~hc/RK356X_SDK_RELEASE.git

..	..	@@ -1,10 +1,6 @@
	1	+// SPDX-License-Identifier: GPL-2.0-or-later
1	2	/*
2	3	* NET3 Protocol independent device support routines.
3		- *
4		- * This program is free software; you can redistribute it and/or
5		- * modify it under the terms of the GNU General Public License
6		- * as published by the Free Software Foundation; either version
7		- * 2 of the License, or (at your option) any later version.
8	4	*
9	5	* Derived from the non IP parts of dev.c 1.0.19
10	6	* Authors: Ross Biro
..	..	@@ -102,6 +98,7 @@
102	98	#include <net/busy_poll.h>
103	99	#include <linux/rtnetlink.h>
104	100	#include <linux/stat.h>
	101	+#include <net/dsa.h>
105	102	#include <net/dst.h>
106	103	#include <net/dst_metadata.h>
107	104	#include <net/pkt_sched.h>
..	..	@@ -132,7 +129,6 @@
132	129	#include <trace/events/napi.h>
133	130	#include <trace/events/net.h>
134	131	#include <trace/events/skb.h>
135		-#include <linux/pci.h>
136	132	#include <linux/inetdevice.h>
137	133	#include <linux/cpu_rmap.h>
138	134	#include <linux/static_key.h>
..	..	@@ -146,11 +142,15 @@
146	142	#include <linux/sctp.h>
147	143	#include <net/udp_tunnel.h>
148	144	#include <linux/net_namespace.h>
	145	+#include <linux/indirect_call_wrapper.h>
	146	+#include <net/devlink.h>
	147	+#include <linux/pm_runtime.h>
	148	+#include <linux/prandom.h>
	149	+#include <trace/hooks/net.h>
149	150
150	151	#include "net-sysfs.h"
151	152
152	153	#define MAX_GRO_SKBS 8
153		-#define MAX_NEST_DEV 8
154	154
155	155	/* This should be increased if a protocol with a bigger head is added. */
156	156	#define GRO_MAX_HEAD (MAX_HEADER + 128)
..	..	@@ -164,6 +164,9 @@
164	164	static int netif_rx_internal(struct sk_buff *skb);
165	165	static int call_netdevice_notifiers_info(unsigned long val,
166	166	struct netdev_notifier_info *info);
	167	+static int call_netdevice_notifiers_extack(unsigned long val,
	168	+ struct net_device *dev,
	169	+ struct netlink_ext_ack *extack);
167	170	static struct napi_struct *napi_by_id(unsigned int napi_id);
168	171
169	172	/*
..	..	@@ -219,15 +222,137 @@
219	222	static inline void rps_lock(struct softnet_data *sd)
220	223	{
221	224	#ifdef CONFIG_RPS
222		- raw_spin_lock(&sd->input_pkt_queue.raw_lock);
	225	+ spin_lock(&sd->input_pkt_queue.lock);
223	226	#endif
224	227	}
225	228
226	229	static inline void rps_unlock(struct softnet_data *sd)
227	230	{
228	231	#ifdef CONFIG_RPS
229		- raw_spin_unlock(&sd->input_pkt_queue.raw_lock);
	232	+ spin_unlock(&sd->input_pkt_queue.lock);
230	233	#endif
	234	+}
	235	+
	236	+static struct netdev_name_node netdev_name_node_alloc(struct net_device dev,
	237	+ const char *name)
	238	+{
	239	+ struct netdev_name_node *name_node;
	240	+
	241	+ name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
	242	+ if (!name_node)
	243	+ return NULL;
	244	+ INIT_HLIST_NODE(&name_node->hlist);
	245	+ name_node->dev = dev;
	246	+ name_node->name = name;
	247	+ return name_node;
	248	+}
	249	+
	250	+static struct netdev_name_node *
	251	+netdev_name_node_head_alloc(struct net_device *dev)
	252	+{
	253	+ struct netdev_name_node *name_node;
	254	+
	255	+ name_node = netdev_name_node_alloc(dev, dev->name);
	256	+ if (!name_node)
	257	+ return NULL;
	258	+ INIT_LIST_HEAD(&name_node->list);
	259	+ return name_node;
	260	+}
	261	+
	262	+static void netdev_name_node_free(struct netdev_name_node *name_node)
	263	+{
	264	+ kfree(name_node);
	265	+}
	266	+
	267	+static void netdev_name_node_add(struct net *net,
	268	+ struct netdev_name_node *name_node)
	269	+{
	270	+ hlist_add_head_rcu(&name_node->hlist,
	271	+ dev_name_hash(net, name_node->name));
	272	+}
	273	+
	274	+static void netdev_name_node_del(struct netdev_name_node *name_node)
	275	+{
	276	+ hlist_del_rcu(&name_node->hlist);
	277	+}
	278	+
	279	+static struct netdev_name_node netdev_name_node_lookup(struct net net,
	280	+ const char *name)
	281	+{
	282	+ struct hlist_head *head = dev_name_hash(net, name);
	283	+ struct netdev_name_node *name_node;
	284	+
	285	+ hlist_for_each_entry(name_node, head, hlist)
	286	+ if (!strcmp(name_node->name, name))
	287	+ return name_node;
	288	+ return NULL;
	289	+}
	290	+
	291	+static struct netdev_name_node netdev_name_node_lookup_rcu(struct net net,
	292	+ const char *name)
	293	+{
	294	+ struct hlist_head *head = dev_name_hash(net, name);
	295	+ struct netdev_name_node *name_node;
	296	+
	297	+ hlist_for_each_entry_rcu(name_node, head, hlist)
	298	+ if (!strcmp(name_node->name, name))
	299	+ return name_node;
	300	+ return NULL;
	301	+}
	302	+
	303	+int netdev_name_node_alt_create(struct net_device dev, const char name)
	304	+{
	305	+ struct netdev_name_node *name_node;
	306	+ struct net *net = dev_net(dev);
	307	+
	308	+ name_node = netdev_name_node_lookup(net, name);
	309	+ if (name_node)
	310	+ return -EEXIST;
	311	+ name_node = netdev_name_node_alloc(dev, name);
	312	+ if (!name_node)
	313	+ return -ENOMEM;
	314	+ netdev_name_node_add(net, name_node);
	315	+ /* The node that holds dev->name acts as a head of per-device list. */
	316	+ list_add_tail(&name_node->list, &dev->name_node->list);
	317	+
	318	+ return 0;
	319	+}
	320	+EXPORT_SYMBOL(netdev_name_node_alt_create);
	321	+
	322	+static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
	323	+{
	324	+ list_del(&name_node->list);
	325	+ netdev_name_node_del(name_node);
	326	+ kfree(name_node->name);
	327	+ netdev_name_node_free(name_node);
	328	+}
	329	+
	330	+int netdev_name_node_alt_destroy(struct net_device dev, const char name)
	331	+{
	332	+ struct netdev_name_node *name_node;
	333	+ struct net *net = dev_net(dev);
	334	+
	335	+ name_node = netdev_name_node_lookup(net, name);
	336	+ if (!name_node)
	337	+ return -ENOENT;
	338	+ /* lookup might have found our primary name or a name belonging
	339	+ * to another device.
	340	+ */
	341	+ if (name_node == dev->name_node \|\| name_node->dev != dev)
	342	+ return -EINVAL;
	343	+
	344	+ __netdev_name_node_alt_destroy(name_node);
	345	+
	346	+ return 0;
	347	+}
	348	+EXPORT_SYMBOL(netdev_name_node_alt_destroy);
	349	+
	350	+static void netdev_name_node_alt_flush(struct net_device *dev)
	351	+{
	352	+ struct netdev_name_node name_node, tmp;
	353	+
	354	+ list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
	355	+ __netdev_name_node_alt_destroy(name_node);
231	356	}
232	357
233	358	/* Device list insertion */
..	..	@@ -239,7 +364,7 @@
239	364
240	365	write_lock_bh(&dev_base_lock);
241	366	list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
242		- hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
	367	+ netdev_name_node_add(net, dev->name_node);
243	368	hlist_add_head_rcu(&dev->index_hlist,
244	369	dev_index_hash(net, dev->ifindex));
245	370	write_unlock_bh(&dev_base_lock);
..	..	@@ -257,7 +382,7 @@
257	382	/* Unlink dev from the device chain */
258	383	write_lock_bh(&dev_base_lock);
259	384	list_del_rcu(&dev->dev_list);
260		- hlist_del_rcu(&dev->name_hlist);
	385	+ netdev_name_node_del(dev->name_node);
261	386	hlist_del_rcu(&dev->index_hlist);
262	387	write_unlock_bh(&dev_base_lock);
263	388
..	..	@@ -355,6 +480,7 @@
355	480	unsigned short dev_type)
356	481	{
357	482	}
	483	+
358	484	static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
359	485	{
360	486	}
..	..	@@ -385,6 +511,12 @@
385	511
386	512	static inline struct list_head ptype_head(const struct packet_type pt)
387	513	{
	514	+ struct list_head vendor_pt = { .next = NULL, };
	515	+
	516	+ trace_android_vh_ptype_head(pt, &vendor_pt);
	517	+ if (vendor_pt.next)
	518	+ return vendor_pt.next;
	519	+
388	520	if (pt->type == htons(ETH_P_ALL))
389	521	return pt->dev ? &pt->dev->ptype_all : &ptype_all;
390	522	else
..	..	@@ -735,14 +867,10 @@
735	867
736	868	struct net_device __dev_get_by_name(struct net net, const char *name)
737	869	{
738		- struct net_device *dev;
739		- struct hlist_head *head = dev_name_hash(net, name);
	870	+ struct netdev_name_node *node_name;
740	871
741		- hlist_for_each_entry(dev, head, name_hlist)
742		- if (!strncmp(dev->name, name, IFNAMSIZ))
743		- return dev;
744		-
745		- return NULL;
	872	+ node_name = netdev_name_node_lookup(net, name);
	873	+ return node_name ? node_name->dev : NULL;
746	874	}
747	875	EXPORT_SYMBOL(__dev_get_by_name);
748	876
..	..	@@ -760,14 +888,10 @@
760	888
761	889	struct net_device dev_get_by_name_rcu(struct net net, const char *name)
762	890	{
763		- struct net_device *dev;
764		- struct hlist_head *head = dev_name_hash(net, name);
	891	+ struct netdev_name_node *node_name;
765	892
766		- hlist_for_each_entry_rcu(dev, head, name_hlist)
767		- if (!strncmp(dev->name, name, IFNAMSIZ))
768		- return dev;
769		-
770		- return NULL;
	893	+ node_name = netdev_name_node_lookup_rcu(net, name);
	894	+ return node_name ? node_name->dev : NULL;
771	895	}
772	896	EXPORT_SYMBOL(dev_get_by_name_rcu);
773	897
..	..	@@ -1015,7 +1139,7 @@
1015	1139	* @name: name string
1016	1140	*
1017	1141	* Network device names need to be valid file names to
1018		- * to allow sysfs to work. We also disallow any kind of
	1142	+ * allow sysfs to work. We also disallow any kind of
1019	1143	* whitespace.
1020	1144	*/
1021	1145	bool dev_valid_name(const char *name)
..	..	@@ -1078,6 +1202,18 @@
1078	1202	return -ENOMEM;
1079	1203
1080	1204	for_each_netdev(net, d) {
	1205	+ struct netdev_name_node *name_node;
	1206	+ list_for_each_entry(name_node, &d->name_node->list, list) {
	1207	+ if (!sscanf(name_node->name, name, &i))
	1208	+ continue;
	1209	+ if (i < 0 \|\| i >= max_netdevices)
	1210	+ continue;
	1211	+
	1212	+ /* avoid cases where sscanf is not exact inverse of printf */
	1213	+ snprintf(buf, IFNAMSIZ, name, i);
	1214	+ if (!strncmp(buf, name_node->name, IFNAMSIZ))
	1215	+ set_bit(i, inuse);
	1216	+ }
1081	1217	if (!sscanf(d->name, name, &i))
1082	1218	continue;
1083	1219	if (i < 0 \|\| i >= max_netdevices)
..	..	@@ -1138,8 +1274,8 @@
1138	1274	}
1139	1275	EXPORT_SYMBOL(dev_alloc_name);
1140	1276
1141		-int dev_get_valid_name(struct net net, struct net_device dev,
1142		- const char *name)
	1277	+static int dev_get_valid_name(struct net net, struct net_device dev,
	1278	+ const char *name)
1143	1279	{
1144	1280	BUG_ON(!net);
1145	1281
..	..	@@ -1155,7 +1291,6 @@
1155	1291
1156	1292	return 0;
1157	1293	}
1158		-EXPORT_SYMBOL(dev_get_valid_name);
1159	1294
1160	1295	/**
1161	1296	* dev_change_name - change name of a device
..	..	@@ -1229,13 +1364,13 @@
1229	1364	netdev_adjacent_rename_links(dev, oldname);
1230	1365
1231	1366	write_lock_bh(&dev_base_lock);
1232		- hlist_del_rcu(&dev->name_hlist);
	1367	+ netdev_name_node_del(dev->name_node);
1233	1368	write_unlock_bh(&dev_base_lock);
1234	1369
1235	1370	synchronize_rcu();
1236	1371
1237	1372	write_lock_bh(&dev_base_lock);
1238		- hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
	1373	+ netdev_name_node_add(net, dev->name_node);
1239	1374	write_unlock_bh(&dev_base_lock);
1240	1375
1241	1376	ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
..	..	@@ -1285,8 +1420,8 @@
1285	1420	}
1286	1421
1287	1422	mutex_lock(&ifalias_mutex);
1288		- rcu_swap_protected(dev->ifalias, new_alias,
1289		- mutex_is_locked(&ifalias_mutex));
	1423	+ new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
	1424	+ mutex_is_locked(&ifalias_mutex));
1290	1425	mutex_unlock(&ifalias_mutex);
1291	1426
1292	1427	if (new_alias)
..	..	@@ -1372,15 +1507,20 @@
1372	1507	}
1373	1508	EXPORT_SYMBOL(netdev_notify_peers);
1374	1509
1375		-static int __dev_open(struct net_device *dev)
	1510	+static int __dev_open(struct net_device dev, struct netlink_ext_ack extack)
1376	1511	{
1377	1512	const struct net_device_ops *ops = dev->netdev_ops;
1378	1513	int ret;
1379	1514
1380	1515	ASSERT_RTNL();
1381	1516
1382		- if (!netif_device_present(dev))
1383		- return -ENODEV;
	1517	+ if (!netif_device_present(dev)) {
	1518	+ /* may be detached because parent is runtime-suspended */
	1519	+ if (dev->dev.parent)
	1520	+ pm_runtime_resume(dev->dev.parent);
	1521	+ if (!netif_device_present(dev))
	1522	+ return -ENODEV;
	1523	+ }
1384	1524
1385	1525	/* Block netpoll from trying to do any rx path servicing.
1386	1526	* If we don't do this there is a chance ndo_poll_controller
..	..	@@ -1388,7 +1528,7 @@
1388	1528	*/
1389	1529	netpoll_poll_disable(dev);
1390	1530
1391		- ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
	1531	+ ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1392	1532	ret = notifier_to_errno(ret);
1393	1533	if (ret)
1394	1534	return ret;
..	..	@@ -1417,7 +1557,8 @@
1417	1557
1418	1558	/**
1419	1559	* dev_open - prepare an interface for use.
1420		- * @dev: device to open
	1560	+ * @dev: device to open
	1561	+ * @extack: netlink extended ack
1421	1562	*
1422	1563	* Takes a device from down to up state. The device's private open
1423	1564	* function is invoked and then the multicast lists are loaded. Finally
..	..	@@ -1427,14 +1568,14 @@
1427	1568	* Calling this function on an active interface is a nop. On a failure
1428	1569	* a negative errno code is returned.
1429	1570	*/
1430		-int dev_open(struct net_device *dev)
	1571	+int dev_open(struct net_device dev, struct netlink_ext_ack extack)
1431	1572	{
1432	1573	int ret;
1433	1574
1434	1575	if (dev->flags & IFF_UP)
1435	1576	return 0;
1436	1577
1437		- ret = __dev_open(dev);
	1578	+ ret = __dev_open(dev, extack);
1438	1579	if (ret < 0)
1439	1580	return ret;
1440	1581
..	..	@@ -1596,6 +1737,7 @@
1596	1737	N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1597	1738	N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1598	1739	N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
	1740	+ N(PRE_CHANGEADDR)
1599	1741	}
1600	1742	#undef N
1601	1743	return "UNKNOWN_NETDEV_EVENT";
..	..	@@ -1610,6 +1752,62 @@
1610	1752	};
1611	1753
1612	1754	return nb->notifier_call(nb, val, &info);
	1755	+}
	1756	+
	1757	+static int call_netdevice_register_notifiers(struct notifier_block *nb,
	1758	+ struct net_device *dev)
	1759	+{
	1760	+ int err;
	1761	+
	1762	+ err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
	1763	+ err = notifier_to_errno(err);
	1764	+ if (err)
	1765	+ return err;
	1766	+
	1767	+ if (!(dev->flags & IFF_UP))
	1768	+ return 0;
	1769	+
	1770	+ call_netdevice_notifier(nb, NETDEV_UP, dev);
	1771	+ return 0;
	1772	+}
	1773	+
	1774	+static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
	1775	+ struct net_device *dev)
	1776	+{
	1777	+ if (dev->flags & IFF_UP) {
	1778	+ call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
	1779	+ dev);
	1780	+ call_netdevice_notifier(nb, NETDEV_DOWN, dev);
	1781	+ }
	1782	+ call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
	1783	+}
	1784	+
	1785	+static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
	1786	+ struct net *net)
	1787	+{
	1788	+ struct net_device *dev;
	1789	+ int err;
	1790	+
	1791	+ for_each_netdev(net, dev) {
	1792	+ err = call_netdevice_register_notifiers(nb, dev);
	1793	+ if (err)
	1794	+ goto rollback;
	1795	+ }
	1796	+ return 0;
	1797	+
	1798	+rollback:
	1799	+ for_each_netdev_continue_reverse(net, dev)
	1800	+ call_netdevice_unregister_notifiers(nb, dev);
	1801	+ return err;
	1802	+}
	1803	+
	1804	+static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
	1805	+ struct net *net)
	1806	+{
	1807	+ struct net_device *dev;
	1808	+
	1809	+ for_each_netdev(net, dev)
	1810	+ call_netdevice_unregister_notifiers(nb, dev);
1613	1811	}
1614	1812
1615	1813	static int dev_boot_phase = 1;
..	..	@@ -1630,8 +1828,6 @@
1630	1828
1631	1829	int register_netdevice_notifier(struct notifier_block *nb)
1632	1830	{
1633		- struct net_device *dev;
1634		- struct net_device *last;
1635	1831	struct net *net;
1636	1832	int err;
1637	1833
..	..	@@ -1644,17 +1840,9 @@
1644	1840	if (dev_boot_phase)
1645	1841	goto unlock;
1646	1842	for_each_net(net) {
1647		- for_each_netdev(net, dev) {
1648		- err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1649		- err = notifier_to_errno(err);
1650		- if (err)
1651		- goto rollback;
1652		-
1653		- if (!(dev->flags & IFF_UP))
1654		- continue;
1655		-
1656		- call_netdevice_notifier(nb, NETDEV_UP, dev);
1657		- }
	1843	+ err = call_netdevice_register_net_notifiers(nb, net);
	1844	+ if (err)
	1845	+ goto rollback;
1658	1846	}
1659	1847
1660	1848	unlock:
..	..	@@ -1663,22 +1851,9 @@
1663	1851	return err;
1664	1852
1665	1853	rollback:
1666		- last = dev;
1667		- for_each_net(net) {
1668		- for_each_netdev(net, dev) {
1669		- if (dev == last)
1670		- goto outroll;
	1854	+ for_each_net_continue_reverse(net)
	1855	+ call_netdevice_unregister_net_notifiers(nb, net);
1671	1856
1672		- if (dev->flags & IFF_UP) {
1673		- call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1674		- dev);
1675		- call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1676		- }
1677		- call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1678		- }
1679		- }
1680		-
1681		-outroll:
1682	1857	raw_notifier_chain_unregister(&netdev_chain, nb);
1683	1858	goto unlock;
1684	1859	}
..	..	@@ -1700,7 +1875,6 @@
1700	1875
1701	1876	int unregister_netdevice_notifier(struct notifier_block *nb)
1702	1877	{
1703		- struct net_device *dev;
1704	1878	struct net *net;
1705	1879	int err;
1706	1880
..	..	@@ -1711,22 +1885,147 @@
1711	1885	if (err)
1712	1886	goto unlock;
1713	1887
1714		- for_each_net(net) {
1715		- for_each_netdev(net, dev) {
1716		- if (dev->flags & IFF_UP) {
1717		- call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1718		- dev);
1719		- call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1720		- }
1721		- call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1722		- }
1723		- }
	1888	+ for_each_net(net)
	1889	+ call_netdevice_unregister_net_notifiers(nb, net);
	1890	+
1724	1891	unlock:
1725	1892	rtnl_unlock();
1726	1893	up_write(&pernet_ops_rwsem);
1727	1894	return err;
1728	1895	}
1729	1896	EXPORT_SYMBOL(unregister_netdevice_notifier);
	1897	+
	1898	+static int __register_netdevice_notifier_net(struct net *net,
	1899	+ struct notifier_block *nb,
	1900	+ bool ignore_call_fail)
	1901	+{
	1902	+ int err;
	1903	+
	1904	+ err = raw_notifier_chain_register(&net->netdev_chain, nb);
	1905	+ if (err)
	1906	+ return err;
	1907	+ if (dev_boot_phase)
	1908	+ return 0;
	1909	+
	1910	+ err = call_netdevice_register_net_notifiers(nb, net);
	1911	+ if (err && !ignore_call_fail)
	1912	+ goto chain_unregister;
	1913	+
	1914	+ return 0;
	1915	+
	1916	+chain_unregister:
	1917	+ raw_notifier_chain_unregister(&net->netdev_chain, nb);
	1918	+ return err;
	1919	+}
	1920	+
	1921	+static int __unregister_netdevice_notifier_net(struct net *net,
	1922	+ struct notifier_block *nb)
	1923	+{
	1924	+ int err;
	1925	+
	1926	+ err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
	1927	+ if (err)
	1928	+ return err;
	1929	+
	1930	+ call_netdevice_unregister_net_notifiers(nb, net);
	1931	+ return 0;
	1932	+}
	1933	+
	1934	+/**
	1935	+ * register_netdevice_notifier_net - register a per-netns network notifier block
	1936	+ * @net: network namespace
	1937	+ * @nb: notifier
	1938	+ *
	1939	+ * Register a notifier to be called when network device events occur.
	1940	+ * The notifier passed is linked into the kernel structures and must
	1941	+ * not be reused until it has been unregistered. A negative errno code
	1942	+ * is returned on a failure.
	1943	+ *
	1944	+ * When registered all registration and up events are replayed
	1945	+ * to the new notifier to allow device to have a race free
	1946	+ * view of the network device list.
	1947	+ */
	1948	+
	1949	+int register_netdevice_notifier_net(struct net net, struct notifier_block nb)
	1950	+{
	1951	+ int err;
	1952	+
	1953	+ rtnl_lock();
	1954	+ err = __register_netdevice_notifier_net(net, nb, false);
	1955	+ rtnl_unlock();
	1956	+ return err;
	1957	+}
	1958	+EXPORT_SYMBOL(register_netdevice_notifier_net);
	1959	+
	1960	+/**
	1961	+ * unregister_netdevice_notifier_net - unregister a per-netns
	1962	+ * network notifier block
	1963	+ * @net: network namespace
	1964	+ * @nb: notifier
	1965	+ *
	1966	+ * Unregister a notifier previously registered by
	1967	+ * register_netdevice_notifier(). The notifier is unlinked into the
	1968	+ * kernel structures and may then be reused. A negative errno code
	1969	+ * is returned on a failure.
	1970	+ *
	1971	+ * After unregistering unregister and down device events are synthesized
	1972	+ * for all devices on the device list to the removed notifier to remove
	1973	+ * the need for special case cleanup code.
	1974	+ */
	1975	+
	1976	+int unregister_netdevice_notifier_net(struct net *net,
	1977	+ struct notifier_block *nb)
	1978	+{
	1979	+ int err;
	1980	+
	1981	+ rtnl_lock();
	1982	+ err = __unregister_netdevice_notifier_net(net, nb);
	1983	+ rtnl_unlock();
	1984	+ return err;
	1985	+}
	1986	+EXPORT_SYMBOL(unregister_netdevice_notifier_net);
	1987	+
	1988	+int register_netdevice_notifier_dev_net(struct net_device *dev,
	1989	+ struct notifier_block *nb,
	1990	+ struct netdev_net_notifier *nn)
	1991	+{
	1992	+ int err;
	1993	+
	1994	+ rtnl_lock();
	1995	+ err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
	1996	+ if (!err) {
	1997	+ nn->nb = nb;
	1998	+ list_add(&nn->list, &dev->net_notifier_list);
	1999	+ }
	2000	+ rtnl_unlock();
	2001	+ return err;
	2002	+}
	2003	+EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
	2004	+
	2005	+int unregister_netdevice_notifier_dev_net(struct net_device *dev,
	2006	+ struct notifier_block *nb,
	2007	+ struct netdev_net_notifier *nn)
	2008	+{
	2009	+ int err;
	2010	+
	2011	+ rtnl_lock();
	2012	+ list_del(&nn->list);
	2013	+ err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
	2014	+ rtnl_unlock();
	2015	+ return err;
	2016	+}
	2017	+EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
	2018	+
	2019	+static void move_netdevice_notifiers_dev_net(struct net_device *dev,
	2020	+ struct net *net)
	2021	+{
	2022	+ struct netdev_net_notifier *nn;
	2023	+
	2024	+ list_for_each_entry(nn, &dev->net_notifier_list, list) {
	2025	+ __unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
	2026	+ __register_netdevice_notifier_net(net, nn->nb, true);
	2027	+ }
	2028	+}
1730	2029
1731	2030	/**
1732	2031	* call_netdevice_notifiers_info - call all network notifier blocks
..	..	@@ -1740,8 +2039,31 @@
1740	2039	static int call_netdevice_notifiers_info(unsigned long val,
1741	2040	struct netdev_notifier_info *info)
1742	2041	{
	2042	+ struct net *net = dev_net(info->dev);
	2043	+ int ret;
	2044	+
1743	2045	ASSERT_RTNL();
	2046	+
	2047	+ /* Run per-netns notifier block chain first, then run the global one.
	2048	+ * Hopefully, one day, the global one is going to be removed after
	2049	+ * all notifier block registrators get converted to be per-netns.
	2050	+ */
	2051	+ ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
	2052	+ if (ret & NOTIFY_STOP_MASK)
	2053	+ return ret;
1744	2054	return raw_notifier_call_chain(&netdev_chain, val, info);
	2055	+}
	2056	+
	2057	+static int call_netdevice_notifiers_extack(unsigned long val,
	2058	+ struct net_device *dev,
	2059	+ struct netlink_ext_ack *extack)
	2060	+{
	2061	+ struct netdev_notifier_info info = {
	2062	+ .dev = dev,
	2063	+ .extack = extack,
	2064	+ };
	2065	+
	2066	+ return call_netdevice_notifiers_info(val, &info);
1745	2067	}
1746	2068
1747	2069	/**
..	..	@@ -1755,11 +2077,7 @@
1755	2077
1756	2078	int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1757	2079	{
1758		- struct netdev_notifier_info info = {
1759		- .dev = dev,
1760		- };
1761		-
1762		- return call_netdevice_notifiers_info(val, &info);
	2080	+ return call_netdevice_notifiers_extack(val, dev, NULL);
1763	2081	}
1764	2082	EXPORT_SYMBOL(call_netdevice_notifiers);
1765	2083
..	..	@@ -1987,6 +2305,17 @@
1987	2305	return false;
1988	2306	}
1989	2307
	2308	+/**
	2309	+ * dev_nit_active - return true if any network interface taps are in use
	2310	+ *
	2311	+ * @dev: network device to check for the presence of taps
	2312	+ */
	2313	+bool dev_nit_active(struct net_device *dev)
	2314	+{
	2315	+ return !list_empty(&ptype_all) \|\| !list_empty(&dev->ptype_all);
	2316	+}
	2317	+EXPORT_SYMBOL_GPL(dev_nit_active);
	2318	+
1990	2319	/*
1991	2320	* Support routine. Sends outgoing frames to any network
1992	2321	* taps currently in use.
..	..	@@ -2002,6 +2331,9 @@
2002	2331	rcu_read_lock();
2003	2332	again:
2004	2333	list_for_each_entry_rcu(ptype, ptype_list, list) {
	2334	+ if (ptype->ignore_outgoing)
	2335	+ continue;
	2336	+
2005	2337	/* Never send packets back to the socket
2006	2338	* they originated from - MvS (miquels@drinkel.ow.org)
2007	2339	*/
..	..	@@ -2723,7 +3055,6 @@
2723	3055	sd->output_queue_tailp = &q->next_sched;
2724	3056	raise_softirq_irqoff(NET_TX_SOFTIRQ);
2725	3057	local_irq_restore(flags);
2726		- preempt_check_resched_rt();
2727	3058	}
2728	3059
2729	3060	void __netif_schedule(struct Qdisc *q)
..	..	@@ -2745,7 +3076,7 @@
2745	3076	void netif_schedule_queue(struct netdev_queue *txq)
2746	3077	{
2747	3078	rcu_read_lock();
2748		- if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
	3079	+ if (!netif_xmit_stopped(txq)) {
2749	3080	struct Qdisc *q = rcu_dereference(txq->qdisc);
2750	3081
2751	3082	__netif_schedule(q);
..	..	@@ -2786,7 +3117,6 @@
2786	3117	__this_cpu_write(softnet_data.completion_queue, skb);
2787	3118	raise_softirq_irqoff(NET_TX_SOFTIRQ);
2788	3119	local_irq_restore(flags);
2789		- preempt_check_resched_rt();
2790	3120	}
2791	3121	EXPORT_SYMBOL(__dev_kfree_skb_irq);
2792	3122
..	..	@@ -2883,12 +3213,10 @@
2883	3213	else
2884	3214	name = netdev_name(dev);
2885	3215	}
2886		- WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2887		- "gso_type=%d ip_summed=%d\n",
	3216	+ skb_dump(KERN_WARNING, skb, false);
	3217	+ WARN(1, "%s: caps=(%pNF, %pNF)\n",
2888	3218	name, dev ? &dev->features : &null_features,
2889		- skb->sk ? &skb->sk->sk_route_caps : &null_features,
2890		- skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2891		- skb_shinfo(skb)->gso_type, skb->ip_summed);
	3219	+ skb->sk ? &skb->sk->sk_route_caps : &null_features);
2892	3220	}
2893	3221
2894	3222	/*
..	..	@@ -2918,18 +3246,19 @@
2918	3246	}
2919	3247
2920	3248	offset = skb_checksum_start_offset(skb);
2921		- BUG_ON(offset >= skb_headlen(skb));
	3249	+ ret = -EINVAL;
	3250	+ if (WARN_ON_ONCE(offset >= skb_headlen(skb)))
	3251	+ goto out;
	3252	+
2922	3253	csum = skb_checksum(skb, offset, skb->len - offset, 0);
2923	3254
2924	3255	offset += skb->csum_offset;
2925		- BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
	3256	+ if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb)))
	3257	+ goto out;
2926	3258
2927		- if (skb_cloned(skb) &&
2928		- !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2929		- ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2930		- if (ret)
2931		- goto out;
2932		- }
	3259	+ ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
	3260	+ if (ret)
	3261	+ goto out;
2933	3262
2934	3263	(__sum16 )(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2935	3264	out_set_summed:
..	..	@@ -2964,12 +3293,11 @@
2964	3293	ret = -EINVAL;
2965	3294	goto out;
2966	3295	}
2967		- if (skb_cloned(skb) &&
2968		- !skb_clone_writable(skb, offset + sizeof(__le32))) {
2969		- ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2970		- if (ret)
2971		- goto out;
2972		- }
	3296	+
	3297	+ ret = skb_ensure_writable(skb, offset + sizeof(__le32));
	3298	+ if (ret)
	3299	+ goto out;
	3300	+
2973	3301	crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
2974	3302	skb->len - start, ~(__u32)0,
2975	3303	crc32c_csum_stub));
..	..	@@ -3054,7 +3382,7 @@
3054	3382	* It may return NULL if the skb requires no segmentation. This is
3055	3383	* only possible when GSO is used for verifying header integrity.
3056	3384	*
3057		- * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
	3385	+ * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3058	3386	*/
3059	3387	struct sk_buff __skb_gso_segment(struct sk_buff skb,
3060	3388	netdev_features_t features, bool tx_path)
..	..	@@ -3083,7 +3411,7 @@
3083	3411	features &= ~NETIF_F_GSO_PARTIAL;
3084	3412	}
3085	3413
3086		- BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
	3414	+ BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3087	3415	sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3088	3416
3089	3417	SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
..	..	@@ -3094,7 +3422,7 @@
3094	3422
3095	3423	segs = skb_mac_gso_segment(skb, features);
3096	3424
3097		- if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
	3425	+ if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3098	3426	skb_warn_bad_offload(skb);
3099	3427
3100	3428	return segs;
..	..	@@ -3103,10 +3431,11 @@
3103	3431
3104	3432	/* Take action when hardware reception checksum errors are detected. */
3105	3433	#ifdef CONFIG_BUG
3106		-void netdev_rx_csum_fault(struct net_device *dev)
	3434	+void netdev_rx_csum_fault(struct net_device dev, struct sk_buff skb)
3107	3435	{
3108	3436	if (net_ratelimit()) {
3109	3437	pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
	3438	+ skb_dump(KERN_ERR, skb, true);
3110	3439	dump_stack();
3111	3440	}
3112	3441	}
..	..	@@ -3156,10 +3485,9 @@
3156	3485	static netdev_features_t harmonize_features(struct sk_buff *skb,
3157	3486	netdev_features_t features)
3158	3487	{
3159		- int tmp;
3160	3488	__be16 type;
3161	3489
3162		- type = skb_network_protocol(skb, &tmp);
	3490	+ type = skb_network_protocol(skb, NULL);
3163	3491	features = net_mpls_features(skb, features, type);
3164	3492
3165	3493	if (skb->ip_summed != CHECKSUM_NONE &&
..	..	@@ -3256,10 +3584,11 @@
3256	3584	unsigned int len;
3257	3585	int rc;
3258	3586
3259		- if (!list_empty(&ptype_all) \|\| !list_empty(&dev->ptype_all))
	3587	+ if (dev_nit_active(dev))
3260	3588	dev_queue_xmit_nit(skb, dev);
3261	3589
3262	3590	len = skb->len;
	3591	+ PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies);
3263	3592	trace_net_dev_start_xmit(skb, dev);
3264	3593	rc = netdev_start_xmit(skb, dev, txq, more);
3265	3594	trace_net_dev_xmit(skb, rc, dev, len);
..	..	@@ -3276,7 +3605,7 @@
3276	3605	while (skb) {
3277	3606	struct sk_buff *next = skb->next;
3278	3607
3279		- skb->next = NULL;
	3608	+ skb_mark_not_on_list(skb);
3280	3609	rc = xmit_one(skb, dev, txq, next != NULL);
3281	3610	if (unlikely(!dev_xmit_complete(rc))) {
3282	3611	skb->next = next;
..	..	@@ -3376,7 +3705,7 @@
3376	3705
3377	3706	for (; skb != NULL; skb = next) {
3378	3707	next = skb->next;
3379		- skb->next = NULL;
	3708	+ skb_mark_not_on_list(skb);
3380	3709
3381	3710	/* in case skb wont be segmented, point to itself */
3382	3711	skb->prev = skb;
..	..	@@ -3407,7 +3736,7 @@
3407	3736	/* To get more precise estimation of bytes sent on wire,
3408	3737	* we add to pkt_len the headers size of all segments
3409	3738	*/
3410		- if (shinfo->gso_size) {
	3739	+ if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3411	3740	unsigned int hdr_len;
3412	3741	u16 gso_segs = shinfo->gso_segs;
3413	3742
..	..	@@ -3451,13 +3780,9 @@
3451	3780	qdisc_calculate_pkt_len(skb, q);
3452	3781
3453	3782	if (q->flags & TCQ_F_NOLOCK) {
3454		- if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3455		- __qdisc_drop(skb, &to_free);
3456		- rc = NET_XMIT_DROP;
3457		- } else {
3458		- rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
	3783	+ rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
	3784	+ if (likely(!netif_xmit_frozen_or_stopped(txq)))
3459	3785	qdisc_run(q);
3460		- }
3461	3786
3462	3787	if (unlikely(to_free))
3463	3788	kfree_skb_list(to_free);
..	..	@@ -3470,11 +3795,7 @@
3470	3795	* This permits qdisc->running owner to get the lock more
3471	3796	* often and dequeue packets faster.
3472	3797	*/
3473		-#ifdef CONFIG_PREEMPT_RT_FULL
3474		- contended = true;
3475		-#else
3476	3798	contended = qdisc_is_running(q);
3477		-#endif
3478	3799	if (unlikely(contended))
3479	3800	spin_lock(&q->busylock);
3480	3801
..	..	@@ -3557,7 +3878,8 @@
3557	3878	skb_reset_mac_header(skb);
3558	3879	__skb_pull(skb, skb_network_offset(skb));
3559	3880	skb->pkt_type = PACKET_LOOPBACK;
3560		- skb->ip_summed = CHECKSUM_UNNECESSARY;
	3881	+ if (skb->ip_summed == CHECKSUM_NONE)
	3882	+ skb->ip_summed = CHECKSUM_UNNECESSARY;
3561	3883	WARN_ON(!skb_dst(skb));
3562	3884	skb_dst_force(skb);
3563	3885	netif_rx_ni(skb);
..	..	@@ -3576,6 +3898,7 @@
3576	3898	return skb;
3577	3899
3578	3900	/* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
	3901	+ qdisc_skb_cb(skb)->mru = 0;
3579	3902	mini_qdisc_bstats_cpu_update(miniq, skb);
3580	3903
3581	3904	switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
..	..	@@ -3676,23 +3999,21 @@
3676	3999	}
3677	4000
3678	4001	u16 dev_pick_tx_zero(struct net_device dev, struct sk_buff skb,
3679		- struct net_device *sb_dev,
3680		- select_queue_fallback_t fallback)
	4002	+ struct net_device *sb_dev)
3681	4003	{
3682	4004	return 0;
3683	4005	}
3684	4006	EXPORT_SYMBOL(dev_pick_tx_zero);
3685	4007
3686	4008	u16 dev_pick_tx_cpu_id(struct net_device dev, struct sk_buff skb,
3687		- struct net_device *sb_dev,
3688		- select_queue_fallback_t fallback)
	4009	+ struct net_device *sb_dev)
3689	4010	{
3690	4011	return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3691	4012	}
3692	4013	EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3693	4014
3694		-static u16 __netdev_pick_tx(struct net_device dev, struct sk_buff skb,
3695		- struct net_device *sb_dev)
	4015	+u16 netdev_pick_tx(struct net_device dev, struct sk_buff skb,
	4016	+ struct net_device *sb_dev)
3696	4017	{
3697	4018	struct sock *sk = skb->sk;
3698	4019	int queue_index = sk_tx_queue_get(sk);
..	..	@@ -3716,10 +4037,11 @@
3716	4037
3717	4038	return queue_index;
3718	4039	}
	4040	+EXPORT_SYMBOL(netdev_pick_tx);
3719	4041
3720		-struct netdev_queue netdev_pick_tx(struct net_device dev,
3721		- struct sk_buff *skb,
3722		- struct net_device *sb_dev)
	4042	+struct netdev_queue netdev_core_pick_tx(struct net_device dev,
	4043	+ struct sk_buff *skb,
	4044	+ struct net_device *sb_dev)
3723	4045	{
3724	4046	int queue_index = 0;
3725	4047
..	..	@@ -3734,10 +4056,9 @@
3734	4056	const struct net_device_ops *ops = dev->netdev_ops;
3735	4057
3736	4058	if (ops->ndo_select_queue)
3737		- queue_index = ops->ndo_select_queue(dev, skb, sb_dev,
3738		- __netdev_pick_tx);
	4059	+ queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
3739	4060	else
3740		- queue_index = __netdev_pick_tx(dev, skb, sb_dev);
	4061	+ queue_index = netdev_pick_tx(dev, skb, sb_dev);
3741	4062
3742	4063	queue_index = netdev_cap_txqueue(dev, queue_index);
3743	4064	}
..	..	@@ -3781,6 +4102,7 @@
3781	4102	bool again = false;
3782	4103
3783	4104	skb_reset_mac_header(skb);
	4105	+ skb_assert_len(skb);
3784	4106
3785	4107	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3786	4108	__skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
..	..	@@ -3811,7 +4133,7 @@
3811	4133	else
3812	4134	skb_dst_force(skb);
3813	4135
3814		- txq = netdev_pick_tx(dev, skb, sb_dev);
	4136	+ txq = netdev_core_pick_tx(dev, skb, sb_dev);
3815	4137	q = rcu_dereference_bh(txq->qdisc);
3816	4138
3817	4139	trace_net_dev_queue(skb);
..	..	@@ -3835,14 +4157,10 @@
3835	4157	if (dev->flags & IFF_UP) {
3836	4158	int cpu = smp_processor_id(); /* ok because BHs are off */
3837	4159
3838		-#ifdef CONFIG_PREEMPT_RT_FULL
3839		- if (READ_ONCE(txq->xmit_lock_owner) != current) {
3840		-#else
3841	4160	/* Other cpus might concurrently change txq->xmit_lock_owner
3842	4161	* to -1 or to their cpu id, but not to our id.
3843	4162	*/
3844	4163	if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
3845		-#endif
3846	4164	if (dev_xmit_recursion())
3847	4165	goto recursion_alert;
3848	4166
..	..	@@ -3850,6 +4168,7 @@
3850	4168	if (!skb)
3851	4169	goto out;
3852	4170
	4171	+ PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
3853	4172	HARD_TX_LOCK(dev, txq, cpu);
3854	4173
3855	4174	if (!netif_xmit_stopped(txq)) {
..	..	@@ -3897,7 +4216,7 @@
3897	4216	}
3898	4217	EXPORT_SYMBOL(dev_queue_xmit_accel);
3899	4218
3900		-int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
	4219	+int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
3901	4220	{
3902	4221	struct net_device *dev = skb->dev;
3903	4222	struct sk_buff *orig_skb = skb;
..	..	@@ -3915,6 +4234,7 @@
3915	4234
3916	4235	skb_set_queue_mapping(skb, queue_id);
3917	4236	txq = skb_get_tx_queue(dev, skb);
	4237	+ PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
3918	4238
3919	4239	local_bh_disable();
3920	4240
..	..	@@ -3926,17 +4246,13 @@
3926	4246	dev_xmit_recursion_dec();
3927	4247
3928	4248	local_bh_enable();
3929		-
3930		- if (!dev_xmit_complete(ret))
3931		- kfree_skb(skb);
3932		-
3933	4249	return ret;
3934	4250	drop:
3935	4251	atomic_long_inc(&dev->tx_dropped);
3936	4252	kfree_skb_list(skb);
3937	4253	return NET_XMIT_DROP;
3938	4254	}
3939		-EXPORT_SYMBOL(dev_direct_xmit);
	4255	+EXPORT_SYMBOL(__dev_direct_xmit);
3940	4256
3941	4257	/*************************************************************************
3942	4258	* Receiver routines
..	..	@@ -3954,6 +4270,8 @@
3954	4270	int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
3955	4271	int dev_rx_weight __read_mostly = 64;
3956	4272	int dev_tx_weight __read_mostly = 64;
	4273	+/* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
	4274	+int gro_normal_batch __read_mostly = 8;
3957	4275
3958	4276	/* Called with irq disabled */
3959	4277	static inline void ____napi_schedule(struct softnet_data *sd,
..	..	@@ -3971,9 +4289,9 @@
3971	4289	u32 rps_cpu_mask __read_mostly;
3972	4290	EXPORT_SYMBOL(rps_cpu_mask);
3973	4291
3974		-struct static_key rps_needed __read_mostly;
	4292	+struct static_key_false rps_needed __read_mostly;
3975	4293	EXPORT_SYMBOL(rps_needed);
3976		-struct static_key rfs_needed __read_mostly;
	4294	+struct static_key_false rfs_needed __read_mostly;
3977	4295	EXPORT_SYMBOL(rfs_needed);
3978	4296
3979	4297	static struct rps_dev_flow *
..	..	@@ -4204,7 +4522,7 @@
4204	4522	struct softnet_data *sd;
4205	4523	unsigned int old_flow, new_flow;
4206	4524
4207		- if (qlen < (netdev_max_backlog >> 1))
	4525	+ if (qlen < (READ_ONCE(netdev_max_backlog) >> 1))
4208	4526	return false;
4209	4527
4210	4528	sd = this_cpu_ptr(&softnet_data);
..	..	@@ -4252,7 +4570,7 @@
4252	4570	if (!netif_running(skb->dev))
4253	4571	goto drop;
4254	4572	qlen = skb_queue_len(&sd->input_pkt_queue);
4255		- if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
	4573	+ if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) {
4256	4574	if (qlen) {
4257	4575	enqueue:
4258	4576	__skb_queue_tail(&sd->input_pkt_queue, skb);
..	..	@@ -4277,7 +4595,6 @@
4277	4595	rps_unlock(sd);
4278	4596
4279	4597	local_irq_restore(flags);
4280		- preempt_check_resched_rt();
4281	4598
4282	4599	atomic_long_inc(&skb->dev->rx_dropped);
4283	4600	kfree_skb(skb);
..	..	@@ -4323,7 +4640,7 @@
4323	4640	/* Reinjected packets coming from act_mirred or similar should
4324	4641	* not get XDP generic processing.
4325	4642	*/
4326		- if (skb_is_tc_redirected(skb))
	4643	+ if (skb_is_redirected(skb))
4327	4644	return XDP_PASS;
4328	4645
4329	4646	/* XDP packets must be linear and must have sufficient headroom
..	..	@@ -4355,6 +4672,11 @@
4355	4672	xdp->data_meta = xdp->data;
4356	4673	xdp->data_end = xdp->data + hlen;
4357	4674	xdp->data_hard_start = skb->data - skb_headroom(skb);
	4675	+
	4676	+ /* SKB "head" area always have tailroom for skb_shared_info */
	4677	+ xdp->frame_sz = (void *)skb_end_pointer(skb) - xdp->data_hard_start;
	4678	+ xdp->frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
	4679	+
4358	4680	orig_data_end = xdp->data_end;
4359	4681	orig_data = xdp->data;
4360	4682	eth = (struct ethhdr *)xdp->data;
..	..	@@ -4378,14 +4700,11 @@
4378	4700	skb_reset_network_header(skb);
4379	4701	}
4380	4702
4381		- /* check if bpf_xdp_adjust_tail was used. it can only "shrink"
4382		- * pckt.
4383		- */
4384		- off = orig_data_end - xdp->data_end;
	4703	+ /* check if bpf_xdp_adjust_tail was used */
	4704	+ off = xdp->data_end - orig_data_end;
4385	4705	if (off != 0) {
4386	4706	skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4387		- skb->len -= off;
4388		-
	4707	+ skb->len += off; /* positive on grow, negative on shrink */
4389	4708	}
4390	4709
4391	4710	/* check if XDP changed eth hdr such SKB needs update */
..	..	@@ -4408,10 +4727,10 @@
4408	4727	break;
4409	4728	default:
4410	4729	bpf_warn_invalid_xdp_action(act);
4411		- /* fall through */
	4730	+ fallthrough;
4412	4731	case XDP_ABORTED:
4413	4732	trace_xdp_exception(skb->dev, xdp_prog, act);
4414		- /* fall through */
	4733	+ fallthrough;
4415	4734	case XDP_DROP:
4416	4735	do_drop:
4417	4736	kfree_skb(skb);
..	..	@@ -4431,7 +4750,7 @@
4431	4750	bool free_skb = true;
4432	4751	int cpu, rc;
4433	4752
4434		- txq = netdev_pick_tx(dev, skb, NULL);
	4753	+ txq = netdev_core_pick_tx(dev, skb, NULL);
4435	4754	cpu = smp_processor_id();
4436	4755	HARD_TX_LOCK(dev, txq, cpu);
4437	4756	if (!netif_xmit_stopped(txq)) {
..	..	@@ -4445,7 +4764,6 @@
4445	4764	kfree_skb(skb);
4446	4765	}
4447	4766	}
4448		-EXPORT_SYMBOL_GPL(generic_xdp_tx);
4449	4767
4450	4768	static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4451	4769
..	..	@@ -4483,16 +4801,16 @@
4483	4801	{
4484	4802	int ret;
4485	4803
4486		- net_timestamp_check(netdev_tstamp_prequeue, skb);
	4804	+ net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
4487	4805
4488	4806	trace_netif_rx(skb);
4489	4807
4490	4808	#ifdef CONFIG_RPS
4491		- if (static_key_false(&rps_needed)) {
	4809	+ if (static_branch_unlikely(&rps_needed)) {
4492	4810	struct rps_dev_flow voidflow, *rflow = &voidflow;
4493	4811	int cpu;
4494	4812
4495		- migrate_disable();
	4813	+ preempt_disable();
4496	4814	rcu_read_lock();
4497	4815
4498	4816	cpu = get_rps_cpu(skb->dev, skb, &rflow);
..	..	@@ -4502,14 +4820,14 @@
4502	4820	ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4503	4821
4504	4822	rcu_read_unlock();
4505		- migrate_enable();
	4823	+ preempt_enable();
4506	4824	} else
4507	4825	#endif
4508	4826	{
4509	4827	unsigned int qtail;
4510	4828
4511		- ret = enqueue_to_backlog(skb, get_cpu_light(), &qtail);
4512		- put_cpu_light();
	4829	+ ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
	4830	+ put_cpu();
4513	4831	}
4514	4832	return ret;
4515	4833	}
..	..	@@ -4531,9 +4849,14 @@
4531	4849
4532	4850	int netif_rx(struct sk_buff *skb)
4533	4851	{
	4852	+ int ret;
	4853	+
4534	4854	trace_netif_rx_entry(skb);
4535	4855
4536		- return netif_rx_internal(skb);
	4856	+ ret = netif_rx_internal(skb);
	4857	+ trace_netif_rx_exit(ret);
	4858	+
	4859	+ return ret;
4537	4860	}
4538	4861	EXPORT_SYMBOL(netif_rx);
4539	4862
..	..	@@ -4543,13 +4866,31 @@
4543	4866
4544	4867	trace_netif_rx_ni_entry(skb);
4545	4868
4546		- local_bh_disable();
	4869	+ preempt_disable();
4547	4870	err = netif_rx_internal(skb);
4548		- local_bh_enable();
	4871	+ if (local_softirq_pending())
	4872	+ do_softirq();
	4873	+ preempt_enable();
	4874	+ trace_netif_rx_ni_exit(err);
4549	4875
4550	4876	return err;
4551	4877	}
4552	4878	EXPORT_SYMBOL(netif_rx_ni);
	4879	+
	4880	+int netif_rx_any_context(struct sk_buff *skb)
	4881	+{
	4882	+ /*
	4883	+ * If invoked from contexts which do not invoke bottom half
	4884	+ * processing either at return from interrupt or when softrqs are
	4885	+ * reenabled, use netif_rx_ni() which invokes bottomhalf processing
	4886	+ * directly.
	4887	+ */
	4888	+ if (in_interrupt())
	4889	+ return netif_rx(skb);
	4890	+ else
	4891	+ return netif_rx_ni(skb);
	4892	+}
	4893	+EXPORT_SYMBOL(netif_rx_any_context);
4553	4894
4554	4895	static __latent_entropy void net_tx_action(struct softirq_action *h)
4555	4896	{
..	..	@@ -4592,25 +4933,43 @@
4592	4933	sd->output_queue_tailp = &sd->output_queue;
4593	4934	local_irq_enable();
4594	4935
	4936	+ rcu_read_lock();
	4937	+
4595	4938	while (head) {
4596	4939	struct Qdisc *q = head;
4597	4940	spinlock_t *root_lock = NULL;
4598	4941
4599	4942	head = head->next_sched;
4600	4943
4601		- if (!(q->flags & TCQ_F_NOLOCK)) {
4602		- root_lock = qdisc_lock(q);
4603		- spin_lock(root_lock);
4604		- }
4605	4944	/* We need to make sure head->next_sched is read
4606	4945	* before clearing __QDISC_STATE_SCHED
4607	4946	*/
4608	4947	smp_mb__before_atomic();
	4948	+
	4949	+ if (!(q->flags & TCQ_F_NOLOCK)) {
	4950	+ root_lock = qdisc_lock(q);
	4951	+ spin_lock(root_lock);
	4952	+ } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
	4953	+ &q->state))) {
	4954	+ /* There is a synchronize_net() between
	4955	+ * STATE_DEACTIVATED flag being set and
	4956	+ * qdisc_reset()/some_qdisc_is_busy() in
	4957	+ * dev_deactivate(), so we can safely bail out
	4958	+ * early here to avoid data race between
	4959	+ * qdisc_deactivate() and some_qdisc_is_busy()
	4960	+ * for lockless qdisc.
	4961	+ */
	4962	+ clear_bit(__QDISC_STATE_SCHED, &q->state);
	4963	+ continue;
	4964	+ }
	4965	+
4609	4966	clear_bit(__QDISC_STATE_SCHED, &q->state);
4610	4967	qdisc_run(q);
4611	4968	if (root_lock)
4612	4969	spin_unlock(root_lock);
4613	4970	}
	4971	+
	4972	+ rcu_read_unlock();
4614	4973	}
4615	4974
4616	4975	xfrm_dev_backlog(sd);
..	..	@@ -4625,7 +4984,7 @@
4625	4984
4626	4985	static inline struct sk_buff *
4627	4986	sch_handle_ingress(struct sk_buff skb, struct packet_type pt_prev, int ret,
4628		- struct net_device *orig_dev)
	4987	+ struct net_device orig_dev, bool another)
4629	4988	{
4630	4989	#ifdef CONFIG_NET_CLS_ACT
4631	4990	struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
..	..	@@ -4645,10 +5004,12 @@
4645	5004	}
4646	5005
4647	5006	qdisc_skb_cb(skb)->pkt_len = skb->len;
	5007	+ qdisc_skb_cb(skb)->mru = 0;
4648	5008	skb->tc_at_ingress = 1;
4649	5009	mini_qdisc_bstats_cpu_update(miniq, skb);
4650	5010
4651		- switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
	5011	+ switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list,
	5012	+ &cl_res, false)) {
4652	5013	case TC_ACT_OK:
4653	5014	case TC_ACT_RECLASSIFY:
4654	5015	skb->tc_index = TC_H_MIN(cl_res.classid);
..	..	@@ -4668,11 +5029,13 @@
4668	5029	* redirecting to another netdev
4669	5030	*/
4670	5031	__skb_push(skb, skb->mac_len);
4671		- skb_do_redirect(skb);
	5032	+ if (skb_do_redirect(skb) == -EAGAIN) {
	5033	+ __skb_pull(skb, skb->mac_len);
	5034	+ *another = true;
	5035	+ break;
	5036	+ }
4672	5037	return NULL;
4673		- case TC_ACT_REINSERT:
4674		- /* this does not scrub the packet, and updates stats on error */
4675		- skb_tc_reinsert(skb, &cl_res);
	5038	+ case TC_ACT_CONSUMED:
4676	5039	return NULL;
4677	5040	default:
4678	5041	break;
..	..	@@ -4772,7 +5135,6 @@
4772	5135	static inline int nf_ingress(struct sk_buff skb, struct packet_type *pt_prev,
4773	5136	int ret, struct net_device orig_dev)
4774	5137	{
4775		-#ifdef CONFIG_NETFILTER_INGRESS
4776	5138	if (nf_hook_ingress_active(skb)) {
4777	5139	int ingress_retval;
4778	5140
..	..	@@ -4786,7 +5148,6 @@
4786	5148	rcu_read_unlock();
4787	5149	return ingress_retval;
4788	5150	}
4789		-#endif /* CONFIG_NETFILTER_INGRESS */
4790	5151	return 0;
4791	5152	}
4792	5153
..	..	@@ -4801,7 +5162,7 @@
4801	5162	int ret = NET_RX_DROP;
4802	5163	__be16 type;
4803	5164
4804		- net_timestamp_check(!netdev_tstamp_prequeue, skb);
	5165	+ net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb);
4805	5166
4806	5167	trace_netif_receive_skb(skb);
4807	5168
..	..	@@ -4861,7 +5222,12 @@
4861	5222	skip_taps:
4862	5223	#ifdef CONFIG_NET_INGRESS
4863	5224	if (static_branch_unlikely(&ingress_needed_key)) {
4864		- skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
	5225	+ bool another = false;
	5226	+
	5227	+ skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
	5228	+ &another);
	5229	+ if (another)
	5230	+ goto another_round;
4865	5231	if (!skb)
4866	5232	goto out;
4867	5233
..	..	@@ -4869,7 +5235,7 @@
4869	5235	goto out;
4870	5236	}
4871	5237	#endif
4872		- skb_reset_tc(skb);
	5238	+ skb_reset_redirect(skb);
4873	5239	skip_classify:
4874	5240	if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4875	5241	goto drop;
..	..	@@ -4906,14 +5272,42 @@
4906	5272	}
4907	5273	}
4908	5274
4909		- if (unlikely(skb_vlan_tag_present(skb))) {
4910		- if (skb_vlan_tag_get_id(skb))
	5275	+ if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
	5276	+check_vlan_id:
	5277	+ if (skb_vlan_tag_get_id(skb)) {
	5278	+ /* Vlan id is non 0 and vlan_do_receive() above couldn't
	5279	+ * find vlan device.
	5280	+ */
4911	5281	skb->pkt_type = PACKET_OTHERHOST;
	5282	+ } else if (skb->protocol == cpu_to_be16(ETH_P_8021Q) \|\|
	5283	+ skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
	5284	+ /* Outer header is 802.1P with vlan 0, inner header is
	5285	+ * 802.1Q or 802.1AD and vlan_do_receive() above could
	5286	+ * not find vlan dev for vlan id 0.
	5287	+ */
	5288	+ __vlan_hwaccel_clear_tag(skb);
	5289	+ skb = skb_vlan_untag(skb);
	5290	+ if (unlikely(!skb))
	5291	+ goto out;
	5292	+ if (vlan_do_receive(&skb))
	5293	+ /* After stripping off 802.1P header with vlan 0
	5294	+ * vlan dev is found for inner header.
	5295	+ */
	5296	+ goto another_round;
	5297	+ else if (unlikely(!skb))
	5298	+ goto out;
	5299	+ else
	5300	+ /* We have stripped outer 802.1P vlan 0 header.
	5301	+ * But could not find vlan dev.
	5302	+ * check again for vlan id to set OTHERHOST.
	5303	+ */
	5304	+ goto check_vlan_id;
	5305	+ }
4912	5306	/* Note: we might in the future use prio bits
4913	5307	* and set skb->priority like in vlan_do_receive()
4914	5308	* For the time being, just ignore Priority Code Point
4915	5309	*/
4916		- skb->vlan_tci = 0;
	5310	+ __vlan_hwaccel_clear_tag(skb);
4917	5311	}
4918	5312
4919	5313	type = skb->protocol;
..	..	@@ -4969,7 +5363,8 @@
4969	5363
4970	5364	ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
4971	5365	if (pt_prev)
4972		- ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
	5366	+ ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
	5367	+ skb->dev, pt_prev, orig_dev);
4973	5368	return ret;
4974	5369	}
4975	5370
..	..	@@ -4979,7 +5374,7 @@
4979	5374	*
4980	5375	* More direct receive version of netif_receive_skb(). It should
4981	5376	* only be used by callers that have a need to skip RPS and Generic XDP.
4982		- * Caller must also take care of handling if (page_is_)pfmemalloc.
	5377	+ * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
4983	5378	*
4984	5379	* This function may only be called from softirq context and interrupts
4985	5380	* should be enabled.
..	..	@@ -5011,7 +5406,8 @@
5011	5406	if (list_empty(head))
5012	5407	return;
5013	5408	if (pt_prev->list_func != NULL)
5014		- pt_prev->list_func(head, pt_prev, orig_dev);
	5409	+ INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
	5410	+ ip_list_rcv, head, pt_prev, orig_dev);
5015	5411	else
5016	5412	list_for_each_entry_safe(skb, next, head, list) {
5017	5413	skb_list_del_init(skb);
..	..	@@ -5122,6 +5518,25 @@
5122	5518	struct bpf_prog *new = xdp->prog;
5123	5519	int ret = 0;
5124	5520
	5521	+ if (new) {
	5522	+ u32 i;
	5523	+
	5524	+ mutex_lock(&new->aux->used_maps_mutex);
	5525	+
	5526	+ /* generic XDP does not work with DEVMAPs that can
	5527	+ * have a bpf_prog installed on an entry
	5528	+ */
	5529	+ for (i = 0; i < new->aux->used_map_cnt; i++) {
	5530	+ if (dev_map_can_have_prog(new->aux->used_maps[i]) \|\|
	5531	+ cpu_map_prog_allowed(new->aux->used_maps[i])) {
	5532	+ mutex_unlock(&new->aux->used_maps_mutex);
	5533	+ return -EINVAL;
	5534	+ }
	5535	+ }
	5536	+
	5537	+ mutex_unlock(&new->aux->used_maps_mutex);
	5538	+ }
	5539	+
5125	5540	switch (xdp->command) {
5126	5541	case XDP_SETUP_PROG:
5127	5542	rcu_assign_pointer(dev->xdp_prog, new);
..	..	@@ -5137,10 +5552,6 @@
5137	5552	}
5138	5553	break;
5139	5554
5140		- case XDP_QUERY_PROG:
5141		- xdp->prog_id = old ? old->aux->id : 0;
5142		- break;
5143		-
5144	5555	default:
5145	5556	ret = -EINVAL;
5146	5557	break;
..	..	@@ -5153,14 +5564,14 @@
5153	5564	{
5154	5565	int ret;
5155	5566
5156		- net_timestamp_check(netdev_tstamp_prequeue, skb);
	5567	+ net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5157	5568
5158	5569	if (skb_defer_rx_timestamp(skb))
5159	5570	return NET_RX_SUCCESS;
5160	5571
5161	5572	rcu_read_lock();
5162	5573	#ifdef CONFIG_RPS
5163		- if (static_key_false(&rps_needed)) {
	5574	+ if (static_branch_unlikely(&rps_needed)) {
5164	5575	struct rps_dev_flow voidflow, *rflow = &voidflow;
5165	5576	int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5166	5577
..	..	@@ -5183,7 +5594,7 @@
5183	5594
5184	5595	INIT_LIST_HEAD(&sublist);
5185	5596	list_for_each_entry_safe(skb, next, head, list) {
5186		- net_timestamp_check(netdev_tstamp_prequeue, skb);
	5597	+ net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5187	5598	skb_list_del_init(skb);
5188	5599	if (!skb_defer_rx_timestamp(skb))
5189	5600	list_add_tail(&skb->list, &sublist);
..	..	@@ -5192,7 +5603,7 @@
5192	5603
5193	5604	rcu_read_lock();
5194	5605	#ifdef CONFIG_RPS
5195		- if (static_key_false(&rps_needed)) {
	5606	+ if (static_branch_unlikely(&rps_needed)) {
5196	5607	list_for_each_entry_safe(skb, next, head, list) {
5197	5608	struct rps_dev_flow voidflow, *rflow = &voidflow;
5198	5609	int cpu = get_rps_cpu(skb->dev, skb, &rflow);
..	..	@@ -5226,9 +5637,14 @@
5226	5637	*/
5227	5638	int netif_receive_skb(struct sk_buff *skb)
5228	5639	{
	5640	+ int ret;
	5641	+
5229	5642	trace_netif_receive_skb_entry(skb);
5230	5643
5231		- return netif_receive_skb_internal(skb);
	5644	+ ret = netif_receive_skb_internal(skb);
	5645	+ trace_netif_receive_skb_exit(ret);
	5646	+
	5647	+ return ret;
5232	5648	}
5233	5649	EXPORT_SYMBOL(netif_receive_skb);
5234	5650
..	..	@@ -5248,13 +5664,16 @@
5248	5664
5249	5665	if (list_empty(head))
5250	5666	return;
5251		- list_for_each_entry(skb, head, list)
5252		- trace_netif_receive_skb_list_entry(skb);
	5667	+ if (trace_netif_receive_skb_list_entry_enabled()) {
	5668	+ list_for_each_entry(skb, head, list)
	5669	+ trace_netif_receive_skb_list_entry(skb);
	5670	+ }
5253	5671	netif_receive_skb_list_internal(head);
	5672	+ trace_netif_receive_skb_list_exit(0);
5254	5673	}
5255	5674	EXPORT_SYMBOL(netif_receive_skb_list);
5256	5675
5257		-DEFINE_PER_CPU(struct work_struct, flush_works);
	5676	+static DEFINE_PER_CPU(struct work_struct, flush_works);
5258	5677
5259	5678	/* Network device is going away, flush any packets still pending */
5260	5679	static void flush_backlog(struct work_struct *work)
..	..	@@ -5287,23 +5706,89 @@
5287	5706	local_bh_enable();
5288	5707	}
5289	5708
	5709	+static bool flush_required(int cpu)
	5710	+{
	5711	+#if IS_ENABLED(CONFIG_RPS)
	5712	+ struct softnet_data *sd = &per_cpu(softnet_data, cpu);
	5713	+ bool do_flush;
	5714	+
	5715	+ local_irq_disable();
	5716	+ rps_lock(sd);
	5717	+
	5718	+ /* as insertion into process_queue happens with the rps lock held,
	5719	+ * process_queue access may race only with dequeue
	5720	+ */
	5721	+ do_flush = !skb_queue_empty(&sd->input_pkt_queue) \|\|
	5722	+ !skb_queue_empty_lockless(&sd->process_queue);
	5723	+ rps_unlock(sd);
	5724	+ local_irq_enable();
	5725	+
	5726	+ return do_flush;
	5727	+#endif
	5728	+ /* without RPS we can't safely check input_pkt_queue: during a
	5729	+ * concurrent remote skb_queue_splice() we can detect as empty both
	5730	+ * input_pkt_queue and process_queue even if the latter could end-up
	5731	+ * containing a lot of packets.
	5732	+ */
	5733	+ return true;
	5734	+}
	5735	+
5290	5736	static void flush_all_backlogs(void)
5291	5737	{
	5738	+ static cpumask_t flush_cpus;
5292	5739	unsigned int cpu;
	5740	+
	5741	+ /* since we are under rtnl lock protection we can use static data
	5742	+ * for the cpumask and avoid allocating on stack the possibly
	5743	+ * large mask
	5744	+ */
	5745	+ ASSERT_RTNL();
5293	5746
5294	5747	get_online_cpus();
5295	5748
5296		- for_each_online_cpu(cpu)
5297		- queue_work_on(cpu, system_highpri_wq,
5298		- per_cpu_ptr(&flush_works, cpu));
	5749	+ cpumask_clear(&flush_cpus);
	5750	+ for_each_online_cpu(cpu) {
	5751	+ if (flush_required(cpu)) {
	5752	+ queue_work_on(cpu, system_highpri_wq,
	5753	+ per_cpu_ptr(&flush_works, cpu));
	5754	+ cpumask_set_cpu(cpu, &flush_cpus);
	5755	+ }
	5756	+ }
5299	5757
5300		- for_each_online_cpu(cpu)
	5758	+ /* we can have in flight packet[s] on the cpus we are not flushing,
	5759	+ * synchronize_net() in unregister_netdevice_many() will take care of
	5760	+ * them
	5761	+ */
	5762	+ for_each_cpu(cpu, &flush_cpus)
5301	5763	flush_work(per_cpu_ptr(&flush_works, cpu));
5302	5764
5303	5765	put_online_cpus();
5304	5766	}
5305	5767
5306		-static int napi_gro_complete(struct sk_buff *skb)
	5768	+/* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
	5769	+static void gro_normal_list(struct napi_struct *napi)
	5770	+{
	5771	+ if (!napi->rx_count)
	5772	+ return;
	5773	+ netif_receive_skb_list_internal(&napi->rx_list);
	5774	+ INIT_LIST_HEAD(&napi->rx_list);
	5775	+ napi->rx_count = 0;
	5776	+}
	5777	+
	5778	+/* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
	5779	+ * pass the whole batch up to the stack.
	5780	+ */
	5781	+static void gro_normal_one(struct napi_struct napi, struct sk_buff skb, int segs)
	5782	+{
	5783	+ list_add_tail(&skb->list, &napi->rx_list);
	5784	+ napi->rx_count += segs;
	5785	+ if (napi->rx_count >= gro_normal_batch)
	5786	+ gro_normal_list(napi);
	5787	+}
	5788	+
	5789	+INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int));
	5790	+INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int));
	5791	+static int napi_gro_complete(struct napi_struct napi, struct sk_buff skb)
5307	5792	{
5308	5793	struct packet_offload *ptype;
5309	5794	__be16 type = skb->protocol;
..	..	@@ -5322,7 +5807,9 @@
5322	5807	if (ptype->type != type \|\| !ptype->callbacks.gro_complete)
5323	5808	continue;
5324	5809
5325		- err = ptype->callbacks.gro_complete(skb, 0);
	5810	+ err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
	5811	+ ipv6_gro_complete, inet_gro_complete,
	5812	+ skb, 0);
5326	5813	break;
5327	5814	}
5328	5815	rcu_read_unlock();
..	..	@@ -5334,7 +5821,8 @@
5334	5821	}
5335	5822
5336	5823	out:
5337		- return netif_receive_skb_internal(skb);
	5824	+ gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count);
	5825	+ return NET_RX_SUCCESS;
5338	5826	}
5339	5827
5340	5828	static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
..	..	@@ -5346,9 +5834,8 @@
5346	5834	list_for_each_entry_safe_reverse(skb, p, head, list) {
5347	5835	if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5348	5836	return;
5349		- list_del(&skb->list);
5350		- skb->next = NULL;
5351		- napi_gro_complete(skb);
	5837	+ skb_list_del_init(skb);
	5838	+ napi_gro_complete(napi, skb);
5352	5839	napi->gro_hash[index].count--;
5353	5840	}
5354	5841
..	..	@@ -5362,11 +5849,13 @@
5362	5849	*/
5363	5850	void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5364	5851	{
5365		- u32 i;
	5852	+ unsigned long bitmask = napi->gro_bitmask;
	5853	+ unsigned int i, base = ~0U;
5366	5854
5367		- for (i = 0; i < GRO_HASH_BUCKETS; i++) {
5368		- if (test_bit(i, &napi->gro_bitmask))
5369		- __napi_gro_flush_chain(napi, i, flush_old);
	5855	+ while ((i = ffs(bitmask)) != 0) {
	5856	+ bitmask >>= i;
	5857	+ base += i;
	5858	+ __napi_gro_flush_chain(napi, base, flush_old);
5370	5859	}
5371	5860	}
5372	5861	EXPORT_SYMBOL(napi_gro_flush);
..	..	@@ -5391,7 +5880,9 @@
5391	5880	}
5392	5881
5393	5882	diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5394		- diffs \|= p->vlan_tci ^ skb->vlan_tci;
	5883	+ diffs \|= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
	5884	+ if (skb_vlan_tag_present(p))
	5885	+ diffs \|= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb);
5395	5886	diffs \|= skb_metadata_dst_cmp(p, skb);
5396	5887	diffs \|= skb_metadata_differs(p, skb);
5397	5888	if (maclen == ETH_HLEN)
..	..	@@ -5401,13 +5892,26 @@
5401	5892	diffs = memcmp(skb_mac_header(p),
5402	5893	skb_mac_header(skb),
5403	5894	maclen);
	5895	+
	5896	+ diffs \|= skb_get_nfct(p) ^ skb_get_nfct(skb);
	5897	+#if IS_ENABLED(CONFIG_SKB_EXTENSIONS) && IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
	5898	+ if (!diffs) {
	5899	+ struct tc_skb_ext *skb_ext = skb_ext_find(skb, TC_SKB_EXT);
	5900	+ struct tc_skb_ext *p_ext = skb_ext_find(p, TC_SKB_EXT);
	5901	+
	5902	+ diffs \|= (!!p_ext) ^ (!!skb_ext);
	5903	+ if (!diffs && unlikely(skb_ext))
	5904	+ diffs \|= p_ext->chain ^ skb_ext->chain;
	5905	+ }
	5906	+#endif
	5907	+
5404	5908	NAPI_GRO_CB(p)->same_flow = !diffs;
5405	5909	}
5406	5910
5407	5911	return head;
5408	5912	}
5409	5913
5410		-static void skb_gro_reset_offset(struct sk_buff *skb)
	5914	+static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff)
5411	5915	{
5412	5916	const struct skb_shared_info *pinfo = skb_shinfo(skb);
5413	5917	const skb_frag_t *frag0 = &pinfo->frags[0];
..	..	@@ -5416,10 +5920,9 @@
5416	5920	NAPI_GRO_CB(skb)->frag0 = NULL;
5417	5921	NAPI_GRO_CB(skb)->frag0_len = 0;
5418	5922
5419		- if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
5420		- pinfo->nr_frags &&
	5923	+ if (!skb_headlen(skb) && pinfo->nr_frags &&
5421	5924	!PageHighMem(skb_frag_page(frag0)) &&
5422		- (!NET_IP_ALIGN \|\| !(skb_frag_off(frag0) & 3))) {
	5925	+ (!NET_IP_ALIGN \|\| !((skb_frag_off(frag0) + nhoff) & 3))) {
5423	5926	NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5424	5927	NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5425	5928	skb_frag_size(frag0),
..	..	@@ -5438,7 +5941,7 @@
5438	5941	skb->data_len -= grow;
5439	5942	skb->tail += grow;
5440	5943
5441		- pinfo->frags[0].page_offset += grow;
	5944	+ skb_frag_off_add(&pinfo->frags[0], grow);
5442	5945	skb_frag_size_sub(&pinfo->frags[0], grow);
5443	5946
5444	5947	if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
..	..	@@ -5448,7 +5951,7 @@
5448	5951	}
5449	5952	}
5450	5953
5451		-static void gro_flush_oldest(struct list_head *head)
	5954	+static void gro_flush_oldest(struct napi_struct napi, struct list_head head)
5452	5955	{
5453	5956	struct sk_buff *oldest;
5454	5957
..	..	@@ -5463,11 +5966,14 @@
5463	5966	/* Do not adjust napi->gro_hash[].count, caller is adding a new
5464	5967	* SKB to the chain.
5465	5968	*/
5466		- list_del(&oldest->list);
5467		- oldest->next = NULL;
5468		- napi_gro_complete(oldest);
	5969	+ skb_list_del_init(oldest);
	5970	+ napi_gro_complete(napi, oldest);
5469	5971	}
5470	5972
	5973	+INDIRECT_CALLABLE_DECLARE(struct sk_buff inet_gro_receive(struct list_head ,
	5974	+ struct sk_buff *));
	5975	+INDIRECT_CALLABLE_DECLARE(struct sk_buff ipv6_gro_receive(struct list_head ,
	5976	+ struct sk_buff *));
5471	5977	static enum gro_result dev_gro_receive(struct napi_struct napi, struct sk_buff skb)
5472	5978	{
5473	5979	u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
..	..	@@ -5517,7 +6023,9 @@
5517	6023	NAPI_GRO_CB(skb)->csum_valid = 0;
5518	6024	}
5519	6025
5520		- pp = ptype->callbacks.gro_receive(gro_head, skb);
	6026	+ pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
	6027	+ ipv6_gro_receive, inet_gro_receive,
	6028	+ gro_head, skb);
5521	6029	break;
5522	6030	}
5523	6031	rcu_read_unlock();
..	..	@@ -5525,7 +6033,7 @@
5525	6033	if (&ptype->list == head)
5526	6034	goto normal;
5527	6035
5528		- if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
	6036	+ if (PTR_ERR(pp) == -EINPROGRESS) {
5529	6037	ret = GRO_CONSUMED;
5530	6038	goto ok;
5531	6039	}
..	..	@@ -5534,9 +6042,8 @@
5534	6042	ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5535	6043
5536	6044	if (pp) {
5537		- list_del(&pp->list);
5538		- pp->next = NULL;
5539		- napi_gro_complete(pp);
	6045	+ skb_list_del_init(pp);
	6046	+ napi_gro_complete(napi, pp);
5540	6047	napi->gro_hash[hash].count--;
5541	6048	}
5542	6049
..	..	@@ -5547,7 +6054,7 @@
5547	6054	goto normal;
5548	6055
5549	6056	if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5550		- gro_flush_oldest(gro_head);
	6057	+ gro_flush_oldest(napi, gro_head);
5551	6058	} else {
5552	6059	napi->gro_hash[hash].count++;
5553	6060	}
..	..	@@ -5608,16 +6115,17 @@
5608	6115	static void napi_skb_free_stolen_head(struct sk_buff *skb)
5609	6116	{
5610	6117	skb_dst_drop(skb);
5611		- secpath_reset(skb);
	6118	+ skb_ext_put(skb);
5612	6119	kmem_cache_free(skbuff_head_cache, skb);
5613	6120	}
5614	6121
5615		-static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
	6122	+static gro_result_t napi_skb_finish(struct napi_struct *napi,
	6123	+ struct sk_buff *skb,
	6124	+ gro_result_t ret)
5616	6125	{
5617	6126	switch (ret) {
5618	6127	case GRO_NORMAL:
5619		- if (netif_receive_skb_internal(skb))
5620		- ret = GRO_DROP;
	6128	+ gro_normal_one(napi, skb, 1);
5621	6129	break;
5622	6130
5623	6131	case GRO_DROP:
..	..	@@ -5642,12 +6150,17 @@
5642	6150
5643	6151	gro_result_t napi_gro_receive(struct napi_struct napi, struct sk_buff skb)
5644	6152	{
	6153	+ gro_result_t ret;
	6154	+
5645	6155	skb_mark_napi_id(skb, napi);
5646	6156	trace_napi_gro_receive_entry(skb);
5647	6157
5648		- skb_gro_reset_offset(skb);
	6158	+ skb_gro_reset_offset(skb, 0);
5649	6159
5650		- return napi_skb_finish(dev_gro_receive(napi, skb), skb);
	6160	+ ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
	6161	+ trace_napi_gro_receive_exit(ret);
	6162	+
	6163	+ return ret;
5651	6164	}
5652	6165	EXPORT_SYMBOL(napi_gro_receive);
5653	6166
..	..	@@ -5660,7 +6173,7 @@
5660	6173	__skb_pull(skb, skb_headlen(skb));
5661	6174	/* restore the reserve we had after netdev_alloc_skb_ip_align() */
5662	6175	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
5663		- skb->vlan_tci = 0;
	6176	+ __vlan_hwaccel_clear_tag(skb);
5664	6177	skb->dev = napi->dev;
5665	6178	skb->skb_iif = 0;
5666	6179
..	..	@@ -5670,7 +6183,8 @@
5670	6183	skb->encapsulation = 0;
5671	6184	skb_shinfo(skb)->gso_type = 0;
5672	6185	skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5673		- secpath_reset(skb);
	6186	+ skb_ext_reset(skb);
	6187	+ nf_reset_ct(skb);
5674	6188
5675	6189	napi->skb = skb;
5676	6190	}
..	..	@@ -5699,8 +6213,8 @@
5699	6213	case GRO_HELD:
5700	6214	__skb_push(skb, ETH_HLEN);
5701	6215	skb->protocol = eth_type_trans(skb, skb->dev);
5702		- if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
5703		- ret = GRO_DROP;
	6216	+ if (ret == GRO_NORMAL)
	6217	+ gro_normal_one(napi, skb, 1);
5704	6218	break;
5705	6219
5706	6220	case GRO_DROP:
..	..	@@ -5735,7 +6249,7 @@
5735	6249	napi->skb = NULL;
5736	6250
5737	6251	skb_reset_mac_header(skb);
5738		- skb_gro_reset_offset(skb);
	6252	+ skb_gro_reset_offset(skb, hlen);
5739	6253
5740	6254	if (unlikely(skb_gro_header_hard(skb, hlen))) {
5741	6255	eth = skb_gro_header_slow(skb, hlen, 0);
..	..	@@ -5765,6 +6279,7 @@
5765	6279
5766	6280	gro_result_t napi_gro_frags(struct napi_struct *napi)
5767	6281	{
	6282	+ gro_result_t ret;
5768	6283	struct sk_buff *skb = napi_frags_skb(napi);
5769	6284
5770	6285	if (!skb)
..	..	@@ -5772,7 +6287,10 @@
5772	6287
5773	6288	trace_napi_gro_frags_entry(skb);
5774	6289
5775		- return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
	6290	+ ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
	6291	+ trace_napi_gro_frags_exit(ret);
	6292	+
	6293	+ return ret;
5776	6294	}
5777	6295	EXPORT_SYMBOL(napi_gro_frags);
5778	6296
..	..	@@ -5788,10 +6306,11 @@
5788	6306
5789	6307	/* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
5790	6308	sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
	6309	+ /* See comments in __skb_checksum_complete(). */
5791	6310	if (likely(!sum)) {
5792	6311	if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
5793	6312	!skb->csum_complete_sw)
5794		- netdev_rx_csum_fault(skb->dev);
	6313	+ netdev_rx_csum_fault(skb->dev, skb);
5795	6314	}
5796	6315
5797	6316	NAPI_GRO_CB(skb)->csum = wsum;
..	..	@@ -5827,14 +6346,12 @@
5827	6346	sd->rps_ipi_list = NULL;
5828	6347
5829	6348	local_irq_enable();
5830		- preempt_check_resched_rt();
5831	6349
5832	6350	/* Send pending IPI's to kick RPS processing on remote cpus. */
5833	6351	net_rps_send_ipi(remsd);
5834	6352	} else
5835	6353	#endif
5836	6354	local_irq_enable();
5837		- preempt_check_resched_rt();
5838	6355	}
5839	6356
5840	6357	static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
..	..	@@ -5860,22 +6377,21 @@
5860	6377	net_rps_action_and_irq_enable(sd);
5861	6378	}
5862	6379
5863		- napi->weight = dev_rx_weight;
	6380	+ napi->weight = READ_ONCE(dev_rx_weight);
5864	6381	while (again) {
5865	6382	struct sk_buff *skb;
5866	6383
5867		- local_irq_disable();
5868	6384	while ((skb = __skb_dequeue(&sd->process_queue))) {
5869		- local_irq_enable();
5870	6385	rcu_read_lock();
5871	6386	__netif_receive_skb(skb);
5872	6387	rcu_read_unlock();
5873	6388	input_queue_head_incr(sd);
5874	6389	if (++work >= quota)
5875		- goto state_changed;
5876		- local_irq_disable();
	6390	+ return work;
	6391	+
5877	6392	}
5878	6393
	6394	+ local_irq_disable();
5879	6395	rps_lock(sd);
5880	6396	if (skb_queue_empty(&sd->input_pkt_queue)) {
5881	6397	/*
..	..	@@ -5896,10 +6412,6 @@
5896	6412	local_irq_enable();
5897	6413	}
5898	6414
5899		-state_changed:
5900		- napi_gro_flush(napi, false);
5901		- sd->current_napi = NULL;
5902		-
5903	6415	return work;
5904	6416	}
5905	6417
..	..	@@ -5917,7 +6429,6 @@
5917	6429	local_irq_save(flags);
5918	6430	____napi_schedule(this_cpu_ptr(&softnet_data), n);
5919	6431	local_irq_restore(flags);
5920		- preempt_check_resched_rt();
5921	6432	}
5922	6433	EXPORT_SYMBOL(__napi_schedule);
5923	6434
..	..	@@ -5926,7 +6437,7 @@
5926	6437	* @n: napi context
5927	6438	*
5928	6439	* Test if NAPI routine is already running, and if not mark
5929		- * it as running. This is used as a condition variable
	6440	+ * it as running. This is used as a condition variable to
5930	6441	* insure only one NAPI poll instance runs. We also make
5931	6442	* sure there is no pending NAPI disable.
5932	6443	*/
..	..	@@ -5954,7 +6465,6 @@
5954	6465	}
5955	6466	EXPORT_SYMBOL(napi_schedule_prep);
5956	6467
5957		-#ifndef CONFIG_PREEMPT_RT_FULL
5958	6468	/**
5959	6469	* __napi_schedule_irqoff - schedule for receive
5960	6470	* @n: entry to schedule
..	..	@@ -5973,11 +6483,11 @@
5973	6483	__napi_schedule(n);
5974	6484	}
5975	6485	EXPORT_SYMBOL(__napi_schedule_irqoff);
5976		-#endif
5977	6486
5978	6487	bool napi_complete_done(struct napi_struct *n, int work_done)
5979	6488	{
5980		- unsigned long flags, val, new;
	6489	+ unsigned long flags, val, new, timeout = 0;
	6490	+ bool ret = true;
5981	6491
5982	6492	/*
5983	6493	* 1) Don't let napi dequeue from the cpu poll list
..	..	@@ -5989,28 +6499,31 @@
5989	6499	NAPIF_STATE_IN_BUSY_POLL)))
5990	6500	return false;
5991	6501
	6502	+ if (work_done) {
	6503	+ if (n->gro_bitmask)
	6504	+ timeout = READ_ONCE(n->dev->gro_flush_timeout);
	6505	+ n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
	6506	+ }
	6507	+ if (n->defer_hard_irqs_count > 0) {
	6508	+ n->defer_hard_irqs_count--;
	6509	+ timeout = READ_ONCE(n->dev->gro_flush_timeout);
	6510	+ if (timeout)
	6511	+ ret = false;
	6512	+ }
5992	6513	if (n->gro_bitmask) {
5993		- unsigned long timeout = 0;
5994		-
5995		- if (work_done)
5996		- timeout = n->dev->gro_flush_timeout;
5997		-
5998	6514	/* When the NAPI instance uses a timeout and keeps postponing
5999	6515	* it, we need to bound somehow the time packets are kept in
6000	6516	* the GRO layer
6001	6517	*/
6002	6518	napi_gro_flush(n, !!timeout);
6003		- if (timeout)
6004		- hrtimer_start(&n->timer, ns_to_ktime(timeout),
6005		- HRTIMER_MODE_REL_PINNED);
6006	6519	}
6007		- if (unlikely(!list_empty(&n->poll_list))) {
6008		- struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6009	6520
	6521	+ gro_normal_list(n);
	6522	+
	6523	+ if (unlikely(!list_empty(&n->poll_list))) {
6010	6524	/* If n->poll_list is not empty, we need to mask irqs */
6011	6525	local_irq_save(flags);
6012	6526	list_del_init(&n->poll_list);
6013		- sd->current_napi = NULL;
6014	6527	local_irq_restore(flags);
6015	6528	}
6016	6529
..	..	@@ -6034,7 +6547,10 @@
6034	6547	return false;
6035	6548	}
6036	6549
6037		- return true;
	6550	+ if (timeout)
	6551	+ hrtimer_start(&n->timer, ns_to_ktime(timeout),
	6552	+ HRTIMER_MODE_REL_PINNED);
	6553	+ return ret;
6038	6554	}
6039	6555	EXPORT_SYMBOL(napi_complete_done);
6040	6556
..	..	@@ -6077,10 +6593,19 @@
6077	6593	* Ideally, a new ndo_busy_poll_stop() could avoid another round.
6078	6594	*/
6079	6595	rc = napi->poll(napi, BUSY_POLL_BUDGET);
	6596	+ /* We can't gro_normal_list() here, because napi->poll() might have
	6597	+ * rearmed the napi (napi_complete_done()) in which case it could
	6598	+ * already be running on another CPU.
	6599	+ */
6080	6600	trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6081	6601	netpoll_poll_unlock(have_poll_lock);
6082		- if (rc == BUSY_POLL_BUDGET)
	6602	+ if (rc == BUSY_POLL_BUDGET) {
	6603	+ /* As the whole budget was spent, we still own the napi so can
	6604	+ * safely handle the rx_list.
	6605	+ */
	6606	+ gro_normal_list(napi);
6083	6607	__napi_schedule(napi);
	6608	+ }
6084	6609	local_bh_enable();
6085	6610	}
6086	6611
..	..	@@ -6125,6 +6650,7 @@
6125	6650	}
6126	6651	work = napi_poll(napi, BUSY_POLL_BUDGET);
6127	6652	trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
	6653	+ gro_normal_list(napi);
6128	6654	count:
6129	6655	if (work > 0)
6130	6656	__NET_ADD_STATS(dev_net(napi->dev),
..	..	@@ -6158,8 +6684,7 @@
6158	6684
6159	6685	static void napi_hash_add(struct napi_struct *napi)
6160	6686	{
6161		- if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) \|\|
6162		- test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
	6687	+ if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6163	6688	return;
6164	6689
6165	6690	spin_lock(&napi_hash_lock);
..	..	@@ -6180,20 +6705,14 @@
6180	6705	/* Warning : caller is responsible to make sure rcu grace period
6181	6706	* is respected before freeing memory containing @napi
6182	6707	*/
6183		-bool napi_hash_del(struct napi_struct *napi)
	6708	+static void napi_hash_del(struct napi_struct *napi)
6184	6709	{
6185		- bool rcu_sync_needed = false;
6186		-
6187	6710	spin_lock(&napi_hash_lock);
6188	6711
6189		- if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
6190		- rcu_sync_needed = true;
6191		- hlist_del_rcu(&napi->napi_hash_node);
6192		- }
	6712	+ hlist_del_init_rcu(&napi->napi_hash_node);
	6713	+
6193	6714	spin_unlock(&napi_hash_lock);
6194		- return rcu_sync_needed;
6195	6715	}
6196		-EXPORT_SYMBOL_GPL(napi_hash_del);
6197	6716
6198	6717	static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6199	6718	{
..	..	@@ -6204,7 +6723,7 @@
6204	6723	/* Note : we use a relaxed variant of napi_schedule_prep() not setting
6205	6724	* NAPI_STATE_MISSED, since we do not react to a device IRQ.
6206	6725	*/
6207		- if (napi->gro_bitmask && !napi_disable_pending(napi) &&
	6726	+ if (!napi_disable_pending(napi) &&
6208	6727	!test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6209	6728	__napi_schedule_irqoff(napi);
6210	6729
..	..	@@ -6225,15 +6744,21 @@
6225	6744	void netif_napi_add(struct net_device dev, struct napi_struct napi,
6226	6745	int (poll)(struct napi_struct , int), int weight)
6227	6746	{
	6747	+ if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
	6748	+ return;
	6749	+
6228	6750	INIT_LIST_HEAD(&napi->poll_list);
	6751	+ INIT_HLIST_NODE(&napi->napi_hash_node);
6229	6752	hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6230	6753	napi->timer.function = napi_watchdog;
6231	6754	init_gro_hash(napi);
6232	6755	napi->skb = NULL;
	6756	+ INIT_LIST_HEAD(&napi->rx_list);
	6757	+ napi->rx_count = 0;
6233	6758	napi->poll = poll;
6234	6759	if (weight > NAPI_POLL_WEIGHT)
6235		- pr_err_once("netif_napi_add() called with weight %d on device %s\n",
6236		- weight, dev->name);
	6760	+ netdev_err_once(dev, "%s() called with weight %d\n", __func__,
	6761	+ weight);
6237	6762	napi->weight = weight;
6238	6763	napi->dev = dev;
6239	6764	#ifdef CONFIG_NETPOLL
..	..	@@ -6276,26 +6801,19 @@
6276	6801	}
6277	6802
6278	6803	/* Must be called in process context */
6279		-void netif_napi_del(struct napi_struct *napi)
	6804	+void __netif_napi_del(struct napi_struct *napi)
6280	6805	{
6281		- might_sleep();
6282		- if (napi_hash_del(napi))
6283		- synchronize_net();
6284		- list_del_init(&napi->dev_list);
	6806	+ if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
	6807	+ return;
	6808	+
	6809	+ napi_hash_del(napi);
	6810	+ list_del_rcu(&napi->dev_list);
6285	6811	napi_free_frags(napi);
6286	6812
6287	6813	flush_gro_hash(napi);
6288	6814	napi->gro_bitmask = 0;
6289	6815	}
6290		-EXPORT_SYMBOL(netif_napi_del);
6291		-
6292		-struct napi_struct *get_current_napi_context(void)
6293		-{
6294		- struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6295		-
6296		- return sd->current_napi;
6297		-}
6298		-EXPORT_SYMBOL(get_current_napi_context);
	6816	+EXPORT_SYMBOL(__netif_napi_del);
6299	6817
6300	6818	static int napi_poll(struct napi_struct n, struct list_head repoll)
6301	6819	{
..	..	@@ -6316,14 +6834,13 @@
6316	6834	*/
6317	6835	work = 0;
6318	6836	if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6319		- struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6320		-
6321		- sd->current_napi = n;
6322	6837	work = n->poll(n, weight);
6323	6838	trace_napi_poll(n, work, weight);
6324	6839	}
6325	6840
6326		- WARN_ON_ONCE(work > weight);
	6841	+ if (unlikely(work > weight))
	6842	+ pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
	6843	+ n->poll, work, weight);
6327	6844
6328	6845	if (likely(work < weight))
6329	6846	goto out_unlock;
..	..	@@ -6344,6 +6861,8 @@
6344	6861	*/
6345	6862	napi_gro_flush(n, HZ >= 1000);
6346	6863	}
	6864	+
	6865	+ gro_normal_list(n);
6347	6866
6348	6867	/* Some drivers may have called napi_schedule
6349	6868	* prior to exhausting their budget.
..	..	@@ -6366,22 +6885,14 @@
6366	6885	{
6367	6886	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6368	6887	unsigned long time_limit = jiffies +
6369		- usecs_to_jiffies(netdev_budget_usecs);
6370		- int budget = netdev_budget;
6371		- struct sk_buff_head tofree_q;
6372		- struct sk_buff *skb;
	6888	+ usecs_to_jiffies(READ_ONCE(netdev_budget_usecs));
	6889	+ int budget = READ_ONCE(netdev_budget);
6373	6890	LIST_HEAD(list);
6374	6891	LIST_HEAD(repoll);
6375	6892
6376		- __skb_queue_head_init(&tofree_q);
6377		-
6378	6893	local_irq_disable();
6379		- skb_queue_splice_init(&sd->tofree_queue, &tofree_q);
6380	6894	list_splice_init(&sd->poll_list, &list);
6381	6895	local_irq_enable();
6382		-
6383		- while ((skb = __skb_dequeue(&tofree_q)))
6384		- kfree_skb(skb);
6385	6896
6386	6897	for (;;) {
6387	6898	struct napi_struct *n;
..	..	@@ -6412,7 +6923,7 @@
6412	6923	list_splice_tail(&repoll, &list);
6413	6924	list_splice(&list, &sd->poll_list);
6414	6925	if (!list_empty(&sd->poll_list))
6415		- __raise_softirq_irqoff_ksoft(NET_RX_SOFTIRQ);
	6926	+ __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6416	6927
6417	6928	net_rps_action_and_irq_enable(sd);
6418	6929	out:
..	..	@@ -6424,6 +6935,9 @@
6424	6935
6425	6936	/* upper master flag, there can only be one master device per list */
6426	6937	bool master;
	6938	+
	6939	+ /* lookup ignore flag */
	6940	+ bool ignore;
6427	6941
6428	6942	/* counter for the number of times this device was added to us */
6429	6943	u16 ref_nr;
..	..	@@ -6447,9 +6961,10 @@
6447	6961	return NULL;
6448	6962	}
6449	6963
6450		-static int __netdev_has_upper_dev(struct net_device upper_dev, void data)
	6964	+static int ____netdev_has_upper_dev(struct net_device *upper_dev,
	6965	+ struct netdev_nested_priv *priv)
6451	6966	{
6452		- struct net_device *dev = data;
	6967	+ struct net_device dev = (struct net_device )priv->data;
6453	6968
6454	6969	return upper_dev == dev;
6455	6970	}
..	..	@@ -6466,10 +6981,14 @@
6466	6981	bool netdev_has_upper_dev(struct net_device *dev,
6467	6982	struct net_device *upper_dev)
6468	6983	{
	6984	+ struct netdev_nested_priv priv = {
	6985	+ .data = (void *)upper_dev,
	6986	+ };
	6987	+
6469	6988	ASSERT_RTNL();
6470	6989
6471		- return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6472		- upper_dev);
	6990	+ return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
	6991	+ &priv);
6473	6992	}
6474	6993	EXPORT_SYMBOL(netdev_has_upper_dev);
6475	6994
..	..	@@ -6486,8 +7005,12 @@
6486	7005	bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6487	7006	struct net_device *upper_dev)
6488	7007	{
6489		- return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
6490		- upper_dev);
	7008	+ struct netdev_nested_priv priv = {
	7009	+ .data = (void *)upper_dev,
	7010	+ };
	7011	+
	7012	+ return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
	7013	+ &priv);
6491	7014	}
6492	7015	EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6493	7016
..	..	@@ -6529,6 +7052,22 @@
6529	7052	return NULL;
6530	7053	}
6531	7054	EXPORT_SYMBOL(netdev_master_upper_dev_get);
	7055	+
	7056	+static struct net_device __netdev_master_upper_dev_get(struct net_device dev)
	7057	+{
	7058	+ struct netdev_adjacent *upper;
	7059	+
	7060	+ ASSERT_RTNL();
	7061	+
	7062	+ if (list_empty(&dev->adj_list.upper))
	7063	+ return NULL;
	7064	+
	7065	+ upper = list_first_entry(&dev->adj_list.upper,
	7066	+ struct netdev_adjacent, list);
	7067	+ if (likely(upper->master) && !upper->ignore)
	7068	+ return upper->dev;
	7069	+ return NULL;
	7070	+}
6532	7071
6533	7072	/**
6534	7073	* netdev_has_any_lower_dev - Check if device is linked to some device
..	..	@@ -6580,8 +7119,9 @@
6580	7119	}
6581	7120	EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6582	7121
6583		-static struct net_device netdev_next_upper_dev(struct net_device dev,
6584		- struct list_head **iter)
	7122	+static struct net_device __netdev_next_upper_dev(struct net_device dev,
	7123	+ struct list_head **iter,
	7124	+ bool *ignore)
6585	7125	{
6586	7126	struct netdev_adjacent *upper;
6587	7127
..	..	@@ -6591,6 +7131,7 @@
6591	7131	return NULL;
6592	7132
6593	7133	*iter = &upper->list;
	7134	+ *ignore = upper->ignore;
6594	7135
6595	7136	return upper->dev;
6596	7137	}
..	..	@@ -6612,30 +7153,33 @@
6612	7153	return upper->dev;
6613	7154	}
6614	7155
6615		-static int netdev_walk_all_upper_dev(struct net_device *dev,
6616		- int (fn)(struct net_device dev,
6617		- void *data),
6618		- void *data)
	7156	+static int __netdev_walk_all_upper_dev(struct net_device *dev,
	7157	+ int (fn)(struct net_device dev,
	7158	+ struct netdev_nested_priv *priv),
	7159	+ struct netdev_nested_priv *priv)
6619	7160	{
6620	7161	struct net_device udev, next, now, dev_stack[MAX_NEST_DEV + 1];
6621	7162	struct list_head niter, iter, *iter_stack[MAX_NEST_DEV + 1];
6622	7163	int ret, cur = 0;
	7164	+ bool ignore;
6623	7165
6624	7166	now = dev;
6625	7167	iter = &dev->adj_list.upper;
6626	7168
6627	7169	while (1) {
6628	7170	if (now != dev) {
6629		- ret = fn(now, data);
	7171	+ ret = fn(now, priv);
6630	7172	if (ret)
6631	7173	return ret;
6632	7174	}
6633	7175
6634	7176	next = NULL;
6635	7177	while (1) {
6636		- udev = netdev_next_upper_dev(now, &iter);
	7178	+ udev = __netdev_next_upper_dev(now, &iter, &ignore);
6637	7179	if (!udev)
6638	7180	break;
	7181	+ if (ignore)
	7182	+ continue;
6639	7183
6640	7184	next = udev;
6641	7185	niter = &udev->adj_list.upper;
..	..	@@ -6660,8 +7204,8 @@
6660	7204
6661	7205	int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6662	7206	int (fn)(struct net_device dev,
6663		- void *data),
6664		- void *data)
	7207	+ struct netdev_nested_priv *priv),
	7208	+ struct netdev_nested_priv *priv)
6665	7209	{
6666	7210	struct net_device udev, next, now, dev_stack[MAX_NEST_DEV + 1];
6667	7211	struct list_head niter, iter, *iter_stack[MAX_NEST_DEV + 1];
..	..	@@ -6672,7 +7216,7 @@
6672	7216
6673	7217	while (1) {
6674	7218	if (now != dev) {
6675		- ret = fn(now, data);
	7219	+ ret = fn(now, priv);
6676	7220	if (ret)
6677	7221	return ret;
6678	7222	}
..	..	@@ -6704,6 +7248,20 @@
6704	7248	return 0;
6705	7249	}
6706	7250	EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
	7251	+
	7252	+static bool __netdev_has_upper_dev(struct net_device *dev,
	7253	+ struct net_device *upper_dev)
	7254	+{
	7255	+ struct netdev_nested_priv priv = {
	7256	+ .flags = 0,
	7257	+ .data = (void *)upper_dev,
	7258	+ };
	7259	+
	7260	+ ASSERT_RTNL();
	7261	+
	7262	+ return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
	7263	+ &priv);
	7264	+}
6707	7265
6708	7266	/**
6709	7267	* netdev_lower_get_next_private - Get the next ->private from the
..	..	@@ -6801,10 +7359,27 @@
6801	7359	return lower->dev;
6802	7360	}
6803	7361
	7362	+static struct net_device __netdev_next_lower_dev(struct net_device dev,
	7363	+ struct list_head **iter,
	7364	+ bool *ignore)
	7365	+{
	7366	+ struct netdev_adjacent *lower;
	7367	+
	7368	+ lower = list_entry((*iter)->next, struct netdev_adjacent, list);
	7369	+
	7370	+ if (&lower->list == &dev->adj_list.lower)
	7371	+ return NULL;
	7372	+
	7373	+ *iter = &lower->list;
	7374	+ *ignore = lower->ignore;
	7375	+
	7376	+ return lower->dev;
	7377	+}
	7378	+
6804	7379	int netdev_walk_all_lower_dev(struct net_device *dev,
6805	7380	int (fn)(struct net_device dev,
6806		- void *data),
6807		- void *data)
	7381	+ struct netdev_nested_priv *priv),
	7382	+ struct netdev_nested_priv *priv)
6808	7383	{
6809	7384	struct net_device ldev, next, now, dev_stack[MAX_NEST_DEV + 1];
6810	7385	struct list_head niter, iter, *iter_stack[MAX_NEST_DEV + 1];
..	..	@@ -6815,7 +7390,7 @@
6815	7390
6816	7391	while (1) {
6817	7392	if (now != dev) {
6818		- ret = fn(now, data);
	7393	+ ret = fn(now, priv);
6819	7394	if (ret)
6820	7395	return ret;
6821	7396	}
..	..	@@ -6848,8 +7423,57 @@
6848	7423	}
6849	7424	EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
6850	7425
6851		-static struct net_device netdev_next_lower_dev_rcu(struct net_device dev,
6852		- struct list_head **iter)
	7426	+static int __netdev_walk_all_lower_dev(struct net_device *dev,
	7427	+ int (fn)(struct net_device dev,
	7428	+ struct netdev_nested_priv *priv),
	7429	+ struct netdev_nested_priv *priv)
	7430	+{
	7431	+ struct net_device ldev, next, now, dev_stack[MAX_NEST_DEV + 1];
	7432	+ struct list_head niter, iter, *iter_stack[MAX_NEST_DEV + 1];
	7433	+ int ret, cur = 0;
	7434	+ bool ignore;
	7435	+
	7436	+ now = dev;
	7437	+ iter = &dev->adj_list.lower;
	7438	+
	7439	+ while (1) {
	7440	+ if (now != dev) {
	7441	+ ret = fn(now, priv);
	7442	+ if (ret)
	7443	+ return ret;
	7444	+ }
	7445	+
	7446	+ next = NULL;
	7447	+ while (1) {
	7448	+ ldev = __netdev_next_lower_dev(now, &iter, &ignore);
	7449	+ if (!ldev)
	7450	+ break;
	7451	+ if (ignore)
	7452	+ continue;
	7453	+
	7454	+ next = ldev;
	7455	+ niter = &ldev->adj_list.lower;
	7456	+ dev_stack[cur] = now;
	7457	+ iter_stack[cur++] = iter;
	7458	+ break;
	7459	+ }
	7460	+
	7461	+ if (!next) {
	7462	+ if (!cur)
	7463	+ return 0;
	7464	+ next = dev_stack[--cur];
	7465	+ niter = iter_stack[cur];
	7466	+ }
	7467	+
	7468	+ now = next;
	7469	+ iter = niter;
	7470	+ }
	7471	+
	7472	+ return 0;
	7473	+}
	7474	+
	7475	+struct net_device netdev_next_lower_dev_rcu(struct net_device dev,
	7476	+ struct list_head **iter)
6853	7477	{
6854	7478	struct netdev_adjacent *lower;
6855	7479
..	..	@@ -6861,17 +7485,21 @@
6861	7485
6862	7486	return lower->dev;
6863	7487	}
	7488	+EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
6864	7489
6865	7490	static u8 __netdev_upper_depth(struct net_device *dev)
6866	7491	{
6867	7492	struct net_device *udev;
6868	7493	struct list_head *iter;
6869	7494	u8 max_depth = 0;
	7495	+ bool ignore;
6870	7496
6871	7497	for (iter = &dev->adj_list.upper,
6872		- udev = netdev_next_upper_dev(dev, &iter);
	7498	+ udev = __netdev_next_upper_dev(dev, &iter, &ignore);
6873	7499	udev;
6874		- udev = netdev_next_upper_dev(dev, &iter)) {
	7500	+ udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
	7501	+ if (ignore)
	7502	+ continue;
6875	7503	if (max_depth < udev->upper_level)
6876	7504	max_depth = udev->upper_level;
6877	7505	}
..	..	@@ -6884,11 +7512,14 @@
6884	7512	struct net_device *ldev;
6885	7513	struct list_head *iter;
6886	7514	u8 max_depth = 0;
	7515	+ bool ignore;
6887	7516
6888	7517	for (iter = &dev->adj_list.lower,
6889		- ldev = netdev_next_lower_dev(dev, &iter);
	7518	+ ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
6890	7519	ldev;
6891		- ldev = netdev_next_lower_dev(dev, &iter)) {
	7520	+ ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
	7521	+ if (ignore)
	7522	+ continue;
6892	7523	if (max_depth < ldev->lower_level)
6893	7524	max_depth = ldev->lower_level;
6894	7525	}
..	..	@@ -6896,22 +7527,34 @@
6896	7527	return max_depth;
6897	7528	}
6898	7529
6899		-static int __netdev_update_upper_level(struct net_device dev, void data)
	7530	+static int __netdev_update_upper_level(struct net_device *dev,
	7531	+ struct netdev_nested_priv *__unused)
6900	7532	{
6901	7533	dev->upper_level = __netdev_upper_depth(dev) + 1;
6902	7534	return 0;
6903	7535	}
6904	7536
6905		-static int __netdev_update_lower_level(struct net_device dev, void data)
	7537	+static int __netdev_update_lower_level(struct net_device *dev,
	7538	+ struct netdev_nested_priv *priv)
6906	7539	{
6907	7540	dev->lower_level = __netdev_lower_depth(dev) + 1;
	7541	+
	7542	+#ifdef CONFIG_LOCKDEP
	7543	+ if (!priv)
	7544	+ return 0;
	7545	+
	7546	+ if (priv->flags & NESTED_SYNC_IMM)
	7547	+ dev->nested_level = dev->lower_level - 1;
	7548	+ if (priv->flags & NESTED_SYNC_TODO)
	7549	+ net_unlink_todo(dev);
	7550	+#endif
6908	7551	return 0;
6909	7552	}
6910	7553
6911	7554	int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
6912	7555	int (fn)(struct net_device dev,
6913		- void *data),
6914		- void *data)
	7556	+ struct netdev_nested_priv *priv),
	7557	+ struct netdev_nested_priv *priv)
6915	7558	{
6916	7559	struct net_device ldev, next, now, dev_stack[MAX_NEST_DEV + 1];
6917	7560	struct list_head niter, iter, *iter_stack[MAX_NEST_DEV + 1];
..	..	@@ -6922,7 +7565,7 @@
6922	7565
6923	7566	while (1) {
6924	7567	if (now != dev) {
6925		- ret = fn(now, data);
	7568	+ ret = fn(now, priv);
6926	7569	if (ret)
6927	7570	return ret;
6928	7571	}
..	..	@@ -7052,6 +7695,7 @@
7052	7695	adj->master = master;
7053	7696	adj->ref_nr = 1;
7054	7697	adj->private = private;
	7698	+ adj->ignore = false;
7055	7699	dev_hold(adj_dev);
7056	7700
7057	7701	pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
..	..	@@ -7181,6 +7825,7 @@
7181	7825	static int __netdev_upper_dev_link(struct net_device *dev,
7182	7826	struct net_device *upper_dev, bool master,
7183	7827	void upper_priv, void upper_info,
	7828	+ struct netdev_nested_priv *priv,
7184	7829	struct netlink_ext_ack *extack)
7185	7830	{
7186	7831	struct netdev_notifier_changeupper_info changeupper_info = {
..	..	@@ -7202,17 +7847,17 @@
7202	7847	return -EBUSY;
7203	7848
7204	7849	/* To prevent loops, check if dev is not upper device to upper_dev. */
7205		- if (netdev_has_upper_dev(upper_dev, dev))
	7850	+ if (__netdev_has_upper_dev(upper_dev, dev))
7206	7851	return -EBUSY;
7207	7852
7208	7853	if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7209	7854	return -EMLINK;
7210	7855
7211	7856	if (!master) {
7212		- if (netdev_has_upper_dev(dev, upper_dev))
	7857	+ if (__netdev_has_upper_dev(dev, upper_dev))
7213	7858	return -EEXIST;
7214	7859	} else {
7215		- master_dev = netdev_master_upper_dev_get(dev);
	7860	+ master_dev = __netdev_master_upper_dev_get(dev);
7216	7861	if (master_dev)
7217	7862	return master_dev == upper_dev ? -EEXIST : -EBUSY;
7218	7863	}
..	..	@@ -7235,10 +7880,11 @@
7235	7880	goto rollback;
7236	7881
7237	7882	__netdev_update_upper_level(dev, NULL);
7238		- netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
	7883	+ __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7239	7884
7240		- __netdev_update_lower_level(upper_dev, NULL);
7241		- netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, NULL);
	7885	+ __netdev_update_lower_level(upper_dev, priv);
	7886	+ __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
	7887	+ priv);
7242	7888
7243	7889	return 0;
7244	7890
..	..	@@ -7263,8 +7909,13 @@
7263	7909	struct net_device *upper_dev,
7264	7910	struct netlink_ext_ack *extack)
7265	7911	{
	7912	+ struct netdev_nested_priv priv = {
	7913	+ .flags = NESTED_SYNC_IMM \| NESTED_SYNC_TODO,
	7914	+ .data = NULL,
	7915	+ };
	7916	+
7266	7917	return __netdev_upper_dev_link(dev, upper_dev, false,
7267		- NULL, NULL, extack);
	7918	+ NULL, NULL, &priv, extack);
7268	7919	}
7269	7920	EXPORT_SYMBOL(netdev_upper_dev_link);
7270	7921
..	..	@@ -7287,21 +7938,19 @@
7287	7938	void upper_priv, void upper_info,
7288	7939	struct netlink_ext_ack *extack)
7289	7940	{
	7941	+ struct netdev_nested_priv priv = {
	7942	+ .flags = NESTED_SYNC_IMM \| NESTED_SYNC_TODO,
	7943	+ .data = NULL,
	7944	+ };
	7945	+
7290	7946	return __netdev_upper_dev_link(dev, upper_dev, true,
7291		- upper_priv, upper_info, extack);
	7947	+ upper_priv, upper_info, &priv, extack);
7292	7948	}
7293	7949	EXPORT_SYMBOL(netdev_master_upper_dev_link);
7294	7950
7295		-/**
7296		- * netdev_upper_dev_unlink - Removes a link to upper device
7297		- * @dev: device
7298		- * @upper_dev: new upper device
7299		- *
7300		- * Removes a link to device which is upper to this one. The caller must hold
7301		- * the RTNL lock.
7302		- */
7303		-void netdev_upper_dev_unlink(struct net_device *dev,
7304		- struct net_device *upper_dev)
	7951	+static void __netdev_upper_dev_unlink(struct net_device *dev,
	7952	+ struct net_device *upper_dev,
	7953	+ struct netdev_nested_priv *priv)
7305	7954	{
7306	7955	struct netdev_notifier_changeupper_info changeupper_info = {
7307	7956	.info = {
..	..	@@ -7324,12 +7973,126 @@
7324	7973	&changeupper_info.info);
7325	7974
7326	7975	__netdev_update_upper_level(dev, NULL);
7327		- netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
	7976	+ __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7328	7977
7329		- __netdev_update_lower_level(upper_dev, NULL);
7330		- netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level, NULL);
	7978	+ __netdev_update_lower_level(upper_dev, priv);
	7979	+ __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
	7980	+ priv);
	7981	+}
	7982	+
	7983	+/**
	7984	+ * netdev_upper_dev_unlink - Removes a link to upper device
	7985	+ * @dev: device
	7986	+ * @upper_dev: new upper device
	7987	+ *
	7988	+ * Removes a link to device which is upper to this one. The caller must hold
	7989	+ * the RTNL lock.
	7990	+ */
	7991	+void netdev_upper_dev_unlink(struct net_device *dev,
	7992	+ struct net_device *upper_dev)
	7993	+{
	7994	+ struct netdev_nested_priv priv = {
	7995	+ .flags = NESTED_SYNC_TODO,
	7996	+ .data = NULL,
	7997	+ };
	7998	+
	7999	+ __netdev_upper_dev_unlink(dev, upper_dev, &priv);
7331	8000	}
7332	8001	EXPORT_SYMBOL(netdev_upper_dev_unlink);
	8002	+
	8003	+static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
	8004	+ struct net_device *lower_dev,
	8005	+ bool val)
	8006	+{
	8007	+ struct netdev_adjacent *adj;
	8008	+
	8009	+ adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
	8010	+ if (adj)
	8011	+ adj->ignore = val;
	8012	+
	8013	+ adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
	8014	+ if (adj)
	8015	+ adj->ignore = val;
	8016	+}
	8017	+
	8018	+static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
	8019	+ struct net_device *lower_dev)
	8020	+{
	8021	+ __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
	8022	+}
	8023	+
	8024	+static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
	8025	+ struct net_device *lower_dev)
	8026	+{
	8027	+ __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
	8028	+}
	8029	+
	8030	+int netdev_adjacent_change_prepare(struct net_device *old_dev,
	8031	+ struct net_device *new_dev,
	8032	+ struct net_device *dev,
	8033	+ struct netlink_ext_ack *extack)
	8034	+{
	8035	+ struct netdev_nested_priv priv = {
	8036	+ .flags = 0,
	8037	+ .data = NULL,
	8038	+ };
	8039	+ int err;
	8040	+
	8041	+ if (!new_dev)
	8042	+ return 0;
	8043	+
	8044	+ if (old_dev && new_dev != old_dev)
	8045	+ netdev_adjacent_dev_disable(dev, old_dev);
	8046	+ err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
	8047	+ extack);
	8048	+ if (err) {
	8049	+ if (old_dev && new_dev != old_dev)
	8050	+ netdev_adjacent_dev_enable(dev, old_dev);
	8051	+ return err;
	8052	+ }
	8053	+
	8054	+ return 0;
	8055	+}
	8056	+EXPORT_SYMBOL(netdev_adjacent_change_prepare);
	8057	+
	8058	+void netdev_adjacent_change_commit(struct net_device *old_dev,
	8059	+ struct net_device *new_dev,
	8060	+ struct net_device *dev)
	8061	+{
	8062	+ struct netdev_nested_priv priv = {
	8063	+ .flags = NESTED_SYNC_IMM \| NESTED_SYNC_TODO,
	8064	+ .data = NULL,
	8065	+ };
	8066	+
	8067	+ if (!new_dev \|\| !old_dev)
	8068	+ return;
	8069	+
	8070	+ if (new_dev == old_dev)
	8071	+ return;
	8072	+
	8073	+ netdev_adjacent_dev_enable(dev, old_dev);
	8074	+ __netdev_upper_dev_unlink(old_dev, dev, &priv);
	8075	+}
	8076	+EXPORT_SYMBOL(netdev_adjacent_change_commit);
	8077	+
	8078	+void netdev_adjacent_change_abort(struct net_device *old_dev,
	8079	+ struct net_device *new_dev,
	8080	+ struct net_device *dev)
	8081	+{
	8082	+ struct netdev_nested_priv priv = {
	8083	+ .flags = 0,
	8084	+ .data = NULL,
	8085	+ };
	8086	+
	8087	+ if (!new_dev)
	8088	+ return;
	8089	+
	8090	+ if (old_dev && new_dev != old_dev)
	8091	+ netdev_adjacent_dev_enable(dev, old_dev);
	8092	+
	8093	+ __netdev_upper_dev_unlink(new_dev, dev, &priv);
	8094	+}
	8095	+EXPORT_SYMBOL(netdev_adjacent_change_abort);
7333	8096
7334	8097	/**
7335	8098	* netdev_bonding_info_change - Dispatch event about slave change
..	..	@@ -7352,6 +8115,29 @@
7352	8115	&info.info);
7353	8116	}
7354	8117	EXPORT_SYMBOL(netdev_bonding_info_change);
	8118	+
	8119	+/**
	8120	+ * netdev_get_xmit_slave - Get the xmit slave of master device
	8121	+ * @dev: device
	8122	+ * @skb: The packet
	8123	+ * @all_slaves: assume all the slaves are active
	8124	+ *
	8125	+ * The reference counters are not incremented so the caller must be
	8126	+ * careful with locks. The caller must hold RCU lock.
	8127	+ * %NULL is returned if no slave is found.
	8128	+ */
	8129	+
	8130	+struct net_device netdev_get_xmit_slave(struct net_device dev,
	8131	+ struct sk_buff *skb,
	8132	+ bool all_slaves)
	8133	+{
	8134	+ const struct net_device_ops *ops = dev->netdev_ops;
	8135	+
	8136	+ if (!ops->ndo_get_xmit_slave)
	8137	+ return NULL;
	8138	+ return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
	8139	+}
	8140	+EXPORT_SYMBOL(netdev_get_xmit_slave);
7355	8141
7356	8142	static void netdev_adjacent_add_links(struct net_device *dev)
7357	8143	{
..	..	@@ -7443,25 +8229,6 @@
7443	8229	}
7444	8230	EXPORT_SYMBOL(netdev_lower_dev_get_private);
7445	8231
7446		-
7447		-int dev_get_nest_level(struct net_device *dev)
7448		-{
7449		- struct net_device *lower = NULL;
7450		- struct list_head *iter;
7451		- int max_nest = -1;
7452		- int nest;
7453		-
7454		- ASSERT_RTNL();
7455		-
7456		- netdev_for_each_lower_dev(dev, lower, iter) {
7457		- nest = dev_get_nest_level(lower);
7458		- if (max_nest < nest)
7459		- max_nest = nest;
7460		- }
7461		-
7462		- return max_nest + 1;
7463		-}
7464		-EXPORT_SYMBOL(dev_get_nest_level);
7465	8232
7466	8233	/**
7467	8234	* netdev_lower_change - Dispatch event about lower device state change
..	..	@@ -7689,7 +8456,8 @@
7689	8456	}
7690	8457	EXPORT_SYMBOL(dev_get_flags);
7691	8458
7692		-int __dev_change_flags(struct net_device *dev, unsigned int flags)
	8459	+int __dev_change_flags(struct net_device *dev, unsigned int flags,
	8460	+ struct netlink_ext_ack *extack)
7693	8461	{
7694	8462	unsigned int old_flags = dev->flags;
7695	8463	int ret;
..	..	@@ -7726,7 +8494,7 @@
7726	8494	if (old_flags & IFF_UP)
7727	8495	__dev_close(dev);
7728	8496	else
7729		- ret = __dev_open(dev);
	8497	+ ret = __dev_open(dev, extack);
7730	8498	}
7731	8499
7732	8500	if ((flags ^ dev->gflags) & IFF_PROMISC) {
..	..	@@ -7786,16 +8554,18 @@
7786	8554	* dev_change_flags - change device settings
7787	8555	* @dev: device
7788	8556	* @flags: device state flags
	8557	+ * @extack: netlink extended ack
7789	8558	*
7790	8559	* Change settings on device based state flags. The flags are
7791	8560	* in the userspace exported format.
7792	8561	*/
7793		-int dev_change_flags(struct net_device *dev, unsigned int flags)
	8562	+int dev_change_flags(struct net_device *dev, unsigned int flags,
	8563	+ struct netlink_ext_ack *extack)
7794	8564	{
7795	8565	int ret;
7796	8566	unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
7797	8567
7798		- ret = __dev_change_flags(dev, flags);
	8568	+ ret = __dev_change_flags(dev, flags, extack);
7799	8569	if (ret < 0)
7800	8570	return ret;
7801	8571
..	..	@@ -7938,13 +8708,36 @@
7938	8708	EXPORT_SYMBOL(dev_set_group);
7939	8709
7940	8710	/**
	8711	+ * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
	8712	+ * @dev: device
	8713	+ * @addr: new address
	8714	+ * @extack: netlink extended ack
	8715	+ */
	8716	+int dev_pre_changeaddr_notify(struct net_device dev, const char addr,
	8717	+ struct netlink_ext_ack *extack)
	8718	+{
	8719	+ struct netdev_notifier_pre_changeaddr_info info = {
	8720	+ .info.dev = dev,
	8721	+ .info.extack = extack,
	8722	+ .dev_addr = addr,
	8723	+ };
	8724	+ int rc;
	8725	+
	8726	+ rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
	8727	+ return notifier_to_errno(rc);
	8728	+}
	8729	+EXPORT_SYMBOL(dev_pre_changeaddr_notify);
	8730	+
	8731	+/**
7941	8732	* dev_set_mac_address - Change Media Access Control Address
7942	8733	* @dev: device
7943	8734	* @sa: new address
	8735	+ * @extack: netlink extended ack
7944	8736	*
7945	8737	* Change the hardware (MAC) address of the device
7946	8738	*/
7947		-int dev_set_mac_address(struct net_device dev, struct sockaddr sa)
	8739	+int dev_set_mac_address(struct net_device dev, struct sockaddr sa,
	8740	+ struct netlink_ext_ack *extack)
7948	8741	{
7949	8742	const struct net_device_ops *ops = dev->netdev_ops;
7950	8743	int err;
..	..	@@ -7955,6 +8748,9 @@
7955	8748	return -EINVAL;
7956	8749	if (!netif_device_present(dev))
7957	8750	return -ENODEV;
	8751	+ err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
	8752	+ if (err)
	8753	+ return err;
7958	8754	err = ops->ndo_set_mac_address(dev, sa);
7959	8755	if (err)
7960	8756	return err;
..	..	@@ -7964,6 +8760,48 @@
7964	8760	return 0;
7965	8761	}
7966	8762	EXPORT_SYMBOL(dev_set_mac_address);
	8763	+
	8764	+static DECLARE_RWSEM(dev_addr_sem);
	8765	+
	8766	+int dev_set_mac_address_user(struct net_device dev, struct sockaddr sa,
	8767	+ struct netlink_ext_ack *extack)
	8768	+{
	8769	+ int ret;
	8770	+
	8771	+ down_write(&dev_addr_sem);
	8772	+ ret = dev_set_mac_address(dev, sa, extack);
	8773	+ up_write(&dev_addr_sem);
	8774	+ return ret;
	8775	+}
	8776	+EXPORT_SYMBOL(dev_set_mac_address_user);
	8777	+
	8778	+int dev_get_mac_address(struct sockaddr sa, struct net net, char *dev_name)
	8779	+{
	8780	+ size_t size = sizeof(sa->sa_data);
	8781	+ struct net_device *dev;
	8782	+ int ret = 0;
	8783	+
	8784	+ down_read(&dev_addr_sem);
	8785	+ rcu_read_lock();
	8786	+
	8787	+ dev = dev_get_by_name_rcu(net, dev_name);
	8788	+ if (!dev) {
	8789	+ ret = -ENODEV;
	8790	+ goto unlock;
	8791	+ }
	8792	+ if (!dev->addr_len)
	8793	+ memset(sa->sa_data, 0, size);
	8794	+ else
	8795	+ memcpy(sa->sa_data, dev->dev_addr,
	8796	+ min_t(size_t, size, dev->addr_len));
	8797	+ sa->sa_family = dev->type;
	8798	+
	8799	+unlock:
	8800	+ rcu_read_unlock();
	8801	+ up_read(&dev_addr_sem);
	8802	+ return ret;
	8803	+}
	8804	+EXPORT_SYMBOL(dev_get_mac_address);
7967	8805
7968	8806	/**
7969	8807	* dev_change_carrier - Change device carrier
..	..	@@ -8014,12 +8852,80 @@
8014	8852	char *name, size_t len)
8015	8853	{
8016	8854	const struct net_device_ops *ops = dev->netdev_ops;
	8855	+ int err;
8017	8856
8018		- if (!ops->ndo_get_phys_port_name)
8019		- return -EOPNOTSUPP;
8020		- return ops->ndo_get_phys_port_name(dev, name, len);
	8857	+ if (ops->ndo_get_phys_port_name) {
	8858	+ err = ops->ndo_get_phys_port_name(dev, name, len);
	8859	+ if (err != -EOPNOTSUPP)
	8860	+ return err;
	8861	+ }
	8862	+ return devlink_compat_phys_port_name_get(dev, name, len);
8021	8863	}
8022	8864	EXPORT_SYMBOL(dev_get_phys_port_name);
	8865	+
	8866	+/**
	8867	+ * dev_get_port_parent_id - Get the device's port parent identifier
	8868	+ * @dev: network device
	8869	+ * @ppid: pointer to a storage for the port's parent identifier
	8870	+ * @recurse: allow/disallow recursion to lower devices
	8871	+ *
	8872	+ * Get the devices's port parent identifier
	8873	+ */
	8874	+int dev_get_port_parent_id(struct net_device *dev,
	8875	+ struct netdev_phys_item_id *ppid,
	8876	+ bool recurse)
	8877	+{
	8878	+ const struct net_device_ops *ops = dev->netdev_ops;
	8879	+ struct netdev_phys_item_id first = { };
	8880	+ struct net_device *lower_dev;
	8881	+ struct list_head *iter;
	8882	+ int err;
	8883	+
	8884	+ if (ops->ndo_get_port_parent_id) {
	8885	+ err = ops->ndo_get_port_parent_id(dev, ppid);
	8886	+ if (err != -EOPNOTSUPP)
	8887	+ return err;
	8888	+ }
	8889	+
	8890	+ err = devlink_compat_switch_id_get(dev, ppid);
	8891	+ if (!err \|\| err != -EOPNOTSUPP)
	8892	+ return err;
	8893	+
	8894	+ if (!recurse)
	8895	+ return -EOPNOTSUPP;
	8896	+
	8897	+ netdev_for_each_lower_dev(dev, lower_dev, iter) {
	8898	+ err = dev_get_port_parent_id(lower_dev, ppid, recurse);
	8899	+ if (err)
	8900	+ break;
	8901	+ if (!first.id_len)
	8902	+ first = *ppid;
	8903	+ else if (memcmp(&first, ppid, sizeof(*ppid)))
	8904	+ return -EOPNOTSUPP;
	8905	+ }
	8906	+
	8907	+ return err;
	8908	+}
	8909	+EXPORT_SYMBOL(dev_get_port_parent_id);
	8910	+
	8911	+/**
	8912	+ * netdev_port_same_parent_id - Indicate if two network devices have
	8913	+ * the same port parent identifier
	8914	+ * @a: first network device
	8915	+ * @b: second network device
	8916	+ */
	8917	+bool netdev_port_same_parent_id(struct net_device a, struct net_device b)
	8918	+{
	8919	+ struct netdev_phys_item_id a_id = { };
	8920	+ struct netdev_phys_item_id b_id = { };
	8921	+
	8922	+ if (dev_get_port_parent_id(a, &a_id, true) \|\|
	8923	+ dev_get_port_parent_id(b, &b_id, true))
	8924	+ return false;
	8925	+
	8926	+ return netdev_phys_item_id_same(&a_id, &b_id);
	8927	+}
	8928	+EXPORT_SYMBOL(netdev_port_same_parent_id);
8023	8929
8024	8930	/**
8025	8931	* dev_change_proto_down - update protocol port state information
..	..	@@ -8041,67 +8947,495 @@
8041	8947	}
8042	8948	EXPORT_SYMBOL(dev_change_proto_down);
8043	8949
8044		-u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op,
8045		- enum bpf_netdev_command cmd)
	8950	+/**
	8951	+ * dev_change_proto_down_generic - generic implementation for
	8952	+ * ndo_change_proto_down that sets carrier according to
	8953	+ * proto_down.
	8954	+ *
	8955	+ * @dev: device
	8956	+ * @proto_down: new value
	8957	+ */
	8958	+int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
8046	8959	{
8047		- struct netdev_bpf xdp;
	8960	+ if (proto_down)
	8961	+ netif_carrier_off(dev);
	8962	+ else
	8963	+ netif_carrier_on(dev);
	8964	+ dev->proto_down = proto_down;
	8965	+ return 0;
	8966	+}
	8967	+EXPORT_SYMBOL(dev_change_proto_down_generic);
8048	8968
8049		- if (!bpf_op)
8050		- return 0;
	8969	+/**
	8970	+ * dev_change_proto_down_reason - proto down reason
	8971	+ *
	8972	+ * @dev: device
	8973	+ * @mask: proto down mask
	8974	+ * @value: proto down value
	8975	+ */
	8976	+void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
	8977	+ u32 value)
	8978	+{
	8979	+ int b;
8051	8980
8052		- memset(&xdp, 0, sizeof(xdp));
8053		- xdp.command = cmd;
	8981	+ if (!mask) {
	8982	+ dev->proto_down_reason = value;
	8983	+ } else {
	8984	+ for_each_set_bit(b, &mask, 32) {
	8985	+ if (value & (1 << b))
	8986	+ dev->proto_down_reason \|= BIT(b);
	8987	+ else
	8988	+ dev->proto_down_reason &= ~BIT(b);
	8989	+ }
	8990	+ }
	8991	+}
	8992	+EXPORT_SYMBOL(dev_change_proto_down_reason);
8054	8993
8055		- /* Query must always succeed. */
8056		- WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG);
	8994	+struct bpf_xdp_link {
	8995	+ struct bpf_link link;
	8996	+ struct net_device dev; / protected by rtnl_lock, no refcnt held */
	8997	+ int flags;
	8998	+};
8057	8999
8058		- return xdp.prog_id;
	9000	+static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
	9001	+{
	9002	+ if (flags & XDP_FLAGS_HW_MODE)
	9003	+ return XDP_MODE_HW;
	9004	+ if (flags & XDP_FLAGS_DRV_MODE)
	9005	+ return XDP_MODE_DRV;
	9006	+ if (flags & XDP_FLAGS_SKB_MODE)
	9007	+ return XDP_MODE_SKB;
	9008	+ return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
8059	9009	}
8060	9010
8061		-static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op,
8062		- struct netlink_ext_ack *extack, u32 flags,
8063		- struct bpf_prog *prog)
	9011	+static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
	9012	+{
	9013	+ switch (mode) {
	9014	+ case XDP_MODE_SKB:
	9015	+ return generic_xdp_install;
	9016	+ case XDP_MODE_DRV:
	9017	+ case XDP_MODE_HW:
	9018	+ return dev->netdev_ops->ndo_bpf;
	9019	+ default:
	9020	+ return NULL;
	9021	+ };
	9022	+}
	9023	+
	9024	+static struct bpf_xdp_link dev_xdp_link(struct net_device dev,
	9025	+ enum bpf_xdp_mode mode)
	9026	+{
	9027	+ return dev->xdp_state[mode].link;
	9028	+}
	9029	+
	9030	+static struct bpf_prog dev_xdp_prog(struct net_device dev,
	9031	+ enum bpf_xdp_mode mode)
	9032	+{
	9033	+ struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
	9034	+
	9035	+ if (link)
	9036	+ return link->link.prog;
	9037	+ return dev->xdp_state[mode].prog;
	9038	+}
	9039	+
	9040	+static u8 dev_xdp_prog_count(struct net_device *dev)
	9041	+{
	9042	+ u8 count = 0;
	9043	+ int i;
	9044	+
	9045	+ for (i = 0; i < __MAX_XDP_MODE; i++)
	9046	+ if (dev->xdp_state[i].prog \|\| dev->xdp_state[i].link)
	9047	+ count++;
	9048	+ return count;
	9049	+}
	9050	+
	9051	+u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
	9052	+{
	9053	+ struct bpf_prog *prog = dev_xdp_prog(dev, mode);
	9054	+
	9055	+ return prog ? prog->aux->id : 0;
	9056	+}
	9057	+
	9058	+static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
	9059	+ struct bpf_xdp_link *link)
	9060	+{
	9061	+ dev->xdp_state[mode].link = link;
	9062	+ dev->xdp_state[mode].prog = NULL;
	9063	+}
	9064	+
	9065	+static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
	9066	+ struct bpf_prog *prog)
	9067	+{
	9068	+ dev->xdp_state[mode].link = NULL;
	9069	+ dev->xdp_state[mode].prog = prog;
	9070	+}
	9071	+
	9072	+static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
	9073	+ bpf_op_t bpf_op, struct netlink_ext_ack *extack,
	9074	+ u32 flags, struct bpf_prog *prog)
8064	9075	{
8065	9076	struct netdev_bpf xdp;
	9077	+ int err;
8066	9078
8067	9079	memset(&xdp, 0, sizeof(xdp));
8068		- if (flags & XDP_FLAGS_HW_MODE)
8069		- xdp.command = XDP_SETUP_PROG_HW;
8070		- else
8071		- xdp.command = XDP_SETUP_PROG;
	9080	+ xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
8072	9081	xdp.extack = extack;
8073	9082	xdp.flags = flags;
8074	9083	xdp.prog = prog;
8075	9084
8076		- return bpf_op(dev, &xdp);
	9085	+ /* Drivers assume refcnt is already incremented (i.e, prog pointer is
	9086	+ * "moved" into driver), so they don't increment it on their own, but
	9087	+ * they do decrement refcnt when program is detached or replaced.
	9088	+ * Given net_device also owns link/prog, we need to bump refcnt here
	9089	+ * to prevent drivers from underflowing it.
	9090	+ */
	9091	+ if (prog)
	9092	+ bpf_prog_inc(prog);
	9093	+ err = bpf_op(dev, &xdp);
	9094	+ if (err) {
	9095	+ if (prog)
	9096	+ bpf_prog_put(prog);
	9097	+ return err;
	9098	+ }
	9099	+
	9100	+ if (mode != XDP_MODE_HW)
	9101	+ bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
	9102	+
	9103	+ return 0;
8077	9104	}
8078	9105
8079	9106	static void dev_xdp_uninstall(struct net_device *dev)
8080	9107	{
8081		- struct netdev_bpf xdp;
8082		- bpf_op_t ndo_bpf;
	9108	+ struct bpf_xdp_link *link;
	9109	+ struct bpf_prog *prog;
	9110	+ enum bpf_xdp_mode mode;
	9111	+ bpf_op_t bpf_op;
8083	9112
8084		- /* Remove generic XDP */
8085		- WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL));
	9113	+ ASSERT_RTNL();
8086	9114
8087		- /* Remove from the driver */
8088		- ndo_bpf = dev->netdev_ops->ndo_bpf;
8089		- if (!ndo_bpf)
8090		- return;
	9115	+ for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
	9116	+ prog = dev_xdp_prog(dev, mode);
	9117	+ if (!prog)
	9118	+ continue;
8091	9119
8092		- memset(&xdp, 0, sizeof(xdp));
8093		- xdp.command = XDP_QUERY_PROG;
8094		- WARN_ON(ndo_bpf(dev, &xdp));
8095		- if (xdp.prog_id)
8096		- WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8097		- NULL));
	9120	+ bpf_op = dev_xdp_bpf_op(dev, mode);
	9121	+ if (!bpf_op)
	9122	+ continue;
8098	9123
8099		- /* Remove HW offload */
8100		- memset(&xdp, 0, sizeof(xdp));
8101		- xdp.command = XDP_QUERY_PROG_HW;
8102		- if (!ndo_bpf(dev, &xdp) && xdp.prog_id)
8103		- WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8104		- NULL));
	9124	+ WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
	9125	+
	9126	+ /* auto-detach link from net device */
	9127	+ link = dev_xdp_link(dev, mode);
	9128	+ if (link)
	9129	+ link->dev = NULL;
	9130	+ else
	9131	+ bpf_prog_put(prog);
	9132	+
	9133	+ dev_xdp_set_link(dev, mode, NULL);
	9134	+ }
	9135	+}
	9136	+
	9137	+static int dev_xdp_attach(struct net_device dev, struct netlink_ext_ack extack,
	9138	+ struct bpf_xdp_link link, struct bpf_prog new_prog,
	9139	+ struct bpf_prog *old_prog, u32 flags)
	9140	+{
	9141	+ unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
	9142	+ struct bpf_prog *cur_prog;
	9143	+ enum bpf_xdp_mode mode;
	9144	+ bpf_op_t bpf_op;
	9145	+ int err;
	9146	+
	9147	+ ASSERT_RTNL();
	9148	+
	9149	+ /* either link or prog attachment, never both */
	9150	+ if (link && (new_prog \|\| old_prog))
	9151	+ return -EINVAL;
	9152	+ /* link supports only XDP mode flags */
	9153	+ if (link && (flags & ~XDP_FLAGS_MODES)) {
	9154	+ NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
	9155	+ return -EINVAL;
	9156	+ }
	9157	+ /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
	9158	+ if (num_modes > 1) {
	9159	+ NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
	9160	+ return -EINVAL;
	9161	+ }
	9162	+ /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
	9163	+ if (!num_modes && dev_xdp_prog_count(dev) > 1) {
	9164	+ NL_SET_ERR_MSG(extack,
	9165	+ "More than one program loaded, unset mode is ambiguous");
	9166	+ return -EINVAL;
	9167	+ }
	9168	+ /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
	9169	+ if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
	9170	+ NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
	9171	+ return -EINVAL;
	9172	+ }
	9173	+
	9174	+ mode = dev_xdp_mode(dev, flags);
	9175	+ /* can't replace attached link */
	9176	+ if (dev_xdp_link(dev, mode)) {
	9177	+ NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
	9178	+ return -EBUSY;
	9179	+ }
	9180	+
	9181	+ cur_prog = dev_xdp_prog(dev, mode);
	9182	+ /* can't replace attached prog with link */
	9183	+ if (link && cur_prog) {
	9184	+ NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
	9185	+ return -EBUSY;
	9186	+ }
	9187	+ if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
	9188	+ NL_SET_ERR_MSG(extack, "Active program does not match expected");
	9189	+ return -EEXIST;
	9190	+ }
	9191	+
	9192	+ /* put effective new program into new_prog */
	9193	+ if (link)
	9194	+ new_prog = link->link.prog;
	9195	+
	9196	+ if (new_prog) {
	9197	+ bool offload = mode == XDP_MODE_HW;
	9198	+ enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
	9199	+ ? XDP_MODE_DRV : XDP_MODE_SKB;
	9200	+
	9201	+ if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
	9202	+ NL_SET_ERR_MSG(extack, "XDP program already attached");
	9203	+ return -EBUSY;
	9204	+ }
	9205	+ if (!offload && dev_xdp_prog(dev, other_mode)) {
	9206	+ NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
	9207	+ return -EEXIST;
	9208	+ }
	9209	+ if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
	9210	+ NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
	9211	+ return -EINVAL;
	9212	+ }
	9213	+ if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
	9214	+ NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
	9215	+ return -EINVAL;
	9216	+ }
	9217	+ if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
	9218	+ NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
	9219	+ return -EINVAL;
	9220	+ }
	9221	+ }
	9222	+
	9223	+ /* don't call drivers if the effective program didn't change */
	9224	+ if (new_prog != cur_prog) {
	9225	+ bpf_op = dev_xdp_bpf_op(dev, mode);
	9226	+ if (!bpf_op) {
	9227	+ NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
	9228	+ return -EOPNOTSUPP;
	9229	+ }
	9230	+
	9231	+ err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
	9232	+ if (err)
	9233	+ return err;
	9234	+ }
	9235	+
	9236	+ if (link)
	9237	+ dev_xdp_set_link(dev, mode, link);
	9238	+ else
	9239	+ dev_xdp_set_prog(dev, mode, new_prog);
	9240	+ if (cur_prog)
	9241	+ bpf_prog_put(cur_prog);
	9242	+
	9243	+ return 0;
	9244	+}
	9245	+
	9246	+static int dev_xdp_attach_link(struct net_device *dev,
	9247	+ struct netlink_ext_ack *extack,
	9248	+ struct bpf_xdp_link *link)
	9249	+{
	9250	+ return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
	9251	+}
	9252	+
	9253	+static int dev_xdp_detach_link(struct net_device *dev,
	9254	+ struct netlink_ext_ack *extack,
	9255	+ struct bpf_xdp_link *link)
	9256	+{
	9257	+ enum bpf_xdp_mode mode;
	9258	+ bpf_op_t bpf_op;
	9259	+
	9260	+ ASSERT_RTNL();
	9261	+
	9262	+ mode = dev_xdp_mode(dev, link->flags);
	9263	+ if (dev_xdp_link(dev, mode) != link)
	9264	+ return -EINVAL;
	9265	+
	9266	+ bpf_op = dev_xdp_bpf_op(dev, mode);
	9267	+ WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
	9268	+ dev_xdp_set_link(dev, mode, NULL);
	9269	+ return 0;
	9270	+}
	9271	+
	9272	+static void bpf_xdp_link_release(struct bpf_link *link)
	9273	+{
	9274	+ struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
	9275	+
	9276	+ rtnl_lock();
	9277	+
	9278	+ /* if racing with net_device's tear down, xdp_link->dev might be
	9279	+ * already NULL, in which case link was already auto-detached
	9280	+ */
	9281	+ if (xdp_link->dev) {
	9282	+ WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
	9283	+ xdp_link->dev = NULL;
	9284	+ }
	9285	+
	9286	+ rtnl_unlock();
	9287	+}
	9288	+
	9289	+static int bpf_xdp_link_detach(struct bpf_link *link)
	9290	+{
	9291	+ bpf_xdp_link_release(link);
	9292	+ return 0;
	9293	+}
	9294	+
	9295	+static void bpf_xdp_link_dealloc(struct bpf_link *link)
	9296	+{
	9297	+ struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
	9298	+
	9299	+ kfree(xdp_link);
	9300	+}
	9301	+
	9302	+static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
	9303	+ struct seq_file *seq)
	9304	+{
	9305	+ struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
	9306	+ u32 ifindex = 0;
	9307	+
	9308	+ rtnl_lock();
	9309	+ if (xdp_link->dev)
	9310	+ ifindex = xdp_link->dev->ifindex;
	9311	+ rtnl_unlock();
	9312	+
	9313	+ seq_printf(seq, "ifindex:\t%u\n", ifindex);
	9314	+}
	9315	+
	9316	+static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
	9317	+ struct bpf_link_info *info)
	9318	+{
	9319	+ struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
	9320	+ u32 ifindex = 0;
	9321	+
	9322	+ rtnl_lock();
	9323	+ if (xdp_link->dev)
	9324	+ ifindex = xdp_link->dev->ifindex;
	9325	+ rtnl_unlock();
	9326	+
	9327	+ info->xdp.ifindex = ifindex;
	9328	+ return 0;
	9329	+}
	9330	+
	9331	+static int bpf_xdp_link_update(struct bpf_link link, struct bpf_prog new_prog,
	9332	+ struct bpf_prog *old_prog)
	9333	+{
	9334	+ struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
	9335	+ enum bpf_xdp_mode mode;
	9336	+ bpf_op_t bpf_op;
	9337	+ int err = 0;
	9338	+
	9339	+ rtnl_lock();
	9340	+
	9341	+ /* link might have been auto-released already, so fail */
	9342	+ if (!xdp_link->dev) {
	9343	+ err = -ENOLINK;
	9344	+ goto out_unlock;
	9345	+ }
	9346	+
	9347	+ if (old_prog && link->prog != old_prog) {
	9348	+ err = -EPERM;
	9349	+ goto out_unlock;
	9350	+ }
	9351	+ old_prog = link->prog;
	9352	+ if (old_prog->type != new_prog->type \|\|
	9353	+ old_prog->expected_attach_type != new_prog->expected_attach_type) {
	9354	+ err = -EINVAL;
	9355	+ goto out_unlock;
	9356	+ }
	9357	+
	9358	+ if (old_prog == new_prog) {
	9359	+ /* no-op, don't disturb drivers */
	9360	+ bpf_prog_put(new_prog);
	9361	+ goto out_unlock;
	9362	+ }
	9363	+
	9364	+ mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
	9365	+ bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
	9366	+ err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
	9367	+ xdp_link->flags, new_prog);
	9368	+ if (err)
	9369	+ goto out_unlock;
	9370	+
	9371	+ old_prog = xchg(&link->prog, new_prog);
	9372	+ bpf_prog_put(old_prog);
	9373	+
	9374	+out_unlock:
	9375	+ rtnl_unlock();
	9376	+ return err;
	9377	+}
	9378	+
	9379	+static const struct bpf_link_ops bpf_xdp_link_lops = {
	9380	+ .release = bpf_xdp_link_release,
	9381	+ .dealloc = bpf_xdp_link_dealloc,
	9382	+ .detach = bpf_xdp_link_detach,
	9383	+ .show_fdinfo = bpf_xdp_link_show_fdinfo,
	9384	+ .fill_link_info = bpf_xdp_link_fill_link_info,
	9385	+ .update_prog = bpf_xdp_link_update,
	9386	+};
	9387	+
	9388	+int bpf_xdp_link_attach(const union bpf_attr attr, struct bpf_prog prog)
	9389	+{
	9390	+ struct net *net = current->nsproxy->net_ns;
	9391	+ struct bpf_link_primer link_primer;
	9392	+ struct bpf_xdp_link *link;
	9393	+ struct net_device *dev;
	9394	+ int err, fd;
	9395	+
	9396	+ rtnl_lock();
	9397	+ dev = dev_get_by_index(net, attr->link_create.target_ifindex);
	9398	+ if (!dev) {
	9399	+ rtnl_unlock();
	9400	+ return -EINVAL;
	9401	+ }
	9402	+
	9403	+ link = kzalloc(sizeof(*link), GFP_USER);
	9404	+ if (!link) {
	9405	+ err = -ENOMEM;
	9406	+ goto unlock;
	9407	+ }
	9408	+
	9409	+ bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
	9410	+ link->dev = dev;
	9411	+ link->flags = attr->link_create.flags;
	9412	+
	9413	+ err = bpf_link_prime(&link->link, &link_primer);
	9414	+ if (err) {
	9415	+ kfree(link);
	9416	+ goto unlock;
	9417	+ }
	9418	+
	9419	+ err = dev_xdp_attach_link(dev, NULL, link);
	9420	+ rtnl_unlock();
	9421	+
	9422	+ if (err) {
	9423	+ link->dev = NULL;
	9424	+ bpf_link_cleanup(&link_primer);
	9425	+ goto out_put_dev;
	9426	+ }
	9427	+
	9428	+ fd = bpf_link_settle(&link_primer);
	9429	+ /* link itself doesn't hold dev's refcnt to not complicate shutdown */
	9430	+ dev_put(dev);
	9431	+ return fd;
	9432	+
	9433	+unlock:
	9434	+ rtnl_unlock();
	9435	+
	9436	+out_put_dev:
	9437	+ dev_put(dev);
	9438	+ return err;
8105	9439	}
8106	9440
8107	9441	/**
..	..	@@ -8109,56 +9443,44 @@
8109	9443	* @dev: device
8110	9444	* @extack: netlink extended ack
8111	9445	* @fd: new program fd or negative value to clear
	9446	+ * @expected_fd: old program fd that userspace expects to replace or clear
8112	9447	* @flags: xdp-related flags
8113	9448	*
8114	9449	* Set or clear a bpf program for a device
8115	9450	*/
8116	9451	int dev_change_xdp_fd(struct net_device dev, struct netlink_ext_ack extack,
8117		- int fd, u32 flags)
	9452	+ int fd, int expected_fd, u32 flags)
8118	9453	{
8119		- const struct net_device_ops *ops = dev->netdev_ops;
8120		- enum bpf_netdev_command query;
8121		- struct bpf_prog *prog = NULL;
8122		- bpf_op_t bpf_op, bpf_chk;
	9454	+ enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
	9455	+ struct bpf_prog new_prog = NULL, old_prog = NULL;
8123	9456	int err;
8124	9457
8125	9458	ASSERT_RTNL();
8126	9459
8127		- query = flags & XDP_FLAGS_HW_MODE ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG;
8128		-
8129		- bpf_op = bpf_chk = ops->ndo_bpf;
8130		- if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE \| XDP_FLAGS_HW_MODE)))
8131		- return -EOPNOTSUPP;
8132		- if (!bpf_op \|\| (flags & XDP_FLAGS_SKB_MODE))
8133		- bpf_op = generic_xdp_install;
8134		- if (bpf_op == bpf_chk)
8135		- bpf_chk = generic_xdp_install;
8136		-
8137	9460	if (fd >= 0) {
8138		- if (__dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG) \|\|
8139		- __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG_HW))
8140		- return -EEXIST;
8141		- if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) &&
8142		- __dev_xdp_query(dev, bpf_op, query))
8143		- return -EBUSY;
	9461	+ new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
	9462	+ mode != XDP_MODE_SKB);
	9463	+ if (IS_ERR(new_prog))
	9464	+ return PTR_ERR(new_prog);
	9465	+ }
8144	9466
8145		- prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
8146		- bpf_op == ops->ndo_bpf);
8147		- if (IS_ERR(prog))
8148		- return PTR_ERR(prog);
8149		-
8150		- if (!(flags & XDP_FLAGS_HW_MODE) &&
8151		- bpf_prog_is_dev_bound(prog->aux)) {
8152		- NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported");
8153		- bpf_prog_put(prog);
8154		- return -EINVAL;
	9467	+ if (expected_fd >= 0) {
	9468	+ old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
	9469	+ mode != XDP_MODE_SKB);
	9470	+ if (IS_ERR(old_prog)) {
	9471	+ err = PTR_ERR(old_prog);
	9472	+ old_prog = NULL;
	9473	+ goto err_out;
8155	9474	}
8156	9475	}
8157	9476
8158		- err = dev_xdp_install(dev, bpf_op, extack, flags, prog);
8159		- if (err < 0 && prog)
8160		- bpf_prog_put(prog);
	9477	+ err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
8161	9478
	9479	+err_out:
	9480	+ if (err && new_prog)
	9481	+ bpf_prog_put(new_prog);
	9482	+ if (old_prog)
	9483	+ bpf_prog_put(old_prog);
8162	9484	return err;
8163	9485	}
8164	9486
..	..	@@ -8190,103 +9512,6 @@
8190	9512	{
8191	9513	list_add_tail(&dev->todo_list, &net_todo_list);
8192	9514	dev_net(dev)->dev_unreg_count++;
8193		-}
8194		-
8195		-static void rollback_registered_many(struct list_head *head)
8196		-{
8197		- struct net_device dev, tmp;
8198		- LIST_HEAD(close_head);
8199		-
8200		- BUG_ON(dev_boot_phase);
8201		- ASSERT_RTNL();
8202		-
8203		- list_for_each_entry_safe(dev, tmp, head, unreg_list) {
8204		- /* Some devices call without registering
8205		- * for initialization unwind. Remove those
8206		- * devices and proceed with the remaining.
8207		- */
8208		- if (dev->reg_state == NETREG_UNINITIALIZED) {
8209		- pr_debug("unregister_netdevice: device %s/%p never was registered\n",
8210		- dev->name, dev);
8211		-
8212		- WARN_ON(1);
8213		- list_del(&dev->unreg_list);
8214		- continue;
8215		- }
8216		- dev->dismantle = true;
8217		- BUG_ON(dev->reg_state != NETREG_REGISTERED);
8218		- }
8219		-
8220		- /* If device is running, close it first. */
8221		- list_for_each_entry(dev, head, unreg_list)
8222		- list_add_tail(&dev->close_list, &close_head);
8223		- dev_close_many(&close_head, true);
8224		-
8225		- list_for_each_entry(dev, head, unreg_list) {
8226		- /* And unlink it from device chain. */
8227		- unlist_netdevice(dev);
8228		-
8229		- dev->reg_state = NETREG_UNREGISTERING;
8230		- }
8231		- flush_all_backlogs();
8232		-
8233		- synchronize_net();
8234		-
8235		- list_for_each_entry(dev, head, unreg_list) {
8236		- struct sk_buff *skb = NULL;
8237		-
8238		- /* Shutdown queueing discipline. */
8239		- dev_shutdown(dev);
8240		-
8241		- dev_xdp_uninstall(dev);
8242		-
8243		- /* Notify protocols, that we are about to destroy
8244		- * this device. They should clean all the things.
8245		- */
8246		- call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8247		-
8248		- if (!dev->rtnl_link_ops \|\|
8249		- dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8250		- skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
8251		- GFP_KERNEL, NULL, 0);
8252		-
8253		- /*
8254		- * Flush the unicast and multicast chains
8255		- */
8256		- dev_uc_flush(dev);
8257		- dev_mc_flush(dev);
8258		-
8259		- if (dev->netdev_ops->ndo_uninit)
8260		- dev->netdev_ops->ndo_uninit(dev);
8261		-
8262		- if (skb)
8263		- rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
8264		-
8265		- /* Notifier chain MUST detach us all upper devices. */
8266		- WARN_ON(netdev_has_any_upper_dev(dev));
8267		- WARN_ON(netdev_has_any_lower_dev(dev));
8268		-
8269		- /* Remove entries from kobject tree */
8270		- netdev_unregister_kobject(dev);
8271		-#ifdef CONFIG_XPS
8272		- /* Remove XPS queueing entries */
8273		- netif_reset_xps_queues_gt(dev, 0);
8274		-#endif
8275		- }
8276		-
8277		- synchronize_net();
8278		-
8279		- list_for_each_entry(dev, head, unreg_list)
8280		- dev_put(dev);
8281		-}
8282		-
8283		-static void rollback_registered(struct net_device *dev)
8284		-{
8285		- LIST_HEAD(single);
8286		-
8287		- list_add(&dev->unreg_list, &single);
8288		- rollback_registered_many(&single);
8289		- list_del(&single);
8290	9515	}
8291	9516
8292	9517	static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
..	..	@@ -8434,7 +9659,7 @@
8434	9659	/* driver might be less strict about feature dependencies */
8435	9660	features = netdev_fix_features(dev, features);
8436	9661
8437		- /* some features can't be enabled if they're off an an upper device */
	9662	+ /* some features can't be enabled if they're off on an upper device */
8438	9663	netdev_for_each_upper_dev_rcu(dev, upper, iter)
8439	9664	features = netdev_sync_upper_features(dev, upper, features);
8440	9665
..	..	@@ -8558,6 +9783,11 @@
8558	9783	else
8559	9784	netif_dormant_off(dev);
8560	9785
	9786	+ if (rootdev->operstate == IF_OPER_TESTING)
	9787	+ netif_testing_on(dev);
	9788	+ else
	9789	+ netif_testing_off(dev);
	9790	+
8561	9791	if (netif_carrier_ok(rootdev))
8562	9792	netif_carrier_on(dev);
8563	9793	else
..	..	@@ -8619,7 +9849,7 @@
8619	9849	/* Initialize queue lock */
8620	9850	spin_lock_init(&queue->_xmit_lock);
8621	9851	netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
8622		- netdev_queue_clear_owner(queue);
	9852	+ queue->xmit_lock_owner = -1;
8623	9853	netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
8624	9854	queue->dev = dev;
8625	9855	#ifdef CONFIG_BQL
..	..	@@ -8698,11 +9928,20 @@
8698	9928	BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
8699	9929	BUG_ON(!net);
8700	9930
	9931	+ ret = ethtool_check_ops(dev->ethtool_ops);
	9932	+ if (ret)
	9933	+ return ret;
	9934	+
8701	9935	spin_lock_init(&dev->addr_list_lock);
8702	9936	netdev_set_addr_lockdep_class(dev);
8703	9937
8704	9938	ret = dev_get_valid_name(net, dev, dev->name);
8705	9939	if (ret < 0)
	9940	+ goto out;
	9941	+
	9942	+ ret = -ENOMEM;
	9943	+ dev->name_node = netdev_name_node_head_alloc(dev);
	9944	+ if (!dev->name_node)
8706	9945	goto out;
8707	9946
8708	9947	/* Init, if this function is available */
..	..	@@ -8711,7 +9950,7 @@
8711	9950	if (ret) {
8712	9951	if (ret > 0)
8713	9952	ret = -EIO;
8714		- goto out;
	9953	+ goto err_free_name;
8715	9954	}
8716	9955	}
8717	9956
..	..	@@ -8733,7 +9972,7 @@
8733	9972	/* Transfer changeable features to wanted_features and enable
8734	9973	* software offloads (GSO and GRO).
8735	9974	*/
8736		- dev->hw_features \|= NETIF_F_SOFT_FEATURES;
	9975	+ dev->hw_features \|= (NETIF_F_SOFT_FEATURES \| NETIF_F_SOFT_FEATURES_OFF);
8737	9976	dev->features \|= NETIF_F_SOFT_FEATURES;
8738	9977
8739	9978	if (dev->netdev_ops->ndo_udp_tunnel_add) {
..	..	@@ -8811,17 +10050,10 @@
8811	10050	ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
8812	10051	ret = notifier_to_errno(ret);
8813	10052	if (ret) {
8814		- rollback_registered(dev);
8815		- rcu_barrier();
8816		-
8817		- dev->reg_state = NETREG_UNREGISTERED;
8818		- /* We should put the kobject that hold in
8819		- * netdev_unregister_kobject(), otherwise
8820		- * the net device cannot be freed when
8821		- * driver calls free_netdev(), because the
8822		- * kobject is being hold.
8823		- */
8824		- kobject_put(&dev->dev.kobj);
	10053	+ /* Expect explicit free_netdev() on failure */
	10054	+ dev->needs_free_netdev = false;
	10055	+ unregister_netdevice_queue(dev, NULL);
	10056	+ goto out;
8825	10057	}
8826	10058	/*
8827	10059	* Prevent userspace races by waiting until the network
..	..	@@ -8839,6 +10071,8 @@
8839	10071	dev->netdev_ops->ndo_uninit(dev);
8840	10072	if (dev->priv_destructor)
8841	10073	dev->priv_destructor(dev);
	10074	+err_free_name:
	10075	+ netdev_name_node_free(dev->name_node);
8842	10076	goto out;
8843	10077	}
8844	10078	EXPORT_SYMBOL(register_netdevice);
..	..	@@ -8922,6 +10156,8 @@
8922	10156	}
8923	10157	EXPORT_SYMBOL(netdev_refcnt_read);
8924	10158
	10159	+#define WAIT_REFS_MIN_MSECS 1
	10160	+#define WAIT_REFS_MAX_MSECS 250
8925	10161	/**
8926	10162	* netdev_wait_allrefs - wait until all references are gone.
8927	10163	* @dev: target net_device
..	..	@@ -8937,7 +10173,7 @@
8937	10173	static void netdev_wait_allrefs(struct net_device *dev)
8938	10174	{
8939	10175	unsigned long rebroadcast_time, warning_time;
8940		- int refcnt;
	10176	+ int wait = 0, refcnt;
8941	10177
8942	10178	linkwatch_forget_dev(dev);
8943	10179
..	..	@@ -8971,7 +10207,13 @@
8971	10207	rebroadcast_time = jiffies;
8972	10208	}
8973	10209
8974		- msleep(250);
	10210	+ if (!wait) {
	10211	+ rcu_barrier();
	10212	+ wait = WAIT_REFS_MIN_MSECS;
	10213	+ } else {
	10214	+ msleep(wait);
	10215	+ wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
	10216	+ }
8975	10217
8976	10218	refcnt = netdev_refcnt_read(dev);
8977	10219
..	..	@@ -9010,6 +10252,19 @@
9010	10252	void netdev_run_todo(void)
9011	10253	{
9012	10254	struct list_head list;
	10255	+#ifdef CONFIG_LOCKDEP
	10256	+ struct list_head unlink_list;
	10257	+
	10258	+ list_replace_init(&net_unlink_list, &unlink_list);
	10259	+
	10260	+ while (!list_empty(&unlink_list)) {
	10261	+ struct net_device *dev = list_first_entry(&unlink_list,
	10262	+ struct net_device,
	10263	+ unlink_list);
	10264	+ list_del_init(&dev->unlink_list);
	10265	+ dev->nested_level = dev->lower_level - 1;
	10266	+ }
	10267	+#endif
9013	10268
9014	10269	/* Snapshot list, allow later requests */
9015	10270	list_replace_init(&net_todo_list, &list);
..	..	@@ -9121,6 +10376,40 @@
9121	10376	}
9122	10377	EXPORT_SYMBOL(dev_get_stats);
9123	10378
	10379	+/**
	10380	+ * dev_fetch_sw_netstats - get per-cpu network device statistics
	10381	+ * @s: place to store stats
	10382	+ * @netstats: per-cpu network stats to read from
	10383	+ *
	10384	+ * Read per-cpu network statistics and populate the related fields in @s.
	10385	+ */
	10386	+void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
	10387	+ const struct pcpu_sw_netstats __percpu *netstats)
	10388	+{
	10389	+ int cpu;
	10390	+
	10391	+ for_each_possible_cpu(cpu) {
	10392	+ const struct pcpu_sw_netstats *stats;
	10393	+ struct pcpu_sw_netstats tmp;
	10394	+ unsigned int start;
	10395	+
	10396	+ stats = per_cpu_ptr(netstats, cpu);
	10397	+ do {
	10398	+ start = u64_stats_fetch_begin_irq(&stats->syncp);
	10399	+ tmp.rx_packets = stats->rx_packets;
	10400	+ tmp.rx_bytes = stats->rx_bytes;
	10401	+ tmp.tx_packets = stats->tx_packets;
	10402	+ tmp.tx_bytes = stats->tx_bytes;
	10403	+ } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
	10404	+
	10405	+ s->rx_packets += tmp.rx_packets;
	10406	+ s->rx_bytes += tmp.rx_bytes;
	10407	+ s->tx_packets += tmp.tx_packets;
	10408	+ s->tx_bytes += tmp.tx_bytes;
	10409	+ }
	10410	+}
	10411	+EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
	10412	+
9124	10413	struct netdev_queue dev_ingress_queue_create(struct net_device dev)
9125	10414	{
9126	10415	struct netdev_queue *queue = dev_ingress_queue(dev);
..	..	@@ -9222,6 +10511,10 @@
9222	10511	dev->gso_max_segs = GSO_MAX_SEGS;
9223	10512	dev->upper_level = 1;
9224	10513	dev->lower_level = 1;
	10514	+#ifdef CONFIG_LOCKDEP
	10515	+ dev->nested_level = 0;
	10516	+ INIT_LIST_HEAD(&dev->unlink_list);
	10517	+#endif
9225	10518
9226	10519	INIT_LIST_HEAD(&dev->napi_list);
9227	10520	INIT_LIST_HEAD(&dev->unreg_list);
..	..	@@ -9231,6 +10524,7 @@
9231	10524	INIT_LIST_HEAD(&dev->adj_list.lower);
9232	10525	INIT_LIST_HEAD(&dev->ptype_all);
9233	10526	INIT_LIST_HEAD(&dev->ptype_specific);
	10527	+ INIT_LIST_HEAD(&dev->net_notifier_list);
9234	10528	#ifdef CONFIG_NET_SCHED
9235	10529	hash_init(dev->qdisc_hash);
9236	10530	#endif
..	..	@@ -9288,6 +10582,17 @@
9288	10582	struct napi_struct p, n;
9289	10583
9290	10584	might_sleep();
	10585	+
	10586	+ /* When called immediately after register_netdevice() failed the unwind
	10587	+ * handling may still be dismantling the device. Handle that case by
	10588	+ * deferring the free.
	10589	+ */
	10590	+ if (dev->reg_state == NETREG_UNREGISTERING) {
	10591	+ ASSERT_RTNL();
	10592	+ dev->needs_free_netdev = true;
	10593	+ return;
	10594	+ }
	10595	+
9291	10596	netif_free_tx_queues(dev);
9292	10597	netif_free_rx_queues(dev);
9293	10598
..	..	@@ -9301,6 +10606,8 @@
9301	10606
9302	10607	free_percpu(dev->pcpu_refcnt);
9303	10608	dev->pcpu_refcnt = NULL;
	10609	+ free_percpu(dev->xdp_bulkq);
	10610	+ dev->xdp_bulkq = NULL;
9304	10611
9305	10612	/* Compatibility with error handling in drivers */
9306	10613	if (dev->reg_state == NETREG_UNINITIALIZED) {
..	..	@@ -9352,9 +10659,10 @@
9352	10659	if (head) {
9353	10660	list_move_tail(&dev->unreg_list, head);
9354	10661	} else {
9355		- rollback_registered(dev);
9356		- /* Finish processing unregister after unlock */
9357		- net_set_todo(dev);
	10662	+ LIST_HEAD(single);
	10663	+
	10664	+ list_add(&dev->unreg_list, &single);
	10665	+ unregister_netdevice_many(&single);
9358	10666	}
9359	10667	}
9360	10668	EXPORT_SYMBOL(unregister_netdevice_queue);
..	..	@@ -9368,14 +10676,100 @@
9368	10676	*/
9369	10677	void unregister_netdevice_many(struct list_head *head)
9370	10678	{
9371		- struct net_device *dev;
	10679	+ struct net_device dev, tmp;
	10680	+ LIST_HEAD(close_head);
9372	10681
9373		- if (!list_empty(head)) {
9374		- rollback_registered_many(head);
9375		- list_for_each_entry(dev, head, unreg_list)
9376		- net_set_todo(dev);
9377		- list_del(head);
	10682	+ BUG_ON(dev_boot_phase);
	10683	+ ASSERT_RTNL();
	10684	+
	10685	+ if (list_empty(head))
	10686	+ return;
	10687	+
	10688	+ list_for_each_entry_safe(dev, tmp, head, unreg_list) {
	10689	+ /* Some devices call without registering
	10690	+ * for initialization unwind. Remove those
	10691	+ * devices and proceed with the remaining.
	10692	+ */
	10693	+ if (dev->reg_state == NETREG_UNINITIALIZED) {
	10694	+ pr_debug("unregister_netdevice: device %s/%p never was registered\n",
	10695	+ dev->name, dev);
	10696	+
	10697	+ WARN_ON(1);
	10698	+ list_del(&dev->unreg_list);
	10699	+ continue;
	10700	+ }
	10701	+ dev->dismantle = true;
	10702	+ BUG_ON(dev->reg_state != NETREG_REGISTERED);
9378	10703	}
	10704	+
	10705	+ /* If device is running, close it first. */
	10706	+ list_for_each_entry(dev, head, unreg_list)
	10707	+ list_add_tail(&dev->close_list, &close_head);
	10708	+ dev_close_many(&close_head, true);
	10709	+
	10710	+ list_for_each_entry(dev, head, unreg_list) {
	10711	+ /* And unlink it from device chain. */
	10712	+ unlist_netdevice(dev);
	10713	+
	10714	+ dev->reg_state = NETREG_UNREGISTERING;
	10715	+ }
	10716	+ flush_all_backlogs();
	10717	+
	10718	+ synchronize_net();
	10719	+
	10720	+ list_for_each_entry(dev, head, unreg_list) {
	10721	+ struct sk_buff *skb = NULL;
	10722	+
	10723	+ /* Shutdown queueing discipline. */
	10724	+ dev_shutdown(dev);
	10725	+
	10726	+ dev_xdp_uninstall(dev);
	10727	+
	10728	+ /* Notify protocols, that we are about to destroy
	10729	+ * this device. They should clean all the things.
	10730	+ */
	10731	+ call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
	10732	+
	10733	+ if (!dev->rtnl_link_ops \|\|
	10734	+ dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
	10735	+ skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
	10736	+ GFP_KERNEL, NULL, 0);
	10737	+
	10738	+ /*
	10739	+ * Flush the unicast and multicast chains
	10740	+ */
	10741	+ dev_uc_flush(dev);
	10742	+ dev_mc_flush(dev);
	10743	+
	10744	+ netdev_name_node_alt_flush(dev);
	10745	+ netdev_name_node_free(dev->name_node);
	10746	+
	10747	+ if (dev->netdev_ops->ndo_uninit)
	10748	+ dev->netdev_ops->ndo_uninit(dev);
	10749	+
	10750	+ if (skb)
	10751	+ rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
	10752	+
	10753	+ /* Notifier chain MUST detach us all upper devices. */
	10754	+ WARN_ON(netdev_has_any_upper_dev(dev));
	10755	+ WARN_ON(netdev_has_any_lower_dev(dev));
	10756	+
	10757	+ /* Remove entries from kobject tree */
	10758	+ netdev_unregister_kobject(dev);
	10759	+#ifdef CONFIG_XPS
	10760	+ /* Remove XPS queueing entries */
	10761	+ netif_reset_xps_queues_gt(dev, 0);
	10762	+#endif
	10763	+ }
	10764	+
	10765	+ synchronize_net();
	10766	+
	10767	+ list_for_each_entry(dev, head, unreg_list) {
	10768	+ dev_put(dev);
	10769	+ net_set_todo(dev);
	10770	+ }
	10771	+
	10772	+ list_del(head);
9379	10773	}
9380	10774	EXPORT_SYMBOL(unregister_netdevice_many);
9381	10775
..	..	@@ -9414,6 +10808,7 @@
9414	10808
9415	10809	int dev_change_net_namespace(struct net_device dev, struct net net, const char *pat)
9416	10810	{
	10811	+ struct net *net_old = dev_net(dev);
9417	10812	int err, new_nsid, new_ifindex;
9418	10813
9419	10814	ASSERT_RTNL();
..	..	@@ -9429,7 +10824,7 @@
9429	10824
9430	10825	/* Get out if there is nothing todo */
9431	10826	err = 0;
9432		- if (net_eq(dev_net(dev), net))
	10827	+ if (net_eq(net_old, net))
9433	10828	goto out;
9434	10829
9435	10830	/* Pick the destination device name, and ensure
..	..	@@ -9490,6 +10885,9 @@
9490	10885	kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
9491	10886	netdev_adjacent_del_links(dev);
9492	10887
	10888	+ /* Move per-net netdevice notifiers that are following the netdevice */
	10889	+ move_netdevice_notifiers_dev_net(dev, net);
	10890	+
9493	10891	/* Actually switch the network namespace */
9494	10892	dev_net_set(dev, net);
9495	10893	dev->ifindex = new_ifindex;
..	..	@@ -9500,6 +10898,12 @@
9500	10898
9501	10899	/* Fixup kobjects */
9502	10900	err = device_rename(&dev->dev, dev->name);
	10901	+ WARN_ON(err);
	10902	+
	10903	+ /* Adapt owner in case owning user namespace of target network
	10904	+ * namespace is different from the original one.
	10905	+ */
	10906	+ err = netdev_change_owner(dev, net_old, net);
9503	10907	WARN_ON(err);
9504	10908
9505	10909	/* Add the device back in the hashes */
..	..	@@ -9566,7 +10970,6 @@
9566	10970
9567	10971	raise_softirq_irqoff(NET_TX_SOFTIRQ);
9568	10972	local_irq_enable();
9569		- preempt_check_resched_rt();
9570	10973
9571	10974	#ifdef CONFIG_RPS
9572	10975	remsd = oldsd->rps_ipi_list;
..	..	@@ -9580,12 +10983,9 @@
9580	10983	netif_rx_ni(skb);
9581	10984	input_queue_head_incr(oldsd);
9582	10985	}
9583		- while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
	10986	+ while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
9584	10987	netif_rx_ni(skb);
9585	10988	input_queue_head_incr(oldsd);
9586		- }
9587		- while ((skb = __skb_dequeue(&oldsd->tofree_queue))) {
9588		- kfree_skb(skb);
9589	10989	}
9590	10990
9591	10991	return 0;
..	..	@@ -9636,7 +11036,7 @@
9636	11036	static int __net_init netdev_init(struct net *net)
9637	11037	{
9638	11038	BUILD_BUG_ON(GRO_HASH_BUCKETS >
9639		- 8 * FIELD_SIZEOF(struct napi_struct, gro_bitmask));
	11039	+ 8 * sizeof_field(struct napi_struct, gro_bitmask));
9640	11040
9641	11041	if (net != &init_net)
9642	11042	INIT_LIST_HEAD(&net->dev_base_head);
..	..	@@ -9648,6 +11048,8 @@
9648	11048	net->dev_index_head = netdev_create_hash();
9649	11049	if (net->dev_index_head == NULL)
9650	11050	goto err_idx;
	11051	+
	11052	+ RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
9651	11053
9652	11054	return 0;
9653	11055
..	..	@@ -9770,7 +11172,7 @@
9770	11172	continue;
9771	11173
9772	11174	/* Leave virtual devices for the generic cleanup */
9773		- if (dev->rtnl_link_ops)
	11175	+ if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
9774	11176	continue;
9775	11177
9776	11178	/* Push remaining network devices to init_net */
..	..	@@ -9897,9 +11299,8 @@
9897	11299
9898	11300	INIT_WORK(flush, flush_backlog);
9899	11301
9900		- skb_queue_head_init_raw(&sd->input_pkt_queue);
9901		- skb_queue_head_init_raw(&sd->process_queue);
9902		- skb_queue_head_init_raw(&sd->tofree_queue);
	11302	+ skb_queue_head_init(&sd->input_pkt_queue);
	11303	+ skb_queue_head_init(&sd->process_queue);
9903	11304	#ifdef CONFIG_XFRM_OFFLOAD
9904	11305	skb_queue_head_init(&sd->xfrm_backlog);
9905	11306	#endif