sqlite/ext/misc/scrub.c

611 lines
18 KiB
C
Raw Normal View History

/*
** 2016-05-05
**
** The author disclaims copyright to this source code. In place of
** a legal notice, here is a blessing:
**
** May you do good and not evil.
** May you find forgiveness for yourself and forgive others.
** May you share freely, never taking more than you give.
**
******************************************************************************
**
** This file implements a utility function (and a utility program) that
** makes a copy of an SQLite database while simultaneously zeroing out all
** deleted content.
**
** Normally (when PRAGMA secure_delete=OFF, which is the default) when SQLite
** deletes content, it does not overwrite the deleted content but rather marks
** the region of the file that held that content as being reusable. This can
** cause deleted content to recoverable from the database file. This stale
** content is removed by the VACUUM command, but VACUUM can be expensive for
** large databases. When in PRAGMA secure_delete=ON mode, the deleted content
** is zeroed, but secure_delete=ON has overhead as well.
**
** This utility attempts to make a copy of a complete SQLite database where
** all of the deleted content is zeroed out in the copy, and it attempts to
** do so while being faster than running VACUUM.
**
** Usage:
**
** int sqlite3_scrub_backup(
** const char *zSourceFile, // Source database filename
** const char *zDestFile, // Destination database filename
** char **pzErrMsg // Write error message here
** );
**
** Simply call the API above specifying the filename of the source database
** and the name of the backup copy. The source database must already exist
** and can be in active use. (A read lock is held during the backup.) The
** destination file should not previously exist. If the pzErrMsg parameter
** is non-NULL and if an error occurs, then an error message might be written
** into memory obtained from sqlite3_malloc() and *pzErrMsg made to point to
** that error message. But if the error is an OOM, the error might not be
** reported. The routine always returns non-zero if there is an error.
**
** If compiled with -DSCRUB_STANDALONE then a main() procedure is added and
** this file becomes a standalone program that can be run as follows:
**
** ./sqlite3scrub SOURCE DEST
*/
#include "sqlite3.h"
#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <stdarg.h>
#include <string.h>
typedef struct ScrubState ScrubState;
typedef unsigned char u8;
typedef unsigned short u16;
typedef unsigned int u32;
/* State information for a scrub-and-backup operation */
struct ScrubState {
const char *zSrcFile; /* Name of the source file */
const char *zDestFile; /* Name of the destination file */
int rcErr; /* Error code */
char *zErr; /* Error message text */
sqlite3 *dbSrc; /* Source database connection */
sqlite3_file *pSrc; /* Source file handle */
sqlite3 *dbDest; /* Destination database connection */
sqlite3_file *pDest; /* Destination file handle */
u32 szPage; /* Page size */
u32 szUsable; /* Usable bytes on each page */
u32 nPage; /* Number of pages */
u32 iLastPage; /* Page number of last page written so far*/
u8 *page1; /* Content of page 1 */
};
/* Store an error message */
static void scrubBackupErr(ScrubState *p, const char *zFormat, ...){
va_list ap;
sqlite3_free(p->zErr);
va_start(ap, zFormat);
p->zErr = sqlite3_vmprintf(zFormat, ap);
va_end(ap);
if( p->rcErr==0 ) p->rcErr = SQLITE_ERROR;
}
/* Allocate memory to hold a single page of content */
static u8 *scrubBackupAllocPage(ScrubState *p){
u8 *pPage;
if( p->rcErr ) return 0;
pPage = sqlite3_malloc( p->szPage );
if( pPage==0 ) p->rcErr = SQLITE_NOMEM;
return pPage;
}
/* Read a page from the source database into memory. Use the memory
** provided by pBuf if not NULL or allocate a new page if pBuf==NULL.
*/
static u8 *scrubBackupRead(ScrubState *p, int pgno, u8 *pBuf){
int rc;
sqlite3_int64 iOff;
u8 *pOut = pBuf;
if( p->rcErr ) return 0;
if( pOut==0 ){
pOut = scrubBackupAllocPage(p);
if( pOut==0 ) return 0;
}
iOff = (pgno-1)*(sqlite3_int64)p->szPage;
rc = p->pSrc->pMethods->xRead(p->pSrc, pOut, p->szPage, iOff);
if( rc!=SQLITE_OK ){
if( pBuf==0 ) sqlite3_free(pOut);
pOut = 0;
scrubBackupErr(p, "read failed for page %d", pgno);
p->rcErr = SQLITE_IOERR;
}
return pOut;
}
/* Write a page to the destination database */
static void scrubBackupWrite(ScrubState *p, int pgno, const u8 *pData){
int rc;
sqlite3_int64 iOff;
if( p->rcErr ) return;
iOff = (pgno-1)*(sqlite3_int64)p->szPage;
rc = p->pDest->pMethods->xWrite(p->pDest, pData, p->szPage, iOff);
if( rc!=SQLITE_OK ){
scrubBackupErr(p, "write failed for page %d", pgno);
p->rcErr = SQLITE_IOERR;
}
if( (u32)pgno>p->iLastPage ) p->iLastPage = pgno;
}
/* Prepare a statement against the "db" database. */
static sqlite3_stmt *scrubBackupPrepare(
ScrubState *p, /* Backup context */
sqlite3 *db, /* Database to prepare against */
const char *zSql /* SQL statement */
){
sqlite3_stmt *pStmt;
if( p->rcErr ) return 0;
p->rcErr = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0);
if( p->rcErr ){
scrubBackupErr(p, "SQL error \"%s\" on \"%s\"",
sqlite3_errmsg(db), zSql);
sqlite3_finalize(pStmt);
return 0;
}
return pStmt;
}
/* Open the source database file */
static void scrubBackupOpenSrc(ScrubState *p){
sqlite3_stmt *pStmt;
int rc;
/* Open the source database file */
p->rcErr = sqlite3_open_v2(p->zSrcFile, &p->dbSrc,
SQLITE_OPEN_READWRITE |
SQLITE_OPEN_URI | SQLITE_OPEN_PRIVATECACHE, 0);
if( p->rcErr ){
scrubBackupErr(p, "cannot open source database: %s",
sqlite3_errmsg(p->dbSrc));
return;
}
p->rcErr = sqlite3_exec(p->dbSrc, "SELECT 1 FROM sqlite_schema; BEGIN;",
0, 0, 0);
if( p->rcErr ){
scrubBackupErr(p,
"cannot start a read transaction on the source database: %s",
sqlite3_errmsg(p->dbSrc));
return;
}
rc = sqlite3_wal_checkpoint_v2(p->dbSrc, "main", SQLITE_CHECKPOINT_FULL,
0, 0);
if( rc ){
scrubBackupErr(p, "cannot checkpoint the source database");
return;
}
pStmt = scrubBackupPrepare(p, p->dbSrc, "PRAGMA page_size");
if( pStmt==0 ) return;
rc = sqlite3_step(pStmt);
if( rc==SQLITE_ROW ){
p->szPage = sqlite3_column_int(pStmt, 0);
}else{
scrubBackupErr(p, "unable to determine the page size");
}
sqlite3_finalize(pStmt);
if( p->rcErr ) return;
pStmt = scrubBackupPrepare(p, p->dbSrc, "PRAGMA page_count");
if( pStmt==0 ) return;
rc = sqlite3_step(pStmt);
if( rc==SQLITE_ROW ){
p->nPage = sqlite3_column_int(pStmt, 0);
}else{
scrubBackupErr(p, "unable to determine the size of the source database");
}
sqlite3_finalize(pStmt);
sqlite3_file_control(p->dbSrc, "main", SQLITE_FCNTL_FILE_POINTER, &p->pSrc);
if( p->pSrc==0 || p->pSrc->pMethods==0 ){
scrubBackupErr(p, "cannot get the source file handle");
p->rcErr = SQLITE_ERROR;
}
}
/* Create and open the destination file */
static void scrubBackupOpenDest(ScrubState *p){
sqlite3_stmt *pStmt;
int rc;
char *zSql;
if( p->rcErr ) return;
p->rcErr = sqlite3_open_v2(p->zDestFile, &p->dbDest,
SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE |
SQLITE_OPEN_URI | SQLITE_OPEN_PRIVATECACHE, 0);
if( p->rcErr ){
scrubBackupErr(p, "cannot open destination database: %s",
sqlite3_errmsg(p->dbDest));
return;
}
zSql = sqlite3_mprintf("PRAGMA page_size(%u);", p->szPage);
if( zSql==0 ){
p->rcErr = SQLITE_NOMEM;
return;
}
p->rcErr = sqlite3_exec(p->dbDest, zSql, 0, 0, 0);
sqlite3_free(zSql);
if( p->rcErr ){
scrubBackupErr(p,
"cannot set the page size on the destination database: %s",
sqlite3_errmsg(p->dbDest));
return;
}
sqlite3_exec(p->dbDest, "PRAGMA journal_mode=OFF;", 0, 0, 0);
p->rcErr = sqlite3_exec(p->dbDest, "BEGIN EXCLUSIVE;", 0, 0, 0);
if( p->rcErr ){
scrubBackupErr(p,
"cannot start a write transaction on the destination database: %s",
sqlite3_errmsg(p->dbDest));
return;
}
pStmt = scrubBackupPrepare(p, p->dbDest, "PRAGMA page_count;");
if( pStmt==0 ) return;
rc = sqlite3_step(pStmt);
if( rc!=SQLITE_ROW ){
scrubBackupErr(p, "cannot measure the size of the destination");
}else if( sqlite3_column_int(pStmt, 0)>1 ){
scrubBackupErr(p, "destination database is not empty - holds %d pages",
sqlite3_column_int(pStmt, 0));
}
sqlite3_finalize(pStmt);
sqlite3_file_control(p->dbDest, "main", SQLITE_FCNTL_FILE_POINTER, &p->pDest);
if( p->pDest==0 || p->pDest->pMethods==0 ){
scrubBackupErr(p, "cannot get the destination file handle");
p->rcErr = SQLITE_ERROR;
}
}
/* Read a 32-bit big-endian integer */
static u32 scrubBackupInt32(const u8 *a){
u32 v = a[3];
v += ((u32)a[2])<<8;
v += ((u32)a[1])<<16;
v += ((u32)a[0])<<24;
return v;
}
/* Read a 16-bit big-endian integer */
static u32 scrubBackupInt16(const u8 *a){
return (a[0]<<8) + a[1];
}
/*
** Read a varint. Put the value in *pVal and return the number of bytes.
*/
static int scrubBackupVarint(const u8 *z, sqlite3_int64 *pVal){
sqlite3_int64 v = 0;
int i;
for(i=0; i<8; i++){
v = (v<<7) + (z[i]&0x7f);
if( (z[i]&0x80)==0 ){ *pVal = v; return i+1; }
}
v = (v<<8) + (z[i]&0xff);
*pVal = v;
return 9;
}
/*
** Return the number of bytes in a varint.
*/
static int scrubBackupVarintSize(const u8 *z){
int i;
for(i=0; i<8; i++){
if( (z[i]&0x80)==0 ){ return i+1; }
}
return 9;
}
/*
** Copy the freelist trunk page given, and all its descendents,
** zeroing out as much as possible in the process.
*/
static void scrubBackupFreelist(ScrubState *p, int pgno, u32 nFree){
u8 *a, *aBuf;
u32 n, mx;
if( p->rcErr ) return;
aBuf = scrubBackupAllocPage(p);
if( aBuf==0 ) return;
while( pgno && nFree){
a = scrubBackupRead(p, pgno, aBuf);
if( a==0 ) break;
n = scrubBackupInt32(&a[4]);
mx = p->szUsable/4 - 2;
if( n<mx ){
memset(&a[n*4+8], 0, 4*(mx-n));
}
scrubBackupWrite(p, pgno, a);
pgno = scrubBackupInt32(a);
#if 0
/* There is really no point in copying the freelist leaf pages.
** Simply leave them uninitialized in the destination database. The
** OS filesystem should zero those pages for us automatically.
*/
for(i=0; i<n && nFree; i++){
u32 iLeaf = scrubBackupInt32(&a[i*4+8]);
if( aZero==0 ){
aZero = scrubBackupAllocPage(p);
if( aZero==0 ){ pgno = 0; break; }
memset(aZero, 0, p->szPage);
}
scrubBackupWrite(p, iLeaf, aZero);
nFree--;
}
#endif
}
sqlite3_free(aBuf);
}
/*
** Copy an overflow chain from source to destination. Zero out any
** unused tail at the end of the overflow chain.
*/
static void scrubBackupOverflow(ScrubState *p, int pgno, u32 nByte){
u8 *a, *aBuf;
aBuf = scrubBackupAllocPage(p);
if( aBuf==0 ) return;
while( nByte>0 && pgno!=0 ){
a = scrubBackupRead(p, pgno, aBuf);
if( a==0 ) break;
if( nByte >= (p->szUsable)-4 ){
nByte -= (p->szUsable) - 4;
}else{
u32 x = (p->szUsable - 4) - nByte;
u32 i = p->szUsable - x;
memset(&a[i], 0, x);
nByte = 0;
}
scrubBackupWrite(p, pgno, a);
pgno = scrubBackupInt32(a);
}
sqlite3_free(aBuf);
}
/*
** Copy B-Tree page pgno, and all of its children, from source to destination.
** Zero out deleted content during the copy.
*/
static void scrubBackupBtree(ScrubState *p, int pgno, int iDepth){
u8 *a;
u32 i, n, pc;
u32 nCell;
u32 nPrefix;
u32 szHdr;
u32 iChild;
u8 *aTop;
u8 *aCell;
u32 x, y;
int ln = 0;
if( p->rcErr ) return;
if( iDepth>50 ){
scrubBackupErr(p, "corrupt: b-tree too deep at page %d", pgno);
return;
}
if( pgno==1 ){
a = p->page1;
}else{
a = scrubBackupRead(p, pgno, 0);
if( a==0 ) return;
}
nPrefix = pgno==1 ? 100 : 0;
aTop = &a[nPrefix];
szHdr = 8 + 4*(aTop[0]==0x02 || aTop[0]==0x05);
aCell = aTop + szHdr;
nCell = scrubBackupInt16(&aTop[3]);
/* Zero out the gap between the cell index and the start of the
** cell content area */
x = scrubBackupInt16(&aTop[5]); /* First byte of cell content area */
if( x>p->szUsable ){ ln=__LINE__; goto btree_corrupt; }
y = szHdr + nPrefix + nCell*2;
if( y>x ){ ln=__LINE__; goto btree_corrupt; }
if( y<x ) memset(a+y, 0, x-y); /* Zero the gap */
/* Zero out all the free blocks */
pc = scrubBackupInt16(&aTop[1]);
if( pc>0 && pc<x ){ ln=__LINE__; goto btree_corrupt; }
while( pc ){
if( pc>(p->szUsable)-4 ){ ln=__LINE__; goto btree_corrupt; }
n = scrubBackupInt16(&a[pc+2]);
if( pc+n>(p->szUsable) ){ ln=__LINE__; goto btree_corrupt; }
if( n>4 ) memset(&a[pc+4], 0, n-4);
x = scrubBackupInt16(&a[pc]);
if( x<pc+4 && x>0 ){ ln=__LINE__; goto btree_corrupt; }
pc = x;
}
/* Write this one page */
scrubBackupWrite(p, pgno, a);
/* Walk the tree and process child pages */
for(i=0; i<nCell; i++){
u32 X, M, K, nLocal;
sqlite3_int64 P;
pc = scrubBackupInt16(&aCell[i*2]);
if( pc <= szHdr ){ ln=__LINE__; goto btree_corrupt; }
if( pc > p->szUsable-3 ){ ln=__LINE__; goto btree_corrupt; }
if( aTop[0]==0x05 || aTop[0]==0x02 ){
if( pc+4 > p->szUsable ){ ln=__LINE__; goto btree_corrupt; }
iChild = scrubBackupInt32(&a[pc]);
pc += 4;
scrubBackupBtree(p, iChild, iDepth+1);
if( aTop[0]==0x05 ) continue;
}
pc += scrubBackupVarint(&a[pc], &P);
if( pc >= p->szUsable ){ ln=__LINE__; goto btree_corrupt; }
if( aTop[0]==0x0d ){
X = p->szUsable - 35;
}else{
X = ((p->szUsable - 12)*64/255) - 23;
}
if( P<=X ){
/* All content is local. No overflow */
continue;
}
M = ((p->szUsable - 12)*32/255)-23;
K = M + ((P-M)%(p->szUsable-4));
if( aTop[0]==0x0d ){
pc += scrubBackupVarintSize(&a[pc]);
if( pc > (p->szUsable-4) ){ ln=__LINE__; goto btree_corrupt; }
}
nLocal = K<=X ? K : M;
if( pc+nLocal > p->szUsable-4 ){ ln=__LINE__; goto btree_corrupt; }
iChild = scrubBackupInt32(&a[pc+nLocal]);
scrubBackupOverflow(p, iChild, (u32)(P-nLocal));
}
/* Walk the right-most tree */
if( aTop[0]==0x05 || aTop[0]==0x02 ){
iChild = scrubBackupInt32(&aTop[8]);
scrubBackupBtree(p, iChild, iDepth+1);
}
/* All done */
if( pgno>1 ) sqlite3_free(a);
return;
btree_corrupt:
scrubBackupErr(p, "corruption on page %d of source database (errid=%d)",
pgno, ln);
if( pgno>1 ) sqlite3_free(a);
}
/*
** Copy all ptrmap pages from source to destination.
** This routine is only called if the source database is in autovacuum
** or incremental vacuum mode.
*/
static void scrubBackupPtrmap(ScrubState *p){
u32 pgno = 2;
u32 J = p->szUsable/5;
u32 iLock = (1073742335/p->szPage)+1;
u8 *a, *pBuf;
if( p->rcErr ) return;
pBuf = scrubBackupAllocPage(p);
if( pBuf==0 ) return;
while( pgno<=p->nPage ){
a = scrubBackupRead(p, pgno, pBuf);
if( a==0 ) break;
scrubBackupWrite(p, pgno, a);
pgno += J+1;
if( pgno==iLock ) pgno++;
}
sqlite3_free(pBuf);
}
int sqlite3_scrub_backup(
const char *zSrcFile, /* Source file */
const char *zDestFile, /* Destination file */
char **pzErr /* Write error here if non-NULL */
){
ScrubState s;
u32 n, i;
sqlite3_stmt *pStmt;
memset(&s, 0, sizeof(s));
s.zSrcFile = zSrcFile;
s.zDestFile = zDestFile;
/* Open both source and destination databases */
scrubBackupOpenSrc(&s);
scrubBackupOpenDest(&s);
/* Read in page 1 */
s.page1 = scrubBackupRead(&s, 1, 0);
if( s.page1==0 ) goto scrub_abort;
s.szUsable = s.szPage - s.page1[20];
/* Copy the freelist */
n = scrubBackupInt32(&s.page1[36]);
i = scrubBackupInt32(&s.page1[32]);
if( n ) scrubBackupFreelist(&s, i, n);
/* Copy ptrmap pages */
n = scrubBackupInt32(&s.page1[52]);
if( n ) scrubBackupPtrmap(&s);
/* Copy all of the btrees */
scrubBackupBtree(&s, 1, 0);
pStmt = scrubBackupPrepare(&s, s.dbSrc,
"SELECT rootpage FROM sqlite_schema WHERE coalesce(rootpage,0)>0");
if( pStmt==0 ) goto scrub_abort;
while( sqlite3_step(pStmt)==SQLITE_ROW ){
i = (u32)sqlite3_column_int(pStmt, 0);
scrubBackupBtree(&s, i, 0);
}
sqlite3_finalize(pStmt);
/* If the last page of the input db file is a free-list leaf, then the
** backup file on disk is still smaller than the size indicated within
** the database header. In this case, write a page of zeroes to the
** last page of the backup database so that SQLite does not mistakenly
** think the db is corrupt. */
if( s.iLastPage<s.nPage ){
u8 *aZero = scrubBackupAllocPage(&s);
if( aZero ){
memset(aZero, 0, s.szPage);
scrubBackupWrite(&s, s.nPage, aZero);
sqlite3_free(aZero);
}
}
scrub_abort:
/* Close the destination database without closing the transaction. If we
** commit, page zero will be overwritten. */
sqlite3_close(s.dbDest);
/* But do close out the read-transaction on the source database */
sqlite3_exec(s.dbSrc, "COMMIT;", 0, 0, 0);
sqlite3_close(s.dbSrc);
sqlite3_free(s.page1);
if( pzErr ){
*pzErr = s.zErr;
}else{
sqlite3_free(s.zErr);
}
return s.rcErr;
}
#ifdef SCRUB_STANDALONE
/* Error and warning log */
static void errorLogCallback(void *pNotUsed, int iErr, const char *zMsg){
const char *zType;
switch( iErr&0xff ){
case SQLITE_WARNING: zType = "WARNING"; break;
case SQLITE_NOTICE: zType = "NOTICE"; break;
default: zType = "ERROR"; break;
}
fprintf(stderr, "%s: %s\n", zType, zMsg);
}
/* The main() routine when this utility is run as a stand-alone program */
int main(int argc, char **argv){
char *zErr = 0;
int rc;
if( argc!=3 ){
fprintf(stderr,"Usage: %s SOURCE DESTINATION\n", argv[0]);
exit(1);
}
sqlite3_config(SQLITE_CONFIG_LOG, errorLogCallback, 0);
rc = sqlite3_scrub_backup(argv[1], argv[2], &zErr);
if( rc==SQLITE_NOMEM ){
fprintf(stderr, "%s: out of memory\n", argv[0]);
exit(1);
}
if( zErr ){
fprintf(stderr, "%s: %s\n", argv[0], zErr);
sqlite3_free(zErr);
exit(1);
}
return 0;
}
#endif