netdev - Re: RE: Network do not works with linux >= 6.1.2. Issue bisected to "425c9bd06b7a70796d880828d15c11321bdfb76d" (RDMA/irdma: Report the correct link speed)

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Tue, 10 Jan 2023 14:02:34 +0800
From:   Zhu Yanjun <yanjun.zhu@...ux.dev>
To:     "Saleem, Shiraz" <shiraz.saleem@...el.com>,
        Jason Gunthorpe <jgg@...pe.ca>,
        Jaroslav Pulchart <jaroslav.pulchart@...ddata.com>,
        "Ertman, David M" <david.m.ertman@...el.com>,
        "Wesierski, DawidX" <dawidx.wesierski@...el.com>
Cc:     "kamalheib1@...il.com" <kamalheib1@...il.com>,
        "leon@...nel.org" <leon@...nel.org>,
        "sashal@...nel.org" <sashal@...nel.org>,
        "linux-rdma@...r.kernel.org" <linux-rdma@...r.kernel.org>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        Igor Raits <igor.raits@...ddata.com>
Subject: Re: RE: Network do not works with linux >= 6.1.2. Issue bisected to
 "425c9bd06b7a70796d880828d15c11321bdfb76d" (RDMA/irdma: Report the correct
 link speed)

在 2023/1/10 3:36, Saleem, Shiraz 写道:
>> Subject: Re: Network do not works with linux >= 6.1.2. Issue bisected to
>> "425c9bd06b7a70796d880828d15c11321bdfb76d" (RDMA/irdma: Report the
>> correct link speed)
>>
>> On Fri, Jan 06, 2023 at 08:55:29AM +0100, Jaroslav Pulchart wrote:
>>> [  257.967099] task:NetworkManager  state:D stack:0     pid:3387
>>> ppid:1      flags:0x00004002
>>> [  257.975446] Call Trace:
>>> [  257.977901]  <TASK>
>>> [  257.980004]  __schedule+0x1eb/0x630 [  257.983498]
>>> schedule+0x5a/0xd0 [  257.986641]  schedule_timeout+0x11d/0x160 [
>>> 257.990654]  __wait_for_common+0x90/0x1e0 [  257.994666]  ?
>>> usleep_range_state+0x90/0x90 [  257.998854]
>>> __flush_workqueue+0x13a/0x3f0 [  258.002955]  ?
>>> __kernfs_remove.part.0+0x11e/0x1e0
>>> [  258.007661]  ib_cache_cleanup_one+0x1c/0xe0 [ib_core] [
>>> 258.012721]  __ib_unregister_device+0x62/0xa0 [ib_core] [  258.017959]
>>> ib_unregister_device+0x22/0x30 [ib_core] [  258.023024]
>>> irdma_remove+0x1a/0x60 [irdma] [  258.027223]
>>> auxiliary_bus_remove+0x18/0x30 [  258.031414]
>>> device_release_driver_internal+0x1aa/0x230
>>> [  258.036643]  bus_remove_device+0xd8/0x150 [  258.040654]
>>> device_del+0x18b/0x3f0 [  258.044149]  ice_unplug_aux_dev+0x42/0x60
>>> [ice]
>>
>> We talked about this already - wasn't it on this series?
> 
> This is yet another path (when ice ports are added to a bond) I believe where the RDMA aux device
> is removed holding the RTNL lock. It's being exposed now with this recent irdma patch - 425c9bd06b7a,
> causing a deadlock.
> 
> ice_lag_event_handler [rtnl_lock]
>   ->ice_lag_changeupper_event
>       ->ice_unplug_aux_dev
>          ->irdma_remove
>              ->ib_unregister_device
>                 ->ib_cache_cleanup_one
>                    ->flush_workqueue(ib)
>                       ->irdma_query_port
>                           -> ib_get_eth_speed [rtnl_lock]

Agree with the above analysis.
Maybe a quick and direct fix is like this.

@@ -74,6 +74,7 @@ static int irdma_query_port(struct ib_device *ibdev, 
u32 port,
  {
         struct irdma_device *iwdev = to_iwdev(ibdev);
         struct net_device *netdev = iwdev->netdev;
+       bool unlock_rtnl = false;

         /* no need to zero out pros here. done by caller */

@@ -91,9 +92,16 @@ static int irdma_query_port(struct ib_device *ibdev, 
u32 port,
                 props->phys_state = IB_PORT_PHYS_STATE_DISABLED;
         }

+       if (rtnl_is_locked()) {
+               rtnl_unlock();
+               unlock_rtnl = true;
+       }
         ib_get_eth_speed(ibdev, port, &props->active_speed,
                          &props->active_width);

+       if (unlock_rtnl) {
+               rtnl_lock();
+       }
         if (rdma_protocol_roce(ibdev, 1)) {
                 props->gid_tbl_len = 32;
                 props->ip_gids = true;

Zhu Yanjun

> 
> Previous discussion was on ethtool channel config change, https://lore.kernel.org/linux-rdma/Y5ES3kmYSINlAQhz@x130/,
> which David E. is taking care of.
> 
> We are working on a patch for this issue.
> 
>>
>> Don't hold locks when removing aux devices.
>>
>>> [  258.048707]  ice_lag_changeupper_event+0x287/0x2a0 [ice] [
>>> 258.054038]  ice_lag_event_handler+0x51/0x130 [ice] [  258.058930]
>>> raw_notifier_call_chain+0x41/0x60 [  258.063381]
>>> __netdev_upper_dev_link+0x1a0/0x370
>>> [  258.068008]  netdev_master_upper_dev_link+0x3d/0x60
>>> [  258.072886]  bond_enslave+0xd16/0x16f0 [bonding] [  258.077517]  ?
>>> nla_put+0x28/0x40 [  258.080756]  do_setlink+0x26c/0xc10 [
>>> 258.084249]  ? avc_alloc_node+0x27/0x180 [  258.088173]  ?
>>> __nla_validate_parse+0x141/0x190 [  258.092708]
>>> __rtnl_newlink+0x53a/0x620 [  258.096549]  rtnl_newlink+0x44/0x70
>>
>> Especially not the rtnl.
>>
>> Jason