Skip to contentSkip to conversation

xen/events: avoid NULL pointer dereference in evtchn_from_irq()

PATCH In review Juergen Gross 3 participants 6 messages Mar 19, 2020 – May 22, 2020

Subsystems (2)
lore
Share & tools
lore

Juergen Gross

author maintainer
#0 lore raw

PATCH 1/1 xen/events: avoid NULL pointer dereference in evtchn_from_irq()

There have been reports of races in evtchn_from_irq() where the info pointer has been NULL.

Avoid that case by testing info before dereferencing it.

In order to avoid accessing a just freed info structure do the kfree() via kfree_rcu().

---
 drivers/xen/events/events_base.c     | 10 ++++++++--
 drivers/xen/events/events_internal.h |  3 +++
 2 files changed, 11 insertions(+), 2 deletions(-)
drivers/xen/events/events_base.c +8 -2
index 499eff7d3f65..838762fe3d6e 100644
@@ -247,10 +247,16 @@ static void xen_irq_info_cleanup(struct irq_info *info)
  */
 unsigned int evtchn_from_irq(unsigned irq)
 {
+	struct irq_info *info;
+
 	if (WARN(irq >= nr_irqs, "Invalid irq %d!\n", irq))
 		return 0;
 
-	return info_for_irq(irq)->evtchn;
+	info = info_for_irq(irq);
+	if (info == NULL)
+		return 0;
+
+	return info->evtchn;
 }
 
 unsigned irq_from_evtchn(unsigned int evtchn)
@@ -436,7 +442,7 @@ static void xen_free_irq(unsigned irq)
 
 	WARN_ON(info->refcnt > 0);
 
-	kfree(info);
+	kfree_rcu(info, rcu);
 
 	/* Legacy IRQ descriptors are managed by the arch. */
 	if (irq < nr_legacy_irqs())
drivers/xen/events/events_internal.h +3 -0
index 82938cff6c7a..c421055843c8 100644
@@ -7,6 +7,8 @@
 #ifndef __EVENTS_INTERNAL_H__
 #define __EVENTS_INTERNAL_H__
 
+#include <linux/rcupdate.h>
+
 /* Interrupt types. */
 enum xen_irq_type {
 	IRQT_UNBOUND = 0,
@@ -30,6 +32,7 @@ enum xen_irq_type {
  */
 struct irq_info {
 	struct list_head list;
+	struct rcu_head rcu;
 	int refcnt;
 	enum xen_irq_type type;	/* type */
 	unsigned irq;
Signature · 1 line
2.16.4
#1 lore raw
Quoted 2 lines · Juergen Gross

There have been reports of races in evtchn_from_irq() where the info pointer has been NULL.

Do you have an example of how this race happens (and how moving kfree to rcu will help)?

(I am mostly wondering whether we'd be masking a more serious problem)

-boris

#2 lore raw
In reply to Juergen Gross
Quoted 7 lines · Juergen Gross

There have been reports of races in evtchn_from_irq() where the info pointer has been NULL.

Avoid that case by testing info before dereferencing it.

In order to avoid accessing a just freed info structure do the kfree() via kfree_rcu().

Looks like noone ever responded to this.

This change looks fine but is there a background on the problem? I looked in the archives and didn't find the relevant discussion.

-boris

Quoted 62 lines
---
 drivers/xen/events/events_base.c     | 10 ++++++++--
 drivers/xen/events/events_internal.h |  3 +++
 2 files changed, 11 insertions(+), 2 deletions(-)
drivers/xen/events/events_base.c +8 -2
index 499eff7d3f65..838762fe3d6e 100644
@@ -247,10 +247,16 @@ static void xen_irq_info_cleanup(struct irq_info *info)
  */
 unsigned int evtchn_from_irq(unsigned irq)
 {
+	struct irq_info *info;
+
 	if (WARN(irq >= nr_irqs, "Invalid irq %d!\n", irq))
 		return 0;
 
-	return info_for_irq(irq)->evtchn;
+	info = info_for_irq(irq);
+	if (info == NULL)
+		return 0;
+
+	return info->evtchn;
 }
 
 unsigned irq_from_evtchn(unsigned int evtchn)
@@ -436,7 +442,7 @@ static void xen_free_irq(unsigned irq)
 
 	WARN_ON(info->refcnt > 0);
 
-	kfree(info);
+	kfree_rcu(info, rcu);
 
 	/* Legacy IRQ descriptors are managed by the arch. */
 	if (irq < nr_legacy_irqs())
drivers/xen/events/events_internal.h +3 -0
index 82938cff6c7a..c421055843c8 100644
@@ -7,6 +7,8 @@
 #ifndef __EVENTS_INTERNAL_H__
 #define __EVENTS_INTERNAL_H__
 
+#include <linux/rcupdate.h>
+
 /* Interrupt types. */
 enum xen_irq_type {
 	IRQT_UNBOUND = 0,
@@ -30,6 +32,7 @@ enum xen_irq_type {
  */
 struct irq_info {
 	struct list_head list;
+	struct rcu_head rcu;
 	int refcnt;
 	enum xen_irq_type type;	/* type */
 	unsigned irq;
#3 lore raw
Quoted 10 lines · Boris Ostrovsky
Quoted 7 lines · Juergen Gross

There have been reports of races in evtchn_from_irq() where the info pointer has been NULL.

Avoid that case by testing info before dereferencing it.

In order to avoid accessing a just freed info structure do the kfree() via kfree_rcu().

Quoted 5 lines

Looks like noone ever responded to this.

This change looks fine but is there a background on the problem? I looked in the archives and didn't find the relevant discussion.

Here is the original bug report:
https://xen.markmail.org/thread/44apwkwzeme4uavo

Signature · 5 lines
Best Regards,
Marek Marczykowski-Górecki
Invisible Things Lab
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?
#4 lore raw
Quoted 10 lines · Marek Marczykowski-Górecki
Quoted 9 lines · Boris Ostrovsky
Quoted 7 lines · Juergen Gross

There have been reports of races in evtchn_from_irq() where the info pointer has been NULL.

Avoid that case by testing info before dereferencing it.

In order to avoid accessing a just freed info structure do the kfree() via kfree_rcu().

Quoted 8 lines
Quoted 5 lines

Looks like noone ever responded to this.

This change looks fine but is there a background on the problem? I looked in the archives and didn't find the relevant discussion.

Here is the original bug report:
https://xen.markmail.org/thread/44apwkwzeme4uavo

Thanks. Do we know what the race is? Is it because an event is being delivered from a dying guest who is in the process of tearing down event channels?

-boris

author maintainer
#5 lore raw
Quoted 16 lines · Boris Ostrovsky
Quoted 15 lines · Marek Marczykowski-Górecki
Quoted 14 lines · Boris Ostrovsky
Quoted 7 lines · Juergen Gross

There have been reports of races in evtchn_from_irq() where the info pointer has been NULL.

Avoid that case by testing info before dereferencing it.

In order to avoid accessing a just freed info structure do the kfree() via kfree_rcu().

Looks like noone ever responded to this.

This change looks fine but is there a background on the problem? I looked in the archives and didn't find the relevant discussion.

Quoted 8 lines
Quoted 3 lines

Here is the original bug report:
https://xen.markmail.org/thread/44apwkwzeme4uavo

Thanks. Do we know what the race is? Is it because an event is being delivered from a dying guest who is in the process of tearing down event channels?

Missing synchronization between event channel (de-)allocation and handling.

I have a patch sitting here, just didn't have the time to do some proper testing and sending it out.

Juergen