From: John Fastabend <john.fastabend@gmail.com>
To: Cong Wang <xiyou.wangcong@gmail.com>,
Yunsheng Lin <linyunsheng@huawei.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>,
Jiri Pirko <jiri@resnulli.us>, David Miller <davem@davemloft.net>,
Jakub Kicinski <kuba@kernel.org>,
Linux Kernel Network Developers <netdev@vger.kernel.org>,
LKML <linux-kernel@vger.kernel.org>,
linuxarm@huawei.com
Subject: Re: [PATCH net-next] net: sch_generic: aviod concurrent reset and enqueue op for lockless qdisc
Date: Thu, 03 Sep 2020 18:30:22 -0700 [thread overview]
Message-ID: <5f5198ae79b98_361ef20824@john-XPS-13-9370.notmuch> (raw)
In-Reply-To: <CAM_iQpWGaTSkg+-Em6u=NSWcyswX-xN=-1p0OAdaR___U1M4rg@mail.gmail.com>
Cong Wang wrote:
> On Wed, Sep 2, 2020 at 7:22 PM Yunsheng Lin <linyunsheng@huawei.com> wrote:
> >
> > On 2020/9/3 9:48, Cong Wang wrote:
> > > On Wed, Sep 2, 2020 at 6:22 PM Yunsheng Lin <linyunsheng@huawei.com> wrote:
> > >>
> > >> On 2020/9/3 8:35, Cong Wang wrote:
> > >>> On Tue, Sep 1, 2020 at 11:35 PM Yunsheng Lin <linyunsheng@huawei.com> wrote:
> > >>>>
> > >>>> On 2020/9/2 12:41, Cong Wang wrote:
> > >>>>> On Tue, Sep 1, 2020 at 6:42 PM Yunsheng Lin <linyunsheng@huawei.com> wrote:
> > >>>>>>
> > >>>>>> On 2020/9/2 2:24, Cong Wang wrote:
> > >>>>>>> On Mon, Aug 31, 2020 at 5:59 PM Yunsheng Lin <linyunsheng@huawei.com> wrote:
> > >>>>>>>>
> > >>>>>>>> Currently there is concurrent reset and enqueue operation for the
> > >>>>>>>> same lockless qdisc when there is no lock to synchronize the
> > >>>>>>>> q->enqueue() in __dev_xmit_skb() with the qdisc reset operation in
> > >>>>>>>> qdisc_deactivate() called by dev_deactivate_queue(), which may cause
> > >>>>>>>> out-of-bounds access for priv->ring[] in hns3 driver if user has
> > >>>>>>>> requested a smaller queue num when __dev_xmit_skb() still enqueue a
> > >>>>>>>> skb with a larger queue_mapping after the corresponding qdisc is
> > >>>>>>>> reset, and call hns3_nic_net_xmit() with that skb later.
> > >>>>>>>
> > >>>>>>> Can you be more specific here? Which call path requests a smaller
> > >>>>>>> tx queue num? If you mean netif_set_real_num_tx_queues(), clearly
> > >>>>>>> we already have a synchronize_net() there.
> > >>>>>>
> > >>>>>> When the netdevice is in active state, the synchronize_net() seems to
> > >>>>>> do the correct work, as below:
> > >>>>>>
> > >>>>>> CPU 0: CPU1:
> > >>>>>> __dev_queue_xmit() netif_set_real_num_tx_queues()
> > >>>>>> rcu_read_lock_bh();
> > >>>>>> netdev_core_pick_tx(dev, skb, sb_dev);
> > >>>>>> .
> > >>>>>> . dev->real_num_tx_queues = txq;
> > >>>>>> . .
> > >>>>>> . .
> > >>>>>> . synchronize_net();
> > >>>>>> . .
> > >>>>>> q->enqueue() .
> > >>>>>> . .
> > >>>>>> rcu_read_unlock_bh() .
> > >>>>>> qdisc_reset_all_tx_gt
> > >>>>>>
> > >>>>>>
> > >>>>>
> > >>>>> Right.
> > >>>>>
> > >>>>>
> > >>>>>> but dev->real_num_tx_queues is not RCU-protected, maybe that is a problem
> > >>>>>> too.
> > >>>>>>
> > >>>>>> The problem we hit is as below:
> > >>>>>> In hns3_set_channels(), hns3_reset_notify(h, HNAE3_DOWN_CLIENT) is called
> > >>>>>> to deactive the netdevice when user requested a smaller queue num, and
> > >>>>>> txq->qdisc is already changed to noop_qdisc when calling
> > >>>>>> netif_set_real_num_tx_queues(), so the synchronize_net() in the function
> > >>>>>> netif_set_real_num_tx_queues() does not help here.
> > >>>>>
> > >>>>> How could qdisc still be running after deactivating the device?
> > >>>>
> > >>>> qdisc could be running during the device deactivating process.
> > >>>>
> > >>>> The main process of changing channel number is as below:
> > >>>>
> > >>>> 1. dev_deactivate()
> > >>>> 2. hns3 handware related setup
> > >>>> 3. netif_set_real_num_tx_queues()
> > >>>> 4. netif_tx_wake_all_queues()
> > >>>> 5. dev_activate()
> > >>>>
> > >>>> During step 1, qdisc could be running while qdisc is resetting, so
> > >>>> there could be skb left in the old qdisc(which will be restored back to
> > >>>> txq->qdisc during dev_activate()), as below:
> > >>>>
> > >>>> CPU 0: CPU1:
> > >>>> __dev_queue_xmit(): dev_deactivate_many():
> > >>>> rcu_read_lock_bh(); qdisc_deactivate(qdisc);
> > >>>> q = rcu_dereference_bh(txq->qdisc); .
> > >>>> netdev_core_pick_tx(dev, skb, sb_dev); .
> > >>>> .
> > >>>> . rcu_assign_pointer(dev_queue->qdisc, qdisc_default);
> > >>>> . .
> > >>>> . .
> > >>>> . .
> > >>>> . .
> > >>>> q->enqueue() .
> > >>>
> > >>>
> > >>> Well, like I said, if the deactivated bit were tested before ->enqueue(),
> > >>> there would be no packet queued after qdisc_deactivate().
Trying to unwind this through git history :/
Original code had a test_bit in dev_xmit_skb(),
if (q->flags & TCQ_F_NOLOCK) {
if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
__qdisc_drop(skb, &to_free);
rc = NET_XMIT_DROP;
} else {
rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
__qdisc_run(q);
}
if (unlikely(to_free))
kfree_skb_list(to_free);
return rc;
}
So we would never enqueue something on top of a deactivated qdisc. And to ensure
we don't have any in-flight enqueues we have to swap qdiscs, wait a grace
period, and then reset the qdisc. That _should_ be OK.
But, I'm still not entirely sure how you got here. In the drivers I did I always
stop the queue before messing with these things with netif_tx_stop_queue(). Then
we really should not get these packets into the driver.
In sch_direct_xmit():
if (likely(skb)) {
HARD_TX_LOCK(dev, txq, smp_processor_id());
if (!netif_xmit_frozen_or_stopped(txq))
skb = dev_hard_start_xmit(skb, dev, txq, &ret);
HARD_TX_UNLOCK(dev, txq);
} else {
if (root_lock)
spin_lock(root_lock);
return true;
}
Maybe I missed something? Does your driver use the netif_tx_stop/start APIs?
> > >>
> > >> Only if the deactivated bit testing is also protected by qdisc->seqlock?
> > >> otherwise there is still window between setting and testing the deactivated bit.
> > >
> > > Can you be more specific here? Why testing or setting a bit is not atomic?
> >
> > testing a bit or setting a bit separately is atomic.
> > But testing a bit and setting a bit is not atomic, right?
> >
> > cpu0: cpu1:
> > testing A bit
> > setting A bit .
> > . .
> > . qdisc enqueuing
> > qdisc reset
> >
>
> Well, this was not a problem until commit d518d2ed8640c1cbbb.
> Prior to that commit, qdsic can still be scheduled even with this
> race condition, that is, the packet just enqueued after resetting can
> still be dequeued with qdisc_run().
>
> It is amazing to see how messy the lockless qdisc is now.
>
> Thanks.
next prev parent reply other threads:[~2020-09-04 1:30 UTC|newest]
Thread overview: 23+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-09-01 0:55 [PATCH net-next] net: sch_generic: aviod concurrent reset and enqueue op for lockless qdisc Yunsheng Lin
2020-09-01 6:48 ` Eric Dumazet
2020-09-01 7:27 ` Yunsheng Lin
2020-09-01 18:34 ` David Miller
2020-09-02 1:43 ` Yunsheng Lin
2020-09-01 18:24 ` Cong Wang
2020-09-02 1:42 ` Yunsheng Lin
2020-09-02 4:41 ` Cong Wang
2020-09-02 6:34 ` Yunsheng Lin
2020-09-02 7:32 ` Eric Dumazet
2020-09-02 8:14 ` Yunsheng Lin
2020-09-02 9:20 ` Eric Dumazet
2020-09-03 1:14 ` Yunsheng Lin
2020-09-03 7:24 ` Eric Dumazet
2020-09-04 8:10 ` Yunsheng Lin
2020-09-03 0:35 ` Cong Wang
2020-09-03 1:21 ` Yunsheng Lin
2020-09-03 1:48 ` Cong Wang
2020-09-03 2:22 ` Yunsheng Lin
2020-09-03 2:53 ` Cong Wang
2020-09-04 1:30 ` John Fastabend [this message]
2020-09-04 8:08 ` Yunsheng Lin
2020-09-06 8:52 ` [net] 6fd0d0dede: hwsim.ap_ht40_5ghz_switch.fail kernel test robot
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=5f5198ae79b98_361ef20824@john-XPS-13-9370.notmuch \
--to=john.fastabend@gmail.com \
--cc=davem@davemloft.net \
--cc=jhs@mojatatu.com \
--cc=jiri@resnulli.us \
--cc=kuba@kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linuxarm@huawei.com \
--cc=linyunsheng@huawei.com \
--cc=netdev@vger.kernel.org \
--cc=xiyou.wangcong@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).