[PATCH] c/r: Add AF_UNIX support (v3)

Oren Laadan orenl at cs.columbia.edu
Tue Jul 7 08:48:20 PDT 2009

Dan Smith wrote:
> OL> Ahh.. and forgot to ask/mention: you do need to call
> OL> sock_unix_unlink() before attempting bind(), for the reasons we
> OL> had discussed earlier (consider same example as above,
> OL> checkpoint/restart done before the unlink(), then restart will
> OL> otherwise fail).
> I thought we agreed that was userspace's job?  That's why I didn't
> unlink() before bind() in this version (4) of the patch either.

I don't recall such a conclusion.

I argued that it it's kernel's job. I suppose we agree that such
pathnames should always be unlinked - so it isn't policy based.
Therefore, I don't see a good reason to have userspace scan the
checkpoint image data just to find such sockets and make them
disappear ...


More information about the Containers mailing list