1/*-------------------------------------------------------------------------
2 *
3 * toasting.c
4 * This file contains routines to support creation of toast tables
5 *
6 *
7 * Portions Copyright (c) 1996-2019, 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/heapam.h"
18#include "access/xact.h"
19#include "catalog/binary_upgrade.h"
20#include "catalog/catalog.h"
21#include "catalog/dependency.h"
22#include "catalog/heap.h"
23#include "catalog/index.h"
24#include "catalog/namespace.h"
25#include "catalog/pg_am.h"
26#include "catalog/pg_namespace.h"
27#include "catalog/pg_opclass.h"
28#include "catalog/pg_type.h"
29#include "catalog/toasting.h"
30#include "miscadmin.h"
31#include "nodes/makefuncs.h"
32#include "storage/lock.h"
33#include "utils/builtins.h"
34#include "utils/rel.h"
35#include "utils/syscache.h"
36
37/* Potentially set by pg_upgrade_support functions */
38Oid binary_upgrade_next_toast_pg_type_oid = InvalidOid;
39
40static void CheckAndCreateToastTable(Oid relOid, Datum reloptions,
41 LOCKMODE lockmode, bool check);
42static bool create_toast_table(Relation rel, Oid toastOid, Oid toastIndexOid,
43 Datum reloptions, LOCKMODE lockmode, bool check);
44static bool needs_toast_table(Relation rel);
45
46
47/*
48 * CreateToastTable variants
49 * If the table needs a toast table, and doesn't already have one,
50 * then create a toast table for it.
51 *
52 * reloptions for the toast table can be passed, too. Pass (Datum) 0
53 * for default reloptions.
54 *
55 * We expect the caller to have verified that the relation is a table and have
56 * already done any necessary permission checks. Callers expect this function
57 * to end with CommandCounterIncrement if it makes any changes.
58 */
59void
60AlterTableCreateToastTable(Oid relOid, Datum reloptions, LOCKMODE lockmode)
61{
62 CheckAndCreateToastTable(relOid, reloptions, lockmode, true);
63}
64
65void
66NewHeapCreateToastTable(Oid relOid, Datum reloptions, LOCKMODE lockmode)
67{
68 CheckAndCreateToastTable(relOid, reloptions, lockmode, false);
69}
70
71void
72NewRelationCreateToastTable(Oid relOid, Datum reloptions)
73{
74 CheckAndCreateToastTable(relOid, reloptions, AccessExclusiveLock, false);
75}
76
77static void
78CheckAndCreateToastTable(Oid relOid, Datum reloptions, LOCKMODE lockmode, bool check)
79{
80 Relation rel;
81
82 rel = table_open(relOid, lockmode);
83
84 /* create_toast_table does all the work */
85 (void) create_toast_table(rel, InvalidOid, InvalidOid, reloptions, lockmode, check);
86
87 table_close(rel, NoLock);
88}
89
90/*
91 * Create a toast table during bootstrap
92 *
93 * Here we need to prespecify the OIDs of the toast table and its index
94 */
95void
96BootstrapToastTable(char *relName, Oid toastOid, Oid toastIndexOid)
97{
98 Relation rel;
99
100 rel = table_openrv(makeRangeVar(NULL, relName, -1), AccessExclusiveLock);
101
102 if (rel->rd_rel->relkind != RELKIND_RELATION &&
103 rel->rd_rel->relkind != RELKIND_MATVIEW)
104 ereport(ERROR,
105 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
106 errmsg("\"%s\" is not a table or materialized view",
107 relName)));
108
109 /* create_toast_table does all the work */
110 if (!create_toast_table(rel, toastOid, toastIndexOid, (Datum) 0,
111 AccessExclusiveLock, false))
112 elog(ERROR, "\"%s\" does not require a toast table",
113 relName);
114
115 table_close(rel, NoLock);
116}
117
118
119/*
120 * create_toast_table --- internal workhorse
121 *
122 * rel is already opened and locked
123 * toastOid and toastIndexOid are normally InvalidOid, but during
124 * bootstrap they can be nonzero to specify hand-assigned OIDs
125 */
126static bool
127create_toast_table(Relation rel, Oid toastOid, Oid toastIndexOid,
128 Datum reloptions, LOCKMODE lockmode, bool check)
129{
130 Oid relOid = RelationGetRelid(rel);
131 HeapTuple reltup;
132 TupleDesc tupdesc;
133 bool shared_relation;
134 bool mapped_relation;
135 Relation toast_rel;
136 Relation class_rel;
137 Oid toast_relid;
138 Oid toast_typid = InvalidOid;
139 Oid namespaceid;
140 char toast_relname[NAMEDATALEN];
141 char toast_idxname[NAMEDATALEN];
142 IndexInfo *indexInfo;
143 Oid collationObjectId[2];
144 Oid classObjectId[2];
145 int16 coloptions[2];
146 ObjectAddress baseobject,
147 toastobject;
148
149 /*
150 * Is it already toasted?
151 */
152 if (rel->rd_rel->reltoastrelid != InvalidOid)
153 return false;
154
155 /*
156 * Check to see whether the table actually needs a TOAST table.
157 */
158 if (!IsBinaryUpgrade)
159 {
160 /* Normal mode, normal check */
161 if (!needs_toast_table(rel))
162 return false;
163 }
164 else
165 {
166 /*
167 * In binary-upgrade mode, create a TOAST table if and only if
168 * pg_upgrade told us to (ie, a TOAST table OID has been provided).
169 *
170 * This indicates that the old cluster had a TOAST table for the
171 * current table. We must create a TOAST table to receive the old
172 * TOAST file, even if the table seems not to need one.
173 *
174 * Contrariwise, if the old cluster did not have a TOAST table, we
175 * should be able to get along without one even if the new version's
176 * needs_toast_table rules suggest we should have one. There is a lot
177 * of daylight between where we will create a TOAST table and where
178 * one is really necessary to avoid failures, so small cross-version
179 * differences in the when-to-create heuristic shouldn't be a problem.
180 * If we tried to create a TOAST table anyway, we would have the
181 * problem that it might take up an OID that will conflict with some
182 * old-cluster table we haven't seen yet.
183 */
184 if (!OidIsValid(binary_upgrade_next_toast_pg_class_oid) ||
185 !OidIsValid(binary_upgrade_next_toast_pg_type_oid))
186 return false;
187 }
188
189 /*
190 * If requested check lockmode is sufficient. This is a cross check in
191 * case of errors or conflicting decisions in earlier code.
192 */
193 if (check && lockmode != AccessExclusiveLock)
194 elog(ERROR, "AccessExclusiveLock required to add toast table.");
195
196 /*
197 * Create the toast table and its index
198 */
199 snprintf(toast_relname, sizeof(toast_relname),
200 "pg_toast_%u", relOid);
201 snprintf(toast_idxname, sizeof(toast_idxname),
202 "pg_toast_%u_index", relOid);
203
204 /* this is pretty painful... need a tuple descriptor */
205 tupdesc = CreateTemplateTupleDesc(3);
206 TupleDescInitEntry(tupdesc, (AttrNumber) 1,
207 "chunk_id",
208 OIDOID,
209 -1, 0);
210 TupleDescInitEntry(tupdesc, (AttrNumber) 2,
211 "chunk_seq",
212 INT4OID,
213 -1, 0);
214 TupleDescInitEntry(tupdesc, (AttrNumber) 3,
215 "chunk_data",
216 BYTEAOID,
217 -1, 0);
218
219 /*
220 * Ensure that the toast table doesn't itself get toasted, or we'll be
221 * toast :-(. This is essential for chunk_data because type bytea is
222 * toastable; hit the other two just to be sure.
223 */
224 TupleDescAttr(tupdesc, 0)->attstorage = 'p';
225 TupleDescAttr(tupdesc, 1)->attstorage = 'p';
226 TupleDescAttr(tupdesc, 2)->attstorage = 'p';
227
228 /*
229 * Toast tables for regular relations go in pg_toast; those for temp
230 * relations go into the per-backend temp-toast-table namespace.
231 */
232 if (isTempOrTempToastNamespace(rel->rd_rel->relnamespace))
233 namespaceid = GetTempToastNamespace();
234 else
235 namespaceid = PG_TOAST_NAMESPACE;
236
237 /*
238 * Use binary-upgrade override for pg_type.oid, if supplied. We might be
239 * in the post-schema-restore phase where we are doing ALTER TABLE to
240 * create TOAST tables that didn't exist in the old cluster.
241 */
242 if (IsBinaryUpgrade && OidIsValid(binary_upgrade_next_toast_pg_type_oid))
243 {
244 toast_typid = binary_upgrade_next_toast_pg_type_oid;
245 binary_upgrade_next_toast_pg_type_oid = InvalidOid;
246 }
247
248 /* Toast table is shared if and only if its parent is. */
249 shared_relation = rel->rd_rel->relisshared;
250
251 /* It's mapped if and only if its parent is, too */
252 mapped_relation = RelationIsMapped(rel);
253
254 toast_relid = heap_create_with_catalog(toast_relname,
255 namespaceid,
256 rel->rd_rel->reltablespace,
257 toastOid,
258 toast_typid,
259 InvalidOid,
260 rel->rd_rel->relowner,
261 rel->rd_rel->relam,
262 tupdesc,
263 NIL,
264 RELKIND_TOASTVALUE,
265 rel->rd_rel->relpersistence,
266 shared_relation,
267 mapped_relation,
268 ONCOMMIT_NOOP,
269 reloptions,
270 false,
271 true,
272 true,
273 InvalidOid,
274 NULL);
275 Assert(toast_relid != InvalidOid);
276
277 /* make the toast relation visible, else table_open will fail */
278 CommandCounterIncrement();
279
280 /* ShareLock is not really needed here, but take it anyway */
281 toast_rel = table_open(toast_relid, ShareLock);
282
283 /*
284 * Create unique index on chunk_id, chunk_seq.
285 *
286 * NOTE: the normal TOAST access routines could actually function with a
287 * single-column index on chunk_id only. However, the slice access
288 * routines use both columns for faster access to an individual chunk. In
289 * addition, we want it to be unique as a check against the possibility of
290 * duplicate TOAST chunk OIDs. The index might also be a little more
291 * efficient this way, since btree isn't all that happy with large numbers
292 * of equal keys.
293 */
294
295 indexInfo = makeNode(IndexInfo);
296 indexInfo->ii_NumIndexAttrs = 2;
297 indexInfo->ii_NumIndexKeyAttrs = 2;
298 indexInfo->ii_IndexAttrNumbers[0] = 1;
299 indexInfo->ii_IndexAttrNumbers[1] = 2;
300 indexInfo->ii_Expressions = NIL;
301 indexInfo->ii_ExpressionsState = NIL;
302 indexInfo->ii_Predicate = NIL;
303 indexInfo->ii_PredicateState = NULL;
304 indexInfo->ii_ExclusionOps = NULL;
305 indexInfo->ii_ExclusionProcs = NULL;
306 indexInfo->ii_ExclusionStrats = NULL;
307 indexInfo->ii_Unique = true;
308 indexInfo->ii_ReadyForInserts = true;
309 indexInfo->ii_Concurrent = false;
310 indexInfo->ii_BrokenHotChain = false;
311 indexInfo->ii_ParallelWorkers = 0;
312 indexInfo->ii_Am = BTREE_AM_OID;
313 indexInfo->ii_AmCache = NULL;
314 indexInfo->ii_Context = CurrentMemoryContext;
315
316 collationObjectId[0] = InvalidOid;
317 collationObjectId[1] = InvalidOid;
318
319 classObjectId[0] = OID_BTREE_OPS_OID;
320 classObjectId[1] = INT4_BTREE_OPS_OID;
321
322 coloptions[0] = 0;
323 coloptions[1] = 0;
324
325 index_create(toast_rel, toast_idxname, toastIndexOid, InvalidOid,
326 InvalidOid, InvalidOid,
327 indexInfo,
328 list_make2("chunk_id", "chunk_seq"),
329 BTREE_AM_OID,
330 rel->rd_rel->reltablespace,
331 collationObjectId, classObjectId, coloptions, (Datum) 0,
332 INDEX_CREATE_IS_PRIMARY, 0, true, true, NULL);
333
334 table_close(toast_rel, NoLock);
335
336 /*
337 * Store the toast table's OID in the parent relation's pg_class row
338 */
339 class_rel = table_open(RelationRelationId, RowExclusiveLock);
340
341 reltup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(relOid));
342 if (!HeapTupleIsValid(reltup))
343 elog(ERROR, "cache lookup failed for relation %u", relOid);
344
345 ((Form_pg_class) GETSTRUCT(reltup))->reltoastrelid = toast_relid;
346
347 if (!IsBootstrapProcessingMode())
348 {
349 /* normal case, use a transactional update */
350 CatalogTupleUpdate(class_rel, &reltup->t_self, reltup);
351 }
352 else
353 {
354 /* While bootstrapping, we cannot UPDATE, so overwrite in-place */
355 heap_inplace_update(class_rel, reltup);
356 }
357
358 heap_freetuple(reltup);
359
360 table_close(class_rel, RowExclusiveLock);
361
362 /*
363 * Register dependency from the toast table to the master, so that the
364 * toast table will be deleted if the master is. Skip this in bootstrap
365 * mode.
366 */
367 if (!IsBootstrapProcessingMode())
368 {
369 baseobject.classId = RelationRelationId;
370 baseobject.objectId = relOid;
371 baseobject.objectSubId = 0;
372 toastobject.classId = RelationRelationId;
373 toastobject.objectId = toast_relid;
374 toastobject.objectSubId = 0;
375
376 recordDependencyOn(&toastobject, &baseobject, DEPENDENCY_INTERNAL);
377 }
378
379 /*
380 * Make changes visible
381 */
382 CommandCounterIncrement();
383
384 return true;
385}
386
387/*
388 * Check to see whether the table needs a TOAST table.
389 */
390static bool
391needs_toast_table(Relation rel)
392{
393 /*
394 * No need to create a TOAST table for partitioned tables.
395 */
396 if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
397 return false;
398
399 /*
400 * We cannot allow toasting a shared relation after initdb (because
401 * there's no way to mark it toasted in other databases' pg_class).
402 */
403 if (rel->rd_rel->relisshared && !IsBootstrapProcessingMode())
404 return false;
405
406 /*
407 * Ignore attempts to create toast tables on catalog tables after initdb.
408 * Which catalogs get toast tables is explicitly chosen in
409 * catalog/toasting.h. (We could get here via some ALTER TABLE command if
410 * the catalog doesn't have a toast table.)
411 */
412 if (IsCatalogRelation(rel) && !IsBootstrapProcessingMode())
413 return false;
414
415 /* Otherwise, let the AM decide. */
416 return table_relation_needs_toast_table(rel);
417}
418