xref: /linux/net/core/dev.c (revision 9b8107553424fd87955fed257a807672c2097297)
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/bitops.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/mm.h>
81 #include <linux/mutex.h>
82 #include <linux/rwsem.h>
83 #include <linux/string.h>
84 #include <linux/mm.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/ethtool.h>
93 #include <linux/skbuff.h>
94 #include <linux/kthread.h>
95 #include <linux/bpf.h>
96 #include <linux/bpf_trace.h>
97 #include <net/net_namespace.h>
98 #include <net/sock.h>
99 #include <net/busy_poll.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/stat.h>
102 #include <net/dsa.h>
103 #include <net/dst.h>
104 #include <net/dst_metadata.h>
105 #include <net/gro.h>
106 #include <net/pkt_sched.h>
107 #include <net/pkt_cls.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/module.h>
113 #include <linux/netpoll.h>
114 #include <linux/rcupdate.h>
115 #include <linux/delay.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
125 #include <net/ip.h>
126 #include <net/mpls.h>
127 #include <linux/ipv6.h>
128 #include <linux/in.h>
129 #include <linux/jhash.h>
130 #include <linux/random.h>
131 #include <trace/events/napi.h>
132 #include <trace/events/net.h>
133 #include <trace/events/skb.h>
134 #include <trace/events/qdisc.h>
135 #include <linux/inetdevice.h>
136 #include <linux/cpu_rmap.h>
137 #include <linux/static_key.h>
138 #include <linux/hashtable.h>
139 #include <linux/vmalloc.h>
140 #include <linux/if_macvlan.h>
141 #include <linux/errqueue.h>
142 #include <linux/hrtimer.h>
143 #include <linux/netfilter_netdev.h>
144 #include <linux/crash_dump.h>
145 #include <linux/sctp.h>
146 #include <net/udp_tunnel.h>
147 #include <linux/net_namespace.h>
148 #include <linux/indirect_call_wrapper.h>
149 #include <net/devlink.h>
150 #include <linux/pm_runtime.h>
151 #include <linux/prandom.h>
152 #include <linux/once_lite.h>
153 
154 #include "dev.h"
155 #include "net-sysfs.h"
156 
157 
158 static DEFINE_SPINLOCK(ptype_lock);
159 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
160 struct list_head ptype_all __read_mostly;	/* Taps */
161 
162 static int netif_rx_internal(struct sk_buff *skb);
163 static int call_netdevice_notifiers_info(unsigned long val,
164 					 struct netdev_notifier_info *info);
165 static int call_netdevice_notifiers_extack(unsigned long val,
166 					   struct net_device *dev,
167 					   struct netlink_ext_ack *extack);
168 static struct napi_struct *napi_by_id(unsigned int napi_id);
169 
170 /*
171  * The @dev_base_head list is protected by @dev_base_lock and the rtnl
172  * semaphore.
173  *
174  * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
175  *
176  * Writers must hold the rtnl semaphore while they loop through the
177  * dev_base_head list, and hold dev_base_lock for writing when they do the
178  * actual updates.  This allows pure readers to access the list even
179  * while a writer is preparing to update it.
180  *
181  * To put it another way, dev_base_lock is held for writing only to
182  * protect against pure readers; the rtnl semaphore provides the
183  * protection against other writers.
184  *
185  * See, for example usages, register_netdevice() and
186  * unregister_netdevice(), which must be called with the rtnl
187  * semaphore held.
188  */
189 DEFINE_RWLOCK(dev_base_lock);
190 EXPORT_SYMBOL(dev_base_lock);
191 
192 static DEFINE_MUTEX(ifalias_mutex);
193 
194 /* protects napi_hash addition/deletion and napi_gen_id */
195 static DEFINE_SPINLOCK(napi_hash_lock);
196 
197 static unsigned int napi_gen_id = NR_CPUS;
198 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
199 
200 static DECLARE_RWSEM(devnet_rename_sem);
201 
202 static inline void dev_base_seq_inc(struct net *net)
203 {
204 	while (++net->dev_base_seq == 0)
205 		;
206 }
207 
208 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
209 {
210 	unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
211 
212 	return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
213 }
214 
215 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
216 {
217 	return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
218 }
219 
220 static inline void rps_lock_irqsave(struct softnet_data *sd,
221 				    unsigned long *flags)
222 {
223 	if (IS_ENABLED(CONFIG_RPS))
224 		spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags);
225 	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
226 		local_irq_save(*flags);
227 }
228 
229 static inline void rps_lock_irq_disable(struct softnet_data *sd)
230 {
231 	if (IS_ENABLED(CONFIG_RPS))
232 		spin_lock_irq(&sd->input_pkt_queue.lock);
233 	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
234 		local_irq_disable();
235 }
236 
237 static inline void rps_unlock_irq_restore(struct softnet_data *sd,
238 					  unsigned long *flags)
239 {
240 	if (IS_ENABLED(CONFIG_RPS))
241 		spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags);
242 	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
243 		local_irq_restore(*flags);
244 }
245 
246 static inline void rps_unlock_irq_enable(struct softnet_data *sd)
247 {
248 	if (IS_ENABLED(CONFIG_RPS))
249 		spin_unlock_irq(&sd->input_pkt_queue.lock);
250 	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
251 		local_irq_enable();
252 }
253 
254 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
255 						       const char *name)
256 {
257 	struct netdev_name_node *name_node;
258 
259 	name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
260 	if (!name_node)
261 		return NULL;
262 	INIT_HLIST_NODE(&name_node->hlist);
263 	name_node->dev = dev;
264 	name_node->name = name;
265 	return name_node;
266 }
267 
268 static struct netdev_name_node *
269 netdev_name_node_head_alloc(struct net_device *dev)
270 {
271 	struct netdev_name_node *name_node;
272 
273 	name_node = netdev_name_node_alloc(dev, dev->name);
274 	if (!name_node)
275 		return NULL;
276 	INIT_LIST_HEAD(&name_node->list);
277 	return name_node;
278 }
279 
280 static void netdev_name_node_free(struct netdev_name_node *name_node)
281 {
282 	kfree(name_node);
283 }
284 
285 static void netdev_name_node_add(struct net *net,
286 				 struct netdev_name_node *name_node)
287 {
288 	hlist_add_head_rcu(&name_node->hlist,
289 			   dev_name_hash(net, name_node->name));
290 }
291 
292 static void netdev_name_node_del(struct netdev_name_node *name_node)
293 {
294 	hlist_del_rcu(&name_node->hlist);
295 }
296 
297 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
298 							const char *name)
299 {
300 	struct hlist_head *head = dev_name_hash(net, name);
301 	struct netdev_name_node *name_node;
302 
303 	hlist_for_each_entry(name_node, head, hlist)
304 		if (!strcmp(name_node->name, name))
305 			return name_node;
306 	return NULL;
307 }
308 
309 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
310 							    const char *name)
311 {
312 	struct hlist_head *head = dev_name_hash(net, name);
313 	struct netdev_name_node *name_node;
314 
315 	hlist_for_each_entry_rcu(name_node, head, hlist)
316 		if (!strcmp(name_node->name, name))
317 			return name_node;
318 	return NULL;
319 }
320 
321 bool netdev_name_in_use(struct net *net, const char *name)
322 {
323 	return netdev_name_node_lookup(net, name);
324 }
325 EXPORT_SYMBOL(netdev_name_in_use);
326 
327 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
328 {
329 	struct netdev_name_node *name_node;
330 	struct net *net = dev_net(dev);
331 
332 	name_node = netdev_name_node_lookup(net, name);
333 	if (name_node)
334 		return -EEXIST;
335 	name_node = netdev_name_node_alloc(dev, name);
336 	if (!name_node)
337 		return -ENOMEM;
338 	netdev_name_node_add(net, name_node);
339 	/* The node that holds dev->name acts as a head of per-device list. */
340 	list_add_tail(&name_node->list, &dev->name_node->list);
341 
342 	return 0;
343 }
344 
345 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
346 {
347 	list_del(&name_node->list);
348 	netdev_name_node_del(name_node);
349 	kfree(name_node->name);
350 	netdev_name_node_free(name_node);
351 }
352 
353 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
354 {
355 	struct netdev_name_node *name_node;
356 	struct net *net = dev_net(dev);
357 
358 	name_node = netdev_name_node_lookup(net, name);
359 	if (!name_node)
360 		return -ENOENT;
361 	/* lookup might have found our primary name or a name belonging
362 	 * to another device.
363 	 */
364 	if (name_node == dev->name_node || name_node->dev != dev)
365 		return -EINVAL;
366 
367 	__netdev_name_node_alt_destroy(name_node);
368 
369 	return 0;
370 }
371 
372 static void netdev_name_node_alt_flush(struct net_device *dev)
373 {
374 	struct netdev_name_node *name_node, *tmp;
375 
376 	list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
377 		__netdev_name_node_alt_destroy(name_node);
378 }
379 
380 /* Device list insertion */
381 static void list_netdevice(struct net_device *dev)
382 {
383 	struct net *net = dev_net(dev);
384 
385 	ASSERT_RTNL();
386 
387 	write_lock(&dev_base_lock);
388 	list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
389 	netdev_name_node_add(net, dev->name_node);
390 	hlist_add_head_rcu(&dev->index_hlist,
391 			   dev_index_hash(net, dev->ifindex));
392 	write_unlock(&dev_base_lock);
393 
394 	dev_base_seq_inc(net);
395 }
396 
397 /* Device list removal
398  * caller must respect a RCU grace period before freeing/reusing dev
399  */
400 static void unlist_netdevice(struct net_device *dev, bool lock)
401 {
402 	ASSERT_RTNL();
403 
404 	/* Unlink dev from the device chain */
405 	if (lock)
406 		write_lock(&dev_base_lock);
407 	list_del_rcu(&dev->dev_list);
408 	netdev_name_node_del(dev->name_node);
409 	hlist_del_rcu(&dev->index_hlist);
410 	if (lock)
411 		write_unlock(&dev_base_lock);
412 
413 	dev_base_seq_inc(dev_net(dev));
414 }
415 
416 /*
417  *	Our notifier list
418  */
419 
420 static RAW_NOTIFIER_HEAD(netdev_chain);
421 
422 /*
423  *	Device drivers call our routines to queue packets here. We empty the
424  *	queue in the local softnet handler.
425  */
426 
427 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
428 EXPORT_PER_CPU_SYMBOL(softnet_data);
429 
430 #ifdef CONFIG_LOCKDEP
431 /*
432  * register_netdevice() inits txq->_xmit_lock and sets lockdep class
433  * according to dev->type
434  */
435 static const unsigned short netdev_lock_type[] = {
436 	 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
437 	 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
438 	 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
439 	 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
440 	 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
441 	 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
442 	 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
443 	 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
444 	 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
445 	 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
446 	 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
447 	 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
448 	 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
449 	 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
450 	 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
451 
452 static const char *const netdev_lock_name[] = {
453 	"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
454 	"_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
455 	"_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
456 	"_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
457 	"_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
458 	"_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
459 	"_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
460 	"_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
461 	"_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
462 	"_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
463 	"_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
464 	"_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
465 	"_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
466 	"_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
467 	"_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
468 
469 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
470 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
471 
472 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
473 {
474 	int i;
475 
476 	for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
477 		if (netdev_lock_type[i] == dev_type)
478 			return i;
479 	/* the last key is used by default */
480 	return ARRAY_SIZE(netdev_lock_type) - 1;
481 }
482 
483 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
484 						 unsigned short dev_type)
485 {
486 	int i;
487 
488 	i = netdev_lock_pos(dev_type);
489 	lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
490 				   netdev_lock_name[i]);
491 }
492 
493 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
494 {
495 	int i;
496 
497 	i = netdev_lock_pos(dev->type);
498 	lockdep_set_class_and_name(&dev->addr_list_lock,
499 				   &netdev_addr_lock_key[i],
500 				   netdev_lock_name[i]);
501 }
502 #else
503 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
504 						 unsigned short dev_type)
505 {
506 }
507 
508 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
509 {
510 }
511 #endif
512 
513 /*******************************************************************************
514  *
515  *		Protocol management and registration routines
516  *
517  *******************************************************************************/
518 
519 
520 /*
521  *	Add a protocol ID to the list. Now that the input handler is
522  *	smarter we can dispense with all the messy stuff that used to be
523  *	here.
524  *
525  *	BEWARE!!! Protocol handlers, mangling input packets,
526  *	MUST BE last in hash buckets and checking protocol handlers
527  *	MUST start from promiscuous ptype_all chain in net_bh.
528  *	It is true now, do not change it.
529  *	Explanation follows: if protocol handler, mangling packet, will
530  *	be the first on list, it is not able to sense, that packet
531  *	is cloned and should be copied-on-write, so that it will
532  *	change it and subsequent readers will get broken packet.
533  *							--ANK (980803)
534  */
535 
536 static inline struct list_head *ptype_head(const struct packet_type *pt)
537 {
538 	if (pt->type == htons(ETH_P_ALL))
539 		return pt->dev ? &pt->dev->ptype_all : &ptype_all;
540 	else
541 		return pt->dev ? &pt->dev->ptype_specific :
542 				 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
543 }
544 
545 /**
546  *	dev_add_pack - add packet handler
547  *	@pt: packet type declaration
548  *
549  *	Add a protocol handler to the networking stack. The passed &packet_type
550  *	is linked into kernel lists and may not be freed until it has been
551  *	removed from the kernel lists.
552  *
553  *	This call does not sleep therefore it can not
554  *	guarantee all CPU's that are in middle of receiving packets
555  *	will see the new packet type (until the next received packet).
556  */
557 
558 void dev_add_pack(struct packet_type *pt)
559 {
560 	struct list_head *head = ptype_head(pt);
561 
562 	spin_lock(&ptype_lock);
563 	list_add_rcu(&pt->list, head);
564 	spin_unlock(&ptype_lock);
565 }
566 EXPORT_SYMBOL(dev_add_pack);
567 
568 /**
569  *	__dev_remove_pack	 - remove packet handler
570  *	@pt: packet type declaration
571  *
572  *	Remove a protocol handler that was previously added to the kernel
573  *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
574  *	from the kernel lists and can be freed or reused once this function
575  *	returns.
576  *
577  *      The packet type might still be in use by receivers
578  *	and must not be freed until after all the CPU's have gone
579  *	through a quiescent state.
580  */
581 void __dev_remove_pack(struct packet_type *pt)
582 {
583 	struct list_head *head = ptype_head(pt);
584 	struct packet_type *pt1;
585 
586 	spin_lock(&ptype_lock);
587 
588 	list_for_each_entry(pt1, head, list) {
589 		if (pt == pt1) {
590 			list_del_rcu(&pt->list);
591 			goto out;
592 		}
593 	}
594 
595 	pr_warn("dev_remove_pack: %p not found\n", pt);
596 out:
597 	spin_unlock(&ptype_lock);
598 }
599 EXPORT_SYMBOL(__dev_remove_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  *	This call sleeps to guarantee that no CPU is looking at the packet
611  *	type after return.
612  */
613 void dev_remove_pack(struct packet_type *pt)
614 {
615 	__dev_remove_pack(pt);
616 
617 	synchronize_net();
618 }
619 EXPORT_SYMBOL(dev_remove_pack);
620 
621 
622 /*******************************************************************************
623  *
624  *			    Device Interface Subroutines
625  *
626  *******************************************************************************/
627 
628 /**
629  *	dev_get_iflink	- get 'iflink' value of a interface
630  *	@dev: targeted interface
631  *
632  *	Indicates the ifindex the interface is linked to.
633  *	Physical interfaces have the same 'ifindex' and 'iflink' values.
634  */
635 
636 int dev_get_iflink(const struct net_device *dev)
637 {
638 	if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
639 		return dev->netdev_ops->ndo_get_iflink(dev);
640 
641 	return dev->ifindex;
642 }
643 EXPORT_SYMBOL(dev_get_iflink);
644 
645 /**
646  *	dev_fill_metadata_dst - Retrieve tunnel egress information.
647  *	@dev: targeted interface
648  *	@skb: The packet.
649  *
650  *	For better visibility of tunnel traffic OVS needs to retrieve
651  *	egress tunnel information for a packet. Following API allows
652  *	user to get this info.
653  */
654 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
655 {
656 	struct ip_tunnel_info *info;
657 
658 	if (!dev->netdev_ops  || !dev->netdev_ops->ndo_fill_metadata_dst)
659 		return -EINVAL;
660 
661 	info = skb_tunnel_info_unclone(skb);
662 	if (!info)
663 		return -ENOMEM;
664 	if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
665 		return -EINVAL;
666 
667 	return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
668 }
669 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
670 
671 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
672 {
673 	int k = stack->num_paths++;
674 
675 	if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
676 		return NULL;
677 
678 	return &stack->path[k];
679 }
680 
681 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
682 			  struct net_device_path_stack *stack)
683 {
684 	const struct net_device *last_dev;
685 	struct net_device_path_ctx ctx = {
686 		.dev	= dev,
687 	};
688 	struct net_device_path *path;
689 	int ret = 0;
690 
691 	memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
692 	stack->num_paths = 0;
693 	while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
694 		last_dev = ctx.dev;
695 		path = dev_fwd_path(stack);
696 		if (!path)
697 			return -1;
698 
699 		memset(path, 0, sizeof(struct net_device_path));
700 		ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
701 		if (ret < 0)
702 			return -1;
703 
704 		if (WARN_ON_ONCE(last_dev == ctx.dev))
705 			return -1;
706 	}
707 
708 	if (!ctx.dev)
709 		return ret;
710 
711 	path = dev_fwd_path(stack);
712 	if (!path)
713 		return -1;
714 	path->type = DEV_PATH_ETHERNET;
715 	path->dev = ctx.dev;
716 
717 	return ret;
718 }
719 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
720 
721 /**
722  *	__dev_get_by_name	- find a device by its name
723  *	@net: the applicable net namespace
724  *	@name: name to find
725  *
726  *	Find an interface by name. Must be called under RTNL semaphore
727  *	or @dev_base_lock. If the name is found a pointer to the device
728  *	is returned. If the name is not found then %NULL is returned. The
729  *	reference counters are not incremented so the caller must be
730  *	careful with locks.
731  */
732 
733 struct net_device *__dev_get_by_name(struct net *net, const char *name)
734 {
735 	struct netdev_name_node *node_name;
736 
737 	node_name = netdev_name_node_lookup(net, name);
738 	return node_name ? node_name->dev : NULL;
739 }
740 EXPORT_SYMBOL(__dev_get_by_name);
741 
742 /**
743  * dev_get_by_name_rcu	- find a device by its name
744  * @net: the applicable net namespace
745  * @name: name to find
746  *
747  * Find an interface by name.
748  * If the name is found a pointer to the device is returned.
749  * If the name is not found then %NULL is returned.
750  * The reference counters are not incremented so the caller must be
751  * careful with locks. The caller must hold RCU lock.
752  */
753 
754 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
755 {
756 	struct netdev_name_node *node_name;
757 
758 	node_name = netdev_name_node_lookup_rcu(net, name);
759 	return node_name ? node_name->dev : NULL;
760 }
761 EXPORT_SYMBOL(dev_get_by_name_rcu);
762 
763 /**
764  *	dev_get_by_name		- find a device by its name
765  *	@net: the applicable net namespace
766  *	@name: name to find
767  *
768  *	Find an interface by name. This can be called from any
769  *	context and does its own locking. The returned handle has
770  *	the usage count incremented and the caller must use dev_put() to
771  *	release it when it is no longer needed. %NULL is returned if no
772  *	matching device is found.
773  */
774 
775 struct net_device *dev_get_by_name(struct net *net, const char *name)
776 {
777 	struct net_device *dev;
778 
779 	rcu_read_lock();
780 	dev = dev_get_by_name_rcu(net, name);
781 	dev_hold(dev);
782 	rcu_read_unlock();
783 	return dev;
784 }
785 EXPORT_SYMBOL(dev_get_by_name);
786 
787 /**
788  *	__dev_get_by_index - find a device by its ifindex
789  *	@net: the applicable net namespace
790  *	@ifindex: index of device
791  *
792  *	Search for an interface by index. Returns %NULL if the device
793  *	is not found or a pointer to the device. The device has not
794  *	had its reference counter increased so the caller must be careful
795  *	about locking. The caller must hold either the RTNL semaphore
796  *	or @dev_base_lock.
797  */
798 
799 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
800 {
801 	struct net_device *dev;
802 	struct hlist_head *head = dev_index_hash(net, ifindex);
803 
804 	hlist_for_each_entry(dev, head, index_hlist)
805 		if (dev->ifindex == ifindex)
806 			return dev;
807 
808 	return NULL;
809 }
810 EXPORT_SYMBOL(__dev_get_by_index);
811 
812 /**
813  *	dev_get_by_index_rcu - find a device by its ifindex
814  *	@net: the applicable net namespace
815  *	@ifindex: index of device
816  *
817  *	Search for an interface by index. Returns %NULL if the device
818  *	is not found or a pointer to the device. The device has not
819  *	had its reference counter increased so the caller must be careful
820  *	about locking. The caller must hold RCU lock.
821  */
822 
823 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
824 {
825 	struct net_device *dev;
826 	struct hlist_head *head = dev_index_hash(net, ifindex);
827 
828 	hlist_for_each_entry_rcu(dev, head, index_hlist)
829 		if (dev->ifindex == ifindex)
830 			return dev;
831 
832 	return NULL;
833 }
834 EXPORT_SYMBOL(dev_get_by_index_rcu);
835 
836 
837 /**
838  *	dev_get_by_index - find a device by its ifindex
839  *	@net: the applicable net namespace
840  *	@ifindex: index of device
841  *
842  *	Search for an interface by index. Returns NULL if the device
843  *	is not found or a pointer to the device. The device returned has
844  *	had a reference added and the pointer is safe until the user calls
845  *	dev_put to indicate they have finished with it.
846  */
847 
848 struct net_device *dev_get_by_index(struct net *net, int ifindex)
849 {
850 	struct net_device *dev;
851 
852 	rcu_read_lock();
853 	dev = dev_get_by_index_rcu(net, ifindex);
854 	dev_hold(dev);
855 	rcu_read_unlock();
856 	return dev;
857 }
858 EXPORT_SYMBOL(dev_get_by_index);
859 
860 /**
861  *	dev_get_by_napi_id - find a device by napi_id
862  *	@napi_id: ID of the NAPI struct
863  *
864  *	Search for an interface by NAPI ID. Returns %NULL if the device
865  *	is not found or a pointer to the device. The device has not had
866  *	its reference counter increased so the caller must be careful
867  *	about locking. The caller must hold RCU lock.
868  */
869 
870 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
871 {
872 	struct napi_struct *napi;
873 
874 	WARN_ON_ONCE(!rcu_read_lock_held());
875 
876 	if (napi_id < MIN_NAPI_ID)
877 		return NULL;
878 
879 	napi = napi_by_id(napi_id);
880 
881 	return napi ? napi->dev : NULL;
882 }
883 EXPORT_SYMBOL(dev_get_by_napi_id);
884 
885 /**
886  *	netdev_get_name - get a netdevice name, knowing its ifindex.
887  *	@net: network namespace
888  *	@name: a pointer to the buffer where the name will be stored.
889  *	@ifindex: the ifindex of the interface to get the name from.
890  */
891 int netdev_get_name(struct net *net, char *name, int ifindex)
892 {
893 	struct net_device *dev;
894 	int ret;
895 
896 	down_read(&devnet_rename_sem);
897 	rcu_read_lock();
898 
899 	dev = dev_get_by_index_rcu(net, ifindex);
900 	if (!dev) {
901 		ret = -ENODEV;
902 		goto out;
903 	}
904 
905 	strcpy(name, dev->name);
906 
907 	ret = 0;
908 out:
909 	rcu_read_unlock();
910 	up_read(&devnet_rename_sem);
911 	return ret;
912 }
913 
914 /**
915  *	dev_getbyhwaddr_rcu - find a device by its hardware address
916  *	@net: the applicable net namespace
917  *	@type: media type of device
918  *	@ha: hardware address
919  *
920  *	Search for an interface by MAC address. Returns NULL if the device
921  *	is not found or a pointer to the device.
922  *	The caller must hold RCU or RTNL.
923  *	The returned device has not had its ref count increased
924  *	and the caller must therefore be careful about locking
925  *
926  */
927 
928 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
929 				       const char *ha)
930 {
931 	struct net_device *dev;
932 
933 	for_each_netdev_rcu(net, dev)
934 		if (dev->type == type &&
935 		    !memcmp(dev->dev_addr, ha, dev->addr_len))
936 			return dev;
937 
938 	return NULL;
939 }
940 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
941 
942 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
943 {
944 	struct net_device *dev, *ret = NULL;
945 
946 	rcu_read_lock();
947 	for_each_netdev_rcu(net, dev)
948 		if (dev->type == type) {
949 			dev_hold(dev);
950 			ret = dev;
951 			break;
952 		}
953 	rcu_read_unlock();
954 	return ret;
955 }
956 EXPORT_SYMBOL(dev_getfirstbyhwtype);
957 
958 /**
959  *	__dev_get_by_flags - find any device with given flags
960  *	@net: the applicable net namespace
961  *	@if_flags: IFF_* values
962  *	@mask: bitmask of bits in if_flags to check
963  *
964  *	Search for any interface with the given flags. Returns NULL if a device
965  *	is not found or a pointer to the device. Must be called inside
966  *	rtnl_lock(), and result refcount is unchanged.
967  */
968 
969 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
970 				      unsigned short mask)
971 {
972 	struct net_device *dev, *ret;
973 
974 	ASSERT_RTNL();
975 
976 	ret = NULL;
977 	for_each_netdev(net, dev) {
978 		if (((dev->flags ^ if_flags) & mask) == 0) {
979 			ret = dev;
980 			break;
981 		}
982 	}
983 	return ret;
984 }
985 EXPORT_SYMBOL(__dev_get_by_flags);
986 
987 /**
988  *	dev_valid_name - check if name is okay for network device
989  *	@name: name string
990  *
991  *	Network device names need to be valid file names to
992  *	allow sysfs to work.  We also disallow any kind of
993  *	whitespace.
994  */
995 bool dev_valid_name(const char *name)
996 {
997 	if (*name == '\0')
998 		return false;
999 	if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1000 		return false;
1001 	if (!strcmp(name, ".") || !strcmp(name, ".."))
1002 		return false;
1003 
1004 	while (*name) {
1005 		if (*name == '/' || *name == ':' || isspace(*name))
1006 			return false;
1007 		name++;
1008 	}
1009 	return true;
1010 }
1011 EXPORT_SYMBOL(dev_valid_name);
1012 
1013 /**
1014  *	__dev_alloc_name - allocate a name for a device
1015  *	@net: network namespace to allocate the device name in
1016  *	@name: name format string
1017  *	@buf:  scratch buffer and result name string
1018  *
1019  *	Passed a format string - eg "lt%d" it will try and find a suitable
1020  *	id. It scans list of devices to build up a free map, then chooses
1021  *	the first empty slot. The caller must hold the dev_base or rtnl lock
1022  *	while allocating the name and adding the device in order to avoid
1023  *	duplicates.
1024  *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1025  *	Returns the number of the unit assigned or a negative errno code.
1026  */
1027 
1028 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1029 {
1030 	int i = 0;
1031 	const char *p;
1032 	const int max_netdevices = 8*PAGE_SIZE;
1033 	unsigned long *inuse;
1034 	struct net_device *d;
1035 
1036 	if (!dev_valid_name(name))
1037 		return -EINVAL;
1038 
1039 	p = strchr(name, '%');
1040 	if (p) {
1041 		/*
1042 		 * Verify the string as this thing may have come from
1043 		 * the user.  There must be either one "%d" and no other "%"
1044 		 * characters.
1045 		 */
1046 		if (p[1] != 'd' || strchr(p + 2, '%'))
1047 			return -EINVAL;
1048 
1049 		/* Use one page as a bit array of possible slots */
1050 		inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1051 		if (!inuse)
1052 			return -ENOMEM;
1053 
1054 		for_each_netdev(net, d) {
1055 			struct netdev_name_node *name_node;
1056 			list_for_each_entry(name_node, &d->name_node->list, list) {
1057 				if (!sscanf(name_node->name, name, &i))
1058 					continue;
1059 				if (i < 0 || i >= max_netdevices)
1060 					continue;
1061 
1062 				/*  avoid cases where sscanf is not exact inverse of printf */
1063 				snprintf(buf, IFNAMSIZ, name, i);
1064 				if (!strncmp(buf, name_node->name, IFNAMSIZ))
1065 					__set_bit(i, inuse);
1066 			}
1067 			if (!sscanf(d->name, name, &i))
1068 				continue;
1069 			if (i < 0 || i >= max_netdevices)
1070 				continue;
1071 
1072 			/*  avoid cases where sscanf is not exact inverse of printf */
1073 			snprintf(buf, IFNAMSIZ, name, i);
1074 			if (!strncmp(buf, d->name, IFNAMSIZ))
1075 				__set_bit(i, inuse);
1076 		}
1077 
1078 		i = find_first_zero_bit(inuse, max_netdevices);
1079 		free_page((unsigned long) inuse);
1080 	}
1081 
1082 	snprintf(buf, IFNAMSIZ, name, i);
1083 	if (!netdev_name_in_use(net, buf))
1084 		return i;
1085 
1086 	/* It is possible to run out of possible slots
1087 	 * when the name is long and there isn't enough space left
1088 	 * for the digits, or if all bits are used.
1089 	 */
1090 	return -ENFILE;
1091 }
1092 
1093 static int dev_alloc_name_ns(struct net *net,
1094 			     struct net_device *dev,
1095 			     const char *name)
1096 {
1097 	char buf[IFNAMSIZ];
1098 	int ret;
1099 
1100 	BUG_ON(!net);
1101 	ret = __dev_alloc_name(net, name, buf);
1102 	if (ret >= 0)
1103 		strscpy(dev->name, buf, IFNAMSIZ);
1104 	return ret;
1105 }
1106 
1107 /**
1108  *	dev_alloc_name - allocate a name for a device
1109  *	@dev: device
1110  *	@name: name format string
1111  *
1112  *	Passed a format string - eg "lt%d" it will try and find a suitable
1113  *	id. It scans list of devices to build up a free map, then chooses
1114  *	the first empty slot. The caller must hold the dev_base or rtnl lock
1115  *	while allocating the name and adding the device in order to avoid
1116  *	duplicates.
1117  *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1118  *	Returns the number of the unit assigned or a negative errno code.
1119  */
1120 
1121 int dev_alloc_name(struct net_device *dev, const char *name)
1122 {
1123 	return dev_alloc_name_ns(dev_net(dev), dev, name);
1124 }
1125 EXPORT_SYMBOL(dev_alloc_name);
1126 
1127 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1128 			      const char *name)
1129 {
1130 	BUG_ON(!net);
1131 
1132 	if (!dev_valid_name(name))
1133 		return -EINVAL;
1134 
1135 	if (strchr(name, '%'))
1136 		return dev_alloc_name_ns(net, dev, name);
1137 	else if (netdev_name_in_use(net, name))
1138 		return -EEXIST;
1139 	else if (dev->name != name)
1140 		strscpy(dev->name, name, IFNAMSIZ);
1141 
1142 	return 0;
1143 }
1144 
1145 /**
1146  *	dev_change_name - change name of a device
1147  *	@dev: device
1148  *	@newname: name (or format string) must be at least IFNAMSIZ
1149  *
1150  *	Change name of a device, can pass format strings "eth%d".
1151  *	for wildcarding.
1152  */
1153 int dev_change_name(struct net_device *dev, const char *newname)
1154 {
1155 	unsigned char old_assign_type;
1156 	char oldname[IFNAMSIZ];
1157 	int err = 0;
1158 	int ret;
1159 	struct net *net;
1160 
1161 	ASSERT_RTNL();
1162 	BUG_ON(!dev_net(dev));
1163 
1164 	net = dev_net(dev);
1165 
1166 	down_write(&devnet_rename_sem);
1167 
1168 	if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1169 		up_write(&devnet_rename_sem);
1170 		return 0;
1171 	}
1172 
1173 	memcpy(oldname, dev->name, IFNAMSIZ);
1174 
1175 	err = dev_get_valid_name(net, dev, newname);
1176 	if (err < 0) {
1177 		up_write(&devnet_rename_sem);
1178 		return err;
1179 	}
1180 
1181 	if (oldname[0] && !strchr(oldname, '%'))
1182 		netdev_info(dev, "renamed from %s%s\n", oldname,
1183 			    dev->flags & IFF_UP ? " (while UP)" : "");
1184 
1185 	old_assign_type = dev->name_assign_type;
1186 	dev->name_assign_type = NET_NAME_RENAMED;
1187 
1188 rollback:
1189 	ret = device_rename(&dev->dev, dev->name);
1190 	if (ret) {
1191 		memcpy(dev->name, oldname, IFNAMSIZ);
1192 		dev->name_assign_type = old_assign_type;
1193 		up_write(&devnet_rename_sem);
1194 		return ret;
1195 	}
1196 
1197 	up_write(&devnet_rename_sem);
1198 
1199 	netdev_adjacent_rename_links(dev, oldname);
1200 
1201 	write_lock(&dev_base_lock);
1202 	netdev_name_node_del(dev->name_node);
1203 	write_unlock(&dev_base_lock);
1204 
1205 	synchronize_rcu();
1206 
1207 	write_lock(&dev_base_lock);
1208 	netdev_name_node_add(net, dev->name_node);
1209 	write_unlock(&dev_base_lock);
1210 
1211 	ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1212 	ret = notifier_to_errno(ret);
1213 
1214 	if (ret) {
1215 		/* err >= 0 after dev_alloc_name() or stores the first errno */
1216 		if (err >= 0) {
1217 			err = ret;
1218 			down_write(&devnet_rename_sem);
1219 			memcpy(dev->name, oldname, IFNAMSIZ);
1220 			memcpy(oldname, newname, IFNAMSIZ);
1221 			dev->name_assign_type = old_assign_type;
1222 			old_assign_type = NET_NAME_RENAMED;
1223 			goto rollback;
1224 		} else {
1225 			netdev_err(dev, "name change rollback failed: %d\n",
1226 				   ret);
1227 		}
1228 	}
1229 
1230 	return err;
1231 }
1232 
1233 /**
1234  *	dev_set_alias - change ifalias of a device
1235  *	@dev: device
1236  *	@alias: name up to IFALIASZ
1237  *	@len: limit of bytes to copy from info
1238  *
1239  *	Set ifalias for a device,
1240  */
1241 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1242 {
1243 	struct dev_ifalias *new_alias = NULL;
1244 
1245 	if (len >= IFALIASZ)
1246 		return -EINVAL;
1247 
1248 	if (len) {
1249 		new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1250 		if (!new_alias)
1251 			return -ENOMEM;
1252 
1253 		memcpy(new_alias->ifalias, alias, len);
1254 		new_alias->ifalias[len] = 0;
1255 	}
1256 
1257 	mutex_lock(&ifalias_mutex);
1258 	new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1259 					mutex_is_locked(&ifalias_mutex));
1260 	mutex_unlock(&ifalias_mutex);
1261 
1262 	if (new_alias)
1263 		kfree_rcu(new_alias, rcuhead);
1264 
1265 	return len;
1266 }
1267 EXPORT_SYMBOL(dev_set_alias);
1268 
1269 /**
1270  *	dev_get_alias - get ifalias of a device
1271  *	@dev: device
1272  *	@name: buffer to store name of ifalias
1273  *	@len: size of buffer
1274  *
1275  *	get ifalias for a device.  Caller must make sure dev cannot go
1276  *	away,  e.g. rcu read lock or own a reference count to device.
1277  */
1278 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1279 {
1280 	const struct dev_ifalias *alias;
1281 	int ret = 0;
1282 
1283 	rcu_read_lock();
1284 	alias = rcu_dereference(dev->ifalias);
1285 	if (alias)
1286 		ret = snprintf(name, len, "%s", alias->ifalias);
1287 	rcu_read_unlock();
1288 
1289 	return ret;
1290 }
1291 
1292 /**
1293  *	netdev_features_change - device changes features
1294  *	@dev: device to cause notification
1295  *
1296  *	Called to indicate a device has changed features.
1297  */
1298 void netdev_features_change(struct net_device *dev)
1299 {
1300 	call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1301 }
1302 EXPORT_SYMBOL(netdev_features_change);
1303 
1304 /**
1305  *	netdev_state_change - device changes state
1306  *	@dev: device to cause notification
1307  *
1308  *	Called to indicate a device has changed state. This function calls
1309  *	the notifier chains for netdev_chain and sends a NEWLINK message
1310  *	to the routing socket.
1311  */
1312 void netdev_state_change(struct net_device *dev)
1313 {
1314 	if (dev->flags & IFF_UP) {
1315 		struct netdev_notifier_change_info change_info = {
1316 			.info.dev = dev,
1317 		};
1318 
1319 		call_netdevice_notifiers_info(NETDEV_CHANGE,
1320 					      &change_info.info);
1321 		rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL);
1322 	}
1323 }
1324 EXPORT_SYMBOL(netdev_state_change);
1325 
1326 /**
1327  * __netdev_notify_peers - notify network peers about existence of @dev,
1328  * to be called when rtnl lock is already held.
1329  * @dev: network device
1330  *
1331  * Generate traffic such that interested network peers are aware of
1332  * @dev, such as by generating a gratuitous ARP. This may be used when
1333  * a device wants to inform the rest of the network about some sort of
1334  * reconfiguration such as a failover event or virtual machine
1335  * migration.
1336  */
1337 void __netdev_notify_peers(struct net_device *dev)
1338 {
1339 	ASSERT_RTNL();
1340 	call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1341 	call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1342 }
1343 EXPORT_SYMBOL(__netdev_notify_peers);
1344 
1345 /**
1346  * netdev_notify_peers - notify network peers about existence of @dev
1347  * @dev: network device
1348  *
1349  * Generate traffic such that interested network peers are aware of
1350  * @dev, such as by generating a gratuitous ARP. This may be used when
1351  * a device wants to inform the rest of the network about some sort of
1352  * reconfiguration such as a failover event or virtual machine
1353  * migration.
1354  */
1355 void netdev_notify_peers(struct net_device *dev)
1356 {
1357 	rtnl_lock();
1358 	__netdev_notify_peers(dev);
1359 	rtnl_unlock();
1360 }
1361 EXPORT_SYMBOL(netdev_notify_peers);
1362 
1363 static int napi_threaded_poll(void *data);
1364 
1365 static int napi_kthread_create(struct napi_struct *n)
1366 {
1367 	int err = 0;
1368 
1369 	/* Create and wake up the kthread once to put it in
1370 	 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1371 	 * warning and work with loadavg.
1372 	 */
1373 	n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1374 				n->dev->name, n->napi_id);
1375 	if (IS_ERR(n->thread)) {
1376 		err = PTR_ERR(n->thread);
1377 		pr_err("kthread_run failed with err %d\n", err);
1378 		n->thread = NULL;
1379 	}
1380 
1381 	return err;
1382 }
1383 
1384 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1385 {
1386 	const struct net_device_ops *ops = dev->netdev_ops;
1387 	int ret;
1388 
1389 	ASSERT_RTNL();
1390 	dev_addr_check(dev);
1391 
1392 	if (!netif_device_present(dev)) {
1393 		/* may be detached because parent is runtime-suspended */
1394 		if (dev->dev.parent)
1395 			pm_runtime_resume(dev->dev.parent);
1396 		if (!netif_device_present(dev))
1397 			return -ENODEV;
1398 	}
1399 
1400 	/* Block netpoll from trying to do any rx path servicing.
1401 	 * If we don't do this there is a chance ndo_poll_controller
1402 	 * or ndo_poll may be running while we open the device
1403 	 */
1404 	netpoll_poll_disable(dev);
1405 
1406 	ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1407 	ret = notifier_to_errno(ret);
1408 	if (ret)
1409 		return ret;
1410 
1411 	set_bit(__LINK_STATE_START, &dev->state);
1412 
1413 	if (ops->ndo_validate_addr)
1414 		ret = ops->ndo_validate_addr(dev);
1415 
1416 	if (!ret && ops->ndo_open)
1417 		ret = ops->ndo_open(dev);
1418 
1419 	netpoll_poll_enable(dev);
1420 
1421 	if (ret)
1422 		clear_bit(__LINK_STATE_START, &dev->state);
1423 	else {
1424 		dev->flags |= IFF_UP;
1425 		dev_set_rx_mode(dev);
1426 		dev_activate(dev);
1427 		add_device_randomness(dev->dev_addr, dev->addr_len);
1428 	}
1429 
1430 	return ret;
1431 }
1432 
1433 /**
1434  *	dev_open	- prepare an interface for use.
1435  *	@dev: device to open
1436  *	@extack: netlink extended ack
1437  *
1438  *	Takes a device from down to up state. The device's private open
1439  *	function is invoked and then the multicast lists are loaded. Finally
1440  *	the device is moved into the up state and a %NETDEV_UP message is
1441  *	sent to the netdev notifier chain.
1442  *
1443  *	Calling this function on an active interface is a nop. On a failure
1444  *	a negative errno code is returned.
1445  */
1446 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1447 {
1448 	int ret;
1449 
1450 	if (dev->flags & IFF_UP)
1451 		return 0;
1452 
1453 	ret = __dev_open(dev, extack);
1454 	if (ret < 0)
1455 		return ret;
1456 
1457 	rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1458 	call_netdevice_notifiers(NETDEV_UP, dev);
1459 
1460 	return ret;
1461 }
1462 EXPORT_SYMBOL(dev_open);
1463 
1464 static void __dev_close_many(struct list_head *head)
1465 {
1466 	struct net_device *dev;
1467 
1468 	ASSERT_RTNL();
1469 	might_sleep();
1470 
1471 	list_for_each_entry(dev, head, close_list) {
1472 		/* Temporarily disable netpoll until the interface is down */
1473 		netpoll_poll_disable(dev);
1474 
1475 		call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1476 
1477 		clear_bit(__LINK_STATE_START, &dev->state);
1478 
1479 		/* Synchronize to scheduled poll. We cannot touch poll list, it
1480 		 * can be even on different cpu. So just clear netif_running().
1481 		 *
1482 		 * dev->stop() will invoke napi_disable() on all of it's
1483 		 * napi_struct instances on this device.
1484 		 */
1485 		smp_mb__after_atomic(); /* Commit netif_running(). */
1486 	}
1487 
1488 	dev_deactivate_many(head);
1489 
1490 	list_for_each_entry(dev, head, close_list) {
1491 		const struct net_device_ops *ops = dev->netdev_ops;
1492 
1493 		/*
1494 		 *	Call the device specific close. This cannot fail.
1495 		 *	Only if device is UP
1496 		 *
1497 		 *	We allow it to be called even after a DETACH hot-plug
1498 		 *	event.
1499 		 */
1500 		if (ops->ndo_stop)
1501 			ops->ndo_stop(dev);
1502 
1503 		dev->flags &= ~IFF_UP;
1504 		netpoll_poll_enable(dev);
1505 	}
1506 }
1507 
1508 static void __dev_close(struct net_device *dev)
1509 {
1510 	LIST_HEAD(single);
1511 
1512 	list_add(&dev->close_list, &single);
1513 	__dev_close_many(&single);
1514 	list_del(&single);
1515 }
1516 
1517 void dev_close_many(struct list_head *head, bool unlink)
1518 {
1519 	struct net_device *dev, *tmp;
1520 
1521 	/* Remove the devices that don't need to be closed */
1522 	list_for_each_entry_safe(dev, tmp, head, close_list)
1523 		if (!(dev->flags & IFF_UP))
1524 			list_del_init(&dev->close_list);
1525 
1526 	__dev_close_many(head);
1527 
1528 	list_for_each_entry_safe(dev, tmp, head, close_list) {
1529 		rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1530 		call_netdevice_notifiers(NETDEV_DOWN, dev);
1531 		if (unlink)
1532 			list_del_init(&dev->close_list);
1533 	}
1534 }
1535 EXPORT_SYMBOL(dev_close_many);
1536 
1537 /**
1538  *	dev_close - shutdown an interface.
1539  *	@dev: device to shutdown
1540  *
1541  *	This function moves an active device into down state. A
1542  *	%NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1543  *	is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1544  *	chain.
1545  */
1546 void dev_close(struct net_device *dev)
1547 {
1548 	if (dev->flags & IFF_UP) {
1549 		LIST_HEAD(single);
1550 
1551 		list_add(&dev->close_list, &single);
1552 		dev_close_many(&single, true);
1553 		list_del(&single);
1554 	}
1555 }
1556 EXPORT_SYMBOL(dev_close);
1557 
1558 
1559 /**
1560  *	dev_disable_lro - disable Large Receive Offload on a device
1561  *	@dev: device
1562  *
1563  *	Disable Large Receive Offload (LRO) on a net device.  Must be
1564  *	called under RTNL.  This is needed if received packets may be
1565  *	forwarded to another interface.
1566  */
1567 void dev_disable_lro(struct net_device *dev)
1568 {
1569 	struct net_device *lower_dev;
1570 	struct list_head *iter;
1571 
1572 	dev->wanted_features &= ~NETIF_F_LRO;
1573 	netdev_update_features(dev);
1574 
1575 	if (unlikely(dev->features & NETIF_F_LRO))
1576 		netdev_WARN(dev, "failed to disable LRO!\n");
1577 
1578 	netdev_for_each_lower_dev(dev, lower_dev, iter)
1579 		dev_disable_lro(lower_dev);
1580 }
1581 EXPORT_SYMBOL(dev_disable_lro);
1582 
1583 /**
1584  *	dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1585  *	@dev: device
1586  *
1587  *	Disable HW Generic Receive Offload (GRO_HW) on a net device.  Must be
1588  *	called under RTNL.  This is needed if Generic XDP is installed on
1589  *	the device.
1590  */
1591 static void dev_disable_gro_hw(struct net_device *dev)
1592 {
1593 	dev->wanted_features &= ~NETIF_F_GRO_HW;
1594 	netdev_update_features(dev);
1595 
1596 	if (unlikely(dev->features & NETIF_F_GRO_HW))
1597 		netdev_WARN(dev, "failed to disable GRO_HW!\n");
1598 }
1599 
1600 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1601 {
1602 #define N(val) 						\
1603 	case NETDEV_##val:				\
1604 		return "NETDEV_" __stringify(val);
1605 	switch (cmd) {
1606 	N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1607 	N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1608 	N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1609 	N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN)
1610 	N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA)
1611 	N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE)
1612 	N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1613 	N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1614 	N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1615 	N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE)
1616 	N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA)
1617 	}
1618 #undef N
1619 	return "UNKNOWN_NETDEV_EVENT";
1620 }
1621 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1622 
1623 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1624 				   struct net_device *dev)
1625 {
1626 	struct netdev_notifier_info info = {
1627 		.dev = dev,
1628 	};
1629 
1630 	return nb->notifier_call(nb, val, &info);
1631 }
1632 
1633 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1634 					     struct net_device *dev)
1635 {
1636 	int err;
1637 
1638 	err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1639 	err = notifier_to_errno(err);
1640 	if (err)
1641 		return err;
1642 
1643 	if (!(dev->flags & IFF_UP))
1644 		return 0;
1645 
1646 	call_netdevice_notifier(nb, NETDEV_UP, dev);
1647 	return 0;
1648 }
1649 
1650 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1651 						struct net_device *dev)
1652 {
1653 	if (dev->flags & IFF_UP) {
1654 		call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1655 					dev);
1656 		call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1657 	}
1658 	call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1659 }
1660 
1661 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1662 						 struct net *net)
1663 {
1664 	struct net_device *dev;
1665 	int err;
1666 
1667 	for_each_netdev(net, dev) {
1668 		err = call_netdevice_register_notifiers(nb, dev);
1669 		if (err)
1670 			goto rollback;
1671 	}
1672 	return 0;
1673 
1674 rollback:
1675 	for_each_netdev_continue_reverse(net, dev)
1676 		call_netdevice_unregister_notifiers(nb, dev);
1677 	return err;
1678 }
1679 
1680 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1681 						    struct net *net)
1682 {
1683 	struct net_device *dev;
1684 
1685 	for_each_netdev(net, dev)
1686 		call_netdevice_unregister_notifiers(nb, dev);
1687 }
1688 
1689 static int dev_boot_phase = 1;
1690 
1691 /**
1692  * register_netdevice_notifier - register a network notifier block
1693  * @nb: notifier
1694  *
1695  * Register a notifier to be called when network device events occur.
1696  * The notifier passed is linked into the kernel structures and must
1697  * not be reused until it has been unregistered. A negative errno code
1698  * is returned on a failure.
1699  *
1700  * When registered all registration and up events are replayed
1701  * to the new notifier to allow device to have a race free
1702  * view of the network device list.
1703  */
1704 
1705 int register_netdevice_notifier(struct notifier_block *nb)
1706 {
1707 	struct net *net;
1708 	int err;
1709 
1710 	/* Close race with setup_net() and cleanup_net() */
1711 	down_write(&pernet_ops_rwsem);
1712 	rtnl_lock();
1713 	err = raw_notifier_chain_register(&netdev_chain, nb);
1714 	if (err)
1715 		goto unlock;
1716 	if (dev_boot_phase)
1717 		goto unlock;
1718 	for_each_net(net) {
1719 		err = call_netdevice_register_net_notifiers(nb, net);
1720 		if (err)
1721 			goto rollback;
1722 	}
1723 
1724 unlock:
1725 	rtnl_unlock();
1726 	up_write(&pernet_ops_rwsem);
1727 	return err;
1728 
1729 rollback:
1730 	for_each_net_continue_reverse(net)
1731 		call_netdevice_unregister_net_notifiers(nb, net);
1732 
1733 	raw_notifier_chain_unregister(&netdev_chain, nb);
1734 	goto unlock;
1735 }
1736 EXPORT_SYMBOL(register_netdevice_notifier);
1737 
1738 /**
1739  * unregister_netdevice_notifier - unregister a network notifier block
1740  * @nb: notifier
1741  *
1742  * Unregister a notifier previously registered by
1743  * register_netdevice_notifier(). The notifier is unlinked into the
1744  * kernel structures and may then be reused. A negative errno code
1745  * is returned on a failure.
1746  *
1747  * After unregistering unregister and down device events are synthesized
1748  * for all devices on the device list to the removed notifier to remove
1749  * the need for special case cleanup code.
1750  */
1751 
1752 int unregister_netdevice_notifier(struct notifier_block *nb)
1753 {
1754 	struct net *net;
1755 	int err;
1756 
1757 	/* Close race with setup_net() and cleanup_net() */
1758 	down_write(&pernet_ops_rwsem);
1759 	rtnl_lock();
1760 	err = raw_notifier_chain_unregister(&netdev_chain, nb);
1761 	if (err)
1762 		goto unlock;
1763 
1764 	for_each_net(net)
1765 		call_netdevice_unregister_net_notifiers(nb, net);
1766 
1767 unlock:
1768 	rtnl_unlock();
1769 	up_write(&pernet_ops_rwsem);
1770 	return err;
1771 }
1772 EXPORT_SYMBOL(unregister_netdevice_notifier);
1773 
1774 static int __register_netdevice_notifier_net(struct net *net,
1775 					     struct notifier_block *nb,
1776 					     bool ignore_call_fail)
1777 {
1778 	int err;
1779 
1780 	err = raw_notifier_chain_register(&net->netdev_chain, nb);
1781 	if (err)
1782 		return err;
1783 	if (dev_boot_phase)
1784 		return 0;
1785 
1786 	err = call_netdevice_register_net_notifiers(nb, net);
1787 	if (err && !ignore_call_fail)
1788 		goto chain_unregister;
1789 
1790 	return 0;
1791 
1792 chain_unregister:
1793 	raw_notifier_chain_unregister(&net->netdev_chain, nb);
1794 	return err;
1795 }
1796 
1797 static int __unregister_netdevice_notifier_net(struct net *net,
1798 					       struct notifier_block *nb)
1799 {
1800 	int err;
1801 
1802 	err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1803 	if (err)
1804 		return err;
1805 
1806 	call_netdevice_unregister_net_notifiers(nb, net);
1807 	return 0;
1808 }
1809 
1810 /**
1811  * register_netdevice_notifier_net - register a per-netns network notifier block
1812  * @net: network namespace
1813  * @nb: notifier
1814  *
1815  * Register a notifier to be called when network device events occur.
1816  * The notifier passed is linked into the kernel structures and must
1817  * not be reused until it has been unregistered. A negative errno code
1818  * is returned on a failure.
1819  *
1820  * When registered all registration and up events are replayed
1821  * to the new notifier to allow device to have a race free
1822  * view of the network device list.
1823  */
1824 
1825 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1826 {
1827 	int err;
1828 
1829 	rtnl_lock();
1830 	err = __register_netdevice_notifier_net(net, nb, false);
1831 	rtnl_unlock();
1832 	return err;
1833 }
1834 EXPORT_SYMBOL(register_netdevice_notifier_net);
1835 
1836 /**
1837  * unregister_netdevice_notifier_net - unregister a per-netns
1838  *                                     network notifier block
1839  * @net: network namespace
1840  * @nb: notifier
1841  *
1842  * Unregister a notifier previously registered by
1843  * register_netdevice_notifier(). The notifier is unlinked into the
1844  * kernel structures and may then be reused. A negative errno code
1845  * is returned on a failure.
1846  *
1847  * After unregistering unregister and down device events are synthesized
1848  * for all devices on the device list to the removed notifier to remove
1849  * the need for special case cleanup code.
1850  */
1851 
1852 int unregister_netdevice_notifier_net(struct net *net,
1853 				      struct notifier_block *nb)
1854 {
1855 	int err;
1856 
1857 	rtnl_lock();
1858 	err = __unregister_netdevice_notifier_net(net, nb);
1859 	rtnl_unlock();
1860 	return err;
1861 }
1862 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1863 
1864 static void __move_netdevice_notifier_net(struct net *src_net,
1865 					  struct net *dst_net,
1866 					  struct notifier_block *nb)
1867 {
1868 	__unregister_netdevice_notifier_net(src_net, nb);
1869 	__register_netdevice_notifier_net(dst_net, nb, true);
1870 }
1871 
1872 void move_netdevice_notifier_net(struct net *src_net, struct net *dst_net,
1873 				 struct notifier_block *nb)
1874 {
1875 	rtnl_lock();
1876 	__move_netdevice_notifier_net(src_net, dst_net, nb);
1877 	rtnl_unlock();
1878 }
1879 
1880 int register_netdevice_notifier_dev_net(struct net_device *dev,
1881 					struct notifier_block *nb,
1882 					struct netdev_net_notifier *nn)
1883 {
1884 	int err;
1885 
1886 	rtnl_lock();
1887 	err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1888 	if (!err) {
1889 		nn->nb = nb;
1890 		list_add(&nn->list, &dev->net_notifier_list);
1891 	}
1892 	rtnl_unlock();
1893 	return err;
1894 }
1895 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1896 
1897 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1898 					  struct notifier_block *nb,
1899 					  struct netdev_net_notifier *nn)
1900 {
1901 	int err;
1902 
1903 	rtnl_lock();
1904 	list_del(&nn->list);
1905 	err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1906 	rtnl_unlock();
1907 	return err;
1908 }
1909 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1910 
1911 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1912 					     struct net *net)
1913 {
1914 	struct netdev_net_notifier *nn;
1915 
1916 	list_for_each_entry(nn, &dev->net_notifier_list, list)
1917 		__move_netdevice_notifier_net(dev_net(dev), net, nn->nb);
1918 }
1919 
1920 /**
1921  *	call_netdevice_notifiers_info - call all network notifier blocks
1922  *	@val: value passed unmodified to notifier function
1923  *	@info: notifier information data
1924  *
1925  *	Call all network notifier blocks.  Parameters and return value
1926  *	are as for raw_notifier_call_chain().
1927  */
1928 
1929 static int call_netdevice_notifiers_info(unsigned long val,
1930 					 struct netdev_notifier_info *info)
1931 {
1932 	struct net *net = dev_net(info->dev);
1933 	int ret;
1934 
1935 	ASSERT_RTNL();
1936 
1937 	/* Run per-netns notifier block chain first, then run the global one.
1938 	 * Hopefully, one day, the global one is going to be removed after
1939 	 * all notifier block registrators get converted to be per-netns.
1940 	 */
1941 	ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1942 	if (ret & NOTIFY_STOP_MASK)
1943 		return ret;
1944 	return raw_notifier_call_chain(&netdev_chain, val, info);
1945 }
1946 
1947 /**
1948  *	call_netdevice_notifiers_info_robust - call per-netns notifier blocks
1949  *	                                       for and rollback on error
1950  *	@val_up: value passed unmodified to notifier function
1951  *	@val_down: value passed unmodified to the notifier function when
1952  *	           recovering from an error on @val_up
1953  *	@info: notifier information data
1954  *
1955  *	Call all per-netns network notifier blocks, but not notifier blocks on
1956  *	the global notifier chain. Parameters and return value are as for
1957  *	raw_notifier_call_chain_robust().
1958  */
1959 
1960 static int
1961 call_netdevice_notifiers_info_robust(unsigned long val_up,
1962 				     unsigned long val_down,
1963 				     struct netdev_notifier_info *info)
1964 {
1965 	struct net *net = dev_net(info->dev);
1966 
1967 	ASSERT_RTNL();
1968 
1969 	return raw_notifier_call_chain_robust(&net->netdev_chain,
1970 					      val_up, val_down, info);
1971 }
1972 
1973 static int call_netdevice_notifiers_extack(unsigned long val,
1974 					   struct net_device *dev,
1975 					   struct netlink_ext_ack *extack)
1976 {
1977 	struct netdev_notifier_info info = {
1978 		.dev = dev,
1979 		.extack = extack,
1980 	};
1981 
1982 	return call_netdevice_notifiers_info(val, &info);
1983 }
1984 
1985 /**
1986  *	call_netdevice_notifiers - call all network notifier blocks
1987  *      @val: value passed unmodified to notifier function
1988  *      @dev: net_device pointer passed unmodified to notifier function
1989  *
1990  *	Call all network notifier blocks.  Parameters and return value
1991  *	are as for raw_notifier_call_chain().
1992  */
1993 
1994 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1995 {
1996 	return call_netdevice_notifiers_extack(val, dev, NULL);
1997 }
1998 EXPORT_SYMBOL(call_netdevice_notifiers);
1999 
2000 /**
2001  *	call_netdevice_notifiers_mtu - call all network notifier blocks
2002  *	@val: value passed unmodified to notifier function
2003  *	@dev: net_device pointer passed unmodified to notifier function
2004  *	@arg: additional u32 argument passed to the notifier function
2005  *
2006  *	Call all network notifier blocks.  Parameters and return value
2007  *	are as for raw_notifier_call_chain().
2008  */
2009 static int call_netdevice_notifiers_mtu(unsigned long val,
2010 					struct net_device *dev, u32 arg)
2011 {
2012 	struct netdev_notifier_info_ext info = {
2013 		.info.dev = dev,
2014 		.ext.mtu = arg,
2015 	};
2016 
2017 	BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2018 
2019 	return call_netdevice_notifiers_info(val, &info.info);
2020 }
2021 
2022 #ifdef CONFIG_NET_INGRESS
2023 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2024 
2025 void net_inc_ingress_queue(void)
2026 {
2027 	static_branch_inc(&ingress_needed_key);
2028 }
2029 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2030 
2031 void net_dec_ingress_queue(void)
2032 {
2033 	static_branch_dec(&ingress_needed_key);
2034 }
2035 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2036 #endif
2037 
2038 #ifdef CONFIG_NET_EGRESS
2039 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2040 
2041 void net_inc_egress_queue(void)
2042 {
2043 	static_branch_inc(&egress_needed_key);
2044 }
2045 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2046 
2047 void net_dec_egress_queue(void)
2048 {
2049 	static_branch_dec(&egress_needed_key);
2050 }
2051 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2052 #endif
2053 
2054 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2055 EXPORT_SYMBOL(netstamp_needed_key);
2056 #ifdef CONFIG_JUMP_LABEL
2057 static atomic_t netstamp_needed_deferred;
2058 static atomic_t netstamp_wanted;
2059 static void netstamp_clear(struct work_struct *work)
2060 {
2061 	int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2062 	int wanted;
2063 
2064 	wanted = atomic_add_return(deferred, &netstamp_wanted);
2065 	if (wanted > 0)
2066 		static_branch_enable(&netstamp_needed_key);
2067 	else
2068 		static_branch_disable(&netstamp_needed_key);
2069 }
2070 static DECLARE_WORK(netstamp_work, netstamp_clear);
2071 #endif
2072 
2073 void net_enable_timestamp(void)
2074 {
2075 #ifdef CONFIG_JUMP_LABEL
2076 	int wanted;
2077 
2078 	while (1) {
2079 		wanted = atomic_read(&netstamp_wanted);
2080 		if (wanted <= 0)
2081 			break;
2082 		if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2083 			return;
2084 	}
2085 	atomic_inc(&netstamp_needed_deferred);
2086 	schedule_work(&netstamp_work);
2087 #else
2088 	static_branch_inc(&netstamp_needed_key);
2089 #endif
2090 }
2091 EXPORT_SYMBOL(net_enable_timestamp);
2092 
2093 void net_disable_timestamp(void)
2094 {
2095 #ifdef CONFIG_JUMP_LABEL
2096 	int wanted;
2097 
2098 	while (1) {
2099 		wanted = atomic_read(&netstamp_wanted);
2100 		if (wanted <= 1)
2101 			break;
2102 		if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2103 			return;
2104 	}
2105 	atomic_dec(&netstamp_needed_deferred);
2106 	schedule_work(&netstamp_work);
2107 #else
2108 	static_branch_dec(&netstamp_needed_key);
2109 #endif
2110 }
2111 EXPORT_SYMBOL(net_disable_timestamp);
2112 
2113 static inline void net_timestamp_set(struct sk_buff *skb)
2114 {
2115 	skb->tstamp = 0;
2116 	skb->mono_delivery_time = 0;
2117 	if (static_branch_unlikely(&netstamp_needed_key))
2118 		skb->tstamp = ktime_get_real();
2119 }
2120 
2121 #define net_timestamp_check(COND, SKB)				\
2122 	if (static_branch_unlikely(&netstamp_needed_key)) {	\
2123 		if ((COND) && !(SKB)->tstamp)			\
2124 			(SKB)->tstamp = ktime_get_real();	\
2125 	}							\
2126 
2127 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2128 {
2129 	return __is_skb_forwardable(dev, skb, true);
2130 }
2131 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2132 
2133 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2134 			      bool check_mtu)
2135 {
2136 	int ret = ____dev_forward_skb(dev, skb, check_mtu);
2137 
2138 	if (likely(!ret)) {
2139 		skb->protocol = eth_type_trans(skb, dev);
2140 		skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2141 	}
2142 
2143 	return ret;
2144 }
2145 
2146 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2147 {
2148 	return __dev_forward_skb2(dev, skb, true);
2149 }
2150 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2151 
2152 /**
2153  * dev_forward_skb - loopback an skb to another netif
2154  *
2155  * @dev: destination network device
2156  * @skb: buffer to forward
2157  *
2158  * return values:
2159  *	NET_RX_SUCCESS	(no congestion)
2160  *	NET_RX_DROP     (packet was dropped, but freed)
2161  *
2162  * dev_forward_skb can be used for injecting an skb from the
2163  * start_xmit function of one device into the receive queue
2164  * of another device.
2165  *
2166  * The receiving device may be in another namespace, so
2167  * we have to clear all information in the skb that could
2168  * impact namespace isolation.
2169  */
2170 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2171 {
2172 	return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2173 }
2174 EXPORT_SYMBOL_GPL(dev_forward_skb);
2175 
2176 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2177 {
2178 	return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2179 }
2180 
2181 static inline int deliver_skb(struct sk_buff *skb,
2182 			      struct packet_type *pt_prev,
2183 			      struct net_device *orig_dev)
2184 {
2185 	if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2186 		return -ENOMEM;
2187 	refcount_inc(&skb->users);
2188 	return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2189 }
2190 
2191 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2192 					  struct packet_type **pt,
2193 					  struct net_device *orig_dev,
2194 					  __be16 type,
2195 					  struct list_head *ptype_list)
2196 {
2197 	struct packet_type *ptype, *pt_prev = *pt;
2198 
2199 	list_for_each_entry_rcu(ptype, ptype_list, list) {
2200 		if (ptype->type != type)
2201 			continue;
2202 		if (pt_prev)
2203 			deliver_skb(skb, pt_prev, orig_dev);
2204 		pt_prev = ptype;
2205 	}
2206 	*pt = pt_prev;
2207 }
2208 
2209 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2210 {
2211 	if (!ptype->af_packet_priv || !skb->sk)
2212 		return false;
2213 
2214 	if (ptype->id_match)
2215 		return ptype->id_match(ptype, skb->sk);
2216 	else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2217 		return true;
2218 
2219 	return false;
2220 }
2221 
2222 /**
2223  * dev_nit_active - return true if any network interface taps are in use
2224  *
2225  * @dev: network device to check for the presence of taps
2226  */
2227 bool dev_nit_active(struct net_device *dev)
2228 {
2229 	return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2230 }
2231 EXPORT_SYMBOL_GPL(dev_nit_active);
2232 
2233 /*
2234  *	Support routine. Sends outgoing frames to any network
2235  *	taps currently in use.
2236  */
2237 
2238 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2239 {
2240 	struct packet_type *ptype;
2241 	struct sk_buff *skb2 = NULL;
2242 	struct packet_type *pt_prev = NULL;
2243 	struct list_head *ptype_list = &ptype_all;
2244 
2245 	rcu_read_lock();
2246 again:
2247 	list_for_each_entry_rcu(ptype, ptype_list, list) {
2248 		if (ptype->ignore_outgoing)
2249 			continue;
2250 
2251 		/* Never send packets back to the socket
2252 		 * they originated from - MvS (miquels@drinkel.ow.org)
2253 		 */
2254 		if (skb_loop_sk(ptype, skb))
2255 			continue;
2256 
2257 		if (pt_prev) {
2258 			deliver_skb(skb2, pt_prev, skb->dev);
2259 			pt_prev = ptype;
2260 			continue;
2261 		}
2262 
2263 		/* need to clone skb, done only once */
2264 		skb2 = skb_clone(skb, GFP_ATOMIC);
2265 		if (!skb2)
2266 			goto out_unlock;
2267 
2268 		net_timestamp_set(skb2);
2269 
2270 		/* skb->nh should be correctly
2271 		 * set by sender, so that the second statement is
2272 		 * just protection against buggy protocols.
2273 		 */
2274 		skb_reset_mac_header(skb2);
2275 
2276 		if (skb_network_header(skb2) < skb2->data ||
2277 		    skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2278 			net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2279 					     ntohs(skb2->protocol),
2280 					     dev->name);
2281 			skb_reset_network_header(skb2);
2282 		}
2283 
2284 		skb2->transport_header = skb2->network_header;
2285 		skb2->pkt_type = PACKET_OUTGOING;
2286 		pt_prev = ptype;
2287 	}
2288 
2289 	if (ptype_list == &ptype_all) {
2290 		ptype_list = &dev->ptype_all;
2291 		goto again;
2292 	}
2293 out_unlock:
2294 	if (pt_prev) {
2295 		if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2296 			pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2297 		else
2298 			kfree_skb(skb2);
2299 	}
2300 	rcu_read_unlock();
2301 }
2302 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2303 
2304 /**
2305  * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2306  * @dev: Network device
2307  * @txq: number of queues available
2308  *
2309  * If real_num_tx_queues is changed the tc mappings may no longer be
2310  * valid. To resolve this verify the tc mapping remains valid and if
2311  * not NULL the mapping. With no priorities mapping to this
2312  * offset/count pair it will no longer be used. In the worst case TC0
2313  * is invalid nothing can be done so disable priority mappings. If is
2314  * expected that drivers will fix this mapping if they can before
2315  * calling netif_set_real_num_tx_queues.
2316  */
2317 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2318 {
2319 	int i;
2320 	struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2321 
2322 	/* If TC0 is invalidated disable TC mapping */
2323 	if (tc->offset + tc->count > txq) {
2324 		netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2325 		dev->num_tc = 0;
2326 		return;
2327 	}
2328 
2329 	/* Invalidated prio to tc mappings set to TC0 */
2330 	for (i = 1; i < TC_BITMASK + 1; i++) {
2331 		int q = netdev_get_prio_tc_map(dev, i);
2332 
2333 		tc = &dev->tc_to_txq[q];
2334 		if (tc->offset + tc->count > txq) {
2335 			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",
2336 				    i, q);
2337 			netdev_set_prio_tc_map(dev, i, 0);
2338 		}
2339 	}
2340 }
2341 
2342 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2343 {
2344 	if (dev->num_tc) {
2345 		struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2346 		int i;
2347 
2348 		/* walk through the TCs and see if it falls into any of them */
2349 		for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2350 			if ((txq - tc->offset) < tc->count)
2351 				return i;
2352 		}
2353 
2354 		/* didn't find it, just return -1 to indicate no match */
2355 		return -1;
2356 	}
2357 
2358 	return 0;
2359 }
2360 EXPORT_SYMBOL(netdev_txq_to_tc);
2361 
2362 #ifdef CONFIG_XPS
2363 static struct static_key xps_needed __read_mostly;
2364 static struct static_key xps_rxqs_needed __read_mostly;
2365 static DEFINE_MUTEX(xps_map_mutex);
2366 #define xmap_dereference(P)		\
2367 	rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2368 
2369 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2370 			     struct xps_dev_maps *old_maps, int tci, u16 index)
2371 {
2372 	struct xps_map *map = NULL;
2373 	int pos;
2374 
2375 	if (dev_maps)
2376 		map = xmap_dereference(dev_maps->attr_map[tci]);
2377 	if (!map)
2378 		return false;
2379 
2380 	for (pos = map->len; pos--;) {
2381 		if (map->queues[pos] != index)
2382 			continue;
2383 
2384 		if (map->len > 1) {
2385 			map->queues[pos] = map->queues[--map->len];
2386 			break;
2387 		}
2388 
2389 		if (old_maps)
2390 			RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2391 		RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2392 		kfree_rcu(map, rcu);
2393 		return false;
2394 	}
2395 
2396 	return true;
2397 }
2398 
2399 static bool remove_xps_queue_cpu(struct net_device *dev,
2400 				 struct xps_dev_maps *dev_maps,
2401 				 int cpu, u16 offset, u16 count)
2402 {
2403 	int num_tc = dev_maps->num_tc;
2404 	bool active = false;
2405 	int tci;
2406 
2407 	for (tci = cpu * num_tc; num_tc--; tci++) {
2408 		int i, j;
2409 
2410 		for (i = count, j = offset; i--; j++) {
2411 			if (!remove_xps_queue(dev_maps, NULL, tci, j))
2412 				break;
2413 		}
2414 
2415 		active |= i < 0;
2416 	}
2417 
2418 	return active;
2419 }
2420 
2421 static void reset_xps_maps(struct net_device *dev,
2422 			   struct xps_dev_maps *dev_maps,
2423 			   enum xps_map_type type)
2424 {
2425 	static_key_slow_dec_cpuslocked(&xps_needed);
2426 	if (type == XPS_RXQS)
2427 		static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2428 
2429 	RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2430 
2431 	kfree_rcu(dev_maps, rcu);
2432 }
2433 
2434 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2435 			   u16 offset, u16 count)
2436 {
2437 	struct xps_dev_maps *dev_maps;
2438 	bool active = false;
2439 	int i, j;
2440 
2441 	dev_maps = xmap_dereference(dev->xps_maps[type]);
2442 	if (!dev_maps)
2443 		return;
2444 
2445 	for (j = 0; j < dev_maps->nr_ids; j++)
2446 		active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2447 	if (!active)
2448 		reset_xps_maps(dev, dev_maps, type);
2449 
2450 	if (type == XPS_CPUS) {
2451 		for (i = offset + (count - 1); count--; i--)
2452 			netdev_queue_numa_node_write(
2453 				netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2454 	}
2455 }
2456 
2457 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2458 				   u16 count)
2459 {
2460 	if (!static_key_false(&xps_needed))
2461 		return;
2462 
2463 	cpus_read_lock();
2464 	mutex_lock(&xps_map_mutex);
2465 
2466 	if (static_key_false(&xps_rxqs_needed))
2467 		clean_xps_maps(dev, XPS_RXQS, offset, count);
2468 
2469 	clean_xps_maps(dev, XPS_CPUS, offset, count);
2470 
2471 	mutex_unlock(&xps_map_mutex);
2472 	cpus_read_unlock();
2473 }
2474 
2475 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2476 {
2477 	netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2478 }
2479 
2480 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2481 				      u16 index, bool is_rxqs_map)
2482 {
2483 	struct xps_map *new_map;
2484 	int alloc_len = XPS_MIN_MAP_ALLOC;
2485 	int i, pos;
2486 
2487 	for (pos = 0; map && pos < map->len; pos++) {
2488 		if (map->queues[pos] != index)
2489 			continue;
2490 		return map;
2491 	}
2492 
2493 	/* Need to add tx-queue to this CPU's/rx-queue's existing map */
2494 	if (map) {
2495 		if (pos < map->alloc_len)
2496 			return map;
2497 
2498 		alloc_len = map->alloc_len * 2;
2499 	}
2500 
2501 	/* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2502 	 *  map
2503 	 */
2504 	if (is_rxqs_map)
2505 		new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2506 	else
2507 		new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2508 				       cpu_to_node(attr_index));
2509 	if (!new_map)
2510 		return NULL;
2511 
2512 	for (i = 0; i < pos; i++)
2513 		new_map->queues[i] = map->queues[i];
2514 	new_map->alloc_len = alloc_len;
2515 	new_map->len = pos;
2516 
2517 	return new_map;
2518 }
2519 
2520 /* Copy xps maps at a given index */
2521 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2522 			      struct xps_dev_maps *new_dev_maps, int index,
2523 			      int tc, bool skip_tc)
2524 {
2525 	int i, tci = index * dev_maps->num_tc;
2526 	struct xps_map *map;
2527 
2528 	/* copy maps belonging to foreign traffic classes */
2529 	for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2530 		if (i == tc && skip_tc)
2531 			continue;
2532 
2533 		/* fill in the new device map from the old device map */
2534 		map = xmap_dereference(dev_maps->attr_map[tci]);
2535 		RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2536 	}
2537 }
2538 
2539 /* Must be called under cpus_read_lock */
2540 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2541 			  u16 index, enum xps_map_type type)
2542 {
2543 	struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2544 	const unsigned long *online_mask = NULL;
2545 	bool active = false, copy = false;
2546 	int i, j, tci, numa_node_id = -2;
2547 	int maps_sz, num_tc = 1, tc = 0;
2548 	struct xps_map *map, *new_map;
2549 	unsigned int nr_ids;
2550 
2551 	if (dev->num_tc) {
2552 		/* Do not allow XPS on subordinate device directly */
2553 		num_tc = dev->num_tc;
2554 		if (num_tc < 0)
2555 			return -EINVAL;
2556 
2557 		/* If queue belongs to subordinate dev use its map */
2558 		dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2559 
2560 		tc = netdev_txq_to_tc(dev, index);
2561 		if (tc < 0)
2562 			return -EINVAL;
2563 	}
2564 
2565 	mutex_lock(&xps_map_mutex);
2566 
2567 	dev_maps = xmap_dereference(dev->xps_maps[type]);
2568 	if (type == XPS_RXQS) {
2569 		maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2570 		nr_ids = dev->num_rx_queues;
2571 	} else {
2572 		maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2573 		if (num_possible_cpus() > 1)
2574 			online_mask = cpumask_bits(cpu_online_mask);
2575 		nr_ids = nr_cpu_ids;
2576 	}
2577 
2578 	if (maps_sz < L1_CACHE_BYTES)
2579 		maps_sz = L1_CACHE_BYTES;
2580 
2581 	/* The old dev_maps could be larger or smaller than the one we're
2582 	 * setting up now, as dev->num_tc or nr_ids could have been updated in
2583 	 * between. We could try to be smart, but let's be safe instead and only
2584 	 * copy foreign traffic classes if the two map sizes match.
2585 	 */
2586 	if (dev_maps &&
2587 	    dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2588 		copy = true;
2589 
2590 	/* allocate memory for queue storage */
2591 	for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2592 	     j < nr_ids;) {
2593 		if (!new_dev_maps) {
2594 			new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2595 			if (!new_dev_maps) {
2596 				mutex_unlock(&xps_map_mutex);
2597 				return -ENOMEM;
2598 			}
2599 
2600 			new_dev_maps->nr_ids = nr_ids;
2601 			new_dev_maps->num_tc = num_tc;
2602 		}
2603 
2604 		tci = j * num_tc + tc;
2605 		map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2606 
2607 		map = expand_xps_map(map, j, index, type == XPS_RXQS);
2608 		if (!map)
2609 			goto error;
2610 
2611 		RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2612 	}
2613 
2614 	if (!new_dev_maps)
2615 		goto out_no_new_maps;
2616 
2617 	if (!dev_maps) {
2618 		/* Increment static keys at most once per type */
2619 		static_key_slow_inc_cpuslocked(&xps_needed);
2620 		if (type == XPS_RXQS)
2621 			static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2622 	}
2623 
2624 	for (j = 0; j < nr_ids; j++) {
2625 		bool skip_tc = false;
2626 
2627 		tci = j * num_tc + tc;
2628 		if (netif_attr_test_mask(j, mask, nr_ids) &&
2629 		    netif_attr_test_online(j, online_mask, nr_ids)) {
2630 			/* add tx-queue to CPU/rx-queue maps */
2631 			int pos = 0;
2632 
2633 			skip_tc = true;
2634 
2635 			map = xmap_dereference(new_dev_maps->attr_map[tci]);
2636 			while ((pos < map->len) && (map->queues[pos] != index))
2637 				pos++;
2638 
2639 			if (pos == map->len)
2640 				map->queues[map->len++] = index;
2641 #ifdef CONFIG_NUMA
2642 			if (type == XPS_CPUS) {
2643 				if (numa_node_id == -2)
2644 					numa_node_id = cpu_to_node(j);
2645 				else if (numa_node_id != cpu_to_node(j))
2646 					numa_node_id = -1;
2647 			}
2648 #endif
2649 		}
2650 
2651 		if (copy)
2652 			xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2653 					  skip_tc);
2654 	}
2655 
2656 	rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2657 
2658 	/* Cleanup old maps */
2659 	if (!dev_maps)
2660 		goto out_no_old_maps;
2661 
2662 	for (j = 0; j < dev_maps->nr_ids; j++) {
2663 		for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2664 			map = xmap_dereference(dev_maps->attr_map[tci]);
2665 			if (!map)
2666 				continue;
2667 
2668 			if (copy) {
2669 				new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2670 				if (map == new_map)
2671 					continue;
2672 			}
2673 
2674 			RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2675 			kfree_rcu(map, rcu);
2676 		}
2677 	}
2678 
2679 	old_dev_maps = dev_maps;
2680 
2681 out_no_old_maps:
2682 	dev_maps = new_dev_maps;
2683 	active = true;
2684 
2685 out_no_new_maps:
2686 	if (type == XPS_CPUS)
2687 		/* update Tx queue numa node */
2688 		netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2689 					     (numa_node_id >= 0) ?
2690 					     numa_node_id : NUMA_NO_NODE);
2691 
2692 	if (!dev_maps)
2693 		goto out_no_maps;
2694 
2695 	/* removes tx-queue from unused CPUs/rx-queues */
2696 	for (j = 0; j < dev_maps->nr_ids; j++) {
2697 		tci = j * dev_maps->num_tc;
2698 
2699 		for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2700 			if (i == tc &&
2701 			    netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2702 			    netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2703 				continue;
2704 
2705 			active |= remove_xps_queue(dev_maps,
2706 						   copy ? old_dev_maps : NULL,
2707 						   tci, index);
2708 		}
2709 	}
2710 
2711 	if (old_dev_maps)
2712 		kfree_rcu(old_dev_maps, rcu);
2713 
2714 	/* free map if not active */
2715 	if (!active)
2716 		reset_xps_maps(dev, dev_maps, type);
2717 
2718 out_no_maps:
2719 	mutex_unlock(&xps_map_mutex);
2720 
2721 	return 0;
2722 error:
2723 	/* remove any maps that we added */
2724 	for (j = 0; j < nr_ids; j++) {
2725 		for (i = num_tc, tci = j * num_tc; i--; tci++) {
2726 			new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2727 			map = copy ?
2728 			      xmap_dereference(dev_maps->attr_map[tci]) :
2729 			      NULL;
2730 			if (new_map && new_map != map)
2731 				kfree(new_map);
2732 		}
2733 	}
2734 
2735 	mutex_unlock(&xps_map_mutex);
2736 
2737 	kfree(new_dev_maps);
2738 	return -ENOMEM;
2739 }
2740 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2741 
2742 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2743 			u16 index)
2744 {
2745 	int ret;
2746 
2747 	cpus_read_lock();
2748 	ret =  __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2749 	cpus_read_unlock();
2750 
2751 	return ret;
2752 }
2753 EXPORT_SYMBOL(netif_set_xps_queue);
2754 
2755 #endif
2756 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2757 {
2758 	struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2759 
2760 	/* Unbind any subordinate channels */
2761 	while (txq-- != &dev->_tx[0]) {
2762 		if (txq->sb_dev)
2763 			netdev_unbind_sb_channel(dev, txq->sb_dev);
2764 	}
2765 }
2766 
2767 void netdev_reset_tc(struct net_device *dev)
2768 {
2769 #ifdef CONFIG_XPS
2770 	netif_reset_xps_queues_gt(dev, 0);
2771 #endif
2772 	netdev_unbind_all_sb_channels(dev);
2773 
2774 	/* Reset TC configuration of device */
2775 	dev->num_tc = 0;
2776 	memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2777 	memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2778 }
2779 EXPORT_SYMBOL(netdev_reset_tc);
2780 
2781 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2782 {
2783 	if (tc >= dev->num_tc)
2784 		return -EINVAL;
2785 
2786 #ifdef CONFIG_XPS
2787 	netif_reset_xps_queues(dev, offset, count);
2788 #endif
2789 	dev->tc_to_txq[tc].count = count;
2790 	dev->tc_to_txq[tc].offset = offset;
2791 	return 0;
2792 }
2793 EXPORT_SYMBOL(netdev_set_tc_queue);
2794 
2795 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2796 {
2797 	if (num_tc > TC_MAX_QUEUE)
2798 		return -EINVAL;
2799 
2800 #ifdef CONFIG_XPS
2801 	netif_reset_xps_queues_gt(dev, 0);
2802 #endif
2803 	netdev_unbind_all_sb_channels(dev);
2804 
2805 	dev->num_tc = num_tc;
2806 	return 0;
2807 }
2808 EXPORT_SYMBOL(netdev_set_num_tc);
2809 
2810 void netdev_unbind_sb_channel(struct net_device *dev,
2811 			      struct net_device *sb_dev)
2812 {
2813 	struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2814 
2815 #ifdef CONFIG_XPS
2816 	netif_reset_xps_queues_gt(sb_dev, 0);
2817 #endif
2818 	memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2819 	memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2820 
2821 	while (txq-- != &dev->_tx[0]) {
2822 		if (txq->sb_dev == sb_dev)
2823 			txq->sb_dev = NULL;
2824 	}
2825 }
2826 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2827 
2828 int netdev_bind_sb_channel_queue(struct net_device *dev,
2829 				 struct net_device *sb_dev,
2830 				 u8 tc, u16 count, u16 offset)
2831 {
2832 	/* Make certain the sb_dev and dev are already configured */
2833 	if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2834 		return -EINVAL;
2835 
2836 	/* We cannot hand out queues we don't have */
2837 	if ((offset + count) > dev->real_num_tx_queues)
2838 		return -EINVAL;
2839 
2840 	/* Record the mapping */
2841 	sb_dev->tc_to_txq[tc].count = count;
2842 	sb_dev->tc_to_txq[tc].offset = offset;
2843 
2844 	/* Provide a way for Tx queue to find the tc_to_txq map or
2845 	 * XPS map for itself.
2846 	 */
2847 	while (count--)
2848 		netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2849 
2850 	return 0;
2851 }
2852 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2853 
2854 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2855 {
2856 	/* Do not use a multiqueue device to represent a subordinate channel */
2857 	if (netif_is_multiqueue(dev))
2858 		return -ENODEV;
2859 
2860 	/* We allow channels 1 - 32767 to be used for subordinate channels.
2861 	 * Channel 0 is meant to be "native" mode and used only to represent
2862 	 * the main root device. We allow writing 0 to reset the device back
2863 	 * to normal mode after being used as a subordinate channel.
2864 	 */
2865 	if (channel > S16_MAX)
2866 		return -EINVAL;
2867 
2868 	dev->num_tc = -channel;
2869 
2870 	return 0;
2871 }
2872 EXPORT_SYMBOL(netdev_set_sb_channel);
2873 
2874 /*
2875  * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2876  * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2877  */
2878 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2879 {
2880 	bool disabling;
2881 	int rc;
2882 
2883 	disabling = txq < dev->real_num_tx_queues;
2884 
2885 	if (txq < 1 || txq > dev->num_tx_queues)
2886 		return -EINVAL;
2887 
2888 	if (dev->reg_state == NETREG_REGISTERED ||
2889 	    dev->reg_state == NETREG_UNREGISTERING) {
2890 		ASSERT_RTNL();
2891 
2892 		rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2893 						  txq);
2894 		if (rc)
2895 			return rc;
2896 
2897 		if (dev->num_tc)
2898 			netif_setup_tc(dev, txq);
2899 
2900 		dev_qdisc_change_real_num_tx(dev, txq);
2901 
2902 		dev->real_num_tx_queues = txq;
2903 
2904 		if (disabling) {
2905 			synchronize_net();
2906 			qdisc_reset_all_tx_gt(dev, txq);
2907 #ifdef CONFIG_XPS
2908 			netif_reset_xps_queues_gt(dev, txq);
2909 #endif
2910 		}
2911 	} else {
2912 		dev->real_num_tx_queues = txq;
2913 	}
2914 
2915 	return 0;
2916 }
2917 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2918 
2919 #ifdef CONFIG_SYSFS
2920 /**
2921  *	netif_set_real_num_rx_queues - set actual number of RX queues used
2922  *	@dev: Network device
2923  *	@rxq: Actual number of RX queues
2924  *
2925  *	This must be called either with the rtnl_lock held or before
2926  *	registration of the net device.  Returns 0 on success, or a
2927  *	negative error code.  If called before registration, it always
2928  *	succeeds.
2929  */
2930 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2931 {
2932 	int rc;
2933 
2934 	if (rxq < 1 || rxq > dev->num_rx_queues)
2935 		return -EINVAL;
2936 
2937 	if (dev->reg_state == NETREG_REGISTERED) {
2938 		ASSERT_RTNL();
2939 
2940 		rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2941 						  rxq);
2942 		if (rc)
2943 			return rc;
2944 	}
2945 
2946 	dev->real_num_rx_queues = rxq;
2947 	return 0;
2948 }
2949 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2950 #endif
2951 
2952 /**
2953  *	netif_set_real_num_queues - set actual number of RX and TX queues used
2954  *	@dev: Network device
2955  *	@txq: Actual number of TX queues
2956  *	@rxq: Actual number of RX queues
2957  *
2958  *	Set the real number of both TX and RX queues.
2959  *	Does nothing if the number of queues is already correct.
2960  */
2961 int netif_set_real_num_queues(struct net_device *dev,
2962 			      unsigned int txq, unsigned int rxq)
2963 {
2964 	unsigned int old_rxq = dev->real_num_rx_queues;
2965 	int err;
2966 
2967 	if (txq < 1 || txq > dev->num_tx_queues ||
2968 	    rxq < 1 || rxq > dev->num_rx_queues)
2969 		return -EINVAL;
2970 
2971 	/* Start from increases, so the error path only does decreases -
2972 	 * decreases can't fail.
2973 	 */
2974 	if (rxq > dev->real_num_rx_queues) {
2975 		err = netif_set_real_num_rx_queues(dev, rxq);
2976 		if (err)
2977 			return err;
2978 	}
2979 	if (txq > dev->real_num_tx_queues) {
2980 		err = netif_set_real_num_tx_queues(dev, txq);
2981 		if (err)
2982 			goto undo_rx;
2983 	}
2984 	if (rxq < dev->real_num_rx_queues)
2985 		WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
2986 	if (txq < dev->real_num_tx_queues)
2987 		WARN_ON(netif_set_real_num_tx_queues(dev, txq));
2988 
2989 	return 0;
2990 undo_rx:
2991 	WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
2992 	return err;
2993 }
2994 EXPORT_SYMBOL(netif_set_real_num_queues);
2995 
2996 /**
2997  * netif_set_tso_max_size() - set the max size of TSO frames supported
2998  * @dev:	netdev to update
2999  * @size:	max skb->len of a TSO frame
3000  *
3001  * Set the limit on the size of TSO super-frames the device can handle.
3002  * Unless explicitly set the stack will assume the value of
3003  * %GSO_LEGACY_MAX_SIZE.
3004  */
3005 void netif_set_tso_max_size(struct net_device *dev, unsigned int size)
3006 {
3007 	dev->tso_max_size = min(GSO_MAX_SIZE, size);
3008 	if (size < READ_ONCE(dev->gso_max_size))
3009 		netif_set_gso_max_size(dev, size);
3010 }
3011 EXPORT_SYMBOL(netif_set_tso_max_size);
3012 
3013 /**
3014  * netif_set_tso_max_segs() - set the max number of segs supported for TSO
3015  * @dev:	netdev to update
3016  * @segs:	max number of TCP segments
3017  *
3018  * Set the limit on the number of TCP segments the device can generate from
3019  * a single TSO super-frame.
3020  * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS.
3021  */
3022 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs)
3023 {
3024 	dev->tso_max_segs = segs;
3025 	if (segs < READ_ONCE(dev->gso_max_segs))
3026 		netif_set_gso_max_segs(dev, segs);
3027 }
3028 EXPORT_SYMBOL(netif_set_tso_max_segs);
3029 
3030 /**
3031  * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper
3032  * @to:		netdev to update
3033  * @from:	netdev from which to copy the limits
3034  */
3035 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from)
3036 {
3037 	netif_set_tso_max_size(to, from->tso_max_size);
3038 	netif_set_tso_max_segs(to, from->tso_max_segs);
3039 }
3040 EXPORT_SYMBOL(netif_inherit_tso_max);
3041 
3042 /**
3043  * netif_get_num_default_rss_queues - default number of RSS queues
3044  *
3045  * Default value is the number of physical cores if there are only 1 or 2, or
3046  * divided by 2 if there are more.
3047  */
3048 int netif_get_num_default_rss_queues(void)
3049 {
3050 	cpumask_var_t cpus;
3051 	int cpu, count = 0;
3052 
3053 	if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL)))
3054 		return 1;
3055 
3056 	cpumask_copy(cpus, cpu_online_mask);
3057 	for_each_cpu(cpu, cpus) {
3058 		++count;
3059 		cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
3060 	}
3061 	free_cpumask_var(cpus);
3062 
3063 	return count > 2 ? DIV_ROUND_UP(count, 2) : count;
3064 }
3065 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3066 
3067 static void __netif_reschedule(struct Qdisc *q)
3068 {
3069 	struct softnet_data *sd;
3070 	unsigned long flags;
3071 
3072 	local_irq_save(flags);
3073 	sd = this_cpu_ptr(&softnet_data);
3074 	q->next_sched = NULL;
3075 	*sd->output_queue_tailp = q;
3076 	sd->output_queue_tailp = &q->next_sched;
3077 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
3078 	local_irq_restore(flags);
3079 }
3080 
3081 void __netif_schedule(struct Qdisc *q)
3082 {
3083 	if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3084 		__netif_reschedule(q);
3085 }
3086 EXPORT_SYMBOL(__netif_schedule);
3087 
3088 struct dev_kfree_skb_cb {
3089 	enum skb_free_reason reason;
3090 };
3091 
3092 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3093 {
3094 	return (struct dev_kfree_skb_cb *)skb->cb;
3095 }
3096 
3097 void netif_schedule_queue(struct netdev_queue *txq)
3098 {
3099 	rcu_read_lock();
3100 	if (!netif_xmit_stopped(txq)) {
3101 		struct Qdisc *q = rcu_dereference(txq->qdisc);
3102 
3103 		__netif_schedule(q);
3104 	}
3105 	rcu_read_unlock();
3106 }
3107 EXPORT_SYMBOL(netif_schedule_queue);
3108 
3109 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3110 {
3111 	if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3112 		struct Qdisc *q;
3113 
3114 		rcu_read_lock();
3115 		q = rcu_dereference(dev_queue->qdisc);
3116 		__netif_schedule(q);
3117 		rcu_read_unlock();
3118 	}
3119 }
3120 EXPORT_SYMBOL(netif_tx_wake_queue);
3121 
3122 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3123 {
3124 	unsigned long flags;
3125 
3126 	if (unlikely(!skb))
3127 		return;
3128 
3129 	if (likely(refcount_read(&skb->users) == 1)) {
3130 		smp_rmb();
3131 		refcount_set(&skb->users, 0);
3132 	} else if (likely(!refcount_dec_and_test(&skb->users))) {
3133 		return;
3134 	}
3135 	get_kfree_skb_cb(skb)->reason = reason;
3136 	local_irq_save(flags);
3137 	skb->next = __this_cpu_read(softnet_data.completion_queue);
3138 	__this_cpu_write(softnet_data.completion_queue, skb);
3139 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
3140 	local_irq_restore(flags);
3141 }
3142 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3143 
3144 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3145 {
3146 	if (in_hardirq() || irqs_disabled())
3147 		__dev_kfree_skb_irq(skb, reason);
3148 	else
3149 		dev_kfree_skb(skb);
3150 }
3151 EXPORT_SYMBOL(__dev_kfree_skb_any);
3152 
3153 
3154 /**
3155  * netif_device_detach - mark device as removed
3156  * @dev: network device
3157  *
3158  * Mark device as removed from system and therefore no longer available.
3159  */
3160 void netif_device_detach(struct net_device *dev)
3161 {
3162 	if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3163 	    netif_running(dev)) {
3164 		netif_tx_stop_all_queues(dev);
3165 	}
3166 }
3167 EXPORT_SYMBOL(netif_device_detach);
3168 
3169 /**
3170  * netif_device_attach - mark device as attached
3171  * @dev: network device
3172  *
3173  * Mark device as attached from system and restart if needed.
3174  */
3175 void netif_device_attach(struct net_device *dev)
3176 {
3177 	if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3178 	    netif_running(dev)) {
3179 		netif_tx_wake_all_queues(dev);
3180 		__netdev_watchdog_up(dev);
3181 	}
3182 }
3183 EXPORT_SYMBOL(netif_device_attach);
3184 
3185 /*
3186  * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3187  * to be used as a distribution range.
3188  */
3189 static u16 skb_tx_hash(const struct net_device *dev,
3190 		       const struct net_device *sb_dev,
3191 		       struct sk_buff *skb)
3192 {
3193 	u32 hash;
3194 	u16 qoffset = 0;
3195 	u16 qcount = dev->real_num_tx_queues;
3196 
3197 	if (dev->num_tc) {
3198 		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3199 
3200 		qoffset = sb_dev->tc_to_txq[tc].offset;
3201 		qcount = sb_dev->tc_to_txq[tc].count;
3202 		if (unlikely(!qcount)) {
3203 			net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3204 					     sb_dev->name, qoffset, tc);
3205 			qoffset = 0;
3206 			qcount = dev->real_num_tx_queues;
3207 		}
3208 	}
3209 
3210 	if (skb_rx_queue_recorded(skb)) {
3211 		hash = skb_get_rx_queue(skb);
3212 		if (hash >= qoffset)
3213 			hash -= qoffset;
3214 		while (unlikely(hash >= qcount))
3215 			hash -= qcount;
3216 		return hash + qoffset;
3217 	}
3218 
3219 	return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3220 }
3221 
3222 static void skb_warn_bad_offload(const struct sk_buff *skb)
3223 {
3224 	static const netdev_features_t null_features;
3225 	struct net_device *dev = skb->dev;
3226 	const char *name = "";
3227 
3228 	if (!net_ratelimit())
3229 		return;
3230 
3231 	if (dev) {
3232 		if (dev->dev.parent)
3233 			name = dev_driver_string(dev->dev.parent);
3234 		else
3235 			name = netdev_name(dev);
3236 	}
3237 	skb_dump(KERN_WARNING, skb, false);
3238 	WARN(1, "%s: caps=(%pNF, %pNF)\n",
3239 	     name, dev ? &dev->features : &null_features,
3240 	     skb->sk ? &skb->sk->sk_route_caps : &null_features);
3241 }
3242 
3243 /*
3244  * Invalidate hardware checksum when packet is to be mangled, and
3245  * complete checksum manually on outgoing path.
3246  */
3247 int skb_checksum_help(struct sk_buff *skb)
3248 {
3249 	__wsum csum;
3250 	int ret = 0, offset;
3251 
3252 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3253 		goto out_set_summed;
3254 
3255 	if (unlikely(skb_is_gso(skb))) {
3256 		skb_warn_bad_offload(skb);
3257 		return -EINVAL;
3258 	}
3259 
3260 	/* Before computing a checksum, we should make sure no frag could
3261 	 * be modified by an external entity : checksum could be wrong.
3262 	 */
3263 	if (skb_has_shared_frag(skb)) {
3264 		ret = __skb_linearize(skb);
3265 		if (ret)
3266 			goto out;
3267 	}
3268 
3269 	offset = skb_checksum_start_offset(skb);
3270 	ret = -EINVAL;
3271 	if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3272 		DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3273 		goto out;
3274 	}
3275 	csum = skb_checksum(skb, offset, skb->len - offset, 0);
3276 
3277 	offset += skb->csum_offset;
3278 	if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) {
3279 		DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3280 		goto out;
3281 	}
3282 	ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3283 	if (ret)
3284 		goto out;
3285 
3286 	*(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3287 out_set_summed:
3288 	skb->ip_summed = CHECKSUM_NONE;
3289 out:
3290 	return ret;
3291 }
3292 EXPORT_SYMBOL(skb_checksum_help);
3293 
3294 int skb_crc32c_csum_help(struct sk_buff *skb)
3295 {
3296 	__le32 crc32c_csum;
3297 	int ret = 0, offset, start;
3298 
3299 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3300 		goto out;
3301 
3302 	if (unlikely(skb_is_gso(skb)))
3303 		goto out;
3304 
3305 	/* Before computing a checksum, we should make sure no frag could
3306 	 * be modified by an external entity : checksum could be wrong.
3307 	 */
3308 	if (unlikely(skb_has_shared_frag(skb))) {
3309 		ret = __skb_linearize(skb);
3310 		if (ret)
3311 			goto out;
3312 	}
3313 	start = skb_checksum_start_offset(skb);
3314 	offset = start + offsetof(struct sctphdr, checksum);
3315 	if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3316 		ret = -EINVAL;
3317 		goto out;
3318 	}
3319 
3320 	ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3321 	if (ret)
3322 		goto out;
3323 
3324 	crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3325 						  skb->len - start, ~(__u32)0,
3326 						  crc32c_csum_stub));
3327 	*(__le32 *)(skb->data + offset) = crc32c_csum;
3328 	skb->ip_summed = CHECKSUM_NONE;
3329 	skb->csum_not_inet = 0;
3330 out:
3331 	return ret;
3332 }
3333 
3334 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3335 {
3336 	__be16 type = skb->protocol;
3337 
3338 	/* Tunnel gso handlers can set protocol to ethernet. */
3339 	if (type == htons(ETH_P_TEB)) {
3340 		struct ethhdr *eth;
3341 
3342 		if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3343 			return 0;
3344 
3345 		eth = (struct ethhdr *)skb->data;
3346 		type = eth->h_proto;
3347 	}
3348 
3349 	return __vlan_get_protocol(skb, type, depth);
3350 }
3351 
3352 /* openvswitch calls this on rx path, so we need a different check.
3353  */
3354 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3355 {
3356 	if (tx_path)
3357 		return skb->ip_summed != CHECKSUM_PARTIAL &&
3358 		       skb->ip_summed != CHECKSUM_UNNECESSARY;
3359 
3360 	return skb->ip_summed == CHECKSUM_NONE;
3361 }
3362 
3363 /**
3364  *	__skb_gso_segment - Perform segmentation on skb.
3365  *	@skb: buffer to segment
3366  *	@features: features for the output path (see dev->features)
3367  *	@tx_path: whether it is called in TX path
3368  *
3369  *	This function segments the given skb and returns a list of segments.
3370  *
3371  *	It may return NULL if the skb requires no segmentation.  This is
3372  *	only possible when GSO is used for verifying header integrity.
3373  *
3374  *	Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3375  */
3376 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3377 				  netdev_features_t features, bool tx_path)
3378 {
3379 	struct sk_buff *segs;
3380 
3381 	if (unlikely(skb_needs_check(skb, tx_path))) {
3382 		int err;
3383 
3384 		/* We're going to init ->check field in TCP or UDP header */
3385 		err = skb_cow_head(skb, 0);
3386 		if (err < 0)
3387 			return ERR_PTR(err);
3388 	}
3389 
3390 	/* Only report GSO partial support if it will enable us to
3391 	 * support segmentation on this frame without needing additional
3392 	 * work.
3393 	 */
3394 	if (features & NETIF_F_GSO_PARTIAL) {
3395 		netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3396 		struct net_device *dev = skb->dev;
3397 
3398 		partial_features |= dev->features & dev->gso_partial_features;
3399 		if (!skb_gso_ok(skb, features | partial_features))
3400 			features &= ~NETIF_F_GSO_PARTIAL;
3401 	}
3402 
3403 	BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3404 		     sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3405 
3406 	SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3407 	SKB_GSO_CB(skb)->encap_level = 0;
3408 
3409 	skb_reset_mac_header(skb);
3410 	skb_reset_mac_len(skb);
3411 
3412 	segs = skb_mac_gso_segment(skb, features);
3413 
3414 	if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3415 		skb_warn_bad_offload(skb);
3416 
3417 	return segs;
3418 }
3419 EXPORT_SYMBOL(__skb_gso_segment);
3420 
3421 /* Take action when hardware reception checksum errors are detected. */
3422 #ifdef CONFIG_BUG
3423 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3424 {
3425 	netdev_err(dev, "hw csum failure\n");
3426 	skb_dump(KERN_ERR, skb, true);
3427 	dump_stack();
3428 }
3429 
3430 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3431 {
3432 	DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3433 }
3434 EXPORT_SYMBOL(netdev_rx_csum_fault);
3435 #endif
3436 
3437 /* XXX: check that highmem exists at all on the given machine. */
3438 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3439 {
3440 #ifdef CONFIG_HIGHMEM
3441 	int i;
3442 
3443 	if (!(dev->features & NETIF_F_HIGHDMA)) {
3444 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3445 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3446 
3447 			if (PageHighMem(skb_frag_page(frag)))
3448 				return 1;
3449 		}
3450 	}
3451 #endif
3452 	return 0;
3453 }
3454 
3455 /* If MPLS offload request, verify we are testing hardware MPLS features
3456  * instead of standard features for the netdev.
3457  */
3458 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3459 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3460 					   netdev_features_t features,
3461 					   __be16 type)
3462 {
3463 	if (eth_p_mpls(type))
3464 		features &= skb->dev->mpls_features;
3465 
3466 	return features;
3467 }
3468 #else
3469 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3470 					   netdev_features_t features,
3471 					   __be16 type)
3472 {
3473 	return features;
3474 }
3475 #endif
3476 
3477 static netdev_features_t harmonize_features(struct sk_buff *skb,
3478 	netdev_features_t features)
3479 {
3480 	__be16 type;
3481 
3482 	type = skb_network_protocol(skb, NULL);
3483 	features = net_mpls_features(skb, features, type);
3484 
3485 	if (skb->ip_summed != CHECKSUM_NONE &&
3486 	    !can_checksum_protocol(features, type)) {
3487 		features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3488 	}
3489 	if (illegal_highdma(skb->dev, skb))
3490 		features &= ~NETIF_F_SG;
3491 
3492 	return features;
3493 }
3494 
3495 netdev_features_t passthru_features_check(struct sk_buff *skb,
3496 					  struct net_device *dev,
3497 					  netdev_features_t features)
3498 {
3499 	return features;
3500 }
3501 EXPORT_SYMBOL(passthru_features_check);
3502 
3503 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3504 					     struct net_device *dev,
3505 					     netdev_features_t features)
3506 {
3507 	return vlan_features_check(skb, features);
3508 }
3509 
3510 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3511 					    struct net_device *dev,
3512 					    netdev_features_t features)
3513 {
3514 	u16 gso_segs = skb_shinfo(skb)->gso_segs;
3515 
3516 	if (gso_segs > READ_ONCE(dev->gso_max_segs))
3517 		return features & ~NETIF_F_GSO_MASK;
3518 
3519 	if (!skb_shinfo(skb)->gso_type) {
3520 		skb_warn_bad_offload(skb);
3521 		return features & ~NETIF_F_GSO_MASK;
3522 	}
3523 
3524 	/* Support for GSO partial features requires software
3525 	 * intervention before we can actually process the packets
3526 	 * so we need to strip support for any partial features now
3527 	 * and we can pull them back in after we have partially
3528 	 * segmented the frame.
3529 	 */
3530 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3531 		features &= ~dev->gso_partial_features;
3532 
3533 	/* Make sure to clear the IPv4 ID mangling feature if the
3534 	 * IPv4 header has the potential to be fragmented.
3535 	 */
3536 	if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3537 		struct iphdr *iph = skb->encapsulation ?
3538 				    inner_ip_hdr(skb) : ip_hdr(skb);
3539 
3540 		if (!(iph->frag_off & htons(IP_DF)))
3541 			features &= ~NETIF_F_TSO_MANGLEID;
3542 	}
3543 
3544 	return features;
3545 }
3546 
3547 netdev_features_t netif_skb_features(struct sk_buff *skb)
3548 {
3549 	struct net_device *dev = skb->dev;
3550 	netdev_features_t features = dev->features;
3551 
3552 	if (skb_is_gso(skb))
3553 		features = gso_features_check(skb, dev, features);
3554 
3555 	/* If encapsulation offload request, verify we are testing
3556 	 * hardware encapsulation features instead of standard
3557 	 * features for the netdev
3558 	 */
3559 	if (skb->encapsulation)
3560 		features &= dev->hw_enc_features;
3561 
3562 	if (skb_vlan_tagged(skb))
3563 		features = netdev_intersect_features(features,
3564 						     dev->vlan_features |
3565 						     NETIF_F_HW_VLAN_CTAG_TX |
3566 						     NETIF_F_HW_VLAN_STAG_TX);
3567 
3568 	if (dev->netdev_ops->ndo_features_check)
3569 		features &= dev->netdev_ops->ndo_features_check(skb, dev,
3570 								features);
3571 	else
3572 		features &= dflt_features_check(skb, dev, features);
3573 
3574 	return harmonize_features(skb, features);
3575 }
3576 EXPORT_SYMBOL(netif_skb_features);
3577 
3578 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3579 		    struct netdev_queue *txq, bool more)
3580 {
3581 	unsigned int len;
3582 	int rc;
3583 
3584 	if (dev_nit_active(dev))
3585 		dev_queue_xmit_nit(skb, dev);
3586 
3587 	len = skb->len;
3588 	trace_net_dev_start_xmit(skb, dev);
3589 	rc = netdev_start_xmit(skb, dev, txq, more);
3590 	trace_net_dev_xmit(skb, rc, dev, len);
3591 
3592 	return rc;
3593 }
3594 
3595 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3596 				    struct netdev_queue *txq, int *ret)
3597 {
3598 	struct sk_buff *skb = first;
3599 	int rc = NETDEV_TX_OK;
3600 
3601 	while (skb) {
3602 		struct sk_buff *next = skb->next;
3603 
3604 		skb_mark_not_on_list(skb);
3605 		rc = xmit_one(skb, dev, txq, next != NULL);
3606 		if (unlikely(!dev_xmit_complete(rc))) {
3607 			skb->next = next;
3608 			goto out;
3609 		}
3610 
3611 		skb = next;
3612 		if (netif_tx_queue_stopped(txq) && skb) {
3613 			rc = NETDEV_TX_BUSY;
3614 			break;
3615 		}
3616 	}
3617 
3618 out:
3619 	*ret = rc;
3620 	return skb;
3621 }
3622 
3623 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3624 					  netdev_features_t features)
3625 {
3626 	if (skb_vlan_tag_present(skb) &&
3627 	    !vlan_hw_offload_capable(features, skb->vlan_proto))
3628 		skb = __vlan_hwaccel_push_inside(skb);
3629 	return skb;
3630 }
3631 
3632 int skb_csum_hwoffload_help(struct sk_buff *skb,
3633 			    const netdev_features_t features)
3634 {
3635 	if (unlikely(skb_csum_is_sctp(skb)))
3636 		return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3637 			skb_crc32c_csum_help(skb);
3638 
3639 	if (features & NETIF_F_HW_CSUM)
3640 		return 0;
3641 
3642 	if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3643 		switch (skb->csum_offset) {
3644 		case offsetof(struct tcphdr, check):
3645 		case offsetof(struct udphdr, check):
3646 			return 0;
3647 		}
3648 	}
3649 
3650 	return skb_checksum_help(skb);
3651 }
3652 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3653 
3654 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3655 {
3656 	netdev_features_t features;
3657 
3658 	features = netif_skb_features(skb);
3659 	skb = validate_xmit_vlan(skb, features);
3660 	if (unlikely(!skb))
3661 		goto out_null;
3662 
3663 	skb = sk_validate_xmit_skb(skb, dev);
3664 	if (unlikely(!skb))
3665 		goto out_null;
3666 
3667 	if (netif_needs_gso(skb, features)) {
3668 		struct sk_buff *segs;
3669 
3670 		segs = skb_gso_segment(skb, features);
3671 		if (IS_ERR(segs)) {
3672 			goto out_kfree_skb;
3673 		} else if (segs) {
3674 			consume_skb(skb);
3675 			skb = segs;
3676 		}
3677 	} else {
3678 		if (skb_needs_linearize(skb, features) &&
3679 		    __skb_linearize(skb))
3680 			goto out_kfree_skb;
3681 
3682 		/* If packet is not checksummed and device does not
3683 		 * support checksumming for this protocol, complete
3684 		 * checksumming here.
3685 		 */
3686 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
3687 			if (skb->encapsulation)
3688 				skb_set_inner_transport_header(skb,
3689 							       skb_checksum_start_offset(skb));
3690 			else
3691 				skb_set_transport_header(skb,
3692 							 skb_checksum_start_offset(skb));
3693 			if (skb_csum_hwoffload_help(skb, features))
3694 				goto out_kfree_skb;
3695 		}
3696 	}
3697 
3698 	skb = validate_xmit_xfrm(skb, features, again);
3699 
3700 	return skb;
3701 
3702 out_kfree_skb:
3703 	kfree_skb(skb);
3704 out_null:
3705 	dev_core_stats_tx_dropped_inc(dev);
3706 	return NULL;
3707 }
3708 
3709 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3710 {
3711 	struct sk_buff *next, *head = NULL, *tail;
3712 
3713 	for (; skb != NULL; skb = next) {
3714 		next = skb->next;
3715 		skb_mark_not_on_list(skb);
3716 
3717 		/* in case skb wont be segmented, point to itself */
3718 		skb->prev = skb;
3719 
3720 		skb = validate_xmit_skb(skb, dev, again);
3721 		if (!skb)
3722 			continue;
3723 
3724 		if (!head)
3725 			head = skb;
3726 		else
3727 			tail->next = skb;
3728 		/* If skb was segmented, skb->prev points to
3729 		 * the last segment. If not, it still contains skb.
3730 		 */
3731 		tail = skb->prev;
3732 	}
3733 	return head;
3734 }
3735 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3736 
3737 static void qdisc_pkt_len_init(struct sk_buff *skb)
3738 {
3739 	const struct skb_shared_info *shinfo = skb_shinfo(skb);
3740 
3741 	qdisc_skb_cb(skb)->pkt_len = skb->len;
3742 
3743 	/* To get more precise estimation of bytes sent on wire,
3744 	 * we add to pkt_len the headers size of all segments
3745 	 */
3746 	if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3747 		unsigned int hdr_len;
3748 		u16 gso_segs = shinfo->gso_segs;
3749 
3750 		/* mac layer + network layer */
3751 		hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3752 
3753 		/* + transport layer */
3754 		if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3755 			const struct tcphdr *th;
3756 			struct tcphdr _tcphdr;
3757 
3758 			th = skb_header_pointer(skb, skb_transport_offset(skb),
3759 						sizeof(_tcphdr), &_tcphdr);
3760 			if (likely(th))
3761 				hdr_len += __tcp_hdrlen(th);
3762 		} else {
3763 			struct udphdr _udphdr;
3764 
3765 			if (skb_header_pointer(skb, skb_transport_offset(skb),
3766 					       sizeof(_udphdr), &_udphdr))
3767 				hdr_len += sizeof(struct udphdr);
3768 		}
3769 
3770 		if (shinfo->gso_type & SKB_GSO_DODGY)
3771 			gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3772 						shinfo->gso_size);
3773 
3774 		qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3775 	}
3776 }
3777 
3778 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3779 			     struct sk_buff **to_free,
3780 			     struct netdev_queue *txq)
3781 {
3782 	int rc;
3783 
3784 	rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3785 	if (rc == NET_XMIT_SUCCESS)
3786 		trace_qdisc_enqueue(q, txq, skb);
3787 	return rc;
3788 }
3789 
3790 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3791 				 struct net_device *dev,
3792 				 struct netdev_queue *txq)
3793 {
3794 	spinlock_t *root_lock = qdisc_lock(q);
3795 	struct sk_buff *to_free = NULL;
3796 	bool contended;
3797 	int rc;
3798 
3799 	qdisc_calculate_pkt_len(skb, q);
3800 
3801 	if (q->flags & TCQ_F_NOLOCK) {
3802 		if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3803 		    qdisc_run_begin(q)) {
3804 			/* Retest nolock_qdisc_is_empty() within the protection
3805 			 * of q->seqlock to protect from racing with requeuing.
3806 			 */
3807 			if (unlikely(!nolock_qdisc_is_empty(q))) {
3808 				rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3809 				__qdisc_run(q);
3810 				qdisc_run_end(q);
3811 
3812 				goto no_lock_out;
3813 			}
3814 
3815 			qdisc_bstats_cpu_update(q, skb);
3816 			if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3817 			    !nolock_qdisc_is_empty(q))
3818 				__qdisc_run(q);
3819 
3820 			qdisc_run_end(q);
3821 			return NET_XMIT_SUCCESS;
3822 		}
3823 
3824 		rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3825 		qdisc_run(q);
3826 
3827 no_lock_out:
3828 		if (unlikely(to_free))
3829 			kfree_skb_list_reason(to_free,
3830 					      SKB_DROP_REASON_QDISC_DROP);
3831 		return rc;
3832 	}
3833 
3834 	/*
3835 	 * Heuristic to force contended enqueues to serialize on a
3836 	 * separate lock before trying to get qdisc main lock.
3837 	 * This permits qdisc->running owner to get the lock more
3838 	 * often and dequeue packets faster.
3839 	 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
3840 	 * and then other tasks will only enqueue packets. The packets will be
3841 	 * sent after the qdisc owner is scheduled again. To prevent this
3842 	 * scenario the task always serialize on the lock.
3843 	 */
3844 	contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
3845 	if (unlikely(contended))
3846 		spin_lock(&q->busylock);
3847 
3848 	spin_lock(root_lock);
3849 	if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3850 		__qdisc_drop(skb, &to_free);
3851 		rc = NET_XMIT_DROP;
3852 	} else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3853 		   qdisc_run_begin(q)) {
3854 		/*
3855 		 * This is a work-conserving queue; there are no old skbs
3856 		 * waiting to be sent out; and the qdisc is not running -
3857 		 * xmit the skb directly.
3858 		 */
3859 
3860 		qdisc_bstats_update(q, skb);
3861 
3862 		if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3863 			if (unlikely(contended)) {
3864 				spin_unlock(&q->busylock);
3865 				contended = false;
3866 			}
3867 			__qdisc_run(q);
3868 		}
3869 
3870 		qdisc_run_end(q);
3871 		rc = NET_XMIT_SUCCESS;
3872 	} else {
3873 		rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3874 		if (qdisc_run_begin(q)) {
3875 			if (unlikely(contended)) {
3876 				spin_unlock(&q->busylock);
3877 				contended = false;
3878 			}
3879 			__qdisc_run(q);
3880 			qdisc_run_end(q);
3881 		}
3882 	}
3883 	spin_unlock(root_lock);
3884 	if (unlikely(to_free))
3885 		kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP);
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 sk_buff *
3938 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3939 {
3940 #ifdef CONFIG_NET_CLS_ACT
3941 	struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3942 	struct tcf_result cl_res;
3943 
3944 	if (!miniq)
3945 		return skb;
3946 
3947 	/* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3948 	tc_skb_cb(skb)->mru = 0;
3949 	tc_skb_cb(skb)->post_ct = false;
3950 	mini_qdisc_bstats_cpu_update(miniq, skb);
3951 
3952 	switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
3953 	case TC_ACT_OK:
3954 	case TC_ACT_RECLASSIFY:
3955 		skb->tc_index = TC_H_MIN(cl_res.classid);
3956 		break;
3957 	case TC_ACT_SHOT:
3958 		mini_qdisc_qstats_cpu_drop(miniq);
3959 		*ret = NET_XMIT_DROP;
3960 		kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS);
3961 		return NULL;
3962 	case TC_ACT_STOLEN:
3963 	case TC_ACT_QUEUED:
3964 	case TC_ACT_TRAP:
3965 		*ret = NET_XMIT_SUCCESS;
3966 		consume_skb(skb);
3967 		return NULL;
3968 	case TC_ACT_REDIRECT:
3969 		/* No need to push/pop skb's mac_header here on egress! */
3970 		skb_do_redirect(skb);
3971 		*ret = NET_XMIT_SUCCESS;
3972 		return NULL;
3973 	default:
3974 		break;
3975 	}
3976 #endif /* CONFIG_NET_CLS_ACT */
3977 
3978 	return skb;
3979 }
3980 
3981 static struct netdev_queue *
3982 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
3983 {
3984 	int qm = skb_get_queue_mapping(skb);
3985 
3986 	return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
3987 }
3988 
3989 static bool netdev_xmit_txqueue_skipped(void)
3990 {
3991 	return __this_cpu_read(softnet_data.xmit.skip_txqueue);
3992 }
3993 
3994 void netdev_xmit_skip_txqueue(bool skip)
3995 {
3996 	__this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
3997 }
3998 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
3999 #endif /* CONFIG_NET_EGRESS */
4000 
4001 #ifdef CONFIG_XPS
4002 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
4003 			       struct xps_dev_maps *dev_maps, unsigned int tci)
4004 {
4005 	int tc = netdev_get_prio_tc_map(dev, skb->priority);
4006 	struct xps_map *map;
4007 	int queue_index = -1;
4008 
4009 	if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4010 		return queue_index;
4011 
4012 	tci *= dev_maps->num_tc;
4013 	tci += tc;
4014 
4015 	map = rcu_dereference(dev_maps->attr_map[tci]);
4016 	if (map) {
4017 		if (map->len == 1)
4018 			queue_index = map->queues[0];
4019 		else
4020 			queue_index = map->queues[reciprocal_scale(
4021 						skb_get_hash(skb), map->len)];
4022 		if (unlikely(queue_index >= dev->real_num_tx_queues))
4023 			queue_index = -1;
4024 	}
4025 	return queue_index;
4026 }
4027 #endif
4028 
4029 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4030 			 struct sk_buff *skb)
4031 {
4032 #ifdef CONFIG_XPS
4033 	struct xps_dev_maps *dev_maps;
4034 	struct sock *sk = skb->sk;
4035 	int queue_index = -1;
4036 
4037 	if (!static_key_false(&xps_needed))
4038 		return -1;
4039 
4040 	rcu_read_lock();
4041 	if (!static_key_false(&xps_rxqs_needed))
4042 		goto get_cpus_map;
4043 
4044 	dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4045 	if (dev_maps) {
4046 		int tci = sk_rx_queue_get(sk);
4047 
4048 		if (tci >= 0)
4049 			queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4050 							  tci);
4051 	}
4052 
4053 get_cpus_map:
4054 	if (queue_index < 0) {
4055 		dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4056 		if (dev_maps) {
4057 			unsigned int tci = skb->sender_cpu - 1;
4058 
4059 			queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4060 							  tci);
4061 		}
4062 	}
4063 	rcu_read_unlock();
4064 
4065 	return queue_index;
4066 #else
4067 	return -1;
4068 #endif
4069 }
4070 
4071 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4072 		     struct net_device *sb_dev)
4073 {
4074 	return 0;
4075 }
4076 EXPORT_SYMBOL(dev_pick_tx_zero);
4077 
4078 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4079 		       struct net_device *sb_dev)
4080 {
4081 	return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4082 }
4083 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4084 
4085 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4086 		     struct net_device *sb_dev)
4087 {
4088 	struct sock *sk = skb->sk;
4089 	int queue_index = sk_tx_queue_get(sk);
4090 
4091 	sb_dev = sb_dev ? : dev;
4092 
4093 	if (queue_index < 0 || skb->ooo_okay ||
4094 	    queue_index >= dev->real_num_tx_queues) {
4095 		int new_index = get_xps_queue(dev, sb_dev, skb);
4096 
4097 		if (new_index < 0)
4098 			new_index = skb_tx_hash(dev, sb_dev, skb);
4099 
4100 		if (queue_index != new_index && sk &&
4101 		    sk_fullsock(sk) &&
4102 		    rcu_access_pointer(sk->sk_dst_cache))
4103 			sk_tx_queue_set(sk, new_index);
4104 
4105 		queue_index = new_index;
4106 	}
4107 
4108 	return queue_index;
4109 }
4110 EXPORT_SYMBOL(netdev_pick_tx);
4111 
4112 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4113 					 struct sk_buff *skb,
4114 					 struct net_device *sb_dev)
4115 {
4116 	int queue_index = 0;
4117 
4118 #ifdef CONFIG_XPS
4119 	u32 sender_cpu = skb->sender_cpu - 1;
4120 
4121 	if (sender_cpu >= (u32)NR_CPUS)
4122 		skb->sender_cpu = raw_smp_processor_id() + 1;
4123 #endif
4124 
4125 	if (dev->real_num_tx_queues != 1) {
4126 		const struct net_device_ops *ops = dev->netdev_ops;
4127 
4128 		if (ops->ndo_select_queue)
4129 			queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4130 		else
4131 			queue_index = netdev_pick_tx(dev, skb, sb_dev);
4132 
4133 		queue_index = netdev_cap_txqueue(dev, queue_index);
4134 	}
4135 
4136 	skb_set_queue_mapping(skb, queue_index);
4137 	return netdev_get_tx_queue(dev, queue_index);
4138 }
4139 
4140 /**
4141  * __dev_queue_xmit() - transmit a buffer
4142  * @skb:	buffer to transmit
4143  * @sb_dev:	suboordinate device used for L2 forwarding offload
4144  *
4145  * Queue a buffer for transmission to a network device. The caller must
4146  * have set the device and priority and built the buffer before calling
4147  * this function. The function can be called from an interrupt.
4148  *
4149  * When calling this method, interrupts MUST be enabled. This is because
4150  * the BH enable code must have IRQs enabled so that it will not deadlock.
4151  *
4152  * Regardless of the return value, the skb is consumed, so it is currently
4153  * difficult to retry a send to this method. (You can bump the ref count
4154  * before sending to hold a reference for retry if you are careful.)
4155  *
4156  * Return:
4157  * * 0				- buffer successfully transmitted
4158  * * positive qdisc return code	- NET_XMIT_DROP etc.
4159  * * negative errno		- other errors
4160  */
4161 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4162 {
4163 	struct net_device *dev = skb->dev;
4164 	struct netdev_queue *txq = NULL;
4165 	struct Qdisc *q;
4166 	int rc = -ENOMEM;
4167 	bool again = false;
4168 
4169 	skb_reset_mac_header(skb);
4170 	skb_assert_len(skb);
4171 
4172 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4173 		__skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4174 
4175 	/* Disable soft irqs for various locks below. Also
4176 	 * stops preemption for RCU.
4177 	 */
4178 	rcu_read_lock_bh();
4179 
4180 	skb_update_prio(skb);
4181 
4182 	qdisc_pkt_len_init(skb);
4183 #ifdef CONFIG_NET_CLS_ACT
4184 	skb->tc_at_ingress = 0;
4185 #endif
4186 #ifdef CONFIG_NET_EGRESS
4187 	if (static_branch_unlikely(&egress_needed_key)) {
4188 		if (nf_hook_egress_active()) {
4189 			skb = nf_hook_egress(skb, &rc, dev);
4190 			if (!skb)
4191 				goto out;
4192 		}
4193 
4194 		netdev_xmit_skip_txqueue(false);
4195 
4196 		nf_skip_egress(skb, true);
4197 		skb = sch_handle_egress(skb, &rc, dev);
4198 		if (!skb)
4199 			goto out;
4200 		nf_skip_egress(skb, false);
4201 
4202 		if (netdev_xmit_txqueue_skipped())
4203 			txq = netdev_tx_queue_mapping(dev, skb);
4204 	}
4205 #endif
4206 	/* If device/qdisc don't need skb->dst, release it right now while
4207 	 * its hot in this cpu cache.
4208 	 */
4209 	if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4210 		skb_dst_drop(skb);
4211 	else
4212 		skb_dst_force(skb);
4213 
4214 	if (!txq)
4215 		txq = netdev_core_pick_tx(dev, skb, sb_dev);
4216 
4217 	q = rcu_dereference_bh(txq->qdisc);
4218 
4219 	trace_net_dev_queue(skb);
4220 	if (q->enqueue) {
4221 		rc = __dev_xmit_skb(skb, q, dev, txq);
4222 		goto out;
4223 	}
4224 
4225 	/* The device has no queue. Common case for software devices:
4226 	 * loopback, all the sorts of tunnels...
4227 
4228 	 * Really, it is unlikely that netif_tx_lock protection is necessary
4229 	 * here.  (f.e. loopback and IP tunnels are clean ignoring statistics
4230 	 * counters.)
4231 	 * However, it is possible, that they rely on protection
4232 	 * made by us here.
4233 
4234 	 * Check this and shot the lock. It is not prone from deadlocks.
4235 	 *Either shot noqueue qdisc, it is even simpler 8)
4236 	 */
4237 	if (dev->flags & IFF_UP) {
4238 		int cpu = smp_processor_id(); /* ok because BHs are off */
4239 
4240 		/* Other cpus might concurrently change txq->xmit_lock_owner
4241 		 * to -1 or to their cpu id, but not to our id.
4242 		 */
4243 		if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4244 			if (dev_xmit_recursion())
4245 				goto recursion_alert;
4246 
4247 			skb = validate_xmit_skb(skb, dev, &again);
4248 			if (!skb)
4249 				goto out;
4250 
4251 			HARD_TX_LOCK(dev, txq, cpu);
4252 
4253 			if (!netif_xmit_stopped(txq)) {
4254 				dev_xmit_recursion_inc();
4255 				skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4256 				dev_xmit_recursion_dec();
4257 				if (dev_xmit_complete(rc)) {
4258 					HARD_TX_UNLOCK(dev, txq);
4259 					goto out;
4260 				}
4261 			}
4262 			HARD_TX_UNLOCK(dev, txq);
4263 			net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4264 					     dev->name);
4265 		} else {
4266 			/* Recursion is detected! It is possible,
4267 			 * unfortunately
4268 			 */
4269 recursion_alert:
4270 			net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4271 					     dev->name);
4272 		}
4273 	}
4274 
4275 	rc = -ENETDOWN;
4276 	rcu_read_unlock_bh();
4277 
4278 	dev_core_stats_tx_dropped_inc(dev);
4279 	kfree_skb_list(skb);
4280 	return rc;
4281 out:
4282 	rcu_read_unlock_bh();
4283 	return rc;
4284 }
4285 EXPORT_SYMBOL(__dev_queue_xmit);
4286 
4287 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4288 {
4289 	struct net_device *dev = skb->dev;
4290 	struct sk_buff *orig_skb = skb;
4291 	struct netdev_queue *txq;
4292 	int ret = NETDEV_TX_BUSY;
4293 	bool again = false;
4294 
4295 	if (unlikely(!netif_running(dev) ||
4296 		     !netif_carrier_ok(dev)))
4297 		goto drop;
4298 
4299 	skb = validate_xmit_skb_list(skb, dev, &again);
4300 	if (skb != orig_skb)
4301 		goto drop;
4302 
4303 	skb_set_queue_mapping(skb, queue_id);
4304 	txq = skb_get_tx_queue(dev, skb);
4305 
4306 	local_bh_disable();
4307 
4308 	dev_xmit_recursion_inc();
4309 	HARD_TX_LOCK(dev, txq, smp_processor_id());
4310 	if (!netif_xmit_frozen_or_drv_stopped(txq))
4311 		ret = netdev_start_xmit(skb, dev, txq, false);
4312 	HARD_TX_UNLOCK(dev, txq);
4313 	dev_xmit_recursion_dec();
4314 
4315 	local_bh_enable();
4316 	return ret;
4317 drop:
4318 	dev_core_stats_tx_dropped_inc(dev);
4319 	kfree_skb_list(skb);
4320 	return NET_XMIT_DROP;
4321 }
4322 EXPORT_SYMBOL(__dev_direct_xmit);
4323 
4324 /*************************************************************************
4325  *			Receiver routines
4326  *************************************************************************/
4327 
4328 int netdev_max_backlog __read_mostly = 1000;
4329 EXPORT_SYMBOL(netdev_max_backlog);
4330 
4331 int netdev_tstamp_prequeue __read_mostly = 1;
4332 unsigned int sysctl_skb_defer_max __read_mostly = 64;
4333 int netdev_budget __read_mostly = 300;
4334 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4335 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4336 int weight_p __read_mostly = 64;           /* old backlog weight */
4337 int dev_weight_rx_bias __read_mostly = 1;  /* bias for backlog weight */
4338 int dev_weight_tx_bias __read_mostly = 1;  /* bias for output_queue quota */
4339 int dev_rx_weight __read_mostly = 64;
4340 int dev_tx_weight __read_mostly = 64;
4341 
4342 /* Called with irq disabled */
4343 static inline void ____napi_schedule(struct softnet_data *sd,
4344 				     struct napi_struct *napi)
4345 {
4346 	struct task_struct *thread;
4347 
4348 	lockdep_assert_irqs_disabled();
4349 
4350 	if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4351 		/* Paired with smp_mb__before_atomic() in
4352 		 * napi_enable()/dev_set_threaded().
4353 		 * Use READ_ONCE() to guarantee a complete
4354 		 * read on napi->thread. Only call
4355 		 * wake_up_process() when it's not NULL.
4356 		 */
4357 		thread = READ_ONCE(napi->thread);
4358 		if (thread) {
4359 			/* Avoid doing set_bit() if the thread is in
4360 			 * INTERRUPTIBLE state, cause napi_thread_wait()
4361 			 * makes sure to proceed with napi polling
4362 			 * if the thread is explicitly woken from here.
4363 			 */
4364 			if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4365 				set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4366 			wake_up_process(thread);
4367 			return;
4368 		}
4369 	}
4370 
4371 	list_add_tail(&napi->poll_list, &sd->poll_list);
4372 	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4373 }
4374 
4375 #ifdef CONFIG_RPS
4376 
4377 /* One global table that all flow-based protocols share. */
4378 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4379 EXPORT_SYMBOL(rps_sock_flow_table);
4380 u32 rps_cpu_mask __read_mostly;
4381 EXPORT_SYMBOL(rps_cpu_mask);
4382 
4383 struct static_key_false rps_needed __read_mostly;
4384 EXPORT_SYMBOL(rps_needed);
4385 struct static_key_false rfs_needed __read_mostly;
4386 EXPORT_SYMBOL(rfs_needed);
4387 
4388 static struct rps_dev_flow *
4389 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4390 	    struct rps_dev_flow *rflow, u16 next_cpu)
4391 {
4392 	if (next_cpu < nr_cpu_ids) {
4393 #ifdef CONFIG_RFS_ACCEL
4394 		struct netdev_rx_queue *rxqueue;
4395 		struct rps_dev_flow_table *flow_table;
4396 		struct rps_dev_flow *old_rflow;
4397 		u32 flow_id;
4398 		u16 rxq_index;
4399 		int rc;
4400 
4401 		/* Should we steer this flow to a different hardware queue? */
4402 		if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4403 		    !(dev->features & NETIF_F_NTUPLE))
4404 			goto out;
4405 		rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4406 		if (rxq_index == skb_get_rx_queue(skb))
4407 			goto out;
4408 
4409 		rxqueue = dev->_rx + rxq_index;
4410 		flow_table = rcu_dereference(rxqueue->rps_flow_table);
4411 		if (!flow_table)
4412 			goto out;
4413 		flow_id = skb_get_hash(skb) & flow_table->mask;
4414 		rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4415 							rxq_index, flow_id);
4416 		if (rc < 0)
4417 			goto out;
4418 		old_rflow = rflow;
4419 		rflow = &flow_table->flows[flow_id];
4420 		rflow->filter = rc;
4421 		if (old_rflow->filter == rflow->filter)
4422 			old_rflow->filter = RPS_NO_FILTER;
4423 	out:
4424 #endif
4425 		rflow->last_qtail =
4426 			per_cpu(softnet_data, next_cpu).input_queue_head;
4427 	}
4428 
4429 	rflow->cpu = next_cpu;
4430 	return rflow;
4431 }
4432 
4433 /*
4434  * get_rps_cpu is called from netif_receive_skb and returns the target
4435  * CPU from the RPS map of the receiving queue for a given skb.
4436  * rcu_read_lock must be held on entry.
4437  */
4438 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4439 		       struct rps_dev_flow **rflowp)
4440 {
4441 	const struct rps_sock_flow_table *sock_flow_table;
4442 	struct netdev_rx_queue *rxqueue = dev->_rx;
4443 	struct rps_dev_flow_table *flow_table;
4444 	struct rps_map *map;
4445 	int cpu = -1;
4446 	u32 tcpu;
4447 	u32 hash;
4448 
4449 	if (skb_rx_queue_recorded(skb)) {
4450 		u16 index = skb_get_rx_queue(skb);
4451 
4452 		if (unlikely(index >= dev->real_num_rx_queues)) {
4453 			WARN_ONCE(dev->real_num_rx_queues > 1,
4454 				  "%s received packet on queue %u, but number "
4455 				  "of RX queues is %u\n",
4456 				  dev->name, index, dev->real_num_rx_queues);
4457 			goto done;
4458 		}
4459 		rxqueue += index;
4460 	}
4461 
4462 	/* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4463 
4464 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
4465 	map = rcu_dereference(rxqueue->rps_map);
4466 	if (!flow_table && !map)
4467 		goto done;
4468 
4469 	skb_reset_network_header(skb);
4470 	hash = skb_get_hash(skb);
4471 	if (!hash)
4472 		goto done;
4473 
4474 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
4475 	if (flow_table && sock_flow_table) {
4476 		struct rps_dev_flow *rflow;
4477 		u32 next_cpu;
4478 		u32 ident;
4479 
4480 		/* First check into global flow table if there is a match */
4481 		ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4482 		if ((ident ^ hash) & ~rps_cpu_mask)
4483 			goto try_rps;
4484 
4485 		next_cpu = ident & rps_cpu_mask;
4486 
4487 		/* OK, now we know there is a match,
4488 		 * we can look at the local (per receive queue) flow table
4489 		 */
4490 		rflow = &flow_table->flows[hash & flow_table->mask];
4491 		tcpu = rflow->cpu;
4492 
4493 		/*
4494 		 * If the desired CPU (where last recvmsg was done) is
4495 		 * different from current CPU (one in the rx-queue flow
4496 		 * table entry), switch if one of the following holds:
4497 		 *   - Current CPU is unset (>= nr_cpu_ids).
4498 		 *   - Current CPU is offline.
4499 		 *   - The current CPU's queue tail has advanced beyond the
4500 		 *     last packet that was enqueued using this table entry.
4501 		 *     This guarantees that all previous packets for the flow
4502 		 *     have been dequeued, thus preserving in order delivery.
4503 		 */
4504 		if (unlikely(tcpu != next_cpu) &&
4505 		    (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4506 		     ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4507 		      rflow->last_qtail)) >= 0)) {
4508 			tcpu = next_cpu;
4509 			rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4510 		}
4511 
4512 		if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4513 			*rflowp = rflow;
4514 			cpu = tcpu;
4515 			goto done;
4516 		}
4517 	}
4518 
4519 try_rps:
4520 
4521 	if (map) {
4522 		tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4523 		if (cpu_online(tcpu)) {
4524 			cpu = tcpu;
4525 			goto done;
4526 		}
4527 	}
4528 
4529 done:
4530 	return cpu;
4531 }
4532 
4533 #ifdef CONFIG_RFS_ACCEL
4534 
4535 /**
4536  * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4537  * @dev: Device on which the filter was set
4538  * @rxq_index: RX queue index
4539  * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4540  * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4541  *
4542  * Drivers that implement ndo_rx_flow_steer() should periodically call
4543  * this function for each installed filter and remove the filters for
4544  * which it returns %true.
4545  */
4546 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4547 			 u32 flow_id, u16 filter_id)
4548 {
4549 	struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4550 	struct rps_dev_flow_table *flow_table;
4551 	struct rps_dev_flow *rflow;
4552 	bool expire = true;
4553 	unsigned int cpu;
4554 
4555 	rcu_read_lock();
4556 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
4557 	if (flow_table && flow_id <= flow_table->mask) {
4558 		rflow = &flow_table->flows[flow_id];
4559 		cpu = READ_ONCE(rflow->cpu);
4560 		if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4561 		    ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4562 			   rflow->last_qtail) <
4563 		     (int)(10 * flow_table->mask)))
4564 			expire = false;
4565 	}
4566 	rcu_read_unlock();
4567 	return expire;
4568 }
4569 EXPORT_SYMBOL(rps_may_expire_flow);
4570 
4571 #endif /* CONFIG_RFS_ACCEL */
4572 
4573 /* Called from hardirq (IPI) context */
4574 static void rps_trigger_softirq(void *data)
4575 {
4576 	struct softnet_data *sd = data;
4577 
4578 	____napi_schedule(sd, &sd->backlog);
4579 	sd->received_rps++;
4580 }
4581 
4582 #endif /* CONFIG_RPS */
4583 
4584 /* Called from hardirq (IPI) context */
4585 static void trigger_rx_softirq(void *data)
4586 {
4587 	struct softnet_data *sd = data;
4588 
4589 	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4590 	smp_store_release(&sd->defer_ipi_scheduled, 0);
4591 }
4592 
4593 /*
4594  * Check if this softnet_data structure is another cpu one
4595  * If yes, queue it to our IPI list and return 1
4596  * If no, return 0
4597  */
4598 static int napi_schedule_rps(struct softnet_data *sd)
4599 {
4600 	struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4601 
4602 #ifdef CONFIG_RPS
4603 	if (sd != mysd) {
4604 		sd->rps_ipi_next = mysd->rps_ipi_list;
4605 		mysd->rps_ipi_list = sd;
4606 
4607 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4608 		return 1;
4609 	}
4610 #endif /* CONFIG_RPS */
4611 	__napi_schedule_irqoff(&mysd->backlog);
4612 	return 0;
4613 }
4614 
4615 #ifdef CONFIG_NET_FLOW_LIMIT
4616 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4617 #endif
4618 
4619 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4620 {
4621 #ifdef CONFIG_NET_FLOW_LIMIT
4622 	struct sd_flow_limit *fl;
4623 	struct softnet_data *sd;
4624 	unsigned int old_flow, new_flow;
4625 
4626 	if (qlen < (READ_ONCE(netdev_max_backlog) >> 1))
4627 		return false;
4628 
4629 	sd = this_cpu_ptr(&softnet_data);
4630 
4631 	rcu_read_lock();
4632 	fl = rcu_dereference(sd->flow_limit);
4633 	if (fl) {
4634 		new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4635 		old_flow = fl->history[fl->history_head];
4636 		fl->history[fl->history_head] = new_flow;
4637 
4638 		fl->history_head++;
4639 		fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4640 
4641 		if (likely(fl->buckets[old_flow]))
4642 			fl->buckets[old_flow]--;
4643 
4644 		if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4645 			fl->count++;
4646 			rcu_read_unlock();
4647 			return true;
4648 		}
4649 	}
4650 	rcu_read_unlock();
4651 #endif
4652 	return false;
4653 }
4654 
4655 /*
4656  * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4657  * queue (may be a remote CPU queue).
4658  */
4659 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4660 			      unsigned int *qtail)
4661 {
4662 	enum skb_drop_reason reason;
4663 	struct softnet_data *sd;
4664 	unsigned long flags;
4665 	unsigned int qlen;
4666 
4667 	reason = SKB_DROP_REASON_NOT_SPECIFIED;
4668 	sd = &per_cpu(softnet_data, cpu);
4669 
4670 	rps_lock_irqsave(sd, &flags);
4671 	if (!netif_running(skb->dev))
4672 		goto drop;
4673 	qlen = skb_queue_len(&sd->input_pkt_queue);
4674 	if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) {
4675 		if (qlen) {
4676 enqueue:
4677 			__skb_queue_tail(&sd->input_pkt_queue, skb);
4678 			input_queue_tail_incr_save(sd, qtail);
4679 			rps_unlock_irq_restore(sd, &flags);
4680 			return NET_RX_SUCCESS;
4681 		}
4682 
4683 		/* Schedule NAPI for backlog device
4684 		 * We can use non atomic operation since we own the queue lock
4685 		 */
4686 		if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
4687 			napi_schedule_rps(sd);
4688 		goto enqueue;
4689 	}
4690 	reason = SKB_DROP_REASON_CPU_BACKLOG;
4691 
4692 drop:
4693 	sd->dropped++;
4694 	rps_unlock_irq_restore(sd, &flags);
4695 
4696 	dev_core_stats_rx_dropped_inc(skb->dev);
4697 	kfree_skb_reason(skb, reason);
4698 	return NET_RX_DROP;
4699 }
4700 
4701 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4702 {
4703 	struct net_device *dev = skb->dev;
4704 	struct netdev_rx_queue *rxqueue;
4705 
4706 	rxqueue = dev->_rx;
4707 
4708 	if (skb_rx_queue_recorded(skb)) {
4709 		u16 index = skb_get_rx_queue(skb);
4710 
4711 		if (unlikely(index >= dev->real_num_rx_queues)) {
4712 			WARN_ONCE(dev->real_num_rx_queues > 1,
4713 				  "%s received packet on queue %u, but number "
4714 				  "of RX queues is %u\n",
4715 				  dev->name, index, dev->real_num_rx_queues);
4716 
4717 			return rxqueue; /* Return first rxqueue */
4718 		}
4719 		rxqueue += index;
4720 	}
4721 	return rxqueue;
4722 }
4723 
4724 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4725 			     struct bpf_prog *xdp_prog)
4726 {
4727 	void *orig_data, *orig_data_end, *hard_start;
4728 	struct netdev_rx_queue *rxqueue;
4729 	bool orig_bcast, orig_host;
4730 	u32 mac_len, frame_sz;
4731 	__be16 orig_eth_type;
4732 	struct ethhdr *eth;
4733 	u32 metalen, act;
4734 	int off;
4735 
4736 	/* The XDP program wants to see the packet starting at the MAC
4737 	 * header.
4738 	 */
4739 	mac_len = skb->data - skb_mac_header(skb);
4740 	hard_start = skb->data - skb_headroom(skb);
4741 
4742 	/* SKB "head" area always have tailroom for skb_shared_info */
4743 	frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4744 	frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4745 
4746 	rxqueue = netif_get_rxqueue(skb);
4747 	xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4748 	xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4749 			 skb_headlen(skb) + mac_len, true);
4750 
4751 	orig_data_end = xdp->data_end;
4752 	orig_data = xdp->data;
4753 	eth = (struct ethhdr *)xdp->data;
4754 	orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4755 	orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4756 	orig_eth_type = eth->h_proto;
4757 
4758 	act = bpf_prog_run_xdp(xdp_prog, xdp);
4759 
4760 	/* check if bpf_xdp_adjust_head was used */
4761 	off = xdp->data - orig_data;
4762 	if (off) {
4763 		if (off > 0)
4764 			__skb_pull(skb, off);
4765 		else if (off < 0)
4766 			__skb_push(skb, -off);
4767 
4768 		skb->mac_header += off;
4769 		skb_reset_network_header(skb);
4770 	}
4771 
4772 	/* check if bpf_xdp_adjust_tail was used */
4773 	off = xdp->data_end - orig_data_end;
4774 	if (off != 0) {
4775 		skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4776 		skb->len += off; /* positive on grow, negative on shrink */
4777 	}
4778 
4779 	/* check if XDP changed eth hdr such SKB needs update */
4780 	eth = (struct ethhdr *)xdp->data;
4781 	if ((orig_eth_type != eth->h_proto) ||
4782 	    (orig_host != ether_addr_equal_64bits(eth->h_dest,
4783 						  skb->dev->dev_addr)) ||
4784 	    (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4785 		__skb_push(skb, ETH_HLEN);
4786 		skb->pkt_type = PACKET_HOST;
4787 		skb->protocol = eth_type_trans(skb, skb->dev);
4788 	}
4789 
4790 	/* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4791 	 * before calling us again on redirect path. We do not call do_redirect
4792 	 * as we leave that up to the caller.
4793 	 *
4794 	 * Caller is responsible for managing lifetime of skb (i.e. calling
4795 	 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4796 	 */
4797 	switch (act) {
4798 	case XDP_REDIRECT:
4799 	case XDP_TX:
4800 		__skb_push(skb, mac_len);
4801 		break;
4802 	case XDP_PASS:
4803 		metalen = xdp->data - xdp->data_meta;
4804 		if (metalen)
4805 			skb_metadata_set(skb, metalen);
4806 		break;
4807 	}
4808 
4809 	return act;
4810 }
4811 
4812 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4813 				     struct xdp_buff *xdp,
4814 				     struct bpf_prog *xdp_prog)
4815 {
4816 	u32 act = XDP_DROP;
4817 
4818 	/* Reinjected packets coming from act_mirred or similar should
4819 	 * not get XDP generic processing.
4820 	 */
4821 	if (skb_is_redirected(skb))
4822 		return XDP_PASS;
4823 
4824 	/* XDP packets must be linear and must have sufficient headroom
4825 	 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4826 	 * native XDP provides, thus we need to do it here as well.
4827 	 */
4828 	if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4829 	    skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4830 		int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4831 		int troom = skb->tail + skb->data_len - skb->end;
4832 
4833 		/* In case we have to go down the path and also linearize,
4834 		 * then lets do the pskb_expand_head() work just once here.
4835 		 */
4836 		if (pskb_expand_head(skb,
4837 				     hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4838 				     troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4839 			goto do_drop;
4840 		if (skb_linearize(skb))
4841 			goto do_drop;
4842 	}
4843 
4844 	act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4845 	switch (act) {
4846 	case XDP_REDIRECT:
4847 	case XDP_TX:
4848 	case XDP_PASS:
4849 		break;
4850 	default:
4851 		bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act);
4852 		fallthrough;
4853 	case XDP_ABORTED:
4854 		trace_xdp_exception(skb->dev, xdp_prog, act);
4855 		fallthrough;
4856 	case XDP_DROP:
4857 	do_drop:
4858 		kfree_skb(skb);
4859 		break;
4860 	}
4861 
4862 	return act;
4863 }
4864 
4865 /* When doing generic XDP we have to bypass the qdisc layer and the
4866  * network taps in order to match in-driver-XDP behavior. This also means
4867  * that XDP packets are able to starve other packets going through a qdisc,
4868  * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX
4869  * queues, so they do not have this starvation issue.
4870  */
4871 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4872 {
4873 	struct net_device *dev = skb->dev;
4874 	struct netdev_queue *txq;
4875 	bool free_skb = true;
4876 	int cpu, rc;
4877 
4878 	txq = netdev_core_pick_tx(dev, skb, NULL);
4879 	cpu = smp_processor_id();
4880 	HARD_TX_LOCK(dev, txq, cpu);
4881 	if (!netif_xmit_frozen_or_drv_stopped(txq)) {
4882 		rc = netdev_start_xmit(skb, dev, txq, 0);
4883 		if (dev_xmit_complete(rc))
4884 			free_skb = false;
4885 	}
4886 	HARD_TX_UNLOCK(dev, txq);
4887 	if (free_skb) {
4888 		trace_xdp_exception(dev, xdp_prog, XDP_TX);
4889 		dev_core_stats_tx_dropped_inc(dev);
4890 		kfree_skb(skb);
4891 	}
4892 }
4893 
4894 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4895 
4896 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4897 {
4898 	if (xdp_prog) {
4899 		struct xdp_buff xdp;
4900 		u32 act;
4901 		int err;
4902 
4903 		act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4904 		if (act != XDP_PASS) {
4905 			switch (act) {
4906 			case XDP_REDIRECT:
4907 				err = xdp_do_generic_redirect(skb->dev, skb,
4908 							      &xdp, xdp_prog);
4909 				if (err)
4910 					goto out_redir;
4911 				break;
4912 			case XDP_TX:
4913 				generic_xdp_tx(skb, xdp_prog);
4914 				break;
4915 			}
4916 			return XDP_DROP;
4917 		}
4918 	}
4919 	return XDP_PASS;
4920 out_redir:
4921 	kfree_skb_reason(skb, SKB_DROP_REASON_XDP);
4922 	return XDP_DROP;
4923 }
4924 EXPORT_SYMBOL_GPL(do_xdp_generic);
4925 
4926 static int netif_rx_internal(struct sk_buff *skb)
4927 {
4928 	int ret;
4929 
4930 	net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
4931 
4932 	trace_netif_rx(skb);
4933 
4934 #ifdef CONFIG_RPS
4935 	if (static_branch_unlikely(&rps_needed)) {
4936 		struct rps_dev_flow voidflow, *rflow = &voidflow;
4937 		int cpu;
4938 
4939 		rcu_read_lock();
4940 
4941 		cpu = get_rps_cpu(skb->dev, skb, &rflow);
4942 		if (cpu < 0)
4943 			cpu = smp_processor_id();
4944 
4945 		ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4946 
4947 		rcu_read_unlock();
4948 	} else
4949 #endif
4950 	{
4951 		unsigned int qtail;
4952 
4953 		ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
4954 	}
4955 	return ret;
4956 }
4957 
4958 /**
4959  *	__netif_rx	-	Slightly optimized version of netif_rx
4960  *	@skb: buffer to post
4961  *
4962  *	This behaves as netif_rx except that it does not disable bottom halves.
4963  *	As a result this function may only be invoked from the interrupt context
4964  *	(either hard or soft interrupt).
4965  */
4966 int __netif_rx(struct sk_buff *skb)
4967 {
4968 	int ret;
4969 
4970 	lockdep_assert_once(hardirq_count() | softirq_count());
4971 
4972 	trace_netif_rx_entry(skb);
4973 	ret = netif_rx_internal(skb);
4974 	trace_netif_rx_exit(ret);
4975 	return ret;
4976 }
4977 EXPORT_SYMBOL(__netif_rx);
4978 
4979 /**
4980  *	netif_rx	-	post buffer to the network code
4981  *	@skb: buffer to post
4982  *
4983  *	This function receives a packet from a device driver and queues it for
4984  *	the upper (protocol) levels to process via the backlog NAPI device. It
4985  *	always succeeds. The buffer may be dropped during processing for
4986  *	congestion control or by the protocol layers.
4987  *	The network buffer is passed via the backlog NAPI device. Modern NIC
4988  *	driver should use NAPI and GRO.
4989  *	This function can used from interrupt and from process context. The
4990  *	caller from process context must not disable interrupts before invoking
4991  *	this function.
4992  *
4993  *	return values:
4994  *	NET_RX_SUCCESS	(no congestion)
4995  *	NET_RX_DROP     (packet was dropped)
4996  *
4997  */
4998 int netif_rx(struct sk_buff *skb)
4999 {
5000 	bool need_bh_off = !(hardirq_count() | softirq_count());
5001 	int ret;
5002 
5003 	if (need_bh_off)
5004 		local_bh_disable();
5005 	trace_netif_rx_entry(skb);
5006 	ret = netif_rx_internal(skb);
5007 	trace_netif_rx_exit(ret);
5008 	if (need_bh_off)
5009 		local_bh_enable();
5010 	return ret;
5011 }
5012 EXPORT_SYMBOL(netif_rx);
5013 
5014 static __latent_entropy void net_tx_action(struct softirq_action *h)
5015 {
5016 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5017 
5018 	if (sd->completion_queue) {
5019 		struct sk_buff *clist;
5020 
5021 		local_irq_disable();
5022 		clist = sd->completion_queue;
5023 		sd->completion_queue = NULL;
5024 		local_irq_enable();
5025 
5026 		while (clist) {
5027 			struct sk_buff *skb = clist;
5028 
5029 			clist = clist->next;
5030 
5031 			WARN_ON(refcount_read(&skb->users));
5032 			if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
5033 				trace_consume_skb(skb);
5034 			else
5035 				trace_kfree_skb(skb, net_tx_action,
5036 						SKB_DROP_REASON_NOT_SPECIFIED);
5037 
5038 			if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5039 				__kfree_skb(skb);
5040 			else
5041 				__kfree_skb_defer(skb);
5042 		}
5043 	}
5044 
5045 	if (sd->output_queue) {
5046 		struct Qdisc *head;
5047 
5048 		local_irq_disable();
5049 		head = sd->output_queue;
5050 		sd->output_queue = NULL;
5051 		sd->output_queue_tailp = &sd->output_queue;
5052 		local_irq_enable();
5053 
5054 		rcu_read_lock();
5055 
5056 		while (head) {
5057 			struct Qdisc *q = head;
5058 			spinlock_t *root_lock = NULL;
5059 
5060 			head = head->next_sched;
5061 
5062 			/* We need to make sure head->next_sched is read
5063 			 * before clearing __QDISC_STATE_SCHED
5064 			 */
5065 			smp_mb__before_atomic();
5066 
5067 			if (!(q->flags & TCQ_F_NOLOCK)) {
5068 				root_lock = qdisc_lock(q);
5069 				spin_lock(root_lock);
5070 			} else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5071 						     &q->state))) {
5072 				/* There is a synchronize_net() between
5073 				 * STATE_DEACTIVATED flag being set and
5074 				 * qdisc_reset()/some_qdisc_is_busy() in
5075 				 * dev_deactivate(), so we can safely bail out
5076 				 * early here to avoid data race between
5077 				 * qdisc_deactivate() and some_qdisc_is_busy()
5078 				 * for lockless qdisc.
5079 				 */
5080 				clear_bit(__QDISC_STATE_SCHED, &q->state);
5081 				continue;
5082 			}
5083 
5084 			clear_bit(__QDISC_STATE_SCHED, &q->state);
5085 			qdisc_run(q);
5086 			if (root_lock)
5087 				spin_unlock(root_lock);
5088 		}
5089 
5090 		rcu_read_unlock();
5091 	}
5092 
5093 	xfrm_dev_backlog(sd);
5094 }
5095 
5096 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5097 /* This hook is defined here for ATM LANE */
5098 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5099 			     unsigned char *addr) __read_mostly;
5100 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5101 #endif
5102 
5103 static inline struct sk_buff *
5104 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5105 		   struct net_device *orig_dev, bool *another)
5106 {
5107 #ifdef CONFIG_NET_CLS_ACT
5108 	struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5109 	struct tcf_result cl_res;
5110 
5111 	/* If there's at least one ingress present somewhere (so
5112 	 * we get here via enabled static key), remaining devices
5113 	 * that are not configured with an ingress qdisc will bail
5114 	 * out here.
5115 	 */
5116 	if (!miniq)
5117 		return skb;
5118 
5119 	if (*pt_prev) {
5120 		*ret = deliver_skb(skb, *pt_prev, orig_dev);
5121 		*pt_prev = NULL;
5122 	}
5123 
5124 	qdisc_skb_cb(skb)->pkt_len = skb->len;
5125 	tc_skb_cb(skb)->mru = 0;
5126 	tc_skb_cb(skb)->post_ct = false;
5127 	skb->tc_at_ingress = 1;
5128 	mini_qdisc_bstats_cpu_update(miniq, skb);
5129 
5130 	switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
5131 	case TC_ACT_OK:
5132 	case TC_ACT_RECLASSIFY:
5133 		skb->tc_index = TC_H_MIN(cl_res.classid);
5134 		break;
5135 	case TC_ACT_SHOT:
5136 		mini_qdisc_qstats_cpu_drop(miniq);
5137 		kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS);
5138 		*ret = NET_RX_DROP;
5139 		return NULL;
5140 	case TC_ACT_STOLEN:
5141 	case TC_ACT_QUEUED:
5142 	case TC_ACT_TRAP:
5143 		consume_skb(skb);
5144 		*ret = NET_RX_SUCCESS;
5145 		return NULL;
5146 	case TC_ACT_REDIRECT:
5147 		/* skb_mac_header check was done by cls/act_bpf, so
5148 		 * we can safely push the L2 header back before
5149 		 * redirecting to another netdev
5150 		 */
5151 		__skb_push(skb, skb->mac_len);
5152 		if (skb_do_redirect(skb) == -EAGAIN) {
5153 			__skb_pull(skb, skb->mac_len);
5154 			*another = true;
5155 			break;
5156 		}
5157 		*ret = NET_RX_SUCCESS;
5158 		return NULL;
5159 	case TC_ACT_CONSUMED:
5160 		*ret = NET_RX_SUCCESS;
5161 		return NULL;
5162 	default:
5163 		break;
5164 	}
5165 #endif /* CONFIG_NET_CLS_ACT */
5166 	return skb;
5167 }
5168 
5169 /**
5170  *	netdev_is_rx_handler_busy - check if receive handler is registered
5171  *	@dev: device to check
5172  *
5173  *	Check if a receive handler is already registered for a given device.
5174  *	Return true if there one.
5175  *
5176  *	The caller must hold the rtnl_mutex.
5177  */
5178 bool netdev_is_rx_handler_busy(struct net_device *dev)
5179 {
5180 	ASSERT_RTNL();
5181 	return dev && rtnl_dereference(dev->rx_handler);
5182 }
5183 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5184 
5185 /**
5186  *	netdev_rx_handler_register - register receive handler
5187  *	@dev: device to register a handler for
5188  *	@rx_handler: receive handler to register
5189  *	@rx_handler_data: data pointer that is used by rx handler
5190  *
5191  *	Register a receive handler for a device. This handler will then be
5192  *	called from __netif_receive_skb. A negative errno code is returned
5193  *	on a failure.
5194  *
5195  *	The caller must hold the rtnl_mutex.
5196  *
5197  *	For a general description of rx_handler, see enum rx_handler_result.
5198  */
5199 int netdev_rx_handler_register(struct net_device *dev,
5200 			       rx_handler_func_t *rx_handler,
5201 			       void *rx_handler_data)
5202 {
5203 	if (netdev_is_rx_handler_busy(dev))
5204 		return -EBUSY;
5205 
5206 	if (dev->priv_flags & IFF_NO_RX_HANDLER)
5207 		return -EINVAL;
5208 
5209 	/* Note: rx_handler_data must be set before rx_handler */
5210 	rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5211 	rcu_assign_pointer(dev->rx_handler, rx_handler);
5212 
5213 	return 0;
5214 }
5215 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5216 
5217 /**
5218  *	netdev_rx_handler_unregister - unregister receive handler
5219  *	@dev: device to unregister a handler from
5220  *
5221  *	Unregister a receive handler from a device.
5222  *
5223  *	The caller must hold the rtnl_mutex.
5224  */
5225 void netdev_rx_handler_unregister(struct net_device *dev)
5226 {
5227 
5228 	ASSERT_RTNL();
5229 	RCU_INIT_POINTER(dev->rx_handler, NULL);
5230 	/* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5231 	 * section has a guarantee to see a non NULL rx_handler_data
5232 	 * as well.
5233 	 */
5234 	synchronize_net();
5235 	RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5236 }
5237 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5238 
5239 /*
5240  * Limit the use of PFMEMALLOC reserves to those protocols that implement
5241  * the special handling of PFMEMALLOC skbs.
5242  */
5243 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5244 {
5245 	switch (skb->protocol) {
5246 	case htons(ETH_P_ARP):
5247 	case htons(ETH_P_IP):
5248 	case htons(ETH_P_IPV6):
5249 	case htons(ETH_P_8021Q):
5250 	case htons(ETH_P_8021AD):
5251 		return true;
5252 	default:
5253 		return false;
5254 	}
5255 }
5256 
5257 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5258 			     int *ret, struct net_device *orig_dev)
5259 {
5260 	if (nf_hook_ingress_active(skb)) {
5261 		int ingress_retval;
5262 
5263 		if (*pt_prev) {
5264 			*ret = deliver_skb(skb, *pt_prev, orig_dev);
5265 			*pt_prev = NULL;
5266 		}
5267 
5268 		rcu_read_lock();
5269 		ingress_retval = nf_hook_ingress(skb);
5270 		rcu_read_unlock();
5271 		return ingress_retval;
5272 	}
5273 	return 0;
5274 }
5275 
5276 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5277 				    struct packet_type **ppt_prev)
5278 {
5279 	struct packet_type *ptype, *pt_prev;
5280 	rx_handler_func_t *rx_handler;
5281 	struct sk_buff *skb = *pskb;
5282 	struct net_device *orig_dev;
5283 	bool deliver_exact = false;
5284 	int ret = NET_RX_DROP;
5285 	__be16 type;
5286 
5287 	net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb);
5288 
5289 	trace_netif_receive_skb(skb);
5290 
5291 	orig_dev = skb->dev;
5292 
5293 	skb_reset_network_header(skb);
5294 	if (!skb_transport_header_was_set(skb))
5295 		skb_reset_transport_header(skb);
5296 	skb_reset_mac_len(skb);
5297 
5298 	pt_prev = NULL;
5299 
5300 another_round:
5301 	skb->skb_iif = skb->dev->ifindex;
5302 
5303 	__this_cpu_inc(softnet_data.processed);
5304 
5305 	if (static_branch_unlikely(&generic_xdp_needed_key)) {
5306 		int ret2;
5307 
5308 		migrate_disable();
5309 		ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5310 		migrate_enable();
5311 
5312 		if (ret2 != XDP_PASS) {
5313 			ret = NET_RX_DROP;
5314 			goto out;
5315 		}
5316 	}
5317 
5318 	if (eth_type_vlan(skb->protocol)) {
5319 		skb = skb_vlan_untag(skb);
5320 		if (unlikely(!skb))
5321 			goto out;
5322 	}
5323 
5324 	if (skb_skip_tc_classify(skb))
5325 		goto skip_classify;
5326 
5327 	if (pfmemalloc)
5328 		goto skip_taps;
5329 
5330 	list_for_each_entry_rcu(ptype, &ptype_all, list) {
5331 		if (pt_prev)
5332 			ret = deliver_skb(skb, pt_prev, orig_dev);
5333 		pt_prev = ptype;
5334 	}
5335 
5336 	list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5337 		if (pt_prev)
5338 			ret = deliver_skb(skb, pt_prev, orig_dev);
5339 		pt_prev = ptype;
5340 	}
5341 
5342 skip_taps:
5343 #ifdef CONFIG_NET_INGRESS
5344 	if (static_branch_unlikely(&ingress_needed_key)) {
5345 		bool another = false;
5346 
5347 		nf_skip_egress(skb, true);
5348 		skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5349 					 &another);
5350 		if (another)
5351 			goto another_round;
5352 		if (!skb)
5353 			goto out;
5354 
5355 		nf_skip_egress(skb, false);
5356 		if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5357 			goto out;
5358 	}
5359 #endif
5360 	skb_reset_redirect(skb);
5361 skip_classify:
5362 	if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5363 		goto drop;
5364 
5365 	if (skb_vlan_tag_present(skb)) {
5366 		if (pt_prev) {
5367 			ret = deliver_skb(skb, pt_prev, orig_dev);
5368 			pt_prev = NULL;
5369 		}
5370 		if (vlan_do_receive(&skb))
5371 			goto another_round;
5372 		else if (unlikely(!skb))
5373 			goto out;
5374 	}
5375 
5376 	rx_handler = rcu_dereference(skb->dev->rx_handler);
5377 	if (rx_handler) {
5378 		if (pt_prev) {
5379 			ret = deliver_skb(skb, pt_prev, orig_dev);
5380 			pt_prev = NULL;
5381 		}
5382 		switch (rx_handler(&skb)) {
5383 		case RX_HANDLER_CONSUMED:
5384 			ret = NET_RX_SUCCESS;
5385 			goto out;
5386 		case RX_HANDLER_ANOTHER:
5387 			goto another_round;
5388 		case RX_HANDLER_EXACT:
5389 			deliver_exact = true;
5390 			break;
5391 		case RX_HANDLER_PASS:
5392 			break;
5393 		default:
5394 			BUG();
5395 		}
5396 	}
5397 
5398 	if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5399 check_vlan_id:
5400 		if (skb_vlan_tag_get_id(skb)) {
5401 			/* Vlan id is non 0 and vlan_do_receive() above couldn't
5402 			 * find vlan device.
5403 			 */
5404 			skb->pkt_type = PACKET_OTHERHOST;
5405 		} else if (eth_type_vlan(skb->protocol)) {
5406 			/* Outer header is 802.1P with vlan 0, inner header is
5407 			 * 802.1Q or 802.1AD and vlan_do_receive() above could
5408 			 * not find vlan dev for vlan id 0.
5409 			 */
5410 			__vlan_hwaccel_clear_tag(skb);
5411 			skb = skb_vlan_untag(skb);
5412 			if (unlikely(!skb))
5413 				goto out;
5414 			if (vlan_do_receive(&skb))
5415 				/* After stripping off 802.1P header with vlan 0
5416 				 * vlan dev is found for inner header.
5417 				 */
5418 				goto another_round;
5419 			else if (unlikely(!skb))
5420 				goto out;
5421 			else
5422 				/* We have stripped outer 802.1P vlan 0 header.
5423 				 * But could not find vlan dev.
5424 				 * check again for vlan id to set OTHERHOST.
5425 				 */
5426 				goto check_vlan_id;
5427 		}
5428 		/* Note: we might in the future use prio bits
5429 		 * and set skb->priority like in vlan_do_receive()
5430 		 * For the time being, just ignore Priority Code Point
5431 		 */
5432 		__vlan_hwaccel_clear_tag(skb);
5433 	}
5434 
5435 	type = skb->protocol;
5436 
5437 	/* deliver only exact match when indicated */
5438 	if (likely(!deliver_exact)) {
5439 		deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5440 				       &ptype_base[ntohs(type) &
5441 						   PTYPE_HASH_MASK]);
5442 	}
5443 
5444 	deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5445 			       &orig_dev->ptype_specific);
5446 
5447 	if (unlikely(skb->dev != orig_dev)) {
5448 		deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5449 				       &skb->dev->ptype_specific);
5450 	}
5451 
5452 	if (pt_prev) {
5453 		if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5454 			goto drop;
5455 		*ppt_prev = pt_prev;
5456 	} else {
5457 drop:
5458 		if (!deliver_exact)
5459 			dev_core_stats_rx_dropped_inc(skb->dev);
5460 		else
5461 			dev_core_stats_rx_nohandler_inc(skb->dev);
5462 		kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO);
5463 		/* Jamal, now you will not able to escape explaining
5464 		 * me how you were going to use this. :-)
5465 		 */
5466 		ret = NET_RX_DROP;
5467 	}
5468 
5469 out:
5470 	/* The invariant here is that if *ppt_prev is not NULL
5471 	 * then skb should also be non-NULL.
5472 	 *
5473 	 * Apparently *ppt_prev assignment above holds this invariant due to
5474 	 * skb dereferencing near it.
5475 	 */
5476 	*pskb = skb;
5477 	return ret;
5478 }
5479 
5480 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5481 {
5482 	struct net_device *orig_dev = skb->dev;
5483 	struct packet_type *pt_prev = NULL;
5484 	int ret;
5485 
5486 	ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5487 	if (pt_prev)
5488 		ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5489 					 skb->dev, pt_prev, orig_dev);
5490 	return ret;
5491 }
5492 
5493 /**
5494  *	netif_receive_skb_core - special purpose version of netif_receive_skb
5495  *	@skb: buffer to process
5496  *
5497  *	More direct receive version of netif_receive_skb().  It should
5498  *	only be used by callers that have a need to skip RPS and Generic XDP.
5499  *	Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5500  *
5501  *	This function may only be called from softirq context and interrupts
5502  *	should be enabled.
5503  *
5504  *	Return values (usually ignored):
5505  *	NET_RX_SUCCESS: no congestion
5506  *	NET_RX_DROP: packet was dropped
5507  */
5508 int netif_receive_skb_core(struct sk_buff *skb)
5509 {
5510 	int ret;
5511 
5512 	rcu_read_lock();
5513 	ret = __netif_receive_skb_one_core(skb, false);
5514 	rcu_read_unlock();
5515 
5516 	return ret;
5517 }
5518 EXPORT_SYMBOL(netif_receive_skb_core);
5519 
5520 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5521 						  struct packet_type *pt_prev,
5522 						  struct net_device *orig_dev)
5523 {
5524 	struct sk_buff *skb, *next;
5525 
5526 	if (!pt_prev)
5527 		return;
5528 	if (list_empty(head))
5529 		return;
5530 	if (pt_prev->list_func != NULL)
5531 		INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5532 				   ip_list_rcv, head, pt_prev, orig_dev);
5533 	else
5534 		list_for_each_entry_safe(skb, next, head, list) {
5535 			skb_list_del_init(skb);
5536 			pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5537 		}
5538 }
5539 
5540 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5541 {
5542 	/* Fast-path assumptions:
5543 	 * - There is no RX handler.
5544 	 * - Only one packet_type matches.
5545 	 * If either of these fails, we will end up doing some per-packet
5546 	 * processing in-line, then handling the 'last ptype' for the whole
5547 	 * sublist.  This can't cause out-of-order delivery to any single ptype,
5548 	 * because the 'last ptype' must be constant across the sublist, and all
5549 	 * other ptypes are handled per-packet.
5550 	 */
5551 	/* Current (common) ptype of sublist */
5552 	struct packet_type *pt_curr = NULL;
5553 	/* Current (common) orig_dev of sublist */
5554 	struct net_device *od_curr = NULL;
5555 	struct list_head sublist;
5556 	struct sk_buff *skb, *next;
5557 
5558 	INIT_LIST_HEAD(&sublist);
5559 	list_for_each_entry_safe(skb, next, head, list) {
5560 		struct net_device *orig_dev = skb->dev;
5561 		struct packet_type *pt_prev = NULL;
5562 
5563 		skb_list_del_init(skb);
5564 		__netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5565 		if (!pt_prev)
5566 			continue;
5567 		if (pt_curr != pt_prev || od_curr != orig_dev) {
5568 			/* dispatch old sublist */
5569 			__netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5570 			/* start new sublist */
5571 			INIT_LIST_HEAD(&sublist);
5572 			pt_curr = pt_prev;
5573 			od_curr = orig_dev;
5574 		}
5575 		list_add_tail(&skb->list, &sublist);
5576 	}
5577 
5578 	/* dispatch final sublist */
5579 	__netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5580 }
5581 
5582 static int __netif_receive_skb(struct sk_buff *skb)
5583 {
5584 	int ret;
5585 
5586 	if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5587 		unsigned int noreclaim_flag;
5588 
5589 		/*
5590 		 * PFMEMALLOC skbs are special, they should
5591 		 * - be delivered to SOCK_MEMALLOC sockets only
5592 		 * - stay away from userspace
5593 		 * - have bounded memory usage
5594 		 *
5595 		 * Use PF_MEMALLOC as this saves us from propagating the allocation
5596 		 * context down to all allocation sites.
5597 		 */
5598 		noreclaim_flag = memalloc_noreclaim_save();
5599 		ret = __netif_receive_skb_one_core(skb, true);
5600 		memalloc_noreclaim_restore(noreclaim_flag);
5601 	} else
5602 		ret = __netif_receive_skb_one_core(skb, false);
5603 
5604 	return ret;
5605 }
5606 
5607 static void __netif_receive_skb_list(struct list_head *head)
5608 {
5609 	unsigned long noreclaim_flag = 0;
5610 	struct sk_buff *skb, *next;
5611 	bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5612 
5613 	list_for_each_entry_safe(skb, next, head, list) {
5614 		if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5615 			struct list_head sublist;
5616 
5617 			/* Handle the previous sublist */
5618 			list_cut_before(&sublist, head, &skb->list);
5619 			if (!list_empty(&sublist))
5620 				__netif_receive_skb_list_core(&sublist, pfmemalloc);
5621 			pfmemalloc = !pfmemalloc;
5622 			/* See comments in __netif_receive_skb */
5623 			if (pfmemalloc)
5624 				noreclaim_flag = memalloc_noreclaim_save();
5625 			else
5626 				memalloc_noreclaim_restore(noreclaim_flag);
5627 		}
5628 	}
5629 	/* Handle the remaining sublist */
5630 	if (!list_empty(head))
5631 		__netif_receive_skb_list_core(head, pfmemalloc);
5632 	/* Restore pflags */
5633 	if (pfmemalloc)
5634 		memalloc_noreclaim_restore(noreclaim_flag);
5635 }
5636 
5637 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5638 {
5639 	struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5640 	struct bpf_prog *new = xdp->prog;
5641 	int ret = 0;
5642 
5643 	switch (xdp->command) {
5644 	case XDP_SETUP_PROG:
5645 		rcu_assign_pointer(dev->xdp_prog, new);
5646 		if (old)
5647 			bpf_prog_put(old);
5648 
5649 		if (old && !new) {
5650 			static_branch_dec(&generic_xdp_needed_key);
5651 		} else if (new && !old) {
5652 			static_branch_inc(&generic_xdp_needed_key);
5653 			dev_disable_lro(dev);
5654 			dev_disable_gro_hw(dev);
5655 		}
5656 		break;
5657 
5658 	default:
5659 		ret = -EINVAL;
5660 		break;
5661 	}
5662 
5663 	return ret;
5664 }
5665 
5666 static int netif_receive_skb_internal(struct sk_buff *skb)
5667 {
5668 	int ret;
5669 
5670 	net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5671 
5672 	if (skb_defer_rx_timestamp(skb))
5673 		return NET_RX_SUCCESS;
5674 
5675 	rcu_read_lock();
5676 #ifdef CONFIG_RPS
5677 	if (static_branch_unlikely(&rps_needed)) {
5678 		struct rps_dev_flow voidflow, *rflow = &voidflow;
5679 		int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5680 
5681 		if (cpu >= 0) {
5682 			ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5683 			rcu_read_unlock();
5684 			return ret;
5685 		}
5686 	}
5687 #endif
5688 	ret = __netif_receive_skb(skb);
5689 	rcu_read_unlock();
5690 	return ret;
5691 }
5692 
5693 void netif_receive_skb_list_internal(struct list_head *head)
5694 {
5695 	struct sk_buff *skb, *next;
5696 	struct list_head sublist;
5697 
5698 	INIT_LIST_HEAD(&sublist);
5699 	list_for_each_entry_safe(skb, next, head, list) {
5700 		net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5701 		skb_list_del_init(skb);
5702 		if (!skb_defer_rx_timestamp(skb))
5703 			list_add_tail(&skb->list, &sublist);
5704 	}
5705 	list_splice_init(&sublist, head);
5706 
5707 	rcu_read_lock();
5708 #ifdef CONFIG_RPS
5709 	if (static_branch_unlikely(&rps_needed)) {
5710 		list_for_each_entry_safe(skb, next, head, list) {
5711 			struct rps_dev_flow voidflow, *rflow = &voidflow;
5712 			int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5713 
5714 			if (cpu >= 0) {
5715 				/* Will be handled, remove from list */
5716 				skb_list_del_init(skb);
5717 				enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5718 			}
5719 		}
5720 	}
5721 #endif
5722 	__netif_receive_skb_list(head);
5723 	rcu_read_unlock();
5724 }
5725 
5726 /**
5727  *	netif_receive_skb - process receive buffer from network
5728  *	@skb: buffer to process
5729  *
5730  *	netif_receive_skb() is the main receive data processing function.
5731  *	It always succeeds. The buffer may be dropped during processing
5732  *	for congestion control or by the protocol layers.
5733  *
5734  *	This function may only be called from softirq context and interrupts
5735  *	should be enabled.
5736  *
5737  *	Return values (usually ignored):
5738  *	NET_RX_SUCCESS: no congestion
5739  *	NET_RX_DROP: packet was dropped
5740  */
5741 int netif_receive_skb(struct sk_buff *skb)
5742 {
5743 	int ret;
5744 
5745 	trace_netif_receive_skb_entry(skb);
5746 
5747 	ret = netif_receive_skb_internal(skb);
5748 	trace_netif_receive_skb_exit(ret);
5749 
5750 	return ret;
5751 }
5752 EXPORT_SYMBOL(netif_receive_skb);
5753 
5754 /**
5755  *	netif_receive_skb_list - process many receive buffers from network
5756  *	@head: list of skbs to process.
5757  *
5758  *	Since return value of netif_receive_skb() is normally ignored, and
5759  *	wouldn't be meaningful for a list, this function returns void.
5760  *
5761  *	This function may only be called from softirq context and interrupts
5762  *	should be enabled.
5763  */
5764 void netif_receive_skb_list(struct list_head *head)
5765 {
5766 	struct sk_buff *skb;
5767 
5768 	if (list_empty(head))
5769 		return;
5770 	if (trace_netif_receive_skb_list_entry_enabled()) {
5771 		list_for_each_entry(skb, head, list)
5772 			trace_netif_receive_skb_list_entry(skb);
5773 	}
5774 	netif_receive_skb_list_internal(head);
5775 	trace_netif_receive_skb_list_exit(0);
5776 }
5777 EXPORT_SYMBOL(netif_receive_skb_list);
5778 
5779 static DEFINE_PER_CPU(struct work_struct, flush_works);
5780 
5781 /* Network device is going away, flush any packets still pending */
5782 static void flush_backlog(struct work_struct *work)
5783 {
5784 	struct sk_buff *skb, *tmp;
5785 	struct softnet_data *sd;
5786 
5787 	local_bh_disable();
5788 	sd = this_cpu_ptr(&softnet_data);
5789 
5790 	rps_lock_irq_disable(sd);
5791 	skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5792 		if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5793 			__skb_unlink(skb, &sd->input_pkt_queue);
5794 			dev_kfree_skb_irq(skb);
5795 			input_queue_head_incr(sd);
5796 		}
5797 	}
5798 	rps_unlock_irq_enable(sd);
5799 
5800 	skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5801 		if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5802 			__skb_unlink(skb, &sd->process_queue);
5803 			kfree_skb(skb);
5804 			input_queue_head_incr(sd);
5805 		}
5806 	}
5807 	local_bh_enable();
5808 }
5809 
5810 static bool flush_required(int cpu)
5811 {
5812 #if IS_ENABLED(CONFIG_RPS)
5813 	struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5814 	bool do_flush;
5815 
5816 	rps_lock_irq_disable(sd);
5817 
5818 	/* as insertion into process_queue happens with the rps lock held,
5819 	 * process_queue access may race only with dequeue
5820 	 */
5821 	do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5822 		   !skb_queue_empty_lockless(&sd->process_queue);
5823 	rps_unlock_irq_enable(sd);
5824 
5825 	return do_flush;
5826 #endif
5827 	/* without RPS we can't safely check input_pkt_queue: during a
5828 	 * concurrent remote skb_queue_splice() we can detect as empty both
5829 	 * input_pkt_queue and process_queue even if the latter could end-up
5830 	 * containing a lot of packets.
5831 	 */
5832 	return true;
5833 }
5834 
5835 static void flush_all_backlogs(void)
5836 {
5837 	static cpumask_t flush_cpus;
5838 	unsigned int cpu;
5839 
5840 	/* since we are under rtnl lock protection we can use static data
5841 	 * for the cpumask and avoid allocating on stack the possibly
5842 	 * large mask
5843 	 */
5844 	ASSERT_RTNL();
5845 
5846 	cpus_read_lock();
5847 
5848 	cpumask_clear(&flush_cpus);
5849 	for_each_online_cpu(cpu) {
5850 		if (flush_required(cpu)) {
5851 			queue_work_on(cpu, system_highpri_wq,
5852 				      per_cpu_ptr(&flush_works, cpu));
5853 			cpumask_set_cpu(cpu, &flush_cpus);
5854 		}
5855 	}
5856 
5857 	/* we can have in flight packet[s] on the cpus we are not flushing,
5858 	 * synchronize_net() in unregister_netdevice_many() will take care of
5859 	 * them
5860 	 */
5861 	for_each_cpu(cpu, &flush_cpus)
5862 		flush_work(per_cpu_ptr(&flush_works, cpu));
5863 
5864 	cpus_read_unlock();
5865 }
5866 
5867 static void net_rps_send_ipi(struct softnet_data *remsd)
5868 {
5869 #ifdef CONFIG_RPS
5870 	while (remsd) {
5871 		struct softnet_data *next = remsd->rps_ipi_next;
5872 
5873 		if (cpu_online(remsd->cpu))
5874 			smp_call_function_single_async(remsd->cpu, &remsd->csd);
5875 		remsd = next;
5876 	}
5877 #endif
5878 }
5879 
5880 /*
5881  * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5882  * Note: called with local irq disabled, but exits with local irq enabled.
5883  */
5884 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5885 {
5886 #ifdef CONFIG_RPS
5887 	struct softnet_data *remsd = sd->rps_ipi_list;
5888 
5889 	if (remsd) {
5890 		sd->rps_ipi_list = NULL;
5891 
5892 		local_irq_enable();
5893 
5894 		/* Send pending IPI's to kick RPS processing on remote cpus. */
5895 		net_rps_send_ipi(remsd);
5896 	} else
5897 #endif
5898 		local_irq_enable();
5899 }
5900 
5901 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5902 {
5903 #ifdef CONFIG_RPS
5904 	return sd->rps_ipi_list != NULL;
5905 #else
5906 	return false;
5907 #endif
5908 }
5909 
5910 static int process_backlog(struct napi_struct *napi, int quota)
5911 {
5912 	struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5913 	bool again = true;
5914 	int work = 0;
5915 
5916 	/* Check if we have pending ipi, its better to send them now,
5917 	 * not waiting net_rx_action() end.
5918 	 */
5919 	if (sd_has_rps_ipi_waiting(sd)) {
5920 		local_irq_disable();
5921 		net_rps_action_and_irq_enable(sd);
5922 	}
5923 
5924 	napi->weight = READ_ONCE(dev_rx_weight);
5925 	while (again) {
5926 		struct sk_buff *skb;
5927 
5928 		while ((skb = __skb_dequeue(&sd->process_queue))) {
5929 			rcu_read_lock();
5930 			__netif_receive_skb(skb);
5931 			rcu_read_unlock();
5932 			input_queue_head_incr(sd);
5933 			if (++work >= quota)
5934 				return work;
5935 
5936 		}
5937 
5938 		rps_lock_irq_disable(sd);
5939 		if (skb_queue_empty(&sd->input_pkt_queue)) {
5940 			/*
5941 			 * Inline a custom version of __napi_complete().
5942 			 * only current cpu owns and manipulates this napi,
5943 			 * and NAPI_STATE_SCHED is the only possible flag set
5944 			 * on backlog.
5945 			 * We can use a plain write instead of clear_bit(),
5946 			 * and we dont need an smp_mb() memory barrier.
5947 			 */
5948 			napi->state = 0;
5949 			again = false;
5950 		} else {
5951 			skb_queue_splice_tail_init(&sd->input_pkt_queue,
5952 						   &sd->process_queue);
5953 		}
5954 		rps_unlock_irq_enable(sd);
5955 	}
5956 
5957 	return work;
5958 }
5959 
5960 /**
5961  * __napi_schedule - schedule for receive
5962  * @n: entry to schedule
5963  *
5964  * The entry's receive function will be scheduled to run.
5965  * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5966  */
5967 void __napi_schedule(struct napi_struct *n)
5968 {
5969 	unsigned long flags;
5970 
5971 	local_irq_save(flags);
5972 	____napi_schedule(this_cpu_ptr(&softnet_data), n);
5973 	local_irq_restore(flags);
5974 }
5975 EXPORT_SYMBOL(__napi_schedule);
5976 
5977 /**
5978  *	napi_schedule_prep - check if napi can be scheduled
5979  *	@n: napi context
5980  *
5981  * Test if NAPI routine is already running, and if not mark
5982  * it as running.  This is used as a condition variable to
5983  * insure only one NAPI poll instance runs.  We also make
5984  * sure there is no pending NAPI disable.
5985  */
5986 bool napi_schedule_prep(struct napi_struct *n)
5987 {
5988 	unsigned long val, new;
5989 
5990 	do {
5991 		val = READ_ONCE(n->state);
5992 		if (unlikely(val & NAPIF_STATE_DISABLE))
5993 			return false;
5994 		new = val | NAPIF_STATE_SCHED;
5995 
5996 		/* Sets STATE_MISSED bit if STATE_SCHED was already set
5997 		 * This was suggested by Alexander Duyck, as compiler
5998 		 * emits better code than :
5999 		 * if (val & NAPIF_STATE_SCHED)
6000 		 *     new |= NAPIF_STATE_MISSED;
6001 		 */
6002 		new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6003 						   NAPIF_STATE_MISSED;
6004 	} while (cmpxchg(&n->state, val, new) != val);
6005 
6006 	return !(val & NAPIF_STATE_SCHED);
6007 }
6008 EXPORT_SYMBOL(napi_schedule_prep);
6009 
6010 /**
6011  * __napi_schedule_irqoff - schedule for receive
6012  * @n: entry to schedule
6013  *
6014  * Variant of __napi_schedule() assuming hard irqs are masked.
6015  *
6016  * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6017  * because the interrupt disabled assumption might not be true
6018  * due to force-threaded interrupts and spinlock substitution.
6019  */
6020 void __napi_schedule_irqoff(struct napi_struct *n)
6021 {
6022 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6023 		____napi_schedule(this_cpu_ptr(&softnet_data), n);
6024 	else
6025 		__napi_schedule(n);
6026 }
6027 EXPORT_SYMBOL(__napi_schedule_irqoff);
6028 
6029 bool napi_complete_done(struct napi_struct *n, int work_done)
6030 {
6031 	unsigned long flags, val, new, timeout = 0;
6032 	bool ret = true;
6033 
6034 	/*
6035 	 * 1) Don't let napi dequeue from the cpu poll list
6036 	 *    just in case its running on a different cpu.
6037 	 * 2) If we are busy polling, do nothing here, we have
6038 	 *    the guarantee we will be called later.
6039 	 */
6040 	if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6041 				 NAPIF_STATE_IN_BUSY_POLL)))
6042 		return false;
6043 
6044 	if (work_done) {
6045 		if (n->gro_bitmask)
6046 			timeout = READ_ONCE(n->dev->gro_flush_timeout);
6047 		n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6048 	}
6049 	if (n->defer_hard_irqs_count > 0) {
6050 		n->defer_hard_irqs_count--;
6051 		timeout = READ_ONCE(n->dev->gro_flush_timeout);
6052 		if (timeout)
6053 			ret = false;
6054 	}
6055 	if (n->gro_bitmask) {
6056 		/* When the NAPI instance uses a timeout and keeps postponing
6057 		 * it, we need to bound somehow the time packets are kept in
6058 		 * the GRO layer
6059 		 */
6060 		napi_gro_flush(n, !!timeout);
6061 	}
6062 
6063 	gro_normal_list(n);
6064 
6065 	if (unlikely(!list_empty(&n->poll_list))) {
6066 		/* If n->poll_list is not empty, we need to mask irqs */
6067 		local_irq_save(flags);
6068 		list_del_init(&n->poll_list);
6069 		local_irq_restore(flags);
6070 	}
6071 
6072 	do {
6073 		val = READ_ONCE(n->state);
6074 
6075 		WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6076 
6077 		new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6078 			      NAPIF_STATE_SCHED_THREADED |
6079 			      NAPIF_STATE_PREFER_BUSY_POLL);
6080 
6081 		/* If STATE_MISSED was set, leave STATE_SCHED set,
6082 		 * because we will call napi->poll() one more time.
6083 		 * This C code was suggested by Alexander Duyck to help gcc.
6084 		 */
6085 		new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6086 						    NAPIF_STATE_SCHED;
6087 	} while (cmpxchg(&n->state, val, new) != val);
6088 
6089 	if (unlikely(val & NAPIF_STATE_MISSED)) {
6090 		__napi_schedule(n);
6091 		return false;
6092 	}
6093 
6094 	if (timeout)
6095 		hrtimer_start(&n->timer, ns_to_ktime(timeout),
6096 			      HRTIMER_MODE_REL_PINNED);
6097 	return ret;
6098 }
6099 EXPORT_SYMBOL(napi_complete_done);
6100 
6101 /* must be called under rcu_read_lock(), as we dont take a reference */
6102 static struct napi_struct *napi_by_id(unsigned int napi_id)
6103 {
6104 	unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6105 	struct napi_struct *napi;
6106 
6107 	hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6108 		if (napi->napi_id == napi_id)
6109 			return napi;
6110 
6111 	return NULL;
6112 }
6113 
6114 #if defined(CONFIG_NET_RX_BUSY_POLL)
6115 
6116 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6117 {
6118 	if (!skip_schedule) {
6119 		gro_normal_list(napi);
6120 		__napi_schedule(napi);
6121 		return;
6122 	}
6123 
6124 	if (napi->gro_bitmask) {
6125 		/* flush too old packets
6126 		 * If HZ < 1000, flush all packets.
6127 		 */
6128 		napi_gro_flush(napi, HZ >= 1000);
6129 	}
6130 
6131 	gro_normal_list(napi);
6132 	clear_bit(NAPI_STATE_SCHED, &napi->state);
6133 }
6134 
6135 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6136 			   u16 budget)
6137 {
6138 	bool skip_schedule = false;
6139 	unsigned long timeout;
6140 	int rc;
6141 
6142 	/* Busy polling means there is a high chance device driver hard irq
6143 	 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6144 	 * set in napi_schedule_prep().
6145 	 * Since we are about to call napi->poll() once more, we can safely
6146 	 * clear NAPI_STATE_MISSED.
6147 	 *
6148 	 * Note: x86 could use a single "lock and ..." instruction
6149 	 * to perform these two clear_bit()
6150 	 */
6151 	clear_bit(NAPI_STATE_MISSED, &napi->state);
6152 	clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6153 
6154 	local_bh_disable();
6155 
6156 	if (prefer_busy_poll) {
6157 		napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6158 		timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6159 		if (napi->defer_hard_irqs_count && timeout) {
6160 			hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6161 			skip_schedule = true;
6162 		}
6163 	}
6164 
6165 	/* All we really want here is to re-enable device interrupts.
6166 	 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6167 	 */
6168 	rc = napi->poll(napi, budget);
6169 	/* We can't gro_normal_list() here, because napi->poll() might have
6170 	 * rearmed the napi (napi_complete_done()) in which case it could
6171 	 * already be running on another CPU.
6172 	 */
6173 	trace_napi_poll(napi, rc, budget);
6174 	netpoll_poll_unlock(have_poll_lock);
6175 	if (rc == budget)
6176 		__busy_poll_stop(napi, skip_schedule);
6177 	local_bh_enable();
6178 }
6179 
6180 void napi_busy_loop(unsigned int napi_id,
6181 		    bool (*loop_end)(void *, unsigned long),
6182 		    void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6183 {
6184 	unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6185 	int (*napi_poll)(struct napi_struct *napi, int budget);
6186 	void *have_poll_lock = NULL;
6187 	struct napi_struct *napi;
6188 
6189 restart:
6190 	napi_poll = NULL;
6191 
6192 	rcu_read_lock();
6193 
6194 	napi = napi_by_id(napi_id);
6195 	if (!napi)
6196 		goto out;
6197 
6198 	preempt_disable();
6199 	for (;;) {
6200 		int work = 0;
6201 
6202 		local_bh_disable();
6203 		if (!napi_poll) {
6204 			unsigned long val = READ_ONCE(napi->state);
6205 
6206 			/* If multiple threads are competing for this napi,
6207 			 * we avoid dirtying napi->state as much as we can.
6208 			 */
6209 			if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6210 				   NAPIF_STATE_IN_BUSY_POLL)) {
6211 				if (prefer_busy_poll)
6212 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6213 				goto count;
6214 			}
6215 			if (cmpxchg(&napi->state, val,
6216 				    val | NAPIF_STATE_IN_BUSY_POLL |
6217 					  NAPIF_STATE_SCHED) != val) {
6218 				if (prefer_busy_poll)
6219 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6220 				goto count;
6221 			}
6222 			have_poll_lock = netpoll_poll_lock(napi);
6223 			napi_poll = napi->poll;
6224 		}
6225 		work = napi_poll(napi, budget);
6226 		trace_napi_poll(napi, work, budget);
6227 		gro_normal_list(napi);
6228 count:
6229 		if (work > 0)
6230 			__NET_ADD_STATS(dev_net(napi->dev),
6231 					LINUX_MIB_BUSYPOLLRXPACKETS, work);
6232 		local_bh_enable();
6233 
6234 		if (!loop_end || loop_end(loop_end_arg, start_time))
6235 			break;
6236 
6237 		if (unlikely(need_resched())) {
6238 			if (napi_poll)
6239 				busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6240 			preempt_enable();
6241 			rcu_read_unlock();
6242 			cond_resched();
6243 			if (loop_end(loop_end_arg, start_time))
6244 				return;
6245 			goto restart;
6246 		}
6247 		cpu_relax();
6248 	}
6249 	if (napi_poll)
6250 		busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6251 	preempt_enable();
6252 out:
6253 	rcu_read_unlock();
6254 }
6255 EXPORT_SYMBOL(napi_busy_loop);
6256 
6257 #endif /* CONFIG_NET_RX_BUSY_POLL */
6258 
6259 static void napi_hash_add(struct napi_struct *napi)
6260 {
6261 	if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6262 		return;
6263 
6264 	spin_lock(&napi_hash_lock);
6265 
6266 	/* 0..NR_CPUS range is reserved for sender_cpu use */
6267 	do {
6268 		if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6269 			napi_gen_id = MIN_NAPI_ID;
6270 	} while (napi_by_id(napi_gen_id));
6271 	napi->napi_id = napi_gen_id;
6272 
6273 	hlist_add_head_rcu(&napi->napi_hash_node,
6274 			   &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6275 
6276 	spin_unlock(&napi_hash_lock);
6277 }
6278 
6279 /* Warning : caller is responsible to make sure rcu grace period
6280  * is respected before freeing memory containing @napi
6281  */
6282 static void napi_hash_del(struct napi_struct *napi)
6283 {
6284 	spin_lock(&napi_hash_lock);
6285 
6286 	hlist_del_init_rcu(&napi->napi_hash_node);
6287 
6288 	spin_unlock(&napi_hash_lock);
6289 }
6290 
6291 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6292 {
6293 	struct napi_struct *napi;
6294 
6295 	napi = container_of(timer, struct napi_struct, timer);
6296 
6297 	/* Note : we use a relaxed variant of napi_schedule_prep() not setting
6298 	 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6299 	 */
6300 	if (!napi_disable_pending(napi) &&
6301 	    !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6302 		clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6303 		__napi_schedule_irqoff(napi);
6304 	}
6305 
6306 	return HRTIMER_NORESTART;
6307 }
6308 
6309 static void init_gro_hash(struct napi_struct *napi)
6310 {
6311 	int i;
6312 
6313 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6314 		INIT_LIST_HEAD(&napi->gro_hash[i].list);
6315 		napi->gro_hash[i].count = 0;
6316 	}
6317 	napi->gro_bitmask = 0;
6318 }
6319 
6320 int dev_set_threaded(struct net_device *dev, bool threaded)
6321 {
6322 	struct napi_struct *napi;
6323 	int err = 0;
6324 
6325 	if (dev->threaded == threaded)
6326 		return 0;
6327 
6328 	if (threaded) {
6329 		list_for_each_entry(napi, &dev->napi_list, dev_list) {
6330 			if (!napi->thread) {
6331 				err = napi_kthread_create(napi);
6332 				if (err) {
6333 					threaded = false;
6334 					break;
6335 				}
6336 			}
6337 		}
6338 	}
6339 
6340 	dev->threaded = threaded;
6341 
6342 	/* Make sure kthread is created before THREADED bit
6343 	 * is set.
6344 	 */
6345 	smp_mb__before_atomic();
6346 
6347 	/* Setting/unsetting threaded mode on a napi might not immediately
6348 	 * take effect, if the current napi instance is actively being
6349 	 * polled. In this case, the switch between threaded mode and
6350 	 * softirq mode will happen in the next round of napi_schedule().
6351 	 * This should not cause hiccups/stalls to the live traffic.
6352 	 */
6353 	list_for_each_entry(napi, &dev->napi_list, dev_list) {
6354 		if (threaded)
6355 			set_bit(NAPI_STATE_THREADED, &napi->state);
6356 		else
6357 			clear_bit(NAPI_STATE_THREADED, &napi->state);
6358 	}
6359 
6360 	return err;
6361 }
6362 EXPORT_SYMBOL(dev_set_threaded);
6363 
6364 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi,
6365 			   int (*poll)(struct napi_struct *, int), int weight)
6366 {
6367 	if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6368 		return;
6369 
6370 	INIT_LIST_HEAD(&napi->poll_list);
6371 	INIT_HLIST_NODE(&napi->napi_hash_node);
6372 	hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6373 	napi->timer.function = napi_watchdog;
6374 	init_gro_hash(napi);
6375 	napi->skb = NULL;
6376 	INIT_LIST_HEAD(&napi->rx_list);
6377 	napi->rx_count = 0;
6378 	napi->poll = poll;
6379 	if (weight > NAPI_POLL_WEIGHT)
6380 		netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6381 				weight);
6382 	napi->weight = weight;
6383 	napi->dev = dev;
6384 #ifdef CONFIG_NETPOLL
6385 	napi->poll_owner = -1;
6386 #endif
6387 	set_bit(NAPI_STATE_SCHED, &napi->state);
6388 	set_bit(NAPI_STATE_NPSVC, &napi->state);
6389 	list_add_rcu(&napi->dev_list, &dev->napi_list);
6390 	napi_hash_add(napi);
6391 	napi_get_frags_check(napi);
6392 	/* Create kthread for this napi if dev->threaded is set.
6393 	 * Clear dev->threaded if kthread creation failed so that
6394 	 * threaded mode will not be enabled in napi_enable().
6395 	 */
6396 	if (dev->threaded && napi_kthread_create(napi))
6397 		dev->threaded = 0;
6398 }
6399 EXPORT_SYMBOL(netif_napi_add_weight);
6400 
6401 void napi_disable(struct napi_struct *n)
6402 {
6403 	unsigned long val, new;
6404 
6405 	might_sleep();
6406 	set_bit(NAPI_STATE_DISABLE, &n->state);
6407 
6408 	for ( ; ; ) {
6409 		val = READ_ONCE(n->state);
6410 		if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6411 			usleep_range(20, 200);
6412 			continue;
6413 		}
6414 
6415 		new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6416 		new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6417 
6418 		if (cmpxchg(&n->state, val, new) == val)
6419 			break;
6420 	}
6421 
6422 	hrtimer_cancel(&n->timer);
6423 
6424 	clear_bit(NAPI_STATE_DISABLE, &n->state);
6425 }
6426 EXPORT_SYMBOL(napi_disable);
6427 
6428 /**
6429  *	napi_enable - enable NAPI scheduling
6430  *	@n: NAPI context
6431  *
6432  * Resume NAPI from being scheduled on this context.
6433  * Must be paired with napi_disable.
6434  */
6435 void napi_enable(struct napi_struct *n)
6436 {
6437 	unsigned long val, new;
6438 
6439 	do {
6440 		val = READ_ONCE(n->state);
6441 		BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6442 
6443 		new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6444 		if (n->dev->threaded && n->thread)
6445 			new |= NAPIF_STATE_THREADED;
6446 	} while (cmpxchg(&n->state, val, new) != val);
6447 }
6448 EXPORT_SYMBOL(napi_enable);
6449 
6450 static void flush_gro_hash(struct napi_struct *napi)
6451 {
6452 	int i;
6453 
6454 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6455 		struct sk_buff *skb, *n;
6456 
6457 		list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6458 			kfree_skb(skb);
6459 		napi->gro_hash[i].count = 0;
6460 	}
6461 }
6462 
6463 /* Must be called in process context */
6464 void __netif_napi_del(struct napi_struct *napi)
6465 {
6466 	if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6467 		return;
6468 
6469 	napi_hash_del(napi);
6470 	list_del_rcu(&napi->dev_list);
6471 	napi_free_frags(napi);
6472 
6473 	flush_gro_hash(napi);
6474 	napi->gro_bitmask = 0;
6475 
6476 	if (napi->thread) {
6477 		kthread_stop(napi->thread);
6478 		napi->thread = NULL;
6479 	}
6480 }
6481 EXPORT_SYMBOL(__netif_napi_del);
6482 
6483 static int __napi_poll(struct napi_struct *n, bool *repoll)
6484 {
6485 	int work, weight;
6486 
6487 	weight = n->weight;
6488 
6489 	/* This NAPI_STATE_SCHED test is for avoiding a race
6490 	 * with netpoll's poll_napi().  Only the entity which
6491 	 * obtains the lock and sees NAPI_STATE_SCHED set will
6492 	 * actually make the ->poll() call.  Therefore we avoid
6493 	 * accidentally calling ->poll() when NAPI is not scheduled.
6494 	 */
6495 	work = 0;
6496 	if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6497 		work = n->poll(n, weight);
6498 		trace_napi_poll(n, work, weight);
6499 	}
6500 
6501 	if (unlikely(work > weight))
6502 		netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6503 				n->poll, work, weight);
6504 
6505 	if (likely(work < weight))
6506 		return work;
6507 
6508 	/* Drivers must not modify the NAPI state if they
6509 	 * consume the entire weight.  In such cases this code
6510 	 * still "owns" the NAPI instance and therefore can
6511 	 * move the instance around on the list at-will.
6512 	 */
6513 	if (unlikely(napi_disable_pending(n))) {
6514 		napi_complete(n);
6515 		return work;
6516 	}
6517 
6518 	/* The NAPI context has more processing work, but busy-polling
6519 	 * is preferred. Exit early.
6520 	 */
6521 	if (napi_prefer_busy_poll(n)) {
6522 		if (napi_complete_done(n, work)) {
6523 			/* If timeout is not set, we need to make sure
6524 			 * that the NAPI is re-scheduled.
6525 			 */
6526 			napi_schedule(n);
6527 		}
6528 		return work;
6529 	}
6530 
6531 	if (n->gro_bitmask) {
6532 		/* flush too old packets
6533 		 * If HZ < 1000, flush all packets.
6534 		 */
6535 		napi_gro_flush(n, HZ >= 1000);
6536 	}
6537 
6538 	gro_normal_list(n);
6539 
6540 	/* Some drivers may have called napi_schedule
6541 	 * prior to exhausting their budget.
6542 	 */
6543 	if (unlikely(!list_empty(&n->poll_list))) {
6544 		pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6545 			     n->dev ? n->dev->name : "backlog");
6546 		return work;
6547 	}
6548 
6549 	*repoll = true;
6550 
6551 	return work;
6552 }
6553 
6554 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6555 {
6556 	bool do_repoll = false;
6557 	void *have;
6558 	int work;
6559 
6560 	list_del_init(&n->poll_list);
6561 
6562 	have = netpoll_poll_lock(n);
6563 
6564 	work = __napi_poll(n, &do_repoll);
6565 
6566 	if (do_repoll)
6567 		list_add_tail(&n->poll_list, repoll);
6568 
6569 	netpoll_poll_unlock(have);
6570 
6571 	return work;
6572 }
6573 
6574 static int napi_thread_wait(struct napi_struct *napi)
6575 {
6576 	bool woken = false;
6577 
6578 	set_current_state(TASK_INTERRUPTIBLE);
6579 
6580 	while (!kthread_should_stop()) {
6581 		/* Testing SCHED_THREADED bit here to make sure the current
6582 		 * kthread owns this napi and could poll on this napi.
6583 		 * Testing SCHED bit is not enough because SCHED bit might be
6584 		 * set by some other busy poll thread or by napi_disable().
6585 		 */
6586 		if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
6587 			WARN_ON(!list_empty(&napi->poll_list));
6588 			__set_current_state(TASK_RUNNING);
6589 			return 0;
6590 		}
6591 
6592 		schedule();
6593 		/* woken being true indicates this thread owns this napi. */
6594 		woken = true;
6595 		set_current_state(TASK_INTERRUPTIBLE);
6596 	}
6597 	__set_current_state(TASK_RUNNING);
6598 
6599 	return -1;
6600 }
6601 
6602 static int napi_threaded_poll(void *data)
6603 {
6604 	struct napi_struct *napi = data;
6605 	void *have;
6606 
6607 	while (!napi_thread_wait(napi)) {
6608 		for (;;) {
6609 			bool repoll = false;
6610 
6611 			local_bh_disable();
6612 
6613 			have = netpoll_poll_lock(napi);
6614 			__napi_poll(napi, &repoll);
6615 			netpoll_poll_unlock(have);
6616 
6617 			local_bh_enable();
6618 
6619 			if (!repoll)
6620 				break;
6621 
6622 			cond_resched();
6623 		}
6624 	}
6625 	return 0;
6626 }
6627 
6628 static void skb_defer_free_flush(struct softnet_data *sd)
6629 {
6630 	struct sk_buff *skb, *next;
6631 	unsigned long flags;
6632 
6633 	/* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
6634 	if (!READ_ONCE(sd->defer_list))
6635 		return;
6636 
6637 	spin_lock_irqsave(&sd->defer_lock, flags);
6638 	skb = sd->defer_list;
6639 	sd->defer_list = NULL;
6640 	sd->defer_count = 0;
6641 	spin_unlock_irqrestore(&sd->defer_lock, flags);
6642 
6643 	while (skb != NULL) {
6644 		next = skb->next;
6645 		napi_consume_skb(skb, 1);
6646 		skb = next;
6647 	}
6648 }
6649 
6650 static __latent_entropy void net_rx_action(struct softirq_action *h)
6651 {
6652 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6653 	unsigned long time_limit = jiffies +
6654 		usecs_to_jiffies(READ_ONCE(netdev_budget_usecs));
6655 	int budget = READ_ONCE(netdev_budget);
6656 	LIST_HEAD(list);
6657 	LIST_HEAD(repoll);
6658 
6659 	local_irq_disable();
6660 	list_splice_init(&sd->poll_list, &list);
6661 	local_irq_enable();
6662 
6663 	for (;;) {
6664 		struct napi_struct *n;
6665 
6666 		skb_defer_free_flush(sd);
6667 
6668 		if (list_empty(&list)) {
6669 			if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6670 				goto end;
6671 			break;
6672 		}
6673 
6674 		n = list_first_entry(&list, struct napi_struct, poll_list);
6675 		budget -= napi_poll(n, &repoll);
6676 
6677 		/* If softirq window is exhausted then punt.
6678 		 * Allow this to run for 2 jiffies since which will allow
6679 		 * an average latency of 1.5/HZ.
6680 		 */
6681 		if (unlikely(budget <= 0 ||
6682 			     time_after_eq(jiffies, time_limit))) {
6683 			sd->time_squeeze++;
6684 			break;
6685 		}
6686 	}
6687 
6688 	local_irq_disable();
6689 
6690 	list_splice_tail_init(&sd->poll_list, &list);
6691 	list_splice_tail(&repoll, &list);
6692 	list_splice(&list, &sd->poll_list);
6693 	if (!list_empty(&sd->poll_list))
6694 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
6695 
6696 	net_rps_action_and_irq_enable(sd);
6697 end:;
6698 }
6699 
6700 struct netdev_adjacent {
6701 	struct net_device *dev;
6702 	netdevice_tracker dev_tracker;
6703 
6704 	/* upper master flag, there can only be one master device per list */
6705 	bool master;
6706 
6707 	/* lookup ignore flag */
6708 	bool ignore;
6709 
6710 	/* counter for the number of times this device was added to us */
6711 	u16 ref_nr;
6712 
6713 	/* private field for the users */
6714 	void *private;
6715 
6716 	struct list_head list;
6717 	struct rcu_head rcu;
6718 };
6719 
6720 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6721 						 struct list_head *adj_list)
6722 {
6723 	struct netdev_adjacent *adj;
6724 
6725 	list_for_each_entry(adj, adj_list, list) {
6726 		if (adj->dev == adj_dev)
6727 			return adj;
6728 	}
6729 	return NULL;
6730 }
6731 
6732 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
6733 				    struct netdev_nested_priv *priv)
6734 {
6735 	struct net_device *dev = (struct net_device *)priv->data;
6736 
6737 	return upper_dev == dev;
6738 }
6739 
6740 /**
6741  * netdev_has_upper_dev - Check if device is linked to an upper device
6742  * @dev: device
6743  * @upper_dev: upper device to check
6744  *
6745  * Find out if a device is linked to specified upper device and return true
6746  * in case it is. Note that this checks only immediate upper device,
6747  * not through a complete stack of devices. The caller must hold the RTNL lock.
6748  */
6749 bool netdev_has_upper_dev(struct net_device *dev,
6750 			  struct net_device *upper_dev)
6751 {
6752 	struct netdev_nested_priv priv = {
6753 		.data = (void *)upper_dev,
6754 	};
6755 
6756 	ASSERT_RTNL();
6757 
6758 	return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6759 					     &priv);
6760 }
6761 EXPORT_SYMBOL(netdev_has_upper_dev);
6762 
6763 /**
6764  * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
6765  * @dev: device
6766  * @upper_dev: upper device to check
6767  *
6768  * Find out if a device is linked to specified upper device and return true
6769  * in case it is. Note that this checks the entire upper device chain.
6770  * The caller must hold rcu lock.
6771  */
6772 
6773 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6774 				  struct net_device *upper_dev)
6775 {
6776 	struct netdev_nested_priv priv = {
6777 		.data = (void *)upper_dev,
6778 	};
6779 
6780 	return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6781 					       &priv);
6782 }
6783 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6784 
6785 /**
6786  * netdev_has_any_upper_dev - Check if device is linked to some device
6787  * @dev: device
6788  *
6789  * Find out if a device is linked to an upper device and return true in case
6790  * it is. The caller must hold the RTNL lock.
6791  */
6792 bool netdev_has_any_upper_dev(struct net_device *dev)
6793 {
6794 	ASSERT_RTNL();
6795 
6796 	return !list_empty(&dev->adj_list.upper);
6797 }
6798 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6799 
6800 /**
6801  * netdev_master_upper_dev_get - Get master upper device
6802  * @dev: device
6803  *
6804  * Find a master upper device and return pointer to it or NULL in case
6805  * it's not there. The caller must hold the RTNL lock.
6806  */
6807 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6808 {
6809 	struct netdev_adjacent *upper;
6810 
6811 	ASSERT_RTNL();
6812 
6813 	if (list_empty(&dev->adj_list.upper))
6814 		return NULL;
6815 
6816 	upper = list_first_entry(&dev->adj_list.upper,
6817 				 struct netdev_adjacent, list);
6818 	if (likely(upper->master))
6819 		return upper->dev;
6820 	return NULL;
6821 }
6822 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6823 
6824 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6825 {
6826 	struct netdev_adjacent *upper;
6827 
6828 	ASSERT_RTNL();
6829 
6830 	if (list_empty(&dev->adj_list.upper))
6831 		return NULL;
6832 
6833 	upper = list_first_entry(&dev->adj_list.upper,
6834 				 struct netdev_adjacent, list);
6835 	if (likely(upper->master) && !upper->ignore)
6836 		return upper->dev;
6837 	return NULL;
6838 }
6839 
6840 /**
6841  * netdev_has_any_lower_dev - Check if device is linked to some device
6842  * @dev: device
6843  *
6844  * Find out if a device is linked to a lower device and return true in case
6845  * it is. The caller must hold the RTNL lock.
6846  */
6847 static bool netdev_has_any_lower_dev(struct net_device *dev)
6848 {
6849 	ASSERT_RTNL();
6850 
6851 	return !list_empty(&dev->adj_list.lower);
6852 }
6853 
6854 void *netdev_adjacent_get_private(struct list_head *adj_list)
6855 {
6856 	struct netdev_adjacent *adj;
6857 
6858 	adj = list_entry(adj_list, struct netdev_adjacent, list);
6859 
6860 	return adj->private;
6861 }
6862 EXPORT_SYMBOL(netdev_adjacent_get_private);
6863 
6864 /**
6865  * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6866  * @dev: device
6867  * @iter: list_head ** of the current position
6868  *
6869  * Gets the next device from the dev's upper list, starting from iter
6870  * position. The caller must hold RCU read lock.
6871  */
6872 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6873 						 struct list_head **iter)
6874 {
6875 	struct netdev_adjacent *upper;
6876 
6877 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6878 
6879 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6880 
6881 	if (&upper->list == &dev->adj_list.upper)
6882 		return NULL;
6883 
6884 	*iter = &upper->list;
6885 
6886 	return upper->dev;
6887 }
6888 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6889 
6890 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6891 						  struct list_head **iter,
6892 						  bool *ignore)
6893 {
6894 	struct netdev_adjacent *upper;
6895 
6896 	upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6897 
6898 	if (&upper->list == &dev->adj_list.upper)
6899 		return NULL;
6900 
6901 	*iter = &upper->list;
6902 	*ignore = upper->ignore;
6903 
6904 	return upper->dev;
6905 }
6906 
6907 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6908 						    struct list_head **iter)
6909 {
6910 	struct netdev_adjacent *upper;
6911 
6912 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6913 
6914 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6915 
6916 	if (&upper->list == &dev->adj_list.upper)
6917 		return NULL;
6918 
6919 	*iter = &upper->list;
6920 
6921 	return upper->dev;
6922 }
6923 
6924 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6925 				       int (*fn)(struct net_device *dev,
6926 					 struct netdev_nested_priv *priv),
6927 				       struct netdev_nested_priv *priv)
6928 {
6929 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6930 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6931 	int ret, cur = 0;
6932 	bool ignore;
6933 
6934 	now = dev;
6935 	iter = &dev->adj_list.upper;
6936 
6937 	while (1) {
6938 		if (now != dev) {
6939 			ret = fn(now, priv);
6940 			if (ret)
6941 				return ret;
6942 		}
6943 
6944 		next = NULL;
6945 		while (1) {
6946 			udev = __netdev_next_upper_dev(now, &iter, &ignore);
6947 			if (!udev)
6948 				break;
6949 			if (ignore)
6950 				continue;
6951 
6952 			next = udev;
6953 			niter = &udev->adj_list.upper;
6954 			dev_stack[cur] = now;
6955 			iter_stack[cur++] = iter;
6956 			break;
6957 		}
6958 
6959 		if (!next) {
6960 			if (!cur)
6961 				return 0;
6962 			next = dev_stack[--cur];
6963 			niter = iter_stack[cur];
6964 		}
6965 
6966 		now = next;
6967 		iter = niter;
6968 	}
6969 
6970 	return 0;
6971 }
6972 
6973 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6974 				  int (*fn)(struct net_device *dev,
6975 					    struct netdev_nested_priv *priv),
6976 				  struct netdev_nested_priv *priv)
6977 {
6978 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6979 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6980 	int ret, cur = 0;
6981 
6982 	now = dev;
6983 	iter = &dev->adj_list.upper;
6984 
6985 	while (1) {
6986 		if (now != dev) {
6987 			ret = fn(now, priv);
6988 			if (ret)
6989 				return ret;
6990 		}
6991 
6992 		next = NULL;
6993 		while (1) {
6994 			udev = netdev_next_upper_dev_rcu(now, &iter);
6995 			if (!udev)
6996 				break;
6997 
6998 			next = udev;
6999 			niter = &udev->adj_list.upper;
7000 			dev_stack[cur] = now;
7001 			iter_stack[cur++] = iter;
7002 			break;
7003 		}
7004 
7005 		if (!next) {
7006 			if (!cur)
7007 				return 0;
7008 			next = dev_stack[--cur];
7009 			niter = iter_stack[cur];
7010 		}
7011 
7012 		now = next;
7013 		iter = niter;
7014 	}
7015 
7016 	return 0;
7017 }
7018 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7019 
7020 static bool __netdev_has_upper_dev(struct net_device *dev,
7021 				   struct net_device *upper_dev)
7022 {
7023 	struct netdev_nested_priv priv = {
7024 		.flags = 0,
7025 		.data = (void *)upper_dev,
7026 	};
7027 
7028 	ASSERT_RTNL();
7029 
7030 	return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7031 					   &priv);
7032 }
7033 
7034 /**
7035  * netdev_lower_get_next_private - Get the next ->private from the
7036  *				   lower neighbour list
7037  * @dev: device
7038  * @iter: list_head ** of the current position
7039  *
7040  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7041  * list, starting from iter position. The caller must hold either hold the
7042  * RTNL lock or its own locking that guarantees that the neighbour lower
7043  * list will remain unchanged.
7044  */
7045 void *netdev_lower_get_next_private(struct net_device *dev,
7046 				    struct list_head **iter)
7047 {
7048 	struct netdev_adjacent *lower;
7049 
7050 	lower = list_entry(*iter, struct netdev_adjacent, list);
7051 
7052 	if (&lower->list == &dev->adj_list.lower)
7053 		return NULL;
7054 
7055 	*iter = lower->list.next;
7056 
7057 	return lower->private;
7058 }
7059 EXPORT_SYMBOL(netdev_lower_get_next_private);
7060 
7061 /**
7062  * netdev_lower_get_next_private_rcu - Get the next ->private from the
7063  *				       lower neighbour list, RCU
7064  *				       variant
7065  * @dev: device
7066  * @iter: list_head ** of the current position
7067  *
7068  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7069  * list, starting from iter position. The caller must hold RCU read lock.
7070  */
7071 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7072 					struct list_head **iter)
7073 {
7074 	struct netdev_adjacent *lower;
7075 
7076 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7077 
7078 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7079 
7080 	if (&lower->list == &dev->adj_list.lower)
7081 		return NULL;
7082 
7083 	*iter = &lower->list;
7084 
7085 	return lower->private;
7086 }
7087 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7088 
7089 /**
7090  * netdev_lower_get_next - Get the next device from the lower neighbour
7091  *                         list
7092  * @dev: device
7093  * @iter: list_head ** of the current position
7094  *
7095  * Gets the next netdev_adjacent from the dev's lower neighbour
7096  * list, starting from iter position. The caller must hold RTNL lock or
7097  * its own locking that guarantees that the neighbour lower
7098  * list will remain unchanged.
7099  */
7100 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7101 {
7102 	struct netdev_adjacent *lower;
7103 
7104 	lower = list_entry(*iter, struct netdev_adjacent, list);
7105 
7106 	if (&lower->list == &dev->adj_list.lower)
7107 		return NULL;
7108 
7109 	*iter = lower->list.next;
7110 
7111 	return lower->dev;
7112 }
7113 EXPORT_SYMBOL(netdev_lower_get_next);
7114 
7115 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7116 						struct list_head **iter)
7117 {
7118 	struct netdev_adjacent *lower;
7119 
7120 	lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7121 
7122 	if (&lower->list == &dev->adj_list.lower)
7123 		return NULL;
7124 
7125 	*iter = &lower->list;
7126 
7127 	return lower->dev;
7128 }
7129 
7130 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7131 						  struct list_head **iter,
7132 						  bool *ignore)
7133 {
7134 	struct netdev_adjacent *lower;
7135 
7136 	lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7137 
7138 	if (&lower->list == &dev->adj_list.lower)
7139 		return NULL;
7140 
7141 	*iter = &lower->list;
7142 	*ignore = lower->ignore;
7143 
7144 	return lower->dev;
7145 }
7146 
7147 int netdev_walk_all_lower_dev(struct net_device *dev,
7148 			      int (*fn)(struct net_device *dev,
7149 					struct netdev_nested_priv *priv),
7150 			      struct netdev_nested_priv *priv)
7151 {
7152 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7153 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7154 	int ret, cur = 0;
7155 
7156 	now = dev;
7157 	iter = &dev->adj_list.lower;
7158 
7159 	while (1) {
7160 		if (now != dev) {
7161 			ret = fn(now, priv);
7162 			if (ret)
7163 				return ret;
7164 		}
7165 
7166 		next = NULL;
7167 		while (1) {
7168 			ldev = netdev_next_lower_dev(now, &iter);
7169 			if (!ldev)
7170 				break;
7171 
7172 			next = ldev;
7173 			niter = &ldev->adj_list.lower;
7174 			dev_stack[cur] = now;
7175 			iter_stack[cur++] = iter;
7176 			break;
7177 		}
7178 
7179 		if (!next) {
7180 			if (!cur)
7181 				return 0;
7182 			next = dev_stack[--cur];
7183 			niter = iter_stack[cur];
7184 		}
7185 
7186 		now = next;
7187 		iter = niter;
7188 	}
7189 
7190 	return 0;
7191 }
7192 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7193 
7194 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7195 				       int (*fn)(struct net_device *dev,
7196 					 struct netdev_nested_priv *priv),
7197 				       struct netdev_nested_priv *priv)
7198 {
7199 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7200 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7201 	int ret, cur = 0;
7202 	bool ignore;
7203 
7204 	now = dev;
7205 	iter = &dev->adj_list.lower;
7206 
7207 	while (1) {
7208 		if (now != dev) {
7209 			ret = fn(now, priv);
7210 			if (ret)
7211 				return ret;
7212 		}
7213 
7214 		next = NULL;
7215 		while (1) {
7216 			ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7217 			if (!ldev)
7218 				break;
7219 			if (ignore)
7220 				continue;
7221 
7222 			next = ldev;
7223 			niter = &ldev->adj_list.lower;
7224 			dev_stack[cur] = now;
7225 			iter_stack[cur++] = iter;
7226 			break;
7227 		}
7228 
7229 		if (!next) {
7230 			if (!cur)
7231 				return 0;
7232 			next = dev_stack[--cur];
7233 			niter = iter_stack[cur];
7234 		}
7235 
7236 		now = next;
7237 		iter = niter;
7238 	}
7239 
7240 	return 0;
7241 }
7242 
7243 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7244 					     struct list_head **iter)
7245 {
7246 	struct netdev_adjacent *lower;
7247 
7248 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7249 	if (&lower->list == &dev->adj_list.lower)
7250 		return NULL;
7251 
7252 	*iter = &lower->list;
7253 
7254 	return lower->dev;
7255 }
7256 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7257 
7258 static u8 __netdev_upper_depth(struct net_device *dev)
7259 {
7260 	struct net_device *udev;
7261 	struct list_head *iter;
7262 	u8 max_depth = 0;
7263 	bool ignore;
7264 
7265 	for (iter = &dev->adj_list.upper,
7266 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7267 	     udev;
7268 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7269 		if (ignore)
7270 			continue;
7271 		if (max_depth < udev->upper_level)
7272 			max_depth = udev->upper_level;
7273 	}
7274 
7275 	return max_depth;
7276 }
7277 
7278 static u8 __netdev_lower_depth(struct net_device *dev)
7279 {
7280 	struct net_device *ldev;
7281 	struct list_head *iter;
7282 	u8 max_depth = 0;
7283 	bool ignore;
7284 
7285 	for (iter = &dev->adj_list.lower,
7286 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7287 	     ldev;
7288 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7289 		if (ignore)
7290 			continue;
7291 		if (max_depth < ldev->lower_level)
7292 			max_depth = ldev->lower_level;
7293 	}
7294 
7295 	return max_depth;
7296 }
7297 
7298 static int __netdev_update_upper_level(struct net_device *dev,
7299 				       struct netdev_nested_priv *__unused)
7300 {
7301 	dev->upper_level = __netdev_upper_depth(dev) + 1;
7302 	return 0;
7303 }
7304 
7305 #ifdef CONFIG_LOCKDEP
7306 static LIST_HEAD(net_unlink_list);
7307 
7308 static void net_unlink_todo(struct net_device *dev)
7309 {
7310 	if (list_empty(&dev->unlink_list))
7311 		list_add_tail(&dev->unlink_list, &net_unlink_list);
7312 }
7313 #endif
7314 
7315 static int __netdev_update_lower_level(struct net_device *dev,
7316 				       struct netdev_nested_priv *priv)
7317 {
7318 	dev->lower_level = __netdev_lower_depth(dev) + 1;
7319 
7320 #ifdef CONFIG_LOCKDEP
7321 	if (!priv)
7322 		return 0;
7323 
7324 	if (priv->flags & NESTED_SYNC_IMM)
7325 		dev->nested_level = dev->lower_level - 1;
7326 	if (priv->flags & NESTED_SYNC_TODO)
7327 		net_unlink_todo(dev);
7328 #endif
7329 	return 0;
7330 }
7331 
7332 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7333 				  int (*fn)(struct net_device *dev,
7334 					    struct netdev_nested_priv *priv),
7335 				  struct netdev_nested_priv *priv)
7336 {
7337 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7338 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7339 	int ret, cur = 0;
7340 
7341 	now = dev;
7342 	iter = &dev->adj_list.lower;
7343 
7344 	while (1) {
7345 		if (now != dev) {
7346 			ret = fn(now, priv);
7347 			if (ret)
7348 				return ret;
7349 		}
7350 
7351 		next = NULL;
7352 		while (1) {
7353 			ldev = netdev_next_lower_dev_rcu(now, &iter);
7354 			if (!ldev)
7355 				break;
7356 
7357 			next = ldev;
7358 			niter = &ldev->adj_list.lower;
7359 			dev_stack[cur] = now;
7360 			iter_stack[cur++] = iter;
7361 			break;
7362 		}
7363 
7364 		if (!next) {
7365 			if (!cur)
7366 				return 0;
7367 			next = dev_stack[--cur];
7368 			niter = iter_stack[cur];
7369 		}
7370 
7371 		now = next;
7372 		iter = niter;
7373 	}
7374 
7375 	return 0;
7376 }
7377 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7378 
7379 /**
7380  * netdev_lower_get_first_private_rcu - Get the first ->private from the
7381  *				       lower neighbour list, RCU
7382  *				       variant
7383  * @dev: device
7384  *
7385  * Gets the first netdev_adjacent->private from the dev's lower neighbour
7386  * list. The caller must hold RCU read lock.
7387  */
7388 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7389 {
7390 	struct netdev_adjacent *lower;
7391 
7392 	lower = list_first_or_null_rcu(&dev->adj_list.lower,
7393 			struct netdev_adjacent, list);
7394 	if (lower)
7395 		return lower->private;
7396 	return NULL;
7397 }
7398 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7399 
7400 /**
7401  * netdev_master_upper_dev_get_rcu - Get master upper device
7402  * @dev: device
7403  *
7404  * Find a master upper device and return pointer to it or NULL in case
7405  * it's not there. The caller must hold the RCU read lock.
7406  */
7407 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7408 {
7409 	struct netdev_adjacent *upper;
7410 
7411 	upper = list_first_or_null_rcu(&dev->adj_list.upper,
7412 				       struct netdev_adjacent, list);
7413 	if (upper && likely(upper->master))
7414 		return upper->dev;
7415 	return NULL;
7416 }
7417 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7418 
7419 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7420 			      struct net_device *adj_dev,
7421 			      struct list_head *dev_list)
7422 {
7423 	char linkname[IFNAMSIZ+7];
7424 
7425 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7426 		"upper_%s" : "lower_%s", adj_dev->name);
7427 	return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7428 				 linkname);
7429 }
7430 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7431 			       char *name,
7432 			       struct list_head *dev_list)
7433 {
7434 	char linkname[IFNAMSIZ+7];
7435 
7436 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7437 		"upper_%s" : "lower_%s", name);
7438 	sysfs_remove_link(&(dev->dev.kobj), linkname);
7439 }
7440 
7441 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7442 						 struct net_device *adj_dev,
7443 						 struct list_head *dev_list)
7444 {
7445 	return (dev_list == &dev->adj_list.upper ||
7446 		dev_list == &dev->adj_list.lower) &&
7447 		net_eq(dev_net(dev), dev_net(adj_dev));
7448 }
7449 
7450 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7451 					struct net_device *adj_dev,
7452 					struct list_head *dev_list,
7453 					void *private, bool master)
7454 {
7455 	struct netdev_adjacent *adj;
7456 	int ret;
7457 
7458 	adj = __netdev_find_adj(adj_dev, dev_list);
7459 
7460 	if (adj) {
7461 		adj->ref_nr += 1;
7462 		pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7463 			 dev->name, adj_dev->name, adj->ref_nr);
7464 
7465 		return 0;
7466 	}
7467 
7468 	adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7469 	if (!adj)
7470 		return -ENOMEM;
7471 
7472 	adj->dev = adj_dev;
7473 	adj->master = master;
7474 	adj->ref_nr = 1;
7475 	adj->private = private;
7476 	adj->ignore = false;
7477 	netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7478 
7479 	pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7480 		 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7481 
7482 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7483 		ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7484 		if (ret)
7485 			goto free_adj;
7486 	}
7487 
7488 	/* Ensure that master link is always the first item in list. */
7489 	if (master) {
7490 		ret = sysfs_create_link(&(dev->dev.kobj),
7491 					&(adj_dev->dev.kobj), "master");
7492 		if (ret)
7493 			goto remove_symlinks;
7494 
7495 		list_add_rcu(&adj->list, dev_list);
7496 	} else {
7497 		list_add_tail_rcu(&adj->list, dev_list);
7498 	}
7499 
7500 	return 0;
7501 
7502 remove_symlinks:
7503 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7504 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7505 free_adj:
7506 	netdev_put(adj_dev, &adj->dev_tracker);
7507 	kfree(adj);
7508 
7509 	return ret;
7510 }
7511 
7512 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7513 					 struct net_device *adj_dev,
7514 					 u16 ref_nr,
7515 					 struct list_head *dev_list)
7516 {
7517 	struct netdev_adjacent *adj;
7518 
7519 	pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7520 		 dev->name, adj_dev->name, ref_nr);
7521 
7522 	adj = __netdev_find_adj(adj_dev, dev_list);
7523 
7524 	if (!adj) {
7525 		pr_err("Adjacency does not exist for device %s from %s\n",
7526 		       dev->name, adj_dev->name);
7527 		WARN_ON(1);
7528 		return;
7529 	}
7530 
7531 	if (adj->ref_nr > ref_nr) {
7532 		pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7533 			 dev->name, adj_dev->name, ref_nr,
7534 			 adj->ref_nr - ref_nr);
7535 		adj->ref_nr -= ref_nr;
7536 		return;
7537 	}
7538 
7539 	if (adj->master)
7540 		sysfs_remove_link(&(dev->dev.kobj), "master");
7541 
7542 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7543 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7544 
7545 	list_del_rcu(&adj->list);
7546 	pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7547 		 adj_dev->name, dev->name, adj_dev->name);
7548 	netdev_put(adj_dev, &adj->dev_tracker);
7549 	kfree_rcu(adj, rcu);
7550 }
7551 
7552 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7553 					    struct net_device *upper_dev,
7554 					    struct list_head *up_list,
7555 					    struct list_head *down_list,
7556 					    void *private, bool master)
7557 {
7558 	int ret;
7559 
7560 	ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7561 					   private, master);
7562 	if (ret)
7563 		return ret;
7564 
7565 	ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7566 					   private, false);
7567 	if (ret) {
7568 		__netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7569 		return ret;
7570 	}
7571 
7572 	return 0;
7573 }
7574 
7575 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7576 					       struct net_device *upper_dev,
7577 					       u16 ref_nr,
7578 					       struct list_head *up_list,
7579 					       struct list_head *down_list)
7580 {
7581 	__netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7582 	__netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7583 }
7584 
7585 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7586 						struct net_device *upper_dev,
7587 						void *private, bool master)
7588 {
7589 	return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7590 						&dev->adj_list.upper,
7591 						&upper_dev->adj_list.lower,
7592 						private, master);
7593 }
7594 
7595 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7596 						   struct net_device *upper_dev)
7597 {
7598 	__netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7599 					   &dev->adj_list.upper,
7600 					   &upper_dev->adj_list.lower);
7601 }
7602 
7603 static int __netdev_upper_dev_link(struct net_device *dev,
7604 				   struct net_device *upper_dev, bool master,
7605 				   void *upper_priv, void *upper_info,
7606 				   struct netdev_nested_priv *priv,
7607 				   struct netlink_ext_ack *extack)
7608 {
7609 	struct netdev_notifier_changeupper_info changeupper_info = {
7610 		.info = {
7611 			.dev = dev,
7612 			.extack = extack,
7613 		},
7614 		.upper_dev = upper_dev,
7615 		.master = master,
7616 		.linking = true,
7617 		.upper_info = upper_info,
7618 	};
7619 	struct net_device *master_dev;
7620 	int ret = 0;
7621 
7622 	ASSERT_RTNL();
7623 
7624 	if (dev == upper_dev)
7625 		return -EBUSY;
7626 
7627 	/* To prevent loops, check if dev is not upper device to upper_dev. */
7628 	if (__netdev_has_upper_dev(upper_dev, dev))
7629 		return -EBUSY;
7630 
7631 	if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7632 		return -EMLINK;
7633 
7634 	if (!master) {
7635 		if (__netdev_has_upper_dev(dev, upper_dev))
7636 			return -EEXIST;
7637 	} else {
7638 		master_dev = __netdev_master_upper_dev_get(dev);
7639 		if (master_dev)
7640 			return master_dev == upper_dev ? -EEXIST : -EBUSY;
7641 	}
7642 
7643 	ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7644 					    &changeupper_info.info);
7645 	ret = notifier_to_errno(ret);
7646 	if (ret)
7647 		return ret;
7648 
7649 	ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7650 						   master);
7651 	if (ret)
7652 		return ret;
7653 
7654 	ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7655 					    &changeupper_info.info);
7656 	ret = notifier_to_errno(ret);
7657 	if (ret)
7658 		goto rollback;
7659 
7660 	__netdev_update_upper_level(dev, NULL);
7661 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7662 
7663 	__netdev_update_lower_level(upper_dev, priv);
7664 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7665 				    priv);
7666 
7667 	return 0;
7668 
7669 rollback:
7670 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7671 
7672 	return ret;
7673 }
7674 
7675 /**
7676  * netdev_upper_dev_link - Add a link to the upper device
7677  * @dev: device
7678  * @upper_dev: new upper device
7679  * @extack: netlink extended ack
7680  *
7681  * Adds a link to device which is upper to this one. The caller must hold
7682  * the RTNL lock. On a failure a negative errno code is returned.
7683  * On success the reference counts are adjusted and the function
7684  * returns zero.
7685  */
7686 int netdev_upper_dev_link(struct net_device *dev,
7687 			  struct net_device *upper_dev,
7688 			  struct netlink_ext_ack *extack)
7689 {
7690 	struct netdev_nested_priv priv = {
7691 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7692 		.data = NULL,
7693 	};
7694 
7695 	return __netdev_upper_dev_link(dev, upper_dev, false,
7696 				       NULL, NULL, &priv, extack);
7697 }
7698 EXPORT_SYMBOL(netdev_upper_dev_link);
7699 
7700 /**
7701  * netdev_master_upper_dev_link - Add a master link to the upper device
7702  * @dev: device
7703  * @upper_dev: new upper device
7704  * @upper_priv: upper device private
7705  * @upper_info: upper info to be passed down via notifier
7706  * @extack: netlink extended ack
7707  *
7708  * Adds a link to device which is upper to this one. In this case, only
7709  * one master upper device can be linked, although other non-master devices
7710  * might be linked as well. The caller must hold the RTNL lock.
7711  * On a failure a negative errno code is returned. On success the reference
7712  * counts are adjusted and the function returns zero.
7713  */
7714 int netdev_master_upper_dev_link(struct net_device *dev,
7715 				 struct net_device *upper_dev,
7716 				 void *upper_priv, void *upper_info,
7717 				 struct netlink_ext_ack *extack)
7718 {
7719 	struct netdev_nested_priv priv = {
7720 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7721 		.data = NULL,
7722 	};
7723 
7724 	return __netdev_upper_dev_link(dev, upper_dev, true,
7725 				       upper_priv, upper_info, &priv, extack);
7726 }
7727 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7728 
7729 static void __netdev_upper_dev_unlink(struct net_device *dev,
7730 				      struct net_device *upper_dev,
7731 				      struct netdev_nested_priv *priv)
7732 {
7733 	struct netdev_notifier_changeupper_info changeupper_info = {
7734 		.info = {
7735 			.dev = dev,
7736 		},
7737 		.upper_dev = upper_dev,
7738 		.linking = false,
7739 	};
7740 
7741 	ASSERT_RTNL();
7742 
7743 	changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7744 
7745 	call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7746 				      &changeupper_info.info);
7747 
7748 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7749 
7750 	call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7751 				      &changeupper_info.info);
7752 
7753 	__netdev_update_upper_level(dev, NULL);
7754 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7755 
7756 	__netdev_update_lower_level(upper_dev, priv);
7757 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7758 				    priv);
7759 }
7760 
7761 /**
7762  * netdev_upper_dev_unlink - Removes a link to upper device
7763  * @dev: device
7764  * @upper_dev: new upper device
7765  *
7766  * Removes a link to device which is upper to this one. The caller must hold
7767  * the RTNL lock.
7768  */
7769 void netdev_upper_dev_unlink(struct net_device *dev,
7770 			     struct net_device *upper_dev)
7771 {
7772 	struct netdev_nested_priv priv = {
7773 		.flags = NESTED_SYNC_TODO,
7774 		.data = NULL,
7775 	};
7776 
7777 	__netdev_upper_dev_unlink(dev, upper_dev, &priv);
7778 }
7779 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7780 
7781 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7782 				      struct net_device *lower_dev,
7783 				      bool val)
7784 {
7785 	struct netdev_adjacent *adj;
7786 
7787 	adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7788 	if (adj)
7789 		adj->ignore = val;
7790 
7791 	adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7792 	if (adj)
7793 		adj->ignore = val;
7794 }
7795 
7796 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7797 					struct net_device *lower_dev)
7798 {
7799 	__netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7800 }
7801 
7802 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7803 				       struct net_device *lower_dev)
7804 {
7805 	__netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7806 }
7807 
7808 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7809 				   struct net_device *new_dev,
7810 				   struct net_device *dev,
7811 				   struct netlink_ext_ack *extack)
7812 {
7813 	struct netdev_nested_priv priv = {
7814 		.flags = 0,
7815 		.data = NULL,
7816 	};
7817 	int err;
7818 
7819 	if (!new_dev)
7820 		return 0;
7821 
7822 	if (old_dev && new_dev != old_dev)
7823 		netdev_adjacent_dev_disable(dev, old_dev);
7824 	err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
7825 				      extack);
7826 	if (err) {
7827 		if (old_dev && new_dev != old_dev)
7828 			netdev_adjacent_dev_enable(dev, old_dev);
7829 		return err;
7830 	}
7831 
7832 	return 0;
7833 }
7834 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7835 
7836 void netdev_adjacent_change_commit(struct net_device *old_dev,
7837 				   struct net_device *new_dev,
7838 				   struct net_device *dev)
7839 {
7840 	struct netdev_nested_priv priv = {
7841 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7842 		.data = NULL,
7843 	};
7844 
7845 	if (!new_dev || !old_dev)
7846 		return;
7847 
7848 	if (new_dev == old_dev)
7849 		return;
7850 
7851 	netdev_adjacent_dev_enable(dev, old_dev);
7852 	__netdev_upper_dev_unlink(old_dev, dev, &priv);
7853 }
7854 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7855 
7856 void netdev_adjacent_change_abort(struct net_device *old_dev,
7857 				  struct net_device *new_dev,
7858 				  struct net_device *dev)
7859 {
7860 	struct netdev_nested_priv priv = {
7861 		.flags = 0,
7862 		.data = NULL,
7863 	};
7864 
7865 	if (!new_dev)
7866 		return;
7867 
7868 	if (old_dev && new_dev != old_dev)
7869 		netdev_adjacent_dev_enable(dev, old_dev);
7870 
7871 	__netdev_upper_dev_unlink(new_dev, dev, &priv);
7872 }
7873 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7874 
7875 /**
7876  * netdev_bonding_info_change - Dispatch event about slave change
7877  * @dev: device
7878  * @bonding_info: info to dispatch
7879  *
7880  * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7881  * The caller must hold the RTNL lock.
7882  */
7883 void netdev_bonding_info_change(struct net_device *dev,
7884 				struct netdev_bonding_info *bonding_info)
7885 {
7886 	struct netdev_notifier_bonding_info info = {
7887 		.info.dev = dev,
7888 	};
7889 
7890 	memcpy(&info.bonding_info, bonding_info,
7891 	       sizeof(struct netdev_bonding_info));
7892 	call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7893 				      &info.info);
7894 }
7895 EXPORT_SYMBOL(netdev_bonding_info_change);
7896 
7897 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
7898 					   struct netlink_ext_ack *extack)
7899 {
7900 	struct netdev_notifier_offload_xstats_info info = {
7901 		.info.dev = dev,
7902 		.info.extack = extack,
7903 		.type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7904 	};
7905 	int err;
7906 	int rc;
7907 
7908 	dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
7909 					 GFP_KERNEL);
7910 	if (!dev->offload_xstats_l3)
7911 		return -ENOMEM;
7912 
7913 	rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
7914 						  NETDEV_OFFLOAD_XSTATS_DISABLE,
7915 						  &info.info);
7916 	err = notifier_to_errno(rc);
7917 	if (err)
7918 		goto free_stats;
7919 
7920 	return 0;
7921 
7922 free_stats:
7923 	kfree(dev->offload_xstats_l3);
7924 	dev->offload_xstats_l3 = NULL;
7925 	return err;
7926 }
7927 
7928 int netdev_offload_xstats_enable(struct net_device *dev,
7929 				 enum netdev_offload_xstats_type type,
7930 				 struct netlink_ext_ack *extack)
7931 {
7932 	ASSERT_RTNL();
7933 
7934 	if (netdev_offload_xstats_enabled(dev, type))
7935 		return -EALREADY;
7936 
7937 	switch (type) {
7938 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7939 		return netdev_offload_xstats_enable_l3(dev, extack);
7940 	}
7941 
7942 	WARN_ON(1);
7943 	return -EINVAL;
7944 }
7945 EXPORT_SYMBOL(netdev_offload_xstats_enable);
7946 
7947 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
7948 {
7949 	struct netdev_notifier_offload_xstats_info info = {
7950 		.info.dev = dev,
7951 		.type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7952 	};
7953 
7954 	call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
7955 				      &info.info);
7956 	kfree(dev->offload_xstats_l3);
7957 	dev->offload_xstats_l3 = NULL;
7958 }
7959 
7960 int netdev_offload_xstats_disable(struct net_device *dev,
7961 				  enum netdev_offload_xstats_type type)
7962 {
7963 	ASSERT_RTNL();
7964 
7965 	if (!netdev_offload_xstats_enabled(dev, type))
7966 		return -EALREADY;
7967 
7968 	switch (type) {
7969 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7970 		netdev_offload_xstats_disable_l3(dev);
7971 		return 0;
7972 	}
7973 
7974 	WARN_ON(1);
7975 	return -EINVAL;
7976 }
7977 EXPORT_SYMBOL(netdev_offload_xstats_disable);
7978 
7979 static void netdev_offload_xstats_disable_all(struct net_device *dev)
7980 {
7981 	netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
7982 }
7983 
7984 static struct rtnl_hw_stats64 *
7985 netdev_offload_xstats_get_ptr(const struct net_device *dev,
7986 			      enum netdev_offload_xstats_type type)
7987 {
7988 	switch (type) {
7989 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7990 		return dev->offload_xstats_l3;
7991 	}
7992 
7993 	WARN_ON(1);
7994 	return NULL;
7995 }
7996 
7997 bool netdev_offload_xstats_enabled(const struct net_device *dev,
7998 				   enum netdev_offload_xstats_type type)
7999 {
8000 	ASSERT_RTNL();
8001 
8002 	return netdev_offload_xstats_get_ptr(dev, type);
8003 }
8004 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
8005 
8006 struct netdev_notifier_offload_xstats_ru {
8007 	bool used;
8008 };
8009 
8010 struct netdev_notifier_offload_xstats_rd {
8011 	struct rtnl_hw_stats64 stats;
8012 	bool used;
8013 };
8014 
8015 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
8016 				  const struct rtnl_hw_stats64 *src)
8017 {
8018 	dest->rx_packets	  += src->rx_packets;
8019 	dest->tx_packets	  += src->tx_packets;
8020 	dest->rx_bytes		  += src->rx_bytes;
8021 	dest->tx_bytes		  += src->tx_bytes;
8022 	dest->rx_errors		  += src->rx_errors;
8023 	dest->tx_errors		  += src->tx_errors;
8024 	dest->rx_dropped	  += src->rx_dropped;
8025 	dest->tx_dropped	  += src->tx_dropped;
8026 	dest->multicast		  += src->multicast;
8027 }
8028 
8029 static int netdev_offload_xstats_get_used(struct net_device *dev,
8030 					  enum netdev_offload_xstats_type type,
8031 					  bool *p_used,
8032 					  struct netlink_ext_ack *extack)
8033 {
8034 	struct netdev_notifier_offload_xstats_ru report_used = {};
8035 	struct netdev_notifier_offload_xstats_info info = {
8036 		.info.dev = dev,
8037 		.info.extack = extack,
8038 		.type = type,
8039 		.report_used = &report_used,
8040 	};
8041 	int rc;
8042 
8043 	WARN_ON(!netdev_offload_xstats_enabled(dev, type));
8044 	rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
8045 					   &info.info);
8046 	*p_used = report_used.used;
8047 	return notifier_to_errno(rc);
8048 }
8049 
8050 static int netdev_offload_xstats_get_stats(struct net_device *dev,
8051 					   enum netdev_offload_xstats_type type,
8052 					   struct rtnl_hw_stats64 *p_stats,
8053 					   bool *p_used,
8054 					   struct netlink_ext_ack *extack)
8055 {
8056 	struct netdev_notifier_offload_xstats_rd report_delta = {};
8057 	struct netdev_notifier_offload_xstats_info info = {
8058 		.info.dev = dev,
8059 		.info.extack = extack,
8060 		.type = type,
8061 		.report_delta = &report_delta,
8062 	};
8063 	struct rtnl_hw_stats64 *stats;
8064 	int rc;
8065 
8066 	stats = netdev_offload_xstats_get_ptr(dev, type);
8067 	if (WARN_ON(!stats))
8068 		return -EINVAL;
8069 
8070 	rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
8071 					   &info.info);
8072 
8073 	/* Cache whatever we got, even if there was an error, otherwise the
8074 	 * successful stats retrievals would get lost.
8075 	 */
8076 	netdev_hw_stats64_add(stats, &report_delta.stats);
8077 
8078 	if (p_stats)
8079 		*p_stats = *stats;
8080 	*p_used = report_delta.used;
8081 
8082 	return notifier_to_errno(rc);
8083 }
8084 
8085 int netdev_offload_xstats_get(struct net_device *dev,
8086 			      enum netdev_offload_xstats_type type,
8087 			      struct rtnl_hw_stats64 *p_stats, bool *p_used,
8088 			      struct netlink_ext_ack *extack)
8089 {
8090 	ASSERT_RTNL();
8091 
8092 	if (p_stats)
8093 		return netdev_offload_xstats_get_stats(dev, type, p_stats,
8094 						       p_used, extack);
8095 	else
8096 		return netdev_offload_xstats_get_used(dev, type, p_used,
8097 						      extack);
8098 }
8099 EXPORT_SYMBOL(netdev_offload_xstats_get);
8100 
8101 void
8102 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
8103 				   const struct rtnl_hw_stats64 *stats)
8104 {
8105 	report_delta->used = true;
8106 	netdev_hw_stats64_add(&report_delta->stats, stats);
8107 }
8108 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
8109 
8110 void
8111 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
8112 {
8113 	report_used->used = true;
8114 }
8115 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
8116 
8117 void netdev_offload_xstats_push_delta(struct net_device *dev,
8118 				      enum netdev_offload_xstats_type type,
8119 				      const struct rtnl_hw_stats64 *p_stats)
8120 {
8121 	struct rtnl_hw_stats64 *stats;
8122 
8123 	ASSERT_RTNL();
8124 
8125 	stats = netdev_offload_xstats_get_ptr(dev, type);
8126 	if (WARN_ON(!stats))
8127 		return;
8128 
8129 	netdev_hw_stats64_add(stats, p_stats);
8130 }
8131 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
8132 
8133 /**
8134  * netdev_get_xmit_slave - Get the xmit slave of master device
8135  * @dev: device
8136  * @skb: The packet
8137  * @all_slaves: assume all the slaves are active
8138  *
8139  * The reference counters are not incremented so the caller must be
8140  * careful with locks. The caller must hold RCU lock.
8141  * %NULL is returned if no slave is found.
8142  */
8143 
8144 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8145 					 struct sk_buff *skb,
8146 					 bool all_slaves)
8147 {
8148 	const struct net_device_ops *ops = dev->netdev_ops;
8149 
8150 	if (!ops->ndo_get_xmit_slave)
8151 		return NULL;
8152 	return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8153 }
8154 EXPORT_SYMBOL(netdev_get_xmit_slave);
8155 
8156 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8157 						  struct sock *sk)
8158 {
8159 	const struct net_device_ops *ops = dev->netdev_ops;
8160 
8161 	if (!ops->ndo_sk_get_lower_dev)
8162 		return NULL;
8163 	return ops->ndo_sk_get_lower_dev(dev, sk);
8164 }
8165 
8166 /**
8167  * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8168  * @dev: device
8169  * @sk: the socket
8170  *
8171  * %NULL is returned if no lower device is found.
8172  */
8173 
8174 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8175 					    struct sock *sk)
8176 {
8177 	struct net_device *lower;
8178 
8179 	lower = netdev_sk_get_lower_dev(dev, sk);
8180 	while (lower) {
8181 		dev = lower;
8182 		lower = netdev_sk_get_lower_dev(dev, sk);
8183 	}
8184 
8185 	return dev;
8186 }
8187 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8188 
8189 static void netdev_adjacent_add_links(struct net_device *dev)
8190 {
8191 	struct netdev_adjacent *iter;
8192 
8193 	struct net *net = dev_net(dev);
8194 
8195 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8196 		if (!net_eq(net, dev_net(iter->dev)))
8197 			continue;
8198 		netdev_adjacent_sysfs_add(iter->dev, dev,
8199 					  &iter->dev->adj_list.lower);
8200 		netdev_adjacent_sysfs_add(dev, iter->dev,
8201 					  &dev->adj_list.upper);
8202 	}
8203 
8204 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8205 		if (!net_eq(net, dev_net(iter->dev)))
8206 			continue;
8207 		netdev_adjacent_sysfs_add(iter->dev, dev,
8208 					  &iter->dev->adj_list.upper);
8209 		netdev_adjacent_sysfs_add(dev, iter->dev,
8210 					  &dev->adj_list.lower);
8211 	}
8212 }
8213 
8214 static void netdev_adjacent_del_links(struct net_device *dev)
8215 {
8216 	struct netdev_adjacent *iter;
8217 
8218 	struct net *net = dev_net(dev);
8219 
8220 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8221 		if (!net_eq(net, dev_net(iter->dev)))
8222 			continue;
8223 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8224 					  &iter->dev->adj_list.lower);
8225 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8226 					  &dev->adj_list.upper);
8227 	}
8228 
8229 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8230 		if (!net_eq(net, dev_net(iter->dev)))
8231 			continue;
8232 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8233 					  &iter->dev->adj_list.upper);
8234 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8235 					  &dev->adj_list.lower);
8236 	}
8237 }
8238 
8239 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8240 {
8241 	struct netdev_adjacent *iter;
8242 
8243 	struct net *net = dev_net(dev);
8244 
8245 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8246 		if (!net_eq(net, dev_net(iter->dev)))
8247 			continue;
8248 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8249 					  &iter->dev->adj_list.lower);
8250 		netdev_adjacent_sysfs_add(iter->dev, dev,
8251 					  &iter->dev->adj_list.lower);
8252 	}
8253 
8254 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8255 		if (!net_eq(net, dev_net(iter->dev)))
8256 			continue;
8257 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8258 					  &iter->dev->adj_list.upper);
8259 		netdev_adjacent_sysfs_add(iter->dev, dev,
8260 					  &iter->dev->adj_list.upper);
8261 	}
8262 }
8263 
8264 void *netdev_lower_dev_get_private(struct net_device *dev,
8265 				   struct net_device *lower_dev)
8266 {
8267 	struct netdev_adjacent *lower;
8268 
8269 	if (!lower_dev)
8270 		return NULL;
8271 	lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8272 	if (!lower)
8273 		return NULL;
8274 
8275 	return lower->private;
8276 }
8277 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8278 
8279 
8280 /**
8281  * netdev_lower_state_changed - Dispatch event about lower device state change
8282  * @lower_dev: device
8283  * @lower_state_info: state to dispatch
8284  *
8285  * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8286  * The caller must hold the RTNL lock.
8287  */
8288 void netdev_lower_state_changed(struct net_device *lower_dev,
8289 				void *lower_state_info)
8290 {
8291 	struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8292 		.info.dev = lower_dev,
8293 	};
8294 
8295 	ASSERT_RTNL();
8296 	changelowerstate_info.lower_state_info = lower_state_info;
8297 	call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8298 				      &changelowerstate_info.info);
8299 }
8300 EXPORT_SYMBOL(netdev_lower_state_changed);
8301 
8302 static void dev_change_rx_flags(struct net_device *dev, int flags)
8303 {
8304 	const struct net_device_ops *ops = dev->netdev_ops;
8305 
8306 	if (ops->ndo_change_rx_flags)
8307 		ops->ndo_change_rx_flags(dev, flags);
8308 }
8309 
8310 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8311 {
8312 	unsigned int old_flags = dev->flags;
8313 	kuid_t uid;
8314 	kgid_t gid;
8315 
8316 	ASSERT_RTNL();
8317 
8318 	dev->flags |= IFF_PROMISC;
8319 	dev->promiscuity += inc;
8320 	if (dev->promiscuity == 0) {
8321 		/*
8322 		 * Avoid overflow.
8323 		 * If inc causes overflow, untouch promisc and return error.
8324 		 */
8325 		if (inc < 0)
8326 			dev->flags &= ~IFF_PROMISC;
8327 		else {
8328 			dev->promiscuity -= inc;
8329 			netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8330 			return -EOVERFLOW;
8331 		}
8332 	}
8333 	if (dev->flags != old_flags) {
8334 		pr_info("device %s %s promiscuous mode\n",
8335 			dev->name,
8336 			dev->flags & IFF_PROMISC ? "entered" : "left");
8337 		if (audit_enabled) {
8338 			current_uid_gid(&uid, &gid);
8339 			audit_log(audit_context(), GFP_ATOMIC,
8340 				  AUDIT_ANOM_PROMISCUOUS,
8341 				  "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8342 				  dev->name, (dev->flags & IFF_PROMISC),
8343 				  (old_flags & IFF_PROMISC),
8344 				  from_kuid(&init_user_ns, audit_get_loginuid(current)),
8345 				  from_kuid(&init_user_ns, uid),
8346 				  from_kgid(&init_user_ns, gid),
8347 				  audit_get_sessionid(current));
8348 		}
8349 
8350 		dev_change_rx_flags(dev, IFF_PROMISC);
8351 	}
8352 	if (notify)
8353 		__dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL);
8354 	return 0;
8355 }
8356 
8357 /**
8358  *	dev_set_promiscuity	- update promiscuity count on a device
8359  *	@dev: device
8360  *	@inc: modifier
8361  *
8362  *	Add or remove promiscuity from a device. While the count in the device
8363  *	remains above zero the interface remains promiscuous. Once it hits zero
8364  *	the device reverts back to normal filtering operation. A negative inc
8365  *	value is used to drop promiscuity on the device.
8366  *	Return 0 if successful or a negative errno code on error.
8367  */
8368 int dev_set_promiscuity(struct net_device *dev, int inc)
8369 {
8370 	unsigned int old_flags = dev->flags;
8371 	int err;
8372 
8373 	err = __dev_set_promiscuity(dev, inc, true);
8374 	if (err < 0)
8375 		return err;
8376 	if (dev->flags != old_flags)
8377 		dev_set_rx_mode(dev);
8378 	return err;
8379 }
8380 EXPORT_SYMBOL(dev_set_promiscuity);
8381 
8382 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8383 {
8384 	unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8385 
8386 	ASSERT_RTNL();
8387 
8388 	dev->flags |= IFF_ALLMULTI;
8389 	dev->allmulti += inc;
8390 	if (dev->allmulti == 0) {
8391 		/*
8392 		 * Avoid overflow.
8393 		 * If inc causes overflow, untouch allmulti and return error.
8394 		 */
8395 		if (inc < 0)
8396 			dev->flags &= ~IFF_ALLMULTI;
8397 		else {
8398 			dev->allmulti -= inc;
8399 			netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8400 			return -EOVERFLOW;
8401 		}
8402 	}
8403 	if (dev->flags ^ old_flags) {
8404 		dev_change_rx_flags(dev, IFF_ALLMULTI);
8405 		dev_set_rx_mode(dev);
8406 		if (notify)
8407 			__dev_notify_flags(dev, old_flags,
8408 					   dev->gflags ^ old_gflags, 0, NULL);
8409 	}
8410 	return 0;
8411 }
8412 
8413 /**
8414  *	dev_set_allmulti	- update allmulti count on a device
8415  *	@dev: device
8416  *	@inc: modifier
8417  *
8418  *	Add or remove reception of all multicast frames to a device. While the
8419  *	count in the device remains above zero the interface remains listening
8420  *	to all interfaces. Once it hits zero the device reverts back to normal
8421  *	filtering operation. A negative @inc value is used to drop the counter
8422  *	when releasing a resource needing all multicasts.
8423  *	Return 0 if successful or a negative errno code on error.
8424  */
8425 
8426 int dev_set_allmulti(struct net_device *dev, int inc)
8427 {
8428 	return __dev_set_allmulti(dev, inc, true);
8429 }
8430 EXPORT_SYMBOL(dev_set_allmulti);
8431 
8432 /*
8433  *	Upload unicast and multicast address lists to device and
8434  *	configure RX filtering. When the device doesn't support unicast
8435  *	filtering it is put in promiscuous mode while unicast addresses
8436  *	are present.
8437  */
8438 void __dev_set_rx_mode(struct net_device *dev)
8439 {
8440 	const struct net_device_ops *ops = dev->netdev_ops;
8441 
8442 	/* dev_open will call this function so the list will stay sane. */
8443 	if (!(dev->flags&IFF_UP))
8444 		return;
8445 
8446 	if (!netif_device_present(dev))
8447 		return;
8448 
8449 	if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8450 		/* Unicast addresses changes may only happen under the rtnl,
8451 		 * therefore calling __dev_set_promiscuity here is safe.
8452 		 */
8453 		if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8454 			__dev_set_promiscuity(dev, 1, false);
8455 			dev->uc_promisc = true;
8456 		} else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8457 			__dev_set_promiscuity(dev, -1, false);
8458 			dev->uc_promisc = false;
8459 		}
8460 	}
8461 
8462 	if (ops->ndo_set_rx_mode)
8463 		ops->ndo_set_rx_mode(dev);
8464 }
8465 
8466 void dev_set_rx_mode(struct net_device *dev)
8467 {
8468 	netif_addr_lock_bh(dev);
8469 	__dev_set_rx_mode(dev);
8470 	netif_addr_unlock_bh(dev);
8471 }
8472 
8473 /**
8474  *	dev_get_flags - get flags reported to userspace
8475  *	@dev: device
8476  *
8477  *	Get the combination of flag bits exported through APIs to userspace.
8478  */
8479 unsigned int dev_get_flags(const struct net_device *dev)
8480 {
8481 	unsigned int flags;
8482 
8483 	flags = (dev->flags & ~(IFF_PROMISC |
8484 				IFF_ALLMULTI |
8485 				IFF_RUNNING |
8486 				IFF_LOWER_UP |
8487 				IFF_DORMANT)) |
8488 		(dev->gflags & (IFF_PROMISC |
8489 				IFF_ALLMULTI));
8490 
8491 	if (netif_running(dev)) {
8492 		if (netif_oper_up(dev))
8493 			flags |= IFF_RUNNING;
8494 		if (netif_carrier_ok(dev))
8495 			flags |= IFF_LOWER_UP;
8496 		if (netif_dormant(dev))
8497 			flags |= IFF_DORMANT;
8498 	}
8499 
8500 	return flags;
8501 }
8502 EXPORT_SYMBOL(dev_get_flags);
8503 
8504 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8505 		       struct netlink_ext_ack *extack)
8506 {
8507 	unsigned int old_flags = dev->flags;
8508 	int ret;
8509 
8510 	ASSERT_RTNL();
8511 
8512 	/*
8513 	 *	Set the flags on our device.
8514 	 */
8515 
8516 	dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8517 			       IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8518 			       IFF_AUTOMEDIA)) |
8519 		     (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8520 				    IFF_ALLMULTI));
8521 
8522 	/*
8523 	 *	Load in the correct multicast list now the flags have changed.
8524 	 */
8525 
8526 	if ((old_flags ^ flags) & IFF_MULTICAST)
8527 		dev_change_rx_flags(dev, IFF_MULTICAST);
8528 
8529 	dev_set_rx_mode(dev);
8530 
8531 	/*
8532 	 *	Have we downed the interface. We handle IFF_UP ourselves
8533 	 *	according to user attempts to set it, rather than blindly
8534 	 *	setting it.
8535 	 */
8536 
8537 	ret = 0;
8538 	if ((old_flags ^ flags) & IFF_UP) {
8539 		if (old_flags & IFF_UP)
8540 			__dev_close(dev);
8541 		else
8542 			ret = __dev_open(dev, extack);
8543 	}
8544 
8545 	if ((flags ^ dev->gflags) & IFF_PROMISC) {
8546 		int inc = (flags & IFF_PROMISC) ? 1 : -1;
8547 		unsigned int old_flags = dev->flags;
8548 
8549 		dev->gflags ^= IFF_PROMISC;
8550 
8551 		if (__dev_set_promiscuity(dev, inc, false) >= 0)
8552 			if (dev->flags != old_flags)
8553 				dev_set_rx_mode(dev);
8554 	}
8555 
8556 	/* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8557 	 * is important. Some (broken) drivers set IFF_PROMISC, when
8558 	 * IFF_ALLMULTI is requested not asking us and not reporting.
8559 	 */
8560 	if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8561 		int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8562 
8563 		dev->gflags ^= IFF_ALLMULTI;
8564 		__dev_set_allmulti(dev, inc, false);
8565 	}
8566 
8567 	return ret;
8568 }
8569 
8570 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8571 			unsigned int gchanges, u32 portid,
8572 			const struct nlmsghdr *nlh)
8573 {
8574 	unsigned int changes = dev->flags ^ old_flags;
8575 
8576 	if (gchanges)
8577 		rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh);
8578 
8579 	if (changes & IFF_UP) {
8580 		if (dev->flags & IFF_UP)
8581 			call_netdevice_notifiers(NETDEV_UP, dev);
8582 		else
8583 			call_netdevice_notifiers(NETDEV_DOWN, dev);
8584 	}
8585 
8586 	if (dev->flags & IFF_UP &&
8587 	    (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8588 		struct netdev_notifier_change_info change_info = {
8589 			.info = {
8590 				.dev = dev,
8591 			},
8592 			.flags_changed = changes,
8593 		};
8594 
8595 		call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8596 	}
8597 }
8598 
8599 /**
8600  *	dev_change_flags - change device settings
8601  *	@dev: device
8602  *	@flags: device state flags
8603  *	@extack: netlink extended ack
8604  *
8605  *	Change settings on device based state flags. The flags are
8606  *	in the userspace exported format.
8607  */
8608 int dev_change_flags(struct net_device *dev, unsigned int flags,
8609 		     struct netlink_ext_ack *extack)
8610 {
8611 	int ret;
8612 	unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8613 
8614 	ret = __dev_change_flags(dev, flags, extack);
8615 	if (ret < 0)
8616 		return ret;
8617 
8618 	changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8619 	__dev_notify_flags(dev, old_flags, changes, 0, NULL);
8620 	return ret;
8621 }
8622 EXPORT_SYMBOL(dev_change_flags);
8623 
8624 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8625 {
8626 	const struct net_device_ops *ops = dev->netdev_ops;
8627 
8628 	if (ops->ndo_change_mtu)
8629 		return ops->ndo_change_mtu(dev, new_mtu);
8630 
8631 	/* Pairs with all the lockless reads of dev->mtu in the stack */
8632 	WRITE_ONCE(dev->mtu, new_mtu);
8633 	return 0;
8634 }
8635 EXPORT_SYMBOL(__dev_set_mtu);
8636 
8637 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8638 		     struct netlink_ext_ack *extack)
8639 {
8640 	/* MTU must be positive, and in range */
8641 	if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8642 		NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8643 		return -EINVAL;
8644 	}
8645 
8646 	if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8647 		NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8648 		return -EINVAL;
8649 	}
8650 	return 0;
8651 }
8652 
8653 /**
8654  *	dev_set_mtu_ext - Change maximum transfer unit
8655  *	@dev: device
8656  *	@new_mtu: new transfer unit
8657  *	@extack: netlink extended ack
8658  *
8659  *	Change the maximum transfer size of the network device.
8660  */
8661 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8662 		    struct netlink_ext_ack *extack)
8663 {
8664 	int err, orig_mtu;
8665 
8666 	if (new_mtu == dev->mtu)
8667 		return 0;
8668 
8669 	err = dev_validate_mtu(dev, new_mtu, extack);
8670 	if (err)
8671 		return err;
8672 
8673 	if (!netif_device_present(dev))
8674 		return -ENODEV;
8675 
8676 	err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8677 	err = notifier_to_errno(err);
8678 	if (err)
8679 		return err;
8680 
8681 	orig_mtu = dev->mtu;
8682 	err = __dev_set_mtu(dev, new_mtu);
8683 
8684 	if (!err) {
8685 		err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8686 						   orig_mtu);
8687 		err = notifier_to_errno(err);
8688 		if (err) {
8689 			/* setting mtu back and notifying everyone again,
8690 			 * so that they have a chance to revert changes.
8691 			 */
8692 			__dev_set_mtu(dev, orig_mtu);
8693 			call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8694 						     new_mtu);
8695 		}
8696 	}
8697 	return err;
8698 }
8699 
8700 int dev_set_mtu(struct net_device *dev, int new_mtu)
8701 {
8702 	struct netlink_ext_ack extack;
8703 	int err;
8704 
8705 	memset(&extack, 0, sizeof(extack));
8706 	err = dev_set_mtu_ext(dev, new_mtu, &extack);
8707 	if (err && extack._msg)
8708 		net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8709 	return err;
8710 }
8711 EXPORT_SYMBOL(dev_set_mtu);
8712 
8713 /**
8714  *	dev_change_tx_queue_len - Change TX queue length of a netdevice
8715  *	@dev: device
8716  *	@new_len: new tx queue length
8717  */
8718 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8719 {
8720 	unsigned int orig_len = dev->tx_queue_len;
8721 	int res;
8722 
8723 	if (new_len != (unsigned int)new_len)
8724 		return -ERANGE;
8725 
8726 	if (new_len != orig_len) {
8727 		dev->tx_queue_len = new_len;
8728 		res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8729 		res = notifier_to_errno(res);
8730 		if (res)
8731 			goto err_rollback;
8732 		res = dev_qdisc_change_tx_queue_len(dev);
8733 		if (res)
8734 			goto err_rollback;
8735 	}
8736 
8737 	return 0;
8738 
8739 err_rollback:
8740 	netdev_err(dev, "refused to change device tx_queue_len\n");
8741 	dev->tx_queue_len = orig_len;
8742 	return res;
8743 }
8744 
8745 /**
8746  *	dev_set_group - Change group this device belongs to
8747  *	@dev: device
8748  *	@new_group: group this device should belong to
8749  */
8750 void dev_set_group(struct net_device *dev, int new_group)
8751 {
8752 	dev->group = new_group;
8753 }
8754 
8755 /**
8756  *	dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8757  *	@dev: device
8758  *	@addr: new address
8759  *	@extack: netlink extended ack
8760  */
8761 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8762 			      struct netlink_ext_ack *extack)
8763 {
8764 	struct netdev_notifier_pre_changeaddr_info info = {
8765 		.info.dev = dev,
8766 		.info.extack = extack,
8767 		.dev_addr = addr,
8768 	};
8769 	int rc;
8770 
8771 	rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8772 	return notifier_to_errno(rc);
8773 }
8774 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8775 
8776 /**
8777  *	dev_set_mac_address - Change Media Access Control Address
8778  *	@dev: device
8779  *	@sa: new address
8780  *	@extack: netlink extended ack
8781  *
8782  *	Change the hardware (MAC) address of the device
8783  */
8784 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8785 			struct netlink_ext_ack *extack)
8786 {
8787 	const struct net_device_ops *ops = dev->netdev_ops;
8788 	int err;
8789 
8790 	if (!ops->ndo_set_mac_address)
8791 		return -EOPNOTSUPP;
8792 	if (sa->sa_family != dev->type)
8793 		return -EINVAL;
8794 	if (!netif_device_present(dev))
8795 		return -ENODEV;
8796 	err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8797 	if (err)
8798 		return err;
8799 	err = ops->ndo_set_mac_address(dev, sa);
8800 	if (err)
8801 		return err;
8802 	dev->addr_assign_type = NET_ADDR_SET;
8803 	call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8804 	add_device_randomness(dev->dev_addr, dev->addr_len);
8805 	return 0;
8806 }
8807 EXPORT_SYMBOL(dev_set_mac_address);
8808 
8809 static DECLARE_RWSEM(dev_addr_sem);
8810 
8811 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
8812 			     struct netlink_ext_ack *extack)
8813 {
8814 	int ret;
8815 
8816 	down_write(&dev_addr_sem);
8817 	ret = dev_set_mac_address(dev, sa, extack);
8818 	up_write(&dev_addr_sem);
8819 	return ret;
8820 }
8821 EXPORT_SYMBOL(dev_set_mac_address_user);
8822 
8823 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
8824 {
8825 	size_t size = sizeof(sa->sa_data_min);
8826 	struct net_device *dev;
8827 	int ret = 0;
8828 
8829 	down_read(&dev_addr_sem);
8830 	rcu_read_lock();
8831 
8832 	dev = dev_get_by_name_rcu(net, dev_name);
8833 	if (!dev) {
8834 		ret = -ENODEV;
8835 		goto unlock;
8836 	}
8837 	if (!dev->addr_len)
8838 		memset(sa->sa_data, 0, size);
8839 	else
8840 		memcpy(sa->sa_data, dev->dev_addr,
8841 		       min_t(size_t, size, dev->addr_len));
8842 	sa->sa_family = dev->type;
8843 
8844 unlock:
8845 	rcu_read_unlock();
8846 	up_read(&dev_addr_sem);
8847 	return ret;
8848 }
8849 EXPORT_SYMBOL(dev_get_mac_address);
8850 
8851 /**
8852  *	dev_change_carrier - Change device carrier
8853  *	@dev: device
8854  *	@new_carrier: new value
8855  *
8856  *	Change device carrier
8857  */
8858 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8859 {
8860 	const struct net_device_ops *ops = dev->netdev_ops;
8861 
8862 	if (!ops->ndo_change_carrier)
8863 		return -EOPNOTSUPP;
8864 	if (!netif_device_present(dev))
8865 		return -ENODEV;
8866 	return ops->ndo_change_carrier(dev, new_carrier);
8867 }
8868 
8869 /**
8870  *	dev_get_phys_port_id - Get device physical port ID
8871  *	@dev: device
8872  *	@ppid: port ID
8873  *
8874  *	Get device physical port ID
8875  */
8876 int dev_get_phys_port_id(struct net_device *dev,
8877 			 struct netdev_phys_item_id *ppid)
8878 {
8879 	const struct net_device_ops *ops = dev->netdev_ops;
8880 
8881 	if (!ops->ndo_get_phys_port_id)
8882 		return -EOPNOTSUPP;
8883 	return ops->ndo_get_phys_port_id(dev, ppid);
8884 }
8885 
8886 /**
8887  *	dev_get_phys_port_name - Get device physical port name
8888  *	@dev: device
8889  *	@name: port name
8890  *	@len: limit of bytes to copy to name
8891  *
8892  *	Get device physical port name
8893  */
8894 int dev_get_phys_port_name(struct net_device *dev,
8895 			   char *name, size_t len)
8896 {
8897 	const struct net_device_ops *ops = dev->netdev_ops;
8898 	int err;
8899 
8900 	if (ops->ndo_get_phys_port_name) {
8901 		err = ops->ndo_get_phys_port_name(dev, name, len);
8902 		if (err != -EOPNOTSUPP)
8903 			return err;
8904 	}
8905 	return devlink_compat_phys_port_name_get(dev, name, len);
8906 }
8907 
8908 /**
8909  *	dev_get_port_parent_id - Get the device's port parent identifier
8910  *	@dev: network device
8911  *	@ppid: pointer to a storage for the port's parent identifier
8912  *	@recurse: allow/disallow recursion to lower devices
8913  *
8914  *	Get the devices's port parent identifier
8915  */
8916 int dev_get_port_parent_id(struct net_device *dev,
8917 			   struct netdev_phys_item_id *ppid,
8918 			   bool recurse)
8919 {
8920 	const struct net_device_ops *ops = dev->netdev_ops;
8921 	struct netdev_phys_item_id first = { };
8922 	struct net_device *lower_dev;
8923 	struct list_head *iter;
8924 	int err;
8925 
8926 	if (ops->ndo_get_port_parent_id) {
8927 		err = ops->ndo_get_port_parent_id(dev, ppid);
8928 		if (err != -EOPNOTSUPP)
8929 			return err;
8930 	}
8931 
8932 	err = devlink_compat_switch_id_get(dev, ppid);
8933 	if (!recurse || err != -EOPNOTSUPP)
8934 		return err;
8935 
8936 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
8937 		err = dev_get_port_parent_id(lower_dev, ppid, true);
8938 		if (err)
8939 			break;
8940 		if (!first.id_len)
8941 			first = *ppid;
8942 		else if (memcmp(&first, ppid, sizeof(*ppid)))
8943 			return -EOPNOTSUPP;
8944 	}
8945 
8946 	return err;
8947 }
8948 EXPORT_SYMBOL(dev_get_port_parent_id);
8949 
8950 /**
8951  *	netdev_port_same_parent_id - Indicate if two network devices have
8952  *	the same port parent identifier
8953  *	@a: first network device
8954  *	@b: second network device
8955  */
8956 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8957 {
8958 	struct netdev_phys_item_id a_id = { };
8959 	struct netdev_phys_item_id b_id = { };
8960 
8961 	if (dev_get_port_parent_id(a, &a_id, true) ||
8962 	    dev_get_port_parent_id(b, &b_id, true))
8963 		return false;
8964 
8965 	return netdev_phys_item_id_same(&a_id, &b_id);
8966 }
8967 EXPORT_SYMBOL(netdev_port_same_parent_id);
8968 
8969 /**
8970  *	dev_change_proto_down - set carrier according to proto_down.
8971  *
8972  *	@dev: device
8973  *	@proto_down: new value
8974  */
8975 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8976 {
8977 	if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN))
8978 		return -EOPNOTSUPP;
8979 	if (!netif_device_present(dev))
8980 		return -ENODEV;
8981 	if (proto_down)
8982 		netif_carrier_off(dev);
8983 	else
8984 		netif_carrier_on(dev);
8985 	dev->proto_down = proto_down;
8986 	return 0;
8987 }
8988 
8989 /**
8990  *	dev_change_proto_down_reason - proto down reason
8991  *
8992  *	@dev: device
8993  *	@mask: proto down mask
8994  *	@value: proto down value
8995  */
8996 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
8997 				  u32 value)
8998 {
8999 	int b;
9000 
9001 	if (!mask) {
9002 		dev->proto_down_reason = value;
9003 	} else {
9004 		for_each_set_bit(b, &mask, 32) {
9005 			if (value & (1 << b))
9006 				dev->proto_down_reason |= BIT(b);
9007 			else
9008 				dev->proto_down_reason &= ~BIT(b);
9009 		}
9010 	}
9011 }
9012 
9013 struct bpf_xdp_link {
9014 	struct bpf_link link;
9015 	struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9016 	int flags;
9017 };
9018 
9019 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9020 {
9021 	if (flags & XDP_FLAGS_HW_MODE)
9022 		return XDP_MODE_HW;
9023 	if (flags & XDP_FLAGS_DRV_MODE)
9024 		return XDP_MODE_DRV;
9025 	if (flags & XDP_FLAGS_SKB_MODE)
9026 		return XDP_MODE_SKB;
9027 	return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9028 }
9029 
9030 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9031 {
9032 	switch (mode) {
9033 	case XDP_MODE_SKB:
9034 		return generic_xdp_install;
9035 	case XDP_MODE_DRV:
9036 	case XDP_MODE_HW:
9037 		return dev->netdev_ops->ndo_bpf;
9038 	default:
9039 		return NULL;
9040 	}
9041 }
9042 
9043 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9044 					 enum bpf_xdp_mode mode)
9045 {
9046 	return dev->xdp_state[mode].link;
9047 }
9048 
9049 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9050 				     enum bpf_xdp_mode mode)
9051 {
9052 	struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9053 
9054 	if (link)
9055 		return link->link.prog;
9056 	return dev->xdp_state[mode].prog;
9057 }
9058 
9059 u8 dev_xdp_prog_count(struct net_device *dev)
9060 {
9061 	u8 count = 0;
9062 	int i;
9063 
9064 	for (i = 0; i < __MAX_XDP_MODE; i++)
9065 		if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9066 			count++;
9067 	return count;
9068 }
9069 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9070 
9071 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9072 {
9073 	struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9074 
9075 	return prog ? prog->aux->id : 0;
9076 }
9077 
9078 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9079 			     struct bpf_xdp_link *link)
9080 {
9081 	dev->xdp_state[mode].link = link;
9082 	dev->xdp_state[mode].prog = NULL;
9083 }
9084 
9085 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9086 			     struct bpf_prog *prog)
9087 {
9088 	dev->xdp_state[mode].link = NULL;
9089 	dev->xdp_state[mode].prog = prog;
9090 }
9091 
9092 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9093 			   bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9094 			   u32 flags, struct bpf_prog *prog)
9095 {
9096 	struct netdev_bpf xdp;
9097 	int err;
9098 
9099 	memset(&xdp, 0, sizeof(xdp));
9100 	xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9101 	xdp.extack = extack;
9102 	xdp.flags = flags;
9103 	xdp.prog = prog;
9104 
9105 	/* Drivers assume refcnt is already incremented (i.e, prog pointer is
9106 	 * "moved" into driver), so they don't increment it on their own, but
9107 	 * they do decrement refcnt when program is detached or replaced.
9108 	 * Given net_device also owns link/prog, we need to bump refcnt here
9109 	 * to prevent drivers from underflowing it.
9110 	 */
9111 	if (prog)
9112 		bpf_prog_inc(prog);
9113 	err = bpf_op(dev, &xdp);
9114 	if (err) {
9115 		if (prog)
9116 			bpf_prog_put(prog);
9117 		return err;
9118 	}
9119 
9120 	if (mode != XDP_MODE_HW)
9121 		bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9122 
9123 	return 0;
9124 }
9125 
9126 static void dev_xdp_uninstall(struct net_device *dev)
9127 {
9128 	struct bpf_xdp_link *link;
9129 	struct bpf_prog *prog;
9130 	enum bpf_xdp_mode mode;
9131 	bpf_op_t bpf_op;
9132 
9133 	ASSERT_RTNL();
9134 
9135 	for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9136 		prog = dev_xdp_prog(dev, mode);
9137 		if (!prog)
9138 			continue;
9139 
9140 		bpf_op = dev_xdp_bpf_op(dev, mode);
9141 		if (!bpf_op)
9142 			continue;
9143 
9144 		WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9145 
9146 		/* auto-detach link from net device */
9147 		link = dev_xdp_link(dev, mode);
9148 		if (link)
9149 			link->dev = NULL;
9150 		else
9151 			bpf_prog_put(prog);
9152 
9153 		dev_xdp_set_link(dev, mode, NULL);
9154 	}
9155 }
9156 
9157 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9158 			  struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9159 			  struct bpf_prog *old_prog, u32 flags)
9160 {
9161 	unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9162 	struct bpf_prog *cur_prog;
9163 	struct net_device *upper;
9164 	struct list_head *iter;
9165 	enum bpf_xdp_mode mode;
9166 	bpf_op_t bpf_op;
9167 	int err;
9168 
9169 	ASSERT_RTNL();
9170 
9171 	/* either link or prog attachment, never both */
9172 	if (link && (new_prog || old_prog))
9173 		return -EINVAL;
9174 	/* link supports only XDP mode flags */
9175 	if (link && (flags & ~XDP_FLAGS_MODES)) {
9176 		NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9177 		return -EINVAL;
9178 	}
9179 	/* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9180 	if (num_modes > 1) {
9181 		NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9182 		return -EINVAL;
9183 	}
9184 	/* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9185 	if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9186 		NL_SET_ERR_MSG(extack,
9187 			       "More than one program loaded, unset mode is ambiguous");
9188 		return -EINVAL;
9189 	}
9190 	/* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9191 	if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9192 		NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9193 		return -EINVAL;
9194 	}
9195 
9196 	mode = dev_xdp_mode(dev, flags);
9197 	/* can't replace attached link */
9198 	if (dev_xdp_link(dev, mode)) {
9199 		NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9200 		return -EBUSY;
9201 	}
9202 
9203 	/* don't allow if an upper device already has a program */
9204 	netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9205 		if (dev_xdp_prog_count(upper) > 0) {
9206 			NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9207 			return -EEXIST;
9208 		}
9209 	}
9210 
9211 	cur_prog = dev_xdp_prog(dev, mode);
9212 	/* can't replace attached prog with link */
9213 	if (link && cur_prog) {
9214 		NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9215 		return -EBUSY;
9216 	}
9217 	if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9218 		NL_SET_ERR_MSG(extack, "Active program does not match expected");
9219 		return -EEXIST;
9220 	}
9221 
9222 	/* put effective new program into new_prog */
9223 	if (link)
9224 		new_prog = link->link.prog;
9225 
9226 	if (new_prog) {
9227 		bool offload = mode == XDP_MODE_HW;
9228 		enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9229 					       ? XDP_MODE_DRV : XDP_MODE_SKB;
9230 
9231 		if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9232 			NL_SET_ERR_MSG(extack, "XDP program already attached");
9233 			return -EBUSY;
9234 		}
9235 		if (!offload && dev_xdp_prog(dev, other_mode)) {
9236 			NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9237 			return -EEXIST;
9238 		}
9239 		if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9240 			NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9241 			return -EINVAL;
9242 		}
9243 		if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9244 			NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9245 			return -EINVAL;
9246 		}
9247 		if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9248 			NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9249 			return -EINVAL;
9250 		}
9251 	}
9252 
9253 	/* don't call drivers if the effective program didn't change */
9254 	if (new_prog != cur_prog) {
9255 		bpf_op = dev_xdp_bpf_op(dev, mode);
9256 		if (!bpf_op) {
9257 			NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9258 			return -EOPNOTSUPP;
9259 		}
9260 
9261 		err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9262 		if (err)
9263 			return err;
9264 	}
9265 
9266 	if (link)
9267 		dev_xdp_set_link(dev, mode, link);
9268 	else
9269 		dev_xdp_set_prog(dev, mode, new_prog);
9270 	if (cur_prog)
9271 		bpf_prog_put(cur_prog);
9272 
9273 	return 0;
9274 }
9275 
9276 static int dev_xdp_attach_link(struct net_device *dev,
9277 			       struct netlink_ext_ack *extack,
9278 			       struct bpf_xdp_link *link)
9279 {
9280 	return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9281 }
9282 
9283 static int dev_xdp_detach_link(struct net_device *dev,
9284 			       struct netlink_ext_ack *extack,
9285 			       struct bpf_xdp_link *link)
9286 {
9287 	enum bpf_xdp_mode mode;
9288 	bpf_op_t bpf_op;
9289 
9290 	ASSERT_RTNL();
9291 
9292 	mode = dev_xdp_mode(dev, link->flags);
9293 	if (dev_xdp_link(dev, mode) != link)
9294 		return -EINVAL;
9295 
9296 	bpf_op = dev_xdp_bpf_op(dev, mode);
9297 	WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9298 	dev_xdp_set_link(dev, mode, NULL);
9299 	return 0;
9300 }
9301 
9302 static void bpf_xdp_link_release(struct bpf_link *link)
9303 {
9304 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9305 
9306 	rtnl_lock();
9307 
9308 	/* if racing with net_device's tear down, xdp_link->dev might be
9309 	 * already NULL, in which case link was already auto-detached
9310 	 */
9311 	if (xdp_link->dev) {
9312 		WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9313 		xdp_link->dev = NULL;
9314 	}
9315 
9316 	rtnl_unlock();
9317 }
9318 
9319 static int bpf_xdp_link_detach(struct bpf_link *link)
9320 {
9321 	bpf_xdp_link_release(link);
9322 	return 0;
9323 }
9324 
9325 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9326 {
9327 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9328 
9329 	kfree(xdp_link);
9330 }
9331 
9332 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9333 				     struct seq_file *seq)
9334 {
9335 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9336 	u32 ifindex = 0;
9337 
9338 	rtnl_lock();
9339 	if (xdp_link->dev)
9340 		ifindex = xdp_link->dev->ifindex;
9341 	rtnl_unlock();
9342 
9343 	seq_printf(seq, "ifindex:\t%u\n", ifindex);
9344 }
9345 
9346 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9347 				       struct bpf_link_info *info)
9348 {
9349 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9350 	u32 ifindex = 0;
9351 
9352 	rtnl_lock();
9353 	if (xdp_link->dev)
9354 		ifindex = xdp_link->dev->ifindex;
9355 	rtnl_unlock();
9356 
9357 	info->xdp.ifindex = ifindex;
9358 	return 0;
9359 }
9360 
9361 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9362 			       struct bpf_prog *old_prog)
9363 {
9364 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9365 	enum bpf_xdp_mode mode;
9366 	bpf_op_t bpf_op;
9367 	int err = 0;
9368 
9369 	rtnl_lock();
9370 
9371 	/* link might have been auto-released already, so fail */
9372 	if (!xdp_link->dev) {
9373 		err = -ENOLINK;
9374 		goto out_unlock;
9375 	}
9376 
9377 	if (old_prog && link->prog != old_prog) {
9378 		err = -EPERM;
9379 		goto out_unlock;
9380 	}
9381 	old_prog = link->prog;
9382 	if (old_prog->type != new_prog->type ||
9383 	    old_prog->expected_attach_type != new_prog->expected_attach_type) {
9384 		err = -EINVAL;
9385 		goto out_unlock;
9386 	}
9387 
9388 	if (old_prog == new_prog) {
9389 		/* no-op, don't disturb drivers */
9390 		bpf_prog_put(new_prog);
9391 		goto out_unlock;
9392 	}
9393 
9394 	mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9395 	bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9396 	err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9397 			      xdp_link->flags, new_prog);
9398 	if (err)
9399 		goto out_unlock;
9400 
9401 	old_prog = xchg(&link->prog, new_prog);
9402 	bpf_prog_put(old_prog);
9403 
9404 out_unlock:
9405 	rtnl_unlock();
9406 	return err;
9407 }
9408 
9409 static const struct bpf_link_ops bpf_xdp_link_lops = {
9410 	.release = bpf_xdp_link_release,
9411 	.dealloc = bpf_xdp_link_dealloc,
9412 	.detach = bpf_xdp_link_detach,
9413 	.show_fdinfo = bpf_xdp_link_show_fdinfo,
9414 	.fill_link_info = bpf_xdp_link_fill_link_info,
9415 	.update_prog = bpf_xdp_link_update,
9416 };
9417 
9418 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9419 {
9420 	struct net *net = current->nsproxy->net_ns;
9421 	struct bpf_link_primer link_primer;
9422 	struct bpf_xdp_link *link;
9423 	struct net_device *dev;
9424 	int err, fd;
9425 
9426 	rtnl_lock();
9427 	dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9428 	if (!dev) {
9429 		rtnl_unlock();
9430 		return -EINVAL;
9431 	}
9432 
9433 	link = kzalloc(sizeof(*link), GFP_USER);
9434 	if (!link) {
9435 		err = -ENOMEM;
9436 		goto unlock;
9437 	}
9438 
9439 	bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9440 	link->dev = dev;
9441 	link->flags = attr->link_create.flags;
9442 
9443 	err = bpf_link_prime(&link->link, &link_primer);
9444 	if (err) {
9445 		kfree(link);
9446 		goto unlock;
9447 	}
9448 
9449 	err = dev_xdp_attach_link(dev, NULL, link);
9450 	rtnl_unlock();
9451 
9452 	if (err) {
9453 		link->dev = NULL;
9454 		bpf_link_cleanup(&link_primer);
9455 		goto out_put_dev;
9456 	}
9457 
9458 	fd = bpf_link_settle(&link_primer);
9459 	/* link itself doesn't hold dev's refcnt to not complicate shutdown */
9460 	dev_put(dev);
9461 	return fd;
9462 
9463 unlock:
9464 	rtnl_unlock();
9465 
9466 out_put_dev:
9467 	dev_put(dev);
9468 	return err;
9469 }
9470 
9471 /**
9472  *	dev_change_xdp_fd - set or clear a bpf program for a device rx path
9473  *	@dev: device
9474  *	@extack: netlink extended ack
9475  *	@fd: new program fd or negative value to clear
9476  *	@expected_fd: old program fd that userspace expects to replace or clear
9477  *	@flags: xdp-related flags
9478  *
9479  *	Set or clear a bpf program for a device
9480  */
9481 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9482 		      int fd, int expected_fd, u32 flags)
9483 {
9484 	enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9485 	struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9486 	int err;
9487 
9488 	ASSERT_RTNL();
9489 
9490 	if (fd >= 0) {
9491 		new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9492 						 mode != XDP_MODE_SKB);
9493 		if (IS_ERR(new_prog))
9494 			return PTR_ERR(new_prog);
9495 	}
9496 
9497 	if (expected_fd >= 0) {
9498 		old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9499 						 mode != XDP_MODE_SKB);
9500 		if (IS_ERR(old_prog)) {
9501 			err = PTR_ERR(old_prog);
9502 			old_prog = NULL;
9503 			goto err_out;
9504 		}
9505 	}
9506 
9507 	err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9508 
9509 err_out:
9510 	if (err && new_prog)
9511 		bpf_prog_put(new_prog);
9512 	if (old_prog)
9513 		bpf_prog_put(old_prog);
9514 	return err;
9515 }
9516 
9517 /**
9518  *	dev_new_index	-	allocate an ifindex
9519  *	@net: the applicable net namespace
9520  *
9521  *	Returns a suitable unique value for a new device interface
9522  *	number.  The caller must hold the rtnl semaphore or the
9523  *	dev_base_lock to be sure it remains unique.
9524  */
9525 static int dev_new_index(struct net *net)
9526 {
9527 	int ifindex = net->ifindex;
9528 
9529 	for (;;) {
9530 		if (++ifindex <= 0)
9531 			ifindex = 1;
9532 		if (!__dev_get_by_index(net, ifindex))
9533 			return net->ifindex = ifindex;
9534 	}
9535 }
9536 
9537 /* Delayed registration/unregisteration */
9538 LIST_HEAD(net_todo_list);
9539 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9540 
9541 static void net_set_todo(struct net_device *dev)
9542 {
9543 	list_add_tail(&dev->todo_list, &net_todo_list);
9544 	atomic_inc(&dev_net(dev)->dev_unreg_count);
9545 }
9546 
9547 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9548 	struct net_device *upper, netdev_features_t features)
9549 {
9550 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9551 	netdev_features_t feature;
9552 	int feature_bit;
9553 
9554 	for_each_netdev_feature(upper_disables, feature_bit) {
9555 		feature = __NETIF_F_BIT(feature_bit);
9556 		if (!(upper->wanted_features & feature)
9557 		    && (features & feature)) {
9558 			netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9559 				   &feature, upper->name);
9560 			features &= ~feature;
9561 		}
9562 	}
9563 
9564 	return features;
9565 }
9566 
9567 static void netdev_sync_lower_features(struct net_device *upper,
9568 	struct net_device *lower, netdev_features_t features)
9569 {
9570 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9571 	netdev_features_t feature;
9572 	int feature_bit;
9573 
9574 	for_each_netdev_feature(upper_disables, feature_bit) {
9575 		feature = __NETIF_F_BIT(feature_bit);
9576 		if (!(features & feature) && (lower->features & feature)) {
9577 			netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9578 				   &feature, lower->name);
9579 			lower->wanted_features &= ~feature;
9580 			__netdev_update_features(lower);
9581 
9582 			if (unlikely(lower->features & feature))
9583 				netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9584 					    &feature, lower->name);
9585 			else
9586 				netdev_features_change(lower);
9587 		}
9588 	}
9589 }
9590 
9591 static netdev_features_t netdev_fix_features(struct net_device *dev,
9592 	netdev_features_t features)
9593 {
9594 	/* Fix illegal checksum combinations */
9595 	if ((features & NETIF_F_HW_CSUM) &&
9596 	    (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9597 		netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9598 		features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9599 	}
9600 
9601 	/* TSO requires that SG is present as well. */
9602 	if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9603 		netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9604 		features &= ~NETIF_F_ALL_TSO;
9605 	}
9606 
9607 	if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9608 					!(features & NETIF_F_IP_CSUM)) {
9609 		netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9610 		features &= ~NETIF_F_TSO;
9611 		features &= ~NETIF_F_TSO_ECN;
9612 	}
9613 
9614 	if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9615 					 !(features & NETIF_F_IPV6_CSUM)) {
9616 		netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9617 		features &= ~NETIF_F_TSO6;
9618 	}
9619 
9620 	/* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9621 	if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9622 		features &= ~NETIF_F_TSO_MANGLEID;
9623 
9624 	/* TSO ECN requires that TSO is present as well. */
9625 	if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9626 		features &= ~NETIF_F_TSO_ECN;
9627 
9628 	/* Software GSO depends on SG. */
9629 	if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9630 		netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9631 		features &= ~NETIF_F_GSO;
9632 	}
9633 
9634 	/* GSO partial features require GSO partial be set */
9635 	if ((features & dev->gso_partial_features) &&
9636 	    !(features & NETIF_F_GSO_PARTIAL)) {
9637 		netdev_dbg(dev,
9638 			   "Dropping partially supported GSO features since no GSO partial.\n");
9639 		features &= ~dev->gso_partial_features;
9640 	}
9641 
9642 	if (!(features & NETIF_F_RXCSUM)) {
9643 		/* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9644 		 * successfully merged by hardware must also have the
9645 		 * checksum verified by hardware.  If the user does not
9646 		 * want to enable RXCSUM, logically, we should disable GRO_HW.
9647 		 */
9648 		if (features & NETIF_F_GRO_HW) {
9649 			netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9650 			features &= ~NETIF_F_GRO_HW;
9651 		}
9652 	}
9653 
9654 	/* LRO/HW-GRO features cannot be combined with RX-FCS */
9655 	if (features & NETIF_F_RXFCS) {
9656 		if (features & NETIF_F_LRO) {
9657 			netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9658 			features &= ~NETIF_F_LRO;
9659 		}
9660 
9661 		if (features & NETIF_F_GRO_HW) {
9662 			netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9663 			features &= ~NETIF_F_GRO_HW;
9664 		}
9665 	}
9666 
9667 	if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9668 		netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9669 		features &= ~NETIF_F_LRO;
9670 	}
9671 
9672 	if (features & NETIF_F_HW_TLS_TX) {
9673 		bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9674 			(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9675 		bool hw_csum = features & NETIF_F_HW_CSUM;
9676 
9677 		if (!ip_csum && !hw_csum) {
9678 			netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9679 			features &= ~NETIF_F_HW_TLS_TX;
9680 		}
9681 	}
9682 
9683 	if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9684 		netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9685 		features &= ~NETIF_F_HW_TLS_RX;
9686 	}
9687 
9688 	return features;
9689 }
9690 
9691 int __netdev_update_features(struct net_device *dev)
9692 {
9693 	struct net_device *upper, *lower;
9694 	netdev_features_t features;
9695 	struct list_head *iter;
9696 	int err = -1;
9697 
9698 	ASSERT_RTNL();
9699 
9700 	features = netdev_get_wanted_features(dev);
9701 
9702 	if (dev->netdev_ops->ndo_fix_features)
9703 		features = dev->netdev_ops->ndo_fix_features(dev, features);
9704 
9705 	/* driver might be less strict about feature dependencies */
9706 	features = netdev_fix_features(dev, features);
9707 
9708 	/* some features can't be enabled if they're off on an upper device */
9709 	netdev_for_each_upper_dev_rcu(dev, upper, iter)
9710 		features = netdev_sync_upper_features(dev, upper, features);
9711 
9712 	if (dev->features == features)
9713 		goto sync_lower;
9714 
9715 	netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9716 		&dev->features, &features);
9717 
9718 	if (dev->netdev_ops->ndo_set_features)
9719 		err = dev->netdev_ops->ndo_set_features(dev, features);
9720 	else
9721 		err = 0;
9722 
9723 	if (unlikely(err < 0)) {
9724 		netdev_err(dev,
9725 			"set_features() failed (%d); wanted %pNF, left %pNF\n",
9726 			err, &features, &dev->features);
9727 		/* return non-0 since some features might have changed and
9728 		 * it's better to fire a spurious notification than miss it
9729 		 */
9730 		return -1;
9731 	}
9732 
9733 sync_lower:
9734 	/* some features must be disabled on lower devices when disabled
9735 	 * on an upper device (think: bonding master or bridge)
9736 	 */
9737 	netdev_for_each_lower_dev(dev, lower, iter)
9738 		netdev_sync_lower_features(dev, lower, features);
9739 
9740 	if (!err) {
9741 		netdev_features_t diff = features ^ dev->features;
9742 
9743 		if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9744 			/* udp_tunnel_{get,drop}_rx_info both need
9745 			 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9746 			 * device, or they won't do anything.
9747 			 * Thus we need to update dev->features
9748 			 * *before* calling udp_tunnel_get_rx_info,
9749 			 * but *after* calling udp_tunnel_drop_rx_info.
9750 			 */
9751 			if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9752 				dev->features = features;
9753 				udp_tunnel_get_rx_info(dev);
9754 			} else {
9755 				udp_tunnel_drop_rx_info(dev);
9756 			}
9757 		}
9758 
9759 		if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9760 			if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9761 				dev->features = features;
9762 				err |= vlan_get_rx_ctag_filter_info(dev);
9763 			} else {
9764 				vlan_drop_rx_ctag_filter_info(dev);
9765 			}
9766 		}
9767 
9768 		if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9769 			if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9770 				dev->features = features;
9771 				err |= vlan_get_rx_stag_filter_info(dev);
9772 			} else {
9773 				vlan_drop_rx_stag_filter_info(dev);
9774 			}
9775 		}
9776 
9777 		dev->features = features;
9778 	}
9779 
9780 	return err < 0 ? 0 : 1;
9781 }
9782 
9783 /**
9784  *	netdev_update_features - recalculate device features
9785  *	@dev: the device to check
9786  *
9787  *	Recalculate dev->features set and send notifications if it
9788  *	has changed. Should be called after driver or hardware dependent
9789  *	conditions might have changed that influence the features.
9790  */
9791 void netdev_update_features(struct net_device *dev)
9792 {
9793 	if (__netdev_update_features(dev))
9794 		netdev_features_change(dev);
9795 }
9796 EXPORT_SYMBOL(netdev_update_features);
9797 
9798 /**
9799  *	netdev_change_features - recalculate device features
9800  *	@dev: the device to check
9801  *
9802  *	Recalculate dev->features set and send notifications even
9803  *	if they have not changed. Should be called instead of
9804  *	netdev_update_features() if also dev->vlan_features might
9805  *	have changed to allow the changes to be propagated to stacked
9806  *	VLAN devices.
9807  */
9808 void netdev_change_features(struct net_device *dev)
9809 {
9810 	__netdev_update_features(dev);
9811 	netdev_features_change(dev);
9812 }
9813 EXPORT_SYMBOL(netdev_change_features);
9814 
9815 /**
9816  *	netif_stacked_transfer_operstate -	transfer operstate
9817  *	@rootdev: the root or lower level device to transfer state from
9818  *	@dev: the device to transfer operstate to
9819  *
9820  *	Transfer operational state from root to device. This is normally
9821  *	called when a stacking relationship exists between the root
9822  *	device and the device(a leaf device).
9823  */
9824 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9825 					struct net_device *dev)
9826 {
9827 	if (rootdev->operstate == IF_OPER_DORMANT)
9828 		netif_dormant_on(dev);
9829 	else
9830 		netif_dormant_off(dev);
9831 
9832 	if (rootdev->operstate == IF_OPER_TESTING)
9833 		netif_testing_on(dev);
9834 	else
9835 		netif_testing_off(dev);
9836 
9837 	if (netif_carrier_ok(rootdev))
9838 		netif_carrier_on(dev);
9839 	else
9840 		netif_carrier_off(dev);
9841 }
9842 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9843 
9844 static int netif_alloc_rx_queues(struct net_device *dev)
9845 {
9846 	unsigned int i, count = dev->num_rx_queues;
9847 	struct netdev_rx_queue *rx;
9848 	size_t sz = count * sizeof(*rx);
9849 	int err = 0;
9850 
9851 	BUG_ON(count < 1);
9852 
9853 	rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9854 	if (!rx)
9855 		return -ENOMEM;
9856 
9857 	dev->_rx = rx;
9858 
9859 	for (i = 0; i < count; i++) {
9860 		rx[i].dev = dev;
9861 
9862 		/* XDP RX-queue setup */
9863 		err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
9864 		if (err < 0)
9865 			goto err_rxq_info;
9866 	}
9867 	return 0;
9868 
9869 err_rxq_info:
9870 	/* Rollback successful reg's and free other resources */
9871 	while (i--)
9872 		xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9873 	kvfree(dev->_rx);
9874 	dev->_rx = NULL;
9875 	return err;
9876 }
9877 
9878 static void netif_free_rx_queues(struct net_device *dev)
9879 {
9880 	unsigned int i, count = dev->num_rx_queues;
9881 
9882 	/* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9883 	if (!dev->_rx)
9884 		return;
9885 
9886 	for (i = 0; i < count; i++)
9887 		xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9888 
9889 	kvfree(dev->_rx);
9890 }
9891 
9892 static void netdev_init_one_queue(struct net_device *dev,
9893 				  struct netdev_queue *queue, void *_unused)
9894 {
9895 	/* Initialize queue lock */
9896 	spin_lock_init(&queue->_xmit_lock);
9897 	netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9898 	queue->xmit_lock_owner = -1;
9899 	netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9900 	queue->dev = dev;
9901 #ifdef CONFIG_BQL
9902 	dql_init(&queue->dql, HZ);
9903 #endif
9904 }
9905 
9906 static void netif_free_tx_queues(struct net_device *dev)
9907 {
9908 	kvfree(dev->_tx);
9909 }
9910 
9911 static int netif_alloc_netdev_queues(struct net_device *dev)
9912 {
9913 	unsigned int count = dev->num_tx_queues;
9914 	struct netdev_queue *tx;
9915 	size_t sz = count * sizeof(*tx);
9916 
9917 	if (count < 1 || count > 0xffff)
9918 		return -EINVAL;
9919 
9920 	tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9921 	if (!tx)
9922 		return -ENOMEM;
9923 
9924 	dev->_tx = tx;
9925 
9926 	netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9927 	spin_lock_init(&dev->tx_global_lock);
9928 
9929 	return 0;
9930 }
9931 
9932 void netif_tx_stop_all_queues(struct net_device *dev)
9933 {
9934 	unsigned int i;
9935 
9936 	for (i = 0; i < dev->num_tx_queues; i++) {
9937 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9938 
9939 		netif_tx_stop_queue(txq);
9940 	}
9941 }
9942 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9943 
9944 /**
9945  * register_netdevice() - register a network device
9946  * @dev: device to register
9947  *
9948  * Take a prepared network device structure and make it externally accessible.
9949  * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
9950  * Callers must hold the rtnl lock - you may want register_netdev()
9951  * instead of this.
9952  */
9953 int register_netdevice(struct net_device *dev)
9954 {
9955 	int ret;
9956 	struct net *net = dev_net(dev);
9957 
9958 	BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9959 		     NETDEV_FEATURE_COUNT);
9960 	BUG_ON(dev_boot_phase);
9961 	ASSERT_RTNL();
9962 
9963 	might_sleep();
9964 
9965 	/* When net_device's are persistent, this will be fatal. */
9966 	BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9967 	BUG_ON(!net);
9968 
9969 	ret = ethtool_check_ops(dev->ethtool_ops);
9970 	if (ret)
9971 		return ret;
9972 
9973 	spin_lock_init(&dev->addr_list_lock);
9974 	netdev_set_addr_lockdep_class(dev);
9975 
9976 	ret = dev_get_valid_name(net, dev, dev->name);
9977 	if (ret < 0)
9978 		goto out;
9979 
9980 	ret = -ENOMEM;
9981 	dev->name_node = netdev_name_node_head_alloc(dev);
9982 	if (!dev->name_node)
9983 		goto out;
9984 
9985 	/* Init, if this function is available */
9986 	if (dev->netdev_ops->ndo_init) {
9987 		ret = dev->netdev_ops->ndo_init(dev);
9988 		if (ret) {
9989 			if (ret > 0)
9990 				ret = -EIO;
9991 			goto err_free_name;
9992 		}
9993 	}
9994 
9995 	if (((dev->hw_features | dev->features) &
9996 	     NETIF_F_HW_VLAN_CTAG_FILTER) &&
9997 	    (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9998 	     !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9999 		netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10000 		ret = -EINVAL;
10001 		goto err_uninit;
10002 	}
10003 
10004 	ret = -EBUSY;
10005 	if (!dev->ifindex)
10006 		dev->ifindex = dev_new_index(net);
10007 	else if (__dev_get_by_index(net, dev->ifindex))
10008 		goto err_uninit;
10009 
10010 	/* Transfer changeable features to wanted_features and enable
10011 	 * software offloads (GSO and GRO).
10012 	 */
10013 	dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10014 	dev->features |= NETIF_F_SOFT_FEATURES;
10015 
10016 	if (dev->udp_tunnel_nic_info) {
10017 		dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10018 		dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10019 	}
10020 
10021 	dev->wanted_features = dev->features & dev->hw_features;
10022 
10023 	if (!(dev->flags & IFF_LOOPBACK))
10024 		dev->hw_features |= NETIF_F_NOCACHE_COPY;
10025 
10026 	/* If IPv4 TCP segmentation offload is supported we should also
10027 	 * allow the device to enable segmenting the frame with the option
10028 	 * of ignoring a static IP ID value.  This doesn't enable the
10029 	 * feature itself but allows the user to enable it later.
10030 	 */
10031 	if (dev->hw_features & NETIF_F_TSO)
10032 		dev->hw_features |= NETIF_F_TSO_MANGLEID;
10033 	if (dev->vlan_features & NETIF_F_TSO)
10034 		dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10035 	if (dev->mpls_features & NETIF_F_TSO)
10036 		dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10037 	if (dev->hw_enc_features & NETIF_F_TSO)
10038 		dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10039 
10040 	/* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10041 	 */
10042 	dev->vlan_features |= NETIF_F_HIGHDMA;
10043 
10044 	/* Make NETIF_F_SG inheritable to tunnel devices.
10045 	 */
10046 	dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10047 
10048 	/* Make NETIF_F_SG inheritable to MPLS.
10049 	 */
10050 	dev->mpls_features |= NETIF_F_SG;
10051 
10052 	ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10053 	ret = notifier_to_errno(ret);
10054 	if (ret)
10055 		goto err_uninit;
10056 
10057 	ret = netdev_register_kobject(dev);
10058 	write_lock(&dev_base_lock);
10059 	dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED;
10060 	write_unlock(&dev_base_lock);
10061 	if (ret)
10062 		goto err_uninit_notify;
10063 
10064 	__netdev_update_features(dev);
10065 
10066 	/*
10067 	 *	Default initial state at registry is that the
10068 	 *	device is present.
10069 	 */
10070 
10071 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10072 
10073 	linkwatch_init_dev(dev);
10074 
10075 	dev_init_scheduler(dev);
10076 
10077 	netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
10078 	list_netdevice(dev);
10079 
10080 	add_device_randomness(dev->dev_addr, dev->addr_len);
10081 
10082 	/* If the device has permanent device address, driver should
10083 	 * set dev_addr and also addr_assign_type should be set to
10084 	 * NET_ADDR_PERM (default value).
10085 	 */
10086 	if (dev->addr_assign_type == NET_ADDR_PERM)
10087 		memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10088 
10089 	/* Notify protocols, that a new device appeared. */
10090 	ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10091 	ret = notifier_to_errno(ret);
10092 	if (ret) {
10093 		/* Expect explicit free_netdev() on failure */
10094 		dev->needs_free_netdev = false;
10095 		unregister_netdevice_queue(dev, NULL);
10096 		goto out;
10097 	}
10098 	/*
10099 	 *	Prevent userspace races by waiting until the network
10100 	 *	device is fully setup before sending notifications.
10101 	 */
10102 	if (!dev->rtnl_link_ops ||
10103 	    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10104 		rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
10105 
10106 out:
10107 	return ret;
10108 
10109 err_uninit_notify:
10110 	call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10111 err_uninit:
10112 	if (dev->netdev_ops->ndo_uninit)
10113 		dev->netdev_ops->ndo_uninit(dev);
10114 	if (dev->priv_destructor)
10115 		dev->priv_destructor(dev);
10116 err_free_name:
10117 	netdev_name_node_free(dev->name_node);
10118 	goto out;
10119 }
10120 EXPORT_SYMBOL(register_netdevice);
10121 
10122 /**
10123  *	init_dummy_netdev	- init a dummy network device for NAPI
10124  *	@dev: device to init
10125  *
10126  *	This takes a network device structure and initialize the minimum
10127  *	amount of fields so it can be used to schedule NAPI polls without
10128  *	registering a full blown interface. This is to be used by drivers
10129  *	that need to tie several hardware interfaces to a single NAPI
10130  *	poll scheduler due to HW limitations.
10131  */
10132 int init_dummy_netdev(struct net_device *dev)
10133 {
10134 	/* Clear everything. Note we don't initialize spinlocks
10135 	 * are they aren't supposed to be taken by any of the
10136 	 * NAPI code and this dummy netdev is supposed to be
10137 	 * only ever used for NAPI polls
10138 	 */
10139 	memset(dev, 0, sizeof(struct net_device));
10140 
10141 	/* make sure we BUG if trying to hit standard
10142 	 * register/unregister code path
10143 	 */
10144 	dev->reg_state = NETREG_DUMMY;
10145 
10146 	/* NAPI wants this */
10147 	INIT_LIST_HEAD(&dev->napi_list);
10148 
10149 	/* a dummy interface is started by default */
10150 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10151 	set_bit(__LINK_STATE_START, &dev->state);
10152 
10153 	/* napi_busy_loop stats accounting wants this */
10154 	dev_net_set(dev, &init_net);
10155 
10156 	/* Note : We dont allocate pcpu_refcnt for dummy devices,
10157 	 * because users of this 'device' dont need to change
10158 	 * its refcount.
10159 	 */
10160 
10161 	return 0;
10162 }
10163 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10164 
10165 
10166 /**
10167  *	register_netdev	- register a network device
10168  *	@dev: device to register
10169  *
10170  *	Take a completed network device structure and add it to the kernel
10171  *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10172  *	chain. 0 is returned on success. A negative errno code is returned
10173  *	on a failure to set up the device, or if the name is a duplicate.
10174  *
10175  *	This is a wrapper around register_netdevice that takes the rtnl semaphore
10176  *	and expands the device name if you passed a format string to
10177  *	alloc_netdev.
10178  */
10179 int register_netdev(struct net_device *dev)
10180 {
10181 	int err;
10182 
10183 	if (rtnl_lock_killable())
10184 		return -EINTR;
10185 	err = register_netdevice(dev);
10186 	rtnl_unlock();
10187 	return err;
10188 }
10189 EXPORT_SYMBOL(register_netdev);
10190 
10191 int netdev_refcnt_read(const struct net_device *dev)
10192 {
10193 #ifdef CONFIG_PCPU_DEV_REFCNT
10194 	int i, refcnt = 0;
10195 
10196 	for_each_possible_cpu(i)
10197 		refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10198 	return refcnt;
10199 #else
10200 	return refcount_read(&dev->dev_refcnt);
10201 #endif
10202 }
10203 EXPORT_SYMBOL(netdev_refcnt_read);
10204 
10205 int netdev_unregister_timeout_secs __read_mostly = 10;
10206 
10207 #define WAIT_REFS_MIN_MSECS 1
10208 #define WAIT_REFS_MAX_MSECS 250
10209 /**
10210  * netdev_wait_allrefs_any - wait until all references are gone.
10211  * @list: list of net_devices to wait on
10212  *
10213  * This is called when unregistering network devices.
10214  *
10215  * Any protocol or device that holds a reference should register
10216  * for netdevice notification, and cleanup and put back the
10217  * reference if they receive an UNREGISTER event.
10218  * We can get stuck here if buggy protocols don't correctly
10219  * call dev_put.
10220  */
10221 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
10222 {
10223 	unsigned long rebroadcast_time, warning_time;
10224 	struct net_device *dev;
10225 	int wait = 0;
10226 
10227 	rebroadcast_time = warning_time = jiffies;
10228 
10229 	list_for_each_entry(dev, list, todo_list)
10230 		if (netdev_refcnt_read(dev) == 1)
10231 			return dev;
10232 
10233 	while (true) {
10234 		if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10235 			rtnl_lock();
10236 
10237 			/* Rebroadcast unregister notification */
10238 			list_for_each_entry(dev, list, todo_list)
10239 				call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10240 
10241 			__rtnl_unlock();
10242 			rcu_barrier();
10243 			rtnl_lock();
10244 
10245 			list_for_each_entry(dev, list, todo_list)
10246 				if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10247 					     &dev->state)) {
10248 					/* We must not have linkwatch events
10249 					 * pending on unregister. If this
10250 					 * happens, we simply run the queue
10251 					 * unscheduled, resulting in a noop
10252 					 * for this device.
10253 					 */
10254 					linkwatch_run_queue();
10255 					break;
10256 				}
10257 
10258 			__rtnl_unlock();
10259 
10260 			rebroadcast_time = jiffies;
10261 		}
10262 
10263 		if (!wait) {
10264 			rcu_barrier();
10265 			wait = WAIT_REFS_MIN_MSECS;
10266 		} else {
10267 			msleep(wait);
10268 			wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10269 		}
10270 
10271 		list_for_each_entry(dev, list, todo_list)
10272 			if (netdev_refcnt_read(dev) == 1)
10273 				return dev;
10274 
10275 		if (time_after(jiffies, warning_time +
10276 			       READ_ONCE(netdev_unregister_timeout_secs) * HZ)) {
10277 			list_for_each_entry(dev, list, todo_list) {
10278 				pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10279 					 dev->name, netdev_refcnt_read(dev));
10280 				ref_tracker_dir_print(&dev->refcnt_tracker, 10);
10281 			}
10282 
10283 			warning_time = jiffies;
10284 		}
10285 	}
10286 }
10287 
10288 /* The sequence is:
10289  *
10290  *	rtnl_lock();
10291  *	...
10292  *	register_netdevice(x1);
10293  *	register_netdevice(x2);
10294  *	...
10295  *	unregister_netdevice(y1);
10296  *	unregister_netdevice(y2);
10297  *      ...
10298  *	rtnl_unlock();
10299  *	free_netdev(y1);
10300  *	free_netdev(y2);
10301  *
10302  * We are invoked by rtnl_unlock().
10303  * This allows us to deal with problems:
10304  * 1) We can delete sysfs objects which invoke hotplug
10305  *    without deadlocking with linkwatch via keventd.
10306  * 2) Since we run with the RTNL semaphore not held, we can sleep
10307  *    safely in order to wait for the netdev refcnt to drop to zero.
10308  *
10309  * We must not return until all unregister events added during
10310  * the interval the lock was held have been completed.
10311  */
10312 void netdev_run_todo(void)
10313 {
10314 	struct net_device *dev, *tmp;
10315 	struct list_head list;
10316 #ifdef CONFIG_LOCKDEP
10317 	struct list_head unlink_list;
10318 
10319 	list_replace_init(&net_unlink_list, &unlink_list);
10320 
10321 	while (!list_empty(&unlink_list)) {
10322 		struct net_device *dev = list_first_entry(&unlink_list,
10323 							  struct net_device,
10324 							  unlink_list);
10325 		list_del_init(&dev->unlink_list);
10326 		dev->nested_level = dev->lower_level - 1;
10327 	}
10328 #endif
10329 
10330 	/* Snapshot list, allow later requests */
10331 	list_replace_init(&net_todo_list, &list);
10332 
10333 	__rtnl_unlock();
10334 
10335 	/* Wait for rcu callbacks to finish before next phase */
10336 	if (!list_empty(&list))
10337 		rcu_barrier();
10338 
10339 	list_for_each_entry_safe(dev, tmp, &list, todo_list) {
10340 		if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10341 			netdev_WARN(dev, "run_todo but not unregistering\n");
10342 			list_del(&dev->todo_list);
10343 			continue;
10344 		}
10345 
10346 		write_lock(&dev_base_lock);
10347 		dev->reg_state = NETREG_UNREGISTERED;
10348 		write_unlock(&dev_base_lock);
10349 		linkwatch_forget_dev(dev);
10350 	}
10351 
10352 	while (!list_empty(&list)) {
10353 		dev = netdev_wait_allrefs_any(&list);
10354 		list_del(&dev->todo_list);
10355 
10356 		/* paranoia */
10357 		BUG_ON(netdev_refcnt_read(dev) != 1);
10358 		BUG_ON(!list_empty(&dev->ptype_all));
10359 		BUG_ON(!list_empty(&dev->ptype_specific));
10360 		WARN_ON(rcu_access_pointer(dev->ip_ptr));
10361 		WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10362 
10363 		if (dev->priv_destructor)
10364 			dev->priv_destructor(dev);
10365 		if (dev->needs_free_netdev)
10366 			free_netdev(dev);
10367 
10368 		if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count))
10369 			wake_up(&netdev_unregistering_wq);
10370 
10371 		/* Free network device */
10372 		kobject_put(&dev->dev.kobj);
10373 	}
10374 }
10375 
10376 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10377  * all the same fields in the same order as net_device_stats, with only
10378  * the type differing, but rtnl_link_stats64 may have additional fields
10379  * at the end for newer counters.
10380  */
10381 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10382 			     const struct net_device_stats *netdev_stats)
10383 {
10384 #if BITS_PER_LONG == 64
10385 	BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
10386 	memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
10387 	/* zero out counters that only exist in rtnl_link_stats64 */
10388 	memset((char *)stats64 + sizeof(*netdev_stats), 0,
10389 	       sizeof(*stats64) - sizeof(*netdev_stats));
10390 #else
10391 	size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
10392 	const unsigned long *src = (const unsigned long *)netdev_stats;
10393 	u64 *dst = (u64 *)stats64;
10394 
10395 	BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10396 	for (i = 0; i < n; i++)
10397 		dst[i] = src[i];
10398 	/* zero out counters that only exist in rtnl_link_stats64 */
10399 	memset((char *)stats64 + n * sizeof(u64), 0,
10400 	       sizeof(*stats64) - n * sizeof(u64));
10401 #endif
10402 }
10403 EXPORT_SYMBOL(netdev_stats_to_stats64);
10404 
10405 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev)
10406 {
10407 	struct net_device_core_stats __percpu *p;
10408 
10409 	p = alloc_percpu_gfp(struct net_device_core_stats,
10410 			     GFP_ATOMIC | __GFP_NOWARN);
10411 
10412 	if (p && cmpxchg(&dev->core_stats, NULL, p))
10413 		free_percpu(p);
10414 
10415 	/* This READ_ONCE() pairs with the cmpxchg() above */
10416 	return READ_ONCE(dev->core_stats);
10417 }
10418 EXPORT_SYMBOL(netdev_core_stats_alloc);
10419 
10420 /**
10421  *	dev_get_stats	- get network device statistics
10422  *	@dev: device to get statistics from
10423  *	@storage: place to store stats
10424  *
10425  *	Get network statistics from device. Return @storage.
10426  *	The device driver may provide its own method by setting
10427  *	dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10428  *	otherwise the internal statistics structure is used.
10429  */
10430 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10431 					struct rtnl_link_stats64 *storage)
10432 {
10433 	const struct net_device_ops *ops = dev->netdev_ops;
10434 	const struct net_device_core_stats __percpu *p;
10435 
10436 	if (ops->ndo_get_stats64) {
10437 		memset(storage, 0, sizeof(*storage));
10438 		ops->ndo_get_stats64(dev, storage);
10439 	} else if (ops->ndo_get_stats) {
10440 		netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10441 	} else {
10442 		netdev_stats_to_stats64(storage, &dev->stats);
10443 	}
10444 
10445 	/* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10446 	p = READ_ONCE(dev->core_stats);
10447 	if (p) {
10448 		const struct net_device_core_stats *core_stats;
10449 		int i;
10450 
10451 		for_each_possible_cpu(i) {
10452 			core_stats = per_cpu_ptr(p, i);
10453 			storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
10454 			storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
10455 			storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
10456 			storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
10457 		}
10458 	}
10459 	return storage;
10460 }
10461 EXPORT_SYMBOL(dev_get_stats);
10462 
10463 /**
10464  *	dev_fetch_sw_netstats - get per-cpu network device statistics
10465  *	@s: place to store stats
10466  *	@netstats: per-cpu network stats to read from
10467  *
10468  *	Read per-cpu network statistics and populate the related fields in @s.
10469  */
10470 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10471 			   const struct pcpu_sw_netstats __percpu *netstats)
10472 {
10473 	int cpu;
10474 
10475 	for_each_possible_cpu(cpu) {
10476 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
10477 		const struct pcpu_sw_netstats *stats;
10478 		unsigned int start;
10479 
10480 		stats = per_cpu_ptr(netstats, cpu);
10481 		do {
10482 			start = u64_stats_fetch_begin(&stats->syncp);
10483 			rx_packets = u64_stats_read(&stats->rx_packets);
10484 			rx_bytes   = u64_stats_read(&stats->rx_bytes);
10485 			tx_packets = u64_stats_read(&stats->tx_packets);
10486 			tx_bytes   = u64_stats_read(&stats->tx_bytes);
10487 		} while (u64_stats_fetch_retry(&stats->syncp, start));
10488 
10489 		s->rx_packets += rx_packets;
10490 		s->rx_bytes   += rx_bytes;
10491 		s->tx_packets += tx_packets;
10492 		s->tx_bytes   += tx_bytes;
10493 	}
10494 }
10495 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10496 
10497 /**
10498  *	dev_get_tstats64 - ndo_get_stats64 implementation
10499  *	@dev: device to get statistics from
10500  *	@s: place to store stats
10501  *
10502  *	Populate @s from dev->stats and dev->tstats. Can be used as
10503  *	ndo_get_stats64() callback.
10504  */
10505 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10506 {
10507 	netdev_stats_to_stats64(s, &dev->stats);
10508 	dev_fetch_sw_netstats(s, dev->tstats);
10509 }
10510 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10511 
10512 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10513 {
10514 	struct netdev_queue *queue = dev_ingress_queue(dev);
10515 
10516 #ifdef CONFIG_NET_CLS_ACT
10517 	if (queue)
10518 		return queue;
10519 	queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10520 	if (!queue)
10521 		return NULL;
10522 	netdev_init_one_queue(dev, queue, NULL);
10523 	RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10524 	queue->qdisc_sleeping = &noop_qdisc;
10525 	rcu_assign_pointer(dev->ingress_queue, queue);
10526 #endif
10527 	return queue;
10528 }
10529 
10530 static const struct ethtool_ops default_ethtool_ops;
10531 
10532 void netdev_set_default_ethtool_ops(struct net_device *dev,
10533 				    const struct ethtool_ops *ops)
10534 {
10535 	if (dev->ethtool_ops == &default_ethtool_ops)
10536 		dev->ethtool_ops = ops;
10537 }
10538 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10539 
10540 void netdev_freemem(struct net_device *dev)
10541 {
10542 	char *addr = (char *)dev - dev->padded;
10543 
10544 	kvfree(addr);
10545 }
10546 
10547 /**
10548  * alloc_netdev_mqs - allocate network device
10549  * @sizeof_priv: size of private data to allocate space for
10550  * @name: device name format string
10551  * @name_assign_type: origin of device name
10552  * @setup: callback to initialize device
10553  * @txqs: the number of TX subqueues to allocate
10554  * @rxqs: the number of RX subqueues to allocate
10555  *
10556  * Allocates a struct net_device with private data area for driver use
10557  * and performs basic initialization.  Also allocates subqueue structs
10558  * for each queue on the device.
10559  */
10560 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10561 		unsigned char name_assign_type,
10562 		void (*setup)(struct net_device *),
10563 		unsigned int txqs, unsigned int rxqs)
10564 {
10565 	struct net_device *dev;
10566 	unsigned int alloc_size;
10567 	struct net_device *p;
10568 
10569 	BUG_ON(strlen(name) >= sizeof(dev->name));
10570 
10571 	if (txqs < 1) {
10572 		pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10573 		return NULL;
10574 	}
10575 
10576 	if (rxqs < 1) {
10577 		pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10578 		return NULL;
10579 	}
10580 
10581 	alloc_size = sizeof(struct net_device);
10582 	if (sizeof_priv) {
10583 		/* ensure 32-byte alignment of private area */
10584 		alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10585 		alloc_size += sizeof_priv;
10586 	}
10587 	/* ensure 32-byte alignment of whole construct */
10588 	alloc_size += NETDEV_ALIGN - 1;
10589 
10590 	p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10591 	if (!p)
10592 		return NULL;
10593 
10594 	dev = PTR_ALIGN(p, NETDEV_ALIGN);
10595 	dev->padded = (char *)dev - (char *)p;
10596 
10597 	ref_tracker_dir_init(&dev->refcnt_tracker, 128);
10598 #ifdef CONFIG_PCPU_DEV_REFCNT
10599 	dev->pcpu_refcnt = alloc_percpu(int);
10600 	if (!dev->pcpu_refcnt)
10601 		goto free_dev;
10602 	__dev_hold(dev);
10603 #else
10604 	refcount_set(&dev->dev_refcnt, 1);
10605 #endif
10606 
10607 	if (dev_addr_init(dev))
10608 		goto free_pcpu;
10609 
10610 	dev_mc_init(dev);
10611 	dev_uc_init(dev);
10612 
10613 	dev_net_set(dev, &init_net);
10614 
10615 	dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
10616 	dev->gso_max_segs = GSO_MAX_SEGS;
10617 	dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
10618 	dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
10619 	dev->tso_max_segs = TSO_MAX_SEGS;
10620 	dev->upper_level = 1;
10621 	dev->lower_level = 1;
10622 #ifdef CONFIG_LOCKDEP
10623 	dev->nested_level = 0;
10624 	INIT_LIST_HEAD(&dev->unlink_list);
10625 #endif
10626 
10627 	INIT_LIST_HEAD(&dev->napi_list);
10628 	INIT_LIST_HEAD(&dev->unreg_list);
10629 	INIT_LIST_HEAD(&dev->close_list);
10630 	INIT_LIST_HEAD(&dev->link_watch_list);
10631 	INIT_LIST_HEAD(&dev->adj_list.upper);
10632 	INIT_LIST_HEAD(&dev->adj_list.lower);
10633 	INIT_LIST_HEAD(&dev->ptype_all);
10634 	INIT_LIST_HEAD(&dev->ptype_specific);
10635 	INIT_LIST_HEAD(&dev->net_notifier_list);
10636 #ifdef CONFIG_NET_SCHED
10637 	hash_init(dev->qdisc_hash);
10638 #endif
10639 	dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10640 	setup(dev);
10641 
10642 	if (!dev->tx_queue_len) {
10643 		dev->priv_flags |= IFF_NO_QUEUE;
10644 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10645 	}
10646 
10647 	dev->num_tx_queues = txqs;
10648 	dev->real_num_tx_queues = txqs;
10649 	if (netif_alloc_netdev_queues(dev))
10650 		goto free_all;
10651 
10652 	dev->num_rx_queues = rxqs;
10653 	dev->real_num_rx_queues = rxqs;
10654 	if (netif_alloc_rx_queues(dev))
10655 		goto free_all;
10656 
10657 	strcpy(dev->name, name);
10658 	dev->name_assign_type = name_assign_type;
10659 	dev->group = INIT_NETDEV_GROUP;
10660 	if (!dev->ethtool_ops)
10661 		dev->ethtool_ops = &default_ethtool_ops;
10662 
10663 	nf_hook_netdev_init(dev);
10664 
10665 	return dev;
10666 
10667 free_all:
10668 	free_netdev(dev);
10669 	return NULL;
10670 
10671 free_pcpu:
10672 #ifdef CONFIG_PCPU_DEV_REFCNT
10673 	free_percpu(dev->pcpu_refcnt);
10674 free_dev:
10675 #endif
10676 	netdev_freemem(dev);
10677 	return NULL;
10678 }
10679 EXPORT_SYMBOL(alloc_netdev_mqs);
10680 
10681 /**
10682  * free_netdev - free network device
10683  * @dev: device
10684  *
10685  * This function does the last stage of destroying an allocated device
10686  * interface. The reference to the device object is released. If this
10687  * is the last reference then it will be freed.Must be called in process
10688  * context.
10689  */
10690 void free_netdev(struct net_device *dev)
10691 {
10692 	struct napi_struct *p, *n;
10693 
10694 	might_sleep();
10695 
10696 	/* When called immediately after register_netdevice() failed the unwind
10697 	 * handling may still be dismantling the device. Handle that case by
10698 	 * deferring the free.
10699 	 */
10700 	if (dev->reg_state == NETREG_UNREGISTERING) {
10701 		ASSERT_RTNL();
10702 		dev->needs_free_netdev = true;
10703 		return;
10704 	}
10705 
10706 	netif_free_tx_queues(dev);
10707 	netif_free_rx_queues(dev);
10708 
10709 	kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10710 
10711 	/* Flush device addresses */
10712 	dev_addr_flush(dev);
10713 
10714 	list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10715 		netif_napi_del(p);
10716 
10717 	ref_tracker_dir_exit(&dev->refcnt_tracker);
10718 #ifdef CONFIG_PCPU_DEV_REFCNT
10719 	free_percpu(dev->pcpu_refcnt);
10720 	dev->pcpu_refcnt = NULL;
10721 #endif
10722 	free_percpu(dev->core_stats);
10723 	dev->core_stats = NULL;
10724 	free_percpu(dev->xdp_bulkq);
10725 	dev->xdp_bulkq = NULL;
10726 
10727 	/*  Compatibility with error handling in drivers */
10728 	if (dev->reg_state == NETREG_UNINITIALIZED) {
10729 		netdev_freemem(dev);
10730 		return;
10731 	}
10732 
10733 	BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10734 	dev->reg_state = NETREG_RELEASED;
10735 
10736 	/* will free via device release */
10737 	put_device(&dev->dev);
10738 }
10739 EXPORT_SYMBOL(free_netdev);
10740 
10741 /**
10742  *	synchronize_net -  Synchronize with packet receive processing
10743  *
10744  *	Wait for packets currently being received to be done.
10745  *	Does not block later packets from starting.
10746  */
10747 void synchronize_net(void)
10748 {
10749 	might_sleep();
10750 	if (rtnl_is_locked())
10751 		synchronize_rcu_expedited();
10752 	else
10753 		synchronize_rcu();
10754 }
10755 EXPORT_SYMBOL(synchronize_net);
10756 
10757 /**
10758  *	unregister_netdevice_queue - remove device from the kernel
10759  *	@dev: device
10760  *	@head: list
10761  *
10762  *	This function shuts down a device interface and removes it
10763  *	from the kernel tables.
10764  *	If head not NULL, device is queued to be unregistered later.
10765  *
10766  *	Callers must hold the rtnl semaphore.  You may want
10767  *	unregister_netdev() instead of this.
10768  */
10769 
10770 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10771 {
10772 	ASSERT_RTNL();
10773 
10774 	if (head) {
10775 		list_move_tail(&dev->unreg_list, head);
10776 	} else {
10777 		LIST_HEAD(single);
10778 
10779 		list_add(&dev->unreg_list, &single);
10780 		unregister_netdevice_many(&single);
10781 	}
10782 }
10783 EXPORT_SYMBOL(unregister_netdevice_queue);
10784 
10785 void unregister_netdevice_many_notify(struct list_head *head,
10786 				      u32 portid, const struct nlmsghdr *nlh)
10787 {
10788 	struct net_device *dev, *tmp;
10789 	LIST_HEAD(close_head);
10790 
10791 	BUG_ON(dev_boot_phase);
10792 	ASSERT_RTNL();
10793 
10794 	if (list_empty(head))
10795 		return;
10796 
10797 	list_for_each_entry_safe(dev, tmp, head, unreg_list) {
10798 		/* Some devices call without registering
10799 		 * for initialization unwind. Remove those
10800 		 * devices and proceed with the remaining.
10801 		 */
10802 		if (dev->reg_state == NETREG_UNINITIALIZED) {
10803 			pr_debug("unregister_netdevice: device %s/%p never was registered\n",
10804 				 dev->name, dev);
10805 
10806 			WARN_ON(1);
10807 			list_del(&dev->unreg_list);
10808 			continue;
10809 		}
10810 		dev->dismantle = true;
10811 		BUG_ON(dev->reg_state != NETREG_REGISTERED);
10812 	}
10813 
10814 	/* If device is running, close it first. */
10815 	list_for_each_entry(dev, head, unreg_list)
10816 		list_add_tail(&dev->close_list, &close_head);
10817 	dev_close_many(&close_head, true);
10818 
10819 	list_for_each_entry(dev, head, unreg_list) {
10820 		/* And unlink it from device chain. */
10821 		write_lock(&dev_base_lock);
10822 		unlist_netdevice(dev, false);
10823 		dev->reg_state = NETREG_UNREGISTERING;
10824 		write_unlock(&dev_base_lock);
10825 	}
10826 	flush_all_backlogs();
10827 
10828 	synchronize_net();
10829 
10830 	list_for_each_entry(dev, head, unreg_list) {
10831 		struct sk_buff *skb = NULL;
10832 
10833 		/* Shutdown queueing discipline. */
10834 		dev_shutdown(dev);
10835 
10836 		dev_xdp_uninstall(dev);
10837 
10838 		netdev_offload_xstats_disable_all(dev);
10839 
10840 		/* Notify protocols, that we are about to destroy
10841 		 * this device. They should clean all the things.
10842 		 */
10843 		call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10844 
10845 		if (!dev->rtnl_link_ops ||
10846 		    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10847 			skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
10848 						     GFP_KERNEL, NULL, 0,
10849 						     portid, nlmsg_seq(nlh));
10850 
10851 		/*
10852 		 *	Flush the unicast and multicast chains
10853 		 */
10854 		dev_uc_flush(dev);
10855 		dev_mc_flush(dev);
10856 
10857 		netdev_name_node_alt_flush(dev);
10858 		netdev_name_node_free(dev->name_node);
10859 
10860 		call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10861 
10862 		if (dev->netdev_ops->ndo_uninit)
10863 			dev->netdev_ops->ndo_uninit(dev);
10864 
10865 		if (skb)
10866 			rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh);
10867 
10868 		/* Notifier chain MUST detach us all upper devices. */
10869 		WARN_ON(netdev_has_any_upper_dev(dev));
10870 		WARN_ON(netdev_has_any_lower_dev(dev));
10871 
10872 		/* Remove entries from kobject tree */
10873 		netdev_unregister_kobject(dev);
10874 #ifdef CONFIG_XPS
10875 		/* Remove XPS queueing entries */
10876 		netif_reset_xps_queues_gt(dev, 0);
10877 #endif
10878 	}
10879 
10880 	synchronize_net();
10881 
10882 	list_for_each_entry(dev, head, unreg_list) {
10883 		netdev_put(dev, &dev->dev_registered_tracker);
10884 		net_set_todo(dev);
10885 	}
10886 
10887 	list_del(head);
10888 }
10889 
10890 /**
10891  *	unregister_netdevice_many - unregister many devices
10892  *	@head: list of devices
10893  *
10894  *  Note: As most callers use a stack allocated list_head,
10895  *  we force a list_del() to make sure stack wont be corrupted later.
10896  */
10897 void unregister_netdevice_many(struct list_head *head)
10898 {
10899 	unregister_netdevice_many_notify(head, 0, NULL);
10900 }
10901 EXPORT_SYMBOL(unregister_netdevice_many);
10902 
10903 /**
10904  *	unregister_netdev - remove device from the kernel
10905  *	@dev: device
10906  *
10907  *	This function shuts down a device interface and removes it
10908  *	from the kernel tables.
10909  *
10910  *	This is just a wrapper for unregister_netdevice that takes
10911  *	the rtnl semaphore.  In general you want to use this and not
10912  *	unregister_netdevice.
10913  */
10914 void unregister_netdev(struct net_device *dev)
10915 {
10916 	rtnl_lock();
10917 	unregister_netdevice(dev);
10918 	rtnl_unlock();
10919 }
10920 EXPORT_SYMBOL(unregister_netdev);
10921 
10922 /**
10923  *	__dev_change_net_namespace - move device to different nethost namespace
10924  *	@dev: device
10925  *	@net: network namespace
10926  *	@pat: If not NULL name pattern to try if the current device name
10927  *	      is already taken in the destination network namespace.
10928  *	@new_ifindex: If not zero, specifies device index in the target
10929  *	              namespace.
10930  *
10931  *	This function shuts down a device interface and moves it
10932  *	to a new network namespace. On success 0 is returned, on
10933  *	a failure a netagive errno code is returned.
10934  *
10935  *	Callers must hold the rtnl semaphore.
10936  */
10937 
10938 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
10939 			       const char *pat, int new_ifindex)
10940 {
10941 	struct net *net_old = dev_net(dev);
10942 	int err, new_nsid;
10943 
10944 	ASSERT_RTNL();
10945 
10946 	/* Don't allow namespace local devices to be moved. */
10947 	err = -EINVAL;
10948 	if (dev->features & NETIF_F_NETNS_LOCAL)
10949 		goto out;
10950 
10951 	/* Ensure the device has been registrered */
10952 	if (dev->reg_state != NETREG_REGISTERED)
10953 		goto out;
10954 
10955 	/* Get out if there is nothing todo */
10956 	err = 0;
10957 	if (net_eq(net_old, net))
10958 		goto out;
10959 
10960 	/* Pick the destination device name, and ensure
10961 	 * we can use it in the destination network namespace.
10962 	 */
10963 	err = -EEXIST;
10964 	if (netdev_name_in_use(net, dev->name)) {
10965 		/* We get here if we can't use the current device name */
10966 		if (!pat)
10967 			goto out;
10968 		err = dev_get_valid_name(net, dev, pat);
10969 		if (err < 0)
10970 			goto out;
10971 	}
10972 
10973 	/* Check that new_ifindex isn't used yet. */
10974 	err = -EBUSY;
10975 	if (new_ifindex && __dev_get_by_index(net, new_ifindex))
10976 		goto out;
10977 
10978 	/*
10979 	 * And now a mini version of register_netdevice unregister_netdevice.
10980 	 */
10981 
10982 	/* If device is running close it first. */
10983 	dev_close(dev);
10984 
10985 	/* And unlink it from device chain */
10986 	unlist_netdevice(dev, true);
10987 
10988 	synchronize_net();
10989 
10990 	/* Shutdown queueing discipline. */
10991 	dev_shutdown(dev);
10992 
10993 	/* Notify protocols, that we are about to destroy
10994 	 * this device. They should clean all the things.
10995 	 *
10996 	 * Note that dev->reg_state stays at NETREG_REGISTERED.
10997 	 * This is wanted because this way 8021q and macvlan know
10998 	 * the device is just moving and can keep their slaves up.
10999 	 */
11000 	call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11001 	rcu_barrier();
11002 
11003 	new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11004 	/* If there is an ifindex conflict assign a new one */
11005 	if (!new_ifindex) {
11006 		if (__dev_get_by_index(net, dev->ifindex))
11007 			new_ifindex = dev_new_index(net);
11008 		else
11009 			new_ifindex = dev->ifindex;
11010 	}
11011 
11012 	rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11013 			    new_ifindex);
11014 
11015 	/*
11016 	 *	Flush the unicast and multicast chains
11017 	 */
11018 	dev_uc_flush(dev);
11019 	dev_mc_flush(dev);
11020 
11021 	/* Send a netdev-removed uevent to the old namespace */
11022 	kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11023 	netdev_adjacent_del_links(dev);
11024 
11025 	/* Move per-net netdevice notifiers that are following the netdevice */
11026 	move_netdevice_notifiers_dev_net(dev, net);
11027 
11028 	/* Actually switch the network namespace */
11029 	dev_net_set(dev, net);
11030 	dev->ifindex = new_ifindex;
11031 
11032 	/* Send a netdev-add uevent to the new namespace */
11033 	kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11034 	netdev_adjacent_add_links(dev);
11035 
11036 	/* Fixup kobjects */
11037 	err = device_rename(&dev->dev, dev->name);
11038 	WARN_ON(err);
11039 
11040 	/* Adapt owner in case owning user namespace of target network
11041 	 * namespace is different from the original one.
11042 	 */
11043 	err = netdev_change_owner(dev, net_old, net);
11044 	WARN_ON(err);
11045 
11046 	/* Add the device back in the hashes */
11047 	list_netdevice(dev);
11048 
11049 	/* Notify protocols, that a new device appeared. */
11050 	call_netdevice_notifiers(NETDEV_REGISTER, dev);
11051 
11052 	/*
11053 	 *	Prevent userspace races by waiting until the network
11054 	 *	device is fully setup before sending notifications.
11055 	 */
11056 	rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
11057 
11058 	synchronize_net();
11059 	err = 0;
11060 out:
11061 	return err;
11062 }
11063 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11064 
11065 static int dev_cpu_dead(unsigned int oldcpu)
11066 {
11067 	struct sk_buff **list_skb;
11068 	struct sk_buff *skb;
11069 	unsigned int cpu;
11070 	struct softnet_data *sd, *oldsd, *remsd = NULL;
11071 
11072 	local_irq_disable();
11073 	cpu = smp_processor_id();
11074 	sd = &per_cpu(softnet_data, cpu);
11075 	oldsd = &per_cpu(softnet_data, oldcpu);
11076 
11077 	/* Find end of our completion_queue. */
11078 	list_skb = &sd->completion_queue;
11079 	while (*list_skb)
11080 		list_skb = &(*list_skb)->next;
11081 	/* Append completion queue from offline CPU. */
11082 	*list_skb = oldsd->completion_queue;
11083 	oldsd->completion_queue = NULL;
11084 
11085 	/* Append output queue from offline CPU. */
11086 	if (oldsd->output_queue) {
11087 		*sd->output_queue_tailp = oldsd->output_queue;
11088 		sd->output_queue_tailp = oldsd->output_queue_tailp;
11089 		oldsd->output_queue = NULL;
11090 		oldsd->output_queue_tailp = &oldsd->output_queue;
11091 	}
11092 	/* Append NAPI poll list from offline CPU, with one exception :
11093 	 * process_backlog() must be called by cpu owning percpu backlog.
11094 	 * We properly handle process_queue & input_pkt_queue later.
11095 	 */
11096 	while (!list_empty(&oldsd->poll_list)) {
11097 		struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11098 							    struct napi_struct,
11099 							    poll_list);
11100 
11101 		list_del_init(&napi->poll_list);
11102 		if (napi->poll == process_backlog)
11103 			napi->state = 0;
11104 		else
11105 			____napi_schedule(sd, napi);
11106 	}
11107 
11108 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
11109 	local_irq_enable();
11110 
11111 #ifdef CONFIG_RPS
11112 	remsd = oldsd->rps_ipi_list;
11113 	oldsd->rps_ipi_list = NULL;
11114 #endif
11115 	/* send out pending IPI's on offline CPU */
11116 	net_rps_send_ipi(remsd);
11117 
11118 	/* Process offline CPU's input_pkt_queue */
11119 	while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11120 		netif_rx(skb);
11121 		input_queue_head_incr(oldsd);
11122 	}
11123 	while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11124 		netif_rx(skb);
11125 		input_queue_head_incr(oldsd);
11126 	}
11127 
11128 	return 0;
11129 }
11130 
11131 /**
11132  *	netdev_increment_features - increment feature set by one
11133  *	@all: current feature set
11134  *	@one: new feature set
11135  *	@mask: mask feature set
11136  *
11137  *	Computes a new feature set after adding a device with feature set
11138  *	@one to the master device with current feature set @all.  Will not
11139  *	enable anything that is off in @mask. Returns the new feature set.
11140  */
11141 netdev_features_t netdev_increment_features(netdev_features_t all,
11142 	netdev_features_t one, netdev_features_t mask)
11143 {
11144 	if (mask & NETIF_F_HW_CSUM)
11145 		mask |= NETIF_F_CSUM_MASK;
11146 	mask |= NETIF_F_VLAN_CHALLENGED;
11147 
11148 	all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11149 	all &= one | ~NETIF_F_ALL_FOR_ALL;
11150 
11151 	/* If one device supports hw checksumming, set for all. */
11152 	if (all & NETIF_F_HW_CSUM)
11153 		all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11154 
11155 	return all;
11156 }
11157 EXPORT_SYMBOL(netdev_increment_features);
11158 
11159 static struct hlist_head * __net_init netdev_create_hash(void)
11160 {
11161 	int i;
11162 	struct hlist_head *hash;
11163 
11164 	hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11165 	if (hash != NULL)
11166 		for (i = 0; i < NETDEV_HASHENTRIES; i++)
11167 			INIT_HLIST_HEAD(&hash[i]);
11168 
11169 	return hash;
11170 }
11171 
11172 /* Initialize per network namespace state */
11173 static int __net_init netdev_init(struct net *net)
11174 {
11175 	BUILD_BUG_ON(GRO_HASH_BUCKETS >
11176 		     8 * sizeof_field(struct napi_struct, gro_bitmask));
11177 
11178 	INIT_LIST_HEAD(&net->dev_base_head);
11179 
11180 	net->dev_name_head = netdev_create_hash();
11181 	if (net->dev_name_head == NULL)
11182 		goto err_name;
11183 
11184 	net->dev_index_head = netdev_create_hash();
11185 	if (net->dev_index_head == NULL)
11186 		goto err_idx;
11187 
11188 	RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11189 
11190 	return 0;
11191 
11192 err_idx:
11193 	kfree(net->dev_name_head);
11194 err_name:
11195 	return -ENOMEM;
11196 }
11197 
11198 /**
11199  *	netdev_drivername - network driver for the device
11200  *	@dev: network device
11201  *
11202  *	Determine network driver for device.
11203  */
11204 const char *netdev_drivername(const struct net_device *dev)
11205 {
11206 	const struct device_driver *driver;
11207 	const struct device *parent;
11208 	const char *empty = "";
11209 
11210 	parent = dev->dev.parent;
11211 	if (!parent)
11212 		return empty;
11213 
11214 	driver = parent->driver;
11215 	if (driver && driver->name)
11216 		return driver->name;
11217 	return empty;
11218 }
11219 
11220 static void __netdev_printk(const char *level, const struct net_device *dev,
11221 			    struct va_format *vaf)
11222 {
11223 	if (dev && dev->dev.parent) {
11224 		dev_printk_emit(level[1] - '0',
11225 				dev->dev.parent,
11226 				"%s %s %s%s: %pV",
11227 				dev_driver_string(dev->dev.parent),
11228 				dev_name(dev->dev.parent),
11229 				netdev_name(dev), netdev_reg_state(dev),
11230 				vaf);
11231 	} else if (dev) {
11232 		printk("%s%s%s: %pV",
11233 		       level, netdev_name(dev), netdev_reg_state(dev), vaf);
11234 	} else {
11235 		printk("%s(NULL net_device): %pV", level, vaf);
11236 	}
11237 }
11238 
11239 void netdev_printk(const char *level, const struct net_device *dev,
11240 		   const char *format, ...)
11241 {
11242 	struct va_format vaf;
11243 	va_list args;
11244 
11245 	va_start(args, format);
11246 
11247 	vaf.fmt = format;
11248 	vaf.va = &args;
11249 
11250 	__netdev_printk(level, dev, &vaf);
11251 
11252 	va_end(args);
11253 }
11254 EXPORT_SYMBOL(netdev_printk);
11255 
11256 #define define_netdev_printk_level(func, level)			\
11257 void func(const struct net_device *dev, const char *fmt, ...)	\
11258 {								\
11259 	struct va_format vaf;					\
11260 	va_list args;						\
11261 								\
11262 	va_start(args, fmt);					\
11263 								\
11264 	vaf.fmt = fmt;						\
11265 	vaf.va = &args;						\
11266 								\
11267 	__netdev_printk(level, dev, &vaf);			\
11268 								\
11269 	va_end(args);						\
11270 }								\
11271 EXPORT_SYMBOL(func);
11272 
11273 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11274 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11275 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11276 define_netdev_printk_level(netdev_err, KERN_ERR);
11277 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11278 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11279 define_netdev_printk_level(netdev_info, KERN_INFO);
11280 
11281 static void __net_exit netdev_exit(struct net *net)
11282 {
11283 	kfree(net->dev_name_head);
11284 	kfree(net->dev_index_head);
11285 	if (net != &init_net)
11286 		WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11287 }
11288 
11289 static struct pernet_operations __net_initdata netdev_net_ops = {
11290 	.init = netdev_init,
11291 	.exit = netdev_exit,
11292 };
11293 
11294 static void __net_exit default_device_exit_net(struct net *net)
11295 {
11296 	struct net_device *dev, *aux;
11297 	/*
11298 	 * Push all migratable network devices back to the
11299 	 * initial network namespace
11300 	 */
11301 	ASSERT_RTNL();
11302 	for_each_netdev_safe(net, dev, aux) {
11303 		int err;
11304 		char fb_name[IFNAMSIZ];
11305 
11306 		/* Ignore unmoveable devices (i.e. loopback) */
11307 		if (dev->features & NETIF_F_NETNS_LOCAL)
11308 			continue;
11309 
11310 		/* Leave virtual devices for the generic cleanup */
11311 		if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11312 			continue;
11313 
11314 		/* Push remaining network devices to init_net */
11315 		snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11316 		if (netdev_name_in_use(&init_net, fb_name))
11317 			snprintf(fb_name, IFNAMSIZ, "dev%%d");
11318 		err = dev_change_net_namespace(dev, &init_net, fb_name);
11319 		if (err) {
11320 			pr_emerg("%s: failed to move %s to init_net: %d\n",
11321 				 __func__, dev->name, err);
11322 			BUG();
11323 		}
11324 	}
11325 }
11326 
11327 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11328 {
11329 	/* At exit all network devices most be removed from a network
11330 	 * namespace.  Do this in the reverse order of registration.
11331 	 * Do this across as many network namespaces as possible to
11332 	 * improve batching efficiency.
11333 	 */
11334 	struct net_device *dev;
11335 	struct net *net;
11336 	LIST_HEAD(dev_kill_list);
11337 
11338 	rtnl_lock();
11339 	list_for_each_entry(net, net_list, exit_list) {
11340 		default_device_exit_net(net);
11341 		cond_resched();
11342 	}
11343 
11344 	list_for_each_entry(net, net_list, exit_list) {
11345 		for_each_netdev_reverse(net, dev) {
11346 			if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11347 				dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11348 			else
11349 				unregister_netdevice_queue(dev, &dev_kill_list);
11350 		}
11351 	}
11352 	unregister_netdevice_many(&dev_kill_list);
11353 	rtnl_unlock();
11354 }
11355 
11356 static struct pernet_operations __net_initdata default_device_ops = {
11357 	.exit_batch = default_device_exit_batch,
11358 };
11359 
11360 /*
11361  *	Initialize the DEV module. At boot time this walks the device list and
11362  *	unhooks any devices that fail to initialise (normally hardware not
11363  *	present) and leaves us with a valid list of present and active devices.
11364  *
11365  */
11366 
11367 /*
11368  *       This is called single threaded during boot, so no need
11369  *       to take the rtnl semaphore.
11370  */
11371 static int __init net_dev_init(void)
11372 {
11373 	int i, rc = -ENOMEM;
11374 
11375 	BUG_ON(!dev_boot_phase);
11376 
11377 	if (dev_proc_init())
11378 		goto out;
11379 
11380 	if (netdev_kobject_init())
11381 		goto out;
11382 
11383 	INIT_LIST_HEAD(&ptype_all);
11384 	for (i = 0; i < PTYPE_HASH_SIZE; i++)
11385 		INIT_LIST_HEAD(&ptype_base[i]);
11386 
11387 	if (register_pernet_subsys(&netdev_net_ops))
11388 		goto out;
11389 
11390 	/*
11391 	 *	Initialise the packet receive queues.
11392 	 */
11393 
11394 	for_each_possible_cpu(i) {
11395 		struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11396 		struct softnet_data *sd = &per_cpu(softnet_data, i);
11397 
11398 		INIT_WORK(flush, flush_backlog);
11399 
11400 		skb_queue_head_init(&sd->input_pkt_queue);
11401 		skb_queue_head_init(&sd->process_queue);
11402 #ifdef CONFIG_XFRM_OFFLOAD
11403 		skb_queue_head_init(&sd->xfrm_backlog);
11404 #endif
11405 		INIT_LIST_HEAD(&sd->poll_list);
11406 		sd->output_queue_tailp = &sd->output_queue;
11407 #ifdef CONFIG_RPS
11408 		INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11409 		sd->cpu = i;
11410 #endif
11411 		INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
11412 		spin_lock_init(&sd->defer_lock);
11413 
11414 		init_gro_hash(&sd->backlog);
11415 		sd->backlog.poll = process_backlog;
11416 		sd->backlog.weight = weight_p;
11417 	}
11418 
11419 	dev_boot_phase = 0;
11420 
11421 	/* The loopback device is special if any other network devices
11422 	 * is present in a network namespace the loopback device must
11423 	 * be present. Since we now dynamically allocate and free the
11424 	 * loopback device ensure this invariant is maintained by
11425 	 * keeping the loopback device as the first device on the
11426 	 * list of network devices.  Ensuring the loopback devices
11427 	 * is the first device that appears and the last network device
11428 	 * that disappears.
11429 	 */
11430 	if (register_pernet_device(&loopback_net_ops))
11431 		goto out;
11432 
11433 	if (register_pernet_device(&default_device_ops))
11434 		goto out;
11435 
11436 	open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11437 	open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11438 
11439 	rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11440 				       NULL, dev_cpu_dead);
11441 	WARN_ON(rc < 0);
11442 	rc = 0;
11443 out:
11444 	return rc;
11445 }
11446 
11447 subsys_initcall(net_dev_init);
11448