[BProc] Announce: BProc 3.1.6

[Erik A. H. <er...@he...>]

BProc 3.1.6 is up on sourceforge.

http://sourceforge.net/project/showfiles.php?group_id=24453&release_id=72421

3.1.6  ---------------------------------------------------------------------

Transparent Remote Exec

  This release adds a facility to allow sys_execve on slave nodes to
  transparently use their ghost to provide a binary image.
  Essentially, this looks like bproc_rexec(-1, ...) followed by
  bproc_move(original_slave) except that the process never really
  leaves the slave and therefore maintains stuff like its current
  working directory and open files.

  This is a big step in being able to do script-like things on slave
  nodes.  Shells don't quite work like they do on the front end since
  it's not possible to walk a path and find a binary before exec()ing
  it.  Anything with a full path name on the binary works though.

Hot Reconnect for Slaves

  The hot reconnect feature allows slaves to reconnect to the front
  end without resetting their state.  This is *not* a fail-over
  feature.  The old connection needs to be shutdown cleanly to avoid
  loss of messages in flight.  Hot reconnect is intended to allow
  slaves to switch networks after startup.

Bugs

  There's also the usual round of bug fixes.  See the change log for
  more info on those.


Changes from 3.1.5 to 3.1.6

        * Added bproc_execve which allows processes on slave nodes to
        perform execve on the front end machine using their ghost process.

        * Added a hook in sys_execve to transparently use the ghost for
        execve if the local execve fails.  This can behavior can be
        switched on and off via /proc/sys/bproc/execve_hook.

        * Updated kernel patch for new features.  Included patch is
        against 2.4.17.
        
        * Added hot reconnect for slave daemons.  The slave can now
        reconnect to the master at runtime w/o affecting the slave's
        state.  (bpctl --reconnect)

        * Added another work-around for another Linux TCP bug.  TCP sure
        does seem awfully broken lately.

        * Added "async" versions of the cache mangement calls.

        * Fixed bproc.o so that sysctl table registration can fail w/o
        causing insmod to fail.  This is incase the kernel doesn't support
        sysctl.

        * Fixed master daemon bug which resulted in failure to note a
        process's new location during a move and resulted in message loops
        between the kernel and the master daemon.

        * Fixed a security hole involving ptrace and setuid binaries on
        slave nodes.

        * Fixed a security that would cause the master default to allowing
        connections from unreserved ports.
        
        * Fixed BProc's CLONE_PARENT, CLONE_THREAD, CLONE_PTRACE handling
        to some extent.  It's still got a few race conditions but at least
        it does something approaching "correct" at this point.

        * Reworked VMADump to reduce its foot print on the caller's
        kernel stack.  Kernel stack overflows have been observed here.