Re: fix to my SEGFAULTs on redhat 7.2

[Mark . Burgess]

Thanks for the tip -- looks like this could be a remnant of
an old code revision...I'll get to this Monday

Mark



On 15 Jun, Ben F wrote:
> I posted a while back about having SEGFAULT problems with
> cfengine (2.0.1 back then) on redhat 7.2 machines. I had managed to
> stop it by not using purge, but recently I needed to turn purge
> back on, and the problem came back in force. 
> [This is with 2.0.2 now]
> 
> If I don't use purge in my /etc copyfiles section, I'm getting SEGFAULT 
> maybe 30% of the time. With purge on, I was getting it 100% of the 
> time at the same place! Which is why I finally had to look at it since
> no one else seems to be having this level of trouble with cfengine. 
> I've been using cfengine for several years FYI.
> 
> So I've ended up going through the code with gdb, and this
> is what I've found. Please note that I haven't done any real
> C programming since the 80's except for the occasional
> porting of things.
> 
> Anyway, the problem seems to be with the global variable
> 'DESTINATION'.
> My crash was occuring in the AbsLinkPath() call in the
> LinkFiles() routine. Evidentally the DESTINATION variable
> is used as a place to hold stuff w/o having to setup
> and the free the space all the time, at least that is what this
> comment leads me to think:
> 
> strcpy(DESTINATION,from);  /* reuse to save stack space */
> 
> DESTINATION is malloc'd in NewParser(),
> and free'd in DeleteParser(). In addition this step is done in
> InitializeAction():
>     *DESTINATION = '\0';
> However, my debugging indicates that the DeleteParser()
> is happening after the various config files are parsed - no
> suprise there - except that DESTINATION is also used in
> the AbsLinkPath() routine farther down the road - so to speak.
> 
> DESTINATION is also used in various routines in install.c.
> 
> 
> As I don't remember much about scope of variables in C, and
> certainly know nothing about pthreads, I just added the following
> lines in link.c:
> 
> diff -r1.1 link.c
> 1179a1180,1182
>> 
>>   DESTINATION = (char *) malloc(bufsize);
>>   bzero(DESTINATION,bufsize);
> 
> Presto chango, no more core dumps. However, I'm pretty sure
> the malloc should be done somewhere more global since 
> DESTINATION is used in other places. And probably the
> routines in parse.c shouldn't be doing stuff with 
> DESTINATION since it is a global.
> 
> Can someone take a look over this and if this seems
> correct?
> 
> thanks,
> Edgar Nielsen
> 
> 
> 
> 
> 
> _______________________________________________
> Bug-cfengine mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/bug-cfengine



~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Work: +47 22453272            Email:  address@hidden
Fax : +47 22453205            WWW  :  http://www.iu.hio.no/~mark
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~