[RFC 1/1] Use ioctl selinux callback io_uring commands that implement the ioctl op convention

[Paul Moore]

On Wed, Nov 16, 2022 at 12:49 PM Kanchan Joshi <joshi.k at samsung.com> wrote:
> On Wed, Nov 16, 2022 at 01:50:51PM +0100, Joel Granados wrote:
> >Signed-off-by: Joel Granados <j.granados at samsung.com>
> >---
> > security/selinux/hooks.c | 15 +++++++++++++--
> > 1 file changed, 13 insertions(+), 2 deletions(-)
> >
> >diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
> >index f553c370397e..a3f37ae5a980 100644
> >--- a/security/selinux/hooks.c
> >+++ b/security/selinux/hooks.c
> >@@ -21,6 +21,7 @@
> >  *  Copyright (C) 2016 Mellanox Technologies
> >  */
> >
> >+#include "linux/nvme_ioctl.h"
> > #include <linux/init.h>
> > #include <linux/kd.h>
> > #include <linux/kernel.h>
> >@@ -7005,12 +7006,22 @@ static int selinux_uring_cmd(struct io_uring_cmd *ioucmd)
> >       struct inode *inode = file_inode(file);
> >       struct inode_security_struct *isec = selinux_inode(inode);
> >       struct common_audit_data ad;
> >+      const struct cred *cred = current_cred();
> >
> >       ad.type = LSM_AUDIT_DATA_FILE;
> >       ad.u.file = file;
> >
> >-      return avc_has_perm(&selinux_state, current_sid(), isec->sid,
> >-                          SECCLASS_IO_URING, IO_URING__CMD, &ad);
> >+      switch (ioucmd->cmd_op) {
> >+      case NVME_URING_CMD_IO:
> >+      case NVME_URING_CMD_IO_VEC:
> >+      case NVME_URING_CMD_ADMIN:
> >+      case NVME_URING_CMD_ADMIN_VEC:
>
> We do not have to spell out these opcodes here.
> How about this instead:
>
> +       /*
> +        * nvme uring-cmd continue to follow the ioctl format, so reuse what
> +        * we do for ioctl.
> +        */
> +       if(_IOC_TYPE(ioucmd->cmd_op) == 'N')
> +               return ioctl_has_perm(cred, file, FILE__IOCTL, (u16) ioucmd->cmd_op);
> +       else
> +               return avc_has_perm(&selinux_state, current_sid(), isec->sid,
> +                                   SECCLASS_IO_URING, IO_URING__CMD, &ad);
> +       }
> +
>
> Now, if we write the above fragment this way -
>
> if (__IOC_TYPE(ioucmd->cmd_op) != 0)
>         reuse_what_is_done_for_ioctl;
> else
>         current_check;
>
> That will be bit more generic and can support more opcodes than nvme.
> ublk will continue to fall into else case, but something else (of
> future) may go into the if-part and be as fine-granular as ioctl hook
> has been.
> Although we defined new nvme opcodes to be used with uring-cmd, it is
> also possible that some other provider decides to work with existing
> ioctl-opcode packaged inside uring-cmd and turns it async. It's just
> another implmentation choice.
>
> Not so nice with the above could be that driver-type being 0 seems
> under conflict already. The table in this page:
> https://www.kernel.org/doc/html/latest/userspace-api/ioctl/ioctl-number.html
> But that is first four out of many others. So those four will fall into
> else-part (if ever we get there) and everything else will go into the
> if-part.
>
> Let's see whether Paul considers all this an improvement from what is
> present now.

There are a two things that need consideration:

* The current access control enforces the SELinux io_uring/cmd
permission on all io_uring command passthrough operations, that would
need to be preserved using something we call "policy capabilities".
The quick summary is that policy capabilities are a way for the
SELinux policy to signal to the kernel that it is aware of a breaking
change and the policy is written to take this change into account;
when the kernel loads this newly capable policy it sets a flag which
triggers a different behavior in the kernel.  A simple example can be
found in selinux_file_ioctl(FIONCLEX)/selinux_policycap_ioctl_skip_cloexec(),
but we can talk more about this later if/when we resolve the other
issue.

* As we discussed previously, the real problem is the fact that we are
missing the necessary context in the LSM hook to separate the
different types of command targets.  With traditional ioctls we can
look at the ioctl number and determine both the type of
device/subsystem/etc. as well as the operation being requested; there
is no such information available with the io_uring command
passthrough.  In this sense, the io_uring command passthrough is
actually worse than traditional ioctls from an access control
perspective.  Until we have an easy(ish)[1] way to determine the
io_uring command target type, changes like the one suggested here are
going to be doomed as each target type is free to define their own
io_uring commands.

[1] Yes, one could theoretically make some determination of the target
type by inspecting io_uring_cmd::file::f_op (or similar), but checking
file_operations' function pointers is both a pretty awful layering
violation and downright ugly; I don't want to have to maintain that
long-term in a LSM.

--
paul-moore.com