[Mhash-dev] sigBUS in mutils_word32nswap() within lib/stdfns.c

[Dennis C. <dc...@bl...>]

On Solaris 10 and compiled with gcc 4.8.0 I see this during testing : 



Checking fragmentation capabilities of HAVAL256: /usr/local/bin/bash: line 4: 21260 Bus Error               (core dumped) ${dir}$tst
FAIL: frag_test
============================================
1 of 5 tests failed
Please report to mha...@so...
============================================


Looking into this I see : 

node002$ dbx ./src/.libs/frag_test time_1372300240-pid_21260-uid_16411-gid_1-fid_frag_test.core
Reading frag_test
core file header read successfully
Reading ld.so.1
Reading libmhash.so.2.0.1
Reading libc.so.1
Reading libgcc_s.so.1
Reading libc_psr.so.1
program terminated by signal BUS (invalid address alignment)
0xffffffff7f005a28: mutils_word32nswap+0x0094:  ld       [%g1], %g1
(dbx) where                                                                                                                      
=>[1] mutils_word32nswap(0xffffffff7fffeb09, 0x20, 0x0, 0x0, 0x0, 0x0), at 0xffffffff7f005a28 
  [2] havalTransform3(0x1001045d4, 0xffffffff7fffeb09, 0x100104680, 0x18c07f, 0xd2355034, 0x143367), at 0xffffffff7f029228 
  [3] havalUpdate(0x1001045d0, 0xffffffff7fffeb08, 0x81, 0xfffffffffffffff9, 0x0, 0xffffffff7fffeb89), at 0xffffffff7f040c4c 
  [4] mhash(0x100101dd0, 0xffffffff7fffeb08, 0x81, 0xffffffff7ef4fc81, 0x0, 0x140), at 0xffffffff7f004684 
  [5] frag_test(0x3, 0x0, 0x182ba0, 0x7fffffff, 0x0, 0x0), at 0x100001160 
  [6] main(0x1, 0xffffffff7fffefd8, 0xffffffff7fffefe8, 0x100101da0, 0x100000000, 0xffffffff7ea00200), at 0x100001568 
(dbx) quit


Problem is here in  lib/stdfns.c : 

/*
 * Byte swap a series of 32-bit integers. If destructive is set to false, the
 * output will be placed in a freshly malloc()'ed buffer and the original
 * data will remain intact.
 */
WIN32DLL_DEFINE
mutils_word32 *
mutils_word32nswap(mutils_word32 *x, mutils_word32 n, mutils_boolean destructive)
{
        mutils_word32 loop;
        mutils_word32 *buffer;
        mutils_word32 *ptrIn;
        mutils_word32 *ptrOut;
        mutils_word32 count = n * 4;

        if (destructive == MUTILS_FALSE)
        {
                buffer = mutils_malloc(count);
                if (buffer == NULL)
                {
                        return(NULL);
                }
        }
        else
        {
                buffer = x;
        }

/*
 * This loop is totally useless for destructive processing of little-endian
 * data on a little-endian machine.
 */

        for (loop = 0, ptrIn = x, ptrOut = buffer; loop < n; loop++, ptrOut++, ptrIn++)
        {
                *ptrOut = mutils_lend32(*ptrIn);
        }

        return(buffer);
}

First thought I have is "why use a non-standard datatype where uint32_t works fine?" followed by "why not use calloc to zero fill the buffer ? 

Anyways .. I see this mail list is essentially dead of traffic so I may just look into this, fix it and post the patch all to myself. 

Dennis