[PATCH] perf/core: fix possible spectre-v1 write

[PATCH] perf/core: fix possible spectre-v1 write

[Mark Rutland]

It's possible for userspace to control event_id. Sanitize event_id when
using it as an array index, to inhibit the potential spectre-v1 write
gadget.

This class of issue is also known as CVE-2018-3693, or "bounds check bypass
store".

Found by smatch.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
---
 kernel/events/core.c | 2 ++
 1 file changed, 2 insertions(+)

For Arm CPUs, more details can be found in the Arm Cache Speculation
Side-channels whitepaper, available from the Arm security updates site [1].

Mark.

[1] https://developer.arm.com/support/arm-security-updates/speculative-processor-vulnerability

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 8f0434a9951a..eece719bd18e 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -8155,6 +8155,7 @@ struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX];
 static void sw_perf_event_destroy(struct perf_event *event)
 {
 	u64 event_id = event->attr.config;
+	event_id = array_index_nospec(event_id, PERF_COUNT_SW_MAX);
 
 	WARN_ON(event->parent);
 
@@ -8186,6 +8187,7 @@ static int perf_swevent_init(struct perf_event *event)
 
 	if (event_id >= PERF_COUNT_SW_MAX)
 		return -ENOENT;
+	event_id = array_index_nospec(event_id, PERF_COUNT_SW_MAX);
 
 	if (!event->parent) {
 		int err;
-- 
2.11.0

Re: [PATCH] perf/core: fix possible spectre-v1 write

[kbuild test robot]

[-- Attachment #1: Type: text/plain, Size: 1563 bytes --]

Hi Mark,

I love your patch! Yet something to improve:

[auto build test ERROR on tip/perf/core]
[also build test ERROR on v4.18-rc4 next-20180710]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

url:    https://github.com/0day-ci/linux/commits/Mark-Rutland/perf-core-fix-possible-spectre-v1-write/20180711-023735
config: i386-randconfig-x009-201827 (attached as .config)
compiler: gcc-7 (Debian 7.3.0-16) 7.3.0
reproduce:
        # save the attached .config to linux build tree
        make ARCH=i386 

All errors (new ones prefixed by >>):

   kernel/events/core.c: In function 'sw_perf_event_destroy':
>> kernel/events/core.c:8158:13: error: implicit declaration of function 'array_index_nospec'; did you mean 'array_index_mask_nospec'? [-Werror=implicit-function-declaration]
     event_id = array_index_nospec(event_id, PERF_COUNT_SW_MAX);
                ^~~~~~~~~~~~~~~~~~
                array_index_mask_nospec
   cc1: some warnings being treated as errors

vim +8158 kernel/events/core.c

  8154	
  8155	static void sw_perf_event_destroy(struct perf_event *event)
  8156	{
  8157		u64 event_id = event->attr.config;
> 8158		event_id = array_index_nospec(event_id, PERF_COUNT_SW_MAX);
  8159	
  8160		WARN_ON(event->parent);
  8161	
  8162		static_key_slow_dec(&perf_swevent_enabled[event_id]);
  8163		swevent_hlist_put();
  8164	}
  8165	

---
0-DAY kernel test infrastructure                Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all                   Intel Corporation

[-- Attachment #2: .config.gz --]
[-- Type: application/gzip, Size: 30435 bytes --]

Re: [PATCH] perf/core: fix possible spectre-v1 write

[kbuild test robot]

[-- Attachment #1: Type: text/plain, Size: 1509 bytes --]

Hi Mark,

I love your patch! Yet something to improve:

[auto build test ERROR on tip/perf/core]
[also build test ERROR on v4.18-rc4 next-20180710]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

url:    https://github.com/0day-ci/linux/commits/Mark-Rutland/perf-core-fix-possible-spectre-v1-write/20180711-023735
config: i386-randconfig-s0-201827 (attached as .config)
compiler: gcc-6 (Debian 6.4.0-9) 6.4.0 20171026
reproduce:
        # save the attached .config to linux build tree
        make ARCH=i386 

All errors (new ones prefixed by >>):

   kernel/events/core.c: In function 'sw_perf_event_destroy':
>> kernel/events/core.c:8158:13: error: implicit declaration of function 'array_index_nospec' [-Werror=implicit-function-declaration]
     event_id = array_index_nospec(event_id, PERF_COUNT_SW_MAX);
                ^~~~~~~~~~~~~~~~~~
   cc1: some warnings being treated as errors

vim +/array_index_nospec +8158 kernel/events/core.c

  8154	
  8155	static void sw_perf_event_destroy(struct perf_event *event)
  8156	{
  8157		u64 event_id = event->attr.config;
> 8158		event_id = array_index_nospec(event_id, PERF_COUNT_SW_MAX);
  8159	
  8160		WARN_ON(event->parent);
  8161	
  8162		static_key_slow_dec(&perf_swevent_enabled[event_id]);
  8163		swevent_hlist_put();
  8164	}
  8165	

---
0-DAY kernel test infrastructure                Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all                   Intel Corporation

[-- Attachment #2: .config.gz --]
[-- Type: application/gzip, Size: 30766 bytes --]

Re: [PATCH] perf/core: fix possible spectre-v1 write

[Mark Rutland]

On Tue, Jul 10, 2018 at 07:06:07PM +0100, Mark Rutland wrote:
> It's possible for userspace to control event_id. Sanitize event_id when
> using it as an array index, to inhibit the potential spectre-v1 write
> gadget.
> 
> This class of issue is also known as CVE-2018-3693, or "bounds check bypass
> store".
> 
> Found by smatch.
> 
> Signed-off-by: Mark Rutland <mark.rutland@arm.com>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Ingo Molnar <mingo@redhat.com>
> ---
>  kernel/events/core.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> For Arm CPUs, more details can be found in the Arm Cache Speculation
> Side-channels whitepaper, available from the Arm security updates site [1].
> 
> Mark.
> 
> [1] https://developer.arm.com/support/arm-security-updates/speculative-processor-vulnerability
> 
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 8f0434a9951a..eece719bd18e 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -8155,6 +8155,7 @@ struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX];
>  static void sw_perf_event_destroy(struct perf_event *event)
>  {
>  	u64 event_id = event->attr.config;
> +	event_id = array_index_nospec(event_id, PERF_COUNT_SW_MAX);

As the kbuild test robot has pointed out, I've failed to include
<linux/nospec.h> for this to compile.

I'll spin a v2 with that added, and the result tested.

Thanks,
Mark.