Commit | Line | Data |
---|---|---|
c988f1f4 | 1 | /* $Cambridge: exim/src/src/retry.c,v 1.2 2005/01/04 10:00:42 ph10 Exp $ */ |
059ec3d9 PH |
2 | |
3 | /************************************************* | |
4 | * Exim - an Internet mail transport agent * | |
5 | *************************************************/ | |
6 | ||
c988f1f4 | 7 | /* Copyright (c) University of Cambridge 1995 - 2005 */ |
059ec3d9 PH |
8 | /* See the file NOTICE for conditions of use and distribution. */ |
9 | ||
10 | /* Functions concerned with retrying unsuccessful deliveries. */ | |
11 | ||
12 | ||
13 | #include "exim.h" | |
14 | ||
15 | ||
16 | ||
17 | /************************************************* | |
18 | * Check the ultimate address timeout * | |
19 | *************************************************/ | |
20 | ||
21 | /* This function tests whether a message has been on the queue longer than | |
22 | the maximum retry time for a particular host. | |
23 | ||
24 | Arguments: | |
25 | host_key the key to look up a host retry rule | |
26 | domain the domain to look up a domain retry rule | |
27 | basic_errno a specific error number, or zero if none | |
28 | more_errno additional data for the error | |
29 | now the time | |
30 | ||
31 | Returns: TRUE if the ultimate timeout has been reached | |
32 | */ | |
33 | ||
34 | static BOOL | |
35 | ultimate_address_timeout(uschar *host_key, uschar *domain, int basic_errno, | |
36 | int more_errno, time_t now) | |
37 | { | |
38 | BOOL address_timeout = TRUE; /* no rule => timed out */ | |
39 | ||
40 | retry_config *retry = | |
41 | retry_find_config(host_key+2, domain, basic_errno, more_errno); | |
42 | ||
43 | if (retry != NULL && retry->rules != NULL) | |
44 | { | |
45 | retry_rule *last_rule; | |
46 | for (last_rule = retry->rules; | |
47 | last_rule->next != NULL; | |
48 | last_rule = last_rule->next); | |
49 | address_timeout = (now - received_time > last_rule->timeout); | |
50 | } | |
51 | ||
52 | return address_timeout; | |
53 | } | |
54 | ||
55 | ||
56 | ||
57 | /************************************************* | |
58 | * Set status of a host+address item * | |
59 | *************************************************/ | |
60 | ||
61 | /* This function is passed a host_item which contains a host name and an | |
62 | IP address string. Its job is to set the status of the address if it is not | |
63 | already set (indicated by hstatus_unknown). The possible values are: | |
64 | ||
65 | hstatus_usable the address is not listed in the unusable tree, and does | |
66 | not have a retry record, OR the time is past the next | |
67 | try time, OR the message has been on the queue for more | |
68 | than the maximum retry time for a failing host | |
69 | ||
70 | hstatus_unusable the address is listed in the unusable tree, or does have | |
71 | a retry record, and the time is not yet at the next retry | |
72 | time. | |
73 | ||
74 | hstatus_unusable_expired as above, but also the retry time has expired | |
75 | for this address. | |
76 | ||
77 | The reason a delivery is permitted when a message has been around for a very | |
78 | long time is to allow the ultimate address timeout to operate after a delivery | |
79 | failure. Otherwise some messages may stick around without being tried for too | |
80 | long. | |
81 | ||
82 | If a host retry record is retrieved from the hints database, the time of last | |
83 | trying is filled into the last_try field of the host block. If a host is | |
84 | generally usable, a check is made to see if there is a retry delay on this | |
85 | specific message at this host. | |
86 | ||
87 | If a non-standard port is being used, it is added to the retry key. | |
88 | ||
89 | Arguments: | |
90 | domain the address domain | |
91 | host pointer to a host item | |
92 | portstring "" for standard port, ":xxxx" for a non-standard port | |
93 | include_ip_address TRUE to include the address in the key - this is | |
94 | usual, but sometimes is not wanted | |
95 | retry_host_key where to put a pointer to the key for the host-specific | |
96 | retry record, if one is read and the host is usable | |
97 | retry_message_key where to put a pointer to the key for the message+host | |
98 | retry record, if one is read and the host is usable | |
99 | ||
100 | Returns: TRUE if the host has expired but is usable because | |
101 | its retry time has come | |
102 | */ | |
103 | ||
104 | BOOL | |
105 | retry_check_address(uschar *domain, host_item *host, uschar *portstring, | |
106 | BOOL include_ip_address, uschar **retry_host_key, uschar **retry_message_key) | |
107 | { | |
108 | BOOL yield = FALSE; | |
109 | time_t now = time(NULL); | |
110 | uschar *host_key, *message_key; | |
111 | open_db dbblock; | |
112 | open_db *dbm_file; | |
113 | tree_node *node; | |
114 | dbdata_retry *host_retry_record, *message_retry_record; | |
115 | ||
116 | *retry_host_key = *retry_message_key = NULL; | |
117 | ||
118 | DEBUG(D_transport|D_retry) debug_printf("checking status of %s\n", host->name); | |
119 | ||
120 | /* Do nothing if status already set; otherwise initialize status as usable. */ | |
121 | ||
122 | if (host->status != hstatus_unknown) return FALSE; | |
123 | host->status = hstatus_usable; | |
124 | ||
125 | /* Generate the host key for the unusable tree and the retry database. Ensure | |
126 | host names are lower cased (that's what %S does). */ | |
127 | ||
128 | host_key = include_ip_address? | |
129 | string_sprintf("T:%S:%s%s", host->name, host->address, portstring) : | |
130 | string_sprintf("T:%S%s", host->name, portstring); | |
131 | ||
132 | /* Generate the message-specific key */ | |
133 | ||
134 | message_key = string_sprintf("%s:%s", host_key, message_id); | |
135 | ||
136 | /* Search the tree of unusable IP addresses. This is filled in when deliveries | |
137 | fail, because the retry database itself is not updated until the end of all | |
138 | deliveries (so as to do it all in one go). The tree records addresses that have | |
139 | become unusable during this delivery process (i.e. those that will get put into | |
140 | the retry database when it is updated). */ | |
141 | ||
142 | node = tree_search(tree_unusable, host_key); | |
143 | if (node != NULL) | |
144 | { | |
145 | DEBUG(D_transport|D_retry) debug_printf("found in tree of unusables\n"); | |
146 | host->status = (node->data.val > 255)? | |
147 | hstatus_unusable_expired : hstatus_unusable; | |
148 | host->why = node->data.val & 255; | |
149 | return FALSE; | |
150 | } | |
151 | ||
152 | /* Open the retry database, giving up if there isn't one. Otherwise, search for | |
153 | the retry records, and then close the database again. */ | |
154 | ||
155 | if ((dbm_file = dbfn_open(US"retry", O_RDONLY, &dbblock, FALSE)) == NULL) | |
156 | { | |
157 | DEBUG(D_deliver|D_retry|D_hints_lookup) | |
158 | debug_printf("no retry data available\n"); | |
159 | return FALSE; | |
160 | } | |
161 | host_retry_record = dbfn_read(dbm_file, host_key); | |
162 | message_retry_record = dbfn_read(dbm_file, message_key); | |
163 | dbfn_close(dbm_file); | |
164 | ||
165 | /* Ignore the data if it is too old - too long since it was written */ | |
166 | ||
167 | if (host_retry_record == NULL) | |
168 | { | |
169 | DEBUG(D_transport|D_retry) debug_printf("no host retry record\n"); | |
170 | } | |
171 | else if (now - host_retry_record->time_stamp > retry_data_expire) | |
172 | { | |
173 | host_retry_record = NULL; | |
174 | DEBUG(D_transport|D_retry) debug_printf("host retry record too old\n"); | |
175 | } | |
176 | ||
177 | if (message_retry_record == NULL) | |
178 | { | |
179 | DEBUG(D_transport|D_retry) debug_printf("no message retry record\n"); | |
180 | } | |
181 | else if (now - message_retry_record->time_stamp > retry_data_expire) | |
182 | { | |
183 | message_retry_record = NULL; | |
184 | DEBUG(D_transport|D_retry) debug_printf("message retry record too old\n"); | |
185 | } | |
186 | ||
187 | /* If there's a host-specific retry record, check for reaching the retry | |
188 | time (or forcing). If not, and the host is not expired, check for the message | |
189 | having been around for longer than the maximum retry time for this host or | |
190 | address. Allow the delivery if it has. Otherwise set the appropriate unusable | |
191 | flag and return FALSE. Otherwise arrange to return TRUE if this is an expired | |
192 | host. */ | |
193 | ||
194 | if (host_retry_record != NULL) | |
195 | { | |
196 | *retry_host_key = host_key; | |
197 | ||
198 | /* We have not reached the next try time. Check for the ultimate address | |
199 | timeout if the host has not expired. */ | |
200 | ||
201 | if (now < host_retry_record->next_try && !deliver_force) | |
202 | { | |
203 | DEBUG(D_transport|D_retry) | |
204 | debug_printf("host retry time not reached: checking ultimate address " | |
205 | "timeout\n"); | |
206 | ||
207 | if (!host_retry_record->expired && | |
208 | ultimate_address_timeout(host_key, domain, | |
209 | host_retry_record->basic_errno, host_retry_record->more_errno, now)) | |
210 | { | |
211 | DEBUG(D_transport|D_retry) | |
212 | debug_printf("on queue longer than maximum retry for " | |
213 | "address - allowing delivery\n"); | |
214 | return FALSE; | |
215 | } | |
216 | ||
217 | /* We have not hit the ultimate address timeout; host is unusable. */ | |
218 | ||
219 | host->status = (host_retry_record->expired)? | |
220 | hstatus_unusable_expired : hstatus_unusable; | |
221 | host->why = hwhy_retry; | |
222 | host->last_try = host_retry_record->last_try; | |
223 | return FALSE; | |
224 | } | |
225 | ||
226 | /* Host is usable; set return TRUE if expired. */ | |
227 | ||
228 | yield = host_retry_record->expired; | |
229 | } | |
230 | ||
231 | /* It's OK to try the host. If there's a message-specific retry record, check | |
232 | for reaching its retry time (or forcing). If not, mark the host unusable, | |
233 | unless the ultimate address timeout has been reached. */ | |
234 | ||
235 | if (message_retry_record != NULL) | |
236 | { | |
237 | *retry_message_key = message_key; | |
238 | if (now < message_retry_record->next_try && !deliver_force) | |
239 | { | |
240 | DEBUG(D_transport|D_retry) | |
241 | debug_printf("host+message retry time not reached: checking ultimate " | |
242 | "address timeout\n"); | |
243 | if (!ultimate_address_timeout(host_key, domain, 0, 0, now)) | |
244 | { | |
245 | host->status = hstatus_unusable; | |
246 | host->why = hwhy_retry; | |
247 | } | |
248 | else | |
249 | { | |
250 | DEBUG(D_transport|D_retry) | |
251 | debug_printf("on queue longer than maximum retry for " | |
252 | "address - allowing delivery\n"); | |
253 | } | |
254 | return FALSE; | |
255 | } | |
256 | } | |
257 | ||
258 | return yield; | |
259 | } | |
260 | ||
261 | ||
262 | ||
263 | ||
264 | /************************************************* | |
265 | * Add a retry item to an address * | |
266 | *************************************************/ | |
267 | ||
268 | /* Retry items are chained onto an address when it is deferred either by router | |
269 | or by a transport, or if it succeeds or fails and there was a previous retry | |
270 | item that now needs to be deleted. Sometimes there can be both kinds of item: | |
271 | for example, if routing was deferred but then succeeded, and delivery then | |
272 | deferred. In that case there is a delete item for the routing retry, and an | |
273 | updating item for the delivery. | |
274 | ||
275 | (But note that that is only visible at the outer level, because in remote | |
276 | delivery subprocesses, the address starts "clean", with no retry items carried | |
277 | in.) | |
278 | ||
279 | These items are used at the end of a delivery attempt to update the retry | |
280 | database. The keys start R: for routing delays and T: for transport delays. | |
281 | ||
282 | Arguments: | |
283 | addr the address block onto which to hang the item | |
284 | key the retry key | |
285 | flags delete, host, and message flags, copied into the block | |
286 | ||
287 | Returns: nothing | |
288 | */ | |
289 | ||
290 | void | |
291 | retry_add_item(address_item *addr, uschar *key, int flags) | |
292 | { | |
293 | retry_item *rti = store_get(sizeof(retry_item)); | |
294 | rti->next = addr->retries; | |
295 | addr->retries = rti; | |
296 | rti->key = key; | |
297 | rti->basic_errno = addr->basic_errno; | |
298 | rti->more_errno = addr->more_errno; | |
299 | rti->message = addr->message; | |
300 | rti->flags = flags; | |
301 | ||
302 | DEBUG(D_transport|D_retry) | |
303 | { | |
304 | int letter = rti->more_errno & 255; | |
305 | debug_printf("added retry item for %s: errno=%d more_errno=", rti->key, | |
306 | rti->basic_errno); | |
307 | if (letter == 'A' || letter == 'M') | |
308 | debug_printf("%d,%c", (rti->more_errno >> 8) & 255, letter); | |
309 | else | |
310 | debug_printf("%d", rti->more_errno); | |
311 | debug_printf(" flags=%d\n", flags); | |
312 | } | |
313 | } | |
314 | ||
315 | ||
316 | ||
317 | /************************************************* | |
318 | * Find retry configuration data * | |
319 | *************************************************/ | |
320 | ||
321 | /* Search the in-store retry information for the first retry item that applies | |
322 | to a given destination. If the key contains an @ we are probably handling a | |
323 | local delivery and have a complete address to search for; this happens when | |
324 | retry_use_local_part is set on a router. Otherwise, the key is likely to be a | |
325 | host name for a remote delivery, or a domain name for a local delivery. We | |
326 | prepend *@ on the front of it so that it will match a retry item whose address | |
327 | item pattern is independent of the local part. The alternate key, if set, is | |
328 | always just a domain, so we treat it likewise. | |
329 | ||
330 | Arguments: | |
331 | key key for which retry info is wanted | |
332 | alternate alternative key, always just a domain | |
333 | basic_errno specific error predicate on the retry rule, or zero | |
334 | more_errno additional data for errno predicate | |
335 | ||
336 | Returns: pointer to retry rule, or NULL | |
337 | */ | |
338 | ||
339 | retry_config * | |
340 | retry_find_config(uschar *key, uschar *alternate, int basic_errno, | |
341 | int more_errno) | |
342 | { | |
343 | int replace; | |
344 | uschar *use_key, *use_alternate; | |
345 | uschar *colon = Ustrchr(key, ':'); | |
346 | retry_config *yield; | |
347 | ||
348 | /* If there's a colon in the key, temporarily replace it with | |
349 | a zero to terminate the string there. */ | |
350 | ||
351 | if (colon != NULL) | |
352 | { | |
353 | replace = ':'; | |
354 | } | |
355 | else | |
356 | { | |
357 | colon = key + Ustrlen(key); | |
358 | replace = 0; | |
359 | } | |
360 | *colon = 0; | |
361 | ||
362 | /* Sort out the keys */ | |
363 | ||
364 | use_key = (Ustrchr(key, '@') != NULL)? key : string_sprintf("*@%s", key); | |
365 | use_alternate = (alternate == NULL)? NULL : string_sprintf("*@%s", alternate); | |
366 | ||
367 | /* Scan the configured retry items. */ | |
368 | ||
369 | for (yield = retries; yield != NULL; yield = yield->next) | |
370 | { | |
371 | uschar *plist = yield->pattern; | |
372 | uschar *slist = yield->senders; | |
373 | ||
374 | /* If a specific error is set for this item, check that we are handling that | |
375 | specific error, and if so, check any additional error information if | |
376 | required. */ | |
377 | ||
378 | if (yield->basic_errno != 0) | |
379 | { | |
380 | /* Special code is required for quota errors, as these can either be system | |
381 | quota errors, or Exim's own quota imposition, which has a different error | |
382 | number. Full partitions are also treated in the same way as quota errors. | |
383 | */ | |
384 | ||
385 | if (yield->basic_errno == ERRNO_EXIMQUOTA) | |
386 | { | |
387 | if ((basic_errno != ERRNO_EXIMQUOTA && basic_errno != errno_quota && | |
388 | basic_errno != ENOSPC) || | |
389 | (yield->more_errno != 0 && yield->more_errno > more_errno)) | |
390 | continue; | |
391 | } | |
392 | ||
393 | /* Handle 4xx responses to RCPT. The code that was received is in the 2nd | |
394 | least significant byte of more_errno (with 400 subtracted). The required | |
395 | value is coded in the 2nd least significant byte of the yield->more_errno | |
396 | field as follows: | |
397 | ||
398 | 255 => any 4xx code | |
399 | >= 100 => the decade must match the value less 100 | |
400 | < 100 => the exact value must match | |
401 | */ | |
402 | ||
403 | else if (yield->basic_errno == ERRNO_RCPT4XX) | |
404 | { | |
405 | int wanted; | |
406 | if (basic_errno != ERRNO_RCPT4XX) continue; | |
407 | wanted = (yield->more_errno >> 8) & 255; | |
408 | if (wanted != 255) | |
409 | { | |
410 | int evalue = (more_errno >> 8) & 255; | |
411 | if (wanted >= 100) | |
412 | { | |
413 | if ((evalue/10)*10 != wanted - 100) continue; | |
414 | } | |
415 | else if (evalue != wanted) continue; | |
416 | } | |
417 | } | |
418 | ||
419 | /* There are some special cases for timeouts */ | |
420 | ||
421 | else if (yield->basic_errno == ETIMEDOUT) | |
422 | { | |
423 | if (basic_errno != ETIMEDOUT) continue; | |
424 | ||
425 | /* Just RTEF_CTOUT in the rule => don't care about 'A'/'M' addresses */ | |
426 | if (yield->more_errno == RTEF_CTOUT) | |
427 | { | |
428 | if ((more_errno & RTEF_CTOUT) == 0) continue; | |
429 | } | |
430 | ||
431 | else if (yield->more_errno != 0) | |
432 | { | |
433 | int cf_errno = more_errno; | |
434 | if ((yield->more_errno & RTEF_CTOUT) == 0) cf_errno &= ~RTEF_CTOUT; | |
435 | if (yield->more_errno != cf_errno) continue; | |
436 | } | |
437 | } | |
438 | ||
439 | /* Default checks for exact match */ | |
440 | ||
441 | else | |
442 | { | |
443 | if (yield->basic_errno != basic_errno || | |
444 | (yield->more_errno != 0 && yield->more_errno != more_errno)) | |
445 | continue; | |
446 | } | |
447 | } | |
448 | ||
449 | /* If the "senders" condition is set, check it. Note that sender_address may | |
450 | be null during -brt checking, in which case we do not use this rule. */ | |
451 | ||
452 | if (slist != NULL && (sender_address == NULL || | |
453 | match_address_list(sender_address, TRUE, TRUE, &slist, NULL, -1, 0, | |
454 | NULL) != OK)) | |
455 | continue; | |
456 | ||
457 | /* Check for a match between the address list item at the start of this retry | |
458 | rule and either the main or alternate keys. */ | |
459 | ||
460 | if (match_address_list(use_key, TRUE, TRUE, &plist, NULL, -1, UCHAR_MAX+1, | |
461 | NULL) == OK || | |
462 | (use_alternate != NULL && | |
463 | match_address_list(use_alternate, TRUE, TRUE, &plist, NULL, -1, | |
464 | UCHAR_MAX+1, NULL) == OK)) | |
465 | break; | |
466 | } | |
467 | ||
468 | *colon = replace; | |
469 | return yield; | |
470 | } | |
471 | ||
472 | ||
473 | ||
474 | ||
475 | /************************************************* | |
476 | * Update retry database * | |
477 | *************************************************/ | |
478 | ||
479 | /* Update the retry data for any directing/routing/transporting that was | |
480 | deferred, or delete it for those that succeeded after a previous defer. This is | |
481 | done all in one go to minimize opening/closing/locking of the database file. | |
482 | ||
483 | Note that, because SMTP delivery involves a list of destinations to try, there | |
484 | may be defer-type retry information for some of them even when the message was | |
485 | successfully delivered. Likewise if it eventually failed. | |
486 | ||
487 | This function may move addresses from the defer to the failed queue if the | |
488 | ultimate retry time has expired. | |
489 | ||
490 | Arguments: | |
491 | addr_defer queue of deferred addresses | |
492 | addr_failed queue of failed addresses | |
493 | addr_succeed queue of successful addresses | |
494 | ||
495 | Returns: nothing | |
496 | */ | |
497 | ||
498 | void | |
499 | retry_update(address_item **addr_defer, address_item **addr_failed, | |
500 | address_item **addr_succeed) | |
501 | { | |
502 | open_db dbblock; | |
503 | open_db *dbm_file = NULL; | |
504 | time_t now = time(NULL); | |
505 | int i; | |
506 | ||
507 | DEBUG(D_retry) debug_printf("Processing retry items\n"); | |
508 | ||
509 | /* Three-times loop to handle succeeded, failed, and deferred addresses. | |
510 | Deferred addresses must be handled after failed ones, because some may be moved | |
511 | to the failed chain if they have timed out. */ | |
512 | ||
513 | for (i = 0; i < 3; i++) | |
514 | { | |
515 | address_item *endaddr, *addr; | |
516 | address_item *last_first = NULL; | |
517 | address_item **paddr = (i==0)? addr_succeed : | |
518 | (i==1)? addr_failed : addr_defer; | |
519 | address_item **saved_paddr = NULL; | |
520 | ||
521 | DEBUG(D_retry) debug_printf("%s addresses:\n", (i == 0)? "Succeeded" : | |
522 | (i == 1)? "Failed" : "Deferred"); | |
523 | ||
524 | /* Loop for each address on the chain. For deferred addresses, the whole | |
525 | address times out unless one of its retry addresses has a retry rule that | |
526 | hasn't yet timed out. Deferred addresses should not be requesting deletion | |
527 | of retry items, but just in case they do by accident, treat that case | |
528 | as "not timed out". | |
529 | ||
530 | As well as handling the addresses themselves, we must also process any | |
531 | retry items for any parent addresses - these are typically "delete" items, | |
532 | because the parent must have succeeded in order to generate the child. */ | |
533 | ||
534 | while ((endaddr = *paddr) != NULL) | |
535 | { | |
536 | BOOL timed_out = FALSE; | |
537 | retry_item *rti; | |
538 | ||
539 | for (addr = endaddr; addr != NULL; addr = addr->parent) | |
540 | { | |
541 | int update_count = 0; | |
542 | int timedout_count = 0; | |
543 | ||
544 | DEBUG(D_retry) debug_printf("%s%s\n", addr->address, (addr->retries == NULL)? | |
545 | ": no retry items" : ""); | |
546 | ||
547 | /* Loop for each retry item. */ | |
548 | ||
549 | for (rti = addr->retries; rti != NULL; rti = rti->next) | |
550 | { | |
551 | uschar *message; | |
552 | int message_length, message_space, failing_interval, next_try; | |
553 | retry_rule *rule, *final_rule; | |
554 | retry_config *retry; | |
555 | dbdata_retry *retry_record; | |
556 | ||
557 | /* Open the retry database if it is not already open; failure to open | |
558 | the file is logged, but otherwise ignored - deferred addresses will | |
559 | get retried at the next opportunity. Not opening earlier than this saves | |
560 | opening if no addresses have retry items - common when none have yet | |
561 | reached their retry next try time. */ | |
562 | ||
563 | if (dbm_file == NULL) | |
564 | dbm_file = dbfn_open(US"retry", O_RDWR, &dbblock, TRUE); | |
565 | ||
566 | if (dbm_file == NULL) | |
567 | { | |
568 | DEBUG(D_deliver|D_retry|D_hints_lookup) | |
569 | debug_printf("retry database not available for updating\n"); | |
570 | return; | |
571 | } | |
572 | ||
573 | /* If there are no deferred addresses, that is, if this message is | |
574 | completing, and the retry item is for a message-specific SMTP error, | |
575 | force it to be deleted, because there's no point in keeping data for | |
576 | no-longer-existing messages. This situation can occur when a domain has | |
577 | two hosts and a message-specific error occurs for the first of them, | |
578 | but the address gets delivered to the second one. This optimization | |
579 | doesn't succeed in cleaning out all the dead entries, but it helps. */ | |
580 | ||
581 | if (*addr_defer == NULL && (rti->flags & rf_message) != 0) | |
582 | rti->flags |= rf_delete; | |
583 | ||
584 | /* Handle the case of a request to delete the retry info for this | |
585 | destination. */ | |
586 | ||
587 | if ((rti->flags & rf_delete) != 0) | |
588 | { | |
589 | (void)dbfn_delete(dbm_file, rti->key); | |
590 | DEBUG(D_retry) | |
591 | debug_printf("deleted retry information for %s\n", rti->key); | |
592 | continue; | |
593 | } | |
594 | ||
595 | /* Count the number of non-delete retry items. This is so that we | |
596 | can compare it to the count of timed_out ones, to check whether | |
597 | all are timed out. */ | |
598 | ||
599 | update_count++; | |
600 | ||
601 | /* Get the retry information for this destination and error code, if | |
602 | any. If this item is for a remote host with ip address, then pass | |
603 | the domain name as an alternative to search for. If no retry | |
604 | information is found, we can't generate a retry time, so there is | |
605 | no point updating the database. This retry item is timed out. */ | |
606 | ||
607 | if ((retry = retry_find_config(rti->key + 2, | |
608 | ((rti->flags & rf_host) != 0)? addr->domain : NULL, | |
609 | rti->basic_errno, rti->more_errno)) == NULL) | |
610 | { | |
611 | DEBUG(D_retry) debug_printf("No configured retry item for %s%s%s\n", | |
612 | rti->key, | |
613 | ((rti->flags & rf_host) != 0)? US" or " : US"", | |
614 | ((rti->flags & rf_host) != 0)? addr->domain : US""); | |
615 | if (addr == endaddr) timedout_count++; | |
616 | continue; | |
617 | } | |
618 | ||
619 | DEBUG(D_retry) | |
620 | { | |
621 | if ((rti->flags & rf_host) != 0) | |
622 | debug_printf("retry for %s (%s) = %s\n", rti->key, | |
623 | addr->domain, retry->pattern); | |
624 | else | |
625 | debug_printf("retry for %s = %s\n", rti->key, retry->pattern); | |
626 | } | |
627 | ||
628 | /* Set up the message for the database retry record. Because DBM | |
629 | records have a maximum data length, we enforce a limit. There isn't | |
630 | much point in keeping a huge message here, anyway. */ | |
631 | ||
632 | message = (rti->basic_errno > 0)? US strerror(rti->basic_errno) : | |
633 | (rti->message == NULL)? | |
634 | US"unknown error" : string_printing(rti->message); | |
635 | message_length = Ustrlen(message); | |
636 | if (message_length > 150) message_length = 150; | |
637 | ||
638 | /* Read a retry record from the database or construct a new one. | |
639 | Ignore an old one if it is too old since it was last updated. */ | |
640 | ||
641 | retry_record = dbfn_read(dbm_file, rti->key); | |
642 | if (retry_record != NULL && | |
643 | now - retry_record->time_stamp > retry_data_expire) | |
644 | retry_record = NULL; | |
645 | ||
646 | if (retry_record == NULL) | |
647 | { | |
648 | retry_record = store_get(sizeof(dbdata_retry) + message_length); | |
649 | message_space = message_length; | |
650 | retry_record->first_failed = now; | |
651 | retry_record->last_try = now; | |
652 | retry_record->next_try = now; | |
653 | retry_record->expired = FALSE; | |
654 | retry_record->text[0] = 0; /* just in case */ | |
655 | } | |
656 | else message_space = Ustrlen(retry_record->text); | |
657 | ||
658 | /* Compute how long this destination has been failing */ | |
659 | ||
660 | failing_interval = now - retry_record->first_failed; | |
661 | ||
662 | /* Search for the current retry rule. The cutoff time of the | |
663 | last rule is handled differently to the others. The rule continues | |
664 | to operate for ever (the global maximum interval will eventually | |
665 | limit the gaps) but its cutoff time determines when an individual | |
666 | destination times out. If there are no retry rules, the destination | |
667 | always times out, but we can't compute a retry time. */ | |
668 | ||
669 | final_rule = NULL; | |
670 | for (rule = retry->rules; rule != NULL; rule = rule->next) | |
671 | { | |
672 | if (failing_interval <= rule->timeout) break; | |
673 | final_rule = rule; | |
674 | } | |
675 | ||
676 | /* If there's an un-timed out rule, the destination has not | |
677 | yet timed out, so the address as a whole has not timed out (but we are | |
678 | interested in this only for the end address). Make sure the expired | |
679 | flag is false (can be forced via fixdb from outside, but ensure it is | |
680 | consistent with the rules whenever we go through here). */ | |
681 | ||
682 | if (rule != NULL) | |
683 | { | |
684 | retry_record->expired = FALSE; | |
685 | } | |
686 | ||
687 | /* Otherwise, set the retry timeout expired, and set the final rule | |
688 | as the one from which to compute the next retry time. Subsequent | |
689 | messages will fail immediately until the retry time is reached (unless | |
690 | there are other, still active, retries). */ | |
691 | ||
692 | else | |
693 | { | |
694 | rule = final_rule; | |
695 | retry_record->expired = TRUE; | |
696 | if (addr == endaddr) timedout_count++; | |
697 | } | |
698 | ||
699 | /* There is a special case to consider when some messages get through | |
700 | to a destination and others don't. This can happen locally when a | |
701 | large message pushes a user over quota, and it can happen remotely | |
702 | when a machine is on a dodgy Internet connection. The messages that | |
703 | get through wipe the retry information, causing those that don't to | |
704 | stay on the queue longer than the final retry time. In order to | |
705 | avoid this, we check, using the time of arrival of the message, to | |
706 | see if it has been on the queue for more than the final cutoff time, | |
707 | and if so, cause this retry item to time out, and the retry time to | |
708 | be set to "now" so that any subsequent messages in the same condition | |
709 | also get tried. We search for the last rule onwards from the one that | |
710 | is in use. If there are no retry rules for the item, rule will be null | |
711 | and timedout_count will already have been updated. | |
712 | ||
713 | This implements "timeout this rule if EITHER the host (or routing or | |
714 | directing) has been failing for more than the maximum time, OR if the | |
715 | message has been on the queue for more than the maximum time." */ | |
716 | ||
717 | if (received_time <= retry_record->first_failed && | |
718 | addr == endaddr && !retry_record->expired && rule != NULL) | |
719 | { | |
720 | retry_rule *last_rule; | |
721 | for (last_rule = rule; | |
722 | last_rule->next != NULL; | |
723 | last_rule = last_rule->next); | |
724 | if (now - received_time > last_rule->timeout) | |
725 | { | |
726 | DEBUG(D_retry) debug_printf("on queue longer than maximum retry\n"); | |
727 | timedout_count++; | |
728 | rule = NULL; | |
729 | } | |
730 | } | |
731 | ||
732 | /* Compute the next try time from the rule, subject to the global | |
733 | maximum, and update the retry database. If rule == NULL it means | |
734 | there were no rules at all (and the timeout will be set expired), | |
735 | or we have a message that is older than the final timeout. In this | |
736 | case set the next retry time to now, so that one delivery attempt | |
737 | happens for subsequent messages. */ | |
738 | ||
739 | if (rule == NULL) next_try = now; else | |
740 | { | |
741 | if (rule->rule == 'F') next_try = now + rule->p1; | |
742 | else /* assume rule = 'G' */ | |
743 | { | |
744 | int last_predicted_gap = | |
745 | retry_record->next_try - retry_record->last_try; | |
746 | int last_actual_gap = now - retry_record->last_try; | |
747 | int lastgap = (last_predicted_gap < last_actual_gap)? | |
748 | last_predicted_gap : last_actual_gap; | |
749 | next_try = now + ((lastgap < rule->p1)? rule->p1 : | |
750 | (lastgap * rule->p2)/1000); | |
751 | } | |
752 | } | |
753 | ||
754 | /* Impose a global retry max */ | |
755 | ||
756 | if (next_try - now > retry_interval_max) | |
757 | next_try = now + retry_interval_max; | |
758 | ||
759 | /* If the new message length is greater than the previous one, we | |
760 | have to copy the record first. */ | |
761 | ||
762 | if (message_length > message_space) | |
763 | { | |
764 | dbdata_retry *newr = store_get(sizeof(dbdata_retry) + message_length); | |
765 | memcpy(newr, retry_record, sizeof(dbdata_retry)); | |
766 | retry_record = newr; | |
767 | } | |
768 | ||
769 | /* Set up the retry record; message_length may be less than the string | |
770 | length for very long error strings. */ | |
771 | ||
772 | retry_record->last_try = now; | |
773 | retry_record->next_try = next_try; | |
774 | retry_record->basic_errno = rti->basic_errno; | |
775 | retry_record->more_errno = rti->more_errno; | |
776 | Ustrncpy(retry_record->text, message, message_length); | |
777 | retry_record->text[message_length] = 0; | |
778 | ||
779 | DEBUG(D_retry) | |
780 | { | |
781 | int letter = retry_record->more_errno & 255; | |
782 | debug_printf("Writing retry data for %s\n", rti->key); | |
783 | debug_printf(" first failed=%d last try=%d next try=%d expired=%d\n", | |
784 | (int)retry_record->first_failed, (int)retry_record->last_try, | |
785 | (int)retry_record->next_try, retry_record->expired); | |
786 | debug_printf(" errno=%d more_errno=", retry_record->basic_errno); | |
787 | if (letter == 'A' || letter == 'M') | |
788 | debug_printf("%d,%c", (retry_record->more_errno >> 8) & 255, | |
789 | letter); | |
790 | else | |
791 | debug_printf("%d", retry_record->more_errno); | |
792 | debug_printf(" %s\n", retry_record->text); | |
793 | } | |
794 | ||
795 | (void)dbfn_write(dbm_file, rti->key, retry_record, | |
796 | sizeof(dbdata_retry) + message_length); | |
797 | } /* Loop for each retry item */ | |
798 | ||
799 | /* If all the non-delete retry items are timed out, the address is | |
800 | timed out, provided that we didn't skip any hosts because their retry | |
801 | time was not reached (or because of hosts_max_try). */ | |
802 | ||
803 | if (update_count > 0 && update_count == timedout_count) | |
804 | { | |
805 | if (!testflag(endaddr, af_retry_skipped)) | |
806 | { | |
807 | DEBUG(D_retry) debug_printf("timed out: all retries expired\n"); | |
808 | timed_out = TRUE; | |
809 | } | |
810 | else | |
811 | { | |
812 | DEBUG(D_retry) | |
813 | debug_printf("timed out but some hosts were skipped\n"); | |
814 | } | |
815 | } | |
816 | } /* Loop for an address and its parents */ | |
817 | ||
818 | /* If this is a deferred address, and retry processing was requested by | |
819 | means of one or more retry items, and they all timed out, move the address | |
820 | to the failed queue, and restart this loop without updating paddr. | |
821 | ||
822 | If there were several addresses batched in the same remote delivery, only | |
823 | the original top one will have host retry items attached to it, but we want | |
824 | to handle all the same. Each will have a pointer back to its "top" address, | |
825 | and they will now precede the item with the retries because addresses are | |
826 | inverted when added to these final queues. We have saved information about | |
827 | them in passing (below) so they can all be cut out at once. */ | |
828 | ||
829 | if (i == 2) /* Handling defers */ | |
830 | { | |
831 | if (endaddr->retries != NULL && timed_out) | |
832 | { | |
833 | if (last_first == endaddr) paddr = saved_paddr; | |
834 | addr = *paddr; | |
835 | *paddr = endaddr->next; | |
836 | ||
837 | endaddr->next = *addr_failed; | |
838 | *addr_failed = addr; | |
839 | ||
840 | for (;; addr = addr->next) | |
841 | { | |
842 | setflag(addr, af_retry_timedout); | |
843 | addr->message = (addr->message == NULL)? US"retry timeout exceeded" : | |
844 | string_sprintf("%s: retry timeout exceeded", addr->message); | |
845 | log_write(0, LOG_MAIN, "** %s%s%s%s: retry timeout exceeded", | |
846 | addr->address, | |
847 | (addr->parent == NULL)? US"" : US" <", | |
848 | (addr->parent == NULL)? US"" : addr->parent->address, | |
849 | (addr->parent == NULL)? US"" : US">"); | |
850 | ||
851 | if (addr == endaddr) break; | |
852 | } | |
853 | ||
854 | continue; /* Restart from changed *paddr */ | |
855 | } | |
856 | ||
857 | /* This address is to remain on the defer chain. If it has a "first" | |
858 | pointer, save the pointer to it in case we want to fail the set of | |
859 | addresses when we get to the first one. */ | |
860 | ||
861 | if (endaddr->first != last_first) | |
862 | { | |
863 | last_first = endaddr->first; | |
864 | saved_paddr = paddr; | |
865 | } | |
866 | } | |
867 | ||
868 | /* All cases (succeed, fail, defer left on queue) */ | |
869 | ||
870 | paddr = &(endaddr->next); /* Advance to next address */ | |
871 | } /* Loop for all addresses */ | |
872 | } /* Loop for succeed, fail, defer */ | |
873 | ||
874 | /* Close and unlock the database */ | |
875 | ||
876 | if (dbm_file != NULL) dbfn_close(dbm_file); | |
877 | ||
878 | DEBUG(D_retry) debug_printf("end of retry processing\n"); | |
879 | } | |
880 | ||
881 | /* End of retry.c */ |