Re: [PATCH] cgroup: Fix crash with CLONE_INTO_CGROUP and v1 cgroups

From: Martin KaFai Lau <martin.lau@linux.dev>
To: Yosry Ahmed <yosryahmed@google.com>
Cc: Christian Brauner <brauner@kernel.org>, Tejun Heo <tj@kernel.org>,
	syzbot <syzbot+534ee3d24c37c411f37f@syzkaller.appspotmail.com>,
	gregkh@linuxfoundation.org,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	syzkaller-bugs@googlegroups.com,
	"Christian A. Ehrhardt" <lk@c--e.de>,
	Andrii Nakryiko <andrii@kernel.org>,
	Alexei Starovoitov <ast@kernel.org>, bpf <bpf@vger.kernel.org>
Subject: Re: [PATCH] cgroup: Fix crash with CLONE_INTO_CGROUP and v1 cgroups
Date: Mon, 10 Oct 2022 11:38:39 -0700	[thread overview]
Message-ID: <5abb2728-9867-175c-911c-b0bb49b8babd@linux.dev> (raw)
In-Reply-To: <CAJD7tkaQSMSmrb3Nt17-NPAPkvEoUp5tJBg8e4UYn0eU6x-Gqw@mail.gmail.com>

On 10/9/22 11:42 AM, Yosry Ahmed wrote:
> On Sun, Oct 9, 2022 at 6:10 AM Christian A. Ehrhardt <lk@c--e.de> wrote:
>>
>>
>> Since commit f3a2aebdd6, Version 1 cgroups no longer cause an
>> error when used with CLONE_INTO_CGROUP. However, the permission
>> checks performed during clone assume a Version 2 cgroup.
>>
>> Restore the error check for V1 cgroups in the clone() path.
>>
>> Reported-by: syzbot+534ee3d24c37c411f37f@syzkaller.appspotmail.com
>> Link: https://lore.kernel.org/lkml/000000000000385cbf05ea3f1862@google.com/
>> Fixes: f3a2aebdd6 ("cgroup: enable cgroup_get_from_file() on cgroup1")
> 
> Thanks for fixing this, and sorry if this caused a mess.
> 
> cgroup_get_from_file() independently seemed like it can support
> cgroup1, I didn't realize that some of the callers depend on the fact
> that it only supports cgroup2.
> 
> +Andrii Nakryiko +Alexei Starovoitov +Martin KaFai Lau +bpf
> I wonder if BPF users have this dependency. Does cgroup_bpf_attach()
> also depend on cgroup_get_from_fd() (which calls
> cgroup_get_from_file()) eliminating v1 cgroups?

Yes, cgroup_bpf_{prog,link}_attach() depends on cgroup_get_from_fd() only 
returning v2 cgroup.  Thus, it needs a fix to get back this filtering after 
commit f3a2aebdd6.

> 
> It seems like cgroup storages (and some other places) use cgroup ids.
> Collisions can happen in cgroup1 ids so I am assuming we want to add a
> check there as well. Perhaps in cgroup_bpf_attach() ?

 From a quick look, cgroup storage should be fine.  The insertion (where the 
cgrp is required for the key purpose) has already passed the attach filtering.

Where 'other places' might have problem with the cgroup id?

> 
> I can send a patch for this if that's the case.
> 
>> Signed-off-by: Christian A. Ehrhardt <lk@c--e.de>
>> ---
>>   kernel/cgroup/cgroup.c | 5 +++++
>>   1 file changed, 5 insertions(+)
>>
>> diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c
>> index b6e3110b3ea7..f7fc3afa88c1 100644
>> --- a/kernel/cgroup/cgroup.c
>> +++ b/kernel/cgroup/cgroup.c
>> @@ -6244,6 +6244,11 @@ static int cgroup_css_set_fork(struct kernel_clone_args *kargs)
>>                  goto err;
>>          }
>>
>> +       if (!cgroup_on_dfl(dst_cgrp)) {
>> +               ret = -EBADF;
>> +               goto err;
>> +       }
>> +
>>          if (cgroup_is_dead(dst_cgrp)) {
>>                  ret = -ENODEV;
>>                  goto err;
>> --
>> 2.34.1
>>