Commit | Line | Data |
---|---|---|
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 | 20 | the maximum retry time for a particular host or address. |
059ec3d9 PH |
21 | |
22 | Arguments: | |
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 | ||
28 | Returns: TRUE if the ultimate timeout has been reached | |
29 | */ | |
30 | ||
ba9af0af TF |
31 | BOOL |
32 | retry_ultimate_address_timeout(uschar *retry_key, uschar *domain, | |
33 | dbdata_retry *retry_record, time_t now) | |
059ec3d9 | 34 | { |
ba9af0af TF |
35 | BOOL address_timeout; |
36 | ||
37 | DEBUG(D_retry) | |
38 | { | |
39 | debug_printf("retry time not reached: checking ultimate address timeout\n"); | |
40 | debug_printf(" now=%d first_failed=%d next_try=%d expired=%d\n", | |
41 | (int)now, (int)retry_record->first_failed, | |
42 | (int)retry_record->next_try, retry_record->expired); | |
43 | } | |
059ec3d9 PH |
44 | |
45 | retry_config *retry = | |
ba9af0af TF |
46 | retry_find_config(retry_key+2, domain, |
47 | retry_record->basic_errno, retry_record->more_errno); | |
059ec3d9 PH |
48 | |
49 | if (retry != NULL && retry->rules != NULL) | |
50 | { | |
51 | retry_rule *last_rule; | |
52 | for (last_rule = retry->rules; | |
53 | last_rule->next != NULL; | |
54 | last_rule = last_rule->next); | |
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 |
60 | else |
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 |
67 | DEBUG(D_retry) |
68 | if (address_timeout) | |
69 | debug_printf("on queue longer than maximum retry for address - " | |
70 | "allowing delivery\n"); | |
71 | ||
059ec3d9 PH |
72 | return address_timeout; |
73 | } | |
74 | ||
75 | ||
76 | ||
77 | /************************************************* | |
78 | * Set status of a host+address item * | |
79 | *************************************************/ | |
80 | ||
81 | /* This function is passed a host_item which contains a host name and an | |
82 | IP address string. Its job is to set the status of the address if it is not | |
83 | already set (indicated by hstatus_unknown). The possible values are: | |
84 | ||
85 | hstatus_usable the address is not listed in the unusable tree, and does | |
86 | not have a retry record, OR the time is past the next | |
87 | try time, OR the message has been on the queue for more | |
88 | than the maximum retry time for a failing host | |
89 | ||
90 | hstatus_unusable the address is listed in the unusable tree, or does have | |
91 | a retry record, and the time is not yet at the next retry | |
92 | time. | |
93 | ||
94 | hstatus_unusable_expired as above, but also the retry time has expired | |
95 | for this address. | |
96 | ||
97 | The reason a delivery is permitted when a message has been around for a very | |
98 | long time is to allow the ultimate address timeout to operate after a delivery | |
99 | failure. Otherwise some messages may stick around without being tried for too | |
100 | long. | |
101 | ||
102 | If a host retry record is retrieved from the hints database, the time of last | |
103 | trying is filled into the last_try field of the host block. If a host is | |
104 | generally usable, a check is made to see if there is a retry delay on this | |
105 | specific message at this host. | |
106 | ||
107 | If a non-standard port is being used, it is added to the retry key. | |
108 | ||
109 | Arguments: | |
110 | domain the address domain | |
111 | host pointer to a host item | |
112 | portstring "" for standard port, ":xxxx" for a non-standard port | |
113 | include_ip_address TRUE to include the address in the key - this is | |
114 | usual, but sometimes is not wanted | |
115 | retry_host_key where to put a pointer to the key for the host-specific | |
116 | retry record, if one is read and the host is usable | |
117 | retry_message_key where to put a pointer to the key for the message+host | |
118 | retry record, if one is read and the host is usable | |
119 | ||
120 | Returns: TRUE if the host has expired but is usable because | |
121 | its retry time has come | |
122 | */ | |
123 | ||
124 | BOOL | |
125 | retry_check_address(uschar *domain, host_item *host, uschar *portstring, | |
126 | BOOL include_ip_address, uschar **retry_host_key, uschar **retry_message_key) | |
127 | { | |
128 | BOOL yield = FALSE; | |
129 | time_t now = time(NULL); | |
130 | uschar *host_key, *message_key; | |
131 | open_db dbblock; | |
132 | open_db *dbm_file; | |
133 | tree_node *node; | |
134 | dbdata_retry *host_retry_record, *message_retry_record; | |
135 | ||
136 | *retry_host_key = *retry_message_key = NULL; | |
137 | ||
138 | DEBUG(D_transport|D_retry) debug_printf("checking status of %s\n", host->name); | |
139 | ||
140 | /* Do nothing if status already set; otherwise initialize status as usable. */ | |
141 | ||
142 | if (host->status != hstatus_unknown) return FALSE; | |
143 | host->status = hstatus_usable; | |
144 | ||
145 | /* Generate the host key for the unusable tree and the retry database. Ensure | |
146 | host names are lower cased (that's what %S does). */ | |
147 | ||
148 | host_key = include_ip_address? | |
149 | string_sprintf("T:%S:%s%s", host->name, host->address, portstring) : | |
150 | string_sprintf("T:%S%s", host->name, portstring); | |
151 | ||
152 | /* Generate the message-specific key */ | |
153 | ||
154 | message_key = string_sprintf("%s:%s", host_key, message_id); | |
155 | ||
156 | /* Search the tree of unusable IP addresses. This is filled in when deliveries | |
157 | fail, because the retry database itself is not updated until the end of all | |
158 | deliveries (so as to do it all in one go). The tree records addresses that have | |
159 | become unusable during this delivery process (i.e. those that will get put into | |
160 | the retry database when it is updated). */ | |
161 | ||
162 | node = tree_search(tree_unusable, host_key); | |
163 | if (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 | |
173 | the retry records, and then close the database again. */ | |
174 | ||
175 | if ((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 | } | |
181 | host_retry_record = dbfn_read(dbm_file, host_key); | |
182 | message_retry_record = dbfn_read(dbm_file, message_key); | |
183 | dbfn_close(dbm_file); | |
184 | ||
185 | /* Ignore the data if it is too old - too long since it was written */ | |
186 | ||
187 | if (host_retry_record == NULL) | |
188 | { | |
189 | DEBUG(D_transport|D_retry) debug_printf("no host retry record\n"); | |
190 | } | |
191 | else 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 | ||
197 | if (message_retry_record == NULL) | |
198 | { | |
199 | DEBUG(D_transport|D_retry) debug_printf("no message retry record\n"); | |
200 | } | |
201 | else 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 | |
208 | time (or forcing). If not, and the host is not expired, check for the message | |
209 | having been around for longer than the maximum retry time for this host or | |
210 | address. Allow the delivery if it has. Otherwise set the appropriate unusable | |
211 | flag and return FALSE. Otherwise arrange to return TRUE if this is an expired | |
212 | host. */ | |
213 | ||
214 | if (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 | |
243 | for reaching its retry time (or forcing). If not, mark the host unusable, | |
244 | unless the ultimate address timeout has been reached. */ | |
245 | ||
246 | if (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 | ||
261 | return 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 | |
272 | or by a transport, or if it succeeds or fails and there was a previous retry | |
273 | item that now needs to be deleted. Sometimes there can be both kinds of item: | |
274 | for example, if routing was deferred but then succeeded, and delivery then | |
275 | deferred. In that case there is a delete item for the routing retry, and an | |
276 | updating item for the delivery. | |
277 | ||
278 | (But note that that is only visible at the outer level, because in remote | |
279 | delivery subprocesses, the address starts "clean", with no retry items carried | |
280 | in.) | |
281 | ||
282 | These items are used at the end of a delivery attempt to update the retry | |
283 | database. The keys start R: for routing delays and T: for transport delays. | |
284 | ||
285 | Arguments: | |
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 | ||
290 | Returns: nothing | |
291 | */ | |
292 | ||
293 | void | |
294 | retry_add_item(address_item *addr, uschar *key, int flags) | |
295 | { | |
296 | retry_item *rti = store_get(sizeof(retry_item)); | |
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 = addr->message; | |
303 | rti->flags = flags; | |
304 | ||
305 | DEBUG(D_transport|D_retry) | |
306 | { | |
307 | int letter = rti->more_errno & 255; | |
308 | debug_printf("added retry item for %s: errno=%d more_errno=", rti->key, | |
309 | rti->basic_errno); | |
310 | if (letter == 'A' || letter == 'M') | |
311 | debug_printf("%d,%c", (rti->more_errno >> 8) & 255, letter); | |
312 | else | |
313 | debug_printf("%d", rti->more_errno); | |
314 | debug_printf(" flags=%d\n", flags); | |
315 | } | |
316 | } | |
317 | ||
318 | ||
319 | ||
320 | /************************************************* | |
321 | * Find retry configuration data * | |
322 | *************************************************/ | |
323 | ||
324 | /* Search the in-store retry information for the first retry item that applies | |
325 | to a given destination. If the key contains an @ we are probably handling a | |
326 | local delivery and have a complete address to search for; this happens when | |
327 | retry_use_local_part is set on a router. Otherwise, the key is likely to be a | |
328 | host name for a remote delivery, or a domain name for a local delivery. We | |
329 | prepend *@ on the front of it so that it will match a retry item whose address | |
330 | item pattern is independent of the local part. The alternate key, if set, is | |
331 | always just a domain, so we treat it likewise. | |
332 | ||
333 | Arguments: | |
334 | key key for which retry info is wanted | |
335 | alternate alternative key, always just a domain | |
336 | basic_errno specific error predicate on the retry rule, or zero | |
337 | more_errno additional data for errno predicate | |
338 | ||
339 | Returns: pointer to retry rule, or NULL | |
340 | */ | |
341 | ||
342 | retry_config * | |
343 | retry_find_config(uschar *key, uschar *alternate, int basic_errno, | |
344 | int more_errno) | |
345 | { | |
ea49d0e1 | 346 | int replace = 0; |
059ec3d9 PH |
347 | uschar *use_key, *use_alternate; |
348 | uschar *colon = Ustrchr(key, ':'); | |
349 | retry_config *yield; | |
350 | ||
ea49d0e1 PH |
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 temporarily replace the colon with a zero, to terminate | |
358 | the string after the host name. | |
359 | ||
360 | (2) This is a key for a pipe, file, or autoreply delivery, in the format | |
361 | ||
362 | pipe-or-file-or-auto:x@y | |
363 | ||
364 | where x@y is the original address that provoked the delivery. The pipe or | |
365 | file or auto will start with | or / or >, whereas a host name will start | |
366 | with a letter or a digit. In this case we want to use the original address | |
367 | to search for a retry rule. */ | |
059ec3d9 PH |
368 | |
369 | if (colon != NULL) | |
370 | { | |
ea49d0e1 PH |
371 | if (isalnum(*key)) |
372 | replace = ':'; | |
373 | else | |
374 | key = Ustrrchr(key, ':') + 1; /* Take from the last colon */ | |
059ec3d9 | 375 | } |
ea49d0e1 PH |
376 | |
377 | if (replace == 0) colon = key + Ustrlen(key); | |
059ec3d9 PH |
378 | *colon = 0; |
379 | ||
380 | /* Sort out the keys */ | |
381 | ||
382 | use_key = (Ustrchr(key, '@') != NULL)? key : string_sprintf("*@%s", key); | |
383 | use_alternate = (alternate == NULL)? NULL : string_sprintf("*@%s", alternate); | |
384 | ||
385 | /* Scan the configured retry items. */ | |
386 | ||
387 | for (yield = retries; yield != NULL; yield = yield->next) | |
388 | { | |
389 | uschar *plist = yield->pattern; | |
390 | uschar *slist = yield->senders; | |
391 | ||
392 | /* If a specific error is set for this item, check that we are handling that | |
393 | specific error, and if so, check any additional error information if | |
394 | required. */ | |
395 | ||
396 | if (yield->basic_errno != 0) | |
397 | { | |
398 | /* Special code is required for quota errors, as these can either be system | |
399 | quota errors, or Exim's own quota imposition, which has a different error | |
400 | number. Full partitions are also treated in the same way as quota errors. | |
401 | */ | |
402 | ||
403 | if (yield->basic_errno == ERRNO_EXIMQUOTA) | |
404 | { | |
405 | if ((basic_errno != ERRNO_EXIMQUOTA && basic_errno != errno_quota && | |
406 | basic_errno != ENOSPC) || | |
407 | (yield->more_errno != 0 && yield->more_errno > more_errno)) | |
408 | continue; | |
409 | } | |
410 | ||
e97957bc PH |
411 | /* The TLSREQUIRED error also covers TLSFAILURE. These are subtly different |
412 | errors, but not worth separating at this level. */ | |
413 | ||
414 | else if (yield->basic_errno == ERRNO_TLSREQUIRED) | |
415 | { | |
416 | if (basic_errno != ERRNO_TLSREQUIRED && basic_errno != ERRNO_TLSFAILURE) | |
417 | continue; | |
418 | } | |
419 | ||
420 | /* Handle 4xx responses to MAIL, RCPT, or DATA. The code that was received | |
421 | is in the 2nd least significant byte of more_errno (with 400 subtracted). | |
422 | The required value is coded in the 2nd least significant byte of the | |
423 | yield->more_errno field as follows: | |
059ec3d9 PH |
424 | |
425 | 255 => any 4xx code | |
426 | >= 100 => the decade must match the value less 100 | |
427 | < 100 => the exact value must match | |
428 | */ | |
429 | ||
e97957bc PH |
430 | else if (yield->basic_errno == ERRNO_MAIL4XX || |
431 | yield->basic_errno == ERRNO_RCPT4XX || | |
432 | yield->basic_errno == ERRNO_DATA4XX) | |
059ec3d9 PH |
433 | { |
434 | int wanted; | |
e97957bc | 435 | if (basic_errno != yield->basic_errno) continue; |
059ec3d9 PH |
436 | wanted = (yield->more_errno >> 8) & 255; |
437 | if (wanted != 255) | |
438 | { | |
439 | int evalue = (more_errno >> 8) & 255; | |
440 | if (wanted >= 100) | |
441 | { | |
442 | if ((evalue/10)*10 != wanted - 100) continue; | |
443 | } | |
444 | else if (evalue != wanted) continue; | |
445 | } | |
446 | } | |
447 | ||
448 | /* There are some special cases for timeouts */ | |
449 | ||
450 | else if (yield->basic_errno == ETIMEDOUT) | |
451 | { | |
452 | if (basic_errno != ETIMEDOUT) continue; | |
453 | ||
454 | /* Just RTEF_CTOUT in the rule => don't care about 'A'/'M' addresses */ | |
455 | if (yield->more_errno == RTEF_CTOUT) | |
456 | { | |
457 | if ((more_errno & RTEF_CTOUT) == 0) continue; | |
458 | } | |
459 | ||
460 | else if (yield->more_errno != 0) | |
461 | { | |
462 | int cf_errno = more_errno; | |
463 | if ((yield->more_errno & RTEF_CTOUT) == 0) cf_errno &= ~RTEF_CTOUT; | |
464 | if (yield->more_errno != cf_errno) continue; | |
465 | } | |
466 | } | |
467 | ||
468 | /* Default checks for exact match */ | |
469 | ||
470 | else | |
471 | { | |
472 | if (yield->basic_errno != basic_errno || | |
473 | (yield->more_errno != 0 && yield->more_errno != more_errno)) | |
474 | continue; | |
475 | } | |
476 | } | |
477 | ||
478 | /* If the "senders" condition is set, check it. Note that sender_address may | |
479 | be null during -brt checking, in which case we do not use this rule. */ | |
480 | ||
481 | if (slist != NULL && (sender_address == NULL || | |
482 | match_address_list(sender_address, TRUE, TRUE, &slist, NULL, -1, 0, | |
483 | NULL) != OK)) | |
484 | continue; | |
485 | ||
486 | /* Check for a match between the address list item at the start of this retry | |
487 | rule and either the main or alternate keys. */ | |
488 | ||
489 | if (match_address_list(use_key, TRUE, TRUE, &plist, NULL, -1, UCHAR_MAX+1, | |
490 | NULL) == OK || | |
491 | (use_alternate != NULL && | |
492 | match_address_list(use_alternate, TRUE, TRUE, &plist, NULL, -1, | |
493 | UCHAR_MAX+1, NULL) == OK)) | |
494 | break; | |
495 | } | |
496 | ||
497 | *colon = replace; | |
498 | return yield; | |
499 | } | |
500 | ||
501 | ||
502 | ||
503 | ||
504 | /************************************************* | |
505 | * Update retry database * | |
506 | *************************************************/ | |
507 | ||
508 | /* Update the retry data for any directing/routing/transporting that was | |
509 | deferred, or delete it for those that succeeded after a previous defer. This is | |
510 | done all in one go to minimize opening/closing/locking of the database file. | |
511 | ||
512 | Note that, because SMTP delivery involves a list of destinations to try, there | |
513 | may be defer-type retry information for some of them even when the message was | |
514 | successfully delivered. Likewise if it eventually failed. | |
515 | ||
516 | This function may move addresses from the defer to the failed queue if the | |
517 | ultimate retry time has expired. | |
518 | ||
519 | Arguments: | |
520 | addr_defer queue of deferred addresses | |
521 | addr_failed queue of failed addresses | |
522 | addr_succeed queue of successful addresses | |
523 | ||
524 | Returns: nothing | |
525 | */ | |
526 | ||
527 | void | |
528 | retry_update(address_item **addr_defer, address_item **addr_failed, | |
529 | address_item **addr_succeed) | |
530 | { | |
531 | open_db dbblock; | |
532 | open_db *dbm_file = NULL; | |
533 | time_t now = time(NULL); | |
534 | int i; | |
535 | ||
536 | DEBUG(D_retry) debug_printf("Processing retry items\n"); | |
537 | ||
538 | /* Three-times loop to handle succeeded, failed, and deferred addresses. | |
539 | Deferred addresses must be handled after failed ones, because some may be moved | |
540 | to the failed chain if they have timed out. */ | |
541 | ||
542 | for (i = 0; i < 3; i++) | |
543 | { | |
544 | address_item *endaddr, *addr; | |
545 | address_item *last_first = NULL; | |
546 | address_item **paddr = (i==0)? addr_succeed : | |
547 | (i==1)? addr_failed : addr_defer; | |
548 | address_item **saved_paddr = NULL; | |
549 | ||
550 | DEBUG(D_retry) debug_printf("%s addresses:\n", (i == 0)? "Succeeded" : | |
551 | (i == 1)? "Failed" : "Deferred"); | |
552 | ||
553 | /* Loop for each address on the chain. For deferred addresses, the whole | |
554 | address times out unless one of its retry addresses has a retry rule that | |
555 | hasn't yet timed out. Deferred addresses should not be requesting deletion | |
556 | of retry items, but just in case they do by accident, treat that case | |
557 | as "not timed out". | |
558 | ||
559 | As well as handling the addresses themselves, we must also process any | |
560 | retry items for any parent addresses - these are typically "delete" items, | |
561 | because the parent must have succeeded in order to generate the child. */ | |
562 | ||
563 | while ((endaddr = *paddr) != NULL) | |
564 | { | |
565 | BOOL timed_out = FALSE; | |
566 | retry_item *rti; | |
567 | ||
568 | for (addr = endaddr; addr != NULL; addr = addr->parent) | |
569 | { | |
570 | int update_count = 0; | |
571 | int timedout_count = 0; | |
572 | ||
573 | DEBUG(D_retry) debug_printf("%s%s\n", addr->address, (addr->retries == NULL)? | |
574 | ": no retry items" : ""); | |
575 | ||
576 | /* Loop for each retry item. */ | |
577 | ||
578 | for (rti = addr->retries; rti != NULL; rti = rti->next) | |
579 | { | |
580 | uschar *message; | |
581 | int message_length, message_space, failing_interval, next_try; | |
582 | retry_rule *rule, *final_rule; | |
583 | retry_config *retry; | |
584 | dbdata_retry *retry_record; | |
585 | ||
586 | /* Open the retry database if it is not already open; failure to open | |
587 | the file is logged, but otherwise ignored - deferred addresses will | |
588 | get retried at the next opportunity. Not opening earlier than this saves | |
589 | opening if no addresses have retry items - common when none have yet | |
590 | reached their retry next try time. */ | |
591 | ||
592 | if (dbm_file == NULL) | |
593 | dbm_file = dbfn_open(US"retry", O_RDWR, &dbblock, TRUE); | |
594 | ||
595 | if (dbm_file == NULL) | |
596 | { | |
597 | DEBUG(D_deliver|D_retry|D_hints_lookup) | |
598 | debug_printf("retry database not available for updating\n"); | |
599 | return; | |
600 | } | |
601 | ||
602 | /* If there are no deferred addresses, that is, if this message is | |
603 | completing, and the retry item is for a message-specific SMTP error, | |
604 | force it to be deleted, because there's no point in keeping data for | |
605 | no-longer-existing messages. This situation can occur when a domain has | |
606 | two hosts and a message-specific error occurs for the first of them, | |
607 | but the address gets delivered to the second one. This optimization | |
608 | doesn't succeed in cleaning out all the dead entries, but it helps. */ | |
609 | ||
610 | if (*addr_defer == NULL && (rti->flags & rf_message) != 0) | |
611 | rti->flags |= rf_delete; | |
612 | ||
613 | /* Handle the case of a request to delete the retry info for this | |
614 | destination. */ | |
615 | ||
616 | if ((rti->flags & rf_delete) != 0) | |
617 | { | |
618 | (void)dbfn_delete(dbm_file, rti->key); | |
619 | DEBUG(D_retry) | |
620 | debug_printf("deleted retry information for %s\n", rti->key); | |
621 | continue; | |
622 | } | |
623 | ||
624 | /* Count the number of non-delete retry items. This is so that we | |
625 | can compare it to the count of timed_out ones, to check whether | |
626 | all are timed out. */ | |
627 | ||
628 | update_count++; | |
629 | ||
630 | /* Get the retry information for this destination and error code, if | |
631 | any. If this item is for a remote host with ip address, then pass | |
632 | the domain name as an alternative to search for. If no retry | |
633 | information is found, we can't generate a retry time, so there is | |
634 | no point updating the database. This retry item is timed out. */ | |
635 | ||
636 | if ((retry = retry_find_config(rti->key + 2, | |
637 | ((rti->flags & rf_host) != 0)? addr->domain : NULL, | |
638 | rti->basic_errno, rti->more_errno)) == NULL) | |
639 | { | |
640 | DEBUG(D_retry) debug_printf("No configured retry item for %s%s%s\n", | |
641 | rti->key, | |
642 | ((rti->flags & rf_host) != 0)? US" or " : US"", | |
643 | ((rti->flags & rf_host) != 0)? addr->domain : US""); | |
644 | if (addr == endaddr) timedout_count++; | |
645 | continue; | |
646 | } | |
647 | ||
648 | DEBUG(D_retry) | |
649 | { | |
650 | if ((rti->flags & rf_host) != 0) | |
ea49d0e1 PH |
651 | debug_printf("retry for %s (%s) = %s %d %d\n", rti->key, |
652 | addr->domain, retry->pattern, retry->basic_errno, | |
653 | retry->more_errno); | |
059ec3d9 | 654 | else |
ea49d0e1 PH |
655 | debug_printf("retry for %s = %s %d %d\n", rti->key, retry->pattern, |
656 | retry->basic_errno, retry->more_errno); | |
059ec3d9 PH |
657 | } |
658 | ||
659 | /* Set up the message for the database retry record. Because DBM | |
660 | records have a maximum data length, we enforce a limit. There isn't | |
661 | much point in keeping a huge message here, anyway. */ | |
662 | ||
663 | message = (rti->basic_errno > 0)? US strerror(rti->basic_errno) : | |
664 | (rti->message == NULL)? | |
665 | US"unknown error" : string_printing(rti->message); | |
666 | message_length = Ustrlen(message); | |
667 | if (message_length > 150) message_length = 150; | |
668 | ||
669 | /* Read a retry record from the database or construct a new one. | |
670 | Ignore an old one if it is too old since it was last updated. */ | |
671 | ||
672 | retry_record = dbfn_read(dbm_file, rti->key); | |
673 | if (retry_record != NULL && | |
674 | now - retry_record->time_stamp > retry_data_expire) | |
675 | retry_record = NULL; | |
676 | ||
677 | if (retry_record == NULL) | |
678 | { | |
679 | retry_record = store_get(sizeof(dbdata_retry) + message_length); | |
680 | message_space = message_length; | |
681 | retry_record->first_failed = now; | |
682 | retry_record->last_try = now; | |
683 | retry_record->next_try = now; | |
684 | retry_record->expired = FALSE; | |
685 | retry_record->text[0] = 0; /* just in case */ | |
686 | } | |
687 | else message_space = Ustrlen(retry_record->text); | |
688 | ||
689 | /* Compute how long this destination has been failing */ | |
690 | ||
691 | failing_interval = now - retry_record->first_failed; | |
727071f8 PH |
692 | DEBUG(D_retry) debug_printf("failing_interval=%d message_age=%d\n", |
693 | failing_interval, message_age); | |
694 | ||
dd16e114 PH |
695 | /* For a non-host error, if the message has been on the queue longer |
696 | than the recorded time of failure, use the message's age instead. This | |
697 | can happen when some messages can be delivered and others cannot; a | |
698 | successful delivery will reset the first_failed time, and this can lead | |
699 | to a failing message being retried too often. */ | |
727071f8 | 700 | |
dd16e114 PH |
701 | if ((rti->flags & rf_host) == 0 && message_age > failing_interval) |
702 | failing_interval = message_age; | |
059ec3d9 PH |
703 | |
704 | /* Search for the current retry rule. The cutoff time of the | |
705 | last rule is handled differently to the others. The rule continues | |
706 | to operate for ever (the global maximum interval will eventually | |
707 | limit the gaps) but its cutoff time determines when an individual | |
708 | destination times out. If there are no retry rules, the destination | |
709 | always times out, but we can't compute a retry time. */ | |
710 | ||
711 | final_rule = NULL; | |
712 | for (rule = retry->rules; rule != NULL; rule = rule->next) | |
713 | { | |
714 | if (failing_interval <= rule->timeout) break; | |
715 | final_rule = rule; | |
716 | } | |
717 | ||
718 | /* If there's an un-timed out rule, the destination has not | |
719 | yet timed out, so the address as a whole has not timed out (but we are | |
720 | interested in this only for the end address). Make sure the expired | |
721 | flag is false (can be forced via fixdb from outside, but ensure it is | |
722 | consistent with the rules whenever we go through here). */ | |
723 | ||
724 | if (rule != NULL) | |
725 | { | |
726 | retry_record->expired = FALSE; | |
727 | } | |
728 | ||
729 | /* Otherwise, set the retry timeout expired, and set the final rule | |
730 | as the one from which to compute the next retry time. Subsequent | |
731 | messages will fail immediately until the retry time is reached (unless | |
732 | there are other, still active, retries). */ | |
733 | ||
734 | else | |
735 | { | |
736 | rule = final_rule; | |
737 | retry_record->expired = TRUE; | |
738 | if (addr == endaddr) timedout_count++; | |
739 | } | |
740 | ||
741 | /* There is a special case to consider when some messages get through | |
742 | to a destination and others don't. This can happen locally when a | |
743 | large message pushes a user over quota, and it can happen remotely | |
744 | when a machine is on a dodgy Internet connection. The messages that | |
745 | get through wipe the retry information, causing those that don't to | |
746 | stay on the queue longer than the final retry time. In order to | |
747 | avoid this, we check, using the time of arrival of the message, to | |
748 | see if it has been on the queue for more than the final cutoff time, | |
749 | and if so, cause this retry item to time out, and the retry time to | |
750 | be set to "now" so that any subsequent messages in the same condition | |
751 | also get tried. We search for the last rule onwards from the one that | |
752 | is in use. If there are no retry rules for the item, rule will be null | |
753 | and timedout_count will already have been updated. | |
754 | ||
755 | This implements "timeout this rule if EITHER the host (or routing or | |
756 | directing) has been failing for more than the maximum time, OR if the | |
727071f8 PH |
757 | message has been on the queue for more than the maximum time." |
758 | ||
759 | February 2006: It is possible that this code is no longer needed | |
760 | following the change to the retry calculation to use the message age if | |
761 | it is larger than the time since first failure. It may be that the | |
762 | expired flag is always set when the other conditions are met. However, | |
763 | this is a small bit of code, and it does no harm to leave it in place, | |
764 | just in case. */ | |
059ec3d9 PH |
765 | |
766 | if (received_time <= retry_record->first_failed && | |
767 | addr == endaddr && !retry_record->expired && rule != NULL) | |
768 | { | |
769 | retry_rule *last_rule; | |
770 | for (last_rule = rule; | |
771 | last_rule->next != NULL; | |
772 | last_rule = last_rule->next); | |
773 | if (now - received_time > last_rule->timeout) | |
774 | { | |
775 | DEBUG(D_retry) debug_printf("on queue longer than maximum retry\n"); | |
776 | timedout_count++; | |
777 | rule = NULL; | |
778 | } | |
779 | } | |
780 | ||
781 | /* Compute the next try time from the rule, subject to the global | |
782 | maximum, and update the retry database. If rule == NULL it means | |
783 | there were no rules at all (and the timeout will be set expired), | |
784 | or we have a message that is older than the final timeout. In this | |
785 | case set the next retry time to now, so that one delivery attempt | |
786 | happens for subsequent messages. */ | |
787 | ||
788 | if (rule == NULL) next_try = now; else | |
789 | { | |
790 | if (rule->rule == 'F') next_try = now + rule->p1; | |
6af56900 | 791 | else /* rule = 'G' or 'H' */ |
059ec3d9 PH |
792 | { |
793 | int last_predicted_gap = | |
794 | retry_record->next_try - retry_record->last_try; | |
795 | int last_actual_gap = now - retry_record->last_try; | |
796 | int lastgap = (last_predicted_gap < last_actual_gap)? | |
797 | last_predicted_gap : last_actual_gap; | |
6af56900 PH |
798 | int next_gap = (lastgap * rule->p2)/1000; |
799 | if (rule->rule == 'G') | |
800 | { | |
801 | next_try = now + ((lastgap < rule->p1)? rule->p1 : next_gap); | |
802 | } | |
803 | else /* The 'H' rule */ | |
804 | { | |
805 | next_try = now + rule->p1; | |
806 | if (next_gap > rule->p1) | |
3cd34f13 PH |
807 | next_try += random_number(next_gap - rule->p1)/2 + |
808 | (next_gap - rule->p1)/2; | |
6af56900 | 809 | } |
059ec3d9 PH |
810 | } |
811 | } | |
812 | ||
813 | /* Impose a global retry max */ | |
814 | ||
815 | if (next_try - now > retry_interval_max) | |
816 | next_try = now + retry_interval_max; | |
817 | ||
818 | /* If the new message length is greater than the previous one, we | |
819 | have to copy the record first. */ | |
820 | ||
821 | if (message_length > message_space) | |
822 | { | |
823 | dbdata_retry *newr = store_get(sizeof(dbdata_retry) + message_length); | |
824 | memcpy(newr, retry_record, sizeof(dbdata_retry)); | |
825 | retry_record = newr; | |
826 | } | |
827 | ||
828 | /* Set up the retry record; message_length may be less than the string | |
829 | length for very long error strings. */ | |
830 | ||
831 | retry_record->last_try = now; | |
832 | retry_record->next_try = next_try; | |
833 | retry_record->basic_errno = rti->basic_errno; | |
834 | retry_record->more_errno = rti->more_errno; | |
835 | Ustrncpy(retry_record->text, message, message_length); | |
836 | retry_record->text[message_length] = 0; | |
837 | ||
838 | DEBUG(D_retry) | |
839 | { | |
840 | int letter = retry_record->more_errno & 255; | |
841 | debug_printf("Writing retry data for %s\n", rti->key); | |
842 | debug_printf(" first failed=%d last try=%d next try=%d expired=%d\n", | |
843 | (int)retry_record->first_failed, (int)retry_record->last_try, | |
844 | (int)retry_record->next_try, retry_record->expired); | |
845 | debug_printf(" errno=%d more_errno=", retry_record->basic_errno); | |
846 | if (letter == 'A' || letter == 'M') | |
847 | debug_printf("%d,%c", (retry_record->more_errno >> 8) & 255, | |
848 | letter); | |
849 | else | |
850 | debug_printf("%d", retry_record->more_errno); | |
851 | debug_printf(" %s\n", retry_record->text); | |
852 | } | |
853 | ||
854 | (void)dbfn_write(dbm_file, rti->key, retry_record, | |
855 | sizeof(dbdata_retry) + message_length); | |
856 | } /* Loop for each retry item */ | |
857 | ||
858 | /* If all the non-delete retry items are timed out, the address is | |
859 | timed out, provided that we didn't skip any hosts because their retry | |
860 | time was not reached (or because of hosts_max_try). */ | |
861 | ||
862 | if (update_count > 0 && update_count == timedout_count) | |
863 | { | |
864 | if (!testflag(endaddr, af_retry_skipped)) | |
865 | { | |
866 | DEBUG(D_retry) debug_printf("timed out: all retries expired\n"); | |
867 | timed_out = TRUE; | |
868 | } | |
869 | else | |
870 | { | |
871 | DEBUG(D_retry) | |
872 | debug_printf("timed out but some hosts were skipped\n"); | |
873 | } | |
874 | } | |
875 | } /* Loop for an address and its parents */ | |
876 | ||
877 | /* If this is a deferred address, and retry processing was requested by | |
878 | means of one or more retry items, and they all timed out, move the address | |
879 | to the failed queue, and restart this loop without updating paddr. | |
880 | ||
881 | If there were several addresses batched in the same remote delivery, only | |
882 | the original top one will have host retry items attached to it, but we want | |
883 | to handle all the same. Each will have a pointer back to its "top" address, | |
884 | and they will now precede the item with the retries because addresses are | |
885 | inverted when added to these final queues. We have saved information about | |
886 | them in passing (below) so they can all be cut out at once. */ | |
887 | ||
888 | if (i == 2) /* Handling defers */ | |
889 | { | |
890 | if (endaddr->retries != NULL && timed_out) | |
891 | { | |
892 | if (last_first == endaddr) paddr = saved_paddr; | |
893 | addr = *paddr; | |
894 | *paddr = endaddr->next; | |
895 | ||
896 | endaddr->next = *addr_failed; | |
897 | *addr_failed = addr; | |
898 | ||
899 | for (;; addr = addr->next) | |
900 | { | |
901 | setflag(addr, af_retry_timedout); | |
902 | addr->message = (addr->message == NULL)? US"retry timeout exceeded" : | |
903 | string_sprintf("%s: retry timeout exceeded", addr->message); | |
fffffe4c PH |
904 | addr->user_message = (addr->user_message == NULL)? |
905 | US"retry timeout exceeded" : | |
906 | string_sprintf("%s: retry timeout exceeded", addr->user_message); | |
059ec3d9 PH |
907 | log_write(0, LOG_MAIN, "** %s%s%s%s: retry timeout exceeded", |
908 | addr->address, | |
909 | (addr->parent == NULL)? US"" : US" <", | |
910 | (addr->parent == NULL)? US"" : addr->parent->address, | |
911 | (addr->parent == NULL)? US"" : US">"); | |
912 | ||
913 | if (addr == endaddr) break; | |
914 | } | |
915 | ||
916 | continue; /* Restart from changed *paddr */ | |
917 | } | |
918 | ||
919 | /* This address is to remain on the defer chain. If it has a "first" | |
920 | pointer, save the pointer to it in case we want to fail the set of | |
921 | addresses when we get to the first one. */ | |
922 | ||
923 | if (endaddr->first != last_first) | |
924 | { | |
925 | last_first = endaddr->first; | |
926 | saved_paddr = paddr; | |
927 | } | |
928 | } | |
929 | ||
930 | /* All cases (succeed, fail, defer left on queue) */ | |
931 | ||
932 | paddr = &(endaddr->next); /* Advance to next address */ | |
933 | } /* Loop for all addresses */ | |
934 | } /* Loop for succeed, fail, defer */ | |
935 | ||
936 | /* Close and unlock the database */ | |
937 | ||
938 | if (dbm_file != NULL) dbfn_close(dbm_file); | |
939 | ||
940 | DEBUG(D_retry) debug_printf("end of retry processing\n"); | |
941 | } | |
942 | ||
943 | /* End of retry.c */ |