linux-kernel - Re: [RFC] cgroup: Avoid thousands of -Wflex-array-member-not-at-end warnings

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <5fb74444-2fbb-476e-b1bf-3f3e279d0ced@embeddedor.com>
Date: Tue, 2 Sep 2025 14:37:40 +0200
From: "Gustavo A. R. Silva" <gustavo@...eddedor.com>
To: Michal Koutný <mkoutny@...e.com>
Cc: Tejun Heo <tj@...nel.org>, Johannes Weiner <hannes@...xchg.org>,
 cgroups@...r.kernel.org, LKML <linux-kernel@...r.kernel.org>,
 linux-hardening@...r.kernel.org, "Gustavo A. R. Silva"
 <gustavoars@...nel.org>, Chen Ridong <chenridong@...weicloud.com>
Subject: Re: [RFC] cgroup: Avoid thousands of -Wflex-array-member-not-at-end
 warnings

On 9/2/25 13:17, Michal Koutný wrote:
> On Tue, Sep 02, 2025 at 09:56:34AM +0200, "Gustavo A. R. Silva" <gustavo@...eddedor.com> wrote:
>> If the increase in size is not a problem, then something like this
>> works fine (unless there is a problem with moving those two members
>> at the end of cgroup_root?):
> 
> Please don't forget to tackle cgroup_root allocators. IIUC, this move
> towards the end shifts the burden to them.

I don't see how placing the TRAILING_OVERLAP() change at the end
of cgroup_root would cause problems in cgroup_create(). I see
this allocation for `struct cgroup *cgrp`:

cgrp = kzalloc(struct_size(cgrp, ancestors, (level + 1)), GFP_KERNEL);

but I don't see why struct cgroup cgrp; and struct cgroup *cgrp_ancestor_storage;
cannot be placed at the end (as long as they're enclosed in TRAILING_OVERLAP()
of course) of cgroup_root. In the end, it seems you're only interested in
having cgrp->ancestors[0] overlap `cgrp_ancestor_storage` so that the latter
points to the start of the FAM in struct cgroup.

> 
> There's only the rcu_head we care about.

Based on this commit a7fb0423c201 ("cgroup: Move rcu_head up near the
top of cgroup_root"), as long as rcu_head is not after struct cgroup,
all's fine.

However, this tells me that people were aware of the possibility of
`cgrp.ancestors[]` growing even beyond `cgrp_ancestor_storage`, which
is yet another reason not to have that flex array in the middle of
cgroup_root.

> 
> (You seem to be well versed with flex arrays, I was wondering if
> something like this could be rearranged to make it work (assuming the
> union is at the end of its containers):
> 
> 	union {
> 		struct cgroup *ancestors[];
> 		struct {
> 			struct cgroup *_root_ancestor;
> 			struct cgroup *_low_ancestors[];
> 		};
> 	};
> )

Yep, that works (as long as it's always at the very end of any container
or ends last in any nested structs, for instance in struct cgroup_root,
it must also be at the end) for GCC-15+, but for older versions of GCC we
have to use the DECLARE_FLEX_ARRAY() helper as below:

         union {
                 /* All ancestors including self */
                 DECLARE_FLEX_ARRAY(struct cgroup *, ancestors);
                 struct {
                         struct cgroup *_root_ancestor;
                         struct cgroup *_low_ancestors[];
                 };
         };

Thanks
-Gustavo