Re: [PATCH] early init: open /dev/console with O_LARGEFILE

[Dominik Brodowski]

On Wed, Jan 01, 2020 at 01:39:00PM -0800, Linus Torvalds wrote:
> I'm not saying that the revert is wrong at this point,

Linus, I'd like you to revert the patch nonetheless -- it was quite broken,
it is still broken and probably needs, besides your fix, the O_LARGEFILE
patch I sent out a few days ago.

What is more, I think a different approach is saner, leads to smaller code,
and creates far less risks. And maybe Al Viro likes it a bit more as well :)

At

	https://git.kernel.org/pub/scm/linux/kernel/git/brodo/linux.git ksys-next

there are a few proof-of-concept patches for three in-kernel-syscalls:

(1) Both ksys_open() and ksys_unlink() can trivially operate properly on a
kernelspace pointer (by means of calling getname_kernel() instead of
getname()), such as:

-static inline long ksys_unlink(const char __user *pathname)
+/* note: operates on a kernelspace pointer to pathname */
+static inline long ksys_unlink(const char *pathname)
 {
-       return do_unlinkat(AT_FDCWD, getname(pathname));
+       return do_unlinkat(AT_FDCWD, getname_kernel(pathname));
 }

(with all callers of ksys_unlinked() checked, of course).


(2) For ksys_mkdir(), we already have a fully vfs-in-kernelspace variant in
devtmpfs::dev_mkdir(), which we can use for ksys_mkdir() as well.


What do you (and others) think of these alternative approaches? If we should
work in this direction, please revert the patch -- the new patches should go
into v5.6 at earliest, and probably be routed via Al Viro (unless he and/or
you want me to have this tree managed independently, and included in
linux-next separately).

Thanks,
	Dominik