contrib/pageinspect/btreefuncs.c

   1 /*
   2  * $PostgreSQL:$
   3  *
   4  *
   5  * btreefuncs.c
   6  *
   7  * Copyright (c) 2006 Satoshi Nagayasu <nagayasus@nttdata.co.jp>
   8  *
   9  * Permission to use, copy, modify, and distribute this software and
  10  * its documentation for any purpose, without fee, and without a
  11  * written agreement is hereby granted, provided that the above
  12  * copyright notice and this paragraph and the following two
  13  * paragraphs appear in all copies.
  14  *
  15  * IN NO EVENT SHALL THE AUTHOR BE LIABLE TO ANY PARTY FOR DIRECT,
  16  * INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES, INCLUDING
  17  * LOST PROFITS, ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS
  18  * DOCUMENTATION, EVEN IF THE UNIVERSITY OF CALIFORNIA HAS BEEN ADVISED
  19  * OF THE POSSIBILITY OF SUCH DAMAGE.
  20  *
  21  * THE AUTHOR SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING, BUT NOT
  22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  23  * A PARTICULAR PURPOSE.  THE SOFTWARE PROVIDED HEREUNDER IS ON AN "AS
  24  * IS" BASIS, AND THE AUTHOR HAS NO OBLIGATIONS TO PROVIDE MAINTENANCE,
  25  * SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
  26  */
  27
  28 #include "postgres.h"
  29
  30 #include "access/heapam.h"
  31 #include "access/nbtree.h"
  32 #include "catalog/namespace.h"
  33 #include "catalog/pg_type.h"
  34 #include "funcapi.h"
  35 #include "miscadmin.h"
  36 #include "storage/bufmgr.h"
  37 #include "utils/builtins.h"
  38
  39
  40 extern Datum bt_metap(PG_FUNCTION_ARGS);
  41 extern Datum bt_page_items(PG_FUNCTION_ARGS);
  42 extern Datum bt_page_stats(PG_FUNCTION_ARGS);
  43
  44 PG_FUNCTION_INFO_V1(bt_metap);
  45 PG_FUNCTION_INFO_V1(bt_page_items);
  46 PG_FUNCTION_INFO_V1(bt_page_stats);
  47
  48 #define IS_INDEX(r) ((r)->rd_rel->relkind == RELKIND_INDEX)
  49 #define IS_BTREE(r) ((r)->rd_rel->relam == BTREE_AM_OID)
  50
  51 #define CHECK_PAGE_OFFSET_RANGE(pg, offnum) { \
  52                 if ( !(FirstOffsetNumber <= (offnum) && \
  53                                                 (offnum) <= PageGetMaxOffsetNumber(pg)) ) \
  54                          elog(ERROR, "page offset number out of range"); }
  55
  56 /* note: BlockNumber is unsigned, hence can't be negative */
  57 #define CHECK_RELATION_BLOCK_RANGE(rel, blkno) { \
  58                 if ( RelationGetNumberOfBlocks(rel) <= (BlockNumber) (blkno) ) \
  59                          elog(ERROR, "block number out of range"); }
  60
  61 /* ------------------------------------------------
  62  * structure for single btree page statistics
  63  * ------------------------------------------------
  64  */
  65 typedef struct BTPageStat
  66 {
  67         uint32          blkno;
  68         uint32          live_items;
  69         uint32          dead_items;
  70         uint32          page_size;
  71         uint32          max_avail;
  72         uint32          free_size;
  73         uint32          avg_item_size;
  74         char            type;
  75
  76         /* opaque data */
  77         BlockNumber btpo_prev;
  78         BlockNumber btpo_next;
  79         union
  80         {
  81                 uint32          level;
  82                 TransactionId xact;
  83         }                       btpo;
  84         uint16          btpo_flags;
  85         BTCycleId       btpo_cycleid;
  86 }       BTPageStat;
  87
  88
  89 /* -------------------------------------------------
  90  * GetBTPageStatistics()
  91  *
  92  * Collect statistics of single b-tree page
  93  * -------------------------------------------------
  94  */
  95 static void
  96 GetBTPageStatistics(BlockNumber blkno, Buffer buffer, BTPageStat * stat)
  97 {
  98         Page            page = BufferGetPage(buffer);
  99         PageHeader      phdr = (PageHeader) page;
 100         OffsetNumber maxoff = PageGetMaxOffsetNumber(page);
 101         BTPageOpaque opaque = (BTPageOpaque) PageGetSpecialPointer(page);
 102         int                     item_size = 0;
 103         int                     off;
 104
 105         stat->blkno = blkno;
 106
 107         stat->max_avail = BLCKSZ - (BLCKSZ - phdr->pd_special + SizeOfPageHeaderData);
 108
 109         stat->dead_items = stat->live_items = 0;
 110
 111         stat->page_size = PageGetPageSize(page);
 112
 113         /* page type (flags) */
 114         if (P_ISDELETED(opaque))
 115         {
 116                 stat->type = 'd';
 117                 stat->btpo.xact = opaque->btpo.xact;
 118                 return;
 119         }
 120         else if (P_IGNORE(opaque))
 121                 stat->type = 'e';
 122         else if (P_ISLEAF(opaque))
 123                 stat->type = 'l';
 124         else if (P_ISROOT(opaque))
 125                 stat->type = 'r';
 126         else
 127                 stat->type = 'i';
 128
 129         /* btpage opaque data */
 130         stat->btpo_prev = opaque->btpo_prev;
 131         stat->btpo_next = opaque->btpo_next;
 132         stat->btpo.level = opaque->btpo.level;
 133         stat->btpo_flags = opaque->btpo_flags;
 134         stat->btpo_cycleid = opaque->btpo_cycleid;
 135
 136         /* count live and dead tuples, and free space */
 137         for (off = FirstOffsetNumber; off <= maxoff; off++)
 138         {
 139                 IndexTuple      itup;
 140
 141                 ItemId          id = PageGetItemId(page, off);
 142
 143                 itup = (IndexTuple) PageGetItem(page, id);
 144
 145                 item_size += IndexTupleSize(itup);
 146
 147                 if (!ItemIdIsDead(id))
 148                         stat->live_items++;
 149                 else
 150                         stat->dead_items++;
 151         }
 152         stat->free_size = PageGetFreeSpace(page);
 153
 154         if ((stat->live_items + stat->dead_items) > 0)
 155                 stat->avg_item_size = item_size / (stat->live_items + stat->dead_items);
 156         else
 157                 stat->avg_item_size = 0;
 158 }
 159
 160 /* -----------------------------------------------
 161  * bt_page()
 162  *
 163  * Usage: SELECT * FROM bt_page('t1_pkey', 1);
 164  * -----------------------------------------------
 165  */
 166 Datum
 167 bt_page_stats(PG_FUNCTION_ARGS)
 168 {
 169         text       *relname = PG_GETARG_TEXT_P(0);
 170         uint32          blkno = PG_GETARG_UINT32(1);
 171         Buffer          buffer;
 172         Relation        rel;
 173         RangeVar   *relrv;
 174         Datum           result;
 175         HeapTuple       tuple;
 176         TupleDesc       tupleDesc;
 177         int                     j;
 178         char       *values[11];
 179         BTPageStat      stat;
 180
 181         if (!superuser())
 182                 ereport(ERROR,
 183                                 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
 184                                  (errmsg("must be superuser to use pageinspect functions"))));
 185
 186         relrv = makeRangeVarFromNameList(textToQualifiedNameList(relname));
 187         rel = relation_openrv(relrv, AccessShareLock);
 188
 189         if (!IS_INDEX(rel) || !IS_BTREE(rel))
 190                 elog(ERROR, "relation \"%s\" is not a btree index",
 191                          RelationGetRelationName(rel));
 192
 193         if (blkno == 0)
 194                 elog(ERROR, "block 0 is a meta page");
 195
 196         CHECK_RELATION_BLOCK_RANGE(rel, blkno);
 197
 198         buffer = ReadBuffer(rel, blkno);
 199
 200         /* keep compiler quiet */
 201         stat.btpo_prev = stat.btpo_next = InvalidBlockNumber;
 202         stat.btpo_flags = stat.free_size = stat.avg_item_size = 0;
 203
 204         GetBTPageStatistics(blkno, buffer, &stat);
 205
 206         /* Build a tuple descriptor for our result type */
 207         if (get_call_result_type(fcinfo, NULL, &tupleDesc) != TYPEFUNC_COMPOSITE)
 208                 elog(ERROR, "return type must be a row type");
 209
 210         j = 0;
 211         values[j] = palloc(32);
 212         snprintf(values[j++], 32, "%d", stat.blkno);
 213         values[j] = palloc(32);
 214         snprintf(values[j++], 32, "%c", stat.type);
 215         values[j] = palloc(32);
 216         snprintf(values[j++], 32, "%d", stat.live_items);
 217         values[j] = palloc(32);
 218         snprintf(values[j++], 32, "%d", stat.dead_items);
 219         values[j] = palloc(32);
 220         snprintf(values[j++], 32, "%d", stat.avg_item_size);
 221         values[j] = palloc(32);
 222         snprintf(values[j++], 32, "%d", stat.page_size);
 223         values[j] = palloc(32);
 224         snprintf(values[j++], 32, "%d", stat.free_size);
 225         values[j] = palloc(32);
 226         snprintf(values[j++], 32, "%d", stat.btpo_prev);
 227         values[j] = palloc(32);
 228         snprintf(values[j++], 32, "%d", stat.btpo_next);
 229         values[j] = palloc(32);
 230         if (stat.type == 'd')
 231                 snprintf(values[j++], 32, "%d", stat.btpo.xact);
 232         else
 233                 snprintf(values[j++], 32, "%d", stat.btpo.level);
 234         values[j] = palloc(32);
 235         snprintf(values[j++], 32, "%d", stat.btpo_flags);
 236
 237         tuple = BuildTupleFromCStrings(TupleDescGetAttInMetadata(tupleDesc),
 238                                                                    values);
 239
 240         result = HeapTupleGetDatum(tuple);
 241
 242         ReleaseBuffer(buffer);
 243
 244         relation_close(rel, AccessShareLock);
 245
 246         PG_RETURN_DATUM(result);
 247 }
 248
 249 /*-------------------------------------------------------
 250  * bt_page_items()
 251  *
 252  * Get IndexTupleData set in a btree page
 253  *
 254  * Usage: SELECT * FROM bt_page_items('t1_pkey', 1);
 255  *-------------------------------------------------------
 256  */
 257
 258 /*
 259  * cross-call data structure for SRF
 260  */
 261 struct user_args
 262 {
 263         Page            page;
 264         OffsetNumber offset;
 265 };
 266
 267 Datum
 268 bt_page_items(PG_FUNCTION_ARGS)
 269 {
 270         text       *relname = PG_GETARG_TEXT_P(0);
 271         uint32          blkno = PG_GETARG_UINT32(1);
 272         Datum           result;
 273         char       *values[6];
 274         HeapTuple       tuple;
 275         FuncCallContext *fctx;
 276         MemoryContext mctx;
 277         struct user_args *uargs;
 278
 279         if (!superuser())
 280                 ereport(ERROR,
 281                                 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
 282                                  (errmsg("must be superuser to use pageinspect functions"))));
 283
 284         if (SRF_IS_FIRSTCALL())
 285         {
 286                 RangeVar   *relrv;
 287                 Relation        rel;
 288                 Buffer          buffer;
 289                 BTPageOpaque opaque;
 290                 TupleDesc       tupleDesc;
 291
 292                 fctx = SRF_FIRSTCALL_INIT();
 293
 294                 relrv = makeRangeVarFromNameList(textToQualifiedNameList(relname));
 295                 rel = relation_openrv(relrv, AccessShareLock);
 296
 297                 if (!IS_INDEX(rel) || !IS_BTREE(rel))
 298                         elog(ERROR, "relation \"%s\" is not a btree index",
 299                                  RelationGetRelationName(rel));
 300
 301                 if (blkno == 0)
 302                         elog(ERROR, "block 0 is a meta page");
 303
 304                 CHECK_RELATION_BLOCK_RANGE(rel, blkno);
 305
 306                 buffer = ReadBuffer(rel, blkno);
 307
 308                 /*
 309                  * We copy the page into local storage to avoid holding pin on the
 310                  * buffer longer than we must, and possibly failing to release it at
 311                  * all if the calling query doesn't fetch all rows.
 312                  */
 313                 mctx = MemoryContextSwitchTo(fctx->multi_call_memory_ctx);
 314
 315                 uargs = palloc(sizeof(struct user_args));
 316
 317                 uargs->page = palloc(BLCKSZ);
 318                 memcpy(uargs->page, BufferGetPage(buffer), BLCKSZ);
 319
 320                 ReleaseBuffer(buffer);
 321                 relation_close(rel, AccessShareLock);
 322
 323                 uargs->offset = FirstOffsetNumber;
 324
 325                 opaque = (BTPageOpaque) PageGetSpecialPointer(uargs->page);
 326
 327                 if (P_ISDELETED(opaque))
 328                         elog(NOTICE, "page is deleted");
 329
 330                 fctx->max_calls = PageGetMaxOffsetNumber(uargs->page);
 331
 332                 /* Build a tuple descriptor for our result type */
 333                 if (get_call_result_type(fcinfo, NULL, &tupleDesc) != TYPEFUNC_COMPOSITE)
 334                         elog(ERROR, "return type must be a row type");
 335
 336                 fctx->attinmeta = TupleDescGetAttInMetadata(tupleDesc);
 337
 338                 fctx->user_fctx = uargs;
 339
 340                 MemoryContextSwitchTo(mctx);
 341         }
 342
 343         fctx = SRF_PERCALL_SETUP();
 344         uargs = fctx->user_fctx;
 345
 346         if (fctx->call_cntr < fctx->max_calls)
 347         {
 348                 ItemId          id;
 349                 IndexTuple      itup;
 350                 int                     j;
 351                 int                     off;
 352                 int                     dlen;
 353                 char       *dump;
 354                 char       *ptr;
 355
 356                 id = PageGetItemId(uargs->page, uargs->offset);
 357
 358                 if (!ItemIdIsValid(id))
 359                         elog(ERROR, "invalid ItemId");
 360
 361                 itup = (IndexTuple) PageGetItem(uargs->page, id);
 362
 363                 j = 0;
 364                 values[j] = palloc(32);
 365                 snprintf(values[j++], 32, "%d", uargs->offset);
 366                 values[j] = palloc(32);
 367                 snprintf(values[j++], 32, "(%u,%u)",
 368                                  BlockIdGetBlockNumber(&(itup->t_tid.ip_blkid)),
 369                                  itup->t_tid.ip_posid);
 370                 values[j] = palloc(32);
 371                 snprintf(values[j++], 32, "%d", (int) IndexTupleSize(itup));
 372                 values[j] = palloc(32);
 373                 snprintf(values[j++], 32, "%c", IndexTupleHasNulls(itup) ? 't' : 'f');
 374                 values[j] = palloc(32);
 375                 snprintf(values[j++], 32, "%c", IndexTupleHasVarwidths(itup) ? 't' : 'f');
 376
 377                 ptr = (char *) itup + IndexInfoFindDataOffset(itup->t_info);
 378                 dlen = IndexTupleSize(itup) - IndexInfoFindDataOffset(itup->t_info);
 379                 dump = palloc0(dlen * 3 + 1);
 380                 values[j] = dump;
 381                 for (off = 0; off < dlen; off++)
 382                 {
 383                         if (off > 0)
 384                                 *dump++ = ' ';
 385                         sprintf(dump, "%02x", *(ptr + off) & 0xff);
 386                         dump += 2;
 387                 }
 388
 389                 tuple = BuildTupleFromCStrings(fctx->attinmeta, values);
 390                 result = HeapTupleGetDatum(tuple);
 391
 392                 uargs->offset = uargs->offset + 1;
 393
 394                 SRF_RETURN_NEXT(fctx, result);
 395         }
 396         else
 397         {
 398                 pfree(uargs->page);
 399                 pfree(uargs);
 400                 SRF_RETURN_DONE(fctx);
 401         }
 402 }
 403
 404
 405 /* ------------------------------------------------
 406  * bt_metap()
 407  *
 408  * Get a btree's meta-page information
 409  *
 410  * Usage: SELECT * FROM bt_metap('t1_pkey')
 411  * ------------------------------------------------
 412  */
 413 Datum
 414 bt_metap(PG_FUNCTION_ARGS)
 415 {
 416         text       *relname = PG_GETARG_TEXT_P(0);
 417         Datum           result;
 418         Relation        rel;
 419         RangeVar   *relrv;
 420         BTMetaPageData *metad;
 421         TupleDesc       tupleDesc;
 422         int                     j;
 423         char       *values[6];
 424         Buffer          buffer;
 425         Page            page;
 426         HeapTuple       tuple;
 427
 428         if (!superuser())
 429                 ereport(ERROR,
 430                                 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
 431                                  (errmsg("must be superuser to use pageinspect functions"))));
 432
 433         relrv = makeRangeVarFromNameList(textToQualifiedNameList(relname));
 434         rel = relation_openrv(relrv, AccessShareLock);
 435
 436         if (!IS_INDEX(rel) || !IS_BTREE(rel))
 437                 elog(ERROR, "relation \"%s\" is not a btree index",
 438                          RelationGetRelationName(rel));
 439
 440         buffer = ReadBuffer(rel, 0);
 441         page = BufferGetPage(buffer);
 442         metad = BTPageGetMeta(page);
 443
 444         /* Build a tuple descriptor for our result type */
 445         if (get_call_result_type(fcinfo, NULL, &tupleDesc) != TYPEFUNC_COMPOSITE)
 446                 elog(ERROR, "return type must be a row type");
 447
 448         j = 0;
 449         values[j] = palloc(32);
 450         snprintf(values[j++], 32, "%d", metad->btm_magic);
 451         values[j] = palloc(32);
 452         snprintf(values[j++], 32, "%d", metad->btm_version);
 453         values[j] = palloc(32);
 454         snprintf(values[j++], 32, "%d", metad->btm_root);
 455         values[j] = palloc(32);
 456         snprintf(values[j++], 32, "%d", metad->btm_level);
 457         values[j] = palloc(32);
 458         snprintf(values[j++], 32, "%d", metad->btm_fastroot);
 459         values[j] = palloc(32);
 460         snprintf(values[j++], 32, "%d", metad->btm_fastlevel);
 461
 462         tuple = BuildTupleFromCStrings(TupleDescGetAttInMetadata(tupleDesc),
 463                                                                    values);
 464
 465         result = HeapTupleGetDatum(tuple);
 466
 467         ReleaseBuffer(buffer);
 468
 469         relation_close(rel, AccessShareLock);
 470
 471         PG_RETURN_DATUM(result);
 472 }