[PATCH 5/5] c/r: Add AF_UNIX support (v7)

Serge E. Hallyn serue at us.ibm.com
Wed Aug 5 06:29:48 PDT 2009

Quoting Dan Smith (danms at us.ibm.com):
> SH> why only free iov_base if ret!=0?
> Because I was diagnosing a crash that only seemed to happen when I
> free()'d the buffer after it was used by sendmsg() and I forgot to
> remove this :(
> >> +	a->sk_peercred.pid = task_tgid_vnr(current);
> >> +	a->sk_peercred.uid = ctx->realcred->uid;
> SH> I don't know how much it matters, but of course root could be
> SH> restarting a set of tasks owned by several non-root uids, and the
> SH> peercred.uid's might need to be something other than
> ctx-> realcred->uid.  Or not?
> Oh, so you're suggesting I use ctx->ecred instead?  I didn't actually
> notice the double declaration in the ckpt_ctx, but I guess that would
> be better.

No, I'm suggesting that the checkpointed application might have had
tasks owned by uids 0, 3, 55, and 1001, and a.peercred.uid might have
been 1001, right?  current, ctx->realcred->uid, and ctx->ecred might
all be different uids.  I think you just need to checkpoint the uid
(eventually an objref to a checkpointed user struct so we can also
catch the user namespace).


More information about the Containers mailing list