Copyright year bumps for substantive changes 2017
[exim.git] / src / src / lookups / pgsql.c
1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
4
5 /* Copyright (c) University of Cambridge 1995 - 2017 */
6 /* See the file NOTICE for conditions of use and distribution. */
7
8 /* Thanks to Petr Cech for contributing the original code for these
9 functions. Thanks to Joachim Wieland for the initial patch for the Unix domain
10 socket extension. */
11
12 #include "../exim.h"
13 #include "lf_functions.h"
14
15 #include <libpq-fe.h> /* The system header */
16
17 /* Structure and anchor for caching connections. */
18
19 typedef struct pgsql_connection {
20 struct pgsql_connection *next;
21 uschar *server;
22 PGconn *handle;
23 } pgsql_connection;
24
25 static pgsql_connection *pgsql_connections = NULL;
26
27
28
29 /*************************************************
30 * Open entry point *
31 *************************************************/
32
33 /* See local README for interface description. */
34
35 static void *
36 pgsql_open(uschar *filename, uschar **errmsg)
37 {
38 return (void *)(1); /* Just return something non-null */
39 }
40
41
42
43 /*************************************************
44 * Tidy entry point *
45 *************************************************/
46
47 /* See local README for interface description. */
48
49 static void
50 pgsql_tidy(void)
51 {
52 pgsql_connection *cn;
53 while ((cn = pgsql_connections) != NULL)
54 {
55 pgsql_connections = cn->next;
56 DEBUG(D_lookup) debug_printf("close PGSQL connection: %s\n", cn->server);
57 PQfinish(cn->handle);
58 }
59 }
60
61
62 /*************************************************
63 * Notice processor function for pgsql *
64 *************************************************/
65
66 /* This function is passed to pgsql below, and called for any PostgreSQL
67 "notices". By default they are written to stderr, which is undesirable.
68
69 Arguments:
70 arg an opaque user cookie (not used)
71 message the notice
72
73 Returns: nothing
74 */
75
76 static void
77 notice_processor(void *arg, const char *message)
78 {
79 arg = arg; /* Keep compiler happy */
80 DEBUG(D_lookup) debug_printf("PGSQL: %s\n", message);
81 }
82
83
84
85 /*************************************************
86 * Internal search function *
87 *************************************************/
88
89 /* This function is called from the find entry point to do the search for a
90 single server. The server string is of the form "server/dbname/user/password".
91
92 PostgreSQL supports connections through Unix domain sockets. This is usually
93 faster and costs less cpu time than a TCP/IP connection. However it can only be
94 used if the mail server runs on the same machine as the database server. A
95 configuration line for PostgreSQL via Unix domain sockets looks like this:
96
97 hide pgsql_servers = (/tmp/.s.PGSQL.5432)/db/user/password[:<nextserver>]
98
99 We enclose the path name in parentheses so that its slashes aren't visually
100 confused with the delimiters for the other pgsql_server settings.
101
102 For TCP/IP connections, the server is a host name and optional port (with a
103 colon separator).
104
105 NOTE:
106 1) All three '/' must be present.
107 2) If host is omitted the local unix socket is used.
108
109 Arguments:
110 query the query string
111 server the server string; this is in dynamic memory and can be updated
112 resultptr where to store the result
113 errmsg where to point an error message
114 defer_break set TRUE if no more servers are to be tried after DEFER
115 do_cache set FALSE if data is changed
116
117 Returns: OK, FAIL, or DEFER
118 */
119
120 static int
121 perform_pgsql_search(const uschar *query, uschar *server, uschar **resultptr,
122 uschar **errmsg, BOOL *defer_break, uint *do_cache)
123 {
124 PGconn *pg_conn = NULL;
125 PGresult *pg_result = NULL;
126
127 int i;
128 gstring * result = NULL;
129 int yield = DEFER;
130 unsigned int num_fields, num_tuples;
131 pgsql_connection *cn;
132 uschar *server_copy = NULL;
133 uschar *sdata[3];
134
135 /* Disaggregate the parameters from the server argument. The order is host or
136 path, database, user, password. We can write to the string, since it is in a
137 nextinlist temporary buffer. The copy of the string that is used for caching
138 has the password removed. This copy is also used for debugging output. */
139
140 for (i = 2; i >= 0; i--)
141 {
142 uschar *pp = Ustrrchr(server, '/');
143 if (pp == NULL)
144 {
145 *errmsg = string_sprintf("incomplete pgSQL server data: %s",
146 (i == 2)? server : server_copy);
147 *defer_break = TRUE;
148 return DEFER;
149 }
150 *pp++ = 0;
151 sdata[i] = pp;
152 if (i == 2) server_copy = string_copy(server); /* sans password */
153 }
154
155 /* The total server string has now been truncated so that what is left at the
156 start is the identification of the server (host or path). See if we have a
157 cached connection to the server. */
158
159 for (cn = pgsql_connections; cn != NULL; cn = cn->next)
160 {
161 if (Ustrcmp(cn->server, server_copy) == 0)
162 {
163 pg_conn = cn->handle;
164 break;
165 }
166 }
167
168 /* If there is no cached connection, we must set one up. */
169
170 if (cn == NULL)
171 {
172 uschar *port = US"";
173
174 /* For a Unix domain socket connection, the path is in parentheses */
175
176 if (*server == '(')
177 {
178 uschar *last_slash, *last_dot, *p;
179
180 p = ++server;
181 while (*p != 0 && *p != ')') p++;
182 *p = 0;
183
184 last_slash = Ustrrchr(server, '/');
185 last_dot = Ustrrchr(server, '.');
186
187 DEBUG(D_lookup) debug_printf("PGSQL new connection: socket=%s "
188 "database=%s user=%s\n", server, sdata[0], sdata[1]);
189
190 /* A valid socket name looks like this: /var/run/postgresql/.s.PGSQL.5432
191 We have to call PQsetdbLogin with '/var/run/postgresql' as the hostname
192 argument and put '5432' into the port variable. */
193
194 if (last_slash == NULL || last_dot == NULL)
195 {
196 *errmsg = string_sprintf("PGSQL invalid filename for socket: %s",
197 server);
198 *defer_break = TRUE;
199 return DEFER;
200 }
201
202 /* Terminate the path name and set up the port: we'll have something like
203 server = "/var/run/postgresql" and port = "5432". */
204
205 *last_slash = 0;
206 port = last_dot + 1;
207 }
208
209 /* Host connection; sort out the port */
210
211 else
212 {
213 uschar *p;
214 if ((p = Ustrchr(server, ':')) != NULL)
215 {
216 *p++ = 0;
217 port = p;
218 }
219
220 if (Ustrchr(server, '/') != NULL)
221 {
222 *errmsg = string_sprintf("unexpected slash in pgSQL server hostname: %s",
223 server);
224 *defer_break = TRUE;
225 return DEFER;
226 }
227
228 DEBUG(D_lookup) debug_printf("PGSQL new connection: host=%s port=%s "
229 "database=%s user=%s\n", server, port, sdata[0], sdata[1]);
230 }
231
232 /* If the database is the empty string, set it NULL - the query must then
233 define it. */
234
235 if (sdata[0][0] == 0) sdata[0] = NULL;
236
237 /* Get store for a new handle, initialize it, and connect to the server */
238
239 pg_conn=PQsetdbLogin(
240 /* host port options tty database user passwd */
241 CS server, CS port, NULL, NULL, CS sdata[0], CS sdata[1], CS sdata[2]);
242
243 if(PQstatus(pg_conn) == CONNECTION_BAD)
244 {
245 store_reset(server_copy);
246 *errmsg = string_sprintf("PGSQL connection failed: %s",
247 PQerrorMessage(pg_conn));
248 PQfinish(pg_conn);
249 goto PGSQL_EXIT;
250 }
251
252 /* Set the client encoding to SQL_ASCII, which means that the server will
253 not try to interpret the query as being in any fancy encoding such as UTF-8
254 or other multibyte code that might cause problems with escaping. */
255
256 PQsetClientEncoding(pg_conn, "SQL_ASCII");
257
258 /* Set the notice processor to prevent notices from being written to stderr
259 (which is what the default does). Our function (above) just produces debug
260 output. */
261
262 PQsetNoticeProcessor(pg_conn, notice_processor, NULL);
263
264 /* Add the connection to the cache */
265
266 cn = store_get(sizeof(pgsql_connection));
267 cn->server = server_copy;
268 cn->handle = pg_conn;
269 cn->next = pgsql_connections;
270 pgsql_connections = cn;
271 }
272
273 /* Else use a previously cached connection */
274
275 else
276 {
277 DEBUG(D_lookup) debug_printf("PGSQL using cached connection for %s\n",
278 server_copy);
279 }
280
281 /* Run the query */
282
283 pg_result = PQexec(pg_conn, CS query);
284 switch(PQresultStatus(pg_result))
285 {
286 case PGRES_EMPTY_QUERY:
287 case PGRES_COMMAND_OK:
288 /* The command was successful but did not return any data since it was
289 not SELECT but either an INSERT, UPDATE or DELETE statement. Tell the
290 high level code to not cache this query, and clean the current cache for
291 this handle by setting *do_cache zero. */
292
293 result = string_cat(result, US PQcmdTuples(pg_result));
294 *do_cache = 0;
295 DEBUG(D_lookup) debug_printf("PGSQL: command does not return any data "
296 "but was successful. Rows affected: %s\n", result->s);
297 break;
298
299 case PGRES_TUPLES_OK:
300 break;
301
302 default:
303 /* This was the original code:
304 *errmsg = string_sprintf("PGSQL: query failed: %s\n",
305 PQresultErrorMessage(pg_result));
306 This was suggested by a user:
307 */
308
309 *errmsg = string_sprintf("PGSQL: query failed: %s (%s) (%s)\n",
310 PQresultErrorMessage(pg_result),
311 PQresStatus(PQresultStatus(pg_result)), query);
312 goto PGSQL_EXIT;
313 }
314
315 /* Result is in pg_result. Find the number of fields returned. If this is one,
316 we don't add field names to the data. Otherwise we do. If the query did not
317 return anything we skip the for loop; this also applies to the case
318 PGRES_COMMAND_OK. */
319
320 num_fields = PQnfields(pg_result);
321 num_tuples = PQntuples(pg_result);
322
323 /* Get the fields and construct the result string. If there is more than one
324 row, we insert '\n' between them. */
325
326 for (i = 0; i < num_tuples; i++)
327 {
328 if (result)
329 result = string_catn(result, US"\n", 1);
330
331 if (num_fields == 1)
332 result = string_catn(NULL,
333 US PQgetvalue(pg_result, i, 0), PQgetlength(pg_result, i, 0));
334 else
335 {
336 int j;
337 for (j = 0; j < num_fields; j++)
338 {
339 uschar *tmp = US PQgetvalue(pg_result, i, j);
340 result = lf_quote(US PQfname(pg_result, j), tmp, Ustrlen(tmp), result);
341 }
342 }
343 }
344
345 /* If result is NULL then no data has been found and so we return FAIL.
346 Otherwise, we must terminate the string which has been built; string_cat()
347 always leaves enough room for a terminating zero. */
348
349 if (!result)
350 {
351 yield = FAIL;
352 *errmsg = US"PGSQL: no data found";
353 }
354 else
355 store_reset(result->s + result->ptr + 1);
356
357 /* Get here by goto from various error checks. */
358
359 PGSQL_EXIT:
360
361 /* Free store for any result that was got; don't close the connection, as
362 it is cached. */
363
364 if (pg_result) PQclear(pg_result);
365
366 /* Non-NULL result indicates a successful result */
367
368 if (result)
369 {
370 *resultptr = string_from_gstring(result);
371 return OK;
372 }
373 else
374 {
375 DEBUG(D_lookup) debug_printf("%s\n", *errmsg);
376 return yield; /* FAIL or DEFER */
377 }
378 }
379
380
381
382
383 /*************************************************
384 * Find entry point *
385 *************************************************/
386
387 /* See local README for interface description. The handle and filename
388 arguments are not used. The code to loop through a list of servers while the
389 query is deferred with a retryable error is now in a separate function that is
390 shared with other SQL lookups. */
391
392 static int
393 pgsql_find(void *handle, uschar *filename, const uschar *query, int length,
394 uschar **result, uschar **errmsg, uint *do_cache)
395 {
396 return lf_sqlperform(US"PostgreSQL", US"pgsql_servers", pgsql_servers, query,
397 result, errmsg, do_cache, perform_pgsql_search);
398 }
399
400
401
402 /*************************************************
403 * Quote entry point *
404 *************************************************/
405
406 /* The characters that always need to be quoted (with backslash) are newline,
407 tab, carriage return, backspace, backslash itself, and the quote characters.
408
409 The original code quoted single quotes as \' which is documented as valid in
410 the O'Reilly book "Practical PostgreSQL" (first edition) as an alternative to
411 the SQL standard '' way of representing a single quote as data. However, in
412 June 2006 there was some security issue with using \' and so this has been
413 changed.
414
415 [Note: There is a function called PQescapeStringConn() that quotes strings.
416 This cannot be used because it needs a PGconn argument (the connection handle).
417 Why, I don't know. Seems odd for just string escaping...]
418
419 Arguments:
420 s the string to be quoted
421 opt additional option text or NULL if none
422
423 Returns: the processed string or NULL for a bad option
424 */
425
426 static uschar *
427 pgsql_quote(uschar *s, uschar *opt)
428 {
429 register int c;
430 int count = 0;
431 uschar *t = s;
432 uschar *quoted;
433
434 if (opt != NULL) return NULL; /* No options recognized */
435
436 while ((c = *t++) != 0)
437 if (Ustrchr("\n\t\r\b\'\"\\", c) != NULL) count++;
438
439 if (count == 0) return s;
440 t = quoted = store_get(Ustrlen(s) + count + 1);
441
442 while ((c = *s++) != 0)
443 {
444 if (c == '\'')
445 {
446 *t++ = '\'';
447 *t++ = '\'';
448 }
449 else if (Ustrchr("\n\t\r\b\"\\", c) != NULL)
450 {
451 *t++ = '\\';
452 switch(c)
453 {
454 case '\n': *t++ = 'n';
455 break;
456 case '\t': *t++ = 't';
457 break;
458 case '\r': *t++ = 'r';
459 break;
460 case '\b': *t++ = 'b';
461 break;
462 default: *t++ = c;
463 break;
464 }
465 }
466 else *t++ = c;
467 }
468
469 *t = 0;
470 return quoted;
471 }
472
473
474 /*************************************************
475 * Version reporting entry point *
476 *************************************************/
477
478 /* See local README for interface description. */
479
480 #include "../version.h"
481
482 void
483 pgsql_version_report(FILE *f)
484 {
485 #ifdef DYNLOOKUP
486 fprintf(f, "Library version: PostgreSQL: Exim version %s\n", EXIM_VERSION_STR);
487 #endif
488
489 /* Version reporting: there appears to be no available information about
490 the client library in libpq-fe.h; once you have a connection object, you
491 can access the server version and the chosen protocol version, but those
492 aren't really what we want. It might make sense to debug_printf those
493 when the connection is established though? */
494 }
495
496
497 static lookup_info _lookup_info = {
498 US"pgsql", /* lookup name */
499 lookup_querystyle, /* query-style lookup */
500 pgsql_open, /* open function */
501 NULL, /* no check function */
502 pgsql_find, /* find function */
503 NULL, /* no close function */
504 pgsql_tidy, /* tidy function */
505 pgsql_quote, /* quoting function */
506 pgsql_version_report /* version reporting */
507 };
508
509 #ifdef DYNLOOKUP
510 #define pgsql_lookup_module_info _lookup_module_info
511 #endif
512
513 static lookup_info *_lookup_list[] = { &_lookup_info };
514 lookup_module_info pgsql_lookup_module_info = { LOOKUP_MODULE_INFO_MAGIC, _lookup_list, 1 };
515
516 /* End of lookups/pgsql.c */