xref: /linux/net/core/dev.c (revision 2a6b6c9a226279b4f6668450ddb21ae655558087)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *      NET3    Protocol independent device support routines.
4  *
5  *	Derived from the non IP parts of dev.c 1.0.19
6  *              Authors:	Ross Biro
7  *				Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
8  *				Mark Evans, <evansmp@uhura.aston.ac.uk>
9  *
10  *	Additional Authors:
11  *		Florian la Roche <rzsfl@rz.uni-sb.de>
12  *		Alan Cox <gw4pts@gw4pts.ampr.org>
13  *		David Hinds <dahinds@users.sourceforge.net>
14  *		Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
15  *		Adam Sulmicki <adam@cfar.umd.edu>
16  *              Pekka Riikonen <priikone@poesidon.pspt.fi>
17  *
18  *	Changes:
19  *              D.J. Barrow     :       Fixed bug where dev->refcnt gets set
20  *                                      to 2 if register_netdev gets called
21  *                                      before net_dev_init & also removed a
22  *                                      few lines of code in the process.
23  *		Alan Cox	:	device private ioctl copies fields back.
24  *		Alan Cox	:	Transmit queue code does relevant
25  *					stunts to keep the queue safe.
26  *		Alan Cox	:	Fixed double lock.
27  *		Alan Cox	:	Fixed promisc NULL pointer trap
28  *		????????	:	Support the full private ioctl range
29  *		Alan Cox	:	Moved ioctl permission check into
30  *					drivers
31  *		Tim Kordas	:	SIOCADDMULTI/SIOCDELMULTI
32  *		Alan Cox	:	100 backlog just doesn't cut it when
33  *					you start doing multicast video 8)
34  *		Alan Cox	:	Rewrote net_bh and list manager.
35  *              Alan Cox        :       Fix ETH_P_ALL echoback lengths.
36  *		Alan Cox	:	Took out transmit every packet pass
37  *					Saved a few bytes in the ioctl handler
38  *		Alan Cox	:	Network driver sets packet type before
39  *					calling netif_rx. Saves a function
40  *					call a packet.
41  *		Alan Cox	:	Hashed net_bh()
42  *		Richard Kooijman:	Timestamp fixes.
43  *		Alan Cox	:	Wrong field in SIOCGIFDSTADDR
44  *		Alan Cox	:	Device lock protection.
45  *              Alan Cox        :       Fixed nasty side effect of device close
46  *					changes.
47  *		Rudi Cilibrasi	:	Pass the right thing to
48  *					set_mac_address()
49  *		Dave Miller	:	32bit quantity for the device lock to
50  *					make it work out on a Sparc.
51  *		Bjorn Ekwall	:	Added KERNELD hack.
52  *		Alan Cox	:	Cleaned up the backlog initialise.
53  *		Craig Metz	:	SIOCGIFCONF fix if space for under
54  *					1 device.
55  *	    Thomas Bogendoerfer :	Return ENODEV for dev_open, if there
56  *					is no device open function.
57  *		Andi Kleen	:	Fix error reporting for SIOCGIFCONF
58  *	    Michael Chastain	:	Fix signed/unsigned for SIOCGIFCONF
59  *		Cyrus Durgin	:	Cleaned for KMOD
60  *		Adam Sulmicki   :	Bug Fix : Network Device Unload
61  *					A network device unload needs to purge
62  *					the backlog queue.
63  *	Paul Rusty Russell	:	SIOCSIFNAME
64  *              Pekka Riikonen  :	Netdev boot-time settings code
65  *              Andrew Morton   :       Make unregister_netdevice wait
66  *                                      indefinitely on dev->refcnt
67  *              J Hadi Salim    :       - Backlog queue sampling
68  *				        - netif_rx() feedback
69  */
70 
71 #include <linux/uaccess.h>
72 #include <linux/bitmap.h>
73 #include <linux/capability.h>
74 #include <linux/cpu.h>
75 #include <linux/types.h>
76 #include <linux/kernel.h>
77 #include <linux/hash.h>
78 #include <linux/slab.h>
79 #include <linux/sched.h>
80 #include <linux/sched/isolation.h>
81 #include <linux/sched/mm.h>
82 #include <linux/smpboot.h>
83 #include <linux/mutex.h>
84 #include <linux/rwsem.h>
85 #include <linux/string.h>
86 #include <linux/mm.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/skbuff.h>
96 #include <linux/kthread.h>
97 #include <linux/bpf.h>
98 #include <linux/bpf_trace.h>
99 #include <net/net_namespace.h>
100 #include <net/sock.h>
101 #include <net/busy_poll.h>
102 #include <linux/rtnetlink.h>
103 #include <linux/stat.h>
104 #include <net/dsa.h>
105 #include <net/dst.h>
106 #include <net/dst_metadata.h>
107 #include <net/gro.h>
108 #include <net/pkt_sched.h>
109 #include <net/pkt_cls.h>
110 #include <net/checksum.h>
111 #include <net/xfrm.h>
112 #include <net/tcx.h>
113 #include <linux/highmem.h>
114 #include <linux/init.h>
115 #include <linux/module.h>
116 #include <linux/netpoll.h>
117 #include <linux/rcupdate.h>
118 #include <linux/delay.h>
119 #include <net/iw_handler.h>
120 #include <asm/current.h>
121 #include <linux/audit.h>
122 #include <linux/dmaengine.h>
123 #include <linux/err.h>
124 #include <linux/ctype.h>
125 #include <linux/if_arp.h>
126 #include <linux/if_vlan.h>
127 #include <linux/ip.h>
128 #include <net/ip.h>
129 #include <net/mpls.h>
130 #include <linux/ipv6.h>
131 #include <linux/in.h>
132 #include <linux/jhash.h>
133 #include <linux/random.h>
134 #include <trace/events/napi.h>
135 #include <trace/events/net.h>
136 #include <trace/events/skb.h>
137 #include <trace/events/qdisc.h>
138 #include <trace/events/xdp.h>
139 #include <linux/inetdevice.h>
140 #include <linux/cpu_rmap.h>
141 #include <linux/static_key.h>
142 #include <linux/hashtable.h>
143 #include <linux/vmalloc.h>
144 #include <linux/if_macvlan.h>
145 #include <linux/errqueue.h>
146 #include <linux/hrtimer.h>
147 #include <linux/netfilter_netdev.h>
148 #include <linux/crash_dump.h>
149 #include <linux/sctp.h>
150 #include <net/udp_tunnel.h>
151 #include <linux/net_namespace.h>
152 #include <linux/indirect_call_wrapper.h>
153 #include <net/devlink.h>
154 #include <linux/pm_runtime.h>
155 #include <linux/prandom.h>
156 #include <linux/once_lite.h>
157 #include <net/netdev_rx_queue.h>
158 #include <net/page_pool/types.h>
159 #include <net/page_pool/helpers.h>
160 #include <net/rps.h>
161 
162 #include "dev.h"
163 #include "net-sysfs.h"
164 
165 static DEFINE_SPINLOCK(ptype_lock);
166 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
167 
168 static int netif_rx_internal(struct sk_buff *skb);
169 static int call_netdevice_notifiers_extack(unsigned long val,
170 					   struct net_device *dev,
171 					   struct netlink_ext_ack *extack);
172 
173 static DEFINE_MUTEX(ifalias_mutex);
174 
175 /* protects napi_hash addition/deletion and napi_gen_id */
176 static DEFINE_SPINLOCK(napi_hash_lock);
177 
178 static unsigned int napi_gen_id = NR_CPUS;
179 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
180 
181 static DECLARE_RWSEM(devnet_rename_sem);
182 
183 static inline void dev_base_seq_inc(struct net *net)
184 {
185 	unsigned int val = net->dev_base_seq + 1;
186 
187 	WRITE_ONCE(net->dev_base_seq, val ?: 1);
188 }
189 
190 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
191 {
192 	unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
193 
194 	return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
195 }
196 
197 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
198 {
199 	return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
200 }
201 
202 #ifndef CONFIG_PREEMPT_RT
203 
204 static DEFINE_STATIC_KEY_FALSE(use_backlog_threads_key);
205 
206 static int __init setup_backlog_napi_threads(char *arg)
207 {
208 	static_branch_enable(&use_backlog_threads_key);
209 	return 0;
210 }
211 early_param("thread_backlog_napi", setup_backlog_napi_threads);
212 
213 static bool use_backlog_threads(void)
214 {
215 	return static_branch_unlikely(&use_backlog_threads_key);
216 }
217 
218 #else
219 
220 static bool use_backlog_threads(void)
221 {
222 	return true;
223 }
224 
225 #endif
226 
227 static inline void backlog_lock_irq_save(struct softnet_data *sd,
228 					 unsigned long *flags)
229 {
230 	if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
231 		spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags);
232 	else
233 		local_irq_save(*flags);
234 }
235 
236 static inline void backlog_lock_irq_disable(struct softnet_data *sd)
237 {
238 	if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
239 		spin_lock_irq(&sd->input_pkt_queue.lock);
240 	else
241 		local_irq_disable();
242 }
243 
244 static inline void backlog_unlock_irq_restore(struct softnet_data *sd,
245 					      unsigned long *flags)
246 {
247 	if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
248 		spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags);
249 	else
250 		local_irq_restore(*flags);
251 }
252 
253 static inline void backlog_unlock_irq_enable(struct softnet_data *sd)
254 {
255 	if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
256 		spin_unlock_irq(&sd->input_pkt_queue.lock);
257 	else
258 		local_irq_enable();
259 }
260 
261 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
262 						       const char *name)
263 {
264 	struct netdev_name_node *name_node;
265 
266 	name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
267 	if (!name_node)
268 		return NULL;
269 	INIT_HLIST_NODE(&name_node->hlist);
270 	name_node->dev = dev;
271 	name_node->name = name;
272 	return name_node;
273 }
274 
275 static struct netdev_name_node *
276 netdev_name_node_head_alloc(struct net_device *dev)
277 {
278 	struct netdev_name_node *name_node;
279 
280 	name_node = netdev_name_node_alloc(dev, dev->name);
281 	if (!name_node)
282 		return NULL;
283 	INIT_LIST_HEAD(&name_node->list);
284 	return name_node;
285 }
286 
287 static void netdev_name_node_free(struct netdev_name_node *name_node)
288 {
289 	kfree(name_node);
290 }
291 
292 static void netdev_name_node_add(struct net *net,
293 				 struct netdev_name_node *name_node)
294 {
295 	hlist_add_head_rcu(&name_node->hlist,
296 			   dev_name_hash(net, name_node->name));
297 }
298 
299 static void netdev_name_node_del(struct netdev_name_node *name_node)
300 {
301 	hlist_del_rcu(&name_node->hlist);
302 }
303 
304 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
305 							const char *name)
306 {
307 	struct hlist_head *head = dev_name_hash(net, name);
308 	struct netdev_name_node *name_node;
309 
310 	hlist_for_each_entry(name_node, head, hlist)
311 		if (!strcmp(name_node->name, name))
312 			return name_node;
313 	return NULL;
314 }
315 
316 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
317 							    const char *name)
318 {
319 	struct hlist_head *head = dev_name_hash(net, name);
320 	struct netdev_name_node *name_node;
321 
322 	hlist_for_each_entry_rcu(name_node, head, hlist)
323 		if (!strcmp(name_node->name, name))
324 			return name_node;
325 	return NULL;
326 }
327 
328 bool netdev_name_in_use(struct net *net, const char *name)
329 {
330 	return netdev_name_node_lookup(net, name);
331 }
332 EXPORT_SYMBOL(netdev_name_in_use);
333 
334 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
335 {
336 	struct netdev_name_node *name_node;
337 	struct net *net = dev_net(dev);
338 
339 	name_node = netdev_name_node_lookup(net, name);
340 	if (name_node)
341 		return -EEXIST;
342 	name_node = netdev_name_node_alloc(dev, name);
343 	if (!name_node)
344 		return -ENOMEM;
345 	netdev_name_node_add(net, name_node);
346 	/* The node that holds dev->name acts as a head of per-device list. */
347 	list_add_tail_rcu(&name_node->list, &dev->name_node->list);
348 
349 	return 0;
350 }
351 
352 static void netdev_name_node_alt_free(struct rcu_head *head)
353 {
354 	struct netdev_name_node *name_node =
355 		container_of(head, struct netdev_name_node, rcu);
356 
357 	kfree(name_node->name);
358 	netdev_name_node_free(name_node);
359 }
360 
361 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
362 {
363 	netdev_name_node_del(name_node);
364 	list_del(&name_node->list);
365 	call_rcu(&name_node->rcu, netdev_name_node_alt_free);
366 }
367 
368 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
369 {
370 	struct netdev_name_node *name_node;
371 	struct net *net = dev_net(dev);
372 
373 	name_node = netdev_name_node_lookup(net, name);
374 	if (!name_node)
375 		return -ENOENT;
376 	/* lookup might have found our primary name or a name belonging
377 	 * to another device.
378 	 */
379 	if (name_node == dev->name_node || name_node->dev != dev)
380 		return -EINVAL;
381 
382 	__netdev_name_node_alt_destroy(name_node);
383 	return 0;
384 }
385 
386 static void netdev_name_node_alt_flush(struct net_device *dev)
387 {
388 	struct netdev_name_node *name_node, *tmp;
389 
390 	list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list) {
391 		list_del(&name_node->list);
392 		netdev_name_node_alt_free(&name_node->rcu);
393 	}
394 }
395 
396 /* Device list insertion */
397 static void list_netdevice(struct net_device *dev)
398 {
399 	struct netdev_name_node *name_node;
400 	struct net *net = dev_net(dev);
401 
402 	ASSERT_RTNL();
403 
404 	list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
405 	netdev_name_node_add(net, dev->name_node);
406 	hlist_add_head_rcu(&dev->index_hlist,
407 			   dev_index_hash(net, dev->ifindex));
408 
409 	netdev_for_each_altname(dev, name_node)
410 		netdev_name_node_add(net, name_node);
411 
412 	/* We reserved the ifindex, this can't fail */
413 	WARN_ON(xa_store(&net->dev_by_index, dev->ifindex, dev, GFP_KERNEL));
414 
415 	dev_base_seq_inc(net);
416 }
417 
418 /* Device list removal
419  * caller must respect a RCU grace period before freeing/reusing dev
420  */
421 static void unlist_netdevice(struct net_device *dev)
422 {
423 	struct netdev_name_node *name_node;
424 	struct net *net = dev_net(dev);
425 
426 	ASSERT_RTNL();
427 
428 	xa_erase(&net->dev_by_index, dev->ifindex);
429 
430 	netdev_for_each_altname(dev, name_node)
431 		netdev_name_node_del(name_node);
432 
433 	/* Unlink dev from the device chain */
434 	list_del_rcu(&dev->dev_list);
435 	netdev_name_node_del(dev->name_node);
436 	hlist_del_rcu(&dev->index_hlist);
437 
438 	dev_base_seq_inc(dev_net(dev));
439 }
440 
441 /*
442  *	Our notifier list
443  */
444 
445 static RAW_NOTIFIER_HEAD(netdev_chain);
446 
447 /*
448  *	Device drivers call our routines to queue packets here. We empty the
449  *	queue in the local softnet handler.
450  */
451 
452 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data) = {
453 	.process_queue_bh_lock = INIT_LOCAL_LOCK(process_queue_bh_lock),
454 };
455 EXPORT_PER_CPU_SYMBOL(softnet_data);
456 
457 /* Page_pool has a lockless array/stack to alloc/recycle pages.
458  * PP consumers must pay attention to run APIs in the appropriate context
459  * (e.g. NAPI context).
460  */
461 static DEFINE_PER_CPU(struct page_pool *, system_page_pool);
462 
463 #ifdef CONFIG_LOCKDEP
464 /*
465  * register_netdevice() inits txq->_xmit_lock and sets lockdep class
466  * according to dev->type
467  */
468 static const unsigned short netdev_lock_type[] = {
469 	 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
470 	 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
471 	 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
472 	 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
473 	 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
474 	 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
475 	 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
476 	 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
477 	 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
478 	 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
479 	 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
480 	 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
481 	 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
482 	 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
483 	 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
484 
485 static const char *const netdev_lock_name[] = {
486 	"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
487 	"_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
488 	"_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
489 	"_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
490 	"_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
491 	"_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
492 	"_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
493 	"_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
494 	"_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
495 	"_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
496 	"_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
497 	"_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
498 	"_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
499 	"_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
500 	"_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
501 
502 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
503 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
504 
505 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
506 {
507 	int i;
508 
509 	for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
510 		if (netdev_lock_type[i] == dev_type)
511 			return i;
512 	/* the last key is used by default */
513 	return ARRAY_SIZE(netdev_lock_type) - 1;
514 }
515 
516 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
517 						 unsigned short dev_type)
518 {
519 	int i;
520 
521 	i = netdev_lock_pos(dev_type);
522 	lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
523 				   netdev_lock_name[i]);
524 }
525 
526 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
527 {
528 	int i;
529 
530 	i = netdev_lock_pos(dev->type);
531 	lockdep_set_class_and_name(&dev->addr_list_lock,
532 				   &netdev_addr_lock_key[i],
533 				   netdev_lock_name[i]);
534 }
535 #else
536 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
537 						 unsigned short dev_type)
538 {
539 }
540 
541 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
542 {
543 }
544 #endif
545 
546 /*******************************************************************************
547  *
548  *		Protocol management and registration routines
549  *
550  *******************************************************************************/
551 
552 
553 /*
554  *	Add a protocol ID to the list. Now that the input handler is
555  *	smarter we can dispense with all the messy stuff that used to be
556  *	here.
557  *
558  *	BEWARE!!! Protocol handlers, mangling input packets,
559  *	MUST BE last in hash buckets and checking protocol handlers
560  *	MUST start from promiscuous ptype_all chain in net_bh.
561  *	It is true now, do not change it.
562  *	Explanation follows: if protocol handler, mangling packet, will
563  *	be the first on list, it is not able to sense, that packet
564  *	is cloned and should be copied-on-write, so that it will
565  *	change it and subsequent readers will get broken packet.
566  *							--ANK (980803)
567  */
568 
569 static inline struct list_head *ptype_head(const struct packet_type *pt)
570 {
571 	if (pt->type == htons(ETH_P_ALL))
572 		return pt->dev ? &pt->dev->ptype_all : &net_hotdata.ptype_all;
573 	else
574 		return pt->dev ? &pt->dev->ptype_specific :
575 				 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
576 }
577 
578 /**
579  *	dev_add_pack - add packet handler
580  *	@pt: packet type declaration
581  *
582  *	Add a protocol handler to the networking stack. The passed &packet_type
583  *	is linked into kernel lists and may not be freed until it has been
584  *	removed from the kernel lists.
585  *
586  *	This call does not sleep therefore it can not
587  *	guarantee all CPU's that are in middle of receiving packets
588  *	will see the new packet type (until the next received packet).
589  */
590 
591 void dev_add_pack(struct packet_type *pt)
592 {
593 	struct list_head *head = ptype_head(pt);
594 
595 	spin_lock(&ptype_lock);
596 	list_add_rcu(&pt->list, head);
597 	spin_unlock(&ptype_lock);
598 }
599 EXPORT_SYMBOL(dev_add_pack);
600 
601 /**
602  *	__dev_remove_pack	 - remove packet handler
603  *	@pt: packet type declaration
604  *
605  *	Remove a protocol handler that was previously added to the kernel
606  *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
607  *	from the kernel lists and can be freed or reused once this function
608  *	returns.
609  *
610  *      The packet type might still be in use by receivers
611  *	and must not be freed until after all the CPU's have gone
612  *	through a quiescent state.
613  */
614 void __dev_remove_pack(struct packet_type *pt)
615 {
616 	struct list_head *head = ptype_head(pt);
617 	struct packet_type *pt1;
618 
619 	spin_lock(&ptype_lock);
620 
621 	list_for_each_entry(pt1, head, list) {
622 		if (pt == pt1) {
623 			list_del_rcu(&pt->list);
624 			goto out;
625 		}
626 	}
627 
628 	pr_warn("dev_remove_pack: %p not found\n", pt);
629 out:
630 	spin_unlock(&ptype_lock);
631 }
632 EXPORT_SYMBOL(__dev_remove_pack);
633 
634 /**
635  *	dev_remove_pack	 - remove packet handler
636  *	@pt: packet type declaration
637  *
638  *	Remove a protocol handler that was previously added to the kernel
639  *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
640  *	from the kernel lists and can be freed or reused once this function
641  *	returns.
642  *
643  *	This call sleeps to guarantee that no CPU is looking at the packet
644  *	type after return.
645  */
646 void dev_remove_pack(struct packet_type *pt)
647 {
648 	__dev_remove_pack(pt);
649 
650 	synchronize_net();
651 }
652 EXPORT_SYMBOL(dev_remove_pack);
653 
654 
655 /*******************************************************************************
656  *
657  *			    Device Interface Subroutines
658  *
659  *******************************************************************************/
660 
661 /**
662  *	dev_get_iflink	- get 'iflink' value of a interface
663  *	@dev: targeted interface
664  *
665  *	Indicates the ifindex the interface is linked to.
666  *	Physical interfaces have the same 'ifindex' and 'iflink' values.
667  */
668 
669 int dev_get_iflink(const struct net_device *dev)
670 {
671 	if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
672 		return dev->netdev_ops->ndo_get_iflink(dev);
673 
674 	return READ_ONCE(dev->ifindex);
675 }
676 EXPORT_SYMBOL(dev_get_iflink);
677 
678 /**
679  *	dev_fill_metadata_dst - Retrieve tunnel egress information.
680  *	@dev: targeted interface
681  *	@skb: The packet.
682  *
683  *	For better visibility of tunnel traffic OVS needs to retrieve
684  *	egress tunnel information for a packet. Following API allows
685  *	user to get this info.
686  */
687 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
688 {
689 	struct ip_tunnel_info *info;
690 
691 	if (!dev->netdev_ops  || !dev->netdev_ops->ndo_fill_metadata_dst)
692 		return -EINVAL;
693 
694 	info = skb_tunnel_info_unclone(skb);
695 	if (!info)
696 		return -ENOMEM;
697 	if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
698 		return -EINVAL;
699 
700 	return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
701 }
702 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
703 
704 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
705 {
706 	int k = stack->num_paths++;
707 
708 	if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
709 		return NULL;
710 
711 	return &stack->path[k];
712 }
713 
714 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
715 			  struct net_device_path_stack *stack)
716 {
717 	const struct net_device *last_dev;
718 	struct net_device_path_ctx ctx = {
719 		.dev	= dev,
720 	};
721 	struct net_device_path *path;
722 	int ret = 0;
723 
724 	memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
725 	stack->num_paths = 0;
726 	while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
727 		last_dev = ctx.dev;
728 		path = dev_fwd_path(stack);
729 		if (!path)
730 			return -1;
731 
732 		memset(path, 0, sizeof(struct net_device_path));
733 		ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
734 		if (ret < 0)
735 			return -1;
736 
737 		if (WARN_ON_ONCE(last_dev == ctx.dev))
738 			return -1;
739 	}
740 
741 	if (!ctx.dev)
742 		return ret;
743 
744 	path = dev_fwd_path(stack);
745 	if (!path)
746 		return -1;
747 	path->type = DEV_PATH_ETHERNET;
748 	path->dev = ctx.dev;
749 
750 	return ret;
751 }
752 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
753 
754 /**
755  *	__dev_get_by_name	- find a device by its name
756  *	@net: the applicable net namespace
757  *	@name: name to find
758  *
759  *	Find an interface by name. Must be called under RTNL semaphore.
760  *	If the name is found a pointer to the device is returned.
761  *	If the name is not found then %NULL is returned. The
762  *	reference counters are not incremented so the caller must be
763  *	careful with locks.
764  */
765 
766 struct net_device *__dev_get_by_name(struct net *net, const char *name)
767 {
768 	struct netdev_name_node *node_name;
769 
770 	node_name = netdev_name_node_lookup(net, name);
771 	return node_name ? node_name->dev : NULL;
772 }
773 EXPORT_SYMBOL(__dev_get_by_name);
774 
775 /**
776  * dev_get_by_name_rcu	- find a device by its name
777  * @net: the applicable net namespace
778  * @name: name to find
779  *
780  * Find an interface by name.
781  * If the name is found a pointer to the device is returned.
782  * If the name is not found then %NULL is returned.
783  * The reference counters are not incremented so the caller must be
784  * careful with locks. The caller must hold RCU lock.
785  */
786 
787 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
788 {
789 	struct netdev_name_node *node_name;
790 
791 	node_name = netdev_name_node_lookup_rcu(net, name);
792 	return node_name ? node_name->dev : NULL;
793 }
794 EXPORT_SYMBOL(dev_get_by_name_rcu);
795 
796 /* Deprecated for new users, call netdev_get_by_name() instead */
797 struct net_device *dev_get_by_name(struct net *net, const char *name)
798 {
799 	struct net_device *dev;
800 
801 	rcu_read_lock();
802 	dev = dev_get_by_name_rcu(net, name);
803 	dev_hold(dev);
804 	rcu_read_unlock();
805 	return dev;
806 }
807 EXPORT_SYMBOL(dev_get_by_name);
808 
809 /**
810  *	netdev_get_by_name() - find a device by its name
811  *	@net: the applicable net namespace
812  *	@name: name to find
813  *	@tracker: tracking object for the acquired reference
814  *	@gfp: allocation flags for the tracker
815  *
816  *	Find an interface by name. This can be called from any
817  *	context and does its own locking. The returned handle has
818  *	the usage count incremented and the caller must use netdev_put() to
819  *	release it when it is no longer needed. %NULL is returned if no
820  *	matching device is found.
821  */
822 struct net_device *netdev_get_by_name(struct net *net, const char *name,
823 				      netdevice_tracker *tracker, gfp_t gfp)
824 {
825 	struct net_device *dev;
826 
827 	dev = dev_get_by_name(net, name);
828 	if (dev)
829 		netdev_tracker_alloc(dev, tracker, gfp);
830 	return dev;
831 }
832 EXPORT_SYMBOL(netdev_get_by_name);
833 
834 /**
835  *	__dev_get_by_index - find a device by its ifindex
836  *	@net: the applicable net namespace
837  *	@ifindex: index of device
838  *
839  *	Search for an interface by index. Returns %NULL if the device
840  *	is not found or a pointer to the device. The device has not
841  *	had its reference counter increased so the caller must be careful
842  *	about locking. The caller must hold the RTNL semaphore.
843  */
844 
845 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
846 {
847 	struct net_device *dev;
848 	struct hlist_head *head = dev_index_hash(net, ifindex);
849 
850 	hlist_for_each_entry(dev, head, index_hlist)
851 		if (dev->ifindex == ifindex)
852 			return dev;
853 
854 	return NULL;
855 }
856 EXPORT_SYMBOL(__dev_get_by_index);
857 
858 /**
859  *	dev_get_by_index_rcu - find a device by its ifindex
860  *	@net: the applicable net namespace
861  *	@ifindex: index of device
862  *
863  *	Search for an interface by index. Returns %NULL if the device
864  *	is not found or a pointer to the device. The device has not
865  *	had its reference counter increased so the caller must be careful
866  *	about locking. The caller must hold RCU lock.
867  */
868 
869 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
870 {
871 	struct net_device *dev;
872 	struct hlist_head *head = dev_index_hash(net, ifindex);
873 
874 	hlist_for_each_entry_rcu(dev, head, index_hlist)
875 		if (dev->ifindex == ifindex)
876 			return dev;
877 
878 	return NULL;
879 }
880 EXPORT_SYMBOL(dev_get_by_index_rcu);
881 
882 /* Deprecated for new users, call netdev_get_by_index() instead */
883 struct net_device *dev_get_by_index(struct net *net, int ifindex)
884 {
885 	struct net_device *dev;
886 
887 	rcu_read_lock();
888 	dev = dev_get_by_index_rcu(net, ifindex);
889 	dev_hold(dev);
890 	rcu_read_unlock();
891 	return dev;
892 }
893 EXPORT_SYMBOL(dev_get_by_index);
894 
895 /**
896  *	netdev_get_by_index() - find a device by its ifindex
897  *	@net: the applicable net namespace
898  *	@ifindex: index of device
899  *	@tracker: tracking object for the acquired reference
900  *	@gfp: allocation flags for the tracker
901  *
902  *	Search for an interface by index. Returns NULL if the device
903  *	is not found or a pointer to the device. The device returned has
904  *	had a reference added and the pointer is safe until the user calls
905  *	netdev_put() to indicate they have finished with it.
906  */
907 struct net_device *netdev_get_by_index(struct net *net, int ifindex,
908 				       netdevice_tracker *tracker, gfp_t gfp)
909 {
910 	struct net_device *dev;
911 
912 	dev = dev_get_by_index(net, ifindex);
913 	if (dev)
914 		netdev_tracker_alloc(dev, tracker, gfp);
915 	return dev;
916 }
917 EXPORT_SYMBOL(netdev_get_by_index);
918 
919 /**
920  *	dev_get_by_napi_id - find a device by napi_id
921  *	@napi_id: ID of the NAPI struct
922  *
923  *	Search for an interface by NAPI ID. Returns %NULL if the device
924  *	is not found or a pointer to the device. The device has not had
925  *	its reference counter increased so the caller must be careful
926  *	about locking. The caller must hold RCU lock.
927  */
928 
929 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
930 {
931 	struct napi_struct *napi;
932 
933 	WARN_ON_ONCE(!rcu_read_lock_held());
934 
935 	if (napi_id < MIN_NAPI_ID)
936 		return NULL;
937 
938 	napi = napi_by_id(napi_id);
939 
940 	return napi ? napi->dev : NULL;
941 }
942 EXPORT_SYMBOL(dev_get_by_napi_id);
943 
944 static DEFINE_SEQLOCK(netdev_rename_lock);
945 
946 void netdev_copy_name(struct net_device *dev, char *name)
947 {
948 	unsigned int seq;
949 
950 	do {
951 		seq = read_seqbegin(&netdev_rename_lock);
952 		strscpy(name, dev->name, IFNAMSIZ);
953 	} while (read_seqretry(&netdev_rename_lock, seq));
954 }
955 
956 /**
957  *	netdev_get_name - get a netdevice name, knowing its ifindex.
958  *	@net: network namespace
959  *	@name: a pointer to the buffer where the name will be stored.
960  *	@ifindex: the ifindex of the interface to get the name from.
961  */
962 int netdev_get_name(struct net *net, char *name, int ifindex)
963 {
964 	struct net_device *dev;
965 	int ret;
966 
967 	rcu_read_lock();
968 
969 	dev = dev_get_by_index_rcu(net, ifindex);
970 	if (!dev) {
971 		ret = -ENODEV;
972 		goto out;
973 	}
974 
975 	netdev_copy_name(dev, name);
976 
977 	ret = 0;
978 out:
979 	rcu_read_unlock();
980 	return ret;
981 }
982 
983 /**
984  *	dev_getbyhwaddr_rcu - find a device by its hardware address
985  *	@net: the applicable net namespace
986  *	@type: media type of device
987  *	@ha: hardware address
988  *
989  *	Search for an interface by MAC address. Returns NULL if the device
990  *	is not found or a pointer to the device.
991  *	The caller must hold RCU or RTNL.
992  *	The returned device has not had its ref count increased
993  *	and the caller must therefore be careful about locking
994  *
995  */
996 
997 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
998 				       const char *ha)
999 {
1000 	struct net_device *dev;
1001 
1002 	for_each_netdev_rcu(net, dev)
1003 		if (dev->type == type &&
1004 		    !memcmp(dev->dev_addr, ha, dev->addr_len))
1005 			return dev;
1006 
1007 	return NULL;
1008 }
1009 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
1010 
1011 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1012 {
1013 	struct net_device *dev, *ret = NULL;
1014 
1015 	rcu_read_lock();
1016 	for_each_netdev_rcu(net, dev)
1017 		if (dev->type == type) {
1018 			dev_hold(dev);
1019 			ret = dev;
1020 			break;
1021 		}
1022 	rcu_read_unlock();
1023 	return ret;
1024 }
1025 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1026 
1027 /**
1028  *	__dev_get_by_flags - find any device with given flags
1029  *	@net: the applicable net namespace
1030  *	@if_flags: IFF_* values
1031  *	@mask: bitmask of bits in if_flags to check
1032  *
1033  *	Search for any interface with the given flags. Returns NULL if a device
1034  *	is not found or a pointer to the device. Must be called inside
1035  *	rtnl_lock(), and result refcount is unchanged.
1036  */
1037 
1038 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1039 				      unsigned short mask)
1040 {
1041 	struct net_device *dev, *ret;
1042 
1043 	ASSERT_RTNL();
1044 
1045 	ret = NULL;
1046 	for_each_netdev(net, dev) {
1047 		if (((dev->flags ^ if_flags) & mask) == 0) {
1048 			ret = dev;
1049 			break;
1050 		}
1051 	}
1052 	return ret;
1053 }
1054 EXPORT_SYMBOL(__dev_get_by_flags);
1055 
1056 /**
1057  *	dev_valid_name - check if name is okay for network device
1058  *	@name: name string
1059  *
1060  *	Network device names need to be valid file names to
1061  *	allow sysfs to work.  We also disallow any kind of
1062  *	whitespace.
1063  */
1064 bool dev_valid_name(const char *name)
1065 {
1066 	if (*name == '\0')
1067 		return false;
1068 	if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1069 		return false;
1070 	if (!strcmp(name, ".") || !strcmp(name, ".."))
1071 		return false;
1072 
1073 	while (*name) {
1074 		if (*name == '/' || *name == ':' || isspace(*name))
1075 			return false;
1076 		name++;
1077 	}
1078 	return true;
1079 }
1080 EXPORT_SYMBOL(dev_valid_name);
1081 
1082 /**
1083  *	__dev_alloc_name - allocate a name for a device
1084  *	@net: network namespace to allocate the device name in
1085  *	@name: name format string
1086  *	@res: result name string
1087  *
1088  *	Passed a format string - eg "lt%d" it will try and find a suitable
1089  *	id. It scans list of devices to build up a free map, then chooses
1090  *	the first empty slot. The caller must hold the dev_base or rtnl lock
1091  *	while allocating the name and adding the device in order to avoid
1092  *	duplicates.
1093  *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1094  *	Returns the number of the unit assigned or a negative errno code.
1095  */
1096 
1097 static int __dev_alloc_name(struct net *net, const char *name, char *res)
1098 {
1099 	int i = 0;
1100 	const char *p;
1101 	const int max_netdevices = 8*PAGE_SIZE;
1102 	unsigned long *inuse;
1103 	struct net_device *d;
1104 	char buf[IFNAMSIZ];
1105 
1106 	/* Verify the string as this thing may have come from the user.
1107 	 * There must be one "%d" and no other "%" characters.
1108 	 */
1109 	p = strchr(name, '%');
1110 	if (!p || p[1] != 'd' || strchr(p + 2, '%'))
1111 		return -EINVAL;
1112 
1113 	/* Use one page as a bit array of possible slots */
1114 	inuse = bitmap_zalloc(max_netdevices, GFP_ATOMIC);
1115 	if (!inuse)
1116 		return -ENOMEM;
1117 
1118 	for_each_netdev(net, d) {
1119 		struct netdev_name_node *name_node;
1120 
1121 		netdev_for_each_altname(d, name_node) {
1122 			if (!sscanf(name_node->name, name, &i))
1123 				continue;
1124 			if (i < 0 || i >= max_netdevices)
1125 				continue;
1126 
1127 			/* avoid cases where sscanf is not exact inverse of printf */
1128 			snprintf(buf, IFNAMSIZ, name, i);
1129 			if (!strncmp(buf, name_node->name, IFNAMSIZ))
1130 				__set_bit(i, inuse);
1131 		}
1132 		if (!sscanf(d->name, name, &i))
1133 			continue;
1134 		if (i < 0 || i >= max_netdevices)
1135 			continue;
1136 
1137 		/* avoid cases where sscanf is not exact inverse of printf */
1138 		snprintf(buf, IFNAMSIZ, name, i);
1139 		if (!strncmp(buf, d->name, IFNAMSIZ))
1140 			__set_bit(i, inuse);
1141 	}
1142 
1143 	i = find_first_zero_bit(inuse, max_netdevices);
1144 	bitmap_free(inuse);
1145 	if (i == max_netdevices)
1146 		return -ENFILE;
1147 
1148 	/* 'res' and 'name' could overlap, use 'buf' as an intermediate buffer */
1149 	strscpy(buf, name, IFNAMSIZ);
1150 	snprintf(res, IFNAMSIZ, buf, i);
1151 	return i;
1152 }
1153 
1154 /* Returns negative errno or allocated unit id (see __dev_alloc_name()) */
1155 static int dev_prep_valid_name(struct net *net, struct net_device *dev,
1156 			       const char *want_name, char *out_name,
1157 			       int dup_errno)
1158 {
1159 	if (!dev_valid_name(want_name))
1160 		return -EINVAL;
1161 
1162 	if (strchr(want_name, '%'))
1163 		return __dev_alloc_name(net, want_name, out_name);
1164 
1165 	if (netdev_name_in_use(net, want_name))
1166 		return -dup_errno;
1167 	if (out_name != want_name)
1168 		strscpy(out_name, want_name, IFNAMSIZ);
1169 	return 0;
1170 }
1171 
1172 /**
1173  *	dev_alloc_name - allocate a name for a device
1174  *	@dev: device
1175  *	@name: name format string
1176  *
1177  *	Passed a format string - eg "lt%d" it will try and find a suitable
1178  *	id. It scans list of devices to build up a free map, then chooses
1179  *	the first empty slot. The caller must hold the dev_base or rtnl lock
1180  *	while allocating the name and adding the device in order to avoid
1181  *	duplicates.
1182  *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1183  *	Returns the number of the unit assigned or a negative errno code.
1184  */
1185 
1186 int dev_alloc_name(struct net_device *dev, const char *name)
1187 {
1188 	return dev_prep_valid_name(dev_net(dev), dev, name, dev->name, ENFILE);
1189 }
1190 EXPORT_SYMBOL(dev_alloc_name);
1191 
1192 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1193 			      const char *name)
1194 {
1195 	int ret;
1196 
1197 	ret = dev_prep_valid_name(net, dev, name, dev->name, EEXIST);
1198 	return ret < 0 ? ret : 0;
1199 }
1200 
1201 /**
1202  *	dev_change_name - change name of a device
1203  *	@dev: device
1204  *	@newname: name (or format string) must be at least IFNAMSIZ
1205  *
1206  *	Change name of a device, can pass format strings "eth%d".
1207  *	for wildcarding.
1208  */
1209 int dev_change_name(struct net_device *dev, const char *newname)
1210 {
1211 	unsigned char old_assign_type;
1212 	char oldname[IFNAMSIZ];
1213 	int err = 0;
1214 	int ret;
1215 	struct net *net;
1216 
1217 	ASSERT_RTNL();
1218 	BUG_ON(!dev_net(dev));
1219 
1220 	net = dev_net(dev);
1221 
1222 	down_write(&devnet_rename_sem);
1223 
1224 	if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1225 		up_write(&devnet_rename_sem);
1226 		return 0;
1227 	}
1228 
1229 	memcpy(oldname, dev->name, IFNAMSIZ);
1230 
1231 	write_seqlock_bh(&netdev_rename_lock);
1232 	err = dev_get_valid_name(net, dev, newname);
1233 	write_sequnlock_bh(&netdev_rename_lock);
1234 
1235 	if (err < 0) {
1236 		up_write(&devnet_rename_sem);
1237 		return err;
1238 	}
1239 
1240 	if (oldname[0] && !strchr(oldname, '%'))
1241 		netdev_info(dev, "renamed from %s%s\n", oldname,
1242 			    dev->flags & IFF_UP ? " (while UP)" : "");
1243 
1244 	old_assign_type = dev->name_assign_type;
1245 	WRITE_ONCE(dev->name_assign_type, NET_NAME_RENAMED);
1246 
1247 rollback:
1248 	ret = device_rename(&dev->dev, dev->name);
1249 	if (ret) {
1250 		memcpy(dev->name, oldname, IFNAMSIZ);
1251 		WRITE_ONCE(dev->name_assign_type, old_assign_type);
1252 		up_write(&devnet_rename_sem);
1253 		return ret;
1254 	}
1255 
1256 	up_write(&devnet_rename_sem);
1257 
1258 	netdev_adjacent_rename_links(dev, oldname);
1259 
1260 	netdev_name_node_del(dev->name_node);
1261 
1262 	synchronize_net();
1263 
1264 	netdev_name_node_add(net, dev->name_node);
1265 
1266 	ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1267 	ret = notifier_to_errno(ret);
1268 
1269 	if (ret) {
1270 		/* err >= 0 after dev_alloc_name() or stores the first errno */
1271 		if (err >= 0) {
1272 			err = ret;
1273 			down_write(&devnet_rename_sem);
1274 			write_seqlock_bh(&netdev_rename_lock);
1275 			memcpy(dev->name, oldname, IFNAMSIZ);
1276 			write_sequnlock_bh(&netdev_rename_lock);
1277 			memcpy(oldname, newname, IFNAMSIZ);
1278 			WRITE_ONCE(dev->name_assign_type, old_assign_type);
1279 			old_assign_type = NET_NAME_RENAMED;
1280 			goto rollback;
1281 		} else {
1282 			netdev_err(dev, "name change rollback failed: %d\n",
1283 				   ret);
1284 		}
1285 	}
1286 
1287 	return err;
1288 }
1289 
1290 /**
1291  *	dev_set_alias - change ifalias of a device
1292  *	@dev: device
1293  *	@alias: name up to IFALIASZ
1294  *	@len: limit of bytes to copy from info
1295  *
1296  *	Set ifalias for a device,
1297  */
1298 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1299 {
1300 	struct dev_ifalias *new_alias = NULL;
1301 
1302 	if (len >= IFALIASZ)
1303 		return -EINVAL;
1304 
1305 	if (len) {
1306 		new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1307 		if (!new_alias)
1308 			return -ENOMEM;
1309 
1310 		memcpy(new_alias->ifalias, alias, len);
1311 		new_alias->ifalias[len] = 0;
1312 	}
1313 
1314 	mutex_lock(&ifalias_mutex);
1315 	new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1316 					mutex_is_locked(&ifalias_mutex));
1317 	mutex_unlock(&ifalias_mutex);
1318 
1319 	if (new_alias)
1320 		kfree_rcu(new_alias, rcuhead);
1321 
1322 	return len;
1323 }
1324 EXPORT_SYMBOL(dev_set_alias);
1325 
1326 /**
1327  *	dev_get_alias - get ifalias of a device
1328  *	@dev: device
1329  *	@name: buffer to store name of ifalias
1330  *	@len: size of buffer
1331  *
1332  *	get ifalias for a device.  Caller must make sure dev cannot go
1333  *	away,  e.g. rcu read lock or own a reference count to device.
1334  */
1335 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1336 {
1337 	const struct dev_ifalias *alias;
1338 	int ret = 0;
1339 
1340 	rcu_read_lock();
1341 	alias = rcu_dereference(dev->ifalias);
1342 	if (alias)
1343 		ret = snprintf(name, len, "%s", alias->ifalias);
1344 	rcu_read_unlock();
1345 
1346 	return ret;
1347 }
1348 
1349 /**
1350  *	netdev_features_change - device changes features
1351  *	@dev: device to cause notification
1352  *
1353  *	Called to indicate a device has changed features.
1354  */
1355 void netdev_features_change(struct net_device *dev)
1356 {
1357 	call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1358 }
1359 EXPORT_SYMBOL(netdev_features_change);
1360 
1361 /**
1362  *	netdev_state_change - device changes state
1363  *	@dev: device to cause notification
1364  *
1365  *	Called to indicate a device has changed state. This function calls
1366  *	the notifier chains for netdev_chain and sends a NEWLINK message
1367  *	to the routing socket.
1368  */
1369 void netdev_state_change(struct net_device *dev)
1370 {
1371 	if (dev->flags & IFF_UP) {
1372 		struct netdev_notifier_change_info change_info = {
1373 			.info.dev = dev,
1374 		};
1375 
1376 		call_netdevice_notifiers_info(NETDEV_CHANGE,
1377 					      &change_info.info);
1378 		rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL);
1379 	}
1380 }
1381 EXPORT_SYMBOL(netdev_state_change);
1382 
1383 /**
1384  * __netdev_notify_peers - notify network peers about existence of @dev,
1385  * to be called when rtnl lock is already held.
1386  * @dev: network device
1387  *
1388  * Generate traffic such that interested network peers are aware of
1389  * @dev, such as by generating a gratuitous ARP. This may be used when
1390  * a device wants to inform the rest of the network about some sort of
1391  * reconfiguration such as a failover event or virtual machine
1392  * migration.
1393  */
1394 void __netdev_notify_peers(struct net_device *dev)
1395 {
1396 	ASSERT_RTNL();
1397 	call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1398 	call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1399 }
1400 EXPORT_SYMBOL(__netdev_notify_peers);
1401 
1402 /**
1403  * netdev_notify_peers - notify network peers about existence of @dev
1404  * @dev: network device
1405  *
1406  * Generate traffic such that interested network peers are aware of
1407  * @dev, such as by generating a gratuitous ARP. This may be used when
1408  * a device wants to inform the rest of the network about some sort of
1409  * reconfiguration such as a failover event or virtual machine
1410  * migration.
1411  */
1412 void netdev_notify_peers(struct net_device *dev)
1413 {
1414 	rtnl_lock();
1415 	__netdev_notify_peers(dev);
1416 	rtnl_unlock();
1417 }
1418 EXPORT_SYMBOL(netdev_notify_peers);
1419 
1420 static int napi_threaded_poll(void *data);
1421 
1422 static int napi_kthread_create(struct napi_struct *n)
1423 {
1424 	int err = 0;
1425 
1426 	/* Create and wake up the kthread once to put it in
1427 	 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1428 	 * warning and work with loadavg.
1429 	 */
1430 	n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1431 				n->dev->name, n->napi_id);
1432 	if (IS_ERR(n->thread)) {
1433 		err = PTR_ERR(n->thread);
1434 		pr_err("kthread_run failed with err %d\n", err);
1435 		n->thread = NULL;
1436 	}
1437 
1438 	return err;
1439 }
1440 
1441 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1442 {
1443 	const struct net_device_ops *ops = dev->netdev_ops;
1444 	int ret;
1445 
1446 	ASSERT_RTNL();
1447 	dev_addr_check(dev);
1448 
1449 	if (!netif_device_present(dev)) {
1450 		/* may be detached because parent is runtime-suspended */
1451 		if (dev->dev.parent)
1452 			pm_runtime_resume(dev->dev.parent);
1453 		if (!netif_device_present(dev))
1454 			return -ENODEV;
1455 	}
1456 
1457 	/* Block netpoll from trying to do any rx path servicing.
1458 	 * If we don't do this there is a chance ndo_poll_controller
1459 	 * or ndo_poll may be running while we open the device
1460 	 */
1461 	netpoll_poll_disable(dev);
1462 
1463 	ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1464 	ret = notifier_to_errno(ret);
1465 	if (ret)
1466 		return ret;
1467 
1468 	set_bit(__LINK_STATE_START, &dev->state);
1469 
1470 	if (ops->ndo_validate_addr)
1471 		ret = ops->ndo_validate_addr(dev);
1472 
1473 	if (!ret && ops->ndo_open)
1474 		ret = ops->ndo_open(dev);
1475 
1476 	netpoll_poll_enable(dev);
1477 
1478 	if (ret)
1479 		clear_bit(__LINK_STATE_START, &dev->state);
1480 	else {
1481 		dev->flags |= IFF_UP;
1482 		dev_set_rx_mode(dev);
1483 		dev_activate(dev);
1484 		add_device_randomness(dev->dev_addr, dev->addr_len);
1485 	}
1486 
1487 	return ret;
1488 }
1489 
1490 /**
1491  *	dev_open	- prepare an interface for use.
1492  *	@dev: device to open
1493  *	@extack: netlink extended ack
1494  *
1495  *	Takes a device from down to up state. The device's private open
1496  *	function is invoked and then the multicast lists are loaded. Finally
1497  *	the device is moved into the up state and a %NETDEV_UP message is
1498  *	sent to the netdev notifier chain.
1499  *
1500  *	Calling this function on an active interface is a nop. On a failure
1501  *	a negative errno code is returned.
1502  */
1503 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1504 {
1505 	int ret;
1506 
1507 	if (dev->flags & IFF_UP)
1508 		return 0;
1509 
1510 	ret = __dev_open(dev, extack);
1511 	if (ret < 0)
1512 		return ret;
1513 
1514 	rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1515 	call_netdevice_notifiers(NETDEV_UP, dev);
1516 
1517 	return ret;
1518 }
1519 EXPORT_SYMBOL(dev_open);
1520 
1521 static void __dev_close_many(struct list_head *head)
1522 {
1523 	struct net_device *dev;
1524 
1525 	ASSERT_RTNL();
1526 	might_sleep();
1527 
1528 	list_for_each_entry(dev, head, close_list) {
1529 		/* Temporarily disable netpoll until the interface is down */
1530 		netpoll_poll_disable(dev);
1531 
1532 		call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1533 
1534 		clear_bit(__LINK_STATE_START, &dev->state);
1535 
1536 		/* Synchronize to scheduled poll. We cannot touch poll list, it
1537 		 * can be even on different cpu. So just clear netif_running().
1538 		 *
1539 		 * dev->stop() will invoke napi_disable() on all of it's
1540 		 * napi_struct instances on this device.
1541 		 */
1542 		smp_mb__after_atomic(); /* Commit netif_running(). */
1543 	}
1544 
1545 	dev_deactivate_many(head);
1546 
1547 	list_for_each_entry(dev, head, close_list) {
1548 		const struct net_device_ops *ops = dev->netdev_ops;
1549 
1550 		/*
1551 		 *	Call the device specific close. This cannot fail.
1552 		 *	Only if device is UP
1553 		 *
1554 		 *	We allow it to be called even after a DETACH hot-plug
1555 		 *	event.
1556 		 */
1557 		if (ops->ndo_stop)
1558 			ops->ndo_stop(dev);
1559 
1560 		dev->flags &= ~IFF_UP;
1561 		netpoll_poll_enable(dev);
1562 	}
1563 }
1564 
1565 static void __dev_close(struct net_device *dev)
1566 {
1567 	LIST_HEAD(single);
1568 
1569 	list_add(&dev->close_list, &single);
1570 	__dev_close_many(&single);
1571 	list_del(&single);
1572 }
1573 
1574 void dev_close_many(struct list_head *head, bool unlink)
1575 {
1576 	struct net_device *dev, *tmp;
1577 
1578 	/* Remove the devices that don't need to be closed */
1579 	list_for_each_entry_safe(dev, tmp, head, close_list)
1580 		if (!(dev->flags & IFF_UP))
1581 			list_del_init(&dev->close_list);
1582 
1583 	__dev_close_many(head);
1584 
1585 	list_for_each_entry_safe(dev, tmp, head, close_list) {
1586 		rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1587 		call_netdevice_notifiers(NETDEV_DOWN, dev);
1588 		if (unlink)
1589 			list_del_init(&dev->close_list);
1590 	}
1591 }
1592 EXPORT_SYMBOL(dev_close_many);
1593 
1594 /**
1595  *	dev_close - shutdown an interface.
1596  *	@dev: device to shutdown
1597  *
1598  *	This function moves an active device into down state. A
1599  *	%NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1600  *	is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1601  *	chain.
1602  */
1603 void dev_close(struct net_device *dev)
1604 {
1605 	if (dev->flags & IFF_UP) {
1606 		LIST_HEAD(single);
1607 
1608 		list_add(&dev->close_list, &single);
1609 		dev_close_many(&single, true);
1610 		list_del(&single);
1611 	}
1612 }
1613 EXPORT_SYMBOL(dev_close);
1614 
1615 
1616 /**
1617  *	dev_disable_lro - disable Large Receive Offload on a device
1618  *	@dev: device
1619  *
1620  *	Disable Large Receive Offload (LRO) on a net device.  Must be
1621  *	called under RTNL.  This is needed if received packets may be
1622  *	forwarded to another interface.
1623  */
1624 void dev_disable_lro(struct net_device *dev)
1625 {
1626 	struct net_device *lower_dev;
1627 	struct list_head *iter;
1628 
1629 	dev->wanted_features &= ~NETIF_F_LRO;
1630 	netdev_update_features(dev);
1631 
1632 	if (unlikely(dev->features & NETIF_F_LRO))
1633 		netdev_WARN(dev, "failed to disable LRO!\n");
1634 
1635 	netdev_for_each_lower_dev(dev, lower_dev, iter)
1636 		dev_disable_lro(lower_dev);
1637 }
1638 EXPORT_SYMBOL(dev_disable_lro);
1639 
1640 /**
1641  *	dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1642  *	@dev: device
1643  *
1644  *	Disable HW Generic Receive Offload (GRO_HW) on a net device.  Must be
1645  *	called under RTNL.  This is needed if Generic XDP is installed on
1646  *	the device.
1647  */
1648 static void dev_disable_gro_hw(struct net_device *dev)
1649 {
1650 	dev->wanted_features &= ~NETIF_F_GRO_HW;
1651 	netdev_update_features(dev);
1652 
1653 	if (unlikely(dev->features & NETIF_F_GRO_HW))
1654 		netdev_WARN(dev, "failed to disable GRO_HW!\n");
1655 }
1656 
1657 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1658 {
1659 #define N(val) 						\
1660 	case NETDEV_##val:				\
1661 		return "NETDEV_" __stringify(val);
1662 	switch (cmd) {
1663 	N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1664 	N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1665 	N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1666 	N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN)
1667 	N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA)
1668 	N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE)
1669 	N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1670 	N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1671 	N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1672 	N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE)
1673 	N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA)
1674 	N(XDP_FEAT_CHANGE)
1675 	}
1676 #undef N
1677 	return "UNKNOWN_NETDEV_EVENT";
1678 }
1679 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1680 
1681 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1682 				   struct net_device *dev)
1683 {
1684 	struct netdev_notifier_info info = {
1685 		.dev = dev,
1686 	};
1687 
1688 	return nb->notifier_call(nb, val, &info);
1689 }
1690 
1691 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1692 					     struct net_device *dev)
1693 {
1694 	int err;
1695 
1696 	err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1697 	err = notifier_to_errno(err);
1698 	if (err)
1699 		return err;
1700 
1701 	if (!(dev->flags & IFF_UP))
1702 		return 0;
1703 
1704 	call_netdevice_notifier(nb, NETDEV_UP, dev);
1705 	return 0;
1706 }
1707 
1708 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1709 						struct net_device *dev)
1710 {
1711 	if (dev->flags & IFF_UP) {
1712 		call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1713 					dev);
1714 		call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1715 	}
1716 	call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1717 }
1718 
1719 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1720 						 struct net *net)
1721 {
1722 	struct net_device *dev;
1723 	int err;
1724 
1725 	for_each_netdev(net, dev) {
1726 		err = call_netdevice_register_notifiers(nb, dev);
1727 		if (err)
1728 			goto rollback;
1729 	}
1730 	return 0;
1731 
1732 rollback:
1733 	for_each_netdev_continue_reverse(net, dev)
1734 		call_netdevice_unregister_notifiers(nb, dev);
1735 	return err;
1736 }
1737 
1738 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1739 						    struct net *net)
1740 {
1741 	struct net_device *dev;
1742 
1743 	for_each_netdev(net, dev)
1744 		call_netdevice_unregister_notifiers(nb, dev);
1745 }
1746 
1747 static int dev_boot_phase = 1;
1748 
1749 /**
1750  * register_netdevice_notifier - register a network notifier block
1751  * @nb: notifier
1752  *
1753  * Register a notifier to be called when network device events occur.
1754  * The notifier passed is linked into the kernel structures and must
1755  * not be reused until it has been unregistered. A negative errno code
1756  * is returned on a failure.
1757  *
1758  * When registered all registration and up events are replayed
1759  * to the new notifier to allow device to have a race free
1760  * view of the network device list.
1761  */
1762 
1763 int register_netdevice_notifier(struct notifier_block *nb)
1764 {
1765 	struct net *net;
1766 	int err;
1767 
1768 	/* Close race with setup_net() and cleanup_net() */
1769 	down_write(&pernet_ops_rwsem);
1770 	rtnl_lock();
1771 	err = raw_notifier_chain_register(&netdev_chain, nb);
1772 	if (err)
1773 		goto unlock;
1774 	if (dev_boot_phase)
1775 		goto unlock;
1776 	for_each_net(net) {
1777 		err = call_netdevice_register_net_notifiers(nb, net);
1778 		if (err)
1779 			goto rollback;
1780 	}
1781 
1782 unlock:
1783 	rtnl_unlock();
1784 	up_write(&pernet_ops_rwsem);
1785 	return err;
1786 
1787 rollback:
1788 	for_each_net_continue_reverse(net)
1789 		call_netdevice_unregister_net_notifiers(nb, net);
1790 
1791 	raw_notifier_chain_unregister(&netdev_chain, nb);
1792 	goto unlock;
1793 }
1794 EXPORT_SYMBOL(register_netdevice_notifier);
1795 
1796 /**
1797  * unregister_netdevice_notifier - unregister a network notifier block
1798  * @nb: notifier
1799  *
1800  * Unregister a notifier previously registered by
1801  * register_netdevice_notifier(). The notifier is unlinked into the
1802  * kernel structures and may then be reused. A negative errno code
1803  * is returned on a failure.
1804  *
1805  * After unregistering unregister and down device events are synthesized
1806  * for all devices on the device list to the removed notifier to remove
1807  * the need for special case cleanup code.
1808  */
1809 
1810 int unregister_netdevice_notifier(struct notifier_block *nb)
1811 {
1812 	struct net *net;
1813 	int err;
1814 
1815 	/* Close race with setup_net() and cleanup_net() */
1816 	down_write(&pernet_ops_rwsem);
1817 	rtnl_lock();
1818 	err = raw_notifier_chain_unregister(&netdev_chain, nb);
1819 	if (err)
1820 		goto unlock;
1821 
1822 	for_each_net(net)
1823 		call_netdevice_unregister_net_notifiers(nb, net);
1824 
1825 unlock:
1826 	rtnl_unlock();
1827 	up_write(&pernet_ops_rwsem);
1828 	return err;
1829 }
1830 EXPORT_SYMBOL(unregister_netdevice_notifier);
1831 
1832 static int __register_netdevice_notifier_net(struct net *net,
1833 					     struct notifier_block *nb,
1834 					     bool ignore_call_fail)
1835 {
1836 	int err;
1837 
1838 	err = raw_notifier_chain_register(&net->netdev_chain, nb);
1839 	if (err)
1840 		return err;
1841 	if (dev_boot_phase)
1842 		return 0;
1843 
1844 	err = call_netdevice_register_net_notifiers(nb, net);
1845 	if (err && !ignore_call_fail)
1846 		goto chain_unregister;
1847 
1848 	return 0;
1849 
1850 chain_unregister:
1851 	raw_notifier_chain_unregister(&net->netdev_chain, nb);
1852 	return err;
1853 }
1854 
1855 static int __unregister_netdevice_notifier_net(struct net *net,
1856 					       struct notifier_block *nb)
1857 {
1858 	int err;
1859 
1860 	err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1861 	if (err)
1862 		return err;
1863 
1864 	call_netdevice_unregister_net_notifiers(nb, net);
1865 	return 0;
1866 }
1867 
1868 /**
1869  * register_netdevice_notifier_net - register a per-netns network notifier block
1870  * @net: network namespace
1871  * @nb: notifier
1872  *
1873  * Register a notifier to be called when network device events occur.
1874  * The notifier passed is linked into the kernel structures and must
1875  * not be reused until it has been unregistered. A negative errno code
1876  * is returned on a failure.
1877  *
1878  * When registered all registration and up events are replayed
1879  * to the new notifier to allow device to have a race free
1880  * view of the network device list.
1881  */
1882 
1883 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1884 {
1885 	int err;
1886 
1887 	rtnl_lock();
1888 	err = __register_netdevice_notifier_net(net, nb, false);
1889 	rtnl_unlock();
1890 	return err;
1891 }
1892 EXPORT_SYMBOL(register_netdevice_notifier_net);
1893 
1894 /**
1895  * unregister_netdevice_notifier_net - unregister a per-netns
1896  *                                     network notifier block
1897  * @net: network namespace
1898  * @nb: notifier
1899  *
1900  * Unregister a notifier previously registered by
1901  * register_netdevice_notifier_net(). The notifier is unlinked from the
1902  * kernel structures and may then be reused. A negative errno code
1903  * is returned on a failure.
1904  *
1905  * After unregistering unregister and down device events are synthesized
1906  * for all devices on the device list to the removed notifier to remove
1907  * the need for special case cleanup code.
1908  */
1909 
1910 int unregister_netdevice_notifier_net(struct net *net,
1911 				      struct notifier_block *nb)
1912 {
1913 	int err;
1914 
1915 	rtnl_lock();
1916 	err = __unregister_netdevice_notifier_net(net, nb);
1917 	rtnl_unlock();
1918 	return err;
1919 }
1920 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1921 
1922 static void __move_netdevice_notifier_net(struct net *src_net,
1923 					  struct net *dst_net,
1924 					  struct notifier_block *nb)
1925 {
1926 	__unregister_netdevice_notifier_net(src_net, nb);
1927 	__register_netdevice_notifier_net(dst_net, nb, true);
1928 }
1929 
1930 int register_netdevice_notifier_dev_net(struct net_device *dev,
1931 					struct notifier_block *nb,
1932 					struct netdev_net_notifier *nn)
1933 {
1934 	int err;
1935 
1936 	rtnl_lock();
1937 	err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1938 	if (!err) {
1939 		nn->nb = nb;
1940 		list_add(&nn->list, &dev->net_notifier_list);
1941 	}
1942 	rtnl_unlock();
1943 	return err;
1944 }
1945 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1946 
1947 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1948 					  struct notifier_block *nb,
1949 					  struct netdev_net_notifier *nn)
1950 {
1951 	int err;
1952 
1953 	rtnl_lock();
1954 	list_del(&nn->list);
1955 	err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1956 	rtnl_unlock();
1957 	return err;
1958 }
1959 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1960 
1961 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1962 					     struct net *net)
1963 {
1964 	struct netdev_net_notifier *nn;
1965 
1966 	list_for_each_entry(nn, &dev->net_notifier_list, list)
1967 		__move_netdevice_notifier_net(dev_net(dev), net, nn->nb);
1968 }
1969 
1970 /**
1971  *	call_netdevice_notifiers_info - call all network notifier blocks
1972  *	@val: value passed unmodified to notifier function
1973  *	@info: notifier information data
1974  *
1975  *	Call all network notifier blocks.  Parameters and return value
1976  *	are as for raw_notifier_call_chain().
1977  */
1978 
1979 int call_netdevice_notifiers_info(unsigned long val,
1980 				  struct netdev_notifier_info *info)
1981 {
1982 	struct net *net = dev_net(info->dev);
1983 	int ret;
1984 
1985 	ASSERT_RTNL();
1986 
1987 	/* Run per-netns notifier block chain first, then run the global one.
1988 	 * Hopefully, one day, the global one is going to be removed after
1989 	 * all notifier block registrators get converted to be per-netns.
1990 	 */
1991 	ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1992 	if (ret & NOTIFY_STOP_MASK)
1993 		return ret;
1994 	return raw_notifier_call_chain(&netdev_chain, val, info);
1995 }
1996 
1997 /**
1998  *	call_netdevice_notifiers_info_robust - call per-netns notifier blocks
1999  *	                                       for and rollback on error
2000  *	@val_up: value passed unmodified to notifier function
2001  *	@val_down: value passed unmodified to the notifier function when
2002  *	           recovering from an error on @val_up
2003  *	@info: notifier information data
2004  *
2005  *	Call all per-netns network notifier blocks, but not notifier blocks on
2006  *	the global notifier chain. Parameters and return value are as for
2007  *	raw_notifier_call_chain_robust().
2008  */
2009 
2010 static int
2011 call_netdevice_notifiers_info_robust(unsigned long val_up,
2012 				     unsigned long val_down,
2013 				     struct netdev_notifier_info *info)
2014 {
2015 	struct net *net = dev_net(info->dev);
2016 
2017 	ASSERT_RTNL();
2018 
2019 	return raw_notifier_call_chain_robust(&net->netdev_chain,
2020 					      val_up, val_down, info);
2021 }
2022 
2023 static int call_netdevice_notifiers_extack(unsigned long val,
2024 					   struct net_device *dev,
2025 					   struct netlink_ext_ack *extack)
2026 {
2027 	struct netdev_notifier_info info = {
2028 		.dev = dev,
2029 		.extack = extack,
2030 	};
2031 
2032 	return call_netdevice_notifiers_info(val, &info);
2033 }
2034 
2035 /**
2036  *	call_netdevice_notifiers - call all network notifier blocks
2037  *      @val: value passed unmodified to notifier function
2038  *      @dev: net_device pointer passed unmodified to notifier function
2039  *
2040  *	Call all network notifier blocks.  Parameters and return value
2041  *	are as for raw_notifier_call_chain().
2042  */
2043 
2044 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2045 {
2046 	return call_netdevice_notifiers_extack(val, dev, NULL);
2047 }
2048 EXPORT_SYMBOL(call_netdevice_notifiers);
2049 
2050 /**
2051  *	call_netdevice_notifiers_mtu - call all network notifier blocks
2052  *	@val: value passed unmodified to notifier function
2053  *	@dev: net_device pointer passed unmodified to notifier function
2054  *	@arg: additional u32 argument passed to the notifier function
2055  *
2056  *	Call all network notifier blocks.  Parameters and return value
2057  *	are as for raw_notifier_call_chain().
2058  */
2059 static int call_netdevice_notifiers_mtu(unsigned long val,
2060 					struct net_device *dev, u32 arg)
2061 {
2062 	struct netdev_notifier_info_ext info = {
2063 		.info.dev = dev,
2064 		.ext.mtu = arg,
2065 	};
2066 
2067 	BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2068 
2069 	return call_netdevice_notifiers_info(val, &info.info);
2070 }
2071 
2072 #ifdef CONFIG_NET_INGRESS
2073 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2074 
2075 void net_inc_ingress_queue(void)
2076 {
2077 	static_branch_inc(&ingress_needed_key);
2078 }
2079 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2080 
2081 void net_dec_ingress_queue(void)
2082 {
2083 	static_branch_dec(&ingress_needed_key);
2084 }
2085 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2086 #endif
2087 
2088 #ifdef CONFIG_NET_EGRESS
2089 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2090 
2091 void net_inc_egress_queue(void)
2092 {
2093 	static_branch_inc(&egress_needed_key);
2094 }
2095 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2096 
2097 void net_dec_egress_queue(void)
2098 {
2099 	static_branch_dec(&egress_needed_key);
2100 }
2101 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2102 #endif
2103 
2104 #ifdef CONFIG_NET_CLS_ACT
2105 DEFINE_STATIC_KEY_FALSE(tcf_bypass_check_needed_key);
2106 EXPORT_SYMBOL(tcf_bypass_check_needed_key);
2107 #endif
2108 
2109 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2110 EXPORT_SYMBOL(netstamp_needed_key);
2111 #ifdef CONFIG_JUMP_LABEL
2112 static atomic_t netstamp_needed_deferred;
2113 static atomic_t netstamp_wanted;
2114 static void netstamp_clear(struct work_struct *work)
2115 {
2116 	int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2117 	int wanted;
2118 
2119 	wanted = atomic_add_return(deferred, &netstamp_wanted);
2120 	if (wanted > 0)
2121 		static_branch_enable(&netstamp_needed_key);
2122 	else
2123 		static_branch_disable(&netstamp_needed_key);
2124 }
2125 static DECLARE_WORK(netstamp_work, netstamp_clear);
2126 #endif
2127 
2128 void net_enable_timestamp(void)
2129 {
2130 #ifdef CONFIG_JUMP_LABEL
2131 	int wanted = atomic_read(&netstamp_wanted);
2132 
2133 	while (wanted > 0) {
2134 		if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1))
2135 			return;
2136 	}
2137 	atomic_inc(&netstamp_needed_deferred);
2138 	schedule_work(&netstamp_work);
2139 #else
2140 	static_branch_inc(&netstamp_needed_key);
2141 #endif
2142 }
2143 EXPORT_SYMBOL(net_enable_timestamp);
2144 
2145 void net_disable_timestamp(void)
2146 {
2147 #ifdef CONFIG_JUMP_LABEL
2148 	int wanted = atomic_read(&netstamp_wanted);
2149 
2150 	while (wanted > 1) {
2151 		if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1))
2152 			return;
2153 	}
2154 	atomic_dec(&netstamp_needed_deferred);
2155 	schedule_work(&netstamp_work);
2156 #else
2157 	static_branch_dec(&netstamp_needed_key);
2158 #endif
2159 }
2160 EXPORT_SYMBOL(net_disable_timestamp);
2161 
2162 static inline void net_timestamp_set(struct sk_buff *skb)
2163 {
2164 	skb->tstamp = 0;
2165 	skb->tstamp_type = SKB_CLOCK_REALTIME;
2166 	if (static_branch_unlikely(&netstamp_needed_key))
2167 		skb->tstamp = ktime_get_real();
2168 }
2169 
2170 #define net_timestamp_check(COND, SKB)				\
2171 	if (static_branch_unlikely(&netstamp_needed_key)) {	\
2172 		if ((COND) && !(SKB)->tstamp)			\
2173 			(SKB)->tstamp = ktime_get_real();	\
2174 	}							\
2175 
2176 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2177 {
2178 	return __is_skb_forwardable(dev, skb, true);
2179 }
2180 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2181 
2182 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2183 			      bool check_mtu)
2184 {
2185 	int ret = ____dev_forward_skb(dev, skb, check_mtu);
2186 
2187 	if (likely(!ret)) {
2188 		skb->protocol = eth_type_trans(skb, dev);
2189 		skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2190 	}
2191 
2192 	return ret;
2193 }
2194 
2195 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2196 {
2197 	return __dev_forward_skb2(dev, skb, true);
2198 }
2199 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2200 
2201 /**
2202  * dev_forward_skb - loopback an skb to another netif
2203  *
2204  * @dev: destination network device
2205  * @skb: buffer to forward
2206  *
2207  * return values:
2208  *	NET_RX_SUCCESS	(no congestion)
2209  *	NET_RX_DROP     (packet was dropped, but freed)
2210  *
2211  * dev_forward_skb can be used for injecting an skb from the
2212  * start_xmit function of one device into the receive queue
2213  * of another device.
2214  *
2215  * The receiving device may be in another namespace, so
2216  * we have to clear all information in the skb that could
2217  * impact namespace isolation.
2218  */
2219 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2220 {
2221 	return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2222 }
2223 EXPORT_SYMBOL_GPL(dev_forward_skb);
2224 
2225 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2226 {
2227 	return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2228 }
2229 
2230 static inline int deliver_skb(struct sk_buff *skb,
2231 			      struct packet_type *pt_prev,
2232 			      struct net_device *orig_dev)
2233 {
2234 	if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2235 		return -ENOMEM;
2236 	refcount_inc(&skb->users);
2237 	return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2238 }
2239 
2240 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2241 					  struct packet_type **pt,
2242 					  struct net_device *orig_dev,
2243 					  __be16 type,
2244 					  struct list_head *ptype_list)
2245 {
2246 	struct packet_type *ptype, *pt_prev = *pt;
2247 
2248 	list_for_each_entry_rcu(ptype, ptype_list, list) {
2249 		if (ptype->type != type)
2250 			continue;
2251 		if (pt_prev)
2252 			deliver_skb(skb, pt_prev, orig_dev);
2253 		pt_prev = ptype;
2254 	}
2255 	*pt = pt_prev;
2256 }
2257 
2258 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2259 {
2260 	if (!ptype->af_packet_priv || !skb->sk)
2261 		return false;
2262 
2263 	if (ptype->id_match)
2264 		return ptype->id_match(ptype, skb->sk);
2265 	else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2266 		return true;
2267 
2268 	return false;
2269 }
2270 
2271 /**
2272  * dev_nit_active - return true if any network interface taps are in use
2273  *
2274  * @dev: network device to check for the presence of taps
2275  */
2276 bool dev_nit_active(struct net_device *dev)
2277 {
2278 	return !list_empty(&net_hotdata.ptype_all) ||
2279 	       !list_empty(&dev->ptype_all);
2280 }
2281 EXPORT_SYMBOL_GPL(dev_nit_active);
2282 
2283 /*
2284  *	Support routine. Sends outgoing frames to any network
2285  *	taps currently in use.
2286  */
2287 
2288 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2289 {
2290 	struct list_head *ptype_list = &net_hotdata.ptype_all;
2291 	struct packet_type *ptype, *pt_prev = NULL;
2292 	struct sk_buff *skb2 = NULL;
2293 
2294 	rcu_read_lock();
2295 again:
2296 	list_for_each_entry_rcu(ptype, ptype_list, list) {
2297 		if (READ_ONCE(ptype->ignore_outgoing))
2298 			continue;
2299 
2300 		/* Never send packets back to the socket
2301 		 * they originated from - MvS (miquels@drinkel.ow.org)
2302 		 */
2303 		if (skb_loop_sk(ptype, skb))
2304 			continue;
2305 
2306 		if (pt_prev) {
2307 			deliver_skb(skb2, pt_prev, skb->dev);
2308 			pt_prev = ptype;
2309 			continue;
2310 		}
2311 
2312 		/* need to clone skb, done only once */
2313 		skb2 = skb_clone(skb, GFP_ATOMIC);
2314 		if (!skb2)
2315 			goto out_unlock;
2316 
2317 		net_timestamp_set(skb2);
2318 
2319 		/* skb->nh should be correctly
2320 		 * set by sender, so that the second statement is
2321 		 * just protection against buggy protocols.
2322 		 */
2323 		skb_reset_mac_header(skb2);
2324 
2325 		if (skb_network_header(skb2) < skb2->data ||
2326 		    skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2327 			net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2328 					     ntohs(skb2->protocol),
2329 					     dev->name);
2330 			skb_reset_network_header(skb2);
2331 		}
2332 
2333 		skb2->transport_header = skb2->network_header;
2334 		skb2->pkt_type = PACKET_OUTGOING;
2335 		pt_prev = ptype;
2336 	}
2337 
2338 	if (ptype_list == &net_hotdata.ptype_all) {
2339 		ptype_list = &dev->ptype_all;
2340 		goto again;
2341 	}
2342 out_unlock:
2343 	if (pt_prev) {
2344 		if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2345 			pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2346 		else
2347 			kfree_skb(skb2);
2348 	}
2349 	rcu_read_unlock();
2350 }
2351 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2352 
2353 /**
2354  * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2355  * @dev: Network device
2356  * @txq: number of queues available
2357  *
2358  * If real_num_tx_queues is changed the tc mappings may no longer be
2359  * valid. To resolve this verify the tc mapping remains valid and if
2360  * not NULL the mapping. With no priorities mapping to this
2361  * offset/count pair it will no longer be used. In the worst case TC0
2362  * is invalid nothing can be done so disable priority mappings. If is
2363  * expected that drivers will fix this mapping if they can before
2364  * calling netif_set_real_num_tx_queues.
2365  */
2366 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2367 {
2368 	int i;
2369 	struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2370 
2371 	/* If TC0 is invalidated disable TC mapping */
2372 	if (tc->offset + tc->count > txq) {
2373 		netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2374 		dev->num_tc = 0;
2375 		return;
2376 	}
2377 
2378 	/* Invalidated prio to tc mappings set to TC0 */
2379 	for (i = 1; i < TC_BITMASK + 1; i++) {
2380 		int q = netdev_get_prio_tc_map(dev, i);
2381 
2382 		tc = &dev->tc_to_txq[q];
2383 		if (tc->offset + tc->count > txq) {
2384 			netdev_warn(dev, "Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2385 				    i, q);
2386 			netdev_set_prio_tc_map(dev, i, 0);
2387 		}
2388 	}
2389 }
2390 
2391 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2392 {
2393 	if (dev->num_tc) {
2394 		struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2395 		int i;
2396 
2397 		/* walk through the TCs and see if it falls into any of them */
2398 		for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2399 			if ((txq - tc->offset) < tc->count)
2400 				return i;
2401 		}
2402 
2403 		/* didn't find it, just return -1 to indicate no match */
2404 		return -1;
2405 	}
2406 
2407 	return 0;
2408 }
2409 EXPORT_SYMBOL(netdev_txq_to_tc);
2410 
2411 #ifdef CONFIG_XPS
2412 static struct static_key xps_needed __read_mostly;
2413 static struct static_key xps_rxqs_needed __read_mostly;
2414 static DEFINE_MUTEX(xps_map_mutex);
2415 #define xmap_dereference(P)		\
2416 	rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2417 
2418 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2419 			     struct xps_dev_maps *old_maps, int tci, u16 index)
2420 {
2421 	struct xps_map *map = NULL;
2422 	int pos;
2423 
2424 	map = xmap_dereference(dev_maps->attr_map[tci]);
2425 	if (!map)
2426 		return false;
2427 
2428 	for (pos = map->len; pos--;) {
2429 		if (map->queues[pos] != index)
2430 			continue;
2431 
2432 		if (map->len > 1) {
2433 			map->queues[pos] = map->queues[--map->len];
2434 			break;
2435 		}
2436 
2437 		if (old_maps)
2438 			RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2439 		RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2440 		kfree_rcu(map, rcu);
2441 		return false;
2442 	}
2443 
2444 	return true;
2445 }
2446 
2447 static bool remove_xps_queue_cpu(struct net_device *dev,
2448 				 struct xps_dev_maps *dev_maps,
2449 				 int cpu, u16 offset, u16 count)
2450 {
2451 	int num_tc = dev_maps->num_tc;
2452 	bool active = false;
2453 	int tci;
2454 
2455 	for (tci = cpu * num_tc; num_tc--; tci++) {
2456 		int i, j;
2457 
2458 		for (i = count, j = offset; i--; j++) {
2459 			if (!remove_xps_queue(dev_maps, NULL, tci, j))
2460 				break;
2461 		}
2462 
2463 		active |= i < 0;
2464 	}
2465 
2466 	return active;
2467 }
2468 
2469 static void reset_xps_maps(struct net_device *dev,
2470 			   struct xps_dev_maps *dev_maps,
2471 			   enum xps_map_type type)
2472 {
2473 	static_key_slow_dec_cpuslocked(&xps_needed);
2474 	if (type == XPS_RXQS)
2475 		static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2476 
2477 	RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2478 
2479 	kfree_rcu(dev_maps, rcu);
2480 }
2481 
2482 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2483 			   u16 offset, u16 count)
2484 {
2485 	struct xps_dev_maps *dev_maps;
2486 	bool active = false;
2487 	int i, j;
2488 
2489 	dev_maps = xmap_dereference(dev->xps_maps[type]);
2490 	if (!dev_maps)
2491 		return;
2492 
2493 	for (j = 0; j < dev_maps->nr_ids; j++)
2494 		active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2495 	if (!active)
2496 		reset_xps_maps(dev, dev_maps, type);
2497 
2498 	if (type == XPS_CPUS) {
2499 		for (i = offset + (count - 1); count--; i--)
2500 			netdev_queue_numa_node_write(
2501 				netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2502 	}
2503 }
2504 
2505 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2506 				   u16 count)
2507 {
2508 	if (!static_key_false(&xps_needed))
2509 		return;
2510 
2511 	cpus_read_lock();
2512 	mutex_lock(&xps_map_mutex);
2513 
2514 	if (static_key_false(&xps_rxqs_needed))
2515 		clean_xps_maps(dev, XPS_RXQS, offset, count);
2516 
2517 	clean_xps_maps(dev, XPS_CPUS, offset, count);
2518 
2519 	mutex_unlock(&xps_map_mutex);
2520 	cpus_read_unlock();
2521 }
2522 
2523 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2524 {
2525 	netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2526 }
2527 
2528 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2529 				      u16 index, bool is_rxqs_map)
2530 {
2531 	struct xps_map *new_map;
2532 	int alloc_len = XPS_MIN_MAP_ALLOC;
2533 	int i, pos;
2534 
2535 	for (pos = 0; map && pos < map->len; pos++) {
2536 		if (map->queues[pos] != index)
2537 			continue;
2538 		return map;
2539 	}
2540 
2541 	/* Need to add tx-queue to this CPU's/rx-queue's existing map */
2542 	if (map) {
2543 		if (pos < map->alloc_len)
2544 			return map;
2545 
2546 		alloc_len = map->alloc_len * 2;
2547 	}
2548 
2549 	/* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2550 	 *  map
2551 	 */
2552 	if (is_rxqs_map)
2553 		new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2554 	else
2555 		new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2556 				       cpu_to_node(attr_index));
2557 	if (!new_map)
2558 		return NULL;
2559 
2560 	for (i = 0; i < pos; i++)
2561 		new_map->queues[i] = map->queues[i];
2562 	new_map->alloc_len = alloc_len;
2563 	new_map->len = pos;
2564 
2565 	return new_map;
2566 }
2567 
2568 /* Copy xps maps at a given index */
2569 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2570 			      struct xps_dev_maps *new_dev_maps, int index,
2571 			      int tc, bool skip_tc)
2572 {
2573 	int i, tci = index * dev_maps->num_tc;
2574 	struct xps_map *map;
2575 
2576 	/* copy maps belonging to foreign traffic classes */
2577 	for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2578 		if (i == tc && skip_tc)
2579 			continue;
2580 
2581 		/* fill in the new device map from the old device map */
2582 		map = xmap_dereference(dev_maps->attr_map[tci]);
2583 		RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2584 	}
2585 }
2586 
2587 /* Must be called under cpus_read_lock */
2588 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2589 			  u16 index, enum xps_map_type type)
2590 {
2591 	struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2592 	const unsigned long *online_mask = NULL;
2593 	bool active = false, copy = false;
2594 	int i, j, tci, numa_node_id = -2;
2595 	int maps_sz, num_tc = 1, tc = 0;
2596 	struct xps_map *map, *new_map;
2597 	unsigned int nr_ids;
2598 
2599 	WARN_ON_ONCE(index >= dev->num_tx_queues);
2600 
2601 	if (dev->num_tc) {
2602 		/* Do not allow XPS on subordinate device directly */
2603 		num_tc = dev->num_tc;
2604 		if (num_tc < 0)
2605 			return -EINVAL;
2606 
2607 		/* If queue belongs to subordinate dev use its map */
2608 		dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2609 
2610 		tc = netdev_txq_to_tc(dev, index);
2611 		if (tc < 0)
2612 			return -EINVAL;
2613 	}
2614 
2615 	mutex_lock(&xps_map_mutex);
2616 
2617 	dev_maps = xmap_dereference(dev->xps_maps[type]);
2618 	if (type == XPS_RXQS) {
2619 		maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2620 		nr_ids = dev->num_rx_queues;
2621 	} else {
2622 		maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2623 		if (num_possible_cpus() > 1)
2624 			online_mask = cpumask_bits(cpu_online_mask);
2625 		nr_ids = nr_cpu_ids;
2626 	}
2627 
2628 	if (maps_sz < L1_CACHE_BYTES)
2629 		maps_sz = L1_CACHE_BYTES;
2630 
2631 	/* The old dev_maps could be larger or smaller than the one we're
2632 	 * setting up now, as dev->num_tc or nr_ids could have been updated in
2633 	 * between. We could try to be smart, but let's be safe instead and only
2634 	 * copy foreign traffic classes if the two map sizes match.
2635 	 */
2636 	if (dev_maps &&
2637 	    dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2638 		copy = true;
2639 
2640 	/* allocate memory for queue storage */
2641 	for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2642 	     j < nr_ids;) {
2643 		if (!new_dev_maps) {
2644 			new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2645 			if (!new_dev_maps) {
2646 				mutex_unlock(&xps_map_mutex);
2647 				return -ENOMEM;
2648 			}
2649 
2650 			new_dev_maps->nr_ids = nr_ids;
2651 			new_dev_maps->num_tc = num_tc;
2652 		}
2653 
2654 		tci = j * num_tc + tc;
2655 		map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2656 
2657 		map = expand_xps_map(map, j, index, type == XPS_RXQS);
2658 		if (!map)
2659 			goto error;
2660 
2661 		RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2662 	}
2663 
2664 	if (!new_dev_maps)
2665 		goto out_no_new_maps;
2666 
2667 	if (!dev_maps) {
2668 		/* Increment static keys at most once per type */
2669 		static_key_slow_inc_cpuslocked(&xps_needed);
2670 		if (type == XPS_RXQS)
2671 			static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2672 	}
2673 
2674 	for (j = 0; j < nr_ids; j++) {
2675 		bool skip_tc = false;
2676 
2677 		tci = j * num_tc + tc;
2678 		if (netif_attr_test_mask(j, mask, nr_ids) &&
2679 		    netif_attr_test_online(j, online_mask, nr_ids)) {
2680 			/* add tx-queue to CPU/rx-queue maps */
2681 			int pos = 0;
2682 
2683 			skip_tc = true;
2684 
2685 			map = xmap_dereference(new_dev_maps->attr_map[tci]);
2686 			while ((pos < map->len) && (map->queues[pos] != index))
2687 				pos++;
2688 
2689 			if (pos == map->len)
2690 				map->queues[map->len++] = index;
2691 #ifdef CONFIG_NUMA
2692 			if (type == XPS_CPUS) {
2693 				if (numa_node_id == -2)
2694 					numa_node_id = cpu_to_node(j);
2695 				else if (numa_node_id != cpu_to_node(j))
2696 					numa_node_id = -1;
2697 			}
2698 #endif
2699 		}
2700 
2701 		if (copy)
2702 			xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2703 					  skip_tc);
2704 	}
2705 
2706 	rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2707 
2708 	/* Cleanup old maps */
2709 	if (!dev_maps)
2710 		goto out_no_old_maps;
2711 
2712 	for (j = 0; j < dev_maps->nr_ids; j++) {
2713 		for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2714 			map = xmap_dereference(dev_maps->attr_map[tci]);
2715 			if (!map)
2716 				continue;
2717 
2718 			if (copy) {
2719 				new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2720 				if (map == new_map)
2721 					continue;
2722 			}
2723 
2724 			RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2725 			kfree_rcu(map, rcu);
2726 		}
2727 	}
2728 
2729 	old_dev_maps = dev_maps;
2730 
2731 out_no_old_maps:
2732 	dev_maps = new_dev_maps;
2733 	active = true;
2734 
2735 out_no_new_maps:
2736 	if (type == XPS_CPUS)
2737 		/* update Tx queue numa node */
2738 		netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2739 					     (numa_node_id >= 0) ?
2740 					     numa_node_id : NUMA_NO_NODE);
2741 
2742 	if (!dev_maps)
2743 		goto out_no_maps;
2744 
2745 	/* removes tx-queue from unused CPUs/rx-queues */
2746 	for (j = 0; j < dev_maps->nr_ids; j++) {
2747 		tci = j * dev_maps->num_tc;
2748 
2749 		for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2750 			if (i == tc &&
2751 			    netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2752 			    netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2753 				continue;
2754 
2755 			active |= remove_xps_queue(dev_maps,
2756 						   copy ? old_dev_maps : NULL,
2757 						   tci, index);
2758 		}
2759 	}
2760 
2761 	if (old_dev_maps)
2762 		kfree_rcu(old_dev_maps, rcu);
2763 
2764 	/* free map if not active */
2765 	if (!active)
2766 		reset_xps_maps(dev, dev_maps, type);
2767 
2768 out_no_maps:
2769 	mutex_unlock(&xps_map_mutex);
2770 
2771 	return 0;
2772 error:
2773 	/* remove any maps that we added */
2774 	for (j = 0; j < nr_ids; j++) {
2775 		for (i = num_tc, tci = j * num_tc; i--; tci++) {
2776 			new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2777 			map = copy ?
2778 			      xmap_dereference(dev_maps->attr_map[tci]) :
2779 			      NULL;
2780 			if (new_map && new_map != map)
2781 				kfree(new_map);
2782 		}
2783 	}
2784 
2785 	mutex_unlock(&xps_map_mutex);
2786 
2787 	kfree(new_dev_maps);
2788 	return -ENOMEM;
2789 }
2790 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2791 
2792 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2793 			u16 index)
2794 {
2795 	int ret;
2796 
2797 	cpus_read_lock();
2798 	ret =  __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2799 	cpus_read_unlock();
2800 
2801 	return ret;
2802 }
2803 EXPORT_SYMBOL(netif_set_xps_queue);
2804 
2805 #endif
2806 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2807 {
2808 	struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2809 
2810 	/* Unbind any subordinate channels */
2811 	while (txq-- != &dev->_tx[0]) {
2812 		if (txq->sb_dev)
2813 			netdev_unbind_sb_channel(dev, txq->sb_dev);
2814 	}
2815 }
2816 
2817 void netdev_reset_tc(struct net_device *dev)
2818 {
2819 #ifdef CONFIG_XPS
2820 	netif_reset_xps_queues_gt(dev, 0);
2821 #endif
2822 	netdev_unbind_all_sb_channels(dev);
2823 
2824 	/* Reset TC configuration of device */
2825 	dev->num_tc = 0;
2826 	memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2827 	memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2828 }
2829 EXPORT_SYMBOL(netdev_reset_tc);
2830 
2831 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2832 {
2833 	if (tc >= dev->num_tc)
2834 		return -EINVAL;
2835 
2836 #ifdef CONFIG_XPS
2837 	netif_reset_xps_queues(dev, offset, count);
2838 #endif
2839 	dev->tc_to_txq[tc].count = count;
2840 	dev->tc_to_txq[tc].offset = offset;
2841 	return 0;
2842 }
2843 EXPORT_SYMBOL(netdev_set_tc_queue);
2844 
2845 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2846 {
2847 	if (num_tc > TC_MAX_QUEUE)
2848 		return -EINVAL;
2849 
2850 #ifdef CONFIG_XPS
2851 	netif_reset_xps_queues_gt(dev, 0);
2852 #endif
2853 	netdev_unbind_all_sb_channels(dev);
2854 
2855 	dev->num_tc = num_tc;
2856 	return 0;
2857 }
2858 EXPORT_SYMBOL(netdev_set_num_tc);
2859 
2860 void netdev_unbind_sb_channel(struct net_device *dev,
2861 			      struct net_device *sb_dev)
2862 {
2863 	struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2864 
2865 #ifdef CONFIG_XPS
2866 	netif_reset_xps_queues_gt(sb_dev, 0);
2867 #endif
2868 	memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2869 	memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2870 
2871 	while (txq-- != &dev->_tx[0]) {
2872 		if (txq->sb_dev == sb_dev)
2873 			txq->sb_dev = NULL;
2874 	}
2875 }
2876 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2877 
2878 int netdev_bind_sb_channel_queue(struct net_device *dev,
2879 				 struct net_device *sb_dev,
2880 				 u8 tc, u16 count, u16 offset)
2881 {
2882 	/* Make certain the sb_dev and dev are already configured */
2883 	if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2884 		return -EINVAL;
2885 
2886 	/* We cannot hand out queues we don't have */
2887 	if ((offset + count) > dev->real_num_tx_queues)
2888 		return -EINVAL;
2889 
2890 	/* Record the mapping */
2891 	sb_dev->tc_to_txq[tc].count = count;
2892 	sb_dev->tc_to_txq[tc].offset = offset;
2893 
2894 	/* Provide a way for Tx queue to find the tc_to_txq map or
2895 	 * XPS map for itself.
2896 	 */
2897 	while (count--)
2898 		netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2899 
2900 	return 0;
2901 }
2902 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2903 
2904 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2905 {
2906 	/* Do not use a multiqueue device to represent a subordinate channel */
2907 	if (netif_is_multiqueue(dev))
2908 		return -ENODEV;
2909 
2910 	/* We allow channels 1 - 32767 to be used for subordinate channels.
2911 	 * Channel 0 is meant to be "native" mode and used only to represent
2912 	 * the main root device. We allow writing 0 to reset the device back
2913 	 * to normal mode after being used as a subordinate channel.
2914 	 */
2915 	if (channel > S16_MAX)
2916 		return -EINVAL;
2917 
2918 	dev->num_tc = -channel;
2919 
2920 	return 0;
2921 }
2922 EXPORT_SYMBOL(netdev_set_sb_channel);
2923 
2924 /*
2925  * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2926  * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2927  */
2928 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2929 {
2930 	bool disabling;
2931 	int rc;
2932 
2933 	disabling = txq < dev->real_num_tx_queues;
2934 
2935 	if (txq < 1 || txq > dev->num_tx_queues)
2936 		return -EINVAL;
2937 
2938 	if (dev->reg_state == NETREG_REGISTERED ||
2939 	    dev->reg_state == NETREG_UNREGISTERING) {
2940 		ASSERT_RTNL();
2941 
2942 		rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2943 						  txq);
2944 		if (rc)
2945 			return rc;
2946 
2947 		if (dev->num_tc)
2948 			netif_setup_tc(dev, txq);
2949 
2950 		dev_qdisc_change_real_num_tx(dev, txq);
2951 
2952 		dev->real_num_tx_queues = txq;
2953 
2954 		if (disabling) {
2955 			synchronize_net();
2956 			qdisc_reset_all_tx_gt(dev, txq);
2957 #ifdef CONFIG_XPS
2958 			netif_reset_xps_queues_gt(dev, txq);
2959 #endif
2960 		}
2961 	} else {
2962 		dev->real_num_tx_queues = txq;
2963 	}
2964 
2965 	return 0;
2966 }
2967 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2968 
2969 #ifdef CONFIG_SYSFS
2970 /**
2971  *	netif_set_real_num_rx_queues - set actual number of RX queues used
2972  *	@dev: Network device
2973  *	@rxq: Actual number of RX queues
2974  *
2975  *	This must be called either with the rtnl_lock held or before
2976  *	registration of the net device.  Returns 0 on success, or a
2977  *	negative error code.  If called before registration, it always
2978  *	succeeds.
2979  */
2980 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2981 {
2982 	int rc;
2983 
2984 	if (rxq < 1 || rxq > dev->num_rx_queues)
2985 		return -EINVAL;
2986 
2987 	if (dev->reg_state == NETREG_REGISTERED) {
2988 		ASSERT_RTNL();
2989 
2990 		rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2991 						  rxq);
2992 		if (rc)
2993 			return rc;
2994 	}
2995 
2996 	dev->real_num_rx_queues = rxq;
2997 	return 0;
2998 }
2999 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
3000 #endif
3001 
3002 /**
3003  *	netif_set_real_num_queues - set actual number of RX and TX queues used
3004  *	@dev: Network device
3005  *	@txq: Actual number of TX queues
3006  *	@rxq: Actual number of RX queues
3007  *
3008  *	Set the real number of both TX and RX queues.
3009  *	Does nothing if the number of queues is already correct.
3010  */
3011 int netif_set_real_num_queues(struct net_device *dev,
3012 			      unsigned int txq, unsigned int rxq)
3013 {
3014 	unsigned int old_rxq = dev->real_num_rx_queues;
3015 	int err;
3016 
3017 	if (txq < 1 || txq > dev->num_tx_queues ||
3018 	    rxq < 1 || rxq > dev->num_rx_queues)
3019 		return -EINVAL;
3020 
3021 	/* Start from increases, so the error path only does decreases -
3022 	 * decreases can't fail.
3023 	 */
3024 	if (rxq > dev->real_num_rx_queues) {
3025 		err = netif_set_real_num_rx_queues(dev, rxq);
3026 		if (err)
3027 			return err;
3028 	}
3029 	if (txq > dev->real_num_tx_queues) {
3030 		err = netif_set_real_num_tx_queues(dev, txq);
3031 		if (err)
3032 			goto undo_rx;
3033 	}
3034 	if (rxq < dev->real_num_rx_queues)
3035 		WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
3036 	if (txq < dev->real_num_tx_queues)
3037 		WARN_ON(netif_set_real_num_tx_queues(dev, txq));
3038 
3039 	return 0;
3040 undo_rx:
3041 	WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
3042 	return err;
3043 }
3044 EXPORT_SYMBOL(netif_set_real_num_queues);
3045 
3046 /**
3047  * netif_set_tso_max_size() - set the max size of TSO frames supported
3048  * @dev:	netdev to update
3049  * @size:	max skb->len of a TSO frame
3050  *
3051  * Set the limit on the size of TSO super-frames the device can handle.
3052  * Unless explicitly set the stack will assume the value of
3053  * %GSO_LEGACY_MAX_SIZE.
3054  */
3055 void netif_set_tso_max_size(struct net_device *dev, unsigned int size)
3056 {
3057 	dev->tso_max_size = min(GSO_MAX_SIZE, size);
3058 	if (size < READ_ONCE(dev->gso_max_size))
3059 		netif_set_gso_max_size(dev, size);
3060 	if (size < READ_ONCE(dev->gso_ipv4_max_size))
3061 		netif_set_gso_ipv4_max_size(dev, size);
3062 }
3063 EXPORT_SYMBOL(netif_set_tso_max_size);
3064 
3065 /**
3066  * netif_set_tso_max_segs() - set the max number of segs supported for TSO
3067  * @dev:	netdev to update
3068  * @segs:	max number of TCP segments
3069  *
3070  * Set the limit on the number of TCP segments the device can generate from
3071  * a single TSO super-frame.
3072  * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS.
3073  */
3074 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs)
3075 {
3076 	dev->tso_max_segs = segs;
3077 	if (segs < READ_ONCE(dev->gso_max_segs))
3078 		netif_set_gso_max_segs(dev, segs);
3079 }
3080 EXPORT_SYMBOL(netif_set_tso_max_segs);
3081 
3082 /**
3083  * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper
3084  * @to:		netdev to update
3085  * @from:	netdev from which to copy the limits
3086  */
3087 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from)
3088 {
3089 	netif_set_tso_max_size(to, from->tso_max_size);
3090 	netif_set_tso_max_segs(to, from->tso_max_segs);
3091 }
3092 EXPORT_SYMBOL(netif_inherit_tso_max);
3093 
3094 /**
3095  * netif_get_num_default_rss_queues - default number of RSS queues
3096  *
3097  * Default value is the number of physical cores if there are only 1 or 2, or
3098  * divided by 2 if there are more.
3099  */
3100 int netif_get_num_default_rss_queues(void)
3101 {
3102 	cpumask_var_t cpus;
3103 	int cpu, count = 0;
3104 
3105 	if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL)))
3106 		return 1;
3107 
3108 	cpumask_copy(cpus, cpu_online_mask);
3109 	for_each_cpu(cpu, cpus) {
3110 		++count;
3111 		cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
3112 	}
3113 	free_cpumask_var(cpus);
3114 
3115 	return count > 2 ? DIV_ROUND_UP(count, 2) : count;
3116 }
3117 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3118 
3119 static void __netif_reschedule(struct Qdisc *q)
3120 {
3121 	struct softnet_data *sd;
3122 	unsigned long flags;
3123 
3124 	local_irq_save(flags);
3125 	sd = this_cpu_ptr(&softnet_data);
3126 	q->next_sched = NULL;
3127 	*sd->output_queue_tailp = q;
3128 	sd->output_queue_tailp = &q->next_sched;
3129 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
3130 	local_irq_restore(flags);
3131 }
3132 
3133 void __netif_schedule(struct Qdisc *q)
3134 {
3135 	if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3136 		__netif_reschedule(q);
3137 }
3138 EXPORT_SYMBOL(__netif_schedule);
3139 
3140 struct dev_kfree_skb_cb {
3141 	enum skb_drop_reason reason;
3142 };
3143 
3144 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3145 {
3146 	return (struct dev_kfree_skb_cb *)skb->cb;
3147 }
3148 
3149 void netif_schedule_queue(struct netdev_queue *txq)
3150 {
3151 	rcu_read_lock();
3152 	if (!netif_xmit_stopped(txq)) {
3153 		struct Qdisc *q = rcu_dereference(txq->qdisc);
3154 
3155 		__netif_schedule(q);
3156 	}
3157 	rcu_read_unlock();
3158 }
3159 EXPORT_SYMBOL(netif_schedule_queue);
3160 
3161 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3162 {
3163 	if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3164 		struct Qdisc *q;
3165 
3166 		rcu_read_lock();
3167 		q = rcu_dereference(dev_queue->qdisc);
3168 		__netif_schedule(q);
3169 		rcu_read_unlock();
3170 	}
3171 }
3172 EXPORT_SYMBOL(netif_tx_wake_queue);
3173 
3174 void dev_kfree_skb_irq_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3175 {
3176 	unsigned long flags;
3177 
3178 	if (unlikely(!skb))
3179 		return;
3180 
3181 	if (likely(refcount_read(&skb->users) == 1)) {
3182 		smp_rmb();
3183 		refcount_set(&skb->users, 0);
3184 	} else if (likely(!refcount_dec_and_test(&skb->users))) {
3185 		return;
3186 	}
3187 	get_kfree_skb_cb(skb)->reason = reason;
3188 	local_irq_save(flags);
3189 	skb->next = __this_cpu_read(softnet_data.completion_queue);
3190 	__this_cpu_write(softnet_data.completion_queue, skb);
3191 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
3192 	local_irq_restore(flags);
3193 }
3194 EXPORT_SYMBOL(dev_kfree_skb_irq_reason);
3195 
3196 void dev_kfree_skb_any_reason(struct sk_buff *skb, enum skb_drop_reason reason)
3197 {
3198 	if (in_hardirq() || irqs_disabled())
3199 		dev_kfree_skb_irq_reason(skb, reason);
3200 	else
3201 		kfree_skb_reason(skb, reason);
3202 }
3203 EXPORT_SYMBOL(dev_kfree_skb_any_reason);
3204 
3205 
3206 /**
3207  * netif_device_detach - mark device as removed
3208  * @dev: network device
3209  *
3210  * Mark device as removed from system and therefore no longer available.
3211  */
3212 void netif_device_detach(struct net_device *dev)
3213 {
3214 	if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3215 	    netif_running(dev)) {
3216 		netif_tx_stop_all_queues(dev);
3217 	}
3218 }
3219 EXPORT_SYMBOL(netif_device_detach);
3220 
3221 /**
3222  * netif_device_attach - mark device as attached
3223  * @dev: network device
3224  *
3225  * Mark device as attached from system and restart if needed.
3226  */
3227 void netif_device_attach(struct net_device *dev)
3228 {
3229 	if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3230 	    netif_running(dev)) {
3231 		netif_tx_wake_all_queues(dev);
3232 		__netdev_watchdog_up(dev);
3233 	}
3234 }
3235 EXPORT_SYMBOL(netif_device_attach);
3236 
3237 /*
3238  * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3239  * to be used as a distribution range.
3240  */
3241 static u16 skb_tx_hash(const struct net_device *dev,
3242 		       const struct net_device *sb_dev,
3243 		       struct sk_buff *skb)
3244 {
3245 	u32 hash;
3246 	u16 qoffset = 0;
3247 	u16 qcount = dev->real_num_tx_queues;
3248 
3249 	if (dev->num_tc) {
3250 		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3251 
3252 		qoffset = sb_dev->tc_to_txq[tc].offset;
3253 		qcount = sb_dev->tc_to_txq[tc].count;
3254 		if (unlikely(!qcount)) {
3255 			net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3256 					     sb_dev->name, qoffset, tc);
3257 			qoffset = 0;
3258 			qcount = dev->real_num_tx_queues;
3259 		}
3260 	}
3261 
3262 	if (skb_rx_queue_recorded(skb)) {
3263 		DEBUG_NET_WARN_ON_ONCE(qcount == 0);
3264 		hash = skb_get_rx_queue(skb);
3265 		if (hash >= qoffset)
3266 			hash -= qoffset;
3267 		while (unlikely(hash >= qcount))
3268 			hash -= qcount;
3269 		return hash + qoffset;
3270 	}
3271 
3272 	return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3273 }
3274 
3275 void skb_warn_bad_offload(const struct sk_buff *skb)
3276 {
3277 	static const netdev_features_t null_features;
3278 	struct net_device *dev = skb->dev;
3279 	const char *name = "";
3280 
3281 	if (!net_ratelimit())
3282 		return;
3283 
3284 	if (dev) {
3285 		if (dev->dev.parent)
3286 			name = dev_driver_string(dev->dev.parent);
3287 		else
3288 			name = netdev_name(dev);
3289 	}
3290 	skb_dump(KERN_WARNING, skb, false);
3291 	WARN(1, "%s: caps=(%pNF, %pNF)\n",
3292 	     name, dev ? &dev->features : &null_features,
3293 	     skb->sk ? &skb->sk->sk_route_caps : &null_features);
3294 }
3295 
3296 /*
3297  * Invalidate hardware checksum when packet is to be mangled, and
3298  * complete checksum manually on outgoing path.
3299  */
3300 int skb_checksum_help(struct sk_buff *skb)
3301 {
3302 	__wsum csum;
3303 	int ret = 0, offset;
3304 
3305 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3306 		goto out_set_summed;
3307 
3308 	if (unlikely(skb_is_gso(skb))) {
3309 		skb_warn_bad_offload(skb);
3310 		return -EINVAL;
3311 	}
3312 
3313 	/* Before computing a checksum, we should make sure no frag could
3314 	 * be modified by an external entity : checksum could be wrong.
3315 	 */
3316 	if (skb_has_shared_frag(skb)) {
3317 		ret = __skb_linearize(skb);
3318 		if (ret)
3319 			goto out;
3320 	}
3321 
3322 	offset = skb_checksum_start_offset(skb);
3323 	ret = -EINVAL;
3324 	if (unlikely(offset >= skb_headlen(skb))) {
3325 		DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3326 		WARN_ONCE(true, "offset (%d) >= skb_headlen() (%u)\n",
3327 			  offset, skb_headlen(skb));
3328 		goto out;
3329 	}
3330 	csum = skb_checksum(skb, offset, skb->len - offset, 0);
3331 
3332 	offset += skb->csum_offset;
3333 	if (unlikely(offset + sizeof(__sum16) > skb_headlen(skb))) {
3334 		DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3335 		WARN_ONCE(true, "offset+2 (%zu) > skb_headlen() (%u)\n",
3336 			  offset + sizeof(__sum16), skb_headlen(skb));
3337 		goto out;
3338 	}
3339 	ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3340 	if (ret)
3341 		goto out;
3342 
3343 	*(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3344 out_set_summed:
3345 	skb->ip_summed = CHECKSUM_NONE;
3346 out:
3347 	return ret;
3348 }
3349 EXPORT_SYMBOL(skb_checksum_help);
3350 
3351 int skb_crc32c_csum_help(struct sk_buff *skb)
3352 {
3353 	__le32 crc32c_csum;
3354 	int ret = 0, offset, start;
3355 
3356 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3357 		goto out;
3358 
3359 	if (unlikely(skb_is_gso(skb)))
3360 		goto out;
3361 
3362 	/* Before computing a checksum, we should make sure no frag could
3363 	 * be modified by an external entity : checksum could be wrong.
3364 	 */
3365 	if (unlikely(skb_has_shared_frag(skb))) {
3366 		ret = __skb_linearize(skb);
3367 		if (ret)
3368 			goto out;
3369 	}
3370 	start = skb_checksum_start_offset(skb);
3371 	offset = start + offsetof(struct sctphdr, checksum);
3372 	if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3373 		ret = -EINVAL;
3374 		goto out;
3375 	}
3376 
3377 	ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3378 	if (ret)
3379 		goto out;
3380 
3381 	crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3382 						  skb->len - start, ~(__u32)0,
3383 						  crc32c_csum_stub));
3384 	*(__le32 *)(skb->data + offset) = crc32c_csum;
3385 	skb_reset_csum_not_inet(skb);
3386 out:
3387 	return ret;
3388 }
3389 
3390 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3391 {
3392 	__be16 type = skb->protocol;
3393 
3394 	/* Tunnel gso handlers can set protocol to ethernet. */
3395 	if (type == htons(ETH_P_TEB)) {
3396 		struct ethhdr *eth;
3397 
3398 		if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3399 			return 0;
3400 
3401 		eth = (struct ethhdr *)skb->data;
3402 		type = eth->h_proto;
3403 	}
3404 
3405 	return vlan_get_protocol_and_depth(skb, type, depth);
3406 }
3407 
3408 
3409 /* Take action when hardware reception checksum errors are detected. */
3410 #ifdef CONFIG_BUG
3411 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3412 {
3413 	netdev_err(dev, "hw csum failure\n");
3414 	skb_dump(KERN_ERR, skb, true);
3415 	dump_stack();
3416 }
3417 
3418 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3419 {
3420 	DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3421 }
3422 EXPORT_SYMBOL(netdev_rx_csum_fault);
3423 #endif
3424 
3425 /* XXX: check that highmem exists at all on the given machine. */
3426 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3427 {
3428 #ifdef CONFIG_HIGHMEM
3429 	int i;
3430 
3431 	if (!(dev->features & NETIF_F_HIGHDMA)) {
3432 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3433 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3434 
3435 			if (PageHighMem(skb_frag_page(frag)))
3436 				return 1;
3437 		}
3438 	}
3439 #endif
3440 	return 0;
3441 }
3442 
3443 /* If MPLS offload request, verify we are testing hardware MPLS features
3444  * instead of standard features for the netdev.
3445  */
3446 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3447 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3448 					   netdev_features_t features,
3449 					   __be16 type)
3450 {
3451 	if (eth_p_mpls(type))
3452 		features &= skb->dev->mpls_features;
3453 
3454 	return features;
3455 }
3456 #else
3457 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3458 					   netdev_features_t features,
3459 					   __be16 type)
3460 {
3461 	return features;
3462 }
3463 #endif
3464 
3465 static netdev_features_t harmonize_features(struct sk_buff *skb,
3466 	netdev_features_t features)
3467 {
3468 	__be16 type;
3469 
3470 	type = skb_network_protocol(skb, NULL);
3471 	features = net_mpls_features(skb, features, type);
3472 
3473 	if (skb->ip_summed != CHECKSUM_NONE &&
3474 	    !can_checksum_protocol(features, type)) {
3475 		features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3476 	}
3477 	if (illegal_highdma(skb->dev, skb))
3478 		features &= ~NETIF_F_SG;
3479 
3480 	return features;
3481 }
3482 
3483 netdev_features_t passthru_features_check(struct sk_buff *skb,
3484 					  struct net_device *dev,
3485 					  netdev_features_t features)
3486 {
3487 	return features;
3488 }
3489 EXPORT_SYMBOL(passthru_features_check);
3490 
3491 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3492 					     struct net_device *dev,
3493 					     netdev_features_t features)
3494 {
3495 	return vlan_features_check(skb, features);
3496 }
3497 
3498 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3499 					    struct net_device *dev,
3500 					    netdev_features_t features)
3501 {
3502 	u16 gso_segs = skb_shinfo(skb)->gso_segs;
3503 
3504 	if (gso_segs > READ_ONCE(dev->gso_max_segs))
3505 		return features & ~NETIF_F_GSO_MASK;
3506 
3507 	if (unlikely(skb->len >= READ_ONCE(dev->gso_max_size)))
3508 		return features & ~NETIF_F_GSO_MASK;
3509 
3510 	if (!skb_shinfo(skb)->gso_type) {
3511 		skb_warn_bad_offload(skb);
3512 		return features & ~NETIF_F_GSO_MASK;
3513 	}
3514 
3515 	/* Support for GSO partial features requires software
3516 	 * intervention before we can actually process the packets
3517 	 * so we need to strip support for any partial features now
3518 	 * and we can pull them back in after we have partially
3519 	 * segmented the frame.
3520 	 */
3521 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3522 		features &= ~dev->gso_partial_features;
3523 
3524 	/* Make sure to clear the IPv4 ID mangling feature if the
3525 	 * IPv4 header has the potential to be fragmented.
3526 	 */
3527 	if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3528 		struct iphdr *iph = skb->encapsulation ?
3529 				    inner_ip_hdr(skb) : ip_hdr(skb);
3530 
3531 		if (!(iph->frag_off & htons(IP_DF)))
3532 			features &= ~NETIF_F_TSO_MANGLEID;
3533 	}
3534 
3535 	return features;
3536 }
3537 
3538 netdev_features_t netif_skb_features(struct sk_buff *skb)
3539 {
3540 	struct net_device *dev = skb->dev;
3541 	netdev_features_t features = dev->features;
3542 
3543 	if (skb_is_gso(skb))
3544 		features = gso_features_check(skb, dev, features);
3545 
3546 	/* If encapsulation offload request, verify we are testing
3547 	 * hardware encapsulation features instead of standard
3548 	 * features for the netdev
3549 	 */
3550 	if (skb->encapsulation)
3551 		features &= dev->hw_enc_features;
3552 
3553 	if (skb_vlan_tagged(skb))
3554 		features = netdev_intersect_features(features,
3555 						     dev->vlan_features |
3556 						     NETIF_F_HW_VLAN_CTAG_TX |
3557 						     NETIF_F_HW_VLAN_STAG_TX);
3558 
3559 	if (dev->netdev_ops->ndo_features_check)
3560 		features &= dev->netdev_ops->ndo_features_check(skb, dev,
3561 								features);
3562 	else
3563 		features &= dflt_features_check(skb, dev, features);
3564 
3565 	return harmonize_features(skb, features);
3566 }
3567 EXPORT_SYMBOL(netif_skb_features);
3568 
3569 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3570 		    struct netdev_queue *txq, bool more)
3571 {
3572 	unsigned int len;
3573 	int rc;
3574 
3575 	if (dev_nit_active(dev))
3576 		dev_queue_xmit_nit(skb, dev);
3577 
3578 	len = skb->len;
3579 	trace_net_dev_start_xmit(skb, dev);
3580 	rc = netdev_start_xmit(skb, dev, txq, more);
3581 	trace_net_dev_xmit(skb, rc, dev, len);
3582 
3583 	return rc;
3584 }
3585 
3586 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3587 				    struct netdev_queue *txq, int *ret)
3588 {
3589 	struct sk_buff *skb = first;
3590 	int rc = NETDEV_TX_OK;
3591 
3592 	while (skb) {
3593 		struct sk_buff *next = skb->next;
3594 
3595 		skb_mark_not_on_list(skb);
3596 		rc = xmit_one(skb, dev, txq, next != NULL);
3597 		if (unlikely(!dev_xmit_complete(rc))) {
3598 			skb->next = next;
3599 			goto out;
3600 		}
3601 
3602 		skb = next;
3603 		if (netif_tx_queue_stopped(txq) && skb) {
3604 			rc = NETDEV_TX_BUSY;
3605 			break;
3606 		}
3607 	}
3608 
3609 out:
3610 	*ret = rc;
3611 	return skb;
3612 }
3613 
3614 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3615 					  netdev_features_t features)
3616 {
3617 	if (skb_vlan_tag_present(skb) &&
3618 	    !vlan_hw_offload_capable(features, skb->vlan_proto))
3619 		skb = __vlan_hwaccel_push_inside(skb);
3620 	return skb;
3621 }
3622 
3623 int skb_csum_hwoffload_help(struct sk_buff *skb,
3624 			    const netdev_features_t features)
3625 {
3626 	if (unlikely(skb_csum_is_sctp(skb)))
3627 		return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3628 			skb_crc32c_csum_help(skb);
3629 
3630 	if (features & NETIF_F_HW_CSUM)
3631 		return 0;
3632 
3633 	if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3634 		switch (skb->csum_offset) {
3635 		case offsetof(struct tcphdr, check):
3636 		case offsetof(struct udphdr, check):
3637 			return 0;
3638 		}
3639 	}
3640 
3641 	return skb_checksum_help(skb);
3642 }
3643 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3644 
3645 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3646 {
3647 	netdev_features_t features;
3648 
3649 	features = netif_skb_features(skb);
3650 	skb = validate_xmit_vlan(skb, features);
3651 	if (unlikely(!skb))
3652 		goto out_null;
3653 
3654 	skb = sk_validate_xmit_skb(skb, dev);
3655 	if (unlikely(!skb))
3656 		goto out_null;
3657 
3658 	if (netif_needs_gso(skb, features)) {
3659 		struct sk_buff *segs;
3660 
3661 		segs = skb_gso_segment(skb, features);
3662 		if (IS_ERR(segs)) {
3663 			goto out_kfree_skb;
3664 		} else if (segs) {
3665 			consume_skb(skb);
3666 			skb = segs;
3667 		}
3668 	} else {
3669 		if (skb_needs_linearize(skb, features) &&
3670 		    __skb_linearize(skb))
3671 			goto out_kfree_skb;
3672 
3673 		/* If packet is not checksummed and device does not
3674 		 * support checksumming for this protocol, complete
3675 		 * checksumming here.
3676 		 */
3677 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
3678 			if (skb->encapsulation)
3679 				skb_set_inner_transport_header(skb,
3680 							       skb_checksum_start_offset(skb));
3681 			else
3682 				skb_set_transport_header(skb,
3683 							 skb_checksum_start_offset(skb));
3684 			if (skb_csum_hwoffload_help(skb, features))
3685 				goto out_kfree_skb;
3686 		}
3687 	}
3688 
3689 	skb = validate_xmit_xfrm(skb, features, again);
3690 
3691 	return skb;
3692 
3693 out_kfree_skb:
3694 	kfree_skb(skb);
3695 out_null:
3696 	dev_core_stats_tx_dropped_inc(dev);
3697 	return NULL;
3698 }
3699 
3700 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3701 {
3702 	struct sk_buff *next, *head = NULL, *tail;
3703 
3704 	for (; skb != NULL; skb = next) {
3705 		next = skb->next;
3706 		skb_mark_not_on_list(skb);
3707 
3708 		/* in case skb wont be segmented, point to itself */
3709 		skb->prev = skb;
3710 
3711 		skb = validate_xmit_skb(skb, dev, again);
3712 		if (!skb)
3713 			continue;
3714 
3715 		if (!head)
3716 			head = skb;
3717 		else
3718 			tail->next = skb;
3719 		/* If skb was segmented, skb->prev points to
3720 		 * the last segment. If not, it still contains skb.
3721 		 */
3722 		tail = skb->prev;
3723 	}
3724 	return head;
3725 }
3726 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3727 
3728 static void qdisc_pkt_len_init(struct sk_buff *skb)
3729 {
3730 	const struct skb_shared_info *shinfo = skb_shinfo(skb);
3731 
3732 	qdisc_skb_cb(skb)->pkt_len = skb->len;
3733 
3734 	/* To get more precise estimation of bytes sent on wire,
3735 	 * we add to pkt_len the headers size of all segments
3736 	 */
3737 	if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3738 		u16 gso_segs = shinfo->gso_segs;
3739 		unsigned int hdr_len;
3740 
3741 		/* mac layer + network layer */
3742 		hdr_len = skb_transport_offset(skb);
3743 
3744 		/* + transport layer */
3745 		if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3746 			const struct tcphdr *th;
3747 			struct tcphdr _tcphdr;
3748 
3749 			th = skb_header_pointer(skb, hdr_len,
3750 						sizeof(_tcphdr), &_tcphdr);
3751 			if (likely(th))
3752 				hdr_len += __tcp_hdrlen(th);
3753 		} else {
3754 			struct udphdr _udphdr;
3755 
3756 			if (skb_header_pointer(skb, hdr_len,
3757 					       sizeof(_udphdr), &_udphdr))
3758 				hdr_len += sizeof(struct udphdr);
3759 		}
3760 
3761 		if (shinfo->gso_type & SKB_GSO_DODGY)
3762 			gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3763 						shinfo->gso_size);
3764 
3765 		qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3766 	}
3767 }
3768 
3769 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3770 			     struct sk_buff **to_free,
3771 			     struct netdev_queue *txq)
3772 {
3773 	int rc;
3774 
3775 	rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3776 	if (rc == NET_XMIT_SUCCESS)
3777 		trace_qdisc_enqueue(q, txq, skb);
3778 	return rc;
3779 }
3780 
3781 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3782 				 struct net_device *dev,
3783 				 struct netdev_queue *txq)
3784 {
3785 	spinlock_t *root_lock = qdisc_lock(q);
3786 	struct sk_buff *to_free = NULL;
3787 	bool contended;
3788 	int rc;
3789 
3790 	qdisc_calculate_pkt_len(skb, q);
3791 
3792 	tcf_set_drop_reason(skb, SKB_DROP_REASON_QDISC_DROP);
3793 
3794 	if (q->flags & TCQ_F_NOLOCK) {
3795 		if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3796 		    qdisc_run_begin(q)) {
3797 			/* Retest nolock_qdisc_is_empty() within the protection
3798 			 * of q->seqlock to protect from racing with requeuing.
3799 			 */
3800 			if (unlikely(!nolock_qdisc_is_empty(q))) {
3801 				rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3802 				__qdisc_run(q);
3803 				qdisc_run_end(q);
3804 
3805 				goto no_lock_out;
3806 			}
3807 
3808 			qdisc_bstats_cpu_update(q, skb);
3809 			if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3810 			    !nolock_qdisc_is_empty(q))
3811 				__qdisc_run(q);
3812 
3813 			qdisc_run_end(q);
3814 			return NET_XMIT_SUCCESS;
3815 		}
3816 
3817 		rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3818 		qdisc_run(q);
3819 
3820 no_lock_out:
3821 		if (unlikely(to_free))
3822 			kfree_skb_list_reason(to_free,
3823 					      tcf_get_drop_reason(to_free));
3824 		return rc;
3825 	}
3826 
3827 	if (unlikely(READ_ONCE(q->owner) == smp_processor_id())) {
3828 		kfree_skb_reason(skb, SKB_DROP_REASON_TC_RECLASSIFY_LOOP);
3829 		return NET_XMIT_DROP;
3830 	}
3831 	/*
3832 	 * Heuristic to force contended enqueues to serialize on a
3833 	 * separate lock before trying to get qdisc main lock.
3834 	 * This permits qdisc->running owner to get the lock more
3835 	 * often and dequeue packets faster.
3836 	 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
3837 	 * and then other tasks will only enqueue packets. The packets will be
3838 	 * sent after the qdisc owner is scheduled again. To prevent this
3839 	 * scenario the task always serialize on the lock.
3840 	 */
3841 	contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
3842 	if (unlikely(contended))
3843 		spin_lock(&q->busylock);
3844 
3845 	spin_lock(root_lock);
3846 	if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3847 		__qdisc_drop(skb, &to_free);
3848 		rc = NET_XMIT_DROP;
3849 	} else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3850 		   qdisc_run_begin(q)) {
3851 		/*
3852 		 * This is a work-conserving queue; there are no old skbs
3853 		 * waiting to be sent out; and the qdisc is not running -
3854 		 * xmit the skb directly.
3855 		 */
3856 
3857 		qdisc_bstats_update(q, skb);
3858 
3859 		if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3860 			if (unlikely(contended)) {
3861 				spin_unlock(&q->busylock);
3862 				contended = false;
3863 			}
3864 			__qdisc_run(q);
3865 		}
3866 
3867 		qdisc_run_end(q);
3868 		rc = NET_XMIT_SUCCESS;
3869 	} else {
3870 		WRITE_ONCE(q->owner, smp_processor_id());
3871 		rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3872 		WRITE_ONCE(q->owner, -1);
3873 		if (qdisc_run_begin(q)) {
3874 			if (unlikely(contended)) {
3875 				spin_unlock(&q->busylock);
3876 				contended = false;
3877 			}
3878 			__qdisc_run(q);
3879 			qdisc_run_end(q);
3880 		}
3881 	}
3882 	spin_unlock(root_lock);
3883 	if (unlikely(to_free))
3884 		kfree_skb_list_reason(to_free,
3885 				      tcf_get_drop_reason(to_free));
3886 	if (unlikely(contended))
3887 		spin_unlock(&q->busylock);
3888 	return rc;
3889 }
3890 
3891 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3892 static void skb_update_prio(struct sk_buff *skb)
3893 {
3894 	const struct netprio_map *map;
3895 	const struct sock *sk;
3896 	unsigned int prioidx;
3897 
3898 	if (skb->priority)
3899 		return;
3900 	map = rcu_dereference_bh(skb->dev->priomap);
3901 	if (!map)
3902 		return;
3903 	sk = skb_to_full_sk(skb);
3904 	if (!sk)
3905 		return;
3906 
3907 	prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3908 
3909 	if (prioidx < map->priomap_len)
3910 		skb->priority = map->priomap[prioidx];
3911 }
3912 #else
3913 #define skb_update_prio(skb)
3914 #endif
3915 
3916 /**
3917  *	dev_loopback_xmit - loop back @skb
3918  *	@net: network namespace this loopback is happening in
3919  *	@sk:  sk needed to be a netfilter okfn
3920  *	@skb: buffer to transmit
3921  */
3922 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3923 {
3924 	skb_reset_mac_header(skb);
3925 	__skb_pull(skb, skb_network_offset(skb));
3926 	skb->pkt_type = PACKET_LOOPBACK;
3927 	if (skb->ip_summed == CHECKSUM_NONE)
3928 		skb->ip_summed = CHECKSUM_UNNECESSARY;
3929 	DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb));
3930 	skb_dst_force(skb);
3931 	netif_rx(skb);
3932 	return 0;
3933 }
3934 EXPORT_SYMBOL(dev_loopback_xmit);
3935 
3936 #ifdef CONFIG_NET_EGRESS
3937 static struct netdev_queue *
3938 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
3939 {
3940 	int qm = skb_get_queue_mapping(skb);
3941 
3942 	return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
3943 }
3944 
3945 #ifndef CONFIG_PREEMPT_RT
3946 static bool netdev_xmit_txqueue_skipped(void)
3947 {
3948 	return __this_cpu_read(softnet_data.xmit.skip_txqueue);
3949 }
3950 
3951 void netdev_xmit_skip_txqueue(bool skip)
3952 {
3953 	__this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
3954 }
3955 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
3956 
3957 #else
3958 static bool netdev_xmit_txqueue_skipped(void)
3959 {
3960 	return current->net_xmit.skip_txqueue;
3961 }
3962 
3963 void netdev_xmit_skip_txqueue(bool skip)
3964 {
3965 	current->net_xmit.skip_txqueue = skip;
3966 }
3967 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
3968 #endif
3969 #endif /* CONFIG_NET_EGRESS */
3970 
3971 #ifdef CONFIG_NET_XGRESS
3972 static int tc_run(struct tcx_entry *entry, struct sk_buff *skb,
3973 		  enum skb_drop_reason *drop_reason)
3974 {
3975 	int ret = TC_ACT_UNSPEC;
3976 #ifdef CONFIG_NET_CLS_ACT
3977 	struct mini_Qdisc *miniq = rcu_dereference_bh(entry->miniq);
3978 	struct tcf_result res;
3979 
3980 	if (!miniq)
3981 		return ret;
3982 
3983 	if (static_branch_unlikely(&tcf_bypass_check_needed_key)) {
3984 		if (tcf_block_bypass_sw(miniq->block))
3985 			return ret;
3986 	}
3987 
3988 	tc_skb_cb(skb)->mru = 0;
3989 	tc_skb_cb(skb)->post_ct = false;
3990 	tcf_set_drop_reason(skb, *drop_reason);
3991 
3992 	mini_qdisc_bstats_cpu_update(miniq, skb);
3993 	ret = tcf_classify(skb, miniq->block, miniq->filter_list, &res, false);
3994 	/* Only tcf related quirks below. */
3995 	switch (ret) {
3996 	case TC_ACT_SHOT:
3997 		*drop_reason = tcf_get_drop_reason(skb);
3998 		mini_qdisc_qstats_cpu_drop(miniq);
3999 		break;
4000 	case TC_ACT_OK:
4001 	case TC_ACT_RECLASSIFY:
4002 		skb->tc_index = TC_H_MIN(res.classid);
4003 		break;
4004 	}
4005 #endif /* CONFIG_NET_CLS_ACT */
4006 	return ret;
4007 }
4008 
4009 static DEFINE_STATIC_KEY_FALSE(tcx_needed_key);
4010 
4011 void tcx_inc(void)
4012 {
4013 	static_branch_inc(&tcx_needed_key);
4014 }
4015 
4016 void tcx_dec(void)
4017 {
4018 	static_branch_dec(&tcx_needed_key);
4019 }
4020 
4021 static __always_inline enum tcx_action_base
4022 tcx_run(const struct bpf_mprog_entry *entry, struct sk_buff *skb,
4023 	const bool needs_mac)
4024 {
4025 	const struct bpf_mprog_fp *fp;
4026 	const struct bpf_prog *prog;
4027 	int ret = TCX_NEXT;
4028 
4029 	if (needs_mac)
4030 		__skb_push(skb, skb->mac_len);
4031 	bpf_mprog_foreach_prog(entry, fp, prog) {
4032 		bpf_compute_data_pointers(skb);
4033 		ret = bpf_prog_run(prog, skb);
4034 		if (ret != TCX_NEXT)
4035 			break;
4036 	}
4037 	if (needs_mac)
4038 		__skb_pull(skb, skb->mac_len);
4039 	return tcx_action_code(skb, ret);
4040 }
4041 
4042 static __always_inline struct sk_buff *
4043 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4044 		   struct net_device *orig_dev, bool *another)
4045 {
4046 	struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress);
4047 	enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS;
4048 	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
4049 	int sch_ret;
4050 
4051 	if (!entry)
4052 		return skb;
4053 
4054 	bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
4055 	if (*pt_prev) {
4056 		*ret = deliver_skb(skb, *pt_prev, orig_dev);
4057 		*pt_prev = NULL;
4058 	}
4059 
4060 	qdisc_skb_cb(skb)->pkt_len = skb->len;
4061 	tcx_set_ingress(skb, true);
4062 
4063 	if (static_branch_unlikely(&tcx_needed_key)) {
4064 		sch_ret = tcx_run(entry, skb, true);
4065 		if (sch_ret != TC_ACT_UNSPEC)
4066 			goto ingress_verdict;
4067 	}
4068 	sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason);
4069 ingress_verdict:
4070 	switch (sch_ret) {
4071 	case TC_ACT_REDIRECT:
4072 		/* skb_mac_header check was done by BPF, so we can safely
4073 		 * push the L2 header back before redirecting to another
4074 		 * netdev.
4075 		 */
4076 		__skb_push(skb, skb->mac_len);
4077 		if (skb_do_redirect(skb) == -EAGAIN) {
4078 			__skb_pull(skb, skb->mac_len);
4079 			*another = true;
4080 			break;
4081 		}
4082 		*ret = NET_RX_SUCCESS;
4083 		bpf_net_ctx_clear(bpf_net_ctx);
4084 		return NULL;
4085 	case TC_ACT_SHOT:
4086 		kfree_skb_reason(skb, drop_reason);
4087 		*ret = NET_RX_DROP;
4088 		bpf_net_ctx_clear(bpf_net_ctx);
4089 		return NULL;
4090 	/* used by tc_run */
4091 	case TC_ACT_STOLEN:
4092 	case TC_ACT_QUEUED:
4093 	case TC_ACT_TRAP:
4094 		consume_skb(skb);
4095 		fallthrough;
4096 	case TC_ACT_CONSUMED:
4097 		*ret = NET_RX_SUCCESS;
4098 		bpf_net_ctx_clear(bpf_net_ctx);
4099 		return NULL;
4100 	}
4101 	bpf_net_ctx_clear(bpf_net_ctx);
4102 
4103 	return skb;
4104 }
4105 
4106 static __always_inline struct sk_buff *
4107 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
4108 {
4109 	struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress);
4110 	enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS;
4111 	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
4112 	int sch_ret;
4113 
4114 	if (!entry)
4115 		return skb;
4116 
4117 	bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
4118 
4119 	/* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was
4120 	 * already set by the caller.
4121 	 */
4122 	if (static_branch_unlikely(&tcx_needed_key)) {
4123 		sch_ret = tcx_run(entry, skb, false);
4124 		if (sch_ret != TC_ACT_UNSPEC)
4125 			goto egress_verdict;
4126 	}
4127 	sch_ret = tc_run(tcx_entry(entry), skb, &drop_reason);
4128 egress_verdict:
4129 	switch (sch_ret) {
4130 	case TC_ACT_REDIRECT:
4131 		/* No need to push/pop skb's mac_header here on egress! */
4132 		skb_do_redirect(skb);
4133 		*ret = NET_XMIT_SUCCESS;
4134 		bpf_net_ctx_clear(bpf_net_ctx);
4135 		return NULL;
4136 	case TC_ACT_SHOT:
4137 		kfree_skb_reason(skb, drop_reason);
4138 		*ret = NET_XMIT_DROP;
4139 		bpf_net_ctx_clear(bpf_net_ctx);
4140 		return NULL;
4141 	/* used by tc_run */
4142 	case TC_ACT_STOLEN:
4143 	case TC_ACT_QUEUED:
4144 	case TC_ACT_TRAP:
4145 		consume_skb(skb);
4146 		fallthrough;
4147 	case TC_ACT_CONSUMED:
4148 		*ret = NET_XMIT_SUCCESS;
4149 		bpf_net_ctx_clear(bpf_net_ctx);
4150 		return NULL;
4151 	}
4152 	bpf_net_ctx_clear(bpf_net_ctx);
4153 
4154 	return skb;
4155 }
4156 #else
4157 static __always_inline struct sk_buff *
4158 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4159 		   struct net_device *orig_dev, bool *another)
4160 {
4161 	return skb;
4162 }
4163 
4164 static __always_inline struct sk_buff *
4165 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
4166 {
4167 	return skb;
4168 }
4169 #endif /* CONFIG_NET_XGRESS */
4170 
4171 #ifdef CONFIG_XPS
4172 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
4173 			       struct xps_dev_maps *dev_maps, unsigned int tci)
4174 {
4175 	int tc = netdev_get_prio_tc_map(dev, skb->priority);
4176 	struct xps_map *map;
4177 	int queue_index = -1;
4178 
4179 	if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4180 		return queue_index;
4181 
4182 	tci *= dev_maps->num_tc;
4183 	tci += tc;
4184 
4185 	map = rcu_dereference(dev_maps->attr_map[tci]);
4186 	if (map) {
4187 		if (map->len == 1)
4188 			queue_index = map->queues[0];
4189 		else
4190 			queue_index = map->queues[reciprocal_scale(
4191 						skb_get_hash(skb), map->len)];
4192 		if (unlikely(queue_index >= dev->real_num_tx_queues))
4193 			queue_index = -1;
4194 	}
4195 	return queue_index;
4196 }
4197 #endif
4198 
4199 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4200 			 struct sk_buff *skb)
4201 {
4202 #ifdef CONFIG_XPS
4203 	struct xps_dev_maps *dev_maps;
4204 	struct sock *sk = skb->sk;
4205 	int queue_index = -1;
4206 
4207 	if (!static_key_false(&xps_needed))
4208 		return -1;
4209 
4210 	rcu_read_lock();
4211 	if (!static_key_false(&xps_rxqs_needed))
4212 		goto get_cpus_map;
4213 
4214 	dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4215 	if (dev_maps) {
4216 		int tci = sk_rx_queue_get(sk);
4217 
4218 		if (tci >= 0)
4219 			queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4220 							  tci);
4221 	}
4222 
4223 get_cpus_map:
4224 	if (queue_index < 0) {
4225 		dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4226 		if (dev_maps) {
4227 			unsigned int tci = skb->sender_cpu - 1;
4228 
4229 			queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4230 							  tci);
4231 		}
4232 	}
4233 	rcu_read_unlock();
4234 
4235 	return queue_index;
4236 #else
4237 	return -1;
4238 #endif
4239 }
4240 
4241 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4242 		     struct net_device *sb_dev)
4243 {
4244 	return 0;
4245 }
4246 EXPORT_SYMBOL(dev_pick_tx_zero);
4247 
4248 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4249 		       struct net_device *sb_dev)
4250 {
4251 	return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4252 }
4253 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4254 
4255 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4256 		     struct net_device *sb_dev)
4257 {
4258 	struct sock *sk = skb->sk;
4259 	int queue_index = sk_tx_queue_get(sk);
4260 
4261 	sb_dev = sb_dev ? : dev;
4262 
4263 	if (queue_index < 0 || skb->ooo_okay ||
4264 	    queue_index >= dev->real_num_tx_queues) {
4265 		int new_index = get_xps_queue(dev, sb_dev, skb);
4266 
4267 		if (new_index < 0)
4268 			new_index = skb_tx_hash(dev, sb_dev, skb);
4269 
4270 		if (queue_index != new_index && sk &&
4271 		    sk_fullsock(sk) &&
4272 		    rcu_access_pointer(sk->sk_dst_cache))
4273 			sk_tx_queue_set(sk, new_index);
4274 
4275 		queue_index = new_index;
4276 	}
4277 
4278 	return queue_index;
4279 }
4280 EXPORT_SYMBOL(netdev_pick_tx);
4281 
4282 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4283 					 struct sk_buff *skb,
4284 					 struct net_device *sb_dev)
4285 {
4286 	int queue_index = 0;
4287 
4288 #ifdef CONFIG_XPS
4289 	u32 sender_cpu = skb->sender_cpu - 1;
4290 
4291 	if (sender_cpu >= (u32)NR_CPUS)
4292 		skb->sender_cpu = raw_smp_processor_id() + 1;
4293 #endif
4294 
4295 	if (dev->real_num_tx_queues != 1) {
4296 		const struct net_device_ops *ops = dev->netdev_ops;
4297 
4298 		if (ops->ndo_select_queue)
4299 			queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4300 		else
4301 			queue_index = netdev_pick_tx(dev, skb, sb_dev);
4302 
4303 		queue_index = netdev_cap_txqueue(dev, queue_index);
4304 	}
4305 
4306 	skb_set_queue_mapping(skb, queue_index);
4307 	return netdev_get_tx_queue(dev, queue_index);
4308 }
4309 
4310 /**
4311  * __dev_queue_xmit() - transmit a buffer
4312  * @skb:	buffer to transmit
4313  * @sb_dev:	suboordinate device used for L2 forwarding offload
4314  *
4315  * Queue a buffer for transmission to a network device. The caller must
4316  * have set the device and priority and built the buffer before calling
4317  * this function. The function can be called from an interrupt.
4318  *
4319  * When calling this method, interrupts MUST be enabled. This is because
4320  * the BH enable code must have IRQs enabled so that it will not deadlock.
4321  *
4322  * Regardless of the return value, the skb is consumed, so it is currently
4323  * difficult to retry a send to this method. (You can bump the ref count
4324  * before sending to hold a reference for retry if you are careful.)
4325  *
4326  * Return:
4327  * * 0				- buffer successfully transmitted
4328  * * positive qdisc return code	- NET_XMIT_DROP etc.
4329  * * negative errno		- other errors
4330  */
4331 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4332 {
4333 	struct net_device *dev = skb->dev;
4334 	struct netdev_queue *txq = NULL;
4335 	struct Qdisc *q;
4336 	int rc = -ENOMEM;
4337 	bool again = false;
4338 
4339 	skb_reset_mac_header(skb);
4340 	skb_assert_len(skb);
4341 
4342 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4343 		__skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4344 
4345 	/* Disable soft irqs for various locks below. Also
4346 	 * stops preemption for RCU.
4347 	 */
4348 	rcu_read_lock_bh();
4349 
4350 	skb_update_prio(skb);
4351 
4352 	qdisc_pkt_len_init(skb);
4353 	tcx_set_ingress(skb, false);
4354 #ifdef CONFIG_NET_EGRESS
4355 	if (static_branch_unlikely(&egress_needed_key)) {
4356 		if (nf_hook_egress_active()) {
4357 			skb = nf_hook_egress(skb, &rc, dev);
4358 			if (!skb)
4359 				goto out;
4360 		}
4361 
4362 		netdev_xmit_skip_txqueue(false);
4363 
4364 		nf_skip_egress(skb, true);
4365 		skb = sch_handle_egress(skb, &rc, dev);
4366 		if (!skb)
4367 			goto out;
4368 		nf_skip_egress(skb, false);
4369 
4370 		if (netdev_xmit_txqueue_skipped())
4371 			txq = netdev_tx_queue_mapping(dev, skb);
4372 	}
4373 #endif
4374 	/* If device/qdisc don't need skb->dst, release it right now while
4375 	 * its hot in this cpu cache.
4376 	 */
4377 	if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4378 		skb_dst_drop(skb);
4379 	else
4380 		skb_dst_force(skb);
4381 
4382 	if (!txq)
4383 		txq = netdev_core_pick_tx(dev, skb, sb_dev);
4384 
4385 	q = rcu_dereference_bh(txq->qdisc);
4386 
4387 	trace_net_dev_queue(skb);
4388 	if (q->enqueue) {
4389 		rc = __dev_xmit_skb(skb, q, dev, txq);
4390 		goto out;
4391 	}
4392 
4393 	/* The device has no queue. Common case for software devices:
4394 	 * loopback, all the sorts of tunnels...
4395 
4396 	 * Really, it is unlikely that netif_tx_lock protection is necessary
4397 	 * here.  (f.e. loopback and IP tunnels are clean ignoring statistics
4398 	 * counters.)
4399 	 * However, it is possible, that they rely on protection
4400 	 * made by us here.
4401 
4402 	 * Check this and shot the lock. It is not prone from deadlocks.
4403 	 *Either shot noqueue qdisc, it is even simpler 8)
4404 	 */
4405 	if (dev->flags & IFF_UP) {
4406 		int cpu = smp_processor_id(); /* ok because BHs are off */
4407 
4408 		/* Other cpus might concurrently change txq->xmit_lock_owner
4409 		 * to -1 or to their cpu id, but not to our id.
4410 		 */
4411 		if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4412 			if (dev_xmit_recursion())
4413 				goto recursion_alert;
4414 
4415 			skb = validate_xmit_skb(skb, dev, &again);
4416 			if (!skb)
4417 				goto out;
4418 
4419 			HARD_TX_LOCK(dev, txq, cpu);
4420 
4421 			if (!netif_xmit_stopped(txq)) {
4422 				dev_xmit_recursion_inc();
4423 				skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4424 				dev_xmit_recursion_dec();
4425 				if (dev_xmit_complete(rc)) {
4426 					HARD_TX_UNLOCK(dev, txq);
4427 					goto out;
4428 				}
4429 			}
4430 			HARD_TX_UNLOCK(dev, txq);
4431 			net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4432 					     dev->name);
4433 		} else {
4434 			/* Recursion is detected! It is possible,
4435 			 * unfortunately
4436 			 */
4437 recursion_alert:
4438 			net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4439 					     dev->name);
4440 		}
4441 	}
4442 
4443 	rc = -ENETDOWN;
4444 	rcu_read_unlock_bh();
4445 
4446 	dev_core_stats_tx_dropped_inc(dev);
4447 	kfree_skb_list(skb);
4448 	return rc;
4449 out:
4450 	rcu_read_unlock_bh();
4451 	return rc;
4452 }
4453 EXPORT_SYMBOL(__dev_queue_xmit);
4454 
4455 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4456 {
4457 	struct net_device *dev = skb->dev;
4458 	struct sk_buff *orig_skb = skb;
4459 	struct netdev_queue *txq;
4460 	int ret = NETDEV_TX_BUSY;
4461 	bool again = false;
4462 
4463 	if (unlikely(!netif_running(dev) ||
4464 		     !netif_carrier_ok(dev)))
4465 		goto drop;
4466 
4467 	skb = validate_xmit_skb_list(skb, dev, &again);
4468 	if (skb != orig_skb)
4469 		goto drop;
4470 
4471 	skb_set_queue_mapping(skb, queue_id);
4472 	txq = skb_get_tx_queue(dev, skb);
4473 
4474 	local_bh_disable();
4475 
4476 	dev_xmit_recursion_inc();
4477 	HARD_TX_LOCK(dev, txq, smp_processor_id());
4478 	if (!netif_xmit_frozen_or_drv_stopped(txq))
4479 		ret = netdev_start_xmit(skb, dev, txq, false);
4480 	HARD_TX_UNLOCK(dev, txq);
4481 	dev_xmit_recursion_dec();
4482 
4483 	local_bh_enable();
4484 	return ret;
4485 drop:
4486 	dev_core_stats_tx_dropped_inc(dev);
4487 	kfree_skb_list(skb);
4488 	return NET_XMIT_DROP;
4489 }
4490 EXPORT_SYMBOL(__dev_direct_xmit);
4491 
4492 /*************************************************************************
4493  *			Receiver routines
4494  *************************************************************************/
4495 static DEFINE_PER_CPU(struct task_struct *, backlog_napi);
4496 
4497 int weight_p __read_mostly = 64;           /* old backlog weight */
4498 int dev_weight_rx_bias __read_mostly = 1;  /* bias for backlog weight */
4499 int dev_weight_tx_bias __read_mostly = 1;  /* bias for output_queue quota */
4500 
4501 /* Called with irq disabled */
4502 static inline void ____napi_schedule(struct softnet_data *sd,
4503 				     struct napi_struct *napi)
4504 {
4505 	struct task_struct *thread;
4506 
4507 	lockdep_assert_irqs_disabled();
4508 
4509 	if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4510 		/* Paired with smp_mb__before_atomic() in
4511 		 * napi_enable()/dev_set_threaded().
4512 		 * Use READ_ONCE() to guarantee a complete
4513 		 * read on napi->thread. Only call
4514 		 * wake_up_process() when it's not NULL.
4515 		 */
4516 		thread = READ_ONCE(napi->thread);
4517 		if (thread) {
4518 			if (use_backlog_threads() && thread == raw_cpu_read(backlog_napi))
4519 				goto use_local_napi;
4520 
4521 			set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4522 			wake_up_process(thread);
4523 			return;
4524 		}
4525 	}
4526 
4527 use_local_napi:
4528 	list_add_tail(&napi->poll_list, &sd->poll_list);
4529 	WRITE_ONCE(napi->list_owner, smp_processor_id());
4530 	/* If not called from net_rx_action()
4531 	 * we have to raise NET_RX_SOFTIRQ.
4532 	 */
4533 	if (!sd->in_net_rx_action)
4534 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4535 }
4536 
4537 #ifdef CONFIG_RPS
4538 
4539 struct static_key_false rps_needed __read_mostly;
4540 EXPORT_SYMBOL(rps_needed);
4541 struct static_key_false rfs_needed __read_mostly;
4542 EXPORT_SYMBOL(rfs_needed);
4543 
4544 static struct rps_dev_flow *
4545 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4546 	    struct rps_dev_flow *rflow, u16 next_cpu)
4547 {
4548 	if (next_cpu < nr_cpu_ids) {
4549 		u32 head;
4550 #ifdef CONFIG_RFS_ACCEL
4551 		struct netdev_rx_queue *rxqueue;
4552 		struct rps_dev_flow_table *flow_table;
4553 		struct rps_dev_flow *old_rflow;
4554 		u16 rxq_index;
4555 		u32 flow_id;
4556 		int rc;
4557 
4558 		/* Should we steer this flow to a different hardware queue? */
4559 		if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4560 		    !(dev->features & NETIF_F_NTUPLE))
4561 			goto out;
4562 		rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4563 		if (rxq_index == skb_get_rx_queue(skb))
4564 			goto out;
4565 
4566 		rxqueue = dev->_rx + rxq_index;
4567 		flow_table = rcu_dereference(rxqueue->rps_flow_table);
4568 		if (!flow_table)
4569 			goto out;
4570 		flow_id = skb_get_hash(skb) & flow_table->mask;
4571 		rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4572 							rxq_index, flow_id);
4573 		if (rc < 0)
4574 			goto out;
4575 		old_rflow = rflow;
4576 		rflow = &flow_table->flows[flow_id];
4577 		WRITE_ONCE(rflow->filter, rc);
4578 		if (old_rflow->filter == rc)
4579 			WRITE_ONCE(old_rflow->filter, RPS_NO_FILTER);
4580 	out:
4581 #endif
4582 		head = READ_ONCE(per_cpu(softnet_data, next_cpu).input_queue_head);
4583 		rps_input_queue_tail_save(&rflow->last_qtail, head);
4584 	}
4585 
4586 	WRITE_ONCE(rflow->cpu, next_cpu);
4587 	return rflow;
4588 }
4589 
4590 /*
4591  * get_rps_cpu is called from netif_receive_skb and returns the target
4592  * CPU from the RPS map of the receiving queue for a given skb.
4593  * rcu_read_lock must be held on entry.
4594  */
4595 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4596 		       struct rps_dev_flow **rflowp)
4597 {
4598 	const struct rps_sock_flow_table *sock_flow_table;
4599 	struct netdev_rx_queue *rxqueue = dev->_rx;
4600 	struct rps_dev_flow_table *flow_table;
4601 	struct rps_map *map;
4602 	int cpu = -1;
4603 	u32 tcpu;
4604 	u32 hash;
4605 
4606 	if (skb_rx_queue_recorded(skb)) {
4607 		u16 index = skb_get_rx_queue(skb);
4608 
4609 		if (unlikely(index >= dev->real_num_rx_queues)) {
4610 			WARN_ONCE(dev->real_num_rx_queues > 1,
4611 				  "%s received packet on queue %u, but number "
4612 				  "of RX queues is %u\n",
4613 				  dev->name, index, dev->real_num_rx_queues);
4614 			goto done;
4615 		}
4616 		rxqueue += index;
4617 	}
4618 
4619 	/* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4620 
4621 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
4622 	map = rcu_dereference(rxqueue->rps_map);
4623 	if (!flow_table && !map)
4624 		goto done;
4625 
4626 	skb_reset_network_header(skb);
4627 	hash = skb_get_hash(skb);
4628 	if (!hash)
4629 		goto done;
4630 
4631 	sock_flow_table = rcu_dereference(net_hotdata.rps_sock_flow_table);
4632 	if (flow_table && sock_flow_table) {
4633 		struct rps_dev_flow *rflow;
4634 		u32 next_cpu;
4635 		u32 ident;
4636 
4637 		/* First check into global flow table if there is a match.
4638 		 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow().
4639 		 */
4640 		ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]);
4641 		if ((ident ^ hash) & ~net_hotdata.rps_cpu_mask)
4642 			goto try_rps;
4643 
4644 		next_cpu = ident & net_hotdata.rps_cpu_mask;
4645 
4646 		/* OK, now we know there is a match,
4647 		 * we can look at the local (per receive queue) flow table
4648 		 */
4649 		rflow = &flow_table->flows[hash & flow_table->mask];
4650 		tcpu = rflow->cpu;
4651 
4652 		/*
4653 		 * If the desired CPU (where last recvmsg was done) is
4654 		 * different from current CPU (one in the rx-queue flow
4655 		 * table entry), switch if one of the following holds:
4656 		 *   - Current CPU is unset (>= nr_cpu_ids).
4657 		 *   - Current CPU is offline.
4658 		 *   - The current CPU's queue tail has advanced beyond the
4659 		 *     last packet that was enqueued using this table entry.
4660 		 *     This guarantees that all previous packets for the flow
4661 		 *     have been dequeued, thus preserving in order delivery.
4662 		 */
4663 		if (unlikely(tcpu != next_cpu) &&
4664 		    (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4665 		     ((int)(READ_ONCE(per_cpu(softnet_data, tcpu).input_queue_head) -
4666 		      rflow->last_qtail)) >= 0)) {
4667 			tcpu = next_cpu;
4668 			rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4669 		}
4670 
4671 		if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4672 			*rflowp = rflow;
4673 			cpu = tcpu;
4674 			goto done;
4675 		}
4676 	}
4677 
4678 try_rps:
4679 
4680 	if (map) {
4681 		tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4682 		if (cpu_online(tcpu)) {
4683 			cpu = tcpu;
4684 			goto done;
4685 		}
4686 	}
4687 
4688 done:
4689 	return cpu;
4690 }
4691 
4692 #ifdef CONFIG_RFS_ACCEL
4693 
4694 /**
4695  * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4696  * @dev: Device on which the filter was set
4697  * @rxq_index: RX queue index
4698  * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4699  * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4700  *
4701  * Drivers that implement ndo_rx_flow_steer() should periodically call
4702  * this function for each installed filter and remove the filters for
4703  * which it returns %true.
4704  */
4705 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4706 			 u32 flow_id, u16 filter_id)
4707 {
4708 	struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4709 	struct rps_dev_flow_table *flow_table;
4710 	struct rps_dev_flow *rflow;
4711 	bool expire = true;
4712 	unsigned int cpu;
4713 
4714 	rcu_read_lock();
4715 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
4716 	if (flow_table && flow_id <= flow_table->mask) {
4717 		rflow = &flow_table->flows[flow_id];
4718 		cpu = READ_ONCE(rflow->cpu);
4719 		if (READ_ONCE(rflow->filter) == filter_id && cpu < nr_cpu_ids &&
4720 		    ((int)(READ_ONCE(per_cpu(softnet_data, cpu).input_queue_head) -
4721 			   READ_ONCE(rflow->last_qtail)) <
4722 		     (int)(10 * flow_table->mask)))
4723 			expire = false;
4724 	}
4725 	rcu_read_unlock();
4726 	return expire;
4727 }
4728 EXPORT_SYMBOL(rps_may_expire_flow);
4729 
4730 #endif /* CONFIG_RFS_ACCEL */
4731 
4732 /* Called from hardirq (IPI) context */
4733 static void rps_trigger_softirq(void *data)
4734 {
4735 	struct softnet_data *sd = data;
4736 
4737 	____napi_schedule(sd, &sd->backlog);
4738 	sd->received_rps++;
4739 }
4740 
4741 #endif /* CONFIG_RPS */
4742 
4743 /* Called from hardirq (IPI) context */
4744 static void trigger_rx_softirq(void *data)
4745 {
4746 	struct softnet_data *sd = data;
4747 
4748 	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4749 	smp_store_release(&sd->defer_ipi_scheduled, 0);
4750 }
4751 
4752 /*
4753  * After we queued a packet into sd->input_pkt_queue,
4754  * we need to make sure this queue is serviced soon.
4755  *
4756  * - If this is another cpu queue, link it to our rps_ipi_list,
4757  *   and make sure we will process rps_ipi_list from net_rx_action().
4758  *
4759  * - If this is our own queue, NAPI schedule our backlog.
4760  *   Note that this also raises NET_RX_SOFTIRQ.
4761  */
4762 static void napi_schedule_rps(struct softnet_data *sd)
4763 {
4764 	struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4765 
4766 #ifdef CONFIG_RPS
4767 	if (sd != mysd) {
4768 		if (use_backlog_threads()) {
4769 			__napi_schedule_irqoff(&sd->backlog);
4770 			return;
4771 		}
4772 
4773 		sd->rps_ipi_next = mysd->rps_ipi_list;
4774 		mysd->rps_ipi_list = sd;
4775 
4776 		/* If not called from net_rx_action() or napi_threaded_poll()
4777 		 * we have to raise NET_RX_SOFTIRQ.
4778 		 */
4779 		if (!mysd->in_net_rx_action && !mysd->in_napi_threaded_poll)
4780 			__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4781 		return;
4782 	}
4783 #endif /* CONFIG_RPS */
4784 	__napi_schedule_irqoff(&mysd->backlog);
4785 }
4786 
4787 void kick_defer_list_purge(struct softnet_data *sd, unsigned int cpu)
4788 {
4789 	unsigned long flags;
4790 
4791 	if (use_backlog_threads()) {
4792 		backlog_lock_irq_save(sd, &flags);
4793 
4794 		if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
4795 			__napi_schedule_irqoff(&sd->backlog);
4796 
4797 		backlog_unlock_irq_restore(sd, &flags);
4798 
4799 	} else if (!cmpxchg(&sd->defer_ipi_scheduled, 0, 1)) {
4800 		smp_call_function_single_async(cpu, &sd->defer_csd);
4801 	}
4802 }
4803 
4804 #ifdef CONFIG_NET_FLOW_LIMIT
4805 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4806 #endif
4807 
4808 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4809 {
4810 #ifdef CONFIG_NET_FLOW_LIMIT
4811 	struct sd_flow_limit *fl;
4812 	struct softnet_data *sd;
4813 	unsigned int old_flow, new_flow;
4814 
4815 	if (qlen < (READ_ONCE(net_hotdata.max_backlog) >> 1))
4816 		return false;
4817 
4818 	sd = this_cpu_ptr(&softnet_data);
4819 
4820 	rcu_read_lock();
4821 	fl = rcu_dereference(sd->flow_limit);
4822 	if (fl) {
4823 		new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4824 		old_flow = fl->history[fl->history_head];
4825 		fl->history[fl->history_head] = new_flow;
4826 
4827 		fl->history_head++;
4828 		fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4829 
4830 		if (likely(fl->buckets[old_flow]))
4831 			fl->buckets[old_flow]--;
4832 
4833 		if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4834 			fl->count++;
4835 			rcu_read_unlock();
4836 			return true;
4837 		}
4838 	}
4839 	rcu_read_unlock();
4840 #endif
4841 	return false;
4842 }
4843 
4844 /*
4845  * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4846  * queue (may be a remote CPU queue).
4847  */
4848 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4849 			      unsigned int *qtail)
4850 {
4851 	enum skb_drop_reason reason;
4852 	struct softnet_data *sd;
4853 	unsigned long flags;
4854 	unsigned int qlen;
4855 	int max_backlog;
4856 	u32 tail;
4857 
4858 	reason = SKB_DROP_REASON_DEV_READY;
4859 	if (!netif_running(skb->dev))
4860 		goto bad_dev;
4861 
4862 	reason = SKB_DROP_REASON_CPU_BACKLOG;
4863 	sd = &per_cpu(softnet_data, cpu);
4864 
4865 	qlen = skb_queue_len_lockless(&sd->input_pkt_queue);
4866 	max_backlog = READ_ONCE(net_hotdata.max_backlog);
4867 	if (unlikely(qlen > max_backlog))
4868 		goto cpu_backlog_drop;
4869 	backlog_lock_irq_save(sd, &flags);
4870 	qlen = skb_queue_len(&sd->input_pkt_queue);
4871 	if (qlen <= max_backlog && !skb_flow_limit(skb, qlen)) {
4872 		if (!qlen) {
4873 			/* Schedule NAPI for backlog device. We can use
4874 			 * non atomic operation as we own the queue lock.
4875 			 */
4876 			if (!__test_and_set_bit(NAPI_STATE_SCHED,
4877 						&sd->backlog.state))
4878 				napi_schedule_rps(sd);
4879 		}
4880 		__skb_queue_tail(&sd->input_pkt_queue, skb);
4881 		tail = rps_input_queue_tail_incr(sd);
4882 		backlog_unlock_irq_restore(sd, &flags);
4883 
4884 		/* save the tail outside of the critical section */
4885 		rps_input_queue_tail_save(qtail, tail);
4886 		return NET_RX_SUCCESS;
4887 	}
4888 
4889 	backlog_unlock_irq_restore(sd, &flags);
4890 
4891 cpu_backlog_drop:
4892 	atomic_inc(&sd->dropped);
4893 bad_dev:
4894 	dev_core_stats_rx_dropped_inc(skb->dev);
4895 	kfree_skb_reason(skb, reason);
4896 	return NET_RX_DROP;
4897 }
4898 
4899 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4900 {
4901 	struct net_device *dev = skb->dev;
4902 	struct netdev_rx_queue *rxqueue;
4903 
4904 	rxqueue = dev->_rx;
4905 
4906 	if (skb_rx_queue_recorded(skb)) {
4907 		u16 index = skb_get_rx_queue(skb);
4908 
4909 		if (unlikely(index >= dev->real_num_rx_queues)) {
4910 			WARN_ONCE(dev->real_num_rx_queues > 1,
4911 				  "%s received packet on queue %u, but number "
4912 				  "of RX queues is %u\n",
4913 				  dev->name, index, dev->real_num_rx_queues);
4914 
4915 			return rxqueue; /* Return first rxqueue */
4916 		}
4917 		rxqueue += index;
4918 	}
4919 	return rxqueue;
4920 }
4921 
4922 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4923 			     struct bpf_prog *xdp_prog)
4924 {
4925 	void *orig_data, *orig_data_end, *hard_start;
4926 	struct netdev_rx_queue *rxqueue;
4927 	bool orig_bcast, orig_host;
4928 	u32 mac_len, frame_sz;
4929 	__be16 orig_eth_type;
4930 	struct ethhdr *eth;
4931 	u32 metalen, act;
4932 	int off;
4933 
4934 	/* The XDP program wants to see the packet starting at the MAC
4935 	 * header.
4936 	 */
4937 	mac_len = skb->data - skb_mac_header(skb);
4938 	hard_start = skb->data - skb_headroom(skb);
4939 
4940 	/* SKB "head" area always have tailroom for skb_shared_info */
4941 	frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4942 	frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4943 
4944 	rxqueue = netif_get_rxqueue(skb);
4945 	xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4946 	xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4947 			 skb_headlen(skb) + mac_len, true);
4948 	if (skb_is_nonlinear(skb)) {
4949 		skb_shinfo(skb)->xdp_frags_size = skb->data_len;
4950 		xdp_buff_set_frags_flag(xdp);
4951 	} else {
4952 		xdp_buff_clear_frags_flag(xdp);
4953 	}
4954 
4955 	orig_data_end = xdp->data_end;
4956 	orig_data = xdp->data;
4957 	eth = (struct ethhdr *)xdp->data;
4958 	orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4959 	orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4960 	orig_eth_type = eth->h_proto;
4961 
4962 	act = bpf_prog_run_xdp(xdp_prog, xdp);
4963 
4964 	/* check if bpf_xdp_adjust_head was used */
4965 	off = xdp->data - orig_data;
4966 	if (off) {
4967 		if (off > 0)
4968 			__skb_pull(skb, off);
4969 		else if (off < 0)
4970 			__skb_push(skb, -off);
4971 
4972 		skb->mac_header += off;
4973 		skb_reset_network_header(skb);
4974 	}
4975 
4976 	/* check if bpf_xdp_adjust_tail was used */
4977 	off = xdp->data_end - orig_data_end;
4978 	if (off != 0) {
4979 		skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4980 		skb->len += off; /* positive on grow, negative on shrink */
4981 	}
4982 
4983 	/* XDP frag metadata (e.g. nr_frags) are updated in eBPF helpers
4984 	 * (e.g. bpf_xdp_adjust_tail), we need to update data_len here.
4985 	 */
4986 	if (xdp_buff_has_frags(xdp))
4987 		skb->data_len = skb_shinfo(skb)->xdp_frags_size;
4988 	else
4989 		skb->data_len = 0;
4990 
4991 	/* check if XDP changed eth hdr such SKB needs update */
4992 	eth = (struct ethhdr *)xdp->data;
4993 	if ((orig_eth_type != eth->h_proto) ||
4994 	    (orig_host != ether_addr_equal_64bits(eth->h_dest,
4995 						  skb->dev->dev_addr)) ||
4996 	    (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4997 		__skb_push(skb, ETH_HLEN);
4998 		skb->pkt_type = PACKET_HOST;
4999 		skb->protocol = eth_type_trans(skb, skb->dev);
5000 	}
5001 
5002 	/* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
5003 	 * before calling us again on redirect path. We do not call do_redirect
5004 	 * as we leave that up to the caller.
5005 	 *
5006 	 * Caller is responsible for managing lifetime of skb (i.e. calling
5007 	 * kfree_skb in response to actions it cannot handle/XDP_DROP).
5008 	 */
5009 	switch (act) {
5010 	case XDP_REDIRECT:
5011 	case XDP_TX:
5012 		__skb_push(skb, mac_len);
5013 		break;
5014 	case XDP_PASS:
5015 		metalen = xdp->data - xdp->data_meta;
5016 		if (metalen)
5017 			skb_metadata_set(skb, metalen);
5018 		break;
5019 	}
5020 
5021 	return act;
5022 }
5023 
5024 static int
5025 netif_skb_check_for_xdp(struct sk_buff **pskb, struct bpf_prog *prog)
5026 {
5027 	struct sk_buff *skb = *pskb;
5028 	int err, hroom, troom;
5029 
5030 	if (!skb_cow_data_for_xdp(this_cpu_read(system_page_pool), pskb, prog))
5031 		return 0;
5032 
5033 	/* In case we have to go down the path and also linearize,
5034 	 * then lets do the pskb_expand_head() work just once here.
5035 	 */
5036 	hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
5037 	troom = skb->tail + skb->data_len - skb->end;
5038 	err = pskb_expand_head(skb,
5039 			       hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
5040 			       troom > 0 ? troom + 128 : 0, GFP_ATOMIC);
5041 	if (err)
5042 		return err;
5043 
5044 	return skb_linearize(skb);
5045 }
5046 
5047 static u32 netif_receive_generic_xdp(struct sk_buff **pskb,
5048 				     struct xdp_buff *xdp,
5049 				     struct bpf_prog *xdp_prog)
5050 {
5051 	struct sk_buff *skb = *pskb;
5052 	u32 mac_len, act = XDP_DROP;
5053 
5054 	/* Reinjected packets coming from act_mirred or similar should
5055 	 * not get XDP generic processing.
5056 	 */
5057 	if (skb_is_redirected(skb))
5058 		return XDP_PASS;
5059 
5060 	/* XDP packets must have sufficient headroom of XDP_PACKET_HEADROOM
5061 	 * bytes. This is the guarantee that also native XDP provides,
5062 	 * thus we need to do it here as well.
5063 	 */
5064 	mac_len = skb->data - skb_mac_header(skb);
5065 	__skb_push(skb, mac_len);
5066 
5067 	if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
5068 	    skb_headroom(skb) < XDP_PACKET_HEADROOM) {
5069 		if (netif_skb_check_for_xdp(pskb, xdp_prog))
5070 			goto do_drop;
5071 	}
5072 
5073 	__skb_pull(*pskb, mac_len);
5074 
5075 	act = bpf_prog_run_generic_xdp(*pskb, xdp, xdp_prog);
5076 	switch (act) {
5077 	case XDP_REDIRECT:
5078 	case XDP_TX:
5079 	case XDP_PASS:
5080 		break;
5081 	default:
5082 		bpf_warn_invalid_xdp_action((*pskb)->dev, xdp_prog, act);
5083 		fallthrough;
5084 	case XDP_ABORTED:
5085 		trace_xdp_exception((*pskb)->dev, xdp_prog, act);
5086 		fallthrough;
5087 	case XDP_DROP:
5088 	do_drop:
5089 		kfree_skb(*pskb);
5090 		break;
5091 	}
5092 
5093 	return act;
5094 }
5095 
5096 /* When doing generic XDP we have to bypass the qdisc layer and the
5097  * network taps in order to match in-driver-XDP behavior. This also means
5098  * that XDP packets are able to starve other packets going through a qdisc,
5099  * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX
5100  * queues, so they do not have this starvation issue.
5101  */
5102 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
5103 {
5104 	struct net_device *dev = skb->dev;
5105 	struct netdev_queue *txq;
5106 	bool free_skb = true;
5107 	int cpu, rc;
5108 
5109 	txq = netdev_core_pick_tx(dev, skb, NULL);
5110 	cpu = smp_processor_id();
5111 	HARD_TX_LOCK(dev, txq, cpu);
5112 	if (!netif_xmit_frozen_or_drv_stopped(txq)) {
5113 		rc = netdev_start_xmit(skb, dev, txq, 0);
5114 		if (dev_xmit_complete(rc))
5115 			free_skb = false;
5116 	}
5117 	HARD_TX_UNLOCK(dev, txq);
5118 	if (free_skb) {
5119 		trace_xdp_exception(dev, xdp_prog, XDP_TX);
5120 		dev_core_stats_tx_dropped_inc(dev);
5121 		kfree_skb(skb);
5122 	}
5123 }
5124 
5125 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
5126 
5127 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff **pskb)
5128 {
5129 	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
5130 
5131 	if (xdp_prog) {
5132 		struct xdp_buff xdp;
5133 		u32 act;
5134 		int err;
5135 
5136 		bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
5137 		act = netif_receive_generic_xdp(pskb, &xdp, xdp_prog);
5138 		if (act != XDP_PASS) {
5139 			switch (act) {
5140 			case XDP_REDIRECT:
5141 				err = xdp_do_generic_redirect((*pskb)->dev, *pskb,
5142 							      &xdp, xdp_prog);
5143 				if (err)
5144 					goto out_redir;
5145 				break;
5146 			case XDP_TX:
5147 				generic_xdp_tx(*pskb, xdp_prog);
5148 				break;
5149 			}
5150 			bpf_net_ctx_clear(bpf_net_ctx);
5151 			return XDP_DROP;
5152 		}
5153 	}
5154 	return XDP_PASS;
5155 out_redir:
5156 	bpf_net_ctx_clear(bpf_net_ctx);
5157 	kfree_skb_reason(*pskb, SKB_DROP_REASON_XDP);
5158 	return XDP_DROP;
5159 }
5160 EXPORT_SYMBOL_GPL(do_xdp_generic);
5161 
5162 static int netif_rx_internal(struct sk_buff *skb)
5163 {
5164 	int ret;
5165 
5166 	net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb);
5167 
5168 	trace_netif_rx(skb);
5169 
5170 #ifdef CONFIG_RPS
5171 	if (static_branch_unlikely(&rps_needed)) {
5172 		struct rps_dev_flow voidflow, *rflow = &voidflow;
5173 		int cpu;
5174 
5175 		rcu_read_lock();
5176 
5177 		cpu = get_rps_cpu(skb->dev, skb, &rflow);
5178 		if (cpu < 0)
5179 			cpu = smp_processor_id();
5180 
5181 		ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5182 
5183 		rcu_read_unlock();
5184 	} else
5185 #endif
5186 	{
5187 		unsigned int qtail;
5188 
5189 		ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
5190 	}
5191 	return ret;
5192 }
5193 
5194 /**
5195  *	__netif_rx	-	Slightly optimized version of netif_rx
5196  *	@skb: buffer to post
5197  *
5198  *	This behaves as netif_rx except that it does not disable bottom halves.
5199  *	As a result this function may only be invoked from the interrupt context
5200  *	(either hard or soft interrupt).
5201  */
5202 int __netif_rx(struct sk_buff *skb)
5203 {
5204 	int ret;
5205 
5206 	lockdep_assert_once(hardirq_count() | softirq_count());
5207 
5208 	trace_netif_rx_entry(skb);
5209 	ret = netif_rx_internal(skb);
5210 	trace_netif_rx_exit(ret);
5211 	return ret;
5212 }
5213 EXPORT_SYMBOL(__netif_rx);
5214 
5215 /**
5216  *	netif_rx	-	post buffer to the network code
5217  *	@skb: buffer to post
5218  *
5219  *	This function receives a packet from a device driver and queues it for
5220  *	the upper (protocol) levels to process via the backlog NAPI device. It
5221  *	always succeeds. The buffer may be dropped during processing for
5222  *	congestion control or by the protocol layers.
5223  *	The network buffer is passed via the backlog NAPI device. Modern NIC
5224  *	driver should use NAPI and GRO.
5225  *	This function can used from interrupt and from process context. The
5226  *	caller from process context must not disable interrupts before invoking
5227  *	this function.
5228  *
5229  *	return values:
5230  *	NET_RX_SUCCESS	(no congestion)
5231  *	NET_RX_DROP     (packet was dropped)
5232  *
5233  */
5234 int netif_rx(struct sk_buff *skb)
5235 {
5236 	bool need_bh_off = !(hardirq_count() | softirq_count());
5237 	int ret;
5238 
5239 	if (need_bh_off)
5240 		local_bh_disable();
5241 	trace_netif_rx_entry(skb);
5242 	ret = netif_rx_internal(skb);
5243 	trace_netif_rx_exit(ret);
5244 	if (need_bh_off)
5245 		local_bh_enable();
5246 	return ret;
5247 }
5248 EXPORT_SYMBOL(netif_rx);
5249 
5250 static __latent_entropy void net_tx_action(struct softirq_action *h)
5251 {
5252 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5253 
5254 	if (sd->completion_queue) {
5255 		struct sk_buff *clist;
5256 
5257 		local_irq_disable();
5258 		clist = sd->completion_queue;
5259 		sd->completion_queue = NULL;
5260 		local_irq_enable();
5261 
5262 		while (clist) {
5263 			struct sk_buff *skb = clist;
5264 
5265 			clist = clist->next;
5266 
5267 			WARN_ON(refcount_read(&skb->users));
5268 			if (likely(get_kfree_skb_cb(skb)->reason == SKB_CONSUMED))
5269 				trace_consume_skb(skb, net_tx_action);
5270 			else
5271 				trace_kfree_skb(skb, net_tx_action,
5272 						get_kfree_skb_cb(skb)->reason, NULL);
5273 
5274 			if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5275 				__kfree_skb(skb);
5276 			else
5277 				__napi_kfree_skb(skb,
5278 						 get_kfree_skb_cb(skb)->reason);
5279 		}
5280 	}
5281 
5282 	if (sd->output_queue) {
5283 		struct Qdisc *head;
5284 
5285 		local_irq_disable();
5286 		head = sd->output_queue;
5287 		sd->output_queue = NULL;
5288 		sd->output_queue_tailp = &sd->output_queue;
5289 		local_irq_enable();
5290 
5291 		rcu_read_lock();
5292 
5293 		while (head) {
5294 			struct Qdisc *q = head;
5295 			spinlock_t *root_lock = NULL;
5296 
5297 			head = head->next_sched;
5298 
5299 			/* We need to make sure head->next_sched is read
5300 			 * before clearing __QDISC_STATE_SCHED
5301 			 */
5302 			smp_mb__before_atomic();
5303 
5304 			if (!(q->flags & TCQ_F_NOLOCK)) {
5305 				root_lock = qdisc_lock(q);
5306 				spin_lock(root_lock);
5307 			} else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5308 						     &q->state))) {
5309 				/* There is a synchronize_net() between
5310 				 * STATE_DEACTIVATED flag being set and
5311 				 * qdisc_reset()/some_qdisc_is_busy() in
5312 				 * dev_deactivate(), so we can safely bail out
5313 				 * early here to avoid data race between
5314 				 * qdisc_deactivate() and some_qdisc_is_busy()
5315 				 * for lockless qdisc.
5316 				 */
5317 				clear_bit(__QDISC_STATE_SCHED, &q->state);
5318 				continue;
5319 			}
5320 
5321 			clear_bit(__QDISC_STATE_SCHED, &q->state);
5322 			qdisc_run(q);
5323 			if (root_lock)
5324 				spin_unlock(root_lock);
5325 		}
5326 
5327 		rcu_read_unlock();
5328 	}
5329 
5330 	xfrm_dev_backlog(sd);
5331 }
5332 
5333 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5334 /* This hook is defined here for ATM LANE */
5335 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5336 			     unsigned char *addr) __read_mostly;
5337 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5338 #endif
5339 
5340 /**
5341  *	netdev_is_rx_handler_busy - check if receive handler is registered
5342  *	@dev: device to check
5343  *
5344  *	Check if a receive handler is already registered for a given device.
5345  *	Return true if there one.
5346  *
5347  *	The caller must hold the rtnl_mutex.
5348  */
5349 bool netdev_is_rx_handler_busy(struct net_device *dev)
5350 {
5351 	ASSERT_RTNL();
5352 	return dev && rtnl_dereference(dev->rx_handler);
5353 }
5354 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5355 
5356 /**
5357  *	netdev_rx_handler_register - register receive handler
5358  *	@dev: device to register a handler for
5359  *	@rx_handler: receive handler to register
5360  *	@rx_handler_data: data pointer that is used by rx handler
5361  *
5362  *	Register a receive handler for a device. This handler will then be
5363  *	called from __netif_receive_skb. A negative errno code is returned
5364  *	on a failure.
5365  *
5366  *	The caller must hold the rtnl_mutex.
5367  *
5368  *	For a general description of rx_handler, see enum rx_handler_result.
5369  */
5370 int netdev_rx_handler_register(struct net_device *dev,
5371 			       rx_handler_func_t *rx_handler,
5372 			       void *rx_handler_data)
5373 {
5374 	if (netdev_is_rx_handler_busy(dev))
5375 		return -EBUSY;
5376 
5377 	if (dev->priv_flags & IFF_NO_RX_HANDLER)
5378 		return -EINVAL;
5379 
5380 	/* Note: rx_handler_data must be set before rx_handler */
5381 	rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5382 	rcu_assign_pointer(dev->rx_handler, rx_handler);
5383 
5384 	return 0;
5385 }
5386 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5387 
5388 /**
5389  *	netdev_rx_handler_unregister - unregister receive handler
5390  *	@dev: device to unregister a handler from
5391  *
5392  *	Unregister a receive handler from a device.
5393  *
5394  *	The caller must hold the rtnl_mutex.
5395  */
5396 void netdev_rx_handler_unregister(struct net_device *dev)
5397 {
5398 
5399 	ASSERT_RTNL();
5400 	RCU_INIT_POINTER(dev->rx_handler, NULL);
5401 	/* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5402 	 * section has a guarantee to see a non NULL rx_handler_data
5403 	 * as well.
5404 	 */
5405 	synchronize_net();
5406 	RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5407 }
5408 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5409 
5410 /*
5411  * Limit the use of PFMEMALLOC reserves to those protocols that implement
5412  * the special handling of PFMEMALLOC skbs.
5413  */
5414 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5415 {
5416 	switch (skb->protocol) {
5417 	case htons(ETH_P_ARP):
5418 	case htons(ETH_P_IP):
5419 	case htons(ETH_P_IPV6):
5420 	case htons(ETH_P_8021Q):
5421 	case htons(ETH_P_8021AD):
5422 		return true;
5423 	default:
5424 		return false;
5425 	}
5426 }
5427 
5428 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5429 			     int *ret, struct net_device *orig_dev)
5430 {
5431 	if (nf_hook_ingress_active(skb)) {
5432 		int ingress_retval;
5433 
5434 		if (*pt_prev) {
5435 			*ret = deliver_skb(skb, *pt_prev, orig_dev);
5436 			*pt_prev = NULL;
5437 		}
5438 
5439 		rcu_read_lock();
5440 		ingress_retval = nf_hook_ingress(skb);
5441 		rcu_read_unlock();
5442 		return ingress_retval;
5443 	}
5444 	return 0;
5445 }
5446 
5447 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5448 				    struct packet_type **ppt_prev)
5449 {
5450 	struct packet_type *ptype, *pt_prev;
5451 	rx_handler_func_t *rx_handler;
5452 	struct sk_buff *skb = *pskb;
5453 	struct net_device *orig_dev;
5454 	bool deliver_exact = false;
5455 	int ret = NET_RX_DROP;
5456 	__be16 type;
5457 
5458 	net_timestamp_check(!READ_ONCE(net_hotdata.tstamp_prequeue), skb);
5459 
5460 	trace_netif_receive_skb(skb);
5461 
5462 	orig_dev = skb->dev;
5463 
5464 	skb_reset_network_header(skb);
5465 	if (!skb_transport_header_was_set(skb))
5466 		skb_reset_transport_header(skb);
5467 	skb_reset_mac_len(skb);
5468 
5469 	pt_prev = NULL;
5470 
5471 another_round:
5472 	skb->skb_iif = skb->dev->ifindex;
5473 
5474 	__this_cpu_inc(softnet_data.processed);
5475 
5476 	if (static_branch_unlikely(&generic_xdp_needed_key)) {
5477 		int ret2;
5478 
5479 		migrate_disable();
5480 		ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog),
5481 				      &skb);
5482 		migrate_enable();
5483 
5484 		if (ret2 != XDP_PASS) {
5485 			ret = NET_RX_DROP;
5486 			goto out;
5487 		}
5488 	}
5489 
5490 	if (eth_type_vlan(skb->protocol)) {
5491 		skb = skb_vlan_untag(skb);
5492 		if (unlikely(!skb))
5493 			goto out;
5494 	}
5495 
5496 	if (skb_skip_tc_classify(skb))
5497 		goto skip_classify;
5498 
5499 	if (pfmemalloc)
5500 		goto skip_taps;
5501 
5502 	list_for_each_entry_rcu(ptype, &net_hotdata.ptype_all, list) {
5503 		if (pt_prev)
5504 			ret = deliver_skb(skb, pt_prev, orig_dev);
5505 		pt_prev = ptype;
5506 	}
5507 
5508 	list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5509 		if (pt_prev)
5510 			ret = deliver_skb(skb, pt_prev, orig_dev);
5511 		pt_prev = ptype;
5512 	}
5513 
5514 skip_taps:
5515 #ifdef CONFIG_NET_INGRESS
5516 	if (static_branch_unlikely(&ingress_needed_key)) {
5517 		bool another = false;
5518 
5519 		nf_skip_egress(skb, true);
5520 		skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5521 					 &another);
5522 		if (another)
5523 			goto another_round;
5524 		if (!skb)
5525 			goto out;
5526 
5527 		nf_skip_egress(skb, false);
5528 		if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5529 			goto out;
5530 	}
5531 #endif
5532 	skb_reset_redirect(skb);
5533 skip_classify:
5534 	if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5535 		goto drop;
5536 
5537 	if (skb_vlan_tag_present(skb)) {
5538 		if (pt_prev) {
5539 			ret = deliver_skb(skb, pt_prev, orig_dev);
5540 			pt_prev = NULL;
5541 		}
5542 		if (vlan_do_receive(&skb))
5543 			goto another_round;
5544 		else if (unlikely(!skb))
5545 			goto out;
5546 	}
5547 
5548 	rx_handler = rcu_dereference(skb->dev->rx_handler);
5549 	if (rx_handler) {
5550 		if (pt_prev) {
5551 			ret = deliver_skb(skb, pt_prev, orig_dev);
5552 			pt_prev = NULL;
5553 		}
5554 		switch (rx_handler(&skb)) {
5555 		case RX_HANDLER_CONSUMED:
5556 			ret = NET_RX_SUCCESS;
5557 			goto out;
5558 		case RX_HANDLER_ANOTHER:
5559 			goto another_round;
5560 		case RX_HANDLER_EXACT:
5561 			deliver_exact = true;
5562 			break;
5563 		case RX_HANDLER_PASS:
5564 			break;
5565 		default:
5566 			BUG();
5567 		}
5568 	}
5569 
5570 	if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5571 check_vlan_id:
5572 		if (skb_vlan_tag_get_id(skb)) {
5573 			/* Vlan id is non 0 and vlan_do_receive() above couldn't
5574 			 * find vlan device.
5575 			 */
5576 			skb->pkt_type = PACKET_OTHERHOST;
5577 		} else if (eth_type_vlan(skb->protocol)) {
5578 			/* Outer header is 802.1P with vlan 0, inner header is
5579 			 * 802.1Q or 802.1AD and vlan_do_receive() above could
5580 			 * not find vlan dev for vlan id 0.
5581 			 */
5582 			__vlan_hwaccel_clear_tag(skb);
5583 			skb = skb_vlan_untag(skb);
5584 			if (unlikely(!skb))
5585 				goto out;
5586 			if (vlan_do_receive(&skb))
5587 				/* After stripping off 802.1P header with vlan 0
5588 				 * vlan dev is found for inner header.
5589 				 */
5590 				goto another_round;
5591 			else if (unlikely(!skb))
5592 				goto out;
5593 			else
5594 				/* We have stripped outer 802.1P vlan 0 header.
5595 				 * But could not find vlan dev.
5596 				 * check again for vlan id to set OTHERHOST.
5597 				 */
5598 				goto check_vlan_id;
5599 		}
5600 		/* Note: we might in the future use prio bits
5601 		 * and set skb->priority like in vlan_do_receive()
5602 		 * For the time being, just ignore Priority Code Point
5603 		 */
5604 		__vlan_hwaccel_clear_tag(skb);
5605 	}
5606 
5607 	type = skb->protocol;
5608 
5609 	/* deliver only exact match when indicated */
5610 	if (likely(!deliver_exact)) {
5611 		deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5612 				       &ptype_base[ntohs(type) &
5613 						   PTYPE_HASH_MASK]);
5614 	}
5615 
5616 	deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5617 			       &orig_dev->ptype_specific);
5618 
5619 	if (unlikely(skb->dev != orig_dev)) {
5620 		deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5621 				       &skb->dev->ptype_specific);
5622 	}
5623 
5624 	if (pt_prev) {
5625 		if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5626 			goto drop;
5627 		*ppt_prev = pt_prev;
5628 	} else {
5629 drop:
5630 		if (!deliver_exact)
5631 			dev_core_stats_rx_dropped_inc(skb->dev);
5632 		else
5633 			dev_core_stats_rx_nohandler_inc(skb->dev);
5634 		kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO);
5635 		/* Jamal, now you will not able to escape explaining
5636 		 * me how you were going to use this. :-)
5637 		 */
5638 		ret = NET_RX_DROP;
5639 	}
5640 
5641 out:
5642 	/* The invariant here is that if *ppt_prev is not NULL
5643 	 * then skb should also be non-NULL.
5644 	 *
5645 	 * Apparently *ppt_prev assignment above holds this invariant due to
5646 	 * skb dereferencing near it.
5647 	 */
5648 	*pskb = skb;
5649 	return ret;
5650 }
5651 
5652 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5653 {
5654 	struct net_device *orig_dev = skb->dev;
5655 	struct packet_type *pt_prev = NULL;
5656 	int ret;
5657 
5658 	ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5659 	if (pt_prev)
5660 		ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5661 					 skb->dev, pt_prev, orig_dev);
5662 	return ret;
5663 }
5664 
5665 /**
5666  *	netif_receive_skb_core - special purpose version of netif_receive_skb
5667  *	@skb: buffer to process
5668  *
5669  *	More direct receive version of netif_receive_skb().  It should
5670  *	only be used by callers that have a need to skip RPS and Generic XDP.
5671  *	Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5672  *
5673  *	This function may only be called from softirq context and interrupts
5674  *	should be enabled.
5675  *
5676  *	Return values (usually ignored):
5677  *	NET_RX_SUCCESS: no congestion
5678  *	NET_RX_DROP: packet was dropped
5679  */
5680 int netif_receive_skb_core(struct sk_buff *skb)
5681 {
5682 	int ret;
5683 
5684 	rcu_read_lock();
5685 	ret = __netif_receive_skb_one_core(skb, false);
5686 	rcu_read_unlock();
5687 
5688 	return ret;
5689 }
5690 EXPORT_SYMBOL(netif_receive_skb_core);
5691 
5692 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5693 						  struct packet_type *pt_prev,
5694 						  struct net_device *orig_dev)
5695 {
5696 	struct sk_buff *skb, *next;
5697 
5698 	if (!pt_prev)
5699 		return;
5700 	if (list_empty(head))
5701 		return;
5702 	if (pt_prev->list_func != NULL)
5703 		INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5704 				   ip_list_rcv, head, pt_prev, orig_dev);
5705 	else
5706 		list_for_each_entry_safe(skb, next, head, list) {
5707 			skb_list_del_init(skb);
5708 			pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5709 		}
5710 }
5711 
5712 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5713 {
5714 	/* Fast-path assumptions:
5715 	 * - There is no RX handler.
5716 	 * - Only one packet_type matches.
5717 	 * If either of these fails, we will end up doing some per-packet
5718 	 * processing in-line, then handling the 'last ptype' for the whole
5719 	 * sublist.  This can't cause out-of-order delivery to any single ptype,
5720 	 * because the 'last ptype' must be constant across the sublist, and all
5721 	 * other ptypes are handled per-packet.
5722 	 */
5723 	/* Current (common) ptype of sublist */
5724 	struct packet_type *pt_curr = NULL;
5725 	/* Current (common) orig_dev of sublist */
5726 	struct net_device *od_curr = NULL;
5727 	struct list_head sublist;
5728 	struct sk_buff *skb, *next;
5729 
5730 	INIT_LIST_HEAD(&sublist);
5731 	list_for_each_entry_safe(skb, next, head, list) {
5732 		struct net_device *orig_dev = skb->dev;
5733 		struct packet_type *pt_prev = NULL;
5734 
5735 		skb_list_del_init(skb);
5736 		__netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5737 		if (!pt_prev)
5738 			continue;
5739 		if (pt_curr != pt_prev || od_curr != orig_dev) {
5740 			/* dispatch old sublist */
5741 			__netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5742 			/* start new sublist */
5743 			INIT_LIST_HEAD(&sublist);
5744 			pt_curr = pt_prev;
5745 			od_curr = orig_dev;
5746 		}
5747 		list_add_tail(&skb->list, &sublist);
5748 	}
5749 
5750 	/* dispatch final sublist */
5751 	__netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5752 }
5753 
5754 static int __netif_receive_skb(struct sk_buff *skb)
5755 {
5756 	int ret;
5757 
5758 	if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5759 		unsigned int noreclaim_flag;
5760 
5761 		/*
5762 		 * PFMEMALLOC skbs are special, they should
5763 		 * - be delivered to SOCK_MEMALLOC sockets only
5764 		 * - stay away from userspace
5765 		 * - have bounded memory usage
5766 		 *
5767 		 * Use PF_MEMALLOC as this saves us from propagating the allocation
5768 		 * context down to all allocation sites.
5769 		 */
5770 		noreclaim_flag = memalloc_noreclaim_save();
5771 		ret = __netif_receive_skb_one_core(skb, true);
5772 		memalloc_noreclaim_restore(noreclaim_flag);
5773 	} else
5774 		ret = __netif_receive_skb_one_core(skb, false);
5775 
5776 	return ret;
5777 }
5778 
5779 static void __netif_receive_skb_list(struct list_head *head)
5780 {
5781 	unsigned long noreclaim_flag = 0;
5782 	struct sk_buff *skb, *next;
5783 	bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5784 
5785 	list_for_each_entry_safe(skb, next, head, list) {
5786 		if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5787 			struct list_head sublist;
5788 
5789 			/* Handle the previous sublist */
5790 			list_cut_before(&sublist, head, &skb->list);
5791 			if (!list_empty(&sublist))
5792 				__netif_receive_skb_list_core(&sublist, pfmemalloc);
5793 			pfmemalloc = !pfmemalloc;
5794 			/* See comments in __netif_receive_skb */
5795 			if (pfmemalloc)
5796 				noreclaim_flag = memalloc_noreclaim_save();
5797 			else
5798 				memalloc_noreclaim_restore(noreclaim_flag);
5799 		}
5800 	}
5801 	/* Handle the remaining sublist */
5802 	if (!list_empty(head))
5803 		__netif_receive_skb_list_core(head, pfmemalloc);
5804 	/* Restore pflags */
5805 	if (pfmemalloc)
5806 		memalloc_noreclaim_restore(noreclaim_flag);
5807 }
5808 
5809 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5810 {
5811 	struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5812 	struct bpf_prog *new = xdp->prog;
5813 	int ret = 0;
5814 
5815 	switch (xdp->command) {
5816 	case XDP_SETUP_PROG:
5817 		rcu_assign_pointer(dev->xdp_prog, new);
5818 		if (old)
5819 			bpf_prog_put(old);
5820 
5821 		if (old && !new) {
5822 			static_branch_dec(&generic_xdp_needed_key);
5823 		} else if (new && !old) {
5824 			static_branch_inc(&generic_xdp_needed_key);
5825 			dev_disable_lro(dev);
5826 			dev_disable_gro_hw(dev);
5827 		}
5828 		break;
5829 
5830 	default:
5831 		ret = -EINVAL;
5832 		break;
5833 	}
5834 
5835 	return ret;
5836 }
5837 
5838 static int netif_receive_skb_internal(struct sk_buff *skb)
5839 {
5840 	int ret;
5841 
5842 	net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue), skb);
5843 
5844 	if (skb_defer_rx_timestamp(skb))
5845 		return NET_RX_SUCCESS;
5846 
5847 	rcu_read_lock();
5848 #ifdef CONFIG_RPS
5849 	if (static_branch_unlikely(&rps_needed)) {
5850 		struct rps_dev_flow voidflow, *rflow = &voidflow;
5851 		int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5852 
5853 		if (cpu >= 0) {
5854 			ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5855 			rcu_read_unlock();
5856 			return ret;
5857 		}
5858 	}
5859 #endif
5860 	ret = __netif_receive_skb(skb);
5861 	rcu_read_unlock();
5862 	return ret;
5863 }
5864 
5865 void netif_receive_skb_list_internal(struct list_head *head)
5866 {
5867 	struct sk_buff *skb, *next;
5868 	struct list_head sublist;
5869 
5870 	INIT_LIST_HEAD(&sublist);
5871 	list_for_each_entry_safe(skb, next, head, list) {
5872 		net_timestamp_check(READ_ONCE(net_hotdata.tstamp_prequeue),
5873 				    skb);
5874 		skb_list_del_init(skb);
5875 		if (!skb_defer_rx_timestamp(skb))
5876 			list_add_tail(&skb->list, &sublist);
5877 	}
5878 	list_splice_init(&sublist, head);
5879 
5880 	rcu_read_lock();
5881 #ifdef CONFIG_RPS
5882 	if (static_branch_unlikely(&rps_needed)) {
5883 		list_for_each_entry_safe(skb, next, head, list) {
5884 			struct rps_dev_flow voidflow, *rflow = &voidflow;
5885 			int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5886 
5887 			if (cpu >= 0) {
5888 				/* Will be handled, remove from list */
5889 				skb_list_del_init(skb);
5890 				enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5891 			}
5892 		}
5893 	}
5894 #endif
5895 	__netif_receive_skb_list(head);
5896 	rcu_read_unlock();
5897 }
5898 
5899 /**
5900  *	netif_receive_skb - process receive buffer from network
5901  *	@skb: buffer to process
5902  *
5903  *	netif_receive_skb() is the main receive data processing function.
5904  *	It always succeeds. The buffer may be dropped during processing
5905  *	for congestion control or by the protocol layers.
5906  *
5907  *	This function may only be called from softirq context and interrupts
5908  *	should be enabled.
5909  *
5910  *	Return values (usually ignored):
5911  *	NET_RX_SUCCESS: no congestion
5912  *	NET_RX_DROP: packet was dropped
5913  */
5914 int netif_receive_skb(struct sk_buff *skb)
5915 {
5916 	int ret;
5917 
5918 	trace_netif_receive_skb_entry(skb);
5919 
5920 	ret = netif_receive_skb_internal(skb);
5921 	trace_netif_receive_skb_exit(ret);
5922 
5923 	return ret;
5924 }
5925 EXPORT_SYMBOL(netif_receive_skb);
5926 
5927 /**
5928  *	netif_receive_skb_list - process many receive buffers from network
5929  *	@head: list of skbs to process.
5930  *
5931  *	Since return value of netif_receive_skb() is normally ignored, and
5932  *	wouldn't be meaningful for a list, this function returns void.
5933  *
5934  *	This function may only be called from softirq context and interrupts
5935  *	should be enabled.
5936  */
5937 void netif_receive_skb_list(struct list_head *head)
5938 {
5939 	struct sk_buff *skb;
5940 
5941 	if (list_empty(head))
5942 		return;
5943 	if (trace_netif_receive_skb_list_entry_enabled()) {
5944 		list_for_each_entry(skb, head, list)
5945 			trace_netif_receive_skb_list_entry(skb);
5946 	}
5947 	netif_receive_skb_list_internal(head);
5948 	trace_netif_receive_skb_list_exit(0);
5949 }
5950 EXPORT_SYMBOL(netif_receive_skb_list);
5951 
5952 static DEFINE_PER_CPU(struct work_struct, flush_works);
5953 
5954 /* Network device is going away, flush any packets still pending */
5955 static void flush_backlog(struct work_struct *work)
5956 {
5957 	struct sk_buff *skb, *tmp;
5958 	struct softnet_data *sd;
5959 
5960 	local_bh_disable();
5961 	sd = this_cpu_ptr(&softnet_data);
5962 
5963 	backlog_lock_irq_disable(sd);
5964 	skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5965 		if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5966 			__skb_unlink(skb, &sd->input_pkt_queue);
5967 			dev_kfree_skb_irq(skb);
5968 			rps_input_queue_head_incr(sd);
5969 		}
5970 	}
5971 	backlog_unlock_irq_enable(sd);
5972 
5973 	local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
5974 	skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5975 		if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5976 			__skb_unlink(skb, &sd->process_queue);
5977 			kfree_skb(skb);
5978 			rps_input_queue_head_incr(sd);
5979 		}
5980 	}
5981 	local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
5982 	local_bh_enable();
5983 }
5984 
5985 static bool flush_required(int cpu)
5986 {
5987 #if IS_ENABLED(CONFIG_RPS)
5988 	struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5989 	bool do_flush;
5990 
5991 	backlog_lock_irq_disable(sd);
5992 
5993 	/* as insertion into process_queue happens with the rps lock held,
5994 	 * process_queue access may race only with dequeue
5995 	 */
5996 	do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5997 		   !skb_queue_empty_lockless(&sd->process_queue);
5998 	backlog_unlock_irq_enable(sd);
5999 
6000 	return do_flush;
6001 #endif
6002 	/* without RPS we can't safely check input_pkt_queue: during a
6003 	 * concurrent remote skb_queue_splice() we can detect as empty both
6004 	 * input_pkt_queue and process_queue even if the latter could end-up
6005 	 * containing a lot of packets.
6006 	 */
6007 	return true;
6008 }
6009 
6010 static void flush_all_backlogs(void)
6011 {
6012 	static cpumask_t flush_cpus;
6013 	unsigned int cpu;
6014 
6015 	/* since we are under rtnl lock protection we can use static data
6016 	 * for the cpumask and avoid allocating on stack the possibly
6017 	 * large mask
6018 	 */
6019 	ASSERT_RTNL();
6020 
6021 	cpus_read_lock();
6022 
6023 	cpumask_clear(&flush_cpus);
6024 	for_each_online_cpu(cpu) {
6025 		if (flush_required(cpu)) {
6026 			queue_work_on(cpu, system_highpri_wq,
6027 				      per_cpu_ptr(&flush_works, cpu));
6028 			cpumask_set_cpu(cpu, &flush_cpus);
6029 		}
6030 	}
6031 
6032 	/* we can have in flight packet[s] on the cpus we are not flushing,
6033 	 * synchronize_net() in unregister_netdevice_many() will take care of
6034 	 * them
6035 	 */
6036 	for_each_cpu(cpu, &flush_cpus)
6037 		flush_work(per_cpu_ptr(&flush_works, cpu));
6038 
6039 	cpus_read_unlock();
6040 }
6041 
6042 static void net_rps_send_ipi(struct softnet_data *remsd)
6043 {
6044 #ifdef CONFIG_RPS
6045 	while (remsd) {
6046 		struct softnet_data *next = remsd->rps_ipi_next;
6047 
6048 		if (cpu_online(remsd->cpu))
6049 			smp_call_function_single_async(remsd->cpu, &remsd->csd);
6050 		remsd = next;
6051 	}
6052 #endif
6053 }
6054 
6055 /*
6056  * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6057  * Note: called with local irq disabled, but exits with local irq enabled.
6058  */
6059 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6060 {
6061 #ifdef CONFIG_RPS
6062 	struct softnet_data *remsd = sd->rps_ipi_list;
6063 
6064 	if (!use_backlog_threads() && remsd) {
6065 		sd->rps_ipi_list = NULL;
6066 
6067 		local_irq_enable();
6068 
6069 		/* Send pending IPI's to kick RPS processing on remote cpus. */
6070 		net_rps_send_ipi(remsd);
6071 	} else
6072 #endif
6073 		local_irq_enable();
6074 }
6075 
6076 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6077 {
6078 #ifdef CONFIG_RPS
6079 	return !use_backlog_threads() && sd->rps_ipi_list;
6080 #else
6081 	return false;
6082 #endif
6083 }
6084 
6085 static int process_backlog(struct napi_struct *napi, int quota)
6086 {
6087 	struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6088 	bool again = true;
6089 	int work = 0;
6090 
6091 	/* Check if we have pending ipi, its better to send them now,
6092 	 * not waiting net_rx_action() end.
6093 	 */
6094 	if (sd_has_rps_ipi_waiting(sd)) {
6095 		local_irq_disable();
6096 		net_rps_action_and_irq_enable(sd);
6097 	}
6098 
6099 	napi->weight = READ_ONCE(net_hotdata.dev_rx_weight);
6100 	while (again) {
6101 		struct sk_buff *skb;
6102 
6103 		local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
6104 		while ((skb = __skb_dequeue(&sd->process_queue))) {
6105 			local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
6106 			rcu_read_lock();
6107 			__netif_receive_skb(skb);
6108 			rcu_read_unlock();
6109 			if (++work >= quota) {
6110 				rps_input_queue_head_add(sd, work);
6111 				return work;
6112 			}
6113 
6114 			local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
6115 		}
6116 		local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
6117 
6118 		backlog_lock_irq_disable(sd);
6119 		if (skb_queue_empty(&sd->input_pkt_queue)) {
6120 			/*
6121 			 * Inline a custom version of __napi_complete().
6122 			 * only current cpu owns and manipulates this napi,
6123 			 * and NAPI_STATE_SCHED is the only possible flag set
6124 			 * on backlog.
6125 			 * We can use a plain write instead of clear_bit(),
6126 			 * and we dont need an smp_mb() memory barrier.
6127 			 */
6128 			napi->state &= NAPIF_STATE_THREADED;
6129 			again = false;
6130 		} else {
6131 			local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
6132 			skb_queue_splice_tail_init(&sd->input_pkt_queue,
6133 						   &sd->process_queue);
6134 			local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
6135 		}
6136 		backlog_unlock_irq_enable(sd);
6137 	}
6138 
6139 	if (work)
6140 		rps_input_queue_head_add(sd, work);
6141 	return work;
6142 }
6143 
6144 /**
6145  * __napi_schedule - schedule for receive
6146  * @n: entry to schedule
6147  *
6148  * The entry's receive function will be scheduled to run.
6149  * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6150  */
6151 void __napi_schedule(struct napi_struct *n)
6152 {
6153 	unsigned long flags;
6154 
6155 	local_irq_save(flags);
6156 	____napi_schedule(this_cpu_ptr(&softnet_data), n);
6157 	local_irq_restore(flags);
6158 }
6159 EXPORT_SYMBOL(__napi_schedule);
6160 
6161 /**
6162  *	napi_schedule_prep - check if napi can be scheduled
6163  *	@n: napi context
6164  *
6165  * Test if NAPI routine is already running, and if not mark
6166  * it as running.  This is used as a condition variable to
6167  * insure only one NAPI poll instance runs.  We also make
6168  * sure there is no pending NAPI disable.
6169  */
6170 bool napi_schedule_prep(struct napi_struct *n)
6171 {
6172 	unsigned long new, val = READ_ONCE(n->state);
6173 
6174 	do {
6175 		if (unlikely(val & NAPIF_STATE_DISABLE))
6176 			return false;
6177 		new = val | NAPIF_STATE_SCHED;
6178 
6179 		/* Sets STATE_MISSED bit if STATE_SCHED was already set
6180 		 * This was suggested by Alexander Duyck, as compiler
6181 		 * emits better code than :
6182 		 * if (val & NAPIF_STATE_SCHED)
6183 		 *     new |= NAPIF_STATE_MISSED;
6184 		 */
6185 		new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6186 						   NAPIF_STATE_MISSED;
6187 	} while (!try_cmpxchg(&n->state, &val, new));
6188 
6189 	return !(val & NAPIF_STATE_SCHED);
6190 }
6191 EXPORT_SYMBOL(napi_schedule_prep);
6192 
6193 /**
6194  * __napi_schedule_irqoff - schedule for receive
6195  * @n: entry to schedule
6196  *
6197  * Variant of __napi_schedule() assuming hard irqs are masked.
6198  *
6199  * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6200  * because the interrupt disabled assumption might not be true
6201  * due to force-threaded interrupts and spinlock substitution.
6202  */
6203 void __napi_schedule_irqoff(struct napi_struct *n)
6204 {
6205 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6206 		____napi_schedule(this_cpu_ptr(&softnet_data), n);
6207 	else
6208 		__napi_schedule(n);
6209 }
6210 EXPORT_SYMBOL(__napi_schedule_irqoff);
6211 
6212 bool napi_complete_done(struct napi_struct *n, int work_done)
6213 {
6214 	unsigned long flags, val, new, timeout = 0;
6215 	bool ret = true;
6216 
6217 	/*
6218 	 * 1) Don't let napi dequeue from the cpu poll list
6219 	 *    just in case its running on a different cpu.
6220 	 * 2) If we are busy polling, do nothing here, we have
6221 	 *    the guarantee we will be called later.
6222 	 */
6223 	if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6224 				 NAPIF_STATE_IN_BUSY_POLL)))
6225 		return false;
6226 
6227 	if (work_done) {
6228 		if (n->gro_bitmask)
6229 			timeout = READ_ONCE(n->dev->gro_flush_timeout);
6230 		n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6231 	}
6232 	if (n->defer_hard_irqs_count > 0) {
6233 		n->defer_hard_irqs_count--;
6234 		timeout = READ_ONCE(n->dev->gro_flush_timeout);
6235 		if (timeout)
6236 			ret = false;
6237 	}
6238 	if (n->gro_bitmask) {
6239 		/* When the NAPI instance uses a timeout and keeps postponing
6240 		 * it, we need to bound somehow the time packets are kept in
6241 		 * the GRO layer
6242 		 */
6243 		napi_gro_flush(n, !!timeout);
6244 	}
6245 
6246 	gro_normal_list(n);
6247 
6248 	if (unlikely(!list_empty(&n->poll_list))) {
6249 		/* If n->poll_list is not empty, we need to mask irqs */
6250 		local_irq_save(flags);
6251 		list_del_init(&n->poll_list);
6252 		local_irq_restore(flags);
6253 	}
6254 	WRITE_ONCE(n->list_owner, -1);
6255 
6256 	val = READ_ONCE(n->state);
6257 	do {
6258 		WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6259 
6260 		new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6261 			      NAPIF_STATE_SCHED_THREADED |
6262 			      NAPIF_STATE_PREFER_BUSY_POLL);
6263 
6264 		/* If STATE_MISSED was set, leave STATE_SCHED set,
6265 		 * because we will call napi->poll() one more time.
6266 		 * This C code was suggested by Alexander Duyck to help gcc.
6267 		 */
6268 		new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6269 						    NAPIF_STATE_SCHED;
6270 	} while (!try_cmpxchg(&n->state, &val, new));
6271 
6272 	if (unlikely(val & NAPIF_STATE_MISSED)) {
6273 		__napi_schedule(n);
6274 		return false;
6275 	}
6276 
6277 	if (timeout)
6278 		hrtimer_start(&n->timer, ns_to_ktime(timeout),
6279 			      HRTIMER_MODE_REL_PINNED);
6280 	return ret;
6281 }
6282 EXPORT_SYMBOL(napi_complete_done);
6283 
6284 /* must be called under rcu_read_lock(), as we dont take a reference */
6285 struct napi_struct *napi_by_id(unsigned int napi_id)
6286 {
6287 	unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6288 	struct napi_struct *napi;
6289 
6290 	hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6291 		if (napi->napi_id == napi_id)
6292 			return napi;
6293 
6294 	return NULL;
6295 }
6296 
6297 static void skb_defer_free_flush(struct softnet_data *sd)
6298 {
6299 	struct sk_buff *skb, *next;
6300 
6301 	/* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
6302 	if (!READ_ONCE(sd->defer_list))
6303 		return;
6304 
6305 	spin_lock(&sd->defer_lock);
6306 	skb = sd->defer_list;
6307 	sd->defer_list = NULL;
6308 	sd->defer_count = 0;
6309 	spin_unlock(&sd->defer_lock);
6310 
6311 	while (skb != NULL) {
6312 		next = skb->next;
6313 		napi_consume_skb(skb, 1);
6314 		skb = next;
6315 	}
6316 }
6317 
6318 #if defined(CONFIG_NET_RX_BUSY_POLL)
6319 
6320 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6321 {
6322 	if (!skip_schedule) {
6323 		gro_normal_list(napi);
6324 		__napi_schedule(napi);
6325 		return;
6326 	}
6327 
6328 	if (napi->gro_bitmask) {
6329 		/* flush too old packets
6330 		 * If HZ < 1000, flush all packets.
6331 		 */
6332 		napi_gro_flush(napi, HZ >= 1000);
6333 	}
6334 
6335 	gro_normal_list(napi);
6336 	clear_bit(NAPI_STATE_SCHED, &napi->state);
6337 }
6338 
6339 enum {
6340 	NAPI_F_PREFER_BUSY_POLL	= 1,
6341 	NAPI_F_END_ON_RESCHED	= 2,
6342 };
6343 
6344 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock,
6345 			   unsigned flags, u16 budget)
6346 {
6347 	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
6348 	bool skip_schedule = false;
6349 	unsigned long timeout;
6350 	int rc;
6351 
6352 	/* Busy polling means there is a high chance device driver hard irq
6353 	 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6354 	 * set in napi_schedule_prep().
6355 	 * Since we are about to call napi->poll() once more, we can safely
6356 	 * clear NAPI_STATE_MISSED.
6357 	 *
6358 	 * Note: x86 could use a single "lock and ..." instruction
6359 	 * to perform these two clear_bit()
6360 	 */
6361 	clear_bit(NAPI_STATE_MISSED, &napi->state);
6362 	clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6363 
6364 	local_bh_disable();
6365 	bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
6366 
6367 	if (flags & NAPI_F_PREFER_BUSY_POLL) {
6368 		napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6369 		timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6370 		if (napi->defer_hard_irqs_count && timeout) {
6371 			hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6372 			skip_schedule = true;
6373 		}
6374 	}
6375 
6376 	/* All we really want here is to re-enable device interrupts.
6377 	 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6378 	 */
6379 	rc = napi->poll(napi, budget);
6380 	/* We can't gro_normal_list() here, because napi->poll() might have
6381 	 * rearmed the napi (napi_complete_done()) in which case it could
6382 	 * already be running on another CPU.
6383 	 */
6384 	trace_napi_poll(napi, rc, budget);
6385 	netpoll_poll_unlock(have_poll_lock);
6386 	if (rc == budget)
6387 		__busy_poll_stop(napi, skip_schedule);
6388 	bpf_net_ctx_clear(bpf_net_ctx);
6389 	local_bh_enable();
6390 }
6391 
6392 static void __napi_busy_loop(unsigned int napi_id,
6393 		      bool (*loop_end)(void *, unsigned long),
6394 		      void *loop_end_arg, unsigned flags, u16 budget)
6395 {
6396 	unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6397 	int (*napi_poll)(struct napi_struct *napi, int budget);
6398 	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
6399 	void *have_poll_lock = NULL;
6400 	struct napi_struct *napi;
6401 
6402 	WARN_ON_ONCE(!rcu_read_lock_held());
6403 
6404 restart:
6405 	napi_poll = NULL;
6406 
6407 	napi = napi_by_id(napi_id);
6408 	if (!napi)
6409 		return;
6410 
6411 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6412 		preempt_disable();
6413 	for (;;) {
6414 		int work = 0;
6415 
6416 		local_bh_disable();
6417 		bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
6418 		if (!napi_poll) {
6419 			unsigned long val = READ_ONCE(napi->state);
6420 
6421 			/* If multiple threads are competing for this napi,
6422 			 * we avoid dirtying napi->state as much as we can.
6423 			 */
6424 			if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6425 				   NAPIF_STATE_IN_BUSY_POLL)) {
6426 				if (flags & NAPI_F_PREFER_BUSY_POLL)
6427 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6428 				goto count;
6429 			}
6430 			if (cmpxchg(&napi->state, val,
6431 				    val | NAPIF_STATE_IN_BUSY_POLL |
6432 					  NAPIF_STATE_SCHED) != val) {
6433 				if (flags & NAPI_F_PREFER_BUSY_POLL)
6434 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6435 				goto count;
6436 			}
6437 			have_poll_lock = netpoll_poll_lock(napi);
6438 			napi_poll = napi->poll;
6439 		}
6440 		work = napi_poll(napi, budget);
6441 		trace_napi_poll(napi, work, budget);
6442 		gro_normal_list(napi);
6443 count:
6444 		if (work > 0)
6445 			__NET_ADD_STATS(dev_net(napi->dev),
6446 					LINUX_MIB_BUSYPOLLRXPACKETS, work);
6447 		skb_defer_free_flush(this_cpu_ptr(&softnet_data));
6448 		bpf_net_ctx_clear(bpf_net_ctx);
6449 		local_bh_enable();
6450 
6451 		if (!loop_end || loop_end(loop_end_arg, start_time))
6452 			break;
6453 
6454 		if (unlikely(need_resched())) {
6455 			if (flags & NAPI_F_END_ON_RESCHED)
6456 				break;
6457 			if (napi_poll)
6458 				busy_poll_stop(napi, have_poll_lock, flags, budget);
6459 			if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6460 				preempt_enable();
6461 			rcu_read_unlock();
6462 			cond_resched();
6463 			rcu_read_lock();
6464 			if (loop_end(loop_end_arg, start_time))
6465 				return;
6466 			goto restart;
6467 		}
6468 		cpu_relax();
6469 	}
6470 	if (napi_poll)
6471 		busy_poll_stop(napi, have_poll_lock, flags, budget);
6472 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6473 		preempt_enable();
6474 }
6475 
6476 void napi_busy_loop_rcu(unsigned int napi_id,
6477 			bool (*loop_end)(void *, unsigned long),
6478 			void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6479 {
6480 	unsigned flags = NAPI_F_END_ON_RESCHED;
6481 
6482 	if (prefer_busy_poll)
6483 		flags |= NAPI_F_PREFER_BUSY_POLL;
6484 
6485 	__napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget);
6486 }
6487 
6488 void napi_busy_loop(unsigned int napi_id,
6489 		    bool (*loop_end)(void *, unsigned long),
6490 		    void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6491 {
6492 	unsigned flags = prefer_busy_poll ? NAPI_F_PREFER_BUSY_POLL : 0;
6493 
6494 	rcu_read_lock();
6495 	__napi_busy_loop(napi_id, loop_end, loop_end_arg, flags, budget);
6496 	rcu_read_unlock();
6497 }
6498 EXPORT_SYMBOL(napi_busy_loop);
6499 
6500 #endif /* CONFIG_NET_RX_BUSY_POLL */
6501 
6502 static void napi_hash_add(struct napi_struct *napi)
6503 {
6504 	if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6505 		return;
6506 
6507 	spin_lock(&napi_hash_lock);
6508 
6509 	/* 0..NR_CPUS range is reserved for sender_cpu use */
6510 	do {
6511 		if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6512 			napi_gen_id = MIN_NAPI_ID;
6513 	} while (napi_by_id(napi_gen_id));
6514 	napi->napi_id = napi_gen_id;
6515 
6516 	hlist_add_head_rcu(&napi->napi_hash_node,
6517 			   &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6518 
6519 	spin_unlock(&napi_hash_lock);
6520 }
6521 
6522 /* Warning : caller is responsible to make sure rcu grace period
6523  * is respected before freeing memory containing @napi
6524  */
6525 static void napi_hash_del(struct napi_struct *napi)
6526 {
6527 	spin_lock(&napi_hash_lock);
6528 
6529 	hlist_del_init_rcu(&napi->napi_hash_node);
6530 
6531 	spin_unlock(&napi_hash_lock);
6532 }
6533 
6534 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6535 {
6536 	struct napi_struct *napi;
6537 
6538 	napi = container_of(timer, struct napi_struct, timer);
6539 
6540 	/* Note : we use a relaxed variant of napi_schedule_prep() not setting
6541 	 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6542 	 */
6543 	if (!napi_disable_pending(napi) &&
6544 	    !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6545 		clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6546 		__napi_schedule_irqoff(napi);
6547 	}
6548 
6549 	return HRTIMER_NORESTART;
6550 }
6551 
6552 static void init_gro_hash(struct napi_struct *napi)
6553 {
6554 	int i;
6555 
6556 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6557 		INIT_LIST_HEAD(&napi->gro_hash[i].list);
6558 		napi->gro_hash[i].count = 0;
6559 	}
6560 	napi->gro_bitmask = 0;
6561 }
6562 
6563 int dev_set_threaded(struct net_device *dev, bool threaded)
6564 {
6565 	struct napi_struct *napi;
6566 	int err = 0;
6567 
6568 	if (dev->threaded == threaded)
6569 		return 0;
6570 
6571 	if (threaded) {
6572 		list_for_each_entry(napi, &dev->napi_list, dev_list) {
6573 			if (!napi->thread) {
6574 				err = napi_kthread_create(napi);
6575 				if (err) {
6576 					threaded = false;
6577 					break;
6578 				}
6579 			}
6580 		}
6581 	}
6582 
6583 	WRITE_ONCE(dev->threaded, threaded);
6584 
6585 	/* Make sure kthread is created before THREADED bit
6586 	 * is set.
6587 	 */
6588 	smp_mb__before_atomic();
6589 
6590 	/* Setting/unsetting threaded mode on a napi might not immediately
6591 	 * take effect, if the current napi instance is actively being
6592 	 * polled. In this case, the switch between threaded mode and
6593 	 * softirq mode will happen in the next round of napi_schedule().
6594 	 * This should not cause hiccups/stalls to the live traffic.
6595 	 */
6596 	list_for_each_entry(napi, &dev->napi_list, dev_list)
6597 		assign_bit(NAPI_STATE_THREADED, &napi->state, threaded);
6598 
6599 	return err;
6600 }
6601 EXPORT_SYMBOL(dev_set_threaded);
6602 
6603 /**
6604  * netif_queue_set_napi - Associate queue with the napi
6605  * @dev: device to which NAPI and queue belong
6606  * @queue_index: Index of queue
6607  * @type: queue type as RX or TX
6608  * @napi: NAPI context, pass NULL to clear previously set NAPI
6609  *
6610  * Set queue with its corresponding napi context. This should be done after
6611  * registering the NAPI handler for the queue-vector and the queues have been
6612  * mapped to the corresponding interrupt vector.
6613  */
6614 void netif_queue_set_napi(struct net_device *dev, unsigned int queue_index,
6615 			  enum netdev_queue_type type, struct napi_struct *napi)
6616 {
6617 	struct netdev_rx_queue *rxq;
6618 	struct netdev_queue *txq;
6619 
6620 	if (WARN_ON_ONCE(napi && !napi->dev))
6621 		return;
6622 	if (dev->reg_state >= NETREG_REGISTERED)
6623 		ASSERT_RTNL();
6624 
6625 	switch (type) {
6626 	case NETDEV_QUEUE_TYPE_RX:
6627 		rxq = __netif_get_rx_queue(dev, queue_index);
6628 		rxq->napi = napi;
6629 		return;
6630 	case NETDEV_QUEUE_TYPE_TX:
6631 		txq = netdev_get_tx_queue(dev, queue_index);
6632 		txq->napi = napi;
6633 		return;
6634 	default:
6635 		return;
6636 	}
6637 }
6638 EXPORT_SYMBOL(netif_queue_set_napi);
6639 
6640 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi,
6641 			   int (*poll)(struct napi_struct *, int), int weight)
6642 {
6643 	if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6644 		return;
6645 
6646 	INIT_LIST_HEAD(&napi->poll_list);
6647 	INIT_HLIST_NODE(&napi->napi_hash_node);
6648 	hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6649 	napi->timer.function = napi_watchdog;
6650 	init_gro_hash(napi);
6651 	napi->skb = NULL;
6652 	INIT_LIST_HEAD(&napi->rx_list);
6653 	napi->rx_count = 0;
6654 	napi->poll = poll;
6655 	if (weight > NAPI_POLL_WEIGHT)
6656 		netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6657 				weight);
6658 	napi->weight = weight;
6659 	napi->dev = dev;
6660 #ifdef CONFIG_NETPOLL
6661 	napi->poll_owner = -1;
6662 #endif
6663 	napi->list_owner = -1;
6664 	set_bit(NAPI_STATE_SCHED, &napi->state);
6665 	set_bit(NAPI_STATE_NPSVC, &napi->state);
6666 	list_add_rcu(&napi->dev_list, &dev->napi_list);
6667 	napi_hash_add(napi);
6668 	napi_get_frags_check(napi);
6669 	/* Create kthread for this napi if dev->threaded is set.
6670 	 * Clear dev->threaded if kthread creation failed so that
6671 	 * threaded mode will not be enabled in napi_enable().
6672 	 */
6673 	if (dev->threaded && napi_kthread_create(napi))
6674 		dev->threaded = false;
6675 	netif_napi_set_irq(napi, -1);
6676 }
6677 EXPORT_SYMBOL(netif_napi_add_weight);
6678 
6679 void napi_disable(struct napi_struct *n)
6680 {
6681 	unsigned long val, new;
6682 
6683 	might_sleep();
6684 	set_bit(NAPI_STATE_DISABLE, &n->state);
6685 
6686 	val = READ_ONCE(n->state);
6687 	do {
6688 		while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6689 			usleep_range(20, 200);
6690 			val = READ_ONCE(n->state);
6691 		}
6692 
6693 		new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6694 		new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6695 	} while (!try_cmpxchg(&n->state, &val, new));
6696 
6697 	hrtimer_cancel(&n->timer);
6698 
6699 	clear_bit(NAPI_STATE_DISABLE, &n->state);
6700 }
6701 EXPORT_SYMBOL(napi_disable);
6702 
6703 /**
6704  *	napi_enable - enable NAPI scheduling
6705  *	@n: NAPI context
6706  *
6707  * Resume NAPI from being scheduled on this context.
6708  * Must be paired with napi_disable.
6709  */
6710 void napi_enable(struct napi_struct *n)
6711 {
6712 	unsigned long new, val = READ_ONCE(n->state);
6713 
6714 	do {
6715 		BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6716 
6717 		new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6718 		if (n->dev->threaded && n->thread)
6719 			new |= NAPIF_STATE_THREADED;
6720 	} while (!try_cmpxchg(&n->state, &val, new));
6721 }
6722 EXPORT_SYMBOL(napi_enable);
6723 
6724 static void flush_gro_hash(struct napi_struct *napi)
6725 {
6726 	int i;
6727 
6728 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6729 		struct sk_buff *skb, *n;
6730 
6731 		list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6732 			kfree_skb(skb);
6733 		napi->gro_hash[i].count = 0;
6734 	}
6735 }
6736 
6737 /* Must be called in process context */
6738 void __netif_napi_del(struct napi_struct *napi)
6739 {
6740 	if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6741 		return;
6742 
6743 	napi_hash_del(napi);
6744 	list_del_rcu(&napi->dev_list);
6745 	napi_free_frags(napi);
6746 
6747 	flush_gro_hash(napi);
6748 	napi->gro_bitmask = 0;
6749 
6750 	if (napi->thread) {
6751 		kthread_stop(napi->thread);
6752 		napi->thread = NULL;
6753 	}
6754 }
6755 EXPORT_SYMBOL(__netif_napi_del);
6756 
6757 static int __napi_poll(struct napi_struct *n, bool *repoll)
6758 {
6759 	int work, weight;
6760 
6761 	weight = n->weight;
6762 
6763 	/* This NAPI_STATE_SCHED test is for avoiding a race
6764 	 * with netpoll's poll_napi().  Only the entity which
6765 	 * obtains the lock and sees NAPI_STATE_SCHED set will
6766 	 * actually make the ->poll() call.  Therefore we avoid
6767 	 * accidentally calling ->poll() when NAPI is not scheduled.
6768 	 */
6769 	work = 0;
6770 	if (napi_is_scheduled(n)) {
6771 		work = n->poll(n, weight);
6772 		trace_napi_poll(n, work, weight);
6773 
6774 		xdp_do_check_flushed(n);
6775 	}
6776 
6777 	if (unlikely(work > weight))
6778 		netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6779 				n->poll, work, weight);
6780 
6781 	if (likely(work < weight))
6782 		return work;
6783 
6784 	/* Drivers must not modify the NAPI state if they
6785 	 * consume the entire weight.  In such cases this code
6786 	 * still "owns" the NAPI instance and therefore can
6787 	 * move the instance around on the list at-will.
6788 	 */
6789 	if (unlikely(napi_disable_pending(n))) {
6790 		napi_complete(n);
6791 		return work;
6792 	}
6793 
6794 	/* The NAPI context has more processing work, but busy-polling
6795 	 * is preferred. Exit early.
6796 	 */
6797 	if (napi_prefer_busy_poll(n)) {
6798 		if (napi_complete_done(n, work)) {
6799 			/* If timeout is not set, we need to make sure
6800 			 * that the NAPI is re-scheduled.
6801 			 */
6802 			napi_schedule(n);
6803 		}
6804 		return work;
6805 	}
6806 
6807 	if (n->gro_bitmask) {
6808 		/* flush too old packets
6809 		 * If HZ < 1000, flush all packets.
6810 		 */
6811 		napi_gro_flush(n, HZ >= 1000);
6812 	}
6813 
6814 	gro_normal_list(n);
6815 
6816 	/* Some drivers may have called napi_schedule
6817 	 * prior to exhausting their budget.
6818 	 */
6819 	if (unlikely(!list_empty(&n->poll_list))) {
6820 		pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6821 			     n->dev ? n->dev->name : "backlog");
6822 		return work;
6823 	}
6824 
6825 	*repoll = true;
6826 
6827 	return work;
6828 }
6829 
6830 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6831 {
6832 	bool do_repoll = false;
6833 	void *have;
6834 	int work;
6835 
6836 	list_del_init(&n->poll_list);
6837 
6838 	have = netpoll_poll_lock(n);
6839 
6840 	work = __napi_poll(n, &do_repoll);
6841 
6842 	if (do_repoll)
6843 		list_add_tail(&n->poll_list, repoll);
6844 
6845 	netpoll_poll_unlock(have);
6846 
6847 	return work;
6848 }
6849 
6850 static int napi_thread_wait(struct napi_struct *napi)
6851 {
6852 	set_current_state(TASK_INTERRUPTIBLE);
6853 
6854 	while (!kthread_should_stop()) {
6855 		/* Testing SCHED_THREADED bit here to make sure the current
6856 		 * kthread owns this napi and could poll on this napi.
6857 		 * Testing SCHED bit is not enough because SCHED bit might be
6858 		 * set by some other busy poll thread or by napi_disable().
6859 		 */
6860 		if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state)) {
6861 			WARN_ON(!list_empty(&napi->poll_list));
6862 			__set_current_state(TASK_RUNNING);
6863 			return 0;
6864 		}
6865 
6866 		schedule();
6867 		set_current_state(TASK_INTERRUPTIBLE);
6868 	}
6869 	__set_current_state(TASK_RUNNING);
6870 
6871 	return -1;
6872 }
6873 
6874 static void napi_threaded_poll_loop(struct napi_struct *napi)
6875 {
6876 	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
6877 	struct softnet_data *sd;
6878 	unsigned long last_qs = jiffies;
6879 
6880 	for (;;) {
6881 		bool repoll = false;
6882 		void *have;
6883 
6884 		local_bh_disable();
6885 		bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
6886 
6887 		sd = this_cpu_ptr(&softnet_data);
6888 		sd->in_napi_threaded_poll = true;
6889 
6890 		have = netpoll_poll_lock(napi);
6891 		__napi_poll(napi, &repoll);
6892 		netpoll_poll_unlock(have);
6893 
6894 		sd->in_napi_threaded_poll = false;
6895 		barrier();
6896 
6897 		if (sd_has_rps_ipi_waiting(sd)) {
6898 			local_irq_disable();
6899 			net_rps_action_and_irq_enable(sd);
6900 		}
6901 		skb_defer_free_flush(sd);
6902 		bpf_net_ctx_clear(bpf_net_ctx);
6903 		local_bh_enable();
6904 
6905 		if (!repoll)
6906 			break;
6907 
6908 		rcu_softirq_qs_periodic(last_qs);
6909 		cond_resched();
6910 	}
6911 }
6912 
6913 static int napi_threaded_poll(void *data)
6914 {
6915 	struct napi_struct *napi = data;
6916 
6917 	while (!napi_thread_wait(napi))
6918 		napi_threaded_poll_loop(napi);
6919 
6920 	return 0;
6921 }
6922 
6923 static __latent_entropy void net_rx_action(struct softirq_action *h)
6924 {
6925 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6926 	unsigned long time_limit = jiffies +
6927 		usecs_to_jiffies(READ_ONCE(net_hotdata.netdev_budget_usecs));
6928 	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
6929 	int budget = READ_ONCE(net_hotdata.netdev_budget);
6930 	LIST_HEAD(list);
6931 	LIST_HEAD(repoll);
6932 
6933 	bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
6934 start:
6935 	sd->in_net_rx_action = true;
6936 	local_irq_disable();
6937 	list_splice_init(&sd->poll_list, &list);
6938 	local_irq_enable();
6939 
6940 	for (;;) {
6941 		struct napi_struct *n;
6942 
6943 		skb_defer_free_flush(sd);
6944 
6945 		if (list_empty(&list)) {
6946 			if (list_empty(&repoll)) {
6947 				sd->in_net_rx_action = false;
6948 				barrier();
6949 				/* We need to check if ____napi_schedule()
6950 				 * had refilled poll_list while
6951 				 * sd->in_net_rx_action was true.
6952 				 */
6953 				if (!list_empty(&sd->poll_list))
6954 					goto start;
6955 				if (!sd_has_rps_ipi_waiting(sd))
6956 					goto end;
6957 			}
6958 			break;
6959 		}
6960 
6961 		n = list_first_entry(&list, struct napi_struct, poll_list);
6962 		budget -= napi_poll(n, &repoll);
6963 
6964 		/* If softirq window is exhausted then punt.
6965 		 * Allow this to run for 2 jiffies since which will allow
6966 		 * an average latency of 1.5/HZ.
6967 		 */
6968 		if (unlikely(budget <= 0 ||
6969 			     time_after_eq(jiffies, time_limit))) {
6970 			sd->time_squeeze++;
6971 			break;
6972 		}
6973 	}
6974 
6975 	local_irq_disable();
6976 
6977 	list_splice_tail_init(&sd->poll_list, &list);
6978 	list_splice_tail(&repoll, &list);
6979 	list_splice(&list, &sd->poll_list);
6980 	if (!list_empty(&sd->poll_list))
6981 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
6982 	else
6983 		sd->in_net_rx_action = false;
6984 
6985 	net_rps_action_and_irq_enable(sd);
6986 end:
6987 	bpf_net_ctx_clear(bpf_net_ctx);
6988 }
6989 
6990 struct netdev_adjacent {
6991 	struct net_device *dev;
6992 	netdevice_tracker dev_tracker;
6993 
6994 	/* upper master flag, there can only be one master device per list */
6995 	bool master;
6996 
6997 	/* lookup ignore flag */
6998 	bool ignore;
6999 
7000 	/* counter for the number of times this device was added to us */
7001 	u16 ref_nr;
7002 
7003 	/* private field for the users */
7004 	void *private;
7005 
7006 	struct list_head list;
7007 	struct rcu_head rcu;
7008 };
7009 
7010 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
7011 						 struct list_head *adj_list)
7012 {
7013 	struct netdev_adjacent *adj;
7014 
7015 	list_for_each_entry(adj, adj_list, list) {
7016 		if (adj->dev == adj_dev)
7017 			return adj;
7018 	}
7019 	return NULL;
7020 }
7021 
7022 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
7023 				    struct netdev_nested_priv *priv)
7024 {
7025 	struct net_device *dev = (struct net_device *)priv->data;
7026 
7027 	return upper_dev == dev;
7028 }
7029 
7030 /**
7031  * netdev_has_upper_dev - Check if device is linked to an upper device
7032  * @dev: device
7033  * @upper_dev: upper device to check
7034  *
7035  * Find out if a device is linked to specified upper device and return true
7036  * in case it is. Note that this checks only immediate upper device,
7037  * not through a complete stack of devices. The caller must hold the RTNL lock.
7038  */
7039 bool netdev_has_upper_dev(struct net_device *dev,
7040 			  struct net_device *upper_dev)
7041 {
7042 	struct netdev_nested_priv priv = {
7043 		.data = (void *)upper_dev,
7044 	};
7045 
7046 	ASSERT_RTNL();
7047 
7048 	return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7049 					     &priv);
7050 }
7051 EXPORT_SYMBOL(netdev_has_upper_dev);
7052 
7053 /**
7054  * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
7055  * @dev: device
7056  * @upper_dev: upper device to check
7057  *
7058  * Find out if a device is linked to specified upper device and return true
7059  * in case it is. Note that this checks the entire upper device chain.
7060  * The caller must hold rcu lock.
7061  */
7062 
7063 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
7064 				  struct net_device *upper_dev)
7065 {
7066 	struct netdev_nested_priv priv = {
7067 		.data = (void *)upper_dev,
7068 	};
7069 
7070 	return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7071 					       &priv);
7072 }
7073 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
7074 
7075 /**
7076  * netdev_has_any_upper_dev - Check if device is linked to some device
7077  * @dev: device
7078  *
7079  * Find out if a device is linked to an upper device and return true in case
7080  * it is. The caller must hold the RTNL lock.
7081  */
7082 bool netdev_has_any_upper_dev(struct net_device *dev)
7083 {
7084 	ASSERT_RTNL();
7085 
7086 	return !list_empty(&dev->adj_list.upper);
7087 }
7088 EXPORT_SYMBOL(netdev_has_any_upper_dev);
7089 
7090 /**
7091  * netdev_master_upper_dev_get - Get master upper device
7092  * @dev: device
7093  *
7094  * Find a master upper device and return pointer to it or NULL in case
7095  * it's not there. The caller must hold the RTNL lock.
7096  */
7097 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
7098 {
7099 	struct netdev_adjacent *upper;
7100 
7101 	ASSERT_RTNL();
7102 
7103 	if (list_empty(&dev->adj_list.upper))
7104 		return NULL;
7105 
7106 	upper = list_first_entry(&dev->adj_list.upper,
7107 				 struct netdev_adjacent, list);
7108 	if (likely(upper->master))
7109 		return upper->dev;
7110 	return NULL;
7111 }
7112 EXPORT_SYMBOL(netdev_master_upper_dev_get);
7113 
7114 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
7115 {
7116 	struct netdev_adjacent *upper;
7117 
7118 	ASSERT_RTNL();
7119 
7120 	if (list_empty(&dev->adj_list.upper))
7121 		return NULL;
7122 
7123 	upper = list_first_entry(&dev->adj_list.upper,
7124 				 struct netdev_adjacent, list);
7125 	if (likely(upper->master) && !upper->ignore)
7126 		return upper->dev;
7127 	return NULL;
7128 }
7129 
7130 /**
7131  * netdev_has_any_lower_dev - Check if device is linked to some device
7132  * @dev: device
7133  *
7134  * Find out if a device is linked to a lower device and return true in case
7135  * it is. The caller must hold the RTNL lock.
7136  */
7137 static bool netdev_has_any_lower_dev(struct net_device *dev)
7138 {
7139 	ASSERT_RTNL();
7140 
7141 	return !list_empty(&dev->adj_list.lower);
7142 }
7143 
7144 void *netdev_adjacent_get_private(struct list_head *adj_list)
7145 {
7146 	struct netdev_adjacent *adj;
7147 
7148 	adj = list_entry(adj_list, struct netdev_adjacent, list);
7149 
7150 	return adj->private;
7151 }
7152 EXPORT_SYMBOL(netdev_adjacent_get_private);
7153 
7154 /**
7155  * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
7156  * @dev: device
7157  * @iter: list_head ** of the current position
7158  *
7159  * Gets the next device from the dev's upper list, starting from iter
7160  * position. The caller must hold RCU read lock.
7161  */
7162 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
7163 						 struct list_head **iter)
7164 {
7165 	struct netdev_adjacent *upper;
7166 
7167 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7168 
7169 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7170 
7171 	if (&upper->list == &dev->adj_list.upper)
7172 		return NULL;
7173 
7174 	*iter = &upper->list;
7175 
7176 	return upper->dev;
7177 }
7178 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
7179 
7180 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
7181 						  struct list_head **iter,
7182 						  bool *ignore)
7183 {
7184 	struct netdev_adjacent *upper;
7185 
7186 	upper = list_entry((*iter)->next, struct netdev_adjacent, list);
7187 
7188 	if (&upper->list == &dev->adj_list.upper)
7189 		return NULL;
7190 
7191 	*iter = &upper->list;
7192 	*ignore = upper->ignore;
7193 
7194 	return upper->dev;
7195 }
7196 
7197 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
7198 						    struct list_head **iter)
7199 {
7200 	struct netdev_adjacent *upper;
7201 
7202 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7203 
7204 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7205 
7206 	if (&upper->list == &dev->adj_list.upper)
7207 		return NULL;
7208 
7209 	*iter = &upper->list;
7210 
7211 	return upper->dev;
7212 }
7213 
7214 static int __netdev_walk_all_upper_dev(struct net_device *dev,
7215 				       int (*fn)(struct net_device *dev,
7216 					 struct netdev_nested_priv *priv),
7217 				       struct netdev_nested_priv *priv)
7218 {
7219 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7220 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7221 	int ret, cur = 0;
7222 	bool ignore;
7223 
7224 	now = dev;
7225 	iter = &dev->adj_list.upper;
7226 
7227 	while (1) {
7228 		if (now != dev) {
7229 			ret = fn(now, priv);
7230 			if (ret)
7231 				return ret;
7232 		}
7233 
7234 		next = NULL;
7235 		while (1) {
7236 			udev = __netdev_next_upper_dev(now, &iter, &ignore);
7237 			if (!udev)
7238 				break;
7239 			if (ignore)
7240 				continue;
7241 
7242 			next = udev;
7243 			niter = &udev->adj_list.upper;
7244 			dev_stack[cur] = now;
7245 			iter_stack[cur++] = iter;
7246 			break;
7247 		}
7248 
7249 		if (!next) {
7250 			if (!cur)
7251 				return 0;
7252 			next = dev_stack[--cur];
7253 			niter = iter_stack[cur];
7254 		}
7255 
7256 		now = next;
7257 		iter = niter;
7258 	}
7259 
7260 	return 0;
7261 }
7262 
7263 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7264 				  int (*fn)(struct net_device *dev,
7265 					    struct netdev_nested_priv *priv),
7266 				  struct netdev_nested_priv *priv)
7267 {
7268 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7269 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7270 	int ret, cur = 0;
7271 
7272 	now = dev;
7273 	iter = &dev->adj_list.upper;
7274 
7275 	while (1) {
7276 		if (now != dev) {
7277 			ret = fn(now, priv);
7278 			if (ret)
7279 				return ret;
7280 		}
7281 
7282 		next = NULL;
7283 		while (1) {
7284 			udev = netdev_next_upper_dev_rcu(now, &iter);
7285 			if (!udev)
7286 				break;
7287 
7288 			next = udev;
7289 			niter = &udev->adj_list.upper;
7290 			dev_stack[cur] = now;
7291 			iter_stack[cur++] = iter;
7292 			break;
7293 		}
7294 
7295 		if (!next) {
7296 			if (!cur)
7297 				return 0;
7298 			next = dev_stack[--cur];
7299 			niter = iter_stack[cur];
7300 		}
7301 
7302 		now = next;
7303 		iter = niter;
7304 	}
7305 
7306 	return 0;
7307 }
7308 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7309 
7310 static bool __netdev_has_upper_dev(struct net_device *dev,
7311 				   struct net_device *upper_dev)
7312 {
7313 	struct netdev_nested_priv priv = {
7314 		.flags = 0,
7315 		.data = (void *)upper_dev,
7316 	};
7317 
7318 	ASSERT_RTNL();
7319 
7320 	return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7321 					   &priv);
7322 }
7323 
7324 /**
7325  * netdev_lower_get_next_private - Get the next ->private from the
7326  *				   lower neighbour list
7327  * @dev: device
7328  * @iter: list_head ** of the current position
7329  *
7330  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7331  * list, starting from iter position. The caller must hold either hold the
7332  * RTNL lock or its own locking that guarantees that the neighbour lower
7333  * list will remain unchanged.
7334  */
7335 void *netdev_lower_get_next_private(struct net_device *dev,
7336 				    struct list_head **iter)
7337 {
7338 	struct netdev_adjacent *lower;
7339 
7340 	lower = list_entry(*iter, struct netdev_adjacent, list);
7341 
7342 	if (&lower->list == &dev->adj_list.lower)
7343 		return NULL;
7344 
7345 	*iter = lower->list.next;
7346 
7347 	return lower->private;
7348 }
7349 EXPORT_SYMBOL(netdev_lower_get_next_private);
7350 
7351 /**
7352  * netdev_lower_get_next_private_rcu - Get the next ->private from the
7353  *				       lower neighbour list, RCU
7354  *				       variant
7355  * @dev: device
7356  * @iter: list_head ** of the current position
7357  *
7358  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7359  * list, starting from iter position. The caller must hold RCU read lock.
7360  */
7361 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7362 					struct list_head **iter)
7363 {
7364 	struct netdev_adjacent *lower;
7365 
7366 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7367 
7368 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7369 
7370 	if (&lower->list == &dev->adj_list.lower)
7371 		return NULL;
7372 
7373 	*iter = &lower->list;
7374 
7375 	return lower->private;
7376 }
7377 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7378 
7379 /**
7380  * netdev_lower_get_next - Get the next device from the lower neighbour
7381  *                         list
7382  * @dev: device
7383  * @iter: list_head ** of the current position
7384  *
7385  * Gets the next netdev_adjacent from the dev's lower neighbour
7386  * list, starting from iter position. The caller must hold RTNL lock or
7387  * its own locking that guarantees that the neighbour lower
7388  * list will remain unchanged.
7389  */
7390 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7391 {
7392 	struct netdev_adjacent *lower;
7393 
7394 	lower = list_entry(*iter, struct netdev_adjacent, list);
7395 
7396 	if (&lower->list == &dev->adj_list.lower)
7397 		return NULL;
7398 
7399 	*iter = lower->list.next;
7400 
7401 	return lower->dev;
7402 }
7403 EXPORT_SYMBOL(netdev_lower_get_next);
7404 
7405 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7406 						struct list_head **iter)
7407 {
7408 	struct netdev_adjacent *lower;
7409 
7410 	lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7411 
7412 	if (&lower->list == &dev->adj_list.lower)
7413 		return NULL;
7414 
7415 	*iter = &lower->list;
7416 
7417 	return lower->dev;
7418 }
7419 
7420 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7421 						  struct list_head **iter,
7422 						  bool *ignore)
7423 {
7424 	struct netdev_adjacent *lower;
7425 
7426 	lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7427 
7428 	if (&lower->list == &dev->adj_list.lower)
7429 		return NULL;
7430 
7431 	*iter = &lower->list;
7432 	*ignore = lower->ignore;
7433 
7434 	return lower->dev;
7435 }
7436 
7437 int netdev_walk_all_lower_dev(struct net_device *dev,
7438 			      int (*fn)(struct net_device *dev,
7439 					struct netdev_nested_priv *priv),
7440 			      struct netdev_nested_priv *priv)
7441 {
7442 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7443 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7444 	int ret, cur = 0;
7445 
7446 	now = dev;
7447 	iter = &dev->adj_list.lower;
7448 
7449 	while (1) {
7450 		if (now != dev) {
7451 			ret = fn(now, priv);
7452 			if (ret)
7453 				return ret;
7454 		}
7455 
7456 		next = NULL;
7457 		while (1) {
7458 			ldev = netdev_next_lower_dev(now, &iter);
7459 			if (!ldev)
7460 				break;
7461 
7462 			next = ldev;
7463 			niter = &ldev->adj_list.lower;
7464 			dev_stack[cur] = now;
7465 			iter_stack[cur++] = iter;
7466 			break;
7467 		}
7468 
7469 		if (!next) {
7470 			if (!cur)
7471 				return 0;
7472 			next = dev_stack[--cur];
7473 			niter = iter_stack[cur];
7474 		}
7475 
7476 		now = next;
7477 		iter = niter;
7478 	}
7479 
7480 	return 0;
7481 }
7482 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7483 
7484 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7485 				       int (*fn)(struct net_device *dev,
7486 					 struct netdev_nested_priv *priv),
7487 				       struct netdev_nested_priv *priv)
7488 {
7489 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7490 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7491 	int ret, cur = 0;
7492 	bool ignore;
7493 
7494 	now = dev;
7495 	iter = &dev->adj_list.lower;
7496 
7497 	while (1) {
7498 		if (now != dev) {
7499 			ret = fn(now, priv);
7500 			if (ret)
7501 				return ret;
7502 		}
7503 
7504 		next = NULL;
7505 		while (1) {
7506 			ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7507 			if (!ldev)
7508 				break;
7509 			if (ignore)
7510 				continue;
7511 
7512 			next = ldev;
7513 			niter = &ldev->adj_list.lower;
7514 			dev_stack[cur] = now;
7515 			iter_stack[cur++] = iter;
7516 			break;
7517 		}
7518 
7519 		if (!next) {
7520 			if (!cur)
7521 				return 0;
7522 			next = dev_stack[--cur];
7523 			niter = iter_stack[cur];
7524 		}
7525 
7526 		now = next;
7527 		iter = niter;
7528 	}
7529 
7530 	return 0;
7531 }
7532 
7533 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7534 					     struct list_head **iter)
7535 {
7536 	struct netdev_adjacent *lower;
7537 
7538 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7539 	if (&lower->list == &dev->adj_list.lower)
7540 		return NULL;
7541 
7542 	*iter = &lower->list;
7543 
7544 	return lower->dev;
7545 }
7546 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7547 
7548 static u8 __netdev_upper_depth(struct net_device *dev)
7549 {
7550 	struct net_device *udev;
7551 	struct list_head *iter;
7552 	u8 max_depth = 0;
7553 	bool ignore;
7554 
7555 	for (iter = &dev->adj_list.upper,
7556 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7557 	     udev;
7558 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7559 		if (ignore)
7560 			continue;
7561 		if (max_depth < udev->upper_level)
7562 			max_depth = udev->upper_level;
7563 	}
7564 
7565 	return max_depth;
7566 }
7567 
7568 static u8 __netdev_lower_depth(struct net_device *dev)
7569 {
7570 	struct net_device *ldev;
7571 	struct list_head *iter;
7572 	u8 max_depth = 0;
7573 	bool ignore;
7574 
7575 	for (iter = &dev->adj_list.lower,
7576 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7577 	     ldev;
7578 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7579 		if (ignore)
7580 			continue;
7581 		if (max_depth < ldev->lower_level)
7582 			max_depth = ldev->lower_level;
7583 	}
7584 
7585 	return max_depth;
7586 }
7587 
7588 static int __netdev_update_upper_level(struct net_device *dev,
7589 				       struct netdev_nested_priv *__unused)
7590 {
7591 	dev->upper_level = __netdev_upper_depth(dev) + 1;
7592 	return 0;
7593 }
7594 
7595 #ifdef CONFIG_LOCKDEP
7596 static LIST_HEAD(net_unlink_list);
7597 
7598 static void net_unlink_todo(struct net_device *dev)
7599 {
7600 	if (list_empty(&dev->unlink_list))
7601 		list_add_tail(&dev->unlink_list, &net_unlink_list);
7602 }
7603 #endif
7604 
7605 static int __netdev_update_lower_level(struct net_device *dev,
7606 				       struct netdev_nested_priv *priv)
7607 {
7608 	dev->lower_level = __netdev_lower_depth(dev) + 1;
7609 
7610 #ifdef CONFIG_LOCKDEP
7611 	if (!priv)
7612 		return 0;
7613 
7614 	if (priv->flags & NESTED_SYNC_IMM)
7615 		dev->nested_level = dev->lower_level - 1;
7616 	if (priv->flags & NESTED_SYNC_TODO)
7617 		net_unlink_todo(dev);
7618 #endif
7619 	return 0;
7620 }
7621 
7622 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7623 				  int (*fn)(struct net_device *dev,
7624 					    struct netdev_nested_priv *priv),
7625 				  struct netdev_nested_priv *priv)
7626 {
7627 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7628 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7629 	int ret, cur = 0;
7630 
7631 	now = dev;
7632 	iter = &dev->adj_list.lower;
7633 
7634 	while (1) {
7635 		if (now != dev) {
7636 			ret = fn(now, priv);
7637 			if (ret)
7638 				return ret;
7639 		}
7640 
7641 		next = NULL;
7642 		while (1) {
7643 			ldev = netdev_next_lower_dev_rcu(now, &iter);
7644 			if (!ldev)
7645 				break;
7646 
7647 			next = ldev;
7648 			niter = &ldev->adj_list.lower;
7649 			dev_stack[cur] = now;
7650 			iter_stack[cur++] = iter;
7651 			break;
7652 		}
7653 
7654 		if (!next) {
7655 			if (!cur)
7656 				return 0;
7657 			next = dev_stack[--cur];
7658 			niter = iter_stack[cur];
7659 		}
7660 
7661 		now = next;
7662 		iter = niter;
7663 	}
7664 
7665 	return 0;
7666 }
7667 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7668 
7669 /**
7670  * netdev_lower_get_first_private_rcu - Get the first ->private from the
7671  *				       lower neighbour list, RCU
7672  *				       variant
7673  * @dev: device
7674  *
7675  * Gets the first netdev_adjacent->private from the dev's lower neighbour
7676  * list. The caller must hold RCU read lock.
7677  */
7678 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7679 {
7680 	struct netdev_adjacent *lower;
7681 
7682 	lower = list_first_or_null_rcu(&dev->adj_list.lower,
7683 			struct netdev_adjacent, list);
7684 	if (lower)
7685 		return lower->private;
7686 	return NULL;
7687 }
7688 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7689 
7690 /**
7691  * netdev_master_upper_dev_get_rcu - Get master upper device
7692  * @dev: device
7693  *
7694  * Find a master upper device and return pointer to it or NULL in case
7695  * it's not there. The caller must hold the RCU read lock.
7696  */
7697 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7698 {
7699 	struct netdev_adjacent *upper;
7700 
7701 	upper = list_first_or_null_rcu(&dev->adj_list.upper,
7702 				       struct netdev_adjacent, list);
7703 	if (upper && likely(upper->master))
7704 		return upper->dev;
7705 	return NULL;
7706 }
7707 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7708 
7709 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7710 			      struct net_device *adj_dev,
7711 			      struct list_head *dev_list)
7712 {
7713 	char linkname[IFNAMSIZ+7];
7714 
7715 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7716 		"upper_%s" : "lower_%s", adj_dev->name);
7717 	return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7718 				 linkname);
7719 }
7720 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7721 			       char *name,
7722 			       struct list_head *dev_list)
7723 {
7724 	char linkname[IFNAMSIZ+7];
7725 
7726 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7727 		"upper_%s" : "lower_%s", name);
7728 	sysfs_remove_link(&(dev->dev.kobj), linkname);
7729 }
7730 
7731 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7732 						 struct net_device *adj_dev,
7733 						 struct list_head *dev_list)
7734 {
7735 	return (dev_list == &dev->adj_list.upper ||
7736 		dev_list == &dev->adj_list.lower) &&
7737 		net_eq(dev_net(dev), dev_net(adj_dev));
7738 }
7739 
7740 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7741 					struct net_device *adj_dev,
7742 					struct list_head *dev_list,
7743 					void *private, bool master)
7744 {
7745 	struct netdev_adjacent *adj;
7746 	int ret;
7747 
7748 	adj = __netdev_find_adj(adj_dev, dev_list);
7749 
7750 	if (adj) {
7751 		adj->ref_nr += 1;
7752 		pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7753 			 dev->name, adj_dev->name, adj->ref_nr);
7754 
7755 		return 0;
7756 	}
7757 
7758 	adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7759 	if (!adj)
7760 		return -ENOMEM;
7761 
7762 	adj->dev = adj_dev;
7763 	adj->master = master;
7764 	adj->ref_nr = 1;
7765 	adj->private = private;
7766 	adj->ignore = false;
7767 	netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7768 
7769 	pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7770 		 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7771 
7772 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7773 		ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7774 		if (ret)
7775 			goto free_adj;
7776 	}
7777 
7778 	/* Ensure that master link is always the first item in list. */
7779 	if (master) {
7780 		ret = sysfs_create_link(&(dev->dev.kobj),
7781 					&(adj_dev->dev.kobj), "master");
7782 		if (ret)
7783 			goto remove_symlinks;
7784 
7785 		list_add_rcu(&adj->list, dev_list);
7786 	} else {
7787 		list_add_tail_rcu(&adj->list, dev_list);
7788 	}
7789 
7790 	return 0;
7791 
7792 remove_symlinks:
7793 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7794 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7795 free_adj:
7796 	netdev_put(adj_dev, &adj->dev_tracker);
7797 	kfree(adj);
7798 
7799 	return ret;
7800 }
7801 
7802 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7803 					 struct net_device *adj_dev,
7804 					 u16 ref_nr,
7805 					 struct list_head *dev_list)
7806 {
7807 	struct netdev_adjacent *adj;
7808 
7809 	pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7810 		 dev->name, adj_dev->name, ref_nr);
7811 
7812 	adj = __netdev_find_adj(adj_dev, dev_list);
7813 
7814 	if (!adj) {
7815 		pr_err("Adjacency does not exist for device %s from %s\n",
7816 		       dev->name, adj_dev->name);
7817 		WARN_ON(1);
7818 		return;
7819 	}
7820 
7821 	if (adj->ref_nr > ref_nr) {
7822 		pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7823 			 dev->name, adj_dev->name, ref_nr,
7824 			 adj->ref_nr - ref_nr);
7825 		adj->ref_nr -= ref_nr;
7826 		return;
7827 	}
7828 
7829 	if (adj->master)
7830 		sysfs_remove_link(&(dev->dev.kobj), "master");
7831 
7832 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7833 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7834 
7835 	list_del_rcu(&adj->list);
7836 	pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7837 		 adj_dev->name, dev->name, adj_dev->name);
7838 	netdev_put(adj_dev, &adj->dev_tracker);
7839 	kfree_rcu(adj, rcu);
7840 }
7841 
7842 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7843 					    struct net_device *upper_dev,
7844 					    struct list_head *up_list,
7845 					    struct list_head *down_list,
7846 					    void *private, bool master)
7847 {
7848 	int ret;
7849 
7850 	ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7851 					   private, master);
7852 	if (ret)
7853 		return ret;
7854 
7855 	ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7856 					   private, false);
7857 	if (ret) {
7858 		__netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7859 		return ret;
7860 	}
7861 
7862 	return 0;
7863 }
7864 
7865 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7866 					       struct net_device *upper_dev,
7867 					       u16 ref_nr,
7868 					       struct list_head *up_list,
7869 					       struct list_head *down_list)
7870 {
7871 	__netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7872 	__netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7873 }
7874 
7875 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7876 						struct net_device *upper_dev,
7877 						void *private, bool master)
7878 {
7879 	return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7880 						&dev->adj_list.upper,
7881 						&upper_dev->adj_list.lower,
7882 						private, master);
7883 }
7884 
7885 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7886 						   struct net_device *upper_dev)
7887 {
7888 	__netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7889 					   &dev->adj_list.upper,
7890 					   &upper_dev->adj_list.lower);
7891 }
7892 
7893 static int __netdev_upper_dev_link(struct net_device *dev,
7894 				   struct net_device *upper_dev, bool master,
7895 				   void *upper_priv, void *upper_info,
7896 				   struct netdev_nested_priv *priv,
7897 				   struct netlink_ext_ack *extack)
7898 {
7899 	struct netdev_notifier_changeupper_info changeupper_info = {
7900 		.info = {
7901 			.dev = dev,
7902 			.extack = extack,
7903 		},
7904 		.upper_dev = upper_dev,
7905 		.master = master,
7906 		.linking = true,
7907 		.upper_info = upper_info,
7908 	};
7909 	struct net_device *master_dev;
7910 	int ret = 0;
7911 
7912 	ASSERT_RTNL();
7913 
7914 	if (dev == upper_dev)
7915 		return -EBUSY;
7916 
7917 	/* To prevent loops, check if dev is not upper device to upper_dev. */
7918 	if (__netdev_has_upper_dev(upper_dev, dev))
7919 		return -EBUSY;
7920 
7921 	if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7922 		return -EMLINK;
7923 
7924 	if (!master) {
7925 		if (__netdev_has_upper_dev(dev, upper_dev))
7926 			return -EEXIST;
7927 	} else {
7928 		master_dev = __netdev_master_upper_dev_get(dev);
7929 		if (master_dev)
7930 			return master_dev == upper_dev ? -EEXIST : -EBUSY;
7931 	}
7932 
7933 	ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7934 					    &changeupper_info.info);
7935 	ret = notifier_to_errno(ret);
7936 	if (ret)
7937 		return ret;
7938 
7939 	ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7940 						   master);
7941 	if (ret)
7942 		return ret;
7943 
7944 	ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7945 					    &changeupper_info.info);
7946 	ret = notifier_to_errno(ret);
7947 	if (ret)
7948 		goto rollback;
7949 
7950 	__netdev_update_upper_level(dev, NULL);
7951 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7952 
7953 	__netdev_update_lower_level(upper_dev, priv);
7954 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7955 				    priv);
7956 
7957 	return 0;
7958 
7959 rollback:
7960 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7961 
7962 	return ret;
7963 }
7964 
7965 /**
7966  * netdev_upper_dev_link - Add a link to the upper device
7967  * @dev: device
7968  * @upper_dev: new upper device
7969  * @extack: netlink extended ack
7970  *
7971  * Adds a link to device which is upper to this one. The caller must hold
7972  * the RTNL lock. On a failure a negative errno code is returned.
7973  * On success the reference counts are adjusted and the function
7974  * returns zero.
7975  */
7976 int netdev_upper_dev_link(struct net_device *dev,
7977 			  struct net_device *upper_dev,
7978 			  struct netlink_ext_ack *extack)
7979 {
7980 	struct netdev_nested_priv priv = {
7981 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7982 		.data = NULL,
7983 	};
7984 
7985 	return __netdev_upper_dev_link(dev, upper_dev, false,
7986 				       NULL, NULL, &priv, extack);
7987 }
7988 EXPORT_SYMBOL(netdev_upper_dev_link);
7989 
7990 /**
7991  * netdev_master_upper_dev_link - Add a master link to the upper device
7992  * @dev: device
7993  * @upper_dev: new upper device
7994  * @upper_priv: upper device private
7995  * @upper_info: upper info to be passed down via notifier
7996  * @extack: netlink extended ack
7997  *
7998  * Adds a link to device which is upper to this one. In this case, only
7999  * one master upper device can be linked, although other non-master devices
8000  * might be linked as well. The caller must hold the RTNL lock.
8001  * On a failure a negative errno code is returned. On success the reference
8002  * counts are adjusted and the function returns zero.
8003  */
8004 int netdev_master_upper_dev_link(struct net_device *dev,
8005 				 struct net_device *upper_dev,
8006 				 void *upper_priv, void *upper_info,
8007 				 struct netlink_ext_ack *extack)
8008 {
8009 	struct netdev_nested_priv priv = {
8010 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8011 		.data = NULL,
8012 	};
8013 
8014 	return __netdev_upper_dev_link(dev, upper_dev, true,
8015 				       upper_priv, upper_info, &priv, extack);
8016 }
8017 EXPORT_SYMBOL(netdev_master_upper_dev_link);
8018 
8019 static void __netdev_upper_dev_unlink(struct net_device *dev,
8020 				      struct net_device *upper_dev,
8021 				      struct netdev_nested_priv *priv)
8022 {
8023 	struct netdev_notifier_changeupper_info changeupper_info = {
8024 		.info = {
8025 			.dev = dev,
8026 		},
8027 		.upper_dev = upper_dev,
8028 		.linking = false,
8029 	};
8030 
8031 	ASSERT_RTNL();
8032 
8033 	changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
8034 
8035 	call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8036 				      &changeupper_info.info);
8037 
8038 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8039 
8040 	call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8041 				      &changeupper_info.info);
8042 
8043 	__netdev_update_upper_level(dev, NULL);
8044 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8045 
8046 	__netdev_update_lower_level(upper_dev, priv);
8047 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8048 				    priv);
8049 }
8050 
8051 /**
8052  * netdev_upper_dev_unlink - Removes a link to upper device
8053  * @dev: device
8054  * @upper_dev: new upper device
8055  *
8056  * Removes a link to device which is upper to this one. The caller must hold
8057  * the RTNL lock.
8058  */
8059 void netdev_upper_dev_unlink(struct net_device *dev,
8060 			     struct net_device *upper_dev)
8061 {
8062 	struct netdev_nested_priv priv = {
8063 		.flags = NESTED_SYNC_TODO,
8064 		.data = NULL,
8065 	};
8066 
8067 	__netdev_upper_dev_unlink(dev, upper_dev, &priv);
8068 }
8069 EXPORT_SYMBOL(netdev_upper_dev_unlink);
8070 
8071 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
8072 				      struct net_device *lower_dev,
8073 				      bool val)
8074 {
8075 	struct netdev_adjacent *adj;
8076 
8077 	adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
8078 	if (adj)
8079 		adj->ignore = val;
8080 
8081 	adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
8082 	if (adj)
8083 		adj->ignore = val;
8084 }
8085 
8086 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
8087 					struct net_device *lower_dev)
8088 {
8089 	__netdev_adjacent_dev_set(upper_dev, lower_dev, true);
8090 }
8091 
8092 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
8093 				       struct net_device *lower_dev)
8094 {
8095 	__netdev_adjacent_dev_set(upper_dev, lower_dev, false);
8096 }
8097 
8098 int netdev_adjacent_change_prepare(struct net_device *old_dev,
8099 				   struct net_device *new_dev,
8100 				   struct net_device *dev,
8101 				   struct netlink_ext_ack *extack)
8102 {
8103 	struct netdev_nested_priv priv = {
8104 		.flags = 0,
8105 		.data = NULL,
8106 	};
8107 	int err;
8108 
8109 	if (!new_dev)
8110 		return 0;
8111 
8112 	if (old_dev && new_dev != old_dev)
8113 		netdev_adjacent_dev_disable(dev, old_dev);
8114 	err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
8115 				      extack);
8116 	if (err) {
8117 		if (old_dev && new_dev != old_dev)
8118 			netdev_adjacent_dev_enable(dev, old_dev);
8119 		return err;
8120 	}
8121 
8122 	return 0;
8123 }
8124 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
8125 
8126 void netdev_adjacent_change_commit(struct net_device *old_dev,
8127 				   struct net_device *new_dev,
8128 				   struct net_device *dev)
8129 {
8130 	struct netdev_nested_priv priv = {
8131 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8132 		.data = NULL,
8133 	};
8134 
8135 	if (!new_dev || !old_dev)
8136 		return;
8137 
8138 	if (new_dev == old_dev)
8139 		return;
8140 
8141 	netdev_adjacent_dev_enable(dev, old_dev);
8142 	__netdev_upper_dev_unlink(old_dev, dev, &priv);
8143 }
8144 EXPORT_SYMBOL(netdev_adjacent_change_commit);
8145 
8146 void netdev_adjacent_change_abort(struct net_device *old_dev,
8147 				  struct net_device *new_dev,
8148 				  struct net_device *dev)
8149 {
8150 	struct netdev_nested_priv priv = {
8151 		.flags = 0,
8152 		.data = NULL,
8153 	};
8154 
8155 	if (!new_dev)
8156 		return;
8157 
8158 	if (old_dev && new_dev != old_dev)
8159 		netdev_adjacent_dev_enable(dev, old_dev);
8160 
8161 	__netdev_upper_dev_unlink(new_dev, dev, &priv);
8162 }
8163 EXPORT_SYMBOL(netdev_adjacent_change_abort);
8164 
8165 /**
8166  * netdev_bonding_info_change - Dispatch event about slave change
8167  * @dev: device
8168  * @bonding_info: info to dispatch
8169  *
8170  * Send NETDEV_BONDING_INFO to netdev notifiers with info.
8171  * The caller must hold the RTNL lock.
8172  */
8173 void netdev_bonding_info_change(struct net_device *dev,
8174 				struct netdev_bonding_info *bonding_info)
8175 {
8176 	struct netdev_notifier_bonding_info info = {
8177 		.info.dev = dev,
8178 	};
8179 
8180 	memcpy(&info.bonding_info, bonding_info,
8181 	       sizeof(struct netdev_bonding_info));
8182 	call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
8183 				      &info.info);
8184 }
8185 EXPORT_SYMBOL(netdev_bonding_info_change);
8186 
8187 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
8188 					   struct netlink_ext_ack *extack)
8189 {
8190 	struct netdev_notifier_offload_xstats_info info = {
8191 		.info.dev = dev,
8192 		.info.extack = extack,
8193 		.type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
8194 	};
8195 	int err;
8196 	int rc;
8197 
8198 	dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
8199 					 GFP_KERNEL);
8200 	if (!dev->offload_xstats_l3)
8201 		return -ENOMEM;
8202 
8203 	rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
8204 						  NETDEV_OFFLOAD_XSTATS_DISABLE,
8205 						  &info.info);
8206 	err = notifier_to_errno(rc);
8207 	if (err)
8208 		goto free_stats;
8209 
8210 	return 0;
8211 
8212 free_stats:
8213 	kfree(dev->offload_xstats_l3);
8214 	dev->offload_xstats_l3 = NULL;
8215 	return err;
8216 }
8217 
8218 int netdev_offload_xstats_enable(struct net_device *dev,
8219 				 enum netdev_offload_xstats_type type,
8220 				 struct netlink_ext_ack *extack)
8221 {
8222 	ASSERT_RTNL();
8223 
8224 	if (netdev_offload_xstats_enabled(dev, type))
8225 		return -EALREADY;
8226 
8227 	switch (type) {
8228 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8229 		return netdev_offload_xstats_enable_l3(dev, extack);
8230 	}
8231 
8232 	WARN_ON(1);
8233 	return -EINVAL;
8234 }
8235 EXPORT_SYMBOL(netdev_offload_xstats_enable);
8236 
8237 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
8238 {
8239 	struct netdev_notifier_offload_xstats_info info = {
8240 		.info.dev = dev,
8241 		.type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
8242 	};
8243 
8244 	call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
8245 				      &info.info);
8246 	kfree(dev->offload_xstats_l3);
8247 	dev->offload_xstats_l3 = NULL;
8248 }
8249 
8250 int netdev_offload_xstats_disable(struct net_device *dev,
8251 				  enum netdev_offload_xstats_type type)
8252 {
8253 	ASSERT_RTNL();
8254 
8255 	if (!netdev_offload_xstats_enabled(dev, type))
8256 		return -EALREADY;
8257 
8258 	switch (type) {
8259 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8260 		netdev_offload_xstats_disable_l3(dev);
8261 		return 0;
8262 	}
8263 
8264 	WARN_ON(1);
8265 	return -EINVAL;
8266 }
8267 EXPORT_SYMBOL(netdev_offload_xstats_disable);
8268 
8269 static void netdev_offload_xstats_disable_all(struct net_device *dev)
8270 {
8271 	netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
8272 }
8273 
8274 static struct rtnl_hw_stats64 *
8275 netdev_offload_xstats_get_ptr(const struct net_device *dev,
8276 			      enum netdev_offload_xstats_type type)
8277 {
8278 	switch (type) {
8279 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
8280 		return dev->offload_xstats_l3;
8281 	}
8282 
8283 	WARN_ON(1);
8284 	return NULL;
8285 }
8286 
8287 bool netdev_offload_xstats_enabled(const struct net_device *dev,
8288 				   enum netdev_offload_xstats_type type)
8289 {
8290 	ASSERT_RTNL();
8291 
8292 	return netdev_offload_xstats_get_ptr(dev, type);
8293 }
8294 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
8295 
8296 struct netdev_notifier_offload_xstats_ru {
8297 	bool used;
8298 };
8299 
8300 struct netdev_notifier_offload_xstats_rd {
8301 	struct rtnl_hw_stats64 stats;
8302 	bool used;
8303 };
8304 
8305 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
8306 				  const struct rtnl_hw_stats64 *src)
8307 {
8308 	dest->rx_packets	  += src->rx_packets;
8309 	dest->tx_packets	  += src->tx_packets;
8310 	dest->rx_bytes		  += src->rx_bytes;
8311 	dest->tx_bytes		  += src->tx_bytes;
8312 	dest->rx_errors		  += src->rx_errors;
8313 	dest->tx_errors		  += src->tx_errors;
8314 	dest->rx_dropped	  += src->rx_dropped;
8315 	dest->tx_dropped	  += src->tx_dropped;
8316 	dest->multicast		  += src->multicast;
8317 }
8318 
8319 static int netdev_offload_xstats_get_used(struct net_device *dev,
8320 					  enum netdev_offload_xstats_type type,
8321 					  bool *p_used,
8322 					  struct netlink_ext_ack *extack)
8323 {
8324 	struct netdev_notifier_offload_xstats_ru report_used = {};
8325 	struct netdev_notifier_offload_xstats_info info = {
8326 		.info.dev = dev,
8327 		.info.extack = extack,
8328 		.type = type,
8329 		.report_used = &report_used,
8330 	};
8331 	int rc;
8332 
8333 	WARN_ON(!netdev_offload_xstats_enabled(dev, type));
8334 	rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
8335 					   &info.info);
8336 	*p_used = report_used.used;
8337 	return notifier_to_errno(rc);
8338 }
8339 
8340 static int netdev_offload_xstats_get_stats(struct net_device *dev,
8341 					   enum netdev_offload_xstats_type type,
8342 					   struct rtnl_hw_stats64 *p_stats,
8343 					   bool *p_used,
8344 					   struct netlink_ext_ack *extack)
8345 {
8346 	struct netdev_notifier_offload_xstats_rd report_delta = {};
8347 	struct netdev_notifier_offload_xstats_info info = {
8348 		.info.dev = dev,
8349 		.info.extack = extack,
8350 		.type = type,
8351 		.report_delta = &report_delta,
8352 	};
8353 	struct rtnl_hw_stats64 *stats;
8354 	int rc;
8355 
8356 	stats = netdev_offload_xstats_get_ptr(dev, type);
8357 	if (WARN_ON(!stats))
8358 		return -EINVAL;
8359 
8360 	rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
8361 					   &info.info);
8362 
8363 	/* Cache whatever we got, even if there was an error, otherwise the
8364 	 * successful stats retrievals would get lost.
8365 	 */
8366 	netdev_hw_stats64_add(stats, &report_delta.stats);
8367 
8368 	if (p_stats)
8369 		*p_stats = *stats;
8370 	*p_used = report_delta.used;
8371 
8372 	return notifier_to_errno(rc);
8373 }
8374 
8375 int netdev_offload_xstats_get(struct net_device *dev,
8376 			      enum netdev_offload_xstats_type type,
8377 			      struct rtnl_hw_stats64 *p_stats, bool *p_used,
8378 			      struct netlink_ext_ack *extack)
8379 {
8380 	ASSERT_RTNL();
8381 
8382 	if (p_stats)
8383 		return netdev_offload_xstats_get_stats(dev, type, p_stats,
8384 						       p_used, extack);
8385 	else
8386 		return netdev_offload_xstats_get_used(dev, type, p_used,
8387 						      extack);
8388 }
8389 EXPORT_SYMBOL(netdev_offload_xstats_get);
8390 
8391 void
8392 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
8393 				   const struct rtnl_hw_stats64 *stats)
8394 {
8395 	report_delta->used = true;
8396 	netdev_hw_stats64_add(&report_delta->stats, stats);
8397 }
8398 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
8399 
8400 void
8401 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
8402 {
8403 	report_used->used = true;
8404 }
8405 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
8406 
8407 void netdev_offload_xstats_push_delta(struct net_device *dev,
8408 				      enum netdev_offload_xstats_type type,
8409 				      const struct rtnl_hw_stats64 *p_stats)
8410 {
8411 	struct rtnl_hw_stats64 *stats;
8412 
8413 	ASSERT_RTNL();
8414 
8415 	stats = netdev_offload_xstats_get_ptr(dev, type);
8416 	if (WARN_ON(!stats))
8417 		return;
8418 
8419 	netdev_hw_stats64_add(stats, p_stats);
8420 }
8421 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
8422 
8423 /**
8424  * netdev_get_xmit_slave - Get the xmit slave of master device
8425  * @dev: device
8426  * @skb: The packet
8427  * @all_slaves: assume all the slaves are active
8428  *
8429  * The reference counters are not incremented so the caller must be
8430  * careful with locks. The caller must hold RCU lock.
8431  * %NULL is returned if no slave is found.
8432  */
8433 
8434 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8435 					 struct sk_buff *skb,
8436 					 bool all_slaves)
8437 {
8438 	const struct net_device_ops *ops = dev->netdev_ops;
8439 
8440 	if (!ops->ndo_get_xmit_slave)
8441 		return NULL;
8442 	return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8443 }
8444 EXPORT_SYMBOL(netdev_get_xmit_slave);
8445 
8446 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8447 						  struct sock *sk)
8448 {
8449 	const struct net_device_ops *ops = dev->netdev_ops;
8450 
8451 	if (!ops->ndo_sk_get_lower_dev)
8452 		return NULL;
8453 	return ops->ndo_sk_get_lower_dev(dev, sk);
8454 }
8455 
8456 /**
8457  * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8458  * @dev: device
8459  * @sk: the socket
8460  *
8461  * %NULL is returned if no lower device is found.
8462  */
8463 
8464 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8465 					    struct sock *sk)
8466 {
8467 	struct net_device *lower;
8468 
8469 	lower = netdev_sk_get_lower_dev(dev, sk);
8470 	while (lower) {
8471 		dev = lower;
8472 		lower = netdev_sk_get_lower_dev(dev, sk);
8473 	}
8474 
8475 	return dev;
8476 }
8477 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8478 
8479 static void netdev_adjacent_add_links(struct net_device *dev)
8480 {
8481 	struct netdev_adjacent *iter;
8482 
8483 	struct net *net = dev_net(dev);
8484 
8485 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8486 		if (!net_eq(net, dev_net(iter->dev)))
8487 			continue;
8488 		netdev_adjacent_sysfs_add(iter->dev, dev,
8489 					  &iter->dev->adj_list.lower);
8490 		netdev_adjacent_sysfs_add(dev, iter->dev,
8491 					  &dev->adj_list.upper);
8492 	}
8493 
8494 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8495 		if (!net_eq(net, dev_net(iter->dev)))
8496 			continue;
8497 		netdev_adjacent_sysfs_add(iter->dev, dev,
8498 					  &iter->dev->adj_list.upper);
8499 		netdev_adjacent_sysfs_add(dev, iter->dev,
8500 					  &dev->adj_list.lower);
8501 	}
8502 }
8503 
8504 static void netdev_adjacent_del_links(struct net_device *dev)
8505 {
8506 	struct netdev_adjacent *iter;
8507 
8508 	struct net *net = dev_net(dev);
8509 
8510 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8511 		if (!net_eq(net, dev_net(iter->dev)))
8512 			continue;
8513 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8514 					  &iter->dev->adj_list.lower);
8515 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8516 					  &dev->adj_list.upper);
8517 	}
8518 
8519 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8520 		if (!net_eq(net, dev_net(iter->dev)))
8521 			continue;
8522 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8523 					  &iter->dev->adj_list.upper);
8524 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8525 					  &dev->adj_list.lower);
8526 	}
8527 }
8528 
8529 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8530 {
8531 	struct netdev_adjacent *iter;
8532 
8533 	struct net *net = dev_net(dev);
8534 
8535 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8536 		if (!net_eq(net, dev_net(iter->dev)))
8537 			continue;
8538 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8539 					  &iter->dev->adj_list.lower);
8540 		netdev_adjacent_sysfs_add(iter->dev, dev,
8541 					  &iter->dev->adj_list.lower);
8542 	}
8543 
8544 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8545 		if (!net_eq(net, dev_net(iter->dev)))
8546 			continue;
8547 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8548 					  &iter->dev->adj_list.upper);
8549 		netdev_adjacent_sysfs_add(iter->dev, dev,
8550 					  &iter->dev->adj_list.upper);
8551 	}
8552 }
8553 
8554 void *netdev_lower_dev_get_private(struct net_device *dev,
8555 				   struct net_device *lower_dev)
8556 {
8557 	struct netdev_adjacent *lower;
8558 
8559 	if (!lower_dev)
8560 		return NULL;
8561 	lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8562 	if (!lower)
8563 		return NULL;
8564 
8565 	return lower->private;
8566 }
8567 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8568 
8569 
8570 /**
8571  * netdev_lower_state_changed - Dispatch event about lower device state change
8572  * @lower_dev: device
8573  * @lower_state_info: state to dispatch
8574  *
8575  * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8576  * The caller must hold the RTNL lock.
8577  */
8578 void netdev_lower_state_changed(struct net_device *lower_dev,
8579 				void *lower_state_info)
8580 {
8581 	struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8582 		.info.dev = lower_dev,
8583 	};
8584 
8585 	ASSERT_RTNL();
8586 	changelowerstate_info.lower_state_info = lower_state_info;
8587 	call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8588 				      &changelowerstate_info.info);
8589 }
8590 EXPORT_SYMBOL(netdev_lower_state_changed);
8591 
8592 static void dev_change_rx_flags(struct net_device *dev, int flags)
8593 {
8594 	const struct net_device_ops *ops = dev->netdev_ops;
8595 
8596 	if (ops->ndo_change_rx_flags)
8597 		ops->ndo_change_rx_flags(dev, flags);
8598 }
8599 
8600 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8601 {
8602 	unsigned int old_flags = dev->flags;
8603 	unsigned int promiscuity, flags;
8604 	kuid_t uid;
8605 	kgid_t gid;
8606 
8607 	ASSERT_RTNL();
8608 
8609 	promiscuity = dev->promiscuity + inc;
8610 	if (promiscuity == 0) {
8611 		/*
8612 		 * Avoid overflow.
8613 		 * If inc causes overflow, untouch promisc and return error.
8614 		 */
8615 		if (unlikely(inc > 0)) {
8616 			netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8617 			return -EOVERFLOW;
8618 		}
8619 		flags = old_flags & ~IFF_PROMISC;
8620 	} else {
8621 		flags = old_flags | IFF_PROMISC;
8622 	}
8623 	WRITE_ONCE(dev->promiscuity, promiscuity);
8624 	if (flags != old_flags) {
8625 		WRITE_ONCE(dev->flags, flags);
8626 		netdev_info(dev, "%s promiscuous mode\n",
8627 			    dev->flags & IFF_PROMISC ? "entered" : "left");
8628 		if (audit_enabled) {
8629 			current_uid_gid(&uid, &gid);
8630 			audit_log(audit_context(), GFP_ATOMIC,
8631 				  AUDIT_ANOM_PROMISCUOUS,
8632 				  "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8633 				  dev->name, (dev->flags & IFF_PROMISC),
8634 				  (old_flags & IFF_PROMISC),
8635 				  from_kuid(&init_user_ns, audit_get_loginuid(current)),
8636 				  from_kuid(&init_user_ns, uid),
8637 				  from_kgid(&init_user_ns, gid),
8638 				  audit_get_sessionid(current));
8639 		}
8640 
8641 		dev_change_rx_flags(dev, IFF_PROMISC);
8642 	}
8643 	if (notify)
8644 		__dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL);
8645 	return 0;
8646 }
8647 
8648 /**
8649  *	dev_set_promiscuity	- update promiscuity count on a device
8650  *	@dev: device
8651  *	@inc: modifier
8652  *
8653  *	Add or remove promiscuity from a device. While the count in the device
8654  *	remains above zero the interface remains promiscuous. Once it hits zero
8655  *	the device reverts back to normal filtering operation. A negative inc
8656  *	value is used to drop promiscuity on the device.
8657  *	Return 0 if successful or a negative errno code on error.
8658  */
8659 int dev_set_promiscuity(struct net_device *dev, int inc)
8660 {
8661 	unsigned int old_flags = dev->flags;
8662 	int err;
8663 
8664 	err = __dev_set_promiscuity(dev, inc, true);
8665 	if (err < 0)
8666 		return err;
8667 	if (dev->flags != old_flags)
8668 		dev_set_rx_mode(dev);
8669 	return err;
8670 }
8671 EXPORT_SYMBOL(dev_set_promiscuity);
8672 
8673 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8674 {
8675 	unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8676 	unsigned int allmulti, flags;
8677 
8678 	ASSERT_RTNL();
8679 
8680 	allmulti = dev->allmulti + inc;
8681 	if (allmulti == 0) {
8682 		/*
8683 		 * Avoid overflow.
8684 		 * If inc causes overflow, untouch allmulti and return error.
8685 		 */
8686 		if (unlikely(inc > 0)) {
8687 			netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8688 			return -EOVERFLOW;
8689 		}
8690 		flags = old_flags & ~IFF_ALLMULTI;
8691 	} else {
8692 		flags = old_flags | IFF_ALLMULTI;
8693 	}
8694 	WRITE_ONCE(dev->allmulti, allmulti);
8695 	if (flags != old_flags) {
8696 		WRITE_ONCE(dev->flags, flags);
8697 		netdev_info(dev, "%s allmulticast mode\n",
8698 			    dev->flags & IFF_ALLMULTI ? "entered" : "left");
8699 		dev_change_rx_flags(dev, IFF_ALLMULTI);
8700 		dev_set_rx_mode(dev);
8701 		if (notify)
8702 			__dev_notify_flags(dev, old_flags,
8703 					   dev->gflags ^ old_gflags, 0, NULL);
8704 	}
8705 	return 0;
8706 }
8707 
8708 /**
8709  *	dev_set_allmulti	- update allmulti count on a device
8710  *	@dev: device
8711  *	@inc: modifier
8712  *
8713  *	Add or remove reception of all multicast frames to a device. While the
8714  *	count in the device remains above zero the interface remains listening
8715  *	to all interfaces. Once it hits zero the device reverts back to normal
8716  *	filtering operation. A negative @inc value is used to drop the counter
8717  *	when releasing a resource needing all multicasts.
8718  *	Return 0 if successful or a negative errno code on error.
8719  */
8720 
8721 int dev_set_allmulti(struct net_device *dev, int inc)
8722 {
8723 	return __dev_set_allmulti(dev, inc, true);
8724 }
8725 EXPORT_SYMBOL(dev_set_allmulti);
8726 
8727 /*
8728  *	Upload unicast and multicast address lists to device and
8729  *	configure RX filtering. When the device doesn't support unicast
8730  *	filtering it is put in promiscuous mode while unicast addresses
8731  *	are present.
8732  */
8733 void __dev_set_rx_mode(struct net_device *dev)
8734 {
8735 	const struct net_device_ops *ops = dev->netdev_ops;
8736 
8737 	/* dev_open will call this function so the list will stay sane. */
8738 	if (!(dev->flags&IFF_UP))
8739 		return;
8740 
8741 	if (!netif_device_present(dev))
8742 		return;
8743 
8744 	if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8745 		/* Unicast addresses changes may only happen under the rtnl,
8746 		 * therefore calling __dev_set_promiscuity here is safe.
8747 		 */
8748 		if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8749 			__dev_set_promiscuity(dev, 1, false);
8750 			dev->uc_promisc = true;
8751 		} else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8752 			__dev_set_promiscuity(dev, -1, false);
8753 			dev->uc_promisc = false;
8754 		}
8755 	}
8756 
8757 	if (ops->ndo_set_rx_mode)
8758 		ops->ndo_set_rx_mode(dev);
8759 }
8760 
8761 void dev_set_rx_mode(struct net_device *dev)
8762 {
8763 	netif_addr_lock_bh(dev);
8764 	__dev_set_rx_mode(dev);
8765 	netif_addr_unlock_bh(dev);
8766 }
8767 
8768 /**
8769  *	dev_get_flags - get flags reported to userspace
8770  *	@dev: device
8771  *
8772  *	Get the combination of flag bits exported through APIs to userspace.
8773  */
8774 unsigned int dev_get_flags(const struct net_device *dev)
8775 {
8776 	unsigned int flags;
8777 
8778 	flags = (READ_ONCE(dev->flags) & ~(IFF_PROMISC |
8779 				IFF_ALLMULTI |
8780 				IFF_RUNNING |
8781 				IFF_LOWER_UP |
8782 				IFF_DORMANT)) |
8783 		(READ_ONCE(dev->gflags) & (IFF_PROMISC |
8784 				IFF_ALLMULTI));
8785 
8786 	if (netif_running(dev)) {
8787 		if (netif_oper_up(dev))
8788 			flags |= IFF_RUNNING;
8789 		if (netif_carrier_ok(dev))
8790 			flags |= IFF_LOWER_UP;
8791 		if (netif_dormant(dev))
8792 			flags |= IFF_DORMANT;
8793 	}
8794 
8795 	return flags;
8796 }
8797 EXPORT_SYMBOL(dev_get_flags);
8798 
8799 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8800 		       struct netlink_ext_ack *extack)
8801 {
8802 	unsigned int old_flags = dev->flags;
8803 	int ret;
8804 
8805 	ASSERT_RTNL();
8806 
8807 	/*
8808 	 *	Set the flags on our device.
8809 	 */
8810 
8811 	dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8812 			       IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8813 			       IFF_AUTOMEDIA)) |
8814 		     (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8815 				    IFF_ALLMULTI));
8816 
8817 	/*
8818 	 *	Load in the correct multicast list now the flags have changed.
8819 	 */
8820 
8821 	if ((old_flags ^ flags) & IFF_MULTICAST)
8822 		dev_change_rx_flags(dev, IFF_MULTICAST);
8823 
8824 	dev_set_rx_mode(dev);
8825 
8826 	/*
8827 	 *	Have we downed the interface. We handle IFF_UP ourselves
8828 	 *	according to user attempts to set it, rather than blindly
8829 	 *	setting it.
8830 	 */
8831 
8832 	ret = 0;
8833 	if ((old_flags ^ flags) & IFF_UP) {
8834 		if (old_flags & IFF_UP)
8835 			__dev_close(dev);
8836 		else
8837 			ret = __dev_open(dev, extack);
8838 	}
8839 
8840 	if ((flags ^ dev->gflags) & IFF_PROMISC) {
8841 		int inc = (flags & IFF_PROMISC) ? 1 : -1;
8842 		unsigned int old_flags = dev->flags;
8843 
8844 		dev->gflags ^= IFF_PROMISC;
8845 
8846 		if (__dev_set_promiscuity(dev, inc, false) >= 0)
8847 			if (dev->flags != old_flags)
8848 				dev_set_rx_mode(dev);
8849 	}
8850 
8851 	/* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8852 	 * is important. Some (broken) drivers set IFF_PROMISC, when
8853 	 * IFF_ALLMULTI is requested not asking us and not reporting.
8854 	 */
8855 	if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8856 		int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8857 
8858 		dev->gflags ^= IFF_ALLMULTI;
8859 		__dev_set_allmulti(dev, inc, false);
8860 	}
8861 
8862 	return ret;
8863 }
8864 
8865 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8866 			unsigned int gchanges, u32 portid,
8867 			const struct nlmsghdr *nlh)
8868 {
8869 	unsigned int changes = dev->flags ^ old_flags;
8870 
8871 	if (gchanges)
8872 		rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh);
8873 
8874 	if (changes & IFF_UP) {
8875 		if (dev->flags & IFF_UP)
8876 			call_netdevice_notifiers(NETDEV_UP, dev);
8877 		else
8878 			call_netdevice_notifiers(NETDEV_DOWN, dev);
8879 	}
8880 
8881 	if (dev->flags & IFF_UP &&
8882 	    (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8883 		struct netdev_notifier_change_info change_info = {
8884 			.info = {
8885 				.dev = dev,
8886 			},
8887 			.flags_changed = changes,
8888 		};
8889 
8890 		call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8891 	}
8892 }
8893 
8894 /**
8895  *	dev_change_flags - change device settings
8896  *	@dev: device
8897  *	@flags: device state flags
8898  *	@extack: netlink extended ack
8899  *
8900  *	Change settings on device based state flags. The flags are
8901  *	in the userspace exported format.
8902  */
8903 int dev_change_flags(struct net_device *dev, unsigned int flags,
8904 		     struct netlink_ext_ack *extack)
8905 {
8906 	int ret;
8907 	unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8908 
8909 	ret = __dev_change_flags(dev, flags, extack);
8910 	if (ret < 0)
8911 		return ret;
8912 
8913 	changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8914 	__dev_notify_flags(dev, old_flags, changes, 0, NULL);
8915 	return ret;
8916 }
8917 EXPORT_SYMBOL(dev_change_flags);
8918 
8919 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8920 {
8921 	const struct net_device_ops *ops = dev->netdev_ops;
8922 
8923 	if (ops->ndo_change_mtu)
8924 		return ops->ndo_change_mtu(dev, new_mtu);
8925 
8926 	/* Pairs with all the lockless reads of dev->mtu in the stack */
8927 	WRITE_ONCE(dev->mtu, new_mtu);
8928 	return 0;
8929 }
8930 EXPORT_SYMBOL(__dev_set_mtu);
8931 
8932 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8933 		     struct netlink_ext_ack *extack)
8934 {
8935 	/* MTU must be positive, and in range */
8936 	if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8937 		NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8938 		return -EINVAL;
8939 	}
8940 
8941 	if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8942 		NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8943 		return -EINVAL;
8944 	}
8945 	return 0;
8946 }
8947 
8948 /**
8949  *	dev_set_mtu_ext - Change maximum transfer unit
8950  *	@dev: device
8951  *	@new_mtu: new transfer unit
8952  *	@extack: netlink extended ack
8953  *
8954  *	Change the maximum transfer size of the network device.
8955  */
8956 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8957 		    struct netlink_ext_ack *extack)
8958 {
8959 	int err, orig_mtu;
8960 
8961 	if (new_mtu == dev->mtu)
8962 		return 0;
8963 
8964 	err = dev_validate_mtu(dev, new_mtu, extack);
8965 	if (err)
8966 		return err;
8967 
8968 	if (!netif_device_present(dev))
8969 		return -ENODEV;
8970 
8971 	err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8972 	err = notifier_to_errno(err);
8973 	if (err)
8974 		return err;
8975 
8976 	orig_mtu = dev->mtu;
8977 	err = __dev_set_mtu(dev, new_mtu);
8978 
8979 	if (!err) {
8980 		err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8981 						   orig_mtu);
8982 		err = notifier_to_errno(err);
8983 		if (err) {
8984 			/* setting mtu back and notifying everyone again,
8985 			 * so that they have a chance to revert changes.
8986 			 */
8987 			__dev_set_mtu(dev, orig_mtu);
8988 			call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8989 						     new_mtu);
8990 		}
8991 	}
8992 	return err;
8993 }
8994 
8995 int dev_set_mtu(struct net_device *dev, int new_mtu)
8996 {
8997 	struct netlink_ext_ack extack;
8998 	int err;
8999 
9000 	memset(&extack, 0, sizeof(extack));
9001 	err = dev_set_mtu_ext(dev, new_mtu, &extack);
9002 	if (err && extack._msg)
9003 		net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
9004 	return err;
9005 }
9006 EXPORT_SYMBOL(dev_set_mtu);
9007 
9008 /**
9009  *	dev_change_tx_queue_len - Change TX queue length of a netdevice
9010  *	@dev: device
9011  *	@new_len: new tx queue length
9012  */
9013 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
9014 {
9015 	unsigned int orig_len = dev->tx_queue_len;
9016 	int res;
9017 
9018 	if (new_len != (unsigned int)new_len)
9019 		return -ERANGE;
9020 
9021 	if (new_len != orig_len) {
9022 		WRITE_ONCE(dev->tx_queue_len, new_len);
9023 		res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
9024 		res = notifier_to_errno(res);
9025 		if (res)
9026 			goto err_rollback;
9027 		res = dev_qdisc_change_tx_queue_len(dev);
9028 		if (res)
9029 			goto err_rollback;
9030 	}
9031 
9032 	return 0;
9033 
9034 err_rollback:
9035 	netdev_err(dev, "refused to change device tx_queue_len\n");
9036 	WRITE_ONCE(dev->tx_queue_len, orig_len);
9037 	return res;
9038 }
9039 
9040 /**
9041  *	dev_set_group - Change group this device belongs to
9042  *	@dev: device
9043  *	@new_group: group this device should belong to
9044  */
9045 void dev_set_group(struct net_device *dev, int new_group)
9046 {
9047 	dev->group = new_group;
9048 }
9049 
9050 /**
9051  *	dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
9052  *	@dev: device
9053  *	@addr: new address
9054  *	@extack: netlink extended ack
9055  */
9056 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
9057 			      struct netlink_ext_ack *extack)
9058 {
9059 	struct netdev_notifier_pre_changeaddr_info info = {
9060 		.info.dev = dev,
9061 		.info.extack = extack,
9062 		.dev_addr = addr,
9063 	};
9064 	int rc;
9065 
9066 	rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
9067 	return notifier_to_errno(rc);
9068 }
9069 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
9070 
9071 /**
9072  *	dev_set_mac_address - Change Media Access Control Address
9073  *	@dev: device
9074  *	@sa: new address
9075  *	@extack: netlink extended ack
9076  *
9077  *	Change the hardware (MAC) address of the device
9078  */
9079 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
9080 			struct netlink_ext_ack *extack)
9081 {
9082 	const struct net_device_ops *ops = dev->netdev_ops;
9083 	int err;
9084 
9085 	if (!ops->ndo_set_mac_address)
9086 		return -EOPNOTSUPP;
9087 	if (sa->sa_family != dev->type)
9088 		return -EINVAL;
9089 	if (!netif_device_present(dev))
9090 		return -ENODEV;
9091 	err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
9092 	if (err)
9093 		return err;
9094 	if (memcmp(dev->dev_addr, sa->sa_data, dev->addr_len)) {
9095 		err = ops->ndo_set_mac_address(dev, sa);
9096 		if (err)
9097 			return err;
9098 	}
9099 	dev->addr_assign_type = NET_ADDR_SET;
9100 	call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
9101 	add_device_randomness(dev->dev_addr, dev->addr_len);
9102 	return 0;
9103 }
9104 EXPORT_SYMBOL(dev_set_mac_address);
9105 
9106 DECLARE_RWSEM(dev_addr_sem);
9107 
9108 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
9109 			     struct netlink_ext_ack *extack)
9110 {
9111 	int ret;
9112 
9113 	down_write(&dev_addr_sem);
9114 	ret = dev_set_mac_address(dev, sa, extack);
9115 	up_write(&dev_addr_sem);
9116 	return ret;
9117 }
9118 EXPORT_SYMBOL(dev_set_mac_address_user);
9119 
9120 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
9121 {
9122 	size_t size = sizeof(sa->sa_data_min);
9123 	struct net_device *dev;
9124 	int ret = 0;
9125 
9126 	down_read(&dev_addr_sem);
9127 	rcu_read_lock();
9128 
9129 	dev = dev_get_by_name_rcu(net, dev_name);
9130 	if (!dev) {
9131 		ret = -ENODEV;
9132 		goto unlock;
9133 	}
9134 	if (!dev->addr_len)
9135 		memset(sa->sa_data, 0, size);
9136 	else
9137 		memcpy(sa->sa_data, dev->dev_addr,
9138 		       min_t(size_t, size, dev->addr_len));
9139 	sa->sa_family = dev->type;
9140 
9141 unlock:
9142 	rcu_read_unlock();
9143 	up_read(&dev_addr_sem);
9144 	return ret;
9145 }
9146 EXPORT_SYMBOL(dev_get_mac_address);
9147 
9148 /**
9149  *	dev_change_carrier - Change device carrier
9150  *	@dev: device
9151  *	@new_carrier: new value
9152  *
9153  *	Change device carrier
9154  */
9155 int dev_change_carrier(struct net_device *dev, bool new_carrier)
9156 {
9157 	const struct net_device_ops *ops = dev->netdev_ops;
9158 
9159 	if (!ops->ndo_change_carrier)
9160 		return -EOPNOTSUPP;
9161 	if (!netif_device_present(dev))
9162 		return -ENODEV;
9163 	return ops->ndo_change_carrier(dev, new_carrier);
9164 }
9165 
9166 /**
9167  *	dev_get_phys_port_id - Get device physical port ID
9168  *	@dev: device
9169  *	@ppid: port ID
9170  *
9171  *	Get device physical port ID
9172  */
9173 int dev_get_phys_port_id(struct net_device *dev,
9174 			 struct netdev_phys_item_id *ppid)
9175 {
9176 	const struct net_device_ops *ops = dev->netdev_ops;
9177 
9178 	if (!ops->ndo_get_phys_port_id)
9179 		return -EOPNOTSUPP;
9180 	return ops->ndo_get_phys_port_id(dev, ppid);
9181 }
9182 
9183 /**
9184  *	dev_get_phys_port_name - Get device physical port name
9185  *	@dev: device
9186  *	@name: port name
9187  *	@len: limit of bytes to copy to name
9188  *
9189  *	Get device physical port name
9190  */
9191 int dev_get_phys_port_name(struct net_device *dev,
9192 			   char *name, size_t len)
9193 {
9194 	const struct net_device_ops *ops = dev->netdev_ops;
9195 	int err;
9196 
9197 	if (ops->ndo_get_phys_port_name) {
9198 		err = ops->ndo_get_phys_port_name(dev, name, len);
9199 		if (err != -EOPNOTSUPP)
9200 			return err;
9201 	}
9202 	return devlink_compat_phys_port_name_get(dev, name, len);
9203 }
9204 
9205 /**
9206  *	dev_get_port_parent_id - Get the device's port parent identifier
9207  *	@dev: network device
9208  *	@ppid: pointer to a storage for the port's parent identifier
9209  *	@recurse: allow/disallow recursion to lower devices
9210  *
9211  *	Get the devices's port parent identifier
9212  */
9213 int dev_get_port_parent_id(struct net_device *dev,
9214 			   struct netdev_phys_item_id *ppid,
9215 			   bool recurse)
9216 {
9217 	const struct net_device_ops *ops = dev->netdev_ops;
9218 	struct netdev_phys_item_id first = { };
9219 	struct net_device *lower_dev;
9220 	struct list_head *iter;
9221 	int err;
9222 
9223 	if (ops->ndo_get_port_parent_id) {
9224 		err = ops->ndo_get_port_parent_id(dev, ppid);
9225 		if (err != -EOPNOTSUPP)
9226 			return err;
9227 	}
9228 
9229 	err = devlink_compat_switch_id_get(dev, ppid);
9230 	if (!recurse || err != -EOPNOTSUPP)
9231 		return err;
9232 
9233 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
9234 		err = dev_get_port_parent_id(lower_dev, ppid, true);
9235 		if (err)
9236 			break;
9237 		if (!first.id_len)
9238 			first = *ppid;
9239 		else if (memcmp(&first, ppid, sizeof(*ppid)))
9240 			return -EOPNOTSUPP;
9241 	}
9242 
9243 	return err;
9244 }
9245 EXPORT_SYMBOL(dev_get_port_parent_id);
9246 
9247 /**
9248  *	netdev_port_same_parent_id - Indicate if two network devices have
9249  *	the same port parent identifier
9250  *	@a: first network device
9251  *	@b: second network device
9252  */
9253 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
9254 {
9255 	struct netdev_phys_item_id a_id = { };
9256 	struct netdev_phys_item_id b_id = { };
9257 
9258 	if (dev_get_port_parent_id(a, &a_id, true) ||
9259 	    dev_get_port_parent_id(b, &b_id, true))
9260 		return false;
9261 
9262 	return netdev_phys_item_id_same(&a_id, &b_id);
9263 }
9264 EXPORT_SYMBOL(netdev_port_same_parent_id);
9265 
9266 /**
9267  *	dev_change_proto_down - set carrier according to proto_down.
9268  *
9269  *	@dev: device
9270  *	@proto_down: new value
9271  */
9272 int dev_change_proto_down(struct net_device *dev, bool proto_down)
9273 {
9274 	if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN))
9275 		return -EOPNOTSUPP;
9276 	if (!netif_device_present(dev))
9277 		return -ENODEV;
9278 	if (proto_down)
9279 		netif_carrier_off(dev);
9280 	else
9281 		netif_carrier_on(dev);
9282 	WRITE_ONCE(dev->proto_down, proto_down);
9283 	return 0;
9284 }
9285 
9286 /**
9287  *	dev_change_proto_down_reason - proto down reason
9288  *
9289  *	@dev: device
9290  *	@mask: proto down mask
9291  *	@value: proto down value
9292  */
9293 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9294 				  u32 value)
9295 {
9296 	u32 proto_down_reason;
9297 	int b;
9298 
9299 	if (!mask) {
9300 		proto_down_reason = value;
9301 	} else {
9302 		proto_down_reason = dev->proto_down_reason;
9303 		for_each_set_bit(b, &mask, 32) {
9304 			if (value & (1 << b))
9305 				proto_down_reason |= BIT(b);
9306 			else
9307 				proto_down_reason &= ~BIT(b);
9308 		}
9309 	}
9310 	WRITE_ONCE(dev->proto_down_reason, proto_down_reason);
9311 }
9312 
9313 struct bpf_xdp_link {
9314 	struct bpf_link link;
9315 	struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9316 	int flags;
9317 };
9318 
9319 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9320 {
9321 	if (flags & XDP_FLAGS_HW_MODE)
9322 		return XDP_MODE_HW;
9323 	if (flags & XDP_FLAGS_DRV_MODE)
9324 		return XDP_MODE_DRV;
9325 	if (flags & XDP_FLAGS_SKB_MODE)
9326 		return XDP_MODE_SKB;
9327 	return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9328 }
9329 
9330 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9331 {
9332 	switch (mode) {
9333 	case XDP_MODE_SKB:
9334 		return generic_xdp_install;
9335 	case XDP_MODE_DRV:
9336 	case XDP_MODE_HW:
9337 		return dev->netdev_ops->ndo_bpf;
9338 	default:
9339 		return NULL;
9340 	}
9341 }
9342 
9343 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9344 					 enum bpf_xdp_mode mode)
9345 {
9346 	return dev->xdp_state[mode].link;
9347 }
9348 
9349 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9350 				     enum bpf_xdp_mode mode)
9351 {
9352 	struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9353 
9354 	if (link)
9355 		return link->link.prog;
9356 	return dev->xdp_state[mode].prog;
9357 }
9358 
9359 u8 dev_xdp_prog_count(struct net_device *dev)
9360 {
9361 	u8 count = 0;
9362 	int i;
9363 
9364 	for (i = 0; i < __MAX_XDP_MODE; i++)
9365 		if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9366 			count++;
9367 	return count;
9368 }
9369 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9370 
9371 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9372 {
9373 	struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9374 
9375 	return prog ? prog->aux->id : 0;
9376 }
9377 
9378 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9379 			     struct bpf_xdp_link *link)
9380 {
9381 	dev->xdp_state[mode].link = link;
9382 	dev->xdp_state[mode].prog = NULL;
9383 }
9384 
9385 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9386 			     struct bpf_prog *prog)
9387 {
9388 	dev->xdp_state[mode].link = NULL;
9389 	dev->xdp_state[mode].prog = prog;
9390 }
9391 
9392 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9393 			   bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9394 			   u32 flags, struct bpf_prog *prog)
9395 {
9396 	struct netdev_bpf xdp;
9397 	int err;
9398 
9399 	memset(&xdp, 0, sizeof(xdp));
9400 	xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9401 	xdp.extack = extack;
9402 	xdp.flags = flags;
9403 	xdp.prog = prog;
9404 
9405 	/* Drivers assume refcnt is already incremented (i.e, prog pointer is
9406 	 * "moved" into driver), so they don't increment it on their own, but
9407 	 * they do decrement refcnt when program is detached or replaced.
9408 	 * Given net_device also owns link/prog, we need to bump refcnt here
9409 	 * to prevent drivers from underflowing it.
9410 	 */
9411 	if (prog)
9412 		bpf_prog_inc(prog);
9413 	err = bpf_op(dev, &xdp);
9414 	if (err) {
9415 		if (prog)
9416 			bpf_prog_put(prog);
9417 		return err;
9418 	}
9419 
9420 	if (mode != XDP_MODE_HW)
9421 		bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9422 
9423 	return 0;
9424 }
9425 
9426 static void dev_xdp_uninstall(struct net_device *dev)
9427 {
9428 	struct bpf_xdp_link *link;
9429 	struct bpf_prog *prog;
9430 	enum bpf_xdp_mode mode;
9431 	bpf_op_t bpf_op;
9432 
9433 	ASSERT_RTNL();
9434 
9435 	for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9436 		prog = dev_xdp_prog(dev, mode);
9437 		if (!prog)
9438 			continue;
9439 
9440 		bpf_op = dev_xdp_bpf_op(dev, mode);
9441 		if (!bpf_op)
9442 			continue;
9443 
9444 		WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9445 
9446 		/* auto-detach link from net device */
9447 		link = dev_xdp_link(dev, mode);
9448 		if (link)
9449 			link->dev = NULL;
9450 		else
9451 			bpf_prog_put(prog);
9452 
9453 		dev_xdp_set_link(dev, mode, NULL);
9454 	}
9455 }
9456 
9457 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9458 			  struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9459 			  struct bpf_prog *old_prog, u32 flags)
9460 {
9461 	unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9462 	struct bpf_prog *cur_prog;
9463 	struct net_device *upper;
9464 	struct list_head *iter;
9465 	enum bpf_xdp_mode mode;
9466 	bpf_op_t bpf_op;
9467 	int err;
9468 
9469 	ASSERT_RTNL();
9470 
9471 	/* either link or prog attachment, never both */
9472 	if (link && (new_prog || old_prog))
9473 		return -EINVAL;
9474 	/* link supports only XDP mode flags */
9475 	if (link && (flags & ~XDP_FLAGS_MODES)) {
9476 		NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9477 		return -EINVAL;
9478 	}
9479 	/* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9480 	if (num_modes > 1) {
9481 		NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9482 		return -EINVAL;
9483 	}
9484 	/* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9485 	if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9486 		NL_SET_ERR_MSG(extack,
9487 			       "More than one program loaded, unset mode is ambiguous");
9488 		return -EINVAL;
9489 	}
9490 	/* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9491 	if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9492 		NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9493 		return -EINVAL;
9494 	}
9495 
9496 	mode = dev_xdp_mode(dev, flags);
9497 	/* can't replace attached link */
9498 	if (dev_xdp_link(dev, mode)) {
9499 		NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9500 		return -EBUSY;
9501 	}
9502 
9503 	/* don't allow if an upper device already has a program */
9504 	netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9505 		if (dev_xdp_prog_count(upper) > 0) {
9506 			NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9507 			return -EEXIST;
9508 		}
9509 	}
9510 
9511 	cur_prog = dev_xdp_prog(dev, mode);
9512 	/* can't replace attached prog with link */
9513 	if (link && cur_prog) {
9514 		NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9515 		return -EBUSY;
9516 	}
9517 	if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9518 		NL_SET_ERR_MSG(extack, "Active program does not match expected");
9519 		return -EEXIST;
9520 	}
9521 
9522 	/* put effective new program into new_prog */
9523 	if (link)
9524 		new_prog = link->link.prog;
9525 
9526 	if (new_prog) {
9527 		bool offload = mode == XDP_MODE_HW;
9528 		enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9529 					       ? XDP_MODE_DRV : XDP_MODE_SKB;
9530 
9531 		if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9532 			NL_SET_ERR_MSG(extack, "XDP program already attached");
9533 			return -EBUSY;
9534 		}
9535 		if (!offload && dev_xdp_prog(dev, other_mode)) {
9536 			NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9537 			return -EEXIST;
9538 		}
9539 		if (!offload && bpf_prog_is_offloaded(new_prog->aux)) {
9540 			NL_SET_ERR_MSG(extack, "Using offloaded program without HW_MODE flag is not supported");
9541 			return -EINVAL;
9542 		}
9543 		if (bpf_prog_is_dev_bound(new_prog->aux) && !bpf_offload_dev_match(new_prog, dev)) {
9544 			NL_SET_ERR_MSG(extack, "Program bound to different device");
9545 			return -EINVAL;
9546 		}
9547 		if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9548 			NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9549 			return -EINVAL;
9550 		}
9551 		if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9552 			NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9553 			return -EINVAL;
9554 		}
9555 	}
9556 
9557 	/* don't call drivers if the effective program didn't change */
9558 	if (new_prog != cur_prog) {
9559 		bpf_op = dev_xdp_bpf_op(dev, mode);
9560 		if (!bpf_op) {
9561 			NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9562 			return -EOPNOTSUPP;
9563 		}
9564 
9565 		err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9566 		if (err)
9567 			return err;
9568 	}
9569 
9570 	if (link)
9571 		dev_xdp_set_link(dev, mode, link);
9572 	else
9573 		dev_xdp_set_prog(dev, mode, new_prog);
9574 	if (cur_prog)
9575 		bpf_prog_put(cur_prog);
9576 
9577 	return 0;
9578 }
9579 
9580 static int dev_xdp_attach_link(struct net_device *dev,
9581 			       struct netlink_ext_ack *extack,
9582 			       struct bpf_xdp_link *link)
9583 {
9584 	return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9585 }
9586 
9587 static int dev_xdp_detach_link(struct net_device *dev,
9588 			       struct netlink_ext_ack *extack,
9589 			       struct bpf_xdp_link *link)
9590 {
9591 	enum bpf_xdp_mode mode;
9592 	bpf_op_t bpf_op;
9593 
9594 	ASSERT_RTNL();
9595 
9596 	mode = dev_xdp_mode(dev, link->flags);
9597 	if (dev_xdp_link(dev, mode) != link)
9598 		return -EINVAL;
9599 
9600 	bpf_op = dev_xdp_bpf_op(dev, mode);
9601 	WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9602 	dev_xdp_set_link(dev, mode, NULL);
9603 	return 0;
9604 }
9605 
9606 static void bpf_xdp_link_release(struct bpf_link *link)
9607 {
9608 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9609 
9610 	rtnl_lock();
9611 
9612 	/* if racing with net_device's tear down, xdp_link->dev might be
9613 	 * already NULL, in which case link was already auto-detached
9614 	 */
9615 	if (xdp_link->dev) {
9616 		WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9617 		xdp_link->dev = NULL;
9618 	}
9619 
9620 	rtnl_unlock();
9621 }
9622 
9623 static int bpf_xdp_link_detach(struct bpf_link *link)
9624 {
9625 	bpf_xdp_link_release(link);
9626 	return 0;
9627 }
9628 
9629 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9630 {
9631 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9632 
9633 	kfree(xdp_link);
9634 }
9635 
9636 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9637 				     struct seq_file *seq)
9638 {
9639 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9640 	u32 ifindex = 0;
9641 
9642 	rtnl_lock();
9643 	if (xdp_link->dev)
9644 		ifindex = xdp_link->dev->ifindex;
9645 	rtnl_unlock();
9646 
9647 	seq_printf(seq, "ifindex:\t%u\n", ifindex);
9648 }
9649 
9650 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9651 				       struct bpf_link_info *info)
9652 {
9653 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9654 	u32 ifindex = 0;
9655 
9656 	rtnl_lock();
9657 	if (xdp_link->dev)
9658 		ifindex = xdp_link->dev->ifindex;
9659 	rtnl_unlock();
9660 
9661 	info->xdp.ifindex = ifindex;
9662 	return 0;
9663 }
9664 
9665 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9666 			       struct bpf_prog *old_prog)
9667 {
9668 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9669 	enum bpf_xdp_mode mode;
9670 	bpf_op_t bpf_op;
9671 	int err = 0;
9672 
9673 	rtnl_lock();
9674 
9675 	/* link might have been auto-released already, so fail */
9676 	if (!xdp_link->dev) {
9677 		err = -ENOLINK;
9678 		goto out_unlock;
9679 	}
9680 
9681 	if (old_prog && link->prog != old_prog) {
9682 		err = -EPERM;
9683 		goto out_unlock;
9684 	}
9685 	old_prog = link->prog;
9686 	if (old_prog->type != new_prog->type ||
9687 	    old_prog->expected_attach_type != new_prog->expected_attach_type) {
9688 		err = -EINVAL;
9689 		goto out_unlock;
9690 	}
9691 
9692 	if (old_prog == new_prog) {
9693 		/* no-op, don't disturb drivers */
9694 		bpf_prog_put(new_prog);
9695 		goto out_unlock;
9696 	}
9697 
9698 	mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9699 	bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9700 	err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9701 			      xdp_link->flags, new_prog);
9702 	if (err)
9703 		goto out_unlock;
9704 
9705 	old_prog = xchg(&link->prog, new_prog);
9706 	bpf_prog_put(old_prog);
9707 
9708 out_unlock:
9709 	rtnl_unlock();
9710 	return err;
9711 }
9712 
9713 static const struct bpf_link_ops bpf_xdp_link_lops = {
9714 	.release = bpf_xdp_link_release,
9715 	.dealloc = bpf_xdp_link_dealloc,
9716 	.detach = bpf_xdp_link_detach,
9717 	.show_fdinfo = bpf_xdp_link_show_fdinfo,
9718 	.fill_link_info = bpf_xdp_link_fill_link_info,
9719 	.update_prog = bpf_xdp_link_update,
9720 };
9721 
9722 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9723 {
9724 	struct net *net = current->nsproxy->net_ns;
9725 	struct bpf_link_primer link_primer;
9726 	struct netlink_ext_ack extack = {};
9727 	struct bpf_xdp_link *link;
9728 	struct net_device *dev;
9729 	int err, fd;
9730 
9731 	rtnl_lock();
9732 	dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9733 	if (!dev) {
9734 		rtnl_unlock();
9735 		return -EINVAL;
9736 	}
9737 
9738 	link = kzalloc(sizeof(*link), GFP_USER);
9739 	if (!link) {
9740 		err = -ENOMEM;
9741 		goto unlock;
9742 	}
9743 
9744 	bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9745 	link->dev = dev;
9746 	link->flags = attr->link_create.flags;
9747 
9748 	err = bpf_link_prime(&link->link, &link_primer);
9749 	if (err) {
9750 		kfree(link);
9751 		goto unlock;
9752 	}
9753 
9754 	err = dev_xdp_attach_link(dev, &extack, link);
9755 	rtnl_unlock();
9756 
9757 	if (err) {
9758 		link->dev = NULL;
9759 		bpf_link_cleanup(&link_primer);
9760 		trace_bpf_xdp_link_attach_failed(extack._msg);
9761 		goto out_put_dev;
9762 	}
9763 
9764 	fd = bpf_link_settle(&link_primer);
9765 	/* link itself doesn't hold dev's refcnt to not complicate shutdown */
9766 	dev_put(dev);
9767 	return fd;
9768 
9769 unlock:
9770 	rtnl_unlock();
9771 
9772 out_put_dev:
9773 	dev_put(dev);
9774 	return err;
9775 }
9776 
9777 /**
9778  *	dev_change_xdp_fd - set or clear a bpf program for a device rx path
9779  *	@dev: device
9780  *	@extack: netlink extended ack
9781  *	@fd: new program fd or negative value to clear
9782  *	@expected_fd: old program fd that userspace expects to replace or clear
9783  *	@flags: xdp-related flags
9784  *
9785  *	Set or clear a bpf program for a device
9786  */
9787 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9788 		      int fd, int expected_fd, u32 flags)
9789 {
9790 	enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9791 	struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9792 	int err;
9793 
9794 	ASSERT_RTNL();
9795 
9796 	if (fd >= 0) {
9797 		new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9798 						 mode != XDP_MODE_SKB);
9799 		if (IS_ERR(new_prog))
9800 			return PTR_ERR(new_prog);
9801 	}
9802 
9803 	if (expected_fd >= 0) {
9804 		old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9805 						 mode != XDP_MODE_SKB);
9806 		if (IS_ERR(old_prog)) {
9807 			err = PTR_ERR(old_prog);
9808 			old_prog = NULL;
9809 			goto err_out;
9810 		}
9811 	}
9812 
9813 	err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9814 
9815 err_out:
9816 	if (err && new_prog)
9817 		bpf_prog_put(new_prog);
9818 	if (old_prog)
9819 		bpf_prog_put(old_prog);
9820 	return err;
9821 }
9822 
9823 /**
9824  * dev_index_reserve() - allocate an ifindex in a namespace
9825  * @net: the applicable net namespace
9826  * @ifindex: requested ifindex, pass %0 to get one allocated
9827  *
9828  * Allocate a ifindex for a new device. Caller must either use the ifindex
9829  * to store the device (via list_netdevice()) or call dev_index_release()
9830  * to give the index up.
9831  *
9832  * Return: a suitable unique value for a new device interface number or -errno.
9833  */
9834 static int dev_index_reserve(struct net *net, u32 ifindex)
9835 {
9836 	int err;
9837 
9838 	if (ifindex > INT_MAX) {
9839 		DEBUG_NET_WARN_ON_ONCE(1);
9840 		return -EINVAL;
9841 	}
9842 
9843 	if (!ifindex)
9844 		err = xa_alloc_cyclic(&net->dev_by_index, &ifindex, NULL,
9845 				      xa_limit_31b, &net->ifindex, GFP_KERNEL);
9846 	else
9847 		err = xa_insert(&net->dev_by_index, ifindex, NULL, GFP_KERNEL);
9848 	if (err < 0)
9849 		return err;
9850 
9851 	return ifindex;
9852 }
9853 
9854 static void dev_index_release(struct net *net, int ifindex)
9855 {
9856 	/* Expect only unused indexes, unlist_netdevice() removes the used */
9857 	WARN_ON(xa_erase(&net->dev_by_index, ifindex));
9858 }
9859 
9860 /* Delayed registration/unregisteration */
9861 LIST_HEAD(net_todo_list);
9862 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9863 atomic_t dev_unreg_count = ATOMIC_INIT(0);
9864 
9865 static void net_set_todo(struct net_device *dev)
9866 {
9867 	list_add_tail(&dev->todo_list, &net_todo_list);
9868 }
9869 
9870 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9871 	struct net_device *upper, netdev_features_t features)
9872 {
9873 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9874 	netdev_features_t feature;
9875 	int feature_bit;
9876 
9877 	for_each_netdev_feature(upper_disables, feature_bit) {
9878 		feature = __NETIF_F_BIT(feature_bit);
9879 		if (!(upper->wanted_features & feature)
9880 		    && (features & feature)) {
9881 			netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9882 				   &feature, upper->name);
9883 			features &= ~feature;
9884 		}
9885 	}
9886 
9887 	return features;
9888 }
9889 
9890 static void netdev_sync_lower_features(struct net_device *upper,
9891 	struct net_device *lower, netdev_features_t features)
9892 {
9893 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9894 	netdev_features_t feature;
9895 	int feature_bit;
9896 
9897 	for_each_netdev_feature(upper_disables, feature_bit) {
9898 		feature = __NETIF_F_BIT(feature_bit);
9899 		if (!(features & feature) && (lower->features & feature)) {
9900 			netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9901 				   &feature, lower->name);
9902 			lower->wanted_features &= ~feature;
9903 			__netdev_update_features(lower);
9904 
9905 			if (unlikely(lower->features & feature))
9906 				netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9907 					    &feature, lower->name);
9908 			else
9909 				netdev_features_change(lower);
9910 		}
9911 	}
9912 }
9913 
9914 static netdev_features_t netdev_fix_features(struct net_device *dev,
9915 	netdev_features_t features)
9916 {
9917 	/* Fix illegal checksum combinations */
9918 	if ((features & NETIF_F_HW_CSUM) &&
9919 	    (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9920 		netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9921 		features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9922 	}
9923 
9924 	/* TSO requires that SG is present as well. */
9925 	if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9926 		netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9927 		features &= ~NETIF_F_ALL_TSO;
9928 	}
9929 
9930 	if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9931 					!(features & NETIF_F_IP_CSUM)) {
9932 		netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9933 		features &= ~NETIF_F_TSO;
9934 		features &= ~NETIF_F_TSO_ECN;
9935 	}
9936 
9937 	if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9938 					 !(features & NETIF_F_IPV6_CSUM)) {
9939 		netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9940 		features &= ~NETIF_F_TSO6;
9941 	}
9942 
9943 	/* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9944 	if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9945 		features &= ~NETIF_F_TSO_MANGLEID;
9946 
9947 	/* TSO ECN requires that TSO is present as well. */
9948 	if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9949 		features &= ~NETIF_F_TSO_ECN;
9950 
9951 	/* Software GSO depends on SG. */
9952 	if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9953 		netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9954 		features &= ~NETIF_F_GSO;
9955 	}
9956 
9957 	/* GSO partial features require GSO partial be set */
9958 	if ((features & dev->gso_partial_features) &&
9959 	    !(features & NETIF_F_GSO_PARTIAL)) {
9960 		netdev_dbg(dev,
9961 			   "Dropping partially supported GSO features since no GSO partial.\n");
9962 		features &= ~dev->gso_partial_features;
9963 	}
9964 
9965 	if (!(features & NETIF_F_RXCSUM)) {
9966 		/* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9967 		 * successfully merged by hardware must also have the
9968 		 * checksum verified by hardware.  If the user does not
9969 		 * want to enable RXCSUM, logically, we should disable GRO_HW.
9970 		 */
9971 		if (features & NETIF_F_GRO_HW) {
9972 			netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9973 			features &= ~NETIF_F_GRO_HW;
9974 		}
9975 	}
9976 
9977 	/* LRO/HW-GRO features cannot be combined with RX-FCS */
9978 	if (features & NETIF_F_RXFCS) {
9979 		if (features & NETIF_F_LRO) {
9980 			netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9981 			features &= ~NETIF_F_LRO;
9982 		}
9983 
9984 		if (features & NETIF_F_GRO_HW) {
9985 			netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9986 			features &= ~NETIF_F_GRO_HW;
9987 		}
9988 	}
9989 
9990 	if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9991 		netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9992 		features &= ~NETIF_F_LRO;
9993 	}
9994 
9995 	if (features & NETIF_F_HW_TLS_TX) {
9996 		bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9997 			(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9998 		bool hw_csum = features & NETIF_F_HW_CSUM;
9999 
10000 		if (!ip_csum && !hw_csum) {
10001 			netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
10002 			features &= ~NETIF_F_HW_TLS_TX;
10003 		}
10004 	}
10005 
10006 	if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
10007 		netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
10008 		features &= ~NETIF_F_HW_TLS_RX;
10009 	}
10010 
10011 	return features;
10012 }
10013 
10014 int __netdev_update_features(struct net_device *dev)
10015 {
10016 	struct net_device *upper, *lower;
10017 	netdev_features_t features;
10018 	struct list_head *iter;
10019 	int err = -1;
10020 
10021 	ASSERT_RTNL();
10022 
10023 	features = netdev_get_wanted_features(dev);
10024 
10025 	if (dev->netdev_ops->ndo_fix_features)
10026 		features = dev->netdev_ops->ndo_fix_features(dev, features);
10027 
10028 	/* driver might be less strict about feature dependencies */
10029 	features = netdev_fix_features(dev, features);
10030 
10031 	/* some features can't be enabled if they're off on an upper device */
10032 	netdev_for_each_upper_dev_rcu(dev, upper, iter)
10033 		features = netdev_sync_upper_features(dev, upper, features);
10034 
10035 	if (dev->features == features)
10036 		goto sync_lower;
10037 
10038 	netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
10039 		&dev->features, &features);
10040 
10041 	if (dev->netdev_ops->ndo_set_features)
10042 		err = dev->netdev_ops->ndo_set_features(dev, features);
10043 	else
10044 		err = 0;
10045 
10046 	if (unlikely(err < 0)) {
10047 		netdev_err(dev,
10048 			"set_features() failed (%d); wanted %pNF, left %pNF\n",
10049 			err, &features, &dev->features);
10050 		/* return non-0 since some features might have changed and
10051 		 * it's better to fire a spurious notification than miss it
10052 		 */
10053 		return -1;
10054 	}
10055 
10056 sync_lower:
10057 	/* some features must be disabled on lower devices when disabled
10058 	 * on an upper device (think: bonding master or bridge)
10059 	 */
10060 	netdev_for_each_lower_dev(dev, lower, iter)
10061 		netdev_sync_lower_features(dev, lower, features);
10062 
10063 	if (!err) {
10064 		netdev_features_t diff = features ^ dev->features;
10065 
10066 		if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
10067 			/* udp_tunnel_{get,drop}_rx_info both need
10068 			 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
10069 			 * device, or they won't do anything.
10070 			 * Thus we need to update dev->features
10071 			 * *before* calling udp_tunnel_get_rx_info,
10072 			 * but *after* calling udp_tunnel_drop_rx_info.
10073 			 */
10074 			if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
10075 				dev->features = features;
10076 				udp_tunnel_get_rx_info(dev);
10077 			} else {
10078 				udp_tunnel_drop_rx_info(dev);
10079 			}
10080 		}
10081 
10082 		if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
10083 			if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
10084 				dev->features = features;
10085 				err |= vlan_get_rx_ctag_filter_info(dev);
10086 			} else {
10087 				vlan_drop_rx_ctag_filter_info(dev);
10088 			}
10089 		}
10090 
10091 		if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
10092 			if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
10093 				dev->features = features;
10094 				err |= vlan_get_rx_stag_filter_info(dev);
10095 			} else {
10096 				vlan_drop_rx_stag_filter_info(dev);
10097 			}
10098 		}
10099 
10100 		dev->features = features;
10101 	}
10102 
10103 	return err < 0 ? 0 : 1;
10104 }
10105 
10106 /**
10107  *	netdev_update_features - recalculate device features
10108  *	@dev: the device to check
10109  *
10110  *	Recalculate dev->features set and send notifications if it
10111  *	has changed. Should be called after driver or hardware dependent
10112  *	conditions might have changed that influence the features.
10113  */
10114 void netdev_update_features(struct net_device *dev)
10115 {
10116 	if (__netdev_update_features(dev))
10117 		netdev_features_change(dev);
10118 }
10119 EXPORT_SYMBOL(netdev_update_features);
10120 
10121 /**
10122  *	netdev_change_features - recalculate device features
10123  *	@dev: the device to check
10124  *
10125  *	Recalculate dev->features set and send notifications even
10126  *	if they have not changed. Should be called instead of
10127  *	netdev_update_features() if also dev->vlan_features might
10128  *	have changed to allow the changes to be propagated to stacked
10129  *	VLAN devices.
10130  */
10131 void netdev_change_features(struct net_device *dev)
10132 {
10133 	__netdev_update_features(dev);
10134 	netdev_features_change(dev);
10135 }
10136 EXPORT_SYMBOL(netdev_change_features);
10137 
10138 /**
10139  *	netif_stacked_transfer_operstate -	transfer operstate
10140  *	@rootdev: the root or lower level device to transfer state from
10141  *	@dev: the device to transfer operstate to
10142  *
10143  *	Transfer operational state from root to device. This is normally
10144  *	called when a stacking relationship exists between the root
10145  *	device and the device(a leaf device).
10146  */
10147 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
10148 					struct net_device *dev)
10149 {
10150 	if (rootdev->operstate == IF_OPER_DORMANT)
10151 		netif_dormant_on(dev);
10152 	else
10153 		netif_dormant_off(dev);
10154 
10155 	if (rootdev->operstate == IF_OPER_TESTING)
10156 		netif_testing_on(dev);
10157 	else
10158 		netif_testing_off(dev);
10159 
10160 	if (netif_carrier_ok(rootdev))
10161 		netif_carrier_on(dev);
10162 	else
10163 		netif_carrier_off(dev);
10164 }
10165 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
10166 
10167 static int netif_alloc_rx_queues(struct net_device *dev)
10168 {
10169 	unsigned int i, count = dev->num_rx_queues;
10170 	struct netdev_rx_queue *rx;
10171 	size_t sz = count * sizeof(*rx);
10172 	int err = 0;
10173 
10174 	BUG_ON(count < 1);
10175 
10176 	rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10177 	if (!rx)
10178 		return -ENOMEM;
10179 
10180 	dev->_rx = rx;
10181 
10182 	for (i = 0; i < count; i++) {
10183 		rx[i].dev = dev;
10184 
10185 		/* XDP RX-queue setup */
10186 		err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
10187 		if (err < 0)
10188 			goto err_rxq_info;
10189 	}
10190 	return 0;
10191 
10192 err_rxq_info:
10193 	/* Rollback successful reg's and free other resources */
10194 	while (i--)
10195 		xdp_rxq_info_unreg(&rx[i].xdp_rxq);
10196 	kvfree(dev->_rx);
10197 	dev->_rx = NULL;
10198 	return err;
10199 }
10200 
10201 static void netif_free_rx_queues(struct net_device *dev)
10202 {
10203 	unsigned int i, count = dev->num_rx_queues;
10204 
10205 	/* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
10206 	if (!dev->_rx)
10207 		return;
10208 
10209 	for (i = 0; i < count; i++)
10210 		xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
10211 
10212 	kvfree(dev->_rx);
10213 }
10214 
10215 static void netdev_init_one_queue(struct net_device *dev,
10216 				  struct netdev_queue *queue, void *_unused)
10217 {
10218 	/* Initialize queue lock */
10219 	spin_lock_init(&queue->_xmit_lock);
10220 	netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
10221 	queue->xmit_lock_owner = -1;
10222 	netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
10223 	queue->dev = dev;
10224 #ifdef CONFIG_BQL
10225 	dql_init(&queue->dql, HZ);
10226 #endif
10227 }
10228 
10229 static void netif_free_tx_queues(struct net_device *dev)
10230 {
10231 	kvfree(dev->_tx);
10232 }
10233 
10234 static int netif_alloc_netdev_queues(struct net_device *dev)
10235 {
10236 	unsigned int count = dev->num_tx_queues;
10237 	struct netdev_queue *tx;
10238 	size_t sz = count * sizeof(*tx);
10239 
10240 	if (count < 1 || count > 0xffff)
10241 		return -EINVAL;
10242 
10243 	tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10244 	if (!tx)
10245 		return -ENOMEM;
10246 
10247 	dev->_tx = tx;
10248 
10249 	netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
10250 	spin_lock_init(&dev->tx_global_lock);
10251 
10252 	return 0;
10253 }
10254 
10255 void netif_tx_stop_all_queues(struct net_device *dev)
10256 {
10257 	unsigned int i;
10258 
10259 	for (i = 0; i < dev->num_tx_queues; i++) {
10260 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
10261 
10262 		netif_tx_stop_queue(txq);
10263 	}
10264 }
10265 EXPORT_SYMBOL(netif_tx_stop_all_queues);
10266 
10267 static int netdev_do_alloc_pcpu_stats(struct net_device *dev)
10268 {
10269 	void __percpu *v;
10270 
10271 	/* Drivers implementing ndo_get_peer_dev must support tstat
10272 	 * accounting, so that skb_do_redirect() can bump the dev's
10273 	 * RX stats upon network namespace switch.
10274 	 */
10275 	if (dev->netdev_ops->ndo_get_peer_dev &&
10276 	    dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS)
10277 		return -EOPNOTSUPP;
10278 
10279 	switch (dev->pcpu_stat_type) {
10280 	case NETDEV_PCPU_STAT_NONE:
10281 		return 0;
10282 	case NETDEV_PCPU_STAT_LSTATS:
10283 		v = dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats);
10284 		break;
10285 	case NETDEV_PCPU_STAT_TSTATS:
10286 		v = dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
10287 		break;
10288 	case NETDEV_PCPU_STAT_DSTATS:
10289 		v = dev->dstats = netdev_alloc_pcpu_stats(struct pcpu_dstats);
10290 		break;
10291 	default:
10292 		return -EINVAL;
10293 	}
10294 
10295 	return v ? 0 : -ENOMEM;
10296 }
10297 
10298 static void netdev_do_free_pcpu_stats(struct net_device *dev)
10299 {
10300 	switch (dev->pcpu_stat_type) {
10301 	case NETDEV_PCPU_STAT_NONE:
10302 		return;
10303 	case NETDEV_PCPU_STAT_LSTATS:
10304 		free_percpu(dev->lstats);
10305 		break;
10306 	case NETDEV_PCPU_STAT_TSTATS:
10307 		free_percpu(dev->tstats);
10308 		break;
10309 	case NETDEV_PCPU_STAT_DSTATS:
10310 		free_percpu(dev->dstats);
10311 		break;
10312 	}
10313 }
10314 
10315 /**
10316  * register_netdevice() - register a network device
10317  * @dev: device to register
10318  *
10319  * Take a prepared network device structure and make it externally accessible.
10320  * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
10321  * Callers must hold the rtnl lock - you may want register_netdev()
10322  * instead of this.
10323  */
10324 int register_netdevice(struct net_device *dev)
10325 {
10326 	int ret;
10327 	struct net *net = dev_net(dev);
10328 
10329 	BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
10330 		     NETDEV_FEATURE_COUNT);
10331 	BUG_ON(dev_boot_phase);
10332 	ASSERT_RTNL();
10333 
10334 	might_sleep();
10335 
10336 	/* When net_device's are persistent, this will be fatal. */
10337 	BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
10338 	BUG_ON(!net);
10339 
10340 	ret = ethtool_check_ops(dev->ethtool_ops);
10341 	if (ret)
10342 		return ret;
10343 
10344 	/* rss ctx ID 0 is reserved for the default context, start from 1 */
10345 	xa_init_flags(&dev->ethtool->rss_ctx, XA_FLAGS_ALLOC1);
10346 	mutex_init(&dev->ethtool->rss_lock);
10347 
10348 	spin_lock_init(&dev->addr_list_lock);
10349 	netdev_set_addr_lockdep_class(dev);
10350 
10351 	ret = dev_get_valid_name(net, dev, dev->name);
10352 	if (ret < 0)
10353 		goto out;
10354 
10355 	ret = -ENOMEM;
10356 	dev->name_node = netdev_name_node_head_alloc(dev);
10357 	if (!dev->name_node)
10358 		goto out;
10359 
10360 	/* Init, if this function is available */
10361 	if (dev->netdev_ops->ndo_init) {
10362 		ret = dev->netdev_ops->ndo_init(dev);
10363 		if (ret) {
10364 			if (ret > 0)
10365 				ret = -EIO;
10366 			goto err_free_name;
10367 		}
10368 	}
10369 
10370 	if (((dev->hw_features | dev->features) &
10371 	     NETIF_F_HW_VLAN_CTAG_FILTER) &&
10372 	    (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10373 	     !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10374 		netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10375 		ret = -EINVAL;
10376 		goto err_uninit;
10377 	}
10378 
10379 	ret = netdev_do_alloc_pcpu_stats(dev);
10380 	if (ret)
10381 		goto err_uninit;
10382 
10383 	ret = dev_index_reserve(net, dev->ifindex);
10384 	if (ret < 0)
10385 		goto err_free_pcpu;
10386 	dev->ifindex = ret;
10387 
10388 	/* Transfer changeable features to wanted_features and enable
10389 	 * software offloads (GSO and GRO).
10390 	 */
10391 	dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10392 	dev->features |= NETIF_F_SOFT_FEATURES;
10393 
10394 	if (dev->udp_tunnel_nic_info) {
10395 		dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10396 		dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10397 	}
10398 
10399 	dev->wanted_features = dev->features & dev->hw_features;
10400 
10401 	if (!(dev->flags & IFF_LOOPBACK))
10402 		dev->hw_features |= NETIF_F_NOCACHE_COPY;
10403 
10404 	/* If IPv4 TCP segmentation offload is supported we should also
10405 	 * allow the device to enable segmenting the frame with the option
10406 	 * of ignoring a static IP ID value.  This doesn't enable the
10407 	 * feature itself but allows the user to enable it later.
10408 	 */
10409 	if (dev->hw_features & NETIF_F_TSO)
10410 		dev->hw_features |= NETIF_F_TSO_MANGLEID;
10411 	if (dev->vlan_features & NETIF_F_TSO)
10412 		dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10413 	if (dev->mpls_features & NETIF_F_TSO)
10414 		dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10415 	if (dev->hw_enc_features & NETIF_F_TSO)
10416 		dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10417 
10418 	/* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10419 	 */
10420 	dev->vlan_features |= NETIF_F_HIGHDMA;
10421 
10422 	/* Make NETIF_F_SG inheritable to tunnel devices.
10423 	 */
10424 	dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10425 
10426 	/* Make NETIF_F_SG inheritable to MPLS.
10427 	 */
10428 	dev->mpls_features |= NETIF_F_SG;
10429 
10430 	ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10431 	ret = notifier_to_errno(ret);
10432 	if (ret)
10433 		goto err_ifindex_release;
10434 
10435 	ret = netdev_register_kobject(dev);
10436 
10437 	WRITE_ONCE(dev->reg_state, ret ? NETREG_UNREGISTERED : NETREG_REGISTERED);
10438 
10439 	if (ret)
10440 		goto err_uninit_notify;
10441 
10442 	__netdev_update_features(dev);
10443 
10444 	/*
10445 	 *	Default initial state at registry is that the
10446 	 *	device is present.
10447 	 */
10448 
10449 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10450 
10451 	linkwatch_init_dev(dev);
10452 
10453 	dev_init_scheduler(dev);
10454 
10455 	netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
10456 	list_netdevice(dev);
10457 
10458 	add_device_randomness(dev->dev_addr, dev->addr_len);
10459 
10460 	/* If the device has permanent device address, driver should
10461 	 * set dev_addr and also addr_assign_type should be set to
10462 	 * NET_ADDR_PERM (default value).
10463 	 */
10464 	if (dev->addr_assign_type == NET_ADDR_PERM)
10465 		memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10466 
10467 	/* Notify protocols, that a new device appeared. */
10468 	ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10469 	ret = notifier_to_errno(ret);
10470 	if (ret) {
10471 		/* Expect explicit free_netdev() on failure */
10472 		dev->needs_free_netdev = false;
10473 		unregister_netdevice_queue(dev, NULL);
10474 		goto out;
10475 	}
10476 	/*
10477 	 *	Prevent userspace races by waiting until the network
10478 	 *	device is fully setup before sending notifications.
10479 	 */
10480 	if (!dev->rtnl_link_ops ||
10481 	    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10482 		rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
10483 
10484 out:
10485 	return ret;
10486 
10487 err_uninit_notify:
10488 	call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10489 err_ifindex_release:
10490 	dev_index_release(net, dev->ifindex);
10491 err_free_pcpu:
10492 	netdev_do_free_pcpu_stats(dev);
10493 err_uninit:
10494 	if (dev->netdev_ops->ndo_uninit)
10495 		dev->netdev_ops->ndo_uninit(dev);
10496 	if (dev->priv_destructor)
10497 		dev->priv_destructor(dev);
10498 err_free_name:
10499 	netdev_name_node_free(dev->name_node);
10500 	goto out;
10501 }
10502 EXPORT_SYMBOL(register_netdevice);
10503 
10504 /* Initialize the core of a dummy net device.
10505  * This is useful if you are calling this function after alloc_netdev(),
10506  * since it does not memset the net_device fields.
10507  */
10508 static void init_dummy_netdev_core(struct net_device *dev)
10509 {
10510 	/* make sure we BUG if trying to hit standard
10511 	 * register/unregister code path
10512 	 */
10513 	dev->reg_state = NETREG_DUMMY;
10514 
10515 	/* NAPI wants this */
10516 	INIT_LIST_HEAD(&dev->napi_list);
10517 
10518 	/* a dummy interface is started by default */
10519 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10520 	set_bit(__LINK_STATE_START, &dev->state);
10521 
10522 	/* napi_busy_loop stats accounting wants this */
10523 	dev_net_set(dev, &init_net);
10524 
10525 	/* Note : We dont allocate pcpu_refcnt for dummy devices,
10526 	 * because users of this 'device' dont need to change
10527 	 * its refcount.
10528 	 */
10529 }
10530 
10531 /**
10532  *	init_dummy_netdev	- init a dummy network device for NAPI
10533  *	@dev: device to init
10534  *
10535  *	This takes a network device structure and initializes the minimum
10536  *	amount of fields so it can be used to schedule NAPI polls without
10537  *	registering a full blown interface. This is to be used by drivers
10538  *	that need to tie several hardware interfaces to a single NAPI
10539  *	poll scheduler due to HW limitations.
10540  */
10541 void init_dummy_netdev(struct net_device *dev)
10542 {
10543 	/* Clear everything. Note we don't initialize spinlocks
10544 	 * as they aren't supposed to be taken by any of the
10545 	 * NAPI code and this dummy netdev is supposed to be
10546 	 * only ever used for NAPI polls
10547 	 */
10548 	memset(dev, 0, sizeof(struct net_device));
10549 	init_dummy_netdev_core(dev);
10550 }
10551 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10552 
10553 /**
10554  *	register_netdev	- register a network device
10555  *	@dev: device to register
10556  *
10557  *	Take a completed network device structure and add it to the kernel
10558  *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10559  *	chain. 0 is returned on success. A negative errno code is returned
10560  *	on a failure to set up the device, or if the name is a duplicate.
10561  *
10562  *	This is a wrapper around register_netdevice that takes the rtnl semaphore
10563  *	and expands the device name if you passed a format string to
10564  *	alloc_netdev.
10565  */
10566 int register_netdev(struct net_device *dev)
10567 {
10568 	int err;
10569 
10570 	if (rtnl_lock_killable())
10571 		return -EINTR;
10572 	err = register_netdevice(dev);
10573 	rtnl_unlock();
10574 	return err;
10575 }
10576 EXPORT_SYMBOL(register_netdev);
10577 
10578 int netdev_refcnt_read(const struct net_device *dev)
10579 {
10580 #ifdef CONFIG_PCPU_DEV_REFCNT
10581 	int i, refcnt = 0;
10582 
10583 	for_each_possible_cpu(i)
10584 		refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10585 	return refcnt;
10586 #else
10587 	return refcount_read(&dev->dev_refcnt);
10588 #endif
10589 }
10590 EXPORT_SYMBOL(netdev_refcnt_read);
10591 
10592 int netdev_unregister_timeout_secs __read_mostly = 10;
10593 
10594 #define WAIT_REFS_MIN_MSECS 1
10595 #define WAIT_REFS_MAX_MSECS 250
10596 /**
10597  * netdev_wait_allrefs_any - wait until all references are gone.
10598  * @list: list of net_devices to wait on
10599  *
10600  * This is called when unregistering network devices.
10601  *
10602  * Any protocol or device that holds a reference should register
10603  * for netdevice notification, and cleanup and put back the
10604  * reference if they receive an UNREGISTER event.
10605  * We can get stuck here if buggy protocols don't correctly
10606  * call dev_put.
10607  */
10608 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
10609 {
10610 	unsigned long rebroadcast_time, warning_time;
10611 	struct net_device *dev;
10612 	int wait = 0;
10613 
10614 	rebroadcast_time = warning_time = jiffies;
10615 
10616 	list_for_each_entry(dev, list, todo_list)
10617 		if (netdev_refcnt_read(dev) == 1)
10618 			return dev;
10619 
10620 	while (true) {
10621 		if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10622 			rtnl_lock();
10623 
10624 			/* Rebroadcast unregister notification */
10625 			list_for_each_entry(dev, list, todo_list)
10626 				call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10627 
10628 			__rtnl_unlock();
10629 			rcu_barrier();
10630 			rtnl_lock();
10631 
10632 			list_for_each_entry(dev, list, todo_list)
10633 				if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10634 					     &dev->state)) {
10635 					/* We must not have linkwatch events
10636 					 * pending on unregister. If this
10637 					 * happens, we simply run the queue
10638 					 * unscheduled, resulting in a noop
10639 					 * for this device.
10640 					 */
10641 					linkwatch_run_queue();
10642 					break;
10643 				}
10644 
10645 			__rtnl_unlock();
10646 
10647 			rebroadcast_time = jiffies;
10648 		}
10649 
10650 		rcu_barrier();
10651 
10652 		if (!wait) {
10653 			wait = WAIT_REFS_MIN_MSECS;
10654 		} else {
10655 			msleep(wait);
10656 			wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10657 		}
10658 
10659 		list_for_each_entry(dev, list, todo_list)
10660 			if (netdev_refcnt_read(dev) == 1)
10661 				return dev;
10662 
10663 		if (time_after(jiffies, warning_time +
10664 			       READ_ONCE(netdev_unregister_timeout_secs) * HZ)) {
10665 			list_for_each_entry(dev, list, todo_list) {
10666 				pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10667 					 dev->name, netdev_refcnt_read(dev));
10668 				ref_tracker_dir_print(&dev->refcnt_tracker, 10);
10669 			}
10670 
10671 			warning_time = jiffies;
10672 		}
10673 	}
10674 }
10675 
10676 /* The sequence is:
10677  *
10678  *	rtnl_lock();
10679  *	...
10680  *	register_netdevice(x1);
10681  *	register_netdevice(x2);
10682  *	...
10683  *	unregister_netdevice(y1);
10684  *	unregister_netdevice(y2);
10685  *      ...
10686  *	rtnl_unlock();
10687  *	free_netdev(y1);
10688  *	free_netdev(y2);
10689  *
10690  * We are invoked by rtnl_unlock().
10691  * This allows us to deal with problems:
10692  * 1) We can delete sysfs objects which invoke hotplug
10693  *    without deadlocking with linkwatch via keventd.
10694  * 2) Since we run with the RTNL semaphore not held, we can sleep
10695  *    safely in order to wait for the netdev refcnt to drop to zero.
10696  *
10697  * We must not return until all unregister events added during
10698  * the interval the lock was held have been completed.
10699  */
10700 void netdev_run_todo(void)
10701 {
10702 	struct net_device *dev, *tmp;
10703 	struct list_head list;
10704 	int cnt;
10705 #ifdef CONFIG_LOCKDEP
10706 	struct list_head unlink_list;
10707 
10708 	list_replace_init(&net_unlink_list, &unlink_list);
10709 
10710 	while (!list_empty(&unlink_list)) {
10711 		struct net_device *dev = list_first_entry(&unlink_list,
10712 							  struct net_device,
10713 							  unlink_list);
10714 		list_del_init(&dev->unlink_list);
10715 		dev->nested_level = dev->lower_level - 1;
10716 	}
10717 #endif
10718 
10719 	/* Snapshot list, allow later requests */
10720 	list_replace_init(&net_todo_list, &list);
10721 
10722 	__rtnl_unlock();
10723 
10724 	/* Wait for rcu callbacks to finish before next phase */
10725 	if (!list_empty(&list))
10726 		rcu_barrier();
10727 
10728 	list_for_each_entry_safe(dev, tmp, &list, todo_list) {
10729 		if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10730 			netdev_WARN(dev, "run_todo but not unregistering\n");
10731 			list_del(&dev->todo_list);
10732 			continue;
10733 		}
10734 
10735 		WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERED);
10736 		linkwatch_sync_dev(dev);
10737 	}
10738 
10739 	cnt = 0;
10740 	while (!list_empty(&list)) {
10741 		dev = netdev_wait_allrefs_any(&list);
10742 		list_del(&dev->todo_list);
10743 
10744 		/* paranoia */
10745 		BUG_ON(netdev_refcnt_read(dev) != 1);
10746 		BUG_ON(!list_empty(&dev->ptype_all));
10747 		BUG_ON(!list_empty(&dev->ptype_specific));
10748 		WARN_ON(rcu_access_pointer(dev->ip_ptr));
10749 		WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10750 
10751 		netdev_do_free_pcpu_stats(dev);
10752 		if (dev->priv_destructor)
10753 			dev->priv_destructor(dev);
10754 		if (dev->needs_free_netdev)
10755 			free_netdev(dev);
10756 
10757 		cnt++;
10758 
10759 		/* Free network device */
10760 		kobject_put(&dev->dev.kobj);
10761 	}
10762 	if (cnt && atomic_sub_and_test(cnt, &dev_unreg_count))
10763 		wake_up(&netdev_unregistering_wq);
10764 }
10765 
10766 /* Collate per-cpu network dstats statistics
10767  *
10768  * Read per-cpu network statistics from dev->dstats and populate the related
10769  * fields in @s.
10770  */
10771 static void dev_fetch_dstats(struct rtnl_link_stats64 *s,
10772 			     const struct pcpu_dstats __percpu *dstats)
10773 {
10774 	int cpu;
10775 
10776 	for_each_possible_cpu(cpu) {
10777 		u64 rx_packets, rx_bytes, rx_drops;
10778 		u64 tx_packets, tx_bytes, tx_drops;
10779 		const struct pcpu_dstats *stats;
10780 		unsigned int start;
10781 
10782 		stats = per_cpu_ptr(dstats, cpu);
10783 		do {
10784 			start = u64_stats_fetch_begin(&stats->syncp);
10785 			rx_packets = u64_stats_read(&stats->rx_packets);
10786 			rx_bytes   = u64_stats_read(&stats->rx_bytes);
10787 			rx_drops   = u64_stats_read(&stats->rx_drops);
10788 			tx_packets = u64_stats_read(&stats->tx_packets);
10789 			tx_bytes   = u64_stats_read(&stats->tx_bytes);
10790 			tx_drops   = u64_stats_read(&stats->tx_drops);
10791 		} while (u64_stats_fetch_retry(&stats->syncp, start));
10792 
10793 		s->rx_packets += rx_packets;
10794 		s->rx_bytes   += rx_bytes;
10795 		s->rx_dropped += rx_drops;
10796 		s->tx_packets += tx_packets;
10797 		s->tx_bytes   += tx_bytes;
10798 		s->tx_dropped += tx_drops;
10799 	}
10800 }
10801 
10802 /* ndo_get_stats64 implementation for dtstats-based accounting.
10803  *
10804  * Populate @s from dev->stats and dev->dstats. This is used internally by the
10805  * core for NETDEV_PCPU_STAT_DSTAT-type stats collection.
10806  */
10807 static void dev_get_dstats64(const struct net_device *dev,
10808 			     struct rtnl_link_stats64 *s)
10809 {
10810 	netdev_stats_to_stats64(s, &dev->stats);
10811 	dev_fetch_dstats(s, dev->dstats);
10812 }
10813 
10814 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10815  * all the same fields in the same order as net_device_stats, with only
10816  * the type differing, but rtnl_link_stats64 may have additional fields
10817  * at the end for newer counters.
10818  */
10819 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10820 			     const struct net_device_stats *netdev_stats)
10821 {
10822 	size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
10823 	const atomic_long_t *src = (atomic_long_t *)netdev_stats;
10824 	u64 *dst = (u64 *)stats64;
10825 
10826 	BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10827 	for (i = 0; i < n; i++)
10828 		dst[i] = (unsigned long)atomic_long_read(&src[i]);
10829 	/* zero out counters that only exist in rtnl_link_stats64 */
10830 	memset((char *)stats64 + n * sizeof(u64), 0,
10831 	       sizeof(*stats64) - n * sizeof(u64));
10832 }
10833 EXPORT_SYMBOL(netdev_stats_to_stats64);
10834 
10835 static __cold struct net_device_core_stats __percpu *netdev_core_stats_alloc(
10836 		struct net_device *dev)
10837 {
10838 	struct net_device_core_stats __percpu *p;
10839 
10840 	p = alloc_percpu_gfp(struct net_device_core_stats,
10841 			     GFP_ATOMIC | __GFP_NOWARN);
10842 
10843 	if (p && cmpxchg(&dev->core_stats, NULL, p))
10844 		free_percpu(p);
10845 
10846 	/* This READ_ONCE() pairs with the cmpxchg() above */
10847 	return READ_ONCE(dev->core_stats);
10848 }
10849 
10850 noinline void netdev_core_stats_inc(struct net_device *dev, u32 offset)
10851 {
10852 	/* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10853 	struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats);
10854 	unsigned long __percpu *field;
10855 
10856 	if (unlikely(!p)) {
10857 		p = netdev_core_stats_alloc(dev);
10858 		if (!p)
10859 			return;
10860 	}
10861 
10862 	field = (__force unsigned long __percpu *)((__force void *)p + offset);
10863 	this_cpu_inc(*field);
10864 }
10865 EXPORT_SYMBOL_GPL(netdev_core_stats_inc);
10866 
10867 /**
10868  *	dev_get_stats	- get network device statistics
10869  *	@dev: device to get statistics from
10870  *	@storage: place to store stats
10871  *
10872  *	Get network statistics from device. Return @storage.
10873  *	The device driver may provide its own method by setting
10874  *	dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10875  *	otherwise the internal statistics structure is used.
10876  */
10877 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10878 					struct rtnl_link_stats64 *storage)
10879 {
10880 	const struct net_device_ops *ops = dev->netdev_ops;
10881 	const struct net_device_core_stats __percpu *p;
10882 
10883 	if (ops->ndo_get_stats64) {
10884 		memset(storage, 0, sizeof(*storage));
10885 		ops->ndo_get_stats64(dev, storage);
10886 	} else if (ops->ndo_get_stats) {
10887 		netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10888 	} else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_TSTATS) {
10889 		dev_get_tstats64(dev, storage);
10890 	} else if (dev->pcpu_stat_type == NETDEV_PCPU_STAT_DSTATS) {
10891 		dev_get_dstats64(dev, storage);
10892 	} else {
10893 		netdev_stats_to_stats64(storage, &dev->stats);
10894 	}
10895 
10896 	/* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10897 	p = READ_ONCE(dev->core_stats);
10898 	if (p) {
10899 		const struct net_device_core_stats *core_stats;
10900 		int i;
10901 
10902 		for_each_possible_cpu(i) {
10903 			core_stats = per_cpu_ptr(p, i);
10904 			storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
10905 			storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
10906 			storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
10907 			storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
10908 		}
10909 	}
10910 	return storage;
10911 }
10912 EXPORT_SYMBOL(dev_get_stats);
10913 
10914 /**
10915  *	dev_fetch_sw_netstats - get per-cpu network device statistics
10916  *	@s: place to store stats
10917  *	@netstats: per-cpu network stats to read from
10918  *
10919  *	Read per-cpu network statistics and populate the related fields in @s.
10920  */
10921 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10922 			   const struct pcpu_sw_netstats __percpu *netstats)
10923 {
10924 	int cpu;
10925 
10926 	for_each_possible_cpu(cpu) {
10927 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
10928 		const struct pcpu_sw_netstats *stats;
10929 		unsigned int start;
10930 
10931 		stats = per_cpu_ptr(netstats, cpu);
10932 		do {
10933 			start = u64_stats_fetch_begin(&stats->syncp);
10934 			rx_packets = u64_stats_read(&stats->rx_packets);
10935 			rx_bytes   = u64_stats_read(&stats->rx_bytes);
10936 			tx_packets = u64_stats_read(&stats->tx_packets);
10937 			tx_bytes   = u64_stats_read(&stats->tx_bytes);
10938 		} while (u64_stats_fetch_retry(&stats->syncp, start));
10939 
10940 		s->rx_packets += rx_packets;
10941 		s->rx_bytes   += rx_bytes;
10942 		s->tx_packets += tx_packets;
10943 		s->tx_bytes   += tx_bytes;
10944 	}
10945 }
10946 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10947 
10948 /**
10949  *	dev_get_tstats64 - ndo_get_stats64 implementation
10950  *	@dev: device to get statistics from
10951  *	@s: place to store stats
10952  *
10953  *	Populate @s from dev->stats and dev->tstats. Can be used as
10954  *	ndo_get_stats64() callback.
10955  */
10956 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10957 {
10958 	netdev_stats_to_stats64(s, &dev->stats);
10959 	dev_fetch_sw_netstats(s, dev->tstats);
10960 }
10961 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10962 
10963 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10964 {
10965 	struct netdev_queue *queue = dev_ingress_queue(dev);
10966 
10967 #ifdef CONFIG_NET_CLS_ACT
10968 	if (queue)
10969 		return queue;
10970 	queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10971 	if (!queue)
10972 		return NULL;
10973 	netdev_init_one_queue(dev, queue, NULL);
10974 	RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10975 	RCU_INIT_POINTER(queue->qdisc_sleeping, &noop_qdisc);
10976 	rcu_assign_pointer(dev->ingress_queue, queue);
10977 #endif
10978 	return queue;
10979 }
10980 
10981 static const struct ethtool_ops default_ethtool_ops;
10982 
10983 void netdev_set_default_ethtool_ops(struct net_device *dev,
10984 				    const struct ethtool_ops *ops)
10985 {
10986 	if (dev->ethtool_ops == &default_ethtool_ops)
10987 		dev->ethtool_ops = ops;
10988 }
10989 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10990 
10991 /**
10992  * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default
10993  * @dev: netdev to enable the IRQ coalescing on
10994  *
10995  * Sets a conservative default for SW IRQ coalescing. Users can use
10996  * sysfs attributes to override the default values.
10997  */
10998 void netdev_sw_irq_coalesce_default_on(struct net_device *dev)
10999 {
11000 	WARN_ON(dev->reg_state == NETREG_REGISTERED);
11001 
11002 	if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
11003 		dev->gro_flush_timeout = 20000;
11004 		dev->napi_defer_hard_irqs = 1;
11005 	}
11006 }
11007 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on);
11008 
11009 /**
11010  * alloc_netdev_mqs - allocate network device
11011  * @sizeof_priv: size of private data to allocate space for
11012  * @name: device name format string
11013  * @name_assign_type: origin of device name
11014  * @setup: callback to initialize device
11015  * @txqs: the number of TX subqueues to allocate
11016  * @rxqs: the number of RX subqueues to allocate
11017  *
11018  * Allocates a struct net_device with private data area for driver use
11019  * and performs basic initialization.  Also allocates subqueue structs
11020  * for each queue on the device.
11021  */
11022 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
11023 		unsigned char name_assign_type,
11024 		void (*setup)(struct net_device *),
11025 		unsigned int txqs, unsigned int rxqs)
11026 {
11027 	struct net_device *dev;
11028 
11029 	BUG_ON(strlen(name) >= sizeof(dev->name));
11030 
11031 	if (txqs < 1) {
11032 		pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
11033 		return NULL;
11034 	}
11035 
11036 	if (rxqs < 1) {
11037 		pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
11038 		return NULL;
11039 	}
11040 
11041 	dev = kvzalloc(struct_size(dev, priv, sizeof_priv),
11042 		       GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
11043 	if (!dev)
11044 		return NULL;
11045 
11046 	dev->priv_len = sizeof_priv;
11047 
11048 	ref_tracker_dir_init(&dev->refcnt_tracker, 128, name);
11049 #ifdef CONFIG_PCPU_DEV_REFCNT
11050 	dev->pcpu_refcnt = alloc_percpu(int);
11051 	if (!dev->pcpu_refcnt)
11052 		goto free_dev;
11053 	__dev_hold(dev);
11054 #else
11055 	refcount_set(&dev->dev_refcnt, 1);
11056 #endif
11057 
11058 	if (dev_addr_init(dev))
11059 		goto free_pcpu;
11060 
11061 	dev_mc_init(dev);
11062 	dev_uc_init(dev);
11063 
11064 	dev_net_set(dev, &init_net);
11065 
11066 	dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
11067 	dev->xdp_zc_max_segs = 1;
11068 	dev->gso_max_segs = GSO_MAX_SEGS;
11069 	dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
11070 	dev->gso_ipv4_max_size = GSO_LEGACY_MAX_SIZE;
11071 	dev->gro_ipv4_max_size = GRO_LEGACY_MAX_SIZE;
11072 	dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
11073 	dev->tso_max_segs = TSO_MAX_SEGS;
11074 	dev->upper_level = 1;
11075 	dev->lower_level = 1;
11076 #ifdef CONFIG_LOCKDEP
11077 	dev->nested_level = 0;
11078 	INIT_LIST_HEAD(&dev->unlink_list);
11079 #endif
11080 
11081 	INIT_LIST_HEAD(&dev->napi_list);
11082 	INIT_LIST_HEAD(&dev->unreg_list);
11083 	INIT_LIST_HEAD(&dev->close_list);
11084 	INIT_LIST_HEAD(&dev->link_watch_list);
11085 	INIT_LIST_HEAD(&dev->adj_list.upper);
11086 	INIT_LIST_HEAD(&dev->adj_list.lower);
11087 	INIT_LIST_HEAD(&dev->ptype_all);
11088 	INIT_LIST_HEAD(&dev->ptype_specific);
11089 	INIT_LIST_HEAD(&dev->net_notifier_list);
11090 #ifdef CONFIG_NET_SCHED
11091 	hash_init(dev->qdisc_hash);
11092 #endif
11093 	dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
11094 	setup(dev);
11095 
11096 	if (!dev->tx_queue_len) {
11097 		dev->priv_flags |= IFF_NO_QUEUE;
11098 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
11099 	}
11100 
11101 	dev->num_tx_queues = txqs;
11102 	dev->real_num_tx_queues = txqs;
11103 	if (netif_alloc_netdev_queues(dev))
11104 		goto free_all;
11105 
11106 	dev->num_rx_queues = rxqs;
11107 	dev->real_num_rx_queues = rxqs;
11108 	if (netif_alloc_rx_queues(dev))
11109 		goto free_all;
11110 	dev->ethtool = kzalloc(sizeof(*dev->ethtool), GFP_KERNEL_ACCOUNT);
11111 	if (!dev->ethtool)
11112 		goto free_all;
11113 
11114 	strcpy(dev->name, name);
11115 	dev->name_assign_type = name_assign_type;
11116 	dev->group = INIT_NETDEV_GROUP;
11117 	if (!dev->ethtool_ops)
11118 		dev->ethtool_ops = &default_ethtool_ops;
11119 
11120 	nf_hook_netdev_init(dev);
11121 
11122 	return dev;
11123 
11124 free_all:
11125 	free_netdev(dev);
11126 	return NULL;
11127 
11128 free_pcpu:
11129 #ifdef CONFIG_PCPU_DEV_REFCNT
11130 	free_percpu(dev->pcpu_refcnt);
11131 free_dev:
11132 #endif
11133 	kvfree(dev);
11134 	return NULL;
11135 }
11136 EXPORT_SYMBOL(alloc_netdev_mqs);
11137 
11138 /**
11139  * free_netdev - free network device
11140  * @dev: device
11141  *
11142  * This function does the last stage of destroying an allocated device
11143  * interface. The reference to the device object is released. If this
11144  * is the last reference then it will be freed.Must be called in process
11145  * context.
11146  */
11147 void free_netdev(struct net_device *dev)
11148 {
11149 	struct napi_struct *p, *n;
11150 
11151 	might_sleep();
11152 
11153 	/* When called immediately after register_netdevice() failed the unwind
11154 	 * handling may still be dismantling the device. Handle that case by
11155 	 * deferring the free.
11156 	 */
11157 	if (dev->reg_state == NETREG_UNREGISTERING) {
11158 		ASSERT_RTNL();
11159 		dev->needs_free_netdev = true;
11160 		return;
11161 	}
11162 
11163 	kfree(dev->ethtool);
11164 	netif_free_tx_queues(dev);
11165 	netif_free_rx_queues(dev);
11166 
11167 	kfree(rcu_dereference_protected(dev->ingress_queue, 1));
11168 
11169 	/* Flush device addresses */
11170 	dev_addr_flush(dev);
11171 
11172 	list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
11173 		netif_napi_del(p);
11174 
11175 	ref_tracker_dir_exit(&dev->refcnt_tracker);
11176 #ifdef CONFIG_PCPU_DEV_REFCNT
11177 	free_percpu(dev->pcpu_refcnt);
11178 	dev->pcpu_refcnt = NULL;
11179 #endif
11180 	free_percpu(dev->core_stats);
11181 	dev->core_stats = NULL;
11182 	free_percpu(dev->xdp_bulkq);
11183 	dev->xdp_bulkq = NULL;
11184 
11185 	/*  Compatibility with error handling in drivers */
11186 	if (dev->reg_state == NETREG_UNINITIALIZED ||
11187 	    dev->reg_state == NETREG_DUMMY) {
11188 		kvfree(dev);
11189 		return;
11190 	}
11191 
11192 	BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
11193 	WRITE_ONCE(dev->reg_state, NETREG_RELEASED);
11194 
11195 	/* will free via device release */
11196 	put_device(&dev->dev);
11197 }
11198 EXPORT_SYMBOL(free_netdev);
11199 
11200 /**
11201  * alloc_netdev_dummy - Allocate and initialize a dummy net device.
11202  * @sizeof_priv: size of private data to allocate space for
11203  *
11204  * Return: the allocated net_device on success, NULL otherwise
11205  */
11206 struct net_device *alloc_netdev_dummy(int sizeof_priv)
11207 {
11208 	return alloc_netdev(sizeof_priv, "dummy#", NET_NAME_UNKNOWN,
11209 			    init_dummy_netdev_core);
11210 }
11211 EXPORT_SYMBOL_GPL(alloc_netdev_dummy);
11212 
11213 /**
11214  *	synchronize_net -  Synchronize with packet receive processing
11215  *
11216  *	Wait for packets currently being received to be done.
11217  *	Does not block later packets from starting.
11218  */
11219 void synchronize_net(void)
11220 {
11221 	might_sleep();
11222 	if (rtnl_is_locked())
11223 		synchronize_rcu_expedited();
11224 	else
11225 		synchronize_rcu();
11226 }
11227 EXPORT_SYMBOL(synchronize_net);
11228 
11229 static void netdev_rss_contexts_free(struct net_device *dev)
11230 {
11231 	struct ethtool_rxfh_context *ctx;
11232 	unsigned long context;
11233 
11234 	mutex_lock(&dev->ethtool->rss_lock);
11235 	xa_for_each(&dev->ethtool->rss_ctx, context, ctx) {
11236 		struct ethtool_rxfh_param rxfh;
11237 
11238 		rxfh.indir = ethtool_rxfh_context_indir(ctx);
11239 		rxfh.key = ethtool_rxfh_context_key(ctx);
11240 		rxfh.hfunc = ctx->hfunc;
11241 		rxfh.input_xfrm = ctx->input_xfrm;
11242 		rxfh.rss_context = context;
11243 		rxfh.rss_delete = true;
11244 
11245 		xa_erase(&dev->ethtool->rss_ctx, context);
11246 		if (dev->ethtool_ops->create_rxfh_context)
11247 			dev->ethtool_ops->remove_rxfh_context(dev, ctx,
11248 							      context, NULL);
11249 		else
11250 			dev->ethtool_ops->set_rxfh(dev, &rxfh, NULL);
11251 		kfree(ctx);
11252 	}
11253 	xa_destroy(&dev->ethtool->rss_ctx);
11254 	mutex_unlock(&dev->ethtool->rss_lock);
11255 }
11256 
11257 /**
11258  *	unregister_netdevice_queue - remove device from the kernel
11259  *	@dev: device
11260  *	@head: list
11261  *
11262  *	This function shuts down a device interface and removes it
11263  *	from the kernel tables.
11264  *	If head not NULL, device is queued to be unregistered later.
11265  *
11266  *	Callers must hold the rtnl semaphore.  You may want
11267  *	unregister_netdev() instead of this.
11268  */
11269 
11270 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
11271 {
11272 	ASSERT_RTNL();
11273 
11274 	if (head) {
11275 		list_move_tail(&dev->unreg_list, head);
11276 	} else {
11277 		LIST_HEAD(single);
11278 
11279 		list_add(&dev->unreg_list, &single);
11280 		unregister_netdevice_many(&single);
11281 	}
11282 }
11283 EXPORT_SYMBOL(unregister_netdevice_queue);
11284 
11285 void unregister_netdevice_many_notify(struct list_head *head,
11286 				      u32 portid, const struct nlmsghdr *nlh)
11287 {
11288 	struct net_device *dev, *tmp;
11289 	LIST_HEAD(close_head);
11290 	int cnt = 0;
11291 
11292 	BUG_ON(dev_boot_phase);
11293 	ASSERT_RTNL();
11294 
11295 	if (list_empty(head))
11296 		return;
11297 
11298 	list_for_each_entry_safe(dev, tmp, head, unreg_list) {
11299 		/* Some devices call without registering
11300 		 * for initialization unwind. Remove those
11301 		 * devices and proceed with the remaining.
11302 		 */
11303 		if (dev->reg_state == NETREG_UNINITIALIZED) {
11304 			pr_debug("unregister_netdevice: device %s/%p never was registered\n",
11305 				 dev->name, dev);
11306 
11307 			WARN_ON(1);
11308 			list_del(&dev->unreg_list);
11309 			continue;
11310 		}
11311 		dev->dismantle = true;
11312 		BUG_ON(dev->reg_state != NETREG_REGISTERED);
11313 	}
11314 
11315 	/* If device is running, close it first. */
11316 	list_for_each_entry(dev, head, unreg_list)
11317 		list_add_tail(&dev->close_list, &close_head);
11318 	dev_close_many(&close_head, true);
11319 
11320 	list_for_each_entry(dev, head, unreg_list) {
11321 		/* And unlink it from device chain. */
11322 		unlist_netdevice(dev);
11323 		WRITE_ONCE(dev->reg_state, NETREG_UNREGISTERING);
11324 	}
11325 	flush_all_backlogs();
11326 
11327 	synchronize_net();
11328 
11329 	list_for_each_entry(dev, head, unreg_list) {
11330 		struct sk_buff *skb = NULL;
11331 
11332 		/* Shutdown queueing discipline. */
11333 		dev_shutdown(dev);
11334 		dev_tcx_uninstall(dev);
11335 		dev_xdp_uninstall(dev);
11336 		bpf_dev_bound_netdev_unregister(dev);
11337 
11338 		netdev_offload_xstats_disable_all(dev);
11339 
11340 		/* Notify protocols, that we are about to destroy
11341 		 * this device. They should clean all the things.
11342 		 */
11343 		call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11344 
11345 		if (!dev->rtnl_link_ops ||
11346 		    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
11347 			skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
11348 						     GFP_KERNEL, NULL, 0,
11349 						     portid, nlh);
11350 
11351 		/*
11352 		 *	Flush the unicast and multicast chains
11353 		 */
11354 		dev_uc_flush(dev);
11355 		dev_mc_flush(dev);
11356 
11357 		netdev_name_node_alt_flush(dev);
11358 		netdev_name_node_free(dev->name_node);
11359 
11360 		netdev_rss_contexts_free(dev);
11361 
11362 		call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
11363 
11364 		if (dev->netdev_ops->ndo_uninit)
11365 			dev->netdev_ops->ndo_uninit(dev);
11366 
11367 		mutex_destroy(&dev->ethtool->rss_lock);
11368 
11369 		if (skb)
11370 			rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh);
11371 
11372 		/* Notifier chain MUST detach us all upper devices. */
11373 		WARN_ON(netdev_has_any_upper_dev(dev));
11374 		WARN_ON(netdev_has_any_lower_dev(dev));
11375 
11376 		/* Remove entries from kobject tree */
11377 		netdev_unregister_kobject(dev);
11378 #ifdef CONFIG_XPS
11379 		/* Remove XPS queueing entries */
11380 		netif_reset_xps_queues_gt(dev, 0);
11381 #endif
11382 	}
11383 
11384 	synchronize_net();
11385 
11386 	list_for_each_entry(dev, head, unreg_list) {
11387 		netdev_put(dev, &dev->dev_registered_tracker);
11388 		net_set_todo(dev);
11389 		cnt++;
11390 	}
11391 	atomic_add(cnt, &dev_unreg_count);
11392 
11393 	list_del(head);
11394 }
11395 
11396 /**
11397  *	unregister_netdevice_many - unregister many devices
11398  *	@head: list of devices
11399  *
11400  *  Note: As most callers use a stack allocated list_head,
11401  *  we force a list_del() to make sure stack wont be corrupted later.
11402  */
11403 void unregister_netdevice_many(struct list_head *head)
11404 {
11405 	unregister_netdevice_many_notify(head, 0, NULL);
11406 }
11407 EXPORT_SYMBOL(unregister_netdevice_many);
11408 
11409 /**
11410  *	unregister_netdev - remove device from the kernel
11411  *	@dev: device
11412  *
11413  *	This function shuts down a device interface and removes it
11414  *	from the kernel tables.
11415  *
11416  *	This is just a wrapper for unregister_netdevice that takes
11417  *	the rtnl semaphore.  In general you want to use this and not
11418  *	unregister_netdevice.
11419  */
11420 void unregister_netdev(struct net_device *dev)
11421 {
11422 	rtnl_lock();
11423 	unregister_netdevice(dev);
11424 	rtnl_unlock();
11425 }
11426 EXPORT_SYMBOL(unregister_netdev);
11427 
11428 /**
11429  *	__dev_change_net_namespace - move device to different nethost namespace
11430  *	@dev: device
11431  *	@net: network namespace
11432  *	@pat: If not NULL name pattern to try if the current device name
11433  *	      is already taken in the destination network namespace.
11434  *	@new_ifindex: If not zero, specifies device index in the target
11435  *	              namespace.
11436  *
11437  *	This function shuts down a device interface and moves it
11438  *	to a new network namespace. On success 0 is returned, on
11439  *	a failure a netagive errno code is returned.
11440  *
11441  *	Callers must hold the rtnl semaphore.
11442  */
11443 
11444 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
11445 			       const char *pat, int new_ifindex)
11446 {
11447 	struct netdev_name_node *name_node;
11448 	struct net *net_old = dev_net(dev);
11449 	char new_name[IFNAMSIZ] = {};
11450 	int err, new_nsid;
11451 
11452 	ASSERT_RTNL();
11453 
11454 	/* Don't allow namespace local devices to be moved. */
11455 	err = -EINVAL;
11456 	if (dev->features & NETIF_F_NETNS_LOCAL)
11457 		goto out;
11458 
11459 	/* Ensure the device has been registrered */
11460 	if (dev->reg_state != NETREG_REGISTERED)
11461 		goto out;
11462 
11463 	/* Get out if there is nothing todo */
11464 	err = 0;
11465 	if (net_eq(net_old, net))
11466 		goto out;
11467 
11468 	/* Pick the destination device name, and ensure
11469 	 * we can use it in the destination network namespace.
11470 	 */
11471 	err = -EEXIST;
11472 	if (netdev_name_in_use(net, dev->name)) {
11473 		/* We get here if we can't use the current device name */
11474 		if (!pat)
11475 			goto out;
11476 		err = dev_prep_valid_name(net, dev, pat, new_name, EEXIST);
11477 		if (err < 0)
11478 			goto out;
11479 	}
11480 	/* Check that none of the altnames conflicts. */
11481 	err = -EEXIST;
11482 	netdev_for_each_altname(dev, name_node)
11483 		if (netdev_name_in_use(net, name_node->name))
11484 			goto out;
11485 
11486 	/* Check that new_ifindex isn't used yet. */
11487 	if (new_ifindex) {
11488 		err = dev_index_reserve(net, new_ifindex);
11489 		if (err < 0)
11490 			goto out;
11491 	} else {
11492 		/* If there is an ifindex conflict assign a new one */
11493 		err = dev_index_reserve(net, dev->ifindex);
11494 		if (err == -EBUSY)
11495 			err = dev_index_reserve(net, 0);
11496 		if (err < 0)
11497 			goto out;
11498 		new_ifindex = err;
11499 	}
11500 
11501 	/*
11502 	 * And now a mini version of register_netdevice unregister_netdevice.
11503 	 */
11504 
11505 	/* If device is running close it first. */
11506 	dev_close(dev);
11507 
11508 	/* And unlink it from device chain */
11509 	unlist_netdevice(dev);
11510 
11511 	synchronize_net();
11512 
11513 	/* Shutdown queueing discipline. */
11514 	dev_shutdown(dev);
11515 
11516 	/* Notify protocols, that we are about to destroy
11517 	 * this device. They should clean all the things.
11518 	 *
11519 	 * Note that dev->reg_state stays at NETREG_REGISTERED.
11520 	 * This is wanted because this way 8021q and macvlan know
11521 	 * the device is just moving and can keep their slaves up.
11522 	 */
11523 	call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11524 	rcu_barrier();
11525 
11526 	new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11527 
11528 	rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11529 			    new_ifindex);
11530 
11531 	/*
11532 	 *	Flush the unicast and multicast chains
11533 	 */
11534 	dev_uc_flush(dev);
11535 	dev_mc_flush(dev);
11536 
11537 	/* Send a netdev-removed uevent to the old namespace */
11538 	kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11539 	netdev_adjacent_del_links(dev);
11540 
11541 	/* Move per-net netdevice notifiers that are following the netdevice */
11542 	move_netdevice_notifiers_dev_net(dev, net);
11543 
11544 	/* Actually switch the network namespace */
11545 	dev_net_set(dev, net);
11546 	dev->ifindex = new_ifindex;
11547 
11548 	if (new_name[0]) {
11549 		/* Rename the netdev to prepared name */
11550 		write_seqlock_bh(&netdev_rename_lock);
11551 		strscpy(dev->name, new_name, IFNAMSIZ);
11552 		write_sequnlock_bh(&netdev_rename_lock);
11553 	}
11554 
11555 	/* Fixup kobjects */
11556 	dev_set_uevent_suppress(&dev->dev, 1);
11557 	err = device_rename(&dev->dev, dev->name);
11558 	dev_set_uevent_suppress(&dev->dev, 0);
11559 	WARN_ON(err);
11560 
11561 	/* Send a netdev-add uevent to the new namespace */
11562 	kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11563 	netdev_adjacent_add_links(dev);
11564 
11565 	/* Adapt owner in case owning user namespace of target network
11566 	 * namespace is different from the original one.
11567 	 */
11568 	err = netdev_change_owner(dev, net_old, net);
11569 	WARN_ON(err);
11570 
11571 	/* Add the device back in the hashes */
11572 	list_netdevice(dev);
11573 
11574 	/* Notify protocols, that a new device appeared. */
11575 	call_netdevice_notifiers(NETDEV_REGISTER, dev);
11576 
11577 	/*
11578 	 *	Prevent userspace races by waiting until the network
11579 	 *	device is fully setup before sending notifications.
11580 	 */
11581 	rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
11582 
11583 	synchronize_net();
11584 	err = 0;
11585 out:
11586 	return err;
11587 }
11588 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11589 
11590 static int dev_cpu_dead(unsigned int oldcpu)
11591 {
11592 	struct sk_buff **list_skb;
11593 	struct sk_buff *skb;
11594 	unsigned int cpu;
11595 	struct softnet_data *sd, *oldsd, *remsd = NULL;
11596 
11597 	local_irq_disable();
11598 	cpu = smp_processor_id();
11599 	sd = &per_cpu(softnet_data, cpu);
11600 	oldsd = &per_cpu(softnet_data, oldcpu);
11601 
11602 	/* Find end of our completion_queue. */
11603 	list_skb = &sd->completion_queue;
11604 	while (*list_skb)
11605 		list_skb = &(*list_skb)->next;
11606 	/* Append completion queue from offline CPU. */
11607 	*list_skb = oldsd->completion_queue;
11608 	oldsd->completion_queue = NULL;
11609 
11610 	/* Append output queue from offline CPU. */
11611 	if (oldsd->output_queue) {
11612 		*sd->output_queue_tailp = oldsd->output_queue;
11613 		sd->output_queue_tailp = oldsd->output_queue_tailp;
11614 		oldsd->output_queue = NULL;
11615 		oldsd->output_queue_tailp = &oldsd->output_queue;
11616 	}
11617 	/* Append NAPI poll list from offline CPU, with one exception :
11618 	 * process_backlog() must be called by cpu owning percpu backlog.
11619 	 * We properly handle process_queue & input_pkt_queue later.
11620 	 */
11621 	while (!list_empty(&oldsd->poll_list)) {
11622 		struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11623 							    struct napi_struct,
11624 							    poll_list);
11625 
11626 		list_del_init(&napi->poll_list);
11627 		if (napi->poll == process_backlog)
11628 			napi->state &= NAPIF_STATE_THREADED;
11629 		else
11630 			____napi_schedule(sd, napi);
11631 	}
11632 
11633 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
11634 	local_irq_enable();
11635 
11636 	if (!use_backlog_threads()) {
11637 #ifdef CONFIG_RPS
11638 		remsd = oldsd->rps_ipi_list;
11639 		oldsd->rps_ipi_list = NULL;
11640 #endif
11641 		/* send out pending IPI's on offline CPU */
11642 		net_rps_send_ipi(remsd);
11643 	}
11644 
11645 	/* Process offline CPU's input_pkt_queue */
11646 	while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11647 		netif_rx(skb);
11648 		rps_input_queue_head_incr(oldsd);
11649 	}
11650 	while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11651 		netif_rx(skb);
11652 		rps_input_queue_head_incr(oldsd);
11653 	}
11654 
11655 	return 0;
11656 }
11657 
11658 /**
11659  *	netdev_increment_features - increment feature set by one
11660  *	@all: current feature set
11661  *	@one: new feature set
11662  *	@mask: mask feature set
11663  *
11664  *	Computes a new feature set after adding a device with feature set
11665  *	@one to the master device with current feature set @all.  Will not
11666  *	enable anything that is off in @mask. Returns the new feature set.
11667  */
11668 netdev_features_t netdev_increment_features(netdev_features_t all,
11669 	netdev_features_t one, netdev_features_t mask)
11670 {
11671 	if (mask & NETIF_F_HW_CSUM)
11672 		mask |= NETIF_F_CSUM_MASK;
11673 	mask |= NETIF_F_VLAN_CHALLENGED;
11674 
11675 	all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11676 	all &= one | ~NETIF_F_ALL_FOR_ALL;
11677 
11678 	/* If one device supports hw checksumming, set for all. */
11679 	if (all & NETIF_F_HW_CSUM)
11680 		all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11681 
11682 	return all;
11683 }
11684 EXPORT_SYMBOL(netdev_increment_features);
11685 
11686 static struct hlist_head * __net_init netdev_create_hash(void)
11687 {
11688 	int i;
11689 	struct hlist_head *hash;
11690 
11691 	hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11692 	if (hash != NULL)
11693 		for (i = 0; i < NETDEV_HASHENTRIES; i++)
11694 			INIT_HLIST_HEAD(&hash[i]);
11695 
11696 	return hash;
11697 }
11698 
11699 /* Initialize per network namespace state */
11700 static int __net_init netdev_init(struct net *net)
11701 {
11702 	BUILD_BUG_ON(GRO_HASH_BUCKETS >
11703 		     8 * sizeof_field(struct napi_struct, gro_bitmask));
11704 
11705 	INIT_LIST_HEAD(&net->dev_base_head);
11706 
11707 	net->dev_name_head = netdev_create_hash();
11708 	if (net->dev_name_head == NULL)
11709 		goto err_name;
11710 
11711 	net->dev_index_head = netdev_create_hash();
11712 	if (net->dev_index_head == NULL)
11713 		goto err_idx;
11714 
11715 	xa_init_flags(&net->dev_by_index, XA_FLAGS_ALLOC1);
11716 
11717 	RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11718 
11719 	return 0;
11720 
11721 err_idx:
11722 	kfree(net->dev_name_head);
11723 err_name:
11724 	return -ENOMEM;
11725 }
11726 
11727 /**
11728  *	netdev_drivername - network driver for the device
11729  *	@dev: network device
11730  *
11731  *	Determine network driver for device.
11732  */
11733 const char *netdev_drivername(const struct net_device *dev)
11734 {
11735 	const struct device_driver *driver;
11736 	const struct device *parent;
11737 	const char *empty = "";
11738 
11739 	parent = dev->dev.parent;
11740 	if (!parent)
11741 		return empty;
11742 
11743 	driver = parent->driver;
11744 	if (driver && driver->name)
11745 		return driver->name;
11746 	return empty;
11747 }
11748 
11749 static void __netdev_printk(const char *level, const struct net_device *dev,
11750 			    struct va_format *vaf)
11751 {
11752 	if (dev && dev->dev.parent) {
11753 		dev_printk_emit(level[1] - '0',
11754 				dev->dev.parent,
11755 				"%s %s %s%s: %pV",
11756 				dev_driver_string(dev->dev.parent),
11757 				dev_name(dev->dev.parent),
11758 				netdev_name(dev), netdev_reg_state(dev),
11759 				vaf);
11760 	} else if (dev) {
11761 		printk("%s%s%s: %pV",
11762 		       level, netdev_name(dev), netdev_reg_state(dev), vaf);
11763 	} else {
11764 		printk("%s(NULL net_device): %pV", level, vaf);
11765 	}
11766 }
11767 
11768 void netdev_printk(const char *level, const struct net_device *dev,
11769 		   const char *format, ...)
11770 {
11771 	struct va_format vaf;
11772 	va_list args;
11773 
11774 	va_start(args, format);
11775 
11776 	vaf.fmt = format;
11777 	vaf.va = &args;
11778 
11779 	__netdev_printk(level, dev, &vaf);
11780 
11781 	va_end(args);
11782 }
11783 EXPORT_SYMBOL(netdev_printk);
11784 
11785 #define define_netdev_printk_level(func, level)			\
11786 void func(const struct net_device *dev, const char *fmt, ...)	\
11787 {								\
11788 	struct va_format vaf;					\
11789 	va_list args;						\
11790 								\
11791 	va_start(args, fmt);					\
11792 								\
11793 	vaf.fmt = fmt;						\
11794 	vaf.va = &args;						\
11795 								\
11796 	__netdev_printk(level, dev, &vaf);			\
11797 								\
11798 	va_end(args);						\
11799 }								\
11800 EXPORT_SYMBOL(func);
11801 
11802 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11803 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11804 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11805 define_netdev_printk_level(netdev_err, KERN_ERR);
11806 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11807 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11808 define_netdev_printk_level(netdev_info, KERN_INFO);
11809 
11810 static void __net_exit netdev_exit(struct net *net)
11811 {
11812 	kfree(net->dev_name_head);
11813 	kfree(net->dev_index_head);
11814 	xa_destroy(&net->dev_by_index);
11815 	if (net != &init_net)
11816 		WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11817 }
11818 
11819 static struct pernet_operations __net_initdata netdev_net_ops = {
11820 	.init = netdev_init,
11821 	.exit = netdev_exit,
11822 };
11823 
11824 static void __net_exit default_device_exit_net(struct net *net)
11825 {
11826 	struct netdev_name_node *name_node, *tmp;
11827 	struct net_device *dev, *aux;
11828 	/*
11829 	 * Push all migratable network devices back to the
11830 	 * initial network namespace
11831 	 */
11832 	ASSERT_RTNL();
11833 	for_each_netdev_safe(net, dev, aux) {
11834 		int err;
11835 		char fb_name[IFNAMSIZ];
11836 
11837 		/* Ignore unmoveable devices (i.e. loopback) */
11838 		if (dev->features & NETIF_F_NETNS_LOCAL)
11839 			continue;
11840 
11841 		/* Leave virtual devices for the generic cleanup */
11842 		if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11843 			continue;
11844 
11845 		/* Push remaining network devices to init_net */
11846 		snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11847 		if (netdev_name_in_use(&init_net, fb_name))
11848 			snprintf(fb_name, IFNAMSIZ, "dev%%d");
11849 
11850 		netdev_for_each_altname_safe(dev, name_node, tmp)
11851 			if (netdev_name_in_use(&init_net, name_node->name))
11852 				__netdev_name_node_alt_destroy(name_node);
11853 
11854 		err = dev_change_net_namespace(dev, &init_net, fb_name);
11855 		if (err) {
11856 			pr_emerg("%s: failed to move %s to init_net: %d\n",
11857 				 __func__, dev->name, err);
11858 			BUG();
11859 		}
11860 	}
11861 }
11862 
11863 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11864 {
11865 	/* At exit all network devices most be removed from a network
11866 	 * namespace.  Do this in the reverse order of registration.
11867 	 * Do this across as many network namespaces as possible to
11868 	 * improve batching efficiency.
11869 	 */
11870 	struct net_device *dev;
11871 	struct net *net;
11872 	LIST_HEAD(dev_kill_list);
11873 
11874 	rtnl_lock();
11875 	list_for_each_entry(net, net_list, exit_list) {
11876 		default_device_exit_net(net);
11877 		cond_resched();
11878 	}
11879 
11880 	list_for_each_entry(net, net_list, exit_list) {
11881 		for_each_netdev_reverse(net, dev) {
11882 			if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11883 				dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11884 			else
11885 				unregister_netdevice_queue(dev, &dev_kill_list);
11886 		}
11887 	}
11888 	unregister_netdevice_many(&dev_kill_list);
11889 	rtnl_unlock();
11890 }
11891 
11892 static struct pernet_operations __net_initdata default_device_ops = {
11893 	.exit_batch = default_device_exit_batch,
11894 };
11895 
11896 static void __init net_dev_struct_check(void)
11897 {
11898 	/* TX read-mostly hotpath */
11899 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, priv_flags);
11900 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, netdev_ops);
11901 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, header_ops);
11902 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, _tx);
11903 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, real_num_tx_queues);
11904 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_size);
11905 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_ipv4_max_size);
11906 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_max_segs);
11907 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, gso_partial_features);
11908 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, num_tc);
11909 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, mtu);
11910 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, needed_headroom);
11911 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tc_to_txq);
11912 #ifdef CONFIG_XPS
11913 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, xps_maps);
11914 #endif
11915 #ifdef CONFIG_NETFILTER_EGRESS
11916 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, nf_hooks_egress);
11917 #endif
11918 #ifdef CONFIG_NET_XGRESS
11919 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_tx, tcx_egress);
11920 #endif
11921 	CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_tx, 160);
11922 
11923 	/* TXRX read-mostly hotpath */
11924 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, lstats);
11925 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, state);
11926 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, flags);
11927 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, hard_header_len);
11928 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, features);
11929 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_txrx, ip6_ptr);
11930 	CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_txrx, 46);
11931 
11932 	/* RX read-mostly hotpath */
11933 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ptype_specific);
11934 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, ifindex);
11935 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, real_num_rx_queues);
11936 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, _rx);
11937 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_flush_timeout);
11938 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, napi_defer_hard_irqs);
11939 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_max_size);
11940 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, gro_ipv4_max_size);
11941 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler);
11942 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, rx_handler_data);
11943 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, nd_net);
11944 #ifdef CONFIG_NETPOLL
11945 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, npinfo);
11946 #endif
11947 #ifdef CONFIG_NET_XGRESS
11948 	CACHELINE_ASSERT_GROUP_MEMBER(struct net_device, net_device_read_rx, tcx_ingress);
11949 #endif
11950 	CACHELINE_ASSERT_GROUP_SIZE(struct net_device, net_device_read_rx, 104);
11951 }
11952 
11953 /*
11954  *	Initialize the DEV module. At boot time this walks the device list and
11955  *	unhooks any devices that fail to initialise (normally hardware not
11956  *	present) and leaves us with a valid list of present and active devices.
11957  *
11958  */
11959 
11960 /* We allocate 256 pages for each CPU if PAGE_SHIFT is 12 */
11961 #define SYSTEM_PERCPU_PAGE_POOL_SIZE	((1 << 20) / PAGE_SIZE)
11962 
11963 static int net_page_pool_create(int cpuid)
11964 {
11965 #if IS_ENABLED(CONFIG_PAGE_POOL)
11966 	struct page_pool_params page_pool_params = {
11967 		.pool_size = SYSTEM_PERCPU_PAGE_POOL_SIZE,
11968 		.flags = PP_FLAG_SYSTEM_POOL,
11969 		.nid = cpu_to_mem(cpuid),
11970 	};
11971 	struct page_pool *pp_ptr;
11972 
11973 	pp_ptr = page_pool_create_percpu(&page_pool_params, cpuid);
11974 	if (IS_ERR(pp_ptr))
11975 		return -ENOMEM;
11976 
11977 	per_cpu(system_page_pool, cpuid) = pp_ptr;
11978 #endif
11979 	return 0;
11980 }
11981 
11982 static int backlog_napi_should_run(unsigned int cpu)
11983 {
11984 	struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu);
11985 	struct napi_struct *napi = &sd->backlog;
11986 
11987 	return test_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
11988 }
11989 
11990 static void run_backlog_napi(unsigned int cpu)
11991 {
11992 	struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu);
11993 
11994 	napi_threaded_poll_loop(&sd->backlog);
11995 }
11996 
11997 static void backlog_napi_setup(unsigned int cpu)
11998 {
11999 	struct softnet_data *sd = per_cpu_ptr(&softnet_data, cpu);
12000 	struct napi_struct *napi = &sd->backlog;
12001 
12002 	napi->thread = this_cpu_read(backlog_napi);
12003 	set_bit(NAPI_STATE_THREADED, &napi->state);
12004 }
12005 
12006 static struct smp_hotplug_thread backlog_threads = {
12007 	.store			= &backlog_napi,
12008 	.thread_should_run	= backlog_napi_should_run,
12009 	.thread_fn		= run_backlog_napi,
12010 	.thread_comm		= "backlog_napi/%u",
12011 	.setup			= backlog_napi_setup,
12012 };
12013 
12014 /*
12015  *       This is called single threaded during boot, so no need
12016  *       to take the rtnl semaphore.
12017  */
12018 static int __init net_dev_init(void)
12019 {
12020 	int i, rc = -ENOMEM;
12021 
12022 	BUG_ON(!dev_boot_phase);
12023 
12024 	net_dev_struct_check();
12025 
12026 	if (dev_proc_init())
12027 		goto out;
12028 
12029 	if (netdev_kobject_init())
12030 		goto out;
12031 
12032 	for (i = 0; i < PTYPE_HASH_SIZE; i++)
12033 		INIT_LIST_HEAD(&ptype_base[i]);
12034 
12035 	if (register_pernet_subsys(&netdev_net_ops))
12036 		goto out;
12037 
12038 	/*
12039 	 *	Initialise the packet receive queues.
12040 	 */
12041 
12042 	for_each_possible_cpu(i) {
12043 		struct work_struct *flush = per_cpu_ptr(&flush_works, i);
12044 		struct softnet_data *sd = &per_cpu(softnet_data, i);
12045 
12046 		INIT_WORK(flush, flush_backlog);
12047 
12048 		skb_queue_head_init(&sd->input_pkt_queue);
12049 		skb_queue_head_init(&sd->process_queue);
12050 #ifdef CONFIG_XFRM_OFFLOAD
12051 		skb_queue_head_init(&sd->xfrm_backlog);
12052 #endif
12053 		INIT_LIST_HEAD(&sd->poll_list);
12054 		sd->output_queue_tailp = &sd->output_queue;
12055 #ifdef CONFIG_RPS
12056 		INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
12057 		sd->cpu = i;
12058 #endif
12059 		INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
12060 		spin_lock_init(&sd->defer_lock);
12061 
12062 		init_gro_hash(&sd->backlog);
12063 		sd->backlog.poll = process_backlog;
12064 		sd->backlog.weight = weight_p;
12065 		INIT_LIST_HEAD(&sd->backlog.poll_list);
12066 
12067 		if (net_page_pool_create(i))
12068 			goto out;
12069 	}
12070 	if (use_backlog_threads())
12071 		smpboot_register_percpu_thread(&backlog_threads);
12072 
12073 	dev_boot_phase = 0;
12074 
12075 	/* The loopback device is special if any other network devices
12076 	 * is present in a network namespace the loopback device must
12077 	 * be present. Since we now dynamically allocate and free the
12078 	 * loopback device ensure this invariant is maintained by
12079 	 * keeping the loopback device as the first device on the
12080 	 * list of network devices.  Ensuring the loopback devices
12081 	 * is the first device that appears and the last network device
12082 	 * that disappears.
12083 	 */
12084 	if (register_pernet_device(&loopback_net_ops))
12085 		goto out;
12086 
12087 	if (register_pernet_device(&default_device_ops))
12088 		goto out;
12089 
12090 	open_softirq(NET_TX_SOFTIRQ, net_tx_action);
12091 	open_softirq(NET_RX_SOFTIRQ, net_rx_action);
12092 
12093 	rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
12094 				       NULL, dev_cpu_dead);
12095 	WARN_ON(rc < 0);
12096 	rc = 0;
12097 
12098 	/* avoid static key IPIs to isolated CPUs */
12099 	if (housekeeping_enabled(HK_TYPE_MISC))
12100 		net_enable_timestamp();
12101 out:
12102 	if (rc < 0) {
12103 		for_each_possible_cpu(i) {
12104 			struct page_pool *pp_ptr;
12105 
12106 			pp_ptr = per_cpu(system_page_pool, i);
12107 			if (!pp_ptr)
12108 				continue;
12109 
12110 			page_pool_destroy(pp_ptr);
12111 			per_cpu(system_page_pool, i) = NULL;
12112 		}
12113 	}
12114 
12115 	return rc;
12116 }
12117 
12118 subsys_initcall(net_dev_init);
12119