[PATCH][xenial SRU] UBUNTU: SAUCE: (namespace) Bypass sget() capability check for nfs

[Andy Whitcroft]

On Fri, Jul 22, 2016 at 09:44:05AM -0500, Seth Forshee wrote:
> BugLink: http://bugs.launchpad.net/bugs/1603719
> 
> 302cabb "UBUNTU: SAUCE: (namespace) Sync with upstream s_user_ns
> patches" added a capability check to sget() which causes a
> regression for automatic submounts, which may happen in the
> context of an unprivileged user. The capability check is not
> necessary in this case.
> 
> The check can be bypassed by using sget_userns() instead.
> init_user_namespace should be used for the user ns since nfs does
> not support unprivileged mounting. This change makes the nfs
> mount behavior in xenial functionally identical to upstream.
> 
> Signed-off-by: Seth Forshee <seth.forshee at canonical.com>
> ---
>  fs/nfs/super.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/nfs/super.c b/fs/nfs/super.c
> index f126828..8a57020 100644
> --- a/fs/nfs/super.c
> +++ b/fs/nfs/super.c
> @@ -2559,7 +2559,8 @@ struct dentry *nfs_fs_mount_common(struct nfs_server *server,
>  			sb_mntdata.mntflags |= MS_SYNCHRONOUS;
>  
>  	/* Get a superblock - note that we may end up sharing one that already exists */
> -	s = sget(nfs_mod->nfs_fs, compare_super, nfs_set_super, flags, &sb_mntdata);
> +	s = sget_userns(nfs_mod->nfs_fs, compare_super, nfs_set_super, flags,
> +			&init_user_ns, &sb_mntdata);
>  	if (IS_ERR(s)) {
>  		mntroot = ERR_CAST(s);
>  		goto out_err_nosb;

Sounds reasonable, good testing.

-apw