constification
[exim.git] / src / src / retry.c
CommitLineData
059ec3d9
PH
1/*************************************************
2* Exim - an Internet mail transport agent *
3*************************************************/
4
0a49a7a4 5/* Copyright (c) University of Cambridge 1995 - 2009 */
059ec3d9
PH
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
ba9af0af 20the maximum retry time for a particular host or address.
059ec3d9
PH
21
22Arguments:
ba9af0af 23 retry_key the key to look up a retry rule
059ec3d9 24 domain the domain to look up a domain retry rule
ba9af0af 25 retry_record contains error information for finding rule
059ec3d9
PH
26 now the time
27
28Returns: TRUE if the ultimate timeout has been reached
29*/
30
ba9af0af 31BOOL
55414b25 32retry_ultimate_address_timeout(uschar *retry_key, const uschar *domain,
ba9af0af 33 dbdata_retry *retry_record, time_t now)
059ec3d9 34{
ba9af0af
TF
35BOOL address_timeout;
36
37DEBUG(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 }
059ec3d9
PH
44
45retry_config *retry =
ba9af0af
TF
46 retry_find_config(retry_key+2, domain,
47 retry_record->basic_errno, retry_record->more_errno);
059ec3d9
PH
48
49if (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);
ba9af0af 55 DEBUG(D_retry)
c816d124
PH
56 debug_printf(" received_time=%d diff=%d timeout=%d\n",
57 received_time, (int)(now - received_time), last_rule->timeout);
059ec3d9
PH
58 address_timeout = (now - received_time > last_rule->timeout);
59 }
ea49d0e1
PH
60else
61 {
ba9af0af 62 DEBUG(D_retry)
ea49d0e1 63 debug_printf("no retry rule found: assume timed out\n");
ba9af0af 64 address_timeout = TRUE;
ea49d0e1 65 }
059ec3d9 66
ba9af0af
TF
67DEBUG(D_retry)
68 if (address_timeout)
69 debug_printf("on queue longer than maximum retry for address - "
70 "allowing delivery\n");
71
059ec3d9
PH
72return 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
82IP address string. Its job is to set the status of the address if it is not
83already 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
97The reason a delivery is permitted when a message has been around for a very
98long time is to allow the ultimate address timeout to operate after a delivery
99failure. Otherwise some messages may stick around without being tried for too
100long.
101
102If a host retry record is retrieved from the hints database, the time of last
103trying is filled into the last_try field of the host block. If a host is
104generally usable, a check is made to see if there is a retry delay on this
105specific message at this host.
106
107If a non-standard port is being used, it is added to the retry key.
108
109Arguments:
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
120Returns: TRUE if the host has expired but is usable because
121 its retry time has come
122*/
123
124BOOL
55414b25 125retry_check_address(const uschar *domain, host_item *host, uschar *portstring,
059ec3d9
PH
126 BOOL include_ip_address, uschar **retry_host_key, uschar **retry_message_key)
127{
128BOOL yield = FALSE;
129time_t now = time(NULL);
130uschar *host_key, *message_key;
131open_db dbblock;
132open_db *dbm_file;
133tree_node *node;
134dbdata_retry *host_retry_record, *message_retry_record;
135
136*retry_host_key = *retry_message_key = NULL;
137
138DEBUG(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
142if (host->status != hstatus_unknown) return FALSE;
143host->status = hstatus_usable;
144
145/* Generate the host key for the unusable tree and the retry database. Ensure
146host names are lower cased (that's what %S does). */
147
148host_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
154message_key = string_sprintf("%s:%s", host_key, message_id);
155
156/* Search the tree of unusable IP addresses. This is filled in when deliveries
157fail, because the retry database itself is not updated until the end of all
158deliveries (so as to do it all in one go). The tree records addresses that have
159become unusable during this delivery process (i.e. those that will get put into
160the retry database when it is updated). */
161
162node = tree_search(tree_unusable, host_key);
163if (node != NULL)
164 {
165 DEBUG(D_transport|D_retry) debug_printf("found in tree of unusables\n");
166 host->status = (node->data.val > 255)?
167 hstatus_unusable_expired : hstatus_unusable;
168 host->why = node->data.val & 255;
169 return FALSE;
170 }
171
172/* Open the retry database, giving up if there isn't one. Otherwise, search for
173the retry records, and then close the database again. */
174
175if ((dbm_file = dbfn_open(US"retry", O_RDONLY, &dbblock, FALSE)) == NULL)
176 {
177 DEBUG(D_deliver|D_retry|D_hints_lookup)
178 debug_printf("no retry data available\n");
179 return FALSE;
180 }
181host_retry_record = dbfn_read(dbm_file, host_key);
182message_retry_record = dbfn_read(dbm_file, message_key);
183dbfn_close(dbm_file);
184
185/* Ignore the data if it is too old - too long since it was written */
186
187if (host_retry_record == NULL)
188 {
189 DEBUG(D_transport|D_retry) debug_printf("no host retry record\n");
190 }
191else if (now - host_retry_record->time_stamp > retry_data_expire)
192 {
193 host_retry_record = NULL;
194 DEBUG(D_transport|D_retry) debug_printf("host retry record too old\n");
195 }
196
197if (message_retry_record == NULL)
198 {
199 DEBUG(D_transport|D_retry) debug_printf("no message retry record\n");
200 }
201else if (now - message_retry_record->time_stamp > retry_data_expire)
202 {
203 message_retry_record = NULL;
204 DEBUG(D_transport|D_retry) debug_printf("message retry record too old\n");
205 }
206
207/* If there's a host-specific retry record, check for reaching the retry
208time (or forcing). If not, and the host is not expired, check for the message
209having been around for longer than the maximum retry time for this host or
210address. Allow the delivery if it has. Otherwise set the appropriate unusable
211flag and return FALSE. Otherwise arrange to return TRUE if this is an expired
212host. */
213
214if (host_retry_record != NULL)
215 {
216 *retry_host_key = host_key;
217
218 /* We have not reached the next try time. Check for the ultimate address
219 timeout if the host has not expired. */
220
221 if (now < host_retry_record->next_try && !deliver_force)
222 {
059ec3d9 223 if (!host_retry_record->expired &&
ba9af0af
TF
224 retry_ultimate_address_timeout(host_key, domain,
225 host_retry_record, now))
059ec3d9 226 return FALSE;
059ec3d9
PH
227
228 /* We have not hit the ultimate address timeout; host is unusable. */
229
230 host->status = (host_retry_record->expired)?
231 hstatus_unusable_expired : hstatus_unusable;
232 host->why = hwhy_retry;
233 host->last_try = host_retry_record->last_try;
234 return FALSE;
235 }
236
237 /* Host is usable; set return TRUE if expired. */
238
239 yield = host_retry_record->expired;
240 }
241
242/* It's OK to try the host. If there's a message-specific retry record, check
243for reaching its retry time (or forcing). If not, mark the host unusable,
244unless the ultimate address timeout has been reached. */
245
246if (message_retry_record != NULL)
247 {
248 *retry_message_key = message_key;
249 if (now < message_retry_record->next_try && !deliver_force)
250 {
ba9af0af
TF
251 if (!retry_ultimate_address_timeout(host_key, domain,
252 message_retry_record, now))
059ec3d9
PH
253 {
254 host->status = hstatus_unusable;
255 host->why = hwhy_retry;
256 }
059ec3d9
PH
257 return FALSE;
258 }
259 }
260
261return yield;
262}
263
264
265
266
267/*************************************************
268* Add a retry item to an address *
269*************************************************/
270
271/* Retry items are chained onto an address when it is deferred either by router
272or by a transport, or if it succeeds or fails and there was a previous retry
273item that now needs to be deleted. Sometimes there can be both kinds of item:
274for example, if routing was deferred but then succeeded, and delivery then
275deferred. In that case there is a delete item for the routing retry, and an
276updating item for the delivery.
277
278(But note that that is only visible at the outer level, because in remote
279delivery subprocesses, the address starts "clean", with no retry items carried
280in.)
281
282These items are used at the end of a delivery attempt to update the retry
283database. The keys start R: for routing delays and T: for transport delays.
284
285Arguments:
286 addr the address block onto which to hang the item
287 key the retry key
288 flags delete, host, and message flags, copied into the block
289
290Returns: nothing
291*/
292
293void
294retry_add_item(address_item *addr, uschar *key, int flags)
295{
296retry_item *rti = store_get(sizeof(retry_item));
c562fd30 297host_item * host = addr->host_used;
059ec3d9
PH
298rti->next = addr->retries;
299addr->retries = rti;
300rti->key = key;
301rti->basic_errno = addr->basic_errno;
302rti->more_errno = addr->more_errno;
c562fd30
JH
303rti->message = host
304 ? string_sprintf("H=%s [%s]: %s", host->name, host->address, addr->message)
305 : addr->message;
059ec3d9
PH
306rti->flags = flags;
307
308DEBUG(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
328to a given destination. If the key contains an @ we are probably handling a
329local delivery and have a complete address to search for; this happens when
330retry_use_local_part is set on a router. Otherwise, the key is likely to be a
331host name for a remote delivery, or a domain name for a local delivery. We
332prepend *@ on the front of it so that it will match a retry item whose address
333item pattern is independent of the local part. The alternate key, if set, is
334always just a domain, so we treat it likewise.
335
336Arguments:
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
342Returns: pointer to retry rule, or NULL
343*/
344
345retry_config *
55414b25 346retry_find_config(const uschar *key, const uschar *alternate, int basic_errno,
059ec3d9
PH
347 int more_errno)
348{
ea49d0e1 349int replace = 0;
55414b25 350const uschar *colon = Ustrchr(key, ':');
059ec3d9
PH
351retry_config *yield;
352
ea49d0e1
PH
353/* If there's a colon in the key, there are two possibilities:
354
355(1) This is a key for a host, ip address, and possibly port, in the format
356
357 hostname:ip+port
358
55414b25 359 In this case, we copy the host name.
ea49d0e1
PH
360
361(2) This is a key for a pipe, file, or autoreply delivery, in the format
362
363 pipe-or-file-or-auto:x@y
364
365 where x@y is the original address that provoked the delivery. The pipe or
366 file or auto will start with | or / or >, whereas a host name will start
367 with a letter or a digit. In this case we want to use the original address
368 to search for a retry rule. */
059ec3d9 369
55414b25
JH
370if (colon)
371 key = isalnum(*key)
372 ? string_copyn(key, colon-key) /* the hostname */
373 : Ustrrchr(key, ':') + 1; /* Take from the last colon */
059ec3d9
PH
374
375/* Sort out the keys */
376
55414b25
JH
377if (!Ustrchr(key, '@')) key = string_sprintf("*@%s", key);
378if (alternate) alternate = string_sprintf("*@%s", alternate);
059ec3d9
PH
379
380/* Scan the configured retry items. */
381
382for (yield = retries; yield != NULL; yield = yield->next)
383 {
55414b25
JH
384 const uschar *plist = yield->pattern;
385 const uschar *slist = yield->senders;
059ec3d9
PH
386
387 /* If a specific error is set for this item, check that we are handling that
388 specific error, and if so, check any additional error information if
389 required. */
390
391 if (yield->basic_errno != 0)
392 {
393 /* Special code is required for quota errors, as these can either be system
394 quota errors, or Exim's own quota imposition, which has a different error
395 number. Full partitions are also treated in the same way as quota errors.
396 */
397
398 if (yield->basic_errno == ERRNO_EXIMQUOTA)
399 {
400 if ((basic_errno != ERRNO_EXIMQUOTA && basic_errno != errno_quota &&
401 basic_errno != ENOSPC) ||
402 (yield->more_errno != 0 && yield->more_errno > more_errno))
403 continue;
404 }
405
e97957bc
PH
406 /* The TLSREQUIRED error also covers TLSFAILURE. These are subtly different
407 errors, but not worth separating at this level. */
408
409 else if (yield->basic_errno == ERRNO_TLSREQUIRED)
410 {
411 if (basic_errno != ERRNO_TLSREQUIRED && basic_errno != ERRNO_TLSFAILURE)
412 continue;
413 }
414
415 /* Handle 4xx responses to MAIL, RCPT, or DATA. The code that was received
416 is in the 2nd least significant byte of more_errno (with 400 subtracted).
417 The required value is coded in the 2nd least significant byte of the
418 yield->more_errno field as follows:
059ec3d9
PH
419
420 255 => any 4xx code
421 >= 100 => the decade must match the value less 100
422 < 100 => the exact value must match
423 */
424
e97957bc
PH
425 else if (yield->basic_errno == ERRNO_MAIL4XX ||
426 yield->basic_errno == ERRNO_RCPT4XX ||
427 yield->basic_errno == ERRNO_DATA4XX)
059ec3d9
PH
428 {
429 int wanted;
e97957bc 430 if (basic_errno != yield->basic_errno) continue;
059ec3d9
PH
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
55414b25 484 if (match_address_list(key, TRUE, TRUE, &plist, NULL, -1, UCHAR_MAX+1,
059ec3d9 485 NULL) == OK ||
55414b25
JH
486 (alternate != NULL &&
487 match_address_list(alternate, TRUE, TRUE, &plist, NULL, -1,
059ec3d9
PH
488 UCHAR_MAX+1, NULL) == OK))
489 break;
490 }
491
059ec3d9
PH
492return yield;
493}
494
495
496
497
498/*************************************************
499* Update retry database *
500*************************************************/
501
502/* Update the retry data for any directing/routing/transporting that was
503deferred, or delete it for those that succeeded after a previous defer. This is
504done all in one go to minimize opening/closing/locking of the database file.
505
506Note that, because SMTP delivery involves a list of destinations to try, there
507may be defer-type retry information for some of them even when the message was
508successfully delivered. Likewise if it eventually failed.
509
510This function may move addresses from the defer to the failed queue if the
511ultimate retry time has expired.
512
513Arguments:
514 addr_defer queue of deferred addresses
515 addr_failed queue of failed addresses
516 addr_succeed queue of successful addresses
517
518Returns: nothing
519*/
520
521void
522retry_update(address_item **addr_defer, address_item **addr_failed,
523 address_item **addr_succeed)
524{
525open_db dbblock;
526open_db *dbm_file = NULL;
527time_t now = time(NULL);
528int i;
529
530DEBUG(D_retry) debug_printf("Processing retry items\n");
531
532/* Three-times loop to handle succeeded, failed, and deferred addresses.
533Deferred addresses must be handled after failed ones, because some may be moved
534to the failed chain if they have timed out. */
535
536for (i = 0; i < 3; i++)
537 {
538 address_item *endaddr, *addr;
539 address_item *last_first = NULL;
540 address_item **paddr = (i==0)? addr_succeed :
541 (i==1)? addr_failed : addr_defer;
542 address_item **saved_paddr = NULL;
543
544 DEBUG(D_retry) debug_printf("%s addresses:\n", (i == 0)? "Succeeded" :
545 (i == 1)? "Failed" : "Deferred");
546
547 /* Loop for each address on the chain. For deferred addresses, the whole
548 address times out unless one of its retry addresses has a retry rule that
549 hasn't yet timed out. Deferred addresses should not be requesting deletion
550 of retry items, but just in case they do by accident, treat that case
551 as "not timed out".
552
553 As well as handling the addresses themselves, we must also process any
554 retry items for any parent addresses - these are typically "delete" items,
555 because the parent must have succeeded in order to generate the child. */
556
557 while ((endaddr = *paddr) != NULL)
558 {
559 BOOL timed_out = FALSE;
560 retry_item *rti;
561
562 for (addr = endaddr; addr != NULL; addr = addr->parent)
563 {
564 int update_count = 0;
565 int timedout_count = 0;
566
567 DEBUG(D_retry) debug_printf("%s%s\n", addr->address, (addr->retries == NULL)?
568 ": no retry items" : "");
569
570 /* Loop for each retry item. */
571
572 for (rti = addr->retries; rti != NULL; rti = rti->next)
573 {
574 uschar *message;
575 int message_length, message_space, failing_interval, next_try;
576 retry_rule *rule, *final_rule;
577 retry_config *retry;
578 dbdata_retry *retry_record;
579
580 /* Open the retry database if it is not already open; failure to open
581 the file is logged, but otherwise ignored - deferred addresses will
582 get retried at the next opportunity. Not opening earlier than this saves
583 opening if no addresses have retry items - common when none have yet
584 reached their retry next try time. */
585
586 if (dbm_file == NULL)
587 dbm_file = dbfn_open(US"retry", O_RDWR, &dbblock, TRUE);
588
589 if (dbm_file == NULL)
590 {
591 DEBUG(D_deliver|D_retry|D_hints_lookup)
592 debug_printf("retry database not available for updating\n");
593 return;
594 }
595
596 /* If there are no deferred addresses, that is, if this message is
597 completing, and the retry item is for a message-specific SMTP error,
598 force it to be deleted, because there's no point in keeping data for
599 no-longer-existing messages. This situation can occur when a domain has
600 two hosts and a message-specific error occurs for the first of them,
601 but the address gets delivered to the second one. This optimization
602 doesn't succeed in cleaning out all the dead entries, but it helps. */
603
604 if (*addr_defer == NULL && (rti->flags & rf_message) != 0)
605 rti->flags |= rf_delete;
606
607 /* Handle the case of a request to delete the retry info for this
608 destination. */
609
610 if ((rti->flags & rf_delete) != 0)
611 {
612 (void)dbfn_delete(dbm_file, rti->key);
613 DEBUG(D_retry)
614 debug_printf("deleted retry information for %s\n", rti->key);
615 continue;
616 }
617
618 /* Count the number of non-delete retry items. This is so that we
619 can compare it to the count of timed_out ones, to check whether
620 all are timed out. */
621
622 update_count++;
623
624 /* Get the retry information for this destination and error code, if
625 any. If this item is for a remote host with ip address, then pass
626 the domain name as an alternative to search for. If no retry
627 information is found, we can't generate a retry time, so there is
628 no point updating the database. This retry item is timed out. */
629
630 if ((retry = retry_find_config(rti->key + 2,
631 ((rti->flags & rf_host) != 0)? addr->domain : NULL,
632 rti->basic_errno, rti->more_errno)) == NULL)
633 {
634 DEBUG(D_retry) debug_printf("No configured retry item for %s%s%s\n",
635 rti->key,
636 ((rti->flags & rf_host) != 0)? US" or " : US"",
637 ((rti->flags & rf_host) != 0)? addr->domain : US"");
638 if (addr == endaddr) timedout_count++;
639 continue;
640 }
641
642 DEBUG(D_retry)
643 {
644 if ((rti->flags & rf_host) != 0)
ea49d0e1
PH
645 debug_printf("retry for %s (%s) = %s %d %d\n", rti->key,
646 addr->domain, retry->pattern, retry->basic_errno,
647 retry->more_errno);
059ec3d9 648 else
ea49d0e1
PH
649 debug_printf("retry for %s = %s %d %d\n", rti->key, retry->pattern,
650 retry->basic_errno, retry->more_errno);
059ec3d9
PH
651 }
652
653 /* Set up the message for the database retry record. Because DBM
654 records have a maximum data length, we enforce a limit. There isn't
655 much point in keeping a huge message here, anyway. */
656
657 message = (rti->basic_errno > 0)? US strerror(rti->basic_errno) :
658 (rti->message == NULL)?
55414b25 659 US"unknown error" : US string_printing(rti->message);
059ec3d9
PH
660 message_length = Ustrlen(message);
661 if (message_length > 150) message_length = 150;
662
663 /* Read a retry record from the database or construct a new one.
664 Ignore an old one if it is too old since it was last updated. */
665
666 retry_record = dbfn_read(dbm_file, rti->key);
667 if (retry_record != NULL &&
668 now - retry_record->time_stamp > retry_data_expire)
669 retry_record = NULL;
670
671 if (retry_record == NULL)
672 {
673 retry_record = store_get(sizeof(dbdata_retry) + message_length);
674 message_space = message_length;
675 retry_record->first_failed = now;
676 retry_record->last_try = now;
677 retry_record->next_try = now;
678 retry_record->expired = FALSE;
679 retry_record->text[0] = 0; /* just in case */
680 }
681 else message_space = Ustrlen(retry_record->text);
682
683 /* Compute how long this destination has been failing */
684
685 failing_interval = now - retry_record->first_failed;
727071f8
PH
686 DEBUG(D_retry) debug_printf("failing_interval=%d message_age=%d\n",
687 failing_interval, message_age);
688
dd16e114
PH
689 /* For a non-host error, if the message has been on the queue longer
690 than the recorded time of failure, use the message's age instead. This
691 can happen when some messages can be delivered and others cannot; a
692 successful delivery will reset the first_failed time, and this can lead
693 to a failing message being retried too often. */
727071f8 694
dd16e114
PH
695 if ((rti->flags & rf_host) == 0 && message_age > failing_interval)
696 failing_interval = message_age;
059ec3d9
PH
697
698 /* Search for the current retry rule. The cutoff time of the
699 last rule is handled differently to the others. The rule continues
700 to operate for ever (the global maximum interval will eventually
701 limit the gaps) but its cutoff time determines when an individual
702 destination times out. If there are no retry rules, the destination
703 always times out, but we can't compute a retry time. */
704
705 final_rule = NULL;
706 for (rule = retry->rules; rule != NULL; rule = rule->next)
707 {
708 if (failing_interval <= rule->timeout) break;
709 final_rule = rule;
710 }
711
712 /* If there's an un-timed out rule, the destination has not
713 yet timed out, so the address as a whole has not timed out (but we are
714 interested in this only for the end address). Make sure the expired
715 flag is false (can be forced via fixdb from outside, but ensure it is
716 consistent with the rules whenever we go through here). */
717
718 if (rule != NULL)
719 {
720 retry_record->expired = FALSE;
721 }
722
723 /* Otherwise, set the retry timeout expired, and set the final rule
724 as the one from which to compute the next retry time. Subsequent
725 messages will fail immediately until the retry time is reached (unless
726 there are other, still active, retries). */
727
728 else
729 {
730 rule = final_rule;
731 retry_record->expired = TRUE;
732 if (addr == endaddr) timedout_count++;
733 }
734
735 /* There is a special case to consider when some messages get through
736 to a destination and others don't. This can happen locally when a
737 large message pushes a user over quota, and it can happen remotely
738 when a machine is on a dodgy Internet connection. The messages that
739 get through wipe the retry information, causing those that don't to
740 stay on the queue longer than the final retry time. In order to
741 avoid this, we check, using the time of arrival of the message, to
742 see if it has been on the queue for more than the final cutoff time,
743 and if so, cause this retry item to time out, and the retry time to
744 be set to "now" so that any subsequent messages in the same condition
745 also get tried. We search for the last rule onwards from the one that
746 is in use. If there are no retry rules for the item, rule will be null
747 and timedout_count will already have been updated.
748
749 This implements "timeout this rule if EITHER the host (or routing or
750 directing) has been failing for more than the maximum time, OR if the
727071f8
PH
751 message has been on the queue for more than the maximum time."
752
753 February 2006: It is possible that this code is no longer needed
754 following the change to the retry calculation to use the message age if
755 it is larger than the time since first failure. It may be that the
756 expired flag is always set when the other conditions are met. However,
757 this is a small bit of code, and it does no harm to leave it in place,
758 just in case. */
059ec3d9
PH
759
760 if (received_time <= retry_record->first_failed &&
761 addr == endaddr && !retry_record->expired && rule != NULL)
762 {
763 retry_rule *last_rule;
764 for (last_rule = rule;
765 last_rule->next != NULL;
766 last_rule = last_rule->next);
767 if (now - received_time > last_rule->timeout)
768 {
769 DEBUG(D_retry) debug_printf("on queue longer than maximum retry\n");
770 timedout_count++;
771 rule = NULL;
772 }
773 }
774
775 /* Compute the next try time from the rule, subject to the global
776 maximum, and update the retry database. If rule == NULL it means
777 there were no rules at all (and the timeout will be set expired),
778 or we have a message that is older than the final timeout. In this
779 case set the next retry time to now, so that one delivery attempt
780 happens for subsequent messages. */
781
782 if (rule == NULL) next_try = now; else
783 {
784 if (rule->rule == 'F') next_try = now + rule->p1;
6af56900 785 else /* rule = 'G' or 'H' */
059ec3d9
PH
786 {
787 int last_predicted_gap =
788 retry_record->next_try - retry_record->last_try;
789 int last_actual_gap = now - retry_record->last_try;
790 int lastgap = (last_predicted_gap < last_actual_gap)?
791 last_predicted_gap : last_actual_gap;
6af56900
PH
792 int next_gap = (lastgap * rule->p2)/1000;
793 if (rule->rule == 'G')
794 {
795 next_try = now + ((lastgap < rule->p1)? rule->p1 : next_gap);
796 }
797 else /* The 'H' rule */
798 {
799 next_try = now + rule->p1;
800 if (next_gap > rule->p1)
3cd34f13
PH
801 next_try += random_number(next_gap - rule->p1)/2 +
802 (next_gap - rule->p1)/2;
6af56900 803 }
059ec3d9
PH
804 }
805 }
806
807 /* Impose a global retry max */
808
809 if (next_try - now > retry_interval_max)
810 next_try = now + retry_interval_max;
811
812 /* If the new message length is greater than the previous one, we
813 have to copy the record first. */
814
815 if (message_length > message_space)
816 {
817 dbdata_retry *newr = store_get(sizeof(dbdata_retry) + message_length);
818 memcpy(newr, retry_record, sizeof(dbdata_retry));
819 retry_record = newr;
820 }
821
822 /* Set up the retry record; message_length may be less than the string
823 length for very long error strings. */
824
825 retry_record->last_try = now;
826 retry_record->next_try = next_try;
827 retry_record->basic_errno = rti->basic_errno;
828 retry_record->more_errno = rti->more_errno;
829 Ustrncpy(retry_record->text, message, message_length);
830 retry_record->text[message_length] = 0;
831
832 DEBUG(D_retry)
833 {
834 int letter = retry_record->more_errno & 255;
835 debug_printf("Writing retry data for %s\n", rti->key);
836 debug_printf(" first failed=%d last try=%d next try=%d expired=%d\n",
837 (int)retry_record->first_failed, (int)retry_record->last_try,
838 (int)retry_record->next_try, retry_record->expired);
839 debug_printf(" errno=%d more_errno=", retry_record->basic_errno);
840 if (letter == 'A' || letter == 'M')
841 debug_printf("%d,%c", (retry_record->more_errno >> 8) & 255,
842 letter);
843 else
844 debug_printf("%d", retry_record->more_errno);
845 debug_printf(" %s\n", retry_record->text);
846 }
847
848 (void)dbfn_write(dbm_file, rti->key, retry_record,
849 sizeof(dbdata_retry) + message_length);
850 } /* Loop for each retry item */
851
852 /* If all the non-delete retry items are timed out, the address is
853 timed out, provided that we didn't skip any hosts because their retry
854 time was not reached (or because of hosts_max_try). */
855
856 if (update_count > 0 && update_count == timedout_count)
857 {
858 if (!testflag(endaddr, af_retry_skipped))
859 {
860 DEBUG(D_retry) debug_printf("timed out: all retries expired\n");
861 timed_out = TRUE;
862 }
863 else
864 {
865 DEBUG(D_retry)
866 debug_printf("timed out but some hosts were skipped\n");
867 }
868 }
869 } /* Loop for an address and its parents */
870
871 /* If this is a deferred address, and retry processing was requested by
872 means of one or more retry items, and they all timed out, move the address
873 to the failed queue, and restart this loop without updating paddr.
874
875 If there were several addresses batched in the same remote delivery, only
876 the original top one will have host retry items attached to it, but we want
877 to handle all the same. Each will have a pointer back to its "top" address,
878 and they will now precede the item with the retries because addresses are
879 inverted when added to these final queues. We have saved information about
880 them in passing (below) so they can all be cut out at once. */
881
882 if (i == 2) /* Handling defers */
883 {
884 if (endaddr->retries != NULL && timed_out)
885 {
886 if (last_first == endaddr) paddr = saved_paddr;
887 addr = *paddr;
888 *paddr = endaddr->next;
889
890 endaddr->next = *addr_failed;
891 *addr_failed = addr;
892
893 for (;; addr = addr->next)
894 {
895 setflag(addr, af_retry_timedout);
896 addr->message = (addr->message == NULL)? US"retry timeout exceeded" :
897 string_sprintf("%s: retry timeout exceeded", addr->message);
fffffe4c
PH
898 addr->user_message = (addr->user_message == NULL)?
899 US"retry timeout exceeded" :
900 string_sprintf("%s: retry timeout exceeded", addr->user_message);
059ec3d9
PH
901 log_write(0, LOG_MAIN, "** %s%s%s%s: retry timeout exceeded",
902 addr->address,
903 (addr->parent == NULL)? US"" : US" <",
904 (addr->parent == NULL)? US"" : addr->parent->address,
905 (addr->parent == NULL)? US"" : US">");
906
907 if (addr == endaddr) break;
908 }
909
910 continue; /* Restart from changed *paddr */
911 }
912
913 /* This address is to remain on the defer chain. If it has a "first"
914 pointer, save the pointer to it in case we want to fail the set of
915 addresses when we get to the first one. */
916
917 if (endaddr->first != last_first)
918 {
919 last_first = endaddr->first;
920 saved_paddr = paddr;
921 }
922 }
923
924 /* All cases (succeed, fail, defer left on queue) */
925
926 paddr = &(endaddr->next); /* Advance to next address */
927 } /* Loop for all addresses */
928 } /* Loop for succeed, fail, defer */
929
930/* Close and unlock the database */
931
932if (dbm_file != NULL) dbfn_close(dbm_file);
933
934DEBUG(D_retry) debug_printf("end of retry processing\n");
935}
936
937/* End of retry.c */