Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * toasting.c
4 : * This file contains routines to support creation of toast tables
5 : *
6 : *
7 : * Portions Copyright (c) 1996-2017, PostgreSQL Global Development Group
8 : * Portions Copyright (c) 1994, Regents of the University of California
9 : *
10 : * IDENTIFICATION
11 : * src/backend/catalog/toasting.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 : #include "postgres.h"
16 :
17 : #include "access/tuptoaster.h"
18 : #include "access/xact.h"
19 : #include "catalog/binary_upgrade.h"
20 : #include "catalog/dependency.h"
21 : #include "catalog/heap.h"
22 : #include "catalog/index.h"
23 : #include "catalog/namespace.h"
24 : #include "catalog/pg_am.h"
25 : #include "catalog/pg_namespace.h"
26 : #include "catalog/pg_opclass.h"
27 : #include "catalog/pg_type.h"
28 : #include "catalog/toasting.h"
29 : #include "miscadmin.h"
30 : #include "nodes/makefuncs.h"
31 : #include "storage/lock.h"
32 : #include "utils/builtins.h"
33 : #include "utils/rel.h"
34 : #include "utils/syscache.h"
35 :
36 : /* Potentially set by pg_upgrade_support functions */
37 : Oid binary_upgrade_next_toast_pg_type_oid = InvalidOid;
38 :
39 : static void CheckAndCreateToastTable(Oid relOid, Datum reloptions,
40 : LOCKMODE lockmode, bool check);
41 : static bool create_toast_table(Relation rel, Oid toastOid, Oid toastIndexOid,
42 : Datum reloptions, LOCKMODE lockmode, bool check);
43 : static bool needs_toast_table(Relation rel);
44 :
45 :
46 : /*
47 : * CreateToastTable variants
48 : * If the table needs a toast table, and doesn't already have one,
49 : * then create a toast table for it.
50 : *
51 : * reloptions for the toast table can be passed, too. Pass (Datum) 0
52 : * for default reloptions.
53 : *
54 : * We expect the caller to have verified that the relation is a table and have
55 : * already done any necessary permission checks. Callers expect this function
56 : * to end with CommandCounterIncrement if it makes any changes.
57 : */
58 : void
59 785 : AlterTableCreateToastTable(Oid relOid, Datum reloptions, LOCKMODE lockmode)
60 : {
61 785 : CheckAndCreateToastTable(relOid, reloptions, lockmode, true);
62 785 : }
63 :
64 : void
65 53 : NewHeapCreateToastTable(Oid relOid, Datum reloptions, LOCKMODE lockmode)
66 : {
67 53 : CheckAndCreateToastTable(relOid, reloptions, lockmode, false);
68 53 : }
69 :
70 : void
71 1505 : NewRelationCreateToastTable(Oid relOid, Datum reloptions)
72 : {
73 1505 : CheckAndCreateToastTable(relOid, reloptions, AccessExclusiveLock, false);
74 1505 : }
75 :
76 : static void
77 2343 : CheckAndCreateToastTable(Oid relOid, Datum reloptions, LOCKMODE lockmode, bool check)
78 : {
79 : Relation rel;
80 :
81 2343 : rel = heap_open(relOid, lockmode);
82 :
83 : /* create_toast_table does all the work */
84 2343 : (void) create_toast_table(rel, InvalidOid, InvalidOid, reloptions, lockmode, check);
85 :
86 2343 : heap_close(rel, NoLock);
87 2343 : }
88 :
89 : /*
90 : * Create a toast table during bootstrap
91 : *
92 : * Here we need to prespecify the OIDs of the toast table and its index
93 : */
94 : void
95 12 : BootstrapToastTable(char *relName, Oid toastOid, Oid toastIndexOid)
96 : {
97 : Relation rel;
98 :
99 12 : rel = heap_openrv(makeRangeVar(NULL, relName, -1), AccessExclusiveLock);
100 :
101 12 : if (rel->rd_rel->relkind != RELKIND_RELATION &&
102 0 : rel->rd_rel->relkind != RELKIND_MATVIEW)
103 0 : ereport(ERROR,
104 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
105 : errmsg("\"%s\" is not a table or materialized view",
106 : relName)));
107 :
108 : /* create_toast_table does all the work */
109 12 : if (!create_toast_table(rel, toastOid, toastIndexOid, (Datum) 0,
110 : AccessExclusiveLock, false))
111 0 : elog(ERROR, "\"%s\" does not require a toast table",
112 : relName);
113 :
114 12 : heap_close(rel, NoLock);
115 12 : }
116 :
117 :
118 : /*
119 : * create_toast_table --- internal workhorse
120 : *
121 : * rel is already opened and locked
122 : * toastOid and toastIndexOid are normally InvalidOid, but during
123 : * bootstrap they can be nonzero to specify hand-assigned OIDs
124 : */
125 : static bool
126 2355 : create_toast_table(Relation rel, Oid toastOid, Oid toastIndexOid,
127 : Datum reloptions, LOCKMODE lockmode, bool check)
128 : {
129 2355 : Oid relOid = RelationGetRelid(rel);
130 : HeapTuple reltup;
131 : TupleDesc tupdesc;
132 : bool shared_relation;
133 : bool mapped_relation;
134 : Relation toast_rel;
135 : Relation class_rel;
136 : Oid toast_relid;
137 2355 : Oid toast_typid = InvalidOid;
138 : Oid namespaceid;
139 : char toast_relname[NAMEDATALEN];
140 : char toast_idxname[NAMEDATALEN];
141 : IndexInfo *indexInfo;
142 : Oid collationObjectId[2];
143 : Oid classObjectId[2];
144 : int16 coloptions[2];
145 : ObjectAddress baseobject,
146 : toastobject;
147 :
148 : /*
149 : * Toast table is shared if and only if its parent is.
150 : *
151 : * We cannot allow toasting a shared relation after initdb (because
152 : * there's no way to mark it toasted in other databases' pg_class).
153 : */
154 2355 : shared_relation = rel->rd_rel->relisshared;
155 2355 : if (shared_relation && !IsBootstrapProcessingMode())
156 0 : ereport(ERROR,
157 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
158 : errmsg("shared tables cannot be toasted after initdb")));
159 :
160 : /* It's mapped if and only if its parent is, too */
161 2355 : mapped_relation = RelationIsMapped(rel);
162 :
163 : /*
164 : * Is it already toasted?
165 : */
166 2355 : if (rel->rd_rel->reltoastrelid != InvalidOid)
167 323 : return false;
168 :
169 : /*
170 : * Check to see whether the table actually needs a TOAST table.
171 : */
172 2032 : if (!IsBinaryUpgrade)
173 : {
174 : /* Normal mode, normal check */
175 2032 : if (!needs_toast_table(rel))
176 1339 : return false;
177 : }
178 : else
179 : {
180 : /*
181 : * In binary-upgrade mode, create a TOAST table if and only if
182 : * pg_upgrade told us to (ie, a TOAST table OID has been provided).
183 : *
184 : * This indicates that the old cluster had a TOAST table for the
185 : * current table. We must create a TOAST table to receive the old
186 : * TOAST file, even if the table seems not to need one.
187 : *
188 : * Contrariwise, if the old cluster did not have a TOAST table, we
189 : * should be able to get along without one even if the new version's
190 : * needs_toast_table rules suggest we should have one. There is a lot
191 : * of daylight between where we will create a TOAST table and where
192 : * one is really necessary to avoid failures, so small cross-version
193 : * differences in the when-to-create heuristic shouldn't be a problem.
194 : * If we tried to create a TOAST table anyway, we would have the
195 : * problem that it might take up an OID that will conflict with some
196 : * old-cluster table we haven't seen yet.
197 : */
198 0 : if (!OidIsValid(binary_upgrade_next_toast_pg_class_oid) ||
199 0 : !OidIsValid(binary_upgrade_next_toast_pg_type_oid))
200 0 : return false;
201 : }
202 :
203 : /*
204 : * If requested check lockmode is sufficient. This is a cross check in
205 : * case of errors or conflicting decisions in earlier code.
206 : */
207 693 : if (check && lockmode != AccessExclusiveLock)
208 0 : elog(ERROR, "AccessExclusiveLock required to add toast table.");
209 :
210 : /*
211 : * Create the toast table and its index
212 : */
213 693 : snprintf(toast_relname, sizeof(toast_relname),
214 : "pg_toast_%u", relOid);
215 693 : snprintf(toast_idxname, sizeof(toast_idxname),
216 : "pg_toast_%u_index", relOid);
217 :
218 : /* this is pretty painful... need a tuple descriptor */
219 693 : tupdesc = CreateTemplateTupleDesc(3, false);
220 693 : TupleDescInitEntry(tupdesc, (AttrNumber) 1,
221 : "chunk_id",
222 : OIDOID,
223 : -1, 0);
224 693 : TupleDescInitEntry(tupdesc, (AttrNumber) 2,
225 : "chunk_seq",
226 : INT4OID,
227 : -1, 0);
228 693 : TupleDescInitEntry(tupdesc, (AttrNumber) 3,
229 : "chunk_data",
230 : BYTEAOID,
231 : -1, 0);
232 :
233 : /*
234 : * Ensure that the toast table doesn't itself get toasted, or we'll be
235 : * toast :-(. This is essential for chunk_data because type bytea is
236 : * toastable; hit the other two just to be sure.
237 : */
238 693 : TupleDescAttr(tupdesc, 0)->attstorage = 'p';
239 693 : TupleDescAttr(tupdesc, 1)->attstorage = 'p';
240 693 : TupleDescAttr(tupdesc, 2)->attstorage = 'p';
241 :
242 : /*
243 : * Toast tables for regular relations go in pg_toast; those for temp
244 : * relations go into the per-backend temp-toast-table namespace.
245 : */
246 693 : if (isTempOrTempToastNamespace(rel->rd_rel->relnamespace))
247 108 : namespaceid = GetTempToastNamespace();
248 : else
249 585 : namespaceid = PG_TOAST_NAMESPACE;
250 :
251 : /*
252 : * Use binary-upgrade override for pg_type.oid, if supplied. We might be
253 : * in the post-schema-restore phase where we are doing ALTER TABLE to
254 : * create TOAST tables that didn't exist in the old cluster.
255 : */
256 693 : if (IsBinaryUpgrade && OidIsValid(binary_upgrade_next_toast_pg_type_oid))
257 : {
258 0 : toast_typid = binary_upgrade_next_toast_pg_type_oid;
259 0 : binary_upgrade_next_toast_pg_type_oid = InvalidOid;
260 : }
261 :
262 2772 : toast_relid = heap_create_with_catalog(toast_relname,
263 : namespaceid,
264 693 : rel->rd_rel->reltablespace,
265 : toastOid,
266 : toast_typid,
267 : InvalidOid,
268 693 : rel->rd_rel->relowner,
269 : tupdesc,
270 : NIL,
271 : RELKIND_TOASTVALUE,
272 693 : rel->rd_rel->relpersistence,
273 : shared_relation,
274 : mapped_relation,
275 : true,
276 : 0,
277 : ONCOMMIT_NOOP,
278 : reloptions,
279 : false,
280 : true,
281 : true,
282 : NULL);
283 693 : Assert(toast_relid != InvalidOid);
284 :
285 : /* make the toast relation visible, else heap_open will fail */
286 693 : CommandCounterIncrement();
287 :
288 : /* ShareLock is not really needed here, but take it anyway */
289 693 : toast_rel = heap_open(toast_relid, ShareLock);
290 :
291 : /*
292 : * Create unique index on chunk_id, chunk_seq.
293 : *
294 : * NOTE: the normal TOAST access routines could actually function with a
295 : * single-column index on chunk_id only. However, the slice access
296 : * routines use both columns for faster access to an individual chunk. In
297 : * addition, we want it to be unique as a check against the possibility of
298 : * duplicate TOAST chunk OIDs. The index might also be a little more
299 : * efficient this way, since btree isn't all that happy with large numbers
300 : * of equal keys.
301 : */
302 :
303 693 : indexInfo = makeNode(IndexInfo);
304 693 : indexInfo->ii_NumIndexAttrs = 2;
305 693 : indexInfo->ii_KeyAttrNumbers[0] = 1;
306 693 : indexInfo->ii_KeyAttrNumbers[1] = 2;
307 693 : indexInfo->ii_Expressions = NIL;
308 693 : indexInfo->ii_ExpressionsState = NIL;
309 693 : indexInfo->ii_Predicate = NIL;
310 693 : indexInfo->ii_PredicateState = NULL;
311 693 : indexInfo->ii_ExclusionOps = NULL;
312 693 : indexInfo->ii_ExclusionProcs = NULL;
313 693 : indexInfo->ii_ExclusionStrats = NULL;
314 693 : indexInfo->ii_Unique = true;
315 693 : indexInfo->ii_ReadyForInserts = true;
316 693 : indexInfo->ii_Concurrent = false;
317 693 : indexInfo->ii_BrokenHotChain = false;
318 693 : indexInfo->ii_AmCache = NULL;
319 693 : indexInfo->ii_Context = CurrentMemoryContext;
320 :
321 693 : collationObjectId[0] = InvalidOid;
322 693 : collationObjectId[1] = InvalidOid;
323 :
324 693 : classObjectId[0] = OID_BTREE_OPS_OID;
325 693 : classObjectId[1] = INT4_BTREE_OPS_OID;
326 :
327 693 : coloptions[0] = 0;
328 693 : coloptions[1] = 0;
329 :
330 693 : index_create(toast_rel, toast_idxname, toastIndexOid, InvalidOid,
331 : indexInfo,
332 : list_make2("chunk_id", "chunk_seq"),
333 : BTREE_AM_OID,
334 693 : rel->rd_rel->reltablespace,
335 : collationObjectId, classObjectId, coloptions, (Datum) 0,
336 : true, false, false, false,
337 : true, false, false, true, false);
338 :
339 693 : heap_close(toast_rel, NoLock);
340 :
341 : /*
342 : * Store the toast table's OID in the parent relation's pg_class row
343 : */
344 693 : class_rel = heap_open(RelationRelationId, RowExclusiveLock);
345 :
346 693 : reltup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relOid));
347 693 : if (!HeapTupleIsValid(reltup))
348 0 : elog(ERROR, "cache lookup failed for relation %u", relOid);
349 :
350 693 : ((Form_pg_class) GETSTRUCT(reltup))->reltoastrelid = toast_relid;
351 :
352 693 : if (!IsBootstrapProcessingMode())
353 : {
354 : /* normal case, use a transactional update */
355 681 : CatalogTupleUpdate(class_rel, &reltup->t_self, reltup);
356 : }
357 : else
358 : {
359 : /* While bootstrapping, we cannot UPDATE, so overwrite in-place */
360 12 : heap_inplace_update(class_rel, reltup);
361 : }
362 :
363 693 : heap_freetuple(reltup);
364 :
365 693 : heap_close(class_rel, RowExclusiveLock);
366 :
367 : /*
368 : * Register dependency from the toast table to the master, so that the
369 : * toast table will be deleted if the master is. Skip this in bootstrap
370 : * mode.
371 : */
372 693 : if (!IsBootstrapProcessingMode())
373 : {
374 681 : baseobject.classId = RelationRelationId;
375 681 : baseobject.objectId = relOid;
376 681 : baseobject.objectSubId = 0;
377 681 : toastobject.classId = RelationRelationId;
378 681 : toastobject.objectId = toast_relid;
379 681 : toastobject.objectSubId = 0;
380 :
381 681 : recordDependencyOn(&toastobject, &baseobject, DEPENDENCY_INTERNAL);
382 : }
383 :
384 : /*
385 : * Make changes visible
386 : */
387 693 : CommandCounterIncrement();
388 :
389 693 : return true;
390 : }
391 :
392 : /*
393 : * Check to see whether the table needs a TOAST table. It does only if
394 : * (1) there are any toastable attributes, and (2) the maximum length
395 : * of a tuple could exceed TOAST_TUPLE_THRESHOLD. (We don't want to
396 : * create a toast table for something like "f1 varchar(20)".)
397 : */
398 : static bool
399 2032 : needs_toast_table(Relation rel)
400 : {
401 2032 : int32 data_length = 0;
402 2032 : bool maxlength_unknown = false;
403 2032 : bool has_toastable_attrs = false;
404 : TupleDesc tupdesc;
405 : int32 tuple_length;
406 : int i;
407 :
408 2032 : tupdesc = rel->rd_att;
409 :
410 6680 : for (i = 0; i < tupdesc->natts; i++)
411 : {
412 4648 : Form_pg_attribute att = TupleDescAttr(tupdesc, i);
413 :
414 4648 : if (att->attisdropped)
415 97 : continue;
416 4551 : data_length = att_align_nominal(data_length, att->attalign);
417 4551 : if (att->attlen > 0)
418 : {
419 : /* Fixed-length types are never toastable */
420 3365 : data_length += att->attlen;
421 : }
422 : else
423 : {
424 1186 : int32 maxlen = type_maximum_size(att->atttypid,
425 : att->atttypmod);
426 :
427 1186 : if (maxlen < 0)
428 963 : maxlength_unknown = true;
429 : else
430 223 : data_length += maxlen;
431 1186 : if (att->attstorage != 'p')
432 1181 : has_toastable_attrs = true;
433 : }
434 : }
435 2032 : if (!has_toastable_attrs)
436 1175 : return false; /* nothing to toast? */
437 857 : if (maxlength_unknown)
438 692 : return true; /* any unlimited-length attrs? */
439 330 : tuple_length = MAXALIGN(SizeofHeapTupleHeader +
440 165 : BITMAPLEN(tupdesc->natts)) +
441 165 : MAXALIGN(data_length);
442 165 : return (tuple_length > TOAST_TUPLE_THRESHOLD);
443 : }
|
