xref: /openbmc/linux/net/core/dev.c (revision 3098cb65)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *      NET3    Protocol independent device support routines.
4  *
5  *	Derived from the non IP parts of dev.c 1.0.19
6  *              Authors:	Ross Biro
7  *				Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
8  *				Mark Evans, <evansmp@uhura.aston.ac.uk>
9  *
10  *	Additional Authors:
11  *		Florian la Roche <rzsfl@rz.uni-sb.de>
12  *		Alan Cox <gw4pts@gw4pts.ampr.org>
13  *		David Hinds <dahinds@users.sourceforge.net>
14  *		Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
15  *		Adam Sulmicki <adam@cfar.umd.edu>
16  *              Pekka Riikonen <priikone@poesidon.pspt.fi>
17  *
18  *	Changes:
19  *              D.J. Barrow     :       Fixed bug where dev->refcnt gets set
20  *                                      to 2 if register_netdev gets called
21  *                                      before net_dev_init & also removed a
22  *                                      few lines of code in the process.
23  *		Alan Cox	:	device private ioctl copies fields back.
24  *		Alan Cox	:	Transmit queue code does relevant
25  *					stunts to keep the queue safe.
26  *		Alan Cox	:	Fixed double lock.
27  *		Alan Cox	:	Fixed promisc NULL pointer trap
28  *		????????	:	Support the full private ioctl range
29  *		Alan Cox	:	Moved ioctl permission check into
30  *					drivers
31  *		Tim Kordas	:	SIOCADDMULTI/SIOCDELMULTI
32  *		Alan Cox	:	100 backlog just doesn't cut it when
33  *					you start doing multicast video 8)
34  *		Alan Cox	:	Rewrote net_bh and list manager.
35  *              Alan Cox        :       Fix ETH_P_ALL echoback lengths.
36  *		Alan Cox	:	Took out transmit every packet pass
37  *					Saved a few bytes in the ioctl handler
38  *		Alan Cox	:	Network driver sets packet type before
39  *					calling netif_rx. Saves a function
40  *					call a packet.
41  *		Alan Cox	:	Hashed net_bh()
42  *		Richard Kooijman:	Timestamp fixes.
43  *		Alan Cox	:	Wrong field in SIOCGIFDSTADDR
44  *		Alan Cox	:	Device lock protection.
45  *              Alan Cox        :       Fixed nasty side effect of device close
46  *					changes.
47  *		Rudi Cilibrasi	:	Pass the right thing to
48  *					set_mac_address()
49  *		Dave Miller	:	32bit quantity for the device lock to
50  *					make it work out on a Sparc.
51  *		Bjorn Ekwall	:	Added KERNELD hack.
52  *		Alan Cox	:	Cleaned up the backlog initialise.
53  *		Craig Metz	:	SIOCGIFCONF fix if space for under
54  *					1 device.
55  *	    Thomas Bogendoerfer :	Return ENODEV for dev_open, if there
56  *					is no device open function.
57  *		Andi Kleen	:	Fix error reporting for SIOCGIFCONF
58  *	    Michael Chastain	:	Fix signed/unsigned for SIOCGIFCONF
59  *		Cyrus Durgin	:	Cleaned for KMOD
60  *		Adam Sulmicki   :	Bug Fix : Network Device Unload
61  *					A network device unload needs to purge
62  *					the backlog queue.
63  *	Paul Rusty Russell	:	SIOCSIFNAME
64  *              Pekka Riikonen  :	Netdev boot-time settings code
65  *              Andrew Morton   :       Make unregister_netdevice wait
66  *                                      indefinitely on dev->refcnt
67  *              J Hadi Salim    :       - Backlog queue sampling
68  *				        - netif_rx() feedback
69  */
70 
71 #include <linux/uaccess.h>
72 #include <linux/bitops.h>
73 #include <linux/capability.h>
74 #include <linux/cpu.h>
75 #include <linux/types.h>
76 #include <linux/kernel.h>
77 #include <linux/hash.h>
78 #include <linux/slab.h>
79 #include <linux/sched.h>
80 #include <linux/sched/mm.h>
81 #include <linux/mutex.h>
82 #include <linux/rwsem.h>
83 #include <linux/string.h>
84 #include <linux/mm.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/ethtool.h>
93 #include <linux/skbuff.h>
94 #include <linux/kthread.h>
95 #include <linux/bpf.h>
96 #include <linux/bpf_trace.h>
97 #include <net/net_namespace.h>
98 #include <net/sock.h>
99 #include <net/busy_poll.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/stat.h>
102 #include <net/dsa.h>
103 #include <net/dst.h>
104 #include <net/dst_metadata.h>
105 #include <net/gro.h>
106 #include <net/pkt_sched.h>
107 #include <net/pkt_cls.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/module.h>
113 #include <linux/netpoll.h>
114 #include <linux/rcupdate.h>
115 #include <linux/delay.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
125 #include <net/ip.h>
126 #include <net/mpls.h>
127 #include <linux/ipv6.h>
128 #include <linux/in.h>
129 #include <linux/jhash.h>
130 #include <linux/random.h>
131 #include <trace/events/napi.h>
132 #include <trace/events/net.h>
133 #include <trace/events/skb.h>
134 #include <trace/events/qdisc.h>
135 #include <linux/inetdevice.h>
136 #include <linux/cpu_rmap.h>
137 #include <linux/static_key.h>
138 #include <linux/hashtable.h>
139 #include <linux/vmalloc.h>
140 #include <linux/if_macvlan.h>
141 #include <linux/errqueue.h>
142 #include <linux/hrtimer.h>
143 #include <linux/netfilter_netdev.h>
144 #include <linux/crash_dump.h>
145 #include <linux/sctp.h>
146 #include <net/udp_tunnel.h>
147 #include <linux/net_namespace.h>
148 #include <linux/indirect_call_wrapper.h>
149 #include <net/devlink.h>
150 #include <linux/pm_runtime.h>
151 #include <linux/prandom.h>
152 #include <linux/once_lite.h>
153 
154 #include "dev.h"
155 #include "net-sysfs.h"
156 
157 
158 static DEFINE_SPINLOCK(ptype_lock);
159 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
160 struct list_head ptype_all __read_mostly;	/* Taps */
161 
162 static int netif_rx_internal(struct sk_buff *skb);
163 static int call_netdevice_notifiers_info(unsigned long val,
164 					 struct netdev_notifier_info *info);
165 static int call_netdevice_notifiers_extack(unsigned long val,
166 					   struct net_device *dev,
167 					   struct netlink_ext_ack *extack);
168 static struct napi_struct *napi_by_id(unsigned int napi_id);
169 
170 /*
171  * The @dev_base_head list is protected by @dev_base_lock and the rtnl
172  * semaphore.
173  *
174  * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
175  *
176  * Writers must hold the rtnl semaphore while they loop through the
177  * dev_base_head list, and hold dev_base_lock for writing when they do the
178  * actual updates.  This allows pure readers to access the list even
179  * while a writer is preparing to update it.
180  *
181  * To put it another way, dev_base_lock is held for writing only to
182  * protect against pure readers; the rtnl semaphore provides the
183  * protection against other writers.
184  *
185  * See, for example usages, register_netdevice() and
186  * unregister_netdevice(), which must be called with the rtnl
187  * semaphore held.
188  */
189 DEFINE_RWLOCK(dev_base_lock);
190 EXPORT_SYMBOL(dev_base_lock);
191 
192 static DEFINE_MUTEX(ifalias_mutex);
193 
194 /* protects napi_hash addition/deletion and napi_gen_id */
195 static DEFINE_SPINLOCK(napi_hash_lock);
196 
197 static unsigned int napi_gen_id = NR_CPUS;
198 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
199 
200 static DECLARE_RWSEM(devnet_rename_sem);
201 
202 static inline void dev_base_seq_inc(struct net *net)
203 {
204 	while (++net->dev_base_seq == 0)
205 		;
206 }
207 
208 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
209 {
210 	unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
211 
212 	return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
213 }
214 
215 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
216 {
217 	return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
218 }
219 
220 static inline void rps_lock_irqsave(struct softnet_data *sd,
221 				    unsigned long *flags)
222 {
223 	if (IS_ENABLED(CONFIG_RPS))
224 		spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags);
225 	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
226 		local_irq_save(*flags);
227 }
228 
229 static inline void rps_lock_irq_disable(struct softnet_data *sd)
230 {
231 	if (IS_ENABLED(CONFIG_RPS))
232 		spin_lock_irq(&sd->input_pkt_queue.lock);
233 	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
234 		local_irq_disable();
235 }
236 
237 static inline void rps_unlock_irq_restore(struct softnet_data *sd,
238 					  unsigned long *flags)
239 {
240 	if (IS_ENABLED(CONFIG_RPS))
241 		spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags);
242 	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
243 		local_irq_restore(*flags);
244 }
245 
246 static inline void rps_unlock_irq_enable(struct softnet_data *sd)
247 {
248 	if (IS_ENABLED(CONFIG_RPS))
249 		spin_unlock_irq(&sd->input_pkt_queue.lock);
250 	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
251 		local_irq_enable();
252 }
253 
254 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
255 						       const char *name)
256 {
257 	struct netdev_name_node *name_node;
258 
259 	name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
260 	if (!name_node)
261 		return NULL;
262 	INIT_HLIST_NODE(&name_node->hlist);
263 	name_node->dev = dev;
264 	name_node->name = name;
265 	return name_node;
266 }
267 
268 static struct netdev_name_node *
269 netdev_name_node_head_alloc(struct net_device *dev)
270 {
271 	struct netdev_name_node *name_node;
272 
273 	name_node = netdev_name_node_alloc(dev, dev->name);
274 	if (!name_node)
275 		return NULL;
276 	INIT_LIST_HEAD(&name_node->list);
277 	return name_node;
278 }
279 
280 static void netdev_name_node_free(struct netdev_name_node *name_node)
281 {
282 	kfree(name_node);
283 }
284 
285 static void netdev_name_node_add(struct net *net,
286 				 struct netdev_name_node *name_node)
287 {
288 	hlist_add_head_rcu(&name_node->hlist,
289 			   dev_name_hash(net, name_node->name));
290 }
291 
292 static void netdev_name_node_del(struct netdev_name_node *name_node)
293 {
294 	hlist_del_rcu(&name_node->hlist);
295 }
296 
297 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
298 							const char *name)
299 {
300 	struct hlist_head *head = dev_name_hash(net, name);
301 	struct netdev_name_node *name_node;
302 
303 	hlist_for_each_entry(name_node, head, hlist)
304 		if (!strcmp(name_node->name, name))
305 			return name_node;
306 	return NULL;
307 }
308 
309 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
310 							    const char *name)
311 {
312 	struct hlist_head *head = dev_name_hash(net, name);
313 	struct netdev_name_node *name_node;
314 
315 	hlist_for_each_entry_rcu(name_node, head, hlist)
316 		if (!strcmp(name_node->name, name))
317 			return name_node;
318 	return NULL;
319 }
320 
321 bool netdev_name_in_use(struct net *net, const char *name)
322 {
323 	return netdev_name_node_lookup(net, name);
324 }
325 EXPORT_SYMBOL(netdev_name_in_use);
326 
327 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
328 {
329 	struct netdev_name_node *name_node;
330 	struct net *net = dev_net(dev);
331 
332 	name_node = netdev_name_node_lookup(net, name);
333 	if (name_node)
334 		return -EEXIST;
335 	name_node = netdev_name_node_alloc(dev, name);
336 	if (!name_node)
337 		return -ENOMEM;
338 	netdev_name_node_add(net, name_node);
339 	/* The node that holds dev->name acts as a head of per-device list. */
340 	list_add_tail(&name_node->list, &dev->name_node->list);
341 
342 	return 0;
343 }
344 
345 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
346 {
347 	list_del(&name_node->list);
348 	netdev_name_node_del(name_node);
349 	kfree(name_node->name);
350 	netdev_name_node_free(name_node);
351 }
352 
353 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
354 {
355 	struct netdev_name_node *name_node;
356 	struct net *net = dev_net(dev);
357 
358 	name_node = netdev_name_node_lookup(net, name);
359 	if (!name_node)
360 		return -ENOENT;
361 	/* lookup might have found our primary name or a name belonging
362 	 * to another device.
363 	 */
364 	if (name_node == dev->name_node || name_node->dev != dev)
365 		return -EINVAL;
366 
367 	__netdev_name_node_alt_destroy(name_node);
368 
369 	return 0;
370 }
371 
372 static void netdev_name_node_alt_flush(struct net_device *dev)
373 {
374 	struct netdev_name_node *name_node, *tmp;
375 
376 	list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
377 		__netdev_name_node_alt_destroy(name_node);
378 }
379 
380 /* Device list insertion */
381 static void list_netdevice(struct net_device *dev)
382 {
383 	struct net *net = dev_net(dev);
384 
385 	ASSERT_RTNL();
386 
387 	write_lock(&dev_base_lock);
388 	list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
389 	netdev_name_node_add(net, dev->name_node);
390 	hlist_add_head_rcu(&dev->index_hlist,
391 			   dev_index_hash(net, dev->ifindex));
392 	write_unlock(&dev_base_lock);
393 
394 	dev_base_seq_inc(net);
395 }
396 
397 /* Device list removal
398  * caller must respect a RCU grace period before freeing/reusing dev
399  */
400 static void unlist_netdevice(struct net_device *dev, bool lock)
401 {
402 	ASSERT_RTNL();
403 
404 	/* Unlink dev from the device chain */
405 	if (lock)
406 		write_lock(&dev_base_lock);
407 	list_del_rcu(&dev->dev_list);
408 	netdev_name_node_del(dev->name_node);
409 	hlist_del_rcu(&dev->index_hlist);
410 	if (lock)
411 		write_unlock(&dev_base_lock);
412 
413 	dev_base_seq_inc(dev_net(dev));
414 }
415 
416 /*
417  *	Our notifier list
418  */
419 
420 static RAW_NOTIFIER_HEAD(netdev_chain);
421 
422 /*
423  *	Device drivers call our routines to queue packets here. We empty the
424  *	queue in the local softnet handler.
425  */
426 
427 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
428 EXPORT_PER_CPU_SYMBOL(softnet_data);
429 
430 #ifdef CONFIG_LOCKDEP
431 /*
432  * register_netdevice() inits txq->_xmit_lock and sets lockdep class
433  * according to dev->type
434  */
435 static const unsigned short netdev_lock_type[] = {
436 	 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
437 	 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
438 	 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
439 	 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
440 	 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
441 	 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
442 	 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
443 	 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
444 	 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
445 	 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
446 	 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
447 	 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
448 	 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
449 	 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
450 	 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
451 
452 static const char *const netdev_lock_name[] = {
453 	"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
454 	"_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
455 	"_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
456 	"_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
457 	"_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
458 	"_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
459 	"_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
460 	"_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
461 	"_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
462 	"_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
463 	"_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
464 	"_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
465 	"_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
466 	"_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
467 	"_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
468 
469 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
470 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
471 
472 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
473 {
474 	int i;
475 
476 	for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
477 		if (netdev_lock_type[i] == dev_type)
478 			return i;
479 	/* the last key is used by default */
480 	return ARRAY_SIZE(netdev_lock_type) - 1;
481 }
482 
483 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
484 						 unsigned short dev_type)
485 {
486 	int i;
487 
488 	i = netdev_lock_pos(dev_type);
489 	lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
490 				   netdev_lock_name[i]);
491 }
492 
493 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
494 {
495 	int i;
496 
497 	i = netdev_lock_pos(dev->type);
498 	lockdep_set_class_and_name(&dev->addr_list_lock,
499 				   &netdev_addr_lock_key[i],
500 				   netdev_lock_name[i]);
501 }
502 #else
503 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
504 						 unsigned short dev_type)
505 {
506 }
507 
508 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
509 {
510 }
511 #endif
512 
513 /*******************************************************************************
514  *
515  *		Protocol management and registration routines
516  *
517  *******************************************************************************/
518 
519 
520 /*
521  *	Add a protocol ID to the list. Now that the input handler is
522  *	smarter we can dispense with all the messy stuff that used to be
523  *	here.
524  *
525  *	BEWARE!!! Protocol handlers, mangling input packets,
526  *	MUST BE last in hash buckets and checking protocol handlers
527  *	MUST start from promiscuous ptype_all chain in net_bh.
528  *	It is true now, do not change it.
529  *	Explanation follows: if protocol handler, mangling packet, will
530  *	be the first on list, it is not able to sense, that packet
531  *	is cloned and should be copied-on-write, so that it will
532  *	change it and subsequent readers will get broken packet.
533  *							--ANK (980803)
534  */
535 
536 static inline struct list_head *ptype_head(const struct packet_type *pt)
537 {
538 	if (pt->type == htons(ETH_P_ALL))
539 		return pt->dev ? &pt->dev->ptype_all : &ptype_all;
540 	else
541 		return pt->dev ? &pt->dev->ptype_specific :
542 				 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
543 }
544 
545 /**
546  *	dev_add_pack - add packet handler
547  *	@pt: packet type declaration
548  *
549  *	Add a protocol handler to the networking stack. The passed &packet_type
550  *	is linked into kernel lists and may not be freed until it has been
551  *	removed from the kernel lists.
552  *
553  *	This call does not sleep therefore it can not
554  *	guarantee all CPU's that are in middle of receiving packets
555  *	will see the new packet type (until the next received packet).
556  */
557 
558 void dev_add_pack(struct packet_type *pt)
559 {
560 	struct list_head *head = ptype_head(pt);
561 
562 	spin_lock(&ptype_lock);
563 	list_add_rcu(&pt->list, head);
564 	spin_unlock(&ptype_lock);
565 }
566 EXPORT_SYMBOL(dev_add_pack);
567 
568 /**
569  *	__dev_remove_pack	 - remove packet handler
570  *	@pt: packet type declaration
571  *
572  *	Remove a protocol handler that was previously added to the kernel
573  *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
574  *	from the kernel lists and can be freed or reused once this function
575  *	returns.
576  *
577  *      The packet type might still be in use by receivers
578  *	and must not be freed until after all the CPU's have gone
579  *	through a quiescent state.
580  */
581 void __dev_remove_pack(struct packet_type *pt)
582 {
583 	struct list_head *head = ptype_head(pt);
584 	struct packet_type *pt1;
585 
586 	spin_lock(&ptype_lock);
587 
588 	list_for_each_entry(pt1, head, list) {
589 		if (pt == pt1) {
590 			list_del_rcu(&pt->list);
591 			goto out;
592 		}
593 	}
594 
595 	pr_warn("dev_remove_pack: %p not found\n", pt);
596 out:
597 	spin_unlock(&ptype_lock);
598 }
599 EXPORT_SYMBOL(__dev_remove_pack);
600 
601 /**
602  *	dev_remove_pack	 - remove packet handler
603  *	@pt: packet type declaration
604  *
605  *	Remove a protocol handler that was previously added to the kernel
606  *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
607  *	from the kernel lists and can be freed or reused once this function
608  *	returns.
609  *
610  *	This call sleeps to guarantee that no CPU is looking at the packet
611  *	type after return.
612  */
613 void dev_remove_pack(struct packet_type *pt)
614 {
615 	__dev_remove_pack(pt);
616 
617 	synchronize_net();
618 }
619 EXPORT_SYMBOL(dev_remove_pack);
620 
621 
622 /*******************************************************************************
623  *
624  *			    Device Interface Subroutines
625  *
626  *******************************************************************************/
627 
628 /**
629  *	dev_get_iflink	- get 'iflink' value of a interface
630  *	@dev: targeted interface
631  *
632  *	Indicates the ifindex the interface is linked to.
633  *	Physical interfaces have the same 'ifindex' and 'iflink' values.
634  */
635 
636 int dev_get_iflink(const struct net_device *dev)
637 {
638 	if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
639 		return dev->netdev_ops->ndo_get_iflink(dev);
640 
641 	return dev->ifindex;
642 }
643 EXPORT_SYMBOL(dev_get_iflink);
644 
645 /**
646  *	dev_fill_metadata_dst - Retrieve tunnel egress information.
647  *	@dev: targeted interface
648  *	@skb: The packet.
649  *
650  *	For better visibility of tunnel traffic OVS needs to retrieve
651  *	egress tunnel information for a packet. Following API allows
652  *	user to get this info.
653  */
654 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
655 {
656 	struct ip_tunnel_info *info;
657 
658 	if (!dev->netdev_ops  || !dev->netdev_ops->ndo_fill_metadata_dst)
659 		return -EINVAL;
660 
661 	info = skb_tunnel_info_unclone(skb);
662 	if (!info)
663 		return -ENOMEM;
664 	if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
665 		return -EINVAL;
666 
667 	return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
668 }
669 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
670 
671 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
672 {
673 	int k = stack->num_paths++;
674 
675 	if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
676 		return NULL;
677 
678 	return &stack->path[k];
679 }
680 
681 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
682 			  struct net_device_path_stack *stack)
683 {
684 	const struct net_device *last_dev;
685 	struct net_device_path_ctx ctx = {
686 		.dev	= dev,
687 	};
688 	struct net_device_path *path;
689 	int ret = 0;
690 
691 	memcpy(ctx.daddr, daddr, sizeof(ctx.daddr));
692 	stack->num_paths = 0;
693 	while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
694 		last_dev = ctx.dev;
695 		path = dev_fwd_path(stack);
696 		if (!path)
697 			return -1;
698 
699 		memset(path, 0, sizeof(struct net_device_path));
700 		ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
701 		if (ret < 0)
702 			return -1;
703 
704 		if (WARN_ON_ONCE(last_dev == ctx.dev))
705 			return -1;
706 	}
707 
708 	if (!ctx.dev)
709 		return ret;
710 
711 	path = dev_fwd_path(stack);
712 	if (!path)
713 		return -1;
714 	path->type = DEV_PATH_ETHERNET;
715 	path->dev = ctx.dev;
716 
717 	return ret;
718 }
719 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
720 
721 /**
722  *	__dev_get_by_name	- find a device by its name
723  *	@net: the applicable net namespace
724  *	@name: name to find
725  *
726  *	Find an interface by name. Must be called under RTNL semaphore
727  *	or @dev_base_lock. If the name is found a pointer to the device
728  *	is returned. If the name is not found then %NULL is returned. The
729  *	reference counters are not incremented so the caller must be
730  *	careful with locks.
731  */
732 
733 struct net_device *__dev_get_by_name(struct net *net, const char *name)
734 {
735 	struct netdev_name_node *node_name;
736 
737 	node_name = netdev_name_node_lookup(net, name);
738 	return node_name ? node_name->dev : NULL;
739 }
740 EXPORT_SYMBOL(__dev_get_by_name);
741 
742 /**
743  * dev_get_by_name_rcu	- find a device by its name
744  * @net: the applicable net namespace
745  * @name: name to find
746  *
747  * Find an interface by name.
748  * If the name is found a pointer to the device is returned.
749  * If the name is not found then %NULL is returned.
750  * The reference counters are not incremented so the caller must be
751  * careful with locks. The caller must hold RCU lock.
752  */
753 
754 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
755 {
756 	struct netdev_name_node *node_name;
757 
758 	node_name = netdev_name_node_lookup_rcu(net, name);
759 	return node_name ? node_name->dev : NULL;
760 }
761 EXPORT_SYMBOL(dev_get_by_name_rcu);
762 
763 /**
764  *	dev_get_by_name		- find a device by its name
765  *	@net: the applicable net namespace
766  *	@name: name to find
767  *
768  *	Find an interface by name. This can be called from any
769  *	context and does its own locking. The returned handle has
770  *	the usage count incremented and the caller must use dev_put() to
771  *	release it when it is no longer needed. %NULL is returned if no
772  *	matching device is found.
773  */
774 
775 struct net_device *dev_get_by_name(struct net *net, const char *name)
776 {
777 	struct net_device *dev;
778 
779 	rcu_read_lock();
780 	dev = dev_get_by_name_rcu(net, name);
781 	dev_hold(dev);
782 	rcu_read_unlock();
783 	return dev;
784 }
785 EXPORT_SYMBOL(dev_get_by_name);
786 
787 /**
788  *	__dev_get_by_index - find a device by its ifindex
789  *	@net: the applicable net namespace
790  *	@ifindex: index of device
791  *
792  *	Search for an interface by index. Returns %NULL if the device
793  *	is not found or a pointer to the device. The device has not
794  *	had its reference counter increased so the caller must be careful
795  *	about locking. The caller must hold either the RTNL semaphore
796  *	or @dev_base_lock.
797  */
798 
799 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
800 {
801 	struct net_device *dev;
802 	struct hlist_head *head = dev_index_hash(net, ifindex);
803 
804 	hlist_for_each_entry(dev, head, index_hlist)
805 		if (dev->ifindex == ifindex)
806 			return dev;
807 
808 	return NULL;
809 }
810 EXPORT_SYMBOL(__dev_get_by_index);
811 
812 /**
813  *	dev_get_by_index_rcu - find a device by its ifindex
814  *	@net: the applicable net namespace
815  *	@ifindex: index of device
816  *
817  *	Search for an interface by index. Returns %NULL if the device
818  *	is not found or a pointer to the device. The device has not
819  *	had its reference counter increased so the caller must be careful
820  *	about locking. The caller must hold RCU lock.
821  */
822 
823 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
824 {
825 	struct net_device *dev;
826 	struct hlist_head *head = dev_index_hash(net, ifindex);
827 
828 	hlist_for_each_entry_rcu(dev, head, index_hlist)
829 		if (dev->ifindex == ifindex)
830 			return dev;
831 
832 	return NULL;
833 }
834 EXPORT_SYMBOL(dev_get_by_index_rcu);
835 
836 
837 /**
838  *	dev_get_by_index - find a device by its ifindex
839  *	@net: the applicable net namespace
840  *	@ifindex: index of device
841  *
842  *	Search for an interface by index. Returns NULL if the device
843  *	is not found or a pointer to the device. The device returned has
844  *	had a reference added and the pointer is safe until the user calls
845  *	dev_put to indicate they have finished with it.
846  */
847 
848 struct net_device *dev_get_by_index(struct net *net, int ifindex)
849 {
850 	struct net_device *dev;
851 
852 	rcu_read_lock();
853 	dev = dev_get_by_index_rcu(net, ifindex);
854 	dev_hold(dev);
855 	rcu_read_unlock();
856 	return dev;
857 }
858 EXPORT_SYMBOL(dev_get_by_index);
859 
860 /**
861  *	dev_get_by_napi_id - find a device by napi_id
862  *	@napi_id: ID of the NAPI struct
863  *
864  *	Search for an interface by NAPI ID. Returns %NULL if the device
865  *	is not found or a pointer to the device. The device has not had
866  *	its reference counter increased so the caller must be careful
867  *	about locking. The caller must hold RCU lock.
868  */
869 
870 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
871 {
872 	struct napi_struct *napi;
873 
874 	WARN_ON_ONCE(!rcu_read_lock_held());
875 
876 	if (napi_id < MIN_NAPI_ID)
877 		return NULL;
878 
879 	napi = napi_by_id(napi_id);
880 
881 	return napi ? napi->dev : NULL;
882 }
883 EXPORT_SYMBOL(dev_get_by_napi_id);
884 
885 /**
886  *	netdev_get_name - get a netdevice name, knowing its ifindex.
887  *	@net: network namespace
888  *	@name: a pointer to the buffer where the name will be stored.
889  *	@ifindex: the ifindex of the interface to get the name from.
890  */
891 int netdev_get_name(struct net *net, char *name, int ifindex)
892 {
893 	struct net_device *dev;
894 	int ret;
895 
896 	down_read(&devnet_rename_sem);
897 	rcu_read_lock();
898 
899 	dev = dev_get_by_index_rcu(net, ifindex);
900 	if (!dev) {
901 		ret = -ENODEV;
902 		goto out;
903 	}
904 
905 	strcpy(name, dev->name);
906 
907 	ret = 0;
908 out:
909 	rcu_read_unlock();
910 	up_read(&devnet_rename_sem);
911 	return ret;
912 }
913 
914 /**
915  *	dev_getbyhwaddr_rcu - find a device by its hardware address
916  *	@net: the applicable net namespace
917  *	@type: media type of device
918  *	@ha: hardware address
919  *
920  *	Search for an interface by MAC address. Returns NULL if the device
921  *	is not found or a pointer to the device.
922  *	The caller must hold RCU or RTNL.
923  *	The returned device has not had its ref count increased
924  *	and the caller must therefore be careful about locking
925  *
926  */
927 
928 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
929 				       const char *ha)
930 {
931 	struct net_device *dev;
932 
933 	for_each_netdev_rcu(net, dev)
934 		if (dev->type == type &&
935 		    !memcmp(dev->dev_addr, ha, dev->addr_len))
936 			return dev;
937 
938 	return NULL;
939 }
940 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
941 
942 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
943 {
944 	struct net_device *dev, *ret = NULL;
945 
946 	rcu_read_lock();
947 	for_each_netdev_rcu(net, dev)
948 		if (dev->type == type) {
949 			dev_hold(dev);
950 			ret = dev;
951 			break;
952 		}
953 	rcu_read_unlock();
954 	return ret;
955 }
956 EXPORT_SYMBOL(dev_getfirstbyhwtype);
957 
958 /**
959  *	__dev_get_by_flags - find any device with given flags
960  *	@net: the applicable net namespace
961  *	@if_flags: IFF_* values
962  *	@mask: bitmask of bits in if_flags to check
963  *
964  *	Search for any interface with the given flags. Returns NULL if a device
965  *	is not found or a pointer to the device. Must be called inside
966  *	rtnl_lock(), and result refcount is unchanged.
967  */
968 
969 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
970 				      unsigned short mask)
971 {
972 	struct net_device *dev, *ret;
973 
974 	ASSERT_RTNL();
975 
976 	ret = NULL;
977 	for_each_netdev(net, dev) {
978 		if (((dev->flags ^ if_flags) & mask) == 0) {
979 			ret = dev;
980 			break;
981 		}
982 	}
983 	return ret;
984 }
985 EXPORT_SYMBOL(__dev_get_by_flags);
986 
987 /**
988  *	dev_valid_name - check if name is okay for network device
989  *	@name: name string
990  *
991  *	Network device names need to be valid file names to
992  *	allow sysfs to work.  We also disallow any kind of
993  *	whitespace.
994  */
995 bool dev_valid_name(const char *name)
996 {
997 	if (*name == '\0')
998 		return false;
999 	if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1000 		return false;
1001 	if (!strcmp(name, ".") || !strcmp(name, ".."))
1002 		return false;
1003 
1004 	while (*name) {
1005 		if (*name == '/' || *name == ':' || isspace(*name))
1006 			return false;
1007 		name++;
1008 	}
1009 	return true;
1010 }
1011 EXPORT_SYMBOL(dev_valid_name);
1012 
1013 /**
1014  *	__dev_alloc_name - allocate a name for a device
1015  *	@net: network namespace to allocate the device name in
1016  *	@name: name format string
1017  *	@buf:  scratch buffer and result name string
1018  *
1019  *	Passed a format string - eg "lt%d" it will try and find a suitable
1020  *	id. It scans list of devices to build up a free map, then chooses
1021  *	the first empty slot. The caller must hold the dev_base or rtnl lock
1022  *	while allocating the name and adding the device in order to avoid
1023  *	duplicates.
1024  *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1025  *	Returns the number of the unit assigned or a negative errno code.
1026  */
1027 
1028 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1029 {
1030 	int i = 0;
1031 	const char *p;
1032 	const int max_netdevices = 8*PAGE_SIZE;
1033 	unsigned long *inuse;
1034 	struct net_device *d;
1035 
1036 	if (!dev_valid_name(name))
1037 		return -EINVAL;
1038 
1039 	p = strchr(name, '%');
1040 	if (p) {
1041 		/*
1042 		 * Verify the string as this thing may have come from
1043 		 * the user.  There must be either one "%d" and no other "%"
1044 		 * characters.
1045 		 */
1046 		if (p[1] != 'd' || strchr(p + 2, '%'))
1047 			return -EINVAL;
1048 
1049 		/* Use one page as a bit array of possible slots */
1050 		inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1051 		if (!inuse)
1052 			return -ENOMEM;
1053 
1054 		for_each_netdev(net, d) {
1055 			struct netdev_name_node *name_node;
1056 			list_for_each_entry(name_node, &d->name_node->list, list) {
1057 				if (!sscanf(name_node->name, name, &i))
1058 					continue;
1059 				if (i < 0 || i >= max_netdevices)
1060 					continue;
1061 
1062 				/*  avoid cases where sscanf is not exact inverse of printf */
1063 				snprintf(buf, IFNAMSIZ, name, i);
1064 				if (!strncmp(buf, name_node->name, IFNAMSIZ))
1065 					__set_bit(i, inuse);
1066 			}
1067 			if (!sscanf(d->name, name, &i))
1068 				continue;
1069 			if (i < 0 || i >= max_netdevices)
1070 				continue;
1071 
1072 			/*  avoid cases where sscanf is not exact inverse of printf */
1073 			snprintf(buf, IFNAMSIZ, name, i);
1074 			if (!strncmp(buf, d->name, IFNAMSIZ))
1075 				__set_bit(i, inuse);
1076 		}
1077 
1078 		i = find_first_zero_bit(inuse, max_netdevices);
1079 		free_page((unsigned long) inuse);
1080 	}
1081 
1082 	snprintf(buf, IFNAMSIZ, name, i);
1083 	if (!netdev_name_in_use(net, buf))
1084 		return i;
1085 
1086 	/* It is possible to run out of possible slots
1087 	 * when the name is long and there isn't enough space left
1088 	 * for the digits, or if all bits are used.
1089 	 */
1090 	return -ENFILE;
1091 }
1092 
1093 static int dev_alloc_name_ns(struct net *net,
1094 			     struct net_device *dev,
1095 			     const char *name)
1096 {
1097 	char buf[IFNAMSIZ];
1098 	int ret;
1099 
1100 	BUG_ON(!net);
1101 	ret = __dev_alloc_name(net, name, buf);
1102 	if (ret >= 0)
1103 		strscpy(dev->name, buf, IFNAMSIZ);
1104 	return ret;
1105 }
1106 
1107 /**
1108  *	dev_alloc_name - allocate a name for a device
1109  *	@dev: device
1110  *	@name: name format string
1111  *
1112  *	Passed a format string - eg "lt%d" it will try and find a suitable
1113  *	id. It scans list of devices to build up a free map, then chooses
1114  *	the first empty slot. The caller must hold the dev_base or rtnl lock
1115  *	while allocating the name and adding the device in order to avoid
1116  *	duplicates.
1117  *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1118  *	Returns the number of the unit assigned or a negative errno code.
1119  */
1120 
1121 int dev_alloc_name(struct net_device *dev, const char *name)
1122 {
1123 	return dev_alloc_name_ns(dev_net(dev), dev, name);
1124 }
1125 EXPORT_SYMBOL(dev_alloc_name);
1126 
1127 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1128 			      const char *name)
1129 {
1130 	BUG_ON(!net);
1131 
1132 	if (!dev_valid_name(name))
1133 		return -EINVAL;
1134 
1135 	if (strchr(name, '%'))
1136 		return dev_alloc_name_ns(net, dev, name);
1137 	else if (netdev_name_in_use(net, name))
1138 		return -EEXIST;
1139 	else if (dev->name != name)
1140 		strscpy(dev->name, name, IFNAMSIZ);
1141 
1142 	return 0;
1143 }
1144 
1145 /**
1146  *	dev_change_name - change name of a device
1147  *	@dev: device
1148  *	@newname: name (or format string) must be at least IFNAMSIZ
1149  *
1150  *	Change name of a device, can pass format strings "eth%d".
1151  *	for wildcarding.
1152  */
1153 int dev_change_name(struct net_device *dev, const char *newname)
1154 {
1155 	unsigned char old_assign_type;
1156 	char oldname[IFNAMSIZ];
1157 	int err = 0;
1158 	int ret;
1159 	struct net *net;
1160 
1161 	ASSERT_RTNL();
1162 	BUG_ON(!dev_net(dev));
1163 
1164 	net = dev_net(dev);
1165 
1166 	down_write(&devnet_rename_sem);
1167 
1168 	if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1169 		up_write(&devnet_rename_sem);
1170 		return 0;
1171 	}
1172 
1173 	memcpy(oldname, dev->name, IFNAMSIZ);
1174 
1175 	err = dev_get_valid_name(net, dev, newname);
1176 	if (err < 0) {
1177 		up_write(&devnet_rename_sem);
1178 		return err;
1179 	}
1180 
1181 	if (oldname[0] && !strchr(oldname, '%'))
1182 		netdev_info(dev, "renamed from %s%s\n", oldname,
1183 			    dev->flags & IFF_UP ? " (while UP)" : "");
1184 
1185 	old_assign_type = dev->name_assign_type;
1186 	dev->name_assign_type = NET_NAME_RENAMED;
1187 
1188 rollback:
1189 	ret = device_rename(&dev->dev, dev->name);
1190 	if (ret) {
1191 		memcpy(dev->name, oldname, IFNAMSIZ);
1192 		dev->name_assign_type = old_assign_type;
1193 		up_write(&devnet_rename_sem);
1194 		return ret;
1195 	}
1196 
1197 	up_write(&devnet_rename_sem);
1198 
1199 	netdev_adjacent_rename_links(dev, oldname);
1200 
1201 	write_lock(&dev_base_lock);
1202 	netdev_name_node_del(dev->name_node);
1203 	write_unlock(&dev_base_lock);
1204 
1205 	synchronize_rcu();
1206 
1207 	write_lock(&dev_base_lock);
1208 	netdev_name_node_add(net, dev->name_node);
1209 	write_unlock(&dev_base_lock);
1210 
1211 	ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1212 	ret = notifier_to_errno(ret);
1213 
1214 	if (ret) {
1215 		/* err >= 0 after dev_alloc_name() or stores the first errno */
1216 		if (err >= 0) {
1217 			err = ret;
1218 			down_write(&devnet_rename_sem);
1219 			memcpy(dev->name, oldname, IFNAMSIZ);
1220 			memcpy(oldname, newname, IFNAMSIZ);
1221 			dev->name_assign_type = old_assign_type;
1222 			old_assign_type = NET_NAME_RENAMED;
1223 			goto rollback;
1224 		} else {
1225 			netdev_err(dev, "name change rollback failed: %d\n",
1226 				   ret);
1227 		}
1228 	}
1229 
1230 	return err;
1231 }
1232 
1233 /**
1234  *	dev_set_alias - change ifalias of a device
1235  *	@dev: device
1236  *	@alias: name up to IFALIASZ
1237  *	@len: limit of bytes to copy from info
1238  *
1239  *	Set ifalias for a device,
1240  */
1241 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1242 {
1243 	struct dev_ifalias *new_alias = NULL;
1244 
1245 	if (len >= IFALIASZ)
1246 		return -EINVAL;
1247 
1248 	if (len) {
1249 		new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1250 		if (!new_alias)
1251 			return -ENOMEM;
1252 
1253 		memcpy(new_alias->ifalias, alias, len);
1254 		new_alias->ifalias[len] = 0;
1255 	}
1256 
1257 	mutex_lock(&ifalias_mutex);
1258 	new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1259 					mutex_is_locked(&ifalias_mutex));
1260 	mutex_unlock(&ifalias_mutex);
1261 
1262 	if (new_alias)
1263 		kfree_rcu(new_alias, rcuhead);
1264 
1265 	return len;
1266 }
1267 EXPORT_SYMBOL(dev_set_alias);
1268 
1269 /**
1270  *	dev_get_alias - get ifalias of a device
1271  *	@dev: device
1272  *	@name: buffer to store name of ifalias
1273  *	@len: size of buffer
1274  *
1275  *	get ifalias for a device.  Caller must make sure dev cannot go
1276  *	away,  e.g. rcu read lock or own a reference count to device.
1277  */
1278 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1279 {
1280 	const struct dev_ifalias *alias;
1281 	int ret = 0;
1282 
1283 	rcu_read_lock();
1284 	alias = rcu_dereference(dev->ifalias);
1285 	if (alias)
1286 		ret = snprintf(name, len, "%s", alias->ifalias);
1287 	rcu_read_unlock();
1288 
1289 	return ret;
1290 }
1291 
1292 /**
1293  *	netdev_features_change - device changes features
1294  *	@dev: device to cause notification
1295  *
1296  *	Called to indicate a device has changed features.
1297  */
1298 void netdev_features_change(struct net_device *dev)
1299 {
1300 	call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1301 }
1302 EXPORT_SYMBOL(netdev_features_change);
1303 
1304 /**
1305  *	netdev_state_change - device changes state
1306  *	@dev: device to cause notification
1307  *
1308  *	Called to indicate a device has changed state. This function calls
1309  *	the notifier chains for netdev_chain and sends a NEWLINK message
1310  *	to the routing socket.
1311  */
1312 void netdev_state_change(struct net_device *dev)
1313 {
1314 	if (dev->flags & IFF_UP) {
1315 		struct netdev_notifier_change_info change_info = {
1316 			.info.dev = dev,
1317 		};
1318 
1319 		call_netdevice_notifiers_info(NETDEV_CHANGE,
1320 					      &change_info.info);
1321 		rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL, 0, NULL);
1322 	}
1323 }
1324 EXPORT_SYMBOL(netdev_state_change);
1325 
1326 /**
1327  * __netdev_notify_peers - notify network peers about existence of @dev,
1328  * to be called when rtnl lock is already held.
1329  * @dev: network device
1330  *
1331  * Generate traffic such that interested network peers are aware of
1332  * @dev, such as by generating a gratuitous ARP. This may be used when
1333  * a device wants to inform the rest of the network about some sort of
1334  * reconfiguration such as a failover event or virtual machine
1335  * migration.
1336  */
1337 void __netdev_notify_peers(struct net_device *dev)
1338 {
1339 	ASSERT_RTNL();
1340 	call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1341 	call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1342 }
1343 EXPORT_SYMBOL(__netdev_notify_peers);
1344 
1345 /**
1346  * netdev_notify_peers - notify network peers about existence of @dev
1347  * @dev: network device
1348  *
1349  * Generate traffic such that interested network peers are aware of
1350  * @dev, such as by generating a gratuitous ARP. This may be used when
1351  * a device wants to inform the rest of the network about some sort of
1352  * reconfiguration such as a failover event or virtual machine
1353  * migration.
1354  */
1355 void netdev_notify_peers(struct net_device *dev)
1356 {
1357 	rtnl_lock();
1358 	__netdev_notify_peers(dev);
1359 	rtnl_unlock();
1360 }
1361 EXPORT_SYMBOL(netdev_notify_peers);
1362 
1363 static int napi_threaded_poll(void *data);
1364 
1365 static int napi_kthread_create(struct napi_struct *n)
1366 {
1367 	int err = 0;
1368 
1369 	/* Create and wake up the kthread once to put it in
1370 	 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1371 	 * warning and work with loadavg.
1372 	 */
1373 	n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1374 				n->dev->name, n->napi_id);
1375 	if (IS_ERR(n->thread)) {
1376 		err = PTR_ERR(n->thread);
1377 		pr_err("kthread_run failed with err %d\n", err);
1378 		n->thread = NULL;
1379 	}
1380 
1381 	return err;
1382 }
1383 
1384 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1385 {
1386 	const struct net_device_ops *ops = dev->netdev_ops;
1387 	int ret;
1388 
1389 	ASSERT_RTNL();
1390 	dev_addr_check(dev);
1391 
1392 	if (!netif_device_present(dev)) {
1393 		/* may be detached because parent is runtime-suspended */
1394 		if (dev->dev.parent)
1395 			pm_runtime_resume(dev->dev.parent);
1396 		if (!netif_device_present(dev))
1397 			return -ENODEV;
1398 	}
1399 
1400 	/* Block netpoll from trying to do any rx path servicing.
1401 	 * If we don't do this there is a chance ndo_poll_controller
1402 	 * or ndo_poll may be running while we open the device
1403 	 */
1404 	netpoll_poll_disable(dev);
1405 
1406 	ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1407 	ret = notifier_to_errno(ret);
1408 	if (ret)
1409 		return ret;
1410 
1411 	set_bit(__LINK_STATE_START, &dev->state);
1412 
1413 	if (ops->ndo_validate_addr)
1414 		ret = ops->ndo_validate_addr(dev);
1415 
1416 	if (!ret && ops->ndo_open)
1417 		ret = ops->ndo_open(dev);
1418 
1419 	netpoll_poll_enable(dev);
1420 
1421 	if (ret)
1422 		clear_bit(__LINK_STATE_START, &dev->state);
1423 	else {
1424 		dev->flags |= IFF_UP;
1425 		dev_set_rx_mode(dev);
1426 		dev_activate(dev);
1427 		add_device_randomness(dev->dev_addr, dev->addr_len);
1428 	}
1429 
1430 	return ret;
1431 }
1432 
1433 /**
1434  *	dev_open	- prepare an interface for use.
1435  *	@dev: device to open
1436  *	@extack: netlink extended ack
1437  *
1438  *	Takes a device from down to up state. The device's private open
1439  *	function is invoked and then the multicast lists are loaded. Finally
1440  *	the device is moved into the up state and a %NETDEV_UP message is
1441  *	sent to the netdev notifier chain.
1442  *
1443  *	Calling this function on an active interface is a nop. On a failure
1444  *	a negative errno code is returned.
1445  */
1446 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1447 {
1448 	int ret;
1449 
1450 	if (dev->flags & IFF_UP)
1451 		return 0;
1452 
1453 	ret = __dev_open(dev, extack);
1454 	if (ret < 0)
1455 		return ret;
1456 
1457 	rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1458 	call_netdevice_notifiers(NETDEV_UP, dev);
1459 
1460 	return ret;
1461 }
1462 EXPORT_SYMBOL(dev_open);
1463 
1464 static void __dev_close_many(struct list_head *head)
1465 {
1466 	struct net_device *dev;
1467 
1468 	ASSERT_RTNL();
1469 	might_sleep();
1470 
1471 	list_for_each_entry(dev, head, close_list) {
1472 		/* Temporarily disable netpoll until the interface is down */
1473 		netpoll_poll_disable(dev);
1474 
1475 		call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1476 
1477 		clear_bit(__LINK_STATE_START, &dev->state);
1478 
1479 		/* Synchronize to scheduled poll. We cannot touch poll list, it
1480 		 * can be even on different cpu. So just clear netif_running().
1481 		 *
1482 		 * dev->stop() will invoke napi_disable() on all of it's
1483 		 * napi_struct instances on this device.
1484 		 */
1485 		smp_mb__after_atomic(); /* Commit netif_running(). */
1486 	}
1487 
1488 	dev_deactivate_many(head);
1489 
1490 	list_for_each_entry(dev, head, close_list) {
1491 		const struct net_device_ops *ops = dev->netdev_ops;
1492 
1493 		/*
1494 		 *	Call the device specific close. This cannot fail.
1495 		 *	Only if device is UP
1496 		 *
1497 		 *	We allow it to be called even after a DETACH hot-plug
1498 		 *	event.
1499 		 */
1500 		if (ops->ndo_stop)
1501 			ops->ndo_stop(dev);
1502 
1503 		dev->flags &= ~IFF_UP;
1504 		netpoll_poll_enable(dev);
1505 	}
1506 }
1507 
1508 static void __dev_close(struct net_device *dev)
1509 {
1510 	LIST_HEAD(single);
1511 
1512 	list_add(&dev->close_list, &single);
1513 	__dev_close_many(&single);
1514 	list_del(&single);
1515 }
1516 
1517 void dev_close_many(struct list_head *head, bool unlink)
1518 {
1519 	struct net_device *dev, *tmp;
1520 
1521 	/* Remove the devices that don't need to be closed */
1522 	list_for_each_entry_safe(dev, tmp, head, close_list)
1523 		if (!(dev->flags & IFF_UP))
1524 			list_del_init(&dev->close_list);
1525 
1526 	__dev_close_many(head);
1527 
1528 	list_for_each_entry_safe(dev, tmp, head, close_list) {
1529 		rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP | IFF_RUNNING, GFP_KERNEL, 0, NULL);
1530 		call_netdevice_notifiers(NETDEV_DOWN, dev);
1531 		if (unlink)
1532 			list_del_init(&dev->close_list);
1533 	}
1534 }
1535 EXPORT_SYMBOL(dev_close_many);
1536 
1537 /**
1538  *	dev_close - shutdown an interface.
1539  *	@dev: device to shutdown
1540  *
1541  *	This function moves an active device into down state. A
1542  *	%NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1543  *	is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1544  *	chain.
1545  */
1546 void dev_close(struct net_device *dev)
1547 {
1548 	if (dev->flags & IFF_UP) {
1549 		LIST_HEAD(single);
1550 
1551 		list_add(&dev->close_list, &single);
1552 		dev_close_many(&single, true);
1553 		list_del(&single);
1554 	}
1555 }
1556 EXPORT_SYMBOL(dev_close);
1557 
1558 
1559 /**
1560  *	dev_disable_lro - disable Large Receive Offload on a device
1561  *	@dev: device
1562  *
1563  *	Disable Large Receive Offload (LRO) on a net device.  Must be
1564  *	called under RTNL.  This is needed if received packets may be
1565  *	forwarded to another interface.
1566  */
1567 void dev_disable_lro(struct net_device *dev)
1568 {
1569 	struct net_device *lower_dev;
1570 	struct list_head *iter;
1571 
1572 	dev->wanted_features &= ~NETIF_F_LRO;
1573 	netdev_update_features(dev);
1574 
1575 	if (unlikely(dev->features & NETIF_F_LRO))
1576 		netdev_WARN(dev, "failed to disable LRO!\n");
1577 
1578 	netdev_for_each_lower_dev(dev, lower_dev, iter)
1579 		dev_disable_lro(lower_dev);
1580 }
1581 EXPORT_SYMBOL(dev_disable_lro);
1582 
1583 /**
1584  *	dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1585  *	@dev: device
1586  *
1587  *	Disable HW Generic Receive Offload (GRO_HW) on a net device.  Must be
1588  *	called under RTNL.  This is needed if Generic XDP is installed on
1589  *	the device.
1590  */
1591 static void dev_disable_gro_hw(struct net_device *dev)
1592 {
1593 	dev->wanted_features &= ~NETIF_F_GRO_HW;
1594 	netdev_update_features(dev);
1595 
1596 	if (unlikely(dev->features & NETIF_F_GRO_HW))
1597 		netdev_WARN(dev, "failed to disable GRO_HW!\n");
1598 }
1599 
1600 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1601 {
1602 #define N(val) 						\
1603 	case NETDEV_##val:				\
1604 		return "NETDEV_" __stringify(val);
1605 	switch (cmd) {
1606 	N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1607 	N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1608 	N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1609 	N(POST_INIT) N(PRE_UNINIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN)
1610 	N(CHANGEUPPER) N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA)
1611 	N(BONDING_INFO) N(PRECHANGEUPPER) N(CHANGELOWERSTATE)
1612 	N(UDP_TUNNEL_PUSH_INFO) N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1613 	N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1614 	N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1615 	N(PRE_CHANGEADDR) N(OFFLOAD_XSTATS_ENABLE) N(OFFLOAD_XSTATS_DISABLE)
1616 	N(OFFLOAD_XSTATS_REPORT_USED) N(OFFLOAD_XSTATS_REPORT_DELTA)
1617 	}
1618 #undef N
1619 	return "UNKNOWN_NETDEV_EVENT";
1620 }
1621 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1622 
1623 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1624 				   struct net_device *dev)
1625 {
1626 	struct netdev_notifier_info info = {
1627 		.dev = dev,
1628 	};
1629 
1630 	return nb->notifier_call(nb, val, &info);
1631 }
1632 
1633 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1634 					     struct net_device *dev)
1635 {
1636 	int err;
1637 
1638 	err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1639 	err = notifier_to_errno(err);
1640 	if (err)
1641 		return err;
1642 
1643 	if (!(dev->flags & IFF_UP))
1644 		return 0;
1645 
1646 	call_netdevice_notifier(nb, NETDEV_UP, dev);
1647 	return 0;
1648 }
1649 
1650 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1651 						struct net_device *dev)
1652 {
1653 	if (dev->flags & IFF_UP) {
1654 		call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1655 					dev);
1656 		call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1657 	}
1658 	call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1659 }
1660 
1661 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1662 						 struct net *net)
1663 {
1664 	struct net_device *dev;
1665 	int err;
1666 
1667 	for_each_netdev(net, dev) {
1668 		err = call_netdevice_register_notifiers(nb, dev);
1669 		if (err)
1670 			goto rollback;
1671 	}
1672 	return 0;
1673 
1674 rollback:
1675 	for_each_netdev_continue_reverse(net, dev)
1676 		call_netdevice_unregister_notifiers(nb, dev);
1677 	return err;
1678 }
1679 
1680 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1681 						    struct net *net)
1682 {
1683 	struct net_device *dev;
1684 
1685 	for_each_netdev(net, dev)
1686 		call_netdevice_unregister_notifiers(nb, dev);
1687 }
1688 
1689 static int dev_boot_phase = 1;
1690 
1691 /**
1692  * register_netdevice_notifier - register a network notifier block
1693  * @nb: notifier
1694  *
1695  * Register a notifier to be called when network device events occur.
1696  * The notifier passed is linked into the kernel structures and must
1697  * not be reused until it has been unregistered. A negative errno code
1698  * is returned on a failure.
1699  *
1700  * When registered all registration and up events are replayed
1701  * to the new notifier to allow device to have a race free
1702  * view of the network device list.
1703  */
1704 
1705 int register_netdevice_notifier(struct notifier_block *nb)
1706 {
1707 	struct net *net;
1708 	int err;
1709 
1710 	/* Close race with setup_net() and cleanup_net() */
1711 	down_write(&pernet_ops_rwsem);
1712 	rtnl_lock();
1713 	err = raw_notifier_chain_register(&netdev_chain, nb);
1714 	if (err)
1715 		goto unlock;
1716 	if (dev_boot_phase)
1717 		goto unlock;
1718 	for_each_net(net) {
1719 		err = call_netdevice_register_net_notifiers(nb, net);
1720 		if (err)
1721 			goto rollback;
1722 	}
1723 
1724 unlock:
1725 	rtnl_unlock();
1726 	up_write(&pernet_ops_rwsem);
1727 	return err;
1728 
1729 rollback:
1730 	for_each_net_continue_reverse(net)
1731 		call_netdevice_unregister_net_notifiers(nb, net);
1732 
1733 	raw_notifier_chain_unregister(&netdev_chain, nb);
1734 	goto unlock;
1735 }
1736 EXPORT_SYMBOL(register_netdevice_notifier);
1737 
1738 /**
1739  * unregister_netdevice_notifier - unregister a network notifier block
1740  * @nb: notifier
1741  *
1742  * Unregister a notifier previously registered by
1743  * register_netdevice_notifier(). The notifier is unlinked into the
1744  * kernel structures and may then be reused. A negative errno code
1745  * is returned on a failure.
1746  *
1747  * After unregistering unregister and down device events are synthesized
1748  * for all devices on the device list to the removed notifier to remove
1749  * the need for special case cleanup code.
1750  */
1751 
1752 int unregister_netdevice_notifier(struct notifier_block *nb)
1753 {
1754 	struct net *net;
1755 	int err;
1756 
1757 	/* Close race with setup_net() and cleanup_net() */
1758 	down_write(&pernet_ops_rwsem);
1759 	rtnl_lock();
1760 	err = raw_notifier_chain_unregister(&netdev_chain, nb);
1761 	if (err)
1762 		goto unlock;
1763 
1764 	for_each_net(net)
1765 		call_netdevice_unregister_net_notifiers(nb, net);
1766 
1767 unlock:
1768 	rtnl_unlock();
1769 	up_write(&pernet_ops_rwsem);
1770 	return err;
1771 }
1772 EXPORT_SYMBOL(unregister_netdevice_notifier);
1773 
1774 static int __register_netdevice_notifier_net(struct net *net,
1775 					     struct notifier_block *nb,
1776 					     bool ignore_call_fail)
1777 {
1778 	int err;
1779 
1780 	err = raw_notifier_chain_register(&net->netdev_chain, nb);
1781 	if (err)
1782 		return err;
1783 	if (dev_boot_phase)
1784 		return 0;
1785 
1786 	err = call_netdevice_register_net_notifiers(nb, net);
1787 	if (err && !ignore_call_fail)
1788 		goto chain_unregister;
1789 
1790 	return 0;
1791 
1792 chain_unregister:
1793 	raw_notifier_chain_unregister(&net->netdev_chain, nb);
1794 	return err;
1795 }
1796 
1797 static int __unregister_netdevice_notifier_net(struct net *net,
1798 					       struct notifier_block *nb)
1799 {
1800 	int err;
1801 
1802 	err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1803 	if (err)
1804 		return err;
1805 
1806 	call_netdevice_unregister_net_notifiers(nb, net);
1807 	return 0;
1808 }
1809 
1810 /**
1811  * register_netdevice_notifier_net - register a per-netns network notifier block
1812  * @net: network namespace
1813  * @nb: notifier
1814  *
1815  * Register a notifier to be called when network device events occur.
1816  * The notifier passed is linked into the kernel structures and must
1817  * not be reused until it has been unregistered. A negative errno code
1818  * is returned on a failure.
1819  *
1820  * When registered all registration and up events are replayed
1821  * to the new notifier to allow device to have a race free
1822  * view of the network device list.
1823  */
1824 
1825 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1826 {
1827 	int err;
1828 
1829 	rtnl_lock();
1830 	err = __register_netdevice_notifier_net(net, nb, false);
1831 	rtnl_unlock();
1832 	return err;
1833 }
1834 EXPORT_SYMBOL(register_netdevice_notifier_net);
1835 
1836 /**
1837  * unregister_netdevice_notifier_net - unregister a per-netns
1838  *                                     network notifier block
1839  * @net: network namespace
1840  * @nb: notifier
1841  *
1842  * Unregister a notifier previously registered by
1843  * register_netdevice_notifier(). The notifier is unlinked into the
1844  * kernel structures and may then be reused. A negative errno code
1845  * is returned on a failure.
1846  *
1847  * After unregistering unregister and down device events are synthesized
1848  * for all devices on the device list to the removed notifier to remove
1849  * the need for special case cleanup code.
1850  */
1851 
1852 int unregister_netdevice_notifier_net(struct net *net,
1853 				      struct notifier_block *nb)
1854 {
1855 	int err;
1856 
1857 	rtnl_lock();
1858 	err = __unregister_netdevice_notifier_net(net, nb);
1859 	rtnl_unlock();
1860 	return err;
1861 }
1862 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1863 
1864 static void __move_netdevice_notifier_net(struct net *src_net,
1865 					  struct net *dst_net,
1866 					  struct notifier_block *nb)
1867 {
1868 	__unregister_netdevice_notifier_net(src_net, nb);
1869 	__register_netdevice_notifier_net(dst_net, nb, true);
1870 }
1871 
1872 int register_netdevice_notifier_dev_net(struct net_device *dev,
1873 					struct notifier_block *nb,
1874 					struct netdev_net_notifier *nn)
1875 {
1876 	int err;
1877 
1878 	rtnl_lock();
1879 	err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1880 	if (!err) {
1881 		nn->nb = nb;
1882 		list_add(&nn->list, &dev->net_notifier_list);
1883 	}
1884 	rtnl_unlock();
1885 	return err;
1886 }
1887 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1888 
1889 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1890 					  struct notifier_block *nb,
1891 					  struct netdev_net_notifier *nn)
1892 {
1893 	int err;
1894 
1895 	rtnl_lock();
1896 	list_del(&nn->list);
1897 	err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1898 	rtnl_unlock();
1899 	return err;
1900 }
1901 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1902 
1903 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1904 					     struct net *net)
1905 {
1906 	struct netdev_net_notifier *nn;
1907 
1908 	list_for_each_entry(nn, &dev->net_notifier_list, list)
1909 		__move_netdevice_notifier_net(dev_net(dev), net, nn->nb);
1910 }
1911 
1912 /**
1913  *	call_netdevice_notifiers_info - call all network notifier blocks
1914  *	@val: value passed unmodified to notifier function
1915  *	@info: notifier information data
1916  *
1917  *	Call all network notifier blocks.  Parameters and return value
1918  *	are as for raw_notifier_call_chain().
1919  */
1920 
1921 static int call_netdevice_notifiers_info(unsigned long val,
1922 					 struct netdev_notifier_info *info)
1923 {
1924 	struct net *net = dev_net(info->dev);
1925 	int ret;
1926 
1927 	ASSERT_RTNL();
1928 
1929 	/* Run per-netns notifier block chain first, then run the global one.
1930 	 * Hopefully, one day, the global one is going to be removed after
1931 	 * all notifier block registrators get converted to be per-netns.
1932 	 */
1933 	ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1934 	if (ret & NOTIFY_STOP_MASK)
1935 		return ret;
1936 	return raw_notifier_call_chain(&netdev_chain, val, info);
1937 }
1938 
1939 /**
1940  *	call_netdevice_notifiers_info_robust - call per-netns notifier blocks
1941  *	                                       for and rollback on error
1942  *	@val_up: value passed unmodified to notifier function
1943  *	@val_down: value passed unmodified to the notifier function when
1944  *	           recovering from an error on @val_up
1945  *	@info: notifier information data
1946  *
1947  *	Call all per-netns network notifier blocks, but not notifier blocks on
1948  *	the global notifier chain. Parameters and return value are as for
1949  *	raw_notifier_call_chain_robust().
1950  */
1951 
1952 static int
1953 call_netdevice_notifiers_info_robust(unsigned long val_up,
1954 				     unsigned long val_down,
1955 				     struct netdev_notifier_info *info)
1956 {
1957 	struct net *net = dev_net(info->dev);
1958 
1959 	ASSERT_RTNL();
1960 
1961 	return raw_notifier_call_chain_robust(&net->netdev_chain,
1962 					      val_up, val_down, info);
1963 }
1964 
1965 static int call_netdevice_notifiers_extack(unsigned long val,
1966 					   struct net_device *dev,
1967 					   struct netlink_ext_ack *extack)
1968 {
1969 	struct netdev_notifier_info info = {
1970 		.dev = dev,
1971 		.extack = extack,
1972 	};
1973 
1974 	return call_netdevice_notifiers_info(val, &info);
1975 }
1976 
1977 /**
1978  *	call_netdevice_notifiers - call all network notifier blocks
1979  *      @val: value passed unmodified to notifier function
1980  *      @dev: net_device pointer passed unmodified to notifier function
1981  *
1982  *	Call all network notifier blocks.  Parameters and return value
1983  *	are as for raw_notifier_call_chain().
1984  */
1985 
1986 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1987 {
1988 	return call_netdevice_notifiers_extack(val, dev, NULL);
1989 }
1990 EXPORT_SYMBOL(call_netdevice_notifiers);
1991 
1992 /**
1993  *	call_netdevice_notifiers_mtu - call all network notifier blocks
1994  *	@val: value passed unmodified to notifier function
1995  *	@dev: net_device pointer passed unmodified to notifier function
1996  *	@arg: additional u32 argument passed to the notifier function
1997  *
1998  *	Call all network notifier blocks.  Parameters and return value
1999  *	are as for raw_notifier_call_chain().
2000  */
2001 static int call_netdevice_notifiers_mtu(unsigned long val,
2002 					struct net_device *dev, u32 arg)
2003 {
2004 	struct netdev_notifier_info_ext info = {
2005 		.info.dev = dev,
2006 		.ext.mtu = arg,
2007 	};
2008 
2009 	BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2010 
2011 	return call_netdevice_notifiers_info(val, &info.info);
2012 }
2013 
2014 #ifdef CONFIG_NET_INGRESS
2015 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2016 
2017 void net_inc_ingress_queue(void)
2018 {
2019 	static_branch_inc(&ingress_needed_key);
2020 }
2021 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2022 
2023 void net_dec_ingress_queue(void)
2024 {
2025 	static_branch_dec(&ingress_needed_key);
2026 }
2027 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2028 #endif
2029 
2030 #ifdef CONFIG_NET_EGRESS
2031 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2032 
2033 void net_inc_egress_queue(void)
2034 {
2035 	static_branch_inc(&egress_needed_key);
2036 }
2037 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2038 
2039 void net_dec_egress_queue(void)
2040 {
2041 	static_branch_dec(&egress_needed_key);
2042 }
2043 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2044 #endif
2045 
2046 DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2047 EXPORT_SYMBOL(netstamp_needed_key);
2048 #ifdef CONFIG_JUMP_LABEL
2049 static atomic_t netstamp_needed_deferred;
2050 static atomic_t netstamp_wanted;
2051 static void netstamp_clear(struct work_struct *work)
2052 {
2053 	int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2054 	int wanted;
2055 
2056 	wanted = atomic_add_return(deferred, &netstamp_wanted);
2057 	if (wanted > 0)
2058 		static_branch_enable(&netstamp_needed_key);
2059 	else
2060 		static_branch_disable(&netstamp_needed_key);
2061 }
2062 static DECLARE_WORK(netstamp_work, netstamp_clear);
2063 #endif
2064 
2065 void net_enable_timestamp(void)
2066 {
2067 #ifdef CONFIG_JUMP_LABEL
2068 	int wanted = atomic_read(&netstamp_wanted);
2069 
2070 	while (wanted > 0) {
2071 		if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted + 1))
2072 			return;
2073 	}
2074 	atomic_inc(&netstamp_needed_deferred);
2075 	schedule_work(&netstamp_work);
2076 #else
2077 	static_branch_inc(&netstamp_needed_key);
2078 #endif
2079 }
2080 EXPORT_SYMBOL(net_enable_timestamp);
2081 
2082 void net_disable_timestamp(void)
2083 {
2084 #ifdef CONFIG_JUMP_LABEL
2085 	int wanted = atomic_read(&netstamp_wanted);
2086 
2087 	while (wanted > 1) {
2088 		if (atomic_try_cmpxchg(&netstamp_wanted, &wanted, wanted - 1))
2089 			return;
2090 	}
2091 	atomic_dec(&netstamp_needed_deferred);
2092 	schedule_work(&netstamp_work);
2093 #else
2094 	static_branch_dec(&netstamp_needed_key);
2095 #endif
2096 }
2097 EXPORT_SYMBOL(net_disable_timestamp);
2098 
2099 static inline void net_timestamp_set(struct sk_buff *skb)
2100 {
2101 	skb->tstamp = 0;
2102 	skb->mono_delivery_time = 0;
2103 	if (static_branch_unlikely(&netstamp_needed_key))
2104 		skb->tstamp = ktime_get_real();
2105 }
2106 
2107 #define net_timestamp_check(COND, SKB)				\
2108 	if (static_branch_unlikely(&netstamp_needed_key)) {	\
2109 		if ((COND) && !(SKB)->tstamp)			\
2110 			(SKB)->tstamp = ktime_get_real();	\
2111 	}							\
2112 
2113 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2114 {
2115 	return __is_skb_forwardable(dev, skb, true);
2116 }
2117 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2118 
2119 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2120 			      bool check_mtu)
2121 {
2122 	int ret = ____dev_forward_skb(dev, skb, check_mtu);
2123 
2124 	if (likely(!ret)) {
2125 		skb->protocol = eth_type_trans(skb, dev);
2126 		skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2127 	}
2128 
2129 	return ret;
2130 }
2131 
2132 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2133 {
2134 	return __dev_forward_skb2(dev, skb, true);
2135 }
2136 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2137 
2138 /**
2139  * dev_forward_skb - loopback an skb to another netif
2140  *
2141  * @dev: destination network device
2142  * @skb: buffer to forward
2143  *
2144  * return values:
2145  *	NET_RX_SUCCESS	(no congestion)
2146  *	NET_RX_DROP     (packet was dropped, but freed)
2147  *
2148  * dev_forward_skb can be used for injecting an skb from the
2149  * start_xmit function of one device into the receive queue
2150  * of another device.
2151  *
2152  * The receiving device may be in another namespace, so
2153  * we have to clear all information in the skb that could
2154  * impact namespace isolation.
2155  */
2156 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2157 {
2158 	return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2159 }
2160 EXPORT_SYMBOL_GPL(dev_forward_skb);
2161 
2162 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2163 {
2164 	return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2165 }
2166 
2167 static inline int deliver_skb(struct sk_buff *skb,
2168 			      struct packet_type *pt_prev,
2169 			      struct net_device *orig_dev)
2170 {
2171 	if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2172 		return -ENOMEM;
2173 	refcount_inc(&skb->users);
2174 	return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2175 }
2176 
2177 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2178 					  struct packet_type **pt,
2179 					  struct net_device *orig_dev,
2180 					  __be16 type,
2181 					  struct list_head *ptype_list)
2182 {
2183 	struct packet_type *ptype, *pt_prev = *pt;
2184 
2185 	list_for_each_entry_rcu(ptype, ptype_list, list) {
2186 		if (ptype->type != type)
2187 			continue;
2188 		if (pt_prev)
2189 			deliver_skb(skb, pt_prev, orig_dev);
2190 		pt_prev = ptype;
2191 	}
2192 	*pt = pt_prev;
2193 }
2194 
2195 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2196 {
2197 	if (!ptype->af_packet_priv || !skb->sk)
2198 		return false;
2199 
2200 	if (ptype->id_match)
2201 		return ptype->id_match(ptype, skb->sk);
2202 	else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2203 		return true;
2204 
2205 	return false;
2206 }
2207 
2208 /**
2209  * dev_nit_active - return true if any network interface taps are in use
2210  *
2211  * @dev: network device to check for the presence of taps
2212  */
2213 bool dev_nit_active(struct net_device *dev)
2214 {
2215 	return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2216 }
2217 EXPORT_SYMBOL_GPL(dev_nit_active);
2218 
2219 /*
2220  *	Support routine. Sends outgoing frames to any network
2221  *	taps currently in use.
2222  */
2223 
2224 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2225 {
2226 	struct packet_type *ptype;
2227 	struct sk_buff *skb2 = NULL;
2228 	struct packet_type *pt_prev = NULL;
2229 	struct list_head *ptype_list = &ptype_all;
2230 
2231 	rcu_read_lock();
2232 again:
2233 	list_for_each_entry_rcu(ptype, ptype_list, list) {
2234 		if (ptype->ignore_outgoing)
2235 			continue;
2236 
2237 		/* Never send packets back to the socket
2238 		 * they originated from - MvS (miquels@drinkel.ow.org)
2239 		 */
2240 		if (skb_loop_sk(ptype, skb))
2241 			continue;
2242 
2243 		if (pt_prev) {
2244 			deliver_skb(skb2, pt_prev, skb->dev);
2245 			pt_prev = ptype;
2246 			continue;
2247 		}
2248 
2249 		/* need to clone skb, done only once */
2250 		skb2 = skb_clone(skb, GFP_ATOMIC);
2251 		if (!skb2)
2252 			goto out_unlock;
2253 
2254 		net_timestamp_set(skb2);
2255 
2256 		/* skb->nh should be correctly
2257 		 * set by sender, so that the second statement is
2258 		 * just protection against buggy protocols.
2259 		 */
2260 		skb_reset_mac_header(skb2);
2261 
2262 		if (skb_network_header(skb2) < skb2->data ||
2263 		    skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2264 			net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2265 					     ntohs(skb2->protocol),
2266 					     dev->name);
2267 			skb_reset_network_header(skb2);
2268 		}
2269 
2270 		skb2->transport_header = skb2->network_header;
2271 		skb2->pkt_type = PACKET_OUTGOING;
2272 		pt_prev = ptype;
2273 	}
2274 
2275 	if (ptype_list == &ptype_all) {
2276 		ptype_list = &dev->ptype_all;
2277 		goto again;
2278 	}
2279 out_unlock:
2280 	if (pt_prev) {
2281 		if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2282 			pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2283 		else
2284 			kfree_skb(skb2);
2285 	}
2286 	rcu_read_unlock();
2287 }
2288 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2289 
2290 /**
2291  * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2292  * @dev: Network device
2293  * @txq: number of queues available
2294  *
2295  * If real_num_tx_queues is changed the tc mappings may no longer be
2296  * valid. To resolve this verify the tc mapping remains valid and if
2297  * not NULL the mapping. With no priorities mapping to this
2298  * offset/count pair it will no longer be used. In the worst case TC0
2299  * is invalid nothing can be done so disable priority mappings. If is
2300  * expected that drivers will fix this mapping if they can before
2301  * calling netif_set_real_num_tx_queues.
2302  */
2303 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2304 {
2305 	int i;
2306 	struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2307 
2308 	/* If TC0 is invalidated disable TC mapping */
2309 	if (tc->offset + tc->count > txq) {
2310 		netdev_warn(dev, "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2311 		dev->num_tc = 0;
2312 		return;
2313 	}
2314 
2315 	/* Invalidated prio to tc mappings set to TC0 */
2316 	for (i = 1; i < TC_BITMASK + 1; i++) {
2317 		int q = netdev_get_prio_tc_map(dev, i);
2318 
2319 		tc = &dev->tc_to_txq[q];
2320 		if (tc->offset + tc->count > txq) {
2321 			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",
2322 				    i, q);
2323 			netdev_set_prio_tc_map(dev, i, 0);
2324 		}
2325 	}
2326 }
2327 
2328 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2329 {
2330 	if (dev->num_tc) {
2331 		struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2332 		int i;
2333 
2334 		/* walk through the TCs and see if it falls into any of them */
2335 		for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2336 			if ((txq - tc->offset) < tc->count)
2337 				return i;
2338 		}
2339 
2340 		/* didn't find it, just return -1 to indicate no match */
2341 		return -1;
2342 	}
2343 
2344 	return 0;
2345 }
2346 EXPORT_SYMBOL(netdev_txq_to_tc);
2347 
2348 #ifdef CONFIG_XPS
2349 static struct static_key xps_needed __read_mostly;
2350 static struct static_key xps_rxqs_needed __read_mostly;
2351 static DEFINE_MUTEX(xps_map_mutex);
2352 #define xmap_dereference(P)		\
2353 	rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2354 
2355 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2356 			     struct xps_dev_maps *old_maps, int tci, u16 index)
2357 {
2358 	struct xps_map *map = NULL;
2359 	int pos;
2360 
2361 	if (dev_maps)
2362 		map = xmap_dereference(dev_maps->attr_map[tci]);
2363 	if (!map)
2364 		return false;
2365 
2366 	for (pos = map->len; pos--;) {
2367 		if (map->queues[pos] != index)
2368 			continue;
2369 
2370 		if (map->len > 1) {
2371 			map->queues[pos] = map->queues[--map->len];
2372 			break;
2373 		}
2374 
2375 		if (old_maps)
2376 			RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2377 		RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2378 		kfree_rcu(map, rcu);
2379 		return false;
2380 	}
2381 
2382 	return true;
2383 }
2384 
2385 static bool remove_xps_queue_cpu(struct net_device *dev,
2386 				 struct xps_dev_maps *dev_maps,
2387 				 int cpu, u16 offset, u16 count)
2388 {
2389 	int num_tc = dev_maps->num_tc;
2390 	bool active = false;
2391 	int tci;
2392 
2393 	for (tci = cpu * num_tc; num_tc--; tci++) {
2394 		int i, j;
2395 
2396 		for (i = count, j = offset; i--; j++) {
2397 			if (!remove_xps_queue(dev_maps, NULL, tci, j))
2398 				break;
2399 		}
2400 
2401 		active |= i < 0;
2402 	}
2403 
2404 	return active;
2405 }
2406 
2407 static void reset_xps_maps(struct net_device *dev,
2408 			   struct xps_dev_maps *dev_maps,
2409 			   enum xps_map_type type)
2410 {
2411 	static_key_slow_dec_cpuslocked(&xps_needed);
2412 	if (type == XPS_RXQS)
2413 		static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2414 
2415 	RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2416 
2417 	kfree_rcu(dev_maps, rcu);
2418 }
2419 
2420 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2421 			   u16 offset, u16 count)
2422 {
2423 	struct xps_dev_maps *dev_maps;
2424 	bool active = false;
2425 	int i, j;
2426 
2427 	dev_maps = xmap_dereference(dev->xps_maps[type]);
2428 	if (!dev_maps)
2429 		return;
2430 
2431 	for (j = 0; j < dev_maps->nr_ids; j++)
2432 		active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2433 	if (!active)
2434 		reset_xps_maps(dev, dev_maps, type);
2435 
2436 	if (type == XPS_CPUS) {
2437 		for (i = offset + (count - 1); count--; i--)
2438 			netdev_queue_numa_node_write(
2439 				netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2440 	}
2441 }
2442 
2443 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2444 				   u16 count)
2445 {
2446 	if (!static_key_false(&xps_needed))
2447 		return;
2448 
2449 	cpus_read_lock();
2450 	mutex_lock(&xps_map_mutex);
2451 
2452 	if (static_key_false(&xps_rxqs_needed))
2453 		clean_xps_maps(dev, XPS_RXQS, offset, count);
2454 
2455 	clean_xps_maps(dev, XPS_CPUS, offset, count);
2456 
2457 	mutex_unlock(&xps_map_mutex);
2458 	cpus_read_unlock();
2459 }
2460 
2461 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2462 {
2463 	netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2464 }
2465 
2466 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2467 				      u16 index, bool is_rxqs_map)
2468 {
2469 	struct xps_map *new_map;
2470 	int alloc_len = XPS_MIN_MAP_ALLOC;
2471 	int i, pos;
2472 
2473 	for (pos = 0; map && pos < map->len; pos++) {
2474 		if (map->queues[pos] != index)
2475 			continue;
2476 		return map;
2477 	}
2478 
2479 	/* Need to add tx-queue to this CPU's/rx-queue's existing map */
2480 	if (map) {
2481 		if (pos < map->alloc_len)
2482 			return map;
2483 
2484 		alloc_len = map->alloc_len * 2;
2485 	}
2486 
2487 	/* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2488 	 *  map
2489 	 */
2490 	if (is_rxqs_map)
2491 		new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2492 	else
2493 		new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2494 				       cpu_to_node(attr_index));
2495 	if (!new_map)
2496 		return NULL;
2497 
2498 	for (i = 0; i < pos; i++)
2499 		new_map->queues[i] = map->queues[i];
2500 	new_map->alloc_len = alloc_len;
2501 	new_map->len = pos;
2502 
2503 	return new_map;
2504 }
2505 
2506 /* Copy xps maps at a given index */
2507 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2508 			      struct xps_dev_maps *new_dev_maps, int index,
2509 			      int tc, bool skip_tc)
2510 {
2511 	int i, tci = index * dev_maps->num_tc;
2512 	struct xps_map *map;
2513 
2514 	/* copy maps belonging to foreign traffic classes */
2515 	for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2516 		if (i == tc && skip_tc)
2517 			continue;
2518 
2519 		/* fill in the new device map from the old device map */
2520 		map = xmap_dereference(dev_maps->attr_map[tci]);
2521 		RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2522 	}
2523 }
2524 
2525 /* Must be called under cpus_read_lock */
2526 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2527 			  u16 index, enum xps_map_type type)
2528 {
2529 	struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2530 	const unsigned long *online_mask = NULL;
2531 	bool active = false, copy = false;
2532 	int i, j, tci, numa_node_id = -2;
2533 	int maps_sz, num_tc = 1, tc = 0;
2534 	struct xps_map *map, *new_map;
2535 	unsigned int nr_ids;
2536 
2537 	if (dev->num_tc) {
2538 		/* Do not allow XPS on subordinate device directly */
2539 		num_tc = dev->num_tc;
2540 		if (num_tc < 0)
2541 			return -EINVAL;
2542 
2543 		/* If queue belongs to subordinate dev use its map */
2544 		dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2545 
2546 		tc = netdev_txq_to_tc(dev, index);
2547 		if (tc < 0)
2548 			return -EINVAL;
2549 	}
2550 
2551 	mutex_lock(&xps_map_mutex);
2552 
2553 	dev_maps = xmap_dereference(dev->xps_maps[type]);
2554 	if (type == XPS_RXQS) {
2555 		maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2556 		nr_ids = dev->num_rx_queues;
2557 	} else {
2558 		maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2559 		if (num_possible_cpus() > 1)
2560 			online_mask = cpumask_bits(cpu_online_mask);
2561 		nr_ids = nr_cpu_ids;
2562 	}
2563 
2564 	if (maps_sz < L1_CACHE_BYTES)
2565 		maps_sz = L1_CACHE_BYTES;
2566 
2567 	/* The old dev_maps could be larger or smaller than the one we're
2568 	 * setting up now, as dev->num_tc or nr_ids could have been updated in
2569 	 * between. We could try to be smart, but let's be safe instead and only
2570 	 * copy foreign traffic classes if the two map sizes match.
2571 	 */
2572 	if (dev_maps &&
2573 	    dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2574 		copy = true;
2575 
2576 	/* allocate memory for queue storage */
2577 	for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2578 	     j < nr_ids;) {
2579 		if (!new_dev_maps) {
2580 			new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2581 			if (!new_dev_maps) {
2582 				mutex_unlock(&xps_map_mutex);
2583 				return -ENOMEM;
2584 			}
2585 
2586 			new_dev_maps->nr_ids = nr_ids;
2587 			new_dev_maps->num_tc = num_tc;
2588 		}
2589 
2590 		tci = j * num_tc + tc;
2591 		map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2592 
2593 		map = expand_xps_map(map, j, index, type == XPS_RXQS);
2594 		if (!map)
2595 			goto error;
2596 
2597 		RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2598 	}
2599 
2600 	if (!new_dev_maps)
2601 		goto out_no_new_maps;
2602 
2603 	if (!dev_maps) {
2604 		/* Increment static keys at most once per type */
2605 		static_key_slow_inc_cpuslocked(&xps_needed);
2606 		if (type == XPS_RXQS)
2607 			static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2608 	}
2609 
2610 	for (j = 0; j < nr_ids; j++) {
2611 		bool skip_tc = false;
2612 
2613 		tci = j * num_tc + tc;
2614 		if (netif_attr_test_mask(j, mask, nr_ids) &&
2615 		    netif_attr_test_online(j, online_mask, nr_ids)) {
2616 			/* add tx-queue to CPU/rx-queue maps */
2617 			int pos = 0;
2618 
2619 			skip_tc = true;
2620 
2621 			map = xmap_dereference(new_dev_maps->attr_map[tci]);
2622 			while ((pos < map->len) && (map->queues[pos] != index))
2623 				pos++;
2624 
2625 			if (pos == map->len)
2626 				map->queues[map->len++] = index;
2627 #ifdef CONFIG_NUMA
2628 			if (type == XPS_CPUS) {
2629 				if (numa_node_id == -2)
2630 					numa_node_id = cpu_to_node(j);
2631 				else if (numa_node_id != cpu_to_node(j))
2632 					numa_node_id = -1;
2633 			}
2634 #endif
2635 		}
2636 
2637 		if (copy)
2638 			xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2639 					  skip_tc);
2640 	}
2641 
2642 	rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2643 
2644 	/* Cleanup old maps */
2645 	if (!dev_maps)
2646 		goto out_no_old_maps;
2647 
2648 	for (j = 0; j < dev_maps->nr_ids; j++) {
2649 		for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2650 			map = xmap_dereference(dev_maps->attr_map[tci]);
2651 			if (!map)
2652 				continue;
2653 
2654 			if (copy) {
2655 				new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2656 				if (map == new_map)
2657 					continue;
2658 			}
2659 
2660 			RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2661 			kfree_rcu(map, rcu);
2662 		}
2663 	}
2664 
2665 	old_dev_maps = dev_maps;
2666 
2667 out_no_old_maps:
2668 	dev_maps = new_dev_maps;
2669 	active = true;
2670 
2671 out_no_new_maps:
2672 	if (type == XPS_CPUS)
2673 		/* update Tx queue numa node */
2674 		netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2675 					     (numa_node_id >= 0) ?
2676 					     numa_node_id : NUMA_NO_NODE);
2677 
2678 	if (!dev_maps)
2679 		goto out_no_maps;
2680 
2681 	/* removes tx-queue from unused CPUs/rx-queues */
2682 	for (j = 0; j < dev_maps->nr_ids; j++) {
2683 		tci = j * dev_maps->num_tc;
2684 
2685 		for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2686 			if (i == tc &&
2687 			    netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2688 			    netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2689 				continue;
2690 
2691 			active |= remove_xps_queue(dev_maps,
2692 						   copy ? old_dev_maps : NULL,
2693 						   tci, index);
2694 		}
2695 	}
2696 
2697 	if (old_dev_maps)
2698 		kfree_rcu(old_dev_maps, rcu);
2699 
2700 	/* free map if not active */
2701 	if (!active)
2702 		reset_xps_maps(dev, dev_maps, type);
2703 
2704 out_no_maps:
2705 	mutex_unlock(&xps_map_mutex);
2706 
2707 	return 0;
2708 error:
2709 	/* remove any maps that we added */
2710 	for (j = 0; j < nr_ids; j++) {
2711 		for (i = num_tc, tci = j * num_tc; i--; tci++) {
2712 			new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2713 			map = copy ?
2714 			      xmap_dereference(dev_maps->attr_map[tci]) :
2715 			      NULL;
2716 			if (new_map && new_map != map)
2717 				kfree(new_map);
2718 		}
2719 	}
2720 
2721 	mutex_unlock(&xps_map_mutex);
2722 
2723 	kfree(new_dev_maps);
2724 	return -ENOMEM;
2725 }
2726 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2727 
2728 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2729 			u16 index)
2730 {
2731 	int ret;
2732 
2733 	cpus_read_lock();
2734 	ret =  __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2735 	cpus_read_unlock();
2736 
2737 	return ret;
2738 }
2739 EXPORT_SYMBOL(netif_set_xps_queue);
2740 
2741 #endif
2742 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2743 {
2744 	struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2745 
2746 	/* Unbind any subordinate channels */
2747 	while (txq-- != &dev->_tx[0]) {
2748 		if (txq->sb_dev)
2749 			netdev_unbind_sb_channel(dev, txq->sb_dev);
2750 	}
2751 }
2752 
2753 void netdev_reset_tc(struct net_device *dev)
2754 {
2755 #ifdef CONFIG_XPS
2756 	netif_reset_xps_queues_gt(dev, 0);
2757 #endif
2758 	netdev_unbind_all_sb_channels(dev);
2759 
2760 	/* Reset TC configuration of device */
2761 	dev->num_tc = 0;
2762 	memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2763 	memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2764 }
2765 EXPORT_SYMBOL(netdev_reset_tc);
2766 
2767 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2768 {
2769 	if (tc >= dev->num_tc)
2770 		return -EINVAL;
2771 
2772 #ifdef CONFIG_XPS
2773 	netif_reset_xps_queues(dev, offset, count);
2774 #endif
2775 	dev->tc_to_txq[tc].count = count;
2776 	dev->tc_to_txq[tc].offset = offset;
2777 	return 0;
2778 }
2779 EXPORT_SYMBOL(netdev_set_tc_queue);
2780 
2781 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2782 {
2783 	if (num_tc > TC_MAX_QUEUE)
2784 		return -EINVAL;
2785 
2786 #ifdef CONFIG_XPS
2787 	netif_reset_xps_queues_gt(dev, 0);
2788 #endif
2789 	netdev_unbind_all_sb_channels(dev);
2790 
2791 	dev->num_tc = num_tc;
2792 	return 0;
2793 }
2794 EXPORT_SYMBOL(netdev_set_num_tc);
2795 
2796 void netdev_unbind_sb_channel(struct net_device *dev,
2797 			      struct net_device *sb_dev)
2798 {
2799 	struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2800 
2801 #ifdef CONFIG_XPS
2802 	netif_reset_xps_queues_gt(sb_dev, 0);
2803 #endif
2804 	memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2805 	memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2806 
2807 	while (txq-- != &dev->_tx[0]) {
2808 		if (txq->sb_dev == sb_dev)
2809 			txq->sb_dev = NULL;
2810 	}
2811 }
2812 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2813 
2814 int netdev_bind_sb_channel_queue(struct net_device *dev,
2815 				 struct net_device *sb_dev,
2816 				 u8 tc, u16 count, u16 offset)
2817 {
2818 	/* Make certain the sb_dev and dev are already configured */
2819 	if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2820 		return -EINVAL;
2821 
2822 	/* We cannot hand out queues we don't have */
2823 	if ((offset + count) > dev->real_num_tx_queues)
2824 		return -EINVAL;
2825 
2826 	/* Record the mapping */
2827 	sb_dev->tc_to_txq[tc].count = count;
2828 	sb_dev->tc_to_txq[tc].offset = offset;
2829 
2830 	/* Provide a way for Tx queue to find the tc_to_txq map or
2831 	 * XPS map for itself.
2832 	 */
2833 	while (count--)
2834 		netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2835 
2836 	return 0;
2837 }
2838 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2839 
2840 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2841 {
2842 	/* Do not use a multiqueue device to represent a subordinate channel */
2843 	if (netif_is_multiqueue(dev))
2844 		return -ENODEV;
2845 
2846 	/* We allow channels 1 - 32767 to be used for subordinate channels.
2847 	 * Channel 0 is meant to be "native" mode and used only to represent
2848 	 * the main root device. We allow writing 0 to reset the device back
2849 	 * to normal mode after being used as a subordinate channel.
2850 	 */
2851 	if (channel > S16_MAX)
2852 		return -EINVAL;
2853 
2854 	dev->num_tc = -channel;
2855 
2856 	return 0;
2857 }
2858 EXPORT_SYMBOL(netdev_set_sb_channel);
2859 
2860 /*
2861  * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2862  * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2863  */
2864 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2865 {
2866 	bool disabling;
2867 	int rc;
2868 
2869 	disabling = txq < dev->real_num_tx_queues;
2870 
2871 	if (txq < 1 || txq > dev->num_tx_queues)
2872 		return -EINVAL;
2873 
2874 	if (dev->reg_state == NETREG_REGISTERED ||
2875 	    dev->reg_state == NETREG_UNREGISTERING) {
2876 		ASSERT_RTNL();
2877 
2878 		rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2879 						  txq);
2880 		if (rc)
2881 			return rc;
2882 
2883 		if (dev->num_tc)
2884 			netif_setup_tc(dev, txq);
2885 
2886 		dev_qdisc_change_real_num_tx(dev, txq);
2887 
2888 		dev->real_num_tx_queues = txq;
2889 
2890 		if (disabling) {
2891 			synchronize_net();
2892 			qdisc_reset_all_tx_gt(dev, txq);
2893 #ifdef CONFIG_XPS
2894 			netif_reset_xps_queues_gt(dev, txq);
2895 #endif
2896 		}
2897 	} else {
2898 		dev->real_num_tx_queues = txq;
2899 	}
2900 
2901 	return 0;
2902 }
2903 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2904 
2905 #ifdef CONFIG_SYSFS
2906 /**
2907  *	netif_set_real_num_rx_queues - set actual number of RX queues used
2908  *	@dev: Network device
2909  *	@rxq: Actual number of RX queues
2910  *
2911  *	This must be called either with the rtnl_lock held or before
2912  *	registration of the net device.  Returns 0 on success, or a
2913  *	negative error code.  If called before registration, it always
2914  *	succeeds.
2915  */
2916 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2917 {
2918 	int rc;
2919 
2920 	if (rxq < 1 || rxq > dev->num_rx_queues)
2921 		return -EINVAL;
2922 
2923 	if (dev->reg_state == NETREG_REGISTERED) {
2924 		ASSERT_RTNL();
2925 
2926 		rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2927 						  rxq);
2928 		if (rc)
2929 			return rc;
2930 	}
2931 
2932 	dev->real_num_rx_queues = rxq;
2933 	return 0;
2934 }
2935 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2936 #endif
2937 
2938 /**
2939  *	netif_set_real_num_queues - set actual number of RX and TX queues used
2940  *	@dev: Network device
2941  *	@txq: Actual number of TX queues
2942  *	@rxq: Actual number of RX queues
2943  *
2944  *	Set the real number of both TX and RX queues.
2945  *	Does nothing if the number of queues is already correct.
2946  */
2947 int netif_set_real_num_queues(struct net_device *dev,
2948 			      unsigned int txq, unsigned int rxq)
2949 {
2950 	unsigned int old_rxq = dev->real_num_rx_queues;
2951 	int err;
2952 
2953 	if (txq < 1 || txq > dev->num_tx_queues ||
2954 	    rxq < 1 || rxq > dev->num_rx_queues)
2955 		return -EINVAL;
2956 
2957 	/* Start from increases, so the error path only does decreases -
2958 	 * decreases can't fail.
2959 	 */
2960 	if (rxq > dev->real_num_rx_queues) {
2961 		err = netif_set_real_num_rx_queues(dev, rxq);
2962 		if (err)
2963 			return err;
2964 	}
2965 	if (txq > dev->real_num_tx_queues) {
2966 		err = netif_set_real_num_tx_queues(dev, txq);
2967 		if (err)
2968 			goto undo_rx;
2969 	}
2970 	if (rxq < dev->real_num_rx_queues)
2971 		WARN_ON(netif_set_real_num_rx_queues(dev, rxq));
2972 	if (txq < dev->real_num_tx_queues)
2973 		WARN_ON(netif_set_real_num_tx_queues(dev, txq));
2974 
2975 	return 0;
2976 undo_rx:
2977 	WARN_ON(netif_set_real_num_rx_queues(dev, old_rxq));
2978 	return err;
2979 }
2980 EXPORT_SYMBOL(netif_set_real_num_queues);
2981 
2982 /**
2983  * netif_set_tso_max_size() - set the max size of TSO frames supported
2984  * @dev:	netdev to update
2985  * @size:	max skb->len of a TSO frame
2986  *
2987  * Set the limit on the size of TSO super-frames the device can handle.
2988  * Unless explicitly set the stack will assume the value of
2989  * %GSO_LEGACY_MAX_SIZE.
2990  */
2991 void netif_set_tso_max_size(struct net_device *dev, unsigned int size)
2992 {
2993 	dev->tso_max_size = min(GSO_MAX_SIZE, size);
2994 	if (size < READ_ONCE(dev->gso_max_size))
2995 		netif_set_gso_max_size(dev, size);
2996 }
2997 EXPORT_SYMBOL(netif_set_tso_max_size);
2998 
2999 /**
3000  * netif_set_tso_max_segs() - set the max number of segs supported for TSO
3001  * @dev:	netdev to update
3002  * @segs:	max number of TCP segments
3003  *
3004  * Set the limit on the number of TCP segments the device can generate from
3005  * a single TSO super-frame.
3006  * Unless explicitly set the stack will assume the value of %GSO_MAX_SEGS.
3007  */
3008 void netif_set_tso_max_segs(struct net_device *dev, unsigned int segs)
3009 {
3010 	dev->tso_max_segs = segs;
3011 	if (segs < READ_ONCE(dev->gso_max_segs))
3012 		netif_set_gso_max_segs(dev, segs);
3013 }
3014 EXPORT_SYMBOL(netif_set_tso_max_segs);
3015 
3016 /**
3017  * netif_inherit_tso_max() - copy all TSO limits from a lower device to an upper
3018  * @to:		netdev to update
3019  * @from:	netdev from which to copy the limits
3020  */
3021 void netif_inherit_tso_max(struct net_device *to, const struct net_device *from)
3022 {
3023 	netif_set_tso_max_size(to, from->tso_max_size);
3024 	netif_set_tso_max_segs(to, from->tso_max_segs);
3025 }
3026 EXPORT_SYMBOL(netif_inherit_tso_max);
3027 
3028 /**
3029  * netif_get_num_default_rss_queues - default number of RSS queues
3030  *
3031  * Default value is the number of physical cores if there are only 1 or 2, or
3032  * divided by 2 if there are more.
3033  */
3034 int netif_get_num_default_rss_queues(void)
3035 {
3036 	cpumask_var_t cpus;
3037 	int cpu, count = 0;
3038 
3039 	if (unlikely(is_kdump_kernel() || !zalloc_cpumask_var(&cpus, GFP_KERNEL)))
3040 		return 1;
3041 
3042 	cpumask_copy(cpus, cpu_online_mask);
3043 	for_each_cpu(cpu, cpus) {
3044 		++count;
3045 		cpumask_andnot(cpus, cpus, topology_sibling_cpumask(cpu));
3046 	}
3047 	free_cpumask_var(cpus);
3048 
3049 	return count > 2 ? DIV_ROUND_UP(count, 2) : count;
3050 }
3051 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3052 
3053 static void __netif_reschedule(struct Qdisc *q)
3054 {
3055 	struct softnet_data *sd;
3056 	unsigned long flags;
3057 
3058 	local_irq_save(flags);
3059 	sd = this_cpu_ptr(&softnet_data);
3060 	q->next_sched = NULL;
3061 	*sd->output_queue_tailp = q;
3062 	sd->output_queue_tailp = &q->next_sched;
3063 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
3064 	local_irq_restore(flags);
3065 }
3066 
3067 void __netif_schedule(struct Qdisc *q)
3068 {
3069 	if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3070 		__netif_reschedule(q);
3071 }
3072 EXPORT_SYMBOL(__netif_schedule);
3073 
3074 struct dev_kfree_skb_cb {
3075 	enum skb_free_reason reason;
3076 };
3077 
3078 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3079 {
3080 	return (struct dev_kfree_skb_cb *)skb->cb;
3081 }
3082 
3083 void netif_schedule_queue(struct netdev_queue *txq)
3084 {
3085 	rcu_read_lock();
3086 	if (!netif_xmit_stopped(txq)) {
3087 		struct Qdisc *q = rcu_dereference(txq->qdisc);
3088 
3089 		__netif_schedule(q);
3090 	}
3091 	rcu_read_unlock();
3092 }
3093 EXPORT_SYMBOL(netif_schedule_queue);
3094 
3095 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3096 {
3097 	if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3098 		struct Qdisc *q;
3099 
3100 		rcu_read_lock();
3101 		q = rcu_dereference(dev_queue->qdisc);
3102 		__netif_schedule(q);
3103 		rcu_read_unlock();
3104 	}
3105 }
3106 EXPORT_SYMBOL(netif_tx_wake_queue);
3107 
3108 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3109 {
3110 	unsigned long flags;
3111 
3112 	if (unlikely(!skb))
3113 		return;
3114 
3115 	if (likely(refcount_read(&skb->users) == 1)) {
3116 		smp_rmb();
3117 		refcount_set(&skb->users, 0);
3118 	} else if (likely(!refcount_dec_and_test(&skb->users))) {
3119 		return;
3120 	}
3121 	get_kfree_skb_cb(skb)->reason = reason;
3122 	local_irq_save(flags);
3123 	skb->next = __this_cpu_read(softnet_data.completion_queue);
3124 	__this_cpu_write(softnet_data.completion_queue, skb);
3125 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
3126 	local_irq_restore(flags);
3127 }
3128 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3129 
3130 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3131 {
3132 	if (in_hardirq() || irqs_disabled())
3133 		__dev_kfree_skb_irq(skb, reason);
3134 	else
3135 		dev_kfree_skb(skb);
3136 }
3137 EXPORT_SYMBOL(__dev_kfree_skb_any);
3138 
3139 
3140 /**
3141  * netif_device_detach - mark device as removed
3142  * @dev: network device
3143  *
3144  * Mark device as removed from system and therefore no longer available.
3145  */
3146 void netif_device_detach(struct net_device *dev)
3147 {
3148 	if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3149 	    netif_running(dev)) {
3150 		netif_tx_stop_all_queues(dev);
3151 	}
3152 }
3153 EXPORT_SYMBOL(netif_device_detach);
3154 
3155 /**
3156  * netif_device_attach - mark device as attached
3157  * @dev: network device
3158  *
3159  * Mark device as attached from system and restart if needed.
3160  */
3161 void netif_device_attach(struct net_device *dev)
3162 {
3163 	if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3164 	    netif_running(dev)) {
3165 		netif_tx_wake_all_queues(dev);
3166 		__netdev_watchdog_up(dev);
3167 	}
3168 }
3169 EXPORT_SYMBOL(netif_device_attach);
3170 
3171 /*
3172  * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3173  * to be used as a distribution range.
3174  */
3175 static u16 skb_tx_hash(const struct net_device *dev,
3176 		       const struct net_device *sb_dev,
3177 		       struct sk_buff *skb)
3178 {
3179 	u32 hash;
3180 	u16 qoffset = 0;
3181 	u16 qcount = dev->real_num_tx_queues;
3182 
3183 	if (dev->num_tc) {
3184 		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3185 
3186 		qoffset = sb_dev->tc_to_txq[tc].offset;
3187 		qcount = sb_dev->tc_to_txq[tc].count;
3188 		if (unlikely(!qcount)) {
3189 			net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
3190 					     sb_dev->name, qoffset, tc);
3191 			qoffset = 0;
3192 			qcount = dev->real_num_tx_queues;
3193 		}
3194 	}
3195 
3196 	if (skb_rx_queue_recorded(skb)) {
3197 		hash = skb_get_rx_queue(skb);
3198 		if (hash >= qoffset)
3199 			hash -= qoffset;
3200 		while (unlikely(hash >= qcount))
3201 			hash -= qcount;
3202 		return hash + qoffset;
3203 	}
3204 
3205 	return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3206 }
3207 
3208 static void skb_warn_bad_offload(const struct sk_buff *skb)
3209 {
3210 	static const netdev_features_t null_features;
3211 	struct net_device *dev = skb->dev;
3212 	const char *name = "";
3213 
3214 	if (!net_ratelimit())
3215 		return;
3216 
3217 	if (dev) {
3218 		if (dev->dev.parent)
3219 			name = dev_driver_string(dev->dev.parent);
3220 		else
3221 			name = netdev_name(dev);
3222 	}
3223 	skb_dump(KERN_WARNING, skb, false);
3224 	WARN(1, "%s: caps=(%pNF, %pNF)\n",
3225 	     name, dev ? &dev->features : &null_features,
3226 	     skb->sk ? &skb->sk->sk_route_caps : &null_features);
3227 }
3228 
3229 /*
3230  * Invalidate hardware checksum when packet is to be mangled, and
3231  * complete checksum manually on outgoing path.
3232  */
3233 int skb_checksum_help(struct sk_buff *skb)
3234 {
3235 	__wsum csum;
3236 	int ret = 0, offset;
3237 
3238 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3239 		goto out_set_summed;
3240 
3241 	if (unlikely(skb_is_gso(skb))) {
3242 		skb_warn_bad_offload(skb);
3243 		return -EINVAL;
3244 	}
3245 
3246 	/* Before computing a checksum, we should make sure no frag could
3247 	 * be modified by an external entity : checksum could be wrong.
3248 	 */
3249 	if (skb_has_shared_frag(skb)) {
3250 		ret = __skb_linearize(skb);
3251 		if (ret)
3252 			goto out;
3253 	}
3254 
3255 	offset = skb_checksum_start_offset(skb);
3256 	ret = -EINVAL;
3257 	if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3258 		DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3259 		goto out;
3260 	}
3261 	csum = skb_checksum(skb, offset, skb->len - offset, 0);
3262 
3263 	offset += skb->csum_offset;
3264 	if (WARN_ON_ONCE(offset + sizeof(__sum16) > skb_headlen(skb))) {
3265 		DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false);
3266 		goto out;
3267 	}
3268 	ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3269 	if (ret)
3270 		goto out;
3271 
3272 	*(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3273 out_set_summed:
3274 	skb->ip_summed = CHECKSUM_NONE;
3275 out:
3276 	return ret;
3277 }
3278 EXPORT_SYMBOL(skb_checksum_help);
3279 
3280 int skb_crc32c_csum_help(struct sk_buff *skb)
3281 {
3282 	__le32 crc32c_csum;
3283 	int ret = 0, offset, start;
3284 
3285 	if (skb->ip_summed != CHECKSUM_PARTIAL)
3286 		goto out;
3287 
3288 	if (unlikely(skb_is_gso(skb)))
3289 		goto out;
3290 
3291 	/* Before computing a checksum, we should make sure no frag could
3292 	 * be modified by an external entity : checksum could be wrong.
3293 	 */
3294 	if (unlikely(skb_has_shared_frag(skb))) {
3295 		ret = __skb_linearize(skb);
3296 		if (ret)
3297 			goto out;
3298 	}
3299 	start = skb_checksum_start_offset(skb);
3300 	offset = start + offsetof(struct sctphdr, checksum);
3301 	if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3302 		ret = -EINVAL;
3303 		goto out;
3304 	}
3305 
3306 	ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3307 	if (ret)
3308 		goto out;
3309 
3310 	crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3311 						  skb->len - start, ~(__u32)0,
3312 						  crc32c_csum_stub));
3313 	*(__le32 *)(skb->data + offset) = crc32c_csum;
3314 	skb->ip_summed = CHECKSUM_NONE;
3315 	skb->csum_not_inet = 0;
3316 out:
3317 	return ret;
3318 }
3319 
3320 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3321 {
3322 	__be16 type = skb->protocol;
3323 
3324 	/* Tunnel gso handlers can set protocol to ethernet. */
3325 	if (type == htons(ETH_P_TEB)) {
3326 		struct ethhdr *eth;
3327 
3328 		if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3329 			return 0;
3330 
3331 		eth = (struct ethhdr *)skb->data;
3332 		type = eth->h_proto;
3333 	}
3334 
3335 	return __vlan_get_protocol(skb, type, depth);
3336 }
3337 
3338 /* openvswitch calls this on rx path, so we need a different check.
3339  */
3340 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3341 {
3342 	if (tx_path)
3343 		return skb->ip_summed != CHECKSUM_PARTIAL &&
3344 		       skb->ip_summed != CHECKSUM_UNNECESSARY;
3345 
3346 	return skb->ip_summed == CHECKSUM_NONE;
3347 }
3348 
3349 /**
3350  *	__skb_gso_segment - Perform segmentation on skb.
3351  *	@skb: buffer to segment
3352  *	@features: features for the output path (see dev->features)
3353  *	@tx_path: whether it is called in TX path
3354  *
3355  *	This function segments the given skb and returns a list of segments.
3356  *
3357  *	It may return NULL if the skb requires no segmentation.  This is
3358  *	only possible when GSO is used for verifying header integrity.
3359  *
3360  *	Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3361  */
3362 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3363 				  netdev_features_t features, bool tx_path)
3364 {
3365 	struct sk_buff *segs;
3366 
3367 	if (unlikely(skb_needs_check(skb, tx_path))) {
3368 		int err;
3369 
3370 		/* We're going to init ->check field in TCP or UDP header */
3371 		err = skb_cow_head(skb, 0);
3372 		if (err < 0)
3373 			return ERR_PTR(err);
3374 	}
3375 
3376 	/* Only report GSO partial support if it will enable us to
3377 	 * support segmentation on this frame without needing additional
3378 	 * work.
3379 	 */
3380 	if (features & NETIF_F_GSO_PARTIAL) {
3381 		netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3382 		struct net_device *dev = skb->dev;
3383 
3384 		partial_features |= dev->features & dev->gso_partial_features;
3385 		if (!skb_gso_ok(skb, features | partial_features))
3386 			features &= ~NETIF_F_GSO_PARTIAL;
3387 	}
3388 
3389 	BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3390 		     sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3391 
3392 	SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3393 	SKB_GSO_CB(skb)->encap_level = 0;
3394 
3395 	skb_reset_mac_header(skb);
3396 	skb_reset_mac_len(skb);
3397 
3398 	segs = skb_mac_gso_segment(skb, features);
3399 
3400 	if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3401 		skb_warn_bad_offload(skb);
3402 
3403 	return segs;
3404 }
3405 EXPORT_SYMBOL(__skb_gso_segment);
3406 
3407 /* Take action when hardware reception checksum errors are detected. */
3408 #ifdef CONFIG_BUG
3409 static void do_netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3410 {
3411 	netdev_err(dev, "hw csum failure\n");
3412 	skb_dump(KERN_ERR, skb, true);
3413 	dump_stack();
3414 }
3415 
3416 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3417 {
3418 	DO_ONCE_LITE(do_netdev_rx_csum_fault, dev, skb);
3419 }
3420 EXPORT_SYMBOL(netdev_rx_csum_fault);
3421 #endif
3422 
3423 /* XXX: check that highmem exists at all on the given machine. */
3424 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3425 {
3426 #ifdef CONFIG_HIGHMEM
3427 	int i;
3428 
3429 	if (!(dev->features & NETIF_F_HIGHDMA)) {
3430 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3431 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3432 
3433 			if (PageHighMem(skb_frag_page(frag)))
3434 				return 1;
3435 		}
3436 	}
3437 #endif
3438 	return 0;
3439 }
3440 
3441 /* If MPLS offload request, verify we are testing hardware MPLS features
3442  * instead of standard features for the netdev.
3443  */
3444 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3445 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3446 					   netdev_features_t features,
3447 					   __be16 type)
3448 {
3449 	if (eth_p_mpls(type))
3450 		features &= skb->dev->mpls_features;
3451 
3452 	return features;
3453 }
3454 #else
3455 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3456 					   netdev_features_t features,
3457 					   __be16 type)
3458 {
3459 	return features;
3460 }
3461 #endif
3462 
3463 static netdev_features_t harmonize_features(struct sk_buff *skb,
3464 	netdev_features_t features)
3465 {
3466 	__be16 type;
3467 
3468 	type = skb_network_protocol(skb, NULL);
3469 	features = net_mpls_features(skb, features, type);
3470 
3471 	if (skb->ip_summed != CHECKSUM_NONE &&
3472 	    !can_checksum_protocol(features, type)) {
3473 		features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3474 	}
3475 	if (illegal_highdma(skb->dev, skb))
3476 		features &= ~NETIF_F_SG;
3477 
3478 	return features;
3479 }
3480 
3481 netdev_features_t passthru_features_check(struct sk_buff *skb,
3482 					  struct net_device *dev,
3483 					  netdev_features_t features)
3484 {
3485 	return features;
3486 }
3487 EXPORT_SYMBOL(passthru_features_check);
3488 
3489 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3490 					     struct net_device *dev,
3491 					     netdev_features_t features)
3492 {
3493 	return vlan_features_check(skb, features);
3494 }
3495 
3496 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3497 					    struct net_device *dev,
3498 					    netdev_features_t features)
3499 {
3500 	u16 gso_segs = skb_shinfo(skb)->gso_segs;
3501 
3502 	if (gso_segs > READ_ONCE(dev->gso_max_segs))
3503 		return features & ~NETIF_F_GSO_MASK;
3504 
3505 	if (!skb_shinfo(skb)->gso_type) {
3506 		skb_warn_bad_offload(skb);
3507 		return features & ~NETIF_F_GSO_MASK;
3508 	}
3509 
3510 	/* Support for GSO partial features requires software
3511 	 * intervention before we can actually process the packets
3512 	 * so we need to strip support for any partial features now
3513 	 * and we can pull them back in after we have partially
3514 	 * segmented the frame.
3515 	 */
3516 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3517 		features &= ~dev->gso_partial_features;
3518 
3519 	/* Make sure to clear the IPv4 ID mangling feature if the
3520 	 * IPv4 header has the potential to be fragmented.
3521 	 */
3522 	if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3523 		struct iphdr *iph = skb->encapsulation ?
3524 				    inner_ip_hdr(skb) : ip_hdr(skb);
3525 
3526 		if (!(iph->frag_off & htons(IP_DF)))
3527 			features &= ~NETIF_F_TSO_MANGLEID;
3528 	}
3529 
3530 	return features;
3531 }
3532 
3533 netdev_features_t netif_skb_features(struct sk_buff *skb)
3534 {
3535 	struct net_device *dev = skb->dev;
3536 	netdev_features_t features = dev->features;
3537 
3538 	if (skb_is_gso(skb))
3539 		features = gso_features_check(skb, dev, features);
3540 
3541 	/* If encapsulation offload request, verify we are testing
3542 	 * hardware encapsulation features instead of standard
3543 	 * features for the netdev
3544 	 */
3545 	if (skb->encapsulation)
3546 		features &= dev->hw_enc_features;
3547 
3548 	if (skb_vlan_tagged(skb))
3549 		features = netdev_intersect_features(features,
3550 						     dev->vlan_features |
3551 						     NETIF_F_HW_VLAN_CTAG_TX |
3552 						     NETIF_F_HW_VLAN_STAG_TX);
3553 
3554 	if (dev->netdev_ops->ndo_features_check)
3555 		features &= dev->netdev_ops->ndo_features_check(skb, dev,
3556 								features);
3557 	else
3558 		features &= dflt_features_check(skb, dev, features);
3559 
3560 	return harmonize_features(skb, features);
3561 }
3562 EXPORT_SYMBOL(netif_skb_features);
3563 
3564 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3565 		    struct netdev_queue *txq, bool more)
3566 {
3567 	unsigned int len;
3568 	int rc;
3569 
3570 	if (dev_nit_active(dev))
3571 		dev_queue_xmit_nit(skb, dev);
3572 
3573 	len = skb->len;
3574 	trace_net_dev_start_xmit(skb, dev);
3575 	rc = netdev_start_xmit(skb, dev, txq, more);
3576 	trace_net_dev_xmit(skb, rc, dev, len);
3577 
3578 	return rc;
3579 }
3580 
3581 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3582 				    struct netdev_queue *txq, int *ret)
3583 {
3584 	struct sk_buff *skb = first;
3585 	int rc = NETDEV_TX_OK;
3586 
3587 	while (skb) {
3588 		struct sk_buff *next = skb->next;
3589 
3590 		skb_mark_not_on_list(skb);
3591 		rc = xmit_one(skb, dev, txq, next != NULL);
3592 		if (unlikely(!dev_xmit_complete(rc))) {
3593 			skb->next = next;
3594 			goto out;
3595 		}
3596 
3597 		skb = next;
3598 		if (netif_tx_queue_stopped(txq) && skb) {
3599 			rc = NETDEV_TX_BUSY;
3600 			break;
3601 		}
3602 	}
3603 
3604 out:
3605 	*ret = rc;
3606 	return skb;
3607 }
3608 
3609 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3610 					  netdev_features_t features)
3611 {
3612 	if (skb_vlan_tag_present(skb) &&
3613 	    !vlan_hw_offload_capable(features, skb->vlan_proto))
3614 		skb = __vlan_hwaccel_push_inside(skb);
3615 	return skb;
3616 }
3617 
3618 int skb_csum_hwoffload_help(struct sk_buff *skb,
3619 			    const netdev_features_t features)
3620 {
3621 	if (unlikely(skb_csum_is_sctp(skb)))
3622 		return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3623 			skb_crc32c_csum_help(skb);
3624 
3625 	if (features & NETIF_F_HW_CSUM)
3626 		return 0;
3627 
3628 	if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3629 		switch (skb->csum_offset) {
3630 		case offsetof(struct tcphdr, check):
3631 		case offsetof(struct udphdr, check):
3632 			return 0;
3633 		}
3634 	}
3635 
3636 	return skb_checksum_help(skb);
3637 }
3638 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3639 
3640 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3641 {
3642 	netdev_features_t features;
3643 
3644 	features = netif_skb_features(skb);
3645 	skb = validate_xmit_vlan(skb, features);
3646 	if (unlikely(!skb))
3647 		goto out_null;
3648 
3649 	skb = sk_validate_xmit_skb(skb, dev);
3650 	if (unlikely(!skb))
3651 		goto out_null;
3652 
3653 	if (netif_needs_gso(skb, features)) {
3654 		struct sk_buff *segs;
3655 
3656 		segs = skb_gso_segment(skb, features);
3657 		if (IS_ERR(segs)) {
3658 			goto out_kfree_skb;
3659 		} else if (segs) {
3660 			consume_skb(skb);
3661 			skb = segs;
3662 		}
3663 	} else {
3664 		if (skb_needs_linearize(skb, features) &&
3665 		    __skb_linearize(skb))
3666 			goto out_kfree_skb;
3667 
3668 		/* If packet is not checksummed and device does not
3669 		 * support checksumming for this protocol, complete
3670 		 * checksumming here.
3671 		 */
3672 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
3673 			if (skb->encapsulation)
3674 				skb_set_inner_transport_header(skb,
3675 							       skb_checksum_start_offset(skb));
3676 			else
3677 				skb_set_transport_header(skb,
3678 							 skb_checksum_start_offset(skb));
3679 			if (skb_csum_hwoffload_help(skb, features))
3680 				goto out_kfree_skb;
3681 		}
3682 	}
3683 
3684 	skb = validate_xmit_xfrm(skb, features, again);
3685 
3686 	return skb;
3687 
3688 out_kfree_skb:
3689 	kfree_skb(skb);
3690 out_null:
3691 	dev_core_stats_tx_dropped_inc(dev);
3692 	return NULL;
3693 }
3694 
3695 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3696 {
3697 	struct sk_buff *next, *head = NULL, *tail;
3698 
3699 	for (; skb != NULL; skb = next) {
3700 		next = skb->next;
3701 		skb_mark_not_on_list(skb);
3702 
3703 		/* in case skb wont be segmented, point to itself */
3704 		skb->prev = skb;
3705 
3706 		skb = validate_xmit_skb(skb, dev, again);
3707 		if (!skb)
3708 			continue;
3709 
3710 		if (!head)
3711 			head = skb;
3712 		else
3713 			tail->next = skb;
3714 		/* If skb was segmented, skb->prev points to
3715 		 * the last segment. If not, it still contains skb.
3716 		 */
3717 		tail = skb->prev;
3718 	}
3719 	return head;
3720 }
3721 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3722 
3723 static void qdisc_pkt_len_init(struct sk_buff *skb)
3724 {
3725 	const struct skb_shared_info *shinfo = skb_shinfo(skb);
3726 
3727 	qdisc_skb_cb(skb)->pkt_len = skb->len;
3728 
3729 	/* To get more precise estimation of bytes sent on wire,
3730 	 * we add to pkt_len the headers size of all segments
3731 	 */
3732 	if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3733 		unsigned int hdr_len;
3734 		u16 gso_segs = shinfo->gso_segs;
3735 
3736 		/* mac layer + network layer */
3737 		hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3738 
3739 		/* + transport layer */
3740 		if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3741 			const struct tcphdr *th;
3742 			struct tcphdr _tcphdr;
3743 
3744 			th = skb_header_pointer(skb, skb_transport_offset(skb),
3745 						sizeof(_tcphdr), &_tcphdr);
3746 			if (likely(th))
3747 				hdr_len += __tcp_hdrlen(th);
3748 		} else {
3749 			struct udphdr _udphdr;
3750 
3751 			if (skb_header_pointer(skb, skb_transport_offset(skb),
3752 					       sizeof(_udphdr), &_udphdr))
3753 				hdr_len += sizeof(struct udphdr);
3754 		}
3755 
3756 		if (shinfo->gso_type & SKB_GSO_DODGY)
3757 			gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3758 						shinfo->gso_size);
3759 
3760 		qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3761 	}
3762 }
3763 
3764 static int dev_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *q,
3765 			     struct sk_buff **to_free,
3766 			     struct netdev_queue *txq)
3767 {
3768 	int rc;
3769 
3770 	rc = q->enqueue(skb, q, to_free) & NET_XMIT_MASK;
3771 	if (rc == NET_XMIT_SUCCESS)
3772 		trace_qdisc_enqueue(q, txq, skb);
3773 	return rc;
3774 }
3775 
3776 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3777 				 struct net_device *dev,
3778 				 struct netdev_queue *txq)
3779 {
3780 	spinlock_t *root_lock = qdisc_lock(q);
3781 	struct sk_buff *to_free = NULL;
3782 	bool contended;
3783 	int rc;
3784 
3785 	qdisc_calculate_pkt_len(skb, q);
3786 
3787 	if (q->flags & TCQ_F_NOLOCK) {
3788 		if (q->flags & TCQ_F_CAN_BYPASS && nolock_qdisc_is_empty(q) &&
3789 		    qdisc_run_begin(q)) {
3790 			/* Retest nolock_qdisc_is_empty() within the protection
3791 			 * of q->seqlock to protect from racing with requeuing.
3792 			 */
3793 			if (unlikely(!nolock_qdisc_is_empty(q))) {
3794 				rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3795 				__qdisc_run(q);
3796 				qdisc_run_end(q);
3797 
3798 				goto no_lock_out;
3799 			}
3800 
3801 			qdisc_bstats_cpu_update(q, skb);
3802 			if (sch_direct_xmit(skb, q, dev, txq, NULL, true) &&
3803 			    !nolock_qdisc_is_empty(q))
3804 				__qdisc_run(q);
3805 
3806 			qdisc_run_end(q);
3807 			return NET_XMIT_SUCCESS;
3808 		}
3809 
3810 		rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3811 		qdisc_run(q);
3812 
3813 no_lock_out:
3814 		if (unlikely(to_free))
3815 			kfree_skb_list_reason(to_free,
3816 					      SKB_DROP_REASON_QDISC_DROP);
3817 		return rc;
3818 	}
3819 
3820 	/*
3821 	 * Heuristic to force contended enqueues to serialize on a
3822 	 * separate lock before trying to get qdisc main lock.
3823 	 * This permits qdisc->running owner to get the lock more
3824 	 * often and dequeue packets faster.
3825 	 * On PREEMPT_RT it is possible to preempt the qdisc owner during xmit
3826 	 * and then other tasks will only enqueue packets. The packets will be
3827 	 * sent after the qdisc owner is scheduled again. To prevent this
3828 	 * scenario the task always serialize on the lock.
3829 	 */
3830 	contended = qdisc_is_running(q) || IS_ENABLED(CONFIG_PREEMPT_RT);
3831 	if (unlikely(contended))
3832 		spin_lock(&q->busylock);
3833 
3834 	spin_lock(root_lock);
3835 	if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3836 		__qdisc_drop(skb, &to_free);
3837 		rc = NET_XMIT_DROP;
3838 	} else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3839 		   qdisc_run_begin(q)) {
3840 		/*
3841 		 * This is a work-conserving queue; there are no old skbs
3842 		 * waiting to be sent out; and the qdisc is not running -
3843 		 * xmit the skb directly.
3844 		 */
3845 
3846 		qdisc_bstats_update(q, skb);
3847 
3848 		if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3849 			if (unlikely(contended)) {
3850 				spin_unlock(&q->busylock);
3851 				contended = false;
3852 			}
3853 			__qdisc_run(q);
3854 		}
3855 
3856 		qdisc_run_end(q);
3857 		rc = NET_XMIT_SUCCESS;
3858 	} else {
3859 		rc = dev_qdisc_enqueue(skb, q, &to_free, txq);
3860 		if (qdisc_run_begin(q)) {
3861 			if (unlikely(contended)) {
3862 				spin_unlock(&q->busylock);
3863 				contended = false;
3864 			}
3865 			__qdisc_run(q);
3866 			qdisc_run_end(q);
3867 		}
3868 	}
3869 	spin_unlock(root_lock);
3870 	if (unlikely(to_free))
3871 		kfree_skb_list_reason(to_free, SKB_DROP_REASON_QDISC_DROP);
3872 	if (unlikely(contended))
3873 		spin_unlock(&q->busylock);
3874 	return rc;
3875 }
3876 
3877 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3878 static void skb_update_prio(struct sk_buff *skb)
3879 {
3880 	const struct netprio_map *map;
3881 	const struct sock *sk;
3882 	unsigned int prioidx;
3883 
3884 	if (skb->priority)
3885 		return;
3886 	map = rcu_dereference_bh(skb->dev->priomap);
3887 	if (!map)
3888 		return;
3889 	sk = skb_to_full_sk(skb);
3890 	if (!sk)
3891 		return;
3892 
3893 	prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3894 
3895 	if (prioidx < map->priomap_len)
3896 		skb->priority = map->priomap[prioidx];
3897 }
3898 #else
3899 #define skb_update_prio(skb)
3900 #endif
3901 
3902 /**
3903  *	dev_loopback_xmit - loop back @skb
3904  *	@net: network namespace this loopback is happening in
3905  *	@sk:  sk needed to be a netfilter okfn
3906  *	@skb: buffer to transmit
3907  */
3908 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3909 {
3910 	skb_reset_mac_header(skb);
3911 	__skb_pull(skb, skb_network_offset(skb));
3912 	skb->pkt_type = PACKET_LOOPBACK;
3913 	if (skb->ip_summed == CHECKSUM_NONE)
3914 		skb->ip_summed = CHECKSUM_UNNECESSARY;
3915 	DEBUG_NET_WARN_ON_ONCE(!skb_dst(skb));
3916 	skb_dst_force(skb);
3917 	netif_rx(skb);
3918 	return 0;
3919 }
3920 EXPORT_SYMBOL(dev_loopback_xmit);
3921 
3922 #ifdef CONFIG_NET_EGRESS
3923 static struct sk_buff *
3924 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3925 {
3926 #ifdef CONFIG_NET_CLS_ACT
3927 	struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3928 	struct tcf_result cl_res;
3929 
3930 	if (!miniq)
3931 		return skb;
3932 
3933 	/* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3934 	tc_skb_cb(skb)->mru = 0;
3935 	tc_skb_cb(skb)->post_ct = false;
3936 	mini_qdisc_bstats_cpu_update(miniq, skb);
3937 
3938 	switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
3939 	case TC_ACT_OK:
3940 	case TC_ACT_RECLASSIFY:
3941 		skb->tc_index = TC_H_MIN(cl_res.classid);
3942 		break;
3943 	case TC_ACT_SHOT:
3944 		mini_qdisc_qstats_cpu_drop(miniq);
3945 		*ret = NET_XMIT_DROP;
3946 		kfree_skb_reason(skb, SKB_DROP_REASON_TC_EGRESS);
3947 		return NULL;
3948 	case TC_ACT_STOLEN:
3949 	case TC_ACT_QUEUED:
3950 	case TC_ACT_TRAP:
3951 		*ret = NET_XMIT_SUCCESS;
3952 		consume_skb(skb);
3953 		return NULL;
3954 	case TC_ACT_REDIRECT:
3955 		/* No need to push/pop skb's mac_header here on egress! */
3956 		skb_do_redirect(skb);
3957 		*ret = NET_XMIT_SUCCESS;
3958 		return NULL;
3959 	default:
3960 		break;
3961 	}
3962 #endif /* CONFIG_NET_CLS_ACT */
3963 
3964 	return skb;
3965 }
3966 
3967 static struct netdev_queue *
3968 netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
3969 {
3970 	int qm = skb_get_queue_mapping(skb);
3971 
3972 	return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
3973 }
3974 
3975 static bool netdev_xmit_txqueue_skipped(void)
3976 {
3977 	return __this_cpu_read(softnet_data.xmit.skip_txqueue);
3978 }
3979 
3980 void netdev_xmit_skip_txqueue(bool skip)
3981 {
3982 	__this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
3983 }
3984 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
3985 #endif /* CONFIG_NET_EGRESS */
3986 
3987 #ifdef CONFIG_XPS
3988 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3989 			       struct xps_dev_maps *dev_maps, unsigned int tci)
3990 {
3991 	int tc = netdev_get_prio_tc_map(dev, skb->priority);
3992 	struct xps_map *map;
3993 	int queue_index = -1;
3994 
3995 	if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
3996 		return queue_index;
3997 
3998 	tci *= dev_maps->num_tc;
3999 	tci += tc;
4000 
4001 	map = rcu_dereference(dev_maps->attr_map[tci]);
4002 	if (map) {
4003 		if (map->len == 1)
4004 			queue_index = map->queues[0];
4005 		else
4006 			queue_index = map->queues[reciprocal_scale(
4007 						skb_get_hash(skb), map->len)];
4008 		if (unlikely(queue_index >= dev->real_num_tx_queues))
4009 			queue_index = -1;
4010 	}
4011 	return queue_index;
4012 }
4013 #endif
4014 
4015 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4016 			 struct sk_buff *skb)
4017 {
4018 #ifdef CONFIG_XPS
4019 	struct xps_dev_maps *dev_maps;
4020 	struct sock *sk = skb->sk;
4021 	int queue_index = -1;
4022 
4023 	if (!static_key_false(&xps_needed))
4024 		return -1;
4025 
4026 	rcu_read_lock();
4027 	if (!static_key_false(&xps_rxqs_needed))
4028 		goto get_cpus_map;
4029 
4030 	dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4031 	if (dev_maps) {
4032 		int tci = sk_rx_queue_get(sk);
4033 
4034 		if (tci >= 0)
4035 			queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4036 							  tci);
4037 	}
4038 
4039 get_cpus_map:
4040 	if (queue_index < 0) {
4041 		dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4042 		if (dev_maps) {
4043 			unsigned int tci = skb->sender_cpu - 1;
4044 
4045 			queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4046 							  tci);
4047 		}
4048 	}
4049 	rcu_read_unlock();
4050 
4051 	return queue_index;
4052 #else
4053 	return -1;
4054 #endif
4055 }
4056 
4057 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4058 		     struct net_device *sb_dev)
4059 {
4060 	return 0;
4061 }
4062 EXPORT_SYMBOL(dev_pick_tx_zero);
4063 
4064 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4065 		       struct net_device *sb_dev)
4066 {
4067 	return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4068 }
4069 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4070 
4071 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4072 		     struct net_device *sb_dev)
4073 {
4074 	struct sock *sk = skb->sk;
4075 	int queue_index = sk_tx_queue_get(sk);
4076 
4077 	sb_dev = sb_dev ? : dev;
4078 
4079 	if (queue_index < 0 || skb->ooo_okay ||
4080 	    queue_index >= dev->real_num_tx_queues) {
4081 		int new_index = get_xps_queue(dev, sb_dev, skb);
4082 
4083 		if (new_index < 0)
4084 			new_index = skb_tx_hash(dev, sb_dev, skb);
4085 
4086 		if (queue_index != new_index && sk &&
4087 		    sk_fullsock(sk) &&
4088 		    rcu_access_pointer(sk->sk_dst_cache))
4089 			sk_tx_queue_set(sk, new_index);
4090 
4091 		queue_index = new_index;
4092 	}
4093 
4094 	return queue_index;
4095 }
4096 EXPORT_SYMBOL(netdev_pick_tx);
4097 
4098 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4099 					 struct sk_buff *skb,
4100 					 struct net_device *sb_dev)
4101 {
4102 	int queue_index = 0;
4103 
4104 #ifdef CONFIG_XPS
4105 	u32 sender_cpu = skb->sender_cpu - 1;
4106 
4107 	if (sender_cpu >= (u32)NR_CPUS)
4108 		skb->sender_cpu = raw_smp_processor_id() + 1;
4109 #endif
4110 
4111 	if (dev->real_num_tx_queues != 1) {
4112 		const struct net_device_ops *ops = dev->netdev_ops;
4113 
4114 		if (ops->ndo_select_queue)
4115 			queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4116 		else
4117 			queue_index = netdev_pick_tx(dev, skb, sb_dev);
4118 
4119 		queue_index = netdev_cap_txqueue(dev, queue_index);
4120 	}
4121 
4122 	skb_set_queue_mapping(skb, queue_index);
4123 	return netdev_get_tx_queue(dev, queue_index);
4124 }
4125 
4126 /**
4127  * __dev_queue_xmit() - transmit a buffer
4128  * @skb:	buffer to transmit
4129  * @sb_dev:	suboordinate device used for L2 forwarding offload
4130  *
4131  * Queue a buffer for transmission to a network device. The caller must
4132  * have set the device and priority and built the buffer before calling
4133  * this function. The function can be called from an interrupt.
4134  *
4135  * When calling this method, interrupts MUST be enabled. This is because
4136  * the BH enable code must have IRQs enabled so that it will not deadlock.
4137  *
4138  * Regardless of the return value, the skb is consumed, so it is currently
4139  * difficult to retry a send to this method. (You can bump the ref count
4140  * before sending to hold a reference for retry if you are careful.)
4141  *
4142  * Return:
4143  * * 0				- buffer successfully transmitted
4144  * * positive qdisc return code	- NET_XMIT_DROP etc.
4145  * * negative errno		- other errors
4146  */
4147 int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4148 {
4149 	struct net_device *dev = skb->dev;
4150 	struct netdev_queue *txq = NULL;
4151 	struct Qdisc *q;
4152 	int rc = -ENOMEM;
4153 	bool again = false;
4154 
4155 	skb_reset_mac_header(skb);
4156 	skb_assert_len(skb);
4157 
4158 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4159 		__skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4160 
4161 	/* Disable soft irqs for various locks below. Also
4162 	 * stops preemption for RCU.
4163 	 */
4164 	rcu_read_lock_bh();
4165 
4166 	skb_update_prio(skb);
4167 
4168 	qdisc_pkt_len_init(skb);
4169 #ifdef CONFIG_NET_CLS_ACT
4170 	skb->tc_at_ingress = 0;
4171 #endif
4172 #ifdef CONFIG_NET_EGRESS
4173 	if (static_branch_unlikely(&egress_needed_key)) {
4174 		if (nf_hook_egress_active()) {
4175 			skb = nf_hook_egress(skb, &rc, dev);
4176 			if (!skb)
4177 				goto out;
4178 		}
4179 
4180 		netdev_xmit_skip_txqueue(false);
4181 
4182 		nf_skip_egress(skb, true);
4183 		skb = sch_handle_egress(skb, &rc, dev);
4184 		if (!skb)
4185 			goto out;
4186 		nf_skip_egress(skb, false);
4187 
4188 		if (netdev_xmit_txqueue_skipped())
4189 			txq = netdev_tx_queue_mapping(dev, skb);
4190 	}
4191 #endif
4192 	/* If device/qdisc don't need skb->dst, release it right now while
4193 	 * its hot in this cpu cache.
4194 	 */
4195 	if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4196 		skb_dst_drop(skb);
4197 	else
4198 		skb_dst_force(skb);
4199 
4200 	if (!txq)
4201 		txq = netdev_core_pick_tx(dev, skb, sb_dev);
4202 
4203 	q = rcu_dereference_bh(txq->qdisc);
4204 
4205 	trace_net_dev_queue(skb);
4206 	if (q->enqueue) {
4207 		rc = __dev_xmit_skb(skb, q, dev, txq);
4208 		goto out;
4209 	}
4210 
4211 	/* The device has no queue. Common case for software devices:
4212 	 * loopback, all the sorts of tunnels...
4213 
4214 	 * Really, it is unlikely that netif_tx_lock protection is necessary
4215 	 * here.  (f.e. loopback and IP tunnels are clean ignoring statistics
4216 	 * counters.)
4217 	 * However, it is possible, that they rely on protection
4218 	 * made by us here.
4219 
4220 	 * Check this and shot the lock. It is not prone from deadlocks.
4221 	 *Either shot noqueue qdisc, it is even simpler 8)
4222 	 */
4223 	if (dev->flags & IFF_UP) {
4224 		int cpu = smp_processor_id(); /* ok because BHs are off */
4225 
4226 		/* Other cpus might concurrently change txq->xmit_lock_owner
4227 		 * to -1 or to their cpu id, but not to our id.
4228 		 */
4229 		if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
4230 			if (dev_xmit_recursion())
4231 				goto recursion_alert;
4232 
4233 			skb = validate_xmit_skb(skb, dev, &again);
4234 			if (!skb)
4235 				goto out;
4236 
4237 			HARD_TX_LOCK(dev, txq, cpu);
4238 
4239 			if (!netif_xmit_stopped(txq)) {
4240 				dev_xmit_recursion_inc();
4241 				skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4242 				dev_xmit_recursion_dec();
4243 				if (dev_xmit_complete(rc)) {
4244 					HARD_TX_UNLOCK(dev, txq);
4245 					goto out;
4246 				}
4247 			}
4248 			HARD_TX_UNLOCK(dev, txq);
4249 			net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4250 					     dev->name);
4251 		} else {
4252 			/* Recursion is detected! It is possible,
4253 			 * unfortunately
4254 			 */
4255 recursion_alert:
4256 			net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4257 					     dev->name);
4258 		}
4259 	}
4260 
4261 	rc = -ENETDOWN;
4262 	rcu_read_unlock_bh();
4263 
4264 	dev_core_stats_tx_dropped_inc(dev);
4265 	kfree_skb_list(skb);
4266 	return rc;
4267 out:
4268 	rcu_read_unlock_bh();
4269 	return rc;
4270 }
4271 EXPORT_SYMBOL(__dev_queue_xmit);
4272 
4273 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4274 {
4275 	struct net_device *dev = skb->dev;
4276 	struct sk_buff *orig_skb = skb;
4277 	struct netdev_queue *txq;
4278 	int ret = NETDEV_TX_BUSY;
4279 	bool again = false;
4280 
4281 	if (unlikely(!netif_running(dev) ||
4282 		     !netif_carrier_ok(dev)))
4283 		goto drop;
4284 
4285 	skb = validate_xmit_skb_list(skb, dev, &again);
4286 	if (skb != orig_skb)
4287 		goto drop;
4288 
4289 	skb_set_queue_mapping(skb, queue_id);
4290 	txq = skb_get_tx_queue(dev, skb);
4291 
4292 	local_bh_disable();
4293 
4294 	dev_xmit_recursion_inc();
4295 	HARD_TX_LOCK(dev, txq, smp_processor_id());
4296 	if (!netif_xmit_frozen_or_drv_stopped(txq))
4297 		ret = netdev_start_xmit(skb, dev, txq, false);
4298 	HARD_TX_UNLOCK(dev, txq);
4299 	dev_xmit_recursion_dec();
4300 
4301 	local_bh_enable();
4302 	return ret;
4303 drop:
4304 	dev_core_stats_tx_dropped_inc(dev);
4305 	kfree_skb_list(skb);
4306 	return NET_XMIT_DROP;
4307 }
4308 EXPORT_SYMBOL(__dev_direct_xmit);
4309 
4310 /*************************************************************************
4311  *			Receiver routines
4312  *************************************************************************/
4313 
4314 int netdev_max_backlog __read_mostly = 1000;
4315 EXPORT_SYMBOL(netdev_max_backlog);
4316 
4317 int netdev_tstamp_prequeue __read_mostly = 1;
4318 unsigned int sysctl_skb_defer_max __read_mostly = 64;
4319 int netdev_budget __read_mostly = 300;
4320 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4321 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4322 int weight_p __read_mostly = 64;           /* old backlog weight */
4323 int dev_weight_rx_bias __read_mostly = 1;  /* bias for backlog weight */
4324 int dev_weight_tx_bias __read_mostly = 1;  /* bias for output_queue quota */
4325 int dev_rx_weight __read_mostly = 64;
4326 int dev_tx_weight __read_mostly = 64;
4327 
4328 /* Called with irq disabled */
4329 static inline void ____napi_schedule(struct softnet_data *sd,
4330 				     struct napi_struct *napi)
4331 {
4332 	struct task_struct *thread;
4333 
4334 	lockdep_assert_irqs_disabled();
4335 
4336 	if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4337 		/* Paired with smp_mb__before_atomic() in
4338 		 * napi_enable()/dev_set_threaded().
4339 		 * Use READ_ONCE() to guarantee a complete
4340 		 * read on napi->thread. Only call
4341 		 * wake_up_process() when it's not NULL.
4342 		 */
4343 		thread = READ_ONCE(napi->thread);
4344 		if (thread) {
4345 			/* Avoid doing set_bit() if the thread is in
4346 			 * INTERRUPTIBLE state, cause napi_thread_wait()
4347 			 * makes sure to proceed with napi polling
4348 			 * if the thread is explicitly woken from here.
4349 			 */
4350 			if (READ_ONCE(thread->__state) != TASK_INTERRUPTIBLE)
4351 				set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4352 			wake_up_process(thread);
4353 			return;
4354 		}
4355 	}
4356 
4357 	list_add_tail(&napi->poll_list, &sd->poll_list);
4358 	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4359 }
4360 
4361 #ifdef CONFIG_RPS
4362 
4363 /* One global table that all flow-based protocols share. */
4364 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4365 EXPORT_SYMBOL(rps_sock_flow_table);
4366 u32 rps_cpu_mask __read_mostly;
4367 EXPORT_SYMBOL(rps_cpu_mask);
4368 
4369 struct static_key_false rps_needed __read_mostly;
4370 EXPORT_SYMBOL(rps_needed);
4371 struct static_key_false rfs_needed __read_mostly;
4372 EXPORT_SYMBOL(rfs_needed);
4373 
4374 static struct rps_dev_flow *
4375 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4376 	    struct rps_dev_flow *rflow, u16 next_cpu)
4377 {
4378 	if (next_cpu < nr_cpu_ids) {
4379 #ifdef CONFIG_RFS_ACCEL
4380 		struct netdev_rx_queue *rxqueue;
4381 		struct rps_dev_flow_table *flow_table;
4382 		struct rps_dev_flow *old_rflow;
4383 		u32 flow_id;
4384 		u16 rxq_index;
4385 		int rc;
4386 
4387 		/* Should we steer this flow to a different hardware queue? */
4388 		if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4389 		    !(dev->features & NETIF_F_NTUPLE))
4390 			goto out;
4391 		rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4392 		if (rxq_index == skb_get_rx_queue(skb))
4393 			goto out;
4394 
4395 		rxqueue = dev->_rx + rxq_index;
4396 		flow_table = rcu_dereference(rxqueue->rps_flow_table);
4397 		if (!flow_table)
4398 			goto out;
4399 		flow_id = skb_get_hash(skb) & flow_table->mask;
4400 		rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4401 							rxq_index, flow_id);
4402 		if (rc < 0)
4403 			goto out;
4404 		old_rflow = rflow;
4405 		rflow = &flow_table->flows[flow_id];
4406 		rflow->filter = rc;
4407 		if (old_rflow->filter == rflow->filter)
4408 			old_rflow->filter = RPS_NO_FILTER;
4409 	out:
4410 #endif
4411 		rflow->last_qtail =
4412 			per_cpu(softnet_data, next_cpu).input_queue_head;
4413 	}
4414 
4415 	rflow->cpu = next_cpu;
4416 	return rflow;
4417 }
4418 
4419 /*
4420  * get_rps_cpu is called from netif_receive_skb and returns the target
4421  * CPU from the RPS map of the receiving queue for a given skb.
4422  * rcu_read_lock must be held on entry.
4423  */
4424 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4425 		       struct rps_dev_flow **rflowp)
4426 {
4427 	const struct rps_sock_flow_table *sock_flow_table;
4428 	struct netdev_rx_queue *rxqueue = dev->_rx;
4429 	struct rps_dev_flow_table *flow_table;
4430 	struct rps_map *map;
4431 	int cpu = -1;
4432 	u32 tcpu;
4433 	u32 hash;
4434 
4435 	if (skb_rx_queue_recorded(skb)) {
4436 		u16 index = skb_get_rx_queue(skb);
4437 
4438 		if (unlikely(index >= dev->real_num_rx_queues)) {
4439 			WARN_ONCE(dev->real_num_rx_queues > 1,
4440 				  "%s received packet on queue %u, but number "
4441 				  "of RX queues is %u\n",
4442 				  dev->name, index, dev->real_num_rx_queues);
4443 			goto done;
4444 		}
4445 		rxqueue += index;
4446 	}
4447 
4448 	/* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4449 
4450 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
4451 	map = rcu_dereference(rxqueue->rps_map);
4452 	if (!flow_table && !map)
4453 		goto done;
4454 
4455 	skb_reset_network_header(skb);
4456 	hash = skb_get_hash(skb);
4457 	if (!hash)
4458 		goto done;
4459 
4460 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
4461 	if (flow_table && sock_flow_table) {
4462 		struct rps_dev_flow *rflow;
4463 		u32 next_cpu;
4464 		u32 ident;
4465 
4466 		/* First check into global flow table if there is a match */
4467 		ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4468 		if ((ident ^ hash) & ~rps_cpu_mask)
4469 			goto try_rps;
4470 
4471 		next_cpu = ident & rps_cpu_mask;
4472 
4473 		/* OK, now we know there is a match,
4474 		 * we can look at the local (per receive queue) flow table
4475 		 */
4476 		rflow = &flow_table->flows[hash & flow_table->mask];
4477 		tcpu = rflow->cpu;
4478 
4479 		/*
4480 		 * If the desired CPU (where last recvmsg was done) is
4481 		 * different from current CPU (one in the rx-queue flow
4482 		 * table entry), switch if one of the following holds:
4483 		 *   - Current CPU is unset (>= nr_cpu_ids).
4484 		 *   - Current CPU is offline.
4485 		 *   - The current CPU's queue tail has advanced beyond the
4486 		 *     last packet that was enqueued using this table entry.
4487 		 *     This guarantees that all previous packets for the flow
4488 		 *     have been dequeued, thus preserving in order delivery.
4489 		 */
4490 		if (unlikely(tcpu != next_cpu) &&
4491 		    (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4492 		     ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4493 		      rflow->last_qtail)) >= 0)) {
4494 			tcpu = next_cpu;
4495 			rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4496 		}
4497 
4498 		if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4499 			*rflowp = rflow;
4500 			cpu = tcpu;
4501 			goto done;
4502 		}
4503 	}
4504 
4505 try_rps:
4506 
4507 	if (map) {
4508 		tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4509 		if (cpu_online(tcpu)) {
4510 			cpu = tcpu;
4511 			goto done;
4512 		}
4513 	}
4514 
4515 done:
4516 	return cpu;
4517 }
4518 
4519 #ifdef CONFIG_RFS_ACCEL
4520 
4521 /**
4522  * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4523  * @dev: Device on which the filter was set
4524  * @rxq_index: RX queue index
4525  * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4526  * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4527  *
4528  * Drivers that implement ndo_rx_flow_steer() should periodically call
4529  * this function for each installed filter and remove the filters for
4530  * which it returns %true.
4531  */
4532 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4533 			 u32 flow_id, u16 filter_id)
4534 {
4535 	struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4536 	struct rps_dev_flow_table *flow_table;
4537 	struct rps_dev_flow *rflow;
4538 	bool expire = true;
4539 	unsigned int cpu;
4540 
4541 	rcu_read_lock();
4542 	flow_table = rcu_dereference(rxqueue->rps_flow_table);
4543 	if (flow_table && flow_id <= flow_table->mask) {
4544 		rflow = &flow_table->flows[flow_id];
4545 		cpu = READ_ONCE(rflow->cpu);
4546 		if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4547 		    ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4548 			   rflow->last_qtail) <
4549 		     (int)(10 * flow_table->mask)))
4550 			expire = false;
4551 	}
4552 	rcu_read_unlock();
4553 	return expire;
4554 }
4555 EXPORT_SYMBOL(rps_may_expire_flow);
4556 
4557 #endif /* CONFIG_RFS_ACCEL */
4558 
4559 /* Called from hardirq (IPI) context */
4560 static void rps_trigger_softirq(void *data)
4561 {
4562 	struct softnet_data *sd = data;
4563 
4564 	____napi_schedule(sd, &sd->backlog);
4565 	sd->received_rps++;
4566 }
4567 
4568 #endif /* CONFIG_RPS */
4569 
4570 /* Called from hardirq (IPI) context */
4571 static void trigger_rx_softirq(void *data)
4572 {
4573 	struct softnet_data *sd = data;
4574 
4575 	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4576 	smp_store_release(&sd->defer_ipi_scheduled, 0);
4577 }
4578 
4579 /*
4580  * Check if this softnet_data structure is another cpu one
4581  * If yes, queue it to our IPI list and return 1
4582  * If no, return 0
4583  */
4584 static int napi_schedule_rps(struct softnet_data *sd)
4585 {
4586 	struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4587 
4588 #ifdef CONFIG_RPS
4589 	if (sd != mysd) {
4590 		sd->rps_ipi_next = mysd->rps_ipi_list;
4591 		mysd->rps_ipi_list = sd;
4592 
4593 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
4594 		return 1;
4595 	}
4596 #endif /* CONFIG_RPS */
4597 	__napi_schedule_irqoff(&mysd->backlog);
4598 	return 0;
4599 }
4600 
4601 #ifdef CONFIG_NET_FLOW_LIMIT
4602 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4603 #endif
4604 
4605 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4606 {
4607 #ifdef CONFIG_NET_FLOW_LIMIT
4608 	struct sd_flow_limit *fl;
4609 	struct softnet_data *sd;
4610 	unsigned int old_flow, new_flow;
4611 
4612 	if (qlen < (READ_ONCE(netdev_max_backlog) >> 1))
4613 		return false;
4614 
4615 	sd = this_cpu_ptr(&softnet_data);
4616 
4617 	rcu_read_lock();
4618 	fl = rcu_dereference(sd->flow_limit);
4619 	if (fl) {
4620 		new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4621 		old_flow = fl->history[fl->history_head];
4622 		fl->history[fl->history_head] = new_flow;
4623 
4624 		fl->history_head++;
4625 		fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4626 
4627 		if (likely(fl->buckets[old_flow]))
4628 			fl->buckets[old_flow]--;
4629 
4630 		if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4631 			fl->count++;
4632 			rcu_read_unlock();
4633 			return true;
4634 		}
4635 	}
4636 	rcu_read_unlock();
4637 #endif
4638 	return false;
4639 }
4640 
4641 /*
4642  * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4643  * queue (may be a remote CPU queue).
4644  */
4645 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4646 			      unsigned int *qtail)
4647 {
4648 	enum skb_drop_reason reason;
4649 	struct softnet_data *sd;
4650 	unsigned long flags;
4651 	unsigned int qlen;
4652 
4653 	reason = SKB_DROP_REASON_NOT_SPECIFIED;
4654 	sd = &per_cpu(softnet_data, cpu);
4655 
4656 	rps_lock_irqsave(sd, &flags);
4657 	if (!netif_running(skb->dev))
4658 		goto drop;
4659 	qlen = skb_queue_len(&sd->input_pkt_queue);
4660 	if (qlen <= READ_ONCE(netdev_max_backlog) && !skb_flow_limit(skb, qlen)) {
4661 		if (qlen) {
4662 enqueue:
4663 			__skb_queue_tail(&sd->input_pkt_queue, skb);
4664 			input_queue_tail_incr_save(sd, qtail);
4665 			rps_unlock_irq_restore(sd, &flags);
4666 			return NET_RX_SUCCESS;
4667 		}
4668 
4669 		/* Schedule NAPI for backlog device
4670 		 * We can use non atomic operation since we own the queue lock
4671 		 */
4672 		if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state))
4673 			napi_schedule_rps(sd);
4674 		goto enqueue;
4675 	}
4676 	reason = SKB_DROP_REASON_CPU_BACKLOG;
4677 
4678 drop:
4679 	sd->dropped++;
4680 	rps_unlock_irq_restore(sd, &flags);
4681 
4682 	dev_core_stats_rx_dropped_inc(skb->dev);
4683 	kfree_skb_reason(skb, reason);
4684 	return NET_RX_DROP;
4685 }
4686 
4687 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4688 {
4689 	struct net_device *dev = skb->dev;
4690 	struct netdev_rx_queue *rxqueue;
4691 
4692 	rxqueue = dev->_rx;
4693 
4694 	if (skb_rx_queue_recorded(skb)) {
4695 		u16 index = skb_get_rx_queue(skb);
4696 
4697 		if (unlikely(index >= dev->real_num_rx_queues)) {
4698 			WARN_ONCE(dev->real_num_rx_queues > 1,
4699 				  "%s received packet on queue %u, but number "
4700 				  "of RX queues is %u\n",
4701 				  dev->name, index, dev->real_num_rx_queues);
4702 
4703 			return rxqueue; /* Return first rxqueue */
4704 		}
4705 		rxqueue += index;
4706 	}
4707 	return rxqueue;
4708 }
4709 
4710 u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
4711 			     struct bpf_prog *xdp_prog)
4712 {
4713 	void *orig_data, *orig_data_end, *hard_start;
4714 	struct netdev_rx_queue *rxqueue;
4715 	bool orig_bcast, orig_host;
4716 	u32 mac_len, frame_sz;
4717 	__be16 orig_eth_type;
4718 	struct ethhdr *eth;
4719 	u32 metalen, act;
4720 	int off;
4721 
4722 	/* The XDP program wants to see the packet starting at the MAC
4723 	 * header.
4724 	 */
4725 	mac_len = skb->data - skb_mac_header(skb);
4726 	hard_start = skb->data - skb_headroom(skb);
4727 
4728 	/* SKB "head" area always have tailroom for skb_shared_info */
4729 	frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4730 	frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4731 
4732 	rxqueue = netif_get_rxqueue(skb);
4733 	xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4734 	xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4735 			 skb_headlen(skb) + mac_len, true);
4736 
4737 	orig_data_end = xdp->data_end;
4738 	orig_data = xdp->data;
4739 	eth = (struct ethhdr *)xdp->data;
4740 	orig_host = ether_addr_equal_64bits(eth->h_dest, skb->dev->dev_addr);
4741 	orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4742 	orig_eth_type = eth->h_proto;
4743 
4744 	act = bpf_prog_run_xdp(xdp_prog, xdp);
4745 
4746 	/* check if bpf_xdp_adjust_head was used */
4747 	off = xdp->data - orig_data;
4748 	if (off) {
4749 		if (off > 0)
4750 			__skb_pull(skb, off);
4751 		else if (off < 0)
4752 			__skb_push(skb, -off);
4753 
4754 		skb->mac_header += off;
4755 		skb_reset_network_header(skb);
4756 	}
4757 
4758 	/* check if bpf_xdp_adjust_tail was used */
4759 	off = xdp->data_end - orig_data_end;
4760 	if (off != 0) {
4761 		skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4762 		skb->len += off; /* positive on grow, negative on shrink */
4763 	}
4764 
4765 	/* check if XDP changed eth hdr such SKB needs update */
4766 	eth = (struct ethhdr *)xdp->data;
4767 	if ((orig_eth_type != eth->h_proto) ||
4768 	    (orig_host != ether_addr_equal_64bits(eth->h_dest,
4769 						  skb->dev->dev_addr)) ||
4770 	    (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4771 		__skb_push(skb, ETH_HLEN);
4772 		skb->pkt_type = PACKET_HOST;
4773 		skb->protocol = eth_type_trans(skb, skb->dev);
4774 	}
4775 
4776 	/* Redirect/Tx gives L2 packet, code that will reuse skb must __skb_pull
4777 	 * before calling us again on redirect path. We do not call do_redirect
4778 	 * as we leave that up to the caller.
4779 	 *
4780 	 * Caller is responsible for managing lifetime of skb (i.e. calling
4781 	 * kfree_skb in response to actions it cannot handle/XDP_DROP).
4782 	 */
4783 	switch (act) {
4784 	case XDP_REDIRECT:
4785 	case XDP_TX:
4786 		__skb_push(skb, mac_len);
4787 		break;
4788 	case XDP_PASS:
4789 		metalen = xdp->data - xdp->data_meta;
4790 		if (metalen)
4791 			skb_metadata_set(skb, metalen);
4792 		break;
4793 	}
4794 
4795 	return act;
4796 }
4797 
4798 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4799 				     struct xdp_buff *xdp,
4800 				     struct bpf_prog *xdp_prog)
4801 {
4802 	u32 act = XDP_DROP;
4803 
4804 	/* Reinjected packets coming from act_mirred or similar should
4805 	 * not get XDP generic processing.
4806 	 */
4807 	if (skb_is_redirected(skb))
4808 		return XDP_PASS;
4809 
4810 	/* XDP packets must be linear and must have sufficient headroom
4811 	 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4812 	 * native XDP provides, thus we need to do it here as well.
4813 	 */
4814 	if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4815 	    skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4816 		int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4817 		int troom = skb->tail + skb->data_len - skb->end;
4818 
4819 		/* In case we have to go down the path and also linearize,
4820 		 * then lets do the pskb_expand_head() work just once here.
4821 		 */
4822 		if (pskb_expand_head(skb,
4823 				     hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4824 				     troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4825 			goto do_drop;
4826 		if (skb_linearize(skb))
4827 			goto do_drop;
4828 	}
4829 
4830 	act = bpf_prog_run_generic_xdp(skb, xdp, xdp_prog);
4831 	switch (act) {
4832 	case XDP_REDIRECT:
4833 	case XDP_TX:
4834 	case XDP_PASS:
4835 		break;
4836 	default:
4837 		bpf_warn_invalid_xdp_action(skb->dev, xdp_prog, act);
4838 		fallthrough;
4839 	case XDP_ABORTED:
4840 		trace_xdp_exception(skb->dev, xdp_prog, act);
4841 		fallthrough;
4842 	case XDP_DROP:
4843 	do_drop:
4844 		kfree_skb(skb);
4845 		break;
4846 	}
4847 
4848 	return act;
4849 }
4850 
4851 /* When doing generic XDP we have to bypass the qdisc layer and the
4852  * network taps in order to match in-driver-XDP behavior. This also means
4853  * that XDP packets are able to starve other packets going through a qdisc,
4854  * and DDOS attacks will be more effective. In-driver-XDP use dedicated TX
4855  * queues, so they do not have this starvation issue.
4856  */
4857 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4858 {
4859 	struct net_device *dev = skb->dev;
4860 	struct netdev_queue *txq;
4861 	bool free_skb = true;
4862 	int cpu, rc;
4863 
4864 	txq = netdev_core_pick_tx(dev, skb, NULL);
4865 	cpu = smp_processor_id();
4866 	HARD_TX_LOCK(dev, txq, cpu);
4867 	if (!netif_xmit_frozen_or_drv_stopped(txq)) {
4868 		rc = netdev_start_xmit(skb, dev, txq, 0);
4869 		if (dev_xmit_complete(rc))
4870 			free_skb = false;
4871 	}
4872 	HARD_TX_UNLOCK(dev, txq);
4873 	if (free_skb) {
4874 		trace_xdp_exception(dev, xdp_prog, XDP_TX);
4875 		dev_core_stats_tx_dropped_inc(dev);
4876 		kfree_skb(skb);
4877 	}
4878 }
4879 
4880 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4881 
4882 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4883 {
4884 	if (xdp_prog) {
4885 		struct xdp_buff xdp;
4886 		u32 act;
4887 		int err;
4888 
4889 		act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4890 		if (act != XDP_PASS) {
4891 			switch (act) {
4892 			case XDP_REDIRECT:
4893 				err = xdp_do_generic_redirect(skb->dev, skb,
4894 							      &xdp, xdp_prog);
4895 				if (err)
4896 					goto out_redir;
4897 				break;
4898 			case XDP_TX:
4899 				generic_xdp_tx(skb, xdp_prog);
4900 				break;
4901 			}
4902 			return XDP_DROP;
4903 		}
4904 	}
4905 	return XDP_PASS;
4906 out_redir:
4907 	kfree_skb_reason(skb, SKB_DROP_REASON_XDP);
4908 	return XDP_DROP;
4909 }
4910 EXPORT_SYMBOL_GPL(do_xdp_generic);
4911 
4912 static int netif_rx_internal(struct sk_buff *skb)
4913 {
4914 	int ret;
4915 
4916 	net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
4917 
4918 	trace_netif_rx(skb);
4919 
4920 #ifdef CONFIG_RPS
4921 	if (static_branch_unlikely(&rps_needed)) {
4922 		struct rps_dev_flow voidflow, *rflow = &voidflow;
4923 		int cpu;
4924 
4925 		rcu_read_lock();
4926 
4927 		cpu = get_rps_cpu(skb->dev, skb, &rflow);
4928 		if (cpu < 0)
4929 			cpu = smp_processor_id();
4930 
4931 		ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4932 
4933 		rcu_read_unlock();
4934 	} else
4935 #endif
4936 	{
4937 		unsigned int qtail;
4938 
4939 		ret = enqueue_to_backlog(skb, smp_processor_id(), &qtail);
4940 	}
4941 	return ret;
4942 }
4943 
4944 /**
4945  *	__netif_rx	-	Slightly optimized version of netif_rx
4946  *	@skb: buffer to post
4947  *
4948  *	This behaves as netif_rx except that it does not disable bottom halves.
4949  *	As a result this function may only be invoked from the interrupt context
4950  *	(either hard or soft interrupt).
4951  */
4952 int __netif_rx(struct sk_buff *skb)
4953 {
4954 	int ret;
4955 
4956 	lockdep_assert_once(hardirq_count() | softirq_count());
4957 
4958 	trace_netif_rx_entry(skb);
4959 	ret = netif_rx_internal(skb);
4960 	trace_netif_rx_exit(ret);
4961 	return ret;
4962 }
4963 EXPORT_SYMBOL(__netif_rx);
4964 
4965 /**
4966  *	netif_rx	-	post buffer to the network code
4967  *	@skb: buffer to post
4968  *
4969  *	This function receives a packet from a device driver and queues it for
4970  *	the upper (protocol) levels to process via the backlog NAPI device. It
4971  *	always succeeds. The buffer may be dropped during processing for
4972  *	congestion control or by the protocol layers.
4973  *	The network buffer is passed via the backlog NAPI device. Modern NIC
4974  *	driver should use NAPI and GRO.
4975  *	This function can used from interrupt and from process context. The
4976  *	caller from process context must not disable interrupts before invoking
4977  *	this function.
4978  *
4979  *	return values:
4980  *	NET_RX_SUCCESS	(no congestion)
4981  *	NET_RX_DROP     (packet was dropped)
4982  *
4983  */
4984 int netif_rx(struct sk_buff *skb)
4985 {
4986 	bool need_bh_off = !(hardirq_count() | softirq_count());
4987 	int ret;
4988 
4989 	if (need_bh_off)
4990 		local_bh_disable();
4991 	trace_netif_rx_entry(skb);
4992 	ret = netif_rx_internal(skb);
4993 	trace_netif_rx_exit(ret);
4994 	if (need_bh_off)
4995 		local_bh_enable();
4996 	return ret;
4997 }
4998 EXPORT_SYMBOL(netif_rx);
4999 
5000 static __latent_entropy void net_tx_action(struct softirq_action *h)
5001 {
5002 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5003 
5004 	if (sd->completion_queue) {
5005 		struct sk_buff *clist;
5006 
5007 		local_irq_disable();
5008 		clist = sd->completion_queue;
5009 		sd->completion_queue = NULL;
5010 		local_irq_enable();
5011 
5012 		while (clist) {
5013 			struct sk_buff *skb = clist;
5014 
5015 			clist = clist->next;
5016 
5017 			WARN_ON(refcount_read(&skb->users));
5018 			if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
5019 				trace_consume_skb(skb);
5020 			else
5021 				trace_kfree_skb(skb, net_tx_action,
5022 						SKB_DROP_REASON_NOT_SPECIFIED);
5023 
5024 			if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5025 				__kfree_skb(skb);
5026 			else
5027 				__kfree_skb_defer(skb);
5028 		}
5029 	}
5030 
5031 	if (sd->output_queue) {
5032 		struct Qdisc *head;
5033 
5034 		local_irq_disable();
5035 		head = sd->output_queue;
5036 		sd->output_queue = NULL;
5037 		sd->output_queue_tailp = &sd->output_queue;
5038 		local_irq_enable();
5039 
5040 		rcu_read_lock();
5041 
5042 		while (head) {
5043 			struct Qdisc *q = head;
5044 			spinlock_t *root_lock = NULL;
5045 
5046 			head = head->next_sched;
5047 
5048 			/* We need to make sure head->next_sched is read
5049 			 * before clearing __QDISC_STATE_SCHED
5050 			 */
5051 			smp_mb__before_atomic();
5052 
5053 			if (!(q->flags & TCQ_F_NOLOCK)) {
5054 				root_lock = qdisc_lock(q);
5055 				spin_lock(root_lock);
5056 			} else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
5057 						     &q->state))) {
5058 				/* There is a synchronize_net() between
5059 				 * STATE_DEACTIVATED flag being set and
5060 				 * qdisc_reset()/some_qdisc_is_busy() in
5061 				 * dev_deactivate(), so we can safely bail out
5062 				 * early here to avoid data race between
5063 				 * qdisc_deactivate() and some_qdisc_is_busy()
5064 				 * for lockless qdisc.
5065 				 */
5066 				clear_bit(__QDISC_STATE_SCHED, &q->state);
5067 				continue;
5068 			}
5069 
5070 			clear_bit(__QDISC_STATE_SCHED, &q->state);
5071 			qdisc_run(q);
5072 			if (root_lock)
5073 				spin_unlock(root_lock);
5074 		}
5075 
5076 		rcu_read_unlock();
5077 	}
5078 
5079 	xfrm_dev_backlog(sd);
5080 }
5081 
5082 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5083 /* This hook is defined here for ATM LANE */
5084 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5085 			     unsigned char *addr) __read_mostly;
5086 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5087 #endif
5088 
5089 static inline struct sk_buff *
5090 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5091 		   struct net_device *orig_dev, bool *another)
5092 {
5093 #ifdef CONFIG_NET_CLS_ACT
5094 	struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5095 	struct tcf_result cl_res;
5096 
5097 	/* If there's at least one ingress present somewhere (so
5098 	 * we get here via enabled static key), remaining devices
5099 	 * that are not configured with an ingress qdisc will bail
5100 	 * out here.
5101 	 */
5102 	if (!miniq)
5103 		return skb;
5104 
5105 	if (*pt_prev) {
5106 		*ret = deliver_skb(skb, *pt_prev, orig_dev);
5107 		*pt_prev = NULL;
5108 	}
5109 
5110 	qdisc_skb_cb(skb)->pkt_len = skb->len;
5111 	tc_skb_cb(skb)->mru = 0;
5112 	tc_skb_cb(skb)->post_ct = false;
5113 	skb->tc_at_ingress = 1;
5114 	mini_qdisc_bstats_cpu_update(miniq, skb);
5115 
5116 	switch (tcf_classify(skb, miniq->block, miniq->filter_list, &cl_res, false)) {
5117 	case TC_ACT_OK:
5118 	case TC_ACT_RECLASSIFY:
5119 		skb->tc_index = TC_H_MIN(cl_res.classid);
5120 		break;
5121 	case TC_ACT_SHOT:
5122 		mini_qdisc_qstats_cpu_drop(miniq);
5123 		kfree_skb_reason(skb, SKB_DROP_REASON_TC_INGRESS);
5124 		*ret = NET_RX_DROP;
5125 		return NULL;
5126 	case TC_ACT_STOLEN:
5127 	case TC_ACT_QUEUED:
5128 	case TC_ACT_TRAP:
5129 		consume_skb(skb);
5130 		*ret = NET_RX_SUCCESS;
5131 		return NULL;
5132 	case TC_ACT_REDIRECT:
5133 		/* skb_mac_header check was done by cls/act_bpf, so
5134 		 * we can safely push the L2 header back before
5135 		 * redirecting to another netdev
5136 		 */
5137 		__skb_push(skb, skb->mac_len);
5138 		if (skb_do_redirect(skb) == -EAGAIN) {
5139 			__skb_pull(skb, skb->mac_len);
5140 			*another = true;
5141 			break;
5142 		}
5143 		*ret = NET_RX_SUCCESS;
5144 		return NULL;
5145 	case TC_ACT_CONSUMED:
5146 		*ret = NET_RX_SUCCESS;
5147 		return NULL;
5148 	default:
5149 		break;
5150 	}
5151 #endif /* CONFIG_NET_CLS_ACT */
5152 	return skb;
5153 }
5154 
5155 /**
5156  *	netdev_is_rx_handler_busy - check if receive handler is registered
5157  *	@dev: device to check
5158  *
5159  *	Check if a receive handler is already registered for a given device.
5160  *	Return true if there one.
5161  *
5162  *	The caller must hold the rtnl_mutex.
5163  */
5164 bool netdev_is_rx_handler_busy(struct net_device *dev)
5165 {
5166 	ASSERT_RTNL();
5167 	return dev && rtnl_dereference(dev->rx_handler);
5168 }
5169 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5170 
5171 /**
5172  *	netdev_rx_handler_register - register receive handler
5173  *	@dev: device to register a handler for
5174  *	@rx_handler: receive handler to register
5175  *	@rx_handler_data: data pointer that is used by rx handler
5176  *
5177  *	Register a receive handler for a device. This handler will then be
5178  *	called from __netif_receive_skb. A negative errno code is returned
5179  *	on a failure.
5180  *
5181  *	The caller must hold the rtnl_mutex.
5182  *
5183  *	For a general description of rx_handler, see enum rx_handler_result.
5184  */
5185 int netdev_rx_handler_register(struct net_device *dev,
5186 			       rx_handler_func_t *rx_handler,
5187 			       void *rx_handler_data)
5188 {
5189 	if (netdev_is_rx_handler_busy(dev))
5190 		return -EBUSY;
5191 
5192 	if (dev->priv_flags & IFF_NO_RX_HANDLER)
5193 		return -EINVAL;
5194 
5195 	/* Note: rx_handler_data must be set before rx_handler */
5196 	rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5197 	rcu_assign_pointer(dev->rx_handler, rx_handler);
5198 
5199 	return 0;
5200 }
5201 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5202 
5203 /**
5204  *	netdev_rx_handler_unregister - unregister receive handler
5205  *	@dev: device to unregister a handler from
5206  *
5207  *	Unregister a receive handler from a device.
5208  *
5209  *	The caller must hold the rtnl_mutex.
5210  */
5211 void netdev_rx_handler_unregister(struct net_device *dev)
5212 {
5213 
5214 	ASSERT_RTNL();
5215 	RCU_INIT_POINTER(dev->rx_handler, NULL);
5216 	/* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5217 	 * section has a guarantee to see a non NULL rx_handler_data
5218 	 * as well.
5219 	 */
5220 	synchronize_net();
5221 	RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5222 }
5223 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5224 
5225 /*
5226  * Limit the use of PFMEMALLOC reserves to those protocols that implement
5227  * the special handling of PFMEMALLOC skbs.
5228  */
5229 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5230 {
5231 	switch (skb->protocol) {
5232 	case htons(ETH_P_ARP):
5233 	case htons(ETH_P_IP):
5234 	case htons(ETH_P_IPV6):
5235 	case htons(ETH_P_8021Q):
5236 	case htons(ETH_P_8021AD):
5237 		return true;
5238 	default:
5239 		return false;
5240 	}
5241 }
5242 
5243 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5244 			     int *ret, struct net_device *orig_dev)
5245 {
5246 	if (nf_hook_ingress_active(skb)) {
5247 		int ingress_retval;
5248 
5249 		if (*pt_prev) {
5250 			*ret = deliver_skb(skb, *pt_prev, orig_dev);
5251 			*pt_prev = NULL;
5252 		}
5253 
5254 		rcu_read_lock();
5255 		ingress_retval = nf_hook_ingress(skb);
5256 		rcu_read_unlock();
5257 		return ingress_retval;
5258 	}
5259 	return 0;
5260 }
5261 
5262 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5263 				    struct packet_type **ppt_prev)
5264 {
5265 	struct packet_type *ptype, *pt_prev;
5266 	rx_handler_func_t *rx_handler;
5267 	struct sk_buff *skb = *pskb;
5268 	struct net_device *orig_dev;
5269 	bool deliver_exact = false;
5270 	int ret = NET_RX_DROP;
5271 	__be16 type;
5272 
5273 	net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb);
5274 
5275 	trace_netif_receive_skb(skb);
5276 
5277 	orig_dev = skb->dev;
5278 
5279 	skb_reset_network_header(skb);
5280 	if (!skb_transport_header_was_set(skb))
5281 		skb_reset_transport_header(skb);
5282 	skb_reset_mac_len(skb);
5283 
5284 	pt_prev = NULL;
5285 
5286 another_round:
5287 	skb->skb_iif = skb->dev->ifindex;
5288 
5289 	__this_cpu_inc(softnet_data.processed);
5290 
5291 	if (static_branch_unlikely(&generic_xdp_needed_key)) {
5292 		int ret2;
5293 
5294 		migrate_disable();
5295 		ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5296 		migrate_enable();
5297 
5298 		if (ret2 != XDP_PASS) {
5299 			ret = NET_RX_DROP;
5300 			goto out;
5301 		}
5302 	}
5303 
5304 	if (eth_type_vlan(skb->protocol)) {
5305 		skb = skb_vlan_untag(skb);
5306 		if (unlikely(!skb))
5307 			goto out;
5308 	}
5309 
5310 	if (skb_skip_tc_classify(skb))
5311 		goto skip_classify;
5312 
5313 	if (pfmemalloc)
5314 		goto skip_taps;
5315 
5316 	list_for_each_entry_rcu(ptype, &ptype_all, list) {
5317 		if (pt_prev)
5318 			ret = deliver_skb(skb, pt_prev, orig_dev);
5319 		pt_prev = ptype;
5320 	}
5321 
5322 	list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5323 		if (pt_prev)
5324 			ret = deliver_skb(skb, pt_prev, orig_dev);
5325 		pt_prev = ptype;
5326 	}
5327 
5328 skip_taps:
5329 #ifdef CONFIG_NET_INGRESS
5330 	if (static_branch_unlikely(&ingress_needed_key)) {
5331 		bool another = false;
5332 
5333 		nf_skip_egress(skb, true);
5334 		skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5335 					 &another);
5336 		if (another)
5337 			goto another_round;
5338 		if (!skb)
5339 			goto out;
5340 
5341 		nf_skip_egress(skb, false);
5342 		if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5343 			goto out;
5344 	}
5345 #endif
5346 	skb_reset_redirect(skb);
5347 skip_classify:
5348 	if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5349 		goto drop;
5350 
5351 	if (skb_vlan_tag_present(skb)) {
5352 		if (pt_prev) {
5353 			ret = deliver_skb(skb, pt_prev, orig_dev);
5354 			pt_prev = NULL;
5355 		}
5356 		if (vlan_do_receive(&skb))
5357 			goto another_round;
5358 		else if (unlikely(!skb))
5359 			goto out;
5360 	}
5361 
5362 	rx_handler = rcu_dereference(skb->dev->rx_handler);
5363 	if (rx_handler) {
5364 		if (pt_prev) {
5365 			ret = deliver_skb(skb, pt_prev, orig_dev);
5366 			pt_prev = NULL;
5367 		}
5368 		switch (rx_handler(&skb)) {
5369 		case RX_HANDLER_CONSUMED:
5370 			ret = NET_RX_SUCCESS;
5371 			goto out;
5372 		case RX_HANDLER_ANOTHER:
5373 			goto another_round;
5374 		case RX_HANDLER_EXACT:
5375 			deliver_exact = true;
5376 			break;
5377 		case RX_HANDLER_PASS:
5378 			break;
5379 		default:
5380 			BUG();
5381 		}
5382 	}
5383 
5384 	if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5385 check_vlan_id:
5386 		if (skb_vlan_tag_get_id(skb)) {
5387 			/* Vlan id is non 0 and vlan_do_receive() above couldn't
5388 			 * find vlan device.
5389 			 */
5390 			skb->pkt_type = PACKET_OTHERHOST;
5391 		} else if (eth_type_vlan(skb->protocol)) {
5392 			/* Outer header is 802.1P with vlan 0, inner header is
5393 			 * 802.1Q or 802.1AD and vlan_do_receive() above could
5394 			 * not find vlan dev for vlan id 0.
5395 			 */
5396 			__vlan_hwaccel_clear_tag(skb);
5397 			skb = skb_vlan_untag(skb);
5398 			if (unlikely(!skb))
5399 				goto out;
5400 			if (vlan_do_receive(&skb))
5401 				/* After stripping off 802.1P header with vlan 0
5402 				 * vlan dev is found for inner header.
5403 				 */
5404 				goto another_round;
5405 			else if (unlikely(!skb))
5406 				goto out;
5407 			else
5408 				/* We have stripped outer 802.1P vlan 0 header.
5409 				 * But could not find vlan dev.
5410 				 * check again for vlan id to set OTHERHOST.
5411 				 */
5412 				goto check_vlan_id;
5413 		}
5414 		/* Note: we might in the future use prio bits
5415 		 * and set skb->priority like in vlan_do_receive()
5416 		 * For the time being, just ignore Priority Code Point
5417 		 */
5418 		__vlan_hwaccel_clear_tag(skb);
5419 	}
5420 
5421 	type = skb->protocol;
5422 
5423 	/* deliver only exact match when indicated */
5424 	if (likely(!deliver_exact)) {
5425 		deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5426 				       &ptype_base[ntohs(type) &
5427 						   PTYPE_HASH_MASK]);
5428 	}
5429 
5430 	deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5431 			       &orig_dev->ptype_specific);
5432 
5433 	if (unlikely(skb->dev != orig_dev)) {
5434 		deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5435 				       &skb->dev->ptype_specific);
5436 	}
5437 
5438 	if (pt_prev) {
5439 		if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5440 			goto drop;
5441 		*ppt_prev = pt_prev;
5442 	} else {
5443 drop:
5444 		if (!deliver_exact)
5445 			dev_core_stats_rx_dropped_inc(skb->dev);
5446 		else
5447 			dev_core_stats_rx_nohandler_inc(skb->dev);
5448 		kfree_skb_reason(skb, SKB_DROP_REASON_UNHANDLED_PROTO);
5449 		/* Jamal, now you will not able to escape explaining
5450 		 * me how you were going to use this. :-)
5451 		 */
5452 		ret = NET_RX_DROP;
5453 	}
5454 
5455 out:
5456 	/* The invariant here is that if *ppt_prev is not NULL
5457 	 * then skb should also be non-NULL.
5458 	 *
5459 	 * Apparently *ppt_prev assignment above holds this invariant due to
5460 	 * skb dereferencing near it.
5461 	 */
5462 	*pskb = skb;
5463 	return ret;
5464 }
5465 
5466 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5467 {
5468 	struct net_device *orig_dev = skb->dev;
5469 	struct packet_type *pt_prev = NULL;
5470 	int ret;
5471 
5472 	ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5473 	if (pt_prev)
5474 		ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5475 					 skb->dev, pt_prev, orig_dev);
5476 	return ret;
5477 }
5478 
5479 /**
5480  *	netif_receive_skb_core - special purpose version of netif_receive_skb
5481  *	@skb: buffer to process
5482  *
5483  *	More direct receive version of netif_receive_skb().  It should
5484  *	only be used by callers that have a need to skip RPS and Generic XDP.
5485  *	Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5486  *
5487  *	This function may only be called from softirq context and interrupts
5488  *	should be enabled.
5489  *
5490  *	Return values (usually ignored):
5491  *	NET_RX_SUCCESS: no congestion
5492  *	NET_RX_DROP: packet was dropped
5493  */
5494 int netif_receive_skb_core(struct sk_buff *skb)
5495 {
5496 	int ret;
5497 
5498 	rcu_read_lock();
5499 	ret = __netif_receive_skb_one_core(skb, false);
5500 	rcu_read_unlock();
5501 
5502 	return ret;
5503 }
5504 EXPORT_SYMBOL(netif_receive_skb_core);
5505 
5506 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5507 						  struct packet_type *pt_prev,
5508 						  struct net_device *orig_dev)
5509 {
5510 	struct sk_buff *skb, *next;
5511 
5512 	if (!pt_prev)
5513 		return;
5514 	if (list_empty(head))
5515 		return;
5516 	if (pt_prev->list_func != NULL)
5517 		INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5518 				   ip_list_rcv, head, pt_prev, orig_dev);
5519 	else
5520 		list_for_each_entry_safe(skb, next, head, list) {
5521 			skb_list_del_init(skb);
5522 			pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5523 		}
5524 }
5525 
5526 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5527 {
5528 	/* Fast-path assumptions:
5529 	 * - There is no RX handler.
5530 	 * - Only one packet_type matches.
5531 	 * If either of these fails, we will end up doing some per-packet
5532 	 * processing in-line, then handling the 'last ptype' for the whole
5533 	 * sublist.  This can't cause out-of-order delivery to any single ptype,
5534 	 * because the 'last ptype' must be constant across the sublist, and all
5535 	 * other ptypes are handled per-packet.
5536 	 */
5537 	/* Current (common) ptype of sublist */
5538 	struct packet_type *pt_curr = NULL;
5539 	/* Current (common) orig_dev of sublist */
5540 	struct net_device *od_curr = NULL;
5541 	struct list_head sublist;
5542 	struct sk_buff *skb, *next;
5543 
5544 	INIT_LIST_HEAD(&sublist);
5545 	list_for_each_entry_safe(skb, next, head, list) {
5546 		struct net_device *orig_dev = skb->dev;
5547 		struct packet_type *pt_prev = NULL;
5548 
5549 		skb_list_del_init(skb);
5550 		__netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5551 		if (!pt_prev)
5552 			continue;
5553 		if (pt_curr != pt_prev || od_curr != orig_dev) {
5554 			/* dispatch old sublist */
5555 			__netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5556 			/* start new sublist */
5557 			INIT_LIST_HEAD(&sublist);
5558 			pt_curr = pt_prev;
5559 			od_curr = orig_dev;
5560 		}
5561 		list_add_tail(&skb->list, &sublist);
5562 	}
5563 
5564 	/* dispatch final sublist */
5565 	__netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5566 }
5567 
5568 static int __netif_receive_skb(struct sk_buff *skb)
5569 {
5570 	int ret;
5571 
5572 	if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5573 		unsigned int noreclaim_flag;
5574 
5575 		/*
5576 		 * PFMEMALLOC skbs are special, they should
5577 		 * - be delivered to SOCK_MEMALLOC sockets only
5578 		 * - stay away from userspace
5579 		 * - have bounded memory usage
5580 		 *
5581 		 * Use PF_MEMALLOC as this saves us from propagating the allocation
5582 		 * context down to all allocation sites.
5583 		 */
5584 		noreclaim_flag = memalloc_noreclaim_save();
5585 		ret = __netif_receive_skb_one_core(skb, true);
5586 		memalloc_noreclaim_restore(noreclaim_flag);
5587 	} else
5588 		ret = __netif_receive_skb_one_core(skb, false);
5589 
5590 	return ret;
5591 }
5592 
5593 static void __netif_receive_skb_list(struct list_head *head)
5594 {
5595 	unsigned long noreclaim_flag = 0;
5596 	struct sk_buff *skb, *next;
5597 	bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5598 
5599 	list_for_each_entry_safe(skb, next, head, list) {
5600 		if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5601 			struct list_head sublist;
5602 
5603 			/* Handle the previous sublist */
5604 			list_cut_before(&sublist, head, &skb->list);
5605 			if (!list_empty(&sublist))
5606 				__netif_receive_skb_list_core(&sublist, pfmemalloc);
5607 			pfmemalloc = !pfmemalloc;
5608 			/* See comments in __netif_receive_skb */
5609 			if (pfmemalloc)
5610 				noreclaim_flag = memalloc_noreclaim_save();
5611 			else
5612 				memalloc_noreclaim_restore(noreclaim_flag);
5613 		}
5614 	}
5615 	/* Handle the remaining sublist */
5616 	if (!list_empty(head))
5617 		__netif_receive_skb_list_core(head, pfmemalloc);
5618 	/* Restore pflags */
5619 	if (pfmemalloc)
5620 		memalloc_noreclaim_restore(noreclaim_flag);
5621 }
5622 
5623 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5624 {
5625 	struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5626 	struct bpf_prog *new = xdp->prog;
5627 	int ret = 0;
5628 
5629 	switch (xdp->command) {
5630 	case XDP_SETUP_PROG:
5631 		rcu_assign_pointer(dev->xdp_prog, new);
5632 		if (old)
5633 			bpf_prog_put(old);
5634 
5635 		if (old && !new) {
5636 			static_branch_dec(&generic_xdp_needed_key);
5637 		} else if (new && !old) {
5638 			static_branch_inc(&generic_xdp_needed_key);
5639 			dev_disable_lro(dev);
5640 			dev_disable_gro_hw(dev);
5641 		}
5642 		break;
5643 
5644 	default:
5645 		ret = -EINVAL;
5646 		break;
5647 	}
5648 
5649 	return ret;
5650 }
5651 
5652 static int netif_receive_skb_internal(struct sk_buff *skb)
5653 {
5654 	int ret;
5655 
5656 	net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5657 
5658 	if (skb_defer_rx_timestamp(skb))
5659 		return NET_RX_SUCCESS;
5660 
5661 	rcu_read_lock();
5662 #ifdef CONFIG_RPS
5663 	if (static_branch_unlikely(&rps_needed)) {
5664 		struct rps_dev_flow voidflow, *rflow = &voidflow;
5665 		int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5666 
5667 		if (cpu >= 0) {
5668 			ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5669 			rcu_read_unlock();
5670 			return ret;
5671 		}
5672 	}
5673 #endif
5674 	ret = __netif_receive_skb(skb);
5675 	rcu_read_unlock();
5676 	return ret;
5677 }
5678 
5679 void netif_receive_skb_list_internal(struct list_head *head)
5680 {
5681 	struct sk_buff *skb, *next;
5682 	struct list_head sublist;
5683 
5684 	INIT_LIST_HEAD(&sublist);
5685 	list_for_each_entry_safe(skb, next, head, list) {
5686 		net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5687 		skb_list_del_init(skb);
5688 		if (!skb_defer_rx_timestamp(skb))
5689 			list_add_tail(&skb->list, &sublist);
5690 	}
5691 	list_splice_init(&sublist, head);
5692 
5693 	rcu_read_lock();
5694 #ifdef CONFIG_RPS
5695 	if (static_branch_unlikely(&rps_needed)) {
5696 		list_for_each_entry_safe(skb, next, head, list) {
5697 			struct rps_dev_flow voidflow, *rflow = &voidflow;
5698 			int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5699 
5700 			if (cpu >= 0) {
5701 				/* Will be handled, remove from list */
5702 				skb_list_del_init(skb);
5703 				enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5704 			}
5705 		}
5706 	}
5707 #endif
5708 	__netif_receive_skb_list(head);
5709 	rcu_read_unlock();
5710 }
5711 
5712 /**
5713  *	netif_receive_skb - process receive buffer from network
5714  *	@skb: buffer to process
5715  *
5716  *	netif_receive_skb() is the main receive data processing function.
5717  *	It always succeeds. The buffer may be dropped during processing
5718  *	for congestion control or by the protocol layers.
5719  *
5720  *	This function may only be called from softirq context and interrupts
5721  *	should be enabled.
5722  *
5723  *	Return values (usually ignored):
5724  *	NET_RX_SUCCESS: no congestion
5725  *	NET_RX_DROP: packet was dropped
5726  */
5727 int netif_receive_skb(struct sk_buff *skb)
5728 {
5729 	int ret;
5730 
5731 	trace_netif_receive_skb_entry(skb);
5732 
5733 	ret = netif_receive_skb_internal(skb);
5734 	trace_netif_receive_skb_exit(ret);
5735 
5736 	return ret;
5737 }
5738 EXPORT_SYMBOL(netif_receive_skb);
5739 
5740 /**
5741  *	netif_receive_skb_list - process many receive buffers from network
5742  *	@head: list of skbs to process.
5743  *
5744  *	Since return value of netif_receive_skb() is normally ignored, and
5745  *	wouldn't be meaningful for a list, this function returns void.
5746  *
5747  *	This function may only be called from softirq context and interrupts
5748  *	should be enabled.
5749  */
5750 void netif_receive_skb_list(struct list_head *head)
5751 {
5752 	struct sk_buff *skb;
5753 
5754 	if (list_empty(head))
5755 		return;
5756 	if (trace_netif_receive_skb_list_entry_enabled()) {
5757 		list_for_each_entry(skb, head, list)
5758 			trace_netif_receive_skb_list_entry(skb);
5759 	}
5760 	netif_receive_skb_list_internal(head);
5761 	trace_netif_receive_skb_list_exit(0);
5762 }
5763 EXPORT_SYMBOL(netif_receive_skb_list);
5764 
5765 static DEFINE_PER_CPU(struct work_struct, flush_works);
5766 
5767 /* Network device is going away, flush any packets still pending */
5768 static void flush_backlog(struct work_struct *work)
5769 {
5770 	struct sk_buff *skb, *tmp;
5771 	struct softnet_data *sd;
5772 
5773 	local_bh_disable();
5774 	sd = this_cpu_ptr(&softnet_data);
5775 
5776 	rps_lock_irq_disable(sd);
5777 	skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5778 		if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5779 			__skb_unlink(skb, &sd->input_pkt_queue);
5780 			dev_kfree_skb_irq(skb);
5781 			input_queue_head_incr(sd);
5782 		}
5783 	}
5784 	rps_unlock_irq_enable(sd);
5785 
5786 	skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5787 		if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5788 			__skb_unlink(skb, &sd->process_queue);
5789 			kfree_skb(skb);
5790 			input_queue_head_incr(sd);
5791 		}
5792 	}
5793 	local_bh_enable();
5794 }
5795 
5796 static bool flush_required(int cpu)
5797 {
5798 #if IS_ENABLED(CONFIG_RPS)
5799 	struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5800 	bool do_flush;
5801 
5802 	rps_lock_irq_disable(sd);
5803 
5804 	/* as insertion into process_queue happens with the rps lock held,
5805 	 * process_queue access may race only with dequeue
5806 	 */
5807 	do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5808 		   !skb_queue_empty_lockless(&sd->process_queue);
5809 	rps_unlock_irq_enable(sd);
5810 
5811 	return do_flush;
5812 #endif
5813 	/* without RPS we can't safely check input_pkt_queue: during a
5814 	 * concurrent remote skb_queue_splice() we can detect as empty both
5815 	 * input_pkt_queue and process_queue even if the latter could end-up
5816 	 * containing a lot of packets.
5817 	 */
5818 	return true;
5819 }
5820 
5821 static void flush_all_backlogs(void)
5822 {
5823 	static cpumask_t flush_cpus;
5824 	unsigned int cpu;
5825 
5826 	/* since we are under rtnl lock protection we can use static data
5827 	 * for the cpumask and avoid allocating on stack the possibly
5828 	 * large mask
5829 	 */
5830 	ASSERT_RTNL();
5831 
5832 	cpus_read_lock();
5833 
5834 	cpumask_clear(&flush_cpus);
5835 	for_each_online_cpu(cpu) {
5836 		if (flush_required(cpu)) {
5837 			queue_work_on(cpu, system_highpri_wq,
5838 				      per_cpu_ptr(&flush_works, cpu));
5839 			cpumask_set_cpu(cpu, &flush_cpus);
5840 		}
5841 	}
5842 
5843 	/* we can have in flight packet[s] on the cpus we are not flushing,
5844 	 * synchronize_net() in unregister_netdevice_many() will take care of
5845 	 * them
5846 	 */
5847 	for_each_cpu(cpu, &flush_cpus)
5848 		flush_work(per_cpu_ptr(&flush_works, cpu));
5849 
5850 	cpus_read_unlock();
5851 }
5852 
5853 static void net_rps_send_ipi(struct softnet_data *remsd)
5854 {
5855 #ifdef CONFIG_RPS
5856 	while (remsd) {
5857 		struct softnet_data *next = remsd->rps_ipi_next;
5858 
5859 		if (cpu_online(remsd->cpu))
5860 			smp_call_function_single_async(remsd->cpu, &remsd->csd);
5861 		remsd = next;
5862 	}
5863 #endif
5864 }
5865 
5866 /*
5867  * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5868  * Note: called with local irq disabled, but exits with local irq enabled.
5869  */
5870 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5871 {
5872 #ifdef CONFIG_RPS
5873 	struct softnet_data *remsd = sd->rps_ipi_list;
5874 
5875 	if (remsd) {
5876 		sd->rps_ipi_list = NULL;
5877 
5878 		local_irq_enable();
5879 
5880 		/* Send pending IPI's to kick RPS processing on remote cpus. */
5881 		net_rps_send_ipi(remsd);
5882 	} else
5883 #endif
5884 		local_irq_enable();
5885 }
5886 
5887 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5888 {
5889 #ifdef CONFIG_RPS
5890 	return sd->rps_ipi_list != NULL;
5891 #else
5892 	return false;
5893 #endif
5894 }
5895 
5896 static int process_backlog(struct napi_struct *napi, int quota)
5897 {
5898 	struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5899 	bool again = true;
5900 	int work = 0;
5901 
5902 	/* Check if we have pending ipi, its better to send them now,
5903 	 * not waiting net_rx_action() end.
5904 	 */
5905 	if (sd_has_rps_ipi_waiting(sd)) {
5906 		local_irq_disable();
5907 		net_rps_action_and_irq_enable(sd);
5908 	}
5909 
5910 	napi->weight = READ_ONCE(dev_rx_weight);
5911 	while (again) {
5912 		struct sk_buff *skb;
5913 
5914 		while ((skb = __skb_dequeue(&sd->process_queue))) {
5915 			rcu_read_lock();
5916 			__netif_receive_skb(skb);
5917 			rcu_read_unlock();
5918 			input_queue_head_incr(sd);
5919 			if (++work >= quota)
5920 				return work;
5921 
5922 		}
5923 
5924 		rps_lock_irq_disable(sd);
5925 		if (skb_queue_empty(&sd->input_pkt_queue)) {
5926 			/*
5927 			 * Inline a custom version of __napi_complete().
5928 			 * only current cpu owns and manipulates this napi,
5929 			 * and NAPI_STATE_SCHED is the only possible flag set
5930 			 * on backlog.
5931 			 * We can use a plain write instead of clear_bit(),
5932 			 * and we dont need an smp_mb() memory barrier.
5933 			 */
5934 			napi->state = 0;
5935 			again = false;
5936 		} else {
5937 			skb_queue_splice_tail_init(&sd->input_pkt_queue,
5938 						   &sd->process_queue);
5939 		}
5940 		rps_unlock_irq_enable(sd);
5941 	}
5942 
5943 	return work;
5944 }
5945 
5946 /**
5947  * __napi_schedule - schedule for receive
5948  * @n: entry to schedule
5949  *
5950  * The entry's receive function will be scheduled to run.
5951  * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5952  */
5953 void __napi_schedule(struct napi_struct *n)
5954 {
5955 	unsigned long flags;
5956 
5957 	local_irq_save(flags);
5958 	____napi_schedule(this_cpu_ptr(&softnet_data), n);
5959 	local_irq_restore(flags);
5960 }
5961 EXPORT_SYMBOL(__napi_schedule);
5962 
5963 /**
5964  *	napi_schedule_prep - check if napi can be scheduled
5965  *	@n: napi context
5966  *
5967  * Test if NAPI routine is already running, and if not mark
5968  * it as running.  This is used as a condition variable to
5969  * insure only one NAPI poll instance runs.  We also make
5970  * sure there is no pending NAPI disable.
5971  */
5972 bool napi_schedule_prep(struct napi_struct *n)
5973 {
5974 	unsigned long new, val = READ_ONCE(n->state);
5975 
5976 	do {
5977 		if (unlikely(val & NAPIF_STATE_DISABLE))
5978 			return false;
5979 		new = val | NAPIF_STATE_SCHED;
5980 
5981 		/* Sets STATE_MISSED bit if STATE_SCHED was already set
5982 		 * This was suggested by Alexander Duyck, as compiler
5983 		 * emits better code than :
5984 		 * if (val & NAPIF_STATE_SCHED)
5985 		 *     new |= NAPIF_STATE_MISSED;
5986 		 */
5987 		new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5988 						   NAPIF_STATE_MISSED;
5989 	} while (!try_cmpxchg(&n->state, &val, new));
5990 
5991 	return !(val & NAPIF_STATE_SCHED);
5992 }
5993 EXPORT_SYMBOL(napi_schedule_prep);
5994 
5995 /**
5996  * __napi_schedule_irqoff - schedule for receive
5997  * @n: entry to schedule
5998  *
5999  * Variant of __napi_schedule() assuming hard irqs are masked.
6000  *
6001  * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
6002  * because the interrupt disabled assumption might not be true
6003  * due to force-threaded interrupts and spinlock substitution.
6004  */
6005 void __napi_schedule_irqoff(struct napi_struct *n)
6006 {
6007 	if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6008 		____napi_schedule(this_cpu_ptr(&softnet_data), n);
6009 	else
6010 		__napi_schedule(n);
6011 }
6012 EXPORT_SYMBOL(__napi_schedule_irqoff);
6013 
6014 bool napi_complete_done(struct napi_struct *n, int work_done)
6015 {
6016 	unsigned long flags, val, new, timeout = 0;
6017 	bool ret = true;
6018 
6019 	/*
6020 	 * 1) Don't let napi dequeue from the cpu poll list
6021 	 *    just in case its running on a different cpu.
6022 	 * 2) If we are busy polling, do nothing here, we have
6023 	 *    the guarantee we will be called later.
6024 	 */
6025 	if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6026 				 NAPIF_STATE_IN_BUSY_POLL)))
6027 		return false;
6028 
6029 	if (work_done) {
6030 		if (n->gro_bitmask)
6031 			timeout = READ_ONCE(n->dev->gro_flush_timeout);
6032 		n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6033 	}
6034 	if (n->defer_hard_irqs_count > 0) {
6035 		n->defer_hard_irqs_count--;
6036 		timeout = READ_ONCE(n->dev->gro_flush_timeout);
6037 		if (timeout)
6038 			ret = false;
6039 	}
6040 	if (n->gro_bitmask) {
6041 		/* When the NAPI instance uses a timeout and keeps postponing
6042 		 * it, we need to bound somehow the time packets are kept in
6043 		 * the GRO layer
6044 		 */
6045 		napi_gro_flush(n, !!timeout);
6046 	}
6047 
6048 	gro_normal_list(n);
6049 
6050 	if (unlikely(!list_empty(&n->poll_list))) {
6051 		/* If n->poll_list is not empty, we need to mask irqs */
6052 		local_irq_save(flags);
6053 		list_del_init(&n->poll_list);
6054 		local_irq_restore(flags);
6055 	}
6056 
6057 	val = READ_ONCE(n->state);
6058 	do {
6059 		WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6060 
6061 		new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6062 			      NAPIF_STATE_SCHED_THREADED |
6063 			      NAPIF_STATE_PREFER_BUSY_POLL);
6064 
6065 		/* If STATE_MISSED was set, leave STATE_SCHED set,
6066 		 * because we will call napi->poll() one more time.
6067 		 * This C code was suggested by Alexander Duyck to help gcc.
6068 		 */
6069 		new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6070 						    NAPIF_STATE_SCHED;
6071 	} while (!try_cmpxchg(&n->state, &val, new));
6072 
6073 	if (unlikely(val & NAPIF_STATE_MISSED)) {
6074 		__napi_schedule(n);
6075 		return false;
6076 	}
6077 
6078 	if (timeout)
6079 		hrtimer_start(&n->timer, ns_to_ktime(timeout),
6080 			      HRTIMER_MODE_REL_PINNED);
6081 	return ret;
6082 }
6083 EXPORT_SYMBOL(napi_complete_done);
6084 
6085 /* must be called under rcu_read_lock(), as we dont take a reference */
6086 static struct napi_struct *napi_by_id(unsigned int napi_id)
6087 {
6088 	unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6089 	struct napi_struct *napi;
6090 
6091 	hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6092 		if (napi->napi_id == napi_id)
6093 			return napi;
6094 
6095 	return NULL;
6096 }
6097 
6098 #if defined(CONFIG_NET_RX_BUSY_POLL)
6099 
6100 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6101 {
6102 	if (!skip_schedule) {
6103 		gro_normal_list(napi);
6104 		__napi_schedule(napi);
6105 		return;
6106 	}
6107 
6108 	if (napi->gro_bitmask) {
6109 		/* flush too old packets
6110 		 * If HZ < 1000, flush all packets.
6111 		 */
6112 		napi_gro_flush(napi, HZ >= 1000);
6113 	}
6114 
6115 	gro_normal_list(napi);
6116 	clear_bit(NAPI_STATE_SCHED, &napi->state);
6117 }
6118 
6119 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6120 			   u16 budget)
6121 {
6122 	bool skip_schedule = false;
6123 	unsigned long timeout;
6124 	int rc;
6125 
6126 	/* Busy polling means there is a high chance device driver hard irq
6127 	 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6128 	 * set in napi_schedule_prep().
6129 	 * Since we are about to call napi->poll() once more, we can safely
6130 	 * clear NAPI_STATE_MISSED.
6131 	 *
6132 	 * Note: x86 could use a single "lock and ..." instruction
6133 	 * to perform these two clear_bit()
6134 	 */
6135 	clear_bit(NAPI_STATE_MISSED, &napi->state);
6136 	clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6137 
6138 	local_bh_disable();
6139 
6140 	if (prefer_busy_poll) {
6141 		napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6142 		timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6143 		if (napi->defer_hard_irqs_count && timeout) {
6144 			hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6145 			skip_schedule = true;
6146 		}
6147 	}
6148 
6149 	/* All we really want here is to re-enable device interrupts.
6150 	 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6151 	 */
6152 	rc = napi->poll(napi, budget);
6153 	/* We can't gro_normal_list() here, because napi->poll() might have
6154 	 * rearmed the napi (napi_complete_done()) in which case it could
6155 	 * already be running on another CPU.
6156 	 */
6157 	trace_napi_poll(napi, rc, budget);
6158 	netpoll_poll_unlock(have_poll_lock);
6159 	if (rc == budget)
6160 		__busy_poll_stop(napi, skip_schedule);
6161 	local_bh_enable();
6162 }
6163 
6164 void napi_busy_loop(unsigned int napi_id,
6165 		    bool (*loop_end)(void *, unsigned long),
6166 		    void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6167 {
6168 	unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6169 	int (*napi_poll)(struct napi_struct *napi, int budget);
6170 	void *have_poll_lock = NULL;
6171 	struct napi_struct *napi;
6172 
6173 restart:
6174 	napi_poll = NULL;
6175 
6176 	rcu_read_lock();
6177 
6178 	napi = napi_by_id(napi_id);
6179 	if (!napi)
6180 		goto out;
6181 
6182 	preempt_disable();
6183 	for (;;) {
6184 		int work = 0;
6185 
6186 		local_bh_disable();
6187 		if (!napi_poll) {
6188 			unsigned long val = READ_ONCE(napi->state);
6189 
6190 			/* If multiple threads are competing for this napi,
6191 			 * we avoid dirtying napi->state as much as we can.
6192 			 */
6193 			if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6194 				   NAPIF_STATE_IN_BUSY_POLL)) {
6195 				if (prefer_busy_poll)
6196 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6197 				goto count;
6198 			}
6199 			if (cmpxchg(&napi->state, val,
6200 				    val | NAPIF_STATE_IN_BUSY_POLL |
6201 					  NAPIF_STATE_SCHED) != val) {
6202 				if (prefer_busy_poll)
6203 					set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6204 				goto count;
6205 			}
6206 			have_poll_lock = netpoll_poll_lock(napi);
6207 			napi_poll = napi->poll;
6208 		}
6209 		work = napi_poll(napi, budget);
6210 		trace_napi_poll(napi, work, budget);
6211 		gro_normal_list(napi);
6212 count:
6213 		if (work > 0)
6214 			__NET_ADD_STATS(dev_net(napi->dev),
6215 					LINUX_MIB_BUSYPOLLRXPACKETS, work);
6216 		local_bh_enable();
6217 
6218 		if (!loop_end || loop_end(loop_end_arg, start_time))
6219 			break;
6220 
6221 		if (unlikely(need_resched())) {
6222 			if (napi_poll)
6223 				busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6224 			preempt_enable();
6225 			rcu_read_unlock();
6226 			cond_resched();
6227 			if (loop_end(loop_end_arg, start_time))
6228 				return;
6229 			goto restart;
6230 		}
6231 		cpu_relax();
6232 	}
6233 	if (napi_poll)
6234 		busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6235 	preempt_enable();
6236 out:
6237 	rcu_read_unlock();
6238 }
6239 EXPORT_SYMBOL(napi_busy_loop);
6240 
6241 #endif /* CONFIG_NET_RX_BUSY_POLL */
6242 
6243 static void napi_hash_add(struct napi_struct *napi)
6244 {
6245 	if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6246 		return;
6247 
6248 	spin_lock(&napi_hash_lock);
6249 
6250 	/* 0..NR_CPUS range is reserved for sender_cpu use */
6251 	do {
6252 		if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6253 			napi_gen_id = MIN_NAPI_ID;
6254 	} while (napi_by_id(napi_gen_id));
6255 	napi->napi_id = napi_gen_id;
6256 
6257 	hlist_add_head_rcu(&napi->napi_hash_node,
6258 			   &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6259 
6260 	spin_unlock(&napi_hash_lock);
6261 }
6262 
6263 /* Warning : caller is responsible to make sure rcu grace period
6264  * is respected before freeing memory containing @napi
6265  */
6266 static void napi_hash_del(struct napi_struct *napi)
6267 {
6268 	spin_lock(&napi_hash_lock);
6269 
6270 	hlist_del_init_rcu(&napi->napi_hash_node);
6271 
6272 	spin_unlock(&napi_hash_lock);
6273 }
6274 
6275 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6276 {
6277 	struct napi_struct *napi;
6278 
6279 	napi = container_of(timer, struct napi_struct, timer);
6280 
6281 	/* Note : we use a relaxed variant of napi_schedule_prep() not setting
6282 	 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6283 	 */
6284 	if (!napi_disable_pending(napi) &&
6285 	    !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6286 		clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6287 		__napi_schedule_irqoff(napi);
6288 	}
6289 
6290 	return HRTIMER_NORESTART;
6291 }
6292 
6293 static void init_gro_hash(struct napi_struct *napi)
6294 {
6295 	int i;
6296 
6297 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6298 		INIT_LIST_HEAD(&napi->gro_hash[i].list);
6299 		napi->gro_hash[i].count = 0;
6300 	}
6301 	napi->gro_bitmask = 0;
6302 }
6303 
6304 int dev_set_threaded(struct net_device *dev, bool threaded)
6305 {
6306 	struct napi_struct *napi;
6307 	int err = 0;
6308 
6309 	if (dev->threaded == threaded)
6310 		return 0;
6311 
6312 	if (threaded) {
6313 		list_for_each_entry(napi, &dev->napi_list, dev_list) {
6314 			if (!napi->thread) {
6315 				err = napi_kthread_create(napi);
6316 				if (err) {
6317 					threaded = false;
6318 					break;
6319 				}
6320 			}
6321 		}
6322 	}
6323 
6324 	dev->threaded = threaded;
6325 
6326 	/* Make sure kthread is created before THREADED bit
6327 	 * is set.
6328 	 */
6329 	smp_mb__before_atomic();
6330 
6331 	/* Setting/unsetting threaded mode on a napi might not immediately
6332 	 * take effect, if the current napi instance is actively being
6333 	 * polled. In this case, the switch between threaded mode and
6334 	 * softirq mode will happen in the next round of napi_schedule().
6335 	 * This should not cause hiccups/stalls to the live traffic.
6336 	 */
6337 	list_for_each_entry(napi, &dev->napi_list, dev_list) {
6338 		if (threaded)
6339 			set_bit(NAPI_STATE_THREADED, &napi->state);
6340 		else
6341 			clear_bit(NAPI_STATE_THREADED, &napi->state);
6342 	}
6343 
6344 	return err;
6345 }
6346 EXPORT_SYMBOL(dev_set_threaded);
6347 
6348 void netif_napi_add_weight(struct net_device *dev, struct napi_struct *napi,
6349 			   int (*poll)(struct napi_struct *, int), int weight)
6350 {
6351 	if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6352 		return;
6353 
6354 	INIT_LIST_HEAD(&napi->poll_list);
6355 	INIT_HLIST_NODE(&napi->napi_hash_node);
6356 	hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6357 	napi->timer.function = napi_watchdog;
6358 	init_gro_hash(napi);
6359 	napi->skb = NULL;
6360 	INIT_LIST_HEAD(&napi->rx_list);
6361 	napi->rx_count = 0;
6362 	napi->poll = poll;
6363 	if (weight > NAPI_POLL_WEIGHT)
6364 		netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6365 				weight);
6366 	napi->weight = weight;
6367 	napi->dev = dev;
6368 #ifdef CONFIG_NETPOLL
6369 	napi->poll_owner = -1;
6370 #endif
6371 	set_bit(NAPI_STATE_SCHED, &napi->state);
6372 	set_bit(NAPI_STATE_NPSVC, &napi->state);
6373 	list_add_rcu(&napi->dev_list, &dev->napi_list);
6374 	napi_hash_add(napi);
6375 	napi_get_frags_check(napi);
6376 	/* Create kthread for this napi if dev->threaded is set.
6377 	 * Clear dev->threaded if kthread creation failed so that
6378 	 * threaded mode will not be enabled in napi_enable().
6379 	 */
6380 	if (dev->threaded && napi_kthread_create(napi))
6381 		dev->threaded = 0;
6382 }
6383 EXPORT_SYMBOL(netif_napi_add_weight);
6384 
6385 void napi_disable(struct napi_struct *n)
6386 {
6387 	unsigned long val, new;
6388 
6389 	might_sleep();
6390 	set_bit(NAPI_STATE_DISABLE, &n->state);
6391 
6392 	val = READ_ONCE(n->state);
6393 	do {
6394 		while (val & (NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC)) {
6395 			usleep_range(20, 200);
6396 			val = READ_ONCE(n->state);
6397 		}
6398 
6399 		new = val | NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC;
6400 		new &= ~(NAPIF_STATE_THREADED | NAPIF_STATE_PREFER_BUSY_POLL);
6401 	} while (!try_cmpxchg(&n->state, &val, new));
6402 
6403 	hrtimer_cancel(&n->timer);
6404 
6405 	clear_bit(NAPI_STATE_DISABLE, &n->state);
6406 }
6407 EXPORT_SYMBOL(napi_disable);
6408 
6409 /**
6410  *	napi_enable - enable NAPI scheduling
6411  *	@n: NAPI context
6412  *
6413  * Resume NAPI from being scheduled on this context.
6414  * Must be paired with napi_disable.
6415  */
6416 void napi_enable(struct napi_struct *n)
6417 {
6418 	unsigned long new, val = READ_ONCE(n->state);
6419 
6420 	do {
6421 		BUG_ON(!test_bit(NAPI_STATE_SCHED, &val));
6422 
6423 		new = val & ~(NAPIF_STATE_SCHED | NAPIF_STATE_NPSVC);
6424 		if (n->dev->threaded && n->thread)
6425 			new |= NAPIF_STATE_THREADED;
6426 	} while (!try_cmpxchg(&n->state, &val, new));
6427 }
6428 EXPORT_SYMBOL(napi_enable);
6429 
6430 static void flush_gro_hash(struct napi_struct *napi)
6431 {
6432 	int i;
6433 
6434 	for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6435 		struct sk_buff *skb, *n;
6436 
6437 		list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6438 			kfree_skb(skb);
6439 		napi->gro_hash[i].count = 0;
6440 	}
6441 }
6442 
6443 /* Must be called in process context */
6444 void __netif_napi_del(struct napi_struct *napi)
6445 {
6446 	if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6447 		return;
6448 
6449 	napi_hash_del(napi);
6450 	list_del_rcu(&napi->dev_list);
6451 	napi_free_frags(napi);
6452 
6453 	flush_gro_hash(napi);
6454 	napi->gro_bitmask = 0;
6455 
6456 	if (napi->thread) {
6457 		kthread_stop(napi->thread);
6458 		napi->thread = NULL;
6459 	}
6460 }
6461 EXPORT_SYMBOL(__netif_napi_del);
6462 
6463 static int __napi_poll(struct napi_struct *n, bool *repoll)
6464 {
6465 	int work, weight;
6466 
6467 	weight = n->weight;
6468 
6469 	/* This NAPI_STATE_SCHED test is for avoiding a race
6470 	 * with netpoll's poll_napi().  Only the entity which
6471 	 * obtains the lock and sees NAPI_STATE_SCHED set will
6472 	 * actually make the ->poll() call.  Therefore we avoid
6473 	 * accidentally calling ->poll() when NAPI is not scheduled.
6474 	 */
6475 	work = 0;
6476 	if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6477 		work = n->poll(n, weight);
6478 		trace_napi_poll(n, work, weight);
6479 	}
6480 
6481 	if (unlikely(work > weight))
6482 		netdev_err_once(n->dev, "NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6483 				n->poll, work, weight);
6484 
6485 	if (likely(work < weight))
6486 		return work;
6487 
6488 	/* Drivers must not modify the NAPI state if they
6489 	 * consume the entire weight.  In such cases this code
6490 	 * still "owns" the NAPI instance and therefore can
6491 	 * move the instance around on the list at-will.
6492 	 */
6493 	if (unlikely(napi_disable_pending(n))) {
6494 		napi_complete(n);
6495 		return work;
6496 	}
6497 
6498 	/* The NAPI context has more processing work, but busy-polling
6499 	 * is preferred. Exit early.
6500 	 */
6501 	if (napi_prefer_busy_poll(n)) {
6502 		if (napi_complete_done(n, work)) {
6503 			/* If timeout is not set, we need to make sure
6504 			 * that the NAPI is re-scheduled.
6505 			 */
6506 			napi_schedule(n);
6507 		}
6508 		return work;
6509 	}
6510 
6511 	if (n->gro_bitmask) {
6512 		/* flush too old packets
6513 		 * If HZ < 1000, flush all packets.
6514 		 */
6515 		napi_gro_flush(n, HZ >= 1000);
6516 	}
6517 
6518 	gro_normal_list(n);
6519 
6520 	/* Some drivers may have called napi_schedule
6521 	 * prior to exhausting their budget.
6522 	 */
6523 	if (unlikely(!list_empty(&n->poll_list))) {
6524 		pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6525 			     n->dev ? n->dev->name : "backlog");
6526 		return work;
6527 	}
6528 
6529 	*repoll = true;
6530 
6531 	return work;
6532 }
6533 
6534 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6535 {
6536 	bool do_repoll = false;
6537 	void *have;
6538 	int work;
6539 
6540 	list_del_init(&n->poll_list);
6541 
6542 	have = netpoll_poll_lock(n);
6543 
6544 	work = __napi_poll(n, &do_repoll);
6545 
6546 	if (do_repoll)
6547 		list_add_tail(&n->poll_list, repoll);
6548 
6549 	netpoll_poll_unlock(have);
6550 
6551 	return work;
6552 }
6553 
6554 static int napi_thread_wait(struct napi_struct *napi)
6555 {
6556 	bool woken = false;
6557 
6558 	set_current_state(TASK_INTERRUPTIBLE);
6559 
6560 	while (!kthread_should_stop()) {
6561 		/* Testing SCHED_THREADED bit here to make sure the current
6562 		 * kthread owns this napi and could poll on this napi.
6563 		 * Testing SCHED bit is not enough because SCHED bit might be
6564 		 * set by some other busy poll thread or by napi_disable().
6565 		 */
6566 		if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
6567 			WARN_ON(!list_empty(&napi->poll_list));
6568 			__set_current_state(TASK_RUNNING);
6569 			return 0;
6570 		}
6571 
6572 		schedule();
6573 		/* woken being true indicates this thread owns this napi. */
6574 		woken = true;
6575 		set_current_state(TASK_INTERRUPTIBLE);
6576 	}
6577 	__set_current_state(TASK_RUNNING);
6578 
6579 	return -1;
6580 }
6581 
6582 static int napi_threaded_poll(void *data)
6583 {
6584 	struct napi_struct *napi = data;
6585 	void *have;
6586 
6587 	while (!napi_thread_wait(napi)) {
6588 		for (;;) {
6589 			bool repoll = false;
6590 
6591 			local_bh_disable();
6592 
6593 			have = netpoll_poll_lock(napi);
6594 			__napi_poll(napi, &repoll);
6595 			netpoll_poll_unlock(have);
6596 
6597 			local_bh_enable();
6598 
6599 			if (!repoll)
6600 				break;
6601 
6602 			cond_resched();
6603 		}
6604 	}
6605 	return 0;
6606 }
6607 
6608 static void skb_defer_free_flush(struct softnet_data *sd)
6609 {
6610 	struct sk_buff *skb, *next;
6611 	unsigned long flags;
6612 
6613 	/* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
6614 	if (!READ_ONCE(sd->defer_list))
6615 		return;
6616 
6617 	spin_lock_irqsave(&sd->defer_lock, flags);
6618 	skb = sd->defer_list;
6619 	sd->defer_list = NULL;
6620 	sd->defer_count = 0;
6621 	spin_unlock_irqrestore(&sd->defer_lock, flags);
6622 
6623 	while (skb != NULL) {
6624 		next = skb->next;
6625 		napi_consume_skb(skb, 1);
6626 		skb = next;
6627 	}
6628 }
6629 
6630 static __latent_entropy void net_rx_action(struct softirq_action *h)
6631 {
6632 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6633 	unsigned long time_limit = jiffies +
6634 		usecs_to_jiffies(READ_ONCE(netdev_budget_usecs));
6635 	int budget = READ_ONCE(netdev_budget);
6636 	LIST_HEAD(list);
6637 	LIST_HEAD(repoll);
6638 
6639 	local_irq_disable();
6640 	list_splice_init(&sd->poll_list, &list);
6641 	local_irq_enable();
6642 
6643 	for (;;) {
6644 		struct napi_struct *n;
6645 
6646 		skb_defer_free_flush(sd);
6647 
6648 		if (list_empty(&list)) {
6649 			if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6650 				goto end;
6651 			break;
6652 		}
6653 
6654 		n = list_first_entry(&list, struct napi_struct, poll_list);
6655 		budget -= napi_poll(n, &repoll);
6656 
6657 		/* If softirq window is exhausted then punt.
6658 		 * Allow this to run for 2 jiffies since which will allow
6659 		 * an average latency of 1.5/HZ.
6660 		 */
6661 		if (unlikely(budget <= 0 ||
6662 			     time_after_eq(jiffies, time_limit))) {
6663 			sd->time_squeeze++;
6664 			break;
6665 		}
6666 	}
6667 
6668 	local_irq_disable();
6669 
6670 	list_splice_tail_init(&sd->poll_list, &list);
6671 	list_splice_tail(&repoll, &list);
6672 	list_splice(&list, &sd->poll_list);
6673 	if (!list_empty(&sd->poll_list))
6674 		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
6675 
6676 	net_rps_action_and_irq_enable(sd);
6677 end:;
6678 }
6679 
6680 struct netdev_adjacent {
6681 	struct net_device *dev;
6682 	netdevice_tracker dev_tracker;
6683 
6684 	/* upper master flag, there can only be one master device per list */
6685 	bool master;
6686 
6687 	/* lookup ignore flag */
6688 	bool ignore;
6689 
6690 	/* counter for the number of times this device was added to us */
6691 	u16 ref_nr;
6692 
6693 	/* private field for the users */
6694 	void *private;
6695 
6696 	struct list_head list;
6697 	struct rcu_head rcu;
6698 };
6699 
6700 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6701 						 struct list_head *adj_list)
6702 {
6703 	struct netdev_adjacent *adj;
6704 
6705 	list_for_each_entry(adj, adj_list, list) {
6706 		if (adj->dev == adj_dev)
6707 			return adj;
6708 	}
6709 	return NULL;
6710 }
6711 
6712 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
6713 				    struct netdev_nested_priv *priv)
6714 {
6715 	struct net_device *dev = (struct net_device *)priv->data;
6716 
6717 	return upper_dev == dev;
6718 }
6719 
6720 /**
6721  * netdev_has_upper_dev - Check if device is linked to an upper device
6722  * @dev: device
6723  * @upper_dev: upper device to check
6724  *
6725  * Find out if a device is linked to specified upper device and return true
6726  * in case it is. Note that this checks only immediate upper device,
6727  * not through a complete stack of devices. The caller must hold the RTNL lock.
6728  */
6729 bool netdev_has_upper_dev(struct net_device *dev,
6730 			  struct net_device *upper_dev)
6731 {
6732 	struct netdev_nested_priv priv = {
6733 		.data = (void *)upper_dev,
6734 	};
6735 
6736 	ASSERT_RTNL();
6737 
6738 	return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6739 					     &priv);
6740 }
6741 EXPORT_SYMBOL(netdev_has_upper_dev);
6742 
6743 /**
6744  * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
6745  * @dev: device
6746  * @upper_dev: upper device to check
6747  *
6748  * Find out if a device is linked to specified upper device and return true
6749  * in case it is. Note that this checks the entire upper device chain.
6750  * The caller must hold rcu lock.
6751  */
6752 
6753 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6754 				  struct net_device *upper_dev)
6755 {
6756 	struct netdev_nested_priv priv = {
6757 		.data = (void *)upper_dev,
6758 	};
6759 
6760 	return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6761 					       &priv);
6762 }
6763 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6764 
6765 /**
6766  * netdev_has_any_upper_dev - Check if device is linked to some device
6767  * @dev: device
6768  *
6769  * Find out if a device is linked to an upper device and return true in case
6770  * it is. The caller must hold the RTNL lock.
6771  */
6772 bool netdev_has_any_upper_dev(struct net_device *dev)
6773 {
6774 	ASSERT_RTNL();
6775 
6776 	return !list_empty(&dev->adj_list.upper);
6777 }
6778 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6779 
6780 /**
6781  * netdev_master_upper_dev_get - Get master upper device
6782  * @dev: device
6783  *
6784  * Find a master upper device and return pointer to it or NULL in case
6785  * it's not there. The caller must hold the RTNL lock.
6786  */
6787 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6788 {
6789 	struct netdev_adjacent *upper;
6790 
6791 	ASSERT_RTNL();
6792 
6793 	if (list_empty(&dev->adj_list.upper))
6794 		return NULL;
6795 
6796 	upper = list_first_entry(&dev->adj_list.upper,
6797 				 struct netdev_adjacent, list);
6798 	if (likely(upper->master))
6799 		return upper->dev;
6800 	return NULL;
6801 }
6802 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6803 
6804 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6805 {
6806 	struct netdev_adjacent *upper;
6807 
6808 	ASSERT_RTNL();
6809 
6810 	if (list_empty(&dev->adj_list.upper))
6811 		return NULL;
6812 
6813 	upper = list_first_entry(&dev->adj_list.upper,
6814 				 struct netdev_adjacent, list);
6815 	if (likely(upper->master) && !upper->ignore)
6816 		return upper->dev;
6817 	return NULL;
6818 }
6819 
6820 /**
6821  * netdev_has_any_lower_dev - Check if device is linked to some device
6822  * @dev: device
6823  *
6824  * Find out if a device is linked to a lower device and return true in case
6825  * it is. The caller must hold the RTNL lock.
6826  */
6827 static bool netdev_has_any_lower_dev(struct net_device *dev)
6828 {
6829 	ASSERT_RTNL();
6830 
6831 	return !list_empty(&dev->adj_list.lower);
6832 }
6833 
6834 void *netdev_adjacent_get_private(struct list_head *adj_list)
6835 {
6836 	struct netdev_adjacent *adj;
6837 
6838 	adj = list_entry(adj_list, struct netdev_adjacent, list);
6839 
6840 	return adj->private;
6841 }
6842 EXPORT_SYMBOL(netdev_adjacent_get_private);
6843 
6844 /**
6845  * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6846  * @dev: device
6847  * @iter: list_head ** of the current position
6848  *
6849  * Gets the next device from the dev's upper list, starting from iter
6850  * position. The caller must hold RCU read lock.
6851  */
6852 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6853 						 struct list_head **iter)
6854 {
6855 	struct netdev_adjacent *upper;
6856 
6857 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6858 
6859 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6860 
6861 	if (&upper->list == &dev->adj_list.upper)
6862 		return NULL;
6863 
6864 	*iter = &upper->list;
6865 
6866 	return upper->dev;
6867 }
6868 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6869 
6870 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6871 						  struct list_head **iter,
6872 						  bool *ignore)
6873 {
6874 	struct netdev_adjacent *upper;
6875 
6876 	upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6877 
6878 	if (&upper->list == &dev->adj_list.upper)
6879 		return NULL;
6880 
6881 	*iter = &upper->list;
6882 	*ignore = upper->ignore;
6883 
6884 	return upper->dev;
6885 }
6886 
6887 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6888 						    struct list_head **iter)
6889 {
6890 	struct netdev_adjacent *upper;
6891 
6892 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6893 
6894 	upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6895 
6896 	if (&upper->list == &dev->adj_list.upper)
6897 		return NULL;
6898 
6899 	*iter = &upper->list;
6900 
6901 	return upper->dev;
6902 }
6903 
6904 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6905 				       int (*fn)(struct net_device *dev,
6906 					 struct netdev_nested_priv *priv),
6907 				       struct netdev_nested_priv *priv)
6908 {
6909 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6910 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6911 	int ret, cur = 0;
6912 	bool ignore;
6913 
6914 	now = dev;
6915 	iter = &dev->adj_list.upper;
6916 
6917 	while (1) {
6918 		if (now != dev) {
6919 			ret = fn(now, priv);
6920 			if (ret)
6921 				return ret;
6922 		}
6923 
6924 		next = NULL;
6925 		while (1) {
6926 			udev = __netdev_next_upper_dev(now, &iter, &ignore);
6927 			if (!udev)
6928 				break;
6929 			if (ignore)
6930 				continue;
6931 
6932 			next = udev;
6933 			niter = &udev->adj_list.upper;
6934 			dev_stack[cur] = now;
6935 			iter_stack[cur++] = iter;
6936 			break;
6937 		}
6938 
6939 		if (!next) {
6940 			if (!cur)
6941 				return 0;
6942 			next = dev_stack[--cur];
6943 			niter = iter_stack[cur];
6944 		}
6945 
6946 		now = next;
6947 		iter = niter;
6948 	}
6949 
6950 	return 0;
6951 }
6952 
6953 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6954 				  int (*fn)(struct net_device *dev,
6955 					    struct netdev_nested_priv *priv),
6956 				  struct netdev_nested_priv *priv)
6957 {
6958 	struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6959 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6960 	int ret, cur = 0;
6961 
6962 	now = dev;
6963 	iter = &dev->adj_list.upper;
6964 
6965 	while (1) {
6966 		if (now != dev) {
6967 			ret = fn(now, priv);
6968 			if (ret)
6969 				return ret;
6970 		}
6971 
6972 		next = NULL;
6973 		while (1) {
6974 			udev = netdev_next_upper_dev_rcu(now, &iter);
6975 			if (!udev)
6976 				break;
6977 
6978 			next = udev;
6979 			niter = &udev->adj_list.upper;
6980 			dev_stack[cur] = now;
6981 			iter_stack[cur++] = iter;
6982 			break;
6983 		}
6984 
6985 		if (!next) {
6986 			if (!cur)
6987 				return 0;
6988 			next = dev_stack[--cur];
6989 			niter = iter_stack[cur];
6990 		}
6991 
6992 		now = next;
6993 		iter = niter;
6994 	}
6995 
6996 	return 0;
6997 }
6998 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6999 
7000 static bool __netdev_has_upper_dev(struct net_device *dev,
7001 				   struct net_device *upper_dev)
7002 {
7003 	struct netdev_nested_priv priv = {
7004 		.flags = 0,
7005 		.data = (void *)upper_dev,
7006 	};
7007 
7008 	ASSERT_RTNL();
7009 
7010 	return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7011 					   &priv);
7012 }
7013 
7014 /**
7015  * netdev_lower_get_next_private - Get the next ->private from the
7016  *				   lower neighbour list
7017  * @dev: device
7018  * @iter: list_head ** of the current position
7019  *
7020  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7021  * list, starting from iter position. The caller must hold either hold the
7022  * RTNL lock or its own locking that guarantees that the neighbour lower
7023  * list will remain unchanged.
7024  */
7025 void *netdev_lower_get_next_private(struct net_device *dev,
7026 				    struct list_head **iter)
7027 {
7028 	struct netdev_adjacent *lower;
7029 
7030 	lower = list_entry(*iter, struct netdev_adjacent, list);
7031 
7032 	if (&lower->list == &dev->adj_list.lower)
7033 		return NULL;
7034 
7035 	*iter = lower->list.next;
7036 
7037 	return lower->private;
7038 }
7039 EXPORT_SYMBOL(netdev_lower_get_next_private);
7040 
7041 /**
7042  * netdev_lower_get_next_private_rcu - Get the next ->private from the
7043  *				       lower neighbour list, RCU
7044  *				       variant
7045  * @dev: device
7046  * @iter: list_head ** of the current position
7047  *
7048  * Gets the next netdev_adjacent->private from the dev's lower neighbour
7049  * list, starting from iter position. The caller must hold RCU read lock.
7050  */
7051 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7052 					struct list_head **iter)
7053 {
7054 	struct netdev_adjacent *lower;
7055 
7056 	WARN_ON_ONCE(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
7057 
7058 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7059 
7060 	if (&lower->list == &dev->adj_list.lower)
7061 		return NULL;
7062 
7063 	*iter = &lower->list;
7064 
7065 	return lower->private;
7066 }
7067 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7068 
7069 /**
7070  * netdev_lower_get_next - Get the next device from the lower neighbour
7071  *                         list
7072  * @dev: device
7073  * @iter: list_head ** of the current position
7074  *
7075  * Gets the next netdev_adjacent from the dev's lower neighbour
7076  * list, starting from iter position. The caller must hold RTNL lock or
7077  * its own locking that guarantees that the neighbour lower
7078  * list will remain unchanged.
7079  */
7080 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7081 {
7082 	struct netdev_adjacent *lower;
7083 
7084 	lower = list_entry(*iter, struct netdev_adjacent, list);
7085 
7086 	if (&lower->list == &dev->adj_list.lower)
7087 		return NULL;
7088 
7089 	*iter = lower->list.next;
7090 
7091 	return lower->dev;
7092 }
7093 EXPORT_SYMBOL(netdev_lower_get_next);
7094 
7095 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7096 						struct list_head **iter)
7097 {
7098 	struct netdev_adjacent *lower;
7099 
7100 	lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7101 
7102 	if (&lower->list == &dev->adj_list.lower)
7103 		return NULL;
7104 
7105 	*iter = &lower->list;
7106 
7107 	return lower->dev;
7108 }
7109 
7110 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7111 						  struct list_head **iter,
7112 						  bool *ignore)
7113 {
7114 	struct netdev_adjacent *lower;
7115 
7116 	lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7117 
7118 	if (&lower->list == &dev->adj_list.lower)
7119 		return NULL;
7120 
7121 	*iter = &lower->list;
7122 	*ignore = lower->ignore;
7123 
7124 	return lower->dev;
7125 }
7126 
7127 int netdev_walk_all_lower_dev(struct net_device *dev,
7128 			      int (*fn)(struct net_device *dev,
7129 					struct netdev_nested_priv *priv),
7130 			      struct netdev_nested_priv *priv)
7131 {
7132 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7133 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7134 	int ret, cur = 0;
7135 
7136 	now = dev;
7137 	iter = &dev->adj_list.lower;
7138 
7139 	while (1) {
7140 		if (now != dev) {
7141 			ret = fn(now, priv);
7142 			if (ret)
7143 				return ret;
7144 		}
7145 
7146 		next = NULL;
7147 		while (1) {
7148 			ldev = netdev_next_lower_dev(now, &iter);
7149 			if (!ldev)
7150 				break;
7151 
7152 			next = ldev;
7153 			niter = &ldev->adj_list.lower;
7154 			dev_stack[cur] = now;
7155 			iter_stack[cur++] = iter;
7156 			break;
7157 		}
7158 
7159 		if (!next) {
7160 			if (!cur)
7161 				return 0;
7162 			next = dev_stack[--cur];
7163 			niter = iter_stack[cur];
7164 		}
7165 
7166 		now = next;
7167 		iter = niter;
7168 	}
7169 
7170 	return 0;
7171 }
7172 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7173 
7174 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7175 				       int (*fn)(struct net_device *dev,
7176 					 struct netdev_nested_priv *priv),
7177 				       struct netdev_nested_priv *priv)
7178 {
7179 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7180 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7181 	int ret, cur = 0;
7182 	bool ignore;
7183 
7184 	now = dev;
7185 	iter = &dev->adj_list.lower;
7186 
7187 	while (1) {
7188 		if (now != dev) {
7189 			ret = fn(now, priv);
7190 			if (ret)
7191 				return ret;
7192 		}
7193 
7194 		next = NULL;
7195 		while (1) {
7196 			ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7197 			if (!ldev)
7198 				break;
7199 			if (ignore)
7200 				continue;
7201 
7202 			next = ldev;
7203 			niter = &ldev->adj_list.lower;
7204 			dev_stack[cur] = now;
7205 			iter_stack[cur++] = iter;
7206 			break;
7207 		}
7208 
7209 		if (!next) {
7210 			if (!cur)
7211 				return 0;
7212 			next = dev_stack[--cur];
7213 			niter = iter_stack[cur];
7214 		}
7215 
7216 		now = next;
7217 		iter = niter;
7218 	}
7219 
7220 	return 0;
7221 }
7222 
7223 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7224 					     struct list_head **iter)
7225 {
7226 	struct netdev_adjacent *lower;
7227 
7228 	lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7229 	if (&lower->list == &dev->adj_list.lower)
7230 		return NULL;
7231 
7232 	*iter = &lower->list;
7233 
7234 	return lower->dev;
7235 }
7236 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7237 
7238 static u8 __netdev_upper_depth(struct net_device *dev)
7239 {
7240 	struct net_device *udev;
7241 	struct list_head *iter;
7242 	u8 max_depth = 0;
7243 	bool ignore;
7244 
7245 	for (iter = &dev->adj_list.upper,
7246 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7247 	     udev;
7248 	     udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7249 		if (ignore)
7250 			continue;
7251 		if (max_depth < udev->upper_level)
7252 			max_depth = udev->upper_level;
7253 	}
7254 
7255 	return max_depth;
7256 }
7257 
7258 static u8 __netdev_lower_depth(struct net_device *dev)
7259 {
7260 	struct net_device *ldev;
7261 	struct list_head *iter;
7262 	u8 max_depth = 0;
7263 	bool ignore;
7264 
7265 	for (iter = &dev->adj_list.lower,
7266 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7267 	     ldev;
7268 	     ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7269 		if (ignore)
7270 			continue;
7271 		if (max_depth < ldev->lower_level)
7272 			max_depth = ldev->lower_level;
7273 	}
7274 
7275 	return max_depth;
7276 }
7277 
7278 static int __netdev_update_upper_level(struct net_device *dev,
7279 				       struct netdev_nested_priv *__unused)
7280 {
7281 	dev->upper_level = __netdev_upper_depth(dev) + 1;
7282 	return 0;
7283 }
7284 
7285 #ifdef CONFIG_LOCKDEP
7286 static LIST_HEAD(net_unlink_list);
7287 
7288 static void net_unlink_todo(struct net_device *dev)
7289 {
7290 	if (list_empty(&dev->unlink_list))
7291 		list_add_tail(&dev->unlink_list, &net_unlink_list);
7292 }
7293 #endif
7294 
7295 static int __netdev_update_lower_level(struct net_device *dev,
7296 				       struct netdev_nested_priv *priv)
7297 {
7298 	dev->lower_level = __netdev_lower_depth(dev) + 1;
7299 
7300 #ifdef CONFIG_LOCKDEP
7301 	if (!priv)
7302 		return 0;
7303 
7304 	if (priv->flags & NESTED_SYNC_IMM)
7305 		dev->nested_level = dev->lower_level - 1;
7306 	if (priv->flags & NESTED_SYNC_TODO)
7307 		net_unlink_todo(dev);
7308 #endif
7309 	return 0;
7310 }
7311 
7312 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7313 				  int (*fn)(struct net_device *dev,
7314 					    struct netdev_nested_priv *priv),
7315 				  struct netdev_nested_priv *priv)
7316 {
7317 	struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7318 	struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7319 	int ret, cur = 0;
7320 
7321 	now = dev;
7322 	iter = &dev->adj_list.lower;
7323 
7324 	while (1) {
7325 		if (now != dev) {
7326 			ret = fn(now, priv);
7327 			if (ret)
7328 				return ret;
7329 		}
7330 
7331 		next = NULL;
7332 		while (1) {
7333 			ldev = netdev_next_lower_dev_rcu(now, &iter);
7334 			if (!ldev)
7335 				break;
7336 
7337 			next = ldev;
7338 			niter = &ldev->adj_list.lower;
7339 			dev_stack[cur] = now;
7340 			iter_stack[cur++] = iter;
7341 			break;
7342 		}
7343 
7344 		if (!next) {
7345 			if (!cur)
7346 				return 0;
7347 			next = dev_stack[--cur];
7348 			niter = iter_stack[cur];
7349 		}
7350 
7351 		now = next;
7352 		iter = niter;
7353 	}
7354 
7355 	return 0;
7356 }
7357 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7358 
7359 /**
7360  * netdev_lower_get_first_private_rcu - Get the first ->private from the
7361  *				       lower neighbour list, RCU
7362  *				       variant
7363  * @dev: device
7364  *
7365  * Gets the first netdev_adjacent->private from the dev's lower neighbour
7366  * list. The caller must hold RCU read lock.
7367  */
7368 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7369 {
7370 	struct netdev_adjacent *lower;
7371 
7372 	lower = list_first_or_null_rcu(&dev->adj_list.lower,
7373 			struct netdev_adjacent, list);
7374 	if (lower)
7375 		return lower->private;
7376 	return NULL;
7377 }
7378 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7379 
7380 /**
7381  * netdev_master_upper_dev_get_rcu - Get master upper device
7382  * @dev: device
7383  *
7384  * Find a master upper device and return pointer to it or NULL in case
7385  * it's not there. The caller must hold the RCU read lock.
7386  */
7387 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7388 {
7389 	struct netdev_adjacent *upper;
7390 
7391 	upper = list_first_or_null_rcu(&dev->adj_list.upper,
7392 				       struct netdev_adjacent, list);
7393 	if (upper && likely(upper->master))
7394 		return upper->dev;
7395 	return NULL;
7396 }
7397 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7398 
7399 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7400 			      struct net_device *adj_dev,
7401 			      struct list_head *dev_list)
7402 {
7403 	char linkname[IFNAMSIZ+7];
7404 
7405 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7406 		"upper_%s" : "lower_%s", adj_dev->name);
7407 	return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7408 				 linkname);
7409 }
7410 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7411 			       char *name,
7412 			       struct list_head *dev_list)
7413 {
7414 	char linkname[IFNAMSIZ+7];
7415 
7416 	sprintf(linkname, dev_list == &dev->adj_list.upper ?
7417 		"upper_%s" : "lower_%s", name);
7418 	sysfs_remove_link(&(dev->dev.kobj), linkname);
7419 }
7420 
7421 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7422 						 struct net_device *adj_dev,
7423 						 struct list_head *dev_list)
7424 {
7425 	return (dev_list == &dev->adj_list.upper ||
7426 		dev_list == &dev->adj_list.lower) &&
7427 		net_eq(dev_net(dev), dev_net(adj_dev));
7428 }
7429 
7430 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7431 					struct net_device *adj_dev,
7432 					struct list_head *dev_list,
7433 					void *private, bool master)
7434 {
7435 	struct netdev_adjacent *adj;
7436 	int ret;
7437 
7438 	adj = __netdev_find_adj(adj_dev, dev_list);
7439 
7440 	if (adj) {
7441 		adj->ref_nr += 1;
7442 		pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7443 			 dev->name, adj_dev->name, adj->ref_nr);
7444 
7445 		return 0;
7446 	}
7447 
7448 	adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7449 	if (!adj)
7450 		return -ENOMEM;
7451 
7452 	adj->dev = adj_dev;
7453 	adj->master = master;
7454 	adj->ref_nr = 1;
7455 	adj->private = private;
7456 	adj->ignore = false;
7457 	netdev_hold(adj_dev, &adj->dev_tracker, GFP_KERNEL);
7458 
7459 	pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7460 		 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7461 
7462 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7463 		ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7464 		if (ret)
7465 			goto free_adj;
7466 	}
7467 
7468 	/* Ensure that master link is always the first item in list. */
7469 	if (master) {
7470 		ret = sysfs_create_link(&(dev->dev.kobj),
7471 					&(adj_dev->dev.kobj), "master");
7472 		if (ret)
7473 			goto remove_symlinks;
7474 
7475 		list_add_rcu(&adj->list, dev_list);
7476 	} else {
7477 		list_add_tail_rcu(&adj->list, dev_list);
7478 	}
7479 
7480 	return 0;
7481 
7482 remove_symlinks:
7483 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7484 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7485 free_adj:
7486 	netdev_put(adj_dev, &adj->dev_tracker);
7487 	kfree(adj);
7488 
7489 	return ret;
7490 }
7491 
7492 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7493 					 struct net_device *adj_dev,
7494 					 u16 ref_nr,
7495 					 struct list_head *dev_list)
7496 {
7497 	struct netdev_adjacent *adj;
7498 
7499 	pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7500 		 dev->name, adj_dev->name, ref_nr);
7501 
7502 	adj = __netdev_find_adj(adj_dev, dev_list);
7503 
7504 	if (!adj) {
7505 		pr_err("Adjacency does not exist for device %s from %s\n",
7506 		       dev->name, adj_dev->name);
7507 		WARN_ON(1);
7508 		return;
7509 	}
7510 
7511 	if (adj->ref_nr > ref_nr) {
7512 		pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7513 			 dev->name, adj_dev->name, ref_nr,
7514 			 adj->ref_nr - ref_nr);
7515 		adj->ref_nr -= ref_nr;
7516 		return;
7517 	}
7518 
7519 	if (adj->master)
7520 		sysfs_remove_link(&(dev->dev.kobj), "master");
7521 
7522 	if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7523 		netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7524 
7525 	list_del_rcu(&adj->list);
7526 	pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7527 		 adj_dev->name, dev->name, adj_dev->name);
7528 	netdev_put(adj_dev, &adj->dev_tracker);
7529 	kfree_rcu(adj, rcu);
7530 }
7531 
7532 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7533 					    struct net_device *upper_dev,
7534 					    struct list_head *up_list,
7535 					    struct list_head *down_list,
7536 					    void *private, bool master)
7537 {
7538 	int ret;
7539 
7540 	ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7541 					   private, master);
7542 	if (ret)
7543 		return ret;
7544 
7545 	ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7546 					   private, false);
7547 	if (ret) {
7548 		__netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7549 		return ret;
7550 	}
7551 
7552 	return 0;
7553 }
7554 
7555 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7556 					       struct net_device *upper_dev,
7557 					       u16 ref_nr,
7558 					       struct list_head *up_list,
7559 					       struct list_head *down_list)
7560 {
7561 	__netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7562 	__netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7563 }
7564 
7565 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7566 						struct net_device *upper_dev,
7567 						void *private, bool master)
7568 {
7569 	return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7570 						&dev->adj_list.upper,
7571 						&upper_dev->adj_list.lower,
7572 						private, master);
7573 }
7574 
7575 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7576 						   struct net_device *upper_dev)
7577 {
7578 	__netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7579 					   &dev->adj_list.upper,
7580 					   &upper_dev->adj_list.lower);
7581 }
7582 
7583 static int __netdev_upper_dev_link(struct net_device *dev,
7584 				   struct net_device *upper_dev, bool master,
7585 				   void *upper_priv, void *upper_info,
7586 				   struct netdev_nested_priv *priv,
7587 				   struct netlink_ext_ack *extack)
7588 {
7589 	struct netdev_notifier_changeupper_info changeupper_info = {
7590 		.info = {
7591 			.dev = dev,
7592 			.extack = extack,
7593 		},
7594 		.upper_dev = upper_dev,
7595 		.master = master,
7596 		.linking = true,
7597 		.upper_info = upper_info,
7598 	};
7599 	struct net_device *master_dev;
7600 	int ret = 0;
7601 
7602 	ASSERT_RTNL();
7603 
7604 	if (dev == upper_dev)
7605 		return -EBUSY;
7606 
7607 	/* To prevent loops, check if dev is not upper device to upper_dev. */
7608 	if (__netdev_has_upper_dev(upper_dev, dev))
7609 		return -EBUSY;
7610 
7611 	if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7612 		return -EMLINK;
7613 
7614 	if (!master) {
7615 		if (__netdev_has_upper_dev(dev, upper_dev))
7616 			return -EEXIST;
7617 	} else {
7618 		master_dev = __netdev_master_upper_dev_get(dev);
7619 		if (master_dev)
7620 			return master_dev == upper_dev ? -EEXIST : -EBUSY;
7621 	}
7622 
7623 	ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7624 					    &changeupper_info.info);
7625 	ret = notifier_to_errno(ret);
7626 	if (ret)
7627 		return ret;
7628 
7629 	ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7630 						   master);
7631 	if (ret)
7632 		return ret;
7633 
7634 	ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7635 					    &changeupper_info.info);
7636 	ret = notifier_to_errno(ret);
7637 	if (ret)
7638 		goto rollback;
7639 
7640 	__netdev_update_upper_level(dev, NULL);
7641 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7642 
7643 	__netdev_update_lower_level(upper_dev, priv);
7644 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7645 				    priv);
7646 
7647 	return 0;
7648 
7649 rollback:
7650 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7651 
7652 	return ret;
7653 }
7654 
7655 /**
7656  * netdev_upper_dev_link - Add a link to the upper device
7657  * @dev: device
7658  * @upper_dev: new upper device
7659  * @extack: netlink extended ack
7660  *
7661  * Adds a link to device which is upper to this one. The caller must hold
7662  * the RTNL lock. On a failure a negative errno code is returned.
7663  * On success the reference counts are adjusted and the function
7664  * returns zero.
7665  */
7666 int netdev_upper_dev_link(struct net_device *dev,
7667 			  struct net_device *upper_dev,
7668 			  struct netlink_ext_ack *extack)
7669 {
7670 	struct netdev_nested_priv priv = {
7671 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7672 		.data = NULL,
7673 	};
7674 
7675 	return __netdev_upper_dev_link(dev, upper_dev, false,
7676 				       NULL, NULL, &priv, extack);
7677 }
7678 EXPORT_SYMBOL(netdev_upper_dev_link);
7679 
7680 /**
7681  * netdev_master_upper_dev_link - Add a master link to the upper device
7682  * @dev: device
7683  * @upper_dev: new upper device
7684  * @upper_priv: upper device private
7685  * @upper_info: upper info to be passed down via notifier
7686  * @extack: netlink extended ack
7687  *
7688  * Adds a link to device which is upper to this one. In this case, only
7689  * one master upper device can be linked, although other non-master devices
7690  * might be linked as well. The caller must hold the RTNL lock.
7691  * On a failure a negative errno code is returned. On success the reference
7692  * counts are adjusted and the function returns zero.
7693  */
7694 int netdev_master_upper_dev_link(struct net_device *dev,
7695 				 struct net_device *upper_dev,
7696 				 void *upper_priv, void *upper_info,
7697 				 struct netlink_ext_ack *extack)
7698 {
7699 	struct netdev_nested_priv priv = {
7700 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7701 		.data = NULL,
7702 	};
7703 
7704 	return __netdev_upper_dev_link(dev, upper_dev, true,
7705 				       upper_priv, upper_info, &priv, extack);
7706 }
7707 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7708 
7709 static void __netdev_upper_dev_unlink(struct net_device *dev,
7710 				      struct net_device *upper_dev,
7711 				      struct netdev_nested_priv *priv)
7712 {
7713 	struct netdev_notifier_changeupper_info changeupper_info = {
7714 		.info = {
7715 			.dev = dev,
7716 		},
7717 		.upper_dev = upper_dev,
7718 		.linking = false,
7719 	};
7720 
7721 	ASSERT_RTNL();
7722 
7723 	changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7724 
7725 	call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7726 				      &changeupper_info.info);
7727 
7728 	__netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7729 
7730 	call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7731 				      &changeupper_info.info);
7732 
7733 	__netdev_update_upper_level(dev, NULL);
7734 	__netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7735 
7736 	__netdev_update_lower_level(upper_dev, priv);
7737 	__netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7738 				    priv);
7739 }
7740 
7741 /**
7742  * netdev_upper_dev_unlink - Removes a link to upper device
7743  * @dev: device
7744  * @upper_dev: new upper device
7745  *
7746  * Removes a link to device which is upper to this one. The caller must hold
7747  * the RTNL lock.
7748  */
7749 void netdev_upper_dev_unlink(struct net_device *dev,
7750 			     struct net_device *upper_dev)
7751 {
7752 	struct netdev_nested_priv priv = {
7753 		.flags = NESTED_SYNC_TODO,
7754 		.data = NULL,
7755 	};
7756 
7757 	__netdev_upper_dev_unlink(dev, upper_dev, &priv);
7758 }
7759 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7760 
7761 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7762 				      struct net_device *lower_dev,
7763 				      bool val)
7764 {
7765 	struct netdev_adjacent *adj;
7766 
7767 	adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7768 	if (adj)
7769 		adj->ignore = val;
7770 
7771 	adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7772 	if (adj)
7773 		adj->ignore = val;
7774 }
7775 
7776 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7777 					struct net_device *lower_dev)
7778 {
7779 	__netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7780 }
7781 
7782 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7783 				       struct net_device *lower_dev)
7784 {
7785 	__netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7786 }
7787 
7788 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7789 				   struct net_device *new_dev,
7790 				   struct net_device *dev,
7791 				   struct netlink_ext_ack *extack)
7792 {
7793 	struct netdev_nested_priv priv = {
7794 		.flags = 0,
7795 		.data = NULL,
7796 	};
7797 	int err;
7798 
7799 	if (!new_dev)
7800 		return 0;
7801 
7802 	if (old_dev && new_dev != old_dev)
7803 		netdev_adjacent_dev_disable(dev, old_dev);
7804 	err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
7805 				      extack);
7806 	if (err) {
7807 		if (old_dev && new_dev != old_dev)
7808 			netdev_adjacent_dev_enable(dev, old_dev);
7809 		return err;
7810 	}
7811 
7812 	return 0;
7813 }
7814 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7815 
7816 void netdev_adjacent_change_commit(struct net_device *old_dev,
7817 				   struct net_device *new_dev,
7818 				   struct net_device *dev)
7819 {
7820 	struct netdev_nested_priv priv = {
7821 		.flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7822 		.data = NULL,
7823 	};
7824 
7825 	if (!new_dev || !old_dev)
7826 		return;
7827 
7828 	if (new_dev == old_dev)
7829 		return;
7830 
7831 	netdev_adjacent_dev_enable(dev, old_dev);
7832 	__netdev_upper_dev_unlink(old_dev, dev, &priv);
7833 }
7834 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7835 
7836 void netdev_adjacent_change_abort(struct net_device *old_dev,
7837 				  struct net_device *new_dev,
7838 				  struct net_device *dev)
7839 {
7840 	struct netdev_nested_priv priv = {
7841 		.flags = 0,
7842 		.data = NULL,
7843 	};
7844 
7845 	if (!new_dev)
7846 		return;
7847 
7848 	if (old_dev && new_dev != old_dev)
7849 		netdev_adjacent_dev_enable(dev, old_dev);
7850 
7851 	__netdev_upper_dev_unlink(new_dev, dev, &priv);
7852 }
7853 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7854 
7855 /**
7856  * netdev_bonding_info_change - Dispatch event about slave change
7857  * @dev: device
7858  * @bonding_info: info to dispatch
7859  *
7860  * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7861  * The caller must hold the RTNL lock.
7862  */
7863 void netdev_bonding_info_change(struct net_device *dev,
7864 				struct netdev_bonding_info *bonding_info)
7865 {
7866 	struct netdev_notifier_bonding_info info = {
7867 		.info.dev = dev,
7868 	};
7869 
7870 	memcpy(&info.bonding_info, bonding_info,
7871 	       sizeof(struct netdev_bonding_info));
7872 	call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7873 				      &info.info);
7874 }
7875 EXPORT_SYMBOL(netdev_bonding_info_change);
7876 
7877 static int netdev_offload_xstats_enable_l3(struct net_device *dev,
7878 					   struct netlink_ext_ack *extack)
7879 {
7880 	struct netdev_notifier_offload_xstats_info info = {
7881 		.info.dev = dev,
7882 		.info.extack = extack,
7883 		.type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7884 	};
7885 	int err;
7886 	int rc;
7887 
7888 	dev->offload_xstats_l3 = kzalloc(sizeof(*dev->offload_xstats_l3),
7889 					 GFP_KERNEL);
7890 	if (!dev->offload_xstats_l3)
7891 		return -ENOMEM;
7892 
7893 	rc = call_netdevice_notifiers_info_robust(NETDEV_OFFLOAD_XSTATS_ENABLE,
7894 						  NETDEV_OFFLOAD_XSTATS_DISABLE,
7895 						  &info.info);
7896 	err = notifier_to_errno(rc);
7897 	if (err)
7898 		goto free_stats;
7899 
7900 	return 0;
7901 
7902 free_stats:
7903 	kfree(dev->offload_xstats_l3);
7904 	dev->offload_xstats_l3 = NULL;
7905 	return err;
7906 }
7907 
7908 int netdev_offload_xstats_enable(struct net_device *dev,
7909 				 enum netdev_offload_xstats_type type,
7910 				 struct netlink_ext_ack *extack)
7911 {
7912 	ASSERT_RTNL();
7913 
7914 	if (netdev_offload_xstats_enabled(dev, type))
7915 		return -EALREADY;
7916 
7917 	switch (type) {
7918 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7919 		return netdev_offload_xstats_enable_l3(dev, extack);
7920 	}
7921 
7922 	WARN_ON(1);
7923 	return -EINVAL;
7924 }
7925 EXPORT_SYMBOL(netdev_offload_xstats_enable);
7926 
7927 static void netdev_offload_xstats_disable_l3(struct net_device *dev)
7928 {
7929 	struct netdev_notifier_offload_xstats_info info = {
7930 		.info.dev = dev,
7931 		.type = NETDEV_OFFLOAD_XSTATS_TYPE_L3,
7932 	};
7933 
7934 	call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_DISABLE,
7935 				      &info.info);
7936 	kfree(dev->offload_xstats_l3);
7937 	dev->offload_xstats_l3 = NULL;
7938 }
7939 
7940 int netdev_offload_xstats_disable(struct net_device *dev,
7941 				  enum netdev_offload_xstats_type type)
7942 {
7943 	ASSERT_RTNL();
7944 
7945 	if (!netdev_offload_xstats_enabled(dev, type))
7946 		return -EALREADY;
7947 
7948 	switch (type) {
7949 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7950 		netdev_offload_xstats_disable_l3(dev);
7951 		return 0;
7952 	}
7953 
7954 	WARN_ON(1);
7955 	return -EINVAL;
7956 }
7957 EXPORT_SYMBOL(netdev_offload_xstats_disable);
7958 
7959 static void netdev_offload_xstats_disable_all(struct net_device *dev)
7960 {
7961 	netdev_offload_xstats_disable(dev, NETDEV_OFFLOAD_XSTATS_TYPE_L3);
7962 }
7963 
7964 static struct rtnl_hw_stats64 *
7965 netdev_offload_xstats_get_ptr(const struct net_device *dev,
7966 			      enum netdev_offload_xstats_type type)
7967 {
7968 	switch (type) {
7969 	case NETDEV_OFFLOAD_XSTATS_TYPE_L3:
7970 		return dev->offload_xstats_l3;
7971 	}
7972 
7973 	WARN_ON(1);
7974 	return NULL;
7975 }
7976 
7977 bool netdev_offload_xstats_enabled(const struct net_device *dev,
7978 				   enum netdev_offload_xstats_type type)
7979 {
7980 	ASSERT_RTNL();
7981 
7982 	return netdev_offload_xstats_get_ptr(dev, type);
7983 }
7984 EXPORT_SYMBOL(netdev_offload_xstats_enabled);
7985 
7986 struct netdev_notifier_offload_xstats_ru {
7987 	bool used;
7988 };
7989 
7990 struct netdev_notifier_offload_xstats_rd {
7991 	struct rtnl_hw_stats64 stats;
7992 	bool used;
7993 };
7994 
7995 static void netdev_hw_stats64_add(struct rtnl_hw_stats64 *dest,
7996 				  const struct rtnl_hw_stats64 *src)
7997 {
7998 	dest->rx_packets	  += src->rx_packets;
7999 	dest->tx_packets	  += src->tx_packets;
8000 	dest->rx_bytes		  += src->rx_bytes;
8001 	dest->tx_bytes		  += src->tx_bytes;
8002 	dest->rx_errors		  += src->rx_errors;
8003 	dest->tx_errors		  += src->tx_errors;
8004 	dest->rx_dropped	  += src->rx_dropped;
8005 	dest->tx_dropped	  += src->tx_dropped;
8006 	dest->multicast		  += src->multicast;
8007 }
8008 
8009 static int netdev_offload_xstats_get_used(struct net_device *dev,
8010 					  enum netdev_offload_xstats_type type,
8011 					  bool *p_used,
8012 					  struct netlink_ext_ack *extack)
8013 {
8014 	struct netdev_notifier_offload_xstats_ru report_used = {};
8015 	struct netdev_notifier_offload_xstats_info info = {
8016 		.info.dev = dev,
8017 		.info.extack = extack,
8018 		.type = type,
8019 		.report_used = &report_used,
8020 	};
8021 	int rc;
8022 
8023 	WARN_ON(!netdev_offload_xstats_enabled(dev, type));
8024 	rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_USED,
8025 					   &info.info);
8026 	*p_used = report_used.used;
8027 	return notifier_to_errno(rc);
8028 }
8029 
8030 static int netdev_offload_xstats_get_stats(struct net_device *dev,
8031 					   enum netdev_offload_xstats_type type,
8032 					   struct rtnl_hw_stats64 *p_stats,
8033 					   bool *p_used,
8034 					   struct netlink_ext_ack *extack)
8035 {
8036 	struct netdev_notifier_offload_xstats_rd report_delta = {};
8037 	struct netdev_notifier_offload_xstats_info info = {
8038 		.info.dev = dev,
8039 		.info.extack = extack,
8040 		.type = type,
8041 		.report_delta = &report_delta,
8042 	};
8043 	struct rtnl_hw_stats64 *stats;
8044 	int rc;
8045 
8046 	stats = netdev_offload_xstats_get_ptr(dev, type);
8047 	if (WARN_ON(!stats))
8048 		return -EINVAL;
8049 
8050 	rc = call_netdevice_notifiers_info(NETDEV_OFFLOAD_XSTATS_REPORT_DELTA,
8051 					   &info.info);
8052 
8053 	/* Cache whatever we got, even if there was an error, otherwise the
8054 	 * successful stats retrievals would get lost.
8055 	 */
8056 	netdev_hw_stats64_add(stats, &report_delta.stats);
8057 
8058 	if (p_stats)
8059 		*p_stats = *stats;
8060 	*p_used = report_delta.used;
8061 
8062 	return notifier_to_errno(rc);
8063 }
8064 
8065 int netdev_offload_xstats_get(struct net_device *dev,
8066 			      enum netdev_offload_xstats_type type,
8067 			      struct rtnl_hw_stats64 *p_stats, bool *p_used,
8068 			      struct netlink_ext_ack *extack)
8069 {
8070 	ASSERT_RTNL();
8071 
8072 	if (p_stats)
8073 		return netdev_offload_xstats_get_stats(dev, type, p_stats,
8074 						       p_used, extack);
8075 	else
8076 		return netdev_offload_xstats_get_used(dev, type, p_used,
8077 						      extack);
8078 }
8079 EXPORT_SYMBOL(netdev_offload_xstats_get);
8080 
8081 void
8082 netdev_offload_xstats_report_delta(struct netdev_notifier_offload_xstats_rd *report_delta,
8083 				   const struct rtnl_hw_stats64 *stats)
8084 {
8085 	report_delta->used = true;
8086 	netdev_hw_stats64_add(&report_delta->stats, stats);
8087 }
8088 EXPORT_SYMBOL(netdev_offload_xstats_report_delta);
8089 
8090 void
8091 netdev_offload_xstats_report_used(struct netdev_notifier_offload_xstats_ru *report_used)
8092 {
8093 	report_used->used = true;
8094 }
8095 EXPORT_SYMBOL(netdev_offload_xstats_report_used);
8096 
8097 void netdev_offload_xstats_push_delta(struct net_device *dev,
8098 				      enum netdev_offload_xstats_type type,
8099 				      const struct rtnl_hw_stats64 *p_stats)
8100 {
8101 	struct rtnl_hw_stats64 *stats;
8102 
8103 	ASSERT_RTNL();
8104 
8105 	stats = netdev_offload_xstats_get_ptr(dev, type);
8106 	if (WARN_ON(!stats))
8107 		return;
8108 
8109 	netdev_hw_stats64_add(stats, p_stats);
8110 }
8111 EXPORT_SYMBOL(netdev_offload_xstats_push_delta);
8112 
8113 /**
8114  * netdev_get_xmit_slave - Get the xmit slave of master device
8115  * @dev: device
8116  * @skb: The packet
8117  * @all_slaves: assume all the slaves are active
8118  *
8119  * The reference counters are not incremented so the caller must be
8120  * careful with locks. The caller must hold RCU lock.
8121  * %NULL is returned if no slave is found.
8122  */
8123 
8124 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8125 					 struct sk_buff *skb,
8126 					 bool all_slaves)
8127 {
8128 	const struct net_device_ops *ops = dev->netdev_ops;
8129 
8130 	if (!ops->ndo_get_xmit_slave)
8131 		return NULL;
8132 	return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8133 }
8134 EXPORT_SYMBOL(netdev_get_xmit_slave);
8135 
8136 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8137 						  struct sock *sk)
8138 {
8139 	const struct net_device_ops *ops = dev->netdev_ops;
8140 
8141 	if (!ops->ndo_sk_get_lower_dev)
8142 		return NULL;
8143 	return ops->ndo_sk_get_lower_dev(dev, sk);
8144 }
8145 
8146 /**
8147  * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8148  * @dev: device
8149  * @sk: the socket
8150  *
8151  * %NULL is returned if no lower device is found.
8152  */
8153 
8154 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8155 					    struct sock *sk)
8156 {
8157 	struct net_device *lower;
8158 
8159 	lower = netdev_sk_get_lower_dev(dev, sk);
8160 	while (lower) {
8161 		dev = lower;
8162 		lower = netdev_sk_get_lower_dev(dev, sk);
8163 	}
8164 
8165 	return dev;
8166 }
8167 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8168 
8169 static void netdev_adjacent_add_links(struct net_device *dev)
8170 {
8171 	struct netdev_adjacent *iter;
8172 
8173 	struct net *net = dev_net(dev);
8174 
8175 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8176 		if (!net_eq(net, dev_net(iter->dev)))
8177 			continue;
8178 		netdev_adjacent_sysfs_add(iter->dev, dev,
8179 					  &iter->dev->adj_list.lower);
8180 		netdev_adjacent_sysfs_add(dev, iter->dev,
8181 					  &dev->adj_list.upper);
8182 	}
8183 
8184 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8185 		if (!net_eq(net, dev_net(iter->dev)))
8186 			continue;
8187 		netdev_adjacent_sysfs_add(iter->dev, dev,
8188 					  &iter->dev->adj_list.upper);
8189 		netdev_adjacent_sysfs_add(dev, iter->dev,
8190 					  &dev->adj_list.lower);
8191 	}
8192 }
8193 
8194 static void netdev_adjacent_del_links(struct net_device *dev)
8195 {
8196 	struct netdev_adjacent *iter;
8197 
8198 	struct net *net = dev_net(dev);
8199 
8200 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8201 		if (!net_eq(net, dev_net(iter->dev)))
8202 			continue;
8203 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8204 					  &iter->dev->adj_list.lower);
8205 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8206 					  &dev->adj_list.upper);
8207 	}
8208 
8209 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8210 		if (!net_eq(net, dev_net(iter->dev)))
8211 			continue;
8212 		netdev_adjacent_sysfs_del(iter->dev, dev->name,
8213 					  &iter->dev->adj_list.upper);
8214 		netdev_adjacent_sysfs_del(dev, iter->dev->name,
8215 					  &dev->adj_list.lower);
8216 	}
8217 }
8218 
8219 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8220 {
8221 	struct netdev_adjacent *iter;
8222 
8223 	struct net *net = dev_net(dev);
8224 
8225 	list_for_each_entry(iter, &dev->adj_list.upper, list) {
8226 		if (!net_eq(net, dev_net(iter->dev)))
8227 			continue;
8228 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8229 					  &iter->dev->adj_list.lower);
8230 		netdev_adjacent_sysfs_add(iter->dev, dev,
8231 					  &iter->dev->adj_list.lower);
8232 	}
8233 
8234 	list_for_each_entry(iter, &dev->adj_list.lower, list) {
8235 		if (!net_eq(net, dev_net(iter->dev)))
8236 			continue;
8237 		netdev_adjacent_sysfs_del(iter->dev, oldname,
8238 					  &iter->dev->adj_list.upper);
8239 		netdev_adjacent_sysfs_add(iter->dev, dev,
8240 					  &iter->dev->adj_list.upper);
8241 	}
8242 }
8243 
8244 void *netdev_lower_dev_get_private(struct net_device *dev,
8245 				   struct net_device *lower_dev)
8246 {
8247 	struct netdev_adjacent *lower;
8248 
8249 	if (!lower_dev)
8250 		return NULL;
8251 	lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8252 	if (!lower)
8253 		return NULL;
8254 
8255 	return lower->private;
8256 }
8257 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8258 
8259 
8260 /**
8261  * netdev_lower_state_changed - Dispatch event about lower device state change
8262  * @lower_dev: device
8263  * @lower_state_info: state to dispatch
8264  *
8265  * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8266  * The caller must hold the RTNL lock.
8267  */
8268 void netdev_lower_state_changed(struct net_device *lower_dev,
8269 				void *lower_state_info)
8270 {
8271 	struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8272 		.info.dev = lower_dev,
8273 	};
8274 
8275 	ASSERT_RTNL();
8276 	changelowerstate_info.lower_state_info = lower_state_info;
8277 	call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8278 				      &changelowerstate_info.info);
8279 }
8280 EXPORT_SYMBOL(netdev_lower_state_changed);
8281 
8282 static void dev_change_rx_flags(struct net_device *dev, int flags)
8283 {
8284 	const struct net_device_ops *ops = dev->netdev_ops;
8285 
8286 	if (ops->ndo_change_rx_flags)
8287 		ops->ndo_change_rx_flags(dev, flags);
8288 }
8289 
8290 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8291 {
8292 	unsigned int old_flags = dev->flags;
8293 	kuid_t uid;
8294 	kgid_t gid;
8295 
8296 	ASSERT_RTNL();
8297 
8298 	dev->flags |= IFF_PROMISC;
8299 	dev->promiscuity += inc;
8300 	if (dev->promiscuity == 0) {
8301 		/*
8302 		 * Avoid overflow.
8303 		 * If inc causes overflow, untouch promisc and return error.
8304 		 */
8305 		if (inc < 0)
8306 			dev->flags &= ~IFF_PROMISC;
8307 		else {
8308 			dev->promiscuity -= inc;
8309 			netdev_warn(dev, "promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n");
8310 			return -EOVERFLOW;
8311 		}
8312 	}
8313 	if (dev->flags != old_flags) {
8314 		pr_info("device %s %s promiscuous mode\n",
8315 			dev->name,
8316 			dev->flags & IFF_PROMISC ? "entered" : "left");
8317 		if (audit_enabled) {
8318 			current_uid_gid(&uid, &gid);
8319 			audit_log(audit_context(), GFP_ATOMIC,
8320 				  AUDIT_ANOM_PROMISCUOUS,
8321 				  "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8322 				  dev->name, (dev->flags & IFF_PROMISC),
8323 				  (old_flags & IFF_PROMISC),
8324 				  from_kuid(&init_user_ns, audit_get_loginuid(current)),
8325 				  from_kuid(&init_user_ns, uid),
8326 				  from_kgid(&init_user_ns, gid),
8327 				  audit_get_sessionid(current));
8328 		}
8329 
8330 		dev_change_rx_flags(dev, IFF_PROMISC);
8331 	}
8332 	if (notify)
8333 		__dev_notify_flags(dev, old_flags, IFF_PROMISC, 0, NULL);
8334 	return 0;
8335 }
8336 
8337 /**
8338  *	dev_set_promiscuity	- update promiscuity count on a device
8339  *	@dev: device
8340  *	@inc: modifier
8341  *
8342  *	Add or remove promiscuity from a device. While the count in the device
8343  *	remains above zero the interface remains promiscuous. Once it hits zero
8344  *	the device reverts back to normal filtering operation. A negative inc
8345  *	value is used to drop promiscuity on the device.
8346  *	Return 0 if successful or a negative errno code on error.
8347  */
8348 int dev_set_promiscuity(struct net_device *dev, int inc)
8349 {
8350 	unsigned int old_flags = dev->flags;
8351 	int err;
8352 
8353 	err = __dev_set_promiscuity(dev, inc, true);
8354 	if (err < 0)
8355 		return err;
8356 	if (dev->flags != old_flags)
8357 		dev_set_rx_mode(dev);
8358 	return err;
8359 }
8360 EXPORT_SYMBOL(dev_set_promiscuity);
8361 
8362 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8363 {
8364 	unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8365 
8366 	ASSERT_RTNL();
8367 
8368 	dev->flags |= IFF_ALLMULTI;
8369 	dev->allmulti += inc;
8370 	if (dev->allmulti == 0) {
8371 		/*
8372 		 * Avoid overflow.
8373 		 * If inc causes overflow, untouch allmulti and return error.
8374 		 */
8375 		if (inc < 0)
8376 			dev->flags &= ~IFF_ALLMULTI;
8377 		else {
8378 			dev->allmulti -= inc;
8379 			netdev_warn(dev, "allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n");
8380 			return -EOVERFLOW;
8381 		}
8382 	}
8383 	if (dev->flags ^ old_flags) {
8384 		dev_change_rx_flags(dev, IFF_ALLMULTI);
8385 		dev_set_rx_mode(dev);
8386 		if (notify)
8387 			__dev_notify_flags(dev, old_flags,
8388 					   dev->gflags ^ old_gflags, 0, NULL);
8389 	}
8390 	return 0;
8391 }
8392 
8393 /**
8394  *	dev_set_allmulti	- update allmulti count on a device
8395  *	@dev: device
8396  *	@inc: modifier
8397  *
8398  *	Add or remove reception of all multicast frames to a device. While the
8399  *	count in the device remains above zero the interface remains listening
8400  *	to all interfaces. Once it hits zero the device reverts back to normal
8401  *	filtering operation. A negative @inc value is used to drop the counter
8402  *	when releasing a resource needing all multicasts.
8403  *	Return 0 if successful or a negative errno code on error.
8404  */
8405 
8406 int dev_set_allmulti(struct net_device *dev, int inc)
8407 {
8408 	return __dev_set_allmulti(dev, inc, true);
8409 }
8410 EXPORT_SYMBOL(dev_set_allmulti);
8411 
8412 /*
8413  *	Upload unicast and multicast address lists to device and
8414  *	configure RX filtering. When the device doesn't support unicast
8415  *	filtering it is put in promiscuous mode while unicast addresses
8416  *	are present.
8417  */
8418 void __dev_set_rx_mode(struct net_device *dev)
8419 {
8420 	const struct net_device_ops *ops = dev->netdev_ops;
8421 
8422 	/* dev_open will call this function so the list will stay sane. */
8423 	if (!(dev->flags&IFF_UP))
8424 		return;
8425 
8426 	if (!netif_device_present(dev))
8427 		return;
8428 
8429 	if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8430 		/* Unicast addresses changes may only happen under the rtnl,
8431 		 * therefore calling __dev_set_promiscuity here is safe.
8432 		 */
8433 		if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8434 			__dev_set_promiscuity(dev, 1, false);
8435 			dev->uc_promisc = true;
8436 		} else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8437 			__dev_set_promiscuity(dev, -1, false);
8438 			dev->uc_promisc = false;
8439 		}
8440 	}
8441 
8442 	if (ops->ndo_set_rx_mode)
8443 		ops->ndo_set_rx_mode(dev);
8444 }
8445 
8446 void dev_set_rx_mode(struct net_device *dev)
8447 {
8448 	netif_addr_lock_bh(dev);
8449 	__dev_set_rx_mode(dev);
8450 	netif_addr_unlock_bh(dev);
8451 }
8452 
8453 /**
8454  *	dev_get_flags - get flags reported to userspace
8455  *	@dev: device
8456  *
8457  *	Get the combination of flag bits exported through APIs to userspace.
8458  */
8459 unsigned int dev_get_flags(const struct net_device *dev)
8460 {
8461 	unsigned int flags;
8462 
8463 	flags = (dev->flags & ~(IFF_PROMISC |
8464 				IFF_ALLMULTI |
8465 				IFF_RUNNING |
8466 				IFF_LOWER_UP |
8467 				IFF_DORMANT)) |
8468 		(dev->gflags & (IFF_PROMISC |
8469 				IFF_ALLMULTI));
8470 
8471 	if (netif_running(dev)) {
8472 		if (netif_oper_up(dev))
8473 			flags |= IFF_RUNNING;
8474 		if (netif_carrier_ok(dev))
8475 			flags |= IFF_LOWER_UP;
8476 		if (netif_dormant(dev))
8477 			flags |= IFF_DORMANT;
8478 	}
8479 
8480 	return flags;
8481 }
8482 EXPORT_SYMBOL(dev_get_flags);
8483 
8484 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8485 		       struct netlink_ext_ack *extack)
8486 {
8487 	unsigned int old_flags = dev->flags;
8488 	int ret;
8489 
8490 	ASSERT_RTNL();
8491 
8492 	/*
8493 	 *	Set the flags on our device.
8494 	 */
8495 
8496 	dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8497 			       IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8498 			       IFF_AUTOMEDIA)) |
8499 		     (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8500 				    IFF_ALLMULTI));
8501 
8502 	/*
8503 	 *	Load in the correct multicast list now the flags have changed.
8504 	 */
8505 
8506 	if ((old_flags ^ flags) & IFF_MULTICAST)
8507 		dev_change_rx_flags(dev, IFF_MULTICAST);
8508 
8509 	dev_set_rx_mode(dev);
8510 
8511 	/*
8512 	 *	Have we downed the interface. We handle IFF_UP ourselves
8513 	 *	according to user attempts to set it, rather than blindly
8514 	 *	setting it.
8515 	 */
8516 
8517 	ret = 0;
8518 	if ((old_flags ^ flags) & IFF_UP) {
8519 		if (old_flags & IFF_UP)
8520 			__dev_close(dev);
8521 		else
8522 			ret = __dev_open(dev, extack);
8523 	}
8524 
8525 	if ((flags ^ dev->gflags) & IFF_PROMISC) {
8526 		int inc = (flags & IFF_PROMISC) ? 1 : -1;
8527 		unsigned int old_flags = dev->flags;
8528 
8529 		dev->gflags ^= IFF_PROMISC;
8530 
8531 		if (__dev_set_promiscuity(dev, inc, false) >= 0)
8532 			if (dev->flags != old_flags)
8533 				dev_set_rx_mode(dev);
8534 	}
8535 
8536 	/* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8537 	 * is important. Some (broken) drivers set IFF_PROMISC, when
8538 	 * IFF_ALLMULTI is requested not asking us and not reporting.
8539 	 */
8540 	if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8541 		int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8542 
8543 		dev->gflags ^= IFF_ALLMULTI;
8544 		__dev_set_allmulti(dev, inc, false);
8545 	}
8546 
8547 	return ret;
8548 }
8549 
8550 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8551 			unsigned int gchanges, u32 portid,
8552 			const struct nlmsghdr *nlh)
8553 {
8554 	unsigned int changes = dev->flags ^ old_flags;
8555 
8556 	if (gchanges)
8557 		rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC, portid, nlh);
8558 
8559 	if (changes & IFF_UP) {
8560 		if (dev->flags & IFF_UP)
8561 			call_netdevice_notifiers(NETDEV_UP, dev);
8562 		else
8563 			call_netdevice_notifiers(NETDEV_DOWN, dev);
8564 	}
8565 
8566 	if (dev->flags & IFF_UP &&
8567 	    (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8568 		struct netdev_notifier_change_info change_info = {
8569 			.info = {
8570 				.dev = dev,
8571 			},
8572 			.flags_changed = changes,
8573 		};
8574 
8575 		call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8576 	}
8577 }
8578 
8579 /**
8580  *	dev_change_flags - change device settings
8581  *	@dev: device
8582  *	@flags: device state flags
8583  *	@extack: netlink extended ack
8584  *
8585  *	Change settings on device based state flags. The flags are
8586  *	in the userspace exported format.
8587  */
8588 int dev_change_flags(struct net_device *dev, unsigned int flags,
8589 		     struct netlink_ext_ack *extack)
8590 {
8591 	int ret;
8592 	unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8593 
8594 	ret = __dev_change_flags(dev, flags, extack);
8595 	if (ret < 0)
8596 		return ret;
8597 
8598 	changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8599 	__dev_notify_flags(dev, old_flags, changes, 0, NULL);
8600 	return ret;
8601 }
8602 EXPORT_SYMBOL(dev_change_flags);
8603 
8604 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8605 {
8606 	const struct net_device_ops *ops = dev->netdev_ops;
8607 
8608 	if (ops->ndo_change_mtu)
8609 		return ops->ndo_change_mtu(dev, new_mtu);
8610 
8611 	/* Pairs with all the lockless reads of dev->mtu in the stack */
8612 	WRITE_ONCE(dev->mtu, new_mtu);
8613 	return 0;
8614 }
8615 EXPORT_SYMBOL(__dev_set_mtu);
8616 
8617 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8618 		     struct netlink_ext_ack *extack)
8619 {
8620 	/* MTU must be positive, and in range */
8621 	if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8622 		NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8623 		return -EINVAL;
8624 	}
8625 
8626 	if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8627 		NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8628 		return -EINVAL;
8629 	}
8630 	return 0;
8631 }
8632 
8633 /**
8634  *	dev_set_mtu_ext - Change maximum transfer unit
8635  *	@dev: device
8636  *	@new_mtu: new transfer unit
8637  *	@extack: netlink extended ack
8638  *
8639  *	Change the maximum transfer size of the network device.
8640  */
8641 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8642 		    struct netlink_ext_ack *extack)
8643 {
8644 	int err, orig_mtu;
8645 
8646 	if (new_mtu == dev->mtu)
8647 		return 0;
8648 
8649 	err = dev_validate_mtu(dev, new_mtu, extack);
8650 	if (err)
8651 		return err;
8652 
8653 	if (!netif_device_present(dev))
8654 		return -ENODEV;
8655 
8656 	err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8657 	err = notifier_to_errno(err);
8658 	if (err)
8659 		return err;
8660 
8661 	orig_mtu = dev->mtu;
8662 	err = __dev_set_mtu(dev, new_mtu);
8663 
8664 	if (!err) {
8665 		err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8666 						   orig_mtu);
8667 		err = notifier_to_errno(err);
8668 		if (err) {
8669 			/* setting mtu back and notifying everyone again,
8670 			 * so that they have a chance to revert changes.
8671 			 */
8672 			__dev_set_mtu(dev, orig_mtu);
8673 			call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8674 						     new_mtu);
8675 		}
8676 	}
8677 	return err;
8678 }
8679 
8680 int dev_set_mtu(struct net_device *dev, int new_mtu)
8681 {
8682 	struct netlink_ext_ack extack;
8683 	int err;
8684 
8685 	memset(&extack, 0, sizeof(extack));
8686 	err = dev_set_mtu_ext(dev, new_mtu, &extack);
8687 	if (err && extack._msg)
8688 		net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8689 	return err;
8690 }
8691 EXPORT_SYMBOL(dev_set_mtu);
8692 
8693 /**
8694  *	dev_change_tx_queue_len - Change TX queue length of a netdevice
8695  *	@dev: device
8696  *	@new_len: new tx queue length
8697  */
8698 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8699 {
8700 	unsigned int orig_len = dev->tx_queue_len;
8701 	int res;
8702 
8703 	if (new_len != (unsigned int)new_len)
8704 		return -ERANGE;
8705 
8706 	if (new_len != orig_len) {
8707 		dev->tx_queue_len = new_len;
8708 		res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8709 		res = notifier_to_errno(res);
8710 		if (res)
8711 			goto err_rollback;
8712 		res = dev_qdisc_change_tx_queue_len(dev);
8713 		if (res)
8714 			goto err_rollback;
8715 	}
8716 
8717 	return 0;
8718 
8719 err_rollback:
8720 	netdev_err(dev, "refused to change device tx_queue_len\n");
8721 	dev->tx_queue_len = orig_len;
8722 	return res;
8723 }
8724 
8725 /**
8726  *	dev_set_group - Change group this device belongs to
8727  *	@dev: device
8728  *	@new_group: group this device should belong to
8729  */
8730 void dev_set_group(struct net_device *dev, int new_group)
8731 {
8732 	dev->group = new_group;
8733 }
8734 
8735 /**
8736  *	dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8737  *	@dev: device
8738  *	@addr: new address
8739  *	@extack: netlink extended ack
8740  */
8741 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8742 			      struct netlink_ext_ack *extack)
8743 {
8744 	struct netdev_notifier_pre_changeaddr_info info = {
8745 		.info.dev = dev,
8746 		.info.extack = extack,
8747 		.dev_addr = addr,
8748 	};
8749 	int rc;
8750 
8751 	rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8752 	return notifier_to_errno(rc);
8753 }
8754 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8755 
8756 /**
8757  *	dev_set_mac_address - Change Media Access Control Address
8758  *	@dev: device
8759  *	@sa: new address
8760  *	@extack: netlink extended ack
8761  *
8762  *	Change the hardware (MAC) address of the device
8763  */
8764 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8765 			struct netlink_ext_ack *extack)
8766 {
8767 	const struct net_device_ops *ops = dev->netdev_ops;
8768 	int err;
8769 
8770 	if (!ops->ndo_set_mac_address)
8771 		return -EOPNOTSUPP;
8772 	if (sa->sa_family != dev->type)
8773 		return -EINVAL;
8774 	if (!netif_device_present(dev))
8775 		return -ENODEV;
8776 	err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8777 	if (err)
8778 		return err;
8779 	err = ops->ndo_set_mac_address(dev, sa);
8780 	if (err)
8781 		return err;
8782 	dev->addr_assign_type = NET_ADDR_SET;
8783 	call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8784 	add_device_randomness(dev->dev_addr, dev->addr_len);
8785 	return 0;
8786 }
8787 EXPORT_SYMBOL(dev_set_mac_address);
8788 
8789 static DECLARE_RWSEM(dev_addr_sem);
8790 
8791 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
8792 			     struct netlink_ext_ack *extack)
8793 {
8794 	int ret;
8795 
8796 	down_write(&dev_addr_sem);
8797 	ret = dev_set_mac_address(dev, sa, extack);
8798 	up_write(&dev_addr_sem);
8799 	return ret;
8800 }
8801 EXPORT_SYMBOL(dev_set_mac_address_user);
8802 
8803 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
8804 {
8805 	size_t size = sizeof(sa->sa_data_min);
8806 	struct net_device *dev;
8807 	int ret = 0;
8808 
8809 	down_read(&dev_addr_sem);
8810 	rcu_read_lock();
8811 
8812 	dev = dev_get_by_name_rcu(net, dev_name);
8813 	if (!dev) {
8814 		ret = -ENODEV;
8815 		goto unlock;
8816 	}
8817 	if (!dev->addr_len)
8818 		memset(sa->sa_data, 0, size);
8819 	else
8820 		memcpy(sa->sa_data, dev->dev_addr,
8821 		       min_t(size_t, size, dev->addr_len));
8822 	sa->sa_family = dev->type;
8823 
8824 unlock:
8825 	rcu_read_unlock();
8826 	up_read(&dev_addr_sem);
8827 	return ret;
8828 }
8829 EXPORT_SYMBOL(dev_get_mac_address);
8830 
8831 /**
8832  *	dev_change_carrier - Change device carrier
8833  *	@dev: device
8834  *	@new_carrier: new value
8835  *
8836  *	Change device carrier
8837  */
8838 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8839 {
8840 	const struct net_device_ops *ops = dev->netdev_ops;
8841 
8842 	if (!ops->ndo_change_carrier)
8843 		return -EOPNOTSUPP;
8844 	if (!netif_device_present(dev))
8845 		return -ENODEV;
8846 	return ops->ndo_change_carrier(dev, new_carrier);
8847 }
8848 
8849 /**
8850  *	dev_get_phys_port_id - Get device physical port ID
8851  *	@dev: device
8852  *	@ppid: port ID
8853  *
8854  *	Get device physical port ID
8855  */
8856 int dev_get_phys_port_id(struct net_device *dev,
8857 			 struct netdev_phys_item_id *ppid)
8858 {
8859 	const struct net_device_ops *ops = dev->netdev_ops;
8860 
8861 	if (!ops->ndo_get_phys_port_id)
8862 		return -EOPNOTSUPP;
8863 	return ops->ndo_get_phys_port_id(dev, ppid);
8864 }
8865 
8866 /**
8867  *	dev_get_phys_port_name - Get device physical port name
8868  *	@dev: device
8869  *	@name: port name
8870  *	@len: limit of bytes to copy to name
8871  *
8872  *	Get device physical port name
8873  */
8874 int dev_get_phys_port_name(struct net_device *dev,
8875 			   char *name, size_t len)
8876 {
8877 	const struct net_device_ops *ops = dev->netdev_ops;
8878 	int err;
8879 
8880 	if (ops->ndo_get_phys_port_name) {
8881 		err = ops->ndo_get_phys_port_name(dev, name, len);
8882 		if (err != -EOPNOTSUPP)
8883 			return err;
8884 	}
8885 	return devlink_compat_phys_port_name_get(dev, name, len);
8886 }
8887 
8888 /**
8889  *	dev_get_port_parent_id - Get the device's port parent identifier
8890  *	@dev: network device
8891  *	@ppid: pointer to a storage for the port's parent identifier
8892  *	@recurse: allow/disallow recursion to lower devices
8893  *
8894  *	Get the devices's port parent identifier
8895  */
8896 int dev_get_port_parent_id(struct net_device *dev,
8897 			   struct netdev_phys_item_id *ppid,
8898 			   bool recurse)
8899 {
8900 	const struct net_device_ops *ops = dev->netdev_ops;
8901 	struct netdev_phys_item_id first = { };
8902 	struct net_device *lower_dev;
8903 	struct list_head *iter;
8904 	int err;
8905 
8906 	if (ops->ndo_get_port_parent_id) {
8907 		err = ops->ndo_get_port_parent_id(dev, ppid);
8908 		if (err != -EOPNOTSUPP)
8909 			return err;
8910 	}
8911 
8912 	err = devlink_compat_switch_id_get(dev, ppid);
8913 	if (!recurse || err != -EOPNOTSUPP)
8914 		return err;
8915 
8916 	netdev_for_each_lower_dev(dev, lower_dev, iter) {
8917 		err = dev_get_port_parent_id(lower_dev, ppid, true);
8918 		if (err)
8919 			break;
8920 		if (!first.id_len)
8921 			first = *ppid;
8922 		else if (memcmp(&first, ppid, sizeof(*ppid)))
8923 			return -EOPNOTSUPP;
8924 	}
8925 
8926 	return err;
8927 }
8928 EXPORT_SYMBOL(dev_get_port_parent_id);
8929 
8930 /**
8931  *	netdev_port_same_parent_id - Indicate if two network devices have
8932  *	the same port parent identifier
8933  *	@a: first network device
8934  *	@b: second network device
8935  */
8936 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8937 {
8938 	struct netdev_phys_item_id a_id = { };
8939 	struct netdev_phys_item_id b_id = { };
8940 
8941 	if (dev_get_port_parent_id(a, &a_id, true) ||
8942 	    dev_get_port_parent_id(b, &b_id, true))
8943 		return false;
8944 
8945 	return netdev_phys_item_id_same(&a_id, &b_id);
8946 }
8947 EXPORT_SYMBOL(netdev_port_same_parent_id);
8948 
8949 /**
8950  *	dev_change_proto_down - set carrier according to proto_down.
8951  *
8952  *	@dev: device
8953  *	@proto_down: new value
8954  */
8955 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8956 {
8957 	if (!(dev->priv_flags & IFF_CHANGE_PROTO_DOWN))
8958 		return -EOPNOTSUPP;
8959 	if (!netif_device_present(dev))
8960 		return -ENODEV;
8961 	if (proto_down)
8962 		netif_carrier_off(dev);
8963 	else
8964 		netif_carrier_on(dev);
8965 	dev->proto_down = proto_down;
8966 	return 0;
8967 }
8968 
8969 /**
8970  *	dev_change_proto_down_reason - proto down reason
8971  *
8972  *	@dev: device
8973  *	@mask: proto down mask
8974  *	@value: proto down value
8975  */
8976 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
8977 				  u32 value)
8978 {
8979 	int b;
8980 
8981 	if (!mask) {
8982 		dev->proto_down_reason = value;
8983 	} else {
8984 		for_each_set_bit(b, &mask, 32) {
8985 			if (value & (1 << b))
8986 				dev->proto_down_reason |= BIT(b);
8987 			else
8988 				dev->proto_down_reason &= ~BIT(b);
8989 		}
8990 	}
8991 }
8992 
8993 struct bpf_xdp_link {
8994 	struct bpf_link link;
8995 	struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
8996 	int flags;
8997 };
8998 
8999 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9000 {
9001 	if (flags & XDP_FLAGS_HW_MODE)
9002 		return XDP_MODE_HW;
9003 	if (flags & XDP_FLAGS_DRV_MODE)
9004 		return XDP_MODE_DRV;
9005 	if (flags & XDP_FLAGS_SKB_MODE)
9006 		return XDP_MODE_SKB;
9007 	return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9008 }
9009 
9010 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9011 {
9012 	switch (mode) {
9013 	case XDP_MODE_SKB:
9014 		return generic_xdp_install;
9015 	case XDP_MODE_DRV:
9016 	case XDP_MODE_HW:
9017 		return dev->netdev_ops->ndo_bpf;
9018 	default:
9019 		return NULL;
9020 	}
9021 }
9022 
9023 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9024 					 enum bpf_xdp_mode mode)
9025 {
9026 	return dev->xdp_state[mode].link;
9027 }
9028 
9029 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9030 				     enum bpf_xdp_mode mode)
9031 {
9032 	struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9033 
9034 	if (link)
9035 		return link->link.prog;
9036 	return dev->xdp_state[mode].prog;
9037 }
9038 
9039 u8 dev_xdp_prog_count(struct net_device *dev)
9040 {
9041 	u8 count = 0;
9042 	int i;
9043 
9044 	for (i = 0; i < __MAX_XDP_MODE; i++)
9045 		if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9046 			count++;
9047 	return count;
9048 }
9049 EXPORT_SYMBOL_GPL(dev_xdp_prog_count);
9050 
9051 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9052 {
9053 	struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9054 
9055 	return prog ? prog->aux->id : 0;
9056 }
9057 
9058 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9059 			     struct bpf_xdp_link *link)
9060 {
9061 	dev->xdp_state[mode].link = link;
9062 	dev->xdp_state[mode].prog = NULL;
9063 }
9064 
9065 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9066 			     struct bpf_prog *prog)
9067 {
9068 	dev->xdp_state[mode].link = NULL;
9069 	dev->xdp_state[mode].prog = prog;
9070 }
9071 
9072 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9073 			   bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9074 			   u32 flags, struct bpf_prog *prog)
9075 {
9076 	struct netdev_bpf xdp;
9077 	int err;
9078 
9079 	memset(&xdp, 0, sizeof(xdp));
9080 	xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9081 	xdp.extack = extack;
9082 	xdp.flags = flags;
9083 	xdp.prog = prog;
9084 
9085 	/* Drivers assume refcnt is already incremented (i.e, prog pointer is
9086 	 * "moved" into driver), so they don't increment it on their own, but
9087 	 * they do decrement refcnt when program is detached or replaced.
9088 	 * Given net_device also owns link/prog, we need to bump refcnt here
9089 	 * to prevent drivers from underflowing it.
9090 	 */
9091 	if (prog)
9092 		bpf_prog_inc(prog);
9093 	err = bpf_op(dev, &xdp);
9094 	if (err) {
9095 		if (prog)
9096 			bpf_prog_put(prog);
9097 		return err;
9098 	}
9099 
9100 	if (mode != XDP_MODE_HW)
9101 		bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9102 
9103 	return 0;
9104 }
9105 
9106 static void dev_xdp_uninstall(struct net_device *dev)
9107 {
9108 	struct bpf_xdp_link *link;
9109 	struct bpf_prog *prog;
9110 	enum bpf_xdp_mode mode;
9111 	bpf_op_t bpf_op;
9112 
9113 	ASSERT_RTNL();
9114 
9115 	for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9116 		prog = dev_xdp_prog(dev, mode);
9117 		if (!prog)
9118 			continue;
9119 
9120 		bpf_op = dev_xdp_bpf_op(dev, mode);
9121 		if (!bpf_op)
9122 			continue;
9123 
9124 		WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9125 
9126 		/* auto-detach link from net device */
9127 		link = dev_xdp_link(dev, mode);
9128 		if (link)
9129 			link->dev = NULL;
9130 		else
9131 			bpf_prog_put(prog);
9132 
9133 		dev_xdp_set_link(dev, mode, NULL);
9134 	}
9135 }
9136 
9137 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9138 			  struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9139 			  struct bpf_prog *old_prog, u32 flags)
9140 {
9141 	unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9142 	struct bpf_prog *cur_prog;
9143 	struct net_device *upper;
9144 	struct list_head *iter;
9145 	enum bpf_xdp_mode mode;
9146 	bpf_op_t bpf_op;
9147 	int err;
9148 
9149 	ASSERT_RTNL();
9150 
9151 	/* either link or prog attachment, never both */
9152 	if (link && (new_prog || old_prog))
9153 		return -EINVAL;
9154 	/* link supports only XDP mode flags */
9155 	if (link && (flags & ~XDP_FLAGS_MODES)) {
9156 		NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9157 		return -EINVAL;
9158 	}
9159 	/* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9160 	if (num_modes > 1) {
9161 		NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9162 		return -EINVAL;
9163 	}
9164 	/* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9165 	if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9166 		NL_SET_ERR_MSG(extack,
9167 			       "More than one program loaded, unset mode is ambiguous");
9168 		return -EINVAL;
9169 	}
9170 	/* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9171 	if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9172 		NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9173 		return -EINVAL;
9174 	}
9175 
9176 	mode = dev_xdp_mode(dev, flags);
9177 	/* can't replace attached link */
9178 	if (dev_xdp_link(dev, mode)) {
9179 		NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9180 		return -EBUSY;
9181 	}
9182 
9183 	/* don't allow if an upper device already has a program */
9184 	netdev_for_each_upper_dev_rcu(dev, upper, iter) {
9185 		if (dev_xdp_prog_count(upper) > 0) {
9186 			NL_SET_ERR_MSG(extack, "Cannot attach when an upper device already has a program");
9187 			return -EEXIST;
9188 		}
9189 	}
9190 
9191 	cur_prog = dev_xdp_prog(dev, mode);
9192 	/* can't replace attached prog with link */
9193 	if (link && cur_prog) {
9194 		NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9195 		return -EBUSY;
9196 	}
9197 	if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9198 		NL_SET_ERR_MSG(extack, "Active program does not match expected");
9199 		return -EEXIST;
9200 	}
9201 
9202 	/* put effective new program into new_prog */
9203 	if (link)
9204 		new_prog = link->link.prog;
9205 
9206 	if (new_prog) {
9207 		bool offload = mode == XDP_MODE_HW;
9208 		enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9209 					       ? XDP_MODE_DRV : XDP_MODE_SKB;
9210 
9211 		if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9212 			NL_SET_ERR_MSG(extack, "XDP program already attached");
9213 			return -EBUSY;
9214 		}
9215 		if (!offload && dev_xdp_prog(dev, other_mode)) {
9216 			NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9217 			return -EEXIST;
9218 		}
9219 		if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9220 			NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9221 			return -EINVAL;
9222 		}
9223 		if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9224 			NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9225 			return -EINVAL;
9226 		}
9227 		if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9228 			NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9229 			return -EINVAL;
9230 		}
9231 	}
9232 
9233 	/* don't call drivers if the effective program didn't change */
9234 	if (new_prog != cur_prog) {
9235 		bpf_op = dev_xdp_bpf_op(dev, mode);
9236 		if (!bpf_op) {
9237 			NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9238 			return -EOPNOTSUPP;
9239 		}
9240 
9241 		err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9242 		if (err)
9243 			return err;
9244 	}
9245 
9246 	if (link)
9247 		dev_xdp_set_link(dev, mode, link);
9248 	else
9249 		dev_xdp_set_prog(dev, mode, new_prog);
9250 	if (cur_prog)
9251 		bpf_prog_put(cur_prog);
9252 
9253 	return 0;
9254 }
9255 
9256 static int dev_xdp_attach_link(struct net_device *dev,
9257 			       struct netlink_ext_ack *extack,
9258 			       struct bpf_xdp_link *link)
9259 {
9260 	return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9261 }
9262 
9263 static int dev_xdp_detach_link(struct net_device *dev,
9264 			       struct netlink_ext_ack *extack,
9265 			       struct bpf_xdp_link *link)
9266 {
9267 	enum bpf_xdp_mode mode;
9268 	bpf_op_t bpf_op;
9269 
9270 	ASSERT_RTNL();
9271 
9272 	mode = dev_xdp_mode(dev, link->flags);
9273 	if (dev_xdp_link(dev, mode) != link)
9274 		return -EINVAL;
9275 
9276 	bpf_op = dev_xdp_bpf_op(dev, mode);
9277 	WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9278 	dev_xdp_set_link(dev, mode, NULL);
9279 	return 0;
9280 }
9281 
9282 static void bpf_xdp_link_release(struct bpf_link *link)
9283 {
9284 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9285 
9286 	rtnl_lock();
9287 
9288 	/* if racing with net_device's tear down, xdp_link->dev might be
9289 	 * already NULL, in which case link was already auto-detached
9290 	 */
9291 	if (xdp_link->dev) {
9292 		WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9293 		xdp_link->dev = NULL;
9294 	}
9295 
9296 	rtnl_unlock();
9297 }
9298 
9299 static int bpf_xdp_link_detach(struct bpf_link *link)
9300 {
9301 	bpf_xdp_link_release(link);
9302 	return 0;
9303 }
9304 
9305 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9306 {
9307 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9308 
9309 	kfree(xdp_link);
9310 }
9311 
9312 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9313 				     struct seq_file *seq)
9314 {
9315 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9316 	u32 ifindex = 0;
9317 
9318 	rtnl_lock();
9319 	if (xdp_link->dev)
9320 		ifindex = xdp_link->dev->ifindex;
9321 	rtnl_unlock();
9322 
9323 	seq_printf(seq, "ifindex:\t%u\n", ifindex);
9324 }
9325 
9326 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9327 				       struct bpf_link_info *info)
9328 {
9329 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9330 	u32 ifindex = 0;
9331 
9332 	rtnl_lock();
9333 	if (xdp_link->dev)
9334 		ifindex = xdp_link->dev->ifindex;
9335 	rtnl_unlock();
9336 
9337 	info->xdp.ifindex = ifindex;
9338 	return 0;
9339 }
9340 
9341 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9342 			       struct bpf_prog *old_prog)
9343 {
9344 	struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9345 	enum bpf_xdp_mode mode;
9346 	bpf_op_t bpf_op;
9347 	int err = 0;
9348 
9349 	rtnl_lock();
9350 
9351 	/* link might have been auto-released already, so fail */
9352 	if (!xdp_link->dev) {
9353 		err = -ENOLINK;
9354 		goto out_unlock;
9355 	}
9356 
9357 	if (old_prog && link->prog != old_prog) {
9358 		err = -EPERM;
9359 		goto out_unlock;
9360 	}
9361 	old_prog = link->prog;
9362 	if (old_prog->type != new_prog->type ||
9363 	    old_prog->expected_attach_type != new_prog->expected_attach_type) {
9364 		err = -EINVAL;
9365 		goto out_unlock;
9366 	}
9367 
9368 	if (old_prog == new_prog) {
9369 		/* no-op, don't disturb drivers */
9370 		bpf_prog_put(new_prog);
9371 		goto out_unlock;
9372 	}
9373 
9374 	mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9375 	bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9376 	err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9377 			      xdp_link->flags, new_prog);
9378 	if (err)
9379 		goto out_unlock;
9380 
9381 	old_prog = xchg(&link->prog, new_prog);
9382 	bpf_prog_put(old_prog);
9383 
9384 out_unlock:
9385 	rtnl_unlock();
9386 	return err;
9387 }
9388 
9389 static const struct bpf_link_ops bpf_xdp_link_lops = {
9390 	.release = bpf_xdp_link_release,
9391 	.dealloc = bpf_xdp_link_dealloc,
9392 	.detach = bpf_xdp_link_detach,
9393 	.show_fdinfo = bpf_xdp_link_show_fdinfo,
9394 	.fill_link_info = bpf_xdp_link_fill_link_info,
9395 	.update_prog = bpf_xdp_link_update,
9396 };
9397 
9398 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9399 {
9400 	struct net *net = current->nsproxy->net_ns;
9401 	struct bpf_link_primer link_primer;
9402 	struct bpf_xdp_link *link;
9403 	struct net_device *dev;
9404 	int err, fd;
9405 
9406 	rtnl_lock();
9407 	dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9408 	if (!dev) {
9409 		rtnl_unlock();
9410 		return -EINVAL;
9411 	}
9412 
9413 	link = kzalloc(sizeof(*link), GFP_USER);
9414 	if (!link) {
9415 		err = -ENOMEM;
9416 		goto unlock;
9417 	}
9418 
9419 	bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9420 	link->dev = dev;
9421 	link->flags = attr->link_create.flags;
9422 
9423 	err = bpf_link_prime(&link->link, &link_primer);
9424 	if (err) {
9425 		kfree(link);
9426 		goto unlock;
9427 	}
9428 
9429 	err = dev_xdp_attach_link(dev, NULL, link);
9430 	rtnl_unlock();
9431 
9432 	if (err) {
9433 		link->dev = NULL;
9434 		bpf_link_cleanup(&link_primer);
9435 		goto out_put_dev;
9436 	}
9437 
9438 	fd = bpf_link_settle(&link_primer);
9439 	/* link itself doesn't hold dev's refcnt to not complicate shutdown */
9440 	dev_put(dev);
9441 	return fd;
9442 
9443 unlock:
9444 	rtnl_unlock();
9445 
9446 out_put_dev:
9447 	dev_put(dev);
9448 	return err;
9449 }
9450 
9451 /**
9452  *	dev_change_xdp_fd - set or clear a bpf program for a device rx path
9453  *	@dev: device
9454  *	@extack: netlink extended ack
9455  *	@fd: new program fd or negative value to clear
9456  *	@expected_fd: old program fd that userspace expects to replace or clear
9457  *	@flags: xdp-related flags
9458  *
9459  *	Set or clear a bpf program for a device
9460  */
9461 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9462 		      int fd, int expected_fd, u32 flags)
9463 {
9464 	enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9465 	struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9466 	int err;
9467 
9468 	ASSERT_RTNL();
9469 
9470 	if (fd >= 0) {
9471 		new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9472 						 mode != XDP_MODE_SKB);
9473 		if (IS_ERR(new_prog))
9474 			return PTR_ERR(new_prog);
9475 	}
9476 
9477 	if (expected_fd >= 0) {
9478 		old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9479 						 mode != XDP_MODE_SKB);
9480 		if (IS_ERR(old_prog)) {
9481 			err = PTR_ERR(old_prog);
9482 			old_prog = NULL;
9483 			goto err_out;
9484 		}
9485 	}
9486 
9487 	err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9488 
9489 err_out:
9490 	if (err && new_prog)
9491 		bpf_prog_put(new_prog);
9492 	if (old_prog)
9493 		bpf_prog_put(old_prog);
9494 	return err;
9495 }
9496 
9497 /**
9498  *	dev_new_index	-	allocate an ifindex
9499  *	@net: the applicable net namespace
9500  *
9501  *	Returns a suitable unique value for a new device interface
9502  *	number.  The caller must hold the rtnl semaphore or the
9503  *	dev_base_lock to be sure it remains unique.
9504  */
9505 static int dev_new_index(struct net *net)
9506 {
9507 	int ifindex = net->ifindex;
9508 
9509 	for (;;) {
9510 		if (++ifindex <= 0)
9511 			ifindex = 1;
9512 		if (!__dev_get_by_index(net, ifindex))
9513 			return net->ifindex = ifindex;
9514 	}
9515 }
9516 
9517 /* Delayed registration/unregisteration */
9518 LIST_HEAD(net_todo_list);
9519 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9520 
9521 static void net_set_todo(struct net_device *dev)
9522 {
9523 	list_add_tail(&dev->todo_list, &net_todo_list);
9524 	atomic_inc(&dev_net(dev)->dev_unreg_count);
9525 }
9526 
9527 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9528 	struct net_device *upper, netdev_features_t features)
9529 {
9530 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9531 	netdev_features_t feature;
9532 	int feature_bit;
9533 
9534 	for_each_netdev_feature(upper_disables, feature_bit) {
9535 		feature = __NETIF_F_BIT(feature_bit);
9536 		if (!(upper->wanted_features & feature)
9537 		    && (features & feature)) {
9538 			netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9539 				   &feature, upper->name);
9540 			features &= ~feature;
9541 		}
9542 	}
9543 
9544 	return features;
9545 }
9546 
9547 static void netdev_sync_lower_features(struct net_device *upper,
9548 	struct net_device *lower, netdev_features_t features)
9549 {
9550 	netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9551 	netdev_features_t feature;
9552 	int feature_bit;
9553 
9554 	for_each_netdev_feature(upper_disables, feature_bit) {
9555 		feature = __NETIF_F_BIT(feature_bit);
9556 		if (!(features & feature) && (lower->features & feature)) {
9557 			netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9558 				   &feature, lower->name);
9559 			lower->wanted_features &= ~feature;
9560 			__netdev_update_features(lower);
9561 
9562 			if (unlikely(lower->features & feature))
9563 				netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9564 					    &feature, lower->name);
9565 			else
9566 				netdev_features_change(lower);
9567 		}
9568 	}
9569 }
9570 
9571 static netdev_features_t netdev_fix_features(struct net_device *dev,
9572 	netdev_features_t features)
9573 {
9574 	/* Fix illegal checksum combinations */
9575 	if ((features & NETIF_F_HW_CSUM) &&
9576 	    (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9577 		netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9578 		features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9579 	}
9580 
9581 	/* TSO requires that SG is present as well. */
9582 	if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9583 		netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9584 		features &= ~NETIF_F_ALL_TSO;
9585 	}
9586 
9587 	if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9588 					!(features & NETIF_F_IP_CSUM)) {
9589 		netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9590 		features &= ~NETIF_F_TSO;
9591 		features &= ~NETIF_F_TSO_ECN;
9592 	}
9593 
9594 	if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9595 					 !(features & NETIF_F_IPV6_CSUM)) {
9596 		netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9597 		features &= ~NETIF_F_TSO6;
9598 	}
9599 
9600 	/* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9601 	if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9602 		features &= ~NETIF_F_TSO_MANGLEID;
9603 
9604 	/* TSO ECN requires that TSO is present as well. */
9605 	if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9606 		features &= ~NETIF_F_TSO_ECN;
9607 
9608 	/* Software GSO depends on SG. */
9609 	if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9610 		netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9611 		features &= ~NETIF_F_GSO;
9612 	}
9613 
9614 	/* GSO partial features require GSO partial be set */
9615 	if ((features & dev->gso_partial_features) &&
9616 	    !(features & NETIF_F_GSO_PARTIAL)) {
9617 		netdev_dbg(dev,
9618 			   "Dropping partially supported GSO features since no GSO partial.\n");
9619 		features &= ~dev->gso_partial_features;
9620 	}
9621 
9622 	if (!(features & NETIF_F_RXCSUM)) {
9623 		/* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9624 		 * successfully merged by hardware must also have the
9625 		 * checksum verified by hardware.  If the user does not
9626 		 * want to enable RXCSUM, logically, we should disable GRO_HW.
9627 		 */
9628 		if (features & NETIF_F_GRO_HW) {
9629 			netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9630 			features &= ~NETIF_F_GRO_HW;
9631 		}
9632 	}
9633 
9634 	/* LRO/HW-GRO features cannot be combined with RX-FCS */
9635 	if (features & NETIF_F_RXFCS) {
9636 		if (features & NETIF_F_LRO) {
9637 			netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9638 			features &= ~NETIF_F_LRO;
9639 		}
9640 
9641 		if (features & NETIF_F_GRO_HW) {
9642 			netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9643 			features &= ~NETIF_F_GRO_HW;
9644 		}
9645 	}
9646 
9647 	if ((features & NETIF_F_GRO_HW) && (features & NETIF_F_LRO)) {
9648 		netdev_dbg(dev, "Dropping LRO feature since HW-GRO is requested.\n");
9649 		features &= ~NETIF_F_LRO;
9650 	}
9651 
9652 	if (features & NETIF_F_HW_TLS_TX) {
9653 		bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9654 			(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9655 		bool hw_csum = features & NETIF_F_HW_CSUM;
9656 
9657 		if (!ip_csum && !hw_csum) {
9658 			netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9659 			features &= ~NETIF_F_HW_TLS_TX;
9660 		}
9661 	}
9662 
9663 	if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9664 		netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9665 		features &= ~NETIF_F_HW_TLS_RX;
9666 	}
9667 
9668 	return features;
9669 }
9670 
9671 int __netdev_update_features(struct net_device *dev)
9672 {
9673 	struct net_device *upper, *lower;
9674 	netdev_features_t features;
9675 	struct list_head *iter;
9676 	int err = -1;
9677 
9678 	ASSERT_RTNL();
9679 
9680 	features = netdev_get_wanted_features(dev);
9681 
9682 	if (dev->netdev_ops->ndo_fix_features)
9683 		features = dev->netdev_ops->ndo_fix_features(dev, features);
9684 
9685 	/* driver might be less strict about feature dependencies */
9686 	features = netdev_fix_features(dev, features);
9687 
9688 	/* some features can't be enabled if they're off on an upper device */
9689 	netdev_for_each_upper_dev_rcu(dev, upper, iter)
9690 		features = netdev_sync_upper_features(dev, upper, features);
9691 
9692 	if (dev->features == features)
9693 		goto sync_lower;
9694 
9695 	netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9696 		&dev->features, &features);
9697 
9698 	if (dev->netdev_ops->ndo_set_features)
9699 		err = dev->netdev_ops->ndo_set_features(dev, features);
9700 	else
9701 		err = 0;
9702 
9703 	if (unlikely(err < 0)) {
9704 		netdev_err(dev,
9705 			"set_features() failed (%d); wanted %pNF, left %pNF\n",
9706 			err, &features, &dev->features);
9707 		/* return non-0 since some features might have changed and
9708 		 * it's better to fire a spurious notification than miss it
9709 		 */
9710 		return -1;
9711 	}
9712 
9713 sync_lower:
9714 	/* some features must be disabled on lower devices when disabled
9715 	 * on an upper device (think: bonding master or bridge)
9716 	 */
9717 	netdev_for_each_lower_dev(dev, lower, iter)
9718 		netdev_sync_lower_features(dev, lower, features);
9719 
9720 	if (!err) {
9721 		netdev_features_t diff = features ^ dev->features;
9722 
9723 		if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9724 			/* udp_tunnel_{get,drop}_rx_info both need
9725 			 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9726 			 * device, or they won't do anything.
9727 			 * Thus we need to update dev->features
9728 			 * *before* calling udp_tunnel_get_rx_info,
9729 			 * but *after* calling udp_tunnel_drop_rx_info.
9730 			 */
9731 			if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9732 				dev->features = features;
9733 				udp_tunnel_get_rx_info(dev);
9734 			} else {
9735 				udp_tunnel_drop_rx_info(dev);
9736 			}
9737 		}
9738 
9739 		if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9740 			if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9741 				dev->features = features;
9742 				err |= vlan_get_rx_ctag_filter_info(dev);
9743 			} else {
9744 				vlan_drop_rx_ctag_filter_info(dev);
9745 			}
9746 		}
9747 
9748 		if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9749 			if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9750 				dev->features = features;
9751 				err |= vlan_get_rx_stag_filter_info(dev);
9752 			} else {
9753 				vlan_drop_rx_stag_filter_info(dev);
9754 			}
9755 		}
9756 
9757 		dev->features = features;
9758 	}
9759 
9760 	return err < 0 ? 0 : 1;
9761 }
9762 
9763 /**
9764  *	netdev_update_features - recalculate device features
9765  *	@dev: the device to check
9766  *
9767  *	Recalculate dev->features set and send notifications if it
9768  *	has changed. Should be called after driver or hardware dependent
9769  *	conditions might have changed that influence the features.
9770  */
9771 void netdev_update_features(struct net_device *dev)
9772 {
9773 	if (__netdev_update_features(dev))
9774 		netdev_features_change(dev);
9775 }
9776 EXPORT_SYMBOL(netdev_update_features);
9777 
9778 /**
9779  *	netdev_change_features - recalculate device features
9780  *	@dev: the device to check
9781  *
9782  *	Recalculate dev->features set and send notifications even
9783  *	if they have not changed. Should be called instead of
9784  *	netdev_update_features() if also dev->vlan_features might
9785  *	have changed to allow the changes to be propagated to stacked
9786  *	VLAN devices.
9787  */
9788 void netdev_change_features(struct net_device *dev)
9789 {
9790 	__netdev_update_features(dev);
9791 	netdev_features_change(dev);
9792 }
9793 EXPORT_SYMBOL(netdev_change_features);
9794 
9795 /**
9796  *	netif_stacked_transfer_operstate -	transfer operstate
9797  *	@rootdev: the root or lower level device to transfer state from
9798  *	@dev: the device to transfer operstate to
9799  *
9800  *	Transfer operational state from root to device. This is normally
9801  *	called when a stacking relationship exists between the root
9802  *	device and the device(a leaf device).
9803  */
9804 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9805 					struct net_device *dev)
9806 {
9807 	if (rootdev->operstate == IF_OPER_DORMANT)
9808 		netif_dormant_on(dev);
9809 	else
9810 		netif_dormant_off(dev);
9811 
9812 	if (rootdev->operstate == IF_OPER_TESTING)
9813 		netif_testing_on(dev);
9814 	else
9815 		netif_testing_off(dev);
9816 
9817 	if (netif_carrier_ok(rootdev))
9818 		netif_carrier_on(dev);
9819 	else
9820 		netif_carrier_off(dev);
9821 }
9822 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9823 
9824 static int netif_alloc_rx_queues(struct net_device *dev)
9825 {
9826 	unsigned int i, count = dev->num_rx_queues;
9827 	struct netdev_rx_queue *rx;
9828 	size_t sz = count * sizeof(*rx);
9829 	int err = 0;
9830 
9831 	BUG_ON(count < 1);
9832 
9833 	rx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9834 	if (!rx)
9835 		return -ENOMEM;
9836 
9837 	dev->_rx = rx;
9838 
9839 	for (i = 0; i < count; i++) {
9840 		rx[i].dev = dev;
9841 
9842 		/* XDP RX-queue setup */
9843 		err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
9844 		if (err < 0)
9845 			goto err_rxq_info;
9846 	}
9847 	return 0;
9848 
9849 err_rxq_info:
9850 	/* Rollback successful reg's and free other resources */
9851 	while (i--)
9852 		xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9853 	kvfree(dev->_rx);
9854 	dev->_rx = NULL;
9855 	return err;
9856 }
9857 
9858 static void netif_free_rx_queues(struct net_device *dev)
9859 {
9860 	unsigned int i, count = dev->num_rx_queues;
9861 
9862 	/* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9863 	if (!dev->_rx)
9864 		return;
9865 
9866 	for (i = 0; i < count; i++)
9867 		xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9868 
9869 	kvfree(dev->_rx);
9870 }
9871 
9872 static void netdev_init_one_queue(struct net_device *dev,
9873 				  struct netdev_queue *queue, void *_unused)
9874 {
9875 	/* Initialize queue lock */
9876 	spin_lock_init(&queue->_xmit_lock);
9877 	netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9878 	queue->xmit_lock_owner = -1;
9879 	netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9880 	queue->dev = dev;
9881 #ifdef CONFIG_BQL
9882 	dql_init(&queue->dql, HZ);
9883 #endif
9884 }
9885 
9886 static void netif_free_tx_queues(struct net_device *dev)
9887 {
9888 	kvfree(dev->_tx);
9889 }
9890 
9891 static int netif_alloc_netdev_queues(struct net_device *dev)
9892 {
9893 	unsigned int count = dev->num_tx_queues;
9894 	struct netdev_queue *tx;
9895 	size_t sz = count * sizeof(*tx);
9896 
9897 	if (count < 1 || count > 0xffff)
9898 		return -EINVAL;
9899 
9900 	tx = kvzalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
9901 	if (!tx)
9902 		return -ENOMEM;
9903 
9904 	dev->_tx = tx;
9905 
9906 	netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9907 	spin_lock_init(&dev->tx_global_lock);
9908 
9909 	return 0;
9910 }
9911 
9912 void netif_tx_stop_all_queues(struct net_device *dev)
9913 {
9914 	unsigned int i;
9915 
9916 	for (i = 0; i < dev->num_tx_queues; i++) {
9917 		struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9918 
9919 		netif_tx_stop_queue(txq);
9920 	}
9921 }
9922 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9923 
9924 /**
9925  * register_netdevice() - register a network device
9926  * @dev: device to register
9927  *
9928  * Take a prepared network device structure and make it externally accessible.
9929  * A %NETDEV_REGISTER message is sent to the netdev notifier chain.
9930  * Callers must hold the rtnl lock - you may want register_netdev()
9931  * instead of this.
9932  */
9933 int register_netdevice(struct net_device *dev)
9934 {
9935 	int ret;
9936 	struct net *net = dev_net(dev);
9937 
9938 	BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9939 		     NETDEV_FEATURE_COUNT);
9940 	BUG_ON(dev_boot_phase);
9941 	ASSERT_RTNL();
9942 
9943 	might_sleep();
9944 
9945 	/* When net_device's are persistent, this will be fatal. */
9946 	BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9947 	BUG_ON(!net);
9948 
9949 	ret = ethtool_check_ops(dev->ethtool_ops);
9950 	if (ret)
9951 		return ret;
9952 
9953 	spin_lock_init(&dev->addr_list_lock);
9954 	netdev_set_addr_lockdep_class(dev);
9955 
9956 	ret = dev_get_valid_name(net, dev, dev->name);
9957 	if (ret < 0)
9958 		goto out;
9959 
9960 	ret = -ENOMEM;
9961 	dev->name_node = netdev_name_node_head_alloc(dev);
9962 	if (!dev->name_node)
9963 		goto out;
9964 
9965 	/* Init, if this function is available */
9966 	if (dev->netdev_ops->ndo_init) {
9967 		ret = dev->netdev_ops->ndo_init(dev);
9968 		if (ret) {
9969 			if (ret > 0)
9970 				ret = -EIO;
9971 			goto err_free_name;
9972 		}
9973 	}
9974 
9975 	if (((dev->hw_features | dev->features) &
9976 	     NETIF_F_HW_VLAN_CTAG_FILTER) &&
9977 	    (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9978 	     !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9979 		netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
9980 		ret = -EINVAL;
9981 		goto err_uninit;
9982 	}
9983 
9984 	ret = -EBUSY;
9985 	if (!dev->ifindex)
9986 		dev->ifindex = dev_new_index(net);
9987 	else if (__dev_get_by_index(net, dev->ifindex))
9988 		goto err_uninit;
9989 
9990 	/* Transfer changeable features to wanted_features and enable
9991 	 * software offloads (GSO and GRO).
9992 	 */
9993 	dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
9994 	dev->features |= NETIF_F_SOFT_FEATURES;
9995 
9996 	if (dev->udp_tunnel_nic_info) {
9997 		dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9998 		dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9999 	}
10000 
10001 	dev->wanted_features = dev->features & dev->hw_features;
10002 
10003 	if (!(dev->flags & IFF_LOOPBACK))
10004 		dev->hw_features |= NETIF_F_NOCACHE_COPY;
10005 
10006 	/* If IPv4 TCP segmentation offload is supported we should also
10007 	 * allow the device to enable segmenting the frame with the option
10008 	 * of ignoring a static IP ID value.  This doesn't enable the
10009 	 * feature itself but allows the user to enable it later.
10010 	 */
10011 	if (dev->hw_features & NETIF_F_TSO)
10012 		dev->hw_features |= NETIF_F_TSO_MANGLEID;
10013 	if (dev->vlan_features & NETIF_F_TSO)
10014 		dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10015 	if (dev->mpls_features & NETIF_F_TSO)
10016 		dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10017 	if (dev->hw_enc_features & NETIF_F_TSO)
10018 		dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10019 
10020 	/* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10021 	 */
10022 	dev->vlan_features |= NETIF_F_HIGHDMA;
10023 
10024 	/* Make NETIF_F_SG inheritable to tunnel devices.
10025 	 */
10026 	dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10027 
10028 	/* Make NETIF_F_SG inheritable to MPLS.
10029 	 */
10030 	dev->mpls_features |= NETIF_F_SG;
10031 
10032 	ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10033 	ret = notifier_to_errno(ret);
10034 	if (ret)
10035 		goto err_uninit;
10036 
10037 	ret = netdev_register_kobject(dev);
10038 	write_lock(&dev_base_lock);
10039 	dev->reg_state = ret ? NETREG_UNREGISTERED : NETREG_REGISTERED;
10040 	write_unlock(&dev_base_lock);
10041 	if (ret)
10042 		goto err_uninit_notify;
10043 
10044 	__netdev_update_features(dev);
10045 
10046 	/*
10047 	 *	Default initial state at registry is that the
10048 	 *	device is present.
10049 	 */
10050 
10051 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10052 
10053 	linkwatch_init_dev(dev);
10054 
10055 	dev_init_scheduler(dev);
10056 
10057 	netdev_hold(dev, &dev->dev_registered_tracker, GFP_KERNEL);
10058 	list_netdevice(dev);
10059 
10060 	add_device_randomness(dev->dev_addr, dev->addr_len);
10061 
10062 	/* If the device has permanent device address, driver should
10063 	 * set dev_addr and also addr_assign_type should be set to
10064 	 * NET_ADDR_PERM (default value).
10065 	 */
10066 	if (dev->addr_assign_type == NET_ADDR_PERM)
10067 		memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10068 
10069 	/* Notify protocols, that a new device appeared. */
10070 	ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10071 	ret = notifier_to_errno(ret);
10072 	if (ret) {
10073 		/* Expect explicit free_netdev() on failure */
10074 		dev->needs_free_netdev = false;
10075 		unregister_netdevice_queue(dev, NULL);
10076 		goto out;
10077 	}
10078 	/*
10079 	 *	Prevent userspace races by waiting until the network
10080 	 *	device is fully setup before sending notifications.
10081 	 */
10082 	if (!dev->rtnl_link_ops ||
10083 	    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10084 		rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
10085 
10086 out:
10087 	return ret;
10088 
10089 err_uninit_notify:
10090 	call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10091 err_uninit:
10092 	if (dev->netdev_ops->ndo_uninit)
10093 		dev->netdev_ops->ndo_uninit(dev);
10094 	if (dev->priv_destructor)
10095 		dev->priv_destructor(dev);
10096 err_free_name:
10097 	netdev_name_node_free(dev->name_node);
10098 	goto out;
10099 }
10100 EXPORT_SYMBOL(register_netdevice);
10101 
10102 /**
10103  *	init_dummy_netdev	- init a dummy network device for NAPI
10104  *	@dev: device to init
10105  *
10106  *	This takes a network device structure and initialize the minimum
10107  *	amount of fields so it can be used to schedule NAPI polls without
10108  *	registering a full blown interface. This is to be used by drivers
10109  *	that need to tie several hardware interfaces to a single NAPI
10110  *	poll scheduler due to HW limitations.
10111  */
10112 int init_dummy_netdev(struct net_device *dev)
10113 {
10114 	/* Clear everything. Note we don't initialize spinlocks
10115 	 * are they aren't supposed to be taken by any of the
10116 	 * NAPI code and this dummy netdev is supposed to be
10117 	 * only ever used for NAPI polls
10118 	 */
10119 	memset(dev, 0, sizeof(struct net_device));
10120 
10121 	/* make sure we BUG if trying to hit standard
10122 	 * register/unregister code path
10123 	 */
10124 	dev->reg_state = NETREG_DUMMY;
10125 
10126 	/* NAPI wants this */
10127 	INIT_LIST_HEAD(&dev->napi_list);
10128 
10129 	/* a dummy interface is started by default */
10130 	set_bit(__LINK_STATE_PRESENT, &dev->state);
10131 	set_bit(__LINK_STATE_START, &dev->state);
10132 
10133 	/* napi_busy_loop stats accounting wants this */
10134 	dev_net_set(dev, &init_net);
10135 
10136 	/* Note : We dont allocate pcpu_refcnt for dummy devices,
10137 	 * because users of this 'device' dont need to change
10138 	 * its refcount.
10139 	 */
10140 
10141 	return 0;
10142 }
10143 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10144 
10145 
10146 /**
10147  *	register_netdev	- register a network device
10148  *	@dev: device to register
10149  *
10150  *	Take a completed network device structure and add it to the kernel
10151  *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10152  *	chain. 0 is returned on success. A negative errno code is returned
10153  *	on a failure to set up the device, or if the name is a duplicate.
10154  *
10155  *	This is a wrapper around register_netdevice that takes the rtnl semaphore
10156  *	and expands the device name if you passed a format string to
10157  *	alloc_netdev.
10158  */
10159 int register_netdev(struct net_device *dev)
10160 {
10161 	int err;
10162 
10163 	if (rtnl_lock_killable())
10164 		return -EINTR;
10165 	err = register_netdevice(dev);
10166 	rtnl_unlock();
10167 	return err;
10168 }
10169 EXPORT_SYMBOL(register_netdev);
10170 
10171 int netdev_refcnt_read(const struct net_device *dev)
10172 {
10173 #ifdef CONFIG_PCPU_DEV_REFCNT
10174 	int i, refcnt = 0;
10175 
10176 	for_each_possible_cpu(i)
10177 		refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10178 	return refcnt;
10179 #else
10180 	return refcount_read(&dev->dev_refcnt);
10181 #endif
10182 }
10183 EXPORT_SYMBOL(netdev_refcnt_read);
10184 
10185 int netdev_unregister_timeout_secs __read_mostly = 10;
10186 
10187 #define WAIT_REFS_MIN_MSECS 1
10188 #define WAIT_REFS_MAX_MSECS 250
10189 /**
10190  * netdev_wait_allrefs_any - wait until all references are gone.
10191  * @list: list of net_devices to wait on
10192  *
10193  * This is called when unregistering network devices.
10194  *
10195  * Any protocol or device that holds a reference should register
10196  * for netdevice notification, and cleanup and put back the
10197  * reference if they receive an UNREGISTER event.
10198  * We can get stuck here if buggy protocols don't correctly
10199  * call dev_put.
10200  */
10201 static struct net_device *netdev_wait_allrefs_any(struct list_head *list)
10202 {
10203 	unsigned long rebroadcast_time, warning_time;
10204 	struct net_device *dev;
10205 	int wait = 0;
10206 
10207 	rebroadcast_time = warning_time = jiffies;
10208 
10209 	list_for_each_entry(dev, list, todo_list)
10210 		if (netdev_refcnt_read(dev) == 1)
10211 			return dev;
10212 
10213 	while (true) {
10214 		if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10215 			rtnl_lock();
10216 
10217 			/* Rebroadcast unregister notification */
10218 			list_for_each_entry(dev, list, todo_list)
10219 				call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10220 
10221 			__rtnl_unlock();
10222 			rcu_barrier();
10223 			rtnl_lock();
10224 
10225 			list_for_each_entry(dev, list, todo_list)
10226 				if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10227 					     &dev->state)) {
10228 					/* We must not have linkwatch events
10229 					 * pending on unregister. If this
10230 					 * happens, we simply run the queue
10231 					 * unscheduled, resulting in a noop
10232 					 * for this device.
10233 					 */
10234 					linkwatch_run_queue();
10235 					break;
10236 				}
10237 
10238 			__rtnl_unlock();
10239 
10240 			rebroadcast_time = jiffies;
10241 		}
10242 
10243 		if (!wait) {
10244 			rcu_barrier();
10245 			wait = WAIT_REFS_MIN_MSECS;
10246 		} else {
10247 			msleep(wait);
10248 			wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10249 		}
10250 
10251 		list_for_each_entry(dev, list, todo_list)
10252 			if (netdev_refcnt_read(dev) == 1)
10253 				return dev;
10254 
10255 		if (time_after(jiffies, warning_time +
10256 			       READ_ONCE(netdev_unregister_timeout_secs) * HZ)) {
10257 			list_for_each_entry(dev, list, todo_list) {
10258 				pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10259 					 dev->name, netdev_refcnt_read(dev));
10260 				ref_tracker_dir_print(&dev->refcnt_tracker, 10);
10261 			}
10262 
10263 			warning_time = jiffies;
10264 		}
10265 	}
10266 }
10267 
10268 /* The sequence is:
10269  *
10270  *	rtnl_lock();
10271  *	...
10272  *	register_netdevice(x1);
10273  *	register_netdevice(x2);
10274  *	...
10275  *	unregister_netdevice(y1);
10276  *	unregister_netdevice(y2);
10277  *      ...
10278  *	rtnl_unlock();
10279  *	free_netdev(y1);
10280  *	free_netdev(y2);
10281  *
10282  * We are invoked by rtnl_unlock().
10283  * This allows us to deal with problems:
10284  * 1) We can delete sysfs objects which invoke hotplug
10285  *    without deadlocking with linkwatch via keventd.
10286  * 2) Since we run with the RTNL semaphore not held, we can sleep
10287  *    safely in order to wait for the netdev refcnt to drop to zero.
10288  *
10289  * We must not return until all unregister events added during
10290  * the interval the lock was held have been completed.
10291  */
10292 void netdev_run_todo(void)
10293 {
10294 	struct net_device *dev, *tmp;
10295 	struct list_head list;
10296 #ifdef CONFIG_LOCKDEP
10297 	struct list_head unlink_list;
10298 
10299 	list_replace_init(&net_unlink_list, &unlink_list);
10300 
10301 	while (!list_empty(&unlink_list)) {
10302 		struct net_device *dev = list_first_entry(&unlink_list,
10303 							  struct net_device,
10304 							  unlink_list);
10305 		list_del_init(&dev->unlink_list);
10306 		dev->nested_level = dev->lower_level - 1;
10307 	}
10308 #endif
10309 
10310 	/* Snapshot list, allow later requests */
10311 	list_replace_init(&net_todo_list, &list);
10312 
10313 	__rtnl_unlock();
10314 
10315 	/* Wait for rcu callbacks to finish before next phase */
10316 	if (!list_empty(&list))
10317 		rcu_barrier();
10318 
10319 	list_for_each_entry_safe(dev, tmp, &list, todo_list) {
10320 		if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10321 			netdev_WARN(dev, "run_todo but not unregistering\n");
10322 			list_del(&dev->todo_list);
10323 			continue;
10324 		}
10325 
10326 		write_lock(&dev_base_lock);
10327 		dev->reg_state = NETREG_UNREGISTERED;
10328 		write_unlock(&dev_base_lock);
10329 		linkwatch_forget_dev(dev);
10330 	}
10331 
10332 	while (!list_empty(&list)) {
10333 		dev = netdev_wait_allrefs_any(&list);
10334 		list_del(&dev->todo_list);
10335 
10336 		/* paranoia */
10337 		BUG_ON(netdev_refcnt_read(dev) != 1);
10338 		BUG_ON(!list_empty(&dev->ptype_all));
10339 		BUG_ON(!list_empty(&dev->ptype_specific));
10340 		WARN_ON(rcu_access_pointer(dev->ip_ptr));
10341 		WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10342 
10343 		if (dev->priv_destructor)
10344 			dev->priv_destructor(dev);
10345 		if (dev->needs_free_netdev)
10346 			free_netdev(dev);
10347 
10348 		if (atomic_dec_and_test(&dev_net(dev)->dev_unreg_count))
10349 			wake_up(&netdev_unregistering_wq);
10350 
10351 		/* Free network device */
10352 		kobject_put(&dev->dev.kobj);
10353 	}
10354 }
10355 
10356 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10357  * all the same fields in the same order as net_device_stats, with only
10358  * the type differing, but rtnl_link_stats64 may have additional fields
10359  * at the end for newer counters.
10360  */
10361 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10362 			     const struct net_device_stats *netdev_stats)
10363 {
10364 	size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
10365 	const atomic_long_t *src = (atomic_long_t *)netdev_stats;
10366 	u64 *dst = (u64 *)stats64;
10367 
10368 	BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10369 	for (i = 0; i < n; i++)
10370 		dst[i] = (unsigned long)atomic_long_read(&src[i]);
10371 	/* zero out counters that only exist in rtnl_link_stats64 */
10372 	memset((char *)stats64 + n * sizeof(u64), 0,
10373 	       sizeof(*stats64) - n * sizeof(u64));
10374 }
10375 EXPORT_SYMBOL(netdev_stats_to_stats64);
10376 
10377 struct net_device_core_stats __percpu *netdev_core_stats_alloc(struct net_device *dev)
10378 {
10379 	struct net_device_core_stats __percpu *p;
10380 
10381 	p = alloc_percpu_gfp(struct net_device_core_stats,
10382 			     GFP_ATOMIC | __GFP_NOWARN);
10383 
10384 	if (p && cmpxchg(&dev->core_stats, NULL, p))
10385 		free_percpu(p);
10386 
10387 	/* This READ_ONCE() pairs with the cmpxchg() above */
10388 	return READ_ONCE(dev->core_stats);
10389 }
10390 EXPORT_SYMBOL(netdev_core_stats_alloc);
10391 
10392 /**
10393  *	dev_get_stats	- get network device statistics
10394  *	@dev: device to get statistics from
10395  *	@storage: place to store stats
10396  *
10397  *	Get network statistics from device. Return @storage.
10398  *	The device driver may provide its own method by setting
10399  *	dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10400  *	otherwise the internal statistics structure is used.
10401  */
10402 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10403 					struct rtnl_link_stats64 *storage)
10404 {
10405 	const struct net_device_ops *ops = dev->netdev_ops;
10406 	const struct net_device_core_stats __percpu *p;
10407 
10408 	if (ops->ndo_get_stats64) {
10409 		memset(storage, 0, sizeof(*storage));
10410 		ops->ndo_get_stats64(dev, storage);
10411 	} else if (ops->ndo_get_stats) {
10412 		netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10413 	} else {
10414 		netdev_stats_to_stats64(storage, &dev->stats);
10415 	}
10416 
10417 	/* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
10418 	p = READ_ONCE(dev->core_stats);
10419 	if (p) {
10420 		const struct net_device_core_stats *core_stats;
10421 		int i;
10422 
10423 		for_each_possible_cpu(i) {
10424 			core_stats = per_cpu_ptr(p, i);
10425 			storage->rx_dropped += READ_ONCE(core_stats->rx_dropped);
10426 			storage->tx_dropped += READ_ONCE(core_stats->tx_dropped);
10427 			storage->rx_nohandler += READ_ONCE(core_stats->rx_nohandler);
10428 			storage->rx_otherhost_dropped += READ_ONCE(core_stats->rx_otherhost_dropped);
10429 		}
10430 	}
10431 	return storage;
10432 }
10433 EXPORT_SYMBOL(dev_get_stats);
10434 
10435 /**
10436  *	dev_fetch_sw_netstats - get per-cpu network device statistics
10437  *	@s: place to store stats
10438  *	@netstats: per-cpu network stats to read from
10439  *
10440  *	Read per-cpu network statistics and populate the related fields in @s.
10441  */
10442 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10443 			   const struct pcpu_sw_netstats __percpu *netstats)
10444 {
10445 	int cpu;
10446 
10447 	for_each_possible_cpu(cpu) {
10448 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
10449 		const struct pcpu_sw_netstats *stats;
10450 		unsigned int start;
10451 
10452 		stats = per_cpu_ptr(netstats, cpu);
10453 		do {
10454 			start = u64_stats_fetch_begin(&stats->syncp);
10455 			rx_packets = u64_stats_read(&stats->rx_packets);
10456 			rx_bytes   = u64_stats_read(&stats->rx_bytes);
10457 			tx_packets = u64_stats_read(&stats->tx_packets);
10458 			tx_bytes   = u64_stats_read(&stats->tx_bytes);
10459 		} while (u64_stats_fetch_retry(&stats->syncp, start));
10460 
10461 		s->rx_packets += rx_packets;
10462 		s->rx_bytes   += rx_bytes;
10463 		s->tx_packets += tx_packets;
10464 		s->tx_bytes   += tx_bytes;
10465 	}
10466 }
10467 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10468 
10469 /**
10470  *	dev_get_tstats64 - ndo_get_stats64 implementation
10471  *	@dev: device to get statistics from
10472  *	@s: place to store stats
10473  *
10474  *	Populate @s from dev->stats and dev->tstats. Can be used as
10475  *	ndo_get_stats64() callback.
10476  */
10477 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10478 {
10479 	netdev_stats_to_stats64(s, &dev->stats);
10480 	dev_fetch_sw_netstats(s, dev->tstats);
10481 }
10482 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10483 
10484 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10485 {
10486 	struct netdev_queue *queue = dev_ingress_queue(dev);
10487 
10488 #ifdef CONFIG_NET_CLS_ACT
10489 	if (queue)
10490 		return queue;
10491 	queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10492 	if (!queue)
10493 		return NULL;
10494 	netdev_init_one_queue(dev, queue, NULL);
10495 	RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10496 	queue->qdisc_sleeping = &noop_qdisc;
10497 	rcu_assign_pointer(dev->ingress_queue, queue);
10498 #endif
10499 	return queue;
10500 }
10501 
10502 static const struct ethtool_ops default_ethtool_ops;
10503 
10504 void netdev_set_default_ethtool_ops(struct net_device *dev,
10505 				    const struct ethtool_ops *ops)
10506 {
10507 	if (dev->ethtool_ops == &default_ethtool_ops)
10508 		dev->ethtool_ops = ops;
10509 }
10510 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10511 
10512 /**
10513  * netdev_sw_irq_coalesce_default_on() - enable SW IRQ coalescing by default
10514  * @dev: netdev to enable the IRQ coalescing on
10515  *
10516  * Sets a conservative default for SW IRQ coalescing. Users can use
10517  * sysfs attributes to override the default values.
10518  */
10519 void netdev_sw_irq_coalesce_default_on(struct net_device *dev)
10520 {
10521 	WARN_ON(dev->reg_state == NETREG_REGISTERED);
10522 
10523 	dev->gro_flush_timeout = 20000;
10524 	dev->napi_defer_hard_irqs = 1;
10525 }
10526 EXPORT_SYMBOL_GPL(netdev_sw_irq_coalesce_default_on);
10527 
10528 void netdev_freemem(struct net_device *dev)
10529 {
10530 	char *addr = (char *)dev - dev->padded;
10531 
10532 	kvfree(addr);
10533 }
10534 
10535 /**
10536  * alloc_netdev_mqs - allocate network device
10537  * @sizeof_priv: size of private data to allocate space for
10538  * @name: device name format string
10539  * @name_assign_type: origin of device name
10540  * @setup: callback to initialize device
10541  * @txqs: the number of TX subqueues to allocate
10542  * @rxqs: the number of RX subqueues to allocate
10543  *
10544  * Allocates a struct net_device with private data area for driver use
10545  * and performs basic initialization.  Also allocates subqueue structs
10546  * for each queue on the device.
10547  */
10548 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10549 		unsigned char name_assign_type,
10550 		void (*setup)(struct net_device *),
10551 		unsigned int txqs, unsigned int rxqs)
10552 {
10553 	struct net_device *dev;
10554 	unsigned int alloc_size;
10555 	struct net_device *p;
10556 
10557 	BUG_ON(strlen(name) >= sizeof(dev->name));
10558 
10559 	if (txqs < 1) {
10560 		pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10561 		return NULL;
10562 	}
10563 
10564 	if (rxqs < 1) {
10565 		pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10566 		return NULL;
10567 	}
10568 
10569 	alloc_size = sizeof(struct net_device);
10570 	if (sizeof_priv) {
10571 		/* ensure 32-byte alignment of private area */
10572 		alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10573 		alloc_size += sizeof_priv;
10574 	}
10575 	/* ensure 32-byte alignment of whole construct */
10576 	alloc_size += NETDEV_ALIGN - 1;
10577 
10578 	p = kvzalloc(alloc_size, GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL);
10579 	if (!p)
10580 		return NULL;
10581 
10582 	dev = PTR_ALIGN(p, NETDEV_ALIGN);
10583 	dev->padded = (char *)dev - (char *)p;
10584 
10585 	ref_tracker_dir_init(&dev->refcnt_tracker, 128);
10586 #ifdef CONFIG_PCPU_DEV_REFCNT
10587 	dev->pcpu_refcnt = alloc_percpu(int);
10588 	if (!dev->pcpu_refcnt)
10589 		goto free_dev;
10590 	__dev_hold(dev);
10591 #else
10592 	refcount_set(&dev->dev_refcnt, 1);
10593 #endif
10594 
10595 	if (dev_addr_init(dev))
10596 		goto free_pcpu;
10597 
10598 	dev_mc_init(dev);
10599 	dev_uc_init(dev);
10600 
10601 	dev_net_set(dev, &init_net);
10602 
10603 	dev->gso_max_size = GSO_LEGACY_MAX_SIZE;
10604 	dev->gso_max_segs = GSO_MAX_SEGS;
10605 	dev->gro_max_size = GRO_LEGACY_MAX_SIZE;
10606 	dev->tso_max_size = TSO_LEGACY_MAX_SIZE;
10607 	dev->tso_max_segs = TSO_MAX_SEGS;
10608 	dev->upper_level = 1;
10609 	dev->lower_level = 1;
10610 #ifdef CONFIG_LOCKDEP
10611 	dev->nested_level = 0;
10612 	INIT_LIST_HEAD(&dev->unlink_list);
10613 #endif
10614 
10615 	INIT_LIST_HEAD(&dev->napi_list);
10616 	INIT_LIST_HEAD(&dev->unreg_list);
10617 	INIT_LIST_HEAD(&dev->close_list);
10618 	INIT_LIST_HEAD(&dev->link_watch_list);
10619 	INIT_LIST_HEAD(&dev->adj_list.upper);
10620 	INIT_LIST_HEAD(&dev->adj_list.lower);
10621 	INIT_LIST_HEAD(&dev->ptype_all);
10622 	INIT_LIST_HEAD(&dev->ptype_specific);
10623 	INIT_LIST_HEAD(&dev->net_notifier_list);
10624 #ifdef CONFIG_NET_SCHED
10625 	hash_init(dev->qdisc_hash);
10626 #endif
10627 	dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10628 	setup(dev);
10629 
10630 	if (!dev->tx_queue_len) {
10631 		dev->priv_flags |= IFF_NO_QUEUE;
10632 		dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10633 	}
10634 
10635 	dev->num_tx_queues = txqs;
10636 	dev->real_num_tx_queues = txqs;
10637 	if (netif_alloc_netdev_queues(dev))
10638 		goto free_all;
10639 
10640 	dev->num_rx_queues = rxqs;
10641 	dev->real_num_rx_queues = rxqs;
10642 	if (netif_alloc_rx_queues(dev))
10643 		goto free_all;
10644 
10645 	strcpy(dev->name, name);
10646 	dev->name_assign_type = name_assign_type;
10647 	dev->group = INIT_NETDEV_GROUP;
10648 	if (!dev->ethtool_ops)
10649 		dev->ethtool_ops = &default_ethtool_ops;
10650 
10651 	nf_hook_netdev_init(dev);
10652 
10653 	return dev;
10654 
10655 free_all:
10656 	free_netdev(dev);
10657 	return NULL;
10658 
10659 free_pcpu:
10660 #ifdef CONFIG_PCPU_DEV_REFCNT
10661 	free_percpu(dev->pcpu_refcnt);
10662 free_dev:
10663 #endif
10664 	netdev_freemem(dev);
10665 	return NULL;
10666 }
10667 EXPORT_SYMBOL(alloc_netdev_mqs);
10668 
10669 /**
10670  * free_netdev - free network device
10671  * @dev: device
10672  *
10673  * This function does the last stage of destroying an allocated device
10674  * interface. The reference to the device object is released. If this
10675  * is the last reference then it will be freed.Must be called in process
10676  * context.
10677  */
10678 void free_netdev(struct net_device *dev)
10679 {
10680 	struct napi_struct *p, *n;
10681 
10682 	might_sleep();
10683 
10684 	/* When called immediately after register_netdevice() failed the unwind
10685 	 * handling may still be dismantling the device. Handle that case by
10686 	 * deferring the free.
10687 	 */
10688 	if (dev->reg_state == NETREG_UNREGISTERING) {
10689 		ASSERT_RTNL();
10690 		dev->needs_free_netdev = true;
10691 		return;
10692 	}
10693 
10694 	netif_free_tx_queues(dev);
10695 	netif_free_rx_queues(dev);
10696 
10697 	kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10698 
10699 	/* Flush device addresses */
10700 	dev_addr_flush(dev);
10701 
10702 	list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10703 		netif_napi_del(p);
10704 
10705 	ref_tracker_dir_exit(&dev->refcnt_tracker);
10706 #ifdef CONFIG_PCPU_DEV_REFCNT
10707 	free_percpu(dev->pcpu_refcnt);
10708 	dev->pcpu_refcnt = NULL;
10709 #endif
10710 	free_percpu(dev->core_stats);
10711 	dev->core_stats = NULL;
10712 	free_percpu(dev->xdp_bulkq);
10713 	dev->xdp_bulkq = NULL;
10714 
10715 	/*  Compatibility with error handling in drivers */
10716 	if (dev->reg_state == NETREG_UNINITIALIZED) {
10717 		netdev_freemem(dev);
10718 		return;
10719 	}
10720 
10721 	BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10722 	dev->reg_state = NETREG_RELEASED;
10723 
10724 	/* will free via device release */
10725 	put_device(&dev->dev);
10726 }
10727 EXPORT_SYMBOL(free_netdev);
10728 
10729 /**
10730  *	synchronize_net -  Synchronize with packet receive processing
10731  *
10732  *	Wait for packets currently being received to be done.
10733  *	Does not block later packets from starting.
10734  */
10735 void synchronize_net(void)
10736 {
10737 	might_sleep();
10738 	if (rtnl_is_locked())
10739 		synchronize_rcu_expedited();
10740 	else
10741 		synchronize_rcu();
10742 }
10743 EXPORT_SYMBOL(synchronize_net);
10744 
10745 /**
10746  *	unregister_netdevice_queue - remove device from the kernel
10747  *	@dev: device
10748  *	@head: list
10749  *
10750  *	This function shuts down a device interface and removes it
10751  *	from the kernel tables.
10752  *	If head not NULL, device is queued to be unregistered later.
10753  *
10754  *	Callers must hold the rtnl semaphore.  You may want
10755  *	unregister_netdev() instead of this.
10756  */
10757 
10758 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10759 {
10760 	ASSERT_RTNL();
10761 
10762 	if (head) {
10763 		list_move_tail(&dev->unreg_list, head);
10764 	} else {
10765 		LIST_HEAD(single);
10766 
10767 		list_add(&dev->unreg_list, &single);
10768 		unregister_netdevice_many(&single);
10769 	}
10770 }
10771 EXPORT_SYMBOL(unregister_netdevice_queue);
10772 
10773 void unregister_netdevice_many_notify(struct list_head *head,
10774 				      u32 portid, const struct nlmsghdr *nlh)
10775 {
10776 	struct net_device *dev, *tmp;
10777 	LIST_HEAD(close_head);
10778 
10779 	BUG_ON(dev_boot_phase);
10780 	ASSERT_RTNL();
10781 
10782 	if (list_empty(head))
10783 		return;
10784 
10785 	list_for_each_entry_safe(dev, tmp, head, unreg_list) {
10786 		/* Some devices call without registering
10787 		 * for initialization unwind. Remove those
10788 		 * devices and proceed with the remaining.
10789 		 */
10790 		if (dev->reg_state == NETREG_UNINITIALIZED) {
10791 			pr_debug("unregister_netdevice: device %s/%p never was registered\n",
10792 				 dev->name, dev);
10793 
10794 			WARN_ON(1);
10795 			list_del(&dev->unreg_list);
10796 			continue;
10797 		}
10798 		dev->dismantle = true;
10799 		BUG_ON(dev->reg_state != NETREG_REGISTERED);
10800 	}
10801 
10802 	/* If device is running, close it first. */
10803 	list_for_each_entry(dev, head, unreg_list)
10804 		list_add_tail(&dev->close_list, &close_head);
10805 	dev_close_many(&close_head, true);
10806 
10807 	list_for_each_entry(dev, head, unreg_list) {
10808 		/* And unlink it from device chain. */
10809 		write_lock(&dev_base_lock);
10810 		unlist_netdevice(dev, false);
10811 		dev->reg_state = NETREG_UNREGISTERING;
10812 		write_unlock(&dev_base_lock);
10813 	}
10814 	flush_all_backlogs();
10815 
10816 	synchronize_net();
10817 
10818 	list_for_each_entry(dev, head, unreg_list) {
10819 		struct sk_buff *skb = NULL;
10820 
10821 		/* Shutdown queueing discipline. */
10822 		dev_shutdown(dev);
10823 
10824 		dev_xdp_uninstall(dev);
10825 
10826 		netdev_offload_xstats_disable_all(dev);
10827 
10828 		/* Notify protocols, that we are about to destroy
10829 		 * this device. They should clean all the things.
10830 		 */
10831 		call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10832 
10833 		if (!dev->rtnl_link_ops ||
10834 		    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10835 			skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
10836 						     GFP_KERNEL, NULL, 0,
10837 						     portid, nlmsg_seq(nlh));
10838 
10839 		/*
10840 		 *	Flush the unicast and multicast chains
10841 		 */
10842 		dev_uc_flush(dev);
10843 		dev_mc_flush(dev);
10844 
10845 		netdev_name_node_alt_flush(dev);
10846 		netdev_name_node_free(dev->name_node);
10847 
10848 		call_netdevice_notifiers(NETDEV_PRE_UNINIT, dev);
10849 
10850 		if (dev->netdev_ops->ndo_uninit)
10851 			dev->netdev_ops->ndo_uninit(dev);
10852 
10853 		if (skb)
10854 			rtmsg_ifinfo_send(skb, dev, GFP_KERNEL, portid, nlh);
10855 
10856 		/* Notifier chain MUST detach us all upper devices. */
10857 		WARN_ON(netdev_has_any_upper_dev(dev));
10858 		WARN_ON(netdev_has_any_lower_dev(dev));
10859 
10860 		/* Remove entries from kobject tree */
10861 		netdev_unregister_kobject(dev);
10862 #ifdef CONFIG_XPS
10863 		/* Remove XPS queueing entries */
10864 		netif_reset_xps_queues_gt(dev, 0);
10865 #endif
10866 	}
10867 
10868 	synchronize_net();
10869 
10870 	list_for_each_entry(dev, head, unreg_list) {
10871 		netdev_put(dev, &dev->dev_registered_tracker);
10872 		net_set_todo(dev);
10873 	}
10874 
10875 	list_del(head);
10876 }
10877 
10878 /**
10879  *	unregister_netdevice_many - unregister many devices
10880  *	@head: list of devices
10881  *
10882  *  Note: As most callers use a stack allocated list_head,
10883  *  we force a list_del() to make sure stack wont be corrupted later.
10884  */
10885 void unregister_netdevice_many(struct list_head *head)
10886 {
10887 	unregister_netdevice_many_notify(head, 0, NULL);
10888 }
10889 EXPORT_SYMBOL(unregister_netdevice_many);
10890 
10891 /**
10892  *	unregister_netdev - remove device from the kernel
10893  *	@dev: device
10894  *
10895  *	This function shuts down a device interface and removes it
10896  *	from the kernel tables.
10897  *
10898  *	This is just a wrapper for unregister_netdevice that takes
10899  *	the rtnl semaphore.  In general you want to use this and not
10900  *	unregister_netdevice.
10901  */
10902 void unregister_netdev(struct net_device *dev)
10903 {
10904 	rtnl_lock();
10905 	unregister_netdevice(dev);
10906 	rtnl_unlock();
10907 }
10908 EXPORT_SYMBOL(unregister_netdev);
10909 
10910 /**
10911  *	__dev_change_net_namespace - move device to different nethost namespace
10912  *	@dev: device
10913  *	@net: network namespace
10914  *	@pat: If not NULL name pattern to try if the current device name
10915  *	      is already taken in the destination network namespace.
10916  *	@new_ifindex: If not zero, specifies device index in the target
10917  *	              namespace.
10918  *
10919  *	This function shuts down a device interface and moves it
10920  *	to a new network namespace. On success 0 is returned, on
10921  *	a failure a netagive errno code is returned.
10922  *
10923  *	Callers must hold the rtnl semaphore.
10924  */
10925 
10926 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
10927 			       const char *pat, int new_ifindex)
10928 {
10929 	struct net *net_old = dev_net(dev);
10930 	int err, new_nsid;
10931 
10932 	ASSERT_RTNL();
10933 
10934 	/* Don't allow namespace local devices to be moved. */
10935 	err = -EINVAL;
10936 	if (dev->features & NETIF_F_NETNS_LOCAL)
10937 		goto out;
10938 
10939 	/* Ensure the device has been registrered */
10940 	if (dev->reg_state != NETREG_REGISTERED)
10941 		goto out;
10942 
10943 	/* Get out if there is nothing todo */
10944 	err = 0;
10945 	if (net_eq(net_old, net))
10946 		goto out;
10947 
10948 	/* Pick the destination device name, and ensure
10949 	 * we can use it in the destination network namespace.
10950 	 */
10951 	err = -EEXIST;
10952 	if (netdev_name_in_use(net, dev->name)) {
10953 		/* We get here if we can't use the current device name */
10954 		if (!pat)
10955 			goto out;
10956 		err = dev_get_valid_name(net, dev, pat);
10957 		if (err < 0)
10958 			goto out;
10959 	}
10960 
10961 	/* Check that new_ifindex isn't used yet. */
10962 	err = -EBUSY;
10963 	if (new_ifindex && __dev_get_by_index(net, new_ifindex))
10964 		goto out;
10965 
10966 	/*
10967 	 * And now a mini version of register_netdevice unregister_netdevice.
10968 	 */
10969 
10970 	/* If device is running close it first. */
10971 	dev_close(dev);
10972 
10973 	/* And unlink it from device chain */
10974 	unlist_netdevice(dev, true);
10975 
10976 	synchronize_net();
10977 
10978 	/* Shutdown queueing discipline. */
10979 	dev_shutdown(dev);
10980 
10981 	/* Notify protocols, that we are about to destroy
10982 	 * this device. They should clean all the things.
10983 	 *
10984 	 * Note that dev->reg_state stays at NETREG_REGISTERED.
10985 	 * This is wanted because this way 8021q and macvlan know
10986 	 * the device is just moving and can keep their slaves up.
10987 	 */
10988 	call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10989 	rcu_barrier();
10990 
10991 	new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
10992 	/* If there is an ifindex conflict assign a new one */
10993 	if (!new_ifindex) {
10994 		if (__dev_get_by_index(net, dev->ifindex))
10995 			new_ifindex = dev_new_index(net);
10996 		else
10997 			new_ifindex = dev->ifindex;
10998 	}
10999 
11000 	rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11001 			    new_ifindex);
11002 
11003 	/*
11004 	 *	Flush the unicast and multicast chains
11005 	 */
11006 	dev_uc_flush(dev);
11007 	dev_mc_flush(dev);
11008 
11009 	/* Send a netdev-removed uevent to the old namespace */
11010 	kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11011 	netdev_adjacent_del_links(dev);
11012 
11013 	/* Move per-net netdevice notifiers that are following the netdevice */
11014 	move_netdevice_notifiers_dev_net(dev, net);
11015 
11016 	/* Actually switch the network namespace */
11017 	dev_net_set(dev, net);
11018 	dev->ifindex = new_ifindex;
11019 
11020 	/* Send a netdev-add uevent to the new namespace */
11021 	kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11022 	netdev_adjacent_add_links(dev);
11023 
11024 	/* Fixup kobjects */
11025 	err = device_rename(&dev->dev, dev->name);
11026 	WARN_ON(err);
11027 
11028 	/* Adapt owner in case owning user namespace of target network
11029 	 * namespace is different from the original one.
11030 	 */
11031 	err = netdev_change_owner(dev, net_old, net);
11032 	WARN_ON(err);
11033 
11034 	/* Add the device back in the hashes */
11035 	list_netdevice(dev);
11036 
11037 	/* Notify protocols, that a new device appeared. */
11038 	call_netdevice_notifiers(NETDEV_REGISTER, dev);
11039 
11040 	/*
11041 	 *	Prevent userspace races by waiting until the network
11042 	 *	device is fully setup before sending notifications.
11043 	 */
11044 	rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL, 0, NULL);
11045 
11046 	synchronize_net();
11047 	err = 0;
11048 out:
11049 	return err;
11050 }
11051 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11052 
11053 static int dev_cpu_dead(unsigned int oldcpu)
11054 {
11055 	struct sk_buff **list_skb;
11056 	struct sk_buff *skb;
11057 	unsigned int cpu;
11058 	struct softnet_data *sd, *oldsd, *remsd = NULL;
11059 
11060 	local_irq_disable();
11061 	cpu = smp_processor_id();
11062 	sd = &per_cpu(softnet_data, cpu);
11063 	oldsd = &per_cpu(softnet_data, oldcpu);
11064 
11065 	/* Find end of our completion_queue. */
11066 	list_skb = &sd->completion_queue;
11067 	while (*list_skb)
11068 		list_skb = &(*list_skb)->next;
11069 	/* Append completion queue from offline CPU. */
11070 	*list_skb = oldsd->completion_queue;
11071 	oldsd->completion_queue = NULL;
11072 
11073 	/* Append output queue from offline CPU. */
11074 	if (oldsd->output_queue) {
11075 		*sd->output_queue_tailp = oldsd->output_queue;
11076 		sd->output_queue_tailp = oldsd->output_queue_tailp;
11077 		oldsd->output_queue = NULL;
11078 		oldsd->output_queue_tailp = &oldsd->output_queue;
11079 	}
11080 	/* Append NAPI poll list from offline CPU, with one exception :
11081 	 * process_backlog() must be called by cpu owning percpu backlog.
11082 	 * We properly handle process_queue & input_pkt_queue later.
11083 	 */
11084 	while (!list_empty(&oldsd->poll_list)) {
11085 		struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11086 							    struct napi_struct,
11087 							    poll_list);
11088 
11089 		list_del_init(&napi->poll_list);
11090 		if (napi->poll == process_backlog)
11091 			napi->state = 0;
11092 		else
11093 			____napi_schedule(sd, napi);
11094 	}
11095 
11096 	raise_softirq_irqoff(NET_TX_SOFTIRQ);
11097 	local_irq_enable();
11098 
11099 #ifdef CONFIG_RPS
11100 	remsd = oldsd->rps_ipi_list;
11101 	oldsd->rps_ipi_list = NULL;
11102 #endif
11103 	/* send out pending IPI's on offline CPU */
11104 	net_rps_send_ipi(remsd);
11105 
11106 	/* Process offline CPU's input_pkt_queue */
11107 	while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11108 		netif_rx(skb);
11109 		input_queue_head_incr(oldsd);
11110 	}
11111 	while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11112 		netif_rx(skb);
11113 		input_queue_head_incr(oldsd);
11114 	}
11115 
11116 	return 0;
11117 }
11118 
11119 /**
11120  *	netdev_increment_features - increment feature set by one
11121  *	@all: current feature set
11122  *	@one: new feature set
11123  *	@mask: mask feature set
11124  *
11125  *	Computes a new feature set after adding a device with feature set
11126  *	@one to the master device with current feature set @all.  Will not
11127  *	enable anything that is off in @mask. Returns the new feature set.
11128  */
11129 netdev_features_t netdev_increment_features(netdev_features_t all,
11130 	netdev_features_t one, netdev_features_t mask)
11131 {
11132 	if (mask & NETIF_F_HW_CSUM)
11133 		mask |= NETIF_F_CSUM_MASK;
11134 	mask |= NETIF_F_VLAN_CHALLENGED;
11135 
11136 	all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11137 	all &= one | ~NETIF_F_ALL_FOR_ALL;
11138 
11139 	/* If one device supports hw checksumming, set for all. */
11140 	if (all & NETIF_F_HW_CSUM)
11141 		all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11142 
11143 	return all;
11144 }
11145 EXPORT_SYMBOL(netdev_increment_features);
11146 
11147 static struct hlist_head * __net_init netdev_create_hash(void)
11148 {
11149 	int i;
11150 	struct hlist_head *hash;
11151 
11152 	hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11153 	if (hash != NULL)
11154 		for (i = 0; i < NETDEV_HASHENTRIES; i++)
11155 			INIT_HLIST_HEAD(&hash[i]);
11156 
11157 	return hash;
11158 }
11159 
11160 /* Initialize per network namespace state */
11161 static int __net_init netdev_init(struct net *net)
11162 {
11163 	BUILD_BUG_ON(GRO_HASH_BUCKETS >
11164 		     8 * sizeof_field(struct napi_struct, gro_bitmask));
11165 
11166 	INIT_LIST_HEAD(&net->dev_base_head);
11167 
11168 	net->dev_name_head = netdev_create_hash();
11169 	if (net->dev_name_head == NULL)
11170 		goto err_name;
11171 
11172 	net->dev_index_head = netdev_create_hash();
11173 	if (net->dev_index_head == NULL)
11174 		goto err_idx;
11175 
11176 	RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11177 
11178 	return 0;
11179 
11180 err_idx:
11181 	kfree(net->dev_name_head);
11182 err_name:
11183 	return -ENOMEM;
11184 }
11185 
11186 /**
11187  *	netdev_drivername - network driver for the device
11188  *	@dev: network device
11189  *
11190  *	Determine network driver for device.
11191  */
11192 const char *netdev_drivername(const struct net_device *dev)
11193 {
11194 	const struct device_driver *driver;
11195 	const struct device *parent;
11196 	const char *empty = "";
11197 
11198 	parent = dev->dev.parent;
11199 	if (!parent)
11200 		return empty;
11201 
11202 	driver = parent->driver;
11203 	if (driver && driver->name)
11204 		return driver->name;
11205 	return empty;
11206 }
11207 
11208 static void __netdev_printk(const char *level, const struct net_device *dev,
11209 			    struct va_format *vaf)
11210 {
11211 	if (dev && dev->dev.parent) {
11212 		dev_printk_emit(level[1] - '0',
11213 				dev->dev.parent,
11214 				"%s %s %s%s: %pV",
11215 				dev_driver_string(dev->dev.parent),
11216 				dev_name(dev->dev.parent),
11217 				netdev_name(dev), netdev_reg_state(dev),
11218 				vaf);
11219 	} else if (dev) {
11220 		printk("%s%s%s: %pV",
11221 		       level, netdev_name(dev), netdev_reg_state(dev), vaf);
11222 	} else {
11223 		printk("%s(NULL net_device): %pV", level, vaf);
11224 	}
11225 }
11226 
11227 void netdev_printk(const char *level, const struct net_device *dev,
11228 		   const char *format, ...)
11229 {
11230 	struct va_format vaf;
11231 	va_list args;
11232 
11233 	va_start(args, format);
11234 
11235 	vaf.fmt = format;
11236 	vaf.va = &args;
11237 
11238 	__netdev_printk(level, dev, &vaf);
11239 
11240 	va_end(args);
11241 }
11242 EXPORT_SYMBOL(netdev_printk);
11243 
11244 #define define_netdev_printk_level(func, level)			\
11245 void func(const struct net_device *dev, const char *fmt, ...)	\
11246 {								\
11247 	struct va_format vaf;					\
11248 	va_list args;						\
11249 								\
11250 	va_start(args, fmt);					\
11251 								\
11252 	vaf.fmt = fmt;						\
11253 	vaf.va = &args;						\
11254 								\
11255 	__netdev_printk(level, dev, &vaf);			\
11256 								\
11257 	va_end(args);						\
11258 }								\
11259 EXPORT_SYMBOL(func);
11260 
11261 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11262 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11263 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11264 define_netdev_printk_level(netdev_err, KERN_ERR);
11265 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11266 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11267 define_netdev_printk_level(netdev_info, KERN_INFO);
11268 
11269 static void __net_exit netdev_exit(struct net *net)
11270 {
11271 	kfree(net->dev_name_head);
11272 	kfree(net->dev_index_head);
11273 	if (net != &init_net)
11274 		WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11275 }
11276 
11277 static struct pernet_operations __net_initdata netdev_net_ops = {
11278 	.init = netdev_init,
11279 	.exit = netdev_exit,
11280 };
11281 
11282 static void __net_exit default_device_exit_net(struct net *net)
11283 {
11284 	struct net_device *dev, *aux;
11285 	/*
11286 	 * Push all migratable network devices back to the
11287 	 * initial network namespace
11288 	 */
11289 	ASSERT_RTNL();
11290 	for_each_netdev_safe(net, dev, aux) {
11291 		int err;
11292 		char fb_name[IFNAMSIZ];
11293 
11294 		/* Ignore unmoveable devices (i.e. loopback) */
11295 		if (dev->features & NETIF_F_NETNS_LOCAL)
11296 			continue;
11297 
11298 		/* Leave virtual devices for the generic cleanup */
11299 		if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11300 			continue;
11301 
11302 		/* Push remaining network devices to init_net */
11303 		snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11304 		if (netdev_name_in_use(&init_net, fb_name))
11305 			snprintf(fb_name, IFNAMSIZ, "dev%%d");
11306 		err = dev_change_net_namespace(dev, &init_net, fb_name);
11307 		if (err) {
11308 			pr_emerg("%s: failed to move %s to init_net: %d\n",
11309 				 __func__, dev->name, err);
11310 			BUG();
11311 		}
11312 	}
11313 }
11314 
11315 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11316 {
11317 	/* At exit all network devices most be removed from a network
11318 	 * namespace.  Do this in the reverse order of registration.
11319 	 * Do this across as many network namespaces as possible to
11320 	 * improve batching efficiency.
11321 	 */
11322 	struct net_device *dev;
11323 	struct net *net;
11324 	LIST_HEAD(dev_kill_list);
11325 
11326 	rtnl_lock();
11327 	list_for_each_entry(net, net_list, exit_list) {
11328 		default_device_exit_net(net);
11329 		cond_resched();
11330 	}
11331 
11332 	list_for_each_entry(net, net_list, exit_list) {
11333 		for_each_netdev_reverse(net, dev) {
11334 			if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11335 				dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11336 			else
11337 				unregister_netdevice_queue(dev, &dev_kill_list);
11338 		}
11339 	}
11340 	unregister_netdevice_many(&dev_kill_list);
11341 	rtnl_unlock();
11342 }
11343 
11344 static struct pernet_operations __net_initdata default_device_ops = {
11345 	.exit_batch = default_device_exit_batch,
11346 };
11347 
11348 /*
11349  *	Initialize the DEV module. At boot time this walks the device list and
11350  *	unhooks any devices that fail to initialise (normally hardware not
11351  *	present) and leaves us with a valid list of present and active devices.
11352  *
11353  */
11354 
11355 /*
11356  *       This is called single threaded during boot, so no need
11357  *       to take the rtnl semaphore.
11358  */
11359 static int __init net_dev_init(void)
11360 {
11361 	int i, rc = -ENOMEM;
11362 
11363 	BUG_ON(!dev_boot_phase);
11364 
11365 	if (dev_proc_init())
11366 		goto out;
11367 
11368 	if (netdev_kobject_init())
11369 		goto out;
11370 
11371 	INIT_LIST_HEAD(&ptype_all);
11372 	for (i = 0; i < PTYPE_HASH_SIZE; i++)
11373 		INIT_LIST_HEAD(&ptype_base[i]);
11374 
11375 	if (register_pernet_subsys(&netdev_net_ops))
11376 		goto out;
11377 
11378 	/*
11379 	 *	Initialise the packet receive queues.
11380 	 */
11381 
11382 	for_each_possible_cpu(i) {
11383 		struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11384 		struct softnet_data *sd = &per_cpu(softnet_data, i);
11385 
11386 		INIT_WORK(flush, flush_backlog);
11387 
11388 		skb_queue_head_init(&sd->input_pkt_queue);
11389 		skb_queue_head_init(&sd->process_queue);
11390 #ifdef CONFIG_XFRM_OFFLOAD
11391 		skb_queue_head_init(&sd->xfrm_backlog);
11392 #endif
11393 		INIT_LIST_HEAD(&sd->poll_list);
11394 		sd->output_queue_tailp = &sd->output_queue;
11395 #ifdef CONFIG_RPS
11396 		INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11397 		sd->cpu = i;
11398 #endif
11399 		INIT_CSD(&sd->defer_csd, trigger_rx_softirq, sd);
11400 		spin_lock_init(&sd->defer_lock);
11401 
11402 		init_gro_hash(&sd->backlog);
11403 		sd->backlog.poll = process_backlog;
11404 		sd->backlog.weight = weight_p;
11405 	}
11406 
11407 	dev_boot_phase = 0;
11408 
11409 	/* The loopback device is special if any other network devices
11410 	 * is present in a network namespace the loopback device must
11411 	 * be present. Since we now dynamically allocate and free the
11412 	 * loopback device ensure this invariant is maintained by
11413 	 * keeping the loopback device as the first device on the
11414 	 * list of network devices.  Ensuring the loopback devices
11415 	 * is the first device that appears and the last network device
11416 	 * that disappears.
11417 	 */
11418 	if (register_pernet_device(&loopback_net_ops))
11419 		goto out;
11420 
11421 	if (register_pernet_device(&default_device_ops))
11422 		goto out;
11423 
11424 	open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11425 	open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11426 
11427 	rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11428 				       NULL, dev_cpu_dead);
11429 	WARN_ON(rc < 0);
11430 	rc = 0;
11431 out:
11432 	return rc;
11433 }
11434 
11435 subsys_initcall(net_dev_init);
11436