SpamAssassin: replace "backup" option on server spec with "pri=<N>". Bug 670
[exim.git] / src / src / spam.c
1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
4
5 /* Copyright (c) Tom Kistner <tom@duncanthrax.net> 2003-???? */
6 /* License: GPL */
7
8 /* Code for calling spamassassin's spamd. Called from acl.c. */
9
10 #include "exim.h"
11 #ifdef WITH_CONTENT_SCAN
12 #include "spam.h"
13
14 uschar spam_score_buffer[16];
15 uschar spam_score_int_buffer[16];
16 uschar spam_bar_buffer[128];
17 uschar spam_action_buffer[32];
18 uschar spam_report_buffer[32600];
19 uschar prev_user_name[128] = "";
20 int spam_ok = 0;
21 int spam_rc = 0;
22 uschar *prev_spamd_address_work = NULL;
23
24 static const uschar * loglabel = US"spam acl condition:";
25
26
27 static int
28 spamd_param_init(spamd_address_container *spamd)
29 {
30 /* default spamd server weight, time and priority value */
31 spamd->is_failed = FALSE;
32 spamd->weight = SPAMD_WEIGHT;
33 spamd->timeout = SPAMD_TIMEOUT;
34 spamd->retry = 0;
35 spamd->priority = 1;
36 return 0;
37 }
38
39
40 static int
41 spamd_param(const uschar *param, spamd_address_container *spamd)
42 {
43 static int timesinceday = -1;
44 const uschar * s;
45 const uschar * name;
46
47 /*XXX more clever parsing could discard embedded spaces? */
48
49 if (sscanf(param, "pri=%u", &spamd->priority))
50 return 0; /* OK */
51
52 if (sscanf(param, "weight=%u", &spamd->weight))
53 {
54 if (spamd->weight == 0) /* this server disabled: skip it */
55 return 1;
56 return 0; /* OK */
57 }
58
59 if (Ustrncmp(param, "time=", 5) == 0)
60 {
61 unsigned int start_h = 0, start_m = 0, start_s = 0;
62 unsigned int end_h = 24, end_m = 0, end_s = 0;
63 unsigned int time_start, time_end;
64 const uschar * end_string;
65
66 name = US"time";
67 s = param+5;
68 if ((end_string = Ustrchr(s, '-')))
69 {
70 end_string++;
71 if ( sscanf(CS end_string, "%u.%u.%u", &end_h, &end_m, &end_s) == 0
72 || sscanf(CS s, "%u.%u.%u", &start_h, &start_m, &start_s) == 0
73 )
74 goto badval;
75 }
76 else
77 goto badval;
78
79 if (timesinceday < 0)
80 {
81 time_t now = time(NULL);
82 struct tm *tmp = localtime(&now);
83 timesinceday = tmp->tm_hour*3600 + tmp->tm_min*60 + tmp->tm_sec;
84 }
85
86 time_start = start_h*3600 + start_m*60 + start_s;
87 time_end = end_h*3600 + end_m*60 + end_s;
88
89 if (timesinceday < time_start || timesinceday >= time_end)
90 return 1; /* skip spamd server */
91
92 return 0; /* OK */
93 }
94
95 if (Ustrcmp(param, "variant=rspamd") == 0)
96 {
97 spamd->is_rspamd = TRUE;
98 return 0;
99 }
100
101 if (Ustrncmp(param, "tmo=", 4) == 0)
102 {
103 int sec = readconf_readtime((s = param+4), '\0', FALSE);
104 name = US"timeout";
105 if (sec < 0)
106 goto badval;
107 spamd->timeout = sec;
108 return 0;
109 }
110
111 if (Ustrncmp(param, "retry=", 6) == 0)
112 {
113 int sec = readconf_readtime((s = param+6), '\0', FALSE);
114 name = US"retry";
115 if (sec < 0)
116 goto badval;
117 spamd->retry = sec;
118 return 0;
119 }
120
121 log_write(0, LOG_MAIN, "%s warning - invalid spamd parameter: '%s'",
122 loglabel, param);
123 return -1; /* syntax error */
124
125 badval:
126 log_write(0, LOG_MAIN,
127 "%s warning - invalid spamd %s value: '%s'", loglabel, name, s);
128 return -1; /* syntax error */
129 }
130
131
132 static int
133 spamd_get_server(spamd_address_container ** spamds, int num_servers)
134 {
135 unsigned int i;
136 spamd_address_container * sd;
137 long rnd, weights;
138 unsigned pri;
139 static BOOL srandomed = FALSE;
140
141 /* seedup, if we have only 1 server */
142 if (num_servers == 1)
143 return (spamds[0]->is_failed ? -1 : 0);
144
145 /* init ranmod */
146 if (!srandomed)
147 {
148 struct timeval tv;
149 gettimeofday(&tv, NULL);
150 srandom((unsigned int)(tv.tv_usec/1000));
151 srandomed = TRUE;
152 }
153
154 /* scan for highest pri */
155 for (pri = 0, i = 0; i < num_servers; i++)
156 {
157 sd = spamds[i];
158 if (!sd->is_failed && sd->priority > pri) pri = sd->priority;
159 }
160
161 /* get sum of weights */
162 for (weights = 0, i = 0; i < num_servers; i++)
163 {
164 sd = spamds[i];
165 if (!sd->is_failed && sd->priority == pri) weights += sd->weight;
166 }
167 if (weights == 0) /* all servers failed */
168 return -1;
169
170 for (rnd = random() % weights, i = 0; i < num_servers; i++)
171 {
172 sd = spamds[i];
173 if (!sd->is_failed && sd->priority == pri)
174 if ((rnd -= sd->weight) <= 0)
175 return i;
176 }
177
178 log_write(0, LOG_MAIN|LOG_PANIC,
179 "%s unknown error (memory/cpu corruption?)", loglabel);
180 return -1;
181 }
182
183
184 int
185 spam(const uschar **listptr)
186 {
187 int sep = 0;
188 const uschar *list = *listptr;
189 uschar *user_name;
190 uschar user_name_buffer[128];
191 unsigned long mbox_size;
192 FILE *mbox_file;
193 int spamd_sock = -1;
194 uschar spamd_buffer[32600];
195 int i, j, offset, result;
196 uschar spamd_version[8];
197 uschar spamd_short_result[8];
198 uschar spamd_score_char;
199 double spamd_threshold, spamd_score, spamd_reject_score;
200 int spamd_report_offset;
201 uschar *p,*q;
202 int override = 0;
203 time_t start;
204 size_t read, wrote;
205 struct sockaddr_un server;
206 #ifndef NO_POLL_H
207 struct pollfd pollfd;
208 #else /* Patch posted by Erik ? for OS X */
209 struct timeval select_tv; /* and applied by PH */
210 fd_set select_fd;
211 #endif
212 uschar *spamd_address_work;
213 spamd_address_container * sd;
214
215 /* stop compiler warning */
216 result = 0;
217
218 /* find the username from the option list */
219 if ((user_name = string_nextinlist(&list, &sep,
220 user_name_buffer,
221 sizeof(user_name_buffer))) == NULL)
222 {
223 /* no username given, this means no scanning should be done */
224 return FAIL;
225 }
226
227 /* if username is "0" or "false", do not scan */
228 if ( (Ustrcmp(user_name,"0") == 0) ||
229 (strcmpic(user_name,US"false") == 0) )
230 return FAIL;
231
232 /* if there is an additional option, check if it is "true" */
233 if (strcmpic(list,US"true") == 0)
234 /* in that case, always return true later */
235 override = 1;
236
237 /* expand spamd_address if needed */
238 if (*spamd_address == '$')
239 {
240 spamd_address_work = expand_string(spamd_address);
241 if (spamd_address_work == NULL)
242 {
243 log_write(0, LOG_MAIN|LOG_PANIC,
244 "%s spamd_address starts with $, but expansion failed: %s",
245 loglabel, expand_string_message);
246 return DEFER;
247 }
248 }
249 else
250 spamd_address_work = spamd_address;
251
252 DEBUG(D_acl) debug_printf("spamd: addrlist '%s'\n", spamd_address_work);
253
254 /* check if previous spamd_address was expanded and has changed. dump cached results if so */
255 if ( spam_ok
256 && prev_spamd_address_work != NULL
257 && Ustrcmp(prev_spamd_address_work, spamd_address_work) != 0
258 )
259 spam_ok = 0;
260
261 /* if we scanned for this username last time, just return */
262 if (spam_ok && Ustrcmp(prev_user_name, user_name) == 0)
263 return override ? OK : spam_rc;
264
265 /* make sure the eml mbox file is spooled up */
266 mbox_file = spool_mbox(&mbox_size, NULL);
267
268 if (mbox_file == NULL)
269 {
270 /* error while spooling */
271 log_write(0, LOG_MAIN|LOG_PANIC,
272 "%s error while creating mbox spool file", loglabel);
273 return DEFER;
274 }
275
276 start = time(NULL);
277
278 {
279 int num_servers = 0;
280 int current_server;
281 uschar * address;
282 const uschar * spamd_address_list_ptr = spamd_address_work;
283 spamd_address_container * spamd_address_vector[32];
284
285 /* Check how many spamd servers we have
286 and register their addresses */
287 sep = 0; /* default colon-sep */
288 while ((address = string_nextinlist(&spamd_address_list_ptr, &sep,
289 NULL, 0)) != NULL)
290 {
291 const uschar * sublist;
292 int sublist_sep = -(int)' '; /* default space-sep */
293 unsigned args;
294 uschar * s;
295
296 DEBUG(D_acl) debug_printf("spamd: addr entry '%s'\n", address);
297 sd = (spamd_address_container *)store_get(sizeof(spamd_address_container));
298
299 for (sublist = address, args = 0, spamd_param_init(sd);
300 s = string_nextinlist(&sublist, &sublist_sep, NULL, 0);
301 args++
302 )
303 {
304 DEBUG(D_acl) debug_printf("spamd: addr parm '%s'\n", s);
305 switch (args)
306 {
307 case 0: sd->hostspec = s;
308 if (*s == '/') args++; /* local; no port */
309 break;
310 case 1: sd->hostspec = string_sprintf("%s %s", sd->hostspec, s);
311 break;
312 default: spamd_param(s, sd);
313 break;
314 }
315 }
316 if (args < 2)
317 {
318 log_write(0, LOG_MAIN,
319 "%s warning - invalid spamd address: '%s'", loglabel, address);
320 continue;
321 }
322
323 spamd_address_vector[num_servers] = sd;
324 if (++num_servers > 31)
325 break;
326 }
327
328 /* check if we have at least one server */
329 if (!num_servers)
330 {
331 log_write(0, LOG_MAIN|LOG_PANIC,
332 "%s no useable spamd server addresses in spamd_address configuration option.",
333 loglabel);
334 goto defer;
335 }
336
337 current_server = spamd_get_server(spamd_address_vector, num_servers);
338 sd = spamd_address_vector[current_server];
339 for(;;)
340 {
341 uschar * errstr;
342
343 debug_printf("trying server %s\n", sd->hostspec);
344
345 for (;;)
346 {
347 if ( (spamd_sock = ip_streamsocket(sd->hostspec, &errstr, 5)) >= 0
348 || sd->retry <= 0
349 )
350 break;
351 debug_printf("server %s: retry conn\n", sd->hostspec);
352 while (sd->retry > 0) sd->retry = sleep(sd->retry);
353 }
354 if (spamd_sock >= 0)
355 break;
356
357 log_write(0, LOG_MAIN, "%s spamd: %s", loglabel, errstr);
358 sd->is_failed = TRUE;
359
360 current_server = spamd_get_server(spamd_address_vector, num_servers);
361 if (current_server < 0)
362 {
363 log_write(0, LOG_MAIN|LOG_PANIC, "%s all spamd servers failed", loglabel);
364 goto defer;
365 }
366 sd = spamd_address_vector[current_server];
367 }
368 }
369
370 if (spamd_sock == -1)
371 {
372 log_write(0, LOG_MAIN|LOG_PANIC,
373 "programming fault, spamd_sock unexpectedly unset");
374 goto defer;
375 }
376
377 (void)fcntl(spamd_sock, F_SETFL, O_NONBLOCK);
378 /* now we are connected to spamd on spamd_sock */
379 if (sd->is_rspamd)
380 { /* rspamd variant */
381 uschar *req_str;
382 const uschar * helo;
383 const uschar * fcrdns;
384 const uschar * authid;
385
386 req_str = string_sprintf("CHECK RSPAMC/1.3\r\nContent-length: %lu\r\n"
387 "Queue-Id: %s\r\nFrom: <%s>\r\nRecipient-Number: %d\r\n", mbox_size,
388 message_id, sender_address, recipients_count);
389 for (i = 0; i < recipients_count; i ++)
390 req_str = string_sprintf("%sRcpt: <%s>\r\n", req_str, recipients_list[i].address);
391 if ((helo = expand_string(US"$sender_helo_name")) != NULL && *helo != '\0')
392 req_str = string_sprintf("%sHelo: %s\r\n", req_str, helo);
393 if ((fcrdns = expand_string(US"$sender_host_name")) != NULL && *fcrdns != '\0')
394 req_str = string_sprintf("%sHostname: %s\r\n", req_str, fcrdns);
395 if (sender_host_address != NULL)
396 req_str = string_sprintf("%sIP: %s\r\n", req_str, sender_host_address);
397 if ((authid = expand_string(US"$authenticated_id")) != NULL && *authid != '\0')
398 req_str = string_sprintf("%sUser: %s\r\n", req_str, authid);
399 req_str = string_sprintf("%s\r\n", req_str);
400 wrote = send(spamd_sock, req_str, Ustrlen(req_str), 0);
401 }
402 else
403 { /* spamassassin variant */
404 (void)string_format(spamd_buffer,
405 sizeof(spamd_buffer),
406 "REPORT SPAMC/1.2\r\nUser: %s\r\nContent-length: %ld\r\n\r\n",
407 user_name,
408 mbox_size);
409 /* send our request */
410 wrote = send(spamd_sock, spamd_buffer, Ustrlen(spamd_buffer), 0);
411 }
412
413 if (wrote == -1)
414 {
415 (void)close(spamd_sock);
416 log_write(0, LOG_MAIN|LOG_PANIC,
417 "%s spamd send failed: %s", loglabel, strerror(errno));
418 goto defer;
419 }
420
421 /* now send the file */
422 /* spamd sometimes accepts conections but doesn't read data off
423 * the connection. We make the file descriptor non-blocking so
424 * that the write will only write sufficient data without blocking
425 * and we poll the desciptor to make sure that we can write without
426 * blocking. Short writes are gracefully handled and if the whole
427 * trasaction takes too long it is aborted.
428 * Note: poll() is not supported in OSX 10.2 and is reported to be
429 * broken in more recent versions (up to 10.4).
430 */
431 #ifndef NO_POLL_H
432 pollfd.fd = spamd_sock;
433 pollfd.events = POLLOUT;
434 #endif
435 (void)fcntl(spamd_sock, F_SETFL, O_NONBLOCK);
436 do
437 {
438 read = fread(spamd_buffer,1,sizeof(spamd_buffer),mbox_file);
439 if (read > 0)
440 {
441 offset = 0;
442 again:
443 #ifndef NO_POLL_H
444 result = poll(&pollfd, 1, 1000);
445
446 /* Patch posted by Erik ? for OS X and applied by PH */
447 #else
448 select_tv.tv_sec = 1;
449 select_tv.tv_usec = 0;
450 FD_ZERO(&select_fd);
451 FD_SET(spamd_sock, &select_fd);
452 result = select(spamd_sock+1, NULL, &select_fd, NULL, &select_tv);
453 #endif
454 /* End Erik's patch */
455
456 if (result == -1 && errno == EINTR)
457 goto again;
458 else if (result < 1)
459 {
460 if (result == -1)
461 log_write(0, LOG_MAIN|LOG_PANIC,
462 "%s %s on spamd socket", loglabel, strerror(errno));
463 else
464 {
465 if (time(NULL) - start < sd->timeout)
466 goto again;
467 log_write(0, LOG_MAIN|LOG_PANIC,
468 "%s timed out writing spamd socket", loglabel);
469 }
470 (void)close(spamd_sock);
471 goto defer;
472 }
473
474 wrote = send(spamd_sock,spamd_buffer + offset,read - offset,0);
475 if (wrote == -1)
476 {
477 log_write(0, LOG_MAIN|LOG_PANIC,
478 "%s %s on spamd socket", loglabel, strerror(errno));
479 (void)close(spamd_sock);
480 goto defer;
481 }
482 if (offset + wrote != read)
483 {
484 offset += wrote;
485 goto again;
486 }
487 }
488 }
489 while (!feof(mbox_file) && !ferror(mbox_file));
490
491 if (ferror(mbox_file))
492 {
493 log_write(0, LOG_MAIN|LOG_PANIC,
494 "%s error reading spool file: %s", loglabel, strerror(errno));
495 (void)close(spamd_sock);
496 goto defer;
497 }
498
499 (void)fclose(mbox_file);
500
501 /* we're done sending, close socket for writing */
502 shutdown(spamd_sock,SHUT_WR);
503
504 /* read spamd response using what's left of the timeout. */
505 memset(spamd_buffer, 0, sizeof(spamd_buffer));
506 offset = 0;
507 while ((i = ip_recv(spamd_sock,
508 spamd_buffer + offset,
509 sizeof(spamd_buffer) - offset - 1,
510 sd->timeout - time(NULL) + start)) > 0 )
511 offset += i;
512
513 /* error handling */
514 if (i <= 0 && errno != 0)
515 {
516 log_write(0, LOG_MAIN|LOG_PANIC,
517 "%s error reading from spamd socket: %s", loglabel, strerror(errno));
518 (void)close(spamd_sock);
519 return DEFER;
520 }
521
522 /* reading done */
523 (void)close(spamd_sock);
524
525 if (sd->is_rspamd)
526 { /* rspamd variant of reply */
527 int r;
528 if ((r = sscanf(CS spamd_buffer,
529 "RSPAMD/%7s 0 EX_OK\r\nMetric: default; %7s %lf / %lf / %lf\r\n%n",
530 spamd_version, spamd_short_result, &spamd_score, &spamd_threshold,
531 &spamd_reject_score, &spamd_report_offset)) != 5)
532 {
533 log_write(0, LOG_MAIN|LOG_PANIC,
534 "%s cannot parse spamd output: %d", loglabel, r);
535 return DEFER;
536 }
537 /* now parse action */
538 p = &spamd_buffer[spamd_report_offset];
539
540 if (Ustrncmp(p, "Action: ", sizeof("Action: ") - 1) == 0)
541 {
542 p += sizeof("Action: ") - 1;
543 q = &spam_action_buffer[0];
544 while (*p && *p != '\r' && (q - spam_action_buffer) < sizeof(spam_action_buffer) - 1)
545 *q++ = *p++;
546 *q = '\0';
547 }
548 }
549 else
550 { /* spamassassin */
551 /* dig in the spamd output and put the report in a multiline header,
552 if requested */
553 if (sscanf(CS spamd_buffer,
554 "SPAMD/%7s 0 EX_OK\r\nContent-length: %*u\r\n\r\n%lf/%lf\r\n%n",
555 spamd_version,&spamd_score,&spamd_threshold,&spamd_report_offset) != 3)
556 {
557 /* try to fall back to pre-2.50 spamd output */
558 if (sscanf(CS spamd_buffer,
559 "SPAMD/%7s 0 EX_OK\r\nSpam: %*s ; %lf / %lf\r\n\r\n%n",
560 spamd_version,&spamd_score,&spamd_threshold,&spamd_report_offset) != 3)
561 {
562 log_write(0, LOG_MAIN|LOG_PANIC,
563 "%s cannot parse spamd output", loglabel);
564 return DEFER;
565 }
566 }
567
568 Ustrcpy(spam_action_buffer,
569 spamd_score >= spamd_threshold ? "reject" : "no action");
570 }
571
572 /* Create report. Since this is a multiline string,
573 we must hack it into shape first */
574 p = &spamd_buffer[spamd_report_offset];
575 q = spam_report_buffer;
576 while (*p != '\0')
577 {
578 /* skip \r */
579 if (*p == '\r')
580 {
581 p++;
582 continue;
583 }
584 *q++ = *p;
585 if (*p++ == '\n')
586 {
587 /* add an extra space after the newline to ensure
588 that it is treated as a header continuation line */
589 *q++ = ' ';
590 }
591 }
592 /* NULL-terminate */
593 *q-- = '\0';
594 /* cut off trailing leftovers */
595 while (*q <= ' ')
596 *q-- = '\0';
597
598 spam_report = spam_report_buffer;
599 spam_action = spam_action_buffer;
600
601 /* create spam bar */
602 spamd_score_char = spamd_score > 0 ? '+' : '-';
603 j = abs((int)(spamd_score));
604 i = 0;
605 if (j != 0)
606 while ((i < j) && (i <= MAX_SPAM_BAR_CHARS))
607 spam_bar_buffer[i++] = spamd_score_char;
608 else
609 {
610 spam_bar_buffer[0] = '/';
611 i = 1;
612 }
613 spam_bar_buffer[i] = '\0';
614 spam_bar = spam_bar_buffer;
615
616 /* create "float" spam score */
617 (void)string_format(spam_score_buffer, sizeof(spam_score_buffer),
618 "%.1f", spamd_score);
619 spam_score = spam_score_buffer;
620
621 /* create "int" spam score */
622 j = (int)((spamd_score + 0.001)*10);
623 (void)string_format(spam_score_int_buffer, sizeof(spam_score_int_buffer),
624 "%d", j);
625 spam_score_int = spam_score_int_buffer;
626
627 /* compare threshold against score */
628 spam_rc = spamd_score >= spamd_threshold
629 ? OK /* spam as determined by user's threshold */
630 : FAIL; /* not spam */
631
632 /* remember expanded spamd_address if needed */
633 if (spamd_address_work != spamd_address)
634 prev_spamd_address_work = string_copy(spamd_address_work);
635
636 /* remember user name and "been here" for it */
637 Ustrcpy(prev_user_name, user_name);
638 spam_ok = 1;
639
640 return override
641 ? OK /* always return OK, no matter what the score */
642 : spam_rc;
643
644 defer:
645 (void)fclose(mbox_file);
646 return DEFER;
647 }
648
649 #endif
650 /* vi: aw ai sw=2
651 */