Re: kill safety

loic-dev_at_gmx.net
Date: 06/04/05

  • Next message: Simone Russo: "[ANN] A little database" 
    Date: 4 Jun 2005 13:55:23 -0700

    Hello,

    > We have situation like this.
    >
    > Before accessing a directory, a program creates fcntl lock file which
    > contains its pid value.
    >
    > Now another proc which would like to kill the proc which holds fcntl
    > lock is to read pid, try locking the file using fcntl lock and if the
    > fcntl lock fails, kill the proc which holding the lock.

    That's rather an odd way of using fcntl lock...

    > How safe is this operation so that the reading of pid and killing does
    > not result in killing another proc which got this same pid due to
    > recycling of pids.
    >
    > Even though time gap is very short -- reading pid, try locking and if
    > failed, killing; but still how safe even in multi processor situation
    > considering all process involved are root procs?

    >>From a pure theoritical standpoint, this may happen.
    >>From a practical standpoint, it very very very very very likely won't.

    But why Do you need such a feature? What are your trying to achieve?
    Why Do you need a fcntl lock for a process that you kill anyway? [
    Usually the fcntl lock + pid file are used to ensure that only one
    instance of a process is running. ]

    Cheers,
    Loic.

  • Next message: Simone Russo: "[ANN] A little database"