On Wed, Feb 28, 2024 at 04:56:09PM -0800, Jakub Kicinski wrote:
> On Wed, 28 Feb 2024 16:01:49 -0800 Kees Cook wrote:
> > So, I found several cases where struct net_device is included in the
> > middle of another structure, which makes my proposal more awkward. But I
> > also don't understand why it's in the _middle_. Shouldn't it always be
> > at the beginning (with priv stuff following it?)
> > Quick search and examined manually: git grep 'struct net_device [a-z0-9_]*;'
> >
> > struct rtw89_dev
> > struct ath10k
> > etc.
>
> Ugh, yes, the (ab)use of NAPI.
>
> > Some even have two included (?)
>
> And some seem to be cargo-culted from out-of-tree code and are unused :S
Ah, which can I remove?
> That's... less pretty. I'd rather push the ugly into the questionable
> users. Make them either allocate the netdev dynamically and store
> a pointer, or move the netdev to the end of the struct.
>
> But yeah, that's a bit of a cleanup :(
So far, it's not too bad. I'm doing some test builds now.
As a further aside, this code:
struct net_device *dev;
...
struct net_device *p;
...
/* ensure 32-byte alignment of whole construct */
alloc_size += NETDEV_ALIGN - 1;
p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
...
dev = PTR_ALIGN(p, NETDEV_ALIGN);
Really screams for a dynamic-sized (bucketed) kmem_cache_alloc
API. Alignment constraints can be described in a regular kmem_cache
allocator (rather than this open-coded version). I've been intending to
build that for struct msg_msg for a while now, and here's another user. :)
-Kees
--
Kees Cook
