Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commit096d34c

Browse files
committed
Fix failure to zero-pad the result of bitshiftright().
If the bitstring length is not a multiple of 8, we'd shift therightmost bits into the pad space, which must be zeroes --- bit_cmp,for one, depends on that. This'd lead to the result failing tocompare equal to what it should compare equal to, as reported inbug #16013 from Daryl Waycott.This is, if memory serves, not the first such bug in the bitstringfunctions. In hopes of making it the last one, do a bit more workthan minimally necessary to fix the bug:* Add assertion checks to bit_out() and varbit_out() to complain ifthey are given incorrectly-padded input. This will improve theodds that manual testing of any new patch finds problems.* Encapsulate the padding-related logic in macros to make iteasier to use.Also, remove unnecessary padding logic from bit_or() and bitxor().Somebody had already noted that we need not re-pad the result ofbit_and() since the inputs are required to be the same length,but failed to extrapolate that to the other two.Also, move a comment block that once was near the head of varbit.c(but people kept putting other stuff in front of it), to put it inthe header block.Note for the release notes: if anyone has inconsistent data as aresult of saving the output of bitshiftright() in a table, it'spossible to fix it with something likeUPDATE mytab SET bitcol = ~(~bitcol) WHERE bitcol != ~(~bitcol);This has been broken since day one, so back-patch to all supportedbranches.Discussion:https://postgr.es/m/16013-c2765b6996aacae9@postgresql.org
1 parent769802e commit096d34c

File tree

4 files changed

+154
-89
lines changed

4 files changed

+154
-89
lines changed

‎src/backend/utils/adt/varbit.c

Lines changed: 77 additions & 88 deletions
Original file line numberDiff line numberDiff line change
@@ -3,6 +3,21 @@
33
* varbit.c
44
* Functions for the SQL datatypes BIT() and BIT VARYING().
55
*
6+
* The data structure contains the following elements:
7+
* header -- length of the whole data structure (incl header)
8+
* in bytes (as with all varying length datatypes)
9+
* data section -- private data section for the bits data structures
10+
* bitlength -- length of the bit string in bits
11+
* bitdata -- bit string, most significant byte first
12+
*
13+
* The length of the bitdata vector should always be exactly as many
14+
* bytes as are needed for the given bitlength. If the bitlength is
15+
* not a multiple of 8, the extra low-order padding bits of the last
16+
* byte must be zeroes.
17+
*
18+
* attypmod is defined as the length of the bit string in bits, or for
19+
* varying bits the maximum length.
20+
*
621
* Code originally contributed by Adriaan Joubert.
722
*
823
* Portions Copyright (c) 1996-2017, PostgreSQL Global Development Group
@@ -25,6 +40,40 @@
2540

2641
#defineHEXDIG(z) ((z)<10 ? ((z)+'0') : ((z)-10+'A'))
2742

43+
/* Mask off any bits that should be zero in the last byte of a bitstring */
44+
#defineVARBIT_PAD(vb) \
45+
do { \
46+
int32pad_ = VARBITPAD(vb); \
47+
Assert(pad_ >= 0 && pad_ < BITS_PER_BYTE); \
48+
if (pad_ > 0) \
49+
*(VARBITS(vb) + VARBITBYTES(vb) - 1) &= BITMASK << pad_; \
50+
} while (0)
51+
52+
/*
53+
* Many functions work byte-by-byte, so they have a pointer handy to the
54+
* last-plus-one byte, which saves a cycle or two.
55+
*/
56+
#defineVARBIT_PAD_LAST(vb,ptr) \
57+
do { \
58+
int32pad_ = VARBITPAD(vb); \
59+
Assert(pad_ >= 0 && pad_ < BITS_PER_BYTE); \
60+
if (pad_ > 0) \
61+
*((ptr) - 1) &= BITMASK << pad_; \
62+
} while (0)
63+
64+
/* Assert proper padding of a bitstring */
65+
#ifdefUSE_ASSERT_CHECKING
66+
#defineVARBIT_CORRECTLY_PADDED(vb) \
67+
do { \
68+
int32pad_ = VARBITPAD(vb); \
69+
Assert(pad_ >= 0 && pad_ < BITS_PER_BYTE); \
70+
Assert(pad_ == 0 || \
71+
(*(VARBITS(vb) + VARBITBYTES(vb) - 1) & ~(BITMASK << pad_)) == 0); \
72+
} while (0)
73+
#else
74+
#defineVARBIT_CORRECTLY_PADDED(vb) ((void) 0)
75+
#endif
76+
2877
staticVarBit*bit_catenate(VarBit*arg1,VarBit*arg2);
2978
staticVarBit*bitsubstring(VarBit*arg,int32s,int32l,
3079
boollength_not_specified);
@@ -85,24 +134,6 @@ anybit_typmodout(int32 typmod)
85134
}
86135

87136

88-
/*----------
89-
*attypmod -- contains the length of the bit string in bits, or for
90-
* varying bits the maximum length.
91-
*
92-
*The data structure contains the following elements:
93-
* header -- length of the whole data structure (incl header)
94-
* in bytes. (as with all varying length datatypes)
95-
* data section -- private data section for the bits data structures
96-
*bitlength -- length of the bit string in bits
97-
*bitdata -- bit string, most significant byte first
98-
*
99-
*The length of the bitdata vector should always be exactly as many
100-
*bytes as are needed for the given bitlength. If the bitlength is
101-
*not a multiple of 8, the extra low-order padding bits of the last
102-
*byte must be zeroes.
103-
*----------
104-
*/
105-
106137
/*
107138
* bit_in -
108139
* converts a char string to the internal representation of a bitstring.
@@ -262,6 +293,9 @@ bit_out(PG_FUNCTION_ARGS)
262293
len,
263294
bitlen;
264295

296+
/* Assertion to help catch any bit functions that don't pad correctly */
297+
VARBIT_CORRECTLY_PADDED(s);
298+
265299
bitlen=VARBITLEN(s);
266300
len= (bitlen+3) /4;
267301
result= (char*)palloc(len+2);
@@ -302,8 +336,6 @@ bit_recv(PG_FUNCTION_ARGS)
302336
VarBit*result;
303337
intlen,
304338
bitlen;
305-
intipad;
306-
bits8mask;
307339

308340
bitlen=pq_getmsgint(buf,sizeof(int32));
309341
if (bitlen<0||bitlen>VARBITMAXLEN)
@@ -328,13 +360,8 @@ bit_recv(PG_FUNCTION_ARGS)
328360

329361
pq_copymsgbytes(buf, (char*)VARBITS(result),VARBITBYTES(result));
330362

331-
/* Make sure last byte is zero-padded if needed */
332-
ipad=VARBITPAD(result);
333-
if (ipad>0)
334-
{
335-
mask=BITMASK <<ipad;
336-
*(VARBITS(result)+VARBITBYTES(result)-1) &=mask;
337-
}
363+
/* Make sure last byte is correctly zero-padded */
364+
VARBIT_PAD(result);
338365

339366
PG_RETURN_VARBIT_P(result);
340367
}
@@ -365,8 +392,6 @@ bit(PG_FUNCTION_ARGS)
365392
boolisExplicit=PG_GETARG_BOOL(2);
366393
VarBit*result;
367394
intrlen;
368-
intipad;
369-
bits8mask;
370395

371396
/* No work if typmod is invalid or supplied data matches it already */
372397
if (len <=0||len>VARBITMAXLEN||len==VARBITLEN(arg))
@@ -392,12 +417,7 @@ bit(PG_FUNCTION_ARGS)
392417
* if source data was shorter than target length (we assume the last byte
393418
* of the source data was itself correctly zero-padded).
394419
*/
395-
ipad=VARBITPAD(result);
396-
if (ipad>0)
397-
{
398-
mask=BITMASK <<ipad;
399-
*(VARBITS(result)+VARBITBYTES(result)-1) &=mask;
400-
}
420+
VARBIT_PAD(result);
401421

402422
PG_RETURN_VARBIT_P(result);
403423
}
@@ -572,6 +592,9 @@ varbit_out(PG_FUNCTION_ARGS)
572592
k,
573593
len;
574594

595+
/* Assertion to help catch any bit functions that don't pad correctly */
596+
VARBIT_CORRECTLY_PADDED(s);
597+
575598
len=VARBITLEN(s);
576599
result= (char*)palloc(len+1);
577600
sp=VARBITS(s);
@@ -618,8 +641,6 @@ varbit_recv(PG_FUNCTION_ARGS)
618641
VarBit*result;
619642
intlen,
620643
bitlen;
621-
intipad;
622-
bits8mask;
623644

624645
bitlen=pq_getmsgint(buf,sizeof(int32));
625646
if (bitlen<0||bitlen>VARBITMAXLEN)
@@ -644,13 +665,8 @@ varbit_recv(PG_FUNCTION_ARGS)
644665

645666
pq_copymsgbytes(buf, (char*)VARBITS(result),VARBITBYTES(result));
646667

647-
/* Make sure last byte is zero-padded if needed */
648-
ipad=VARBITPAD(result);
649-
if (ipad>0)
650-
{
651-
mask=BITMASK <<ipad;
652-
*(VARBITS(result)+VARBITBYTES(result)-1) &=mask;
653-
}
668+
/* Make sure last byte is correctly zero-padded */
669+
VARBIT_PAD(result);
654670

655671
PG_RETURN_VARBIT_P(result);
656672
}
@@ -718,8 +734,6 @@ varbit(PG_FUNCTION_ARGS)
718734
boolisExplicit=PG_GETARG_BOOL(2);
719735
VarBit*result;
720736
intrlen;
721-
intipad;
722-
bits8mask;
723737

724738
/* No work if typmod is invalid or supplied data matches it already */
725739
if (len <=0||len >=VARBITLEN(arg))
@@ -738,13 +752,8 @@ varbit(PG_FUNCTION_ARGS)
738752

739753
memcpy(VARBITS(result),VARBITS(arg),VARBITBYTES(result));
740754

741-
/* Make sure last byte is zero-padded if needed */
742-
ipad=VARBITPAD(result);
743-
if (ipad>0)
744-
{
745-
mask=BITMASK <<ipad;
746-
*(VARBITS(result)+VARBITBYTES(result)-1) &=mask;
747-
}
755+
/* Make sure last byte is correctly zero-padded */
756+
VARBIT_PAD(result);
748757

749758
PG_RETURN_VARBIT_P(result);
750759
}
@@ -1002,6 +1011,8 @@ bit_catenate(VarBit *arg1, VarBit *arg2)
10021011
}
10031012
}
10041013

1014+
/* The pad bits should be already zero at this point */
1015+
10051016
returnresult;
10061017
}
10071018

@@ -1035,14 +1046,12 @@ bitsubstring(VarBit *arg, int32 s, int32 l, bool length_not_specified)
10351046
intbitlen,
10361047
rbitlen,
10371048
len,
1038-
ipad=0,
10391049
ishift,
10401050
i;
10411051
inte,
10421052
s1,
10431053
e1;
1044-
bits8mask,
1045-
*r,
1054+
bits8*r,
10461055
*ps;
10471056

10481057
bitlen=VARBITLEN(arg);
@@ -1107,13 +1116,9 @@ bitsubstring(VarBit *arg, int32 s, int32 l, bool length_not_specified)
11071116
r++;
11081117
}
11091118
}
1110-
/* Do we need to pad at the end? */
1111-
ipad=VARBITPAD(result);
1112-
if (ipad>0)
1113-
{
1114-
mask=BITMASK <<ipad;
1115-
*(VARBITS(result)+len-1) &=mask;
1116-
}
1119+
1120+
/* Make sure last byte is correctly zero-padded */
1121+
VARBIT_PAD(result);
11171122
}
11181123

11191124
returnresult;
@@ -1236,7 +1241,7 @@ bit_and(PG_FUNCTION_ARGS)
12361241
for (i=0;i<VARBITBYTES(arg1);i++)
12371242
*r++=*p1++&*p2++;
12381243

1239-
/* Padding is not needed as & of 0pad is 0 */
1244+
/* Padding is not needed as & of 0pads is 0 */
12401245

12411246
PG_RETURN_VARBIT_P(result);
12421247
}
@@ -1258,7 +1263,6 @@ bit_or(PG_FUNCTION_ARGS)
12581263
bits8*p1,
12591264
*p2,
12601265
*r;
1261-
bits8mask;
12621266

12631267
bitlen1=VARBITLEN(arg1);
12641268
bitlen2=VARBITLEN(arg2);
@@ -1277,13 +1281,7 @@ bit_or(PG_FUNCTION_ARGS)
12771281
for (i=0;i<VARBITBYTES(arg1);i++)
12781282
*r++=*p1++ |*p2++;
12791283

1280-
/* Pad the result */
1281-
mask=BITMASK <<VARBITPAD(result);
1282-
if (mask)
1283-
{
1284-
r--;
1285-
*r &=mask;
1286-
}
1284+
/* Padding is not needed as | of 0 pads is 0 */
12871285

12881286
PG_RETURN_VARBIT_P(result);
12891287
}
@@ -1305,7 +1303,6 @@ bitxor(PG_FUNCTION_ARGS)
13051303
bits8*p1,
13061304
*p2,
13071305
*r;
1308-
bits8mask;
13091306

13101307
bitlen1=VARBITLEN(arg1);
13111308
bitlen2=VARBITLEN(arg2);
@@ -1325,13 +1322,7 @@ bitxor(PG_FUNCTION_ARGS)
13251322
for (i=0;i<VARBITBYTES(arg1);i++)
13261323
*r++=*p1++ ^*p2++;
13271324

1328-
/* Pad the result */
1329-
mask=BITMASK <<VARBITPAD(result);
1330-
if (mask)
1331-
{
1332-
r--;
1333-
*r &=mask;
1334-
}
1325+
/* Padding is not needed as ^ of 0 pads is 0 */
13351326

13361327
PG_RETURN_VARBIT_P(result);
13371328
}
@@ -1347,7 +1338,6 @@ bitnot(PG_FUNCTION_ARGS)
13471338
VarBit*result;
13481339
bits8*p,
13491340
*r;
1350-
bits8mask;
13511341

13521342
result= (VarBit*)palloc(VARSIZE(arg));
13531343
SET_VARSIZE(result,VARSIZE(arg));
@@ -1358,13 +1348,8 @@ bitnot(PG_FUNCTION_ARGS)
13581348
for (;p<VARBITEND(arg);p++)
13591349
*r++= ~*p;
13601350

1361-
/* Pad the result */
1362-
mask=BITMASK <<VARBITPAD(result);
1363-
if (mask)
1364-
{
1365-
r--;
1366-
*r &=mask;
1367-
}
1351+
/* Must zero-pad the result, because extra bits are surely 1's here */
1352+
VARBIT_PAD_LAST(result,r);
13681353

13691354
PG_RETURN_VARBIT_P(result);
13701355
}
@@ -1431,6 +1416,8 @@ bitshiftleft(PG_FUNCTION_ARGS)
14311416
*r=0;
14321417
}
14331418

1419+
/* The pad bits should be already zero at this point */
1420+
14341421
PG_RETURN_VARBIT_P(result);
14351422
}
14361423

@@ -1497,6 +1484,8 @@ bitshiftright(PG_FUNCTION_ARGS)
14971484
if ((++r)<VARBITEND(result))
14981485
*r= (*p << (BITS_PER_BYTE-ishift))&BITMASK;
14991486
}
1487+
/* We may have shifted 1's into the pad bits, so fix that */
1488+
VARBIT_PAD_LAST(result,r);
15001489
}
15011490

15021491
PG_RETURN_VARBIT_P(result);

‎src/include/utils/varbit.h

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -21,6 +21,11 @@
2121

2222
/*
2323
* Modeled on struct varlena from postgres.h, but data type is bits8.
24+
*
25+
* Caution: if bit_len is not a multiple of BITS_PER_BYTE, the low-order
26+
* bits of the last byte of bit_dat[] are unused and MUST be zeroes.
27+
* (This allows bit_cmp() to not bother masking the last byte.)
28+
* Also, there should not be any excess bytes counted in the header length.
2429
*/
2530
typedefstruct
2631
{

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp