lkml.org 
[lkml]   [2008]   [Sep]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCH 1/1] system call notification with self_ptrace
On 09/08, Pierre Morel wrote:
>
> --- linux-2.6.26.orig/arch/s390/kernel/signal.c
> +++ linux-2.6.26/arch/s390/kernel/signal.c
> @@ -409,6 +409,11 @@ handle_signal(unsigned long sig, struct
> spin_unlock_irq(&current->sighand->siglock);
> }
>
> + if (current->instrumentation) {
> + clear_thread_flag(TIF_SYSCALL_TRACE);
> + current->instrumentation &= ~PTS_SELF;
> + }
> +
> return ret;
> }

I still think this patch shouldn't change handle_signal().

Once again. The signal handler for SIGSYS can first do
sys_ptrace(PTRACE_SELF_OFF) (which is filtered out), and then use any
other syscall, so this change is not needed, afaics.

The overhead of the additional PTRACE_SELF_OFF syscall is very small,
especially compared to signal delivery. I don't think this functionality
will be widely used, but this change adds the unconditional overhead
to handle_signal().

Btw, the check above looks wrong, shouldn't it be

if (current->instrumentation & PTS_SELF)

?

And. According to the prior discussion, this requires to hook every
signal handler in user space, otherwise we can miss syscall. But every
hook should start with PTRACE_SELF_ON, so I can't see any gain.

> +#define PTS_INSTRUMENTED 0x00000001
> +#define PTS_SELF 0x00000002

I don't really understand why do we need 2 flags, see also below,

> --- linux-2.6.26.orig/kernel/ptrace.c
> +++ linux-2.6.26/kernel/ptrace.c
> @@ -543,6 +543,38 @@ asmlinkage long sys_ptrace(long request,
> * This lock_kernel fixes a subtle race with suid exec
> */
> lock_kernel();
> + if (request == PTRACE_SELF_ON) {
> + task_lock(current);
> + if (current->ptrace) {
> + task_unlock(current);
> + ret = -EPERM;
> + goto out;
> + }
> + set_thread_flag(TIF_SYSCALL_TRACE);
> + current->instrumentation |= PTS_INSTRUMENTED|PTS_SELF;
> + task_unlock(current);
> + ret = 0;
> + goto out;

The code looks strange. How about

if (request == PTRACE_SELF_ON) {
ret = -EPERM;
task_lock(current);
if (!current->ptrace) {
set_thread_flag(TIF_SYSCALL_TRACE);
current->instrumentation |= PTS_INSTRUMENTED|PTS_SELF;
ret = 0;
}
task_unlock(current);
goto out;
}

?

I don't understand how task_lock() can help. This code runs under
lock_kernel(), and without this lock the code is racy anyway.

> + }
> + if (request == PTRACE_SELF_OFF) {
> + task_lock(current);
> + if (current->ptrace) {
> + task_unlock(current);
> + ret = -EPERM;
> + goto out;
> + }
> + clear_thread_flag(TIF_SYSCALL_TRACE);
> + current->instrumentation &= ~PTS_SELF;

So. PTRACE_SELF_OFF doesn't clear PTS_INSTRUMENTED? How can the task
reset ->instrumentation ?
> + if (current->instrumentation) {
> + ret = -EPERM;
> + goto out;
> + }

So, PTRACE_SELF_XXX disables the "normal" ptrace. Not sure this is good.

Oleg.



\
 
 \ /
  Last update: 2008-09-09 14:41    [W:0.095 / U:0.824 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site