[PATCH] markers: bit-field is not thread-safe nor smp-safe

[PATCH] markers: bit-field is not thread-safe nor smp-safe

[Lai Jiangshan]

bit-field is not thread-safe nor smp-safe.

struct marker_entry.rcu_pending is not protected by any lock
in rcu-callback free_old_closure().
so we must turn it into a safe type.

detail:

I suppose rcu_pending and ptype are store in struct marker_entry.tmp1

free_old_closure() side:           change ptype side:

                                |  load struct marker_entry.tmp1
--------------------------------|--------------------------------
                                |  change ptype bit in tmp1
load struct marker_entry.tmp1   |
change rcu_pending bit in tmp1  |
store tmp1                      |
--------------------------------|--------------------------------
                                |  store tmp1

now this result equals that free_old_closure() do not change rcu_pending bit, bug.

see also: http://en.wikipedia.org/wiki/Bit_field

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
---
diff --git a/kernel/marker.c b/kernel/marker.c
index 7d1faec..4777218 100644
--- a/kernel/marker.c
+++ b/kernel/marker.c
@@ -62,7 +62,7 @@ struct marker_entry {
 	int refcount;	/* Number of times armed. 0 if disarmed. */
 	struct rcu_head rcu;
 	void *oldptr;
-	unsigned char rcu_pending:1;
+	unsigned char rcu_pending;
 	unsigned char ptype:1;
 	char name[0];	/* Contains name'\0'format'\0' */
 };

Re: [PATCH] markers: bit-field is not thread-safe nor smp-safe

[KOSAKI Motohiro]

Hi Lai-san,

> 
> bit-field is not thread-safe nor smp-safe.
> 
> struct marker_entry.rcu_pending is not protected by any lock
> in rcu-callback free_old_closure().
> so we must turn it into a safe type.

hmmm
however, char also doesn't smp-safe because some architecture doesn't have
any byte load/store instruction.

It seems bogus solution to me ;)


> detail:
> 
> I suppose rcu_pending and ptype are store in struct marker_entry.tmp1
> 
> free_old_closure() side:           change ptype side:
> 
>                                 |  load struct marker_entry.tmp1
> --------------------------------|--------------------------------
>                                 |  change ptype bit in tmp1
> load struct marker_entry.tmp1   |
> change rcu_pending bit in tmp1  |
> store tmp1                      |
> --------------------------------|--------------------------------
>                                 |  store tmp1
> 
> now this result equals that free_old_closure() do not change rcu_pending bit, bug.
> 
> see also: http://en.wikipedia.org/wiki/Bit_field
> 
> Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
> ---
> diff --git a/kernel/marker.c b/kernel/marker.c
> index 7d1faec..4777218 100644
> --- a/kernel/marker.c
> +++ b/kernel/marker.c
> @@ -62,7 +62,7 @@ struct marker_entry {
>  	int refcount;	/* Number of times armed. 0 if disarmed. */
>  	struct rcu_head rcu;
>  	void *oldptr;
> -	unsigned char rcu_pending:1;
> +	unsigned char rcu_pending;
>  	unsigned char ptype:1;
>  	char name[0];	/* Contains name'\0'format'\0' */
>  };

Re: [PATCH] markers: bit-field is not thread-safe nor smp-safe

[Lai Jiangshan]

KOSAKI Motohiro wrote:
> Hi Lai-san,
> 
>> bit-field is not thread-safe nor smp-safe.
>>
>> struct marker_entry.rcu_pending is not protected by any lock
>> in rcu-callback free_old_closure().
>> so we must turn it into a safe type.
> 
> hmmm
> however, char also doesn't smp-safe because some architecture doesn't have
> any byte load/store instruction.
> 
> It seems bogus solution to me ;)

Hi, KOSAKI-san,

	Thank you very much!

	char also doesn't smp-safe if the architecture doesn't have
any byte load/store instruction.

	We must use int, is it right?

Lai

> 
> 
>> detail:
>>
>> I suppose rcu_pending and ptype are store in struct marker_entry.tmp1
>>
>> free_old_closure() side:           change ptype side:
>>
>>                                 |  load struct marker_entry.tmp1
>> --------------------------------|--------------------------------
>>                                 |  change ptype bit in tmp1
>> load struct marker_entry.tmp1   |
>> change rcu_pending bit in tmp1  |
>> store tmp1                      |
>> --------------------------------|--------------------------------
>>                                 |  store tmp1
>>
>> now this result equals that free_old_closure() do not change rcu_pending bit, bug.
>>
>> see also: http://en.wikipedia.org/wiki/Bit_field
>>
>> Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
>> ---
>> diff --git a/kernel/marker.c b/kernel/marker.c
>> index 7d1faec..4777218 100644
>> --- a/kernel/marker.c
>> +++ b/kernel/marker.c
>> @@ -62,7 +62,7 @@ struct marker_entry {
>>  	int refcount;	/* Number of times armed. 0 if disarmed. */
>>  	struct rcu_head rcu;
>>  	void *oldptr;
>> -	unsigned char rcu_pending:1;
>> +	unsigned char rcu_pending;
>>  	unsigned char ptype:1;
>>  	char name[0];	/* Contains name'\0'format'\0' */
>>  };
> 
> 
> 
> 
> 
> 

Re: [PATCH] markers: bit-field is not thread-safe nor smp-safe

[Mathieu Desnoyers]

* Lai Jiangshan (laijs@cn.fujitsu.com) wrote:
> 
> bit-field is not thread-safe nor smp-safe.
> 
> struct marker_entry.rcu_pending is not protected by any lock
> in rcu-callback free_old_closure().
> so we must turn it into a safe type.
> 

All struct marker_entry.rcu_pending accesses are done with the
markers_mutex held, except the one done in free_old_closure(). Normally,
there should be a 
        if (entry->rcu_pending)
           rcu_barrier_sched();

At the beginning of each markers_mutex section (just after get_marker())
to make sure any pending callback is executed at that point before any
of rcu_pending or ptype are touched.

I just noticed that the "markers: fix unchecked format" patch has a race
with respect to this. I'll post a patch in a jiffy.

Mathieu

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68

[PATCH] markers: bit-field is not thread-safe nor smp-safe

[Lai Jiangshan]

bit-field is not thread-safe nor smp-safe.

struct marker_entry.rcu_pending is not protected by any lock
in rcu-callback free_old_closure().
so we must turn it into a safe type.

detail:

I suppose rcu_pending and ptype are store in struct marker_entry.tmp1

free_old_closure() side:           change ptype side:

                                |  load struct marker_entry.tmp1
--------------------------------|--------------------------------
                                |  change ptype bit in tmp1
load struct marker_entry.tmp1   |
change rcu_pending bit in tmp1  |
store tmp1                      |
--------------------------------|--------------------------------
                                |  store tmp1

now this result equals that free_old_closure() do not change rcu_pending
bit, bug! This bug will cause redundant rcu_barrier_sched() called.
not too harmful.

----- corresponding:

free_old_closure() side:           change ptype side:

load struct marker_entry.tmp1   |
--------------------------------|--------------------------------
                                |  load struct marker_entry.tmp1
change rcu_pending bit in tmp1  |
                                |  change ptype bit in tmp1
                                |  store tmp1
--------------------------------|--------------------------------
store tmp1                      |

now this result equals that change ptype side do not change ptype
bit, bug! this bug cause marker_probe_cb() access to invalid memory.
oops!

see also: http://en.wikipedia.org/wiki/Bit_field

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
---
diff --git a/kernel/marker.c b/kernel/marker.c
index 7d1faec..95c62da 100644
--- a/kernel/marker.c
+++ b/kernel/marker.c
@@ -62,7 +62,7 @@ struct marker_entry {
 	int refcount;	/* Number of times armed. 0 if disarmed. */
 	struct rcu_head rcu;
 	void *oldptr;
-	unsigned char rcu_pending:1;
+	int rcu_pending;
 	unsigned char ptype:1;
 	char name[0];	/* Contains name'\0'format'\0' */
 };

Re: [PATCH] markers: bit-field is not thread-safe nor smp-safe

[Mathieu Desnoyers]

* Lai Jiangshan (laijs@cn.fujitsu.com) wrote:
> 
> bit-field is not thread-safe nor smp-safe.
> 
> struct marker_entry.rcu_pending is not protected by any lock
> in rcu-callback free_old_closure().
> so we must turn it into a safe type.
> 


hrm, yes, you are right. I first test for 

        if (entry->rcu_pending)
           rcu_barrier_sched();

To check if I must execute the rcu callback, and _this_ races against
the entry->rcu_pending = 0; within the callback.

Your fix is therefore needed.

Thanks !

Acked-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>

> detail:
> 
> I suppose rcu_pending and ptype are store in struct marker_entry.tmp1
> 
> free_old_closure() side:           change ptype side:
> 
>                                 |  load struct marker_entry.tmp1
> --------------------------------|--------------------------------
>                                 |  change ptype bit in tmp1
> load struct marker_entry.tmp1   |
> change rcu_pending bit in tmp1  |
> store tmp1                      |
> --------------------------------|--------------------------------
>                                 |  store tmp1
> 
> now this result equals that free_old_closure() do not change rcu_pending
> bit, bug! This bug will cause redundant rcu_barrier_sched() called.
> not too harmful.
> 
> ----- corresponding:
> 
> free_old_closure() side:           change ptype side:
> 
> load struct marker_entry.tmp1   |
> --------------------------------|--------------------------------
>                                 |  load struct marker_entry.tmp1
> change rcu_pending bit in tmp1  |
>                                 |  change ptype bit in tmp1
>                                 |  store tmp1
> --------------------------------|--------------------------------
> store tmp1                      |
> 
> now this result equals that change ptype side do not change ptype
> bit, bug! this bug cause marker_probe_cb() access to invalid memory.
> oops!
> 
> see also: http://en.wikipedia.org/wiki/Bit_field
> 
> Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
> ---
> diff --git a/kernel/marker.c b/kernel/marker.c
> index 7d1faec..95c62da 100644
> --- a/kernel/marker.c
> +++ b/kernel/marker.c
> @@ -62,7 +62,7 @@ struct marker_entry {
>  	int refcount;	/* Number of times armed. 0 if disarmed. */
>  	struct rcu_head rcu;
>  	void *oldptr;
> -	unsigned char rcu_pending:1;
> +	int rcu_pending;
>  	unsigned char ptype:1;
>  	char name[0];	/* Contains name'\0'format'\0' */
>  };
> 
> 
> 

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68

Re: [PATCH] markers: bit-field is not thread-safe nor smp-safe

[Ingo Molnar]

* Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca> wrote:

> * Lai Jiangshan (laijs@cn.fujitsu.com) wrote:
> > 
> > bit-field is not thread-safe nor smp-safe.
> > 
> > struct marker_entry.rcu_pending is not protected by any lock
> > in rcu-callback free_old_closure().
> > so we must turn it into a safe type.
> > 
> 
> 
> hrm, yes, you are right. I first test for 
> 
>         if (entry->rcu_pending)
>            rcu_barrier_sched();
> 
> To check if I must execute the rcu callback, and _this_ races against
> the entry->rcu_pending = 0; within the callback.
> 
> Your fix is therefore needed.
> 
> Thanks !
> 
> Acked-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>

applied to tip/tracing/markers, thanks guys!

	Ingo