Re: [GIT PULL] percpu fix for v5.9-rc6

From: Arvind Sankar
Date: Fri Sep 18 2020 - 18:40:04 EST


On Fri, Sep 18, 2020 at 02:18:20PM -0700, Linus Torvalds wrote:
> On Fri, Sep 18, 2020 at 2:00 PM Arvind Sankar <nivedita@xxxxxxxxxxxx> wrote:
> >
> > You could just assert that offsetof(typeof(s),flex) == sizeof(s), no?
>
> No, because the whole point is that I want that "sizeof(s)" to *WARN*.
>
> It's a nonsensical thing to do. That 's' has no statically known size.
>
> The C standard is being very confused here, in that it tries to claim
> that the flexible arrays are somehow fundamentally different from a
> zero-sized one. But then it acts as if they are exactly the same wrt
> sizeof() and structure copies.
>
> It should warn, exactly because right now it causes potential bugs
> like the one that started this thread.
>
> You can't have both "zero-sized arrays are bad and shouldn't be used"
> and "flexible arrays are good, and work exactly like zero-sized
> arrays".
>
> Either zero-sized arrays are bad or they aren't. And if they are bad,
> then flexible arrays shouldn't work *exactly* like them apart from
> some UBSAN warnings.
>
> See my point?
>
> Linus

Ouch, offsetof() and sizeof() will give different results in the
presence of alignment padding.

https://godbolt.org/z/rqnxTK

I think, grepping at random, that at least struct scsi_vpd is like this,
size is 24 but data[] starts at offset 20.

struct scsi_vpd {
struct rcu_head rcu;
int len;
unsigned char data[];
};