The functions vring_new_virtqueue() and __vring_new_virtqueue() are used
with split rings, and any allocations within these functions are managed
outside of the .we_own_ring flag. The commit cbeedb72b97a ("virtio_ring:
allocate desc state for split ring separately") allocates the desc state
within the __vring_new_virtqueue() but frees it only when the .we_own_ring
flag is set. This leads to a memory leak when freeing such allocated
virtqueues with the vring_del_virtqueue() function.
Fix this by moving the desc_state free code outside the flag and only
for split rings. Issue was discovered during testing with remoteproc
and virtio_rpmsg.
Fixes: cbeedb72b97a ("virtio_ring: allocate desc state for split ring separately")
Signed-off-by: Suman Anna <[email protected]>
---
drivers/virtio/virtio_ring.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index 867c7ebd3f10..58b96baa8d48 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -2203,10 +2203,10 @@ void vring_del_virtqueue(struct virtqueue *_vq)
vq->split.queue_size_in_bytes,
vq->split.vring.desc,
vq->split.queue_dma_addr);
-
- kfree(vq->split.desc_state);
}
}
+ if (!vq->packed_ring)
+ kfree(vq->split.desc_state);
list_del(&_vq->list);
kfree(vq);
}
--
2.23.0
On 2020/2/25 上午5:26, Suman Anna wrote:
> The functions vring_new_virtqueue() and __vring_new_virtqueue() are used
> with split rings, and any allocations within these functions are managed
> outside of the .we_own_ring flag. The commit cbeedb72b97a ("virtio_ring:
> allocate desc state for split ring separately") allocates the desc state
> within the __vring_new_virtqueue() but frees it only when the .we_own_ring
> flag is set. This leads to a memory leak when freeing such allocated
> virtqueues with the vring_del_virtqueue() function.
>
> Fix this by moving the desc_state free code outside the flag and only
> for split rings. Issue was discovered during testing with remoteproc
> and virtio_rpmsg.
>
> Fixes: cbeedb72b97a ("virtio_ring: allocate desc state for split ring separately")
> Signed-off-by: Suman Anna <[email protected]>
> ---
> drivers/virtio/virtio_ring.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
> index 867c7ebd3f10..58b96baa8d48 100644
> --- a/drivers/virtio/virtio_ring.c
> +++ b/drivers/virtio/virtio_ring.c
> @@ -2203,10 +2203,10 @@ void vring_del_virtqueue(struct virtqueue *_vq)
> vq->split.queue_size_in_bytes,
> vq->split.vring.desc,
> vq->split.queue_dma_addr);
> -
> - kfree(vq->split.desc_state);
> }
> }
> + if (!vq->packed_ring)
> + kfree(vq->split.desc_state);
Nitpick, it looks to me it would be more clear if we just free
desc_state unconditionally here (and remove the kfree for packed above).
Anyway desc_state will be allocated by use even if !we_own_ring.
Thanks
> list_del(&_vq->list);
> kfree(vq);
> }
Hi Jason,
On 2/24/20 11:39 PM, Jason Wang wrote:
>
> On 2020/2/25 上午5:26, Suman Anna wrote:
>> The functions vring_new_virtqueue() and __vring_new_virtqueue() are used
>> with split rings, and any allocations within these functions are managed
>> outside of the .we_own_ring flag. The commit cbeedb72b97a ("virtio_ring:
>> allocate desc state for split ring separately") allocates the desc state
>> within the __vring_new_virtqueue() but frees it only when the
>> .we_own_ring
>> flag is set. This leads to a memory leak when freeing such allocated
>> virtqueues with the vring_del_virtqueue() function.
>>
>> Fix this by moving the desc_state free code outside the flag and only
>> for split rings. Issue was discovered during testing with remoteproc
>> and virtio_rpmsg.
>>
>> Fixes: cbeedb72b97a ("virtio_ring: allocate desc state for split ring
>> separately")
>> Signed-off-by: Suman Anna <[email protected]>
>> ---
>> drivers/virtio/virtio_ring.c | 4 ++--
>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
>> index 867c7ebd3f10..58b96baa8d48 100644
>> --- a/drivers/virtio/virtio_ring.c
>> +++ b/drivers/virtio/virtio_ring.c
>> @@ -2203,10 +2203,10 @@ void vring_del_virtqueue(struct virtqueue *_vq)
>> vq->split.queue_size_in_bytes,
>> vq->split.vring.desc,
>> vq->split.queue_dma_addr);
>> -
>> - kfree(vq->split.desc_state);
>> }
>> }
>> + if (!vq->packed_ring)
>> + kfree(vq->split.desc_state);
>
>
> Nitpick, it looks to me it would be more clear if we just free
> desc_state unconditionally here (and remove the kfree for packed above).
OK, are you sure you want that to be folded into this patch? It looks to
me a separate cleanup/consolidation patch, and packed desc_state does
not suffer this memleak, and need not be backported into stable kernels.
regards
Suman
> Anyway desc_state will be allocated by use even if !we_own_ring.
>
> Thanks
>
>
>> list_del(&_vq->list);
>> kfree(vq);
>> }
>
On 2020/2/26 上午12:51, Suman Anna wrote:
> Hi Jason,
>
> On 2/24/20 11:39 PM, Jason Wang wrote:
>> On 2020/2/25 上午5:26, Suman Anna wrote:
>>> The functions vring_new_virtqueue() and __vring_new_virtqueue() are used
>>> with split rings, and any allocations within these functions are managed
>>> outside of the .we_own_ring flag. The commit cbeedb72b97a ("virtio_ring:
>>> allocate desc state for split ring separately") allocates the desc state
>>> within the __vring_new_virtqueue() but frees it only when the
>>> .we_own_ring
>>> flag is set. This leads to a memory leak when freeing such allocated
>>> virtqueues with the vring_del_virtqueue() function.
>>>
>>> Fix this by moving the desc_state free code outside the flag and only
>>> for split rings. Issue was discovered during testing with remoteproc
>>> and virtio_rpmsg.
>>>
>>> Fixes: cbeedb72b97a ("virtio_ring: allocate desc state for split ring
>>> separately")
>>> Signed-off-by: Suman Anna<[email protected]>
>>> ---
>>> drivers/virtio/virtio_ring.c | 4 ++--
>>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
>>> index 867c7ebd3f10..58b96baa8d48 100644
>>> --- a/drivers/virtio/virtio_ring.c
>>> +++ b/drivers/virtio/virtio_ring.c
>>> @@ -2203,10 +2203,10 @@ void vring_del_virtqueue(struct virtqueue *_vq)
>>> vq->split.queue_size_in_bytes,
>>> vq->split.vring.desc,
>>> vq->split.queue_dma_addr);
>>> -
>>> - kfree(vq->split.desc_state);
>>> }
>>> }
>>> + if (!vq->packed_ring)
>>> + kfree(vq->split.desc_state);
>> Nitpick, it looks to me it would be more clear if we just free
>> desc_state unconditionally here (and remove the kfree for packed above).
> OK, are you sure you want that to be folded into this patch? It looks to
> me a separate cleanup/consolidation patch, and packed desc_state does
> not suffer this memleak, and need not be backported into stable kernels.
>
> regards
> Suman
Though it's just a small tweak, I'm fine for leaving it for future.
So
Acked-by: Jason Wang <[email protected]>
Thanks
On 2/25/20 9:13 PM, Jason Wang wrote:
>
> On 2020/2/26 上午12:51, Suman Anna wrote:
>> Hi Jason,
>>
>> On 2/24/20 11:39 PM, Jason Wang wrote:
>>> On 2020/2/25 上午5:26, Suman Anna wrote:
>>>> The functions vring_new_virtqueue() and __vring_new_virtqueue() are
>>>> used
>>>> with split rings, and any allocations within these functions are
>>>> managed
>>>> outside of the .we_own_ring flag. The commit cbeedb72b97a
>>>> ("virtio_ring:
>>>> allocate desc state for split ring separately") allocates the desc
>>>> state
>>>> within the __vring_new_virtqueue() but frees it only when the
>>>> .we_own_ring
>>>> flag is set. This leads to a memory leak when freeing such allocated
>>>> virtqueues with the vring_del_virtqueue() function.
>>>>
>>>> Fix this by moving the desc_state free code outside the flag and only
>>>> for split rings. Issue was discovered during testing with remoteproc
>>>> and virtio_rpmsg.
>>>>
>>>> Fixes: cbeedb72b97a ("virtio_ring: allocate desc state for split ring
>>>> separately")
>>>> Signed-off-by: Suman Anna<[email protected]>
>>>> ---
>>>> drivers/virtio/virtio_ring.c | 4 ++--
>>>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/virtio/virtio_ring.c
>>>> b/drivers/virtio/virtio_ring.c
>>>> index 867c7ebd3f10..58b96baa8d48 100644
>>>> --- a/drivers/virtio/virtio_ring.c
>>>> +++ b/drivers/virtio/virtio_ring.c
>>>> @@ -2203,10 +2203,10 @@ void vring_del_virtqueue(struct virtqueue *_vq)
>>>> vq->split.queue_size_in_bytes,
>>>> vq->split.vring.desc,
>>>> vq->split.queue_dma_addr);
>>>> -
>>>> - kfree(vq->split.desc_state);
>>>> }
>>>> }
>>>> + if (!vq->packed_ring)
>>>> + kfree(vq->split.desc_state);
>>> Nitpick, it looks to me it would be more clear if we just free
>>> desc_state unconditionally here (and remove the kfree for packed above).
>> OK, are you sure you want that to be folded into this patch? It looks to
>> me a separate cleanup/consolidation patch, and packed desc_state does
>> not suffer this memleak, and need not be backported into stable kernels.
>>
>> regards
>> Suman
>
>
> Though it's just a small tweak, I'm fine for leaving it for future.
>
> So
>
> Acked-by: Jason Wang <[email protected]>
Thanks Jason, will post a patch for the same once this is merged.
regards
Suman
On 2/25/20 9:13 PM, Jason Wang wrote:
>
> On 2020/2/26 上午12:51, Suman Anna wrote:
>> Hi Jason,
>>
>> On 2/24/20 11:39 PM, Jason Wang wrote:
>>> On 2020/2/25 上午5:26, Suman Anna wrote:
>>>> The functions vring_new_virtqueue() and __vring_new_virtqueue() are
>>>> used
>>>> with split rings, and any allocations within these functions are
>>>> managed
>>>> outside of the .we_own_ring flag. The commit cbeedb72b97a
>>>> ("virtio_ring:
>>>> allocate desc state for split ring separately") allocates the desc
>>>> state
>>>> within the __vring_new_virtqueue() but frees it only when the
>>>> .we_own_ring
>>>> flag is set. This leads to a memory leak when freeing such allocated
>>>> virtqueues with the vring_del_virtqueue() function.
>>>>
>>>> Fix this by moving the desc_state free code outside the flag and only
>>>> for split rings. Issue was discovered during testing with remoteproc
>>>> and virtio_rpmsg.
>>>>
>>>> Fixes: cbeedb72b97a ("virtio_ring: allocate desc state for split ring
>>>> separately")
>>>> Signed-off-by: Suman Anna<[email protected]>
>>>> ---
>>>> drivers/virtio/virtio_ring.c | 4 ++--
>>>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/virtio/virtio_ring.c
>>>> b/drivers/virtio/virtio_ring.c
>>>> index 867c7ebd3f10..58b96baa8d48 100644
>>>> --- a/drivers/virtio/virtio_ring.c
>>>> +++ b/drivers/virtio/virtio_ring.c
>>>> @@ -2203,10 +2203,10 @@ void vring_del_virtqueue(struct virtqueue *_vq)
>>>> vq->split.queue_size_in_bytes,
>>>> vq->split.vring.desc,
>>>> vq->split.queue_dma_addr);
>>>> -
>>>> - kfree(vq->split.desc_state);
>>>> }
>>>> }
>>>> + if (!vq->packed_ring)
>>>> + kfree(vq->split.desc_state);
>>> Nitpick, it looks to me it would be more clear if we just free
>>> desc_state unconditionally here (and remove the kfree for packed above).
>> OK, are you sure you want that to be folded into this patch? It looks to
>> me a separate cleanup/consolidation patch, and packed desc_state does
>> not suffer this memleak, and need not be backported into stable kernels.
>>
>> regards
>> Suman
>
>
> Though it's just a small tweak, I'm fine for leaving it for future.
>
> So
>
> Acked-by: Jason Wang <[email protected]>
>
Mike,
Ping on this. I don't see the patch in -next yet. Can we get this into
the current -rc please?
regards
Suman
On Thu, Mar 05, 2020 at 06:27:53PM -0600, Suman Anna wrote:
> On 2/25/20 9:13 PM, Jason Wang wrote:
> >
> > On 2020/2/26 上午12:51, Suman Anna wrote:
> >> Hi Jason,
> >>
> >> On 2/24/20 11:39 PM, Jason Wang wrote:
> >>> On 2020/2/25 上午5:26, Suman Anna wrote:
> >>>> The functions vring_new_virtqueue() and __vring_new_virtqueue() are
> >>>> used
> >>>> with split rings, and any allocations within these functions are
> >>>> managed
> >>>> outside of the .we_own_ring flag. The commit cbeedb72b97a
> >>>> ("virtio_ring:
> >>>> allocate desc state for split ring separately") allocates the desc
> >>>> state
> >>>> within the __vring_new_virtqueue() but frees it only when the
> >>>> .we_own_ring
> >>>> flag is set. This leads to a memory leak when freeing such allocated
> >>>> virtqueues with the vring_del_virtqueue() function.
> >>>>
> >>>> Fix this by moving the desc_state free code outside the flag and only
> >>>> for split rings. Issue was discovered during testing with remoteproc
> >>>> and virtio_rpmsg.
> >>>>
> >>>> Fixes: cbeedb72b97a ("virtio_ring: allocate desc state for split ring
> >>>> separately")
> >>>> Signed-off-by: Suman Anna<[email protected]>
> >>>> ---
> >>>> drivers/virtio/virtio_ring.c | 4 ++--
> >>>> 1 file changed, 2 insertions(+), 2 deletions(-)
> >>>>
> >>>> diff --git a/drivers/virtio/virtio_ring.c
> >>>> b/drivers/virtio/virtio_ring.c
> >>>> index 867c7ebd3f10..58b96baa8d48 100644
> >>>> --- a/drivers/virtio/virtio_ring.c
> >>>> +++ b/drivers/virtio/virtio_ring.c
> >>>> @@ -2203,10 +2203,10 @@ void vring_del_virtqueue(struct virtqueue *_vq)
> >>>> vq->split.queue_size_in_bytes,
> >>>> vq->split.vring.desc,
> >>>> vq->split.queue_dma_addr);
> >>>> -
> >>>> - kfree(vq->split.desc_state);
> >>>> }
> >>>> }
> >>>> + if (!vq->packed_ring)
> >>>> + kfree(vq->split.desc_state);
> >>> Nitpick, it looks to me it would be more clear if we just free
> >>> desc_state unconditionally here (and remove the kfree for packed above).
> >> OK, are you sure you want that to be folded into this patch? It looks to
> >> me a separate cleanup/consolidation patch, and packed desc_state does
> >> not suffer this memleak, and need not be backported into stable kernels.
> >>
> >> regards
> >> Suman
> >
> >
> > Though it's just a small tweak, I'm fine for leaving it for future.
> >
> > So
> >
> > Acked-by: Jason Wang <[email protected]>
> >
>
> Mike,
> Ping on this. I don't see the patch in -next yet. Can we get this into
> the current -rc please?
>
> regards
> Suman
Yes will queue it shortly, thanks!