Message-ID: <da1eae25-1df0-36f0-3d4f-12ba52f1f62f@gmail.com>
Date:   Fri, 10 Feb 2023 09:52:27 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.7.1
Subject: Re: [ovs-dev] [PATCH v2] net: openvswitch: fix possible memory leak
 in ovs_meter_cmd_set()
To:     Eelco Chaudron <echaudro@redhat.com>,
        Simon Horman <simon.horman@corigine.com>
Cc:     pshelar@ovn.org, davem@davemloft.net, edumazet@google.com,
        kuba@kernel.org, pabeni@redhat.com, xiangxia.m.yue@gmail.com,
        dev@openvswitch.org, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org
References: <20230209093240.14685-1-hbh25y@gmail.com>
 <Y+TR3sB4X5Yt79Tx@corigine.com>
 <5C180FB7-5EAA-4AEB-BD69-9522F2CD73B5@redhat.com>
Content-Language: en-US
From:   Hangyu Hua <hbh25y@gmail.com>
In-Reply-To: <5C180FB7-5EAA-4AEB-BD69-9522F2CD73B5@redhat.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 9/2/2023 21:42, Eelco Chaudron wrote:
> 
> 
> On 9 Feb 2023, at 11:58, Simon Horman wrote:
> 
>> On Thu, Feb 09, 2023 at 05:32:40PM +0800, Hangyu Hua wrote:
>>> old_meter needs to be free after it is detached regardless of whether
>>> the new meter is successfully attached.
>>>
>>> Fixes: c7c4c44c9a95 ("net: openvswitch: expand the meters supported number")
>>> Signed-off-by: Hangyu Hua <hbh25y@gmail.com>
>>> ---
>>>
>>> v2: use goto label and free old_meter outside of ovs lock.
>>>
>>>   net/openvswitch/meter.c | 12 +++++++++---
>>>   1 file changed, 9 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/net/openvswitch/meter.c b/net/openvswitch/meter.c
>>> index 6e38f68f88c2..9b680f0894f1 100644
>>> --- a/net/openvswitch/meter.c
>>> +++ b/net/openvswitch/meter.c
>>> @@ -417,6 +417,7 @@ static int ovs_meter_cmd_set(struct sk_buff *skb, struct genl_info *info)
>>>   	int err;
>>>   	u32 meter_id;
>>>   	bool failed;
>>> +	bool locked = true;
>>>
>>>   	if (!a[OVS_METER_ATTR_ID])
>>>   		return -EINVAL;
>>> @@ -448,11 +449,13 @@ static int ovs_meter_cmd_set(struct sk_buff *skb, struct genl_info *info)
>>>   		goto exit_unlock;
>>>
>>>   	err = attach_meter(meter_tbl, meter);
>>> -	if (err)
>>> -		goto exit_unlock;
>>>
>>>   	ovs_unlock();
>>>
>>> +	if (err) {
>>> +		locked = false;
>>> +		goto exit_free_old_meter;
>>> +	}
>>>   	/* Build response with the meter_id and stats from
>>>   	 * the old meter, if any.
>>>   	 */
>>> @@ -472,8 +475,11 @@ static int ovs_meter_cmd_set(struct sk_buff *skb, struct genl_info *info)
>>>   	genlmsg_end(reply, ovs_reply_header);
>>>   	return genlmsg_reply(reply, info);
>>>
>>> +exit_free_old_meter:
>>> +	ovs_meter_free(old_meter);
>>>   exit_unlock:
>>> -	ovs_unlock();
>>> +	if (locked)
>>> +		ovs_unlock();
>>
>> I see where you are going here, but is the complexity of the
>> locked variable worth the benefit of freeing old_meter outside
>> the lock?
> 
> Looking at the error path, I would agree with Simon, and just add an “exit_free_old_meter” label as suggested in v1 and keep the lock in place to make the error path more straightforward.
> 
> //Eelco

I see. I will send a v3 later.

Thanks,
Hangyu

> 
>>>   	nlmsg_free(reply);
>>>   exit_free_meter:
>>>   	kfree(meter);
>>> -- 
>>> 2.34.1
>>>
>>> _______________________________________________
>>> dev mailing list
>>> dev@openvswitch.org
>>> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
>>>
>