Message-ID: <20180406125758.GC19345@nanopsycho>
Date:   Fri, 6 Apr 2018 14:57:58 +0200
From:   Jiri Pirko <jiri@...nulli.us>
To:     Sridhar Samudrala <sridhar.samudrala@...el.com>
Cc:     mst@...hat.com, stephen@...workplumber.org, davem@...emloft.net,
        netdev@...r.kernel.org, virtualization@...ts.linux-foundation.org,
        virtio-dev@...ts.oasis-open.org, jesse.brandeburg@...el.com,
        alexander.h.duyck@...el.com, kubakici@...pl, jasowang@...hat.com,
        loseweigh@...il.com
Subject: Re: [RFC PATCH net-next v5 2/4] net: Introduce generic bypass module

Thu, Apr 05, 2018 at 11:08:21PM CEST, sridhar.samudrala@...el.com wrote:
>This provides a generic interface for paravirtual drivers to listen
>for netdev register/unregister/link change events from pci ethernet
>devices with the same MAC and takeover their datapath. The notifier and
>event handling code is based on the existing netvsc implementation. A
>paravirtual driver can use this module by registering a set of ops and
>each instance of the device when it is probed.
>
>Signed-off-by: Sridhar Samudrala <sridhar.samudrala@...el.com>
>---
> include/net/bypass.h |  80 ++++++++++
> net/Kconfig          |  18 +++
> net/core/Makefile    |   1 +
> net/core/bypass.c    | 406 +++++++++++++++++++++++++++++++++++++++++++++++++++
> 4 files changed, 505 insertions(+)
> create mode 100644 include/net/bypass.h
> create mode 100644 net/core/bypass.c
>
>diff --git a/include/net/bypass.h b/include/net/bypass.h
>new file mode 100644
>index 000000000000..e2dd122f951a
>--- /dev/null
>+++ b/include/net/bypass.h
>@@ -0,0 +1,80 @@
>+// SPDX-License-Identifier: GPL-2.0
>+/* Copyright (c) 2018, Intel Corporation. */
>+
>+#ifndef _NET_BYPASS_H
>+#define _NET_BYPASS_H
>+
>+#include <linux/netdevice.h>
>+
>+struct bypass_ops {

Perhaps "net_bypass_" would be better prefix for this module structs
and functions. No strong opinion though.


>+	int (*register_child)(struct net_device *bypass_netdev,
>+			      struct net_device *child_netdev);

We have master/slave upper/lower netdevices. This adds "child". Consider
using some existing names. Not sure if possible without loss of meaning.


>+	int (*join_child)(struct net_device *bypass_netdev,
>+			  struct net_device *child_netdev);
>+	int (*unregister_child)(struct net_device *bypass_netdev,
>+				struct net_device *child_netdev);
>+	int (*release_child)(struct net_device *bypass_netdev,
>+			     struct net_device *child_netdev);
>+	int (*update_link)(struct net_device *bypass_netdev,
>+			   struct net_device *child_netdev);
>+	rx_handler_result_t (*handle_frame)(struct sk_buff **pskb);
>+};
>+
>+struct bypass_instance {
>+	struct list_head list;
>+	struct net_device __rcu *bypass_netdev;
>+	struct bypass *bypass;
>+};
>+
>+struct bypass {
>+	struct list_head list;
>+	const struct bypass_ops *ops;
>+	const struct net_device_ops *netdev_ops;
>+	struct list_head instance_list;
>+	struct mutex lock;
>+};
>+
>+#if IS_ENABLED(CONFIG_NET_BYPASS)
>+
>+struct bypass *bypass_register_driver(const struct bypass_ops *ops,
>+				      const struct net_device_ops *netdev_ops);
>+void bypass_unregister_driver(struct bypass *bypass);
>+
>+int bypass_register_instance(struct bypass *bypass, struct net_device *dev);
>+int bypass_unregister_instance(struct bypass *bypass, struct net_device	*dev);
>+
>+int bypass_unregister_child(struct net_device *child_netdev);
>+
>+#else
>+
>+static inline
>+struct bypass *bypass_register_driver(const struct bypass_ops *ops,
>+				      const struct net_device_ops *netdev_ops)
>+{
>+	return NULL;
>+}
>+
>+static inline void bypass_unregister_driver(struct bypass *bypass)
>+{
>+}
>+
>+static inline int bypass_register_instance(struct bypass *bypass,
>+					   struct net_device *dev)
>+{
>+	return 0;
>+}
>+
>+static inline int bypass_unregister_instance(struct bypass *bypass,
>+					     struct net_device *dev)
>+{
>+	return 0;
>+}
>+
>+static inline int bypass_unregister_child(struct net_device *child_netdev)
>+{
>+	return 0;
>+}
>+
>+#endif
>+
>+#endif /* _NET_BYPASS_H */
>diff --git a/net/Kconfig b/net/Kconfig
>index 0428f12c25c2..994445f4a96a 100644
>--- a/net/Kconfig
>+++ b/net/Kconfig
>@@ -423,6 +423,24 @@ config MAY_USE_DEVLINK
> 	  on MAY_USE_DEVLINK to ensure they do not cause link errors when
> 	  devlink is a loadable module and the driver using it is built-in.
> 
>+config NET_BYPASS
>+	tristate "Bypass interface"
>+	---help---
>+	  This provides a generic interface for paravirtual drivers to listen
>+	  for netdev register/unregister/link change events from pci ethernet
>+	  devices with the same MAC and takeover their datapath. This also
>+	  enables live migration of a VM with direct attached VF by failing
>+	  over to the paravirtual datapath when the VF is unplugged.
>+
>+config MAY_USE_BYPASS
>+	tristate
>+	default m if NET_BYPASS=m
>+	default y if NET_BYPASS=y || NET_BYPASS=n
>+	help
>+	  Drivers using the bypass infrastructure should have a dependency
>+	  on MAY_USE_BYPASS to ensure they do not cause link errors when
>+	  bypass is a loadable module and the driver using it is built-in.
>+
> endif   # if NET
> 
> # Used by archs to tell that they support BPF JIT compiler plus which flavour.
>diff --git a/net/core/Makefile b/net/core/Makefile
>index 6dbbba8c57ae..a9727ed1c8fc 100644
>--- a/net/core/Makefile
>+++ b/net/core/Makefile
>@@ -30,3 +30,4 @@ obj-$(CONFIG_DST_CACHE) += dst_cache.o
> obj-$(CONFIG_HWBM) += hwbm.o
> obj-$(CONFIG_NET_DEVLINK) += devlink.o
> obj-$(CONFIG_GRO_CELLS) += gro_cells.o
>+obj-$(CONFIG_NET_BYPASS) += bypass.o
>diff --git a/net/core/bypass.c b/net/core/bypass.c
>new file mode 100644
>index 000000000000..7bde962ec3d4
>--- /dev/null
>+++ b/net/core/bypass.c
>@@ -0,0 +1,406 @@
>+// SPDX-License-Identifier: GPL-2.0
>+/* Copyright (c) 2018, Intel Corporation. */
>+
>+/* A common module to handle registrations and notifications for paravirtual
>+ * drivers to enable accelerated datapath and support VF live migration.
>+ *
>+ * The notifier and event handling code is based on netvsc driver.
>+ */
>+
>+#include <linux/netdevice.h>
>+#include <linux/etherdevice.h>
>+#include <linux/ethtool.h>
>+#include <linux/module.h>
>+#include <linux/slab.h>
>+#include <linux/netdevice.h>
>+#include <linux/netpoll.h>
>+#include <linux/rtnetlink.h>
>+#include <linux/if_vlan.h>
>+#include <net/sch_generic.h>
>+#include <uapi/linux/if_arp.h>
>+#include <net/bypass.h>
>+
>+static LIST_HEAD(bypass_list);
>+
>+static DEFINE_MUTEX(bypass_mutex);

Why mutex? Apparently you don't need to sleep while holding a lock.
Simple spinlock would do.


>+
>+struct bypass_instance *bypass_instance_alloc(struct net_device *dev)
>+{
>+	struct bypass_instance *bypass_instance;
>+
>+	bypass_instance = kzalloc(sizeof(*bypass_instance), GFP_KERNEL);
>+	if (!bypass_instance)
>+		return NULL;
>+
>+	dev_hold(dev);
>+	rcu_assign_pointer(bypass_instance->bypass_netdev, dev);
>+
>+	return bypass_instance;
>+}
>+
>+void bypass_instance_free(struct bypass_instance *bypass_instance)
>+{
>+	struct net_device *bypass_netdev;
>+
>+	bypass_netdev = rcu_dereference(bypass_instance->bypass_netdev);
>+
>+	dev_put(bypass_netdev);
>+	kfree(bypass_instance);
>+}
>+
>+static struct bypass_instance *bypass_get_instance_bymac(u8 *mac)
>+{
>+	struct bypass_instance *bypass_instance;
>+	struct net_device *bypass_netdev;
>+	struct bypass *bypass;
>+
>+	list_for_each_entry(bypass, &bypass_list, list) {
>+		mutex_lock(&bypass->lock);
>+		list_for_each_entry(bypass_instance, &bypass->instance_list,
>+				    list) {
>+			bypass_netdev =
>+				rcu_dereference(bypass_instance->bypass_netdev);
>+			if (ether_addr_equal(bypass_netdev->perm_addr, mac)) {
>+				mutex_unlock(&bypass->lock);
>+				goto out;
>+			}
>+		}
>+		mutex_unlock(&bypass->lock);
>+	}
>+
>+	bypass_instance = NULL;
>+out:
>+	return bypass_instance;
>+}
>+
>+static int bypass_register_child(struct net_device *child_netdev)
>+{
>+	struct bypass_instance *bypass_instance;
>+	struct bypass *bypass;
>+	struct net_device *bypass_netdev;
>+	int ret, orig_mtu;
>+
>+	ASSERT_RTNL();
>+
>+	mutex_lock(&bypass_mutex);
>+	bypass_instance = bypass_get_instance_bymac(child_netdev->perm_addr);
>+	if (!bypass_instance) {
>+		mutex_unlock(&bypass_mutex);
>+		goto done;
>+	}
>+
>+	bypass_netdev = rcu_dereference(bypass_instance->bypass_netdev);
>+	bypass = bypass_instance->bypass;
>+	mutex_unlock(&bypass_mutex);
>+
>+	if (!bypass->ops->register_child)
>+		goto done;
>+
>+	ret = bypass->ops->register_child(bypass_netdev, child_netdev);
>+	if (ret != 0)
>+		goto done;
>+
>+	ret = netdev_rx_handler_register(child_netdev,
>+					 bypass->ops->handle_frame,
>+					 bypass_netdev);
>+	if (ret != 0) {
>+		netdev_err(child_netdev,
>+			   "can not register bypass rx handler (err = %d)\n",
>+			   ret);
>+		goto rx_handler_failed;
>+	}
>+
>+	ret = netdev_upper_dev_link(child_netdev, bypass_netdev, NULL);
>+	if (ret != 0) {
>+		netdev_err(child_netdev,

No line-wrap is needed here and in other cases like this.


>+			   "can not set master device %s (err = %d)\n",
>+			   bypass_netdev->name, ret);
>+		goto upper_link_failed;
>+	}
>+
>+	child_netdev->flags |= IFF_SLAVE;

Don't reuse IFF_SLAVE. That is bonding-specific thing. I know that
netvsc uses it, it is wrong.
Please rather introduce:
IFF_BYPASS for master
and IFF_BYPASS_SLAVE for slaves.




>+
>+	if (netif_running(bypass_netdev)) {
>+		ret = dev_open(child_netdev);
>+		if (ret && (ret != -EBUSY)) {
>+			netdev_err(bypass_netdev,
>+				   "Opening child %s failed ret:%d\n",
>+				   child_netdev->name, ret);
>+			goto err_interface_up;
>+		}
>+	}
>+
>+	/* Align MTU of child with master */
>+	orig_mtu = child_netdev->mtu;
>+	ret = dev_set_mtu(child_netdev, bypass_netdev->mtu);
>+	if (ret != 0) {
>+		netdev_err(bypass_netdev,
>+			   "unable to change mtu of %s to %u register failed\n",
>+			   child_netdev->name, bypass_netdev->mtu);
>+		goto err_set_mtu;
>+	}
>+
>+	ret = bypass->ops->join_child(bypass_netdev, child_netdev);
>+	if (ret != 0)
>+		goto err_join;
>+
>+	call_netdevice_notifiers(NETDEV_JOIN, child_netdev);
>+
>+	goto done;
>+
>+err_join:
>+	dev_set_mtu(child_netdev, orig_mtu);
>+err_set_mtu:
>+	dev_close(child_netdev);
>+err_interface_up:
>+	netdev_upper_dev_unlink(child_netdev, bypass_netdev);
>+	child_netdev->flags &= ~IFF_SLAVE;
>+upper_link_failed:
>+	netdev_rx_handler_unregister(child_netdev);
>+rx_handler_failed:
>+	bypass->ops->unregister_child(bypass_netdev, child_netdev);
>+
>+done:
>+	return NOTIFY_DONE;
>+}
>+
>+int bypass_unregister_child(struct net_device *child_netdev)
>+{
>+	struct bypass_instance *bypass_instance;
>+	struct net_device *bypass_netdev;
>+	struct bypass *bypass;
>+	int ret;
>+
>+	ASSERT_RTNL();
>+
>+	mutex_lock(&bypass_mutex);
>+	bypass_instance = bypass_get_instance_bymac(child_netdev->perm_addr);
>+	if (!bypass_instance) {
>+		mutex_unlock(&bypass_mutex);
>+		goto done;
>+	}
>+
>+	bypass_netdev = rcu_dereference(bypass_instance->bypass_netdev);
>+	bypass = bypass_instance->bypass;
>+	mutex_unlock(&bypass_mutex);
>+
>+	ret = bypass->ops->release_child(bypass_netdev, child_netdev);
>+	if (ret != 0)
>+		goto done;
>+
>+	netdev_rx_handler_unregister(child_netdev);
>+	netdev_upper_dev_unlink(child_netdev, bypass_netdev);
>+	child_netdev->flags &= ~IFF_SLAVE;
>+
>+	if (!bypass->ops->unregister_child)
>+		goto done;
>+
>+	bypass->ops->unregister_child(bypass_netdev, child_netdev);
>+
>+done:
>+	return NOTIFY_DONE;
>+}
>+EXPORT_SYMBOL(bypass_unregister_child);

Please use "EXPORT_SYMBOL_GPL" for all exported symbols.


>+
>+static int bypass_update_link(struct net_device *child_netdev)
>+{
>+	struct bypass_instance *bypass_instance;
>+	struct net_device *bypass_netdev;
>+	struct bypass *bypass;
>+
>+	ASSERT_RTNL();
>+
>+	mutex_lock(&bypass_mutex);
>+	bypass_instance = bypass_get_instance_bymac(child_netdev->perm_addr);

You don't really need this lookup. The kernel knows about the master
device, you can just use netdev_master_upper_dev_get_rcu() to get it.


>+	if (!bypass_instance) {
>+		mutex_unlock(&bypass_mutex);
>+		goto done;
>+	}
>+
>+	bypass_netdev = rcu_dereference(bypass_instance->bypass_netdev);
>+	bypass = bypass_instance->bypass;
>+	mutex_unlock(&bypass_mutex);
>+
>+	if (!bypass->ops->update_link)
>+		goto done;
>+
>+	bypass->ops->update_link(bypass_netdev, child_netdev);
>+
>+done:
>+	return NOTIFY_DONE;
>+}
>+
>+static bool bypass_validate_child_dev(struct net_device *dev)
>+{
>+	/* Avoid non-Ethernet type devices */
>+	if (dev->type != ARPHRD_ETHER)
>+		return false;
>+
>+	/* Avoid Vlan dev with same MAC registering as VF */
>+	if (is_vlan_dev(dev))
>+		return false;
>+
>+	/* Avoid Bonding master dev with same MAC registering as child dev */
>+	if ((dev->priv_flags & IFF_BONDING) && (dev->flags & IFF_MASTER))
>+		return false;
>+
>+	return true;
>+}
>+
>+static int
>+bypass_event(struct notifier_block *this, unsigned long event, void *ptr)
>+{
>+	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
>+	struct bypass *bypass;
>+
>+	/* Skip Parent events */
>+	mutex_lock(&bypass_mutex);
>+	list_for_each_entry(bypass, &bypass_list, list) {
>+		if (event_dev->netdev_ops == bypass->netdev_ops) {
>+			mutex_unlock(&bypass_mutex);
>+			return NOTIFY_DONE;
>+		}

What you need instead of this is an identification helper
netif_is_bypass_master()
similar to
netif_is_team_master()
netif_is_bridge_master()
etc

 

>+	}
>+	mutex_unlock(&bypass_mutex);
>+
>+	if (!bypass_validate_child_dev(event_dev))
>+		return NOTIFY_DONE;
>+
>+	switch (event) {
>+	case NETDEV_REGISTER:
>+		return bypass_register_child(event_dev);
>+	case NETDEV_UNREGISTER:
>+		return bypass_unregister_child(event_dev);
>+	case NETDEV_UP:
>+	case NETDEV_DOWN:
>+	case NETDEV_CHANGE:
>+		return bypass_update_link(event_dev);
>+	default:
>+		return NOTIFY_DONE;
>+	}
>+}
>+
>+static struct notifier_block bypass_notifier = {
>+	.notifier_call = bypass_event,
>+};
>+
>+static void bypass_register_existing_child(struct net_device *bypass_netdev)
>+{
>+	struct net *net = dev_net(bypass_netdev);
>+	struct net_device *dev;
>+
>+	rtnl_lock();
>+	for_each_netdev(net, dev) {
>+		if (dev == bypass_netdev)
>+			continue;
>+		if (!bypass_validate_child_dev(dev))
>+			continue;
>+		if (ether_addr_equal(bypass_netdev->perm_addr, dev->perm_addr))
>+			bypass_register_child(dev);
>+	}
>+	rtnl_unlock();
>+}
>+
>+int bypass_register_instance(struct bypass *bypass, struct net_device *dev)
>+{
>+	struct bypass_instance *bypass_instance;

No need to allocate this instace here. You can just have is embedded
inside netdevice priv and pass pointer to it here. You can pass the
pointer back to the driver when you call ops as the driver can get priv
back by it.

I would also call it "struct bypass_master" and this function
"bypass_master_register".

It should contain the ops pointer too.


>+	struct net_device *bypass_netdev;
>+	int ret = 0;
>+
>+	mutex_lock(&bypass->lock);
>+	list_for_each_entry(bypass_instance, &bypass->instance_list, list) {
>+		bypass_netdev = rcu_dereference(bypass_instance->bypass_netdev);
>+		if (bypass_netdev == dev) {

This means the driver registered one netdev twice. That is a bug in
driver, so WARN_ON would be nice here to point that out.


>+			ret = -EEXIST;
>+			goto done;
>+		}
>+	}
>+
>+	bypass_instance = bypass_instance_alloc(dev);
>+	if (!bypass_instance) {
>+		ret = -ENOMEM;
>+		goto done;
>+	}
>+
>+	bypass_instance->bypass = bypass;
>+	list_add_tail(&bypass_instance->list, &bypass->instance_list);
>+
>+done:
>+	mutex_unlock(&bypass->lock);
>+	bypass_register_existing_child(dev);
>+	return ret;
>+}
>+EXPORT_SYMBOL(bypass_register_instance);
>+
>+int bypass_unregister_instance(struct bypass *bypass, struct net_device *dev)
>+{
>+	struct bypass_instance *bypass_instance;
>+	struct net_device *bypass_netdev;
>+	int ret = 0;
>+
>+	mutex_lock(&bypass->lock);
>+	list_for_each_entry(bypass_instance, &bypass->instance_list, list) {
>+		bypass_netdev = rcu_dereference(bypass_instance->bypass_netdev);
>+		if (bypass_netdev == dev) {
>+			list_del(&bypass_instance->list);
>+			bypass_instance_free(bypass_instance);
>+			goto done;
>+		}
>+	}
>+
>+	ret = -ENOENT;
>+done:
>+	mutex_unlock(&bypass->lock);
>+	return ret;
>+}
>+EXPORT_SYMBOL(bypass_unregister_instance);
>+
>+struct bypass *bypass_register_driver(const struct bypass_ops *ops,
>+				      const struct net_device_ops *netdev_ops)

I don't see why you need a list of drivers. What you need is just a list
of instances - bypass masters (probably to call them like that in the
code as well). Well, you can use the common netdevice list for that
purpose with the identification helper I mentioned above. Then you need
no lists and no mutexes/spinlocks.


>+{
>+	struct bypass *bypass;
>+
>+	bypass = kzalloc(sizeof(*bypass), GFP_KERNEL);
>+	if (!bypass)
>+		return NULL;
>+
>+	bypass->ops = ops;
>+	bypass->netdev_ops = netdev_ops;
>+	INIT_LIST_HEAD(&bypass->instance_list);
>+
>+	mutex_lock(&bypass_mutex);
>+	list_add_tail(&bypass->list, &bypass_list);
>+	mutex_unlock(&bypass_mutex);
>+
>+	return bypass;
>+}
>+EXPORT_SYMBOL_GPL(bypass_register_driver);
>+
>+void bypass_unregister_driver(struct bypass *bypass)
>+{
>+	mutex_lock(&bypass_mutex);
>+	list_del(&bypass->list);
>+	mutex_unlock(&bypass_mutex);
>+
>+	kfree(bypass);
>+}
>+EXPORT_SYMBOL_GPL(bypass_unregister_driver);
>+
>+static __init int
>+bypass_init(void)
>+{
>+	register_netdevice_notifier(&bypass_notifier);
>+
>+	return 0;
>+}
>+module_init(bypass_init);
>+
>+static __exit
>+void bypass_exit(void)
>+{
>+	unregister_netdevice_notifier(&bypass_notifier);
>+}
>+module_exit(bypass_exit);
>+
>+MODULE_DESCRIPTION("Bypass infrastructure/interface for Paravirtual drivers");
>+MODULE_LICENSE("GPL v2");
>-- 
>2.14.3
>

Powered by blists - more mailing lists