Copyright updates:
[exim.git] / src / src / lookups / pgsql.c
1 /*************************************************
2 * Exim - an Internet mail transport agent *
3 *************************************************/
4
5 /* Copyright (c) University of Cambridge 1995 - 2018 */
6 /* Copyright (c) The Exim Maintainers 2020 */
7 /* See the file NOTICE for conditions of use and distribution. */
8
9 /* Thanks to Petr Cech for contributing the original code for these
10 functions. Thanks to Joachim Wieland for the initial patch for the Unix domain
11 socket extension. */
12
13 #include "../exim.h"
14 #include "lf_functions.h"
15
16 #include <libpq-fe.h> /* The system header */
17
18 /* Structure and anchor for caching connections. */
19
20 typedef struct pgsql_connection {
21 struct pgsql_connection *next;
22 uschar *server;
23 PGconn *handle;
24 } pgsql_connection;
25
26 static pgsql_connection *pgsql_connections = NULL;
27
28
29
30 /*************************************************
31 * Open entry point *
32 *************************************************/
33
34 /* See local README for interface description. */
35
36 static void *
37 pgsql_open(const uschar * filename, uschar ** errmsg)
38 {
39 return (void *)(1); /* Just return something non-null */
40 }
41
42
43
44 /*************************************************
45 * Tidy entry point *
46 *************************************************/
47
48 /* See local README for interface description. */
49
50 static void
51 pgsql_tidy(void)
52 {
53 pgsql_connection *cn;
54 while ((cn = pgsql_connections) != NULL)
55 {
56 pgsql_connections = cn->next;
57 DEBUG(D_lookup) debug_printf_indent("close PGSQL connection: %s\n", cn->server);
58 PQfinish(cn->handle);
59 }
60 }
61
62
63 /*************************************************
64 * Notice processor function for pgsql *
65 *************************************************/
66
67 /* This function is passed to pgsql below, and called for any PostgreSQL
68 "notices". By default they are written to stderr, which is undesirable.
69
70 Arguments:
71 arg an opaque user cookie (not used)
72 message the notice
73
74 Returns: nothing
75 */
76
77 static void
78 notice_processor(void *arg, const char *message)
79 {
80 arg = arg; /* Keep compiler happy */
81 DEBUG(D_lookup) debug_printf_indent("PGSQL: %s\n", message);
82 }
83
84
85
86 /*************************************************
87 * Internal search function *
88 *************************************************/
89
90 /* This function is called from the find entry point to do the search for a
91 single server. The server string is of the form "server/dbname/user/password".
92
93 PostgreSQL supports connections through Unix domain sockets. This is usually
94 faster and costs less cpu time than a TCP/IP connection. However it can only be
95 used if the mail server runs on the same machine as the database server. A
96 configuration line for PostgreSQL via Unix domain sockets looks like this:
97
98 hide pgsql_servers = (/tmp/.s.PGSQL.5432)/db/user/password[:<nextserver>]
99
100 We enclose the path name in parentheses so that its slashes aren't visually
101 confused with the delimiters for the other pgsql_server settings.
102
103 For TCP/IP connections, the server is a host name and optional port (with a
104 colon separator).
105
106 NOTE:
107 1) All three '/' must be present.
108 2) If host is omitted the local unix socket is used.
109
110 Arguments:
111 query the query string
112 server the server string; this is in dynamic memory and can be updated
113 resultptr where to store the result
114 errmsg where to point an error message
115 defer_break set TRUE if no more servers are to be tried after DEFER
116 do_cache set FALSE if data is changed
117 opts options list
118
119 Returns: OK, FAIL, or DEFER
120 */
121
122 static int
123 perform_pgsql_search(const uschar *query, uschar *server, uschar **resultptr,
124 uschar **errmsg, BOOL *defer_break, uint *do_cache, const uschar * opts)
125 {
126 PGconn *pg_conn = NULL;
127 PGresult *pg_result = NULL;
128
129 gstring * result = NULL;
130 int yield = DEFER;
131 unsigned int num_fields, num_tuples;
132 pgsql_connection *cn;
133 rmark reset_point = store_mark();
134 uschar *server_copy = NULL;
135 uschar *sdata[3];
136
137 /* Disaggregate the parameters from the server argument. The order is host or
138 path, database, user, password. We can write to the string, since it is in a
139 nextinlist temporary buffer. The copy of the string that is used for caching
140 has the password removed. This copy is also used for debugging output. */
141
142 for (int i = 2; i >= 0; i--)
143 {
144 uschar *pp = Ustrrchr(server, '/');
145 if (!pp)
146 {
147 *errmsg = string_sprintf("incomplete pgSQL server data: %s",
148 (i == 2)? server : server_copy);
149 *defer_break = TRUE;
150 return DEFER;
151 }
152 *pp++ = 0;
153 sdata[i] = pp;
154 if (i == 2) server_copy = string_copy(server); /* sans password */
155 }
156
157 /* The total server string has now been truncated so that what is left at the
158 start is the identification of the server (host or path). See if we have a
159 cached connection to the server. */
160
161 for (cn = pgsql_connections; cn; cn = cn->next)
162 if (Ustrcmp(cn->server, server_copy) == 0)
163 {
164 pg_conn = cn->handle;
165 break;
166 }
167
168 /* If there is no cached connection, we must set one up. */
169
170 if (!cn)
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 && *p != ')') p++;
182 *p = 0;
183
184 last_slash = Ustrrchr(server, '/');
185 last_dot = Ustrrchr(server, '.');
186
187 DEBUG(D_lookup) debug_printf_indent("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 || !last_dot)
195 {
196 *errmsg = string_sprintf("PGSQL invalid filename for socket: %s", server);
197 *defer_break = TRUE;
198 return DEFER;
199 }
200
201 /* Terminate the path name and set up the port: we'll have something like
202 server = "/var/run/postgresql" and port = "5432". */
203
204 *last_slash = 0;
205 port = last_dot + 1;
206 }
207
208 /* Host connection; sort out the port */
209
210 else
211 {
212 uschar *p;
213 if ((p = Ustrchr(server, ':')))
214 {
215 *p++ = 0;
216 port = p;
217 }
218
219 if (Ustrchr(server, '/'))
220 {
221 *errmsg = string_sprintf("unexpected slash in pgSQL server hostname: %s",
222 server);
223 *defer_break = TRUE;
224 return DEFER;
225 }
226
227 DEBUG(D_lookup) debug_printf_indent("PGSQL new connection: host=%s port=%s "
228 "database=%s user=%s\n", server, port, sdata[0], sdata[1]);
229 }
230
231 /* If the database is the empty string, set it NULL - the query must then
232 define it. */
233
234 if (sdata[0][0] == 0) sdata[0] = NULL;
235
236 /* Get store for a new handle, initialize it, and connect to the server */
237
238 pg_conn=PQsetdbLogin(
239 /* host port options tty database user passwd */
240 CS server, CS port, NULL, NULL, CS sdata[0], CS sdata[1], CS sdata[2]);
241
242 if(PQstatus(pg_conn) == CONNECTION_BAD)
243 {
244 reset_point = store_reset(reset_point);
245 *errmsg = string_sprintf("PGSQL connection failed: %s",
246 PQerrorMessage(pg_conn));
247 PQfinish(pg_conn);
248 goto PGSQL_EXIT;
249 }
250
251 /* Set the client encoding to SQL_ASCII, which means that the server will
252 not try to interpret the query as being in any fancy encoding such as UTF-8
253 or other multibyte code that might cause problems with escaping. */
254
255 PQsetClientEncoding(pg_conn, "SQL_ASCII");
256
257 /* Set the notice processor to prevent notices from being written to stderr
258 (which is what the default does). Our function (above) just produces debug
259 output. */
260
261 PQsetNoticeProcessor(pg_conn, notice_processor, NULL);
262
263 /* Add the connection to the cache */
264
265 cn = store_get(sizeof(pgsql_connection), FALSE);
266 cn->server = server_copy;
267 cn->handle = pg_conn;
268 cn->next = pgsql_connections;
269 pgsql_connections = cn;
270 }
271
272 /* Else use a previously cached connection */
273
274 else
275 {
276 DEBUG(D_lookup) debug_printf_indent("PGSQL using cached connection for %s\n",
277 server_copy);
278 }
279
280 /* Run the query */
281
282 pg_result = PQexec(pg_conn, CS query);
283 switch(PQresultStatus(pg_result))
284 {
285 case PGRES_EMPTY_QUERY:
286 case PGRES_COMMAND_OK:
287 /* The command was successful but did not return any data since it was
288 not SELECT but either an INSERT, UPDATE or DELETE statement. Tell the
289 high level code to not cache this query, and clean the current cache for
290 this handle by setting *do_cache zero. */
291
292 result = string_cat(result, US PQcmdTuples(pg_result));
293 *do_cache = 0;
294 DEBUG(D_lookup) debug_printf_indent("PGSQL: command does not return any data "
295 "but was successful. Rows affected: %s\n", string_from_gstring(result));
296 break;
297
298 case PGRES_TUPLES_OK:
299 break;
300
301 default:
302 /* This was the original code:
303 *errmsg = string_sprintf("PGSQL: query failed: %s\n",
304 PQresultErrorMessage(pg_result));
305 This was suggested by a user:
306 */
307
308 *errmsg = string_sprintf("PGSQL: query failed: %s (%s) (%s)\n",
309 PQresultErrorMessage(pg_result),
310 PQresStatus(PQresultStatus(pg_result)), query);
311 goto PGSQL_EXIT;
312 }
313
314 /* Result is in pg_result. Find the number of fields returned. If this is one,
315 we don't add field names to the data. Otherwise we do. If the query did not
316 return anything we skip the for loop; this also applies to the case
317 PGRES_COMMAND_OK. */
318
319 num_fields = PQnfields(pg_result);
320 num_tuples = PQntuples(pg_result);
321
322 /* Get the fields and construct the result string. If there is more than one
323 row, we insert '\n' between them. */
324
325 for (int i = 0; i < num_tuples; i++)
326 {
327 if (result)
328 result = string_catn(result, US"\n", 1);
329
330 if (num_fields == 1)
331 result = string_catn(result,
332 US PQgetvalue(pg_result, i, 0), PQgetlength(pg_result, i, 0));
333 else
334 for (int j = 0; j < num_fields; j++)
335 {
336 uschar *tmp = US PQgetvalue(pg_result, i, j);
337 result = lf_quote(US PQfname(pg_result, j), tmp, Ustrlen(tmp), result);
338 }
339 }
340
341 /* If result is NULL then no data has been found and so we return FAIL. */
342
343 if (!result)
344 {
345 yield = FAIL;
346 *errmsg = US"PGSQL: no data found";
347 }
348
349 /* Get here by goto from various error checks. */
350
351 PGSQL_EXIT:
352
353 /* Free store for any result that was got; don't close the connection, as
354 it is cached. */
355
356 if (pg_result) PQclear(pg_result);
357
358 /* Non-NULL result indicates a successful result */
359
360 if (result)
361 {
362 gstring_release_unused(result);
363 *resultptr = string_from_gstring(result);
364 return OK;
365 }
366 else
367 {
368 DEBUG(D_lookup) debug_printf_indent("%s\n", *errmsg);
369 return yield; /* FAIL or DEFER */
370 }
371 }
372
373
374
375
376 /*************************************************
377 * Find entry point *
378 *************************************************/
379
380 /* See local README for interface description. The handle and filename
381 arguments are not used. The code to loop through a list of servers while the
382 query is deferred with a retryable error is now in a separate function that is
383 shared with other SQL lookups. */
384
385 static int
386 pgsql_find(void * handle, const uschar * filename, const uschar * query,
387 int length, uschar ** result, uschar ** errmsg, uint * do_cache,
388 const uschar * opts)
389 {
390 return lf_sqlperform(US"PostgreSQL", US"pgsql_servers", pgsql_servers, query,
391 result, errmsg, do_cache, opts, perform_pgsql_search);
392 }
393
394
395
396 /*************************************************
397 * Quote entry point *
398 *************************************************/
399
400 /* The characters that always need to be quoted (with backslash) are newline,
401 tab, carriage return, backspace, backslash itself, and the quote characters.
402
403 The original code quoted single quotes as \' which is documented as valid in
404 the O'Reilly book "Practical PostgreSQL" (first edition) as an alternative to
405 the SQL standard '' way of representing a single quote as data. However, in
406 June 2006 there was some security issue with using \' and so this has been
407 changed.
408
409 [Note: There is a function called PQescapeStringConn() that quotes strings.
410 This cannot be used because it needs a PGconn argument (the connection handle).
411 Why, I don't know. Seems odd for just string escaping...]
412
413 Arguments:
414 s the string to be quoted
415 opt additional option text or NULL if none
416
417 Returns: the processed string or NULL for a bad option
418 */
419
420 static uschar *
421 pgsql_quote(uschar *s, uschar *opt)
422 {
423 register int c;
424 int count = 0;
425 uschar *t = s;
426 uschar *quoted;
427
428 if (opt != NULL) return NULL; /* No options recognized */
429
430 while ((c = *t++) != 0)
431 if (Ustrchr("\n\t\r\b\'\"\\", c) != NULL) count++;
432
433 if (count == 0) return s;
434 t = quoted = store_get(Ustrlen(s) + count + 1, is_tainted(s));
435
436 while ((c = *s++) != 0)
437 {
438 if (c == '\'')
439 {
440 *t++ = '\'';
441 *t++ = '\'';
442 }
443 else if (Ustrchr("\n\t\r\b\"\\", c) != NULL)
444 {
445 *t++ = '\\';
446 switch(c)
447 {
448 case '\n': *t++ = 'n';
449 break;
450 case '\t': *t++ = 't';
451 break;
452 case '\r': *t++ = 'r';
453 break;
454 case '\b': *t++ = 'b';
455 break;
456 default: *t++ = c;
457 break;
458 }
459 }
460 else *t++ = c;
461 }
462
463 *t = 0;
464 return quoted;
465 }
466
467
468 /*************************************************
469 * Version reporting entry point *
470 *************************************************/
471
472 /* See local README for interface description. */
473
474 #include "../version.h"
475
476 void
477 pgsql_version_report(FILE *f)
478 {
479 #ifdef DYNLOOKUP
480 fprintf(f, "Library version: PostgreSQL: Exim version %s\n", EXIM_VERSION_STR);
481 #endif
482
483 /* Version reporting: there appears to be no available information about
484 the client library in libpq-fe.h; once you have a connection object, you
485 can access the server version and the chosen protocol version, but those
486 aren't really what we want. It might make sense to debug_printf those
487 when the connection is established though? */
488 }
489
490
491 static lookup_info _lookup_info = {
492 .name = US"pgsql", /* lookup name */
493 .type = lookup_querystyle, /* query-style lookup */
494 .open = pgsql_open, /* open function */
495 .check = NULL, /* no check function */
496 .find = pgsql_find, /* find function */
497 .close = NULL, /* no close function */
498 .tidy = pgsql_tidy, /* tidy function */
499 .quote = pgsql_quote, /* quoting function */
500 .version_report = pgsql_version_report /* version reporting */
501 };
502
503 #ifdef DYNLOOKUP
504 #define pgsql_lookup_module_info _lookup_module_info
505 #endif
506
507 static lookup_info *_lookup_list[] = { &_lookup_info };
508 lookup_module_info pgsql_lookup_module_info = { LOOKUP_MODULE_INFO_MAGIC, _lookup_list, 1 };
509
510 /* End of lookups/pgsql.c */