Re: (ITS#3456) test018 consumer segfault

[Kurt@OpenLDAP.org]

I think this might be a red herring as well.  I rebuilt
--with-multiple-precision=no and still see the original
fault (when not using efence).  I just rebuilt with
--with-multiple-precision=gmp and still see the original
fault (when not using efence).  So, the version (0.9.7d)
of OpenSSL I'm using here should be moot.

At 10:58 PM 1/27/2005, Pierangelo Masarati wrote:
>Kurt@OpenLDAP.org wrote:
>
>>At 03:28 PM 1/27/2005, you wrote:
>> 
>>
>>>That looks like an Efence error, do you have the multithreading patches for Efence applied?
>>>   
>>
>>With your patch with -DUSE_PTHREADS, eftest fails.
>>With your patch without -DUSE_PTHREAD (and without -DUSE_SEMAPHORE)
>>I got:
>>
>>Program received signal SIGBUS, Bus error.
>>EF_memalign (alignment=4, userSize=8, dolock=1) at efence.c:505
>>505                     if ( slot->mode == FREE
>>(gdb) where
>>#0  EF_memalign (alignment=4, userSize=8, dolock=1) at efence.c:505
>>#1  0x80e6e04 in memalign (alignment=4, userSize=8) at efence.c:653
>>#2  0x80e7231 in malloc (size=8) at efence.c:860
>>#3  0x2832ef05 in bn_cmp_part_words () from /usr/lib/libcrypto.so.3
>>#4  0x2832f4e4 in CRYPTO_malloc () from /usr/lib/libcrypto.so.3
>>#5  0x2832e423 in BN_new () from /usr/lib/libcrypto.so.3
>>#6  0x2832e621 in bn_expand2 () from /usr/lib/libcrypto.so.3
>>#7  0x2832442f in BN_add_word () from /usr/lib/libcrypto.so.3
>>#8  0x8061abf in send_ldap_response (op=0x2d4e7e5c, rs=0x2de64db0)
>>   at result.c:456
>>#9  0x8061f43 in slap_send_ldap_result (op=0x2d4e7e5c, rs=0x2de64db0)
>>   at result.c:607
>>#10 0x806bb24 in fe_op_bind (op=0x2d4e7e5c, rs=0x2de64db0) at bind.c:402
>>#11 0x806b4a0 in do_bind (op=0x2d4e7e5c, rs=0x2de64db0) at bind.c:204
>>#12 0x8056b24 in connection_operation (ctx=0x2de64e3c, arg_v=0x2d4e7e5c)
>>   at connection.c:1038
>>#13 0x80c71b9 in ldap_int_thread_pool_wrapper (xpool=0x29146fc8) at tpool.c:467
>>#14 0x2838c4c8 in _thread_start () from /usr/lib/libc_r.so.4
>>#15 0x0 in ?? ()
>>
>>
>> 
>I note that that code is using ldap_pvt_mp_add_ulong(), i.e. BN_add_word(), with a var "bytes" that's actually a "long".  Could this be part of the problem?  Could you try casting it to (long)?
>
>diff -u -r1.245 result.c
>--- result.c    1 Jan 2005 19:49:50 -0000       1.245
>+++ result.c    28 Jan 2005 06:55:14 -0000
>@@ -453,7 +453,7 @@
>       ldap_pvt_thread_mutex_lock( &slap_counters.sc_sent_mutex );
>       ldap_pvt_mp_add_ulong( slap_counters.sc_pdu, 1 );
>-       ldap_pvt_mp_add_ulong( slap_counters.sc_bytes, bytes );
>+       ldap_pvt_mp_add_ulong( slap_counters.sc_bytes, (unsigned long)bytes );
>       ldap_pvt_thread_mutex_unlock( &slap_counters.sc_sent_mutex );
>cleanup:;
>
>I doubt this can do any harm, but just in case...  What version of openssl are you using?  I wonder if this may be part of the problem (i.e. a bug in openssl allocation?)
>
>p.  
>
>
>
>
>
>   SysNet - via Dossi,8 27100 Pavia Tel: +390382573859 Fax: +390382476497