This applies on top of yesterday's patchset. thanks, -serge >From f70a08dc6abdba372e1b3e1f25aadb919d07b171 Mon Sep 17 00:00:00 2001 From: Serge E. Hallyn <serue@xxxxxxxxxx> Date: Tue, 12 May 2009 08:17:34 -0400 Subject: [PATCH 1/1] cr: child user namespace must grab ref to creator With this change plus Oren's proposed fix to obj_put() after obj_new() during restore_obj(), refcounting on user_ns and user_struct should be completely correct. Signed-off-by: Serge E. Hallyn <serue@xxxxxxxxxx> --- checkpoint/process.c | 24 +++--------------------- 1 files changed, 3 insertions(+), 21 deletions(-) diff --git a/checkpoint/process.c b/checkpoint/process.c index a469f46..a8cd674 100644 --- a/checkpoint/process.c +++ b/checkpoint/process.c @@ -569,27 +569,9 @@ static struct user_namespace *restore_read_userns(struct ckpt_ctx *ctx) if (IS_ERR(ns)) return ns; - /* we need a way to keep track of the new_root just - * until we alloc the uid inthe userns which we - * actually want. Then we can do: - * if (uid == 0) - * new_user = new_root; - * else - * new_user = alloc_uid(ns, uid); - * free_uid(new_root); - * cred->user = new_user; - * This is because new_root is right now the only - * thing pinning the user_ns. - * BUT I don't think I can just add it to the - * objhash, bc then we use up an objref which we'll - * need for the next real objhash object, right? - * I suppose I could just add them to the top of - * the objref space :) (MAX_INT-1) - * - * For now, this code is just plain wrong bc it will - * leak the user_ns and its root_user when the task - * exits. But, a leak is better than an OOPS... - */ + /* new_user_ns() doesn't bump creator's refcount */ + get_uid(creator); + return ns; } -- 1.6.1 _______________________________________________ Containers mailing list Containers@xxxxxxxxxxxxxxxxxxxxxxxxxx https://lists.linux-foundation.org/mailman/listinfo/containers
