Memory leak is possible for UDF array arguments

[dmitry-lipetsk]

FB v4.0.4.2988

If this line throws

firebird/src/jrd/fun.epp

Line 1018 in 9591891

UCHAR* const temp = FB_NEW_POOL(*getDefaultMemoryPool()) UCHAR[static_cast<size_t>(to.dsc_length * n)];

you do not release the allocated memory in 'data'

firebird/src/jrd/fun.epp

Line 1005 in 9591891

UCHAR* data = FB_NEW_POOL(*getDefaultMemoryPool()) UCHAR[static_cast<size_t>(array_desc->iad_total_length)];

Bricks for reproducing this problem:

DECLARE EXTERNAL FUNCTION MY_UDF__GET_DIM
 INTEGER BY SCALAR_ARRAY
RETURNS INTEGER BY VALUE
ENTRY_POINT 'fn_get_dim__i4__sa'
MODULE_NAME 'lcpi.test.udf.01.dll'

select MY_UDF__GET_DIM(cast('12311111111111111111111111111111111111111111111111111111111111111111111111111111111' as blob)) from rdb$database

Memory releases after disconnect.

[dyemanov]

Please verify the fix and I will port it to other versions.

[dmitry-lipetsk]

Good time of day.

Fix looks ok.

Another simple tests work fine, too.

Source code of test UDF:

#include "source/fn.h"
#include <cstddef>

////////////////////////////////////////////////////////////////////////////////

using SSHORT=std::int16_t;
using SLONG =std::int32_t;
using SCHAR =char;

////////////////////////////////////////////////////////////////////////////////

typedef struct dsc
{
 UCHAR  dsc_dtype;
 SCHAR  dsc_scale;
 USHORT dsc_length;
 SSHORT dsc_sub_type;
 USHORT dsc_flags;
 UCHAR* dsc_address; // Used either as offset in a message or as a pointer
} DSC;

struct scalar_array_desc
{
 DSC   sad_desc;
 SLONG sad_dimensions;
 
 struct sad_repeat
 {
  SLONG sad_lower;
  SLONG sad_upper;
 } sad_rpt[1];
};//struct scalar_array_desc

////////////////////////////////////////////////////////////////////////////////

std::int32_t FBUDF_API fn_get_dim__i4__sa(const scalar_array_desc* const descr)
{
 if(descr->sad_desc.dsc_address==nullptr)
  return 0;

 return descr->sad_dimensions;
}//fn_get_dim__i4__sa

////////////////////////////////////////////////////////////////////////////////

[onlylunix]

Will there be a fix for version 3.0.11?

[dyemanov]

Will there be a fix for version 3.0.11?

Backported.