1 /*
2 * hashfuncs.c
3 * Functions to investigate the content of HASH indexes
4 *
5 * Copyright (c) 2017-2019, PostgreSQL Global Development Group
6 *
7 * IDENTIFICATION
8 * contrib/pageinspect/hashfuncs.c
9 */
10
11 #include "postgres.h"
12
13 #include "pageinspect.h"
14
15 #include "access/hash.h"
16 #include "access/htup_details.h"
17 #include "catalog/pg_type.h"
18 #include "catalog/pg_am.h"
19 #include "funcapi.h"
20 #include "miscadmin.h"
21 #include "utils/builtins.h"
22 #include "utils/rel.h"
23
24 PG_FUNCTION_INFO_V1(hash_page_type);
25 PG_FUNCTION_INFO_V1(hash_page_stats);
26 PG_FUNCTION_INFO_V1(hash_page_items);
27 PG_FUNCTION_INFO_V1(hash_bitmap_info);
28 PG_FUNCTION_INFO_V1(hash_metapage_info);
29
30 #define IS_HASH(r) ((r)->rd_rel->relam == HASH_AM_OID)
31
32 /* ------------------------------------------------
33 * structure for single hash page statistics
34 * ------------------------------------------------
35 */
36 typedef struct HashPageStat
37 {
38 int live_items;
39 int dead_items;
40 int page_size;
41 int free_size;
42
43 /* opaque data */
44 BlockNumber hasho_prevblkno;
45 BlockNumber hasho_nextblkno;
46 Bucket hasho_bucket;
47 uint16 hasho_flag;
48 uint16 hasho_page_id;
49 } HashPageStat;
50
51
52 /*
53 * Verify that the given bytea contains a HASH page, or die in the attempt.
54 * A pointer to a palloc'd, properly aligned copy of the page is returned.
55 */
56 static Page
verify_hash_page(bytea * raw_page,int flags)57 verify_hash_page(bytea *raw_page, int flags)
58 {
59 Page page = get_page_from_raw(raw_page);
60 int pagetype = LH_UNUSED_PAGE;
61
62 /* Treat new pages as unused. */
63 if (!PageIsNew(page))
64 {
65 HashPageOpaque pageopaque;
66
67 if (PageGetSpecialSize(page) != MAXALIGN(sizeof(HashPageOpaqueData)))
68 ereport(ERROR,
69 (errcode(ERRCODE_INDEX_CORRUPTED),
70 errmsg("index table contains corrupted page")));
71
72 pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
73 if (pageopaque->hasho_page_id != HASHO_PAGE_ID)
74 ereport(ERROR,
75 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
76 errmsg("page is not a hash page"),
77 errdetail("Expected %08x, got %08x.",
78 HASHO_PAGE_ID, pageopaque->hasho_page_id)));
79
80 pagetype = pageopaque->hasho_flag & LH_PAGE_TYPE;
81 }
82
83 /* Check that page type is sane. */
84 if (pagetype != LH_OVERFLOW_PAGE && pagetype != LH_BUCKET_PAGE &&
85 pagetype != LH_BITMAP_PAGE && pagetype != LH_META_PAGE &&
86 pagetype != LH_UNUSED_PAGE)
87 ereport(ERROR,
88 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
89 errmsg("invalid hash page type %08x", pagetype)));
90
91 /* If requested, verify page type. */
92 if (flags != 0 && (pagetype & flags) == 0)
93 {
94 switch (flags)
95 {
96 case LH_META_PAGE:
97 ereport(ERROR,
98 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
99 errmsg("page is not a hash meta page")));
100 break;
101 case LH_BUCKET_PAGE | LH_OVERFLOW_PAGE:
102 ereport(ERROR,
103 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
104 errmsg("page is not a hash bucket or overflow page")));
105 break;
106 case LH_OVERFLOW_PAGE:
107 ereport(ERROR,
108 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
109 errmsg("page is not a hash overflow page")));
110 break;
111 default:
112 elog(ERROR,
113 "hash page of type %08x not in mask %08x",
114 pagetype, flags);
115 break;
116 }
117 }
118
119 /*
120 * If it is the metapage, also verify magic number and version.
121 */
122 if (pagetype == LH_META_PAGE)
123 {
124 HashMetaPage metap = HashPageGetMeta(page);
125
126 if (metap->hashm_magic != HASH_MAGIC)
127 ereport(ERROR,
128 (errcode(ERRCODE_INDEX_CORRUPTED),
129 errmsg("invalid magic number for metadata"),
130 errdetail("Expected 0x%08x, got 0x%08x.",
131 HASH_MAGIC, metap->hashm_magic)));
132
133 if (metap->hashm_version != HASH_VERSION)
134 ereport(ERROR,
135 (errcode(ERRCODE_INDEX_CORRUPTED),
136 errmsg("invalid version for metadata"),
137 errdetail("Expected %d, got %d",
138 HASH_VERSION, metap->hashm_version)));
139 }
140
141 return page;
142 }
143
144 /* -------------------------------------------------
145 * GetHashPageStatistics()
146 *
147 * Collect statistics of single hash page
148 * -------------------------------------------------
149 */
150 static void
GetHashPageStatistics(Page page,HashPageStat * stat)151 GetHashPageStatistics(Page page, HashPageStat *stat)
152 {
153 OffsetNumber maxoff = PageGetMaxOffsetNumber(page);
154 HashPageOpaque opaque = (HashPageOpaque) PageGetSpecialPointer(page);
155 int off;
156
157 stat->dead_items = stat->live_items = 0;
158 stat->page_size = PageGetPageSize(page);
159
160 /* hash page opaque data */
161 stat->hasho_prevblkno = opaque->hasho_prevblkno;
162 stat->hasho_nextblkno = opaque->hasho_nextblkno;
163 stat->hasho_bucket = opaque->hasho_bucket;
164 stat->hasho_flag = opaque->hasho_flag;
165 stat->hasho_page_id = opaque->hasho_page_id;
166
167 /* count live and dead tuples, and free space */
168 for (off = FirstOffsetNumber; off <= maxoff; off++)
169 {
170 ItemId id = PageGetItemId(page, off);
171
172 if (!ItemIdIsDead(id))
173 stat->live_items++;
174 else
175 stat->dead_items++;
176 }
177 stat->free_size = PageGetFreeSpace(page);
178 }
179
180 /* ---------------------------------------------------
181 * hash_page_type()
182 *
183 * Usage: SELECT hash_page_type(get_raw_page('con_hash_index', 1));
184 * ---------------------------------------------------
185 */
186 Datum
hash_page_type(PG_FUNCTION_ARGS)187 hash_page_type(PG_FUNCTION_ARGS)
188 {
189 bytea *raw_page = PG_GETARG_BYTEA_P(0);
190 Page page;
191 HashPageOpaque opaque;
192 int pagetype;
193 const char *type;
194
195 if (!superuser())
196 ereport(ERROR,
197 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
198 (errmsg("must be superuser to use raw page functions"))));
199
200 page = verify_hash_page(raw_page, 0);
201
202 if (PageIsNew(page))
203 type = "unused";
204 else
205 {
206 opaque = (HashPageOpaque) PageGetSpecialPointer(page);
207
208 /* page type (flags) */
209 pagetype = opaque->hasho_flag & LH_PAGE_TYPE;
210 if (pagetype == LH_META_PAGE)
211 type = "metapage";
212 else if (pagetype == LH_OVERFLOW_PAGE)
213 type = "overflow";
214 else if (pagetype == LH_BUCKET_PAGE)
215 type = "bucket";
216 else if (pagetype == LH_BITMAP_PAGE)
217 type = "bitmap";
218 else
219 type = "unused";
220 }
221
222 PG_RETURN_TEXT_P(cstring_to_text(type));
223 }
224
225 /* ---------------------------------------------------
226 * hash_page_stats()
227 *
228 * Usage: SELECT * FROM hash_page_stats(get_raw_page('con_hash_index', 1));
229 * ---------------------------------------------------
230 */
231 Datum
hash_page_stats(PG_FUNCTION_ARGS)232 hash_page_stats(PG_FUNCTION_ARGS)
233 {
234 bytea *raw_page = PG_GETARG_BYTEA_P(0);
235 Page page;
236 int j;
237 Datum values[9];
238 bool nulls[9];
239 HashPageStat stat;
240 HeapTuple tuple;
241 TupleDesc tupleDesc;
242
243 if (!superuser())
244 ereport(ERROR,
245 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
246 (errmsg("must be superuser to use raw page functions"))));
247
248 page = verify_hash_page(raw_page, LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
249
250 /* keep compiler quiet */
251 stat.hasho_prevblkno = stat.hasho_nextblkno = InvalidBlockNumber;
252 stat.hasho_flag = stat.hasho_page_id = stat.free_size = 0;
253
254 GetHashPageStatistics(page, &stat);
255
256 /* Build a tuple descriptor for our result type */
257 if (get_call_result_type(fcinfo, NULL, &tupleDesc) != TYPEFUNC_COMPOSITE)
258 elog(ERROR, "return type must be a row type");
259 tupleDesc = BlessTupleDesc(tupleDesc);
260
261 MemSet(nulls, 0, sizeof(nulls));
262
263 j = 0;
264 values[j++] = Int32GetDatum(stat.live_items);
265 values[j++] = Int32GetDatum(stat.dead_items);
266 values[j++] = Int32GetDatum(stat.page_size);
267 values[j++] = Int32GetDatum(stat.free_size);
268 values[j++] = Int64GetDatum((int64) stat.hasho_prevblkno);
269 values[j++] = Int64GetDatum((int64) stat.hasho_nextblkno);
270 values[j++] = Int64GetDatum((int64) stat.hasho_bucket);
271 values[j++] = Int32GetDatum((int32) stat.hasho_flag);
272 values[j++] = Int32GetDatum((int32) stat.hasho_page_id);
273
274 tuple = heap_form_tuple(tupleDesc, values, nulls);
275
276 PG_RETURN_DATUM(HeapTupleGetDatum(tuple));
277 }
278
279 /*
280 * cross-call data structure for SRF
281 */
282 struct user_args
283 {
284 Page page;
285 OffsetNumber offset;
286 };
287
288 /*-------------------------------------------------------
289 * hash_page_items()
290 *
291 * Get IndexTupleData set in a hash page
292 *
293 * Usage: SELECT * FROM hash_page_items(get_raw_page('con_hash_index', 1));
294 *-------------------------------------------------------
295 */
296 Datum
hash_page_items(PG_FUNCTION_ARGS)297 hash_page_items(PG_FUNCTION_ARGS)
298 {
299 bytea *raw_page = PG_GETARG_BYTEA_P(0);
300 Page page;
301 Datum result;
302 Datum values[3];
303 bool nulls[3];
304 uint32 hashkey;
305 HeapTuple tuple;
306 FuncCallContext *fctx;
307 MemoryContext mctx;
308 struct user_args *uargs;
309
310 if (!superuser())
311 ereport(ERROR,
312 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
313 (errmsg("must be superuser to use raw page functions"))));
314
315 if (SRF_IS_FIRSTCALL())
316 {
317 TupleDesc tupleDesc;
318
319 fctx = SRF_FIRSTCALL_INIT();
320
321 mctx = MemoryContextSwitchTo(fctx->multi_call_memory_ctx);
322
323 page = verify_hash_page(raw_page, LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
324
325 uargs = palloc(sizeof(struct user_args));
326
327 uargs->page = page;
328
329 uargs->offset = FirstOffsetNumber;
330
331 fctx->max_calls = PageGetMaxOffsetNumber(uargs->page);
332
333 /* Build a tuple descriptor for our result type */
334 if (get_call_result_type(fcinfo, NULL, &tupleDesc) != TYPEFUNC_COMPOSITE)
335 elog(ERROR, "return type must be a row type");
336 tupleDesc = BlessTupleDesc(tupleDesc);
337
338 fctx->attinmeta = TupleDescGetAttInMetadata(tupleDesc);
339
340 fctx->user_fctx = uargs;
341
342 MemoryContextSwitchTo(mctx);
343 }
344
345 fctx = SRF_PERCALL_SETUP();
346 uargs = fctx->user_fctx;
347
348 if (fctx->call_cntr < fctx->max_calls)
349 {
350 ItemId id;
351 IndexTuple itup;
352 int j;
353
354 id = PageGetItemId(uargs->page, uargs->offset);
355
356 if (!ItemIdIsValid(id))
357 elog(ERROR, "invalid ItemId");
358
359 itup = (IndexTuple) PageGetItem(uargs->page, id);
360
361 MemSet(nulls, 0, sizeof(nulls));
362
363 j = 0;
364 values[j++] = Int32GetDatum((int32) uargs->offset);
365 values[j++] = PointerGetDatum(&itup->t_tid);
366
367 hashkey = _hash_get_indextuple_hashkey(itup);
368 values[j] = Int64GetDatum((int64) hashkey);
369
370 tuple = heap_form_tuple(fctx->attinmeta->tupdesc, values, nulls);
371 result = HeapTupleGetDatum(tuple);
372
373 uargs->offset = uargs->offset + 1;
374
375 SRF_RETURN_NEXT(fctx, result);
376 }
377 else
378 {
379 pfree(uargs);
380 SRF_RETURN_DONE(fctx);
381 }
382 }
383
384 /* ------------------------------------------------
385 * hash_bitmap_info()
386 *
387 * Get bitmap information for a particular overflow page
388 *
389 * Usage: SELECT * FROM hash_bitmap_info('con_hash_index'::regclass, 5);
390 * ------------------------------------------------
391 */
392 Datum
hash_bitmap_info(PG_FUNCTION_ARGS)393 hash_bitmap_info(PG_FUNCTION_ARGS)
394 {
395 Oid indexRelid = PG_GETARG_OID(0);
396 uint64 ovflblkno = PG_GETARG_INT64(1);
397 HashMetaPage metap;
398 Buffer metabuf,
399 mapbuf;
400 BlockNumber bitmapblkno;
401 Page mappage;
402 bool bit = false;
403 TupleDesc tupleDesc;
404 Relation indexRel;
405 uint32 ovflbitno;
406 int32 bitmappage,
407 bitmapbit;
408 HeapTuple tuple;
409 int i,
410 j;
411 Datum values[3];
412 bool nulls[3];
413 uint32 *freep;
414
415 if (!superuser())
416 ereport(ERROR,
417 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
418 (errmsg("must be superuser to use raw page functions"))));
419
420 indexRel = index_open(indexRelid, AccessShareLock);
421
422 if (!IS_HASH(indexRel))
423 elog(ERROR, "relation \"%s\" is not a hash index",
424 RelationGetRelationName(indexRel));
425
426 if (RELATION_IS_OTHER_TEMP(indexRel))
427 ereport(ERROR,
428 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
429 errmsg("cannot access temporary tables of other sessions")));
430
431 if (ovflblkno >= RelationGetNumberOfBlocks(indexRel))
432 ereport(ERROR,
433 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
434 errmsg("block number " UINT64_FORMAT " is out of range for relation \"%s\"",
435 ovflblkno, RelationGetRelationName(indexRel))));
436
437 /* Read the metapage so we can determine which bitmap page to use */
438 metabuf = _hash_getbuf(indexRel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
439 metap = HashPageGetMeta(BufferGetPage(metabuf));
440
441 /*
442 * Reject attempt to read the bit for a metapage or bitmap page; this is
443 * only meaningful for overflow pages.
444 */
445 if (ovflblkno == 0)
446 ereport(ERROR,
447 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
448 errmsg("invalid overflow block number %u",
449 (BlockNumber) ovflblkno)));
450 for (i = 0; i < metap->hashm_nmaps; i++)
451 if (metap->hashm_mapp[i] == ovflblkno)
452 ereport(ERROR,
453 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
454 errmsg("invalid overflow block number %u",
455 (BlockNumber) ovflblkno)));
456
457 /*
458 * Identify overflow bit number. This will error out for primary bucket
459 * pages, and we've already rejected the metapage and bitmap pages above.
460 */
461 ovflbitno = _hash_ovflblkno_to_bitno(metap, (BlockNumber) ovflblkno);
462
463 bitmappage = ovflbitno >> BMPG_SHIFT(metap);
464 bitmapbit = ovflbitno & BMPG_MASK(metap);
465
466 if (bitmappage >= metap->hashm_nmaps)
467 ereport(ERROR,
468 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
469 errmsg("invalid overflow block number %u",
470 (BlockNumber) ovflblkno)));
471
472 bitmapblkno = metap->hashm_mapp[bitmappage];
473
474 _hash_relbuf(indexRel, metabuf);
475
476 /* Check the status of bitmap bit for overflow page */
477 mapbuf = _hash_getbuf(indexRel, bitmapblkno, HASH_READ, LH_BITMAP_PAGE);
478 mappage = BufferGetPage(mapbuf);
479 freep = HashPageGetBitmap(mappage);
480
481 bit = ISSET(freep, bitmapbit) != 0;
482
483 _hash_relbuf(indexRel, mapbuf);
484 index_close(indexRel, AccessShareLock);
485
486 /* Build a tuple descriptor for our result type */
487 if (get_call_result_type(fcinfo, NULL, &tupleDesc) != TYPEFUNC_COMPOSITE)
488 elog(ERROR, "return type must be a row type");
489 tupleDesc = BlessTupleDesc(tupleDesc);
490
491 MemSet(nulls, 0, sizeof(nulls));
492
493 j = 0;
494 values[j++] = Int64GetDatum((int64) bitmapblkno);
495 values[j++] = Int32GetDatum(bitmapbit);
496 values[j++] = BoolGetDatum(bit);
497
498 tuple = heap_form_tuple(tupleDesc, values, nulls);
499
500 PG_RETURN_DATUM(HeapTupleGetDatum(tuple));
501 }
502
503 /* ------------------------------------------------
504 * hash_metapage_info()
505 *
506 * Get the meta-page information for a hash index
507 *
508 * Usage: SELECT * FROM hash_metapage_info(get_raw_page('con_hash_index', 0))
509 * ------------------------------------------------
510 */
511 Datum
hash_metapage_info(PG_FUNCTION_ARGS)512 hash_metapage_info(PG_FUNCTION_ARGS)
513 {
514 bytea *raw_page = PG_GETARG_BYTEA_P(0);
515 Page page;
516 HashMetaPageData *metad;
517 TupleDesc tupleDesc;
518 HeapTuple tuple;
519 int i,
520 j;
521 Datum values[16];
522 bool nulls[16];
523 Datum spares[HASH_MAX_SPLITPOINTS];
524 Datum mapp[HASH_MAX_BITMAPS];
525
526 if (!superuser())
527 ereport(ERROR,
528 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
529 (errmsg("must be superuser to use raw page functions"))));
530
531 page = verify_hash_page(raw_page, LH_META_PAGE);
532
533 /* Build a tuple descriptor for our result type */
534 if (get_call_result_type(fcinfo, NULL, &tupleDesc) != TYPEFUNC_COMPOSITE)
535 elog(ERROR, "return type must be a row type");
536 tupleDesc = BlessTupleDesc(tupleDesc);
537
538 metad = HashPageGetMeta(page);
539
540 MemSet(nulls, 0, sizeof(nulls));
541
542 j = 0;
543 values[j++] = Int64GetDatum((int64) metad->hashm_magic);
544 values[j++] = Int64GetDatum((int64) metad->hashm_version);
545 values[j++] = Float8GetDatum(metad->hashm_ntuples);
546 values[j++] = Int32GetDatum((int32) metad->hashm_ffactor);
547 values[j++] = Int32GetDatum((int32) metad->hashm_bsize);
548 values[j++] = Int32GetDatum((int32) metad->hashm_bmsize);
549 values[j++] = Int32GetDatum((int32) metad->hashm_bmshift);
550 values[j++] = Int64GetDatum((int64) metad->hashm_maxbucket);
551 values[j++] = Int64GetDatum((int64) metad->hashm_highmask);
552 values[j++] = Int64GetDatum((int64) metad->hashm_lowmask);
553 values[j++] = Int64GetDatum((int64) metad->hashm_ovflpoint);
554 values[j++] = Int64GetDatum((int64) metad->hashm_firstfree);
555 values[j++] = Int64GetDatum((int64) metad->hashm_nmaps);
556 values[j++] = ObjectIdGetDatum((Oid) metad->hashm_procid);
557
558 for (i = 0; i < HASH_MAX_SPLITPOINTS; i++)
559 spares[i] = Int64GetDatum((int64) metad->hashm_spares[i]);
560 values[j++] = PointerGetDatum(construct_array(spares,
561 HASH_MAX_SPLITPOINTS,
562 INT8OID,
563 8, FLOAT8PASSBYVAL, 'd'));
564
565 for (i = 0; i < HASH_MAX_BITMAPS; i++)
566 mapp[i] = Int64GetDatum((int64) metad->hashm_mapp[i]);
567 values[j++] = PointerGetDatum(construct_array(mapp,
568 HASH_MAX_BITMAPS,
569 INT8OID,
570 8, FLOAT8PASSBYVAL, 'd'));
571
572 tuple = heap_form_tuple(tupleDesc, values, nulls);
573
574 PG_RETURN_DATUM(HeapTupleGetDatum(tuple));
575 }
576