]> git.ipfire.org Git - thirdparty/kernel/stable.git/commitdiff
audit: Syscall rules are not applied to existing processes on non-x86
authorAnton Blanchard <anton@samba.org>
Tue, 8 Jan 2013 23:46:17 +0000 (10:46 +1100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 19 May 2013 18:38:24 +0000 (11:38 -0700)
commit cdee3904b4ce7c03d1013ed6dd704b43ae7fc2e9 upstream.

Commit b05d8447e782 (audit: inline audit_syscall_entry to reduce
burden on archs) changed audit_syscall_entry to check for a dummy
context before calling __audit_syscall_entry. Unfortunately the dummy
context state is maintained in __audit_syscall_entry so once set it
never gets cleared, even if the audit rules change.

As a result, if there are no auditing rules when a process starts
then it will never be subject to any rules added later. x86 doesn't
see this because it has an assembly fast path that calls directly into
__audit_syscall_entry.

I noticed this issue when working on audit performance optimisations.
I wrote a set of simple test cases available at:

http://ozlabs.org/~anton/junkcode/audit_tests.tar.gz

02_new_rule.py fails without the patch and passes with it. The
test case clears all rules, starts a process, adds a rule then
verifies the process produces a syscall audit record.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Eric Paris <eparis@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
include/linux/audit.h

index 5a6d718adf34825eb11bc4dd20cbcb8f1dced0ad..37464c592c92819ec3305138dc57cbedf522c545 100644 (file)
@@ -120,7 +120,7 @@ static inline void audit_syscall_entry(int arch, int major, unsigned long a0,
                                       unsigned long a1, unsigned long a2,
                                       unsigned long a3)
 {
-       if (unlikely(!audit_dummy_context()))
+       if (unlikely(current->audit_context))
                __audit_syscall_entry(arch, major, a0, a1, a2, a3);
 }
 static inline void audit_syscall_exit(void *pt_regs)