On Fri, Dec 17, 2021 at 3:29 PM Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Fri, Dec 17, 2021 at 12:02 PM Paul Moore <paul@xxxxxxxxxxxxxx> wrote:
> >
> > Another small SELinux fix for v5.16 to ensure that we don't block on
> > memory allocations while holding a spinlock. This passes all our
> > tests without problem, please merge this for the next v5.16-rcX
> > release.
>
> Ugh, pulled.
>
> GFP_NOWAIT can very easily fail, so I'm not convinced your tests would
> catch any of the interesting cases.
>
> There is only one single caller of the new
> security_sb_mnt_opts_compat() callback, and I get the feeling that
> maybe we could parse the options first - into a temporary new
> superblock, and then at "test" time (when we're under that sb_lock) it
> could compare that temporary sb with pre-existing ones?
>
> That would also avoid the need for doing that mount option parsing
> over and over and over again for each sb on the 'fs_supers' lists.
>
> I've pulled this, bit it does smell bad to me, and I think that
> original commit 69c4a42d72eb ("lsm,selinux: add new hook to compare
> new mount to an existing mount") and ec1ade6a0448 ("nfs: account for
> selinux security context when deciding to share superblock") may not
> have been fully thought out.
Can you please elaborate on what is problematic with the two patches
you've highlighted. NFS needs a way to determine if the security mount
options have changed between the two mounts in order to determine if
superblock can be shared.
> It may have *looked* like just adding that check to
> 'nfs_compare_super' was a simple and good idea, but it really doesn't
> look great.
>
> Adding a few more people to the cc.
>
> Linus
