Re: bug in killing processes of user that contain multiple digits

[Bas van der Vlies]

Mark Burgess wrote:
> Bas van der Vlies wrote:
> > Hello,
> >
> >  we are using cfengine 2.1.20.
> >
> > We kill all the processes of user when the job has finished on our 
> > cluster. For some reasons not all processes are killed for some 
> > usernames.
> >
> > example script:
> > #!/usr/sbin/cfagent -K -x -f
> > #
> > # This is the cfengine clean up script for cleaning up login
> > #
> >
> > control:
> >
> >   actionsequence = ( processes )
> >
> > processes:
> >   ".*"
> >         signal=kill
> >         action=do
> >         filter=kill_users
> >
> > filters:
> >
> > { kill_users
> >   Owner: "root|postfix|daemon|ganglia|jaap|willem|bas|walter|bert|ramon"
> >   Result: "!Owner"
> > }
> >
> >
> > I have patches process.c. The diff is included as patch.
> >
> > Regards
> >
> > ------------------------------------------------------------------------
> >
> > --- process.c.org    2005-02-08 12:50:02.000000000 +0100
> > +++ process.c    2006-07-11 12:29:30.243631590 +0200
> > @@ -330,46 +330,34 @@
> >               Debug("Matched proc[%s]\n",ip->name);
> >  
> > -      one_space = false;
> > -             for (sp = ip->name; *sp != '\0'; sp++) /* if first field 
> > contains alpha, skip */
> >           {
> >           while (true)
> >              {
> > -            while (!isdigit((int)*sp) && (*sp != '\0'))
> > -               {
> > -               if (*sp == ' ')
> > -                  {
> > -                  one_space = true;
> > -                  }
> > -               sp++;
> > -               }
> > -
> > -            if (!one_space)
> > -               {
> > -               while(*sp != ' ' && *sp != '\t')
> > +            /* HvB If start with alpha then skip it till the first 
> > space */
> > +        if (isalnum((int)*sp))
> > +           {
> > +           while ((*sp != ' ') && (*sp != '\0'))
> >                    {
> >                    sp++;
> > -                  }
> > -               }
> > -            -            if ((sp > ip->name) && 
> > isalnum((int)*(sp-1))) /* Username contains number*/
> > -               {
> > -               sp++;
> > -               }
> > -            else
> > -               {
> > -               break;
> > -               }
> > -            }
> > +          }
> > +           }
> > +
> > +           /* skip spaces and tabs */
> > +               while ((*sp == ' ') && (*sp == '\t'))
> > +              {
> > +          sp++;
> > +          }
> > +             break;
> > +             } /* end while */
> >              sscanf(sp,"%d",&pid);
> > -         +   
> >           if (pid != -1)
> >              {
> >              break;
> >              }
> > -         }  +         } /* end for */
> >               if (pid == -1)
> >           {
> >   
> > ------------------------------------------------------------------------
> >
> > _______________________________________________
> > Bug-cfengine mailing list
> > address@hidden
> > http://lists.gnu.org/mailman/listinfo/bug-cfengine
> >   
>
> Bas - thanks for the patch. It looks susicously simplistic. HAve you 
> considered the case in which a PID can overap into a user name that 
> contains a number? Can you tell me exactly what the problem was that 
> this patch fixes? Without more info, I do not want to accept the
> patch as I am suspicious of its simplicity.
>
> M

I always i have learned that it must be simplistic ;-). It has todo that 
it finds something the wrong pif if a username contains 8 chars and ends 
with a char i think. Just in time i found an example, because the 
cluster nodes are now cleaned ;-)

--- -d2 ----Applying filter kill_users
FilterProcMatch(UID,USER,root|postfix|daemon|ganglia|jaap|willem|bas|walter|bert|ramon|labraham,<bfdaed80>,<bfdaec60>)
Match 
(root|postfix|daemon|ganglia|jaap|willem|bas|walter|bert|ramon|labraham) 
to (v61u0055)
No match of 
(root|postfix|daemon|ganglia|jaap|willem|bas|walter|bert|ramon|labraham) 
to (v61u0055)
Prepending [PID]
Prepending [PPID]
Prepending [PGID]
Prepending [VSize]
Prepending [RSize]
Prepending [Status]
Prepending [Command]
Prepending [TTime]
Prepending [STime]
Prepending [TTY]
Prepending [Priority]
Prepending [Threads]
Matched proc[v61u0055 18258 18257  0 Jun03 ?        00:00:00 
/usr/local/adf2004.01/bin/adf.exe ib-r6n7.irc.sara.nl 10833 -p4amslave 
-p4yourname ib-r5n3 -p4rmrank 1]
Found matching pid 0
cfengine:ib-r5n3: will not signal or restart processes 0,1,2,3
cfengine:ib-r5n3: occurred while looking for .

That is for all process of this user. There are a loot on this node. The 
patch fixes this problem. Is this enough info

in the cfrc we have a psopt: -ef



--
--
********************************************************************
*                                                                  *
*  Bas van der Vlies                     e-mail: address@hidden      *
*  SARA - Academic Computing Services    phone:  +31 20 592 8012   *
*  Kruislaan 415                         fax:    +31 20 6683167    *
*  1098 SJ Amsterdam                                               *
*                                                                  *
********************************************************************