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

[Peter Zijlstra]

On Fri, Aug 02, 2019 at 06:21:04PM +0100, Alessio Balsini wrote:
> Hi Peter,
> 
> This is indeed an important missing piece.
> 
> I think it would be worth having some simple checks, e.g.:
> - period_min <= period_max;
> - period_min >= (1ULL << DL_SCALE).
> 
> To be even more picky, I'm wondering if it would be a good practice to
> fail the write operation if there are already dl-tasks in the system
> that violate the new constraint.  I'm afraid there is no cheap way of
> achieving such check, so, I think we can skip this last tricky thing for
> now.

Like so?

---
Subject: sched/deadline: Impose global limits on sched_attr::sched_period
From: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Date: Tue Jul 23 16:01:29 CEST 2019

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>
Cc: Daniel Bristot de Oliveira <bristot@xxxxxxxxxx>
Cc: Luca Abeni <luca.abeni@xxxxxxxxxxxxxxx>
Cc: Juri Lelli <juri.lelli@xxxxxxxxxx>
Cc: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
---
 include/linux/sched/sysctl.h |    7 ++++
 kernel/sched/deadline.c      |   61 +++++++++++++++++++++++++++++++++++++++++--
 kernel/sysctl.c              |   14 +++++++++
 3 files changed, 80 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
@@ -2660,6 +2660,52 @@ 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 old_min, old_max;
+	static DEFINE_MUTEX(mutex);
+	int ret;
+
+	mutex_lock(&mutex);
+	old_min = sysctl_sched_dl_period_min;
+	old_max = sysctl_sched_dl_period_max;
+
+	ret = proc_dointvec(table, write, buffer, lenp, ppos);
+
+	if (!ret && write) {
+		u64 min = (u64)sysctl_sched_dl_period_min * NSEC_PER_USEC;
+		u64 max = (u64)sysctl_sched_dl_period_man * NSEC_PER_USEC;
+
+		ret = -EINVAL;
+		if (min < 1ULL << DL_SCALE)
+			goto undo;
+
+		if (max < min)
+			goto undo;
+
+		ret = 0;
+	}
+
+	if (0) {
+undo:
+		sysctl_sched_rt_period = old_period;
+		sysctl_sched_rt_runtime = old_runtime;
+	}
+	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
@@ -2671,6 +2717,8 @@ 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;
@@ -2694,12 +2742,21 @@ 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),