xref: /sqlite-3.40.0/test/fuzzinvariants.c (revision 053bb22f)
1 /*
2 ** 2022-06-14
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 **
13 ** This library is used by fuzzcheck to test query invariants.
14 **
15 ** An sqlite3_stmt is passed in that has just returned SQLITE_ROW.  This
16 ** routine does:
17 **
18 **     *   Record the output of the current row
19 **     *   Construct an alternative query that should return the same row
20 **     *   Run the alternative query and verify that it does in fact return
21 **         the same row
22 **
23 */
24 #include "sqlite3.h"
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <ctype.h>
29 
30 /* Forward references */
31 static char *fuzz_invariant_sql(sqlite3_stmt*, int);
32 static int sameValue(sqlite3_stmt*,int,sqlite3_stmt*,int,sqlite3_stmt*);
33 static void reportInvariantFailed(sqlite3_stmt*,sqlite3_stmt*,int);
34 
35 /*
36 ** Do an invariant check on pStmt.  iCnt determines which invariant check to
37 ** perform.  The first check is iCnt==0.
38 **
39 ** *pbCorrupt is a flag that, if true, indicates that the database file
40 ** is known to be corrupt.  A value of non-zero means "yes, the database
41 ** is corrupt".  A zero value means "we do not know whether or not the
42 ** database is corrupt".  The value might be set prior to entry, or this
43 ** routine might set the value.
44 **
45 ** Return values:
46 **
47 **     SQLITE_OK          This check was successful.
48 **
49 **     SQLITE_DONE        iCnt is out of range.  The caller typically sets
50 **                        up a loop on iCnt starting with zero, and increments
51 **                        iCnt until this code is returned.
52 **
53 **     SQLITE_CORRUPT     The invariant failed, but the underlying database
54 **                        file is indicating that it is corrupt, which might
55 **                        be the cause of the malfunction.  The *pCorrupt
56 **                        value will also be set.
57 **
58 **     SQLITE_INTERNAL    The invariant failed, and the database file is not
59 **                        corrupt.  (This never happens because this function
60 **                        will call abort() following an invariant failure.)
61 **
62 **     (other)            Some other kind of error occurred.
63 */
fuzz_invariant(sqlite3 * db,sqlite3_stmt * pStmt,int iCnt,int iRow,int nRow,int * pbCorrupt,int eVerbosity)64 int fuzz_invariant(
65   sqlite3 *db,            /* The database connection */
66   sqlite3_stmt *pStmt,    /* Test statement stopped on an SQLITE_ROW */
67   int iCnt,               /* Invariant sequence number, starting at 0 */
68   int iRow,               /* Current row number */
69   int nRow,               /* Number of output rows from pStmt */
70   int *pbCorrupt,         /* IN/OUT: Flag indicating a corrupt database file */
71   int eVerbosity          /* How much debugging output */
72 ){
73   char *zTest;
74   sqlite3_stmt *pTestStmt = 0;
75   int rc;
76   int i;
77   int nCol;
78   int nParam;
79 
80   if( *pbCorrupt ) return SQLITE_DONE;
81   nParam = sqlite3_bind_parameter_count(pStmt);
82   if( nParam>100 ) return SQLITE_DONE;
83   zTest = fuzz_invariant_sql(pStmt, iCnt);
84   if( zTest==0 ) return SQLITE_DONE;
85   rc = sqlite3_prepare_v2(db, zTest, -1, &pTestStmt, 0);
86   if( rc ){
87     if( eVerbosity ){
88       printf("invariant compile failed: %s\n%s\n",
89              sqlite3_errmsg(db), zTest);
90     }
91     sqlite3_free(zTest);
92     sqlite3_finalize(pTestStmt);
93     return rc;
94   }
95   sqlite3_free(zTest);
96   nCol = sqlite3_column_count(pStmt);
97   for(i=0; i<nCol; i++){
98     rc = sqlite3_bind_value(pTestStmt,i+1+nParam,sqlite3_column_value(pStmt,i));
99     if( rc!=SQLITE_OK && rc!=SQLITE_RANGE ){
100       sqlite3_finalize(pTestStmt);
101       return rc;
102     }
103   }
104   if( eVerbosity>=2 ){
105     char *zSql = sqlite3_expanded_sql(pTestStmt);
106     printf("invariant-sql #%d:\n%s\n", iCnt, zSql);
107     sqlite3_free(zSql);
108   }
109   while( (rc = sqlite3_step(pTestStmt))==SQLITE_ROW ){
110     for(i=0; i<nCol; i++){
111       if( !sameValue(pStmt, i, pTestStmt, i, 0) ) break;
112     }
113     if( i>=nCol ) break;
114   }
115   if( rc==SQLITE_DONE ){
116     /* No matching output row found */
117     sqlite3_stmt *pCk = 0;
118 
119     /* This is not a fault if the database file is corrupt, because anything
120     ** can happen with a corrupt database file */
121     rc = sqlite3_prepare_v2(db, "PRAGMA integrity_check", -1, &pCk, 0);
122     if( rc ){
123       sqlite3_finalize(pCk);
124       sqlite3_finalize(pTestStmt);
125       return rc;
126     }
127     rc = sqlite3_step(pCk);
128     if( rc!=SQLITE_ROW
129      || sqlite3_column_text(pCk, 0)==0
130      || strcmp((const char*)sqlite3_column_text(pCk,0),"ok")!=0
131     ){
132       *pbCorrupt = 1;
133       sqlite3_finalize(pCk);
134       sqlite3_finalize(pTestStmt);
135       return SQLITE_CORRUPT;
136     }
137     sqlite3_finalize(pCk);
138 
139     if( sqlite3_strlike("%group%by%order%by%desc%",sqlite3_sql(pStmt),0)==0 ){
140       /* dbsqlfuzz crash-647c162051c9b23ce091b7bbbe5125ce5f00e922
141       ** Original statement is:
142       **
143       **    SELECT a,c,d,b,'' FROM t1 GROUP BY 1 HAVING d<>345 ORDER BY a DESC;
144       **
145       ** The values of c, d, and b are indeterminate and change when the
146       ** enclosed in the test query because the DESC is dropped.
147       **
148       **    SELECT * FROM (...) WHERE "a"==0
149       */
150       goto not_a_fault;
151     }
152 
153     if( sqlite3_strlike("%limit%)%order%by%", sqlite3_sql(pTestStmt),0)==0 ){
154       /* crash-89bd6a6f8c6166e9a4c5f47b3e70b225f69b76c6
155       ** Original statement is:
156       **
157       **    SELECT a,b,c* FROM t1 LIMIT 1%5<4
158       **
159       ** When running:
160       **
161       **    SELECT * FROM (...) ORDER BY 1
162       **
163       ** A different subset of the rows come out
164       */
165       goto not_a_fault;
166     }
167 
168     /* The original sameValue() comparison assumed a collating sequence
169     ** of "binary".  It can sometimes get an incorrect result for different
170     ** collating sequences.  So rerun the test with no assumptions about
171     ** collations.
172     */
173     rc = sqlite3_prepare_v2(db,
174        "SELECT ?1=?2 OR ?1=?2 COLLATE nocase OR ?1=?2 COLLATE rtrim",
175        -1, &pCk, 0);
176     if( rc==SQLITE_OK ){
177       sqlite3_reset(pTestStmt);
178       while( (rc = sqlite3_step(pTestStmt))==SQLITE_ROW ){
179         for(i=0; i<nCol; i++){
180           if( !sameValue(pStmt, i, pTestStmt, i, pCk) ) break;
181         }
182         if( i>=nCol ){
183           sqlite3_finalize(pCk);
184           goto not_a_fault;
185         }
186       }
187     }
188     sqlite3_finalize(pCk);
189 
190     /* Invariants do not necessarily work if there are virtual tables
191     ** involved in the query */
192     rc = sqlite3_prepare_v2(db,
193             "SELECT 1 FROM bytecode(?1) WHERE opcode='VOpen'", -1, &pCk, 0);
194     if( rc==SQLITE_OK ){
195       sqlite3_bind_pointer(pCk, 1, pStmt, "stmt-pointer", 0);
196       rc = sqlite3_step(pCk);
197     }
198     sqlite3_finalize(pCk);
199     if( rc==SQLITE_DONE ){
200       reportInvariantFailed(pStmt, pTestStmt, iRow);
201       return SQLITE_INTERNAL;
202     }else if( eVerbosity>0 ){
203       printf("invariant-error ignored due to the use of virtual tables\n");
204     }
205   }
206 not_a_fault:
207   sqlite3_finalize(pTestStmt);
208   return SQLITE_OK;
209 }
210 
211 
212 /*
213 ** Generate SQL used to test a statement invariant.
214 **
215 ** Return 0 if the iCnt is out of range.
216 **
217 ** iCnt meanings:
218 **
219 **   0     SELECT * FROM (<query>)
220 **   1     SELECT DISTINCT * FROM (<query>)
221 **   2     SELECT * FROM (<query>) WHERE ORDER BY 1
222 **   3     SELECT DISTINCT * FROM (<query>) ORDER BY 1
223 **   4     SELECT * FROM (<query>) WHERE <all-columns>=<all-values>
224 **   5     SELECT DISTINCT * FROM (<query>) WHERE <all-columns=<all-values
225 **   6     SELECT * FROM (<query>) WHERE <all-column>=<all-value> ORDER BY 1
226 **   7     SELECT DISTINCT * FROM (<query>) WHERE <all-column>=<all-value>
227 **                           ORDER BY 1
228 **   N+0   SELECT * FROM (<query>) WHERE <nth-column>=<value>
229 **   N+1   SELECT DISTINCT * FROM (<query>) WHERE <Nth-column>=<value>
230 **   N+2   SELECT * FROM (<query>) WHERE <Nth-column>=<value> ORDER BY 1
231 **   N+3   SELECT DISTINCT * FROM (<query>) WHERE <Nth-column>=<value>
232 **                           ORDER BY N
233 **
234 */
fuzz_invariant_sql(sqlite3_stmt * pStmt,int iCnt)235 static char *fuzz_invariant_sql(sqlite3_stmt *pStmt, int iCnt){
236   const char *zIn;
237   size_t nIn;
238   const char *zAnd = "WHERE";
239   int i;
240   sqlite3_str *pTest;
241   sqlite3_stmt *pBase = 0;
242   sqlite3 *db = sqlite3_db_handle(pStmt);
243   int rc;
244   int nCol = sqlite3_column_count(pStmt);
245   int mxCnt;
246   int bDistinct = 0;
247   int bOrderBy = 0;
248   int nParam = sqlite3_bind_parameter_count(pStmt);
249 
250   switch( iCnt % 4 ){
251     case 1:  bDistinct = 1;              break;
252     case 2:  bOrderBy = 1;               break;
253     case 3:  bDistinct = bOrderBy = 1;   break;
254   }
255   iCnt /= 4;
256   mxCnt = nCol;
257   if( iCnt<0 || iCnt>mxCnt ) return 0;
258   zIn = sqlite3_sql(pStmt);
259   if( zIn==0 ) return 0;
260   nIn = strlen(zIn);
261   while( nIn>0 && (isspace(zIn[nIn-1]) || zIn[nIn-1]==';') ) nIn--;
262   if( strchr(zIn, '?') ) return 0;
263   pTest = sqlite3_str_new(0);
264   sqlite3_str_appendf(pTest, "SELECT %s* FROM (",
265                       bDistinct ? "DISTINCT " : "");
266   sqlite3_str_append(pTest, zIn, (int)nIn);
267   sqlite3_str_append(pTest, ")", 1);
268   rc = sqlite3_prepare_v2(db, sqlite3_str_value(pTest), -1, &pBase, 0);
269   if( rc ){
270     sqlite3_finalize(pBase);
271     pBase = pStmt;
272   }
273   for(i=0; i<sqlite3_column_count(pStmt); i++){
274     const char *zColName = sqlite3_column_name(pBase,i);
275     const char *zSuffix = zColName ? strrchr(zColName, ':') : 0;
276     if( zSuffix
277      && isdigit(zSuffix[1])
278      && (zSuffix[1]>'3' || isdigit(zSuffix[2]))
279     ){
280       /* This is a randomized column name and so cannot be used in the
281       ** WHERE clause. */
282       continue;
283     }
284     if( iCnt==0 ) continue;
285     if( iCnt>1 && i+2!=iCnt ) continue;
286     if( zColName==0 ) continue;
287     if( sqlite3_column_type(pStmt, i)==SQLITE_NULL ){
288       sqlite3_str_appendf(pTest, " %s \"%w\" ISNULL", zAnd, zColName);
289     }else{
290       sqlite3_str_appendf(pTest, " %s \"%w\"=?%d", zAnd, zColName,
291                           i+1+nParam);
292     }
293     zAnd = "AND";
294   }
295   if( pBase!=pStmt ) sqlite3_finalize(pBase);
296   if( bOrderBy ){
297     sqlite3_str_appendf(pTest, " ORDER BY %d", iCnt>2 ? iCnt-1 : 1);
298   }
299   return sqlite3_str_finish(pTest);
300 }
301 
302 /*
303 ** Return true if and only if v1 and is the same as v2.
304 */
sameValue(sqlite3_stmt * pS1,int i1,sqlite3_stmt * pS2,int i2,sqlite3_stmt * pTestCompare)305 static int sameValue(
306   sqlite3_stmt *pS1, int i1,       /* Value to text on the left */
307   sqlite3_stmt *pS2, int i2,       /* Value to test on the right */
308   sqlite3_stmt *pTestCompare       /* COLLATE comparison statement or NULL */
309 ){
310   int x = 1;
311   int t1 = sqlite3_column_type(pS1,i1);
312   int t2 = sqlite3_column_type(pS2,i2);
313   if( t1!=t2 ){
314     if( (t1==SQLITE_INTEGER && t2==SQLITE_FLOAT)
315      || (t1==SQLITE_FLOAT && t2==SQLITE_INTEGER)
316     ){
317       /* Comparison of numerics is ok */
318     }else{
319       return 0;
320     }
321   }
322   switch( sqlite3_column_type(pS1,i1) ){
323     case SQLITE_INTEGER: {
324       x =  sqlite3_column_int64(pS1,i1)==sqlite3_column_int64(pS2,i2);
325       break;
326     }
327     case SQLITE_FLOAT: {
328       x = sqlite3_column_double(pS1,i1)==sqlite3_column_double(pS2,i2);
329       break;
330     }
331     case SQLITE_TEXT: {
332       int e1 = sqlite3_value_encoding(sqlite3_column_value(pS1,i1));
333       int e2 = sqlite3_value_encoding(sqlite3_column_value(pS2,i2));
334       if( e1!=e2 ){
335         const char *z1 = (const char*)sqlite3_column_text(pS1,i1);
336         const char *z2 = (const char*)sqlite3_column_text(pS2,i2);
337         x = ((z1==0 && z2==0) || (z1!=0 && z2!=0 && strcmp(z1,z1)==0));
338         printf("Encodings differ.  %d on left and %d on right\n", e1, e2);
339         abort();
340       }
341       if( pTestCompare ){
342         sqlite3_bind_value(pTestCompare, 1, sqlite3_column_value(pS1,i1));
343         sqlite3_bind_value(pTestCompare, 2, sqlite3_column_value(pS2,i2));
344         x = sqlite3_step(pTestCompare)==SQLITE_ROW
345                       && sqlite3_column_int(pTestCompare,0)!=0;
346         sqlite3_reset(pTestCompare);
347         break;
348       }
349       if( e1!=SQLITE_UTF8 ){
350         int len1 = sqlite3_column_bytes16(pS1,i1);
351         const unsigned char *b1 = sqlite3_column_blob(pS1,i1);
352         int len2 = sqlite3_column_bytes16(pS2,i2);
353         const unsigned char *b2 = sqlite3_column_blob(pS2,i2);
354         if( len1!=len2 ){
355           x = 0;
356         }else if( len1==0 ){
357           x = 1;
358         }else{
359           x = (b1!=0 && b2!=0 && memcmp(b1,b2,len1)==0);
360         }
361         break;
362       }
363       /* Fall through into the SQLITE_BLOB case */
364     }
365     case SQLITE_BLOB: {
366       int len1 = sqlite3_column_bytes(pS1,i1);
367       const unsigned char *b1 = sqlite3_column_blob(pS1,i1);
368       int len2 = sqlite3_column_bytes(pS2,i2);
369       const unsigned char *b2 = sqlite3_column_blob(pS2,i2);
370       if( len1!=len2 ){
371         x = 0;
372       }else if( len1==0 ){
373         x = 1;
374       }else{
375         x = (b1!=0 && b2!=0 && memcmp(b1,b2,len1)==0);
376       }
377       break;
378     }
379   }
380   return x;
381 }
382 
383 /*
384 ** Print binary data as hex
385 */
printHex(const unsigned char * a,int n,int mx)386 static void printHex(const unsigned char *a, int n, int mx){
387   int j;
388   for(j=0; j<mx && j<n; j++){
389     printf("%02x", a[j]);
390   }
391   if( j<n ) printf("...");
392 }
393 
394 /*
395 ** Print a single row from the prepared statement
396 */
printRow(sqlite3_stmt * pStmt,int iRow)397 static void printRow(sqlite3_stmt *pStmt, int iRow){
398   int i, n, nCol;
399   unsigned const char *data;
400   nCol = sqlite3_column_count(pStmt);
401   for(i=0; i<nCol; i++){
402     printf("row%d.col%d = ", iRow, i);
403     switch( sqlite3_column_type(pStmt, i) ){
404       case SQLITE_NULL: {
405         printf("NULL\n");
406         break;
407       }
408       case SQLITE_INTEGER: {
409         printf("(integer) %lld\n", sqlite3_column_int64(pStmt, i));
410         break;
411       }
412       case SQLITE_FLOAT: {
413         printf("(float) %f\n", sqlite3_column_double(pStmt, i));
414         break;
415       }
416       case SQLITE_TEXT: {
417         switch( sqlite3_value_encoding(sqlite3_column_value(pStmt,i)) ){
418           case SQLITE_UTF8: {
419             printf("(utf8) x'");
420             n = sqlite3_column_bytes(pStmt, i);
421             data = sqlite3_column_blob(pStmt, i);
422             printHex(data, n, 35);
423             printf("'\n");
424             break;
425           }
426           case SQLITE_UTF16BE: {
427             printf("(utf16be) x'");
428             n = sqlite3_column_bytes16(pStmt, i);
429             data = sqlite3_column_blob(pStmt, i);
430             printHex(data, n, 35);
431             printf("'\n");
432             break;
433           }
434           case SQLITE_UTF16LE: {
435             printf("(utf16le) x'");
436             n = sqlite3_column_bytes16(pStmt, i);
437             data = sqlite3_column_blob(pStmt, i);
438             printHex(data, n, 35);
439             printf("'\n");
440             break;
441           }
442           default: {
443             printf("Illegal return from sqlite3_value_encoding(): %d\n",
444                 sqlite3_value_encoding(sqlite3_column_value(pStmt,i)));
445             abort();
446           }
447         }
448         break;
449       }
450       case SQLITE_BLOB: {
451         n = sqlite3_column_bytes(pStmt, i);
452         data = sqlite3_column_blob(pStmt, i);
453         printf("(blob %d bytes) x'", n);
454         printHex(data, n, 35);
455         printf("'\n");
456         break;
457       }
458     }
459   }
460 }
461 
462 /*
463 ** Report a failure of the invariant:  The current output row of pOrig
464 ** does not appear in any row of the output from pTest.
465 */
reportInvariantFailed(sqlite3_stmt * pOrig,sqlite3_stmt * pTest,int iRow)466 static void reportInvariantFailed(
467   sqlite3_stmt *pOrig,   /* The original query */
468   sqlite3_stmt *pTest,   /* The alternative test query with a missing row */
469   int iRow               /* Row number in pOrig */
470 ){
471   int iTestRow = 0;
472   printf("Invariant check failed on row %d.\n", iRow);
473   printf("Original query --------------------------------------------------\n");
474   printf("%s\n", sqlite3_expanded_sql(pOrig));
475   printf("Alternative query -----------------------------------------------\n");
476   printf("%s\n", sqlite3_expanded_sql(pTest));
477   printf("Result row that is missing from the alternative -----------------\n");
478   printRow(pOrig, iRow);
479   printf("Complete results from the alternative query ---------------------\n");
480   sqlite3_reset(pTest);
481   while( sqlite3_step(pTest)==SQLITE_ROW ){
482     iTestRow++;
483     printRow(pTest, iTestRow);
484   }
485   sqlite3_finalize(pTest);
486   abort();
487 }
488