Re: [RFC][PATCH 01/13] sched/deadline: Impose global limits on sched_attr::sched_period

[Peter Zijlstra]

On Thu, Aug 22, 2019 at 01:29:49PM +0100, Alessio Balsini wrote:
> Yes, that's exactly what I meant!
> What about this refactoring?

Makes sense; and now I spot a race (and sched_rt_handler() from which I
copied this is susceptible too):

> +int sched_dl_period_handler(struct ctl_table *table, int write,
> +			    void __user *buffer, size_t *lenp,
> +			    loff_t *ppos)
> +{
> +	unsigned int old_max, old_min;
> +	static DEFINE_MUTEX(mutex);
> +	int ret;
> +
> +	mutex_lock(&mutex);
> +	old_max = sysctl_sched_dl_period_max;
> +	old_min = sysctl_sched_dl_period_min;
> +
> +	ret = proc_douintvec(table, write, buffer, lenp, ppos);

Here the sysctl_* values have been changed, interleave with:

> +	if (!ret && write) {
> +		u64 max = (u64)sysctl_sched_dl_period_max * NSEC_PER_USEC;
> +		u64 min = (u64)sysctl_sched_dl_period_min * NSEC_PER_USEC;
> +
> +		if (min < 1ULL << DL_SCALE || max < min) {
> +			sysctl_sched_dl_period_max = old_max;
> +			sysctl_sched_dl_period_min = old_min;
> +			ret = -EINVAL;
> +		}
> +	}
> +	mutex_unlock(&mutex);
> +
> +	return ret;
> +}

> @@ -2675,12 +2713,21 @@ bool __checkparam_dl(const struct sched_attr *attr)
>  	    attr->sched_period & (1ULL << 63))
>  		return false;
>  
> +	period = attr->sched_period;
> +	if (!period)
> +		period = attr->sched_deadline;
> +
>  	/* runtime <= deadline <= period (if period != 0) */
> -	if ((attr->sched_period != 0 &&
> -	     attr->sched_period < attr->sched_deadline) ||
> +	if (period < attr->sched_deadline ||
>  	    attr->sched_deadline < attr->sched_runtime)
>  		return false;
>  
> +	max = (u64)READ_ONCE(sysctl_sched_dl_period_max) * NSEC_PER_USEC;
> +	min = (u64)READ_ONCE(sysctl_sched_dl_period_min) * NSEC_PER_USEC;

this, and we're using unvalidated numbers.

> +	if (period < min || period > max)
> +		return false;
> +
>  	return true;
>  }

Something like the completely untested (again, sorry) below ought to
cure that I think; the same needs doing to sched_rt_handler() I'm
thinking.

---
Subject: sched/deadline: Impose global limits on sched_attr::sched_period
From: Alessio Balsini <balsini@xxxxxxxxxxx>
Date: Thu, 22 Aug 2019 13:29:49 +0100

There are two DoS scenarios with SCHED_DEADLINE related to
sched_attr::sched_period:

 - since access-control only looks at utilization and density, a very
   large period can allow a very large runtime, which in turn can
   incur a very large latency to lower priority tasks.

 - for very short periods we can end up spending more time programming
   the hardware timer than actually running the task.

Mitigate these by imposing limits on the period.

Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
Signed-off-by: Alessio Balsini <balsini@xxxxxxxxxxx>
Cc: rostedt@xxxxxxxxxxx
Cc: mingo@xxxxxxxxxx
Cc: luca.abeni@xxxxxxxxxxxxxxx
Cc: bristot@xxxxxxxxxx
Cc: vpillai@xxxxxxxxxxxxxxxx
Cc: kernel-team@xxxxxxxxxxx
Cc: juri.lelli@xxxxxxxxxx
Cc: dietmar.eggemann@xxxxxxx
Cc: dvyukov@xxxxxxxxxx
Cc: tglx@xxxxxxxxxxxxx
---
 include/linux/sched/sysctl.h |    7 +++++
 kernel/sched/deadline.c      |   58 +++++++++++++++++++++++++++++++++++++++++--
 kernel/sysctl.c              |   14 ++++++++++
 3 files changed, 77 insertions(+), 2 deletions(-)

--- a/include/linux/sched/sysctl.h
+++ b/include/linux/sched/sysctl.h
@@ -56,6 +56,13 @@ int sched_proc_update_handler(struct ctl
 extern unsigned int sysctl_sched_rt_period;
 extern int sysctl_sched_rt_runtime;
 
+extern unsigned int sysctl_sched_dl_period_max;
+extern unsigned int sysctl_sched_dl_period_min;
+
+extern int sched_dl_period_handler(struct ctl_table *table, int write,
+		void __user *buffer, size_t *lenp,
+		loff_t *ppos);
+
 #ifdef CONFIG_UCLAMP_TASK
 extern unsigned int sysctl_sched_uclamp_util_min;
 extern unsigned int sysctl_sched_uclamp_util_max;
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -2633,6 +2633,49 @@ void __getparam_dl(struct task_struct *p
 }
 
 /*
+ * Default limits for DL period: on the top end we guard against small util
+ * tasks still getting ridiculous long effective runtimes, on the bottom end we
+ * guard against timer DoS.
+ */
+unsigned int sysctl_sched_dl_period_max = 1 << 22; /* ~4 seconds */
+unsigned int sysctl_sched_dl_period_min = 100;     /* 100 us */
+
+int sched_dl_period_handler(struct ctl_table *table, int write,
+			    void __user *buffer, size_t *lenp,
+			    loff_t *ppos)
+{
+	unsigned int new_max, new_min;
+	struct ctl_table new_table;
+	static DEFINE_MUTEX(mutex);
+	int ret;
+
+	mutex_lock(&mutex);
+	new_max = sysctl_sched_dl_period_max;
+	new_min = sysctl_sched_dl_period_min;
+	new_table = *table;
+	if (new_table.data == &sysctl_sched_dl_period_max)
+		new_table.data = &new_max;
+	else
+		new_table.data = &new_min;
+
+	ret = proc_douintvec(&new_table, write, buffer, lenp, ppos);
+	if (!ret && write) {
+		u64 max = (u64)new_max * NSEC_PER_USEC;
+		u64 min = (u64)new_min * NSEC_PER_USEC;
+
+		if (min > 1ULL << DL_SCALE && max > min) {
+			WRITE_ONCE(sysctl_sched_dl_period_max, new_max);
+			WRITE_ONCE(sysctl_sched_dl_period_min, new_min);
+		} else {
+			ret = -EINVAL;
+		}
+	}
+	mutex_unlock(&mutex);
+
+	return ret;
+}
+
+/*
  * This function validates the new parameters of a -deadline task.
  * We ask for the deadline not being zero, and greater or equal
  * than the runtime, as well as the period of being zero or
@@ -2644,6 +2687,8 @@ void __getparam_dl(struct task_struct *p
  */
 bool __checkparam_dl(const struct sched_attr *attr)
 {
+	u64 period, max, min;
+
 	/* special dl tasks don't actually use any parameter */
 	if (attr->sched_flags & SCHED_FLAG_SUGOV)
 		return true;
@@ -2667,12 +2712,21 @@ bool __checkparam_dl(const struct sched_
 	    attr->sched_period & (1ULL << 63))
 		return false;
 
+	period = attr->sched_period;
+	if (!period)
+		period = attr->sched_deadline;
+
 	/* runtime <= deadline <= period (if period != 0) */
-	if ((attr->sched_period != 0 &&
-	     attr->sched_period < attr->sched_deadline) ||
+	if (period < attr->sched_deadline ||
 	    attr->sched_deadline < attr->sched_runtime)
 		return false;
 
+	max = (u64)READ_ONCE(sysctl_sched_dl_period_max) * NSEC_PER_USEC;
+	min = (u64)READ_ONCE(sysctl_sched_dl_period_min) * NSEC_PER_USEC;
+
+	if (period < min || period > max)
+		return false;
+
 	return true;
 }
 
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -443,6 +443,20 @@ static struct ctl_table kern_table[] = {
 		.proc_handler	= sched_rt_handler,
 	},
 	{
+		.procname	= "sched_deadline_period_max_us",
+		.data		= &sysctl_sched_dl_period_max,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= sched_dl_period_handler,
+	},
+	{
+		.procname	= "sched_deadline_period_min_us",
+		.data		= &sysctl_sched_dl_period_min,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= sched_dl_period_handler,
+	},
+	{
 		.procname	= "sched_rr_timeslice_ms",
 		.data		= &sysctl_sched_rr_timeslice,
 		.maxlen		= sizeof(int),