2013-09-09 05:12:38

by Li Bin

Subject: [PATCH] workqueue: fix potential reentrancy issue

From: Li Bin <[email protected]>

When one work starts execution, the high bits of work's data contain
pool ID. It can represent a maximum of WORK_OFFQ_POOL_NONE. Pool ID
is assigned WORK_OFFQ_POOL_NONE when the work being initialized
indicating that no pool is associated and get_work_pool() uses it to
check the associated pool. So if worker_pool_assign_id() assigns a
ID greater than or equal WORK_OFFQ_POOL_NONE to a pool, it may break
the non-reentrance guarantee.

This patch fix this issue by modifying the worker_pool_assign_id()
function to add the WORK_OFFQ_POOL_NONE check condition.

Signed-off-by: Li Bin <[email protected]>
---
kernel/workqueue.c | 11 ++++++++++-
1 file changed, 10 insertions(+), 1 deletion(-)

diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 41019b1..97d9ff7 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -518,7 +518,14 @@ static inline void debug_work_activate(struct work_struct *work) { }
static inline void debug_work_deactivate(struct work_struct *work) { }
#endif

-/* allocate ID and assign it to @pool */
+/**
+ * worker_pool_assign_id - allocate ID and assing it to @pool
+ * @pool: the pool pointer of interest
+ *
+ * Return 0 if ID assigned successful.
+ * Return non-zero if the allocation fails or the ID number out of
+ * %WORK_OFFQ_POOL_NONE range.
+ */
static int worker_pool_assign_id(struct worker_pool *pool)
{
int ret;
@@ -526,6 +533,8 @@ static int worker_pool_assign_id(struct worker_pool *pool)
lockdep_assert_held(&wq_pool_mutex);

ret = idr_alloc(&worker_pool_idr, pool, 0, 0, GFP_KERNEL);
+ if (ret >= WORK_OFFQ_POOL_NONE)
+ return ret;
if (ret >= 0) {
pool->id = ret;
return 0;
--
1.8.2.1

2013-09-09 14:14:35

by Tejun Heo

Subject: Re: [PATCH] workqueue: fix potential reentrancy issue

On Mon, Sep 09, 2013 at 01:12:14PM +0800, Libin wrote:
> From: Li Bin <[email protected]>
>
> When one work starts execution, the high bits of work's data contain
> pool ID. It can represent a maximum of WORK_OFFQ_POOL_NONE. Pool ID
> is assigned WORK_OFFQ_POOL_NONE when the work being initialized
> indicating that no pool is associated and get_work_pool() uses it to
> check the associated pool. So if worker_pool_assign_id() assigns a
> ID greater than or equal WORK_OFFQ_POOL_NONE to a pool, it may break
> the non-reentrance guarantee.
>
> This patch fix this issue by modifying the worker_pool_assign_id()
> function to add the WORK_OFFQ_POOL_NONE check condition.
>
> Signed-off-by: Li Bin <[email protected]>
> ---
> kernel/workqueue.c | 11 ++++++++++-
> 1 file changed, 10 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
> index 41019b1..97d9ff7 100644
> --- a/kernel/workqueue.c
> +++ b/kernel/workqueue.c
> @@ -518,7 +518,14 @@ static inline void debug_work_activate(struct work_struct *work) { }
> static inline void debug_work_deactivate(struct work_struct *work) { }
> #endif
>
> -/* allocate ID and assign it to @pool */
> +/**
> + * worker_pool_assign_id - allocate ID and assing it to @pool
> + * @pool: the pool pointer of interest
> + *
> + * Return 0 if ID assigned successful.
> + * Return non-zero if the allocation fails or the ID number out of
> + * %WORK_OFFQ_POOL_NONE range.
> + */
> static int worker_pool_assign_id(struct worker_pool *pool)
> {
> int ret;
> @@ -526,6 +533,8 @@ static int worker_pool_assign_id(struct worker_pool *pool)
> lockdep_assert_held(&wq_pool_mutex);
>
> ret = idr_alloc(&worker_pool_idr, pool, 0, 0, GFP_KERNEL);
> + if (ret >= WORK_OFFQ_POOL_NONE)
> + return ret;

Hmmm.... this would leak the allocated ID. Just setting @end param to
idr_alloc to WORK_OFFQ_POOL_NONE would work, right?

Thanks.

--
tejun

2013-09-10 00:34:00

by Li Bin

Subject: Re: [PATCH] workqueue: fix potential reentrancy issue

Hello Tejun,
On 2013/9/9 22:14, Tejun Heo wrote:
> On Mon, Sep 09, 2013 at 01:12:14PM +0800, Libin wrote:
>> From: Li Bin <[email protected]>
>>
>> When one work starts execution, the high bits of work's data contain
>> pool ID. It can represent a maximum of WORK_OFFQ_POOL_NONE. Pool ID
>> is assigned WORK_OFFQ_POOL_NONE when the work being initialized
>> indicating that no pool is associated and get_work_pool() uses it to
>> check the associated pool. So if worker_pool_assign_id() assigns a
>> ID greater than or equal WORK_OFFQ_POOL_NONE to a pool, it may break
>> the non-reentrance guarantee.
>>
>> This patch fix this issue by modifying the worker_pool_assign_id()
>> function to add the WORK_OFFQ_POOL_NONE check condition.
>>
>> Signed-off-by: Li Bin <[email protected]>
>> ---
>> kernel/workqueue.c | 11 ++++++++++-
>> 1 file changed, 10 insertions(+), 1 deletion(-)
>>
>> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
>> index 41019b1..97d9ff7 100644
>> --- a/kernel/workqueue.c
>> +++ b/kernel/workqueue.c
>> @@ -518,7 +518,14 @@ static inline void debug_work_activate(struct work_struct *work) { }
>> static inline void debug_work_deactivate(struct work_struct *work) { }
>> #endif
>>
>> -/* allocate ID and assign it to @pool */
>> +/**
>> + * worker_pool_assign_id - allocate ID and assing it to @pool
>> + * @pool: the pool pointer of interest
>> + *
>> + * Return 0 if ID assigned successful.
>> + * Return non-zero if the allocation fails or the ID number out of
>> + * %WORK_OFFQ_POOL_NONE range.
>> + */
>> static int worker_pool_assign_id(struct worker_pool *pool)
>> {
>> int ret;
>> @@ -526,6 +533,8 @@ static int worker_pool_assign_id(struct worker_pool *pool)
>> lockdep_assert_held(&wq_pool_mutex);
>>
>> ret = idr_alloc(&worker_pool_idr, pool, 0, 0, GFP_KERNEL);
>> + if (ret >= WORK_OFFQ_POOL_NONE)
>> + return ret;
>
> Hmmm.... this would leak the allocated ID. Just setting @end param to
> idr_alloc to WORK_OFFQ_POOL_NONE would work, right?
>

Yes, I will update it with the last patch.
Thanks!
Libin

> Thanks.
>