Currently ppid filtering on syscall auditing does not appear to work. An
easy reproducer would be to do the following:
touch ./test
auditctl -a entry,always -S chmod -F ppid=[pid of your shell]
chmod 000 ./test
no audit record will appear! (although !=[pid of your shell] will show
all chmod commands from all processes regardless of the ppid)
With a little instrumentation I found that ctx->ppid == 0 inside
audit_filter_rules(). I originally wanted to set the ppid during the
context creation back in something like audit_alloc_context but that
didn't work. Because at that point the new process had not forked off
so the ppid of the chmod process was actually it's parents parents.
Instead I set the ppid in audit_syscall_entry when we are actually
building the specific context.
After some looking I did not see a way to get into audit_log_exit
without having set the ppid. So I am dropping the set from there and
only doing it at the beginning.
Please comment/ack/nak as soon as possible.
-Eric
kernel/auditsc.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
--- linux-2.6.18.i686/kernel/auditsc.c.orig 2006-09-27 21:53:44.000000000 -0400
+++ linux-2.6.18.i686/kernel/auditsc.c 2006-09-28 15:51:44.000000000 -0400
@@ -795,7 +795,6 @@ static void audit_log_exit(struct audit_
/* tsk == current */
context->pid = tsk->pid;
- context->ppid = sys_getppid(); /* sic. tsk == current in all cases */
context->uid = tsk->uid;
context->gid = tsk->gid;
context->euid = tsk->euid;
@@ -1116,6 +1115,7 @@ void audit_syscall_entry(int arch, int m
context->arch = arch;
context->major = major;
+ context->ppid = sys_getppid();
context->argv[0] = a1;
context->argv[1] = a2;
context->argv[2] = a3;