[Xen-devel] [PATCH 0/3] libxl: Fix deadlock with pygrub

[Ian Jackson <ian.jackson@xxxxxxxxxxxxx>]

Thanks to Michael Young for reporting this, and my apologies for
introducing this bug.

  1/3 libxl: Fix libxl_postfork_child_noexec deadlock etc.
  2/3 libxl: Hold the atfork lock while closing carefd
  3/3 libxl: Fix carefd lock leak in save callout

Patch 1 is the actual bugfix.  The bug is IMO clearly a blocker for
4.4 so the patch should go in if it is correct.  I have checked that
it seems to fix the problem for me.

Patches 2 and 3 are other theoretical locking bugs I discovered while
looking for the cause of the pygrub deadlock.  I haven't tried to
construct test cases that might exercise these bugs; doing so would be
quite difficult.

The atfork carefd race (patch 2) might be relevant to callers which
are multithreaded and also call libxl_postfork_child_noexec.  libvirt
does not make any such call.  xl is single-threaded.  I haven't
investigated other toolstacks, but the race is theoretical rather than
observed.

The carefd lock leak (patch 3) is obvious but theoretical except in
deeply pathological situations.  I don't think it's worth adding risk
to 4.4 to fix it.

So I think patch 1 should go into 4.4.0 after review.
2-3 should wait for 4.4.1 (coming via unstable in the usual way).


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel