Re: gcc4 compilation error

[Jeff Sheltren]

Hi Mark, I finally got some time to do a bit more debugging on this, and it
looks like the buffer overflow may not be related to gcc4, but rather a
compile flag which is used by default on Fedora Core 4 Test rpmbuild:
-Wp,-D_FORTIFY_SOURCE=2

>From what I gather, FORTIFY_SOURCE is supposed to help protect from certain
types of buffer overflows.  So, perhaps there is a possible buffer overflow
happening (on line 52 of crypto.c), and it just isn't seen until compiled
with the FORTIFY_SOURCE flag?  Here is the backtrace of a core dump cause
when I run 'cfagent -qKv':

(gdb) bt
#0  0x00f037e2 in _dl_sysinfo_int80 () from /lib/ld-linux.so.2
#1  0x0040d17c in raise () from /lib/libc.so.6
#2  0x0040e8c8 in abort () from /lib/libc.so.6
#3  0x00441f2a in __libc_message () from /lib/libc.so.6
#4  0x004c2b65 in __chk_fail () from /lib/libc.so.6
#5  0x004c262f in __vsnprintf_chk () from /lib/libc.so.6
#6  0x004c2523 in __snprintf_chk () from /lib/libc.so.6
#7  0x080a2c51 in RandomSeed () at crypto.c:52
#8  0x0804ed0a in Initialize (argc=2, argv=0xbfb27eb4) at cfagent.c:298
#9  0x0804f2d6 in main (argc=0, argv=0x0) at cfagent.c:83
#10 0x003f9dd6 in __libc_start_main () from /lib/libc.so.6
#11 0x0804b131 in _start ()

Any ideas?  If the FORTIFY_SOURCE flag is not used, I can compile and run
cfagent fine, but I'd like to get this working since it will be the default
for Fedora/Redhat future releases, and there may actually be an overflow
happening there...

Thanks,
Jeff

On 3/18/05 8:39 AM, "Mark Burgess" <address@hidden> wrote:

> On Thu, 2005-03-17 at 11:00 -0800, Jeff Sheltren wrote:
>> On 3/17/05 9:46 AM, "Jeff Sheltren" <address@hidden> wrote:
>> 
>>> I added the attached patch and was able to compile again without errors -
>>> although there were quite a few warnings.  Unfortunately, I'm still getting
>>> a buffer overflow when I run any of the binaries and I'm not sure what is
>>> causing that.  I'll try to do some more debugging.
>>> 
>>> -Jeff
>> 
>> Looks like the buffer overflow is happening on line 52 of crypto.c due to
>> the fact that CF_BUFSIZE isn't available there.  From a gdb run:
>> 
>> Breakpoint 1, RandomSeed () at crypto.c:52
>> (gdb) print CF_BUFSIZE
>> No symbol "CF_BUFSIZE" in current context.
>> (gdb) print CF_AVDB_FILE
>> No symbol "CF_AVDB_FILE" in current context.
>> 
>> AVDB and CFWORKDIR are defined OK.
>> 
>> -Jeff
> 
> 
> The symbols you mention are macro values, so they do not have symbols in
> the debugger. This is ok.
> 
> Mark