[PATCH v5] vfs, security: Fix automount superblock LSM init problem, preventing NFS sb sharing

[Paul Moore]

On Thu, Nov 10, 2022 at 6:09 AM David Howells <dhowells at redhat.com> wrote:
>
> When NFS superblocks are created by automounting, their LSM parameters
> aren't set in the fs_context struct prior to sget_fc() being called,
> leading to failure to match existing superblocks.
>
> Fix this by adding a new LSM hook to load fc->security for submount
> creation when alloc_fs_context() is creating the fs_context for it.
>
> However, this uncovers a further bug: nfs_get_root() initialises the
> superblock security manually by calling security_sb_set_mnt_opts() or
> security_sb_clone_mnt_opts() - but then vfs_get_tree() calls
> security_sb_set_mnt_opts(), which can lead to SELinux, at least,
> complaining.
>
> Fix that by adding a flag to the fs_context that suppresses the
> security_sb_set_mnt_opts() call in vfs_get_tree().  This can be set by NFS
> when it sets the LSM context on the new superblock.
>
> The first bug leads to messages like the following appearing in dmesg:
>
>         NFS: Cache volume key already in use (nfs,4.2,2,108,106a8c0,1,,,,100000,100000,2ee,3a98,1d4c,3a98,1)
>
> Changes
> =======
> ver #5)
>  - Removed unused variable.
>  - Only allocate smack_mnt_opts if we're dealing with a submount.
>
> ver #4)
>  - When doing a FOR_SUBMOUNT mount, don't set the root label in SELinux or
>    Smack.
>
> ver #3)
>  - Made LSM parameter extraction dependent on fc->purpose ==
>    FS_CONTEXT_FOR_SUBMOUNT.  Shouldn't happen on FOR_RECONFIGURE.
>
> ver #2)
>  - Added Smack support
>  - Made LSM parameter extraction dependent on reference != NULL.
>
> Signed-off-by: David Howells <dhowells at redhat.com>
> Fixes: 9bc61ab18b1d ("vfs: Introduce fs_context, switch vfs_kern_mount() to it.")
> Fixes: 779df6a5480f ("NFS: Ensure security label is set for root inode)
> Tested-by: Jeff Layton <jlayton at kernel.org>
> Reviewed-by: Jeff Layton <jlayton at kernel.org>
> Acked-by: Casey Schaufler <casey at schaufler-ca.com>
> Acked-by: Christian Brauner (Microsoft) <brauner at kernel.org>
> cc: Trond Myklebust <trond.myklebust at hammerspace.com>
> cc: Anna Schumaker <anna at kernel.org>
> cc: Alexander Viro <viro at zeniv.linux.org.uk>
> cc: Scott Mayhew <smayhew at redhat.com>
> cc: Jeff Layton <jlayton at kernel.org>
> cc: Paul Moore <paul at paul-moore.com>
> cc: linux-nfs at vger.kernel.org
> cc: selinux at vger.kernel.org
> cc: linux-security-module at vger.kernel.org
> cc: linux-fsdevel at vger.kernel.org
> Link: https://lore.kernel.org/r/165962680944.3334508.6610023900349142034.stgit@warthog.procyon.org.uk/ # v1
> Link: https://lore.kernel.org/r/165962729225.3357250.14350728846471527137.stgit@warthog.procyon.org.uk/ # v2
> Link: https://lore.kernel.org/r/165970659095.2812394.6868894171102318796.stgit@warthog.procyon.org.uk/ # v3
> Link: https://lore.kernel.org/r/166133579016.3678898.6283195019480567275.stgit@warthog.procyon.org.uk/ # v4
> Link: https://lore.kernel.org/r/217595.1662033775@warthog.procyon.org.uk/ # v5
> ---
>
>  fs/fs_context.c               |    4 +++
>  fs/nfs/getroot.c              |    1 +
>  fs/super.c                    |   10 +++++---
>  include/linux/fs_context.h    |    1 +
>  include/linux/lsm_hook_defs.h |    1 +
>  include/linux/lsm_hooks.h     |    6 ++++-
>  include/linux/security.h      |    6 +++++
>  security/security.c           |    5 ++++
>  security/selinux/hooks.c      |   25 +++++++++++++++++++
>  security/smack/smack_lsm.c    |   54 +++++++++++++++++++++++++++++++++++++++++
>  10 files changed, 108 insertions(+), 5 deletions(-)
>
> diff --git a/fs/fs_context.c b/fs/fs_context.c
> index 24ce12f0db32..22248b8a88a8 100644
> --- a/fs/fs_context.c
> +++ b/fs/fs_context.c
> @@ -282,6 +282,10 @@ static struct fs_context *alloc_fs_context(struct file_system_type *fs_type,
>                 break;
>         }
>
> +       ret = security_fs_context_init(fc, reference);
> +       if (ret < 0)
> +               goto err_fc;
> +
>         /* TODO: Make all filesystems support this unconditionally */
>         init_fs_context = fc->fs_type->init_fs_context;
>         if (!init_fs_context)
> diff --git a/fs/nfs/getroot.c b/fs/nfs/getroot.c
> index 11ff2b2e060f..651bffb0067e 100644
> --- a/fs/nfs/getroot.c
> +++ b/fs/nfs/getroot.c
> @@ -144,6 +144,7 @@ int nfs_get_root(struct super_block *s, struct fs_context *fc)
>         }
>         if (error)
>                 goto error_splat_root;
> +       fc->lsm_set = true;
>         if (server->caps & NFS_CAP_SECURITY_LABEL &&
>                 !(kflags_out & SECURITY_LSM_NATIVE_LABELS))
>                 server->caps &= ~NFS_CAP_SECURITY_LABEL;
> diff --git a/fs/super.c b/fs/super.c
> index 8d39e4f11cfa..f200ae0549ca 100644
> --- a/fs/super.c
> +++ b/fs/super.c
> @@ -1553,10 +1553,12 @@ int vfs_get_tree(struct fs_context *fc)
>         smp_wmb();
>         sb->s_flags |= SB_BORN;
>
> -       error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
> -       if (unlikely(error)) {
> -               fc_drop_locked(fc);
> -               return error;
> +       if (!(fc->lsm_set)) {
> +               error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
> +               if (unlikely(error)) {
> +                       fc_drop_locked(fc);
> +                       return error;
> +               }
>         }

Thinking about all the different things that an LSM could do, would it
ever be possible that a LSM would want the security_sb_set_mnt_opts()
call to happen here?  I'm wondering if we are better off leaving it up
to the LSM by passing the fs_context in the security_sb_set_mnt_opts()
hook; those that want to effectively skip this call due to a submount
setup already done in security_fs_context_init() can check the
fs_context::purpose value in the security_sb_set_mnt_opts() hook.

Thoughts?

-- 
paul-moore.com