Smack: Don't ignore other bprm->unsafe flags if LSM_UNSAFE_PTRACE is set

There is a logic bug in the current smack_bprm_set_creds():
If LSM_UNSAFE_PTRACE is set, but the ptrace state is deemed to be
acceptable (e.g. because the ptracer detached in the meantime), the other
->unsafe flags aren't checked. As far as I can tell, this means that
something like the following could work (but I haven't tested it):

 - task A: create task B with fork()
 - task B: set NO_NEW_PRIVS
 - task B: install a seccomp filter that makes open() return 0 under some
   conditions
 - task B: replace fd 0 with a malicious library
 - task A: attach to task B with PTRACE_ATTACH
 - task B: execve() a file with an SMACK64EXEC extended attribute
 - task A: while task B is still in the middle of execve(), exit (which
   destroys the ptrace relationship)

Make sure that if any flags other than LSM_UNSAFE_PTRACE are set in
bprm->unsafe, we reject the execve().

Cc: stable@vger.kernel.org
Fixes: 5663884caab1 ("Smack: unify all ptrace accesses in the smack")
Signed-off-by: Jann Horn <jannh@google.com>
Signed-off-by: Casey Schaufler <casey@schaufler-ca.com>

diff --git a/security/smack/smack_lsm.c b/security/smack/smack_lsm.c
index 4c5e5a438f8bdd77f88d4cdfdf8424c910c444ed..d8b59480a01c72813a573b2470895761a6ec78c3 100644 (file)
--- a/security/smack/smack_lsm.c
+++ b/security/smack/smack_lsm.c
@@ -937,7 +937,8 @@ static int smack_bprm_set_creds(struct linux_binprm *bprm)
 
                if (rc != 0)
                        return rc;
-       } else if (bprm->unsafe)
+       }
+       if (bprm->unsafe & ~LSM_UNSAFE_PTRACE)
                return -EPERM;
 
        bsp->smk_task = isp->smk_task;