On Thu, Jun 18, 2020 at 11:40 PM Zefan Li [off-list ref] wrote:

On 2020/6/19 5:09, Cong Wang wrote:

quoted

On Thu, Jun 18, 2020 at 12:36 PM Roman Gushchin [off-list ref] wrote:

quoted

On Thu, Jun 18, 2020 at 12:19:13PM -0700, Cong Wang wrote:

quoted

On Wed, Jun 17, 2020 at 6:44 PM Zefan Li [off-list ref] wrote:

quoted

Cc: Roman Gushchin <redacted>

Thanks for fixing this.

On 2020/6/17 2:03, Cong Wang wrote:

quoted

When we clone a socket in sk_clone_lock(), its sk_cgrp_data is
copied, so the cgroup refcnt must be taken too. And, unlike the
sk_alloc() path, sock_update_netprioidx() is not called here.
Therefore, it is safe and necessary to grab the cgroup refcnt
even when cgroup_sk_alloc is disabled.

sk_clone_lock() is in BH context anyway, the in_interrupt()
would terminate this function if called there. And for sk_alloc()
skcd->val is always zero. So it's safe to factor out the code
to make it more readable.

Fixes: 090e28b229af92dc5b ("netprio_cgroup: Fix unlimited memory leak of v2 cgroups")

but I don't think the bug was introduced by this commit, because there
are already calls to cgroup_sk_alloc_disable() in write_priomap() and
write_classid(), which can be triggered by writing to ifpriomap or
classid in cgroupfs. This commit just made it much easier to happen
with systemd invovled.

I think it's 4bfc0bb2c60e2f4c ("bpf: decouple the lifetime of cgroup_bpf from cgroup itself"),
which added cgroup_bpf_get() in cgroup_sk_alloc().

Good point.

I take a deeper look, it looks like commit d979a39d7242e06
is the one to blame, because it is the first commit that began to
hold cgroup refcnt in cgroup_sk_alloc().

I agree, ut seems that the issue is not related to bpf and probably
can be reproduced without CONFIG_CGROUP_BPF. d979a39d7242e06 indeed
seems closer to the origin.

Yeah, I will update the Fixes tag and send V2.

Commit d979a39d7242e06 looks innocent to me. With this commit when cgroup_sk_alloc
is disabled and then a socket is cloned the cgroup refcnt will not be incremented,
but this is fine, because when the socket is to be freed:

 sk_prot_free()
   cgroup_sk_free()
     cgroup_put(sock_cgroup_ptr(skcd)) == cgroup_put(&cgrp_dfl_root.cgrp)

cgroup_put() does nothing for the default root cgroup, so nothing bad will happen.

But skcd->val can be a pointer to a non-root cgroup:

static inline struct cgroup *sock_cgroup_ptr(struct sock_cgroup_data *skcd)
{
#if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
        unsigned long v;

        /*
         * @skcd->val is 64bit but the following is safe on 32bit too as we
         * just need the lower ulong to be written and read atomically.
         */
        v = READ_ONCE(skcd->val);

        if (v & 1)
                return &cgrp_dfl_root.cgrp;

        return (struct cgroup *)(unsigned long)v ?: &cgrp_dfl_root.cgrp;
#else
        return (struct cgroup *)(unsigned long)skcd->val;
#endif
}