netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH v2 1/2] ibmvnic: fix a race between open and reset
@ 2021-02-03  5:06 Sukadev Bhattiprolu
  2021-02-03  5:06 ` [PATCH v2 2/2] ibmvnic: fix race with multiple open/close Sukadev Bhattiprolu
  0 siblings, 1 reply; 5+ messages in thread
From: Sukadev Bhattiprolu @ 2021-02-03  5:06 UTC (permalink / raw)
  To: netdev; +Cc: Dany Madden, Lijun Pan, Rick Lindsley, abdhalee, sukadev

__ibmvnic_reset() currently reads the adapter->state before getting the
rtnl and saves that state as the "target state" for the reset. If this
read occurs when adapter is in PROBED state, the target state would be
PROBED.

Just after the target state is saved, and before the actual reset process
is started (i.e before rtnl is acquired) if we get an ibmvnic_open() call
we would move the adapter to OPEN state.

But when the reset is processed (after ibmvnic_open()) drops the rtnl),
it will leave the adapter in PROBED state even though we already moved
it to OPEN.

To fix this, use the RTNL to improve the serialization when reading/updating
the adapter state. i.e determine the target state of a reset only after
getting the RTNL. And if a reset is in progress during an open, simply
set the target state of the adapter and let the reset code finish the
open (like we currently do if failover is pending).

One twist to this serialization is if the adapter state changes when we
drop the RTNL to update the link state. Account for this by checking if
there was an intervening open and update the target state for the reset
accordingly (see new comments in the code). Note that only the reset
functions and ibmvnic_open() can set the adapter to OPEN state and this
must happen under rtnl.

Fixes: 7d7195a026ba ("ibmvnic: Do not process device remove during device reset")
Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
Reviewed-by: Dany Madden <drt@linux.ibm.com>
---
Changelog[v2]
	[Jakub Kicinski] Use ASSERT_RTNL() instead of WARN_ON_ONCE()
	and rtnl_is_locked());
---
 drivers/net/ethernet/ibm/ibmvnic.c | 65 ++++++++++++++++++++++++++----
 1 file changed, 58 insertions(+), 7 deletions(-)

diff --git a/drivers/net/ethernet/ibm/ibmvnic.c b/drivers/net/ethernet/ibm/ibmvnic.c
index 0ed169ef1cfc..78d244aeee69 100644
--- a/drivers/net/ethernet/ibm/ibmvnic.c
+++ b/drivers/net/ethernet/ibm/ibmvnic.c
@@ -1197,12 +1197,26 @@ static int ibmvnic_open(struct net_device *netdev)
 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
 	int rc;
 
-	/* If device failover is pending, just set device state and return.
-	 * Device operation will be handled by reset routine.
+	ASSERT_RTNL();
+
+	/**
+	 * If device failover is pending or we are about to reset, just set
+	 * device state and return. Device operation will be handled by reset
+	 * routine.
+	 *
+	 * It should be safe to overwrite the adapter->state here. Since
+	 * we hold the rtnl, either the reset has not actually started or
+	 * the rtnl got dropped during the set_link_state() in do_reset().
+	 * In the former case, no one else is changing the state (again we
+	 * have the rtnl) and in the latter case, do_reset() will detect and
+	 * honor our setting below.
 	 */
-	if (adapter->failover_pending) {
+	if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
+		netdev_dbg(netdev, "[S:%d FOP:%d] Resetting, deferring open\n",
+			   adapter->state, adapter->failover_pending);
 		adapter->state = VNIC_OPEN;
-		return 0;
+		rc = 0;
+		goto out;
 	}
 
 	if (adapter->state != VNIC_CLOSED) {
@@ -1222,10 +1236,12 @@ static int ibmvnic_open(struct net_device *netdev)
 
 out:
 	/*
-	 * If open fails due to a pending failover, set device state and
-	 * return. Device operation will be handled by reset routine.
+	 * If open failed and there is a pending failover or in-progress reset,
+	 * set device state and return. Device operation will be handled by
+	 * reset routine. See also comments above regarding rtnl.
 	 */
-	if (rc && adapter->failover_pending) {
+	if (rc &&
+	    (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
 		adapter->state = VNIC_OPEN;
 		rc = 0;
 	}
@@ -1954,6 +1970,14 @@ static int do_reset(struct ibmvnic_adapter *adapter,
 	if (rwi->reset_reason == VNIC_RESET_FAILOVER)
 		adapter->failover_pending = false;
 
+	/* read the state and check (again) after getting rtnl */
+	reset_state = adapter->state;
+
+	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
+		rc = -EBUSY;
+		goto out;
+	}
+
 	netif_carrier_off(netdev);
 
 	old_num_rx_queues = adapter->req_rx_queues;
@@ -1984,7 +2008,26 @@ static int do_reset(struct ibmvnic_adapter *adapter,
 			if (rc)
 				goto out;
 
+			if (adapter->state == VNIC_OPEN) {
+				/**
+				 * When we dropped rtnl, ibmvnic_open() got
+				 * it and noticed that we are resetting and
+				 * set the adapter state to OPEN. Update our
+				 * new "target" state, and resume the reset
+				 * from VNIC_CLOSING state.
+				 */
+				netdev_dbg(netdev,
+					   "Open changed state from %d, updating.\n",
+					    reset_state);
+				reset_state = VNIC_OPEN;
+				adapter->state = VNIC_CLOSING;
+			}
+
 			if (adapter->state != VNIC_CLOSING) {
+				/**
+				 * If someone else changed the adapter state
+				 * when we dropped the rtnl, fail the reset
+				 */
 				rc = -1;
 				goto out;
 			}
@@ -2133,6 +2176,14 @@ static int do_hard_reset(struct ibmvnic_adapter *adapter,
 	netdev_dbg(adapter->netdev, "Hard resetting driver (%d)\n",
 		   rwi->reset_reason);
 
+	/* read the state and check (again) after getting rtnl */
+	reset_state = adapter->state;
+
+	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
+		rc = -EBUSY;
+		goto out;
+	}
+
 	netif_carrier_off(netdev);
 	adapter->reset_reason = rwi->reset_reason;
 
-- 
2.26.2


^ permalink raw reply related	[flat|nested] 5+ messages in thread

* [PATCH v2 2/2] ibmvnic: fix race with multiple open/close
  2021-02-03  5:06 [PATCH v2 1/2] ibmvnic: fix a race between open and reset Sukadev Bhattiprolu
@ 2021-02-03  5:06 ` Sukadev Bhattiprolu
  2021-02-03 18:40   ` Willem de Bruijn
  0 siblings, 1 reply; 5+ messages in thread
From: Sukadev Bhattiprolu @ 2021-02-03  5:06 UTC (permalink / raw)
  To: netdev; +Cc: Dany Madden, Lijun Pan, Rick Lindsley, abdhalee, sukadev

If two or more instances of 'ip link set' commands race and first one
already brings the interface up (or down), the subsequent instances
can simply return without redoing the up/down operation.

Fixes: ed651a10875f ("ibmvnic: Updated reset handling")
Reported-by: Abdul Haleem <abdhalee@in.ibm.com>
Tested-by: Abdul Haleem <abdhalee@in.ibm.com>
Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>

---
Changelog[v2] For consistency with ibmvnic_open() use "goto out" and return
	      from end of function.
---
 drivers/net/ethernet/ibm/ibmvnic.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/drivers/net/ethernet/ibm/ibmvnic.c b/drivers/net/ethernet/ibm/ibmvnic.c
index 78d244aeee69..df1b4884b4e8 100644
--- a/drivers/net/ethernet/ibm/ibmvnic.c
+++ b/drivers/net/ethernet/ibm/ibmvnic.c
@@ -1219,6 +1219,14 @@ static int ibmvnic_open(struct net_device *netdev)
 		goto out;
 	}
 
+	/* If adapter is already open, we don't have to do anything. */
+	if (adapter->state == VNIC_OPEN) {
+		netdev_dbg(netdev, "[S:%d] adapter already open\n",
+			   adapter->state);
+		rc = 0;
+		goto out;
+	}
+
 	if (adapter->state != VNIC_CLOSED) {
 		rc = ibmvnic_login(netdev);
 		if (rc)
@@ -1392,6 +1400,12 @@ static int ibmvnic_close(struct net_device *netdev)
 		return 0;
 	}
 
+	/* If adapter is already closed, we don't have to do anything. */
+	if (adapter->state == VNIC_CLOSED) {
+		netdev_dbg(netdev, "[S:%d] adapter already closed\n",
+			   adapter->state);
+		return 0;
+	}
 	rc = __ibmvnic_close(netdev);
 	ibmvnic_cleanup(netdev);
 
-- 
2.26.2


^ permalink raw reply related	[flat|nested] 5+ messages in thread

* Re: [PATCH v2 2/2] ibmvnic: fix race with multiple open/close
  2021-02-03  5:06 ` [PATCH v2 2/2] ibmvnic: fix race with multiple open/close Sukadev Bhattiprolu
@ 2021-02-03 18:40   ` Willem de Bruijn
  2021-02-03 20:56     ` Sukadev Bhattiprolu
  0 siblings, 1 reply; 5+ messages in thread
From: Willem de Bruijn @ 2021-02-03 18:40 UTC (permalink / raw)
  To: Sukadev Bhattiprolu
  Cc: Network Development, Dany Madden, Lijun Pan, Rick Lindsley, abdhalee

On Wed, Feb 3, 2021 at 12:10 AM Sukadev Bhattiprolu
<sukadev@linux.ibm.com> wrote:
>
> If two or more instances of 'ip link set' commands race and first one
> already brings the interface up (or down), the subsequent instances
> can simply return without redoing the up/down operation.
>
> Fixes: ed651a10875f ("ibmvnic: Updated reset handling")
> Reported-by: Abdul Haleem <abdhalee@in.ibm.com>
> Tested-by: Abdul Haleem <abdhalee@in.ibm.com>
> Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
>
> ---
> Changelog[v2] For consistency with ibmvnic_open() use "goto out" and return
>               from end of function.

Did you find the code path that triggers this?

In v1 we discussed how the usual ip link path should not call the
driver twice based on IFF_UP.

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH v2 2/2] ibmvnic: fix race with multiple open/close
  2021-02-03 18:40   ` Willem de Bruijn
@ 2021-02-03 20:56     ` Sukadev Bhattiprolu
  2021-02-06 18:45       ` Jakub Kicinski
  0 siblings, 1 reply; 5+ messages in thread
From: Sukadev Bhattiprolu @ 2021-02-03 20:56 UTC (permalink / raw)
  To: Willem de Bruijn
  Cc: Network Development, Dany Madden, Lijun Pan, Rick Lindsley, abdhalee

Willem de Bruijn [willemdebruijn.kernel@gmail.com] wrote:
> On Wed, Feb 3, 2021 at 12:10 AM Sukadev Bhattiprolu
> <sukadev@linux.ibm.com> wrote:
> >
> > If two or more instances of 'ip link set' commands race and first one
> > already brings the interface up (or down), the subsequent instances
> > can simply return without redoing the up/down operation.
> >
> > Fixes: ed651a10875f ("ibmvnic: Updated reset handling")
> > Reported-by: Abdul Haleem <abdhalee@in.ibm.com>
> > Tested-by: Abdul Haleem <abdhalee@in.ibm.com>
> > Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
> >
> > ---
> > Changelog[v2] For consistency with ibmvnic_open() use "goto out" and return
> >               from end of function.
> 
> Did you find the code path that triggers this?

Not yet. I need to find time on the system to repro/debug that
> 
> In v1 we discussed how the usual ip link path should not call the
> driver twice based on IFF_UP.

Yes, we can hold this patch for now if necessary. Hopefully Patch 1/2 is
ok.

Sukadev

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH v2 2/2] ibmvnic: fix race with multiple open/close
  2021-02-03 20:56     ` Sukadev Bhattiprolu
@ 2021-02-06 18:45       ` Jakub Kicinski
  0 siblings, 0 replies; 5+ messages in thread
From: Jakub Kicinski @ 2021-02-06 18:45 UTC (permalink / raw)
  To: Sukadev Bhattiprolu
  Cc: Willem de Bruijn, Network Development, Dany Madden, Lijun Pan,
	Rick Lindsley, abdhalee

On Wed, 3 Feb 2021 12:56:52 -0800 Sukadev Bhattiprolu wrote:
> Willem de Bruijn [willemdebruijn.kernel@gmail.com] wrote:
> > On Wed, Feb 3, 2021 at 12:10 AM Sukadev Bhattiprolu
> > <sukadev@linux.ibm.com> wrote:  
> > >
> > > If two or more instances of 'ip link set' commands race and first one
> > > already brings the interface up (or down), the subsequent instances
> > > can simply return without redoing the up/down operation.
> > >
> > > Fixes: ed651a10875f ("ibmvnic: Updated reset handling")
> > > Reported-by: Abdul Haleem <abdhalee@in.ibm.com>
> > > Tested-by: Abdul Haleem <abdhalee@in.ibm.com>
> > > Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
> > >
> > > ---
> > > Changelog[v2] For consistency with ibmvnic_open() use "goto out" and return
> > >               from end of function.  
> > 
> > Did you find the code path that triggers this?  
> 
> Not yet. I need to find time on the system to repro/debug that
> > 
> > In v1 we discussed how the usual ip link path should not call the
> > driver twice based on IFF_UP.  
> 
> Yes, we can hold this patch for now if necessary. Hopefully Patch 1/2 is
> ok.

Patch 1 does not apply to net as is. Please rebase and fix up the
comments the way I fixed them for

 "ibmvnic: Clear failover_pending if unable to schedule"

Thanks!

^ permalink raw reply	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2021-02-06 18:46 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-02-03  5:06 [PATCH v2 1/2] ibmvnic: fix a race between open and reset Sukadev Bhattiprolu
2021-02-03  5:06 ` [PATCH v2 2/2] ibmvnic: fix race with multiple open/close Sukadev Bhattiprolu
2021-02-03 18:40   ` Willem de Bruijn
2021-02-03 20:56     ` Sukadev Bhattiprolu
2021-02-06 18:45       ` Jakub Kicinski

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).