On Tue, Dec 3, 2019 at 12:59 AM Darrick J. Wong <darrick.wong@xxxxxxxxxx> wrote:
> On Mon, Dec 02, 2019 at 03:22:31PM -0800, Linus Torvalds wrote:
> Yeah, that looks correct to me. Stephen's solution backed out the
> changes that Arnd made for the !x86_64 compat ioctl case, so I or
> someone would have had to re-apply them.
>
> > There was some other minor difference too, but it's also possible I
> > could have messed up, so cc'ing Stephen and Arnd on this just in case
> > they have comments.
>
> <nod> Thanks for sorting this out.
Yes, this is the right solution. Note that this part of the series originally
came from Al Viro, but I added more patches following the same idea.
While working on a follow-up series, I now also noticed that the
FS_IOC_RESVSP compat handler has always been broken for
x32 user space. I don't think anyone cares, but my series in [1]
addresses this as well by handling FS_IOC_RESVSP/
FS_IOC_UNRESVSP/FS_IOC_ZERO_RANGE on x86-64 as
well, in addition to FS_IOC_RESVSP_32 etc.
Arnd
[1] https://git.kernel.org/pub/scm/linux/kernel/git/arnd/playground.git
compat-ioctl-endgame
