Re: [PATCH] add lock hook to prevent race

Previous message: Chris Wright: "Re: [PATCH] add lock hook to prevent race"
Maybe in reply to: Antony Edwards: "[PATCH] add lock hook to prevent race"
Next in thread: Chris Wright: "Re: [PATCH] add lock hook to prevent race"
Reply: Chris Wright: "Re: [PATCH] add lock hook to prevent race"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

aedwardat_private

Hi,

> I realize that copy_from_user can sleep in the fucntions
> fcntl_[gs]etlk(64), but it's not clear to me what could
> be raceable if that (kernel) interface was changed to
> using the already looked up filp instead of the fd.
> any ideas?

If fcntl_*lck* was passed the already looked up filp (as
the other handlers in do_fcntl are) then I don't think
there is any race. The filp has a non-zero ref-count and
is passed on the stack, so there's no way another thread
can mess around with any relationships.

I think changing the kernel interface is a much cleaner
solution (better to prevent the race than detect it),
and would be happy to provide a patch to do it. I just
didn't want to change any base kernel interfaces in
my original patch for acceptance.

BTW, looking around this morning I found a bug in my
patch. When the lock hooks fail they should "goto
out_putf", not "goto out". I've attached the new patch.

Antony

(See attached file: lsm-2.5.2-locks-3.patch)