Re: [PATCH 03/11] fs: add new read_uptr and write_uptr file operations

[Linus Torvalds]

On Mon, Jun 29, 2020 at 8:29 AM Christoph Hellwig <hch@xxxxxx> wrote:
>
> So based on that I'd rather get away without our flag and tag the
> kernel pointer case in setsockopt explicitly.

Yeah, I'd be ok to pass that kind of flag around for setsockopt, in
ways I _don't_ want to do for some very core vfs thing like 'read()'.

That said, is there no practical limit on how big "optlen" can be?
Sure, I realize that a lot of setsockopt users may not use all of the
data, but let's say that "optlen" is 128, but the actual low-level
setsockopt operation only uses the first 16 bytes, maybe we could
always just copy the 128 bytes from user space into kernel space, and
just say "setsockopt() always gets a kernel pointer".

Then the bpf use is even simpler. It would just pass the kernel
pointer natively.

Because that seems to be what the BPF code really wants to do: it
takes the user optval, and munges it into a kernel optval, and then
(if that has been done) runs the low-level sock_setsockopt() under
KERNEL_DS.

Couldn't we switch things around instead, and just *always* copy
things from user space, and sock_setsockopt (and
sock->ops->setsockopt) _always_ get a kernel buffer?

And avoid the set_fs(KERNEL_DS) games entirely that way?

Attached it a RFC patch just for __sys_setsockopt() - note that it
does *not* change all the low-level setsockopt callers to just do the
kernel access instead, so this is completely broken, but you can kind
of see what I mean.

Wouldn't this work? In fact, wouldn't this simplify all the setsockopt
places that now don't need to do "get_user()" etc any more?

It would be better if we could limit "optlen" to something sane, but
right now it just does a kmalloc() of whatever the user claims the opt
len is..

                    Linus

Attachment: patch
Description: Binary data