Re: check device core dump

[Martin Pala]

Thanks for report - bug fixed.


The thing with DeviceInfo_Usage should not be needed, because it is done by:

 if(stat(s->path, &stat_buf) != 0) {
   Event_post(s, EVENT_START, "Event: device '%s' doesn't exist\n", 
s->name);
   return FALSE;
 }

on the beginning of check_device() => it is done for every device 
regardless of space or inode test definition.


Martin

Jan-Henrik Haukeland wrote:

> Martin, I just ran a test (as a non-root user) using the following
> monitrc:
>
> check device cdrom with path /dev/cdrom
>   start "/bin/mount /dev/cdrom"
>   stop  "/bin/umount /dev/cdrom"
>   if space > 1000 MB then alert
>   alert address@hidden
>
> check device disk1 with path /
>   alert address@hidden
>
> Running monit: monit -Iv
>
> Gives the following core dump:
>
>  The service list contains the following entries:
>  
>  Device Name           = cdrom
>  Group                = (not defined)
>  Path                 = /dev/cdrom
>  Monitoring mode      = active
>  Start program        = /bin/mount /dev/cdrom
>  Stop program         = /bin/umount /dev/cdrom
>  
>  Program received signal SIGSEGV, Segmentation fault.
>  [Switching to Thread 1024 (LWP 14146)]
>  0x42080c63 in strlen () from /lib/i686/libc.so.6
>  (gdb) 
>
> The problematic code is in util.c:659:
>
>
>    } else if(dl->resource == RESOURCE_ID_SPACE) {
>
> =>   printf(" %-20s = if %s %ld %s then %s\n",
>           "Space usage limit",
>           operatornames[dl->operator],
>           (dl->limit_absolute > -1)?dl->limit_absolute:dl->limit_percent,
>           (dl->limit_absolute > -1)?"blocks":"%",
>           actionnames[dl->action]);
>
>    }
>
> (gdb) p *dl
> $2 = {resource = 11, operator = 0, limit_absolute = -2147483648, 
>  limit_percent = -1, action = 1000, next = 0x0}
>
>
> The core dump happens because actionnames[1000] points way out of the
> actionnames array. The limit_absolute value looks rotten to. Can you
> please take a look at it?
>
> ------------------------------------------------------------------------------
>
> Also I think that using simply
>
> check device disk1 with path /dev/hda5
>   alert address@hidden
>
> should lead to a call to DeviceInfo_Usage in validate.c:check_device,
> because it's first after this call, when you try to read info from the
> device you will now if the device was mounted or not, the other tests
> in validate.c:check_device almost always will be true since they
> simply test the /dev/device if it exist, which it usually will.
>
> As the parser is now, this function will not be called, unless an
> IF-THEN test was added to the entry.
>
> BTW, I have changed the event to START if monit cannot read from the
> device, so monit will call the entry's start method, and not UNMONITOR
> as I wrongly added before.
>
>