klp_send_signals() and klp_force_transition() do not acquire klp_mutex,
because it seemed to be superfluous. A potential race in
klp_send_signals() was harmless and there was nothing in
klp_force_transition() which needed to be synchronized. That changed
with the addition of klp_forced variable during the review process.
There is a small window now, when klp_complete_transition() does not see
klp_forced set to true while all tasks have been already transitioned to
the target state. module_put() is called and the module can be removed.
Acquire klp_mutex to prevent it. Do the same in klp_send_signals() just
to be sure. There is no real downside to that.
Reported-by: Jason Baron <[email protected]>
Signed-off-by: Miroslav Benes <[email protected]>
---
kernel/livepatch/transition.c | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/kernel/livepatch/transition.c b/kernel/livepatch/transition.c
index be5bfa533ee8..3f932ff607cd 100644
--- a/kernel/livepatch/transition.c
+++ b/kernel/livepatch/transition.c
@@ -625,6 +625,8 @@ void klp_send_signals(void)
pr_notice("signaling remaining tasks\n");
+ mutex_lock(&klp_mutex);
+
read_lock(&tasklist_lock);
for_each_process_thread(g, task) {
if (!klp_patch_pending(task))
@@ -653,6 +655,8 @@ void klp_send_signals(void)
}
}
read_unlock(&tasklist_lock);
+
+ mutex_unlock(&klp_mutex);
}
/*
@@ -671,6 +675,8 @@ void klp_force_transition(void)
pr_warn("forcing remaining tasks to the patched state\n");
+ mutex_lock(&klp_mutex);
+
read_lock(&tasklist_lock);
for_each_process_thread(g, task)
klp_update_patch_state(task);
@@ -680,4 +686,6 @@ void klp_force_transition(void)
klp_update_patch_state(idle_task(cpu));
klp_forced = true;
+
+ mutex_unlock(&klp_mutex);
}
--
2.15.1
On Wed 2017-12-20 10:28:07, Miroslav Benes wrote:
> klp_send_signals() and klp_force_transition() do not acquire klp_mutex,
> because it seemed to be superfluous. A potential race in
> klp_send_signals() was harmless and there was nothing in
> klp_force_transition() which needed to be synchronized. That changed
> with the addition of klp_forced variable during the review process.
>
> There is a small window now, when klp_complete_transition() does not see
> klp_forced set to true while all tasks have been already transitioned to
> the target state. module_put() is called and the module can be removed.
>
> Acquire klp_mutex to prevent it. Do the same in klp_send_signals() just
> to be sure. There is no real downside to that.
>
> Reported-by: Jason Baron <[email protected]>
> Signed-off-by: Miroslav Benes <[email protected]>
> ---
> kernel/livepatch/transition.c | 8 ++++++++
> 1 file changed, 8 insertions(+)
>
> diff --git a/kernel/livepatch/transition.c b/kernel/livepatch/transition.c
> index be5bfa533ee8..3f932ff607cd 100644
> --- a/kernel/livepatch/transition.c
> +++ b/kernel/livepatch/transition.c
> @@ -625,6 +625,8 @@ void klp_send_signals(void)
>
> pr_notice("signaling remaining tasks\n");
>
> + mutex_lock(&klp_mutex);
> +
> read_lock(&tasklist_lock);
> for_each_process_thread(g, task) {
> if (!klp_patch_pending(task))
> @@ -653,6 +655,8 @@ void klp_send_signals(void)
> }
> }
> read_unlock(&tasklist_lock);
> +
> + mutex_unlock(&klp_mutex);
It would be cleaner if the lock guarded also the check:
if (patch != klp_transition_patch)
return -EINVAL;
in signal_store(). Then we could remove also the comment
above this check.
Same is true also for the force part stuff.
Best Regards,
Petr
PS: I am sorry that I hand waved the proposed solution when
we spoke about it yeasterday. I should have looked into
the code.
On Wed, 20 Dec 2017, Petr Mladek wrote:
> On Wed 2017-12-20 10:28:07, Miroslav Benes wrote:
> > klp_send_signals() and klp_force_transition() do not acquire klp_mutex,
> > because it seemed to be superfluous. A potential race in
> > klp_send_signals() was harmless and there was nothing in
> > klp_force_transition() which needed to be synchronized. That changed
> > with the addition of klp_forced variable during the review process.
> >
> > There is a small window now, when klp_complete_transition() does not see
> > klp_forced set to true while all tasks have been already transitioned to
> > the target state. module_put() is called and the module can be removed.
> >
> > Acquire klp_mutex to prevent it. Do the same in klp_send_signals() just
> > to be sure. There is no real downside to that.
> >
> > Reported-by: Jason Baron <[email protected]>
> > Signed-off-by: Miroslav Benes <[email protected]>
> > ---
> > kernel/livepatch/transition.c | 8 ++++++++
> > 1 file changed, 8 insertions(+)
> >
> > diff --git a/kernel/livepatch/transition.c b/kernel/livepatch/transition.c
> > index be5bfa533ee8..3f932ff607cd 100644
> > --- a/kernel/livepatch/transition.c
> > +++ b/kernel/livepatch/transition.c
> > @@ -625,6 +625,8 @@ void klp_send_signals(void)
> >
> > pr_notice("signaling remaining tasks\n");
> >
> > + mutex_lock(&klp_mutex);
> > +
> > read_lock(&tasklist_lock);
> > for_each_process_thread(g, task) {
> > if (!klp_patch_pending(task))
> > @@ -653,6 +655,8 @@ void klp_send_signals(void)
> > }
> > }
> > read_unlock(&tasklist_lock);
> > +
> > + mutex_unlock(&klp_mutex);
>
> It would be cleaner if the lock guarded also the check:
>
> if (patch != klp_transition_patch)
> return -EINVAL;
>
> in signal_store(). Then we could remove also the comment
> above this check.
>
> Same is true also for the force part stuff.
And I even left obsolete comments in sysfs callbacks. Sigh. v2 is
inevitable...
Miroslav