RE: [PATCH] can: ucan: fix alignment constraints

[David Laight <David.Laight@ACULAB.COM>]

From: Marc Kleine-Budde
> Sent: 08 February 2021 13:16
> 
> On 04.02.2021 17:26:13, Arnd Bergmann wrote:
> > From: Arnd Bergmann <arnd@arndb.de>
> >
> > struct ucan_message_in contains member with 4-byte alignment
> > but is itself marked as unaligned, which triggers a warning:
> >
> > drivers/net/can/usb/ucan.c:249:1: warning: alignment 1 of 'struct ucan_message_in' is less than 4 [-
> Wpacked-not-aligned]
> >
> > Mark the outer structure to have the same alignment as the inner
> > one.
> >
> > Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> 
> Applied to linux-can-next/testing.

I've just had a look at that file.

Are any of the __packed or __aligned actually required at all.

AFAICT there is one structure that would have end-padding.
But I didn't actually spot anything validating it's length.
Which may well mean that it is possible to read off the end
of the USB receive buffer - plausibly into unmapped addresses.

I looked because all the patches to 'fix' the compiler warning
are dubious.
Once you've changed the outer alignment (eg of a union) then
the compiler will assume that any pointer to that union is aligned.
So any _packed() attributes that are required for mis-aligned
accesses get ignored - because the compiler knows the pointer
must be aligned.

So while the changes remove the warning, they may be removing
support for misaligned addresses.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)