src/vacuum.c

   1 /*
   2 ** 2003 April 6
   3 **
   4 ** The author disclaims copyright to this source code.  In place of
   5 ** a legal notice, here is a blessing:
   6 **
   7 **    May you do good and not evil.
   8 **    May you find forgiveness for yourself and forgive others.
   9 **    May you share freely, never taking more than you give.
  10 **
  11 *************************************************************************
  12 ** This file contains code used to implement the VACUUM command.
  13 **
  14 ** Most of the code in this file may be omitted by defining the
  15 ** SQLITE_OMIT_VACUUM macro.
  16 */
  17 #include "sqliteInt.h"
  18 #include "vdbeInt.h"
  19
  20 #if !defined(SQLITE_OMIT_VACUUM) && !defined(SQLITE_OMIT_ATTACH)
  21
  22 /*
  23 ** Execute zSql on database db.
  24 **
  25 ** If zSql returns rows, then each row will have exactly one
  26 ** column.  (This will only happen if zSql begins with "SELECT".)
  27 ** Take each row of result and call execSql() again recursively.
  28 **
  29 ** The execSqlF() routine does the same thing, except it accepts
  30 ** a format string as its third argument
  31 */
  32 static int execSql(sqlite3 *db, char **pzErrMsg, const char *zSql){
  33   sqlite3_stmt *pStmt;
  34   int rc;
  35
  36   /* printf("SQL: [%s]\n", zSql); fflush(stdout); */
  37   rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0);
  38   if( rc!=SQLITE_OK ) return rc;
  39   while( SQLITE_ROW==(rc = sqlite3_step(pStmt)) ){
  40     const char *zSubSql = (const char*)sqlite3_column_text(pStmt,0);
  41     assert( sqlite3_strnicmp(zSql,"SELECT",6)==0 );
  42     /* The secondary SQL must be one of CREATE TABLE, CREATE INDEX,
  43     ** or INSERT.  Historically there have been attacks that first
  44     ** corrupt the sqlite_master.sql field with other kinds of statements
  45     ** then run VACUUM to get those statements to execute at inappropriate
  46     ** times. */
  47     if( zSubSql
  48      && (strncmp(zSubSql,"CRE",3)==0 || strncmp(zSubSql,"INS",3)==0)
  49     ){
  50       rc = execSql(db, pzErrMsg, zSubSql);
  51       if( rc!=SQLITE_OK ) break;
  52     }
  53   }
  54   assert( rc!=SQLITE_ROW );
  55   if( rc==SQLITE_DONE ) rc = SQLITE_OK;
  56   if( rc ){
  57     sqlite3SetString(pzErrMsg, db, sqlite3_errmsg(db));
  58   }
  59   (void)sqlite3_finalize(pStmt);
  60   return rc;
  61 }
  62 static int execSqlF(sqlite3 *db, char **pzErrMsg, const char *zSql, ...){
  63   char *z;
  64   va_list ap;
  65   int rc;
  66   va_start(ap, zSql);
  67   z = sqlite3VMPrintf(db, zSql, ap);
  68   va_end(ap);
  69   if( z==0 ) return SQLITE_NOMEM;
  70   rc = execSql(db, pzErrMsg, z);
  71   sqlite3DbFree(db, z);
  72   return rc;
  73 }
  74
  75 /*
  76 ** The VACUUM command is used to clean up the database,
  77 ** collapse free space, etc.  It is modelled after the VACUUM command
  78 ** in PostgreSQL.  The VACUUM command works as follows:
  79 **
  80 **   (1)  Create a new transient database file
  81 **   (2)  Copy all content from the database being vacuumed into
  82 **        the new transient database file
  83 **   (3)  Copy content from the transient database back into the
  84 **        original database.
  85 **
  86 ** The transient database requires temporary disk space approximately
  87 ** equal to the size of the original database.  The copy operation of
  88 ** step (3) requires additional temporary disk space approximately equal
  89 ** to the size of the original database for the rollback journal.
  90 ** Hence, temporary disk space that is approximately 2x the size of the
  91 ** original database is required.  Every page of the database is written
  92 ** approximately 3 times:  Once for step (2) and twice for step (3).
  93 ** Two writes per page are required in step (3) because the original
  94 ** database content must be written into the rollback journal prior to
  95 ** overwriting the database with the vacuumed content.
  96 **
  97 ** Only 1x temporary space and only 1x writes would be required if
  98 ** the copy of step (3) were replaced by deleting the original database
  99 ** and renaming the transient database as the original.  But that will
 100 ** not work if other processes are attached to the original database.
 101 ** And a power loss in between deleting the original and renaming the
 102 ** transient would cause the database file to appear to be deleted
 103 ** following reboot.
 104 */
 105 void sqlite3Vacuum(Parse *pParse, Token *pNm){
 106   Vdbe *v = sqlite3GetVdbe(pParse);
 107   int iDb = 0;
 108   if( v==0 ) return;
 109   if( pNm ){
 110 #ifndef SQLITE_BUG_COMPATIBLE_20160819
 111     /* Default behavior:  Report an error if the argument to VACUUM is
 112     ** not recognized */
 113     iDb = sqlite3TwoPartName(pParse, pNm, pNm, &pNm);
 114     if( iDb<0 ) return;
 115 #else
 116     /* When SQLITE_BUG_COMPATIBLE_20160819 is defined, unrecognized arguments
 117     ** to VACUUM are silently ignored.  This is a back-out of a bug fix that
 118     ** occurred on 2016-08-19 (https://www.sqlite.org/src/info/083f9e6270).
 119     ** The buggy behavior is required for binary compatibility with some
 120     ** legacy applications. */
 121     iDb = sqlite3FindDb(pParse->db, pNm);
 122     if( iDb<0 ) iDb = 0;
 123 #endif
 124   }
 125   if( iDb!=1 ){
 126     sqlite3VdbeAddOp1(v, OP_Vacuum, iDb);
 127     sqlite3VdbeUsesBtree(v, iDb);
 128   }
 129   return;
 130 }
 131
 132 /*
 133 ** This routine implements the OP_Vacuum opcode of the VDBE.
 134 */
 135 int sqlite3RunVacuum(char **pzErrMsg, sqlite3 *db, int iDb){
 136   int rc = SQLITE_OK;     /* Return code from service routines */
 137   Btree *pMain;           /* The database being vacuumed */
 138   Btree *pTemp;           /* The temporary database we vacuum into */
 139   u16 saved_mDbFlags;     /* Saved value of db->mDbFlags */
 140   u32 saved_flags;        /* Saved value of db->flags */
 141   int saved_nChange;      /* Saved value of db->nChange */
 142   int saved_nTotalChange; /* Saved value of db->nTotalChange */
 143   u8 saved_mTrace;        /* Saved trace settings */
 144   Db *pDb = 0;            /* Database to detach at end of vacuum */
 145   int isMemDb;            /* True if vacuuming a :memory: database */
 146   int nRes;               /* Bytes of reserved space at the end of each page */
 147   int nDb;                /* Number of attached databases */
 148   const char *zDbMain;    /* Schema name of database to vacuum */
 149
 150   if( !db->autoCommit ){
 151     sqlite3SetString(pzErrMsg, db, "cannot VACUUM from within a transaction");
 152     return SQLITE_ERROR;
 153   }
 154   if( db->nVdbeActive>1 ){
 155     sqlite3SetString(pzErrMsg, db,"cannot VACUUM - SQL statements in progress");
 156     return SQLITE_ERROR;
 157   }
 158
 159   /* Save the current value of the database flags so that it can be
 160   ** restored before returning. Then set the writable-schema flag, and
 161   ** disable CHECK and foreign key constraints.  */
 162   saved_flags = db->flags;
 163   saved_mDbFlags = db->mDbFlags;
 164   saved_nChange = db->nChange;
 165   saved_nTotalChange = db->nTotalChange;
 166   saved_mTrace = db->mTrace;
 167   db->flags |= SQLITE_WriteSchema | SQLITE_IgnoreChecks;
 168   db->mDbFlags |= DBFLAG_PreferBuiltin | DBFLAG_Vacuum;
 169   db->flags &= ~(SQLITE_ForeignKeys | SQLITE_ReverseOrder | SQLITE_CountRows);
 170   db->mTrace = 0;
 171
 172   zDbMain = db->aDb[iDb].zDbSName;
 173   pMain = db->aDb[iDb].pBt;
 174   isMemDb = sqlite3PagerIsMemdb(sqlite3BtreePager(pMain));
 175
 176   /* Attach the temporary database as 'vacuum_db'. The synchronous pragma
 177   ** can be set to 'off' for this file, as it is not recovered if a crash
 178   ** occurs anyway. The integrity of the database is maintained by a
 179   ** (possibly synchronous) transaction opened on the main database before
 180   ** sqlite3BtreeCopyFile() is called.
 181   **
 182   ** An optimisation would be to use a non-journaled pager.
 183   ** (Later:) I tried setting "PRAGMA vacuum_db.journal_mode=OFF" but
 184   ** that actually made the VACUUM run slower.  Very little journalling
 185   ** actually occurs when doing a vacuum since the vacuum_db is initially
 186   ** empty.  Only the journal header is written.  Apparently it takes more
 187   ** time to parse and run the PRAGMA to turn journalling off than it does
 188   ** to write the journal header file.
 189   */
 190   nDb = db->nDb;
 191   rc = execSql(db, pzErrMsg, "ATTACH''AS vacuum_db");
 192   if( rc!=SQLITE_OK ) goto end_of_vacuum;
 193   assert( (db->nDb-1)==nDb );
 194   pDb = &db->aDb[nDb];
 195   assert( strcmp(pDb->zDbSName,"vacuum_db")==0 );
 196   pTemp = pDb->pBt;
 197
 198   /* The call to execSql() to attach the temp database has left the file
 199   ** locked (as there was more than one active statement when the transaction
 200   ** to read the schema was concluded. Unlock it here so that this doesn't
 201   ** cause problems for the call to BtreeSetPageSize() below.  */
 202   sqlite3BtreeCommit(pTemp);
 203
 204   nRes = sqlite3BtreeGetOptimalReserve(pMain);
 205
 206   /* A VACUUM cannot change the pagesize of an encrypted database. */
 207 #ifdef SQLITE_HAS_CODEC
 208   if( db->nextPagesize ){
 209     extern void sqlite3CodecGetKey(sqlite3*, int, void**, int*);
 210     int nKey;
 211     char *zKey;
 212     sqlite3CodecGetKey(db, iDb, (void**)&zKey, &nKey);
 213     if( nKey ) db->nextPagesize = 0;
 214   }
 215 #endif
 216
 217   sqlite3BtreeSetCacheSize(pTemp, db->aDb[iDb].pSchema->cache_size);
 218   sqlite3BtreeSetSpillSize(pTemp, sqlite3BtreeSetSpillSize(pMain,0));
 219   sqlite3BtreeSetPagerFlags(pTemp, PAGER_SYNCHRONOUS_OFF|PAGER_CACHESPILL);
 220
 221   /* Begin a transaction and take an exclusive lock on the main database
 222   ** file. This is done before the sqlite3BtreeGetPageSize(pMain) call below,
 223   ** to ensure that we do not try to change the page-size on a WAL database.
 224   */
 225   rc = execSql(db, pzErrMsg, "BEGIN");
 226   if( rc!=SQLITE_OK ) goto end_of_vacuum;
 227   rc = sqlite3BtreeBeginTrans(pMain, 2, 0);
 228   if( rc!=SQLITE_OK ) goto end_of_vacuum;
 229
 230   /* Do not attempt to change the page size for a WAL database */
 231   if( sqlite3PagerGetJournalMode(sqlite3BtreePager(pMain))
 232                                                ==PAGER_JOURNALMODE_WAL ){
 233     db->nextPagesize = 0;
 234   }
 235
 236   if( sqlite3BtreeSetPageSize(pTemp, sqlite3BtreeGetPageSize(pMain), nRes, 0)
 237    || (!isMemDb && sqlite3BtreeSetPageSize(pTemp, db->nextPagesize, nRes, 0))
 238    || NEVER(db->mallocFailed)
 239   ){
 240     rc = SQLITE_NOMEM_BKPT;
 241     goto end_of_vacuum;
 242   }
 243
 244 #ifndef SQLITE_OMIT_AUTOVACUUM
 245   sqlite3BtreeSetAutoVacuum(pTemp, db->nextAutovac>=0 ? db->nextAutovac :
 246                                            sqlite3BtreeGetAutoVacuum(pMain));
 247 #endif
 248
 249   /* Query the schema of the main database. Create a mirror schema
 250   ** in the temporary database.
 251   */
 252   db->init.iDb = nDb; /* force new CREATE statements into vacuum_db */
 253   rc = execSqlF(db, pzErrMsg,
 254       "SELECT sql FROM \"%w\".sqlite_master"
 255       " WHERE type='table'AND name<>'sqlite_sequence'"
 256       " AND coalesce(rootpage,1)>0",
 257       zDbMain
 258   );
 259   if( rc!=SQLITE_OK ) goto end_of_vacuum;
 260   rc = execSqlF(db, pzErrMsg,
 261       "SELECT sql FROM \"%w\".sqlite_master"
 262       " WHERE type='index'",
 263       zDbMain
 264   );
 265   if( rc!=SQLITE_OK ) goto end_of_vacuum;
 266   db->init.iDb = 0;
 267
 268   /* Loop through the tables in the main database. For each, do
 269   ** an "INSERT INTO vacuum_db.xxx SELECT * FROM main.xxx;" to copy
 270   ** the contents to the temporary database.
 271   */
 272   rc = execSqlF(db, pzErrMsg,
 273       "SELECT'INSERT INTO vacuum_db.'||quote(name)"
 274       "||' SELECT*FROM\"%w\".'||quote(name)"
 275       "FROM vacuum_db.sqlite_master "
 276       "WHERE type='table'AND coalesce(rootpage,1)>0",
 277       zDbMain
 278   );
 279   assert( (db->mDbFlags & DBFLAG_Vacuum)!=0 );
 280   db->mDbFlags &= ~DBFLAG_Vacuum;
 281   if( rc!=SQLITE_OK ) goto end_of_vacuum;
 282
 283   /* Copy the triggers, views, and virtual tables from the main database
 284   ** over to the temporary database.  None of these objects has any
 285   ** associated storage, so all we have to do is copy their entries
 286   ** from the SQLITE_MASTER table.
 287   */
 288   rc = execSqlF(db, pzErrMsg,
 289       "INSERT INTO vacuum_db.sqlite_master"
 290       " SELECT*FROM \"%w\".sqlite_master"
 291       " WHERE type IN('view','trigger')"
 292       " OR(type='table'AND rootpage=0)",
 293       zDbMain
 294   );
 295   if( rc ) goto end_of_vacuum;
 296
 297   /* At this point, there is a write transaction open on both the
 298   ** vacuum database and the main database. Assuming no error occurs,
 299   ** both transactions are closed by this block - the main database
 300   ** transaction by sqlite3BtreeCopyFile() and the other by an explicit
 301   ** call to sqlite3BtreeCommit().
 302   */
 303   {
 304     u32 meta;
 305     int i;
 306
 307     /* This array determines which meta meta values are preserved in the
 308     ** vacuum.  Even entries are the meta value number and odd entries
 309     ** are an increment to apply to the meta value after the vacuum.
 310     ** The increment is used to increase the schema cookie so that other
 311     ** connections to the same database will know to reread the schema.
 312     */
 313     static const unsigned char aCopy[] = {
 314        BTREE_SCHEMA_VERSION,     1,  /* Add one to the old schema cookie */
 315        BTREE_DEFAULT_CACHE_SIZE, 0,  /* Preserve the default page cache size */
 316        BTREE_TEXT_ENCODING,      0,  /* Preserve the text encoding */
 317        BTREE_USER_VERSION,       0,  /* Preserve the user version */
 318        BTREE_APPLICATION_ID,     0,  /* Preserve the application id */
 319     };
 320
 321     assert( 1==sqlite3BtreeIsInTrans(pTemp) );
 322     assert( 1==sqlite3BtreeIsInTrans(pMain) );
 323
 324     /* Copy Btree meta values */
 325     for(i=0; i<ArraySize(aCopy); i+=2){
 326       /* GetMeta() and UpdateMeta() cannot fail in this context because
 327       ** we already have page 1 loaded into cache and marked dirty. */
 328       sqlite3BtreeGetMeta(pMain, aCopy[i], &meta);
 329       rc = sqlite3BtreeUpdateMeta(pTemp, aCopy[i], meta+aCopy[i+1]);
 330       if( NEVER(rc!=SQLITE_OK) ) goto end_of_vacuum;
 331     }
 332
 333     rc = sqlite3BtreeCopyFile(pMain, pTemp);
 334     if( rc!=SQLITE_OK ) goto end_of_vacuum;
 335     rc = sqlite3BtreeCommit(pTemp);
 336     if( rc!=SQLITE_OK ) goto end_of_vacuum;
 337 #ifndef SQLITE_OMIT_AUTOVACUUM
 338     sqlite3BtreeSetAutoVacuum(pMain, sqlite3BtreeGetAutoVacuum(pTemp));
 339 #endif
 340   }
 341
 342   assert( rc==SQLITE_OK );
 343   rc = sqlite3BtreeSetPageSize(pMain, sqlite3BtreeGetPageSize(pTemp), nRes,1);
 344
 345 end_of_vacuum:
 346   /* Restore the original value of db->flags */
 347   db->init.iDb = 0;
 348   db->mDbFlags = saved_mDbFlags;
 349   db->flags = saved_flags;
 350   db->nChange = saved_nChange;
 351   db->nTotalChange = saved_nTotalChange;
 352   db->mTrace = saved_mTrace;
 353   sqlite3BtreeSetPageSize(pMain, -1, -1, 1);
 354
 355   /* Currently there is an SQL level transaction open on the vacuum
 356   ** database. No locks are held on any other files (since the main file
 357   ** was committed at the btree level). So it safe to end the transaction
 358   ** by manually setting the autoCommit flag to true and detaching the
 359   ** vacuum database. The vacuum_db journal file is deleted when the pager
 360   ** is closed by the DETACH.
 361   */
 362   db->autoCommit = 1;
 363
 364   if( pDb ){
 365     sqlite3BtreeClose(pDb->pBt);
 366     pDb->pBt = 0;
 367     pDb->pSchema = 0;
 368   }
 369
 370   /* This both clears the schemas and reduces the size of the db->aDb[]
 371   ** array. */
 372   sqlite3ResetAllSchemasOfConnection(db);
 373
 374   return rc;
 375 }
 376
 377 #endif  /* SQLITE_OMIT_VACUUM && SQLITE_OMIT_ATTACH */