/src/postgres/src/backend/utils/cache/partcache.c
Line | Count | Source (jump to first uncovered line) |
1 | | /*------------------------------------------------------------------------- |
2 | | * |
3 | | * partcache.c |
4 | | * Support routines for manipulating partition information cached in |
5 | | * relcache |
6 | | * |
7 | | * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
8 | | * Portions Copyright (c) 1994, Regents of the University of California |
9 | | * |
10 | | * IDENTIFICATION |
11 | | * src/backend/utils/cache/partcache.c |
12 | | * |
13 | | *------------------------------------------------------------------------- |
14 | | */ |
15 | | #include "postgres.h" |
16 | | |
17 | | #include "access/hash.h" |
18 | | #include "access/htup_details.h" |
19 | | #include "access/nbtree.h" |
20 | | #include "access/relation.h" |
21 | | #include "catalog/partition.h" |
22 | | #include "catalog/pg_opclass.h" |
23 | | #include "catalog/pg_partitioned_table.h" |
24 | | #include "miscadmin.h" |
25 | | #include "nodes/makefuncs.h" |
26 | | #include "nodes/nodeFuncs.h" |
27 | | #include "optimizer/optimizer.h" |
28 | | #include "partitioning/partbounds.h" |
29 | | #include "utils/builtins.h" |
30 | | #include "utils/lsyscache.h" |
31 | | #include "utils/memutils.h" |
32 | | #include "utils/partcache.h" |
33 | | #include "utils/rel.h" |
34 | | #include "utils/syscache.h" |
35 | | |
36 | | |
37 | | static void RelationBuildPartitionKey(Relation relation); |
38 | | static List *generate_partition_qual(Relation rel); |
39 | | |
40 | | /* |
41 | | * RelationGetPartitionKey -- get partition key, if relation is partitioned |
42 | | * |
43 | | * Note: partition keys are not allowed to change after the partitioned rel |
44 | | * is created. RelationClearRelation knows this and preserves rd_partkey |
45 | | * across relcache rebuilds, as long as the relation is open. Therefore, |
46 | | * even though we hand back a direct pointer into the relcache entry, it's |
47 | | * safe for callers to continue to use that pointer as long as they hold |
48 | | * the relation open. |
49 | | */ |
50 | | PartitionKey |
51 | | RelationGetPartitionKey(Relation rel) |
52 | 0 | { |
53 | 0 | if (rel->rd_rel->relkind != RELKIND_PARTITIONED_TABLE) |
54 | 0 | return NULL; |
55 | | |
56 | 0 | if (unlikely(rel->rd_partkey == NULL)) |
57 | 0 | RelationBuildPartitionKey(rel); |
58 | |
|
59 | 0 | return rel->rd_partkey; |
60 | 0 | } |
61 | | |
62 | | /* |
63 | | * RelationBuildPartitionKey |
64 | | * Build partition key data of relation, and attach to relcache |
65 | | * |
66 | | * Partitioning key data is a complex structure; to avoid complicated logic to |
67 | | * free individual elements whenever the relcache entry is flushed, we give it |
68 | | * its own memory context, a child of CacheMemoryContext, which can easily be |
69 | | * deleted on its own. To avoid leaking memory in that context in case of an |
70 | | * error partway through this function, the context is initially created as a |
71 | | * child of CurTransactionContext and only re-parented to CacheMemoryContext |
72 | | * at the end, when no further errors are possible. Also, we don't make this |
73 | | * context the current context except in very brief code sections, out of fear |
74 | | * that some of our callees allocate memory on their own which would be leaked |
75 | | * permanently. |
76 | | */ |
77 | | static void |
78 | | RelationBuildPartitionKey(Relation relation) |
79 | 0 | { |
80 | 0 | Form_pg_partitioned_table form; |
81 | 0 | HeapTuple tuple; |
82 | 0 | bool isnull; |
83 | 0 | int i; |
84 | 0 | PartitionKey key; |
85 | 0 | AttrNumber *attrs; |
86 | 0 | oidvector *opclass; |
87 | 0 | oidvector *collation; |
88 | 0 | ListCell *partexprs_item; |
89 | 0 | Datum datum; |
90 | 0 | MemoryContext partkeycxt, |
91 | 0 | oldcxt; |
92 | 0 | int16 procnum; |
93 | |
|
94 | 0 | tuple = SearchSysCache1(PARTRELID, |
95 | 0 | ObjectIdGetDatum(RelationGetRelid(relation))); |
96 | |
|
97 | 0 | if (!HeapTupleIsValid(tuple)) |
98 | 0 | elog(ERROR, "cache lookup failed for partition key of relation %u", |
99 | 0 | RelationGetRelid(relation)); |
100 | | |
101 | 0 | partkeycxt = AllocSetContextCreate(CurTransactionContext, |
102 | 0 | "partition key", |
103 | 0 | ALLOCSET_SMALL_SIZES); |
104 | 0 | MemoryContextCopyAndSetIdentifier(partkeycxt, |
105 | 0 | RelationGetRelationName(relation)); |
106 | |
|
107 | 0 | key = (PartitionKey) MemoryContextAllocZero(partkeycxt, |
108 | 0 | sizeof(PartitionKeyData)); |
109 | | |
110 | | /* Fixed-length attributes */ |
111 | 0 | form = (Form_pg_partitioned_table) GETSTRUCT(tuple); |
112 | 0 | key->strategy = form->partstrat; |
113 | 0 | key->partnatts = form->partnatts; |
114 | | |
115 | | /* Validate partition strategy code */ |
116 | 0 | if (key->strategy != PARTITION_STRATEGY_LIST && |
117 | 0 | key->strategy != PARTITION_STRATEGY_RANGE && |
118 | 0 | key->strategy != PARTITION_STRATEGY_HASH) |
119 | 0 | elog(ERROR, "invalid partition strategy \"%c\"", key->strategy); |
120 | | |
121 | | /* |
122 | | * We can rely on the first variable-length attribute being mapped to the |
123 | | * relevant field of the catalog's C struct, because all previous |
124 | | * attributes are non-nullable and fixed-length. |
125 | | */ |
126 | 0 | attrs = form->partattrs.values; |
127 | | |
128 | | /* But use the hard way to retrieve further variable-length attributes */ |
129 | | /* Operator class */ |
130 | 0 | datum = SysCacheGetAttrNotNull(PARTRELID, tuple, |
131 | 0 | Anum_pg_partitioned_table_partclass); |
132 | 0 | opclass = (oidvector *) DatumGetPointer(datum); |
133 | | |
134 | | /* Collation */ |
135 | 0 | datum = SysCacheGetAttrNotNull(PARTRELID, tuple, |
136 | 0 | Anum_pg_partitioned_table_partcollation); |
137 | 0 | collation = (oidvector *) DatumGetPointer(datum); |
138 | | |
139 | | /* Expressions */ |
140 | 0 | datum = SysCacheGetAttr(PARTRELID, tuple, |
141 | 0 | Anum_pg_partitioned_table_partexprs, &isnull); |
142 | 0 | if (!isnull) |
143 | 0 | { |
144 | 0 | char *exprString; |
145 | 0 | Node *expr; |
146 | |
|
147 | 0 | exprString = TextDatumGetCString(datum); |
148 | 0 | expr = stringToNode(exprString); |
149 | 0 | pfree(exprString); |
150 | | |
151 | | /* |
152 | | * Run the expressions through const-simplification since the planner |
153 | | * will be comparing them to similarly-processed qual clause operands, |
154 | | * and may fail to detect valid matches without this step; fix |
155 | | * opfuncids while at it. We don't need to bother with |
156 | | * canonicalize_qual() though, because partition expressions should be |
157 | | * in canonical form already (ie, no need for OR-merging or constant |
158 | | * elimination). |
159 | | */ |
160 | 0 | expr = eval_const_expressions(NULL, expr); |
161 | 0 | fix_opfuncids(expr); |
162 | |
|
163 | 0 | oldcxt = MemoryContextSwitchTo(partkeycxt); |
164 | 0 | key->partexprs = (List *) copyObject(expr); |
165 | 0 | MemoryContextSwitchTo(oldcxt); |
166 | 0 | } |
167 | | |
168 | | /* Allocate assorted arrays in the partkeycxt, which we'll fill below */ |
169 | 0 | oldcxt = MemoryContextSwitchTo(partkeycxt); |
170 | 0 | key->partattrs = (AttrNumber *) palloc0(key->partnatts * sizeof(AttrNumber)); |
171 | 0 | key->partopfamily = (Oid *) palloc0(key->partnatts * sizeof(Oid)); |
172 | 0 | key->partopcintype = (Oid *) palloc0(key->partnatts * sizeof(Oid)); |
173 | 0 | key->partsupfunc = (FmgrInfo *) palloc0(key->partnatts * sizeof(FmgrInfo)); |
174 | |
|
175 | 0 | key->partcollation = (Oid *) palloc0(key->partnatts * sizeof(Oid)); |
176 | 0 | key->parttypid = (Oid *) palloc0(key->partnatts * sizeof(Oid)); |
177 | 0 | key->parttypmod = (int32 *) palloc0(key->partnatts * sizeof(int32)); |
178 | 0 | key->parttyplen = (int16 *) palloc0(key->partnatts * sizeof(int16)); |
179 | 0 | key->parttypbyval = (bool *) palloc0(key->partnatts * sizeof(bool)); |
180 | 0 | key->parttypalign = (char *) palloc0(key->partnatts * sizeof(char)); |
181 | 0 | key->parttypcoll = (Oid *) palloc0(key->partnatts * sizeof(Oid)); |
182 | 0 | MemoryContextSwitchTo(oldcxt); |
183 | | |
184 | | /* determine support function number to search for */ |
185 | 0 | procnum = (key->strategy == PARTITION_STRATEGY_HASH) ? |
186 | 0 | HASHEXTENDED_PROC : BTORDER_PROC; |
187 | | |
188 | | /* Copy partattrs and fill other per-attribute info */ |
189 | 0 | memcpy(key->partattrs, attrs, key->partnatts * sizeof(int16)); |
190 | 0 | partexprs_item = list_head(key->partexprs); |
191 | 0 | for (i = 0; i < key->partnatts; i++) |
192 | 0 | { |
193 | 0 | AttrNumber attno = key->partattrs[i]; |
194 | 0 | HeapTuple opclasstup; |
195 | 0 | Form_pg_opclass opclassform; |
196 | 0 | Oid funcid; |
197 | | |
198 | | /* Collect opfamily information */ |
199 | 0 | opclasstup = SearchSysCache1(CLAOID, |
200 | 0 | ObjectIdGetDatum(opclass->values[i])); |
201 | 0 | if (!HeapTupleIsValid(opclasstup)) |
202 | 0 | elog(ERROR, "cache lookup failed for opclass %u", opclass->values[i]); |
203 | | |
204 | 0 | opclassform = (Form_pg_opclass) GETSTRUCT(opclasstup); |
205 | 0 | key->partopfamily[i] = opclassform->opcfamily; |
206 | 0 | key->partopcintype[i] = opclassform->opcintype; |
207 | | |
208 | | /* Get a support function for the specified opfamily and datatypes */ |
209 | 0 | funcid = get_opfamily_proc(opclassform->opcfamily, |
210 | 0 | opclassform->opcintype, |
211 | 0 | opclassform->opcintype, |
212 | 0 | procnum); |
213 | 0 | if (!OidIsValid(funcid)) |
214 | 0 | ereport(ERROR, |
215 | 0 | (errcode(ERRCODE_INVALID_OBJECT_DEFINITION), |
216 | 0 | errmsg("operator class \"%s\" of access method %s is missing support function %d for type %s", |
217 | 0 | NameStr(opclassform->opcname), |
218 | 0 | (key->strategy == PARTITION_STRATEGY_HASH) ? |
219 | 0 | "hash" : "btree", |
220 | 0 | procnum, |
221 | 0 | format_type_be(opclassform->opcintype)))); |
222 | | |
223 | 0 | fmgr_info_cxt(funcid, &key->partsupfunc[i], partkeycxt); |
224 | | |
225 | | /* Collation */ |
226 | 0 | key->partcollation[i] = collation->values[i]; |
227 | | |
228 | | /* Collect type information */ |
229 | 0 | if (attno != 0) |
230 | 0 | { |
231 | 0 | Form_pg_attribute att = TupleDescAttr(relation->rd_att, attno - 1); |
232 | |
|
233 | 0 | key->parttypid[i] = att->atttypid; |
234 | 0 | key->parttypmod[i] = att->atttypmod; |
235 | 0 | key->parttypcoll[i] = att->attcollation; |
236 | 0 | } |
237 | 0 | else |
238 | 0 | { |
239 | 0 | if (partexprs_item == NULL) |
240 | 0 | elog(ERROR, "wrong number of partition key expressions"); |
241 | | |
242 | 0 | key->parttypid[i] = exprType(lfirst(partexprs_item)); |
243 | 0 | key->parttypmod[i] = exprTypmod(lfirst(partexprs_item)); |
244 | 0 | key->parttypcoll[i] = exprCollation(lfirst(partexprs_item)); |
245 | |
|
246 | 0 | partexprs_item = lnext(key->partexprs, partexprs_item); |
247 | 0 | } |
248 | 0 | get_typlenbyvalalign(key->parttypid[i], |
249 | 0 | &key->parttyplen[i], |
250 | 0 | &key->parttypbyval[i], |
251 | 0 | &key->parttypalign[i]); |
252 | |
|
253 | 0 | ReleaseSysCache(opclasstup); |
254 | 0 | } |
255 | | |
256 | 0 | ReleaseSysCache(tuple); |
257 | | |
258 | | /* Assert that we're not leaking any old data during assignments below */ |
259 | 0 | Assert(relation->rd_partkeycxt == NULL); |
260 | 0 | Assert(relation->rd_partkey == NULL); |
261 | | |
262 | | /* |
263 | | * Success --- reparent our context and make the relcache point to the |
264 | | * newly constructed key |
265 | | */ |
266 | 0 | MemoryContextSetParent(partkeycxt, CacheMemoryContext); |
267 | 0 | relation->rd_partkeycxt = partkeycxt; |
268 | 0 | relation->rd_partkey = key; |
269 | 0 | } |
270 | | |
271 | | /* |
272 | | * RelationGetPartitionQual |
273 | | * |
274 | | * Returns a list of partition quals |
275 | | */ |
276 | | List * |
277 | | RelationGetPartitionQual(Relation rel) |
278 | 0 | { |
279 | | /* Quick exit */ |
280 | 0 | if (!rel->rd_rel->relispartition) |
281 | 0 | return NIL; |
282 | | |
283 | 0 | return generate_partition_qual(rel); |
284 | 0 | } |
285 | | |
286 | | /* |
287 | | * get_partition_qual_relid |
288 | | * |
289 | | * Returns an expression tree describing the passed-in relation's partition |
290 | | * constraint. |
291 | | * |
292 | | * If the relation is not found, or is not a partition, or there is no |
293 | | * partition constraint, return NULL. We must guard against the first two |
294 | | * cases because this supports a SQL function that could be passed any OID. |
295 | | * The last case can happen even if relispartition is true, when a default |
296 | | * partition is the only partition. |
297 | | */ |
298 | | Expr * |
299 | | get_partition_qual_relid(Oid relid) |
300 | 0 | { |
301 | 0 | Expr *result = NULL; |
302 | | |
303 | | /* Do the work only if this relation exists and is a partition. */ |
304 | 0 | if (get_rel_relispartition(relid)) |
305 | 0 | { |
306 | 0 | Relation rel = relation_open(relid, AccessShareLock); |
307 | 0 | List *and_args; |
308 | |
|
309 | 0 | and_args = generate_partition_qual(rel); |
310 | | |
311 | | /* Convert implicit-AND list format to boolean expression */ |
312 | 0 | if (and_args == NIL) |
313 | 0 | result = NULL; |
314 | 0 | else if (list_length(and_args) > 1) |
315 | 0 | result = makeBoolExpr(AND_EXPR, and_args, -1); |
316 | 0 | else |
317 | 0 | result = linitial(and_args); |
318 | | |
319 | | /* Keep the lock, to allow safe deparsing against the rel by caller. */ |
320 | 0 | relation_close(rel, NoLock); |
321 | 0 | } |
322 | |
|
323 | 0 | return result; |
324 | 0 | } |
325 | | |
326 | | /* |
327 | | * generate_partition_qual |
328 | | * |
329 | | * Generate partition predicate from rel's partition bound expression. The |
330 | | * function returns a NIL list if there is no predicate. |
331 | | * |
332 | | * We cache a copy of the result in the relcache entry, after constructing |
333 | | * it using the caller's context. This approach avoids leaking any data |
334 | | * into long-lived cache contexts, especially if we fail partway through. |
335 | | */ |
336 | | static List * |
337 | | generate_partition_qual(Relation rel) |
338 | 0 | { |
339 | 0 | HeapTuple tuple; |
340 | 0 | MemoryContext oldcxt; |
341 | 0 | Datum boundDatum; |
342 | 0 | bool isnull; |
343 | 0 | List *my_qual = NIL, |
344 | 0 | *result = NIL; |
345 | 0 | Oid parentrelid; |
346 | 0 | Relation parent; |
347 | | |
348 | | /* Guard against stack overflow due to overly deep partition tree */ |
349 | 0 | check_stack_depth(); |
350 | | |
351 | | /* If we already cached the result, just return a copy */ |
352 | 0 | if (rel->rd_partcheckvalid) |
353 | 0 | return copyObject(rel->rd_partcheck); |
354 | | |
355 | | /* |
356 | | * Grab at least an AccessShareLock on the parent table. Must do this |
357 | | * even if the partition has been partially detached, because transactions |
358 | | * concurrent with the detach might still be trying to use a partition |
359 | | * descriptor that includes it. |
360 | | */ |
361 | 0 | parentrelid = get_partition_parent(RelationGetRelid(rel), true); |
362 | 0 | parent = relation_open(parentrelid, AccessShareLock); |
363 | | |
364 | | /* Get pg_class.relpartbound */ |
365 | 0 | tuple = SearchSysCache1(RELOID, |
366 | 0 | ObjectIdGetDatum(RelationGetRelid(rel))); |
367 | 0 | if (!HeapTupleIsValid(tuple)) |
368 | 0 | elog(ERROR, "cache lookup failed for relation %u", |
369 | 0 | RelationGetRelid(rel)); |
370 | | |
371 | 0 | boundDatum = SysCacheGetAttr(RELOID, tuple, |
372 | 0 | Anum_pg_class_relpartbound, |
373 | 0 | &isnull); |
374 | 0 | if (!isnull) |
375 | 0 | { |
376 | 0 | PartitionBoundSpec *bound; |
377 | |
|
378 | 0 | bound = castNode(PartitionBoundSpec, |
379 | 0 | stringToNode(TextDatumGetCString(boundDatum))); |
380 | |
|
381 | 0 | my_qual = get_qual_from_partbound(parent, bound); |
382 | 0 | } |
383 | |
|
384 | 0 | ReleaseSysCache(tuple); |
385 | | |
386 | | /* Add the parent's quals to the list (if any) */ |
387 | 0 | if (parent->rd_rel->relispartition) |
388 | 0 | result = list_concat(generate_partition_qual(parent), my_qual); |
389 | 0 | else |
390 | 0 | result = my_qual; |
391 | | |
392 | | /* |
393 | | * Change Vars to have partition's attnos instead of the parent's. We do |
394 | | * this after we concatenate the parent's quals, because we want every Var |
395 | | * in it to bear this relation's attnos. It's safe to assume varno = 1 |
396 | | * here. |
397 | | */ |
398 | 0 | result = map_partition_varattnos(result, 1, rel, parent); |
399 | | |
400 | | /* Assert that we're not leaking any old data during assignments below */ |
401 | 0 | Assert(rel->rd_partcheckcxt == NULL); |
402 | 0 | Assert(rel->rd_partcheck == NIL); |
403 | | |
404 | | /* |
405 | | * Save a copy in the relcache. The order of these operations is fairly |
406 | | * critical to avoid memory leaks and ensure that we don't leave a corrupt |
407 | | * relcache entry if we fail partway through copyObject. |
408 | | * |
409 | | * If, as is definitely possible, the partcheck list is NIL, then we do |
410 | | * not need to make a context to hold it. |
411 | | */ |
412 | 0 | if (result != NIL) |
413 | 0 | { |
414 | 0 | rel->rd_partcheckcxt = AllocSetContextCreate(CacheMemoryContext, |
415 | 0 | "partition constraint", |
416 | 0 | ALLOCSET_SMALL_SIZES); |
417 | 0 | MemoryContextCopyAndSetIdentifier(rel->rd_partcheckcxt, |
418 | 0 | RelationGetRelationName(rel)); |
419 | 0 | oldcxt = MemoryContextSwitchTo(rel->rd_partcheckcxt); |
420 | 0 | rel->rd_partcheck = copyObject(result); |
421 | 0 | MemoryContextSwitchTo(oldcxt); |
422 | 0 | } |
423 | 0 | else |
424 | 0 | rel->rd_partcheck = NIL; |
425 | 0 | rel->rd_partcheckvalid = true; |
426 | | |
427 | | /* Keep the parent locked until commit */ |
428 | 0 | relation_close(parent, NoLock); |
429 | | |
430 | | /* Return the working copy to the caller */ |
431 | 0 | return result; |
432 | 0 | } |