Received-SPF: pass (google.com: domain of linux-wireless-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <a54581e8-792a-0f54-e886-bca3d1d1063e@wetzel-home.de>
Date:   Tue, 25 Oct 2022 16:42:10 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.3.3
Subject: Re: [PATCH] wifi: mac80211: fix general-protection-fault in
 ieee80211_subif_start_xmit()
Content-Language: en-US
To:     Zhengchao Shao <shaozhengchao@huawei.com>,
        linux-wireless@vger.kernel.org, netdev@vger.kernel.org,
        johannes@sipsolutions.net, davem@davemloft.net,
        edumazet@google.com, kuba@kernel.org, pabeni@redhat.com
Cc:     weiyongjun1@huawei.com, yuehaibing@huawei.com
References: <20221025123250.143952-1-shaozhengchao@huawei.com>
From:   Alexander Wetzel <alexander@wetzel-home.de>
In-Reply-To: <20221025123250.143952-1-shaozhengchao@huawei.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 25.10.22 14:32, Zhengchao Shao wrote:
> When device is running and the interface status is changed, the gpf issue
> is triggered. The problem triggering process is as follows:
> Thread A:                           Thread B
> ieee80211_runtime_change_iftype()   process_one_work()
>      ...                                 ...
>      ieee80211_do_stop()                 ...
>      ...                                 ...
>          sdata->bss = NULL               ...
>          ...                             ieee80211_subif_start_xmit()
>                                              ieee80211_multicast_to_unicast
>                                      //!sdata->bss->multicast_to_unicast
>                                        cause gpf issue
> 
> When the interface status is changed, the sending queue continues to send
> packets. After the bss is set to NULL, the bss is accessed. As a result,
> this causes a general-protection-fault issue.
> 
> The following is the stack information:
> general protection fault, probably for non-canonical address
> 0xdffffc000000002f: 0000 [#1] PREEMPT SMP KASAN
> KASAN: null-ptr-deref in range [0x0000000000000178-0x000000000000017f]
> Workqueue: mld mld_ifc_work
> RIP: 0010:ieee80211_subif_start_xmit+0x25b/0x1310
> Call Trace:
> <TASK>
> dev_hard_start_xmit+0x1be/0x990
> __dev_queue_xmit+0x2c9a/0x3b60
> ip6_finish_output2+0xf92/0x1520
> ip6_finish_output+0x6af/0x11e0
> ip6_output+0x1ed/0x540
> mld_sendpack+0xa09/0xe70
> mld_ifc_work+0x71c/0xdb0
> process_one_work+0x9bf/0x1710
> worker_thread+0x665/0x1080
> kthread+0x2e4/0x3a0
> ret_from_fork+0x1f/0x30
> </TASK>
> 
> Fixes: 107395f9cf44 ("wifi: mac80211: Drop support for TX push path")

Don't think this patch fixes an issue introduced with the patch you 
refer to. This patch changed nothing from a flow perspective and is just 
cleaning up unused code.
It still may still make sense to refer to the series: It next to be sure 
triggered the issue for at least one driver (I assume it was hwsim here.)

That said this seems to be more related to whatever caused this bug:
f856373e2f31 ("wifi: mac80211: do not wake queues on a vif that is being 
stopped")


> Reported-by: syzbot+c6e8fca81c294fd5620a@syzkaller.appspotmail.com
> Signed-off-by: Zhengchao Shao <shaozhengchao@huawei.com>
> ---
>   net/mac80211/iface.c | 6 ++++++
>   1 file changed, 6 insertions(+)
> 
> diff --git a/net/mac80211/iface.c b/net/mac80211/iface.c
> index dd9ac1f7d2ea..5a924459bfd1 100644
> --- a/net/mac80211/iface.c
> +++ b/net/mac80211/iface.c
> @@ -1900,6 +1900,9 @@ static int ieee80211_runtime_change_iftype(struct ieee80211_sub_if_data *sdata,
>   				  IEEE80211_QUEUE_STOP_REASON_IFTYPE_CHANGE);
>   	synchronize_net();
>   
> +	if (sdata->dev)
> +		netif_tx_stop_all_queues(sdata->dev);

All mac80211 interfaces are now non-queuing interfaces.
When you stop the netif queues for a non-queuing interface netdev will 
warn about that.

To avoid that you have to replace the netif call with
	clear_bit(SDATA_STATE_RUNNING, &sdata->state);

Should just work the same for you here.
> +
>   	ieee80211_do_stop(sdata, false);
>   
>   	ieee80211_teardown_sdata(sdata);
> @@ -1922,6 +1925,9 @@ static int ieee80211_runtime_change_iftype(struct ieee80211_sub_if_data *sdata,
>   	err = ieee80211_do_open(&sdata->wdev, false);
>   	WARN(err, "type change: do_open returned %d", err);
>   
> +	if (sdata->dev)
> +		netif_tx_start_all_queues(sdata->dev);

That must then be of course
	set_bit(SDATA_STATE_RUNNING, &sdata->state);


> +
>   	ieee80211_wake_vif_queues(local, sdata,
>   				  IEEE80211_QUEUE_STOP_REASON_IFTYPE_CHANGE);
>   	return ret;

Alexander