Age Owner TLA Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * hashvalidate.c
4 : * Opclass validator for hash.
5 : *
6 : * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : * IDENTIFICATION
10 : * src/backend/access/hash/hashvalidate.c
11 : *
12 : *-------------------------------------------------------------------------
13 : */
14 : #include "postgres.h"
15 :
16 : #include "access/amvalidate.h"
17 : #include "access/hash.h"
18 : #include "access/htup_details.h"
19 : #include "access/xact.h"
20 : #include "catalog/pg_am.h"
21 : #include "catalog/pg_amop.h"
22 : #include "catalog/pg_amproc.h"
23 : #include "catalog/pg_opclass.h"
24 : #include "catalog/pg_opfamily.h"
25 : #include "catalog/pg_proc.h"
26 : #include "catalog/pg_type.h"
27 : #include "parser/parse_coerce.h"
28 : #include "utils/builtins.h"
29 : #include "utils/fmgroids.h"
30 : #include "utils/lsyscache.h"
31 : #include "utils/regproc.h"
32 : #include "utils/syscache.h"
33 :
34 :
35 : static bool check_hash_func_signature(Oid funcid, int16 amprocnum, Oid argtype);
36 :
37 :
38 : /*
39 : * Validator for a hash opclass.
40 : *
41 : * Some of the checks done here cover the whole opfamily, and therefore are
42 : * redundant when checking each opclass in a family. But they don't run long
43 : * enough to be much of a problem, so we accept the duplication rather than
44 : * complicate the amvalidate API.
45 : */
46 : bool
2639 tgl 47 CBC 139 : hashvalidate(Oid opclassoid)
48 : {
2635 49 139 : bool result = true;
50 : HeapTuple classtup;
51 : Form_pg_opclass classform;
52 : Oid opfamilyoid;
53 : Oid opcintype;
54 : char *opclassname;
55 : HeapTuple familytup;
56 : Form_pg_opfamily familyform;
57 : char *opfamilyname;
58 : CatCList *proclist,
59 : *oprlist;
60 : List *grouplist;
61 : OpFamilyOpFuncGroup *opclassgroup;
62 139 : List *hashabletypes = NIL;
63 : int i;
64 : ListCell *lc;
65 :
66 : /* Fetch opclass information */
2639 67 139 : classtup = SearchSysCache1(CLAOID, ObjectIdGetDatum(opclassoid));
68 139 : if (!HeapTupleIsValid(classtup))
2639 tgl 69 UBC 0 : elog(ERROR, "cache lookup failed for operator class %u", opclassoid);
2639 tgl 70 CBC 139 : classform = (Form_pg_opclass) GETSTRUCT(classtup);
71 :
72 139 : opfamilyoid = classform->opcfamily;
73 139 : opcintype = classform->opcintype;
2635 74 139 : opclassname = NameStr(classform->opcname);
75 :
76 : /* Fetch opfamily information */
77 139 : familytup = SearchSysCache1(OPFAMILYOID, ObjectIdGetDatum(opfamilyoid));
78 139 : if (!HeapTupleIsValid(familytup))
2635 tgl 79 UBC 0 : elog(ERROR, "cache lookup failed for operator family %u", opfamilyoid);
2635 tgl 80 CBC 139 : familyform = (Form_pg_opfamily) GETSTRUCT(familytup);
81 :
82 139 : opfamilyname = NameStr(familyform->opfname);
83 :
84 : /* Fetch all operators and support functions of the opfamily */
2639 85 139 : oprlist = SearchSysCacheList1(AMOPSTRATEGY, ObjectIdGetDatum(opfamilyoid));
86 139 : proclist = SearchSysCacheList1(AMPROCNUM, ObjectIdGetDatum(opfamilyoid));
87 :
88 : /* Check individual support functions */
89 525 : for (i = 0; i < proclist->n_members; i++)
90 : {
91 386 : HeapTuple proctup = &proclist->members[i]->tuple;
92 386 : Form_pg_amproc procform = (Form_pg_amproc) GETSTRUCT(proctup);
93 :
94 : /*
95 : * All hash functions should be registered with matching left/right
96 : * types
97 : */
2635 98 386 : if (procform->amproclefttype != procform->amprocrighttype)
99 : {
2635 tgl 100 UBC 0 : ereport(INFO,
101 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
102 : errmsg("operator family \"%s\" of access method %s contains support function %s with different left and right input types",
103 : opfamilyname, "hash",
104 : format_procedure(procform->amproc))));
105 0 : result = false;
106 : }
107 :
108 : /* Check procedure numbers and function signatures */
2635 tgl 109 CBC 386 : switch (procform->amprocnum)
110 : {
2047 rhaas 111 386 : case HASHSTANDARD_PROC:
112 : case HASHEXTENDED_PROC:
113 386 : if (!check_hash_func_signature(procform->amproc, procform->amprocnum,
114 : procform->amproclefttype))
115 : {
2635 tgl 116 UBC 0 : ereport(INFO,
117 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
118 : errmsg("operator family \"%s\" of access method %s contains function %s with wrong signature for support number %d",
119 : opfamilyname, "hash",
120 : format_procedure(procform->amproc),
121 : procform->amprocnum)));
122 0 : result = false;
123 : }
124 : else
125 : {
126 : /* Remember which types we can hash */
127 : hashabletypes =
2635 tgl 128 CBC 386 : list_append_unique_oid(hashabletypes,
129 : procform->amproclefttype);
130 : }
131 386 : break;
1105 akorotkov 132 UBC 0 : case HASHOPTIONS_PROC:
133 0 : if (!check_amoptsproc_signature(procform->amproc))
134 0 : result = false;
135 0 : break;
2635 tgl 136 0 : default:
137 0 : ereport(INFO,
138 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
139 : errmsg("operator family \"%s\" of access method %s contains function %s with invalid support number %d",
140 : opfamilyname, "hash",
141 : format_procedure(procform->amproc),
142 : procform->amprocnum)));
143 0 : result = false;
144 0 : break;
145 : }
146 : }
147 :
148 : /* Check individual operators */
2639 tgl 149 CBC 611 : for (i = 0; i < oprlist->n_members; i++)
150 : {
151 472 : HeapTuple oprtup = &oprlist->members[i]->tuple;
152 472 : Form_pg_amop oprform = (Form_pg_amop) GETSTRUCT(oprtup);
153 :
154 : /* Check that only allowed strategy numbers exist */
155 472 : if (oprform->amopstrategy < 1 ||
156 472 : oprform->amopstrategy > HTMaxStrategyNumber)
157 : {
2635 tgl 158 UBC 0 : ereport(INFO,
159 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
160 : errmsg("operator family \"%s\" of access method %s contains operator %s with invalid strategy number %d",
161 : opfamilyname, "hash",
162 : format_operator(oprform->amopopr),
163 : oprform->amopstrategy)));
164 0 : result = false;
165 : }
166 :
167 : /* hash doesn't support ORDER BY operators */
2635 tgl 168 CBC 472 : if (oprform->amoppurpose != AMOP_SEARCH ||
169 472 : OidIsValid(oprform->amopsortfamily))
170 : {
2635 tgl 171 UBC 0 : ereport(INFO,
172 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
173 : errmsg("operator family \"%s\" of access method %s contains invalid ORDER BY specification for operator %s",
174 : opfamilyname, "hash",
175 : format_operator(oprform->amopopr))));
176 0 : result = false;
177 : }
178 :
179 : /* Check operator signature --- same for all hash strategies */
2635 tgl 180 CBC 472 : if (!check_amop_signature(oprform->amopopr, BOOLOID,
181 : oprform->amoplefttype,
182 : oprform->amoprighttype))
183 : {
2635 tgl 184 UBC 0 : ereport(INFO,
185 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
186 : errmsg("operator family \"%s\" of access method %s contains operator %s with wrong signature",
187 : opfamilyname, "hash",
188 : format_operator(oprform->amopopr))));
189 0 : result = false;
190 : }
191 :
192 : /* There should be relevant hash functions for each datatype */
2635 tgl 193 CBC 472 : if (!list_member_oid(hashabletypes, oprform->amoplefttype) ||
194 472 : !list_member_oid(hashabletypes, oprform->amoprighttype))
195 : {
2635 tgl 196 UBC 0 : ereport(INFO,
197 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
198 : errmsg("operator family \"%s\" of access method %s lacks support function for operator %s",
199 : opfamilyname, "hash",
200 : format_operator(oprform->amopopr))));
201 0 : result = false;
202 : }
203 : }
204 :
205 : /* Now check for inconsistent groups of operators/functions */
2635 tgl 206 CBC 139 : grouplist = identify_opfamily_groups(oprlist, proclist);
207 139 : opclassgroup = NULL;
208 611 : foreach(lc, grouplist)
209 : {
210 472 : OpFamilyOpFuncGroup *thisgroup = (OpFamilyOpFuncGroup *) lfirst(lc);
211 :
212 : /* Remember the group exactly matching the test opclass */
213 472 : if (thisgroup->lefttype == opcintype &&
214 192 : thisgroup->righttype == opcintype)
215 139 : opclassgroup = thisgroup;
216 :
217 : /*
218 : * Complain if there seems to be an incomplete set of operators for
219 : * this datatype pair (implying that we have a hash function but no
220 : * operator).
221 : */
222 472 : if (thisgroup->operatorset != (1 << HTEqualStrategyNumber))
223 : {
2635 tgl 224 UBC 0 : ereport(INFO,
225 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
226 : errmsg("operator family \"%s\" of access method %s is missing operator(s) for types %s and %s",
227 : opfamilyname, "hash",
228 : format_type_be(thisgroup->lefttype),
229 : format_type_be(thisgroup->righttype))));
230 0 : result = false;
231 : }
232 : }
233 :
234 : /* Check that the originally-named opclass is supported */
235 : /* (if group is there, we already checked it adequately above) */
2635 tgl 236 CBC 139 : if (!opclassgroup)
237 : {
2635 tgl 238 UBC 0 : ereport(INFO,
239 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
240 : errmsg("operator class \"%s\" of access method %s is missing operator(s)",
241 : opclassname, "hash")));
242 0 : result = false;
243 : }
244 :
245 : /*
246 : * Complain if the opfamily doesn't have entries for all possible
247 : * combinations of its supported datatypes. While missing cross-type
248 : * operators are not fatal, it seems reasonable to insist that all
249 : * built-in hash opfamilies be complete.
250 : */
2635 tgl 251 CBC 139 : if (list_length(grouplist) !=
252 139 : list_length(hashabletypes) * list_length(hashabletypes))
253 : {
254 8 : ereport(INFO,
255 : (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
256 : errmsg("operator family \"%s\" of access method %s is missing cross-type operator(s)",
257 : opfamilyname, "hash")));
258 8 : result = false;
259 : }
260 :
2639 261 139 : ReleaseCatCacheList(proclist);
262 139 : ReleaseCatCacheList(oprlist);
2635 263 139 : ReleaseSysCache(familytup);
264 139 : ReleaseSysCache(classtup);
265 :
266 139 : return result;
267 : }
268 :
269 :
270 : /*
271 : * We need a custom version of check_amproc_signature because of assorted
272 : * hacks in the core hash opclass definitions.
273 : */
274 : static bool
2047 rhaas 275 386 : check_hash_func_signature(Oid funcid, int16 amprocnum, Oid argtype)
276 : {
2635 tgl 277 386 : bool result = true;
278 : Oid restype;
279 : int16 nargs;
280 : HeapTuple tp;
281 : Form_pg_proc procform;
282 :
2047 rhaas 283 386 : switch (amprocnum)
284 : {
285 222 : case HASHSTANDARD_PROC:
286 222 : restype = INT4OID;
287 222 : nargs = 1;
288 222 : break;
289 :
290 164 : case HASHEXTENDED_PROC:
291 164 : restype = INT8OID;
292 164 : nargs = 2;
293 164 : break;
294 :
2047 rhaas 295 UBC 0 : default:
296 0 : elog(ERROR, "invalid amprocnum");
297 : }
298 :
2635 tgl 299 CBC 386 : tp = SearchSysCache1(PROCOID, ObjectIdGetDatum(funcid));
300 386 : if (!HeapTupleIsValid(tp))
2635 tgl 301 UBC 0 : elog(ERROR, "cache lookup failed for function %u", funcid);
2635 tgl 302 CBC 386 : procform = (Form_pg_proc) GETSTRUCT(tp);
303 :
304 386 : if (procform->prorettype != restype || procform->proretset ||
2047 rhaas 305 386 : procform->pronargs != nargs)
2635 tgl 306 UBC 0 : result = false;
307 :
2635 tgl 308 CBC 386 : if (!IsBinaryCoercible(argtype, procform->proargtypes.values[0]))
309 : {
310 : /*
311 : * Some of the built-in hash opclasses cheat by using hash functions
312 : * that are different from but physically compatible with the opclass
313 : * datatype. In some of these cases, even a "binary coercible" check
314 : * fails because there's no relevant cast. For the moment, fix it by
315 : * having a list of allowed cases. Test the specific function
316 : * identity, not just its input type, because hashvarlena() takes
317 : * INTERNAL and allowing any such function seems too scary.
318 : */
2047 rhaas 319 42 : if ((funcid == F_HASHINT4 || funcid == F_HASHINT4EXTENDED) &&
2635 tgl 320 12 : (argtype == DATEOID ||
321 6 : argtype == XIDOID || argtype == CIDOID))
322 : /* okay, allowed use of hashint4() */ ;
1097 tmunro 323 24 : else if ((funcid == F_HASHINT8 || funcid == F_HASHINT8EXTENDED) &&
324 : (argtype == XID8OID))
325 : /* okay, allowed use of hashint8() */ ;
2047 rhaas 326 18 : else if ((funcid == F_TIMESTAMP_HASH ||
327 6 : funcid == F_TIMESTAMP_HASH_EXTENDED) &&
328 : argtype == TIMESTAMPTZOID)
329 : /* okay, allowed use of timestamp_hash() */ ;
330 12 : else if ((funcid == F_HASHCHAR || funcid == F_HASHCHAREXTENDED) &&
331 : argtype == BOOLOID)
332 : /* okay, allowed use of hashchar() */ ;
333 6 : else if ((funcid == F_HASHVARLENA || funcid == F_HASHVARLENAEXTENDED) &&
334 : argtype == BYTEAOID)
335 : /* okay, allowed use of hashvarlena() */ ;
336 : else
2635 tgl 337 UBC 0 : result = false;
338 : }
339 :
340 : /* If function takes a second argument, it must be for a 64-bit salt. */
2047 rhaas 341 CBC 386 : if (nargs == 2 && procform->proargtypes.values[1] != INT8OID)
2047 rhaas 342 UBC 0 : result = false;
343 :
2635 tgl 344 CBC 386 : ReleaseSysCache(tp);
345 386 : return result;
346 : }
347 :
348 : /*
349 : * Prechecking function for adding operators/functions to a hash opfamily.
350 : */
351 : void
981 352 70 : hashadjustmembers(Oid opfamilyoid,
353 : Oid opclassoid,
354 : List *operators,
355 : List *functions)
356 : {
357 : Oid opcintype;
358 : ListCell *lc;
359 :
360 : /*
361 : * Hash operators and required support functions are always "loose"
362 : * members of the opfamily if they are cross-type. If they are not
363 : * cross-type, we prefer to tie them to the appropriate opclass ... but if
364 : * the user hasn't created one, we can't do that, and must fall back to
365 : * using the opfamily dependency. (We mustn't force creation of an
366 : * opclass in such a case, as leaving an incomplete opclass laying about
367 : * would be bad. Throwing an error is another undesirable alternative.)
368 : *
369 : * This behavior results in a bit of a dump/reload hazard, in that the
370 : * order of restoring objects could affect what dependencies we end up
371 : * with. pg_dump's existing behavior will preserve the dependency choices
372 : * in most cases, but not if a cross-type operator has been bound tightly
373 : * into an opclass. That's a mistake anyway, so silently "fixing" it
374 : * isn't awful.
375 : *
376 : * Optional support functions are always "loose" family members.
377 : *
378 : * To avoid repeated lookups, we remember the most recently used opclass's
379 : * input type.
380 : */
381 70 : if (OidIsValid(opclassoid))
382 : {
383 : /* During CREATE OPERATOR CLASS, need CCI to see the pg_opclass row */
384 50 : CommandCounterIncrement();
385 50 : opcintype = get_opclass_input_type(opclassoid);
386 : }
387 : else
388 20 : opcintype = InvalidOid;
389 :
390 : /*
391 : * We handle operators and support functions almost identically, so rather
392 : * than duplicate this code block, just join the lists.
393 : */
394 172 : foreach(lc, list_concat_copy(operators, functions))
395 : {
396 102 : OpFamilyMember *op = (OpFamilyMember *) lfirst(lc);
397 :
398 102 : if (op->is_func && op->number != HASHSTANDARD_PROC)
399 : {
400 : /* Optional support proc, so always a soft family dependency */
401 24 : op->ref_is_hard = false;
402 24 : op->ref_is_family = true;
403 24 : op->refobjid = opfamilyoid;
404 : }
405 78 : else if (op->lefttype != op->righttype)
406 : {
407 : /* Cross-type, so always a soft family dependency */
408 20 : op->ref_is_hard = false;
409 20 : op->ref_is_family = true;
410 20 : op->refobjid = opfamilyoid;
411 : }
412 : else
413 : {
414 : /* Not cross-type; is there a suitable opclass? */
415 58 : if (op->lefttype != opcintype)
416 : {
417 : /* Avoid repeating this expensive lookup, even if it fails */
981 tgl 418 UBC 0 : opcintype = op->lefttype;
419 0 : opclassoid = opclass_for_family_datatype(HASH_AM_OID,
420 : opfamilyoid,
421 : opcintype);
422 : }
981 tgl 423 CBC 58 : if (OidIsValid(opclassoid))
424 : {
425 : /* Hard dependency on opclass */
426 58 : op->ref_is_hard = true;
427 58 : op->ref_is_family = false;
428 58 : op->refobjid = opclassoid;
429 : }
430 : else
431 : {
432 : /* We're stuck, so make a soft dependency on the opfamily */
981 tgl 433 UBC 0 : op->ref_is_hard = false;
434 0 : op->ref_is_family = true;
435 0 : op->refobjid = opfamilyoid;
436 : }
437 : }
438 : }
981 tgl 439 CBC 70 : }
|
