Re: [PATCH] cgroup: reorder flexible array members of struct cgroup_root

From: Tejun Heo
Date: Thu Oct 26 2017 - 10:32:32 EST


Hello,

On Wed, Oct 25, 2017 at 02:54:23PM -0700, Matthias Kaehlcke wrote:
> From your earlier comment I understand that there is no problem in
> this case because we know that cgroup_root->cgrp will always be
> empty.
>
> However in other instances the warning could point out actual errors
> in the code, so I think it is good to have this warning generally
> enabled. If cgroup_root was defined in a .c file we could consider to
> disable the warning locally, but since the definition is in a header
> that is widely included (indirectly through linux/cgroup.h and
> net/sock.h) this doesn't seem to be an option.
>
> Is there a good reason for the current position of cgrp within
> cgroup_root? If there are no drawbacks in moving it to the end of
> the struct I think Nick's patch is a reasonable solution.

This all sounds really bogus to me. Let's say we have something like
the following.

struct flex_struct {
int array[];
};

And the following two usages.

1.

struct flex_struct *fs =
kmalloc(sizeof(struct flex_struct) + N * sizeof(int));

2.

struct enclosing_struct es {
struct flex_struct fs;
int fs_array_storage[N];
};

struct enclosing_struct *es =
kmalloc(sizeof(struct enclosing_struct));

So, you're saying #1 is okay but #2 is not, which is just silly. The
compiler can't warn correctly about flex array members whether they're
embedded or not. Nothing prevents somebody accessing beyond N in #1
either.

This effort seems really pointless to me. Let's please not waste any
more bandwidth on this.

Thanks.

--
tejun