xref: /openbmc/linux/net/core/dev.c (revision 7b7090b4)
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)
401 {
402 	ASSERT_RTNL();
403 
404 	/* Unlink dev from the device chain */
405 	write_lock(&dev_base_lock);
406 	list_del_rcu(&dev->dev_list);
407 	netdev_name_node_del(dev->name_node);
408 	hlist_del_rcu(&dev->index_hlist);
409 	write_unlock(&dev_base_lock);
410 
411 	dev_base_seq_inc(dev_net(dev));
412 }
413 
414 /*
415  *	Our notifier list
416  */
417 
418 static RAW_NOTIFIER_HEAD(netdev_chain);
419 
420 /*
421  *	Device drivers call our routines to queue packets here. We empty the
422  *	queue in the local softnet handler.
423  */
424 
425 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
426 EXPORT_PER_CPU_SYMBOL(softnet_data);
427 
428 #ifdef CONFIG_LOCKDEP
429 /*
430  * register_netdevice() inits txq->_xmit_lock and sets lockdep class
431  * according to dev->type
432  */
433 static const unsigned short netdev_lock_type[] = {
434 	 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
435 	 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
436 	 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
437 	 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
438 	 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
439 	 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
440 	 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
441 	 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
442 	 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
443 	 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
444 	 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
445 	 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
446 	 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
447 	 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
448 	 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
449 
450 static const char *const netdev_lock_name[] = {
451 	"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
452 	"_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
453 	"_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
454 	"_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
455 	"_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
456 	"_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
457 	"_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
458 	"_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
459 	"_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
460 	"_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
461 	"_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
462 	"_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
463 	"_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
464 	"_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
465 	"_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
466 
467 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
468 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
469 
470 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
471 {
472 	int i;
473 
474 	for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
475 		if (netdev_lock_type[i] == dev_type)
476 			return i;
477 	/* the last key is used by default */
478 	return ARRAY_SIZE(netdev_lock_type) - 1;
479 }
480 
481 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
482 						 unsigned short dev_type)
483 {
484 	int i;
485 
486 	i = netdev_lock_pos(dev_type);
487 	lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
488 				   netdev_lock_name[i]);
489 }
490 
491 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
492 {
493 	int i;
494 
495 	i = netdev_lock_pos(dev->type);
496 	lockdep_set_class_and_name(&dev->addr_list_lock,
497 				   &netdev_addr_lock_key[i],
498 				   netdev_lock_name[i]);
499 }
500 #else
501 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
502 						 unsigned short dev_type)
503 {
504 }
505 
506 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
507 {
508 }
509 #endif
510 
511 /*******************************************************************************
512  *
513  *		Protocol management and registration routines
514  *
515  *******************************************************************************/
516 
517 
518 /*
519  *	Add a protocol ID to the list. Now that the input handler is
520  *	smarter we can dispense with all the messy stuff that used to be
521  *	here.
522  *
523  *	BEWARE!!! Protocol handlers, mangling input packets,
524  *	MUST BE last in hash buckets and checking protocol handlers
525  *	MUST start from promiscuous ptype_all chain in net_bh.
526  *	It is true now, do not change it.
527  *	Explanation follows: if protocol handler, mangling packet, will
528  *	be the first on list, it is not able to sense, that packet
529  *	is cloned and should be copied-on-write, so that it will
530  *	change it and subsequent readers will get broken packet.
531  *							--ANK (980803)
532  */
533 
534 static inline struct list_head *ptype_head(const struct packet_type *pt)
535 {
536 	if (pt->type == htons(ETH_P_ALL))
537 		return pt->dev ? &pt->dev->ptype_all : &ptype_all;
538 	else
539 		return pt->dev ? &pt->dev->ptype_specific :
540 				 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
541 }
542 
543 /**
544  *	dev_add_pack - add packet handler
545  *	@pt: packet type declaration
546  *
547  *	Add a protocol handler to the networking stack. The passed &packet_type
548  *	is linked into kernel lists and may not be freed until it has been
549  *	removed from the kernel lists.
550  *
551  *	This call does not sleep therefore it can not
552  *	guarantee all CPU's that are in middle of receiving packets
553  *	will see the new packet type (until the next received packet).
554  */
555 
556 void dev_add_pack(struct packet_type *pt)
557 {
558 	struct list_head *head = ptype_head(pt);
559 
560 	spin_lock(&ptype_lock);
561 	list_add_rcu(&pt->list, head);
562 	spin_unlock(&ptype_lock);
563 }
564 EXPORT_SYMBOL(dev_add_pack);
565 
566 /**
567  *	__dev_remove_pack	 - remove packet handler
568  *	@pt: packet type declaration
569  *
570  *	Remove a protocol handler that was previously added to the kernel
571  *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
572  *	from the kernel lists and can be freed or reused once this function
573  *	returns.
574  *
575  *      The packet type might still be in use by receivers
576  *	and must not be freed until after all the CPU's have gone
577  *	through a quiescent state.
578  */
579 void __dev_remove_pack(struct packet_type *pt)
580 {
581 	struct list_head *head = ptype_head(pt);
582 	struct packet_type *pt1;
583 
584 	spin_lock(&ptype_lock);
585 
586 	list_for_each_entry(pt1, head, list) {
587 		if (pt == pt1) {
588 			list_del_rcu(&pt->list);
589 			goto out;
590 		}
591 	}
592 
593 	pr_warn("dev_remove_pack: %p not found\n", pt);
594 out:
595 	spin_unlock(&ptype_lock);
596 }
597 EXPORT_SYMBOL(__dev_remove_pack);
598 
599 /**
600  *	dev_remove_pack	 - remove packet handler
601  *	@pt: packet type declaration
602  *
603  *	Remove a protocol handler that was previously added to the kernel
604  *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
605  *	from the kernel lists and can be freed or reused once this function
606  *	returns.
607  *
608  *	This call sleeps to guarantee that no CPU is looking at the packet
609  *	type after return.
610  */
611 void dev_remove_pack(struct packet_type *pt)
612 {
613 	__dev_remove_pack(pt);
614 
615 	synchronize_net();
616 }
617 EXPORT_SYMBOL(dev_remove_pack);
618 
619 
620 /*******************************************************************************
621  *
622  *			    Device Interface Subroutines
623  *
624  *******************************************************************************/
625 
626 /**
627  *	dev_get_iflink	- get 'iflink' value of a interface
628  *	@dev: targeted interface
629  *
630  *	Indicates the ifindex the interface is linked to.
631  *	Physical interfaces have the same 'ifindex' and 'iflink' values.
632  */
633 
634 int dev_get_iflink(const struct net_device *dev)
635 {
636 	if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
637 		return dev->netdev_ops->ndo_get_iflink(dev);
638 
639 	return dev->ifindex;
640 }
641 EXPORT_SYMBOL(dev_get_iflink);
642 
643 /**
644  *	dev_fill_metadata_dst - Retrieve tunnel egress information.
645  *	@dev: targeted interface
646  *	@skb: The packet.
647  *
648  *	For better visibility of tunnel traffic OVS needs to retrieve
649  *	egress tunnel information for a packet. Following API allows
650  *	user to get this info.
651  */
652 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
653 {
654 	struct ip_tunnel_info *info;
655 
656 	if (!dev->netdev_ops  || !dev->netdev_ops->ndo_fill_metadata_dst)
657 		return -EINVAL;
658 
659 	info = skb_tunnel_info_unclone(skb);
660 	if (!info)
661 		return -ENOMEM;
662 	if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
663 		return -EINVAL;
664 
665 	return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
666 }
667 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
668 
669 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
670 {
671 	int k = stack->num_paths++;
672 
673 	if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
674 		return NULL;
675 
676 	return &stack->path[k];
677 }
678 
679 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
680 			  struct net_device_path_stack *stack)
681 {
682 	const struct net_device *last_dev;
683 	struct net_device_path_ctx ctx = {
684 		.dev	= dev,
685 	};
686 	struct net_device_path *path;
687 	int ret = 0;
688 
689 	memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
690 	stack->num_paths = 0;
691 	while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
692 		last_dev = ctx.dev;
693 		path = dev_fwd_path(stack);
694 		if (!path)
695 			return -1;
696 
697 		memset(path, 0, sizeof(struct net_device_path));
698 		ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
699 		if (ret < 0)
700 			return -1;
701 
702 		if (WARN_ON_ONCE(last_dev == ctx.dev))
703 			return -1;
704 	}
705 
706 	if (!ctx.dev)
707 		return ret;
708 
709 	path = dev_fwd_path(stack);
710 	if (!path)
711 		return -1;
712 	path->type = DEV_PATH_ETHERNET;
713 	path->dev = ctx.dev;
714 
715 	return ret;
716 }
717 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
718 
719 /**
720  *	__dev_get_by_name	- find a device by its name
721  *	@net: the applicable net namespace
722  *	@name: name to find
723  *
724  *	Find an interface by name. Must be called under RTNL semaphore
725  *	or @dev_base_lock. If the name is found a pointer to the device
726  *	is returned. If the name is not found then %NULL is returned. The
727  *	reference counters are not incremented so the caller must be
728  *	careful with locks.
729  */
730 
731 struct net_device *__dev_get_by_name(struct net *net, const char *name)
732 {
733 	struct netdev_name_node *node_name;
734 
735 	node_name = netdev_name_node_lookup(net, name);
736 	return node_name ? node_name->dev : NULL;
737 }
738 EXPORT_SYMBOL(__dev_get_by_name);
739 
740 /**
741  * dev_get_by_name_rcu	- find a device by its name
742  * @net: the applicable net namespace
743  * @name: name to find
744  *
745  * Find an interface by name.
746  * If the name is found a pointer to the device is returned.
747  * If the name is not found then %NULL is returned.
748  * The reference counters are not incremented so the caller must be
749  * careful with locks. The caller must hold RCU lock.
750  */
751 
752 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
753 {
754 	struct netdev_name_node *node_name;
755 
756 	node_name = netdev_name_node_lookup_rcu(net, name);
757 	return node_name ? node_name->dev : NULL;
758 }
759 EXPORT_SYMBOL(dev_get_by_name_rcu);
760 
761 /**
762  *	dev_get_by_name		- find a device by its name
763  *	@net: the applicable net namespace
764  *	@name: name to find
765  *
766  *	Find an interface by name. This can be called from any
767  *	context and does its own locking. The returned handle has
768  *	the usage count incremented and the caller must use dev_put() to
769  *	release it when it is no longer needed. %NULL is returned if no
770  *	matching device is found.
771  */
772 
773 struct net_device *dev_get_by_name(struct net *net, const char *name)
774 {
775 	struct net_device *dev;
776 
777 	rcu_read_lock();
778 	dev = dev_get_by_name_rcu(net, name);
779 	dev_hold(dev);
780 	rcu_read_unlock();
781 	return dev;
782 }
783 EXPORT_SYMBOL(dev_get_by_name);
784 
785 /**
786  *	__dev_get_by_index - find a device by its ifindex
787  *	@net: the applicable net namespace
788  *	@ifindex: index of device
789  *
790  *	Search for an interface by index. Returns %NULL if the device
791  *	is not found or a pointer to the device. The device has not
792  *	had its reference counter increased so the caller must be careful
793  *	about locking. The caller must hold either the RTNL semaphore
794  *	or @dev_base_lock.
795  */
796 
797 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
798 {
799 	struct net_device *dev;
800 	struct hlist_head *head = dev_index_hash(net, ifindex);
801 
802 	hlist_for_each_entry(dev, head, index_hlist)
803 		if (dev->ifindex == ifindex)
804 			return dev;
805 
806 	return NULL;
807 }
808 EXPORT_SYMBOL(__dev_get_by_index);
809 
810 /**
811  *	dev_get_by_index_rcu - find a device by its ifindex
812  *	@net: the applicable net namespace
813  *	@ifindex: index of device
814  *
815  *	Search for an interface by index. Returns %NULL if the device
816  *	is not found or a pointer to the device. The device has not
817  *	had its reference counter increased so the caller must be careful
818  *	about locking. The caller must hold RCU lock.
819  */
820 
821 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
822 {
823 	struct net_device *dev;
824 	struct hlist_head *head = dev_index_hash(net, ifindex);
825 
826 	hlist_for_each_entry_rcu(dev, head, index_hlist)
827 		if (dev->ifindex == ifindex)
828 			return dev;
829 
830 	return NULL;
831 }
832 EXPORT_SYMBOL(dev_get_by_index_rcu);
833 
834 
835 /**
836  *	dev_get_by_index - find a device by its ifindex
837  *	@net: the applicable net namespace
838  *	@ifindex: index of device
839  *
840  *	Search for an interface by index. Returns NULL if the device
841  *	is not found or a pointer to the device. The device returned has
842  *	had a reference added and the pointer is safe until the user calls
843  *	dev_put to indicate they have finished with it.
844  */
845 
846 struct net_device *dev_get_by_index(struct net *net, int ifindex)
847 {
848 	struct net_device *dev;
849 
850 	rcu_read_lock();
851 	dev = dev_get_by_index_rcu(net, ifindex);
852 	dev_hold(dev);
853 	rcu_read_unlock();
854 	return dev;
855 }
856 EXPORT_SYMBOL(dev_get_by_index);
857 
858 /**
859  *	dev_get_by_napi_id - find a device by napi_id
860  *	@napi_id: ID of the NAPI struct
861  *
862  *	Search for an interface by NAPI ID. Returns %NULL if the device
863  *	is not found or a pointer to the device. The device has not had
864  *	its reference counter increased so the caller must be careful
865  *	about locking. The caller must hold RCU lock.
866  */
867 
868 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
869 {
870 	struct napi_struct *napi;
871 
872 	WARN_ON_ONCE(!rcu_read_lock_held());
873 
874 	if (napi_id < MIN_NAPI_ID)
875 		return NULL;
876 
877 	napi = napi_by_id(napi_id);
878 
879 	return napi ? napi->dev : NULL;
880 }
881 EXPORT_SYMBOL(dev_get_by_napi_id);
882 
883 /**
884  *	netdev_get_name - get a netdevice name, knowing its ifindex.
885  *	@net: network namespace
886  *	@name: a pointer to the buffer where the name will be stored.
887  *	@ifindex: the ifindex of the interface to get the name from.
888  */
889 int netdev_get_name(struct net *net, char *name, int ifindex)
890 {
891 	struct net_device *dev;
892 	int ret;
893 
894 	down_read(&devnet_rename_sem);
895 	rcu_read_lock();
896 
897 	dev = dev_get_by_index_rcu(net, ifindex);
898 	if (!dev) {
899 		ret = -ENODEV;
900 		goto out;
901 	}
902 
903 	strcpy(name, dev->name);
904 
905 	ret = 0;
906 out:
907 	rcu_read_unlock();
908 	up_read(&devnet_rename_sem);
909 	return ret;
910 }
911 
912 /**
913  *	dev_getbyhwaddr_rcu - find a device by its hardware address
914  *	@net: the applicable net namespace
915  *	@type: media type of device
916  *	@ha: hardware address
917  *
918  *	Search for an interface by MAC address. Returns NULL if the device
919  *	is not found or a pointer to the device.
920  *	The caller must hold RCU or RTNL.
921  *	The returned device has not had its ref count increased
922  *	and the caller must therefore be careful about locking
923  *
924  */
925 
926 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
927 				       const char *ha)
928 {
929 	struct net_device *dev;
930 
931 	for_each_netdev_rcu(net, dev)
932 		if (dev->type == type &&
933 		    !memcmp(dev->dev_addr, ha, dev->addr_len))
934 			return dev;
935 
936 	return NULL;
937 }
938 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
939 
940 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
941 {
942 	struct net_device *dev, *ret = NULL;
943 
944 	rcu_read_lock();
945 	for_each_netdev_rcu(net, dev)
946 		if (dev->type == type) {
947 			dev_hold(dev);
948 			ret = dev;
949 			break;
950 		}
951 	rcu_read_unlock();
952 	return ret;
953 }
954 EXPORT_SYMBOL(dev_getfirstbyhwtype);
955 
956 /**
957  *	__dev_get_by_flags - find any device with given flags
958  *	@net: the applicable net namespace
959  *	@if_flags: IFF_* values
960  *	@mask: bitmask of bits in if_flags to check
961  *
962  *	Search for any interface with the given flags. Returns NULL if a device
963  *	is not found or a pointer to the device. Must be called inside
964  *	rtnl_lock(), and result refcount is unchanged.
965  */
966 
967 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
968 				      unsigned short mask)
969 {
970 	struct net_device *dev, *ret;
971 
972 	ASSERT_RTNL();
973 
974 	ret = NULL;
975 	for_each_netdev(net, dev) {
976 		if (((dev->flags ^ if_flags) & mask) == 0) {
977 			ret = dev;
978 			break;
979 		}
980 	}
981 	return ret;
982 }
983 EXPORT_SYMBOL(__dev_get_by_flags);
984 
985 /**
986  *	dev_valid_name - check if name is okay for network device
987  *	@name: name string
988  *
989  *	Network device names need to be valid file names to
990  *	allow sysfs to work.  We also disallow any kind of
991  *	whitespace.
992  */
993 bool dev_valid_name(const char *name)
994 {
995 	if (*name == '\0')
996 		return false;
997 	if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
998 		return false;
999 	if (!strcmp(name, ".") || !strcmp(name, ".."))
1000 		return false;
1001 
1002 	while (*name) {
1003 		if (*name == '/' || *name == ':' || isspace(*name))
1004 			return false;
1005 		name++;
1006 	}
1007 	return true;
1008 }
1009 EXPORT_SYMBOL(dev_valid_name);
1010 
1011 /**
1012  *	__dev_alloc_name - allocate a name for a device
1013  *	@net: network namespace to allocate the device name in
1014  *	@name: name format string
1015  *	@buf:  scratch buffer and result name string
1016  *
1017  *	Passed a format string - eg "lt%d" it will try and find a suitable
1018  *	id. It scans list of devices to build up a free map, then chooses
1019  *	the first empty slot. The caller must hold the dev_base or rtnl lock
1020  *	while allocating the name and adding the device in order to avoid
1021  *	duplicates.
1022  *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1023  *	Returns the number of the unit assigned or a negative errno code.
1024  */
1025 
1026 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1027 {
1028 	int i = 0;
1029 	const char *p;
1030 	const int max_netdevices = 8*PAGE_SIZE;
1031 	unsigned long *inuse;
1032 	struct net_device *d;
1033 
1034 	if (!dev_valid_name(name))
1035 		return -EINVAL;
1036 
1037 	p = strchr(name, '%');
1038 	if (p) {
1039 		/*
1040 		 * Verify the string as this thing may have come from
1041 		 * the user.  There must be either one "%d" and no other "%"
1042 		 * characters.
1043 		 */
1044 		if (p[1] != 'd' || strchr(p + 2, '%'))
1045 			return -EINVAL;
1046 
1047 		/* Use one page as a bit array of possible slots */
1048 		inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1049 		if (!inuse)
1050 			return -ENOMEM;
1051 
1052 		for_each_netdev(net, d) {
1053 			struct netdev_name_node *name_node;
1054 			list_for_each_entry(name_node, &d->name_node->list, list) {
1055 				if (!sscanf(name_node->name, name, &i))
1056 					continue;
1057 				if (i < 0 || i >= max_netdevices)
1058 					continue;
1059 
1060 				/*  avoid cases where sscanf is not exact inverse of printf */
1061 				snprintf(buf, IFNAMSIZ, name, i);
1062 				if (!strncmp(buf, name_node->name, IFNAMSIZ))
1063 					__set_bit(i, inuse);
1064 			}
1065 			if (!sscanf(d->name, name, &i))
1066 				continue;
1067 			if (i < 0 || i >= max_netdevices)
1068 				continue;
1069 
1070 			/*  avoid cases where sscanf is not exact inverse of printf */
1071 			snprintf(buf, IFNAMSIZ, name, i);
1072 			if (!strncmp(buf, d->name, IFNAMSIZ))
1073 				__set_bit(i, inuse);
1074 		}
1075 
1076 		i = find_first_zero_bit(inuse, max_netdevices);
1077 		free_page((unsigned long) inuse);
1078 	}
1079 
1080 	snprintf(buf, IFNAMSIZ, name, i);
1081 	if (!netdev_name_in_use(net, buf))
1082 		return i;
1083 
1084 	/* It is possible to run out of possible slots
1085 	 * when the name is long and there isn't enough space left
1086 	 * for the digits, or if all bits are used.
1087 	 */
1088 	return -ENFILE;
1089 }
1090 
1091 static int dev_alloc_name_ns(struct net *net,
1092 			     struct net_device *dev,
1093 			     const char *name)
1094 {
1095 	char buf[IFNAMSIZ];
1096 	int ret;
1097 
1098 	BUG_ON(!net);
1099 	ret = __dev_alloc_name(net, name, buf);
1100 	if (ret >= 0)
1101 		strlcpy(dev->name, buf, IFNAMSIZ);
1102 	return ret;
1103 }
1104 
1105 /**
1106  *	dev_alloc_name - allocate a name for a device
1107  *	@dev: device
1108  *	@name: name format string
1109  *
1110  *	Passed a format string - eg "lt%d" it will try and find a suitable
1111  *	id. It scans list of devices to build up a free map, then chooses
1112  *	the first empty slot. The caller must hold the dev_base or rtnl lock
1113  *	while allocating the name and adding the device in order to avoid
1114  *	duplicates.
1115  *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1116  *	Returns the number of the unit assigned or a negative errno code.
1117  */
1118 
1119 int dev_alloc_name(struct net_device *dev, const char *name)
1120 {
1121 	return dev_alloc_name_ns(dev_net(dev), dev, name);
1122 }
1123 EXPORT_SYMBOL(dev_alloc_name);
1124 
1125 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1126 			      const char *name)
1127 {
1128 	BUG_ON(!net);
1129 
1130 	if (!dev_valid_name(name))
1131 		return -EINVAL;
1132 
1133 	if (strchr(name, '%'))
1134 		return dev_alloc_name_ns(net, dev, name);
1135 	else if (netdev_name_in_use(net, name))
1136 		return -EEXIST;
1137 	else if (dev->name != name)
1138 		strlcpy(dev->name, name, IFNAMSIZ);
1139 
1140 	return 0;
1141 }
1142 
1143 /**
1144  *	dev_change_name - change name of a device
1145  *	@dev: device
1146  *	@newname: name (or format string) must be at least IFNAMSIZ
1147  *
1148  *	Change name of a device, can pass format strings "eth%d".
1149  *	for wildcarding.
1150  */
1151 int dev_change_name(struct net_device *dev, const char *newname)
1152 {
1153 	unsigned char old_assign_type;
1154 	char oldname[IFNAMSIZ];
1155 	int err = 0;
1156 	int ret;
1157 	struct net *net;
1158 
1159 	ASSERT_RTNL();
1160 	BUG_ON(!dev_net(dev));
1161 
1162 	net = dev_net(dev);
1163 
1164 	/* Some auto-enslaved devices e.g. failover slaves are
1165 	 * special, as userspace might rename the device after
1166 	 * the interface had been brought up and running since
1167 	 * the point kernel initiated auto-enslavement. Allow
1168 	 * live name change even when these slave devices are
1169 	 * up and running.
1170 	 *
1171 	 * Typically, users of these auto-enslaving devices
1172 	 * don't actually care about slave name change, as
1173 	 * they are supposed to operate on master interface
1174 	 * directly.
1175 	 */
1176 	if (dev->flags & IFF_UP &&
1177 	    likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1178 		return -EBUSY;
1179 
1180 	down_write(&devnet_rename_sem);
1181 
1182 	if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1183 		up_write(&devnet_rename_sem);
1184 		return 0;
1185 	}
1186 
1187 	memcpy(oldname, dev->name, IFNAMSIZ);
1188 
1189 	err = dev_get_valid_name(net, dev, newname);
1190 	if (err < 0) {
1191 		up_write(&devnet_rename_sem);
1192 		return err;
1193 	}
1194 
1195 	if (oldname[0] && !strchr(oldname, '%'))
1196 		netdev_info(dev, "renamed from %s\n", oldname);
1197 
1198 	old_assign_type = dev->name_assign_type;
1199 	dev->name_assign_type = NET_NAME_RENAMED;
1200 
1201 rollback:
1202 	ret = device_rename(&dev->dev, dev->name);
1203 	if (ret) {
1204 		memcpy(dev->name, oldname, IFNAMSIZ);
1205 		dev->name_assign_type = old_assign_type;
1206 		up_write(&devnet_rename_sem);
1207 		return ret;
1208 	}
1209 
1210 	up_write(&devnet_rename_sem);
1211 
1212 	netdev_adjacent_rename_links(dev, oldname);
1213 
1214 	write_lock(&dev_base_lock);
1215 	netdev_name_node_del(dev->name_node);
1216 	write_unlock(&dev_base_lock);
1217 
1218 	synchronize_rcu();
1219 
1220 	write_lock(&dev_base_lock);
1221 	netdev_name_node_add(net, dev->name_node);
1222 	write_unlock(&dev_base_lock);
1223 
1224 	ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1225 	ret = notifier_to_errno(ret);
1226 
1227 	if (ret) {
1228 		/* err >= 0 after dev_alloc_name() or stores the first errno */
1229 		if (err >= 0) {
1230 			err = ret;
1231 			down_write(&devnet_rename_sem);
1232 			memcpy(dev->name, oldname, IFNAMSIZ);
1233 			memcpy(oldname, newname, IFNAMSIZ);
1234 			dev->name_assign_type = old_assign_type;
1235 			old_assign_type = NET_NAME_RENAMED;
1236 			goto rollback;
1237 		} else {
1238 			netdev_err(dev, "name change rollback failed: %d\n",
1239 				   ret);
1240 		}
1241 	}
1242 
1243 	return err;
1244 }
1245 
1246 /**
1247  *	dev_set_alias - change ifalias of a device
1248  *	@dev: device
1249  *	@alias: name up to IFALIASZ
1250  *	@len: limit of bytes to copy from info
1251  *
1252  *	Set ifalias for a device,
1253  */
1254 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1255 {
1256 	struct dev_ifalias *new_alias = NULL;
1257 
1258 	if (len >= IFALIASZ)
1259 		return -EINVAL;
1260 
1261 	if (len) {
1262 		new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1263 		if (!new_alias)
1264 			return -ENOMEM;
1265 
1266 		memcpy(new_alias->ifalias, alias, len);
1267 		new_alias->ifalias[len] = 0;
1268 	}
1269 
1270 	mutex_lock(&ifalias_mutex);
1271 	new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1272 					mutex_is_locked(&ifalias_mutex));
1273 	mutex_unlock(&ifalias_mutex);
1274 
1275 	if (new_alias)
1276 		kfree_rcu(new_alias, rcuhead);
1277 
1278 	return len;
1279 }
1280 EXPORT_SYMBOL(dev_set_alias);
1281 
1282 /**
1283  *	dev_get_alias - get ifalias of a device
1284  *	@dev: device
1285  *	@name: buffer to store name of ifalias
1286  *	@len: size of buffer
1287  *
1288  *	get ifalias for a device.  Caller must make sure dev cannot go
1289  *	away,  e.g. rcu read lock or own a reference count to device.
1290  */
1291 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1292 {
1293 	const struct dev_ifalias *alias;
1294 	int ret = 0;
1295 
1296 	rcu_read_lock();
1297 	alias = rcu_dereference(dev->ifalias);
1298 	if (alias)
1299 		ret = snprintf(name, len, "%s", alias->ifalias);
1300 	rcu_read_unlock();
1301 
1302 	return ret;
1303 }
1304 
1305 /**
1306  *	netdev_features_change - device changes features
1307  *	@dev: device to cause notification
1308  *
1309  *	Called to indicate a device has changed features.
1310  */
1311 void netdev_features_change(struct net_device *dev)
1312 {
1313 	call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1314 }
1315 EXPORT_SYMBOL(netdev_features_change);
1316 
1317 /**
1318  *	netdev_state_change - device changes state
1319  *	@dev: device to cause notification
1320  *
1321  *	Called to indicate a device has changed state. This function calls
1322  *	the notifier chains for netdev_chain and sends a NEWLINK message
1323  *	to the routing socket.
1324  */
1325 void netdev_state_change(struct net_device *dev)
1326 {
1327 	if (dev->flags & IFF_UP) {
1328 		struct netdev_notifier_change_info change_info = {
1329 			.info.dev = dev,
1330 		};
1331 
1332 		call_netdevice_notifiers_info(NETDEV_CHANGE,
1333 					      &change_info.info);
1334 		rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1335 	}
1336 }
1337 EXPORT_SYMBOL(netdev_state_change);
1338 
1339 /**
1340  * __netdev_notify_peers - notify network peers about existence of @dev,
1341  * to be called when rtnl lock is already held.
1342  * @dev: network device
1343  *
1344  * Generate traffic such that interested network peers are aware of
1345  * @dev, such as by generating a gratuitous ARP. This may be used when
1346  * a device wants to inform the rest of the network about some sort of
1347  * reconfiguration such as a failover event or virtual machine
1348  * migration.
1349  */
1350 void __netdev_notify_peers(struct net_device *dev)
1351 {
1352 	ASSERT_RTNL();
1353 	call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1354 	call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1355 }
1356 EXPORT_SYMBOL(__netdev_notify_peers);
1357 
1358 /**
1359  * netdev_notify_peers - notify network peers about existence of @dev
1360  * @dev: network device
1361  *
1362  * Generate traffic such that interested network peers are aware of
1363  * @dev, such as by generating a gratuitous ARP. This may be used when
1364  * a device wants to inform the rest of the network about some sort of
1365  * reconfiguration such as a failover event or virtual machine
1366  * migration.
1367  */
1368 void netdev_notify_peers(struct net_device *dev)
1369 {
1370 	rtnl_lock();
1371 	__netdev_notify_peers(dev);
1372 	rtnl_unlock();
1373 }
1374 EXPORT_SYMBOL(netdev_notify_peers);
1375 
1376 static int napi_threaded_poll(void *data);
1377 
1378 static int napi_kthread_create(struct napi_struct *n)
1379 {
1380 	int err = 0;
1381 
1382 	/* Create and wake up the kthread once to put it in
1383 	 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1384 	 * warning and work with loadavg.
1385 	 */
1386 	n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1387 				n->dev->name, n->napi_id);
1388 	if (IS_ERR(n->thread)) {
1389 		err = PTR_ERR(n->thread);
1390 		pr_err("kthread_run failed with err %d\n", err);
1391 		n->thread = NULL;
1392 	}
1393 
1394 	return err;
1395 }
1396 
1397 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1398 {
1399 	const struct net_device_ops *ops = dev->netdev_ops;
1400 	int ret;
1401 
1402 	ASSERT_RTNL();
1403 	dev_addr_check(dev);
1404 
1405 	if (!netif_device_present(dev)) {
1406 		/* may be detached because parent is runtime-suspended */
1407 		if (dev->dev.parent)
1408 			pm_runtime_resume(dev->dev.parent);
1409 		if (!netif_device_present(dev))
1410 			return -ENODEV;
1411 	}
1412 
1413 	/* Block netpoll from trying to do any rx path servicing.
1414 	 * If we don't do this there is a chance ndo_poll_controller
1415 	 * or ndo_poll may be running while we open the device
1416 	 */
1417 	netpoll_poll_disable(dev);
1418 
1419 	ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1420 	ret = notifier_to_errno(ret);
1421 	if (ret)
1422 		return ret;
1423 
1424 	set_bit(__LINK_STATE_START, &dev->state);
1425 
1426 	if (ops->ndo_validate_addr)
1427 		ret = ops->ndo_validate_addr(dev);
1428 
1429 	if (!ret && ops->ndo_open)
1430 		ret = ops->ndo_open(dev);
1431 
1432 	netpoll_poll_enable(dev);
1433 
1434 	if (ret)
1435 		clear_bit(__LINK_STATE_START, &dev->state);
1436 	else {
1437 		dev->flags |= IFF_UP;
1438 		dev_set_rx_mode(dev);
1439 		dev_activate(dev);
1440 		add_device_randomness(dev->dev_addr, dev->addr_len);
1441 	}
1442 
1443 	return ret;
1444 }
1445 
1446 /**
1447  *	dev_open	- prepare an interface for use.
1448  *	@dev: device to open
1449  *	@extack: netlink extended ack
1450  *
1451  *	Takes a device from down to up state. The device's private open
1452  *	function is invoked and then the multicast lists are loaded. Finally
1453  *	the device is moved into the up state and a %NETDEV_UP message is
1454  *	sent to the netdev notifier chain.
1455  *
1456  *	Calling this function on an active interface is a nop. On a failure
1457  *	a negative errno code is returned.
1458  */
1459 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1460 {
1461 	int ret;
1462 
1463 	if (dev->flags & IFF_UP)
1464 		return 0;
1465 
1466 	ret = __dev_open(dev, extack);
1467 	if (ret < 0)
1468 		return ret;
1469 
1470 	rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1471 	call_netdevice_notifiers(NETDEV_UP, dev);
1472 
1473 	return ret;
1474 }
1475 EXPORT_SYMBOL(dev_open);
1476 
1477 static void __dev_close_many(struct list_head *head)
1478 {
1479 	struct net_device *dev;
1480 
1481 	ASSERT_RTNL();
1482 	might_sleep();
1483 
1484 	list_for_each_entry(dev, head, close_list) {
1485 		/* Temporarily disable netpoll until the interface is down */
1486 		netpoll_poll_disable(dev);
1487 
1488 		call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1489 
1490 		clear_bit(__LINK_STATE_START, &dev->state);
1491 
1492 		/* Synchronize to scheduled poll. We cannot touch poll list, it
1493 		 * can be even on different cpu. So just clear netif_running().
1494 		 *
1495 		 * dev->stop() will invoke napi_disable() on all of it's
1496 		 * napi_struct instances on this device.
1497 		 */
1498 		smp_mb__after_atomic(); /* Commit netif_running(). */
1499 	}
1500 
1501 	dev_deactivate_many(head);
1502 
1503 	list_for_each_entry(dev, head, close_list) {
1504 		const struct net_device_ops *ops = dev->netdev_ops;
1505 
1506 		/*
1507 		 *	Call the device specific close. This cannot fail.
1508 		 *	Only if device is UP
1509 		 *
1510 		 *	We allow it to be called even after a DETACH hot-plug
1511 		 *	event.
1512 		 */
1513 		if (ops->ndo_stop)
1514 			ops->ndo_stop(dev);
1515 
1516 		dev->flags &= ~IFF_UP;
1517 		netpoll_poll_enable(dev);
1518 	}
1519 }
1520 
1521 static void __dev_close(struct net_device *dev)
1522 {
1523 	LIST_HEAD(single);
1524 
1525 	list_add(&dev->close_list, &single);
1526 	__dev_close_many(&single);
1527 	list_del(&single);
1528 }
1529 
1530 void dev_close_many(struct list_head *head, bool unlink)
1531 {
1532 	struct net_device *dev, *tmp;
1533 
1534 	/* Remove the devices that don't need to be closed */
1535 	list_for_each_entry_safe(dev, tmp, head, close_list)
1536 		if (!(dev->flags & IFF_UP))
1537 			list_del_init(&dev->close_list);
1538 
1539 	__dev_close_many(head);
1540 
1541 	list_for_each_entry_safe(dev, tmp, head, close_list) {
1542 		rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1543 		call_netdevice_notifiers(NETDEV_DOWN, dev);
1544 		if (unlink)
1545 			list_del_init(&dev->close_list);
1546 	}
1547 }
1548 EXPORT_SYMBOL(dev_close_many);
1549 
1550 /**
1551  *	dev_close - shutdown an interface.
1552  *	@dev: device to shutdown
1553  *
1554  *	This function moves an active device into down state. A
1555  *	%NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1556  *	is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1557  *	chain.
1558  */
1559 void dev_close(struct net_device *dev)
1560 {
1561 	if (dev->flags & IFF_UP) {
1562 		LIST_HEAD(single);
1563 
1564 		list_add(&dev->close_list, &single);
1565 		dev_close_many(&single, true);
1566 		list_del(&single);
1567 	}
1568 }
1569 EXPORT_SYMBOL(dev_close);
1570 
1571 
1572 /**
1573  *	dev_disable_lro - disable Large Receive Offload on a device
1574  *	@dev: device
1575  *
1576  *	Disable Large Receive Offload (LRO) on a net device.  Must be
1577  *	called under RTNL.  This is needed if received packets may be
1578  *	forwarded to another interface.
1579  */
1580 void dev_disable_lro(struct net_device *dev)
1581 {
1582 	struct net_device *lower_dev;
1583 	struct list_head *iter;
1584 
1585 	dev->wanted_features &= ~NETIF_F_LRO;
1586 	netdev_update_features(dev);
1587 
1588 	if (unlikely(dev->features & NETIF_F_LRO))
1589 		netdev_WARN(dev, "failed to disable LRO!\n");
1590 
1591 	netdev_for_each_lower_dev(dev, lower_dev, iter)
1592 		dev_disable_lro(lower_dev);
1593 }
1594 EXPORT_SYMBOL(dev_disable_lro);
1595 
1596 /**
1597  *	dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1598  *	@dev: device
1599  *
1600  *	Disable HW Generic Receive Offload (GRO_HW) on a net device.  Must be
1601  *	called under RTNL.  This is needed if Generic XDP is installed on
1602  *	the device.
1603  */
1604 static void dev_disable_gro_hw(struct net_device *dev)
1605 {
1606 	dev->wanted_features &= ~NETIF_F_GRO_HW;
1607 	netdev_update_features(dev);
1608 
1609 	if (unlikely(dev->features & NETIF_F_GRO_HW))
1610 		netdev_WARN(dev, "failed to disable GRO_HW!\n");
1611 }
1612 
1613 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1614 {
1615 #define N(val) 						\
1616 	case NETDEV_##val:				\
1617 		return "NETDEV_" __stringify(val);
1618 	switch (cmd) {
1619 	N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1620 	N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1621 	N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1622 	N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1623 	N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1624 	N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1625 	N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1626 	N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1627 	N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1628 	N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE)
1629 	N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA)
1630 	}
1631 #undef N
1632 	return "UNKNOWN_NETDEV_EVENT";
1633 }
1634 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1635 
1636 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1637 				   struct net_device *dev)
1638 {
1639 	struct netdev_notifier_info info = {
1640 		.dev = dev,
1641 	};
1642 
1643 	return nb->notifier_call(nb, val, &info);
1644 }
1645 
1646 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1647 					     struct net_device *dev)
1648 {
1649 	int err;
1650 
1651 	err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1652 	err = notifier_to_errno(err);
1653 	if (err)
1654 		return err;
1655 
1656 	if (!(dev->flags & IFF_UP))
1657 		return 0;
1658 
1659 	call_netdevice_notifier(nb, NETDEV_UP, dev);
1660 	return 0;
1661 }
1662 
1663 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1664 						struct net_device *dev)
1665 {
1666 	if (dev->flags & IFF_UP) {
1667 		call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1668 					dev);
1669 		call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1670 	}
1671 	call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1672 }
1673 
1674 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1675 						 struct net *net)
1676 {
1677 	struct net_device *dev;
1678 	int err;
1679 
1680 	for_each_netdev(net, dev) {
1681 		err = call_netdevice_register_notifiers(nb, dev);
1682 		if (err)
1683 			goto rollback;
1684 	}
1685 	return 0;
1686 
1687 rollback:
1688 	for_each_netdev_continue_reverse(net, dev)
1689 		call_netdevice_unregister_notifiers(nb, dev);
1690 	return err;
1691 }
1692 
1693 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1694 						    struct net *net)
1695 {
1696 	struct net_device *dev;
1697 
1698 	for_each_netdev(net, dev)
1699 		call_netdevice_unregister_notifiers(nb, dev);
1700 }
1701 
1702 static int dev_boot_phase = 1;
1703 
1704 /**
1705  * register_netdevice_notifier - register a network notifier block
1706  * @nb: notifier
1707  *
1708  * Register a notifier to be called when network device events occur.
1709  * The notifier passed is linked into the kernel structures and must
1710  * not be reused until it has been unregistered. A negative errno code
1711  * is returned on a failure.
1712  *
1713  * When registered all registration and up events are replayed
1714  * to the new notifier to allow device to have a race free
1715  * view of the network device list.
1716  */
1717 
1718 int register_netdevice_notifier(struct notifier_block *nb)
1719 {
1720 	struct net *net;
1721 	int err;
1722 
1723 	/* Close race with setup_net() and cleanup_net() */
1724 	down_write(&pernet_ops_rwsem);
1725 	rtnl_lock();
1726 	err = raw_notifier_chain_register(&netdev_chain, nb);
1727 	if (err)
1728 		goto unlock;
1729 	if (dev_boot_phase)
1730 		goto unlock;
1731 	for_each_net(net) {
1732 		err = call_netdevice_register_net_notifiers(nb, net);
1733 		if (err)
1734 			goto rollback;
1735 	}
1736 
1737 unlock:
1738 	rtnl_unlock();
1739 	up_write(&pernet_ops_rwsem);
1740 	return err;
1741 
1742 rollback:
1743 	for_each_net_continue_reverse(net)
1744 		call_netdevice_unregister_net_notifiers(nb, net);
1745 
1746 	raw_notifier_chain_unregister(&netdev_chain, nb);
1747 	goto unlock;
1748 }
1749 EXPORT_SYMBOL(register_netdevice_notifier);
1750 
1751 /**
1752  * unregister_netdevice_notifier - unregister a network notifier block
1753  * @nb: notifier
1754  *
1755  * Unregister a notifier previously registered by
1756  * register_netdevice_notifier(). The notifier is unlinked into the
1757  * kernel structures and may then be reused. A negative errno code
1758  * is returned on a failure.
1759  *
1760  * After unregistering unregister and down device events are synthesized
1761  * for all devices on the device list to the removed notifier to remove
1762  * the need for special case cleanup code.
1763  */
1764 
1765 int unregister_netdevice_notifier(struct notifier_block *nb)
1766 {
1767 	struct net *net;
1768 	int err;
1769 
1770 	/* Close race with setup_net() and cleanup_net() */
1771 	down_write(&pernet_ops_rwsem);
1772 	rtnl_lock();
1773 	err = raw_notifier_chain_unregister(&netdev_chain, nb);
1774 	if (err)
1775 		goto unlock;
1776 
1777 	for_each_net(net)
1778 		call_netdevice_unregister_net_notifiers(nb, net);
1779 
1780 unlock:
1781 	rtnl_unlock();
1782 	up_write(&pernet_ops_rwsem);
1783 	return err;
1784 }
1785 EXPORT_SYMBOL(unregister_netdevice_notifier);
1786 
1787 static int __register_netdevice_notifier_net(struct net *net,
1788 					     struct notifier_block *nb,
1789 					     bool ignore_call_fail)
1790 {
1791 	int err;
1792 
1793 	err = raw_notifier_chain_register(&net->netdev_chain, nb);
1794 	if (err)
1795 		return err;
1796 	if (dev_boot_phase)
1797 		return 0;
1798 
1799 	err = call_netdevice_register_net_notifiers(nb, net);
1800 	if (err && !ignore_call_fail)
1801 		goto chain_unregister;
1802 
1803 	return 0;
1804 
1805 chain_unregister:
1806 	raw_notifier_chain_unregister(&net->netdev_chain, nb);
1807 	return err;
1808 }
1809 
1810 static int __unregister_netdevice_notifier_net(struct net *net,
1811 					       struct notifier_block *nb)
1812 {
1813 	int err;
1814 
1815 	err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1816 	if (err)
1817 		return err;
1818 
1819 	call_netdevice_unregister_net_notifiers(nb, net);
1820 	return 0;
1821 }
1822 
1823 /**
1824  * register_netdevice_notifier_net - register a per-netns network notifier block
1825  * @net: network namespace
1826  * @nb: notifier
1827  *
1828  * Register a notifier to be called when network device events occur.
1829  * The notifier passed is linked into the kernel structures and must
1830  * not be reused until it has been unregistered. A negative errno code
1831  * is returned on a failure.
1832  *
1833  * When registered all registration and up events are replayed
1834  * to the new notifier to allow device to have a race free
1835  * view of the network device list.
1836  */
1837 
1838 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1839 {
1840 	int err;
1841 
1842 	rtnl_lock();
1843 	err = __register_netdevice_notifier_net(net, nb, false);
1844 	rtnl_unlock();
1845 	return err;
1846 }
1847 EXPORT_SYMBOL(register_netdevice_notifier_net);
1848 
1849 /**
1850  * unregister_netdevice_notifier_net - unregister a per-netns
1851  *                                     network notifier block
1852  * @net: network namespace
1853  * @nb: notifier
1854  *
1855  * Unregister a notifier previously registered by
1856  * register_netdevice_notifier(). The notifier is unlinked into the
1857  * kernel structures and may then be reused. A negative errno code
1858  * is returned on a failure.
1859  *
1860  * After unregistering unregister and down device events are synthesized
1861  * for all devices on the device list to the removed notifier to remove
1862  * the need for special case cleanup code.
1863  */
1864 
1865 int unregister_netdevice_notifier_net(struct net *net,
1866 				      struct notifier_block *nb)
1867 {
1868 	int err;
1869 
1870 	rtnl_lock();
1871 	err = __unregister_netdevice_notifier_net(net, nb);
1872 	rtnl_unlock();
1873 	return err;
1874 }
1875 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1876 
1877 int register_netdevice_notifier_dev_net(struct net_device *dev,
1878 					struct notifier_block *nb,
1879 					struct netdev_net_notifier *nn)
1880 {
1881 	int err;
1882 
1883 	rtnl_lock();
1884 	err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1885 	if (!err) {
1886 		nn->nb = nb;
1887 		list_add(&nn->list, &dev->net_notifier_list);
1888 	}
1889 	rtnl_unlock();
1890 	return err;
1891 }
1892 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1893 
1894 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1895 					  struct notifier_block *nb,
1896 					  struct netdev_net_notifier *nn)
1897 {
1898 	int err;
1899 
1900 	rtnl_lock();
1901 	list_del(&nn->list);
1902 	err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1903 	rtnl_unlock();
1904 	return err;
1905 }
1906 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1907 
1908 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1909 					     struct net *net)
1910 {
1911 	struct netdev_net_notifier *nn;
1912 
1913 	list_for_each_entry(nn, &dev->net_notifier_list, list) {
1914 		__unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
1915 		__register_netdevice_notifier_net(net, nn->nb, true);
1916 	}
1917 }
1918 
1919 /**
1920  *	call_netdevice_notifiers_info - call all network notifier blocks
1921  *	@val: value passed unmodified to notifier function
1922  *	@info: notifier information data
1923  *
1924  *	Call all network notifier blocks.  Parameters and return value
1925  *	are as for raw_notifier_call_chain().
1926  */
1927 
1928 static int call_netdevice_notifiers_info(unsigned long val,
1929 					 struct netdev_notifier_info *info)
1930 {
1931 	struct net *net = dev_net(info->dev);
1932 	int ret;
1933 
1934 	ASSERT_RTNL();
1935 
1936 	/* Run per-netns notifier block chain first, then run the global one.
1937 	 * Hopefully, one day, the global one is going to be removed after
1938 	 * all notifier block registrators get converted to be per-netns.
1939 	 */
1940 	ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1941 	if (ret & NOTIFY_STOP_MASK)
1942 		return ret;
1943 	return raw_notifier_call_chain(&netdev_chain, val, info);
1944 }
1945 
1946 /**
1947  *	call_netdevice_notifiers_info_robust - call per-netns notifier blocks
1948  *	                                       for and rollback on error
1949  *	@val_up: value passed unmodified to notifier function
1950  *	@val_down: value passed unmodified to the notifier function when
1951  *	           recovering from an error on @val_up
1952  *	@info: notifier information data
1953  *
1954  *	Call all per-netns network notifier blocks, but not notifier blocks on
1955  *	the global notifier chain. Parameters and return value are as for
1956  *	raw_notifier_call_chain_robust().
1957  */
1958 
1959 static int
1960 call_netdevice_notifiers_info_robust(unsigned long val_up,
1961 				     unsigned long val_down,
1962 				     struct netdev_notifier_info *info)
1963 {
1964 	struct net *net = dev_net(info->dev);
1965 
1966 	ASSERT_RTNL();
1967 
1968 	return raw_notifier_call_chain_robust(&net->netdev_chain,
1969 					      val_up, val_down, info);
1970 }
1971 
1972 static int call_netdevice_notifiers_extack(unsigned long val,
1973 					   struct net_device *dev,
1974 					   struct netlink_ext_ack *extack)
1975 {
1976 	struct netdev_notifier_info info = {
1977 		.dev = dev,
1978 		.extack = extack,
1979 	};
1980 
1981 	return call_netdevice_notifiers_info(val, &info);
1982 }
1983 
1984 /**
1985  *	call_netdevice_notifiers - call all network notifier blocks
1986  *      @val: value passed unmodified to notifier function
1987  *      @dev: net_device pointer passed unmodified to notifier function
1988  *
1989  *	Call all network notifier blocks.  Parameters and return value
1990  *	are as for raw_notifier_call_chain().
1991  */
1992 
1993 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1994 {
1995 	return call_netdevice_notifiers_extack(val, dev, NULL);
1996 }
1997 EXPORT_SYMBOL(call_netdevice_notifiers);
1998 
1999 /**
2000  *	call_netdevice_notifiers_mtu - call all network notifier blocks
2001  *	@val: value passed unmodified to notifier function
2002  *	@dev: net_device pointer passed unmodified to notifier function
2003  *	@arg: additional u32 argument passed to the notifier function
2004  *
2005  *	Call all network notifier blocks.  Parameters and return value
2006  *	are as for raw_notifier_call_chain().
2007  */
2008 static int call_netdevice_notifiers_mtu(unsigned long val,
2009 					struct net_device *dev, u32 arg)
2010 {
2011 	struct netdev_notifier_info_ext info = {
2012 		.info.dev = dev,
2013 		.ext.mtu = arg,
2014 	};
2015 
2016 	BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2017 
2018 	return call_netdevice_notifiers_info(val, &info.info);
2019 }
2020 
2021 #ifdef CONFIG_NET_INGRESS
2022 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2023 
2024 void net_inc_ingress_queue(void)
2025 {
2026 	static_branch_inc(&ingress_needed_key);
2027 }
2028 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2029 
2030 void net_dec_ingress_queue(void)
2031 {
2032 	static_branch_dec(&ingress_needed_key);
2033 }
2034 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2035 #endif
2036 
2037 #ifdef CONFIG_NET_EGRESS
2038 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2039 
2040 void net_inc_egress_queue(void)
2041 {
2042 	static_branch_inc(&egress_needed_key);
2043 }
2044 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2045 
2046 void net_dec_egress_queue(void)
2047 {
2048 	static_branch_dec(&egress_needed_key);
2049 }
2050 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2051 #endif
2052 
2053 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2054 EXPORT_SYMBOL(netstamp_needed_key);
2055 #ifdef CONFIG_JUMP_LABEL
2056 static atomic_t netstamp_needed_deferred;
2057 static atomic_t netstamp_wanted;
2058 static void netstamp_clear(struct work_struct *work)
2059 {
2060 	int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2061 	int wanted;
2062 
2063 	wanted = atomic_add_return(deferred, &netstamp_wanted);
2064 	if (wanted > 0)
2065 		static_branch_enable(&netstamp_needed_key);
2066 	else
2067 		static_branch_disable(&netstamp_needed_key);
2068 }
2069 static DECLARE_WORK(netstamp_work, netstamp_clear);
2070 #endif
2071 
2072 void net_enable_timestamp(void)
2073 {
2074 #ifdef CONFIG_JUMP_LABEL
2075 	int wanted;
2076 
2077 	while (1) {
2078 		wanted = atomic_read(&netstamp_wanted);
2079 		if (wanted <= 0)
2080 			break;
2081 		if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2082 			return;
2083 	}
2084 	atomic_inc(&netstamp_needed_deferred);
2085 	schedule_work(&netstamp_work);
2086 #else
2087 	static_branch_inc(&netstamp_needed_key);
2088 #endif
2089 }
2090 EXPORT_SYMBOL(net_enable_timestamp);
2091 
2092 void net_disable_timestamp(void)
2093 {
2094 #ifdef CONFIG_JUMP_LABEL
2095 	int wanted;
2096 
2097 	while (1) {
2098 		wanted = atomic_read(&netstamp_wanted);
2099 		if (wanted <= 1)
2100 			break;
2101 		if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2102 			return;
2103 	}
2104 	atomic_dec(&netstamp_needed_deferred);
2105 	schedule_work(&netstamp_work);
2106 #else
2107 	static_branch_dec(&netstamp_needed_key);
2108 #endif
2109 }
2110 EXPORT_SYMBOL(net_disable_timestamp);
2111 
2112 static inline void net_timestamp_set(struct sk_buff *skb)
2113 {
2114 	skb->tstamp = 0;
2115 	skb->mono_delivery_time = 0;
2116 	if (static_branch_unlikely(&netstamp_needed_key))
2117 		skb->tstamp = ktime_get_real();
2118 }
2119 
2120 #define net_timestamp_check(COND, SKB)				\
2121 	if (static_branch_unlikely(&netstamp_needed_key)) {	\
2122 		if ((COND) && !(SKB)->tstamp)			\
2123 			(SKB)->tstamp = ktime_get_real();	\
2124 	}							\
2125 
2126 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2127 {
2128 	return __is_skb_forwardable(dev, skb, true);
2129 }
2130 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2131 
2132 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2133 			      bool check_mtu)
2134 {
2135 	int ret = ____dev_forward_skb(dev, skb, check_mtu);
2136 
2137 	if (likely(!ret)) {
2138 		skb->protocol = eth_type_trans(skb, dev);
2139 		skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2140 	}
2141 
2142 	return ret;
2143 }
2144 
2145 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2146 {
2147 	return __dev_forward_skb2(dev, skb, true);
2148 }
2149 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2150 
2151 /**
2152  * dev_forward_skb - loopback an skb to another netif
2153  *
2154  * @dev: destination network device
2155  * @skb: buffer to forward
2156  *
2157  * return values:
2158  *	NET_RX_SUCCESS	(no congestion)
2159  *	NET_RX_DROP     (packet was dropped, but freed)
2160  *
2161  * dev_forward_skb can be used for injecting an skb from the
2162  * start_xmit function of one device into the receive queue
2163  * of another device.
2164  *
2165  * The receiving device may be in another namespace, so
2166  * we have to clear all information in the skb that could
2167  * impact namespace isolation.
2168  */
2169 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2170 {
2171 	return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2172 }
2173 EXPORT_SYMBOL_GPL(dev_forward_skb);
2174 
2175 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2176 {
2177 	return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2178 }
2179 
2180 static inline int deliver_skb(struct sk_buff *skb,
2181 			      struct packet_type *pt_prev,
2182 			      struct net_device *orig_dev)
2183 {
2184 	if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2185 		return -ENOMEM;
2186 	refcount_inc(&skb->users);
2187 	return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2188 }
2189 
2190 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2191 					  struct packet_type **pt,
2192 					  struct net_device *orig_dev,
2193 					  __be16 type,
2194 					  struct list_head *ptype_list)
2195 {
2196 	struct packet_type *ptype, *pt_prev = *pt;
2197 
2198 	list_for_each_entry_rcu(ptype, ptype_list, list) {
2199 		if (ptype->type != type)
2200 			continue;
2201 		if (pt_prev)
2202 			deliver_skb(skb, pt_prev, orig_dev);
2203 		pt_prev = ptype;
2204 	}
2205 	*pt = pt_prev;
2206 }
2207 
2208 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2209 {
2210 	if (!ptype->af_packet_priv || !skb->sk)
2211 		return false;
2212 
2213 	if (ptype->id_match)
2214 		return ptype->id_match(ptype, skb->sk);
2215 	else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2216 		return true;
2217 
2218 	return false;
2219 }
2220 
2221 /**
2222  * dev_nit_active - return true if any network interface taps are in use
2223  *
2224  * @dev: network device to check for the presence of taps
2225  */
2226 bool dev_nit_active(struct net_device *dev)
2227 {
2228 	return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2229 }
2230 EXPORT_SYMBOL_GPL(dev_nit_active);
2231 
2232 /*
2233  *	Support routine. Sends outgoing frames to any network
2234  *	taps currently in use.
2235  */
2236 
2237 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2238 {
2239 	struct packet_type *ptype;
2240 	struct sk_buff *skb2 = NULL;
2241 	struct packet_type *pt_prev = NULL;
2242 	struct list_head *ptype_list = &ptype_all;
2243 
2244 	rcu_read_lock();
2245 again:
2246 	list_for_each_entry_rcu(ptype, ptype_list, list) {
2247 		if (ptype->ignore_outgoing)
2248 			continue;
2249 
2250 		/* Never send packets back to the socket
2251 		 * they originated from - MvS (miquels@drinkel.ow.org)
2252 		 */
2253 		if (skb_loop_sk(ptype, skb))
2254 			continue;
2255 
2256 		if (pt_prev) {
2257 			deliver_skb(skb2, pt_prev, skb->dev);
2258 			pt_prev = ptype;
2259 			continue;
2260 		}
2261 
2262 		/* need to clone skb, done only once */
2263 		skb2 = skb_clone(skb, GFP_ATOMIC);
2264 		if (!skb2)
2265 			goto out_unlock;
2266 
2267 		net_timestamp_set(skb2);
2268 
2269 		/* skb->nh should be correctly
2270 		 * set by sender, so that the second statement is
2271 		 * just protection against buggy protocols.
2272 		 */
2273 		skb_reset_mac_header(skb2);
2274 
2275 		if (skb_network_header(skb2) < skb2->data ||
2276 		    skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2277 			net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2278 					     ntohs(skb2->protocol),
2279 					     dev->name);
2280 			skb_reset_network_header(skb2);
2281 		}
2282 
2283 		skb2->transport_header = skb2->network_header;
2284 		skb2->pkt_type = PACKET_OUTGOING;
2285 		pt_prev = ptype;
2286 	}
2287 
2288 	if (ptype_list == &ptype_all) {
2289 		ptype_list = &dev->ptype_all;
2290 		goto again;
2291 	}
2292 out_unlock:
2293 	if (pt_prev) {
2294 		if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2295 			pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2296 		else
2297 			kfree_skb(skb2);
2298 	}
2299 	rcu_read_unlock();
2300 }
2301 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2302 
2303 /**
2304  * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2305  * @dev: Network device
2306  * @txq: number of queues available
2307  *
2308  * If real_num_tx_queues is changed the tc mappings may no longer be
2309  * valid. To resolve this verify the tc mapping remains valid and if
2310  * not NULL the mapping. With no priorities mapping to this
2311  * offset/count pair it will no longer be used. In the worst case TC0
2312  * is invalid nothing can be done so disable priority mappings. If is
2313  * expected that drivers will fix this mapping if they can before
2314  * calling netif_set_real_num_tx_queues.
2315  */
2316 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2317 {
2318 	int i;
2319 	struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2320 
2321 	/* If TC0 is invalidated disable TC mapping */
2322 	if (tc->offset + tc->count > txq) {
2323 		netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2324 		dev->num_tc = 0;
2325 		return;
2326 	}
2327 
2328 	/* Invalidated prio to tc mappings set to TC0 */
2329 	for (i = 1; i < TC_BITMASK + 1; i++) {
2330 		int q = netdev_get_prio_tc_map(dev, i);
2331 
2332 		tc = &dev->tc_to_txq[q];
2333 		if (tc->offset + tc->count > txq) {
2334 			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",
2335 				    i, q);
2336 			netdev_set_prio_tc_map(dev, i, 0);
2337 		}
2338 	}
2339 }
2340 
2341 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2342 {
2343 	if (dev->num_tc) {
2344 		struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2345 		int i;
2346 
2347 		/* walk through the TCs and see if it falls into any of them */
2348 		for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2349 			if ((txq - tc->offset) < tc->count)
2350 				return i;
2351 		}
2352 
2353 		/* didn't find it, just return -1 to indicate no match */
2354 		return -1;
2355 	}
2356 
2357 	return 0;
2358 }
2359 EXPORT_SYMBOL(netdev_txq_to_tc);
2360 
2361 #ifdef CONFIG_XPS
2362 static struct static_key xps_needed __read_mostly;
2363 static struct static_key xps_rxqs_needed __read_mostly;
2364 static DEFINE_MUTEX(xps_map_mutex);
2365 #define xmap_dereference(P)		\
2366 	rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2367 
2368 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2369 			     struct xps_dev_maps *old_maps, int tci, u16 index)
2370 {
2371 	struct xps_map *map = NULL;
2372 	int pos;
2373 
2374 	if (dev_maps)
2375 		map = xmap_dereference(dev_maps->attr_map[tci]);
2376 	if (!map)
2377 		return false;
2378 
2379 	for (pos = map->len; pos--;) {
2380 		if (map->queues[pos] != index)
2381 			continue;
2382 
2383 		if (map->len > 1) {
2384 			map->queues[pos] = map->queues[--map->len];
2385 			break;
2386 		}
2387 
2388 		if (old_maps)
2389 			RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2390 		RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2391 		kfree_rcu(map, rcu);
2392 		return false;
2393 	}
2394 
2395 	return true;
2396 }
2397 
2398 static bool remove_xps_queue_cpu(struct net_device *dev,
2399 				 struct xps_dev_maps *dev_maps,
2400 				 int cpu, u16 offset, u16 count)
2401 {
2402 	int num_tc = dev_maps->num_tc;
2403 	bool active = false;
2404 	int tci;
2405 
2406 	for (tci = cpu * num_tc; num_tc--; tci++) {
2407 		int i, j;
2408 
2409 		for (i = count, j = offset; i--; j++) {
2410 			if (!remove_xps_queue(dev_maps, NULL, tci, j))
2411 				break;
2412 		}
2413 
2414 		active |= i < 0;
2415 	}
2416 
2417 	return active;
2418 }
2419 
2420 static void reset_xps_maps(struct net_device *dev,
2421 			   struct xps_dev_maps *dev_maps,
2422 			   enum xps_map_type type)
2423 {
2424 	static_key_slow_dec_cpuslocked(&xps_needed);
2425 	if (type == XPS_RXQS)
2426 		static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2427 
2428 	RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2429 
2430 	kfree_rcu(dev_maps, rcu);
2431 }
2432 
2433 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2434 			   u16 offset, u16 count)
2435 {
2436 	struct xps_dev_maps *dev_maps;
2437 	bool active = false;
2438 	int i, j;
2439 
2440 	dev_maps = xmap_dereference(dev->xps_maps[type]);
2441 	if (!dev_maps)
2442 		return;
2443 
2444 	for (j = 0; j < dev_maps->nr_ids; j++)
2445 		active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2446 	if (!active)
2447 		reset_xps_maps(dev, dev_maps, type);
2448 
2449 	if (type == XPS_CPUS) {
2450 		for (i = offset + (count - 1); count--; i--)
2451 			netdev_queue_numa_node_write(
2452 				netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2453 	}
2454 }
2455 
2456 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2457 				   u16 count)
2458 {
2459 	if (!static_key_false(&xps_needed))
2460 		return;
2461 
2462 	cpus_read_lock();
2463 	mutex_lock(&xps_map_mutex);
2464 
2465 	if (static_key_false(&xps_rxqs_needed))
2466 		clean_xps_maps(dev, XPS_RXQS, offset, count);
2467 
2468 	clean_xps_maps(dev, XPS_CPUS, offset, count);
2469 
2470 	mutex_unlock(&xps_map_mutex);
2471 	cpus_read_unlock();
2472 }
2473 
2474 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2475 {
2476 	netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2477 }
2478 
2479 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2480 				      u16 index, bool is_rxqs_map)
2481 {
2482 	struct xps_map *new_map;
2483 	int alloc_len = XPS_MIN_MAP_ALLOC;
2484 	int i, pos;
2485 
2486 	for (pos = 0; map && pos < map->len; pos++) {
2487 		if (map->queues[pos] != index)
2488 			continue;
2489 		return map;
2490 	}
2491 
2492 	/* Need to add tx-queue to this CPU's/rx-queue's existing map */
2493 	if (map) {
2494 		if (pos < map->alloc_len)
2495 			return map;
2496 
2497 		alloc_len = map->alloc_len * 2;
2498 	}
2499 
2500 	/* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2501 	 *  map
2502 	 */
2503 	if (is_rxqs_map)
2504 		new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2505 	else
2506 		new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2507 				       cpu_to_node(attr_index));
2508 	if (!new_map)
2509 		return NULL;
2510 
2511 	for (i = 0; i < pos; i++)
2512 		new_map->queues[i] = map->queues[i];
2513 	new_map->alloc_len = alloc_len;
2514 	new_map->len = pos;
2515 
2516 	return new_map;
2517 }
2518 
2519 /* Copy xps maps at a given index */
2520 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2521 			      struct xps_dev_maps *new_dev_maps, int index,
2522 			      int tc, bool skip_tc)
2523 {
2524 	int i, tci = index * dev_maps->num_tc;
2525 	struct xps_map *map;
2526 
2527 	/* copy maps belonging to foreign traffic classes */
2528 	for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2529 		if (i == tc && skip_tc)
2530 			continue;
2531 
2532 		/* fill in the new device map from the old device map */
2533 		map = xmap_dereference(dev_maps->attr_map[tci]);
2534 		RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2535 	}
2536 }
2537 
2538 /* Must be called under cpus_read_lock */
2539 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2540 			  u16 index, enum xps_map_type type)
2541 {
2542 	struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2543 	const unsigned long *online_mask = NULL;
2544 	bool active = false, copy = false;
2545 	int i, j, tci, numa_node_id = -2;
2546 	int maps_sz, num_tc = 1, tc = 0;
2547 	struct xps_map *map, *new_map;
2548 	unsigned int nr_ids;
2549 
2550 	if (dev->num_tc) {
2551 		/* Do not allow XPS on subordinate device directly */
2552 		num_tc = dev->num_tc;
2553 		if (num_tc < 0)
2554 			return -EINVAL;
2555 
2556 		/* If queue belongs to subordinate dev use its map */
2557 		dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2558 
2559 		tc = netdev_txq_to_tc(dev, index);
2560 		if (tc < 0)
2561 			return -EINVAL;
2562 	}
2563 
2564 	mutex_lock(&xps_map_mutex);
2565 
2566 	dev_maps = xmap_dereference(dev->xps_maps[type]);
2567 	if (type == XPS_RXQS) {
2568 		maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2569 		nr_ids = dev->num_rx_queues;
2570 	} else {
2571 		maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2572 		if (num_possible_cpus() > 1)
2573 			online_mask = cpumask_bits(cpu_online_mask);
2574 		nr_ids = nr_cpu_ids;
2575 	}
2576 
2577 	if (maps_sz < L1_CACHE_BYTES)
2578 		maps_sz = L1_CACHE_BYTES;
2579 
2580 	/* The old dev_maps could be larger or smaller than the one we're
2581 	 * setting up now, as dev->num_tc or nr_ids could have been updated in
2582 	 * between. We could try to be smart, but let's be safe instead and only
2583 	 * copy foreign traffic classes if the two map sizes match.
2584 	 */
2585 	if (dev_maps &&
2586 	    dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2587 		copy = true;
2588 
2589 	/* allocate memory for queue storage */
2590 	for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2591 	     j < nr_ids;) {
2592 		if (!new_dev_maps) {
2593 			new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2594 			if (!new_dev_maps) {
2595 				mutex_unlock(&xps_map_mutex);
2596 				return -ENOMEM;
2597 			}
2598 
2599 			new_dev_maps->nr_ids = nr_ids;
2600 			new_dev_maps->num_tc = num_tc;
2601 		}
2602 
2603 		tci = j * num_tc + tc;
2604 		map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2605 
2606 		map = expand_xps_map(map, j, index, type == XPS_RXQS);
2607 		if (!map)
2608 			goto error;
2609 
2610 		RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2611 	}
2612 
2613 	if (!new_dev_maps)
2614 		goto out_no_new_maps;
2615 
2616 	if (!dev_maps) {
2617 		/* Increment static keys at most once per type */
2618 		static_key_slow_inc_cpuslocked(&xps_needed);
2619 		if (type == XPS_RXQS)
2620 			static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2621 	}
2622 
2623 	for (j = 0; j < nr_ids; j++) {
2624 		bool skip_tc = false;
2625 
2626 		tci = j * num_tc + tc;
2627 		if (netif_attr_test_mask(j, mask, nr_ids) &&
2628 		    netif_attr_test_online(j, online_mask, nr_ids)) {
2629 			/* add tx-queue to CPU/rx-queue maps */
2630 			int pos = 0;
2631 
2632 			skip_tc = true;
2633 
2634 			map = xmap_dereference(new_dev_maps->attr_map[tci]);
2635 			while ((pos < map->len) && (map->queues[pos] != index))
2636 				pos++;
2637 
2638 			if (pos == map->len)
2639 				map->queues[map->len++] = index;
2640 #ifdef CONFIG_NUMA
2641 			if (type == XPS_CPUS) {
2642 				if (numa_node_id == -2)
2643 					numa_node_id = cpu_to_node(j);
2644 				else if (numa_node_id != cpu_to_node(j))
2645 					numa_node_id = -1;
2646 			}
2647 #endif
2648 		}
2649 
2650 		if (copy)
2651 			xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2652 					  skip_tc);
2653 	}
2654 
2655 	rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2656 
2657 	/* Cleanup old maps */
2658 	if (!dev_maps)
2659 		goto out_no_old_maps;
2660 
2661 	for (j = 0; j < dev_maps->nr_ids; j++) {
2662 		for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2663 			map = xmap_dereference(dev_maps->attr_map[tci]);
2664 			if (!map)
2665 				continue;
2666 
2667 			if (copy) {
2668 				new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2669 				if (map == new_map)
2670 					continue;
2671 			}
2672 
2673 			RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2674 			kfree_rcu(map, rcu);
2675 		}
2676 	}
2677 
2678 	old_dev_maps = dev_maps;
2679 
2680 out_no_old_maps:
2681 	dev_maps = new_dev_maps;
2682 	active = true;
2683 
2684 out_no_new_maps:
2685 	if (type == XPS_CPUS)
2686 		/* update Tx queue numa node */
2687 		netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2688 					     (numa_node_id >= 0) ?
2689 					     numa_node_id : NUMA_NO_NODE);
2690 
2691 	if (!dev_maps)
2692 		goto out_no_maps;
2693 
2694 	/* removes tx-queue from unused CPUs/rx-queues */
2695 	for (j = 0; j < dev_maps->nr_ids; j++) {
2696 		tci = j * dev_maps->num_tc;
2697 
2698 		for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2699 			if (i == tc &&
2700 			    netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2701 			    netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2702 				continue;
2703 
2704 			active |= remove_xps_queue(dev_maps,
2705 						   copy ? old_dev_maps : NULL,
2706 						   tci, index);
2707 		}
2708 	}
2709 
2710 	if (old_dev_maps)
2711 		kfree_rcu(old_dev_maps, rcu);
2712 
2713 	/* free map if not active */
2714 	if (!active)
2715 		reset_xps_maps(dev, dev_maps, type);
2716 
2717 out_no_maps:
2718 	mutex_unlock(&xps_map_mutex);
2719 
2720 	return 0;
2721 error:
2722 	/* remove any maps that we added */
2723 	for (j = 0; j < nr_ids; j++) {
2724 		for (i = num_tc, tci = j * num_tc; i--; tci++) {
2725 			new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2726 			map = copy ?
2727 			      xmap_dereference(dev_maps->attr_map[tci]) :
2728 			      NULL;
2729 			if (new_map && new_map != map)
2730 				kfree(new_map);
2731 		}
2732 	}
2733 
2734 	mutex_unlock(&xps_map_mutex);
2735 
2736 	kfree(new_dev_maps);
2737 	return -ENOMEM;
2738 }
2739 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2740 
2741 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2742 			u16 index)
2743 {
2744 	int ret;
2745 
2746 	cpus_read_lock();
2747 	ret =  __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2748 	cpus_read_unlock();
2749 
2750 	return ret;
2751 }
2752 EXPORT_SYMBOL(netif_set_xps_queue);
2753 
2754 #endif
2755 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2756 {
2757 	struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2758 
2759 	/* Unbind any subordinate channels */
2760 	while (txq-- != &dev->_tx[0]) {
2761 		if (txq->sb_dev)
2762 			netdev_unbind_sb_channel(dev, txq->sb_dev);
2763 	}
2764 }
2765 
2766 void netdev_reset_tc(struct net_device *dev)
2767 {
2768 #ifdef CONFIG_XPS
2769 	netif_reset_xps_queues_gt(dev, 0);
2770 #endif
2771 	netdev_unbind_all_sb_channels(dev);
2772 
2773 	/* Reset TC configuration of device */
2774 	dev->num_tc = 0;
2775 	memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2776 	memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2777 }
2778 EXPORT_SYMBOL(netdev_reset_tc);
2779 
2780 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2781 {
2782 	if (tc >= dev->num_tc)
2783 		return -EINVAL;
2784 
2785 #ifdef CONFIG_XPS
2786 	netif_reset_xps_queues(dev, offset, count);
2787 #endif
2788 	dev->tc_to_txq[tc].count = count;
2789 	dev->tc_to_txq[tc].offset = offset;
2790 	return 0;
2791 }
2792 EXPORT_SYMBOL(netdev_set_tc_queue);
2793 
2794 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2795 {
2796 	if (num_tc > TC_MAX_QUEUE)
2797 		return -EINVAL;
2798 
2799 #ifdef CONFIG_XPS
2800 	netif_reset_xps_queues_gt(dev, 0);
2801 #endif
2802 	netdev_unbind_all_sb_channels(dev);
2803 
2804 	dev->num_tc = num_tc;
2805 	return 0;
2806 }
2807 EXPORT_SYMBOL(netdev_set_num_tc);
2808 
2809 void netdev_unbind_sb_channel(struct net_device *dev,
2810 			      struct net_device *sb_dev)
2811 {
2812 	struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2813 
2814 #ifdef CONFIG_XPS
2815 	netif_reset_xps_queues_gt(sb_dev, 0);
2816 #endif
2817 	memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2818 	memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2819 
2820 	while (txq-- != &dev->_tx[0]) {
2821 		if (txq->sb_dev == sb_dev)
2822 			txq->sb_dev = NULL;
2823 	}
2824 }
2825 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2826 
2827 int netdev_bind_sb_channel_queue(struct net_device *dev,
2828 				 struct net_device *sb_dev,
2829 				 u8 tc, u16 count, u16 offset)
2830 {
2831 	/* Make certain the sb_dev and dev are already configured */
2832 	if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2833 		return -EINVAL;
2834 
2835 	/* We cannot hand out queues we don't have */
2836 	if ((offset + count) > dev->real_num_tx_queues)
2837 		return -EINVAL;
2838 
2839 	/* Record the mapping */
2840 	sb_dev->tc_to_txq[tc].count = count;
2841 	sb_dev->tc_to_txq[tc].offset = offset;
2842 
2843 	/* Provide a way for Tx queue to find the tc_to_txq map or
2844 	 * XPS map for itself.
2845 	 */
2846 	while (count--)
2847 		netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2848 
2849 	return 0;
2850 }
2851 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2852 
2853 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2854 {
2855 	/* Do not use a multiqueue device to represent a subordinate channel */
2856 	if (netif_is_multiqueue(dev))
2857 		return -ENODEV;
2858 
2859 	/* We allow channels 1 - 32767 to be used for subordinate channels.
2860 	 * Channel 0 is meant to be "native" mode and used only to represent
2861 	 * the main root device. We allow writing 0 to reset the device back
2862 	 * to normal mode after being used as a subordinate channel.
2863 	 */
2864 	if (channel > S16_MAX)
2865 		return -EINVAL;
2866 
2867 	dev->num_tc = -channel;
2868 
2869 	return 0;
2870 }
2871 EXPORT_SYMBOL(netdev_set_sb_channel);
2872 
2873 /*
2874  * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2875  * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2876  */
2877 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2878 {
2879 	bool disabling;
2880 	int rc;
2881 
2882 	disabling = txq < dev->real_num_tx_queues;
2883 
2884 	if (txq < 1 || txq > dev->num_tx_queues)
2885 		return -EINVAL;
2886 
2887 	if (dev->reg_state == NETREG_REGISTERED ||
2888 	    dev->reg_state == NETREG_UNREGISTERING) {
2889 		ASSERT_RTNL();
2890 
2891 		rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2892 						  txq);
2893 		if (rc)
2894 			return rc;
2895 
2896 		if (dev->num_tc)
2897 			netif_setup_tc(dev, txq);
2898 
2899 		dev_qdisc_change_real_num_tx(dev, txq);
2900 
2901 		dev->real_num_tx_queues = txq;
2902 
2903 		if (disabling) {
2904 			synchronize_net();
2905 			qdisc_reset_all_tx_gt(dev, txq);
2906 #ifdef CONFIG_XPS
2907 			netif_reset_xps_queues_gt(dev, txq);
2908 #endif
2909 		}
2910 	} else {
2911 		dev->real_num_tx_queues = txq;
2912 	}
2913 
2914 	return 0;
2915 }
2916 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2917 
2918 #ifdef CONFIG_SYSFS
2919 /**
2920  *	netif_set_real_num_rx_queues - set actual number of RX queues used
2921  *	@dev: Network device
2922  *	@rxq: Actual number of RX queues
2923  *
2924  *	This must be called either with the rtnl_lock held or before
2925  *	registration of the net device.  Returns 0 on success, or a
2926  *	negative error code.  If called before registration, it always
2927  *	succeeds.
2928  */
2929 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2930 {
2931 	int rc;
2932 
2933 	if (rxq < 1 || rxq > dev->num_rx_queues)
2934 		return -EINVAL;
2935 
2936 	if (dev->reg_state == NETREG_REGISTERED) {
2937 		ASSERT_RTNL();
2938 
2939 		rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2940 						  rxq);
2941 		if (rc)
2942 			return rc;
2943 	}
2944 
2945 	dev->real_num_rx_queues = rxq;
2946 	return 0;
2947 }
2948 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2949 #endif
2950 
2951 /**
2952  *	netif_set_real_num_queues - set actual number of RX and TX queues used
2953  *	@dev: Network device
2954  *	@txq: Actual number of TX queues
2955  *	@rxq: Actual number of RX queues
2956  *
2957  *	Set the real number of both TX and RX queues.
2958  *	Does nothing if the number of queues is already correct.
2959  */
2960 int netif_set_real_num_queues(struct net_device *dev,
2961 			      unsigned int txq, unsigned int rxq)
2962 {
2963 	unsigned int old_rxq = dev->real_num_rx_queues;
2964 	int err;
2965 
2966 	if (txq < 1 || txq > dev->num_tx_queues ||
2967 	    rxq < 1 || rxq > dev->num_rx_queues)
2968 		return -EINVAL;
2969 
2970 	/* Start from increases, so the error path only does decreases -
2971 	 * decreases can't fail.
2972 	 */
2973 	if (rxq > dev->real_num_rx_queues) {
2974 		err = netif_set_real_num_rx_queues(dev, rxq);
2975 		if (err)
2976 			return err;
2977 	}
2978 	if (txq > dev->real_num_tx_queues) {
2979 		err = netif_set_real_num_tx_queues(dev, txq);
2980 		if (err)
2981 			goto undo_rx;
2982 	}
2983 	if (rxq < dev->real_num_rx_queues)
2984 		WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
2985 	if (txq < dev->real_num_tx_queues)
2986 		WARN_ON(netif_set_real_num_tx_queues(dev, txq));
2987 
2988 	return 0;
2989 undo_rx:
2990 	WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
2991 	return err;
2992 }
2993 EXPORT_SYMBOL(netif_set_real_num_queues);
2994 
2995 /**
2996  * netif_set_tso_max_size() - set the max size of TSO frames supported
2997  * @dev:	netdev to update
2998  * @size:	max skb->len of a TSO frame
2999  *
3000  * Set the limit on the size of TSO super-frames the device can handle.
3001  * Unless explicitly set the stack will assume the value of
3002  * %GSO_LEGACY_MAX_SIZE.
3003  */
3004 void netif_set_tso_max_size(struct net_device *dev, unsigned int size)
3005 {
3006 	dev->tso_max_size = min(GSO_MAX_SIZE, size);
3007 	if (size < READ_ONCE(dev->gso_max_size))
3008 		netif_set_gso_max_size(dev, size);
3009 }
3010 EXPORT_SYMBOL(netif_set_tso_max_size);
3011 
3012 /**
3013  * netif_set_tso_max_segs() - set the max number of segs supported for TSO
3014  * @dev:	netdev to update
3015  * @segs:	max number of TCP segments
3016  *
3017  * Set the limit on the number of TCP segments the device can generate from
3018  * a single TSO super-frame.
3019  * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS.
3020  */
3021 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs)
3022 {
3023 	dev->tso_max_segs = segs;
3024 	if (segs < READ_ONCE(dev->gso_max_segs))
3025 		netif_set_gso_max_segs(dev, segs);
3026 }
3027 EXPORT_SYMBOL(netif_set_tso_max_segs);
3028 
3029 /**
3030  * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper
3031  * @to:		netdev to update
3032  * @from:	netdev from which to copy the limits
3033  */
3034 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from)
3035 {
3036 	netif_set_tso_max_size(to, from->tso_max_size);
3037 	netif_set_tso_max_segs(to, from->tso_max_segs);
3038 }
3039 EXPORT_SYMBOL(netif_inherit_tso_max);
3040 
3041 /**
3042  * netif_get_num_default_rss_queues - default number of RSS queues
3043  *
3044  * Default value is the number of physical cores if there are only 1 or 2, or
3045  * divided by 2 if there are more.
3046  */
3047 int netif_get_num_default_rss_queues(void)
3048 {
3049 	cpumask_var_t cpus;
3050 	int cpu, count = 0;
3051 
3052 	if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL)))
3053 		return 1;
3054 
3055 	cpumask_copy(cpus, cpu_online_mask);
3056 	for_each_cpu(cpu, cpus) {
3057 		++count;
3058 		cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
3059 	}
3060 	free_cpumask_var(cpus);
3061 
3062 	return count > 2 ? DIV_ROUND_UP(count, 2) : count;
3063 }
3064 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3065 
3066 static void __netif_reschedule(struct Qdisc *q)
3067 {
3068 	struct softnet_data *sd;
3069 	unsigned long flags;
3070 
3071 	local_irq_save(flags);
3072 	sd = this_cpu_ptr(&softnet_data);
3073 	q->next_sched = NULL;
3074 	*sd->output_queue_tailp = q;
3075 	sd->output_queue_tailp = &q->next_sched;
3076 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
3077 	local_irq_restore(flags);
3078 }
3079 
3080 void __netif_schedule(struct Qdisc *q)
3081 {
3082 	if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3083 		__netif_reschedule(q);
3084 }
3085 EXPORT_SYMBOL(__netif_schedule);
3086 
3087 struct dev_kfree_skb_cb {
3088 	enum skb_free_reason reason;
3089 };
3090 
3091 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3092 {
3093 	return (struct dev_kfree_skb_cb *)skb->cb;
3094 }
3095 
3096 void netif_schedule_queue(struct netdev_queue *txq)
3097 {
3098 	rcu_read_lock();
3099 	if (!netif_xmit_stopped(txq)) {
3100 		struct Qdisc *q = rcu_dereference(txq->qdisc);
3101 
3102 		__netif_schedule(q);
3103 	}
3104 	rcu_read_unlock();
3105 }
3106 EXPORT_SYMBOL(netif_schedule_queue);
3107 
3108 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3109 {
3110 	if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3111 		struct Qdisc *q;
3112 
3113 		rcu_read_lock();
3114 		q = rcu_dereference(dev_queue->qdisc);
3115 		__netif_schedule(q);
3116 		rcu_read_unlock();
3117 	}
3118 }
3119 EXPORT_SYMBOL(netif_tx_wake_queue);
3120 
3121 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3122 {
3123 	unsigned long flags;
3124 
3125 	if (unlikely(!skb))
3126 		return;
3127 
3128 	if (likely(refcount_read(&skb->users) == 1)) {
3129 		smp_rmb();
3130 		refcount_set(&skb->users, 0);
3131 	} else if (likely(!refcount_dec_and_test(&skb->users))) {
3132 		return;
3133 	}
3134 	get_kfree_skb_cb(skb)->reason = reason;
3135 	local_irq_save(flags);
3136 	skb->next = __this_cpu_read(softnet_data.completion_queue);
3137 	__this_cpu_write(softnet_data.completion_queue, skb);
3138 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
3139 	local_irq_restore(flags);
3140 }
3141 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3142 
3143 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3144 {
3145 	if (in_hardirq() || irqs_disabled())
3146 		__dev_kfree_skb_irq(skb, reason);
3147 	else
3148 		dev_kfree_skb(skb);
3149 }
3150 EXPORT_SYMBOL(__dev_kfree_skb_any);
3151 
3152 
3153 /**
3154  * netif_device_detach - mark device as removed
3155  * @dev: network device
3156  *
3157  * Mark device as removed from system and therefore no longer available.
3158  */
3159 void netif_device_detach(struct net_device *dev)
3160 {
3161 	if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3162 	    netif_running(dev)) {
3163 		netif_tx_stop_all_queues(dev);
3164 	}
3165 }
3166 EXPORT_SYMBOL(netif_device_detach);
3167 
3168 /**
3169  * netif_device_attach - mark device as attached
3170  * @dev: network device
3171  *
3172  * Mark device as attached from system and restart if needed.
3173  */
3174 void netif_device_attach(struct net_device *dev)
3175 {
3176 	if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3177 	    netif_running(dev)) {
3178 		netif_tx_wake_all_queues(dev);
3179 		__netdev_watchdog_up(dev);
3180 	}
3181 }
3182 EXPORT_SYMBOL(netif_device_attach);
3183 
3184 /*
3185  * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3186  * to be used as a distribution range.
3187  */
3188 static u16 skb_tx_hash(const struct net_device *dev,
3189 		       const struct net_device *sb_dev,
3190 		       struct sk_buff *skb)
3191 {
3192 	u32 hash;
3193 	u16 qoffset = 0;
3194 	u16 qcount = dev->real_num_tx_queues;
3195 
3196 	if (dev->num_tc) {
3197 		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3198 
3199 		qoffset = sb_dev->tc_to_txq[tc].offset;
3200 		qcount = sb_dev->tc_to_txq[tc].count;
3201 		if (unlikely(!qcount)) {
3202 			net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3203 					     sb_dev->name, qoffset, tc);
3204 			qoffset = 0;
3205 			qcount = dev->real_num_tx_queues;
3206 		}
3207 	}
3208 
3209 	if (skb_rx_queue_recorded(skb)) {
3210 		hash = skb_get_rx_queue(skb);
3211 		if (hash >= qoffset)
3212 			hash -= qoffset;
3213 		while (unlikely(hash >= qcount))
3214 			hash -= qcount;
3215 		return hash + qoffset;
3216 	}
3217 
3218 	return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3219 }
3220 
3221 static void skb_warn_bad_offload(const struct sk_buff *skb)
3222 {
3223 	static const netdev_features_t null_features;
3224 	struct net_device *dev = skb->dev;
3225 	const char *name = "";
3226 
3227 	if (!net_ratelimit())
3228 		return;
3229 
3230 	if (dev) {
3231 		if (dev->dev.parent)
3232 			name = dev_driver_string(dev->dev.parent);
3233 		else
3234 			name = netdev_name(dev);
3235 	}
3236 	skb_dump(KERN_WARNING, skb, false);
3237 	WARN(1, "%s: caps=(%pNF, %pNF)\n",
3238 	     name, dev ? &dev->features : &null_features,
3239 	     skb->sk ? &skb->sk->sk_route_caps : &null_features);
3240 }
3241 
3242 /*
3243  * Invalidate hardware checksum when packet is to be mangled, and
3244  * complete checksum manually on outgoing path.
3245  */
3246 int skb_checksum_help(struct sk_buff *skb)
3247 {
3248 	__wsum csum;
3249 	int ret = 0, offset;
3250 
3251 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3252 		goto out_set_summed;
3253 
3254 	if (unlikely(skb_is_gso(skb))) {
3255 		skb_warn_bad_offload(skb);
3256 		return -EINVAL;
3257 	}
3258 
3259 	/* Before computing a checksum, we should make sure no frag could
3260 	 * be modified by an external entity : checksum could be wrong.
3261 	 */
3262 	if (skb_has_shared_frag(skb)) {
3263 		ret = __skb_linearize(skb);
3264 		if (ret)
3265 			goto out;
3266 	}
3267 
3268 	offset = skb_checksum_start_offset(skb);
3269 	ret = -EINVAL;
3270 	if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3271 		DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3272 		goto out;
3273 	}
3274 	csum = skb_checksum(skb, offset, skb->len - offset, 0);
3275 
3276 	offset += skb->csum_offset;
3277 	if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) {
3278 		DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3279 		goto out;
3280 	}
3281 	ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3282 	if (ret)
3283 		goto out;
3284 
3285 	*(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3286 out_set_summed:
3287 	skb->ip_summed = CHECKSUM_NONE;
3288 out:
3289 	return ret;
3290 }
3291 EXPORT_SYMBOL(skb_checksum_help);
3292 
3293 int skb_crc32c_csum_help(struct sk_buff *skb)
3294 {
3295 	__le32 crc32c_csum;
3296 	int ret = 0, offset, start;
3297 
3298 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3299 		goto out;
3300 
3301 	if (unlikely(skb_is_gso(skb)))
3302 		goto out;
3303 
3304 	/* Before computing a checksum, we should make sure no frag could
3305 	 * be modified by an external entity : checksum could be wrong.
3306 	 */
3307 	if (unlikely(skb_has_shared_frag(skb))) {
3308 		ret = __skb_linearize(skb);
3309 		if (ret)
3310 			goto out;
3311 	}
3312 	start = skb_checksum_start_offset(skb);
3313 	offset = start + offsetof(struct sctphdr, checksum);
3314 	if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3315 		ret = -EINVAL;
3316 		goto out;
3317 	}
3318 
3319 	ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3320 	if (ret)
3321 		goto out;
3322 
3323 	crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3324 						  skb->len - start, ~(__u32)0,
3325 						  crc32c_csum_stub));
3326 	*(__le32 *)(skb->data + offset) = crc32c_csum;
3327 	skb->ip_summed = CHECKSUM_NONE;
3328 	skb->csum_not_inet = 0;
3329 out:
3330 	return ret;
3331 }
3332 
3333 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3334 {
3335 	__be16 type = skb->protocol;
3336 
3337 	/* Tunnel gso handlers can set protocol to ethernet. */
3338 	if (type == htons(ETH_P_TEB)) {
3339 		struct ethhdr *eth;
3340 
3341 		if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3342 			return 0;
3343 
3344 		eth = (struct ethhdr *)skb->data;
3345 		type = eth->h_proto;
3346 	}
3347 
3348 	return __vlan_get_protocol(skb, type, depth);
3349 }
3350 
3351 /* openvswitch calls this on rx path, so we need a different check.
3352  */
3353 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3354 {
3355 	if (tx_path)
3356 		return skb->ip_summed != CHECKSUM_PARTIAL &&
3357 		       skb->ip_summed != CHECKSUM_UNNECESSARY;
3358 
3359 	return skb->ip_summed == CHECKSUM_NONE;
3360 }
3361 
3362 /**
3363  *	__skb_gso_segment - Perform segmentation on skb.
3364  *	@skb: buffer to segment
3365  *	@features: features for the output path (see dev->features)
3366  *	@tx_path: whether it is called in TX path
3367  *
3368  *	This function segments the given skb and returns a list of segments.
3369  *
3370  *	It may return NULL if the skb requires no segmentation.  This is
3371  *	only possible when GSO is used for verifying header integrity.
3372  *
3373  *	Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3374  */
3375 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3376 				  netdev_features_t features, bool tx_path)
3377 {
3378 	struct sk_buff *segs;
3379 
3380 	if (unlikely(skb_needs_check(skb, tx_path))) {
3381 		int err;
3382 
3383 		/* We're going to init ->check field in TCP or UDP header */
3384 		err = skb_cow_head(skb, 0);
3385 		if (err < 0)
3386 			return ERR_PTR(err);
3387 	}
3388 
3389 	/* Only report GSO partial support if it will enable us to
3390 	 * support segmentation on this frame without needing additional
3391 	 * work.
3392 	 */
3393 	if (features & NETIF_F_GSO_PARTIAL) {
3394 		netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3395 		struct net_device *dev = skb->dev;
3396 
3397 		partial_features |= dev->features & dev->gso_partial_features;
3398 		if (!skb_gso_ok(skb, features | partial_features))
3399 			features &= ~NETIF_F_GSO_PARTIAL;
3400 	}
3401 
3402 	BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3403 		     sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3404 
3405 	SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3406 	SKB_GSO_CB(skb)->encap_level = 0;
3407 
3408 	skb_reset_mac_header(skb);
3409 	skb_reset_mac_len(skb);
3410 
3411 	segs = skb_mac_gso_segment(skb, features);
3412 
3413 	if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3414 		skb_warn_bad_offload(skb);
3415 
3416 	return segs;
3417 }
3418 EXPORT_SYMBOL(__skb_gso_segment);
3419 
3420 /* Take action when hardware reception checksum errors are detected. */
3421 #ifdef CONFIG_BUG
3422 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3423 {
3424 	netdev_err(dev, "hw csum failure\n");
3425 	skb_dump(KERN_ERR, skb, true);
3426 	dump_stack();
3427 }
3428 
3429 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3430 {
3431 	DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3432 }
3433 EXPORT_SYMBOL(netdev_rx_csum_fault);
3434 #endif
3435 
3436 /* XXX: check that highmem exists at all on the given machine. */
3437 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3438 {
3439 #ifdef CONFIG_HIGHMEM
3440 	int i;
3441 
3442 	if (!(dev->features & NETIF_F_HIGHDMA)) {
3443 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3444 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3445 
3446 			if (PageHighMem(skb_frag_page(frag)))
3447 				return 1;
3448 		}
3449 	}
3450 #endif
3451 	return 0;
3452 }
3453 
3454 /* If MPLS offload request, verify we are testing hardware MPLS features
3455  * instead of standard features for the netdev.
3456  */
3457 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3458 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3459 					   netdev_features_t features,
3460 					   __be16 type)
3461 {
3462 	if (eth_p_mpls(type))
3463 		features &= skb->dev->mpls_features;
3464 
3465 	return features;
3466 }
3467 #else
3468 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3469 					   netdev_features_t features,
3470 					   __be16 type)
3471 {
3472 	return features;
3473 }
3474 #endif
3475 
3476 static netdev_features_t harmonize_features(struct sk_buff *skb,
3477 	netdev_features_t features)
3478 {
3479 	__be16 type;
3480 
3481 	type = skb_network_protocol(skb, NULL);
3482 	features = net_mpls_features(skb, features, type);
3483 
3484 	if (skb->ip_summed != CHECKSUM_NONE &&
3485 	    !can_checksum_protocol(features, type)) {
3486 		features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3487 	}
3488 	if (illegal_highdma(skb->dev, skb))
3489 		features &= ~NETIF_F_SG;
3490 
3491 	return features;
3492 }
3493 
3494 netdev_features_t passthru_features_check(struct sk_buff *skb,
3495 					  struct net_device *dev,
3496 					  netdev_features_t features)
3497 {
3498 	return features;
3499 }
3500 EXPORT_SYMBOL(passthru_features_check);
3501 
3502 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3503 					     struct net_device *dev,
3504 					     netdev_features_t features)
3505 {
3506 	return vlan_features_check(skb, features);
3507 }
3508 
3509 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3510 					    struct net_device *dev,
3511 					    netdev_features_t features)
3512 {
3513 	u16 gso_segs = skb_shinfo(skb)->gso_segs;
3514 
3515 	if (gso_segs > READ_ONCE(dev->gso_max_segs))
3516 		return features & ~NETIF_F_GSO_MASK;
3517 
3518 	if (!skb_shinfo(skb)->gso_type) {
3519 		skb_warn_bad_offload(skb);
3520 		return features & ~NETIF_F_GSO_MASK;
3521 	}
3522 
3523 	/* Support for GSO partial features requires software
3524 	 * intervention before we can actually process the packets
3525 	 * so we need to strip support for any partial features now
3526 	 * and we can pull them back in after we have partially
3527 	 * segmented the frame.
3528 	 */
3529 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3530 		features &= ~dev->gso_partial_features;
3531 
3532 	/* Make sure to clear the IPv4 ID mangling feature if the
3533 	 * IPv4 header has the potential to be fragmented.
3534 	 */
3535 	if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3536 		struct iphdr *iph = skb->encapsulation ?
3537 				    inner_ip_hdr(skb) : ip_hdr(skb);
3538 
3539 		if (!(iph->frag_off & htons(IP_DF)))
3540 			features &= ~NETIF_F_TSO_MANGLEID;
3541 	}
3542 
3543 	return features;
3544 }
3545 
3546 netdev_features_t netif_skb_features(struct sk_buff *skb)
3547 {
3548 	struct net_device *dev = skb->dev;
3549 	netdev_features_t features = dev->features;
3550 
3551 	if (skb_is_gso(skb))
3552 		features = gso_features_check(skb, dev, features);
3553 
3554 	/* If encapsulation offload request, verify we are testing
3555 	 * hardware encapsulation features instead of standard
3556 	 * features for the netdev
3557 	 */
3558 	if (skb->encapsulation)
3559 		features &= dev->hw_enc_features;
3560 
3561 	if (skb_vlan_tagged(skb))
3562 		features = netdev_intersect_features(features,
3563 						     dev->vlan_features |
3564 						     NETIF_F_HW_VLAN_CTAG_TX |
3565 						     NETIF_F_HW_VLAN_STAG_TX);
3566 
3567 	if (dev->netdev_ops->ndo_features_check)
3568 		features &= dev->netdev_ops->ndo_features_check(skb, dev,
3569 								features);
3570 	else
3571 		features &= dflt_features_check(skb, dev, features);
3572 
3573 	return harmonize_features(skb, features);
3574 }
3575 EXPORT_SYMBOL(netif_skb_features);
3576 
3577 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3578 		    struct netdev_queue *txq, bool more)
3579 {
3580 	unsigned int len;
3581 	int rc;
3582 
3583 	if (dev_nit_active(dev))
3584 		dev_queue_xmit_nit(skb, dev);
3585 
3586 	len = skb->len;
3587 	trace_net_dev_start_xmit(skb, dev);
3588 	rc = netdev_start_xmit(skb, dev, txq, more);
3589 	trace_net_dev_xmit(skb, rc, dev, len);
3590 
3591 	return rc;
3592 }
3593 
3594 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3595 				    struct netdev_queue *txq, int *ret)
3596 {
3597 	struct sk_buff *skb = first;
3598 	int rc = NETDEV_TX_OK;
3599 
3600 	while (skb) {
3601 		struct sk_buff *next = skb->next;
3602 
3603 		skb_mark_not_on_list(skb);
3604 		rc = xmit_one(skb, dev, txq, next != NULL);
3605 		if (unlikely(!dev_xmit_complete(rc))) {
3606 			skb->next = next;
3607 			goto out;
3608 		}
3609 
3610 		skb = next;
3611 		if (netif_tx_queue_stopped(txq) && skb) {
3612 			rc = NETDEV_TX_BUSY;
3613 			break;
3614 		}
3615 	}
3616 
3617 out:
3618 	*ret = rc;
3619 	return skb;
3620 }
3621 
3622 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3623 					  netdev_features_t features)
3624 {
3625 	if (skb_vlan_tag_present(skb) &&
3626 	    !vlan_hw_offload_capable(features, skb->vlan_proto))
3627 		skb = __vlan_hwaccel_push_inside(skb);
3628 	return skb;
3629 }
3630 
3631 int skb_csum_hwoffload_help(struct sk_buff *skb,
3632 			    const netdev_features_t features)
3633 {
3634 	if (unlikely(skb_csum_is_sctp(skb)))
3635 		return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3636 			skb_crc32c_csum_help(skb);
3637 
3638 	if (features & NETIF_F_HW_CSUM)
3639 		return 0;
3640 
3641 	if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3642 		switch (skb->csum_offset) {
3643 		case offsetof(struct tcphdr, check):
3644 		case offsetof(struct udphdr, check):
3645 			return 0;
3646 		}
3647 	}
3648 
3649 	return skb_checksum_help(skb);
3650 }
3651 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3652 
3653 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3654 {
3655 	netdev_features_t features;
3656 
3657 	features = netif_skb_features(skb);
3658 	skb = validate_xmit_vlan(skb, features);
3659 	if (unlikely(!skb))
3660 		goto out_null;
3661 
3662 	skb = sk_validate_xmit_skb(skb, dev);
3663 	if (unlikely(!skb))
3664 		goto out_null;
3665 
3666 	if (netif_needs_gso(skb, features)) {
3667 		struct sk_buff *segs;
3668 
3669 		segs = skb_gso_segment(skb, features);
3670 		if (IS_ERR(segs)) {
3671 			goto out_kfree_skb;
3672 		} else if (segs) {
3673 			consume_skb(skb);
3674 			skb = segs;
3675 		}
3676 	} else {
3677 		if (skb_needs_linearize(skb, features) &&
3678 		    __skb_linearize(skb))
3679 			goto out_kfree_skb;
3680 
3681 		/* If packet is not checksummed and device does not
3682 		 * support checksumming for this protocol, complete
3683 		 * checksumming here.
3684 		 */
3685 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
3686 			if (skb->encapsulation)
3687 				skb_set_inner_transport_header(skb,
3688 							       skb_checksum_start_offset(skb));
3689 			else
3690 				skb_set_transport_header(skb,
3691 							 skb_checksum_start_offset(skb));
3692 			if (skb_csum_hwoffload_help(skb, features))
3693 				goto out_kfree_skb;
3694 		}
3695 	}
3696 
3697 	skb = validate_xmit_xfrm(skb, features, again);
3698 
3699 	return skb;
3700 
3701 out_kfree_skb:
3702 	kfree_skb(skb);
3703 out_null:
3704 	dev_core_stats_tx_dropped_inc(dev);
3705 	return NULL;
3706 }
3707 
3708 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3709 {
3710 	struct sk_buff *next, *head = NULL, *tail;
3711 
3712 	for (; skb != NULL; skb = next) {
3713 		next = skb->next;
3714 		skb_mark_not_on_list(skb);
3715 
3716 		/* in case skb wont be segmented, point to itself */
3717 		skb->prev = skb;
3718 
3719 		skb = validate_xmit_skb(skb, dev, again);
3720 		if (!skb)
3721 			continue;
3722 
3723 		if (!head)
3724 			head = skb;
3725 		else
3726 			tail->next = skb;
3727 		/* If skb was segmented, skb->prev points to
3728 		 * the last segment. If not, it still contains skb.
3729 		 */
3730 		tail = skb->prev;
3731 	}
3732 	return head;
3733 }
3734 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3735 
3736 static void qdisc_pkt_len_init(struct sk_buff *skb)
3737 {
3738 	const struct skb_shared_info *shinfo = skb_shinfo(skb);
3739 
3740 	qdisc_skb_cb(skb)->pkt_len = skb->len;
3741 
3742 	/* To get more precise estimation of bytes sent on wire,
3743 	 * we add to pkt_len the headers size of all segments
3744 	 */
3745 	if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3746 		unsigned int hdr_len;
3747 		u16 gso_segs = shinfo->gso_segs;
3748 
3749 		/* mac layer + network layer */
3750 		hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3751 
3752 		/* + transport layer */
3753 		if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3754 			const struct tcphdr *th;
3755 			struct tcphdr _tcphdr;
3756 
3757 			th = skb_header_pointer(skb, skb_transport_offset(skb),
3758 						sizeof(_tcphdr), &_tcphdr);
3759 			if (likely(th))
3760 				hdr_len += __tcp_hdrlen(th);
3761 		} else {
3762 			struct udphdr _udphdr;
3763 
3764 			if (skb_header_pointer(skb, skb_transport_offset(skb),
3765 					       sizeof(_udphdr), &_udphdr))
3766 				hdr_len += sizeof(struct udphdr);
3767 		}
3768 
3769 		if (shinfo->gso_type & SKB_GSO_DODGY)
3770 			gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3771 						shinfo->gso_size);
3772 
3773 		qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3774 	}
3775 }
3776 
3777 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3778 			     struct sk_buff **to_free,
3779 			     struct netdev_queue *txq)
3780 {
3781 	int rc;
3782 
3783 	rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3784 	if (rc == NET_XMIT_SUCCESS)
3785 		trace_qdisc_enqueue(q, txq, skb);
3786 	return rc;
3787 }
3788 
3789 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3790 				 struct net_device *dev,
3791 				 struct netdev_queue *txq)
3792 {
3793 	spinlock_t *root_lock = qdisc_lock(q);
3794 	struct sk_buff *to_free = NULL;
3795 	bool contended;
3796 	int rc;
3797 
3798 	qdisc_calculate_pkt_len(skb, q);
3799 
3800 	if (q->flags & TCQ_F_NOLOCK) {
3801 		if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3802 		    qdisc_run_begin(q)) {
3803 			/* Retest nolock_qdisc_is_empty() within the protection
3804 			 * of q->seqlock to protect from racing with requeuing.
3805 			 */
3806 			if (unlikely(!nolock_qdisc_is_empty(q))) {
3807 				rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3808 				__qdisc_run(q);
3809 				qdisc_run_end(q);
3810 
3811 				goto no_lock_out;
3812 			}
3813 
3814 			qdisc_bstats_cpu_update(q, skb);
3815 			if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3816 			    !nolock_qdisc_is_empty(q))
3817 				__qdisc_run(q);
3818 
3819 			qdisc_run_end(q);
3820 			return NET_XMIT_SUCCESS;
3821 		}
3822 
3823 		rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3824 		qdisc_run(q);
3825 
3826 no_lock_out:
3827 		if (unlikely(to_free))
3828 			kfree_skb_list_reason(to_free,
3829 					      SKB_DROP_REASON_QDISC_DROP);
3830 		return rc;
3831 	}
3832 
3833 	/*
3834 	 * Heuristic to force contended enqueues to serialize on a
3835 	 * separate lock before trying to get qdisc main lock.
3836 	 * This permits qdisc->running owner to get the lock more
3837 	 * often and dequeue packets faster.
3838 	 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
3839 	 * and then other tasks will only enqueue packets. The packets will be
3840 	 * sent after the qdisc owner is scheduled again. To prevent this
3841 	 * scenario the task always serialize on the lock.
3842 	 */
3843 	contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
3844 	if (unlikely(contended))
3845 		spin_lock(&q->busylock);
3846 
3847 	spin_lock(root_lock);
3848 	if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3849 		__qdisc_drop(skb, &to_free);
3850 		rc = NET_XMIT_DROP;
3851 	} else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3852 		   qdisc_run_begin(q)) {
3853 		/*
3854 		 * This is a work-conserving queue; there are no old skbs
3855 		 * waiting to be sent out; and the qdisc is not running -
3856 		 * xmit the skb directly.
3857 		 */
3858 
3859 		qdisc_bstats_update(q, skb);
3860 
3861 		if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3862 			if (unlikely(contended)) {
3863 				spin_unlock(&q->busylock);
3864 				contended = false;
3865 			}
3866 			__qdisc_run(q);
3867 		}
3868 
3869 		qdisc_run_end(q);
3870 		rc = NET_XMIT_SUCCESS;
3871 	} else {
3872 		rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3873 		if (qdisc_run_begin(q)) {
3874 			if (unlikely(contended)) {
3875 				spin_unlock(&q->busylock);
3876 				contended = false;
3877 			}
3878 			__qdisc_run(q);
3879 			qdisc_run_end(q);
3880 		}
3881 	}
3882 	spin_unlock(root_lock);
3883 	if (unlikely(to_free))
3884 		kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP);
3885 	if (unlikely(contended))
3886 		spin_unlock(&q->busylock);
3887 	return rc;
3888 }
3889 
3890 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3891 static void skb_update_prio(struct sk_buff *skb)
3892 {
3893 	const struct netprio_map *map;
3894 	const struct sock *sk;
3895 	unsigned int prioidx;
3896 
3897 	if (skb->priority)
3898 		return;
3899 	map = rcu_dereference_bh(skb->dev->priomap);
3900 	if (!map)
3901 		return;
3902 	sk = skb_to_full_sk(skb);
3903 	if (!sk)
3904 		return;
3905 
3906 	prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3907 
3908 	if (prioidx < map->priomap_len)
3909 		skb->priority = map->priomap[prioidx];
3910 }
3911 #else
3912 #define skb_update_prio(skb)
3913 #endif
3914 
3915 /**
3916  *	dev_loopback_xmit - loop back @skb
3917  *	@net: network namespace this loopback is happening in
3918  *	@sk:  sk needed to be a netfilter okfn
3919  *	@skb: buffer to transmit
3920  */
3921 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3922 {
3923 	skb_reset_mac_header(skb);
3924 	__skb_pull(skb, skb_network_offset(skb));
3925 	skb->pkt_type = PACKET_LOOPBACK;
3926 	if (skb->ip_summed == CHECKSUM_NONE)
3927 		skb->ip_summed = CHECKSUM_UNNECESSARY;
3928 	DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb));
3929 	skb_dst_force(skb);
3930 	netif_rx(skb);
3931 	return 0;
3932 }
3933 EXPORT_SYMBOL(dev_loopback_xmit);
3934 
3935 #ifdef CONFIG_NET_EGRESS
3936 static struct sk_buff *
3937 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3938 {
3939 #ifdef CONFIG_NET_CLS_ACT
3940 	struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3941 	struct tcf_result cl_res;
3942 
3943 	if (!miniq)
3944 		return skb;
3945 
3946 	/* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3947 	tc_skb_cb(skb)->mru = 0;
3948 	tc_skb_cb(skb)->post_ct = false;
3949 	mini_qdisc_bstats_cpu_update(miniq, skb);
3950 
3951 	switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
3952 	case TC_ACT_OK:
3953 	case TC_ACT_RECLASSIFY:
3954 		skb->tc_index = TC_H_MIN(cl_res.classid);
3955 		break;
3956 	case TC_ACT_SHOT:
3957 		mini_qdisc_qstats_cpu_drop(miniq);
3958 		*ret = NET_XMIT_DROP;
3959 		kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS);
3960 		return NULL;
3961 	case TC_ACT_STOLEN:
3962 	case TC_ACT_QUEUED:
3963 	case TC_ACT_TRAP:
3964 		*ret = NET_XMIT_SUCCESS;
3965 		consume_skb(skb);
3966 		return NULL;
3967 	case TC_ACT_REDIRECT:
3968 		/* No need to push/pop skb's mac_header here on egress! */
3969 		skb_do_redirect(skb);
3970 		*ret = NET_XMIT_SUCCESS;
3971 		return NULL;
3972 	default:
3973 		break;
3974 	}
3975 #endif /* CONFIG_NET_CLS_ACT */
3976 
3977 	return skb;
3978 }
3979 
3980 static struct netdev_queue *
3981 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
3982 {
3983 	int qm = skb_get_queue_mapping(skb);
3984 
3985 	return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
3986 }
3987 
3988 static bool netdev_xmit_txqueue_skipped(void)
3989 {
3990 	return __this_cpu_read(softnet_data.xmit.skip_txqueue);
3991 }
3992 
3993 void netdev_xmit_skip_txqueue(bool skip)
3994 {
3995 	__this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
3996 }
3997 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
3998 #endif /* CONFIG_NET_EGRESS */
3999 
4000 #ifdef CONFIG_XPS
4001 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
4002 			       struct xps_dev_maps *dev_maps, unsigned int tci)
4003 {
4004 	int tc = netdev_get_prio_tc_map(dev, skb->priority);
4005 	struct xps_map *map;
4006 	int queue_index = -1;
4007 
4008 	if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4009 		return queue_index;
4010 
4011 	tci *= dev_maps->num_tc;
4012 	tci += tc;
4013 
4014 	map = rcu_dereference(dev_maps->attr_map[tci]);
4015 	if (map) {
4016 		if (map->len == 1)
4017 			queue_index = map->queues[0];
4018 		else
4019 			queue_index = map->queues[reciprocal_scale(
4020 						skb_get_hash(skb), map->len)];
4021 		if (unlikely(queue_index >= dev->real_num_tx_queues))
4022 			queue_index = -1;
4023 	}
4024 	return queue_index;
4025 }
4026 #endif
4027 
4028 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4029 			 struct sk_buff *skb)
4030 {
4031 #ifdef CONFIG_XPS
4032 	struct xps_dev_maps *dev_maps;
4033 	struct sock *sk = skb->sk;
4034 	int queue_index = -1;
4035 
4036 	if (!static_key_false(&xps_needed))
4037 		return -1;
4038 
4039 	rcu_read_lock();
4040 	if (!static_key_false(&xps_rxqs_needed))
4041 		goto get_cpus_map;
4042 
4043 	dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4044 	if (dev_maps) {
4045 		int tci = sk_rx_queue_get(sk);
4046 
4047 		if (tci >= 0)
4048 			queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4049 							  tci);
4050 	}
4051 
4052 get_cpus_map:
4053 	if (queue_index < 0) {
4054 		dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4055 		if (dev_maps) {
4056 			unsigned int tci = skb->sender_cpu - 1;
4057 
4058 			queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4059 							  tci);
4060 		}
4061 	}
4062 	rcu_read_unlock();
4063 
4064 	return queue_index;
4065 #else
4066 	return -1;
4067 #endif
4068 }
4069 
4070 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4071 		     struct net_device *sb_dev)
4072 {
4073 	return 0;
4074 }
4075 EXPORT_SYMBOL(dev_pick_tx_zero);
4076 
4077 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4078 		       struct net_device *sb_dev)
4079 {
4080 	return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4081 }
4082 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4083 
4084 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4085 		     struct net_device *sb_dev)
4086 {
4087 	struct sock *sk = skb->sk;
4088 	int queue_index = sk_tx_queue_get(sk);
4089 
4090 	sb_dev = sb_dev ? : dev;
4091 
4092 	if (queue_index < 0 || skb->ooo_okay ||
4093 	    queue_index >= dev->real_num_tx_queues) {
4094 		int new_index = get_xps_queue(dev, sb_dev, skb);
4095 
4096 		if (new_index < 0)
4097 			new_index = skb_tx_hash(dev, sb_dev, skb);
4098 
4099 		if (queue_index != new_index && sk &&
4100 		    sk_fullsock(sk) &&
4101 		    rcu_access_pointer(sk->sk_dst_cache))
4102 			sk_tx_queue_set(sk, new_index);
4103 
4104 		queue_index = new_index;
4105 	}
4106 
4107 	return queue_index;
4108 }
4109 EXPORT_SYMBOL(netdev_pick_tx);
4110 
4111 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4112 					 struct sk_buff *skb,
4113 					 struct net_device *sb_dev)
4114 {
4115 	int queue_index = 0;
4116 
4117 #ifdef CONFIG_XPS
4118 	u32 sender_cpu = skb->sender_cpu - 1;
4119 
4120 	if (sender_cpu >= (u32)NR_CPUS)
4121 		skb->sender_cpu = raw_smp_processor_id() + 1;
4122 #endif
4123 
4124 	if (dev->real_num_tx_queues != 1) {
4125 		const struct net_device_ops *ops = dev->netdev_ops;
4126 
4127 		if (ops->ndo_select_queue)
4128 			queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4129 		else
4130 			queue_index = netdev_pick_tx(dev, skb, sb_dev);
4131 
4132 		queue_index = netdev_cap_txqueue(dev, queue_index);
4133 	}
4134 
4135 	skb_set_queue_mapping(skb, queue_index);
4136 	return netdev_get_tx_queue(dev, queue_index);
4137 }
4138 
4139 /**
4140  * __dev_queue_xmit() - transmit a buffer
4141  * @skb:	buffer to transmit
4142  * @sb_dev:	suboordinate device used for L2 forwarding offload
4143  *
4144  * Queue a buffer for transmission to a network device. The caller must
4145  * have set the device and priority and built the buffer before calling
4146  * this function. The function can be called from an interrupt.
4147  *
4148  * When calling this method, interrupts MUST be enabled. This is because
4149  * the BH enable code must have IRQs enabled so that it will not deadlock.
4150  *
4151  * Regardless of the return value, the skb is consumed, so it is currently
4152  * difficult to retry a send to this method. (You can bump the ref count
4153  * before sending to hold a reference for retry if you are careful.)
4154  *
4155  * Return:
4156  * * 0				- buffer successfully transmitted
4157  * * positive qdisc return code	- NET_XMIT_DROP etc.
4158  * * negative errno		- other errors
4159  */
4160 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4161 {
4162 	struct net_device *dev = skb->dev;
4163 	struct netdev_queue *txq = NULL;
4164 	struct Qdisc *q;
4165 	int rc = -ENOMEM;
4166 	bool again = false;
4167 
4168 	skb_reset_mac_header(skb);
4169 
4170 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4171 		__skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4172 
4173 	/* Disable soft irqs for various locks below. Also
4174 	 * stops preemption for RCU.
4175 	 */
4176 	rcu_read_lock_bh();
4177 
4178 	skb_update_prio(skb);
4179 
4180 	qdisc_pkt_len_init(skb);
4181 #ifdef CONFIG_NET_CLS_ACT
4182 	skb->tc_at_ingress = 0;
4183 #endif
4184 #ifdef CONFIG_NET_EGRESS
4185 	if (static_branch_unlikely(&egress_needed_key)) {
4186 		if (nf_hook_egress_active()) {
4187 			skb = nf_hook_egress(skb, &rc, dev);
4188 			if (!skb)
4189 				goto out;
4190 		}
4191 
4192 		netdev_xmit_skip_txqueue(false);
4193 
4194 		nf_skip_egress(skb, true);
4195 		skb = sch_handle_egress(skb, &rc, dev);
4196 		if (!skb)
4197 			goto out;
4198 		nf_skip_egress(skb, false);
4199 
4200 		if (netdev_xmit_txqueue_skipped())
4201 			txq = netdev_tx_queue_mapping(dev, skb);
4202 	}
4203 #endif
4204 	/* If device/qdisc don't need skb->dst, release it right now while
4205 	 * its hot in this cpu cache.
4206 	 */
4207 	if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4208 		skb_dst_drop(skb);
4209 	else
4210 		skb_dst_force(skb);
4211 
4212 	if (!txq)
4213 		txq = netdev_core_pick_tx(dev, skb, sb_dev);
4214 
4215 	q = rcu_dereference_bh(txq->qdisc);
4216 
4217 	trace_net_dev_queue(skb);
4218 	if (q->enqueue) {
4219 		rc = __dev_xmit_skb(skb, q, dev, txq);
4220 		goto out;
4221 	}
4222 
4223 	/* The device has no queue. Common case for software devices:
4224 	 * loopback, all the sorts of tunnels...
4225 
4226 	 * Really, it is unlikely that netif_tx_lock protection is necessary
4227 	 * here.  (f.e. loopback and IP tunnels are clean ignoring statistics
4228 	 * counters.)
4229 	 * However, it is possible, that they rely on protection
4230 	 * made by us here.
4231 
4232 	 * Check this and shot the lock. It is not prone from deadlocks.
4233 	 *Either shot noqueue qdisc, it is even simpler 8)
4234 	 */
4235 	if (dev->flags & IFF_UP) {
4236 		int cpu = smp_processor_id(); /* ok because BHs are off */
4237 
4238 		/* Other cpus might concurrently change txq->xmit_lock_owner
4239 		 * to -1 or to their cpu id, but not to our id.
4240 		 */
4241 		if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4242 			if (dev_xmit_recursion())
4243 				goto recursion_alert;
4244 
4245 			skb = validate_xmit_skb(skb, dev, &again);
4246 			if (!skb)
4247 				goto out;
4248 
4249 			HARD_TX_LOCK(dev, txq, cpu);
4250 
4251 			if (!netif_xmit_stopped(txq)) {
4252 				dev_xmit_recursion_inc();
4253 				skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4254 				dev_xmit_recursion_dec();
4255 				if (dev_xmit_complete(rc)) {
4256 					HARD_TX_UNLOCK(dev, txq);
4257 					goto out;
4258 				}
4259 			}
4260 			HARD_TX_UNLOCK(dev, txq);
4261 			net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4262 					     dev->name);
4263 		} else {
4264 			/* Recursion is detected! It is possible,
4265 			 * unfortunately
4266 			 */
4267 recursion_alert:
4268 			net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4269 					     dev->name);
4270 		}
4271 	}
4272 
4273 	rc = -ENETDOWN;
4274 	rcu_read_unlock_bh();
4275 
4276 	dev_core_stats_tx_dropped_inc(dev);
4277 	kfree_skb_list(skb);
4278 	return rc;
4279 out:
4280 	rcu_read_unlock_bh();
4281 	return rc;
4282 }
4283 EXPORT_SYMBOL(__dev_queue_xmit);
4284 
4285 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4286 {
4287 	struct net_device *dev = skb->dev;
4288 	struct sk_buff *orig_skb = skb;
4289 	struct netdev_queue *txq;
4290 	int ret = NETDEV_TX_BUSY;
4291 	bool again = false;
4292 
4293 	if (unlikely(!netif_running(dev) ||
4294 		     !netif_carrier_ok(dev)))
4295 		goto drop;
4296 
4297 	skb = validate_xmit_skb_list(skb, dev, &again);
4298 	if (skb != orig_skb)
4299 		goto drop;
4300 
4301 	skb_set_queue_mapping(skb, queue_id);
4302 	txq = skb_get_tx_queue(dev, skb);
4303 
4304 	local_bh_disable();
4305 
4306 	dev_xmit_recursion_inc();
4307 	HARD_TX_LOCK(dev, txq, smp_processor_id());
4308 	if (!netif_xmit_frozen_or_drv_stopped(txq))
4309 		ret = netdev_start_xmit(skb, dev, txq, false);
4310 	HARD_TX_UNLOCK(dev, txq);
4311 	dev_xmit_recursion_dec();
4312 
4313 	local_bh_enable();
4314 	return ret;
4315 drop:
4316 	dev_core_stats_tx_dropped_inc(dev);
4317 	kfree_skb_list(skb);
4318 	return NET_XMIT_DROP;
4319 }
4320 EXPORT_SYMBOL(__dev_direct_xmit);
4321 
4322 /*************************************************************************
4323  *			Receiver routines
4324  *************************************************************************/
4325 
4326 int netdev_max_backlog __read_mostly = 1000;
4327 EXPORT_SYMBOL(netdev_max_backlog);
4328 
4329 int netdev_tstamp_prequeue __read_mostly = 1;
4330 unsigned int sysctl_skb_defer_max __read_mostly = 64;
4331 int netdev_budget __read_mostly = 300;
4332 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4333 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4334 int weight_p __read_mostly = 64;           /* old backlog weight */
4335 int dev_weight_rx_bias __read_mostly = 1;  /* bias for backlog weight */
4336 int dev_weight_tx_bias __read_mostly = 1;  /* bias for output_queue quota */
4337 int dev_rx_weight __read_mostly = 64;
4338 int dev_tx_weight __read_mostly = 64;
4339 
4340 /* Called with irq disabled */
4341 static inline void ____napi_schedule(struct softnet_data *sd,
4342 				     struct napi_struct *napi)
4343 {
4344 	struct task_struct *thread;
4345 
4346 	lockdep_assert_irqs_disabled();
4347 
4348 	if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4349 		/* Paired with smp_mb__before_atomic() in
4350 		 * napi_enable()/dev_set_threaded().
4351 		 * Use READ_ONCE() to guarantee a complete
4352 		 * read on napi->thread. Only call
4353 		 * wake_up_process() when it's not NULL.
4354 		 */
4355 		thread = READ_ONCE(napi->thread);
4356 		if (thread) {
4357 			/* Avoid doing set_bit() if the thread is in
4358 			 * INTERRUPTIBLE state, cause napi_thread_wait()
4359 			 * makes sure to proceed with napi polling
4360 			 * if the thread is explicitly woken from here.
4361 			 */
4362 			if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4363 				set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4364 			wake_up_process(thread);
4365 			return;
4366 		}
4367 	}
4368 
4369 	list_add_tail(&napi->poll_list, &sd->poll_list);
4370 	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4371 }
4372 
4373 #ifdef CONFIG_RPS
4374 
4375 /* One global table that all flow-based protocols share. */
4376 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4377 EXPORT_SYMBOL(rps_sock_flow_table);
4378 u32 rps_cpu_mask __read_mostly;
4379 EXPORT_SYMBOL(rps_cpu_mask);
4380 
4381 struct static_key_false rps_needed __read_mostly;
4382 EXPORT_SYMBOL(rps_needed);
4383 struct static_key_false rfs_needed __read_mostly;
4384 EXPORT_SYMBOL(rfs_needed);
4385 
4386 static struct rps_dev_flow *
4387 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4388 	    struct rps_dev_flow *rflow, u16 next_cpu)
4389 {
4390 	if (next_cpu < nr_cpu_ids) {
4391 #ifdef CONFIG_RFS_ACCEL
4392 		struct netdev_rx_queue *rxqueue;
4393 		struct rps_dev_flow_table *flow_table;
4394 		struct rps_dev_flow *old_rflow;
4395 		u32 flow_id;
4396 		u16 rxq_index;
4397 		int rc;
4398 
4399 		/* Should we steer this flow to a different hardware queue? */
4400 		if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4401 		    !(dev->features & NETIF_F_NTUPLE))
4402 			goto out;
4403 		rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4404 		if (rxq_index == skb_get_rx_queue(skb))
4405 			goto out;
4406 
4407 		rxqueue = dev->_rx + rxq_index;
4408 		flow_table = rcu_dereference(rxqueue->rps_flow_table);
4409 		if (!flow_table)
4410 			goto out;
4411 		flow_id = skb_get_hash(skb) & flow_table->mask;
4412 		rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4413 							rxq_index, flow_id);
4414 		if (rc < 0)
4415 			goto out;
4416 		old_rflow = rflow;
4417 		rflow = &flow_table->flows[flow_id];
4418 		rflow->filter = rc;
4419 		if (old_rflow->filter == rflow->filter)
4420 			old_rflow->filter = RPS_NO_FILTER;
4421 	out:
4422 #endif
4423 		rflow->last_qtail =
4424 			per_cpu(softnet_data, next_cpu).input_queue_head;
4425 	}
4426 
4427 	rflow->cpu = next_cpu;
4428 	return rflow;
4429 }
4430 
4431 /*
4432  * get_rps_cpu is called from netif_receive_skb and returns the target
4433  * CPU from the RPS map of the receiving queue for a given skb.
4434  * rcu_read_lock must be held on entry.
4435  */
4436 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4437 		       struct rps_dev_flow **rflowp)
4438 {
4439 	const struct rps_sock_flow_table *sock_flow_table;
4440 	struct netdev_rx_queue *rxqueue = dev->_rx;
4441 	struct rps_dev_flow_table *flow_table;
4442 	struct rps_map *map;
4443 	int cpu = -1;
4444 	u32 tcpu;
4445 	u32 hash;
4446 
4447 	if (skb_rx_queue_recorded(skb)) {
4448 		u16 index = skb_get_rx_queue(skb);
4449 
4450 		if (unlikely(index >= dev->real_num_rx_queues)) {
4451 			WARN_ONCE(dev->real_num_rx_queues > 1,
4452 				  "%s received packet on queue %u, but number "
4453 				  "of RX queues is %u\n",
4454 				  dev->name, index, dev->real_num_rx_queues);
4455 			goto done;
4456 		}
4457 		rxqueue += index;
4458 	}
4459 
4460 	/* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4461 
4462 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
4463 	map = rcu_dereference(rxqueue->rps_map);
4464 	if (!flow_table && !map)
4465 		goto done;
4466 
4467 	skb_reset_network_header(skb);
4468 	hash = skb_get_hash(skb);
4469 	if (!hash)
4470 		goto done;
4471 
4472 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
4473 	if (flow_table && sock_flow_table) {
4474 		struct rps_dev_flow *rflow;
4475 		u32 next_cpu;
4476 		u32 ident;
4477 
4478 		/* First check into global flow table if there is a match */
4479 		ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4480 		if ((ident ^ hash) & ~rps_cpu_mask)
4481 			goto try_rps;
4482 
4483 		next_cpu = ident & rps_cpu_mask;
4484 
4485 		/* OK, now we know there is a match,
4486 		 * we can look at the local (per receive queue) flow table
4487 		 */
4488 		rflow = &flow_table->flows[hash & flow_table->mask];
4489 		tcpu = rflow->cpu;
4490 
4491 		/*
4492 		 * If the desired CPU (where last recvmsg was done) is
4493 		 * different from current CPU (one in the rx-queue flow
4494 		 * table entry), switch if one of the following holds:
4495 		 *   - Current CPU is unset (>= nr_cpu_ids).
4496 		 *   - Current CPU is offline.
4497 		 *   - The current CPU's queue tail has advanced beyond the
4498 		 *     last packet that was enqueued using this table entry.
4499 		 *     This guarantees that all previous packets for the flow
4500 		 *     have been dequeued, thus preserving in order delivery.
4501 		 */
4502 		if (unlikely(tcpu != next_cpu) &&
4503 		    (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4504 		     ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4505 		      rflow->last_qtail)) >= 0)) {
4506 			tcpu = next_cpu;
4507 			rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4508 		}
4509 
4510 		if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4511 			*rflowp = rflow;
4512 			cpu = tcpu;
4513 			goto done;
4514 		}
4515 	}
4516 
4517 try_rps:
4518 
4519 	if (map) {
4520 		tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4521 		if (cpu_online(tcpu)) {
4522 			cpu = tcpu;
4523 			goto done;
4524 		}
4525 	}
4526 
4527 done:
4528 	return cpu;
4529 }
4530 
4531 #ifdef CONFIG_RFS_ACCEL
4532 
4533 /**
4534  * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4535  * @dev: Device on which the filter was set
4536  * @rxq_index: RX queue index
4537  * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4538  * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4539  *
4540  * Drivers that implement ndo_rx_flow_steer() should periodically call
4541  * this function for each installed filter and remove the filters for
4542  * which it returns %true.
4543  */
4544 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4545 			 u32 flow_id, u16 filter_id)
4546 {
4547 	struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4548 	struct rps_dev_flow_table *flow_table;
4549 	struct rps_dev_flow *rflow;
4550 	bool expire = true;
4551 	unsigned int cpu;
4552 
4553 	rcu_read_lock();
4554 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
4555 	if (flow_table && flow_id <= flow_table->mask) {
4556 		rflow = &flow_table->flows[flow_id];
4557 		cpu = READ_ONCE(rflow->cpu);
4558 		if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4559 		    ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4560 			   rflow->last_qtail) <
4561 		     (int)(10 * flow_table->mask)))
4562 			expire = false;
4563 	}
4564 	rcu_read_unlock();
4565 	return expire;
4566 }
4567 EXPORT_SYMBOL(rps_may_expire_flow);
4568 
4569 #endif /* CONFIG_RFS_ACCEL */
4570 
4571 /* Called from hardirq (IPI) context */
4572 static void rps_trigger_softirq(void *data)
4573 {
4574 	struct softnet_data *sd = data;
4575 
4576 	____napi_schedule(sd, &sd->backlog);
4577 	sd->received_rps++;
4578 }
4579 
4580 #endif /* CONFIG_RPS */
4581 
4582 /* Called from hardirq (IPI) context */
4583 static void trigger_rx_softirq(void *data)
4584 {
4585 	struct softnet_data *sd = data;
4586 
4587 	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4588 	smp_store_release(&sd->defer_ipi_scheduled, 0);
4589 }
4590 
4591 /*
4592  * Check if this softnet_data structure is another cpu one
4593  * If yes, queue it to our IPI list and return 1
4594  * If no, return 0
4595  */
4596 static int napi_schedule_rps(struct softnet_data *sd)
4597 {
4598 	struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4599 
4600 #ifdef CONFIG_RPS
4601 	if (sd != mysd) {
4602 		sd->rps_ipi_next = mysd->rps_ipi_list;
4603 		mysd->rps_ipi_list = sd;
4604 
4605 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4606 		return 1;
4607 	}
4608 #endif /* CONFIG_RPS */
4609 	__napi_schedule_irqoff(&mysd->backlog);
4610 	return 0;
4611 }
4612 
4613 #ifdef CONFIG_NET_FLOW_LIMIT
4614 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4615 #endif
4616 
4617 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4618 {
4619 #ifdef CONFIG_NET_FLOW_LIMIT
4620 	struct sd_flow_limit *fl;
4621 	struct softnet_data *sd;
4622 	unsigned int old_flow, new_flow;
4623 
4624 	if (qlen < (netdev_max_backlog >> 1))
4625 		return false;
4626 
4627 	sd = this_cpu_ptr(&softnet_data);
4628 
4629 	rcu_read_lock();
4630 	fl = rcu_dereference(sd->flow_limit);
4631 	if (fl) {
4632 		new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4633 		old_flow = fl->history[fl->history_head];
4634 		fl->history[fl->history_head] = new_flow;
4635 
4636 		fl->history_head++;
4637 		fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4638 
4639 		if (likely(fl->buckets[old_flow]))
4640 			fl->buckets[old_flow]--;
4641 
4642 		if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4643 			fl->count++;
4644 			rcu_read_unlock();
4645 			return true;
4646 		}
4647 	}
4648 	rcu_read_unlock();
4649 #endif
4650 	return false;
4651 }
4652 
4653 /*
4654  * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4655  * queue (may be a remote CPU queue).
4656  */
4657 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4658 			      unsigned int *qtail)
4659 {
4660 	enum skb_drop_reason reason;
4661 	struct softnet_data *sd;
4662 	unsigned long flags;
4663 	unsigned int qlen;
4664 
4665 	reason = SKB_DROP_REASON_NOT_SPECIFIED;
4666 	sd = &per_cpu(softnet_data, cpu);
4667 
4668 	rps_lock_irqsave(sd, &flags);
4669 	if (!netif_running(skb->dev))
4670 		goto drop;
4671 	qlen = skb_queue_len(&sd->input_pkt_queue);
4672 	if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4673 		if (qlen) {
4674 enqueue:
4675 			__skb_queue_tail(&sd->input_pkt_queue, skb);
4676 			input_queue_tail_incr_save(sd, qtail);
4677 			rps_unlock_irq_restore(sd, &flags);
4678 			return NET_RX_SUCCESS;
4679 		}
4680 
4681 		/* Schedule NAPI for backlog device
4682 		 * We can use non atomic operation since we own the queue lock
4683 		 */
4684 		if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
4685 			napi_schedule_rps(sd);
4686 		goto enqueue;
4687 	}
4688 	reason = SKB_DROP_REASON_CPU_BACKLOG;
4689 
4690 drop:
4691 	sd->dropped++;
4692 	rps_unlock_irq_restore(sd, &flags);
4693 
4694 	dev_core_stats_rx_dropped_inc(skb->dev);
4695 	kfree_skb_reason(skb, reason);
4696 	return NET_RX_DROP;
4697 }
4698 
4699 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4700 {
4701 	struct net_device *dev = skb->dev;
4702 	struct netdev_rx_queue *rxqueue;
4703 
4704 	rxqueue = dev->_rx;
4705 
4706 	if (skb_rx_queue_recorded(skb)) {
4707 		u16 index = skb_get_rx_queue(skb);
4708 
4709 		if (unlikely(index >= dev->real_num_rx_queues)) {
4710 			WARN_ONCE(dev->real_num_rx_queues > 1,
4711 				  "%s received packet on queue %u, but number "
4712 				  "of RX queues is %u\n",
4713 				  dev->name, index, dev->real_num_rx_queues);
4714 
4715 			return rxqueue; /* Return first rxqueue */
4716 		}
4717 		rxqueue += index;
4718 	}
4719 	return rxqueue;
4720 }
4721 
4722 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4723 			     struct bpf_prog *xdp_prog)
4724 {
4725 	void *orig_data, *orig_data_end, *hard_start;
4726 	struct netdev_rx_queue *rxqueue;
4727 	bool orig_bcast, orig_host;
4728 	u32 mac_len, frame_sz;
4729 	__be16 orig_eth_type;
4730 	struct ethhdr *eth;
4731 	u32 metalen, act;
4732 	int off;
4733 
4734 	/* The XDP program wants to see the packet starting at the MAC
4735 	 * header.
4736 	 */
4737 	mac_len = skb->data - skb_mac_header(skb);
4738 	hard_start = skb->data - skb_headroom(skb);
4739 
4740 	/* SKB "head" area always have tailroom for skb_shared_info */
4741 	frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4742 	frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4743 
4744 	rxqueue = netif_get_rxqueue(skb);
4745 	xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4746 	xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4747 			 skb_headlen(skb) + mac_len, true);
4748 
4749 	orig_data_end = xdp->data_end;
4750 	orig_data = xdp->data;
4751 	eth = (struct ethhdr *)xdp->data;
4752 	orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4753 	orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4754 	orig_eth_type = eth->h_proto;
4755 
4756 	act = bpf_prog_run_xdp(xdp_prog, xdp);
4757 
4758 	/* check if bpf_xdp_adjust_head was used */
4759 	off = xdp->data - orig_data;
4760 	if (off) {
4761 		if (off > 0)
4762 			__skb_pull(skb, off);
4763 		else if (off < 0)
4764 			__skb_push(skb, -off);
4765 
4766 		skb->mac_header += off;
4767 		skb_reset_network_header(skb);
4768 	}
4769 
4770 	/* check if bpf_xdp_adjust_tail was used */
4771 	off = xdp->data_end - orig_data_end;
4772 	if (off != 0) {
4773 		skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4774 		skb->len += off; /* positive on grow, negative on shrink */
4775 	}
4776 
4777 	/* check if XDP changed eth hdr such SKB needs update */
4778 	eth = (struct ethhdr *)xdp->data;
4779 	if ((orig_eth_type != eth->h_proto) ||
4780 	    (orig_host != ether_addr_equal_64bits(eth->h_dest,
4781 						  skb->dev->dev_addr)) ||
4782 	    (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4783 		__skb_push(skb, ETH_HLEN);
4784 		skb->pkt_type = PACKET_HOST;
4785 		skb->protocol = eth_type_trans(skb, skb->dev);
4786 	}
4787 
4788 	/* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4789 	 * before calling us again on redirect path. We do not call do_redirect
4790 	 * as we leave that up to the caller.
4791 	 *
4792 	 * Caller is responsible for managing lifetime of skb (i.e. calling
4793 	 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4794 	 */
4795 	switch (act) {
4796 	case XDP_REDIRECT:
4797 	case XDP_TX:
4798 		__skb_push(skb, mac_len);
4799 		break;
4800 	case XDP_PASS:
4801 		metalen = xdp->data - xdp->data_meta;
4802 		if (metalen)
4803 			skb_metadata_set(skb, metalen);
4804 		break;
4805 	}
4806 
4807 	return act;
4808 }
4809 
4810 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4811 				     struct xdp_buff *xdp,
4812 				     struct bpf_prog *xdp_prog)
4813 {
4814 	u32 act = XDP_DROP;
4815 
4816 	/* Reinjected packets coming from act_mirred or similar should
4817 	 * not get XDP generic processing.
4818 	 */
4819 	if (skb_is_redirected(skb))
4820 		return XDP_PASS;
4821 
4822 	/* XDP packets must be linear and must have sufficient headroom
4823 	 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4824 	 * native XDP provides, thus we need to do it here as well.
4825 	 */
4826 	if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4827 	    skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4828 		int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4829 		int troom = skb->tail + skb->data_len - skb->end;
4830 
4831 		/* In case we have to go down the path and also linearize,
4832 		 * then lets do the pskb_expand_head() work just once here.
4833 		 */
4834 		if (pskb_expand_head(skb,
4835 				     hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4836 				     troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4837 			goto do_drop;
4838 		if (skb_linearize(skb))
4839 			goto do_drop;
4840 	}
4841 
4842 	act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4843 	switch (act) {
4844 	case XDP_REDIRECT:
4845 	case XDP_TX:
4846 	case XDP_PASS:
4847 		break;
4848 	default:
4849 		bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act);
4850 		fallthrough;
4851 	case XDP_ABORTED:
4852 		trace_xdp_exception(skb->dev, xdp_prog, act);
4853 		fallthrough;
4854 	case XDP_DROP:
4855 	do_drop:
4856 		kfree_skb(skb);
4857 		break;
4858 	}
4859 
4860 	return act;
4861 }
4862 
4863 /* When doing generic XDP we have to bypass the qdisc layer and the
4864  * network taps in order to match in-driver-XDP behavior.
4865  */
4866 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4867 {
4868 	struct net_device *dev = skb->dev;
4869 	struct netdev_queue *txq;
4870 	bool free_skb = true;
4871 	int cpu, rc;
4872 
4873 	txq = netdev_core_pick_tx(dev, skb, NULL);
4874 	cpu = smp_processor_id();
4875 	HARD_TX_LOCK(dev, txq, cpu);
4876 	if (!netif_xmit_stopped(txq)) {
4877 		rc = netdev_start_xmit(skb, dev, txq, 0);
4878 		if (dev_xmit_complete(rc))
4879 			free_skb = false;
4880 	}
4881 	HARD_TX_UNLOCK(dev, txq);
4882 	if (free_skb) {
4883 		trace_xdp_exception(dev, xdp_prog, XDP_TX);
4884 		kfree_skb(skb);
4885 	}
4886 }
4887 
4888 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4889 
4890 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4891 {
4892 	if (xdp_prog) {
4893 		struct xdp_buff xdp;
4894 		u32 act;
4895 		int err;
4896 
4897 		act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4898 		if (act != XDP_PASS) {
4899 			switch (act) {
4900 			case XDP_REDIRECT:
4901 				err = xdp_do_generic_redirect(skb->dev, skb,
4902 							      &xdp, xdp_prog);
4903 				if (err)
4904 					goto out_redir;
4905 				break;
4906 			case XDP_TX:
4907 				generic_xdp_tx(skb, xdp_prog);
4908 				break;
4909 			}
4910 			return XDP_DROP;
4911 		}
4912 	}
4913 	return XDP_PASS;
4914 out_redir:
4915 	kfree_skb_reason(skb, SKB_DROP_REASON_XDP);
4916 	return XDP_DROP;
4917 }
4918 EXPORT_SYMBOL_GPL(do_xdp_generic);
4919 
4920 static int netif_rx_internal(struct sk_buff *skb)
4921 {
4922 	int ret;
4923 
4924 	net_timestamp_check(netdev_tstamp_prequeue, skb);
4925 
4926 	trace_netif_rx(skb);
4927 
4928 #ifdef CONFIG_RPS
4929 	if (static_branch_unlikely(&rps_needed)) {
4930 		struct rps_dev_flow voidflow, *rflow = &voidflow;
4931 		int cpu;
4932 
4933 		rcu_read_lock();
4934 
4935 		cpu = get_rps_cpu(skb->dev, skb, &rflow);
4936 		if (cpu < 0)
4937 			cpu = smp_processor_id();
4938 
4939 		ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4940 
4941 		rcu_read_unlock();
4942 	} else
4943 #endif
4944 	{
4945 		unsigned int qtail;
4946 
4947 		ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
4948 	}
4949 	return ret;
4950 }
4951 
4952 /**
4953  *	__netif_rx	-	Slightly optimized version of netif_rx
4954  *	@skb: buffer to post
4955  *
4956  *	This behaves as netif_rx except that it does not disable bottom halves.
4957  *	As a result this function may only be invoked from the interrupt context
4958  *	(either hard or soft interrupt).
4959  */
4960 int __netif_rx(struct sk_buff *skb)
4961 {
4962 	int ret;
4963 
4964 	lockdep_assert_once(hardirq_count() | softirq_count());
4965 
4966 	trace_netif_rx_entry(skb);
4967 	ret = netif_rx_internal(skb);
4968 	trace_netif_rx_exit(ret);
4969 	return ret;
4970 }
4971 EXPORT_SYMBOL(__netif_rx);
4972 
4973 /**
4974  *	netif_rx	-	post buffer to the network code
4975  *	@skb: buffer to post
4976  *
4977  *	This function receives a packet from a device driver and queues it for
4978  *	the upper (protocol) levels to process via the backlog NAPI device. It
4979  *	always succeeds. The buffer may be dropped during processing for
4980  *	congestion control or by the protocol layers.
4981  *	The network buffer is passed via the backlog NAPI device. Modern NIC
4982  *	driver should use NAPI and GRO.
4983  *	This function can used from interrupt and from process context. The
4984  *	caller from process context must not disable interrupts before invoking
4985  *	this function.
4986  *
4987  *	return values:
4988  *	NET_RX_SUCCESS	(no congestion)
4989  *	NET_RX_DROP     (packet was dropped)
4990  *
4991  */
4992 int netif_rx(struct sk_buff *skb)
4993 {
4994 	bool need_bh_off = !(hardirq_count() | softirq_count());
4995 	int ret;
4996 
4997 	if (need_bh_off)
4998 		local_bh_disable();
4999 	trace_netif_rx_entry(skb);
5000 	ret = netif_rx_internal(skb);
5001 	trace_netif_rx_exit(ret);
5002 	if (need_bh_off)
5003 		local_bh_enable();
5004 	return ret;
5005 }
5006 EXPORT_SYMBOL(netif_rx);
5007 
5008 static __latent_entropy void net_tx_action(struct softirq_action *h)
5009 {
5010 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5011 
5012 	if (sd->completion_queue) {
5013 		struct sk_buff *clist;
5014 
5015 		local_irq_disable();
5016 		clist = sd->completion_queue;
5017 		sd->completion_queue = NULL;
5018 		local_irq_enable();
5019 
5020 		while (clist) {
5021 			struct sk_buff *skb = clist;
5022 
5023 			clist = clist->next;
5024 
5025 			WARN_ON(refcount_read(&skb->users));
5026 			if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
5027 				trace_consume_skb(skb);
5028 			else
5029 				trace_kfree_skb(skb, net_tx_action,
5030 						SKB_DROP_REASON_NOT_SPECIFIED);
5031 
5032 			if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5033 				__kfree_skb(skb);
5034 			else
5035 				__kfree_skb_defer(skb);
5036 		}
5037 	}
5038 
5039 	if (sd->output_queue) {
5040 		struct Qdisc *head;
5041 
5042 		local_irq_disable();
5043 		head = sd->output_queue;
5044 		sd->output_queue = NULL;
5045 		sd->output_queue_tailp = &sd->output_queue;
5046 		local_irq_enable();
5047 
5048 		rcu_read_lock();
5049 
5050 		while (head) {
5051 			struct Qdisc *q = head;
5052 			spinlock_t *root_lock = NULL;
5053 
5054 			head = head->next_sched;
5055 
5056 			/* We need to make sure head->next_sched is read
5057 			 * before clearing __QDISC_STATE_SCHED
5058 			 */
5059 			smp_mb__before_atomic();
5060 
5061 			if (!(q->flags & TCQ_F_NOLOCK)) {
5062 				root_lock = qdisc_lock(q);
5063 				spin_lock(root_lock);
5064 			} else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5065 						     &q->state))) {
5066 				/* There is a synchronize_net() between
5067 				 * STATE_DEACTIVATED flag being set and
5068 				 * qdisc_reset()/some_qdisc_is_busy() in
5069 				 * dev_deactivate(), so we can safely bail out
5070 				 * early here to avoid data race between
5071 				 * qdisc_deactivate() and some_qdisc_is_busy()
5072 				 * for lockless qdisc.
5073 				 */
5074 				clear_bit(__QDISC_STATE_SCHED, &q->state);
5075 				continue;
5076 			}
5077 
5078 			clear_bit(__QDISC_STATE_SCHED, &q->state);
5079 			qdisc_run(q);
5080 			if (root_lock)
5081 				spin_unlock(root_lock);
5082 		}
5083 
5084 		rcu_read_unlock();
5085 	}
5086 
5087 	xfrm_dev_backlog(sd);
5088 }
5089 
5090 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5091 /* This hook is defined here for ATM LANE */
5092 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5093 			     unsigned char *addr) __read_mostly;
5094 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5095 #endif
5096 
5097 static inline struct sk_buff *
5098 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5099 		   struct net_device *orig_dev, bool *another)
5100 {
5101 #ifdef CONFIG_NET_CLS_ACT
5102 	struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5103 	struct tcf_result cl_res;
5104 
5105 	/* If there's at least one ingress present somewhere (so
5106 	 * we get here via enabled static key), remaining devices
5107 	 * that are not configured with an ingress qdisc will bail
5108 	 * out here.
5109 	 */
5110 	if (!miniq)
5111 		return skb;
5112 
5113 	if (*pt_prev) {
5114 		*ret = deliver_skb(skb, *pt_prev, orig_dev);
5115 		*pt_prev = NULL;
5116 	}
5117 
5118 	qdisc_skb_cb(skb)->pkt_len = skb->len;
5119 	tc_skb_cb(skb)->mru = 0;
5120 	tc_skb_cb(skb)->post_ct = false;
5121 	skb->tc_at_ingress = 1;
5122 	mini_qdisc_bstats_cpu_update(miniq, skb);
5123 
5124 	switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
5125 	case TC_ACT_OK:
5126 	case TC_ACT_RECLASSIFY:
5127 		skb->tc_index = TC_H_MIN(cl_res.classid);
5128 		break;
5129 	case TC_ACT_SHOT:
5130 		mini_qdisc_qstats_cpu_drop(miniq);
5131 		kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS);
5132 		return NULL;
5133 	case TC_ACT_STOLEN:
5134 	case TC_ACT_QUEUED:
5135 	case TC_ACT_TRAP:
5136 		consume_skb(skb);
5137 		return NULL;
5138 	case TC_ACT_REDIRECT:
5139 		/* skb_mac_header check was done by cls/act_bpf, so
5140 		 * we can safely push the L2 header back before
5141 		 * redirecting to another netdev
5142 		 */
5143 		__skb_push(skb, skb->mac_len);
5144 		if (skb_do_redirect(skb) == -EAGAIN) {
5145 			__skb_pull(skb, skb->mac_len);
5146 			*another = true;
5147 			break;
5148 		}
5149 		return NULL;
5150 	case TC_ACT_CONSUMED:
5151 		return NULL;
5152 	default:
5153 		break;
5154 	}
5155 #endif /* CONFIG_NET_CLS_ACT */
5156 	return skb;
5157 }
5158 
5159 /**
5160  *	netdev_is_rx_handler_busy - check if receive handler is registered
5161  *	@dev: device to check
5162  *
5163  *	Check if a receive handler is already registered for a given device.
5164  *	Return true if there one.
5165  *
5166  *	The caller must hold the rtnl_mutex.
5167  */
5168 bool netdev_is_rx_handler_busy(struct net_device *dev)
5169 {
5170 	ASSERT_RTNL();
5171 	return dev && rtnl_dereference(dev->rx_handler);
5172 }
5173 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5174 
5175 /**
5176  *	netdev_rx_handler_register - register receive handler
5177  *	@dev: device to register a handler for
5178  *	@rx_handler: receive handler to register
5179  *	@rx_handler_data: data pointer that is used by rx handler
5180  *
5181  *	Register a receive handler for a device. This handler will then be
5182  *	called from __netif_receive_skb. A negative errno code is returned
5183  *	on a failure.
5184  *
5185  *	The caller must hold the rtnl_mutex.
5186  *
5187  *	For a general description of rx_handler, see enum rx_handler_result.
5188  */
5189 int netdev_rx_handler_register(struct net_device *dev,
5190 			       rx_handler_func_t *rx_handler,
5191 			       void *rx_handler_data)
5192 {
5193 	if (netdev_is_rx_handler_busy(dev))
5194 		return -EBUSY;
5195 
5196 	if (dev->priv_flags & IFF_NO_RX_HANDLER)
5197 		return -EINVAL;
5198 
5199 	/* Note: rx_handler_data must be set before rx_handler */
5200 	rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5201 	rcu_assign_pointer(dev->rx_handler, rx_handler);
5202 
5203 	return 0;
5204 }
5205 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5206 
5207 /**
5208  *	netdev_rx_handler_unregister - unregister receive handler
5209  *	@dev: device to unregister a handler from
5210  *
5211  *	Unregister a receive handler from a device.
5212  *
5213  *	The caller must hold the rtnl_mutex.
5214  */
5215 void netdev_rx_handler_unregister(struct net_device *dev)
5216 {
5217 
5218 	ASSERT_RTNL();
5219 	RCU_INIT_POINTER(dev->rx_handler, NULL);
5220 	/* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5221 	 * section has a guarantee to see a non NULL rx_handler_data
5222 	 * as well.
5223 	 */
5224 	synchronize_net();
5225 	RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5226 }
5227 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5228 
5229 /*
5230  * Limit the use of PFMEMALLOC reserves to those protocols that implement
5231  * the special handling of PFMEMALLOC skbs.
5232  */
5233 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5234 {
5235 	switch (skb->protocol) {
5236 	case htons(ETH_P_ARP):
5237 	case htons(ETH_P_IP):
5238 	case htons(ETH_P_IPV6):
5239 	case htons(ETH_P_8021Q):
5240 	case htons(ETH_P_8021AD):
5241 		return true;
5242 	default:
5243 		return false;
5244 	}
5245 }
5246 
5247 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5248 			     int *ret, struct net_device *orig_dev)
5249 {
5250 	if (nf_hook_ingress_active(skb)) {
5251 		int ingress_retval;
5252 
5253 		if (*pt_prev) {
5254 			*ret = deliver_skb(skb, *pt_prev, orig_dev);
5255 			*pt_prev = NULL;
5256 		}
5257 
5258 		rcu_read_lock();
5259 		ingress_retval = nf_hook_ingress(skb);
5260 		rcu_read_unlock();
5261 		return ingress_retval;
5262 	}
5263 	return 0;
5264 }
5265 
5266 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5267 				    struct packet_type **ppt_prev)
5268 {
5269 	struct packet_type *ptype, *pt_prev;
5270 	rx_handler_func_t *rx_handler;
5271 	struct sk_buff *skb = *pskb;
5272 	struct net_device *orig_dev;
5273 	bool deliver_exact = false;
5274 	int ret = NET_RX_DROP;
5275 	__be16 type;
5276 
5277 	net_timestamp_check(!netdev_tstamp_prequeue, skb);
5278 
5279 	trace_netif_receive_skb(skb);
5280 
5281 	orig_dev = skb->dev;
5282 
5283 	skb_reset_network_header(skb);
5284 	if (!skb_transport_header_was_set(skb))
5285 		skb_reset_transport_header(skb);
5286 	skb_reset_mac_len(skb);
5287 
5288 	pt_prev = NULL;
5289 
5290 another_round:
5291 	skb->skb_iif = skb->dev->ifindex;
5292 
5293 	__this_cpu_inc(softnet_data.processed);
5294 
5295 	if (static_branch_unlikely(&generic_xdp_needed_key)) {
5296 		int ret2;
5297 
5298 		migrate_disable();
5299 		ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5300 		migrate_enable();
5301 
5302 		if (ret2 != XDP_PASS) {
5303 			ret = NET_RX_DROP;
5304 			goto out;
5305 		}
5306 	}
5307 
5308 	if (eth_type_vlan(skb->protocol)) {
5309 		skb = skb_vlan_untag(skb);
5310 		if (unlikely(!skb))
5311 			goto out;
5312 	}
5313 
5314 	if (skb_skip_tc_classify(skb))
5315 		goto skip_classify;
5316 
5317 	if (pfmemalloc)
5318 		goto skip_taps;
5319 
5320 	list_for_each_entry_rcu(ptype, &ptype_all, list) {
5321 		if (pt_prev)
5322 			ret = deliver_skb(skb, pt_prev, orig_dev);
5323 		pt_prev = ptype;
5324 	}
5325 
5326 	list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5327 		if (pt_prev)
5328 			ret = deliver_skb(skb, pt_prev, orig_dev);
5329 		pt_prev = ptype;
5330 	}
5331 
5332 skip_taps:
5333 #ifdef CONFIG_NET_INGRESS
5334 	if (static_branch_unlikely(&ingress_needed_key)) {
5335 		bool another = false;
5336 
5337 		nf_skip_egress(skb, true);
5338 		skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5339 					 &another);
5340 		if (another)
5341 			goto another_round;
5342 		if (!skb)
5343 			goto out;
5344 
5345 		nf_skip_egress(skb, false);
5346 		if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5347 			goto out;
5348 	}
5349 #endif
5350 	skb_reset_redirect(skb);
5351 skip_classify:
5352 	if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5353 		goto drop;
5354 
5355 	if (skb_vlan_tag_present(skb)) {
5356 		if (pt_prev) {
5357 			ret = deliver_skb(skb, pt_prev, orig_dev);
5358 			pt_prev = NULL;
5359 		}
5360 		if (vlan_do_receive(&skb))
5361 			goto another_round;
5362 		else if (unlikely(!skb))
5363 			goto out;
5364 	}
5365 
5366 	rx_handler = rcu_dereference(skb->dev->rx_handler);
5367 	if (rx_handler) {
5368 		if (pt_prev) {
5369 			ret = deliver_skb(skb, pt_prev, orig_dev);
5370 			pt_prev = NULL;
5371 		}
5372 		switch (rx_handler(&skb)) {
5373 		case RX_HANDLER_CONSUMED:
5374 			ret = NET_RX_SUCCESS;
5375 			goto out;
5376 		case RX_HANDLER_ANOTHER:
5377 			goto another_round;
5378 		case RX_HANDLER_EXACT:
5379 			deliver_exact = true;
5380 			break;
5381 		case RX_HANDLER_PASS:
5382 			break;
5383 		default:
5384 			BUG();
5385 		}
5386 	}
5387 
5388 	if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5389 check_vlan_id:
5390 		if (skb_vlan_tag_get_id(skb)) {
5391 			/* Vlan id is non 0 and vlan_do_receive() above couldn't
5392 			 * find vlan device.
5393 			 */
5394 			skb->pkt_type = PACKET_OTHERHOST;
5395 		} else if (eth_type_vlan(skb->protocol)) {
5396 			/* Outer header is 802.1P with vlan 0, inner header is
5397 			 * 802.1Q or 802.1AD and vlan_do_receive() above could
5398 			 * not find vlan dev for vlan id 0.
5399 			 */
5400 			__vlan_hwaccel_clear_tag(skb);
5401 			skb = skb_vlan_untag(skb);
5402 			if (unlikely(!skb))
5403 				goto out;
5404 			if (vlan_do_receive(&skb))
5405 				/* After stripping off 802.1P header with vlan 0
5406 				 * vlan dev is found for inner header.
5407 				 */
5408 				goto another_round;
5409 			else if (unlikely(!skb))
5410 				goto out;
5411 			else
5412 				/* We have stripped outer 802.1P vlan 0 header.
5413 				 * But could not find vlan dev.
5414 				 * check again for vlan id to set OTHERHOST.
5415 				 */
5416 				goto check_vlan_id;
5417 		}
5418 		/* Note: we might in the future use prio bits
5419 		 * and set skb->priority like in vlan_do_receive()
5420 		 * For the time being, just ignore Priority Code Point
5421 		 */
5422 		__vlan_hwaccel_clear_tag(skb);
5423 	}
5424 
5425 	type = skb->protocol;
5426 
5427 	/* deliver only exact match when indicated */
5428 	if (likely(!deliver_exact)) {
5429 		deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5430 				       &ptype_base[ntohs(type) &
5431 						   PTYPE_HASH_MASK]);
5432 	}
5433 
5434 	deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5435 			       &orig_dev->ptype_specific);
5436 
5437 	if (unlikely(skb->dev != orig_dev)) {
5438 		deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5439 				       &skb->dev->ptype_specific);
5440 	}
5441 
5442 	if (pt_prev) {
5443 		if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5444 			goto drop;
5445 		*ppt_prev = pt_prev;
5446 	} else {
5447 drop:
5448 		if (!deliver_exact)
5449 			dev_core_stats_rx_dropped_inc(skb->dev);
5450 		else
5451 			dev_core_stats_rx_nohandler_inc(skb->dev);
5452 		kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO);
5453 		/* Jamal, now you will not able to escape explaining
5454 		 * me how you were going to use this. :-)
5455 		 */
5456 		ret = NET_RX_DROP;
5457 	}
5458 
5459 out:
5460 	/* The invariant here is that if *ppt_prev is not NULL
5461 	 * then skb should also be non-NULL.
5462 	 *
5463 	 * Apparently *ppt_prev assignment above holds this invariant due to
5464 	 * skb dereferencing near it.
5465 	 */
5466 	*pskb = skb;
5467 	return ret;
5468 }
5469 
5470 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5471 {
5472 	struct net_device *orig_dev = skb->dev;
5473 	struct packet_type *pt_prev = NULL;
5474 	int ret;
5475 
5476 	ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5477 	if (pt_prev)
5478 		ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5479 					 skb->dev, pt_prev, orig_dev);
5480 	return ret;
5481 }
5482 
5483 /**
5484  *	netif_receive_skb_core - special purpose version of netif_receive_skb
5485  *	@skb: buffer to process
5486  *
5487  *	More direct receive version of netif_receive_skb().  It should
5488  *	only be used by callers that have a need to skip RPS and Generic XDP.
5489  *	Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5490  *
5491  *	This function may only be called from softirq context and interrupts
5492  *	should be enabled.
5493  *
5494  *	Return values (usually ignored):
5495  *	NET_RX_SUCCESS: no congestion
5496  *	NET_RX_DROP: packet was dropped
5497  */
5498 int netif_receive_skb_core(struct sk_buff *skb)
5499 {
5500 	int ret;
5501 
5502 	rcu_read_lock();
5503 	ret = __netif_receive_skb_one_core(skb, false);
5504 	rcu_read_unlock();
5505 
5506 	return ret;
5507 }
5508 EXPORT_SYMBOL(netif_receive_skb_core);
5509 
5510 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5511 						  struct packet_type *pt_prev,
5512 						  struct net_device *orig_dev)
5513 {
5514 	struct sk_buff *skb, *next;
5515 
5516 	if (!pt_prev)
5517 		return;
5518 	if (list_empty(head))
5519 		return;
5520 	if (pt_prev->list_func != NULL)
5521 		INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5522 				   ip_list_rcv, head, pt_prev, orig_dev);
5523 	else
5524 		list_for_each_entry_safe(skb, next, head, list) {
5525 			skb_list_del_init(skb);
5526 			pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5527 		}
5528 }
5529 
5530 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5531 {
5532 	/* Fast-path assumptions:
5533 	 * - There is no RX handler.
5534 	 * - Only one packet_type matches.
5535 	 * If either of these fails, we will end up doing some per-packet
5536 	 * processing in-line, then handling the 'last ptype' for the whole
5537 	 * sublist.  This can't cause out-of-order delivery to any single ptype,
5538 	 * because the 'last ptype' must be constant across the sublist, and all
5539 	 * other ptypes are handled per-packet.
5540 	 */
5541 	/* Current (common) ptype of sublist */
5542 	struct packet_type *pt_curr = NULL;
5543 	/* Current (common) orig_dev of sublist */
5544 	struct net_device *od_curr = NULL;
5545 	struct list_head sublist;
5546 	struct sk_buff *skb, *next;
5547 
5548 	INIT_LIST_HEAD(&sublist);
5549 	list_for_each_entry_safe(skb, next, head, list) {
5550 		struct net_device *orig_dev = skb->dev;
5551 		struct packet_type *pt_prev = NULL;
5552 
5553 		skb_list_del_init(skb);
5554 		__netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5555 		if (!pt_prev)
5556 			continue;
5557 		if (pt_curr != pt_prev || od_curr != orig_dev) {
5558 			/* dispatch old sublist */
5559 			__netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5560 			/* start new sublist */
5561 			INIT_LIST_HEAD(&sublist);
5562 			pt_curr = pt_prev;
5563 			od_curr = orig_dev;
5564 		}
5565 		list_add_tail(&skb->list, &sublist);
5566 	}
5567 
5568 	/* dispatch final sublist */
5569 	__netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5570 }
5571 
5572 static int __netif_receive_skb(struct sk_buff *skb)
5573 {
5574 	int ret;
5575 
5576 	if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5577 		unsigned int noreclaim_flag;
5578 
5579 		/*
5580 		 * PFMEMALLOC skbs are special, they should
5581 		 * - be delivered to SOCK_MEMALLOC sockets only
5582 		 * - stay away from userspace
5583 		 * - have bounded memory usage
5584 		 *
5585 		 * Use PF_MEMALLOC as this saves us from propagating the allocation
5586 		 * context down to all allocation sites.
5587 		 */
5588 		noreclaim_flag = memalloc_noreclaim_save();
5589 		ret = __netif_receive_skb_one_core(skb, true);
5590 		memalloc_noreclaim_restore(noreclaim_flag);
5591 	} else
5592 		ret = __netif_receive_skb_one_core(skb, false);
5593 
5594 	return ret;
5595 }
5596 
5597 static void __netif_receive_skb_list(struct list_head *head)
5598 {
5599 	unsigned long noreclaim_flag = 0;
5600 	struct sk_buff *skb, *next;
5601 	bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5602 
5603 	list_for_each_entry_safe(skb, next, head, list) {
5604 		if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5605 			struct list_head sublist;
5606 
5607 			/* Handle the previous sublist */
5608 			list_cut_before(&sublist, head, &skb->list);
5609 			if (!list_empty(&sublist))
5610 				__netif_receive_skb_list_core(&sublist, pfmemalloc);
5611 			pfmemalloc = !pfmemalloc;
5612 			/* See comments in __netif_receive_skb */
5613 			if (pfmemalloc)
5614 				noreclaim_flag = memalloc_noreclaim_save();
5615 			else
5616 				memalloc_noreclaim_restore(noreclaim_flag);
5617 		}
5618 	}
5619 	/* Handle the remaining sublist */
5620 	if (!list_empty(head))
5621 		__netif_receive_skb_list_core(head, pfmemalloc);
5622 	/* Restore pflags */
5623 	if (pfmemalloc)
5624 		memalloc_noreclaim_restore(noreclaim_flag);
5625 }
5626 
5627 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5628 {
5629 	struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5630 	struct bpf_prog *new = xdp->prog;
5631 	int ret = 0;
5632 
5633 	switch (xdp->command) {
5634 	case XDP_SETUP_PROG:
5635 		rcu_assign_pointer(dev->xdp_prog, new);
5636 		if (old)
5637 			bpf_prog_put(old);
5638 
5639 		if (old && !new) {
5640 			static_branch_dec(&generic_xdp_needed_key);
5641 		} else if (new && !old) {
5642 			static_branch_inc(&generic_xdp_needed_key);
5643 			dev_disable_lro(dev);
5644 			dev_disable_gro_hw(dev);
5645 		}
5646 		break;
5647 
5648 	default:
5649 		ret = -EINVAL;
5650 		break;
5651 	}
5652 
5653 	return ret;
5654 }
5655 
5656 static int netif_receive_skb_internal(struct sk_buff *skb)
5657 {
5658 	int ret;
5659 
5660 	net_timestamp_check(netdev_tstamp_prequeue, skb);
5661 
5662 	if (skb_defer_rx_timestamp(skb))
5663 		return NET_RX_SUCCESS;
5664 
5665 	rcu_read_lock();
5666 #ifdef CONFIG_RPS
5667 	if (static_branch_unlikely(&rps_needed)) {
5668 		struct rps_dev_flow voidflow, *rflow = &voidflow;
5669 		int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5670 
5671 		if (cpu >= 0) {
5672 			ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5673 			rcu_read_unlock();
5674 			return ret;
5675 		}
5676 	}
5677 #endif
5678 	ret = __netif_receive_skb(skb);
5679 	rcu_read_unlock();
5680 	return ret;
5681 }
5682 
5683 void netif_receive_skb_list_internal(struct list_head *head)
5684 {
5685 	struct sk_buff *skb, *next;
5686 	struct list_head sublist;
5687 
5688 	INIT_LIST_HEAD(&sublist);
5689 	list_for_each_entry_safe(skb, next, head, list) {
5690 		net_timestamp_check(netdev_tstamp_prequeue, skb);
5691 		skb_list_del_init(skb);
5692 		if (!skb_defer_rx_timestamp(skb))
5693 			list_add_tail(&skb->list, &sublist);
5694 	}
5695 	list_splice_init(&sublist, head);
5696 
5697 	rcu_read_lock();
5698 #ifdef CONFIG_RPS
5699 	if (static_branch_unlikely(&rps_needed)) {
5700 		list_for_each_entry_safe(skb, next, head, list) {
5701 			struct rps_dev_flow voidflow, *rflow = &voidflow;
5702 			int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5703 
5704 			if (cpu >= 0) {
5705 				/* Will be handled, remove from list */
5706 				skb_list_del_init(skb);
5707 				enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5708 			}
5709 		}
5710 	}
5711 #endif
5712 	__netif_receive_skb_list(head);
5713 	rcu_read_unlock();
5714 }
5715 
5716 /**
5717  *	netif_receive_skb - process receive buffer from network
5718  *	@skb: buffer to process
5719  *
5720  *	netif_receive_skb() is the main receive data processing function.
5721  *	It always succeeds. The buffer may be dropped during processing
5722  *	for congestion control or by the protocol layers.
5723  *
5724  *	This function may only be called from softirq context and interrupts
5725  *	should be enabled.
5726  *
5727  *	Return values (usually ignored):
5728  *	NET_RX_SUCCESS: no congestion
5729  *	NET_RX_DROP: packet was dropped
5730  */
5731 int netif_receive_skb(struct sk_buff *skb)
5732 {
5733 	int ret;
5734 
5735 	trace_netif_receive_skb_entry(skb);
5736 
5737 	ret = netif_receive_skb_internal(skb);
5738 	trace_netif_receive_skb_exit(ret);
5739 
5740 	return ret;
5741 }
5742 EXPORT_SYMBOL(netif_receive_skb);
5743 
5744 /**
5745  *	netif_receive_skb_list - process many receive buffers from network
5746  *	@head: list of skbs to process.
5747  *
5748  *	Since return value of netif_receive_skb() is normally ignored, and
5749  *	wouldn't be meaningful for a list, this function returns void.
5750  *
5751  *	This function may only be called from softirq context and interrupts
5752  *	should be enabled.
5753  */
5754 void netif_receive_skb_list(struct list_head *head)
5755 {
5756 	struct sk_buff *skb;
5757 
5758 	if (list_empty(head))
5759 		return;
5760 	if (trace_netif_receive_skb_list_entry_enabled()) {
5761 		list_for_each_entry(skb, head, list)
5762 			trace_netif_receive_skb_list_entry(skb);
5763 	}
5764 	netif_receive_skb_list_internal(head);
5765 	trace_netif_receive_skb_list_exit(0);
5766 }
5767 EXPORT_SYMBOL(netif_receive_skb_list);
5768 
5769 static DEFINE_PER_CPU(struct work_struct, flush_works);
5770 
5771 /* Network device is going away, flush any packets still pending */
5772 static void flush_backlog(struct work_struct *work)
5773 {
5774 	struct sk_buff *skb, *tmp;
5775 	struct softnet_data *sd;
5776 
5777 	local_bh_disable();
5778 	sd = this_cpu_ptr(&softnet_data);
5779 
5780 	rps_lock_irq_disable(sd);
5781 	skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5782 		if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5783 			__skb_unlink(skb, &sd->input_pkt_queue);
5784 			dev_kfree_skb_irq(skb);
5785 			input_queue_head_incr(sd);
5786 		}
5787 	}
5788 	rps_unlock_irq_enable(sd);
5789 
5790 	skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5791 		if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5792 			__skb_unlink(skb, &sd->process_queue);
5793 			kfree_skb(skb);
5794 			input_queue_head_incr(sd);
5795 		}
5796 	}
5797 	local_bh_enable();
5798 }
5799 
5800 static bool flush_required(int cpu)
5801 {
5802 #if IS_ENABLED(CONFIG_RPS)
5803 	struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5804 	bool do_flush;
5805 
5806 	rps_lock_irq_disable(sd);
5807 
5808 	/* as insertion into process_queue happens with the rps lock held,
5809 	 * process_queue access may race only with dequeue
5810 	 */
5811 	do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5812 		   !skb_queue_empty_lockless(&sd->process_queue);
5813 	rps_unlock_irq_enable(sd);
5814 
5815 	return do_flush;
5816 #endif
5817 	/* without RPS we can't safely check input_pkt_queue: during a
5818 	 * concurrent remote skb_queue_splice() we can detect as empty both
5819 	 * input_pkt_queue and process_queue even if the latter could end-up
5820 	 * containing a lot of packets.
5821 	 */
5822 	return true;
5823 }
5824 
5825 static void flush_all_backlogs(void)
5826 {
5827 	static cpumask_t flush_cpus;
5828 	unsigned int cpu;
5829 
5830 	/* since we are under rtnl lock protection we can use static data
5831 	 * for the cpumask and avoid allocating on stack the possibly
5832 	 * large mask
5833 	 */
5834 	ASSERT_RTNL();
5835 
5836 	cpus_read_lock();
5837 
5838 	cpumask_clear(&flush_cpus);
5839 	for_each_online_cpu(cpu) {
5840 		if (flush_required(cpu)) {
5841 			queue_work_on(cpu, system_highpri_wq,
5842 				      per_cpu_ptr(&flush_works, cpu));
5843 			cpumask_set_cpu(cpu, &flush_cpus);
5844 		}
5845 	}
5846 
5847 	/* we can have in flight packet[s] on the cpus we are not flushing,
5848 	 * synchronize_net() in unregister_netdevice_many() will take care of
5849 	 * them
5850 	 */
5851 	for_each_cpu(cpu, &flush_cpus)
5852 		flush_work(per_cpu_ptr(&flush_works, cpu));
5853 
5854 	cpus_read_unlock();
5855 }
5856 
5857 static void net_rps_send_ipi(struct softnet_data *remsd)
5858 {
5859 #ifdef CONFIG_RPS
5860 	while (remsd) {
5861 		struct softnet_data *next = remsd->rps_ipi_next;
5862 
5863 		if (cpu_online(remsd->cpu))
5864 			smp_call_function_single_async(remsd->cpu, &remsd->csd);
5865 		remsd = next;
5866 	}
5867 #endif
5868 }
5869 
5870 /*
5871  * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5872  * Note: called with local irq disabled, but exits with local irq enabled.
5873  */
5874 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5875 {
5876 #ifdef CONFIG_RPS
5877 	struct softnet_data *remsd = sd->rps_ipi_list;
5878 
5879 	if (remsd) {
5880 		sd->rps_ipi_list = NULL;
5881 
5882 		local_irq_enable();
5883 
5884 		/* Send pending IPI's to kick RPS processing on remote cpus. */
5885 		net_rps_send_ipi(remsd);
5886 	} else
5887 #endif
5888 		local_irq_enable();
5889 }
5890 
5891 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5892 {
5893 #ifdef CONFIG_RPS
5894 	return sd->rps_ipi_list != NULL;
5895 #else
5896 	return false;
5897 #endif
5898 }
5899 
5900 static int process_backlog(struct napi_struct *napi, int quota)
5901 {
5902 	struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5903 	bool again = true;
5904 	int work = 0;
5905 
5906 	/* Check if we have pending ipi, its better to send them now,
5907 	 * not waiting net_rx_action() end.
5908 	 */
5909 	if (sd_has_rps_ipi_waiting(sd)) {
5910 		local_irq_disable();
5911 		net_rps_action_and_irq_enable(sd);
5912 	}
5913 
5914 	napi->weight = dev_rx_weight;
5915 	while (again) {
5916 		struct sk_buff *skb;
5917 
5918 		while ((skb = __skb_dequeue(&sd->process_queue))) {
5919 			rcu_read_lock();
5920 			__netif_receive_skb(skb);
5921 			rcu_read_unlock();
5922 			input_queue_head_incr(sd);
5923 			if (++work >= quota)
5924 				return work;
5925 
5926 		}
5927 
5928 		rps_lock_irq_disable(sd);
5929 		if (skb_queue_empty(&sd->input_pkt_queue)) {
5930 			/*
5931 			 * Inline a custom version of __napi_complete().
5932 			 * only current cpu owns and manipulates this napi,
5933 			 * and NAPI_STATE_SCHED is the only possible flag set
5934 			 * on backlog.
5935 			 * We can use a plain write instead of clear_bit(),
5936 			 * and we dont need an smp_mb() memory barrier.
5937 			 */
5938 			napi->state = 0;
5939 			again = false;
5940 		} else {
5941 			skb_queue_splice_tail_init(&sd->input_pkt_queue,
5942 						   &sd->process_queue);
5943 		}
5944 		rps_unlock_irq_enable(sd);
5945 	}
5946 
5947 	return work;
5948 }
5949 
5950 /**
5951  * __napi_schedule - schedule for receive
5952  * @n: entry to schedule
5953  *
5954  * The entry's receive function will be scheduled to run.
5955  * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5956  */
5957 void __napi_schedule(struct napi_struct *n)
5958 {
5959 	unsigned long flags;
5960 
5961 	local_irq_save(flags);
5962 	____napi_schedule(this_cpu_ptr(&softnet_data), n);
5963 	local_irq_restore(flags);
5964 }
5965 EXPORT_SYMBOL(__napi_schedule);
5966 
5967 /**
5968  *	napi_schedule_prep - check if napi can be scheduled
5969  *	@n: napi context
5970  *
5971  * Test if NAPI routine is already running, and if not mark
5972  * it as running.  This is used as a condition variable to
5973  * insure only one NAPI poll instance runs.  We also make
5974  * sure there is no pending NAPI disable.
5975  */
5976 bool napi_schedule_prep(struct napi_struct *n)
5977 {
5978 	unsigned long val, new;
5979 
5980 	do {
5981 		val = READ_ONCE(n->state);
5982 		if (unlikely(val & NAPIF_STATE_DISABLE))
5983 			return false;
5984 		new = val | NAPIF_STATE_SCHED;
5985 
5986 		/* Sets STATE_MISSED bit if STATE_SCHED was already set
5987 		 * This was suggested by Alexander Duyck, as compiler
5988 		 * emits better code than :
5989 		 * if (val & NAPIF_STATE_SCHED)
5990 		 *     new |= NAPIF_STATE_MISSED;
5991 		 */
5992 		new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5993 						   NAPIF_STATE_MISSED;
5994 	} while (cmpxchg(&n->state, val, new) != val);
5995 
5996 	return !(val & NAPIF_STATE_SCHED);
5997 }
5998 EXPORT_SYMBOL(napi_schedule_prep);
5999 
6000 /**
6001  * __napi_schedule_irqoff - schedule for receive
6002  * @n: entry to schedule
6003  *
6004  * Variant of __napi_schedule() assuming hard irqs are masked.
6005  *
6006  * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6007  * because the interrupt disabled assumption might not be true
6008  * due to force-threaded interrupts and spinlock substitution.
6009  */
6010 void __napi_schedule_irqoff(struct napi_struct *n)
6011 {
6012 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6013 		____napi_schedule(this_cpu_ptr(&softnet_data), n);
6014 	else
6015 		__napi_schedule(n);
6016 }
6017 EXPORT_SYMBOL(__napi_schedule_irqoff);
6018 
6019 bool napi_complete_done(struct napi_struct *n, int work_done)
6020 {
6021 	unsigned long flags, val, new, timeout = 0;
6022 	bool ret = true;
6023 
6024 	/*
6025 	 * 1) Don't let napi dequeue from the cpu poll list
6026 	 *    just in case its running on a different cpu.
6027 	 * 2) If we are busy polling, do nothing here, we have
6028 	 *    the guarantee we will be called later.
6029 	 */
6030 	if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6031 				 NAPIF_STATE_IN_BUSY_POLL)))
6032 		return false;
6033 
6034 	if (work_done) {
6035 		if (n->gro_bitmask)
6036 			timeout = READ_ONCE(n->dev->gro_flush_timeout);
6037 		n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6038 	}
6039 	if (n->defer_hard_irqs_count > 0) {
6040 		n->defer_hard_irqs_count--;
6041 		timeout = READ_ONCE(n->dev->gro_flush_timeout);
6042 		if (timeout)
6043 			ret = false;
6044 	}
6045 	if (n->gro_bitmask) {
6046 		/* When the NAPI instance uses a timeout and keeps postponing
6047 		 * it, we need to bound somehow the time packets are kept in
6048 		 * the GRO layer
6049 		 */
6050 		napi_gro_flush(n, !!timeout);
6051 	}
6052 
6053 	gro_normal_list(n);
6054 
6055 	if (unlikely(!list_empty(&n->poll_list))) {
6056 		/* If n->poll_list is not empty, we need to mask irqs */
6057 		local_irq_save(flags);
6058 		list_del_init(&n->poll_list);
6059 		local_irq_restore(flags);
6060 	}
6061 
6062 	do {
6063 		val = READ_ONCE(n->state);
6064 
6065 		WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6066 
6067 		new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6068 			      NAPIF_STATE_SCHED_THREADED |
6069 			      NAPIF_STATE_PREFER_BUSY_POLL);
6070 
6071 		/* If STATE_MISSED was set, leave STATE_SCHED set,
6072 		 * because we will call napi->poll() one more time.
6073 		 * This C code was suggested by Alexander Duyck to help gcc.
6074 		 */
6075 		new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6076 						    NAPIF_STATE_SCHED;
6077 	} while (cmpxchg(&n->state, val, new) != val);
6078 
6079 	if (unlikely(val & NAPIF_STATE_MISSED)) {
6080 		__napi_schedule(n);
6081 		return false;
6082 	}
6083 
6084 	if (timeout)
6085 		hrtimer_start(&n->timer, ns_to_ktime(timeout),
6086 			      HRTIMER_MODE_REL_PINNED);
6087 	return ret;
6088 }
6089 EXPORT_SYMBOL(napi_complete_done);
6090 
6091 /* must be called under rcu_read_lock(), as we dont take a reference */
6092 static struct napi_struct *napi_by_id(unsigned int napi_id)
6093 {
6094 	unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6095 	struct napi_struct *napi;
6096 
6097 	hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6098 		if (napi->napi_id == napi_id)
6099 			return napi;
6100 
6101 	return NULL;
6102 }
6103 
6104 #if defined(CONFIG_NET_RX_BUSY_POLL)
6105 
6106 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6107 {
6108 	if (!skip_schedule) {
6109 		gro_normal_list(napi);
6110 		__napi_schedule(napi);
6111 		return;
6112 	}
6113 
6114 	if (napi->gro_bitmask) {
6115 		/* flush too old packets
6116 		 * If HZ < 1000, flush all packets.
6117 		 */
6118 		napi_gro_flush(napi, HZ >= 1000);
6119 	}
6120 
6121 	gro_normal_list(napi);
6122 	clear_bit(NAPI_STATE_SCHED, &napi->state);
6123 }
6124 
6125 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6126 			   u16 budget)
6127 {
6128 	bool skip_schedule = false;
6129 	unsigned long timeout;
6130 	int rc;
6131 
6132 	/* Busy polling means there is a high chance device driver hard irq
6133 	 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6134 	 * set in napi_schedule_prep().
6135 	 * Since we are about to call napi->poll() once more, we can safely
6136 	 * clear NAPI_STATE_MISSED.
6137 	 *
6138 	 * Note: x86 could use a single "lock and ..." instruction
6139 	 * to perform these two clear_bit()
6140 	 */
6141 	clear_bit(NAPI_STATE_MISSED, &napi->state);
6142 	clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6143 
6144 	local_bh_disable();
6145 
6146 	if (prefer_busy_poll) {
6147 		napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6148 		timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6149 		if (napi->defer_hard_irqs_count && timeout) {
6150 			hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6151 			skip_schedule = true;
6152 		}
6153 	}
6154 
6155 	/* All we really want here is to re-enable device interrupts.
6156 	 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6157 	 */
6158 	rc = napi->poll(napi, budget);
6159 	/* We can't gro_normal_list() here, because napi->poll() might have
6160 	 * rearmed the napi (napi_complete_done()) in which case it could
6161 	 * already be running on another CPU.
6162 	 */
6163 	trace_napi_poll(napi, rc, budget);
6164 	netpoll_poll_unlock(have_poll_lock);
6165 	if (rc == budget)
6166 		__busy_poll_stop(napi, skip_schedule);
6167 	local_bh_enable();
6168 }
6169 
6170 void napi_busy_loop(unsigned int napi_id,
6171 		    bool (*loop_end)(void *, unsigned long),
6172 		    void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6173 {
6174 	unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6175 	int (*napi_poll)(struct napi_struct *napi, int budget);
6176 	void *have_poll_lock = NULL;
6177 	struct napi_struct *napi;
6178 
6179 restart:
6180 	napi_poll = NULL;
6181 
6182 	rcu_read_lock();
6183 
6184 	napi = napi_by_id(napi_id);
6185 	if (!napi)
6186 		goto out;
6187 
6188 	preempt_disable();
6189 	for (;;) {
6190 		int work = 0;
6191 
6192 		local_bh_disable();
6193 		if (!napi_poll) {
6194 			unsigned long val = READ_ONCE(napi->state);
6195 
6196 			/* If multiple threads are competing for this napi,
6197 			 * we avoid dirtying napi->state as much as we can.
6198 			 */
6199 			if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6200 				   NAPIF_STATE_IN_BUSY_POLL)) {
6201 				if (prefer_busy_poll)
6202 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6203 				goto count;
6204 			}
6205 			if (cmpxchg(&napi->state, val,
6206 				    val | NAPIF_STATE_IN_BUSY_POLL |
6207 					  NAPIF_STATE_SCHED) != val) {
6208 				if (prefer_busy_poll)
6209 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6210 				goto count;
6211 			}
6212 			have_poll_lock = netpoll_poll_lock(napi);
6213 			napi_poll = napi->poll;
6214 		}
6215 		work = napi_poll(napi, budget);
6216 		trace_napi_poll(napi, work, budget);
6217 		gro_normal_list(napi);
6218 count:
6219 		if (work > 0)
6220 			__NET_ADD_STATS(dev_net(napi->dev),
6221 					LINUX_MIB_BUSYPOLLRXPACKETS, work);
6222 		local_bh_enable();
6223 
6224 		if (!loop_end || loop_end(loop_end_arg, start_time))
6225 			break;
6226 
6227 		if (unlikely(need_resched())) {
6228 			if (napi_poll)
6229 				busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6230 			preempt_enable();
6231 			rcu_read_unlock();
6232 			cond_resched();
6233 			if (loop_end(loop_end_arg, start_time))
6234 				return;
6235 			goto restart;
6236 		}
6237 		cpu_relax();
6238 	}
6239 	if (napi_poll)
6240 		busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6241 	preempt_enable();
6242 out:
6243 	rcu_read_unlock();
6244 }
6245 EXPORT_SYMBOL(napi_busy_loop);
6246 
6247 #endif /* CONFIG_NET_RX_BUSY_POLL */
6248 
6249 static void napi_hash_add(struct napi_struct *napi)
6250 {
6251 	if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6252 		return;
6253 
6254 	spin_lock(&napi_hash_lock);
6255 
6256 	/* 0..NR_CPUS range is reserved for sender_cpu use */
6257 	do {
6258 		if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6259 			napi_gen_id = MIN_NAPI_ID;
6260 	} while (napi_by_id(napi_gen_id));
6261 	napi->napi_id = napi_gen_id;
6262 
6263 	hlist_add_head_rcu(&napi->napi_hash_node,
6264 			   &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6265 
6266 	spin_unlock(&napi_hash_lock);
6267 }
6268 
6269 /* Warning : caller is responsible to make sure rcu grace period
6270  * is respected before freeing memory containing @napi
6271  */
6272 static void napi_hash_del(struct napi_struct *napi)
6273 {
6274 	spin_lock(&napi_hash_lock);
6275 
6276 	hlist_del_init_rcu(&napi->napi_hash_node);
6277 
6278 	spin_unlock(&napi_hash_lock);
6279 }
6280 
6281 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6282 {
6283 	struct napi_struct *napi;
6284 
6285 	napi = container_of(timer, struct napi_struct, timer);
6286 
6287 	/* Note : we use a relaxed variant of napi_schedule_prep() not setting
6288 	 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6289 	 */
6290 	if (!napi_disable_pending(napi) &&
6291 	    !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6292 		clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6293 		__napi_schedule_irqoff(napi);
6294 	}
6295 
6296 	return HRTIMER_NORESTART;
6297 }
6298 
6299 static void init_gro_hash(struct napi_struct *napi)
6300 {
6301 	int i;
6302 
6303 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6304 		INIT_LIST_HEAD(&napi->gro_hash[i].list);
6305 		napi->gro_hash[i].count = 0;
6306 	}
6307 	napi->gro_bitmask = 0;
6308 }
6309 
6310 int dev_set_threaded(struct net_device *dev, bool threaded)
6311 {
6312 	struct napi_struct *napi;
6313 	int err = 0;
6314 
6315 	if (dev->threaded == threaded)
6316 		return 0;
6317 
6318 	if (threaded) {
6319 		list_for_each_entry(napi, &dev->napi_list, dev_list) {
6320 			if (!napi->thread) {
6321 				err = napi_kthread_create(napi);
6322 				if (err) {
6323 					threaded = false;
6324 					break;
6325 				}
6326 			}
6327 		}
6328 	}
6329 
6330 	dev->threaded = threaded;
6331 
6332 	/* Make sure kthread is created before THREADED bit
6333 	 * is set.
6334 	 */
6335 	smp_mb__before_atomic();
6336 
6337 	/* Setting/unsetting threaded mode on a napi might not immediately
6338 	 * take effect, if the current napi instance is actively being
6339 	 * polled. In this case, the switch between threaded mode and
6340 	 * softirq mode will happen in the next round of napi_schedule().
6341 	 * This should not cause hiccups/stalls to the live traffic.
6342 	 */
6343 	list_for_each_entry(napi, &dev->napi_list, dev_list) {
6344 		if (threaded)
6345 			set_bit(NAPI_STATE_THREADED, &napi->state);
6346 		else
6347 			clear_bit(NAPI_STATE_THREADED, &napi->state);
6348 	}
6349 
6350 	return err;
6351 }
6352 EXPORT_SYMBOL(dev_set_threaded);
6353 
6354 /* Double check that napi_get_frags() allocates skbs with
6355  * skb->head being backed by slab, not a page fragment.
6356  * This is to make sure bug fixed in 3226b158e67c
6357  * ("net: avoid 32 x truesize under-estimation for tiny skbs")
6358  * does not accidentally come back.
6359  */
6360 static void napi_get_frags_check(struct napi_struct *napi)
6361 {
6362 	struct sk_buff *skb;
6363 
6364 	local_bh_disable();
6365 	skb = napi_get_frags(napi);
6366 	WARN_ON_ONCE(skb && skb->head_frag);
6367 	napi_free_frags(napi);
6368 	local_bh_enable();
6369 }
6370 
6371 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi,
6372 			   int (*poll)(struct napi_struct *, int), int weight)
6373 {
6374 	if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6375 		return;
6376 
6377 	INIT_LIST_HEAD(&napi->poll_list);
6378 	INIT_HLIST_NODE(&napi->napi_hash_node);
6379 	hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6380 	napi->timer.function = napi_watchdog;
6381 	init_gro_hash(napi);
6382 	napi->skb = NULL;
6383 	INIT_LIST_HEAD(&napi->rx_list);
6384 	napi->rx_count = 0;
6385 	napi->poll = poll;
6386 	if (weight > NAPI_POLL_WEIGHT)
6387 		netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6388 				weight);
6389 	napi->weight = weight;
6390 	napi->dev = dev;
6391 #ifdef CONFIG_NETPOLL
6392 	napi->poll_owner = -1;
6393 #endif
6394 	set_bit(NAPI_STATE_SCHED, &napi->state);
6395 	set_bit(NAPI_STATE_NPSVC, &napi->state);
6396 	list_add_rcu(&napi->dev_list, &dev->napi_list);
6397 	napi_hash_add(napi);
6398 	napi_get_frags_check(napi);
6399 	/* Create kthread for this napi if dev->threaded is set.
6400 	 * Clear dev->threaded if kthread creation failed so that
6401 	 * threaded mode will not be enabled in napi_enable().
6402 	 */
6403 	if (dev->threaded && napi_kthread_create(napi))
6404 		dev->threaded = 0;
6405 }
6406 EXPORT_SYMBOL(netif_napi_add_weight);
6407 
6408 void napi_disable(struct napi_struct *n)
6409 {
6410 	unsigned long val, new;
6411 
6412 	might_sleep();
6413 	set_bit(NAPI_STATE_DISABLE, &n->state);
6414 
6415 	for ( ; ; ) {
6416 		val = READ_ONCE(n->state);
6417 		if (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6418 			usleep_range(20, 200);
6419 			continue;
6420 		}
6421 
6422 		new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6423 		new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6424 
6425 		if (cmpxchg(&n->state, val, new) == val)
6426 			break;
6427 	}
6428 
6429 	hrtimer_cancel(&n->timer);
6430 
6431 	clear_bit(NAPI_STATE_DISABLE, &n->state);
6432 }
6433 EXPORT_SYMBOL(napi_disable);
6434 
6435 /**
6436  *	napi_enable - enable NAPI scheduling
6437  *	@n: NAPI context
6438  *
6439  * Resume NAPI from being scheduled on this context.
6440  * Must be paired with napi_disable.
6441  */
6442 void napi_enable(struct napi_struct *n)
6443 {
6444 	unsigned long val, new;
6445 
6446 	do {
6447 		val = READ_ONCE(n->state);
6448 		BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6449 
6450 		new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6451 		if (n->dev->threaded && n->thread)
6452 			new |= NAPIF_STATE_THREADED;
6453 	} while (cmpxchg(&n->state, val, new) != val);
6454 }
6455 EXPORT_SYMBOL(napi_enable);
6456 
6457 static void flush_gro_hash(struct napi_struct *napi)
6458 {
6459 	int i;
6460 
6461 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6462 		struct sk_buff *skb, *n;
6463 
6464 		list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6465 			kfree_skb(skb);
6466 		napi->gro_hash[i].count = 0;
6467 	}
6468 }
6469 
6470 /* Must be called in process context */
6471 void __netif_napi_del(struct napi_struct *napi)
6472 {
6473 	if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6474 		return;
6475 
6476 	napi_hash_del(napi);
6477 	list_del_rcu(&napi->dev_list);
6478 	napi_free_frags(napi);
6479 
6480 	flush_gro_hash(napi);
6481 	napi->gro_bitmask = 0;
6482 
6483 	if (napi->thread) {
6484 		kthread_stop(napi->thread);
6485 		napi->thread = NULL;
6486 	}
6487 }
6488 EXPORT_SYMBOL(__netif_napi_del);
6489 
6490 static int __napi_poll(struct napi_struct *n, bool *repoll)
6491 {
6492 	int work, weight;
6493 
6494 	weight = n->weight;
6495 
6496 	/* This NAPI_STATE_SCHED test is for avoiding a race
6497 	 * with netpoll's poll_napi().  Only the entity which
6498 	 * obtains the lock and sees NAPI_STATE_SCHED set will
6499 	 * actually make the ->poll() call.  Therefore we avoid
6500 	 * accidentally calling ->poll() when NAPI is not scheduled.
6501 	 */
6502 	work = 0;
6503 	if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6504 		work = n->poll(n, weight);
6505 		trace_napi_poll(n, work, weight);
6506 	}
6507 
6508 	if (unlikely(work > weight))
6509 		netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6510 				n->poll, work, weight);
6511 
6512 	if (likely(work < weight))
6513 		return work;
6514 
6515 	/* Drivers must not modify the NAPI state if they
6516 	 * consume the entire weight.  In such cases this code
6517 	 * still "owns" the NAPI instance and therefore can
6518 	 * move the instance around on the list at-will.
6519 	 */
6520 	if (unlikely(napi_disable_pending(n))) {
6521 		napi_complete(n);
6522 		return work;
6523 	}
6524 
6525 	/* The NAPI context has more processing work, but busy-polling
6526 	 * is preferred. Exit early.
6527 	 */
6528 	if (napi_prefer_busy_poll(n)) {
6529 		if (napi_complete_done(n, work)) {
6530 			/* If timeout is not set, we need to make sure
6531 			 * that the NAPI is re-scheduled.
6532 			 */
6533 			napi_schedule(n);
6534 		}
6535 		return work;
6536 	}
6537 
6538 	if (n->gro_bitmask) {
6539 		/* flush too old packets
6540 		 * If HZ < 1000, flush all packets.
6541 		 */
6542 		napi_gro_flush(n, HZ >= 1000);
6543 	}
6544 
6545 	gro_normal_list(n);
6546 
6547 	/* Some drivers may have called napi_schedule
6548 	 * prior to exhausting their budget.
6549 	 */
6550 	if (unlikely(!list_empty(&n->poll_list))) {
6551 		pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6552 			     n->dev ? n->dev->name : "backlog");
6553 		return work;
6554 	}
6555 
6556 	*repoll = true;
6557 
6558 	return work;
6559 }
6560 
6561 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6562 {
6563 	bool do_repoll = false;
6564 	void *have;
6565 	int work;
6566 
6567 	list_del_init(&n->poll_list);
6568 
6569 	have = netpoll_poll_lock(n);
6570 
6571 	work = __napi_poll(n, &do_repoll);
6572 
6573 	if (do_repoll)
6574 		list_add_tail(&n->poll_list, repoll);
6575 
6576 	netpoll_poll_unlock(have);
6577 
6578 	return work;
6579 }
6580 
6581 static int napi_thread_wait(struct napi_struct *napi)
6582 {
6583 	bool woken = false;
6584 
6585 	set_current_state(TASK_INTERRUPTIBLE);
6586 
6587 	while (!kthread_should_stop()) {
6588 		/* Testing SCHED_THREADED bit here to make sure the current
6589 		 * kthread owns this napi and could poll on this napi.
6590 		 * Testing SCHED bit is not enough because SCHED bit might be
6591 		 * set by some other busy poll thread or by napi_disable().
6592 		 */
6593 		if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
6594 			WARN_ON(!list_empty(&napi->poll_list));
6595 			__set_current_state(TASK_RUNNING);
6596 			return 0;
6597 		}
6598 
6599 		schedule();
6600 		/* woken being true indicates this thread owns this napi. */
6601 		woken = true;
6602 		set_current_state(TASK_INTERRUPTIBLE);
6603 	}
6604 	__set_current_state(TASK_RUNNING);
6605 
6606 	return -1;
6607 }
6608 
6609 static int napi_threaded_poll(void *data)
6610 {
6611 	struct napi_struct *napi = data;
6612 	void *have;
6613 
6614 	while (!napi_thread_wait(napi)) {
6615 		for (;;) {
6616 			bool repoll = false;
6617 
6618 			local_bh_disable();
6619 
6620 			have = netpoll_poll_lock(napi);
6621 			__napi_poll(napi, &repoll);
6622 			netpoll_poll_unlock(have);
6623 
6624 			local_bh_enable();
6625 
6626 			if (!repoll)
6627 				break;
6628 
6629 			cond_resched();
6630 		}
6631 	}
6632 	return 0;
6633 }
6634 
6635 static void skb_defer_free_flush(struct softnet_data *sd)
6636 {
6637 	struct sk_buff *skb, *next;
6638 	unsigned long flags;
6639 
6640 	/* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
6641 	if (!READ_ONCE(sd->defer_list))
6642 		return;
6643 
6644 	spin_lock_irqsave(&sd->defer_lock, flags);
6645 	skb = sd->defer_list;
6646 	sd->defer_list = NULL;
6647 	sd->defer_count = 0;
6648 	spin_unlock_irqrestore(&sd->defer_lock, flags);
6649 
6650 	while (skb != NULL) {
6651 		next = skb->next;
6652 		napi_consume_skb(skb, 1);
6653 		skb = next;
6654 	}
6655 }
6656 
6657 static __latent_entropy void net_rx_action(struct softirq_action *h)
6658 {
6659 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6660 	unsigned long time_limit = jiffies +
6661 		usecs_to_jiffies(netdev_budget_usecs);
6662 	int budget = netdev_budget;
6663 	LIST_HEAD(list);
6664 	LIST_HEAD(repoll);
6665 
6666 	local_irq_disable();
6667 	list_splice_init(&sd->poll_list, &list);
6668 	local_irq_enable();
6669 
6670 	for (;;) {
6671 		struct napi_struct *n;
6672 
6673 		skb_defer_free_flush(sd);
6674 
6675 		if (list_empty(&list)) {
6676 			if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6677 				goto end;
6678 			break;
6679 		}
6680 
6681 		n = list_first_entry(&list, struct napi_struct, poll_list);
6682 		budget -= napi_poll(n, &repoll);
6683 
6684 		/* If softirq window is exhausted then punt.
6685 		 * Allow this to run for 2 jiffies since which will allow
6686 		 * an average latency of 1.5/HZ.
6687 		 */
6688 		if (unlikely(budget <= 0 ||
6689 			     time_after_eq(jiffies, time_limit))) {
6690 			sd->time_squeeze++;
6691 			break;
6692 		}
6693 	}
6694 
6695 	local_irq_disable();
6696 
6697 	list_splice_tail_init(&sd->poll_list, &list);
6698 	list_splice_tail(&repoll, &list);
6699 	list_splice(&list, &sd->poll_list);
6700 	if (!list_empty(&sd->poll_list))
6701 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
6702 
6703 	net_rps_action_and_irq_enable(sd);
6704 end:;
6705 }
6706 
6707 struct netdev_adjacent {
6708 	struct net_device *dev;
6709 	netdevice_tracker dev_tracker;
6710 
6711 	/* upper master flag, there can only be one master device per list */
6712 	bool master;
6713 
6714 	/* lookup ignore flag */
6715 	bool ignore;
6716 
6717 	/* counter for the number of times this device was added to us */
6718 	u16 ref_nr;
6719 
6720 	/* private field for the users */
6721 	void *private;
6722 
6723 	struct list_head list;
6724 	struct rcu_head rcu;
6725 };
6726 
6727 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6728 						 struct list_head *adj_list)
6729 {
6730 	struct netdev_adjacent *adj;
6731 
6732 	list_for_each_entry(adj, adj_list, list) {
6733 		if (adj->dev == adj_dev)
6734 			return adj;
6735 	}
6736 	return NULL;
6737 }
6738 
6739 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
6740 				    struct netdev_nested_priv *priv)
6741 {
6742 	struct net_device *dev = (struct net_device *)priv->data;
6743 
6744 	return upper_dev == dev;
6745 }
6746 
6747 /**
6748  * netdev_has_upper_dev - Check if device is linked to an upper device
6749  * @dev: device
6750  * @upper_dev: upper device to check
6751  *
6752  * Find out if a device is linked to specified upper device and return true
6753  * in case it is. Note that this checks only immediate upper device,
6754  * not through a complete stack of devices. The caller must hold the RTNL lock.
6755  */
6756 bool netdev_has_upper_dev(struct net_device *dev,
6757 			  struct net_device *upper_dev)
6758 {
6759 	struct netdev_nested_priv priv = {
6760 		.data = (void *)upper_dev,
6761 	};
6762 
6763 	ASSERT_RTNL();
6764 
6765 	return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6766 					     &priv);
6767 }
6768 EXPORT_SYMBOL(netdev_has_upper_dev);
6769 
6770 /**
6771  * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
6772  * @dev: device
6773  * @upper_dev: upper device to check
6774  *
6775  * Find out if a device is linked to specified upper device and return true
6776  * in case it is. Note that this checks the entire upper device chain.
6777  * The caller must hold rcu lock.
6778  */
6779 
6780 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6781 				  struct net_device *upper_dev)
6782 {
6783 	struct netdev_nested_priv priv = {
6784 		.data = (void *)upper_dev,
6785 	};
6786 
6787 	return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6788 					       &priv);
6789 }
6790 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6791 
6792 /**
6793  * netdev_has_any_upper_dev - Check if device is linked to some device
6794  * @dev: device
6795  *
6796  * Find out if a device is linked to an upper device and return true in case
6797  * it is. The caller must hold the RTNL lock.
6798  */
6799 bool netdev_has_any_upper_dev(struct net_device *dev)
6800 {
6801 	ASSERT_RTNL();
6802 
6803 	return !list_empty(&dev->adj_list.upper);
6804 }
6805 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6806 
6807 /**
6808  * netdev_master_upper_dev_get - Get master upper device
6809  * @dev: device
6810  *
6811  * Find a master upper device and return pointer to it or NULL in case
6812  * it's not there. The caller must hold the RTNL lock.
6813  */
6814 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6815 {
6816 	struct netdev_adjacent *upper;
6817 
6818 	ASSERT_RTNL();
6819 
6820 	if (list_empty(&dev->adj_list.upper))
6821 		return NULL;
6822 
6823 	upper = list_first_entry(&dev->adj_list.upper,
6824 				 struct netdev_adjacent, list);
6825 	if (likely(upper->master))
6826 		return upper->dev;
6827 	return NULL;
6828 }
6829 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6830 
6831 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6832 {
6833 	struct netdev_adjacent *upper;
6834 
6835 	ASSERT_RTNL();
6836 
6837 	if (list_empty(&dev->adj_list.upper))
6838 		return NULL;
6839 
6840 	upper = list_first_entry(&dev->adj_list.upper,
6841 				 struct netdev_adjacent, list);
6842 	if (likely(upper->master) && !upper->ignore)
6843 		return upper->dev;
6844 	return NULL;
6845 }
6846 
6847 /**
6848  * netdev_has_any_lower_dev - Check if device is linked to some device
6849  * @dev: device
6850  *
6851  * Find out if a device is linked to a lower device and return true in case
6852  * it is. The caller must hold the RTNL lock.
6853  */
6854 static bool netdev_has_any_lower_dev(struct net_device *dev)
6855 {
6856 	ASSERT_RTNL();
6857 
6858 	return !list_empty(&dev->adj_list.lower);
6859 }
6860 
6861 void *netdev_adjacent_get_private(struct list_head *adj_list)
6862 {
6863 	struct netdev_adjacent *adj;
6864 
6865 	adj = list_entry(adj_list, struct netdev_adjacent, list);
6866 
6867 	return adj->private;
6868 }
6869 EXPORT_SYMBOL(netdev_adjacent_get_private);
6870 
6871 /**
6872  * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6873  * @dev: device
6874  * @iter: list_head ** of the current position
6875  *
6876  * Gets the next device from the dev's upper list, starting from iter
6877  * position. The caller must hold RCU read lock.
6878  */
6879 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6880 						 struct list_head **iter)
6881 {
6882 	struct netdev_adjacent *upper;
6883 
6884 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6885 
6886 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6887 
6888 	if (&upper->list == &dev->adj_list.upper)
6889 		return NULL;
6890 
6891 	*iter = &upper->list;
6892 
6893 	return upper->dev;
6894 }
6895 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6896 
6897 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6898 						  struct list_head **iter,
6899 						  bool *ignore)
6900 {
6901 	struct netdev_adjacent *upper;
6902 
6903 	upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6904 
6905 	if (&upper->list == &dev->adj_list.upper)
6906 		return NULL;
6907 
6908 	*iter = &upper->list;
6909 	*ignore = upper->ignore;
6910 
6911 	return upper->dev;
6912 }
6913 
6914 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6915 						    struct list_head **iter)
6916 {
6917 	struct netdev_adjacent *upper;
6918 
6919 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6920 
6921 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6922 
6923 	if (&upper->list == &dev->adj_list.upper)
6924 		return NULL;
6925 
6926 	*iter = &upper->list;
6927 
6928 	return upper->dev;
6929 }
6930 
6931 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6932 				       int (*fn)(struct net_device *dev,
6933 					 struct netdev_nested_priv *priv),
6934 				       struct netdev_nested_priv *priv)
6935 {
6936 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6937 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6938 	int ret, cur = 0;
6939 	bool ignore;
6940 
6941 	now = dev;
6942 	iter = &dev->adj_list.upper;
6943 
6944 	while (1) {
6945 		if (now != dev) {
6946 			ret = fn(now, priv);
6947 			if (ret)
6948 				return ret;
6949 		}
6950 
6951 		next = NULL;
6952 		while (1) {
6953 			udev = __netdev_next_upper_dev(now, &iter, &ignore);
6954 			if (!udev)
6955 				break;
6956 			if (ignore)
6957 				continue;
6958 
6959 			next = udev;
6960 			niter = &udev->adj_list.upper;
6961 			dev_stack[cur] = now;
6962 			iter_stack[cur++] = iter;
6963 			break;
6964 		}
6965 
6966 		if (!next) {
6967 			if (!cur)
6968 				return 0;
6969 			next = dev_stack[--cur];
6970 			niter = iter_stack[cur];
6971 		}
6972 
6973 		now = next;
6974 		iter = niter;
6975 	}
6976 
6977 	return 0;
6978 }
6979 
6980 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6981 				  int (*fn)(struct net_device *dev,
6982 					    struct netdev_nested_priv *priv),
6983 				  struct netdev_nested_priv *priv)
6984 {
6985 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6986 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6987 	int ret, cur = 0;
6988 
6989 	now = dev;
6990 	iter = &dev->adj_list.upper;
6991 
6992 	while (1) {
6993 		if (now != dev) {
6994 			ret = fn(now, priv);
6995 			if (ret)
6996 				return ret;
6997 		}
6998 
6999 		next = NULL;
7000 		while (1) {
7001 			udev = netdev_next_upper_dev_rcu(now, &iter);
7002 			if (!udev)
7003 				break;
7004 
7005 			next = udev;
7006 			niter = &udev->adj_list.upper;
7007 			dev_stack[cur] = now;
7008 			iter_stack[cur++] = iter;
7009 			break;
7010 		}
7011 
7012 		if (!next) {
7013 			if (!cur)
7014 				return 0;
7015 			next = dev_stack[--cur];
7016 			niter = iter_stack[cur];
7017 		}
7018 
7019 		now = next;
7020 		iter = niter;
7021 	}
7022 
7023 	return 0;
7024 }
7025 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7026 
7027 static bool __netdev_has_upper_dev(struct net_device *dev,
7028 				   struct net_device *upper_dev)
7029 {
7030 	struct netdev_nested_priv priv = {
7031 		.flags = 0,
7032 		.data = (void *)upper_dev,
7033 	};
7034 
7035 	ASSERT_RTNL();
7036 
7037 	return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7038 					   &priv);
7039 }
7040 
7041 /**
7042  * netdev_lower_get_next_private - Get the next ->private from the
7043  *				   lower neighbour list
7044  * @dev: device
7045  * @iter: list_head ** of the current position
7046  *
7047  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7048  * list, starting from iter position. The caller must hold either hold the
7049  * RTNL lock or its own locking that guarantees that the neighbour lower
7050  * list will remain unchanged.
7051  */
7052 void *netdev_lower_get_next_private(struct net_device *dev,
7053 				    struct list_head **iter)
7054 {
7055 	struct netdev_adjacent *lower;
7056 
7057 	lower = list_entry(*iter, struct netdev_adjacent, list);
7058 
7059 	if (&lower->list == &dev->adj_list.lower)
7060 		return NULL;
7061 
7062 	*iter = lower->list.next;
7063 
7064 	return lower->private;
7065 }
7066 EXPORT_SYMBOL(netdev_lower_get_next_private);
7067 
7068 /**
7069  * netdev_lower_get_next_private_rcu - Get the next ->private from the
7070  *				       lower neighbour list, RCU
7071  *				       variant
7072  * @dev: device
7073  * @iter: list_head ** of the current position
7074  *
7075  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7076  * list, starting from iter position. The caller must hold RCU read lock.
7077  */
7078 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7079 					struct list_head **iter)
7080 {
7081 	struct netdev_adjacent *lower;
7082 
7083 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7084 
7085 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7086 
7087 	if (&lower->list == &dev->adj_list.lower)
7088 		return NULL;
7089 
7090 	*iter = &lower->list;
7091 
7092 	return lower->private;
7093 }
7094 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7095 
7096 /**
7097  * netdev_lower_get_next - Get the next device from the lower neighbour
7098  *                         list
7099  * @dev: device
7100  * @iter: list_head ** of the current position
7101  *
7102  * Gets the next netdev_adjacent from the dev's lower neighbour
7103  * list, starting from iter position. The caller must hold RTNL lock or
7104  * its own locking that guarantees that the neighbour lower
7105  * list will remain unchanged.
7106  */
7107 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7108 {
7109 	struct netdev_adjacent *lower;
7110 
7111 	lower = list_entry(*iter, struct netdev_adjacent, list);
7112 
7113 	if (&lower->list == &dev->adj_list.lower)
7114 		return NULL;
7115 
7116 	*iter = lower->list.next;
7117 
7118 	return lower->dev;
7119 }
7120 EXPORT_SYMBOL(netdev_lower_get_next);
7121 
7122 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7123 						struct list_head **iter)
7124 {
7125 	struct netdev_adjacent *lower;
7126 
7127 	lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7128 
7129 	if (&lower->list == &dev->adj_list.lower)
7130 		return NULL;
7131 
7132 	*iter = &lower->list;
7133 
7134 	return lower->dev;
7135 }
7136 
7137 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7138 						  struct list_head **iter,
7139 						  bool *ignore)
7140 {
7141 	struct netdev_adjacent *lower;
7142 
7143 	lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7144 
7145 	if (&lower->list == &dev->adj_list.lower)
7146 		return NULL;
7147 
7148 	*iter = &lower->list;
7149 	*ignore = lower->ignore;
7150 
7151 	return lower->dev;
7152 }
7153 
7154 int netdev_walk_all_lower_dev(struct net_device *dev,
7155 			      int (*fn)(struct net_device *dev,
7156 					struct netdev_nested_priv *priv),
7157 			      struct netdev_nested_priv *priv)
7158 {
7159 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7160 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7161 	int ret, cur = 0;
7162 
7163 	now = dev;
7164 	iter = &dev->adj_list.lower;
7165 
7166 	while (1) {
7167 		if (now != dev) {
7168 			ret = fn(now, priv);
7169 			if (ret)
7170 				return ret;
7171 		}
7172 
7173 		next = NULL;
7174 		while (1) {
7175 			ldev = netdev_next_lower_dev(now, &iter);
7176 			if (!ldev)
7177 				break;
7178 
7179 			next = ldev;
7180 			niter = &ldev->adj_list.lower;
7181 			dev_stack[cur] = now;
7182 			iter_stack[cur++] = iter;
7183 			break;
7184 		}
7185 
7186 		if (!next) {
7187 			if (!cur)
7188 				return 0;
7189 			next = dev_stack[--cur];
7190 			niter = iter_stack[cur];
7191 		}
7192 
7193 		now = next;
7194 		iter = niter;
7195 	}
7196 
7197 	return 0;
7198 }
7199 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7200 
7201 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7202 				       int (*fn)(struct net_device *dev,
7203 					 struct netdev_nested_priv *priv),
7204 				       struct netdev_nested_priv *priv)
7205 {
7206 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7207 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7208 	int ret, cur = 0;
7209 	bool ignore;
7210 
7211 	now = dev;
7212 	iter = &dev->adj_list.lower;
7213 
7214 	while (1) {
7215 		if (now != dev) {
7216 			ret = fn(now, priv);
7217 			if (ret)
7218 				return ret;
7219 		}
7220 
7221 		next = NULL;
7222 		while (1) {
7223 			ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7224 			if (!ldev)
7225 				break;
7226 			if (ignore)
7227 				continue;
7228 
7229 			next = ldev;
7230 			niter = &ldev->adj_list.lower;
7231 			dev_stack[cur] = now;
7232 			iter_stack[cur++] = iter;
7233 			break;
7234 		}
7235 
7236 		if (!next) {
7237 			if (!cur)
7238 				return 0;
7239 			next = dev_stack[--cur];
7240 			niter = iter_stack[cur];
7241 		}
7242 
7243 		now = next;
7244 		iter = niter;
7245 	}
7246 
7247 	return 0;
7248 }
7249 
7250 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7251 					     struct list_head **iter)
7252 {
7253 	struct netdev_adjacent *lower;
7254 
7255 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7256 	if (&lower->list == &dev->adj_list.lower)
7257 		return NULL;
7258 
7259 	*iter = &lower->list;
7260 
7261 	return lower->dev;
7262 }
7263 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7264 
7265 static u8 __netdev_upper_depth(struct net_device *dev)
7266 {
7267 	struct net_device *udev;
7268 	struct list_head *iter;
7269 	u8 max_depth = 0;
7270 	bool ignore;
7271 
7272 	for (iter = &dev->adj_list.upper,
7273 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7274 	     udev;
7275 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7276 		if (ignore)
7277 			continue;
7278 		if (max_depth < udev->upper_level)
7279 			max_depth = udev->upper_level;
7280 	}
7281 
7282 	return max_depth;
7283 }
7284 
7285 static u8 __netdev_lower_depth(struct net_device *dev)
7286 {
7287 	struct net_device *ldev;
7288 	struct list_head *iter;
7289 	u8 max_depth = 0;
7290 	bool ignore;
7291 
7292 	for (iter = &dev->adj_list.lower,
7293 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7294 	     ldev;
7295 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7296 		if (ignore)
7297 			continue;
7298 		if (max_depth < ldev->lower_level)
7299 			max_depth = ldev->lower_level;
7300 	}
7301 
7302 	return max_depth;
7303 }
7304 
7305 static int __netdev_update_upper_level(struct net_device *dev,
7306 				       struct netdev_nested_priv *__unused)
7307 {
7308 	dev->upper_level = __netdev_upper_depth(dev) + 1;
7309 	return 0;
7310 }
7311 
7312 #ifdef CONFIG_LOCKDEP
7313 static LIST_HEAD(net_unlink_list);
7314 
7315 static void net_unlink_todo(struct net_device *dev)
7316 {
7317 	if (list_empty(&dev->unlink_list))
7318 		list_add_tail(&dev->unlink_list, &net_unlink_list);
7319 }
7320 #endif
7321 
7322 static int __netdev_update_lower_level(struct net_device *dev,
7323 				       struct netdev_nested_priv *priv)
7324 {
7325 	dev->lower_level = __netdev_lower_depth(dev) + 1;
7326 
7327 #ifdef CONFIG_LOCKDEP
7328 	if (!priv)
7329 		return 0;
7330 
7331 	if (priv->flags & NESTED_SYNC_IMM)
7332 		dev->nested_level = dev->lower_level - 1;
7333 	if (priv->flags & NESTED_SYNC_TODO)
7334 		net_unlink_todo(dev);
7335 #endif
7336 	return 0;
7337 }
7338 
7339 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7340 				  int (*fn)(struct net_device *dev,
7341 					    struct netdev_nested_priv *priv),
7342 				  struct netdev_nested_priv *priv)
7343 {
7344 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7345 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7346 	int ret, cur = 0;
7347 
7348 	now = dev;
7349 	iter = &dev->adj_list.lower;
7350 
7351 	while (1) {
7352 		if (now != dev) {
7353 			ret = fn(now, priv);
7354 			if (ret)
7355 				return ret;
7356 		}
7357 
7358 		next = NULL;
7359 		while (1) {
7360 			ldev = netdev_next_lower_dev_rcu(now, &iter);
7361 			if (!ldev)
7362 				break;
7363 
7364 			next = ldev;
7365 			niter = &ldev->adj_list.lower;
7366 			dev_stack[cur] = now;
7367 			iter_stack[cur++] = iter;
7368 			break;
7369 		}
7370 
7371 		if (!next) {
7372 			if (!cur)
7373 				return 0;
7374 			next = dev_stack[--cur];
7375 			niter = iter_stack[cur];
7376 		}
7377 
7378 		now = next;
7379 		iter = niter;
7380 	}
7381 
7382 	return 0;
7383 }
7384 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7385 
7386 /**
7387  * netdev_lower_get_first_private_rcu - Get the first ->private from the
7388  *				       lower neighbour list, RCU
7389  *				       variant
7390  * @dev: device
7391  *
7392  * Gets the first netdev_adjacent->private from the dev's lower neighbour
7393  * list. The caller must hold RCU read lock.
7394  */
7395 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7396 {
7397 	struct netdev_adjacent *lower;
7398 
7399 	lower = list_first_or_null_rcu(&dev->adj_list.lower,
7400 			struct netdev_adjacent, list);
7401 	if (lower)
7402 		return lower->private;
7403 	return NULL;
7404 }
7405 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7406 
7407 /**
7408  * netdev_master_upper_dev_get_rcu - Get master upper device
7409  * @dev: device
7410  *
7411  * Find a master upper device and return pointer to it or NULL in case
7412  * it's not there. The caller must hold the RCU read lock.
7413  */
7414 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7415 {
7416 	struct netdev_adjacent *upper;
7417 
7418 	upper = list_first_or_null_rcu(&dev->adj_list.upper,
7419 				       struct netdev_adjacent, list);
7420 	if (upper && likely(upper->master))
7421 		return upper->dev;
7422 	return NULL;
7423 }
7424 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7425 
7426 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7427 			      struct net_device *adj_dev,
7428 			      struct list_head *dev_list)
7429 {
7430 	char linkname[IFNAMSIZ+7];
7431 
7432 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7433 		"upper_%s" : "lower_%s", adj_dev->name);
7434 	return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7435 				 linkname);
7436 }
7437 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7438 			       char *name,
7439 			       struct list_head *dev_list)
7440 {
7441 	char linkname[IFNAMSIZ+7];
7442 
7443 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7444 		"upper_%s" : "lower_%s", name);
7445 	sysfs_remove_link(&(dev->dev.kobj), linkname);
7446 }
7447 
7448 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7449 						 struct net_device *adj_dev,
7450 						 struct list_head *dev_list)
7451 {
7452 	return (dev_list == &dev->adj_list.upper ||
7453 		dev_list == &dev->adj_list.lower) &&
7454 		net_eq(dev_net(dev), dev_net(adj_dev));
7455 }
7456 
7457 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7458 					struct net_device *adj_dev,
7459 					struct list_head *dev_list,
7460 					void *private, bool master)
7461 {
7462 	struct netdev_adjacent *adj;
7463 	int ret;
7464 
7465 	adj = __netdev_find_adj(adj_dev, dev_list);
7466 
7467 	if (adj) {
7468 		adj->ref_nr += 1;
7469 		pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7470 			 dev->name, adj_dev->name, adj->ref_nr);
7471 
7472 		return 0;
7473 	}
7474 
7475 	adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7476 	if (!adj)
7477 		return -ENOMEM;
7478 
7479 	adj->dev = adj_dev;
7480 	adj->master = master;
7481 	adj->ref_nr = 1;
7482 	adj->private = private;
7483 	adj->ignore = false;
7484 	netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7485 
7486 	pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7487 		 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7488 
7489 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7490 		ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7491 		if (ret)
7492 			goto free_adj;
7493 	}
7494 
7495 	/* Ensure that master link is always the first item in list. */
7496 	if (master) {
7497 		ret = sysfs_create_link(&(dev->dev.kobj),
7498 					&(adj_dev->dev.kobj), "master");
7499 		if (ret)
7500 			goto remove_symlinks;
7501 
7502 		list_add_rcu(&adj->list, dev_list);
7503 	} else {
7504 		list_add_tail_rcu(&adj->list, dev_list);
7505 	}
7506 
7507 	return 0;
7508 
7509 remove_symlinks:
7510 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7511 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7512 free_adj:
7513 	netdev_put(adj_dev, &adj->dev_tracker);
7514 	kfree(adj);
7515 
7516 	return ret;
7517 }
7518 
7519 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7520 					 struct net_device *adj_dev,
7521 					 u16 ref_nr,
7522 					 struct list_head *dev_list)
7523 {
7524 	struct netdev_adjacent *adj;
7525 
7526 	pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7527 		 dev->name, adj_dev->name, ref_nr);
7528 
7529 	adj = __netdev_find_adj(adj_dev, dev_list);
7530 
7531 	if (!adj) {
7532 		pr_err("Adjacency does not exist for device %s from %s\n",
7533 		       dev->name, adj_dev->name);
7534 		WARN_ON(1);
7535 		return;
7536 	}
7537 
7538 	if (adj->ref_nr > ref_nr) {
7539 		pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7540 			 dev->name, adj_dev->name, ref_nr,
7541 			 adj->ref_nr - ref_nr);
7542 		adj->ref_nr -= ref_nr;
7543 		return;
7544 	}
7545 
7546 	if (adj->master)
7547 		sysfs_remove_link(&(dev->dev.kobj), "master");
7548 
7549 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7550 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7551 
7552 	list_del_rcu(&adj->list);
7553 	pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7554 		 adj_dev->name, dev->name, adj_dev->name);
7555 	netdev_put(adj_dev, &adj->dev_tracker);
7556 	kfree_rcu(adj, rcu);
7557 }
7558 
7559 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7560 					    struct net_device *upper_dev,
7561 					    struct list_head *up_list,
7562 					    struct list_head *down_list,
7563 					    void *private, bool master)
7564 {
7565 	int ret;
7566 
7567 	ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7568 					   private, master);
7569 	if (ret)
7570 		return ret;
7571 
7572 	ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7573 					   private, false);
7574 	if (ret) {
7575 		__netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7576 		return ret;
7577 	}
7578 
7579 	return 0;
7580 }
7581 
7582 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7583 					       struct net_device *upper_dev,
7584 					       u16 ref_nr,
7585 					       struct list_head *up_list,
7586 					       struct list_head *down_list)
7587 {
7588 	__netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7589 	__netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7590 }
7591 
7592 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7593 						struct net_device *upper_dev,
7594 						void *private, bool master)
7595 {
7596 	return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7597 						&dev->adj_list.upper,
7598 						&upper_dev->adj_list.lower,
7599 						private, master);
7600 }
7601 
7602 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7603 						   struct net_device *upper_dev)
7604 {
7605 	__netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7606 					   &dev->adj_list.upper,
7607 					   &upper_dev->adj_list.lower);
7608 }
7609 
7610 static int __netdev_upper_dev_link(struct net_device *dev,
7611 				   struct net_device *upper_dev, bool master,
7612 				   void *upper_priv, void *upper_info,
7613 				   struct netdev_nested_priv *priv,
7614 				   struct netlink_ext_ack *extack)
7615 {
7616 	struct netdev_notifier_changeupper_info changeupper_info = {
7617 		.info = {
7618 			.dev = dev,
7619 			.extack = extack,
7620 		},
7621 		.upper_dev = upper_dev,
7622 		.master = master,
7623 		.linking = true,
7624 		.upper_info = upper_info,
7625 	};
7626 	struct net_device *master_dev;
7627 	int ret = 0;
7628 
7629 	ASSERT_RTNL();
7630 
7631 	if (dev == upper_dev)
7632 		return -EBUSY;
7633 
7634 	/* To prevent loops, check if dev is not upper device to upper_dev. */
7635 	if (__netdev_has_upper_dev(upper_dev, dev))
7636 		return -EBUSY;
7637 
7638 	if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7639 		return -EMLINK;
7640 
7641 	if (!master) {
7642 		if (__netdev_has_upper_dev(dev, upper_dev))
7643 			return -EEXIST;
7644 	} else {
7645 		master_dev = __netdev_master_upper_dev_get(dev);
7646 		if (master_dev)
7647 			return master_dev == upper_dev ? -EEXIST : -EBUSY;
7648 	}
7649 
7650 	ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7651 					    &changeupper_info.info);
7652 	ret = notifier_to_errno(ret);
7653 	if (ret)
7654 		return ret;
7655 
7656 	ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7657 						   master);
7658 	if (ret)
7659 		return ret;
7660 
7661 	ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7662 					    &changeupper_info.info);
7663 	ret = notifier_to_errno(ret);
7664 	if (ret)
7665 		goto rollback;
7666 
7667 	__netdev_update_upper_level(dev, NULL);
7668 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7669 
7670 	__netdev_update_lower_level(upper_dev, priv);
7671 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7672 				    priv);
7673 
7674 	return 0;
7675 
7676 rollback:
7677 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7678 
7679 	return ret;
7680 }
7681 
7682 /**
7683  * netdev_upper_dev_link - Add a link to the upper device
7684  * @dev: device
7685  * @upper_dev: new upper device
7686  * @extack: netlink extended ack
7687  *
7688  * Adds a link to device which is upper to this one. The caller must hold
7689  * the RTNL lock. On a failure a negative errno code is returned.
7690  * On success the reference counts are adjusted and the function
7691  * returns zero.
7692  */
7693 int netdev_upper_dev_link(struct net_device *dev,
7694 			  struct net_device *upper_dev,
7695 			  struct netlink_ext_ack *extack)
7696 {
7697 	struct netdev_nested_priv priv = {
7698 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7699 		.data = NULL,
7700 	};
7701 
7702 	return __netdev_upper_dev_link(dev, upper_dev, false,
7703 				       NULL, NULL, &priv, extack);
7704 }
7705 EXPORT_SYMBOL(netdev_upper_dev_link);
7706 
7707 /**
7708  * netdev_master_upper_dev_link - Add a master link to the upper device
7709  * @dev: device
7710  * @upper_dev: new upper device
7711  * @upper_priv: upper device private
7712  * @upper_info: upper info to be passed down via notifier
7713  * @extack: netlink extended ack
7714  *
7715  * Adds a link to device which is upper to this one. In this case, only
7716  * one master upper device can be linked, although other non-master devices
7717  * might be linked as well. The caller must hold the RTNL lock.
7718  * On a failure a negative errno code is returned. On success the reference
7719  * counts are adjusted and the function returns zero.
7720  */
7721 int netdev_master_upper_dev_link(struct net_device *dev,
7722 				 struct net_device *upper_dev,
7723 				 void *upper_priv, void *upper_info,
7724 				 struct netlink_ext_ack *extack)
7725 {
7726 	struct netdev_nested_priv priv = {
7727 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7728 		.data = NULL,
7729 	};
7730 
7731 	return __netdev_upper_dev_link(dev, upper_dev, true,
7732 				       upper_priv, upper_info, &priv, extack);
7733 }
7734 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7735 
7736 static void __netdev_upper_dev_unlink(struct net_device *dev,
7737 				      struct net_device *upper_dev,
7738 				      struct netdev_nested_priv *priv)
7739 {
7740 	struct netdev_notifier_changeupper_info changeupper_info = {
7741 		.info = {
7742 			.dev = dev,
7743 		},
7744 		.upper_dev = upper_dev,
7745 		.linking = false,
7746 	};
7747 
7748 	ASSERT_RTNL();
7749 
7750 	changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7751 
7752 	call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7753 				      &changeupper_info.info);
7754 
7755 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7756 
7757 	call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7758 				      &changeupper_info.info);
7759 
7760 	__netdev_update_upper_level(dev, NULL);
7761 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7762 
7763 	__netdev_update_lower_level(upper_dev, priv);
7764 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7765 				    priv);
7766 }
7767 
7768 /**
7769  * netdev_upper_dev_unlink - Removes a link to upper device
7770  * @dev: device
7771  * @upper_dev: new upper device
7772  *
7773  * Removes a link to device which is upper to this one. The caller must hold
7774  * the RTNL lock.
7775  */
7776 void netdev_upper_dev_unlink(struct net_device *dev,
7777 			     struct net_device *upper_dev)
7778 {
7779 	struct netdev_nested_priv priv = {
7780 		.flags = NESTED_SYNC_TODO,
7781 		.data = NULL,
7782 	};
7783 
7784 	__netdev_upper_dev_unlink(dev, upper_dev, &priv);
7785 }
7786 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7787 
7788 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7789 				      struct net_device *lower_dev,
7790 				      bool val)
7791 {
7792 	struct netdev_adjacent *adj;
7793 
7794 	adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7795 	if (adj)
7796 		adj->ignore = val;
7797 
7798 	adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7799 	if (adj)
7800 		adj->ignore = val;
7801 }
7802 
7803 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7804 					struct net_device *lower_dev)
7805 {
7806 	__netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7807 }
7808 
7809 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7810 				       struct net_device *lower_dev)
7811 {
7812 	__netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7813 }
7814 
7815 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7816 				   struct net_device *new_dev,
7817 				   struct net_device *dev,
7818 				   struct netlink_ext_ack *extack)
7819 {
7820 	struct netdev_nested_priv priv = {
7821 		.flags = 0,
7822 		.data = NULL,
7823 	};
7824 	int err;
7825 
7826 	if (!new_dev)
7827 		return 0;
7828 
7829 	if (old_dev && new_dev != old_dev)
7830 		netdev_adjacent_dev_disable(dev, old_dev);
7831 	err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
7832 				      extack);
7833 	if (err) {
7834 		if (old_dev && new_dev != old_dev)
7835 			netdev_adjacent_dev_enable(dev, old_dev);
7836 		return err;
7837 	}
7838 
7839 	return 0;
7840 }
7841 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7842 
7843 void netdev_adjacent_change_commit(struct net_device *old_dev,
7844 				   struct net_device *new_dev,
7845 				   struct net_device *dev)
7846 {
7847 	struct netdev_nested_priv priv = {
7848 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7849 		.data = NULL,
7850 	};
7851 
7852 	if (!new_dev || !old_dev)
7853 		return;
7854 
7855 	if (new_dev == old_dev)
7856 		return;
7857 
7858 	netdev_adjacent_dev_enable(dev, old_dev);
7859 	__netdev_upper_dev_unlink(old_dev, dev, &priv);
7860 }
7861 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7862 
7863 void netdev_adjacent_change_abort(struct net_device *old_dev,
7864 				  struct net_device *new_dev,
7865 				  struct net_device *dev)
7866 {
7867 	struct netdev_nested_priv priv = {
7868 		.flags = 0,
7869 		.data = NULL,
7870 	};
7871 
7872 	if (!new_dev)
7873 		return;
7874 
7875 	if (old_dev && new_dev != old_dev)
7876 		netdev_adjacent_dev_enable(dev, old_dev);
7877 
7878 	__netdev_upper_dev_unlink(new_dev, dev, &priv);
7879 }
7880 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7881 
7882 /**
7883  * netdev_bonding_info_change - Dispatch event about slave change
7884  * @dev: device
7885  * @bonding_info: info to dispatch
7886  *
7887  * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7888  * The caller must hold the RTNL lock.
7889  */
7890 void netdev_bonding_info_change(struct net_device *dev,
7891 				struct netdev_bonding_info *bonding_info)
7892 {
7893 	struct netdev_notifier_bonding_info info = {
7894 		.info.dev = dev,
7895 	};
7896 
7897 	memcpy(&info.bonding_info, bonding_info,
7898 	       sizeof(struct netdev_bonding_info));
7899 	call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7900 				      &info.info);
7901 }
7902 EXPORT_SYMBOL(netdev_bonding_info_change);
7903 
7904 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
7905 					   struct netlink_ext_ack *extack)
7906 {
7907 	struct netdev_notifier_offload_xstats_info info = {
7908 		.info.dev = dev,
7909 		.info.extack = extack,
7910 		.type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7911 	};
7912 	int err;
7913 	int rc;
7914 
7915 	dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
7916 					 GFP_KERNEL);
7917 	if (!dev->offload_xstats_l3)
7918 		return -ENOMEM;
7919 
7920 	rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
7921 						  NETDEV_OFFLOAD_XSTATS_DISABLE,
7922 						  &info.info);
7923 	err = notifier_to_errno(rc);
7924 	if (err)
7925 		goto free_stats;
7926 
7927 	return 0;
7928 
7929 free_stats:
7930 	kfree(dev->offload_xstats_l3);
7931 	dev->offload_xstats_l3 = NULL;
7932 	return err;
7933 }
7934 
7935 int netdev_offload_xstats_enable(struct net_device *dev,
7936 				 enum netdev_offload_xstats_type type,
7937 				 struct netlink_ext_ack *extack)
7938 {
7939 	ASSERT_RTNL();
7940 
7941 	if (netdev_offload_xstats_enabled(dev, type))
7942 		return -EALREADY;
7943 
7944 	switch (type) {
7945 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7946 		return netdev_offload_xstats_enable_l3(dev, extack);
7947 	}
7948 
7949 	WARN_ON(1);
7950 	return -EINVAL;
7951 }
7952 EXPORT_SYMBOL(netdev_offload_xstats_enable);
7953 
7954 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
7955 {
7956 	struct netdev_notifier_offload_xstats_info info = {
7957 		.info.dev = dev,
7958 		.type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7959 	};
7960 
7961 	call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
7962 				      &info.info);
7963 	kfree(dev->offload_xstats_l3);
7964 	dev->offload_xstats_l3 = NULL;
7965 }
7966 
7967 int netdev_offload_xstats_disable(struct net_device *dev,
7968 				  enum netdev_offload_xstats_type type)
7969 {
7970 	ASSERT_RTNL();
7971 
7972 	if (!netdev_offload_xstats_enabled(dev, type))
7973 		return -EALREADY;
7974 
7975 	switch (type) {
7976 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7977 		netdev_offload_xstats_disable_l3(dev);
7978 		return 0;
7979 	}
7980 
7981 	WARN_ON(1);
7982 	return -EINVAL;
7983 }
7984 EXPORT_SYMBOL(netdev_offload_xstats_disable);
7985 
7986 static void netdev_offload_xstats_disable_all(struct net_device *dev)
7987 {
7988 	netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
7989 }
7990 
7991 static struct rtnl_hw_stats64 *
7992 netdev_offload_xstats_get_ptr(const struct net_device *dev,
7993 			      enum netdev_offload_xstats_type type)
7994 {
7995 	switch (type) {
7996 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7997 		return dev->offload_xstats_l3;
7998 	}
7999 
8000 	WARN_ON(1);
8001 	return NULL;
8002 }
8003 
8004 bool netdev_offload_xstats_enabled(const struct net_device *dev,
8005 				   enum netdev_offload_xstats_type type)
8006 {
8007 	ASSERT_RTNL();
8008 
8009 	return netdev_offload_xstats_get_ptr(dev, type);
8010 }
8011 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
8012 
8013 struct netdev_notifier_offload_xstats_ru {
8014 	bool used;
8015 };
8016 
8017 struct netdev_notifier_offload_xstats_rd {
8018 	struct rtnl_hw_stats64 stats;
8019 	bool used;
8020 };
8021 
8022 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
8023 				  const struct rtnl_hw_stats64 *src)
8024 {
8025 	dest->rx_packets	  += src->rx_packets;
8026 	dest->tx_packets	  += src->tx_packets;
8027 	dest->rx_bytes		  += src->rx_bytes;
8028 	dest->tx_bytes		  += src->tx_bytes;
8029 	dest->rx_errors		  += src->rx_errors;
8030 	dest->tx_errors		  += src->tx_errors;
8031 	dest->rx_dropped	  += src->rx_dropped;
8032 	dest->tx_dropped	  += src->tx_dropped;
8033 	dest->multicast		  += src->multicast;
8034 }
8035 
8036 static int netdev_offload_xstats_get_used(struct net_device *dev,
8037 					  enum netdev_offload_xstats_type type,
8038 					  bool *p_used,
8039 					  struct netlink_ext_ack *extack)
8040 {
8041 	struct netdev_notifier_offload_xstats_ru report_used = {};
8042 	struct netdev_notifier_offload_xstats_info info = {
8043 		.info.dev = dev,
8044 		.info.extack = extack,
8045 		.type = type,
8046 		.report_used = &report_used,
8047 	};
8048 	int rc;
8049 
8050 	WARN_ON(!netdev_offload_xstats_enabled(dev, type));
8051 	rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
8052 					   &info.info);
8053 	*p_used = report_used.used;
8054 	return notifier_to_errno(rc);
8055 }
8056 
8057 static int netdev_offload_xstats_get_stats(struct net_device *dev,
8058 					   enum netdev_offload_xstats_type type,
8059 					   struct rtnl_hw_stats64 *p_stats,
8060 					   bool *p_used,
8061 					   struct netlink_ext_ack *extack)
8062 {
8063 	struct netdev_notifier_offload_xstats_rd report_delta = {};
8064 	struct netdev_notifier_offload_xstats_info info = {
8065 		.info.dev = dev,
8066 		.info.extack = extack,
8067 		.type = type,
8068 		.report_delta = &report_delta,
8069 	};
8070 	struct rtnl_hw_stats64 *stats;
8071 	int rc;
8072 
8073 	stats = netdev_offload_xstats_get_ptr(dev, type);
8074 	if (WARN_ON(!stats))
8075 		return -EINVAL;
8076 
8077 	rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
8078 					   &info.info);
8079 
8080 	/* Cache whatever we got, even if there was an error, otherwise the
8081 	 * successful stats retrievals would get lost.
8082 	 */
8083 	netdev_hw_stats64_add(stats, &report_delta.stats);
8084 
8085 	if (p_stats)
8086 		*p_stats = *stats;
8087 	*p_used = report_delta.used;
8088 
8089 	return notifier_to_errno(rc);
8090 }
8091 
8092 int netdev_offload_xstats_get(struct net_device *dev,
8093 			      enum netdev_offload_xstats_type type,
8094 			      struct rtnl_hw_stats64 *p_stats, bool *p_used,
8095 			      struct netlink_ext_ack *extack)
8096 {
8097 	ASSERT_RTNL();
8098 
8099 	if (p_stats)
8100 		return netdev_offload_xstats_get_stats(dev, type, p_stats,
8101 						       p_used, extack);
8102 	else
8103 		return netdev_offload_xstats_get_used(dev, type, p_used,
8104 						      extack);
8105 }
8106 EXPORT_SYMBOL(netdev_offload_xstats_get);
8107 
8108 void
8109 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
8110 				   const struct rtnl_hw_stats64 *stats)
8111 {
8112 	report_delta->used = true;
8113 	netdev_hw_stats64_add(&report_delta->stats, stats);
8114 }
8115 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
8116 
8117 void
8118 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
8119 {
8120 	report_used->used = true;
8121 }
8122 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
8123 
8124 void netdev_offload_xstats_push_delta(struct net_device *dev,
8125 				      enum netdev_offload_xstats_type type,
8126 				      const struct rtnl_hw_stats64 *p_stats)
8127 {
8128 	struct rtnl_hw_stats64 *stats;
8129 
8130 	ASSERT_RTNL();
8131 
8132 	stats = netdev_offload_xstats_get_ptr(dev, type);
8133 	if (WARN_ON(!stats))
8134 		return;
8135 
8136 	netdev_hw_stats64_add(stats, p_stats);
8137 }
8138 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
8139 
8140 /**
8141  * netdev_get_xmit_slave - Get the xmit slave of master device
8142  * @dev: device
8143  * @skb: The packet
8144  * @all_slaves: assume all the slaves are active
8145  *
8146  * The reference counters are not incremented so the caller must be
8147  * careful with locks. The caller must hold RCU lock.
8148  * %NULL is returned if no slave is found.
8149  */
8150 
8151 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8152 					 struct sk_buff *skb,
8153 					 bool all_slaves)
8154 {
8155 	const struct net_device_ops *ops = dev->netdev_ops;
8156 
8157 	if (!ops->ndo_get_xmit_slave)
8158 		return NULL;
8159 	return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8160 }
8161 EXPORT_SYMBOL(netdev_get_xmit_slave);
8162 
8163 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8164 						  struct sock *sk)
8165 {
8166 	const struct net_device_ops *ops = dev->netdev_ops;
8167 
8168 	if (!ops->ndo_sk_get_lower_dev)
8169 		return NULL;
8170 	return ops->ndo_sk_get_lower_dev(dev, sk);
8171 }
8172 
8173 /**
8174  * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8175  * @dev: device
8176  * @sk: the socket
8177  *
8178  * %NULL is returned if no lower device is found.
8179  */
8180 
8181 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8182 					    struct sock *sk)
8183 {
8184 	struct net_device *lower;
8185 
8186 	lower = netdev_sk_get_lower_dev(dev, sk);
8187 	while (lower) {
8188 		dev = lower;
8189 		lower = netdev_sk_get_lower_dev(dev, sk);
8190 	}
8191 
8192 	return dev;
8193 }
8194 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8195 
8196 static void netdev_adjacent_add_links(struct net_device *dev)
8197 {
8198 	struct netdev_adjacent *iter;
8199 
8200 	struct net *net = dev_net(dev);
8201 
8202 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8203 		if (!net_eq(net, dev_net(iter->dev)))
8204 			continue;
8205 		netdev_adjacent_sysfs_add(iter->dev, dev,
8206 					  &iter->dev->adj_list.lower);
8207 		netdev_adjacent_sysfs_add(dev, iter->dev,
8208 					  &dev->adj_list.upper);
8209 	}
8210 
8211 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8212 		if (!net_eq(net, dev_net(iter->dev)))
8213 			continue;
8214 		netdev_adjacent_sysfs_add(iter->dev, dev,
8215 					  &iter->dev->adj_list.upper);
8216 		netdev_adjacent_sysfs_add(dev, iter->dev,
8217 					  &dev->adj_list.lower);
8218 	}
8219 }
8220 
8221 static void netdev_adjacent_del_links(struct net_device *dev)
8222 {
8223 	struct netdev_adjacent *iter;
8224 
8225 	struct net *net = dev_net(dev);
8226 
8227 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8228 		if (!net_eq(net, dev_net(iter->dev)))
8229 			continue;
8230 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8231 					  &iter->dev->adj_list.lower);
8232 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8233 					  &dev->adj_list.upper);
8234 	}
8235 
8236 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8237 		if (!net_eq(net, dev_net(iter->dev)))
8238 			continue;
8239 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8240 					  &iter->dev->adj_list.upper);
8241 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8242 					  &dev->adj_list.lower);
8243 	}
8244 }
8245 
8246 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8247 {
8248 	struct netdev_adjacent *iter;
8249 
8250 	struct net *net = dev_net(dev);
8251 
8252 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8253 		if (!net_eq(net, dev_net(iter->dev)))
8254 			continue;
8255 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8256 					  &iter->dev->adj_list.lower);
8257 		netdev_adjacent_sysfs_add(iter->dev, dev,
8258 					  &iter->dev->adj_list.lower);
8259 	}
8260 
8261 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8262 		if (!net_eq(net, dev_net(iter->dev)))
8263 			continue;
8264 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8265 					  &iter->dev->adj_list.upper);
8266 		netdev_adjacent_sysfs_add(iter->dev, dev,
8267 					  &iter->dev->adj_list.upper);
8268 	}
8269 }
8270 
8271 void *netdev_lower_dev_get_private(struct net_device *dev,
8272 				   struct net_device *lower_dev)
8273 {
8274 	struct netdev_adjacent *lower;
8275 
8276 	if (!lower_dev)
8277 		return NULL;
8278 	lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8279 	if (!lower)
8280 		return NULL;
8281 
8282 	return lower->private;
8283 }
8284 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8285 
8286 
8287 /**
8288  * netdev_lower_state_changed - Dispatch event about lower device state change
8289  * @lower_dev: device
8290  * @lower_state_info: state to dispatch
8291  *
8292  * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8293  * The caller must hold the RTNL lock.
8294  */
8295 void netdev_lower_state_changed(struct net_device *lower_dev,
8296 				void *lower_state_info)
8297 {
8298 	struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8299 		.info.dev = lower_dev,
8300 	};
8301 
8302 	ASSERT_RTNL();
8303 	changelowerstate_info.lower_state_info = lower_state_info;
8304 	call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8305 				      &changelowerstate_info.info);
8306 }
8307 EXPORT_SYMBOL(netdev_lower_state_changed);
8308 
8309 static void dev_change_rx_flags(struct net_device *dev, int flags)
8310 {
8311 	const struct net_device_ops *ops = dev->netdev_ops;
8312 
8313 	if (ops->ndo_change_rx_flags)
8314 		ops->ndo_change_rx_flags(dev, flags);
8315 }
8316 
8317 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8318 {
8319 	unsigned int old_flags = dev->flags;
8320 	kuid_t uid;
8321 	kgid_t gid;
8322 
8323 	ASSERT_RTNL();
8324 
8325 	dev->flags |= IFF_PROMISC;
8326 	dev->promiscuity += inc;
8327 	if (dev->promiscuity == 0) {
8328 		/*
8329 		 * Avoid overflow.
8330 		 * If inc causes overflow, untouch promisc and return error.
8331 		 */
8332 		if (inc < 0)
8333 			dev->flags &= ~IFF_PROMISC;
8334 		else {
8335 			dev->promiscuity -= inc;
8336 			netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8337 			return -EOVERFLOW;
8338 		}
8339 	}
8340 	if (dev->flags != old_flags) {
8341 		pr_info("device %s %s promiscuous mode\n",
8342 			dev->name,
8343 			dev->flags & IFF_PROMISC ? "entered" : "left");
8344 		if (audit_enabled) {
8345 			current_uid_gid(&uid, &gid);
8346 			audit_log(audit_context(), GFP_ATOMIC,
8347 				  AUDIT_ANOM_PROMISCUOUS,
8348 				  "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8349 				  dev->name, (dev->flags & IFF_PROMISC),
8350 				  (old_flags & IFF_PROMISC),
8351 				  from_kuid(&init_user_ns, audit_get_loginuid(current)),
8352 				  from_kuid(&init_user_ns, uid),
8353 				  from_kgid(&init_user_ns, gid),
8354 				  audit_get_sessionid(current));
8355 		}
8356 
8357 		dev_change_rx_flags(dev, IFF_PROMISC);
8358 	}
8359 	if (notify)
8360 		__dev_notify_flags(dev, old_flags, IFF_PROMISC);
8361 	return 0;
8362 }
8363 
8364 /**
8365  *	dev_set_promiscuity	- update promiscuity count on a device
8366  *	@dev: device
8367  *	@inc: modifier
8368  *
8369  *	Add or remove promiscuity from a device. While the count in the device
8370  *	remains above zero the interface remains promiscuous. Once it hits zero
8371  *	the device reverts back to normal filtering operation. A negative inc
8372  *	value is used to drop promiscuity on the device.
8373  *	Return 0 if successful or a negative errno code on error.
8374  */
8375 int dev_set_promiscuity(struct net_device *dev, int inc)
8376 {
8377 	unsigned int old_flags = dev->flags;
8378 	int err;
8379 
8380 	err = __dev_set_promiscuity(dev, inc, true);
8381 	if (err < 0)
8382 		return err;
8383 	if (dev->flags != old_flags)
8384 		dev_set_rx_mode(dev);
8385 	return err;
8386 }
8387 EXPORT_SYMBOL(dev_set_promiscuity);
8388 
8389 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8390 {
8391 	unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8392 
8393 	ASSERT_RTNL();
8394 
8395 	dev->flags |= IFF_ALLMULTI;
8396 	dev->allmulti += inc;
8397 	if (dev->allmulti == 0) {
8398 		/*
8399 		 * Avoid overflow.
8400 		 * If inc causes overflow, untouch allmulti and return error.
8401 		 */
8402 		if (inc < 0)
8403 			dev->flags &= ~IFF_ALLMULTI;
8404 		else {
8405 			dev->allmulti -= inc;
8406 			netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8407 			return -EOVERFLOW;
8408 		}
8409 	}
8410 	if (dev->flags ^ old_flags) {
8411 		dev_change_rx_flags(dev, IFF_ALLMULTI);
8412 		dev_set_rx_mode(dev);
8413 		if (notify)
8414 			__dev_notify_flags(dev, old_flags,
8415 					   dev->gflags ^ old_gflags);
8416 	}
8417 	return 0;
8418 }
8419 
8420 /**
8421  *	dev_set_allmulti	- update allmulti count on a device
8422  *	@dev: device
8423  *	@inc: modifier
8424  *
8425  *	Add or remove reception of all multicast frames to a device. While the
8426  *	count in the device remains above zero the interface remains listening
8427  *	to all interfaces. Once it hits zero the device reverts back to normal
8428  *	filtering operation. A negative @inc value is used to drop the counter
8429  *	when releasing a resource needing all multicasts.
8430  *	Return 0 if successful or a negative errno code on error.
8431  */
8432 
8433 int dev_set_allmulti(struct net_device *dev, int inc)
8434 {
8435 	return __dev_set_allmulti(dev, inc, true);
8436 }
8437 EXPORT_SYMBOL(dev_set_allmulti);
8438 
8439 /*
8440  *	Upload unicast and multicast address lists to device and
8441  *	configure RX filtering. When the device doesn't support unicast
8442  *	filtering it is put in promiscuous mode while unicast addresses
8443  *	are present.
8444  */
8445 void __dev_set_rx_mode(struct net_device *dev)
8446 {
8447 	const struct net_device_ops *ops = dev->netdev_ops;
8448 
8449 	/* dev_open will call this function so the list will stay sane. */
8450 	if (!(dev->flags&IFF_UP))
8451 		return;
8452 
8453 	if (!netif_device_present(dev))
8454 		return;
8455 
8456 	if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8457 		/* Unicast addresses changes may only happen under the rtnl,
8458 		 * therefore calling __dev_set_promiscuity here is safe.
8459 		 */
8460 		if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8461 			__dev_set_promiscuity(dev, 1, false);
8462 			dev->uc_promisc = true;
8463 		} else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8464 			__dev_set_promiscuity(dev, -1, false);
8465 			dev->uc_promisc = false;
8466 		}
8467 	}
8468 
8469 	if (ops->ndo_set_rx_mode)
8470 		ops->ndo_set_rx_mode(dev);
8471 }
8472 
8473 void dev_set_rx_mode(struct net_device *dev)
8474 {
8475 	netif_addr_lock_bh(dev);
8476 	__dev_set_rx_mode(dev);
8477 	netif_addr_unlock_bh(dev);
8478 }
8479 
8480 /**
8481  *	dev_get_flags - get flags reported to userspace
8482  *	@dev: device
8483  *
8484  *	Get the combination of flag bits exported through APIs to userspace.
8485  */
8486 unsigned int dev_get_flags(const struct net_device *dev)
8487 {
8488 	unsigned int flags;
8489 
8490 	flags = (dev->flags & ~(IFF_PROMISC |
8491 				IFF_ALLMULTI |
8492 				IFF_RUNNING |
8493 				IFF_LOWER_UP |
8494 				IFF_DORMANT)) |
8495 		(dev->gflags & (IFF_PROMISC |
8496 				IFF_ALLMULTI));
8497 
8498 	if (netif_running(dev)) {
8499 		if (netif_oper_up(dev))
8500 			flags |= IFF_RUNNING;
8501 		if (netif_carrier_ok(dev))
8502 			flags |= IFF_LOWER_UP;
8503 		if (netif_dormant(dev))
8504 			flags |= IFF_DORMANT;
8505 	}
8506 
8507 	return flags;
8508 }
8509 EXPORT_SYMBOL(dev_get_flags);
8510 
8511 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8512 		       struct netlink_ext_ack *extack)
8513 {
8514 	unsigned int old_flags = dev->flags;
8515 	int ret;
8516 
8517 	ASSERT_RTNL();
8518 
8519 	/*
8520 	 *	Set the flags on our device.
8521 	 */
8522 
8523 	dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8524 			       IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8525 			       IFF_AUTOMEDIA)) |
8526 		     (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8527 				    IFF_ALLMULTI));
8528 
8529 	/*
8530 	 *	Load in the correct multicast list now the flags have changed.
8531 	 */
8532 
8533 	if ((old_flags ^ flags) & IFF_MULTICAST)
8534 		dev_change_rx_flags(dev, IFF_MULTICAST);
8535 
8536 	dev_set_rx_mode(dev);
8537 
8538 	/*
8539 	 *	Have we downed the interface. We handle IFF_UP ourselves
8540 	 *	according to user attempts to set it, rather than blindly
8541 	 *	setting it.
8542 	 */
8543 
8544 	ret = 0;
8545 	if ((old_flags ^ flags) & IFF_UP) {
8546 		if (old_flags & IFF_UP)
8547 			__dev_close(dev);
8548 		else
8549 			ret = __dev_open(dev, extack);
8550 	}
8551 
8552 	if ((flags ^ dev->gflags) & IFF_PROMISC) {
8553 		int inc = (flags & IFF_PROMISC) ? 1 : -1;
8554 		unsigned int old_flags = dev->flags;
8555 
8556 		dev->gflags ^= IFF_PROMISC;
8557 
8558 		if (__dev_set_promiscuity(dev, inc, false) >= 0)
8559 			if (dev->flags != old_flags)
8560 				dev_set_rx_mode(dev);
8561 	}
8562 
8563 	/* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8564 	 * is important. Some (broken) drivers set IFF_PROMISC, when
8565 	 * IFF_ALLMULTI is requested not asking us and not reporting.
8566 	 */
8567 	if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8568 		int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8569 
8570 		dev->gflags ^= IFF_ALLMULTI;
8571 		__dev_set_allmulti(dev, inc, false);
8572 	}
8573 
8574 	return ret;
8575 }
8576 
8577 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8578 			unsigned int gchanges)
8579 {
8580 	unsigned int changes = dev->flags ^ old_flags;
8581 
8582 	if (gchanges)
8583 		rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8584 
8585 	if (changes & IFF_UP) {
8586 		if (dev->flags & IFF_UP)
8587 			call_netdevice_notifiers(NETDEV_UP, dev);
8588 		else
8589 			call_netdevice_notifiers(NETDEV_DOWN, dev);
8590 	}
8591 
8592 	if (dev->flags & IFF_UP &&
8593 	    (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8594 		struct netdev_notifier_change_info change_info = {
8595 			.info = {
8596 				.dev = dev,
8597 			},
8598 			.flags_changed = changes,
8599 		};
8600 
8601 		call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8602 	}
8603 }
8604 
8605 /**
8606  *	dev_change_flags - change device settings
8607  *	@dev: device
8608  *	@flags: device state flags
8609  *	@extack: netlink extended ack
8610  *
8611  *	Change settings on device based state flags. The flags are
8612  *	in the userspace exported format.
8613  */
8614 int dev_change_flags(struct net_device *dev, unsigned int flags,
8615 		     struct netlink_ext_ack *extack)
8616 {
8617 	int ret;
8618 	unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8619 
8620 	ret = __dev_change_flags(dev, flags, extack);
8621 	if (ret < 0)
8622 		return ret;
8623 
8624 	changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8625 	__dev_notify_flags(dev, old_flags, changes);
8626 	return ret;
8627 }
8628 EXPORT_SYMBOL(dev_change_flags);
8629 
8630 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8631 {
8632 	const struct net_device_ops *ops = dev->netdev_ops;
8633 
8634 	if (ops->ndo_change_mtu)
8635 		return ops->ndo_change_mtu(dev, new_mtu);
8636 
8637 	/* Pairs with all the lockless reads of dev->mtu in the stack */
8638 	WRITE_ONCE(dev->mtu, new_mtu);
8639 	return 0;
8640 }
8641 EXPORT_SYMBOL(__dev_set_mtu);
8642 
8643 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8644 		     struct netlink_ext_ack *extack)
8645 {
8646 	/* MTU must be positive, and in range */
8647 	if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8648 		NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8649 		return -EINVAL;
8650 	}
8651 
8652 	if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8653 		NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8654 		return -EINVAL;
8655 	}
8656 	return 0;
8657 }
8658 
8659 /**
8660  *	dev_set_mtu_ext - Change maximum transfer unit
8661  *	@dev: device
8662  *	@new_mtu: new transfer unit
8663  *	@extack: netlink extended ack
8664  *
8665  *	Change the maximum transfer size of the network device.
8666  */
8667 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8668 		    struct netlink_ext_ack *extack)
8669 {
8670 	int err, orig_mtu;
8671 
8672 	if (new_mtu == dev->mtu)
8673 		return 0;
8674 
8675 	err = dev_validate_mtu(dev, new_mtu, extack);
8676 	if (err)
8677 		return err;
8678 
8679 	if (!netif_device_present(dev))
8680 		return -ENODEV;
8681 
8682 	err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8683 	err = notifier_to_errno(err);
8684 	if (err)
8685 		return err;
8686 
8687 	orig_mtu = dev->mtu;
8688 	err = __dev_set_mtu(dev, new_mtu);
8689 
8690 	if (!err) {
8691 		err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8692 						   orig_mtu);
8693 		err = notifier_to_errno(err);
8694 		if (err) {
8695 			/* setting mtu back and notifying everyone again,
8696 			 * so that they have a chance to revert changes.
8697 			 */
8698 			__dev_set_mtu(dev, orig_mtu);
8699 			call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8700 						     new_mtu);
8701 		}
8702 	}
8703 	return err;
8704 }
8705 
8706 int dev_set_mtu(struct net_device *dev, int new_mtu)
8707 {
8708 	struct netlink_ext_ack extack;
8709 	int err;
8710 
8711 	memset(&extack, 0, sizeof(extack));
8712 	err = dev_set_mtu_ext(dev, new_mtu, &extack);
8713 	if (err && extack._msg)
8714 		net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8715 	return err;
8716 }
8717 EXPORT_SYMBOL(dev_set_mtu);
8718 
8719 /**
8720  *	dev_change_tx_queue_len - Change TX queue length of a netdevice
8721  *	@dev: device
8722  *	@new_len: new tx queue length
8723  */
8724 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8725 {
8726 	unsigned int orig_len = dev->tx_queue_len;
8727 	int res;
8728 
8729 	if (new_len != (unsigned int)new_len)
8730 		return -ERANGE;
8731 
8732 	if (new_len != orig_len) {
8733 		dev->tx_queue_len = new_len;
8734 		res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8735 		res = notifier_to_errno(res);
8736 		if (res)
8737 			goto err_rollback;
8738 		res = dev_qdisc_change_tx_queue_len(dev);
8739 		if (res)
8740 			goto err_rollback;
8741 	}
8742 
8743 	return 0;
8744 
8745 err_rollback:
8746 	netdev_err(dev, "refused to change device tx_queue_len\n");
8747 	dev->tx_queue_len = orig_len;
8748 	return res;
8749 }
8750 
8751 /**
8752  *	dev_set_group - Change group this device belongs to
8753  *	@dev: device
8754  *	@new_group: group this device should belong to
8755  */
8756 void dev_set_group(struct net_device *dev, int new_group)
8757 {
8758 	dev->group = new_group;
8759 }
8760 
8761 /**
8762  *	dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8763  *	@dev: device
8764  *	@addr: new address
8765  *	@extack: netlink extended ack
8766  */
8767 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8768 			      struct netlink_ext_ack *extack)
8769 {
8770 	struct netdev_notifier_pre_changeaddr_info info = {
8771 		.info.dev = dev,
8772 		.info.extack = extack,
8773 		.dev_addr = addr,
8774 	};
8775 	int rc;
8776 
8777 	rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8778 	return notifier_to_errno(rc);
8779 }
8780 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8781 
8782 /**
8783  *	dev_set_mac_address - Change Media Access Control Address
8784  *	@dev: device
8785  *	@sa: new address
8786  *	@extack: netlink extended ack
8787  *
8788  *	Change the hardware (MAC) address of the device
8789  */
8790 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8791 			struct netlink_ext_ack *extack)
8792 {
8793 	const struct net_device_ops *ops = dev->netdev_ops;
8794 	int err;
8795 
8796 	if (!ops->ndo_set_mac_address)
8797 		return -EOPNOTSUPP;
8798 	if (sa->sa_family != dev->type)
8799 		return -EINVAL;
8800 	if (!netif_device_present(dev))
8801 		return -ENODEV;
8802 	err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8803 	if (err)
8804 		return err;
8805 	err = ops->ndo_set_mac_address(dev, sa);
8806 	if (err)
8807 		return err;
8808 	dev->addr_assign_type = NET_ADDR_SET;
8809 	call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8810 	add_device_randomness(dev->dev_addr, dev->addr_len);
8811 	return 0;
8812 }
8813 EXPORT_SYMBOL(dev_set_mac_address);
8814 
8815 static DECLARE_RWSEM(dev_addr_sem);
8816 
8817 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
8818 			     struct netlink_ext_ack *extack)
8819 {
8820 	int ret;
8821 
8822 	down_write(&dev_addr_sem);
8823 	ret = dev_set_mac_address(dev, sa, extack);
8824 	up_write(&dev_addr_sem);
8825 	return ret;
8826 }
8827 EXPORT_SYMBOL(dev_set_mac_address_user);
8828 
8829 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
8830 {
8831 	size_t size = sizeof(sa->sa_data);
8832 	struct net_device *dev;
8833 	int ret = 0;
8834 
8835 	down_read(&dev_addr_sem);
8836 	rcu_read_lock();
8837 
8838 	dev = dev_get_by_name_rcu(net, dev_name);
8839 	if (!dev) {
8840 		ret = -ENODEV;
8841 		goto unlock;
8842 	}
8843 	if (!dev->addr_len)
8844 		memset(sa->sa_data, 0, size);
8845 	else
8846 		memcpy(sa->sa_data, dev->dev_addr,
8847 		       min_t(size_t, size, dev->addr_len));
8848 	sa->sa_family = dev->type;
8849 
8850 unlock:
8851 	rcu_read_unlock();
8852 	up_read(&dev_addr_sem);
8853 	return ret;
8854 }
8855 EXPORT_SYMBOL(dev_get_mac_address);
8856 
8857 /**
8858  *	dev_change_carrier - Change device carrier
8859  *	@dev: device
8860  *	@new_carrier: new value
8861  *
8862  *	Change device carrier
8863  */
8864 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8865 {
8866 	const struct net_device_ops *ops = dev->netdev_ops;
8867 
8868 	if (!ops->ndo_change_carrier)
8869 		return -EOPNOTSUPP;
8870 	if (!netif_device_present(dev))
8871 		return -ENODEV;
8872 	return ops->ndo_change_carrier(dev, new_carrier);
8873 }
8874 
8875 /**
8876  *	dev_get_phys_port_id - Get device physical port ID
8877  *	@dev: device
8878  *	@ppid: port ID
8879  *
8880  *	Get device physical port ID
8881  */
8882 int dev_get_phys_port_id(struct net_device *dev,
8883 			 struct netdev_phys_item_id *ppid)
8884 {
8885 	const struct net_device_ops *ops = dev->netdev_ops;
8886 
8887 	if (!ops->ndo_get_phys_port_id)
8888 		return -EOPNOTSUPP;
8889 	return ops->ndo_get_phys_port_id(dev, ppid);
8890 }
8891 
8892 /**
8893  *	dev_get_phys_port_name - Get device physical port name
8894  *	@dev: device
8895  *	@name: port name
8896  *	@len: limit of bytes to copy to name
8897  *
8898  *	Get device physical port name
8899  */
8900 int dev_get_phys_port_name(struct net_device *dev,
8901 			   char *name, size_t len)
8902 {
8903 	const struct net_device_ops *ops = dev->netdev_ops;
8904 	int err;
8905 
8906 	if (ops->ndo_get_phys_port_name) {
8907 		err = ops->ndo_get_phys_port_name(dev, name, len);
8908 		if (err != -EOPNOTSUPP)
8909 			return err;
8910 	}
8911 	return devlink_compat_phys_port_name_get(dev, name, len);
8912 }
8913 
8914 /**
8915  *	dev_get_port_parent_id - Get the device's port parent identifier
8916  *	@dev: network device
8917  *	@ppid: pointer to a storage for the port's parent identifier
8918  *	@recurse: allow/disallow recursion to lower devices
8919  *
8920  *	Get the devices's port parent identifier
8921  */
8922 int dev_get_port_parent_id(struct net_device *dev,
8923 			   struct netdev_phys_item_id *ppid,
8924 			   bool recurse)
8925 {
8926 	const struct net_device_ops *ops = dev->netdev_ops;
8927 	struct netdev_phys_item_id first = { };
8928 	struct net_device *lower_dev;
8929 	struct list_head *iter;
8930 	int err;
8931 
8932 	if (ops->ndo_get_port_parent_id) {
8933 		err = ops->ndo_get_port_parent_id(dev, ppid);
8934 		if (err != -EOPNOTSUPP)
8935 			return err;
8936 	}
8937 
8938 	err = devlink_compat_switch_id_get(dev, ppid);
8939 	if (!recurse || err != -EOPNOTSUPP)
8940 		return err;
8941 
8942 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
8943 		err = dev_get_port_parent_id(lower_dev, ppid, true);
8944 		if (err)
8945 			break;
8946 		if (!first.id_len)
8947 			first = *ppid;
8948 		else if (memcmp(&first, ppid, sizeof(*ppid)))
8949 			return -EOPNOTSUPP;
8950 	}
8951 
8952 	return err;
8953 }
8954 EXPORT_SYMBOL(dev_get_port_parent_id);
8955 
8956 /**
8957  *	netdev_port_same_parent_id - Indicate if two network devices have
8958  *	the same port parent identifier
8959  *	@a: first network device
8960  *	@b: second network device
8961  */
8962 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8963 {
8964 	struct netdev_phys_item_id a_id = { };
8965 	struct netdev_phys_item_id b_id = { };
8966 
8967 	if (dev_get_port_parent_id(a, &a_id, true) ||
8968 	    dev_get_port_parent_id(b, &b_id, true))
8969 		return false;
8970 
8971 	return netdev_phys_item_id_same(&a_id, &b_id);
8972 }
8973 EXPORT_SYMBOL(netdev_port_same_parent_id);
8974 
8975 /**
8976  *	dev_change_proto_down - set carrier according to proto_down.
8977  *
8978  *	@dev: device
8979  *	@proto_down: new value
8980  */
8981 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8982 {
8983 	if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN))
8984 		return -EOPNOTSUPP;
8985 	if (!netif_device_present(dev))
8986 		return -ENODEV;
8987 	if (proto_down)
8988 		netif_carrier_off(dev);
8989 	else
8990 		netif_carrier_on(dev);
8991 	dev->proto_down = proto_down;
8992 	return 0;
8993 }
8994 
8995 /**
8996  *	dev_change_proto_down_reason - proto down reason
8997  *
8998  *	@dev: device
8999  *	@mask: proto down mask
9000  *	@value: proto down value
9001  */
9002 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9003 				  u32 value)
9004 {
9005 	int b;
9006 
9007 	if (!mask) {
9008 		dev->proto_down_reason = value;
9009 	} else {
9010 		for_each_set_bit(b, &mask, 32) {
9011 			if (value & (1 << b))
9012 				dev->proto_down_reason |= BIT(b);
9013 			else
9014 				dev->proto_down_reason &= ~BIT(b);
9015 		}
9016 	}
9017 }
9018 
9019 struct bpf_xdp_link {
9020 	struct bpf_link link;
9021 	struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9022 	int flags;
9023 };
9024 
9025 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9026 {
9027 	if (flags & XDP_FLAGS_HW_MODE)
9028 		return XDP_MODE_HW;
9029 	if (flags & XDP_FLAGS_DRV_MODE)
9030 		return XDP_MODE_DRV;
9031 	if (flags & XDP_FLAGS_SKB_MODE)
9032 		return XDP_MODE_SKB;
9033 	return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9034 }
9035 
9036 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9037 {
9038 	switch (mode) {
9039 	case XDP_MODE_SKB:
9040 		return generic_xdp_install;
9041 	case XDP_MODE_DRV:
9042 	case XDP_MODE_HW:
9043 		return dev->netdev_ops->ndo_bpf;
9044 	default:
9045 		return NULL;
9046 	}
9047 }
9048 
9049 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9050 					 enum bpf_xdp_mode mode)
9051 {
9052 	return dev->xdp_state[mode].link;
9053 }
9054 
9055 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9056 				     enum bpf_xdp_mode mode)
9057 {
9058 	struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9059 
9060 	if (link)
9061 		return link->link.prog;
9062 	return dev->xdp_state[mode].prog;
9063 }
9064 
9065 u8 dev_xdp_prog_count(struct net_device *dev)
9066 {
9067 	u8 count = 0;
9068 	int i;
9069 
9070 	for (i = 0; i < __MAX_XDP_MODE; i++)
9071 		if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9072 			count++;
9073 	return count;
9074 }
9075 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9076 
9077 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9078 {
9079 	struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9080 
9081 	return prog ? prog->aux->id : 0;
9082 }
9083 
9084 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9085 			     struct bpf_xdp_link *link)
9086 {
9087 	dev->xdp_state[mode].link = link;
9088 	dev->xdp_state[mode].prog = NULL;
9089 }
9090 
9091 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9092 			     struct bpf_prog *prog)
9093 {
9094 	dev->xdp_state[mode].link = NULL;
9095 	dev->xdp_state[mode].prog = prog;
9096 }
9097 
9098 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9099 			   bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9100 			   u32 flags, struct bpf_prog *prog)
9101 {
9102 	struct netdev_bpf xdp;
9103 	int err;
9104 
9105 	memset(&xdp, 0, sizeof(xdp));
9106 	xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9107 	xdp.extack = extack;
9108 	xdp.flags = flags;
9109 	xdp.prog = prog;
9110 
9111 	/* Drivers assume refcnt is already incremented (i.e, prog pointer is
9112 	 * "moved" into driver), so they don't increment it on their own, but
9113 	 * they do decrement refcnt when program is detached or replaced.
9114 	 * Given net_device also owns link/prog, we need to bump refcnt here
9115 	 * to prevent drivers from underflowing it.
9116 	 */
9117 	if (prog)
9118 		bpf_prog_inc(prog);
9119 	err = bpf_op(dev, &xdp);
9120 	if (err) {
9121 		if (prog)
9122 			bpf_prog_put(prog);
9123 		return err;
9124 	}
9125 
9126 	if (mode != XDP_MODE_HW)
9127 		bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9128 
9129 	return 0;
9130 }
9131 
9132 static void dev_xdp_uninstall(struct net_device *dev)
9133 {
9134 	struct bpf_xdp_link *link;
9135 	struct bpf_prog *prog;
9136 	enum bpf_xdp_mode mode;
9137 	bpf_op_t bpf_op;
9138 
9139 	ASSERT_RTNL();
9140 
9141 	for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9142 		prog = dev_xdp_prog(dev, mode);
9143 		if (!prog)
9144 			continue;
9145 
9146 		bpf_op = dev_xdp_bpf_op(dev, mode);
9147 		if (!bpf_op)
9148 			continue;
9149 
9150 		WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9151 
9152 		/* auto-detach link from net device */
9153 		link = dev_xdp_link(dev, mode);
9154 		if (link)
9155 			link->dev = NULL;
9156 		else
9157 			bpf_prog_put(prog);
9158 
9159 		dev_xdp_set_link(dev, mode, NULL);
9160 	}
9161 }
9162 
9163 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9164 			  struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9165 			  struct bpf_prog *old_prog, u32 flags)
9166 {
9167 	unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9168 	struct bpf_prog *cur_prog;
9169 	struct net_device *upper;
9170 	struct list_head *iter;
9171 	enum bpf_xdp_mode mode;
9172 	bpf_op_t bpf_op;
9173 	int err;
9174 
9175 	ASSERT_RTNL();
9176 
9177 	/* either link or prog attachment, never both */
9178 	if (link && (new_prog || old_prog))
9179 		return -EINVAL;
9180 	/* link supports only XDP mode flags */
9181 	if (link && (flags & ~XDP_FLAGS_MODES)) {
9182 		NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9183 		return -EINVAL;
9184 	}
9185 	/* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9186 	if (num_modes > 1) {
9187 		NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9188 		return -EINVAL;
9189 	}
9190 	/* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9191 	if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9192 		NL_SET_ERR_MSG(extack,
9193 			       "More than one program loaded, unset mode is ambiguous");
9194 		return -EINVAL;
9195 	}
9196 	/* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9197 	if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9198 		NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9199 		return -EINVAL;
9200 	}
9201 
9202 	mode = dev_xdp_mode(dev, flags);
9203 	/* can't replace attached link */
9204 	if (dev_xdp_link(dev, mode)) {
9205 		NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9206 		return -EBUSY;
9207 	}
9208 
9209 	/* don't allow if an upper device already has a program */
9210 	netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9211 		if (dev_xdp_prog_count(upper) > 0) {
9212 			NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9213 			return -EEXIST;
9214 		}
9215 	}
9216 
9217 	cur_prog = dev_xdp_prog(dev, mode);
9218 	/* can't replace attached prog with link */
9219 	if (link && cur_prog) {
9220 		NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9221 		return -EBUSY;
9222 	}
9223 	if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9224 		NL_SET_ERR_MSG(extack, "Active program does not match expected");
9225 		return -EEXIST;
9226 	}
9227 
9228 	/* put effective new program into new_prog */
9229 	if (link)
9230 		new_prog = link->link.prog;
9231 
9232 	if (new_prog) {
9233 		bool offload = mode == XDP_MODE_HW;
9234 		enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9235 					       ? XDP_MODE_DRV : XDP_MODE_SKB;
9236 
9237 		if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9238 			NL_SET_ERR_MSG(extack, "XDP program already attached");
9239 			return -EBUSY;
9240 		}
9241 		if (!offload && dev_xdp_prog(dev, other_mode)) {
9242 			NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9243 			return -EEXIST;
9244 		}
9245 		if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9246 			NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9247 			return -EINVAL;
9248 		}
9249 		if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9250 			NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9251 			return -EINVAL;
9252 		}
9253 		if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9254 			NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9255 			return -EINVAL;
9256 		}
9257 	}
9258 
9259 	/* don't call drivers if the effective program didn't change */
9260 	if (new_prog != cur_prog) {
9261 		bpf_op = dev_xdp_bpf_op(dev, mode);
9262 		if (!bpf_op) {
9263 			NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9264 			return -EOPNOTSUPP;
9265 		}
9266 
9267 		err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9268 		if (err)
9269 			return err;
9270 	}
9271 
9272 	if (link)
9273 		dev_xdp_set_link(dev, mode, link);
9274 	else
9275 		dev_xdp_set_prog(dev, mode, new_prog);
9276 	if (cur_prog)
9277 		bpf_prog_put(cur_prog);
9278 
9279 	return 0;
9280 }
9281 
9282 static int dev_xdp_attach_link(struct net_device *dev,
9283 			       struct netlink_ext_ack *extack,
9284 			       struct bpf_xdp_link *link)
9285 {
9286 	return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9287 }
9288 
9289 static int dev_xdp_detach_link(struct net_device *dev,
9290 			       struct netlink_ext_ack *extack,
9291 			       struct bpf_xdp_link *link)
9292 {
9293 	enum bpf_xdp_mode mode;
9294 	bpf_op_t bpf_op;
9295 
9296 	ASSERT_RTNL();
9297 
9298 	mode = dev_xdp_mode(dev, link->flags);
9299 	if (dev_xdp_link(dev, mode) != link)
9300 		return -EINVAL;
9301 
9302 	bpf_op = dev_xdp_bpf_op(dev, mode);
9303 	WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9304 	dev_xdp_set_link(dev, mode, NULL);
9305 	return 0;
9306 }
9307 
9308 static void bpf_xdp_link_release(struct bpf_link *link)
9309 {
9310 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9311 
9312 	rtnl_lock();
9313 
9314 	/* if racing with net_device's tear down, xdp_link->dev might be
9315 	 * already NULL, in which case link was already auto-detached
9316 	 */
9317 	if (xdp_link->dev) {
9318 		WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9319 		xdp_link->dev = NULL;
9320 	}
9321 
9322 	rtnl_unlock();
9323 }
9324 
9325 static int bpf_xdp_link_detach(struct bpf_link *link)
9326 {
9327 	bpf_xdp_link_release(link);
9328 	return 0;
9329 }
9330 
9331 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9332 {
9333 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9334 
9335 	kfree(xdp_link);
9336 }
9337 
9338 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9339 				     struct seq_file *seq)
9340 {
9341 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9342 	u32 ifindex = 0;
9343 
9344 	rtnl_lock();
9345 	if (xdp_link->dev)
9346 		ifindex = xdp_link->dev->ifindex;
9347 	rtnl_unlock();
9348 
9349 	seq_printf(seq, "ifindex:\t%u\n", ifindex);
9350 }
9351 
9352 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9353 				       struct bpf_link_info *info)
9354 {
9355 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9356 	u32 ifindex = 0;
9357 
9358 	rtnl_lock();
9359 	if (xdp_link->dev)
9360 		ifindex = xdp_link->dev->ifindex;
9361 	rtnl_unlock();
9362 
9363 	info->xdp.ifindex = ifindex;
9364 	return 0;
9365 }
9366 
9367 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9368 			       struct bpf_prog *old_prog)
9369 {
9370 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9371 	enum bpf_xdp_mode mode;
9372 	bpf_op_t bpf_op;
9373 	int err = 0;
9374 
9375 	rtnl_lock();
9376 
9377 	/* link might have been auto-released already, so fail */
9378 	if (!xdp_link->dev) {
9379 		err = -ENOLINK;
9380 		goto out_unlock;
9381 	}
9382 
9383 	if (old_prog && link->prog != old_prog) {
9384 		err = -EPERM;
9385 		goto out_unlock;
9386 	}
9387 	old_prog = link->prog;
9388 	if (old_prog->type != new_prog->type ||
9389 	    old_prog->expected_attach_type != new_prog->expected_attach_type) {
9390 		err = -EINVAL;
9391 		goto out_unlock;
9392 	}
9393 
9394 	if (old_prog == new_prog) {
9395 		/* no-op, don't disturb drivers */
9396 		bpf_prog_put(new_prog);
9397 		goto out_unlock;
9398 	}
9399 
9400 	mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9401 	bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9402 	err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9403 			      xdp_link->flags, new_prog);
9404 	if (err)
9405 		goto out_unlock;
9406 
9407 	old_prog = xchg(&link->prog, new_prog);
9408 	bpf_prog_put(old_prog);
9409 
9410 out_unlock:
9411 	rtnl_unlock();
9412 	return err;
9413 }
9414 
9415 static const struct bpf_link_ops bpf_xdp_link_lops = {
9416 	.release = bpf_xdp_link_release,
9417 	.dealloc = bpf_xdp_link_dealloc,
9418 	.detach = bpf_xdp_link_detach,
9419 	.show_fdinfo = bpf_xdp_link_show_fdinfo,
9420 	.fill_link_info = bpf_xdp_link_fill_link_info,
9421 	.update_prog = bpf_xdp_link_update,
9422 };
9423 
9424 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9425 {
9426 	struct net *net = current->nsproxy->net_ns;
9427 	struct bpf_link_primer link_primer;
9428 	struct bpf_xdp_link *link;
9429 	struct net_device *dev;
9430 	int err, fd;
9431 
9432 	rtnl_lock();
9433 	dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9434 	if (!dev) {
9435 		rtnl_unlock();
9436 		return -EINVAL;
9437 	}
9438 
9439 	link = kzalloc(sizeof(*link), GFP_USER);
9440 	if (!link) {
9441 		err = -ENOMEM;
9442 		goto unlock;
9443 	}
9444 
9445 	bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9446 	link->dev = dev;
9447 	link->flags = attr->link_create.flags;
9448 
9449 	err = bpf_link_prime(&link->link, &link_primer);
9450 	if (err) {
9451 		kfree(link);
9452 		goto unlock;
9453 	}
9454 
9455 	err = dev_xdp_attach_link(dev, NULL, link);
9456 	rtnl_unlock();
9457 
9458 	if (err) {
9459 		link->dev = NULL;
9460 		bpf_link_cleanup(&link_primer);
9461 		goto out_put_dev;
9462 	}
9463 
9464 	fd = bpf_link_settle(&link_primer);
9465 	/* link itself doesn't hold dev's refcnt to not complicate shutdown */
9466 	dev_put(dev);
9467 	return fd;
9468 
9469 unlock:
9470 	rtnl_unlock();
9471 
9472 out_put_dev:
9473 	dev_put(dev);
9474 	return err;
9475 }
9476 
9477 /**
9478  *	dev_change_xdp_fd - set or clear a bpf program for a device rx path
9479  *	@dev: device
9480  *	@extack: netlink extended ack
9481  *	@fd: new program fd or negative value to clear
9482  *	@expected_fd: old program fd that userspace expects to replace or clear
9483  *	@flags: xdp-related flags
9484  *
9485  *	Set or clear a bpf program for a device
9486  */
9487 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9488 		      int fd, int expected_fd, u32 flags)
9489 {
9490 	enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9491 	struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9492 	int err;
9493 
9494 	ASSERT_RTNL();
9495 
9496 	if (fd >= 0) {
9497 		new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9498 						 mode != XDP_MODE_SKB);
9499 		if (IS_ERR(new_prog))
9500 			return PTR_ERR(new_prog);
9501 	}
9502 
9503 	if (expected_fd >= 0) {
9504 		old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9505 						 mode != XDP_MODE_SKB);
9506 		if (IS_ERR(old_prog)) {
9507 			err = PTR_ERR(old_prog);
9508 			old_prog = NULL;
9509 			goto err_out;
9510 		}
9511 	}
9512 
9513 	err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9514 
9515 err_out:
9516 	if (err && new_prog)
9517 		bpf_prog_put(new_prog);
9518 	if (old_prog)
9519 		bpf_prog_put(old_prog);
9520 	return err;
9521 }
9522 
9523 /**
9524  *	dev_new_index	-	allocate an ifindex
9525  *	@net: the applicable net namespace
9526  *
9527  *	Returns a suitable unique value for a new device interface
9528  *	number.  The caller must hold the rtnl semaphore or the
9529  *	dev_base_lock to be sure it remains unique.
9530  */
9531 static int dev_new_index(struct net *net)
9532 {
9533 	int ifindex = net->ifindex;
9534 
9535 	for (;;) {
9536 		if (++ifindex <= 0)
9537 			ifindex = 1;
9538 		if (!__dev_get_by_index(net, ifindex))
9539 			return net->ifindex = ifindex;
9540 	}
9541 }
9542 
9543 /* Delayed registration/unregisteration */
9544 LIST_HEAD(net_todo_list);
9545 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9546 
9547 static void net_set_todo(struct net_device *dev)
9548 {
9549 	list_add_tail(&dev->todo_list, &net_todo_list);
9550 	atomic_inc(&dev_net(dev)->dev_unreg_count);
9551 }
9552 
9553 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9554 	struct net_device *upper, netdev_features_t features)
9555 {
9556 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9557 	netdev_features_t feature;
9558 	int feature_bit;
9559 
9560 	for_each_netdev_feature(upper_disables, feature_bit) {
9561 		feature = __NETIF_F_BIT(feature_bit);
9562 		if (!(upper->wanted_features & feature)
9563 		    && (features & feature)) {
9564 			netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9565 				   &feature, upper->name);
9566 			features &= ~feature;
9567 		}
9568 	}
9569 
9570 	return features;
9571 }
9572 
9573 static void netdev_sync_lower_features(struct net_device *upper,
9574 	struct net_device *lower, netdev_features_t features)
9575 {
9576 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9577 	netdev_features_t feature;
9578 	int feature_bit;
9579 
9580 	for_each_netdev_feature(upper_disables, feature_bit) {
9581 		feature = __NETIF_F_BIT(feature_bit);
9582 		if (!(features & feature) && (lower->features & feature)) {
9583 			netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9584 				   &feature, lower->name);
9585 			lower->wanted_features &= ~feature;
9586 			__netdev_update_features(lower);
9587 
9588 			if (unlikely(lower->features & feature))
9589 				netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9590 					    &feature, lower->name);
9591 			else
9592 				netdev_features_change(lower);
9593 		}
9594 	}
9595 }
9596 
9597 static netdev_features_t netdev_fix_features(struct net_device *dev,
9598 	netdev_features_t features)
9599 {
9600 	/* Fix illegal checksum combinations */
9601 	if ((features & NETIF_F_HW_CSUM) &&
9602 	    (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9603 		netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9604 		features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9605 	}
9606 
9607 	/* TSO requires that SG is present as well. */
9608 	if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9609 		netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9610 		features &= ~NETIF_F_ALL_TSO;
9611 	}
9612 
9613 	if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9614 					!(features & NETIF_F_IP_CSUM)) {
9615 		netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9616 		features &= ~NETIF_F_TSO;
9617 		features &= ~NETIF_F_TSO_ECN;
9618 	}
9619 
9620 	if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9621 					 !(features & NETIF_F_IPV6_CSUM)) {
9622 		netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9623 		features &= ~NETIF_F_TSO6;
9624 	}
9625 
9626 	/* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9627 	if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9628 		features &= ~NETIF_F_TSO_MANGLEID;
9629 
9630 	/* TSO ECN requires that TSO is present as well. */
9631 	if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9632 		features &= ~NETIF_F_TSO_ECN;
9633 
9634 	/* Software GSO depends on SG. */
9635 	if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9636 		netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9637 		features &= ~NETIF_F_GSO;
9638 	}
9639 
9640 	/* GSO partial features require GSO partial be set */
9641 	if ((features & dev->gso_partial_features) &&
9642 	    !(features & NETIF_F_GSO_PARTIAL)) {
9643 		netdev_dbg(dev,
9644 			   "Dropping partially supported GSO features since no GSO partial.\n");
9645 		features &= ~dev->gso_partial_features;
9646 	}
9647 
9648 	if (!(features & NETIF_F_RXCSUM)) {
9649 		/* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9650 		 * successfully merged by hardware must also have the
9651 		 * checksum verified by hardware.  If the user does not
9652 		 * want to enable RXCSUM, logically, we should disable GRO_HW.
9653 		 */
9654 		if (features & NETIF_F_GRO_HW) {
9655 			netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9656 			features &= ~NETIF_F_GRO_HW;
9657 		}
9658 	}
9659 
9660 	/* LRO/HW-GRO features cannot be combined with RX-FCS */
9661 	if (features & NETIF_F_RXFCS) {
9662 		if (features & NETIF_F_LRO) {
9663 			netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9664 			features &= ~NETIF_F_LRO;
9665 		}
9666 
9667 		if (features & NETIF_F_GRO_HW) {
9668 			netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9669 			features &= ~NETIF_F_GRO_HW;
9670 		}
9671 	}
9672 
9673 	if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9674 		netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9675 		features &= ~NETIF_F_LRO;
9676 	}
9677 
9678 	if (features & NETIF_F_HW_TLS_TX) {
9679 		bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9680 			(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9681 		bool hw_csum = features & NETIF_F_HW_CSUM;
9682 
9683 		if (!ip_csum && !hw_csum) {
9684 			netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9685 			features &= ~NETIF_F_HW_TLS_TX;
9686 		}
9687 	}
9688 
9689 	if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9690 		netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9691 		features &= ~NETIF_F_HW_TLS_RX;
9692 	}
9693 
9694 	return features;
9695 }
9696 
9697 int __netdev_update_features(struct net_device *dev)
9698 {
9699 	struct net_device *upper, *lower;
9700 	netdev_features_t features;
9701 	struct list_head *iter;
9702 	int err = -1;
9703 
9704 	ASSERT_RTNL();
9705 
9706 	features = netdev_get_wanted_features(dev);
9707 
9708 	if (dev->netdev_ops->ndo_fix_features)
9709 		features = dev->netdev_ops->ndo_fix_features(dev, features);
9710 
9711 	/* driver might be less strict about feature dependencies */
9712 	features = netdev_fix_features(dev, features);
9713 
9714 	/* some features can't be enabled if they're off on an upper device */
9715 	netdev_for_each_upper_dev_rcu(dev, upper, iter)
9716 		features = netdev_sync_upper_features(dev, upper, features);
9717 
9718 	if (dev->features == features)
9719 		goto sync_lower;
9720 
9721 	netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9722 		&dev->features, &features);
9723 
9724 	if (dev->netdev_ops->ndo_set_features)
9725 		err = dev->netdev_ops->ndo_set_features(dev, features);
9726 	else
9727 		err = 0;
9728 
9729 	if (unlikely(err < 0)) {
9730 		netdev_err(dev,
9731 			"set_features() failed (%d); wanted %pNF, left %pNF\n",
9732 			err, &features, &dev->features);
9733 		/* return non-0 since some features might have changed and
9734 		 * it's better to fire a spurious notification than miss it
9735 		 */
9736 		return -1;
9737 	}
9738 
9739 sync_lower:
9740 	/* some features must be disabled on lower devices when disabled
9741 	 * on an upper device (think: bonding master or bridge)
9742 	 */
9743 	netdev_for_each_lower_dev(dev, lower, iter)
9744 		netdev_sync_lower_features(dev, lower, features);
9745 
9746 	if (!err) {
9747 		netdev_features_t diff = features ^ dev->features;
9748 
9749 		if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9750 			/* udp_tunnel_{get,drop}_rx_info both need
9751 			 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9752 			 * device, or they won't do anything.
9753 			 * Thus we need to update dev->features
9754 			 * *before* calling udp_tunnel_get_rx_info,
9755 			 * but *after* calling udp_tunnel_drop_rx_info.
9756 			 */
9757 			if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9758 				dev->features = features;
9759 				udp_tunnel_get_rx_info(dev);
9760 			} else {
9761 				udp_tunnel_drop_rx_info(dev);
9762 			}
9763 		}
9764 
9765 		if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9766 			if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9767 				dev->features = features;
9768 				err |= vlan_get_rx_ctag_filter_info(dev);
9769 			} else {
9770 				vlan_drop_rx_ctag_filter_info(dev);
9771 			}
9772 		}
9773 
9774 		if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9775 			if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9776 				dev->features = features;
9777 				err |= vlan_get_rx_stag_filter_info(dev);
9778 			} else {
9779 				vlan_drop_rx_stag_filter_info(dev);
9780 			}
9781 		}
9782 
9783 		dev->features = features;
9784 	}
9785 
9786 	return err < 0 ? 0 : 1;
9787 }
9788 
9789 /**
9790  *	netdev_update_features - recalculate device features
9791  *	@dev: the device to check
9792  *
9793  *	Recalculate dev->features set and send notifications if it
9794  *	has changed. Should be called after driver or hardware dependent
9795  *	conditions might have changed that influence the features.
9796  */
9797 void netdev_update_features(struct net_device *dev)
9798 {
9799 	if (__netdev_update_features(dev))
9800 		netdev_features_change(dev);
9801 }
9802 EXPORT_SYMBOL(netdev_update_features);
9803 
9804 /**
9805  *	netdev_change_features - recalculate device features
9806  *	@dev: the device to check
9807  *
9808  *	Recalculate dev->features set and send notifications even
9809  *	if they have not changed. Should be called instead of
9810  *	netdev_update_features() if also dev->vlan_features might
9811  *	have changed to allow the changes to be propagated to stacked
9812  *	VLAN devices.
9813  */
9814 void netdev_change_features(struct net_device *dev)
9815 {
9816 	__netdev_update_features(dev);
9817 	netdev_features_change(dev);
9818 }
9819 EXPORT_SYMBOL(netdev_change_features);
9820 
9821 /**
9822  *	netif_stacked_transfer_operstate -	transfer operstate
9823  *	@rootdev: the root or lower level device to transfer state from
9824  *	@dev: the device to transfer operstate to
9825  *
9826  *	Transfer operational state from root to device. This is normally
9827  *	called when a stacking relationship exists between the root
9828  *	device and the device(a leaf device).
9829  */
9830 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9831 					struct net_device *dev)
9832 {
9833 	if (rootdev->operstate == IF_OPER_DORMANT)
9834 		netif_dormant_on(dev);
9835 	else
9836 		netif_dormant_off(dev);
9837 
9838 	if (rootdev->operstate == IF_OPER_TESTING)
9839 		netif_testing_on(dev);
9840 	else
9841 		netif_testing_off(dev);
9842 
9843 	if (netif_carrier_ok(rootdev))
9844 		netif_carrier_on(dev);
9845 	else
9846 		netif_carrier_off(dev);
9847 }
9848 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9849 
9850 static int netif_alloc_rx_queues(struct net_device *dev)
9851 {
9852 	unsigned int i, count = dev->num_rx_queues;
9853 	struct netdev_rx_queue *rx;
9854 	size_t sz = count * sizeof(*rx);
9855 	int err = 0;
9856 
9857 	BUG_ON(count < 1);
9858 
9859 	rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9860 	if (!rx)
9861 		return -ENOMEM;
9862 
9863 	dev->_rx = rx;
9864 
9865 	for (i = 0; i < count; i++) {
9866 		rx[i].dev = dev;
9867 
9868 		/* XDP RX-queue setup */
9869 		err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
9870 		if (err < 0)
9871 			goto err_rxq_info;
9872 	}
9873 	return 0;
9874 
9875 err_rxq_info:
9876 	/* Rollback successful reg's and free other resources */
9877 	while (i--)
9878 		xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9879 	kvfree(dev->_rx);
9880 	dev->_rx = NULL;
9881 	return err;
9882 }
9883 
9884 static void netif_free_rx_queues(struct net_device *dev)
9885 {
9886 	unsigned int i, count = dev->num_rx_queues;
9887 
9888 	/* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9889 	if (!dev->_rx)
9890 		return;
9891 
9892 	for (i = 0; i < count; i++)
9893 		xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9894 
9895 	kvfree(dev->_rx);
9896 }
9897 
9898 static void netdev_init_one_queue(struct net_device *dev,
9899 				  struct netdev_queue *queue, void *_unused)
9900 {
9901 	/* Initialize queue lock */
9902 	spin_lock_init(&queue->_xmit_lock);
9903 	netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9904 	queue->xmit_lock_owner = -1;
9905 	netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9906 	queue->dev = dev;
9907 #ifdef CONFIG_BQL
9908 	dql_init(&queue->dql, HZ);
9909 #endif
9910 }
9911 
9912 static void netif_free_tx_queues(struct net_device *dev)
9913 {
9914 	kvfree(dev->_tx);
9915 }
9916 
9917 static int netif_alloc_netdev_queues(struct net_device *dev)
9918 {
9919 	unsigned int count = dev->num_tx_queues;
9920 	struct netdev_queue *tx;
9921 	size_t sz = count * sizeof(*tx);
9922 
9923 	if (count < 1 || count > 0xffff)
9924 		return -EINVAL;
9925 
9926 	tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9927 	if (!tx)
9928 		return -ENOMEM;
9929 
9930 	dev->_tx = tx;
9931 
9932 	netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9933 	spin_lock_init(&dev->tx_global_lock);
9934 
9935 	return 0;
9936 }
9937 
9938 void netif_tx_stop_all_queues(struct net_device *dev)
9939 {
9940 	unsigned int i;
9941 
9942 	for (i = 0; i < dev->num_tx_queues; i++) {
9943 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9944 
9945 		netif_tx_stop_queue(txq);
9946 	}
9947 }
9948 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9949 
9950 /**
9951  * register_netdevice() - register a network device
9952  * @dev: device to register
9953  *
9954  * Take a prepared network device structure and make it externally accessible.
9955  * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
9956  * Callers must hold the rtnl lock - you may want register_netdev()
9957  * instead of this.
9958  */
9959 int register_netdevice(struct net_device *dev)
9960 {
9961 	int ret;
9962 	struct net *net = dev_net(dev);
9963 
9964 	BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9965 		     NETDEV_FEATURE_COUNT);
9966 	BUG_ON(dev_boot_phase);
9967 	ASSERT_RTNL();
9968 
9969 	might_sleep();
9970 
9971 	/* When net_device's are persistent, this will be fatal. */
9972 	BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9973 	BUG_ON(!net);
9974 
9975 	ret = ethtool_check_ops(dev->ethtool_ops);
9976 	if (ret)
9977 		return ret;
9978 
9979 	spin_lock_init(&dev->addr_list_lock);
9980 	netdev_set_addr_lockdep_class(dev);
9981 
9982 	ret = dev_get_valid_name(net, dev, dev->name);
9983 	if (ret < 0)
9984 		goto out;
9985 
9986 	ret = -ENOMEM;
9987 	dev->name_node = netdev_name_node_head_alloc(dev);
9988 	if (!dev->name_node)
9989 		goto out;
9990 
9991 	/* Init, if this function is available */
9992 	if (dev->netdev_ops->ndo_init) {
9993 		ret = dev->netdev_ops->ndo_init(dev);
9994 		if (ret) {
9995 			if (ret > 0)
9996 				ret = -EIO;
9997 			goto err_free_name;
9998 		}
9999 	}
10000 
10001 	if (((dev->hw_features | dev->features) &
10002 	     NETIF_F_HW_VLAN_CTAG_FILTER) &&
10003 	    (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10004 	     !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10005 		netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10006 		ret = -EINVAL;
10007 		goto err_uninit;
10008 	}
10009 
10010 	ret = -EBUSY;
10011 	if (!dev->ifindex)
10012 		dev->ifindex = dev_new_index(net);
10013 	else if (__dev_get_by_index(net, dev->ifindex))
10014 		goto err_uninit;
10015 
10016 	/* Transfer changeable features to wanted_features and enable
10017 	 * software offloads (GSO and GRO).
10018 	 */
10019 	dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10020 	dev->features |= NETIF_F_SOFT_FEATURES;
10021 
10022 	if (dev->udp_tunnel_nic_info) {
10023 		dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10024 		dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10025 	}
10026 
10027 	dev->wanted_features = dev->features & dev->hw_features;
10028 
10029 	if (!(dev->flags & IFF_LOOPBACK))
10030 		dev->hw_features |= NETIF_F_NOCACHE_COPY;
10031 
10032 	/* If IPv4 TCP segmentation offload is supported we should also
10033 	 * allow the device to enable segmenting the frame with the option
10034 	 * of ignoring a static IP ID value.  This doesn't enable the
10035 	 * feature itself but allows the user to enable it later.
10036 	 */
10037 	if (dev->hw_features & NETIF_F_TSO)
10038 		dev->hw_features |= NETIF_F_TSO_MANGLEID;
10039 	if (dev->vlan_features & NETIF_F_TSO)
10040 		dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10041 	if (dev->mpls_features & NETIF_F_TSO)
10042 		dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10043 	if (dev->hw_enc_features & NETIF_F_TSO)
10044 		dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10045 
10046 	/* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10047 	 */
10048 	dev->vlan_features |= NETIF_F_HIGHDMA;
10049 
10050 	/* Make NETIF_F_SG inheritable to tunnel devices.
10051 	 */
10052 	dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10053 
10054 	/* Make NETIF_F_SG inheritable to MPLS.
10055 	 */
10056 	dev->mpls_features |= NETIF_F_SG;
10057 
10058 	ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10059 	ret = notifier_to_errno(ret);
10060 	if (ret)
10061 		goto err_uninit;
10062 
10063 	ret = netdev_register_kobject(dev);
10064 	if (ret) {
10065 		dev->reg_state = NETREG_UNREGISTERED;
10066 		goto err_uninit;
10067 	}
10068 	dev->reg_state = NETREG_REGISTERED;
10069 
10070 	__netdev_update_features(dev);
10071 
10072 	/*
10073 	 *	Default initial state at registry is that the
10074 	 *	device is present.
10075 	 */
10076 
10077 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10078 
10079 	linkwatch_init_dev(dev);
10080 
10081 	dev_init_scheduler(dev);
10082 
10083 	netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
10084 	list_netdevice(dev);
10085 
10086 	add_device_randomness(dev->dev_addr, dev->addr_len);
10087 
10088 	/* If the device has permanent device address, driver should
10089 	 * set dev_addr and also addr_assign_type should be set to
10090 	 * NET_ADDR_PERM (default value).
10091 	 */
10092 	if (dev->addr_assign_type == NET_ADDR_PERM)
10093 		memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10094 
10095 	/* Notify protocols, that a new device appeared. */
10096 	ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10097 	ret = notifier_to_errno(ret);
10098 	if (ret) {
10099 		/* Expect explicit free_netdev() on failure */
10100 		dev->needs_free_netdev = false;
10101 		unregister_netdevice_queue(dev, NULL);
10102 		goto out;
10103 	}
10104 	/*
10105 	 *	Prevent userspace races by waiting until the network
10106 	 *	device is fully setup before sending notifications.
10107 	 */
10108 	if (!dev->rtnl_link_ops ||
10109 	    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10110 		rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10111 
10112 out:
10113 	return ret;
10114 
10115 err_uninit:
10116 	if (dev->netdev_ops->ndo_uninit)
10117 		dev->netdev_ops->ndo_uninit(dev);
10118 	if (dev->priv_destructor)
10119 		dev->priv_destructor(dev);
10120 err_free_name:
10121 	netdev_name_node_free(dev->name_node);
10122 	goto out;
10123 }
10124 EXPORT_SYMBOL(register_netdevice);
10125 
10126 /**
10127  *	init_dummy_netdev	- init a dummy network device for NAPI
10128  *	@dev: device to init
10129  *
10130  *	This takes a network device structure and initialize the minimum
10131  *	amount of fields so it can be used to schedule NAPI polls without
10132  *	registering a full blown interface. This is to be used by drivers
10133  *	that need to tie several hardware interfaces to a single NAPI
10134  *	poll scheduler due to HW limitations.
10135  */
10136 int init_dummy_netdev(struct net_device *dev)
10137 {
10138 	/* Clear everything. Note we don't initialize spinlocks
10139 	 * are they aren't supposed to be taken by any of the
10140 	 * NAPI code and this dummy netdev is supposed to be
10141 	 * only ever used for NAPI polls
10142 	 */
10143 	memset(dev, 0, sizeof(struct net_device));
10144 
10145 	/* make sure we BUG if trying to hit standard
10146 	 * register/unregister code path
10147 	 */
10148 	dev->reg_state = NETREG_DUMMY;
10149 
10150 	/* NAPI wants this */
10151 	INIT_LIST_HEAD(&dev->napi_list);
10152 
10153 	/* a dummy interface is started by default */
10154 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10155 	set_bit(__LINK_STATE_START, &dev->state);
10156 
10157 	/* napi_busy_loop stats accounting wants this */
10158 	dev_net_set(dev, &init_net);
10159 
10160 	/* Note : We dont allocate pcpu_refcnt for dummy devices,
10161 	 * because users of this 'device' dont need to change
10162 	 * its refcount.
10163 	 */
10164 
10165 	return 0;
10166 }
10167 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10168 
10169 
10170 /**
10171  *	register_netdev	- register a network device
10172  *	@dev: device to register
10173  *
10174  *	Take a completed network device structure and add it to the kernel
10175  *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10176  *	chain. 0 is returned on success. A negative errno code is returned
10177  *	on a failure to set up the device, or if the name is a duplicate.
10178  *
10179  *	This is a wrapper around register_netdevice that takes the rtnl semaphore
10180  *	and expands the device name if you passed a format string to
10181  *	alloc_netdev.
10182  */
10183 int register_netdev(struct net_device *dev)
10184 {
10185 	int err;
10186 
10187 	if (rtnl_lock_killable())
10188 		return -EINTR;
10189 	err = register_netdevice(dev);
10190 	rtnl_unlock();
10191 	return err;
10192 }
10193 EXPORT_SYMBOL(register_netdev);
10194 
10195 int netdev_refcnt_read(const struct net_device *dev)
10196 {
10197 #ifdef CONFIG_PCPU_DEV_REFCNT
10198 	int i, refcnt = 0;
10199 
10200 	for_each_possible_cpu(i)
10201 		refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10202 	return refcnt;
10203 #else
10204 	return refcount_read(&dev->dev_refcnt);
10205 #endif
10206 }
10207 EXPORT_SYMBOL(netdev_refcnt_read);
10208 
10209 int netdev_unregister_timeout_secs __read_mostly = 10;
10210 
10211 #define WAIT_REFS_MIN_MSECS 1
10212 #define WAIT_REFS_MAX_MSECS 250
10213 /**
10214  * netdev_wait_allrefs_any - wait until all references are gone.
10215  * @list: list of net_devices to wait on
10216  *
10217  * This is called when unregistering network devices.
10218  *
10219  * Any protocol or device that holds a reference should register
10220  * for netdevice notification, and cleanup and put back the
10221  * reference if they receive an UNREGISTER event.
10222  * We can get stuck here if buggy protocols don't correctly
10223  * call dev_put.
10224  */
10225 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
10226 {
10227 	unsigned long rebroadcast_time, warning_time;
10228 	struct net_device *dev;
10229 	int wait = 0;
10230 
10231 	rebroadcast_time = warning_time = jiffies;
10232 
10233 	list_for_each_entry(dev, list, todo_list)
10234 		if (netdev_refcnt_read(dev) == 1)
10235 			return dev;
10236 
10237 	while (true) {
10238 		if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10239 			rtnl_lock();
10240 
10241 			/* Rebroadcast unregister notification */
10242 			list_for_each_entry(dev, list, todo_list)
10243 				call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10244 
10245 			__rtnl_unlock();
10246 			rcu_barrier();
10247 			rtnl_lock();
10248 
10249 			list_for_each_entry(dev, list, todo_list)
10250 				if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10251 					     &dev->state)) {
10252 					/* We must not have linkwatch events
10253 					 * pending on unregister. If this
10254 					 * happens, we simply run the queue
10255 					 * unscheduled, resulting in a noop
10256 					 * for this device.
10257 					 */
10258 					linkwatch_run_queue();
10259 					break;
10260 				}
10261 
10262 			__rtnl_unlock();
10263 
10264 			rebroadcast_time = jiffies;
10265 		}
10266 
10267 		if (!wait) {
10268 			rcu_barrier();
10269 			wait = WAIT_REFS_MIN_MSECS;
10270 		} else {
10271 			msleep(wait);
10272 			wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10273 		}
10274 
10275 		list_for_each_entry(dev, list, todo_list)
10276 			if (netdev_refcnt_read(dev) == 1)
10277 				return dev;
10278 
10279 		if (time_after(jiffies, warning_time +
10280 			       netdev_unregister_timeout_secs * HZ)) {
10281 			list_for_each_entry(dev, list, todo_list) {
10282 				pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10283 					 dev->name, netdev_refcnt_read(dev));
10284 				ref_tracker_dir_print(&dev->refcnt_tracker, 10);
10285 			}
10286 
10287 			warning_time = jiffies;
10288 		}
10289 	}
10290 }
10291 
10292 /* The sequence is:
10293  *
10294  *	rtnl_lock();
10295  *	...
10296  *	register_netdevice(x1);
10297  *	register_netdevice(x2);
10298  *	...
10299  *	unregister_netdevice(y1);
10300  *	unregister_netdevice(y2);
10301  *      ...
10302  *	rtnl_unlock();
10303  *	free_netdev(y1);
10304  *	free_netdev(y2);
10305  *
10306  * We are invoked by rtnl_unlock().
10307  * This allows us to deal with problems:
10308  * 1) We can delete sysfs objects which invoke hotplug
10309  *    without deadlocking with linkwatch via keventd.
10310  * 2) Since we run with the RTNL semaphore not held, we can sleep
10311  *    safely in order to wait for the netdev refcnt to drop to zero.
10312  *
10313  * We must not return until all unregister events added during
10314  * the interval the lock was held have been completed.
10315  */
10316 void netdev_run_todo(void)
10317 {
10318 	struct net_device *dev, *tmp;
10319 	struct list_head list;
10320 #ifdef CONFIG_LOCKDEP
10321 	struct list_head unlink_list;
10322 
10323 	list_replace_init(&net_unlink_list, &unlink_list);
10324 
10325 	while (!list_empty(&unlink_list)) {
10326 		struct net_device *dev = list_first_entry(&unlink_list,
10327 							  struct net_device,
10328 							  unlink_list);
10329 		list_del_init(&dev->unlink_list);
10330 		dev->nested_level = dev->lower_level - 1;
10331 	}
10332 #endif
10333 
10334 	/* Snapshot list, allow later requests */
10335 	list_replace_init(&net_todo_list, &list);
10336 
10337 	__rtnl_unlock();
10338 
10339 	/* Wait for rcu callbacks to finish before next phase */
10340 	if (!list_empty(&list))
10341 		rcu_barrier();
10342 
10343 	list_for_each_entry_safe(dev, tmp, &list, todo_list) {
10344 		if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10345 			netdev_WARN(dev, "run_todo but not unregistering\n");
10346 			list_del(&dev->todo_list);
10347 			continue;
10348 		}
10349 
10350 		dev->reg_state = NETREG_UNREGISTERED;
10351 		linkwatch_forget_dev(dev);
10352 	}
10353 
10354 	while (!list_empty(&list)) {
10355 		dev = netdev_wait_allrefs_any(&list);
10356 		list_del(&dev->todo_list);
10357 
10358 		/* paranoia */
10359 		BUG_ON(netdev_refcnt_read(dev) != 1);
10360 		BUG_ON(!list_empty(&dev->ptype_all));
10361 		BUG_ON(!list_empty(&dev->ptype_specific));
10362 		WARN_ON(rcu_access_pointer(dev->ip_ptr));
10363 		WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10364 #if IS_ENABLED(CONFIG_DECNET)
10365 		WARN_ON(dev->dn_ptr);
10366 #endif
10367 		if (dev->priv_destructor)
10368 			dev->priv_destructor(dev);
10369 		if (dev->needs_free_netdev)
10370 			free_netdev(dev);
10371 
10372 		if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count))
10373 			wake_up(&netdev_unregistering_wq);
10374 
10375 		/* Free network device */
10376 		kobject_put(&dev->dev.kobj);
10377 	}
10378 }
10379 
10380 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10381  * all the same fields in the same order as net_device_stats, with only
10382  * the type differing, but rtnl_link_stats64 may have additional fields
10383  * at the end for newer counters.
10384  */
10385 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10386 			     const struct net_device_stats *netdev_stats)
10387 {
10388 #if BITS_PER_LONG == 64
10389 	BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
10390 	memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
10391 	/* zero out counters that only exist in rtnl_link_stats64 */
10392 	memset((char *)stats64 + sizeof(*netdev_stats), 0,
10393 	       sizeof(*stats64) - sizeof(*netdev_stats));
10394 #else
10395 	size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
10396 	const unsigned long *src = (const unsigned long *)netdev_stats;
10397 	u64 *dst = (u64 *)stats64;
10398 
10399 	BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10400 	for (i = 0; i < n; i++)
10401 		dst[i] = src[i];
10402 	/* zero out counters that only exist in rtnl_link_stats64 */
10403 	memset((char *)stats64 + n * sizeof(u64), 0,
10404 	       sizeof(*stats64) - n * sizeof(u64));
10405 #endif
10406 }
10407 EXPORT_SYMBOL(netdev_stats_to_stats64);
10408 
10409 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev)
10410 {
10411 	struct net_device_core_stats __percpu *p;
10412 
10413 	p = alloc_percpu_gfp(struct net_device_core_stats,
10414 			     GFP_ATOMIC | __GFP_NOWARN);
10415 
10416 	if (p && cmpxchg(&dev->core_stats, NULL, p))
10417 		free_percpu(p);
10418 
10419 	/* This READ_ONCE() pairs with the cmpxchg() above */
10420 	return READ_ONCE(dev->core_stats);
10421 }
10422 EXPORT_SYMBOL(netdev_core_stats_alloc);
10423 
10424 /**
10425  *	dev_get_stats	- get network device statistics
10426  *	@dev: device to get statistics from
10427  *	@storage: place to store stats
10428  *
10429  *	Get network statistics from device. Return @storage.
10430  *	The device driver may provide its own method by setting
10431  *	dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10432  *	otherwise the internal statistics structure is used.
10433  */
10434 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10435 					struct rtnl_link_stats64 *storage)
10436 {
10437 	const struct net_device_ops *ops = dev->netdev_ops;
10438 	const struct net_device_core_stats __percpu *p;
10439 
10440 	if (ops->ndo_get_stats64) {
10441 		memset(storage, 0, sizeof(*storage));
10442 		ops->ndo_get_stats64(dev, storage);
10443 	} else if (ops->ndo_get_stats) {
10444 		netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10445 	} else {
10446 		netdev_stats_to_stats64(storage, &dev->stats);
10447 	}
10448 
10449 	/* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10450 	p = READ_ONCE(dev->core_stats);
10451 	if (p) {
10452 		const struct net_device_core_stats *core_stats;
10453 		int i;
10454 
10455 		for_each_possible_cpu(i) {
10456 			core_stats = per_cpu_ptr(p, i);
10457 			storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
10458 			storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
10459 			storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
10460 			storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
10461 		}
10462 	}
10463 	return storage;
10464 }
10465 EXPORT_SYMBOL(dev_get_stats);
10466 
10467 /**
10468  *	dev_fetch_sw_netstats - get per-cpu network device statistics
10469  *	@s: place to store stats
10470  *	@netstats: per-cpu network stats to read from
10471  *
10472  *	Read per-cpu network statistics and populate the related fields in @s.
10473  */
10474 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10475 			   const struct pcpu_sw_netstats __percpu *netstats)
10476 {
10477 	int cpu;
10478 
10479 	for_each_possible_cpu(cpu) {
10480 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
10481 		const struct pcpu_sw_netstats *stats;
10482 		unsigned int start;
10483 
10484 		stats = per_cpu_ptr(netstats, cpu);
10485 		do {
10486 			start = u64_stats_fetch_begin_irq(&stats->syncp);
10487 			rx_packets = u64_stats_read(&stats->rx_packets);
10488 			rx_bytes   = u64_stats_read(&stats->rx_bytes);
10489 			tx_packets = u64_stats_read(&stats->tx_packets);
10490 			tx_bytes   = u64_stats_read(&stats->tx_bytes);
10491 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
10492 
10493 		s->rx_packets += rx_packets;
10494 		s->rx_bytes   += rx_bytes;
10495 		s->tx_packets += tx_packets;
10496 		s->tx_bytes   += tx_bytes;
10497 	}
10498 }
10499 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10500 
10501 /**
10502  *	dev_get_tstats64 - ndo_get_stats64 implementation
10503  *	@dev: device to get statistics from
10504  *	@s: place to store stats
10505  *
10506  *	Populate @s from dev->stats and dev->tstats. Can be used as
10507  *	ndo_get_stats64() callback.
10508  */
10509 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10510 {
10511 	netdev_stats_to_stats64(s, &dev->stats);
10512 	dev_fetch_sw_netstats(s, dev->tstats);
10513 }
10514 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10515 
10516 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10517 {
10518 	struct netdev_queue *queue = dev_ingress_queue(dev);
10519 
10520 #ifdef CONFIG_NET_CLS_ACT
10521 	if (queue)
10522 		return queue;
10523 	queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10524 	if (!queue)
10525 		return NULL;
10526 	netdev_init_one_queue(dev, queue, NULL);
10527 	RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10528 	queue->qdisc_sleeping = &noop_qdisc;
10529 	rcu_assign_pointer(dev->ingress_queue, queue);
10530 #endif
10531 	return queue;
10532 }
10533 
10534 static const struct ethtool_ops default_ethtool_ops;
10535 
10536 void netdev_set_default_ethtool_ops(struct net_device *dev,
10537 				    const struct ethtool_ops *ops)
10538 {
10539 	if (dev->ethtool_ops == &default_ethtool_ops)
10540 		dev->ethtool_ops = ops;
10541 }
10542 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10543 
10544 void netdev_freemem(struct net_device *dev)
10545 {
10546 	char *addr = (char *)dev - dev->padded;
10547 
10548 	kvfree(addr);
10549 }
10550 
10551 /**
10552  * alloc_netdev_mqs - allocate network device
10553  * @sizeof_priv: size of private data to allocate space for
10554  * @name: device name format string
10555  * @name_assign_type: origin of device name
10556  * @setup: callback to initialize device
10557  * @txqs: the number of TX subqueues to allocate
10558  * @rxqs: the number of RX subqueues to allocate
10559  *
10560  * Allocates a struct net_device with private data area for driver use
10561  * and performs basic initialization.  Also allocates subqueue structs
10562  * for each queue on the device.
10563  */
10564 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10565 		unsigned char name_assign_type,
10566 		void (*setup)(struct net_device *),
10567 		unsigned int txqs, unsigned int rxqs)
10568 {
10569 	struct net_device *dev;
10570 	unsigned int alloc_size;
10571 	struct net_device *p;
10572 
10573 	BUG_ON(strlen(name) >= sizeof(dev->name));
10574 
10575 	if (txqs < 1) {
10576 		pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10577 		return NULL;
10578 	}
10579 
10580 	if (rxqs < 1) {
10581 		pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10582 		return NULL;
10583 	}
10584 
10585 	alloc_size = sizeof(struct net_device);
10586 	if (sizeof_priv) {
10587 		/* ensure 32-byte alignment of private area */
10588 		alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10589 		alloc_size += sizeof_priv;
10590 	}
10591 	/* ensure 32-byte alignment of whole construct */
10592 	alloc_size += NETDEV_ALIGN - 1;
10593 
10594 	p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10595 	if (!p)
10596 		return NULL;
10597 
10598 	dev = PTR_ALIGN(p, NETDEV_ALIGN);
10599 	dev->padded = (char *)dev - (char *)p;
10600 
10601 	ref_tracker_dir_init(&dev->refcnt_tracker, 128);
10602 #ifdef CONFIG_PCPU_DEV_REFCNT
10603 	dev->pcpu_refcnt = alloc_percpu(int);
10604 	if (!dev->pcpu_refcnt)
10605 		goto free_dev;
10606 	__dev_hold(dev);
10607 #else
10608 	refcount_set(&dev->dev_refcnt, 1);
10609 #endif
10610 
10611 	if (dev_addr_init(dev))
10612 		goto free_pcpu;
10613 
10614 	dev_mc_init(dev);
10615 	dev_uc_init(dev);
10616 
10617 	dev_net_set(dev, &init_net);
10618 
10619 	dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
10620 	dev->gso_max_segs = GSO_MAX_SEGS;
10621 	dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
10622 	dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
10623 	dev->tso_max_segs = TSO_MAX_SEGS;
10624 	dev->upper_level = 1;
10625 	dev->lower_level = 1;
10626 #ifdef CONFIG_LOCKDEP
10627 	dev->nested_level = 0;
10628 	INIT_LIST_HEAD(&dev->unlink_list);
10629 #endif
10630 
10631 	INIT_LIST_HEAD(&dev->napi_list);
10632 	INIT_LIST_HEAD(&dev->unreg_list);
10633 	INIT_LIST_HEAD(&dev->close_list);
10634 	INIT_LIST_HEAD(&dev->link_watch_list);
10635 	INIT_LIST_HEAD(&dev->adj_list.upper);
10636 	INIT_LIST_HEAD(&dev->adj_list.lower);
10637 	INIT_LIST_HEAD(&dev->ptype_all);
10638 	INIT_LIST_HEAD(&dev->ptype_specific);
10639 	INIT_LIST_HEAD(&dev->net_notifier_list);
10640 #ifdef CONFIG_NET_SCHED
10641 	hash_init(dev->qdisc_hash);
10642 #endif
10643 	dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10644 	setup(dev);
10645 
10646 	if (!dev->tx_queue_len) {
10647 		dev->priv_flags |= IFF_NO_QUEUE;
10648 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10649 	}
10650 
10651 	dev->num_tx_queues = txqs;
10652 	dev->real_num_tx_queues = txqs;
10653 	if (netif_alloc_netdev_queues(dev))
10654 		goto free_all;
10655 
10656 	dev->num_rx_queues = rxqs;
10657 	dev->real_num_rx_queues = rxqs;
10658 	if (netif_alloc_rx_queues(dev))
10659 		goto free_all;
10660 
10661 	strcpy(dev->name, name);
10662 	dev->name_assign_type = name_assign_type;
10663 	dev->group = INIT_NETDEV_GROUP;
10664 	if (!dev->ethtool_ops)
10665 		dev->ethtool_ops = &default_ethtool_ops;
10666 
10667 	nf_hook_netdev_init(dev);
10668 
10669 	return dev;
10670 
10671 free_all:
10672 	free_netdev(dev);
10673 	return NULL;
10674 
10675 free_pcpu:
10676 #ifdef CONFIG_PCPU_DEV_REFCNT
10677 	free_percpu(dev->pcpu_refcnt);
10678 free_dev:
10679 #endif
10680 	netdev_freemem(dev);
10681 	return NULL;
10682 }
10683 EXPORT_SYMBOL(alloc_netdev_mqs);
10684 
10685 /**
10686  * free_netdev - free network device
10687  * @dev: device
10688  *
10689  * This function does the last stage of destroying an allocated device
10690  * interface. The reference to the device object is released. If this
10691  * is the last reference then it will be freed.Must be called in process
10692  * context.
10693  */
10694 void free_netdev(struct net_device *dev)
10695 {
10696 	struct napi_struct *p, *n;
10697 
10698 	might_sleep();
10699 
10700 	/* When called immediately after register_netdevice() failed the unwind
10701 	 * handling may still be dismantling the device. Handle that case by
10702 	 * deferring the free.
10703 	 */
10704 	if (dev->reg_state == NETREG_UNREGISTERING) {
10705 		ASSERT_RTNL();
10706 		dev->needs_free_netdev = true;
10707 		return;
10708 	}
10709 
10710 	netif_free_tx_queues(dev);
10711 	netif_free_rx_queues(dev);
10712 
10713 	kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10714 
10715 	/* Flush device addresses */
10716 	dev_addr_flush(dev);
10717 
10718 	list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10719 		netif_napi_del(p);
10720 
10721 	ref_tracker_dir_exit(&dev->refcnt_tracker);
10722 #ifdef CONFIG_PCPU_DEV_REFCNT
10723 	free_percpu(dev->pcpu_refcnt);
10724 	dev->pcpu_refcnt = NULL;
10725 #endif
10726 	free_percpu(dev->core_stats);
10727 	dev->core_stats = NULL;
10728 	free_percpu(dev->xdp_bulkq);
10729 	dev->xdp_bulkq = NULL;
10730 
10731 	/*  Compatibility with error handling in drivers */
10732 	if (dev->reg_state == NETREG_UNINITIALIZED) {
10733 		netdev_freemem(dev);
10734 		return;
10735 	}
10736 
10737 	BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10738 	dev->reg_state = NETREG_RELEASED;
10739 
10740 	/* will free via device release */
10741 	put_device(&dev->dev);
10742 }
10743 EXPORT_SYMBOL(free_netdev);
10744 
10745 /**
10746  *	synchronize_net -  Synchronize with packet receive processing
10747  *
10748  *	Wait for packets currently being received to be done.
10749  *	Does not block later packets from starting.
10750  */
10751 void synchronize_net(void)
10752 {
10753 	might_sleep();
10754 	if (rtnl_is_locked())
10755 		synchronize_rcu_expedited();
10756 	else
10757 		synchronize_rcu();
10758 }
10759 EXPORT_SYMBOL(synchronize_net);
10760 
10761 /**
10762  *	unregister_netdevice_queue - remove device from the kernel
10763  *	@dev: device
10764  *	@head: list
10765  *
10766  *	This function shuts down a device interface and removes it
10767  *	from the kernel tables.
10768  *	If head not NULL, device is queued to be unregistered later.
10769  *
10770  *	Callers must hold the rtnl semaphore.  You may want
10771  *	unregister_netdev() instead of this.
10772  */
10773 
10774 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10775 {
10776 	ASSERT_RTNL();
10777 
10778 	if (head) {
10779 		list_move_tail(&dev->unreg_list, head);
10780 	} else {
10781 		LIST_HEAD(single);
10782 
10783 		list_add(&dev->unreg_list, &single);
10784 		unregister_netdevice_many(&single);
10785 	}
10786 }
10787 EXPORT_SYMBOL(unregister_netdevice_queue);
10788 
10789 /**
10790  *	unregister_netdevice_many - unregister many devices
10791  *	@head: list of devices
10792  *
10793  *  Note: As most callers use a stack allocated list_head,
10794  *  we force a list_del() to make sure stack wont be corrupted later.
10795  */
10796 void unregister_netdevice_many(struct list_head *head)
10797 {
10798 	struct net_device *dev, *tmp;
10799 	LIST_HEAD(close_head);
10800 
10801 	BUG_ON(dev_boot_phase);
10802 	ASSERT_RTNL();
10803 
10804 	if (list_empty(head))
10805 		return;
10806 
10807 	list_for_each_entry_safe(dev, tmp, head, unreg_list) {
10808 		/* Some devices call without registering
10809 		 * for initialization unwind. Remove those
10810 		 * devices and proceed with the remaining.
10811 		 */
10812 		if (dev->reg_state == NETREG_UNINITIALIZED) {
10813 			pr_debug("unregister_netdevice: device %s/%p never was registered\n",
10814 				 dev->name, dev);
10815 
10816 			WARN_ON(1);
10817 			list_del(&dev->unreg_list);
10818 			continue;
10819 		}
10820 		dev->dismantle = true;
10821 		BUG_ON(dev->reg_state != NETREG_REGISTERED);
10822 	}
10823 
10824 	/* If device is running, close it first. */
10825 	list_for_each_entry(dev, head, unreg_list)
10826 		list_add_tail(&dev->close_list, &close_head);
10827 	dev_close_many(&close_head, true);
10828 
10829 	list_for_each_entry(dev, head, unreg_list) {
10830 		/* And unlink it from device chain. */
10831 		unlist_netdevice(dev);
10832 
10833 		dev->reg_state = NETREG_UNREGISTERING;
10834 	}
10835 	flush_all_backlogs();
10836 
10837 	synchronize_net();
10838 
10839 	list_for_each_entry(dev, head, unreg_list) {
10840 		struct sk_buff *skb = NULL;
10841 
10842 		/* Shutdown queueing discipline. */
10843 		dev_shutdown(dev);
10844 
10845 		dev_xdp_uninstall(dev);
10846 
10847 		netdev_offload_xstats_disable_all(dev);
10848 
10849 		/* Notify protocols, that we are about to destroy
10850 		 * this device. They should clean all the things.
10851 		 */
10852 		call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10853 
10854 		if (!dev->rtnl_link_ops ||
10855 		    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10856 			skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
10857 						     GFP_KERNEL, NULL, 0);
10858 
10859 		/*
10860 		 *	Flush the unicast and multicast chains
10861 		 */
10862 		dev_uc_flush(dev);
10863 		dev_mc_flush(dev);
10864 
10865 		netdev_name_node_alt_flush(dev);
10866 		netdev_name_node_free(dev->name_node);
10867 
10868 		if (dev->netdev_ops->ndo_uninit)
10869 			dev->netdev_ops->ndo_uninit(dev);
10870 
10871 		if (skb)
10872 			rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
10873 
10874 		/* Notifier chain MUST detach us all upper devices. */
10875 		WARN_ON(netdev_has_any_upper_dev(dev));
10876 		WARN_ON(netdev_has_any_lower_dev(dev));
10877 
10878 		/* Remove entries from kobject tree */
10879 		netdev_unregister_kobject(dev);
10880 #ifdef CONFIG_XPS
10881 		/* Remove XPS queueing entries */
10882 		netif_reset_xps_queues_gt(dev, 0);
10883 #endif
10884 	}
10885 
10886 	synchronize_net();
10887 
10888 	list_for_each_entry(dev, head, unreg_list) {
10889 		netdev_put(dev, &dev->dev_registered_tracker);
10890 		net_set_todo(dev);
10891 	}
10892 
10893 	list_del(head);
10894 }
10895 EXPORT_SYMBOL(unregister_netdevice_many);
10896 
10897 /**
10898  *	unregister_netdev - remove device from the kernel
10899  *	@dev: device
10900  *
10901  *	This function shuts down a device interface and removes it
10902  *	from the kernel tables.
10903  *
10904  *	This is just a wrapper for unregister_netdevice that takes
10905  *	the rtnl semaphore.  In general you want to use this and not
10906  *	unregister_netdevice.
10907  */
10908 void unregister_netdev(struct net_device *dev)
10909 {
10910 	rtnl_lock();
10911 	unregister_netdevice(dev);
10912 	rtnl_unlock();
10913 }
10914 EXPORT_SYMBOL(unregister_netdev);
10915 
10916 /**
10917  *	__dev_change_net_namespace - move device to different nethost namespace
10918  *	@dev: device
10919  *	@net: network namespace
10920  *	@pat: If not NULL name pattern to try if the current device name
10921  *	      is already taken in the destination network namespace.
10922  *	@new_ifindex: If not zero, specifies device index in the target
10923  *	              namespace.
10924  *
10925  *	This function shuts down a device interface and moves it
10926  *	to a new network namespace. On success 0 is returned, on
10927  *	a failure a netagive errno code is returned.
10928  *
10929  *	Callers must hold the rtnl semaphore.
10930  */
10931 
10932 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
10933 			       const char *pat, int new_ifindex)
10934 {
10935 	struct net *net_old = dev_net(dev);
10936 	int err, new_nsid;
10937 
10938 	ASSERT_RTNL();
10939 
10940 	/* Don't allow namespace local devices to be moved. */
10941 	err = -EINVAL;
10942 	if (dev->features & NETIF_F_NETNS_LOCAL)
10943 		goto out;
10944 
10945 	/* Ensure the device has been registrered */
10946 	if (dev->reg_state != NETREG_REGISTERED)
10947 		goto out;
10948 
10949 	/* Get out if there is nothing todo */
10950 	err = 0;
10951 	if (net_eq(net_old, net))
10952 		goto out;
10953 
10954 	/* Pick the destination device name, and ensure
10955 	 * we can use it in the destination network namespace.
10956 	 */
10957 	err = -EEXIST;
10958 	if (netdev_name_in_use(net, dev->name)) {
10959 		/* We get here if we can't use the current device name */
10960 		if (!pat)
10961 			goto out;
10962 		err = dev_get_valid_name(net, dev, pat);
10963 		if (err < 0)
10964 			goto out;
10965 	}
10966 
10967 	/* Check that new_ifindex isn't used yet. */
10968 	err = -EBUSY;
10969 	if (new_ifindex && __dev_get_by_index(net, new_ifindex))
10970 		goto out;
10971 
10972 	/*
10973 	 * And now a mini version of register_netdevice unregister_netdevice.
10974 	 */
10975 
10976 	/* If device is running close it first. */
10977 	dev_close(dev);
10978 
10979 	/* And unlink it from device chain */
10980 	unlist_netdevice(dev);
10981 
10982 	synchronize_net();
10983 
10984 	/* Shutdown queueing discipline. */
10985 	dev_shutdown(dev);
10986 
10987 	/* Notify protocols, that we are about to destroy
10988 	 * this device. They should clean all the things.
10989 	 *
10990 	 * Note that dev->reg_state stays at NETREG_REGISTERED.
10991 	 * This is wanted because this way 8021q and macvlan know
10992 	 * the device is just moving and can keep their slaves up.
10993 	 */
10994 	call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10995 	rcu_barrier();
10996 
10997 	new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
10998 	/* If there is an ifindex conflict assign a new one */
10999 	if (!new_ifindex) {
11000 		if (__dev_get_by_index(net, dev->ifindex))
11001 			new_ifindex = dev_new_index(net);
11002 		else
11003 			new_ifindex = dev->ifindex;
11004 	}
11005 
11006 	rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11007 			    new_ifindex);
11008 
11009 	/*
11010 	 *	Flush the unicast and multicast chains
11011 	 */
11012 	dev_uc_flush(dev);
11013 	dev_mc_flush(dev);
11014 
11015 	/* Send a netdev-removed uevent to the old namespace */
11016 	kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11017 	netdev_adjacent_del_links(dev);
11018 
11019 	/* Move per-net netdevice notifiers that are following the netdevice */
11020 	move_netdevice_notifiers_dev_net(dev, net);
11021 
11022 	/* Actually switch the network namespace */
11023 	dev_net_set(dev, net);
11024 	dev->ifindex = new_ifindex;
11025 
11026 	/* Send a netdev-add uevent to the new namespace */
11027 	kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11028 	netdev_adjacent_add_links(dev);
11029 
11030 	/* Fixup kobjects */
11031 	err = device_rename(&dev->dev, dev->name);
11032 	WARN_ON(err);
11033 
11034 	/* Adapt owner in case owning user namespace of target network
11035 	 * namespace is different from the original one.
11036 	 */
11037 	err = netdev_change_owner(dev, net_old, net);
11038 	WARN_ON(err);
11039 
11040 	/* Add the device back in the hashes */
11041 	list_netdevice(dev);
11042 
11043 	/* Notify protocols, that a new device appeared. */
11044 	call_netdevice_notifiers(NETDEV_REGISTER, dev);
11045 
11046 	/*
11047 	 *	Prevent userspace races by waiting until the network
11048 	 *	device is fully setup before sending notifications.
11049 	 */
11050 	rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
11051 
11052 	synchronize_net();
11053 	err = 0;
11054 out:
11055 	return err;
11056 }
11057 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11058 
11059 static int dev_cpu_dead(unsigned int oldcpu)
11060 {
11061 	struct sk_buff **list_skb;
11062 	struct sk_buff *skb;
11063 	unsigned int cpu;
11064 	struct softnet_data *sd, *oldsd, *remsd = NULL;
11065 
11066 	local_irq_disable();
11067 	cpu = smp_processor_id();
11068 	sd = &per_cpu(softnet_data, cpu);
11069 	oldsd = &per_cpu(softnet_data, oldcpu);
11070 
11071 	/* Find end of our completion_queue. */
11072 	list_skb = &sd->completion_queue;
11073 	while (*list_skb)
11074 		list_skb = &(*list_skb)->next;
11075 	/* Append completion queue from offline CPU. */
11076 	*list_skb = oldsd->completion_queue;
11077 	oldsd->completion_queue = NULL;
11078 
11079 	/* Append output queue from offline CPU. */
11080 	if (oldsd->output_queue) {
11081 		*sd->output_queue_tailp = oldsd->output_queue;
11082 		sd->output_queue_tailp = oldsd->output_queue_tailp;
11083 		oldsd->output_queue = NULL;
11084 		oldsd->output_queue_tailp = &oldsd->output_queue;
11085 	}
11086 	/* Append NAPI poll list from offline CPU, with one exception :
11087 	 * process_backlog() must be called by cpu owning percpu backlog.
11088 	 * We properly handle process_queue & input_pkt_queue later.
11089 	 */
11090 	while (!list_empty(&oldsd->poll_list)) {
11091 		struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11092 							    struct napi_struct,
11093 							    poll_list);
11094 
11095 		list_del_init(&napi->poll_list);
11096 		if (napi->poll == process_backlog)
11097 			napi->state = 0;
11098 		else
11099 			____napi_schedule(sd, napi);
11100 	}
11101 
11102 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
11103 	local_irq_enable();
11104 
11105 #ifdef CONFIG_RPS
11106 	remsd = oldsd->rps_ipi_list;
11107 	oldsd->rps_ipi_list = NULL;
11108 #endif
11109 	/* send out pending IPI's on offline CPU */
11110 	net_rps_send_ipi(remsd);
11111 
11112 	/* Process offline CPU's input_pkt_queue */
11113 	while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11114 		netif_rx(skb);
11115 		input_queue_head_incr(oldsd);
11116 	}
11117 	while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11118 		netif_rx(skb);
11119 		input_queue_head_incr(oldsd);
11120 	}
11121 
11122 	return 0;
11123 }
11124 
11125 /**
11126  *	netdev_increment_features - increment feature set by one
11127  *	@all: current feature set
11128  *	@one: new feature set
11129  *	@mask: mask feature set
11130  *
11131  *	Computes a new feature set after adding a device with feature set
11132  *	@one to the master device with current feature set @all.  Will not
11133  *	enable anything that is off in @mask. Returns the new feature set.
11134  */
11135 netdev_features_t netdev_increment_features(netdev_features_t all,
11136 	netdev_features_t one, netdev_features_t mask)
11137 {
11138 	if (mask & NETIF_F_HW_CSUM)
11139 		mask |= NETIF_F_CSUM_MASK;
11140 	mask |= NETIF_F_VLAN_CHALLENGED;
11141 
11142 	all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11143 	all &= one | ~NETIF_F_ALL_FOR_ALL;
11144 
11145 	/* If one device supports hw checksumming, set for all. */
11146 	if (all & NETIF_F_HW_CSUM)
11147 		all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11148 
11149 	return all;
11150 }
11151 EXPORT_SYMBOL(netdev_increment_features);
11152 
11153 static struct hlist_head * __net_init netdev_create_hash(void)
11154 {
11155 	int i;
11156 	struct hlist_head *hash;
11157 
11158 	hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11159 	if (hash != NULL)
11160 		for (i = 0; i < NETDEV_HASHENTRIES; i++)
11161 			INIT_HLIST_HEAD(&hash[i]);
11162 
11163 	return hash;
11164 }
11165 
11166 /* Initialize per network namespace state */
11167 static int __net_init netdev_init(struct net *net)
11168 {
11169 	BUILD_BUG_ON(GRO_HASH_BUCKETS >
11170 		     8 * sizeof_field(struct napi_struct, gro_bitmask));
11171 
11172 	INIT_LIST_HEAD(&net->dev_base_head);
11173 
11174 	net->dev_name_head = netdev_create_hash();
11175 	if (net->dev_name_head == NULL)
11176 		goto err_name;
11177 
11178 	net->dev_index_head = netdev_create_hash();
11179 	if (net->dev_index_head == NULL)
11180 		goto err_idx;
11181 
11182 	RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11183 
11184 	return 0;
11185 
11186 err_idx:
11187 	kfree(net->dev_name_head);
11188 err_name:
11189 	return -ENOMEM;
11190 }
11191 
11192 /**
11193  *	netdev_drivername - network driver for the device
11194  *	@dev: network device
11195  *
11196  *	Determine network driver for device.
11197  */
11198 const char *netdev_drivername(const struct net_device *dev)
11199 {
11200 	const struct device_driver *driver;
11201 	const struct device *parent;
11202 	const char *empty = "";
11203 
11204 	parent = dev->dev.parent;
11205 	if (!parent)
11206 		return empty;
11207 
11208 	driver = parent->driver;
11209 	if (driver && driver->name)
11210 		return driver->name;
11211 	return empty;
11212 }
11213 
11214 static void __netdev_printk(const char *level, const struct net_device *dev,
11215 			    struct va_format *vaf)
11216 {
11217 	if (dev && dev->dev.parent) {
11218 		dev_printk_emit(level[1] - '0',
11219 				dev->dev.parent,
11220 				"%s %s %s%s: %pV",
11221 				dev_driver_string(dev->dev.parent),
11222 				dev_name(dev->dev.parent),
11223 				netdev_name(dev), netdev_reg_state(dev),
11224 				vaf);
11225 	} else if (dev) {
11226 		printk("%s%s%s: %pV",
11227 		       level, netdev_name(dev), netdev_reg_state(dev), vaf);
11228 	} else {
11229 		printk("%s(NULL net_device): %pV", level, vaf);
11230 	}
11231 }
11232 
11233 void netdev_printk(const char *level, const struct net_device *dev,
11234 		   const char *format, ...)
11235 {
11236 	struct va_format vaf;
11237 	va_list args;
11238 
11239 	va_start(args, format);
11240 
11241 	vaf.fmt = format;
11242 	vaf.va = &args;
11243 
11244 	__netdev_printk(level, dev, &vaf);
11245 
11246 	va_end(args);
11247 }
11248 EXPORT_SYMBOL(netdev_printk);
11249 
11250 #define define_netdev_printk_level(func, level)			\
11251 void func(const struct net_device *dev, const char *fmt, ...)	\
11252 {								\
11253 	struct va_format vaf;					\
11254 	va_list args;						\
11255 								\
11256 	va_start(args, fmt);					\
11257 								\
11258 	vaf.fmt = fmt;						\
11259 	vaf.va = &args;						\
11260 								\
11261 	__netdev_printk(level, dev, &vaf);			\
11262 								\
11263 	va_end(args);						\
11264 }								\
11265 EXPORT_SYMBOL(func);
11266 
11267 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11268 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11269 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11270 define_netdev_printk_level(netdev_err, KERN_ERR);
11271 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11272 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11273 define_netdev_printk_level(netdev_info, KERN_INFO);
11274 
11275 static void __net_exit netdev_exit(struct net *net)
11276 {
11277 	kfree(net->dev_name_head);
11278 	kfree(net->dev_index_head);
11279 	if (net != &init_net)
11280 		WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11281 }
11282 
11283 static struct pernet_operations __net_initdata netdev_net_ops = {
11284 	.init = netdev_init,
11285 	.exit = netdev_exit,
11286 };
11287 
11288 static void __net_exit default_device_exit_net(struct net *net)
11289 {
11290 	struct net_device *dev, *aux;
11291 	/*
11292 	 * Push all migratable network devices back to the
11293 	 * initial network namespace
11294 	 */
11295 	ASSERT_RTNL();
11296 	for_each_netdev_safe(net, dev, aux) {
11297 		int err;
11298 		char fb_name[IFNAMSIZ];
11299 
11300 		/* Ignore unmoveable devices (i.e. loopback) */
11301 		if (dev->features & NETIF_F_NETNS_LOCAL)
11302 			continue;
11303 
11304 		/* Leave virtual devices for the generic cleanup */
11305 		if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11306 			continue;
11307 
11308 		/* Push remaining network devices to init_net */
11309 		snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11310 		if (netdev_name_in_use(&init_net, fb_name))
11311 			snprintf(fb_name, IFNAMSIZ, "dev%%d");
11312 		err = dev_change_net_namespace(dev, &init_net, fb_name);
11313 		if (err) {
11314 			pr_emerg("%s: failed to move %s to init_net: %d\n",
11315 				 __func__, dev->name, err);
11316 			BUG();
11317 		}
11318 	}
11319 }
11320 
11321 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11322 {
11323 	/* At exit all network devices most be removed from a network
11324 	 * namespace.  Do this in the reverse order of registration.
11325 	 * Do this across as many network namespaces as possible to
11326 	 * improve batching efficiency.
11327 	 */
11328 	struct net_device *dev;
11329 	struct net *net;
11330 	LIST_HEAD(dev_kill_list);
11331 
11332 	rtnl_lock();
11333 	list_for_each_entry(net, net_list, exit_list) {
11334 		default_device_exit_net(net);
11335 		cond_resched();
11336 	}
11337 
11338 	list_for_each_entry(net, net_list, exit_list) {
11339 		for_each_netdev_reverse(net, dev) {
11340 			if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11341 				dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11342 			else
11343 				unregister_netdevice_queue(dev, &dev_kill_list);
11344 		}
11345 	}
11346 	unregister_netdevice_many(&dev_kill_list);
11347 	rtnl_unlock();
11348 }
11349 
11350 static struct pernet_operations __net_initdata default_device_ops = {
11351 	.exit_batch = default_device_exit_batch,
11352 };
11353 
11354 /*
11355  *	Initialize the DEV module. At boot time this walks the device list and
11356  *	unhooks any devices that fail to initialise (normally hardware not
11357  *	present) and leaves us with a valid list of present and active devices.
11358  *
11359  */
11360 
11361 /*
11362  *       This is called single threaded during boot, so no need
11363  *       to take the rtnl semaphore.
11364  */
11365 static int __init net_dev_init(void)
11366 {
11367 	int i, rc = -ENOMEM;
11368 
11369 	BUG_ON(!dev_boot_phase);
11370 
11371 	if (dev_proc_init())
11372 		goto out;
11373 
11374 	if (netdev_kobject_init())
11375 		goto out;
11376 
11377 	INIT_LIST_HEAD(&ptype_all);
11378 	for (i = 0; i < PTYPE_HASH_SIZE; i++)
11379 		INIT_LIST_HEAD(&ptype_base[i]);
11380 
11381 	if (register_pernet_subsys(&netdev_net_ops))
11382 		goto out;
11383 
11384 	/*
11385 	 *	Initialise the packet receive queues.
11386 	 */
11387 
11388 	for_each_possible_cpu(i) {
11389 		struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11390 		struct softnet_data *sd = &per_cpu(softnet_data, i);
11391 
11392 		INIT_WORK(flush, flush_backlog);
11393 
11394 		skb_queue_head_init(&sd->input_pkt_queue);
11395 		skb_queue_head_init(&sd->process_queue);
11396 #ifdef CONFIG_XFRM_OFFLOAD
11397 		skb_queue_head_init(&sd->xfrm_backlog);
11398 #endif
11399 		INIT_LIST_HEAD(&sd->poll_list);
11400 		sd->output_queue_tailp = &sd->output_queue;
11401 #ifdef CONFIG_RPS
11402 		INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11403 		sd->cpu = i;
11404 #endif
11405 		INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
11406 		spin_lock_init(&sd->defer_lock);
11407 
11408 		init_gro_hash(&sd->backlog);
11409 		sd->backlog.poll = process_backlog;
11410 		sd->backlog.weight = weight_p;
11411 	}
11412 
11413 	dev_boot_phase = 0;
11414 
11415 	/* The loopback device is special if any other network devices
11416 	 * is present in a network namespace the loopback device must
11417 	 * be present. Since we now dynamically allocate and free the
11418 	 * loopback device ensure this invariant is maintained by
11419 	 * keeping the loopback device as the first device on the
11420 	 * list of network devices.  Ensuring the loopback devices
11421 	 * is the first device that appears and the last network device
11422 	 * that disappears.
11423 	 */
11424 	if (register_pernet_device(&loopback_net_ops))
11425 		goto out;
11426 
11427 	if (register_pernet_device(&default_device_ops))
11428 		goto out;
11429 
11430 	open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11431 	open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11432 
11433 	rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11434 				       NULL, dev_cpu_dead);
11435 	WARN_ON(rc < 0);
11436 	rc = 0;
11437 out:
11438 	return rc;
11439 }
11440 
11441 subsys_initcall(net_dev_init);
11442