEF_CTOUT;
457 if (yield->more_errno != cf_errno) continue;
458 }
459 }
460
461 /* Default checks for exact match */
462
463 else
464 {
465 if (yield->basic_errno != basic_errno ||
466 (yield->more_errno != 0 && yield->more_errno != more_errno))
467 continue;
468 }
469 }
470
471 /* If the "senders" condition is set, check it. Note that sender_address may
472 be null during -brt checking, in which case we do not use this rule. */
473
474 if ( slist
475 && ( !sender_address
476 || match_address_list_basic(sender_address, &slist, 0) != OK
477 ) )
478 continue;
479
480 /* Check for a match between the address list item at the start of this retry
481 rule and either the main or alternate keys. */
482
483 if ( match_address_list_basic(key, &plist, UCHAR_MAX+1) == OK
484 || ( alternate
485 && match_address_list_basic(alternate, &plist, UCHAR_MAX+1) == OK
486 ) )
487 break;
488 }
489
490 return yield;
491 }
492
493
494
495
496 /*************************************************
497 * Update retry database *
498 *************************************************/
499
500 /* Update the retry data for any directing/routing/transporting that was
501 deferred, or delete it for those that succeeded after a previous defer. This is
502 done all in one go to minimize opening/closing/locking of the database file.
503
504 Note that, because SMTP delivery involves a list of destinations to try, there
505 may be defer-type retry information for some of them even when the message was
506 successfully delivered. Likewise if it eventually failed.
507
508 This function may move addresses from the defer to the failed queue if the
509 ultimate retry time has expired.
510
511 Arguments:
512 addr_defer queue of deferred addresses
513 addr_failed queue of failed addresses
514 addr_succeed queue of successful addresses
515
516 Returns: nothing
517 */
518
519 void
520 retry_update(address_item **addr_defer, address_item **addr_failed,
521 address_item **addr_succeed)
522 {
523 open_db dbblock;
524 open_db *dbm_file = NULL;
525 time_t now = time(NULL);
526
527 DEBUG(D_retry) debug_printf("Processing retry items\n");
528
529 /* Three-times loop to handle succeeded, failed, and deferred addresses.
530 Deferred addresses must be handled after failed ones, because some may be moved
531 to the failed chain if they have timed out. */
532
533 for (int i = 0; i < 3; i++)
534 {
535 address_item *endaddr, *addr;
536 address_item *last_first = NULL;
537 address_item **paddr = i==0 ? addr_succeed :
538 i==1 ? addr_failed : addr_defer;
539 address_item **saved_paddr = NULL;
540
541 DEBUG(D_retry) debug_printf("%s addresses:\n",
542 i == 0 ? "Succeeded" : i == 1 ? "Failed" : "Deferred");
543
544 /* Loop for each address on the chain. For deferred addresses, the whole
545 address times out unless one of its retry addresses has a retry rule that
546 hasn't yet timed out. Deferred addresses should not be requesting deletion
547 of retry items, but just in case they do by accident, treat that case
548 as "not timed out".
549
550 As well as handling the addresses themselves, we must also process any
551 retry items for any parent addresses - these are typically "delete" items,
552 because the parent must have succeeded in order to generate the child. */
553
554 while ((endaddr = *paddr))
555 {
556 BOOL timed_out = FALSE;
557
558 for (addr = endaddr; addr; addr = addr->parent)
559 {
560 int update_count = 0;
561 int timedout_count = 0;
562
563 DEBUG(D_retry) debug_printf(" %s%s\n", addr->address,
564 addr->retries ? "" : ": no retry items");
565
566 /* Loop for each retry item. */
567
568 for (retry_item * rti = addr->retries; rti; rti = rti->next)
569 {
570 uschar *message;
571 int message_length, message_space, failing_interval, next_try;
572 retry_rule *rule, *final_rule;
573 retry_config *retry;
574 dbdata_retry *retry_record;
575
576 /* Open the retry database if it is not already open; failure to open
577 the file is logged, but otherwise ignored - deferred addresses will
578 get retried at the next opportunity. Not opening earlier than this saves
579 opening if no addresses have retry items - common when none have yet
580 reached their retry next try time. */
581
582 if (!dbm_file)
583 dbm_file = dbfn_open(US"retry", O_RDWR, &dbblock, TRUE);
584
585 if (!dbm_file)
586 {
587 DEBUG(D_deliver|D_retry|D_hints_lookup)
588 debug_printf("retry database not available for updating\n");
589 return;
590 }
591
592 /* If there are no deferred addresses, that is, if this message is
593 completing, and the retry item is for a message-specific SMTP error,
594 force it to be deleted, because there's no point in keeping data for
595 no-longer-existing messages. This situation can occur when a domain has
596 two hosts and a message-specific error occurs for the first of them,
597 but the address gets delivered to the second one. This optimization
598 doesn't succeed in cleaning out all the dead entries, but it helps. */
599
600 if (!*addr_defer && rti->flags & rf_message)
601 rti->flags |= rf_delete;
602
603 /* Handle the case of a request to delete the retry info for this
604 destination. */
605
606 if (rti->flags & rf_delete)
607 {
608 (void)dbfn_delete(dbm_file, rti->key);
609 DEBUG(D_retry)
610 debug_printf("deleted retry information for %s\n", rti->key);
611 continue;
612 }
613
614 /* Count the number of non-delete retry items. This is so that we
615 can compare it to the count of timed_out ones, to check whether
616 all are timed out. */
617
618 update_count++;
619
620 /* Get the retry information for this destination and error code, if
621 any. If this item is for a remote host with ip address, then pass
622 the domain name as an alternative to search for. If no retry
623 information is found, we can't generate a retry time, so there is
624 no point updating the database. This retry item is timed out. */
625
626 if (!(retry = retry_find_config(rti->key + 2,
627 rti->flags & rf_host ? addr->domain : NULL,
628 rti->basic_errno, rti->more_errno)))
629 {
630 DEBUG(D_retry) debug_printf("No configured retry item for %s%s%s\n",
631 rti->key,
632 rti->flags & rf_host ? US" or " : US"",
633 rti->flags & rf_host ? addr->domain : US"");
634 if (addr == endaddr) timedout_count++;
635 continue;
636 }
637
638 DEBUG(D_retry)
639 if (rti->flags & rf_host)
640 debug_printf("retry for %s (%s) = %s %d %d\n", rti->key,
641 addr->domain, retry->pattern, retry->basic_errno,
642 retry->more_errno);
643 else
644 debug_printf("retry for %s = %s %d %d\n", rti->key, retry->pattern,
645 retry->basic_errno, retry->more_errno);
646
647 /* Set up the message for the database retry record. Because DBM
648 records have a maximum data length, we enforce a limit. There isn't
649 much point in keeping a huge message here, anyway. */
650
651 message = rti->basic_errno > 0
652 ? US strerror(rti->basic_errno)
653 : rti->message
654 ? US string_printing(rti->message)
655 : US"unknown error";
656 message_length = Ustrlen(message);
657 if (message_length > 150) message_length = 150;
658
659 /* Read a retry record from the database or construct a new one.
660 Ignore an old one if it is too old since it was last updated. */
661
662 retry_record = dbfn_read(dbm_file, rti->key);
663 if ( retry_record
664 && now - retry_record->time_stamp > retry_data_expire)
665 retry_record = NULL;
666
667 if (!retry_record)
668 {
669 retry_record = store_get(sizeof(dbdata_retry) + message_length);
670 message_space = message_length;
671 retry_record->first_failed = now;
672 retry_record->last_try = now;
673 retry_record->next_try = now;
674 retry_record->expired = FALSE;
675 retry_record->text[0] = 0; /* just in case */
676 }
677 else message_space = Ustrlen(retry_record->text);
678
679 /* Compute how long this destination has been failing */
680
681 failing_interval = now - retry_record->first_failed;
682 DEBUG(D_retry) debug_printf("failing_interval=%d message_age=%d\n",
683 failing_interval, message_age);
684
685 /* For a non-host error, if the message has been on the queue longer
686 than the recorded time of failure, use the message's age instead. This
687 can happen when some messages can be delivered and others cannot; a
688 successful delivery will reset the first_failed time, and this can lead
689 to a failing message being retried too often. */
690
691 if (!(rti->flags & rf_host) && message_age > failing_interval)
692 failing_interval = message_age;
693
694 /* Search for the current retry rule. The cutoff time of the
695 last rule is handled differently to the others. The rule continues
696 to operate for ever (the global maximum interval will eventually
697 limit the gaps) but its cutoff time determines when an individual
698 destination times out. If there are no retry rules, the destination
699 always times out, but we can't compute a retry time. */
700
701 final_rule = NULL;
702 for (rule = retry->rules; rule; rule = rule->next)
703 {
704 if (failing_interval <= rule->timeout) break;
705 final_rule = rule;
706 }
707
708 /* If there's an un-timed out rule, the destination has not
709 yet timed out, so the address as a whole has not timed out (but we are
710 interested in this only for the end address). Make sure the expired
711 flag is false (can be forced via fixdb from outside, but ensure it is
712 consistent with the rules whenever we go through here). */
713
714 if (rule)
715 retry_record->expired = FALSE;
716
717 /* Otherwise, set the retry timeout expired, and set the final rule
718 as the one from which to compute the next retry time. Subsequent
719 messages will fail immediately until the retry time is reached (unless
720 there are other, still active, retries). */
721
722 else
723 {
724 rule = final_rule;
725 retry_record->expired = TRUE;
726 if (addr == endaddr) timedout_count++;
727 }
728
729 /* There is a special case to consider when some messages get through
730 to a destination and others don't. This can happen locally when a
731 large message pushes a user over quota, and it can happen remotely
732 when a machine is on a dodgy Internet connection. The messages that
733 get through wipe the retry information, causing those that don't to
734 stay on the queue longer than the final retry time. In order to
735 avoid this, we check, using the time of arrival of the message, to
736 see if it has been on the queue for more than the final cutoff time,
737 and if so, cause this retry item to time out, and the retry time to
738 be set to "now" so that any subsequent messages in the same condition
739 also get tried. We search for the last rule onwards from the one that
740 is in use. If there are no retry rules for the item, rule will be null
741 and timedout_count will already have been updated.
742
743 This implements "timeout this rule if EITHER the host (or routing or
744 directing) has been failing for more than the maximum time, OR if the
745 message has been on the queue for more than the maximum time."
746
747 February 2006: It is possible that this code is no longer needed
748 following the change to the retry calculation to use the message age if
749 it is larger than the time since first failure. It may be that the
750 expired flag is always set when the other conditions are met. However,
751 this is a small bit of code, and it does no harm to leave it in place,
752 just in case. */
753
754 if ( received_time.tv_sec <= retry_record->first_failed
755 && addr == endaddr
756 && !retry_record->expired
757 && rule)
758 {
759 retry_rule *last_rule;
760 for (last_rule = rule; last_rule->next; last_rule = last_rule->next)
761 ;
762 if (now - received_time.tv_sec > last_rule->timeout)
763 {
764 DEBUG(D_retry) debug_printf("on queue longer than maximum retry\n");
765 timedout_count++;
766 rule = NULL;
767 }
768 }
769
770 /* Compute the next try time from the rule, subject to the global
771 maximum, and update the retry database. If rule == NULL it means
772 there were no rules at all (and the timeout will be set expired),
773 or we have a message that is older than the final timeout. In this
774 case set the next retry time to now, so that one delivery attempt
775 happens for subsequent messages. */
776
777 if (!rule)
778 next_try = now;
779 else
780 {
781 if (rule->rule == 'F')
782 next_try = now + rule->p1;
783 else /* rule = 'G' or 'H' */
784 {
785 int last_predicted_gap =
786 retry_record->next_try - retry_record->last_try;
787 int last_actual_gap = now - retry_record->last_try;
788 int lastgap = (last_predicted_gap < last_actual_gap)?
789 last_predicted_gap : last_actual_gap;
790 int next_gap = (lastgap * rule->p2)/1000;
791 if (rule->rule == 'G')
792 next_try = now + ((lastgap < rule->p1)? rule->p1 : next_gap);
793 else /* The 'H' rule */
794 {
795 next_try = now + rule->p1;
796 if (next_gap > rule->p1)
797 next_try += random_number(next_gap - rule->p1)/2 +
798 (next_gap - rule->p1)/2;
799 }
800 }
801 }
802
803 /* Impose a global retry max */
804
805 if (next_try - now > retry_interval_max)
806 next_try = now + retry_interval_max;
807
808 /* If the new message length is greater than the previous one, we
809 have to copy the record first. */
810
811 if (message_length > message_space)
812 {
813 dbdata_retry *newr = store_get(sizeof(dbdata_retry) + message_length);
814 memcpy(newr, retry_record, sizeof(dbdata_retry));
815 retry_record = newr;
816 }
817
818 /* Set up the retry record; message_length may be less than the string
819 length for very long error strings. */
820
821 retry_record->last_try = now;
822 retry_record->next_try = next_try;
823 retry_record->basic_errno = rti->basic_errno;
824 retry_record->more_errno = rti->more_errno;
825 Ustrncpy(retry_record->text, message, message_length);
826 retry_record->text[message_length] = 0;
827
828 DEBUG(D_retry)
829 {
830 int letter = retry_record->more_errno & 255;
831 debug_printf("Writing retry data for %s\n", rti->key);
832 debug_printf(" first failed=%d last try=%d next try=%d expired=%d\n",
833 (int)retry_record->first_failed, (int)retry_record->last_try,
834 (int)retry_record->next_try, retry_record->expired);
835 debug_printf(" errno=%d more_errno=", retry_record->basic_errno);
836 if (letter == 'A' || letter == 'M')
837 debug_printf("%d,%c", (retry_record->more_errno >> 8) & 255,
838 letter);
839 else
840 debug_printf("%d", retry_record->more_errno);
841 debug_printf(" %s\n", retry_record->text);
842 }
843
844 (void)dbfn_write(dbm_file, rti->key, retry_record,
845 sizeof(dbdata_retry) + message_length);
846 } /* Loop for each retry item */
847
848 /* If all the non-delete retry items are timed out, the address is
849 timed out, provided that we didn't skip any hosts because their retry
850 time was not reached (or because of hosts_max_try). */
851
852 if (update_count > 0 && update_count == timedout_count)
853 if (!testflag(endaddr, af_retry_skipped))
854 {
855 DEBUG(D_retry) debug_printf("timed out: all retries expired\n");
856 timed_out = TRUE;
857 }
858 else
859 DEBUG(D_retry)
860 debug_printf("timed out but some hosts were skipped\n");
861 } /* Loop for an address and its parents */
862
863 /* If this is a deferred address, and retry processing was requested by
864 means of one or more retry items, and they all timed out, move the address
865 to the failed queue, and restart this loop without updating paddr.
866
867 If there were several addresses batched in the same remote delivery, only
868 the original top one will have host retry items attached to it, but we want
869 to handle all the same. Each will have a pointer back to its "top" address,
870 and they will now precede the item with the retries because addresses are
871 inverted when added to these final queues. We have saved information about
872 them in passing (below) so they can all be cut out at once. */
873
874 if (i == 2) /* Handling defers */
875 {
876 if (endaddr->retries && timed_out)
877 {
878 if (last_first == endaddr) paddr = saved_paddr;
879 addr = *paddr;
880 *paddr = endaddr->next;
881
882 endaddr->next = *addr_failed;
883 *addr_failed = addr;
884
885 for (;; addr = addr->next)
886 {
887 setflag(addr, af_retry_timedout);
888 addr->message = addr->message
889 ? string_sprintf("%s: retry timeout exceeded", addr->message)
890 : US"retry timeout exceeded";
891 addr->user_message = addr->user_message
892 ? string_sprintf("%s: retry timeout exceeded", addr->user_message)
893 : US"retry timeout exceeded";
894 log_write(0, LOG_MAIN, "** %s%s%s%s: retry timeout exceeded",
895 addr->address,
896 addr->parent ? US" <" : US"",
897 addr->parent ? addr->parent->address : US"",
898 addr->parent ? US">" : US"");
899
900 if (addr == endaddr) break;
901 }
902
903 continue; /* Restart from changed *paddr */
904 }
905
906 /* This address is to remain on the defer chain. If it has a "first"
907 pointer, save the pointer to it in case we want to fail the set of
908 addresses when we get to the first one. */
909
910 if (endaddr->first != last_first)
911 {
912 last_first = endaddr->first;
913 saved_paddr = paddr;
914 }
915 }
916
917 /* All cases (succeed, fail, defer left on queue) */
918
919 paddr = &(endaddr->next); /* Advance to next address */
920 } /* Loop for all addresses */
921 } /* Loop for succeed, fail, defer */
922
923 /* Close and unlock the database */
924
925 if (dbm_file) dbfn_close(dbm_file);
926
927 DEBUG(D_retry) debug_printf("end of retry processing\n");
928 }
929
930 /* End of retry.c */